blob: 67099bb99a026168f11ce5af7c13c758b87ad658 [file] [log] [blame]
Jens Axboe2b188cc2019-01-07 10:46:33 -07001// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared application/kernel submission and completion ring pairs, for
4 * supporting fast/efficient IO.
5 *
6 * A note on the read/write ordering memory barriers that are matched between
Stefan Bühler1e84b972019-04-24 23:54:16 +02007 * the application and kernel side.
8 *
9 * After the application reads the CQ ring tail, it must use an
10 * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses
11 * before writing the tail (using smp_load_acquire to read the tail will
12 * do). It also needs a smp_mb() before updating CQ head (ordering the
13 * entry load(s) with the head store), pairing with an implicit barrier
Pavel Begunkovd068b502021-05-16 22:58:11 +010014 * through a control-dependency in io_get_cqe (smp_store_release to
Stefan Bühler1e84b972019-04-24 23:54:16 +020015 * store head will do). Failure to do so could lead to reading invalid
16 * CQ entries.
17 *
18 * Likewise, the application must use an appropriate smp_wmb() before
19 * writing the SQ tail (ordering SQ entry stores with the tail store),
20 * which pairs with smp_load_acquire in io_get_sqring (smp_store_release
21 * to store the tail will do). And it needs a barrier ordering the SQ
22 * head load before writing new SQ entries (smp_load_acquire to read
23 * head will do).
24 *
25 * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application
26 * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after*
27 * updating the SQ tail; a full memory barrier smp_mb() is needed
28 * between.
Jens Axboe2b188cc2019-01-07 10:46:33 -070029 *
30 * Also see the examples in the liburing library:
31 *
32 * git://git.kernel.dk/liburing
33 *
34 * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens
35 * from data shared between the kernel and application. This is done both
36 * for ordering purposes, but also to ensure that once a value is loaded from
37 * data that the application could potentially modify, it remains stable.
38 *
39 * Copyright (C) 2018-2019 Jens Axboe
Christoph Hellwigc992fe22019-01-11 09:43:02 -070040 * Copyright (c) 2018-2019 Christoph Hellwig
Jens Axboe2b188cc2019-01-07 10:46:33 -070041 */
42#include <linux/kernel.h>
43#include <linux/init.h>
44#include <linux/errno.h>
45#include <linux/syscalls.h>
46#include <linux/compat.h>
Jens Axboe52de1fe2020-02-27 10:15:42 -070047#include <net/compat.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070048#include <linux/refcount.h>
49#include <linux/uio.h>
Pavel Begunkov6b47ee62020-01-18 20:22:41 +030050#include <linux/bits.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070051
52#include <linux/sched/signal.h>
53#include <linux/fs.h>
54#include <linux/file.h>
55#include <linux/fdtable.h>
56#include <linux/mm.h>
57#include <linux/mman.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070058#include <linux/percpu.h>
59#include <linux/slab.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070060#include <linux/blkdev.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070061#include <linux/bvec.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070062#include <linux/net.h>
63#include <net/sock.h>
64#include <net/af_unix.h>
Jens Axboe6b063142019-01-10 22:13:58 -070065#include <net/scm.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070066#include <linux/anon_inodes.h>
67#include <linux/sched/mm.h>
68#include <linux/uaccess.h>
69#include <linux/nospec.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070070#include <linux/sizes.h>
71#include <linux/hugetlb.h>
Jens Axboeaa4c3962019-11-29 10:14:00 -070072#include <linux/highmem.h>
Jens Axboe15b71ab2019-12-11 11:20:36 -070073#include <linux/namei.h>
74#include <linux/fsnotify.h>
Jens Axboe4840e412019-12-25 22:03:45 -070075#include <linux/fadvise.h>
Jens Axboe3e4827b2020-01-08 15:18:09 -070076#include <linux/eventpoll.h>
Pavel Begunkov7d67af22020-02-24 11:32:45 +030077#include <linux/splice.h>
Jens Axboeb41e9852020-02-17 09:52:41 -070078#include <linux/task_work.h>
Jens Axboebcf5a062020-05-22 09:24:42 -060079#include <linux/pagemap.h>
Jens Axboe0f212202020-09-13 13:09:39 -060080#include <linux/io_uring.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070081
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020082#define CREATE_TRACE_POINTS
83#include <trace/events/io_uring.h>
84
Jens Axboe2b188cc2019-01-07 10:46:33 -070085#include <uapi/linux/io_uring.h>
86
87#include "internal.h"
Jens Axboe561fb042019-10-24 07:25:42 -060088#include "io-wq.h"
Jens Axboe2b188cc2019-01-07 10:46:33 -070089
Daniel Xu5277dea2019-09-14 14:23:45 -070090#define IORING_MAX_ENTRIES 32768
Jens Axboe33a107f2019-10-04 12:10:03 -060091#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
Olivier Langlois4ce8ad92021-06-23 11:50:18 -070092#define IORING_SQPOLL_CAP_ENTRIES_VALUE 8
Jens Axboe65e19f52019-10-26 07:20:21 -060093
94/*
95 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
96 */
97#define IORING_FILE_TABLE_SHIFT 9
98#define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT)
99#define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1)
100#define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE)
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200101#define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \
102 IORING_REGISTER_LAST + IORING_OP_LAST)
Jens Axboe2b188cc2019-01-07 10:46:33 -0700103
Pavel Begunkov2d091d62021-06-14 02:36:21 +0100104#define IO_RSRC_TAG_TABLE_SHIFT 9
105#define IO_RSRC_TAG_TABLE_MAX (1U << IO_RSRC_TAG_TABLE_SHIFT)
106#define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1)
107
Pavel Begunkov489809e2021-05-14 12:06:44 +0100108#define IORING_MAX_REG_BUFFERS (1U << 14)
109
Pavel Begunkovb16fed662021-02-18 18:29:40 +0000110#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
111 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
112 IOSQE_BUFFER_SELECT)
Pavel Begunkovc8543572021-06-17 18:14:04 +0100113#define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \
114 REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS)
Pavel Begunkovb16fed662021-02-18 18:29:40 +0000115
Pavel Begunkov09899b12021-06-14 02:36:22 +0100116#define IO_TCTX_REFS_CACHE_NR (1U << 10)
117
Jens Axboe2b188cc2019-01-07 10:46:33 -0700118struct io_uring {
119 u32 head ____cacheline_aligned_in_smp;
120 u32 tail ____cacheline_aligned_in_smp;
121};
122
Stefan Bühler1e84b972019-04-24 23:54:16 +0200123/*
Hristo Venev75b28af2019-08-26 17:23:46 +0000124 * This data is shared with the application through the mmap at offsets
125 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200126 *
127 * The offsets to the member fields are published through struct
128 * io_sqring_offsets when calling io_uring_setup.
129 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000130struct io_rings {
Stefan Bühler1e84b972019-04-24 23:54:16 +0200131 /*
132 * Head and tail offsets into the ring; the offsets need to be
133 * masked to get valid indices.
134 *
Hristo Venev75b28af2019-08-26 17:23:46 +0000135 * The kernel controls head of the sq ring and the tail of the cq ring,
136 * and the application controls tail of the sq ring and the head of the
137 * cq ring.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200138 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000139 struct io_uring sq, cq;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200140 /*
Hristo Venev75b28af2019-08-26 17:23:46 +0000141 * Bitmasks to apply to head and tail offsets (constant, equals
Stefan Bühler1e84b972019-04-24 23:54:16 +0200142 * ring_entries - 1)
143 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000144 u32 sq_ring_mask, cq_ring_mask;
145 /* Ring sizes (constant, power of 2) */
146 u32 sq_ring_entries, cq_ring_entries;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200147 /*
148 * Number of invalid entries dropped by the kernel due to
149 * invalid index stored in array
150 *
151 * Written by the kernel, shouldn't be modified by the
152 * application (i.e. get number of "new events" by comparing to
153 * cached value).
154 *
155 * After a new SQ head value was read by the application this
156 * counter includes all submissions that were dropped reaching
157 * the new SQ head (and possibly more).
158 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000159 u32 sq_dropped;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200160 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200161 * Runtime SQ flags
Stefan Bühler1e84b972019-04-24 23:54:16 +0200162 *
163 * Written by the kernel, shouldn't be modified by the
164 * application.
165 *
166 * The application needs a full memory barrier before checking
167 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
168 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000169 u32 sq_flags;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200170 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200171 * Runtime CQ flags
172 *
173 * Written by the application, shouldn't be modified by the
174 * kernel.
175 */
Pavel Begunkovfe7e3252021-06-24 15:09:57 +0100176 u32 cq_flags;
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200177 /*
Stefan Bühler1e84b972019-04-24 23:54:16 +0200178 * Number of completion events lost because the queue was full;
179 * this should be avoided by the application by making sure
LimingWu0b4295b2019-12-05 20:18:18 +0800180 * there are not more requests pending than there is space in
Stefan Bühler1e84b972019-04-24 23:54:16 +0200181 * the completion queue.
182 *
183 * Written by the kernel, shouldn't be modified by the
184 * application (i.e. get number of "new events" by comparing to
185 * cached value).
186 *
187 * As completion events come in out of order this counter is not
188 * ordered with any other data.
189 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000190 u32 cq_overflow;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200191 /*
192 * Ring buffer of completion events.
193 *
194 * The kernel writes completion events fresh every time they are
195 * produced, so the application is allowed to modify pending
196 * entries.
197 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000198 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700199};
200
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000201enum io_uring_cmd_flags {
202 IO_URING_F_NONBLOCK = 1,
Pavel Begunkov889fca72021-02-10 00:03:09 +0000203 IO_URING_F_COMPLETE_DEFER = 2,
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000204};
205
Jens Axboeedafcce2019-01-09 09:16:05 -0700206struct io_mapped_ubuf {
207 u64 ubuf;
Pavel Begunkov4751f532021-04-01 15:43:55 +0100208 u64 ubuf_end;
Jens Axboeedafcce2019-01-09 09:16:05 -0700209 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600210 unsigned long acct_pages;
Pavel Begunkov41edf1a2021-04-25 14:32:23 +0100211 struct bio_vec bvec[];
Jens Axboeedafcce2019-01-09 09:16:05 -0700212};
213
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000214struct io_ring_ctx;
215
Pavel Begunkov6c2450a2021-02-23 12:40:22 +0000216struct io_overflow_cqe {
217 struct io_uring_cqe cqe;
218 struct list_head list;
219};
220
Pavel Begunkova04b0ac2021-04-01 15:44:04 +0100221struct io_fixed_file {
222 /* file * with additional FFS_* flags */
223 unsigned long file_ptr;
224};
225
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000226struct io_rsrc_put {
227 struct list_head list;
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +0100228 u64 tag;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000229 union {
230 void *rsrc;
231 struct file *file;
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +0100232 struct io_mapped_ubuf *buf;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000233 };
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000234};
235
Pavel Begunkovaeca2412021-04-11 01:46:37 +0100236struct io_file_table {
237 /* two level table */
238 struct io_fixed_file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700239};
240
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100241struct io_rsrc_node {
Xiaoguang Wang05589552020-03-31 14:05:18 +0800242 struct percpu_ref refs;
243 struct list_head node;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000244 struct list_head rsrc_list;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100245 struct io_rsrc_data *rsrc_data;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600246 struct llist_node llist;
Pavel Begunkove2978222020-11-18 14:56:26 +0000247 bool done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800248};
249
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +0100250typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc);
251
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100252struct io_rsrc_data {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700253 struct io_ring_ctx *ctx;
254
Pavel Begunkov2d091d62021-06-14 02:36:21 +0100255 u64 **tags;
256 unsigned int nr;
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +0100257 rsrc_put_fn *do_put;
Pavel Begunkov3e942492021-04-11 01:46:34 +0100258 atomic_t refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700259 struct completion done;
Hao Xu8bad28d2021-02-19 17:19:36 +0800260 bool quiesce;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700261};
262
Jens Axboe5a2e7452020-02-23 16:23:11 -0700263struct io_buffer {
264 struct list_head list;
265 __u64 addr;
Thadeu Lima de Souza Cascardod1f82802021-05-05 09:47:06 -0300266 __u32 len;
Jens Axboe5a2e7452020-02-23 16:23:11 -0700267 __u16 bid;
268};
269
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200270struct io_restriction {
271 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
272 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
273 u8 sqe_flags_allowed;
274 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200275 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200276};
277
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700278enum {
279 IO_SQ_THREAD_SHOULD_STOP = 0,
280 IO_SQ_THREAD_SHOULD_PARK,
281};
282
Jens Axboe534ca6d2020-09-02 13:52:19 -0600283struct io_sq_data {
284 refcount_t refs;
Pavel Begunkov9e138a42021-03-14 20:57:12 +0000285 atomic_t park_pending;
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +0000286 struct mutex lock;
Jens Axboe69fb2132020-09-14 11:16:23 -0600287
288 /* ctx's that are using this sqd */
289 struct list_head ctx_list;
Jens Axboe69fb2132020-09-14 11:16:23 -0600290
Jens Axboe534ca6d2020-09-02 13:52:19 -0600291 struct task_struct *thread;
292 struct wait_queue_head wait;
Xiaoguang Wang08369242020-11-03 14:15:59 +0800293
294 unsigned sq_thread_idle;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700295 int sq_cpu;
296 pid_t task_pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -0700297 pid_t task_tgid;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700298
299 unsigned long state;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700300 struct completion exited;
Jens Axboe534ca6d2020-09-02 13:52:19 -0600301};
302
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000303#define IO_IOPOLL_BATCH 8
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000304#define IO_COMPL_BATCH 32
Pavel Begunkov6ff119a2021-02-10 00:03:18 +0000305#define IO_REQ_CACHE_SIZE 32
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000306#define IO_REQ_ALLOC_BATCH 8
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000307
308struct io_comp_state {
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000309 struct io_kiocb *reqs[IO_COMPL_BATCH];
Jens Axboe1b4c3512021-02-10 00:03:19 +0000310 unsigned int nr;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700311 /* inline/task_work completion list, under ->uring_lock */
Jens Axboe1b4c3512021-02-10 00:03:19 +0000312 struct list_head free_list;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000313};
314
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000315struct io_submit_link {
316 struct io_kiocb *head;
317 struct io_kiocb *last;
318};
319
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000320struct io_submit_state {
321 struct blk_plug plug;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000322 struct io_submit_link link;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000323
324 /*
325 * io_kiocb alloc cache
326 */
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000327 void *reqs[IO_REQ_CACHE_SIZE];
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000328 unsigned int free_reqs;
329
330 bool plug_started;
331
332 /*
333 * Batch completion logic
334 */
335 struct io_comp_state comp;
336
337 /*
338 * File reference cache
339 */
340 struct file *file;
341 unsigned int fd;
342 unsigned int file_refs;
343 unsigned int ios_left;
344};
345
Jens Axboe2b188cc2019-01-07 10:46:33 -0700346struct io_ring_ctx {
Pavel Begunkovb52ecf82021-06-14 23:37:21 +0100347 /* const or read-mostly hot data */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700348 struct {
349 struct percpu_ref refs;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700350
Pavel Begunkovb52ecf82021-06-14 23:37:21 +0100351 struct io_rings *rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700352 unsigned int flags;
Randy Dunlape1d85332020-02-05 20:57:10 -0800353 unsigned int compat: 1;
Randy Dunlape1d85332020-02-05 20:57:10 -0800354 unsigned int drain_next: 1;
355 unsigned int eventfd_async: 1;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200356 unsigned int restricted: 1;
Pavel Begunkovf18ee4c2021-06-14 23:37:25 +0100357 unsigned int off_timeout_used: 1;
Pavel Begunkov10c66902021-06-15 16:47:56 +0100358 unsigned int drain_active: 1;
Pavel Begunkovb52ecf82021-06-14 23:37:21 +0100359 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700360
Pavel Begunkov7f1129d2021-06-14 23:37:22 +0100361 /* submission data */
Pavel Begunkovb52ecf82021-06-14 23:37:21 +0100362 struct {
Pavel Begunkov0499e582021-06-14 23:37:29 +0100363 struct mutex uring_lock;
364
Hristo Venev75b28af2019-08-26 17:23:46 +0000365 /*
366 * Ring buffer of indices into array of io_uring_sqe, which is
367 * mmapped by the application using the IORING_OFF_SQES offset.
368 *
369 * This indirection could e.g. be used to assign fixed
370 * io_uring_sqe entries to operations and only submit them to
371 * the queue when needed.
372 *
373 * The kernel modifies neither the indices array nor the entries
374 * array.
375 */
376 u32 *sq_array;
Pavel Begunkovc7af47c2021-06-14 23:37:20 +0100377 struct io_uring_sqe *sq_sqes;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700378 unsigned cached_sq_head;
379 unsigned sq_entries;
Jens Axboede0617e2019-04-06 21:51:27 -0600380 struct list_head defer_list;
Pavel Begunkov7f1129d2021-06-14 23:37:22 +0100381
382 /*
383 * Fixed resources fast path, should be accessed only under
384 * uring_lock, and updated through io_uring_register(2)
385 */
386 struct io_rsrc_node *rsrc_node;
387 struct io_file_table file_table;
388 unsigned nr_user_files;
389 unsigned nr_user_bufs;
390 struct io_mapped_ubuf **user_bufs;
391
392 struct io_submit_state submit_state;
Jens Axboe5262f562019-09-17 12:26:57 -0600393 struct list_head timeout_list;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -0700394 struct list_head cq_overflow_list;
Pavel Begunkov7f1129d2021-06-14 23:37:22 +0100395 struct xarray io_buffers;
396 struct xarray personalities;
397 u32 pers_next;
398 unsigned sq_thread_idle;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700399 } ____cacheline_aligned_in_smp;
400
Pavel Begunkovd0acdee2021-05-16 22:58:12 +0100401 /* IRQ completion list, under ->completion_lock */
402 struct list_head locked_free_list;
403 unsigned int locked_free_nr;
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700404
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +0100405 const struct cred *sq_creds; /* cred used for __io_sq_thread() */
Jens Axboe534ca6d2020-09-02 13:52:19 -0600406 struct io_sq_data *sq_data; /* if using sq thread polling */
407
Jens Axboe90554202020-09-03 12:12:41 -0600408 struct wait_queue_head sqo_sq_wait;
Jens Axboe69fb2132020-09-14 11:16:23 -0600409 struct list_head sqd_list;
Hristo Venev75b28af2019-08-26 17:23:46 +0000410
Pavel Begunkov5ed7a372021-06-14 23:37:27 +0100411 unsigned long check_cq_overflow;
412
Jens Axboe206aefd2019-11-07 18:27:42 -0700413 struct {
414 unsigned cached_cq_tail;
415 unsigned cq_entries;
Jens Axboe206aefd2019-11-07 18:27:42 -0700416 struct eventfd_ctx *cq_ev_fd;
Pavel Begunkov0499e582021-06-14 23:37:29 +0100417 struct wait_queue_head poll_wait;
418 struct wait_queue_head cq_wait;
419 unsigned cq_extra;
420 atomic_t cq_timeouts;
421 struct fasync_struct *cq_fasync;
422 unsigned cq_last_tm_flush;
Jens Axboe206aefd2019-11-07 18:27:42 -0700423 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700424
425 struct {
Jens Axboe2b188cc2019-01-07 10:46:33 -0700426 spinlock_t completion_lock;
Jens Axboee94f1412019-12-19 12:06:02 -0700427
Jens Axboedef596e2019-01-09 08:59:42 -0700428 /*
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300429 * ->iopoll_list is protected by the ctx->uring_lock for
Jens Axboedef596e2019-01-09 08:59:42 -0700430 * io_uring instances that don't use IORING_SETUP_SQPOLL.
431 * For SQPOLL, only the single threaded io_sq_thread() will
432 * manipulate the list, hence no extra locking is needed there.
433 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300434 struct list_head iopoll_list;
Jens Axboe78076bb2019-12-04 19:56:40 -0700435 struct hlist_head *cancel_hash;
436 unsigned cancel_hash_bits;
Hao Xu915b3dd2021-06-28 05:37:30 +0800437 bool poll_multi_queue;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700438 } ____cacheline_aligned_in_smp;
Jens Axboe85faa7b2020-04-09 18:14:00 -0600439
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200440 struct io_restriction restrictions;
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700441
Pavel Begunkovb13a8912021-05-16 22:58:07 +0100442 /* slow path rsrc auxilary data, used by update/register */
443 struct {
444 struct io_rsrc_node *rsrc_backup_node;
445 struct io_mapped_ubuf *dummy_ubuf;
446 struct io_rsrc_data *file_data;
447 struct io_rsrc_data *buf_data;
448
449 struct delayed_work rsrc_put_work;
450 struct llist_head rsrc_put_llist;
451 struct list_head rsrc_ref_list;
452 spinlock_t rsrc_ref_lock;
453 };
454
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700455 /* Keep this last, we don't need it for the fast path */
Pavel Begunkovb986af72021-05-16 22:58:06 +0100456 struct {
457 #if defined(CONFIG_UNIX)
458 struct socket *ring_sock;
459 #endif
460 /* hashed buffered write serialization */
461 struct io_wq_hash *hash_map;
462
463 /* Only used for accounting purposes */
464 struct user_struct *user;
465 struct mm_struct *mm_account;
466
467 /* ctx exit and cancelation */
468 struct callback_head *exit_task_work;
469 struct work_struct exit_work;
470 struct list_head tctx_list;
471 struct completion ref_comp;
472 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700473};
474
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100475struct io_uring_task {
476 /* submission side */
Pavel Begunkov09899b12021-06-14 02:36:22 +0100477 int cached_refs;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100478 struct xarray xa;
479 struct wait_queue_head wait;
Stefan Metzmacheree53fb22021-03-15 12:56:57 +0100480 const struct io_ring_ctx *last;
481 struct io_wq *io_wq;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100482 struct percpu_counter inflight;
Pavel Begunkovb303fe22021-04-11 01:46:26 +0100483 atomic_t inflight_tracked;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100484 atomic_t in_idle;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100485
486 spinlock_t task_lock;
487 struct io_wq_work_list task_list;
488 unsigned long task_state;
489 struct callback_head task_work;
490};
491
Jens Axboe09bb8392019-03-13 12:39:28 -0600492/*
493 * First field must be the file pointer in all the
494 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
495 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700496struct io_poll_iocb {
497 struct file *file;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000498 struct wait_queue_head *head;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700499 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600500 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700501 bool canceled;
Jens Axboe392edb42019-12-09 17:52:20 -0700502 struct wait_queue_entry wait;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700503};
504
Pavel Begunkov9d805892021-04-13 02:58:40 +0100505struct io_poll_update {
Pavel Begunkov018043b2020-10-27 23:17:18 +0000506 struct file *file;
Pavel Begunkov9d805892021-04-13 02:58:40 +0100507 u64 old_user_data;
508 u64 new_user_data;
509 __poll_t events;
Jens Axboeb69de282021-03-17 08:37:41 -0600510 bool update_events;
511 bool update_user_data;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000512};
513
Jens Axboeb5dba592019-12-11 14:02:38 -0700514struct io_close {
515 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700516 int fd;
517};
518
Jens Axboead8a48a2019-11-15 08:49:11 -0700519struct io_timeout_data {
520 struct io_kiocb *req;
521 struct hrtimer timer;
522 struct timespec64 ts;
523 enum hrtimer_mode mode;
524};
525
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700526struct io_accept {
527 struct file *file;
528 struct sockaddr __user *addr;
529 int __user *addr_len;
530 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600531 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700532};
533
534struct io_sync {
535 struct file *file;
536 loff_t len;
537 loff_t off;
538 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700539 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700540};
541
Jens Axboefbf23842019-12-17 18:45:56 -0700542struct io_cancel {
543 struct file *file;
544 u64 addr;
545};
546
Jens Axboeb29472e2019-12-17 18:50:29 -0700547struct io_timeout {
548 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300549 u32 off;
550 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300551 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000552 /* head of the link, used by linked timeouts only */
553 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700554};
555
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100556struct io_timeout_rem {
557 struct file *file;
558 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000559
560 /* timeout update */
561 struct timespec64 ts;
562 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100563};
564
Jens Axboe9adbd452019-12-20 08:45:55 -0700565struct io_rw {
566 /* NOTE: kiocb has the file as the first member, so don't do it here */
567 struct kiocb kiocb;
568 u64 addr;
569 u64 len;
570};
571
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700572struct io_connect {
573 struct file *file;
574 struct sockaddr __user *addr;
575 int addr_len;
576};
577
Jens Axboee47293f2019-12-20 08:58:21 -0700578struct io_sr_msg {
579 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700580 union {
Pavel Begunkov4af34172021-04-11 01:46:30 +0100581 struct compat_msghdr __user *umsg_compat;
582 struct user_msghdr __user *umsg;
583 void __user *buf;
Jens Axboefddafac2020-01-04 20:19:44 -0700584 };
Jens Axboee47293f2019-12-20 08:58:21 -0700585 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700586 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700587 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700588 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700589};
590
Jens Axboe15b71ab2019-12-11 11:20:36 -0700591struct io_open {
592 struct file *file;
593 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700594 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700595 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600596 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700597};
598
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000599struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700600 struct file *file;
601 u64 arg;
602 u32 nr_args;
603 u32 offset;
604};
605
Jens Axboe4840e412019-12-25 22:03:45 -0700606struct io_fadvise {
607 struct file *file;
608 u64 offset;
609 u32 len;
610 u32 advice;
611};
612
Jens Axboec1ca7572019-12-25 22:18:28 -0700613struct io_madvise {
614 struct file *file;
615 u64 addr;
616 u32 len;
617 u32 advice;
618};
619
Jens Axboe3e4827b2020-01-08 15:18:09 -0700620struct io_epoll {
621 struct file *file;
622 int epfd;
623 int op;
624 int fd;
625 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700626};
627
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300628struct io_splice {
629 struct file *file_out;
630 struct file *file_in;
631 loff_t off_out;
632 loff_t off_in;
633 u64 len;
634 unsigned int flags;
635};
636
Jens Axboeddf0322d2020-02-23 16:41:33 -0700637struct io_provide_buf {
638 struct file *file;
639 __u64 addr;
Pavel Begunkov38134ad2021-04-15 13:07:39 +0100640 __u32 len;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700641 __u32 bgid;
642 __u16 nbufs;
643 __u16 bid;
644};
645
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700646struct io_statx {
647 struct file *file;
648 int dfd;
649 unsigned int mask;
650 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700651 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700652 struct statx __user *buffer;
653};
654
Jens Axboe36f4fa62020-09-05 11:14:22 -0600655struct io_shutdown {
656 struct file *file;
657 int how;
658};
659
Jens Axboe80a261f2020-09-28 14:23:58 -0600660struct io_rename {
661 struct file *file;
662 int old_dfd;
663 int new_dfd;
664 struct filename *oldpath;
665 struct filename *newpath;
666 int flags;
667};
668
Jens Axboe14a11432020-09-28 14:27:37 -0600669struct io_unlink {
670 struct file *file;
671 int dfd;
672 int flags;
673 struct filename *filename;
674};
675
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300676struct io_completion {
677 struct file *file;
678 struct list_head list;
Pavel Begunkov8c3f9cd2021-02-28 22:35:15 +0000679 u32 cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300680};
681
Jens Axboef499a022019-12-02 16:28:46 -0700682struct io_async_connect {
683 struct sockaddr_storage address;
684};
685
Jens Axboe03b12302019-12-02 18:50:25 -0700686struct io_async_msghdr {
687 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000688 /* points to an allocated iov, if NULL we use fast_iov instead */
689 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700690 struct sockaddr __user *uaddr;
691 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700692 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700693};
694
Jens Axboef67676d2019-12-02 11:03:47 -0700695struct io_async_rw {
696 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600697 const struct iovec *free_iovec;
698 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600699 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600700 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700701};
702
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300703enum {
704 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
705 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
706 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
707 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
708 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700709 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300710
Pavel Begunkovdddca222021-04-27 16:13:52 +0100711 /* first byte is taken by user flags, shift it to not overlap */
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +0100712 REQ_F_FAIL_BIT = 8,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300713 REQ_F_INFLIGHT_BIT,
714 REQ_F_CUR_POS_BIT,
715 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300716 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300717 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700718 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700719 REQ_F_BUFFER_SELECTED_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100720 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000721 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe230d50d2021-04-01 20:41:15 -0600722 REQ_F_REISSUE_BIT,
Pavel Begunkov8c130822021-03-22 01:58:32 +0000723 REQ_F_DONT_REISSUE_BIT,
Pavel Begunkovb8e64b52021-06-17 18:14:02 +0100724 REQ_F_CREDS_BIT,
Jens Axboe7b29f922021-03-12 08:30:14 -0700725 /* keep async read/write and isreg together and in order */
726 REQ_F_ASYNC_READ_BIT,
727 REQ_F_ASYNC_WRITE_BIT,
728 REQ_F_ISREG_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700729
730 /* not a real bit, just to check we're not overflowing the space */
731 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300732};
733
734enum {
735 /* ctx owns file */
736 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
737 /* drain existing IO first */
738 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
739 /* linked sqes */
740 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
741 /* doesn't sever on completion < 0 */
742 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
743 /* IOSQE_ASYNC */
744 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700745 /* IOSQE_BUFFER_SELECT */
746 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300747
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300748 /* fail rest of links */
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +0100749 REQ_F_FAIL = BIT(REQ_F_FAIL_BIT),
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +0000750 /* on inflight list, should be cancelled and waited on exit reliably */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300751 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
752 /* read/write uses file position */
753 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
754 /* must not punt to workers */
755 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100756 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300757 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300758 /* needs cleanup */
759 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700760 /* already went through poll handler */
761 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700762 /* buffer already selected */
763 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100764 /* linked timeout is active, i.e. prepared by link's head */
765 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000766 /* completion is deferred through io_comp_state */
767 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboe230d50d2021-04-01 20:41:15 -0600768 /* caller should reissue async */
769 REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT),
Pavel Begunkov8c130822021-03-22 01:58:32 +0000770 /* don't attempt request reissue, see io_rw_reissue() */
771 REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT),
Jens Axboe7b29f922021-03-12 08:30:14 -0700772 /* supports async reads */
773 REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT),
774 /* supports async writes */
775 REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT),
776 /* regular file */
777 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Pavel Begunkovb8e64b52021-06-17 18:14:02 +0100778 /* has creds assigned */
779 REQ_F_CREDS = BIT(REQ_F_CREDS_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700780};
781
782struct async_poll {
783 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600784 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300785};
786
Jens Axboe7cbf1722021-02-10 00:03:20 +0000787struct io_task_work {
788 struct io_wq_work_node node;
789 task_work_func_t func;
790};
791
Pavel Begunkov992da012021-06-10 16:37:37 +0100792enum {
793 IORING_RSRC_FILE = 0,
794 IORING_RSRC_BUFFER = 1,
795};
796
Jens Axboe09bb8392019-03-13 12:39:28 -0600797/*
798 * NOTE! Each of the iocb union members has the file pointer
799 * as the first entry in their struct definition. So you can
800 * access the file pointer through any of the sub-structs,
801 * or directly as just 'ki_filp' in this struct.
802 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700803struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700804 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600805 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700806 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700807 struct io_poll_iocb poll;
Pavel Begunkov9d805892021-04-13 02:58:40 +0100808 struct io_poll_update poll_update;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700809 struct io_accept accept;
810 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700811 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700812 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100813 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700814 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700815 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700816 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700817 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000818 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700819 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700820 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700821 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300822 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700823 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700824 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600825 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600826 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600827 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300828 /* use only after cleaning per-op data, see io_clean_op() */
829 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700830 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700831
Jens Axboee8c2bc12020-08-15 18:44:09 -0700832 /* opcode allocated if it needs to store data for async defer */
833 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700834 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800835 /* polled IO has completed */
836 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700837
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700838 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300839 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700840
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300841 struct io_ring_ctx *ctx;
842 unsigned int flags;
Jens Axboeabc54d62021-02-24 13:32:30 -0700843 atomic_t refs;
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300844 struct task_struct *task;
845 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700846
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000847 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000848 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700849
Pavel Begunkovb303fe22021-04-11 01:46:26 +0100850 /* used with ctx->iopoll_list with reads/writes */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300851 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000852 union {
853 struct io_task_work io_task_work;
854 struct callback_head task_work;
855 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300856 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
857 struct hlist_node hash_node;
858 struct async_poll *apoll;
859 struct io_wq_work work;
Pavel Begunkovfe7e3252021-06-24 15:09:57 +0100860 const struct cred *creds;
Pavel Begunkovc10d1f92021-06-17 18:14:01 +0100861
Pavel Begunkoveae071c2021-04-25 14:32:24 +0100862 /* store used ubuf, so we can prevent reloading */
863 struct io_mapped_ubuf *imu;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700864};
865
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000866struct io_tctx_node {
867 struct list_head ctx_node;
868 struct task_struct *task;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000869 struct io_ring_ctx *ctx;
870};
871
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300872struct io_defer_entry {
873 struct list_head list;
874 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300875 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300876};
877
Jens Axboed3656342019-12-18 09:50:26 -0700878struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700879 /* needs req->file assigned */
880 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700881 /* hash wq insertion if file is a regular file */
882 unsigned hash_reg_file : 1;
883 /* unbound wq insertion if file is a non-regular file */
884 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700885 /* opcode is not supported by this kernel */
886 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700887 /* set if opcode supports polled "wait" */
888 unsigned pollin : 1;
889 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700890 /* op supports buffer selection */
891 unsigned buffer_select : 1;
Pavel Begunkov26f05052021-02-28 22:35:18 +0000892 /* do prep async if is going to be punted */
893 unsigned needs_async_setup : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600894 /* should block plug */
895 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700896 /* size of async data needed, if any */
897 unsigned short async_size;
Jens Axboed3656342019-12-18 09:50:26 -0700898};
899
Jens Axboe09186822020-10-13 15:01:40 -0600900static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300901 [IORING_OP_NOP] = {},
902 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700903 .needs_file = 1,
904 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700905 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700906 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000907 .needs_async_setup = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600908 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700909 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700910 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300911 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700912 .needs_file = 1,
913 .hash_reg_file = 1,
914 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700915 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000916 .needs_async_setup = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600917 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700918 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700919 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300920 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700921 .needs_file = 1,
922 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300923 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700924 .needs_file = 1,
925 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700926 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600927 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700928 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700929 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300930 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700931 .needs_file = 1,
932 .hash_reg_file = 1,
933 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700934 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600935 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700936 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700937 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300938 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700939 .needs_file = 1,
940 .unbound_nonreg_file = 1,
941 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300942 [IORING_OP_POLL_REMOVE] = {},
943 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700944 .needs_file = 1,
945 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300946 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700947 .needs_file = 1,
948 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700949 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000950 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700951 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700952 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300953 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700954 .needs_file = 1,
955 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700956 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700957 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000958 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700959 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700960 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300961 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700962 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700963 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000964 [IORING_OP_TIMEOUT_REMOVE] = {
965 /* used by timeout updates' prep() */
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000966 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300967 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700968 .needs_file = 1,
969 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700970 .pollin = 1,
Jens Axboed3656342019-12-18 09:50:26 -0700971 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300972 [IORING_OP_ASYNC_CANCEL] = {},
973 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700974 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700975 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300976 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700977 .needs_file = 1,
978 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700979 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000980 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700981 .async_size = sizeof(struct io_async_connect),
Jens Axboed3656342019-12-18 09:50:26 -0700982 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300983 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700984 .needs_file = 1,
985 },
Jens Axboe44526be2021-02-15 13:32:18 -0700986 [IORING_OP_OPENAT] = {},
987 [IORING_OP_CLOSE] = {},
988 [IORING_OP_FILES_UPDATE] = {},
989 [IORING_OP_STATX] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300990 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700991 .needs_file = 1,
992 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700993 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700994 .buffer_select = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600995 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700996 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700997 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300998 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700999 .needs_file = 1,
1000 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -07001001 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -06001002 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -07001003 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -07001004 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +03001005 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -07001006 .needs_file = 1,
1007 },
Jens Axboe44526be2021-02-15 13:32:18 -07001008 [IORING_OP_MADVISE] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +03001009 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -07001010 .needs_file = 1,
1011 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -07001012 .pollout = 1,
Jens Axboefddafac2020-01-04 20:19:44 -07001013 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +03001014 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -07001015 .needs_file = 1,
1016 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -07001017 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -07001018 .buffer_select = 1,
Jens Axboefddafac2020-01-04 20:19:44 -07001019 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +03001020 [IORING_OP_OPENAT2] = {
Jens Axboecebdb982020-01-08 17:59:24 -07001021 },
Jens Axboe3e4827b2020-01-08 15:18:09 -07001022 [IORING_OP_EPOLL_CTL] = {
1023 .unbound_nonreg_file = 1,
Jens Axboe3e4827b2020-01-08 15:18:09 -07001024 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +03001025 [IORING_OP_SPLICE] = {
1026 .needs_file = 1,
1027 .hash_reg_file = 1,
1028 .unbound_nonreg_file = 1,
Jens Axboeddf0322d2020-02-23 16:41:33 -07001029 },
1030 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -07001031 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03001032 [IORING_OP_TEE] = {
1033 .needs_file = 1,
1034 .hash_reg_file = 1,
1035 .unbound_nonreg_file = 1,
1036 },
Jens Axboe36f4fa62020-09-05 11:14:22 -06001037 [IORING_OP_SHUTDOWN] = {
1038 .needs_file = 1,
1039 },
Jens Axboe44526be2021-02-15 13:32:18 -07001040 [IORING_OP_RENAMEAT] = {},
1041 [IORING_OP_UNLINKAT] = {},
Jens Axboed3656342019-12-18 09:50:26 -07001042};
1043
Pavel Begunkov7a612352021-03-09 00:37:59 +00001044static bool io_disarm_next(struct io_kiocb *req);
Pavel Begunkoveef51da2021-06-14 02:36:15 +01001045static void io_uring_del_tctx_node(unsigned long index);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00001046static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
1047 struct task_struct *task,
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01001048 bool cancel_all);
Pavel Begunkov78cc6872021-06-14 02:36:23 +01001049static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01001050static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00001051
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001052static bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data,
1053 long res, unsigned int cflags);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001054static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001055static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001056static void io_dismantle_req(struct io_kiocb *req);
1057static void io_put_task(struct task_struct *task, int nr);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001058static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
1059static void io_queue_linked_timeout(struct io_kiocb *req);
Pavel Begunkovfdecb662021-04-25 14:32:20 +01001060static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type,
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01001061 struct io_uring_rsrc_update2 *up,
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01001062 unsigned nr_args);
Pavel Begunkov68fb8972021-03-19 17:22:41 +00001063static void io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001064static struct file *io_file_get(struct io_submit_state *state,
1065 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001066static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001067static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001068
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001069static void io_req_task_queue(struct io_kiocb *req);
Pavel Begunkov2a2758f2021-06-17 18:14:00 +01001070static void io_submit_flush_completions(struct io_ring_ctx *ctx);
Jens Axboe50826202021-02-23 09:02:26 -07001071static bool io_poll_remove_waitqs(struct io_kiocb *req);
Pavel Begunkov179ae0d2021-02-28 22:35:20 +00001072static int io_req_prep_async(struct io_kiocb *req);
Jens Axboe9a56a232019-01-09 09:06:50 -07001073
Jens Axboe2b188cc2019-01-07 10:46:33 -07001074static struct kmem_cache *req_cachep;
1075
Jens Axboe09186822020-10-13 15:01:40 -06001076static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001077
1078struct sock *io_uring_get_socket(struct file *file)
1079{
1080#if defined(CONFIG_UNIX)
1081 if (file->f_op == &io_uring_fops) {
1082 struct io_ring_ctx *ctx = file->private_data;
1083
1084 return ctx->ring_sock->sk;
1085 }
1086#endif
1087 return NULL;
1088}
1089EXPORT_SYMBOL(io_uring_get_socket);
1090
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001091#define io_for_each_link(pos, head) \
1092 for (pos = (head); pos; pos = pos->link)
1093
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01001094static inline void io_req_set_rsrc_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001095{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001096 struct io_ring_ctx *ctx = req->ctx;
1097
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001098 if (!req->fixed_rsrc_refs) {
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01001099 req->fixed_rsrc_refs = &ctx->rsrc_node->refs;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001100 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001101 }
1102}
1103
Pavel Begunkovf70865d2021-04-11 01:46:40 +01001104static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl)
1105{
1106 bool got = percpu_ref_tryget(ref);
1107
1108 /* already at zero, wait for ->release() */
1109 if (!got)
1110 wait_for_completion(compl);
1111 percpu_ref_resurrect(ref);
1112 if (got)
1113 percpu_ref_put(ref);
1114}
1115
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01001116static bool io_match_task(struct io_kiocb *head, struct task_struct *task,
1117 bool cancel_all)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001118{
1119 struct io_kiocb *req;
1120
Pavel Begunkov68207682021-03-22 01:58:25 +00001121 if (task && head->task != task)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001122 return false;
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01001123 if (cancel_all)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001124 return true;
1125
1126 io_for_each_link(req, head) {
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +00001127 if (req->flags & REQ_F_INFLIGHT)
Jens Axboe02a13672021-01-23 15:49:31 -07001128 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001129 }
1130 return false;
1131}
1132
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01001133static inline void req_set_fail(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001134{
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01001135 req->flags |= REQ_F_FAIL;
Jens Axboec40f6372020-06-25 15:39:59 -06001136}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001137
Jens Axboe2b188cc2019-01-07 10:46:33 -07001138static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1139{
1140 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1141
Jens Axboe0f158b42020-05-14 17:18:39 -06001142 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001143}
1144
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001145static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1146{
1147 return !req->timeout.off;
1148}
1149
Jens Axboe2b188cc2019-01-07 10:46:33 -07001150static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1151{
1152 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001153 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001154
1155 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1156 if (!ctx)
1157 return NULL;
1158
Jens Axboe78076bb2019-12-04 19:56:40 -07001159 /*
1160 * Use 5 bits less than the max cq entries, that should give us around
1161 * 32 entries per hash list if totally full and uniformly spread.
1162 */
1163 hash_bits = ilog2(p->cq_entries);
1164 hash_bits -= 5;
1165 if (hash_bits <= 0)
1166 hash_bits = 1;
1167 ctx->cancel_hash_bits = hash_bits;
1168 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1169 GFP_KERNEL);
1170 if (!ctx->cancel_hash)
1171 goto err;
1172 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1173
Pavel Begunkov62248432021-04-28 13:11:29 +01001174 ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL);
1175 if (!ctx->dummy_ubuf)
1176 goto err;
1177 /* set invalid range, so io_import_fixed() fails meeting it */
1178 ctx->dummy_ubuf->ubuf = -1UL;
1179
Roman Gushchin21482892019-05-07 10:01:48 -07001180 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001181 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1182 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001183
1184 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001185 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001186 INIT_LIST_HEAD(&ctx->sqd_list);
Pavel Begunkov311997b2021-06-14 23:37:28 +01001187 init_waitqueue_head(&ctx->poll_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001188 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001189 init_completion(&ctx->ref_comp);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07001190 xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00001191 xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001192 mutex_init(&ctx->uring_lock);
Pavel Begunkov311997b2021-06-14 23:37:28 +01001193 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001194 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001195 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001196 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001197 INIT_LIST_HEAD(&ctx->timeout_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001198 spin_lock_init(&ctx->rsrc_ref_lock);
1199 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001200 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1201 init_llist_head(&ctx->rsrc_put_llist);
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00001202 INIT_LIST_HEAD(&ctx->tctx_list);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001203 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Pavel Begunkovd0acdee2021-05-16 22:58:12 +01001204 INIT_LIST_HEAD(&ctx->locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001205 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001206err:
Pavel Begunkov62248432021-04-28 13:11:29 +01001207 kfree(ctx->dummy_ubuf);
Jens Axboe78076bb2019-12-04 19:56:40 -07001208 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001209 kfree(ctx);
1210 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001211}
1212
Pavel Begunkov8f6ed492021-05-16 22:58:10 +01001213static void io_account_cq_overflow(struct io_ring_ctx *ctx)
1214{
1215 struct io_rings *r = ctx->rings;
1216
1217 WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1);
1218 ctx->cq_extra--;
1219}
1220
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001221static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001222{
Jens Axboe2bc99302020-07-09 09:43:27 -06001223 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1224 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001225
Pavel Begunkov8f6ed492021-05-16 22:58:10 +01001226 return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail;
Jens Axboe2bc99302020-07-09 09:43:27 -06001227 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001228
Bob Liu9d858b22019-11-13 18:06:25 +08001229 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001230}
1231
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001232static void io_req_track_inflight(struct io_kiocb *req)
1233{
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001234 if (!(req->flags & REQ_F_INFLIGHT)) {
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001235 req->flags |= REQ_F_INFLIGHT;
Pavel Begunkovb303fe22021-04-11 01:46:26 +01001236 atomic_inc(&current->io_uring->inflight_tracked);
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001237 }
1238}
1239
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001240static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001241{
Jens Axboed3656342019-12-18 09:50:26 -07001242 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001243 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001244
Pavel Begunkovb8e64b52021-06-17 18:14:02 +01001245 if (!(req->flags & REQ_F_CREDS)) {
1246 req->flags |= REQ_F_CREDS;
Pavel Begunkovc10d1f92021-06-17 18:14:01 +01001247 req->creds = get_current_cred();
Pavel Begunkovb8e64b52021-06-17 18:14:02 +01001248 }
Jens Axboe003e8dc2021-03-06 09:22:27 -07001249
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001250 req->work.list.next = NULL;
1251 req->work.flags = 0;
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001252 if (req->flags & REQ_F_FORCE_ASYNC)
1253 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1254
Jens Axboed3656342019-12-18 09:50:26 -07001255 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001256 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001257 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboe4b982bd2021-04-01 08:38:34 -06001258 } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) {
Jens Axboed3656342019-12-18 09:50:26 -07001259 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001260 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001261 }
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001262
1263 switch (req->opcode) {
1264 case IORING_OP_SPLICE:
1265 case IORING_OP_TEE:
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001266 if (!S_ISREG(file_inode(req->splice.file_in)->i_mode))
1267 req->work.flags |= IO_WQ_WORK_UNBOUND;
1268 break;
1269 }
Jens Axboe561fb042019-10-24 07:25:42 -06001270}
1271
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001272static void io_prep_async_link(struct io_kiocb *req)
1273{
1274 struct io_kiocb *cur;
1275
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001276 io_for_each_link(cur, req)
1277 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001278}
1279
Pavel Begunkovebf93662021-03-01 18:20:47 +00001280static void io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001281{
Jackie Liua197f662019-11-08 08:09:12 -07001282 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001283 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001284 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe561fb042019-10-24 07:25:42 -06001285
Jens Axboe3bfe6102021-02-16 14:15:30 -07001286 BUG_ON(!tctx);
1287 BUG_ON(!tctx->io_wq);
Jens Axboe561fb042019-10-24 07:25:42 -06001288
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001289 /* init ->work of the whole link before punting */
1290 io_prep_async_link(req);
Pavel Begunkovd07f1e8a2021-03-22 01:45:58 +00001291 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1292 &req->work, req->flags);
Pavel Begunkovebf93662021-03-01 18:20:47 +00001293 io_wq_enqueue(tctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001294 if (link)
1295 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001296}
1297
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001298static void io_kill_timeout(struct io_kiocb *req, int status)
Pavel Begunkov8c855882021-04-13 02:58:41 +01001299 __must_hold(&req->ctx->completion_lock)
Jens Axboe5262f562019-09-17 12:26:57 -06001300{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001301 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001302
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01001303 if (hrtimer_try_to_cancel(&io->timer) != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001304 atomic_set(&req->ctx->cq_timeouts,
1305 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001306 list_del_init(&req->timeout.list);
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001307 io_cqring_fill_event(req->ctx, req->user_data, status, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001308 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001309 }
1310}
1311
Pavel Begunkov441b8a72021-06-14 23:37:31 +01001312static void io_queue_deferred(struct io_ring_ctx *ctx)
Pavel Begunkov04518942020-05-26 20:34:05 +03001313{
Pavel Begunkov441b8a72021-06-14 23:37:31 +01001314 while (!list_empty(&ctx->defer_list)) {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001315 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1316 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001317
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001318 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001319 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001320 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001321 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001322 kfree(de);
Pavel Begunkov441b8a72021-06-14 23:37:31 +01001323 }
Pavel Begunkov04518942020-05-26 20:34:05 +03001324}
1325
Pavel Begunkov360428f2020-05-30 14:54:17 +03001326static void io_flush_timeouts(struct io_ring_ctx *ctx)
1327{
Pavel Begunkov441b8a72021-06-14 23:37:31 +01001328 u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001329
Pavel Begunkovf18ee4c2021-06-14 23:37:25 +01001330 while (!list_empty(&ctx->timeout_list)) {
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001331 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001332 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001333 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001334
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001335 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001336 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001337
1338 /*
1339 * Since seq can easily wrap around over time, subtract
1340 * the last seq at which timeouts were flushed before comparing.
1341 * Assuming not more than 2^31-1 events have happened since,
1342 * these subtractions won't have wrapped, so we can check if
1343 * target is in [last_seq, current_seq] by comparing the two.
1344 */
1345 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1346 events_got = seq - ctx->cq_last_tm_flush;
1347 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001348 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001349
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001350 list_del_init(&req->timeout.list);
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001351 io_kill_timeout(req, 0);
Pavel Begunkovf18ee4c2021-06-14 23:37:25 +01001352 }
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001353 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001354}
1355
Pavel Begunkov2335f6f2021-06-15 16:47:58 +01001356static void __io_commit_cqring_flush(struct io_ring_ctx *ctx)
Jens Axboede0617e2019-04-06 21:51:27 -06001357{
Pavel Begunkov2335f6f2021-06-15 16:47:58 +01001358 if (ctx->off_timeout_used)
1359 io_flush_timeouts(ctx);
1360 if (ctx->drain_active)
1361 io_queue_deferred(ctx);
1362}
1363
1364static inline void io_commit_cqring(struct io_ring_ctx *ctx)
1365{
1366 if (unlikely(ctx->off_timeout_used || ctx->drain_active))
1367 __io_commit_cqring_flush(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001368 /* order cqe stores with ring update */
1369 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001370}
1371
Jens Axboe90554202020-09-03 12:12:41 -06001372static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1373{
1374 struct io_rings *r = ctx->rings;
1375
Pavel Begunkova566c552021-05-16 22:58:08 +01001376 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries;
Jens Axboe90554202020-09-03 12:12:41 -06001377}
1378
Pavel Begunkov888aae22021-01-19 13:32:39 +00001379static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1380{
1381 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1382}
1383
Pavel Begunkovd068b502021-05-16 22:58:11 +01001384static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001385{
Hristo Venev75b28af2019-08-26 17:23:46 +00001386 struct io_rings *rings = ctx->rings;
Pavel Begunkovea5ab3b2021-05-16 22:58:09 +01001387 unsigned tail, mask = ctx->cq_entries - 1;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001388
Stefan Bühler115e12e2019-04-24 23:54:18 +02001389 /*
1390 * writes to the cq entry need to come after reading head; the
1391 * control dependency is enough as we're using WRITE_ONCE to
1392 * fill the cq entry
1393 */
Pavel Begunkova566c552021-05-16 22:58:08 +01001394 if (__io_cqring_events(ctx) == ctx->cq_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001395 return NULL;
1396
Pavel Begunkov888aae22021-01-19 13:32:39 +00001397 tail = ctx->cached_cq_tail++;
Pavel Begunkovea5ab3b2021-05-16 22:58:09 +01001398 return &rings->cqes[tail & mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001399}
1400
Jens Axboef2842ab2020-01-08 11:04:00 -07001401static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1402{
Pavel Begunkov44c769d2021-04-11 01:46:31 +01001403 if (likely(!ctx->cq_ev_fd))
Jens Axboef0b493e2020-02-01 21:30:11 -07001404 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001405 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1406 return false;
Pavel Begunkov44c769d2021-04-11 01:46:31 +01001407 return !ctx->eventfd_async || io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001408}
1409
Jens Axboeb41e9852020-02-17 09:52:41 -07001410static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001411{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001412 /* see waitqueue_active() comment */
1413 smp_mb();
1414
Pavel Begunkov311997b2021-06-14 23:37:28 +01001415 if (waitqueue_active(&ctx->cq_wait))
1416 wake_up(&ctx->cq_wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001417 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1418 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001419 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001420 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkov311997b2021-06-14 23:37:28 +01001421 if (waitqueue_active(&ctx->poll_wait)) {
1422 wake_up_interruptible(&ctx->poll_wait);
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001423 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1424 }
Jens Axboe8c838782019-03-12 15:48:16 -06001425}
1426
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001427static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1428{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001429 /* see waitqueue_active() comment */
1430 smp_mb();
1431
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001432 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov311997b2021-06-14 23:37:28 +01001433 if (waitqueue_active(&ctx->cq_wait))
1434 wake_up(&ctx->cq_wait);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001435 }
1436 if (io_should_trigger_evfd(ctx))
1437 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkov311997b2021-06-14 23:37:28 +01001438 if (waitqueue_active(&ctx->poll_wait)) {
1439 wake_up_interruptible(&ctx->poll_wait);
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001440 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1441 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001442}
1443
Jens Axboec4a2ed72019-11-21 21:01:26 -07001444/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001445static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001446{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001447 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001448 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001449
Pavel Begunkova566c552021-05-16 22:58:08 +01001450 if (!force && __io_cqring_events(ctx) == ctx->cq_entries)
Pavel Begunkove23de152020-12-17 00:24:37 +00001451 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001452
Jens Axboeb18032b2021-01-24 16:58:56 -07001453 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001454 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001455 while (!list_empty(&ctx->cq_overflow_list)) {
Pavel Begunkovd068b502021-05-16 22:58:11 +01001456 struct io_uring_cqe *cqe = io_get_cqe(ctx);
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001457 struct io_overflow_cqe *ocqe;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001458
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001459 if (!cqe && !force)
1460 break;
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001461 ocqe = list_first_entry(&ctx->cq_overflow_list,
1462 struct io_overflow_cqe, list);
1463 if (cqe)
1464 memcpy(cqe, &ocqe->cqe, sizeof(*cqe));
1465 else
Pavel Begunkov8f6ed492021-05-16 22:58:10 +01001466 io_account_cq_overflow(ctx);
1467
Jens Axboeb18032b2021-01-24 16:58:56 -07001468 posted = true;
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001469 list_del(&ocqe->list);
1470 kfree(ocqe);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001471 }
1472
Pavel Begunkov09e88402020-12-17 00:24:38 +00001473 all_flushed = list_empty(&ctx->cq_overflow_list);
1474 if (all_flushed) {
Pavel Begunkov5ed7a372021-06-14 23:37:27 +01001475 clear_bit(0, &ctx->check_cq_overflow);
Pavel Begunkov09e88402020-12-17 00:24:38 +00001476 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1477 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001478
Jens Axboeb18032b2021-01-24 16:58:56 -07001479 if (posted)
1480 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001481 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001482 if (posted)
1483 io_cqring_ev_posted(ctx);
Pavel Begunkov09e88402020-12-17 00:24:38 +00001484 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001485}
1486
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001487static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
Pavel Begunkov6c503152021-01-04 20:36:36 +00001488{
Jens Axboeca0a2652021-03-04 17:15:48 -07001489 bool ret = true;
1490
Pavel Begunkov5ed7a372021-06-14 23:37:27 +01001491 if (test_bit(0, &ctx->check_cq_overflow)) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00001492 /* iopoll syncs against uring_lock, not completion_lock */
1493 if (ctx->flags & IORING_SETUP_IOPOLL)
1494 mutex_lock(&ctx->uring_lock);
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001495 ret = __io_cqring_overflow_flush(ctx, force);
Pavel Begunkov6c503152021-01-04 20:36:36 +00001496 if (ctx->flags & IORING_SETUP_IOPOLL)
1497 mutex_unlock(&ctx->uring_lock);
1498 }
Jens Axboeca0a2652021-03-04 17:15:48 -07001499
1500 return ret;
Pavel Begunkov6c503152021-01-04 20:36:36 +00001501}
1502
Jens Axboeabc54d62021-02-24 13:32:30 -07001503/*
1504 * Shamelessly stolen from the mm implementation of page reference checking,
1505 * see commit f958d7b528b1 for details.
1506 */
1507#define req_ref_zero_or_close_to_overflow(req) \
1508 ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u)
1509
Jens Axboede9b4cc2021-02-24 13:28:27 -07001510static inline bool req_ref_inc_not_zero(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001511{
Jens Axboeabc54d62021-02-24 13:32:30 -07001512 return atomic_inc_not_zero(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001513}
1514
1515static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs)
1516{
Jens Axboeabc54d62021-02-24 13:32:30 -07001517 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1518 return atomic_sub_and_test(refs, &req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001519}
1520
1521static inline bool req_ref_put_and_test(struct io_kiocb *req)
1522{
Jens Axboeabc54d62021-02-24 13:32:30 -07001523 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1524 return atomic_dec_and_test(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001525}
1526
1527static inline void req_ref_put(struct io_kiocb *req)
1528{
Jens Axboeabc54d62021-02-24 13:32:30 -07001529 WARN_ON_ONCE(req_ref_put_and_test(req));
Jens Axboede9b4cc2021-02-24 13:28:27 -07001530}
1531
1532static inline void req_ref_get(struct io_kiocb *req)
1533{
Jens Axboeabc54d62021-02-24 13:32:30 -07001534 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1535 atomic_inc(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001536}
1537
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001538static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
1539 long res, unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001540{
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001541 struct io_overflow_cqe *ocqe;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001542
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001543 ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
1544 if (!ocqe) {
1545 /*
1546 * If we're in ring overflow flush mode, or in task cancel mode,
1547 * or cannot allocate an overflow entry, then we need to drop it
1548 * on the floor.
1549 */
Pavel Begunkov8f6ed492021-05-16 22:58:10 +01001550 io_account_cq_overflow(ctx);
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001551 return false;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001552 }
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001553 if (list_empty(&ctx->cq_overflow_list)) {
Pavel Begunkov5ed7a372021-06-14 23:37:27 +01001554 set_bit(0, &ctx->check_cq_overflow);
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001555 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
1556 }
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001557 ocqe->cqe.user_data = user_data;
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001558 ocqe->cqe.res = res;
1559 ocqe->cqe.flags = cflags;
1560 list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
1561 return true;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001562}
1563
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001564static inline bool __io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data,
1565 long res, unsigned int cflags)
Pavel Begunkov8d133262021-04-11 01:46:33 +01001566{
Jens Axboe2b188cc2019-01-07 10:46:33 -07001567 struct io_uring_cqe *cqe;
1568
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001569 trace_io_uring_complete(ctx, user_data, res, cflags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001570
1571 /*
1572 * If we can't get a cq entry, userspace overflowed the
1573 * submission (by quite a lot). Increment the overflow count in
1574 * the ring.
1575 */
Pavel Begunkovd068b502021-05-16 22:58:11 +01001576 cqe = io_get_cqe(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001577 if (likely(cqe)) {
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001578 WRITE_ONCE(cqe->user_data, user_data);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001579 WRITE_ONCE(cqe->res, res);
1580 WRITE_ONCE(cqe->flags, cflags);
Pavel Begunkov8d133262021-04-11 01:46:33 +01001581 return true;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001582 }
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001583 return io_cqring_event_overflow(ctx, user_data, res, cflags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001584}
1585
Pavel Begunkov8d133262021-04-11 01:46:33 +01001586/* not as hot to bloat with inlining */
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001587static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data,
1588 long res, unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001589{
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001590 return __io_cqring_fill_event(ctx, user_data, res, cflags);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001591}
1592
Pavel Begunkov7a612352021-03-09 00:37:59 +00001593static void io_req_complete_post(struct io_kiocb *req, long res,
1594 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001595{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001596 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001597 unsigned long flags;
1598
1599 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001600 __io_cqring_fill_event(ctx, req->user_data, res, cflags);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001601 /*
1602 * If we're the last reference to this request, add to our locked
1603 * free_list cache.
1604 */
Jens Axboede9b4cc2021-02-24 13:28:27 -07001605 if (req_ref_put_and_test(req)) {
Pavel Begunkov7a612352021-03-09 00:37:59 +00001606 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01001607 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL))
Pavel Begunkov7a612352021-03-09 00:37:59 +00001608 io_disarm_next(req);
1609 if (req->link) {
1610 io_req_task_queue(req->link);
1611 req->link = NULL;
1612 }
1613 }
Jens Axboec7dae4b2021-02-09 19:53:37 -07001614 io_dismantle_req(req);
1615 io_put_task(req->task, 1);
Pavel Begunkovd0acdee2021-05-16 22:58:12 +01001616 list_add(&req->compl.list, &ctx->locked_free_list);
1617 ctx->locked_free_nr++;
Pavel Begunkov180f8292021-03-14 20:57:09 +00001618 } else {
1619 if (!percpu_ref_tryget(&ctx->refs))
1620 req = NULL;
1621 }
Pavel Begunkov7a612352021-03-09 00:37:59 +00001622 io_commit_cqring(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001623 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7a612352021-03-09 00:37:59 +00001624
Pavel Begunkov180f8292021-03-14 20:57:09 +00001625 if (req) {
1626 io_cqring_ev_posted(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001627 percpu_ref_put(&ctx->refs);
Pavel Begunkov180f8292021-03-14 20:57:09 +00001628 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001629}
1630
Jens Axboe4e3d9ff2021-04-15 17:44:34 -06001631static inline bool io_req_needs_clean(struct io_kiocb *req)
1632{
Pavel Begunkovc8543572021-06-17 18:14:04 +01001633 return req->flags & IO_REQ_CLEAN_FLAGS;
Jens Axboe4e3d9ff2021-04-15 17:44:34 -06001634}
1635
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001636static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001637 unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001638{
Jens Axboe4e3d9ff2021-04-15 17:44:34 -06001639 if (io_req_needs_clean(req))
Pavel Begunkov68fb8972021-03-19 17:22:41 +00001640 io_clean_op(req);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001641 req->result = res;
1642 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001643 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001644}
Jens Axboe2b188cc2019-01-07 10:46:33 -07001645
Pavel Begunkov889fca72021-02-10 00:03:09 +00001646static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1647 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001648{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001649 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1650 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001651 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001652 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001653}
Jens Axboebcda7ba2020-02-23 16:42:51 -07001654
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001655static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001656{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001657 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001658}
1659
Pavel Begunkovf41db2732021-02-28 22:35:12 +00001660static void io_req_complete_failed(struct io_kiocb *req, long res)
1661{
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01001662 req_set_fail(req);
Pavel Begunkovf41db2732021-02-28 22:35:12 +00001663 io_put_req(req);
1664 io_req_complete_post(req, res, 0);
1665}
1666
Pavel Begunkovdac7a092021-03-19 17:22:39 +00001667static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx,
1668 struct io_comp_state *cs)
1669{
1670 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkovd0acdee2021-05-16 22:58:12 +01001671 list_splice_init(&ctx->locked_free_list, &cs->free_list);
1672 ctx->locked_free_nr = 0;
Pavel Begunkovdac7a092021-03-19 17:22:39 +00001673 spin_unlock_irq(&ctx->completion_lock);
1674}
1675
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001676/* Returns true IFF there are requests in the cache */
Jens Axboec7dae4b2021-02-09 19:53:37 -07001677static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001678{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001679 struct io_submit_state *state = &ctx->submit_state;
1680 struct io_comp_state *cs = &state->comp;
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001681 int nr;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001682
Jens Axboec7dae4b2021-02-09 19:53:37 -07001683 /*
1684 * If we have more than a batch's worth of requests in our IRQ side
1685 * locked cache, grab the lock and move them over to our submission
1686 * side cache.
1687 */
Pavel Begunkovd0acdee2021-05-16 22:58:12 +01001688 if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH)
Pavel Begunkovdac7a092021-03-19 17:22:39 +00001689 io_flush_cached_locked_reqs(ctx, cs);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001690
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001691 nr = state->free_reqs;
Jens Axboec7dae4b2021-02-09 19:53:37 -07001692 while (!list_empty(&cs->free_list)) {
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001693 struct io_kiocb *req = list_first_entry(&cs->free_list,
1694 struct io_kiocb, compl.list);
1695
Jens Axboe2b188cc2019-01-07 10:46:33 -07001696 list_del(&req->compl.list);
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001697 state->reqs[nr++] = req;
1698 if (nr == ARRAY_SIZE(state->reqs))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001699 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001700 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001701
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001702 state->free_reqs = nr;
1703 return nr != 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001704}
1705
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001706static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001707{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001708 struct io_submit_state *state = &ctx->submit_state;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001709
Pavel Begunkovfe7e3252021-06-24 15:09:57 +01001710 BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001711
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001712 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001713 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Pavel Begunkov99ebe4e2021-06-26 21:40:49 +01001714 int ret, i;
Jens Axboe2579f912019-01-09 09:10:43 -07001715
Jens Axboec7dae4b2021-02-09 19:53:37 -07001716 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001717 goto got_req;
1718
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001719 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
1720 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06001721
1722 /*
1723 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1724 * retry single alloc to be on the safe side.
1725 */
1726 if (unlikely(ret <= 0)) {
1727 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1728 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00001729 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06001730 ret = 1;
1731 }
Pavel Begunkov99ebe4e2021-06-26 21:40:49 +01001732
1733 /*
1734 * Don't initialise the fields below on every allocation, but
1735 * do that in advance and keep valid on free.
1736 */
1737 for (i = 0; i < ret; i++) {
1738 struct io_kiocb *req = state->reqs[i];
1739
1740 req->ctx = ctx;
1741 req->link = NULL;
1742 req->async_data = NULL;
1743 /* not necessary, but safer to zero */
1744 req->result = 0;
1745 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001746 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001747 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001748got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03001749 state->free_reqs--;
1750 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001751}
1752
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001753static inline void io_put_file(struct file *file)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001754{
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001755 if (file)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001756 fput(file);
1757}
1758
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001759static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001760{
Pavel Begunkov094bae42021-03-19 17:22:42 +00001761 unsigned int flags = req->flags;
Pavel Begunkov929a3af2020-02-19 00:19:09 +03001762
Pavel Begunkov3a0a6902021-04-20 12:03:31 +01001763 if (io_req_needs_clean(req))
1764 io_clean_op(req);
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001765 if (!(flags & REQ_F_FIXED_FILE))
1766 io_put_file(req->file);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001767 if (req->fixed_rsrc_refs)
1768 percpu_ref_put(req->fixed_rsrc_refs);
Pavel Begunkov99ebe4e2021-06-26 21:40:49 +01001769 if (req->async_data) {
Pavel Begunkov094bae42021-03-19 17:22:42 +00001770 kfree(req->async_data);
Pavel Begunkov99ebe4e2021-06-26 21:40:49 +01001771 req->async_data = NULL;
1772 }
Pavel Begunkove6543a82020-06-28 12:52:30 +03001773}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001774
Pavel Begunkovb23fcf42021-03-01 18:20:48 +00001775/* must to be called somewhat shortly after putting a request */
Pavel Begunkov7c660732021-01-25 11:42:21 +00001776static inline void io_put_task(struct task_struct *task, int nr)
1777{
1778 struct io_uring_task *tctx = task->io_uring;
1779
1780 percpu_counter_sub(&tctx->inflight, nr);
1781 if (unlikely(atomic_read(&tctx->in_idle)))
1782 wake_up(&tctx->wait);
1783 put_task_struct_many(task, nr);
1784}
1785
Pavel Begunkov216578e2020-10-13 09:44:00 +01001786static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001787{
Jens Axboe51a4cc12020-08-10 10:55:56 -06001788 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001789
Pavel Begunkov216578e2020-10-13 09:44:00 +01001790 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00001791 io_put_task(req->task, 1);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001792
Pavel Begunkov3893f392021-02-10 00:03:15 +00001793 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001794 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001795}
1796
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001797static inline void io_remove_next_linked(struct io_kiocb *req)
1798{
1799 struct io_kiocb *nxt = req->link;
1800
1801 req->link = nxt->link;
1802 nxt->link = NULL;
1803}
1804
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001805static bool io_kill_linked_timeout(struct io_kiocb *req)
1806 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001807{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001808 struct io_kiocb *link = req->link;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001809
Pavel Begunkov900fad42020-10-19 16:39:16 +01001810 /*
1811 * Can happen if a linked timeout fired and link had been like
1812 * req -> link t-out -> link t-out [-> ...]
1813 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001814 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1815 struct io_timeout_data *io = link->async_data;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001816
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001817 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00001818 link->timeout.head = NULL;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01001819 if (hrtimer_try_to_cancel(&io->timer) != -1) {
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001820 io_cqring_fill_event(link->ctx, link->user_data,
1821 -ECANCELED, 0);
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001822 io_put_req_deferred(link, 1);
Pavel Begunkovd4729fb2021-03-22 01:58:24 +00001823 return true;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001824 }
1825 }
Pavel Begunkovd4729fb2021-03-22 01:58:24 +00001826 return false;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001827}
1828
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001829static void io_fail_links(struct io_kiocb *req)
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001830 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001831{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001832 struct io_kiocb *nxt, *link = req->link;
Jens Axboe9e645e112019-05-10 16:07:28 -06001833
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001834 req->link = NULL;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001835 while (link) {
1836 nxt = link->link;
1837 link->link = NULL;
1838
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001839 trace_io_uring_fail_link(req, link);
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01001840 io_cqring_fill_event(link->ctx, link->user_data, -ECANCELED, 0);
Jens Axboe1575f212021-02-27 15:20:49 -07001841 io_put_req_deferred(link, 2);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001842 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001843 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001844}
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001845
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001846static bool io_disarm_next(struct io_kiocb *req)
1847 __must_hold(&req->ctx->completion_lock)
1848{
1849 bool posted = false;
1850
1851 if (likely(req->flags & REQ_F_LINK_TIMEOUT))
1852 posted = io_kill_linked_timeout(req);
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01001853 if (unlikely((req->flags & REQ_F_FAIL) &&
Pavel Begunkove4335ed2021-04-11 01:46:39 +01001854 !(req->flags & REQ_F_HARDLINK))) {
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001855 posted |= (req->link != NULL);
1856 io_fail_links(req);
1857 }
1858 return posted;
Jens Axboe9e645e112019-05-10 16:07:28 -06001859}
1860
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001861static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001862{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001863 struct io_kiocb *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07001864
Jens Axboe9e645e112019-05-10 16:07:28 -06001865 /*
1866 * If LINK is set, we have dependent requests in this chain. If we
1867 * didn't fail this request, queue the first one up, moving any other
1868 * dependencies to the next request. In case of failure, fail the rest
1869 * of the chain.
1870 */
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01001871 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL)) {
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001872 struct io_ring_ctx *ctx = req->ctx;
1873 unsigned long flags;
1874 bool posted;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001875
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001876 spin_lock_irqsave(&ctx->completion_lock, flags);
1877 posted = io_disarm_next(req);
1878 if (posted)
1879 io_commit_cqring(req->ctx);
1880 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1881 if (posted)
1882 io_cqring_ev_posted(ctx);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001883 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001884 nxt = req->link;
1885 req->link = NULL;
1886 return nxt;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001887}
Jens Axboe2665abf2019-11-05 12:40:47 -07001888
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001889static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001890{
Pavel Begunkovcdbff982021-02-12 18:41:16 +00001891 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK))))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001892 return NULL;
1893 return __io_req_find_next(req);
1894}
1895
Pavel Begunkov2c323952021-02-28 22:04:53 +00001896static void ctx_flush_and_put(struct io_ring_ctx *ctx)
1897{
1898 if (!ctx)
1899 return;
1900 if (ctx->submit_state.comp.nr) {
1901 mutex_lock(&ctx->uring_lock);
Pavel Begunkov2a2758f2021-06-17 18:14:00 +01001902 io_submit_flush_completions(ctx);
Pavel Begunkov2c323952021-02-28 22:04:53 +00001903 mutex_unlock(&ctx->uring_lock);
1904 }
1905 percpu_ref_put(&ctx->refs);
1906}
1907
Jens Axboe7cbf1722021-02-10 00:03:20 +00001908static void tctx_task_work(struct callback_head *cb)
1909{
Pavel Begunkovebd0df22021-06-17 18:14:07 +01001910 struct io_ring_ctx *ctx = NULL;
Pavel Begunkov3f184072021-06-17 18:14:06 +01001911 struct io_uring_task *tctx = container_of(cb, struct io_uring_task,
1912 task_work);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001913
Pavel Begunkov16f72072021-06-17 18:14:09 +01001914 while (1) {
Pavel Begunkov3f184072021-06-17 18:14:06 +01001915 struct io_wq_work_node *node;
1916
1917 spin_lock_irq(&tctx->task_lock);
Pavel Begunkovc6538be2021-06-17 18:14:08 +01001918 node = tctx->task_list.first;
Pavel Begunkov3f184072021-06-17 18:14:06 +01001919 INIT_WQ_LIST(&tctx->task_list);
1920 spin_unlock_irq(&tctx->task_lock);
1921
Pavel Begunkov3f184072021-06-17 18:14:06 +01001922 while (node) {
1923 struct io_wq_work_node *next = node->next;
1924 struct io_kiocb *req = container_of(node, struct io_kiocb,
1925 io_task_work.node);
1926
1927 if (req->ctx != ctx) {
1928 ctx_flush_and_put(ctx);
1929 ctx = req->ctx;
1930 percpu_ref_get(&ctx->refs);
1931 }
1932 req->task_work.func(&req->task_work);
1933 node = next;
1934 }
Pavel Begunkov7a778f92021-06-17 18:14:10 +01001935 if (wq_list_empty(&tctx->task_list)) {
1936 clear_bit(0, &tctx->task_state);
1937 if (wq_list_empty(&tctx->task_list))
1938 break;
1939 /* another tctx_task_work() is enqueued, yield */
1940 if (test_and_set_bit(0, &tctx->task_state))
1941 break;
1942 }
Jens Axboe7cbf1722021-02-10 00:03:20 +00001943 cond_resched();
Pavel Begunkov3f184072021-06-17 18:14:06 +01001944 }
Pavel Begunkovebd0df22021-06-17 18:14:07 +01001945
1946 ctx_flush_and_put(ctx);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001947}
1948
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001949static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboe7cbf1722021-02-10 00:03:20 +00001950{
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001951 struct task_struct *tsk = req->task;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001952 struct io_uring_task *tctx = tsk->io_uring;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001953 enum task_work_notify_mode notify;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001954 struct io_wq_work_node *node, *prev;
Jens Axboe0b81e802021-02-16 10:33:53 -07001955 unsigned long flags;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001956 int ret = 0;
1957
1958 if (unlikely(tsk->flags & PF_EXITING))
1959 return -ESRCH;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001960
1961 WARN_ON_ONCE(!tctx);
1962
Jens Axboe0b81e802021-02-16 10:33:53 -07001963 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001964 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001965 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001966
1967 /* task_work already pending, we're done */
1968 if (test_bit(0, &tctx->task_state) ||
1969 test_and_set_bit(0, &tctx->task_state))
1970 return 0;
1971
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001972 /*
1973 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
1974 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
1975 * processing task_work. There's no reliable way to tell if TWA_RESUME
1976 * will do the job.
1977 */
1978 notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL;
1979
1980 if (!task_work_add(tsk, &tctx->task_work, notify)) {
1981 wake_up_process(tsk);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001982 return 0;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001983 }
Jens Axboe7cbf1722021-02-10 00:03:20 +00001984
1985 /*
1986 * Slow path - we failed, find and delete work. if the work is not
1987 * in the list, it got run and we're fine.
1988 */
Jens Axboe0b81e802021-02-16 10:33:53 -07001989 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001990 wq_list_for_each(node, prev, &tctx->task_list) {
1991 if (&req->io_task_work.node == node) {
1992 wq_list_del(&tctx->task_list, node, prev);
1993 ret = 1;
1994 break;
1995 }
1996 }
Jens Axboe0b81e802021-02-16 10:33:53 -07001997 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001998 clear_bit(0, &tctx->task_state);
1999 return ret;
2000}
2001
Pavel Begunkov9b465712021-03-15 14:23:07 +00002002static bool io_run_task_work_head(struct callback_head **work_head)
2003{
2004 struct callback_head *work, *next;
2005 bool executed = false;
2006
2007 do {
2008 work = xchg(work_head, NULL);
2009 if (!work)
2010 break;
2011
2012 do {
2013 next = work->next;
2014 work->func(work);
2015 work = next;
2016 cond_resched();
2017 } while (work);
2018 executed = true;
2019 } while (1);
2020
2021 return executed;
2022}
2023
2024static void io_task_work_add_head(struct callback_head **work_head,
2025 struct callback_head *task_work)
2026{
2027 struct callback_head *head;
2028
2029 do {
2030 head = READ_ONCE(*work_head);
2031 task_work->next = head;
2032 } while (cmpxchg(work_head, head, task_work) != head);
2033}
2034
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002035static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00002036 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002037{
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002038 init_task_work(&req->task_work, cb);
Pavel Begunkov9b465712021-03-15 14:23:07 +00002039 io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002040}
2041
Jens Axboec40f6372020-06-25 15:39:59 -06002042static void io_req_task_cancel(struct callback_head *cb)
2043{
2044 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002045 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002046
Pavel Begunkove83acd72021-02-28 22:35:09 +00002047 /* ctx is guaranteed to stay alive while we hold uring_lock */
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002048 mutex_lock(&ctx->uring_lock);
Pavel Begunkov25935532021-03-19 17:22:40 +00002049 io_req_complete_failed(req, req->result);
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002050 mutex_unlock(&ctx->uring_lock);
Jens Axboec40f6372020-06-25 15:39:59 -06002051}
2052
2053static void __io_req_task_submit(struct io_kiocb *req)
2054{
2055 struct io_ring_ctx *ctx = req->ctx;
2056
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00002057 /* ctx stays valid until unlock, even if we drop all ours ctx->refs */
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002058 mutex_lock(&ctx->uring_lock);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00002059 if (!(current->flags & PF_EXITING) && !current->in_execve)
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00002060 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002061 else
Pavel Begunkov25935532021-03-19 17:22:40 +00002062 io_req_complete_failed(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002063 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06002064}
2065
Jens Axboec40f6372020-06-25 15:39:59 -06002066static void io_req_task_submit(struct callback_head *cb)
2067{
2068 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2069
2070 __io_req_task_submit(req);
2071}
2072
Pavel Begunkova3df76982021-02-18 22:32:52 +00002073static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
2074{
Pavel Begunkova3df76982021-02-18 22:32:52 +00002075 req->result = ret;
2076 req->task_work.func = io_req_task_cancel;
2077
2078 if (unlikely(io_req_task_work_add(req)))
2079 io_req_task_work_add_fallback(req, io_req_task_cancel);
2080}
2081
Pavel Begunkov2c4b8eb2021-02-28 22:35:10 +00002082static void io_req_task_queue(struct io_kiocb *req)
2083{
2084 req->task_work.func = io_req_task_submit;
2085
2086 if (unlikely(io_req_task_work_add(req)))
2087 io_req_task_queue_fail(req, -ECANCELED);
2088}
2089
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002090static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002091{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002092 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002093
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002094 if (nxt)
2095 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002096}
2097
Jens Axboe9e645e112019-05-10 16:07:28 -06002098static void io_free_req(struct io_kiocb *req)
2099{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002100 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002101 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002102}
2103
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002104struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002105 struct task_struct *task;
2106 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002107 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002108};
2109
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002110static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002111{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002112 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002113 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002114 rb->task = NULL;
2115}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002116
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002117static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2118 struct req_batch *rb)
2119{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002120 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002121 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002122 if (rb->ctx_refs)
2123 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002124}
2125
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002126static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2127 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002128{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002129 io_queue_next(req);
Pavel Begunkov96670652021-03-19 17:22:32 +00002130 io_dismantle_req(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002131
Jens Axboee3bc8e92020-09-24 08:45:57 -06002132 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002133 if (rb->task)
2134 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002135 rb->task = req->task;
2136 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002137 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002138 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002139 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002140
Pavel Begunkovbd759042021-02-12 03:23:50 +00002141 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002142 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002143 else
2144 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002145}
2146
Pavel Begunkov2a2758f2021-06-17 18:14:00 +01002147static void io_submit_flush_completions(struct io_ring_ctx *ctx)
Pavel Begunkov905c1722021-02-10 00:03:14 +00002148{
Pavel Begunkov2a2758f2021-06-17 18:14:00 +01002149 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkov905c1722021-02-10 00:03:14 +00002150 int i, nr = cs->nr;
Pavel Begunkov905c1722021-02-10 00:03:14 +00002151 struct req_batch rb;
2152
Pavel Begunkov905c1722021-02-10 00:03:14 +00002153 spin_lock_irq(&ctx->completion_lock);
2154 for (i = 0; i < nr; i++) {
Pavel Begunkov5182ed22021-06-26 21:40:48 +01002155 struct io_kiocb *req = cs->reqs[i];
2156
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01002157 __io_cqring_fill_event(ctx, req->user_data, req->result,
2158 req->compl.cflags);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002159 }
2160 io_commit_cqring(ctx);
2161 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002162 io_cqring_ev_posted(ctx);
Pavel Begunkov5182ed22021-06-26 21:40:48 +01002163
2164 io_init_req_batch(&rb);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002165 for (i = 0; i < nr; i++) {
Pavel Begunkov5182ed22021-06-26 21:40:48 +01002166 struct io_kiocb *req = cs->reqs[i];
Pavel Begunkov905c1722021-02-10 00:03:14 +00002167
2168 /* submission and completion refs */
Jens Axboede9b4cc2021-02-24 13:28:27 -07002169 if (req_ref_sub_and_test(req, 2))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002170 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002171 }
2172
2173 io_req_free_batch_finish(ctx, &rb);
2174 cs->nr = 0;
Jens Axboee65ef562019-03-12 10:16:44 -06002175}
2176
Jens Axboeba816ad2019-09-28 11:36:45 -06002177/*
2178 * Drop reference to request, return next in chain (if there is one) if this
2179 * was the last reference to this request.
2180 */
Pavel Begunkov0d850352021-03-19 17:22:37 +00002181static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002182{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002183 struct io_kiocb *nxt = NULL;
2184
Jens Axboede9b4cc2021-02-24 13:28:27 -07002185 if (req_ref_put_and_test(req)) {
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002186 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002187 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002188 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002189 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002190}
2191
Pavel Begunkov0d850352021-03-19 17:22:37 +00002192static inline void io_put_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002193{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002194 if (req_ref_put_and_test(req))
Jens Axboedef596e2019-01-09 08:59:42 -07002195 io_free_req(req);
2196}
2197
Pavel Begunkov216578e2020-10-13 09:44:00 +01002198static void io_put_req_deferred_cb(struct callback_head *cb)
2199{
2200 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2201
2202 io_free_req(req);
2203}
2204
2205static void io_free_req_deferred(struct io_kiocb *req)
2206{
Jens Axboe7cbf1722021-02-10 00:03:20 +00002207 req->task_work.func = io_put_req_deferred_cb;
Pavel Begunkova05432f2021-03-19 17:22:38 +00002208 if (unlikely(io_req_task_work_add(req)))
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002209 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002210}
2211
2212static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2213{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002214 if (req_ref_sub_and_test(req, refs))
Pavel Begunkov216578e2020-10-13 09:44:00 +01002215 io_free_req_deferred(req);
2216}
2217
Pavel Begunkov6c503152021-01-04 20:36:36 +00002218static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002219{
2220 /* See comment at the top of this file */
2221 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002222 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002223}
2224
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002225static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2226{
2227 struct io_rings *rings = ctx->rings;
2228
2229 /* make sure SQ entry isn't read before tail */
2230 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2231}
2232
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002233static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002234{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002235 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002236
Jens Axboebcda7ba2020-02-23 16:42:51 -07002237 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2238 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002239 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002240 kfree(kbuf);
2241 return cflags;
2242}
2243
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002244static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2245{
2246 struct io_buffer *kbuf;
2247
2248 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2249 return io_put_kbuf(req, kbuf);
2250}
2251
Jens Axboe4c6e2772020-07-01 11:29:10 -06002252static inline bool io_run_task_work(void)
2253{
2254 if (current->task_works) {
2255 __set_current_state(TASK_RUNNING);
2256 task_work_run();
2257 return true;
2258 }
2259
2260 return false;
2261}
2262
Jens Axboedef596e2019-01-09 08:59:42 -07002263/*
2264 * Find and free completed poll iocbs
2265 */
2266static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2267 struct list_head *done)
2268{
Jens Axboe8237e042019-12-28 10:48:22 -07002269 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002270 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002271
2272 /* order with ->result store in io_complete_rw_iopoll() */
2273 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002274
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002275 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002276 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002277 int cflags = 0;
2278
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002279 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002280 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002281
Pavel Begunkov8c130822021-03-22 01:58:32 +00002282 if (READ_ONCE(req->result) == -EAGAIN &&
2283 !(req->flags & REQ_F_DONT_REISSUE)) {
Pavel Begunkovf1613402021-02-11 18:28:21 +00002284 req->iopoll_completed = 0;
Pavel Begunkov8c130822021-03-22 01:58:32 +00002285 req_ref_get(req);
2286 io_queue_async_work(req);
2287 continue;
Pavel Begunkovf1613402021-02-11 18:28:21 +00002288 }
2289
Jens Axboebcda7ba2020-02-23 16:42:51 -07002290 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002291 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002292
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01002293 __io_cqring_fill_event(ctx, req->user_data, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002294 (*nr_events)++;
2295
Jens Axboede9b4cc2021-02-24 13:28:27 -07002296 if (req_ref_put_and_test(req))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002297 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002298 }
Jens Axboedef596e2019-01-09 08:59:42 -07002299
Jens Axboe09bb8392019-03-13 12:39:28 -06002300 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002301 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002302 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002303}
2304
Jens Axboedef596e2019-01-09 08:59:42 -07002305static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2306 long min)
2307{
2308 struct io_kiocb *req, *tmp;
2309 LIST_HEAD(done);
2310 bool spin;
2311 int ret;
2312
2313 /*
2314 * Only spin for completions if we don't have multiple devices hanging
2315 * off our complete list, and we're under the requested amount.
2316 */
Hao Xu915b3dd2021-06-28 05:37:30 +08002317 spin = !ctx->poll_multi_queue && *nr_events < min;
Jens Axboedef596e2019-01-09 08:59:42 -07002318
2319 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002320 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002321 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002322
2323 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002324 * Move completed and retryable entries to our local lists.
2325 * If we find a request that requires polling, break out
2326 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002327 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002328 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002329 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002330 continue;
2331 }
2332 if (!list_empty(&done))
2333 break;
2334
2335 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2336 if (ret < 0)
2337 break;
2338
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002339 /* iopoll may have completed current req */
2340 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002341 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002342
Jens Axboedef596e2019-01-09 08:59:42 -07002343 if (ret && spin)
2344 spin = false;
2345 ret = 0;
2346 }
2347
2348 if (!list_empty(&done))
2349 io_iopoll_complete(ctx, nr_events, &done);
2350
2351 return ret;
2352}
2353
2354/*
Jens Axboedef596e2019-01-09 08:59:42 -07002355 * We can't just wait for polled events to come to us, we have to actively
2356 * find and complete them.
2357 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002358static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002359{
2360 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2361 return;
2362
2363 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002364 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002365 unsigned int nr_events = 0;
2366
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002367 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002368
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002369 /* let it sleep and repeat later if can't complete a request */
2370 if (nr_events == 0)
2371 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002372 /*
2373 * Ensure we allow local-to-the-cpu processing to take place,
2374 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002375 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002376 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002377 if (need_resched()) {
2378 mutex_unlock(&ctx->uring_lock);
2379 cond_resched();
2380 mutex_lock(&ctx->uring_lock);
2381 }
Jens Axboedef596e2019-01-09 08:59:42 -07002382 }
2383 mutex_unlock(&ctx->uring_lock);
2384}
2385
Pavel Begunkov7668b922020-07-07 16:36:21 +03002386static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002387{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002388 unsigned int nr_events = 0;
Pavel Begunkove9979b32021-04-13 02:58:45 +01002389 int ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002390
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002391 /*
2392 * We disallow the app entering submit/complete with polling, but we
2393 * still need to lock the ring to prevent racing with polled issue
2394 * that got punted to a workqueue.
2395 */
2396 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf39c8a52021-04-13 02:58:46 +01002397 /*
2398 * Don't enter poll loop if we already have events pending.
2399 * If we do, we can potentially be spinning for commands that
2400 * already triggered a CQE (eg in error).
2401 */
Pavel Begunkov5ed7a372021-06-14 23:37:27 +01002402 if (test_bit(0, &ctx->check_cq_overflow))
Pavel Begunkovf39c8a52021-04-13 02:58:46 +01002403 __io_cqring_overflow_flush(ctx, false);
2404 if (io_cqring_events(ctx))
2405 goto out;
Jens Axboedef596e2019-01-09 08:59:42 -07002406 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002407 /*
2408 * If a submit got punted to a workqueue, we can have the
2409 * application entering polling for a command before it gets
2410 * issued. That app will hold the uring_lock for the duration
2411 * of the poll right here, so we need to take a breather every
2412 * now and then to ensure that the issue has a chance to add
2413 * the poll to the issued list. Otherwise we can spin here
2414 * forever, while the workqueue is stuck trying to acquire the
2415 * very same mutex.
2416 */
Pavel Begunkove9979b32021-04-13 02:58:45 +01002417 if (list_empty(&ctx->iopoll_list)) {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002418 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002419 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002420 mutex_lock(&ctx->uring_lock);
Pavel Begunkove9979b32021-04-13 02:58:45 +01002421
2422 if (list_empty(&ctx->iopoll_list))
2423 break;
Jens Axboe500f9fb2019-08-19 12:15:59 -06002424 }
Pavel Begunkovf39c8a52021-04-13 02:58:46 +01002425 ret = io_do_iopoll(ctx, &nr_events, min);
2426 } while (!ret && nr_events < min && !need_resched());
2427out:
Jens Axboe500f9fb2019-08-19 12:15:59 -06002428 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002429 return ret;
2430}
2431
Jens Axboe491381ce2019-10-17 09:20:46 -06002432static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002433{
Jens Axboe491381ce2019-10-17 09:20:46 -06002434 /*
2435 * Tell lockdep we inherited freeze protection from submission
2436 * thread.
2437 */
2438 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov1c986792021-03-22 01:58:31 +00002439 struct super_block *sb = file_inode(req->file)->i_sb;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002440
Pavel Begunkov1c986792021-03-22 01:58:31 +00002441 __sb_writers_acquired(sb, SB_FREEZE_WRITE);
2442 sb_end_write(sb);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002443 }
2444}
2445
Jens Axboeb63534c2020-06-04 11:28:00 -06002446#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002447static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002448{
Pavel Begunkovab454432021-03-22 01:58:33 +00002449 struct io_async_rw *rw = req->async_data;
Jens Axboeb63534c2020-06-04 11:28:00 -06002450
Pavel Begunkovab454432021-03-22 01:58:33 +00002451 if (!rw)
2452 return !io_req_prep_async(req);
2453 /* may have left rw->iter inconsistent on -EIOCBQUEUED */
2454 iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter));
2455 return true;
Jens Axboeb63534c2020-06-04 11:28:00 -06002456}
Jens Axboeb63534c2020-06-04 11:28:00 -06002457
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002458static bool io_rw_should_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002459{
Jens Axboe355afae2020-09-02 09:30:31 -06002460 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002461 struct io_ring_ctx *ctx = req->ctx;
Jens Axboeb63534c2020-06-04 11:28:00 -06002462
Jens Axboe355afae2020-09-02 09:30:31 -06002463 if (!S_ISBLK(mode) && !S_ISREG(mode))
2464 return false;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002465 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() &&
2466 !(ctx->flags & IORING_SETUP_IOPOLL)))
Jens Axboeb63534c2020-06-04 11:28:00 -06002467 return false;
Jens Axboe7c977a52021-02-23 19:17:35 -07002468 /*
2469 * If ref is dying, we might be running poll reap from the exit work.
2470 * Don't attempt to reissue from that path, just let it fail with
2471 * -EAGAIN.
2472 */
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002473 if (percpu_ref_is_dying(&ctx->refs))
2474 return false;
2475 return true;
2476}
Jens Axboee82ad482021-04-02 19:45:34 -06002477#else
Jens Axboea1ff1e32021-04-12 06:40:02 -06002478static bool io_resubmit_prep(struct io_kiocb *req)
2479{
2480 return false;
2481}
Jens Axboee82ad482021-04-02 19:45:34 -06002482static bool io_rw_should_reissue(struct io_kiocb *req)
2483{
2484 return false;
2485}
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002486#endif
2487
Jens Axboea1d7c392020-06-22 11:09:46 -06002488static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002489 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002490{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002491 int cflags = 0;
2492
Pavel Begunkovb65c1282021-03-22 01:45:59 +00002493 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2494 kiocb_end_write(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002495 if (res != req->result) {
2496 if ((res == -EAGAIN || res == -EOPNOTSUPP) &&
2497 io_rw_should_reissue(req)) {
2498 req->flags |= REQ_F_REISSUE;
2499 return;
2500 }
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01002501 req_set_fail(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002502 }
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002503 if (req->flags & REQ_F_BUFFER_SELECTED)
2504 cflags = io_put_rw_kbuf(req);
2505 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002506}
2507
2508static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2509{
Jens Axboe9adbd452019-12-20 08:45:55 -07002510 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002511
Pavel Begunkov889fca72021-02-10 00:03:09 +00002512 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002513}
2514
Jens Axboedef596e2019-01-09 08:59:42 -07002515static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2516{
Jens Axboe9adbd452019-12-20 08:45:55 -07002517 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002518
Jens Axboe491381ce2019-10-17 09:20:46 -06002519 if (kiocb->ki_flags & IOCB_WRITE)
2520 kiocb_end_write(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002521 if (unlikely(res != req->result)) {
Jens Axboea1ff1e32021-04-12 06:40:02 -06002522 if (!(res == -EAGAIN && io_rw_should_reissue(req) &&
2523 io_resubmit_prep(req))) {
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01002524 req_set_fail(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002525 req->flags |= REQ_F_DONT_REISSUE;
2526 }
Pavel Begunkov8c130822021-03-22 01:58:32 +00002527 }
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002528
2529 WRITE_ONCE(req->result, res);
Jens Axboeb9b0e0d2021-02-23 08:18:36 -07002530 /* order with io_iopoll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002531 smp_wmb();
2532 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002533}
2534
2535/*
2536 * After the iocb has been issued, it's safe to be found on the poll list.
2537 * Adding the kiocb to the list AFTER submission ensures that we don't
Pavel Begunkovf39c8a52021-04-13 02:58:46 +01002538 * find it from a io_do_iopoll() thread before the issuer is done
Jens Axboedef596e2019-01-09 08:59:42 -07002539 * accessing the kiocb cookie.
2540 */
Pavel Begunkovcb3d8972021-06-14 02:36:14 +01002541static void io_iopoll_req_issued(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07002542{
2543 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcb3d8972021-06-14 02:36:14 +01002544 const bool in_async = io_wq_current_is_worker();
2545
2546 /* workqueue context doesn't hold uring_lock, grab it now */
2547 if (unlikely(in_async))
2548 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002549
2550 /*
2551 * Track whether we have multiple files in our lists. This will impact
2552 * how we do polling eventually, not spinning if we're on potentially
2553 * different devices.
2554 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002555 if (list_empty(&ctx->iopoll_list)) {
Hao Xu915b3dd2021-06-28 05:37:30 +08002556 ctx->poll_multi_queue = false;
2557 } else if (!ctx->poll_multi_queue) {
Jens Axboedef596e2019-01-09 08:59:42 -07002558 struct io_kiocb *list_req;
Hao Xu915b3dd2021-06-28 05:37:30 +08002559 unsigned int queue_num0, queue_num1;
Jens Axboedef596e2019-01-09 08:59:42 -07002560
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002561 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002562 inflight_entry);
Hao Xu915b3dd2021-06-28 05:37:30 +08002563
2564 if (list_req->file != req->file) {
2565 ctx->poll_multi_queue = true;
2566 } else {
2567 queue_num0 = blk_qc_t_to_queue_num(list_req->rw.kiocb.ki_cookie);
2568 queue_num1 = blk_qc_t_to_queue_num(req->rw.kiocb.ki_cookie);
2569 if (queue_num0 != queue_num1)
2570 ctx->poll_multi_queue = true;
2571 }
Jens Axboedef596e2019-01-09 08:59:42 -07002572 }
2573
2574 /*
2575 * For fast devices, IO may have already completed. If it has, add
2576 * it to the front so we find it first.
2577 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002578 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002579 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002580 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002581 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002582
Pavel Begunkovcb3d8972021-06-14 02:36:14 +01002583 if (unlikely(in_async)) {
2584 /*
2585 * If IORING_SETUP_SQPOLL is enabled, sqes are either handle
2586 * in sq thread task context or in io worker task context. If
2587 * current task context is sq thread, we don't need to check
2588 * whether should wake up sq thread.
2589 */
2590 if ((ctx->flags & IORING_SETUP_SQPOLL) &&
2591 wq_has_sleeper(&ctx->sq_data->wait))
2592 wake_up(&ctx->sq_data->wait);
2593
2594 mutex_unlock(&ctx->uring_lock);
2595 }
Jens Axboedef596e2019-01-09 08:59:42 -07002596}
2597
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002598static inline void io_state_file_put(struct io_submit_state *state)
2599{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002600 if (state->file_refs) {
2601 fput_many(state->file, state->file_refs);
2602 state->file_refs = 0;
2603 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002604}
2605
2606/*
2607 * Get as many references to a file as we have IOs left in this submission,
2608 * assuming most submissions are for one file, or at least that each file
2609 * has more than one submission.
2610 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002611static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002612{
2613 if (!state)
2614 return fget(fd);
2615
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002616 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002617 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002618 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002619 return state->file;
2620 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002621 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002622 }
2623 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002624 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002625 return NULL;
2626
2627 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002628 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002629 return state->file;
2630}
2631
Jens Axboe4503b762020-06-01 10:00:27 -06002632static bool io_bdev_nowait(struct block_device *bdev)
2633{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002634 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002635}
2636
Jens Axboe2b188cc2019-01-07 10:46:33 -07002637/*
2638 * If we tracked the file through the SCM inflight mechanism, we could support
2639 * any file. For now, just ensure that anything potentially problematic is done
2640 * inline.
2641 */
Jens Axboe7b29f922021-03-12 08:30:14 -07002642static bool __io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002643{
2644 umode_t mode = file_inode(file)->i_mode;
2645
Jens Axboe4503b762020-06-01 10:00:27 -06002646 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002647 if (IS_ENABLED(CONFIG_BLOCK) &&
2648 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002649 return true;
2650 return false;
2651 }
Pavel Begunkov976517f2021-06-09 12:07:25 +01002652 if (S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002653 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002654 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002655 if (IS_ENABLED(CONFIG_BLOCK) &&
2656 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002657 file->f_op != &io_uring_fops)
2658 return true;
2659 return false;
2660 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002661
Jens Axboec5b85622020-06-09 19:23:05 -06002662 /* any ->read/write should understand O_NONBLOCK */
2663 if (file->f_flags & O_NONBLOCK)
2664 return true;
2665
Jens Axboeaf197f52020-04-28 13:15:06 -06002666 if (!(file->f_mode & FMODE_NOWAIT))
2667 return false;
2668
2669 if (rw == READ)
2670 return file->f_op->read_iter != NULL;
2671
2672 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002673}
2674
Jens Axboe7b29f922021-03-12 08:30:14 -07002675static bool io_file_supports_async(struct io_kiocb *req, int rw)
2676{
2677 if (rw == READ && (req->flags & REQ_F_ASYNC_READ))
2678 return true;
2679 else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE))
2680 return true;
2681
2682 return __io_file_supports_async(req->file, rw);
2683}
2684
Pavel Begunkova88fc402020-09-30 22:57:53 +03002685static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002686{
Jens Axboedef596e2019-01-09 08:59:42 -07002687 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002688 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002689 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002690 unsigned ioprio;
2691 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002692
Jens Axboe7b29f922021-03-12 08:30:14 -07002693 if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002694 req->flags |= REQ_F_ISREG;
2695
Jens Axboe2b188cc2019-01-07 10:46:33 -07002696 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002697 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002698 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002699 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002700 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002701 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002702 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2703 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2704 if (unlikely(ret))
2705 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002706
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002707 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
2708 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
2709 req->flags |= REQ_F_NOWAIT;
2710
Jens Axboe2b188cc2019-01-07 10:46:33 -07002711 ioprio = READ_ONCE(sqe->ioprio);
2712 if (ioprio) {
2713 ret = ioprio_check_cap(ioprio);
2714 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002715 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002716
2717 kiocb->ki_ioprio = ioprio;
2718 } else
2719 kiocb->ki_ioprio = get_current_ioprio();
2720
Jens Axboedef596e2019-01-09 08:59:42 -07002721 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002722 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2723 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002724 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002725
Jens Axboedef596e2019-01-09 08:59:42 -07002726 kiocb->ki_flags |= IOCB_HIPRI;
2727 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002728 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002729 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002730 if (kiocb->ki_flags & IOCB_HIPRI)
2731 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002732 kiocb->ki_complete = io_complete_rw;
2733 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002734
Pavel Begunkoveae071c2021-04-25 14:32:24 +01002735 if (req->opcode == IORING_OP_READ_FIXED ||
2736 req->opcode == IORING_OP_WRITE_FIXED) {
2737 req->imu = NULL;
2738 io_req_set_rsrc_node(req);
2739 }
2740
Jens Axboe3529d8c2019-12-19 18:24:38 -07002741 req->rw.addr = READ_ONCE(sqe->addr);
2742 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002743 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002744 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002745}
2746
2747static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2748{
2749 switch (ret) {
2750 case -EIOCBQUEUED:
2751 break;
2752 case -ERESTARTSYS:
2753 case -ERESTARTNOINTR:
2754 case -ERESTARTNOHAND:
2755 case -ERESTART_RESTARTBLOCK:
2756 /*
2757 * We can't just restart the syscall, since previously
2758 * submitted sqes may already be in progress. Just fail this
2759 * IO with EINTR.
2760 */
2761 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002762 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002763 default:
2764 kiocb->ki_complete(kiocb, ret, 0);
2765 }
2766}
2767
Jens Axboea1d7c392020-06-22 11:09:46 -06002768static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002769 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06002770{
Jens Axboeba042912019-12-25 16:33:42 -07002771 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002772 struct io_async_rw *io = req->async_data;
Pavel Begunkov97284632021-04-08 19:28:03 +01002773 bool check_reissue = kiocb->ki_complete == io_complete_rw;
Jens Axboeba042912019-12-25 16:33:42 -07002774
Jens Axboe227c0c92020-08-13 11:51:40 -06002775 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002776 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002777 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002778 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002779 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002780 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002781 }
2782
Jens Axboeba042912019-12-25 16:33:42 -07002783 if (req->flags & REQ_F_CUR_POS)
2784 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002785 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00002786 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002787 else
2788 io_rw_done(kiocb, ret);
Pavel Begunkov97284632021-04-08 19:28:03 +01002789
Pavel Begunkovfe7e3252021-06-24 15:09:57 +01002790 if (check_reissue && (req->flags & REQ_F_REISSUE)) {
Pavel Begunkov97284632021-04-08 19:28:03 +01002791 req->flags &= ~REQ_F_REISSUE;
Jens Axboea7be7c22021-04-15 11:31:14 -06002792 if (io_resubmit_prep(req)) {
Pavel Begunkov8c130822021-03-22 01:58:32 +00002793 req_ref_get(req);
2794 io_queue_async_work(req);
2795 } else {
Pavel Begunkov97284632021-04-08 19:28:03 +01002796 int cflags = 0;
2797
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01002798 req_set_fail(req);
Pavel Begunkov97284632021-04-08 19:28:03 +01002799 if (req->flags & REQ_F_BUFFER_SELECTED)
2800 cflags = io_put_rw_kbuf(req);
2801 __io_req_complete(req, issue_flags, ret, cflags);
2802 }
2803 }
Jens Axboeba816ad2019-09-28 11:36:45 -06002804}
2805
Pavel Begunkoveae071c2021-04-25 14:32:24 +01002806static int __io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter,
2807 struct io_mapped_ubuf *imu)
Jens Axboeedafcce2019-01-09 09:16:05 -07002808{
Jens Axboe9adbd452019-12-20 08:45:55 -07002809 size_t len = req->rw.len;
Pavel Begunkov75769e32021-04-01 15:43:54 +01002810 u64 buf_end, buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002811 size_t offset;
Jens Axboeedafcce2019-01-09 09:16:05 -07002812
Pavel Begunkov75769e32021-04-01 15:43:54 +01002813 if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end)))
Jens Axboeedafcce2019-01-09 09:16:05 -07002814 return -EFAULT;
2815 /* not inside the mapped region */
Pavel Begunkov4751f532021-04-01 15:43:55 +01002816 if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end))
Jens Axboeedafcce2019-01-09 09:16:05 -07002817 return -EFAULT;
2818
2819 /*
2820 * May not be a start of buffer, set size appropriately
2821 * and advance us to the beginning.
2822 */
2823 offset = buf_addr - imu->ubuf;
2824 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002825
2826 if (offset) {
2827 /*
2828 * Don't use iov_iter_advance() here, as it's really slow for
2829 * using the latter parts of a big fixed buffer - it iterates
2830 * over each segment manually. We can cheat a bit here, because
2831 * we know that:
2832 *
2833 * 1) it's a BVEC iter, we set it up
2834 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2835 * first and last bvec
2836 *
2837 * So just find our index, and adjust the iterator afterwards.
2838 * If the offset is within the first bvec (or the whole first
2839 * bvec, just use iov_iter_advance(). This makes it easier
2840 * since we can just skip the first segment, which may not
2841 * be PAGE_SIZE aligned.
2842 */
2843 const struct bio_vec *bvec = imu->bvec;
2844
2845 if (offset <= bvec->bv_len) {
2846 iov_iter_advance(iter, offset);
2847 } else {
2848 unsigned long seg_skip;
2849
2850 /* skip first vec */
2851 offset -= bvec->bv_len;
2852 seg_skip = 1 + (offset >> PAGE_SHIFT);
2853
2854 iter->bvec = bvec + seg_skip;
2855 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002856 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002857 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002858 }
2859 }
2860
Pavel Begunkov847595d2021-02-04 13:52:06 +00002861 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07002862}
2863
Pavel Begunkoveae071c2021-04-25 14:32:24 +01002864static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
2865{
2866 struct io_ring_ctx *ctx = req->ctx;
2867 struct io_mapped_ubuf *imu = req->imu;
2868 u16 index, buf_index = req->buf_index;
2869
2870 if (likely(!imu)) {
2871 if (unlikely(buf_index >= ctx->nr_user_bufs))
2872 return -EFAULT;
2873 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2874 imu = READ_ONCE(ctx->user_bufs[index]);
2875 req->imu = imu;
2876 }
2877 return __io_import_fixed(req, rw, iter, imu);
2878}
2879
Jens Axboebcda7ba2020-02-23 16:42:51 -07002880static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2881{
2882 if (needs_lock)
2883 mutex_unlock(&ctx->uring_lock);
2884}
2885
2886static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2887{
2888 /*
2889 * "Normal" inline submissions always hold the uring_lock, since we
2890 * grab it from the system call. Same is true for the SQPOLL offload.
2891 * The only exception is when we've detached the request and issue it
2892 * from an async worker thread, grab the lock for that case.
2893 */
2894 if (needs_lock)
2895 mutex_lock(&ctx->uring_lock);
2896}
2897
2898static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2899 int bgid, struct io_buffer *kbuf,
2900 bool needs_lock)
2901{
2902 struct io_buffer *head;
2903
2904 if (req->flags & REQ_F_BUFFER_SELECTED)
2905 return kbuf;
2906
2907 io_ring_submit_lock(req->ctx, needs_lock);
2908
2909 lockdep_assert_held(&req->ctx->uring_lock);
2910
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002911 head = xa_load(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002912 if (head) {
2913 if (!list_empty(&head->list)) {
2914 kbuf = list_last_entry(&head->list, struct io_buffer,
2915 list);
2916 list_del(&kbuf->list);
2917 } else {
2918 kbuf = head;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002919 xa_erase(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002920 }
2921 if (*len > kbuf->len)
2922 *len = kbuf->len;
2923 } else {
2924 kbuf = ERR_PTR(-ENOBUFS);
2925 }
2926
2927 io_ring_submit_unlock(req->ctx, needs_lock);
2928
2929 return kbuf;
2930}
2931
Jens Axboe4d954c22020-02-27 07:31:19 -07002932static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2933 bool needs_lock)
2934{
2935 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002936 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002937
2938 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002939 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002940 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2941 if (IS_ERR(kbuf))
2942 return kbuf;
2943 req->rw.addr = (u64) (unsigned long) kbuf;
2944 req->flags |= REQ_F_BUFFER_SELECTED;
2945 return u64_to_user_ptr(kbuf->addr);
2946}
2947
2948#ifdef CONFIG_COMPAT
2949static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2950 bool needs_lock)
2951{
2952 struct compat_iovec __user *uiov;
2953 compat_ssize_t clen;
2954 void __user *buf;
2955 ssize_t len;
2956
2957 uiov = u64_to_user_ptr(req->rw.addr);
2958 if (!access_ok(uiov, sizeof(*uiov)))
2959 return -EFAULT;
2960 if (__get_user(clen, &uiov->iov_len))
2961 return -EFAULT;
2962 if (clen < 0)
2963 return -EINVAL;
2964
2965 len = clen;
2966 buf = io_rw_buffer_select(req, &len, needs_lock);
2967 if (IS_ERR(buf))
2968 return PTR_ERR(buf);
2969 iov[0].iov_base = buf;
2970 iov[0].iov_len = (compat_size_t) len;
2971 return 0;
2972}
2973#endif
2974
2975static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2976 bool needs_lock)
2977{
2978 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
2979 void __user *buf;
2980 ssize_t len;
2981
2982 if (copy_from_user(iov, uiov, sizeof(*uiov)))
2983 return -EFAULT;
2984
2985 len = iov[0].iov_len;
2986 if (len < 0)
2987 return -EINVAL;
2988 buf = io_rw_buffer_select(req, &len, needs_lock);
2989 if (IS_ERR(buf))
2990 return PTR_ERR(buf);
2991 iov[0].iov_base = buf;
2992 iov[0].iov_len = len;
2993 return 0;
2994}
2995
2996static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2997 bool needs_lock)
2998{
Jens Axboedddb3e22020-06-04 11:27:01 -06002999 if (req->flags & REQ_F_BUFFER_SELECTED) {
3000 struct io_buffer *kbuf;
3001
3002 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
3003 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
3004 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07003005 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06003006 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00003007 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07003008 return -EINVAL;
3009
3010#ifdef CONFIG_COMPAT
3011 if (req->ctx->compat)
3012 return io_compat_import(req, iov, needs_lock);
3013#endif
3014
3015 return __io_iov_buffer_select(req, iov, needs_lock);
3016}
3017
Pavel Begunkov847595d2021-02-04 13:52:06 +00003018static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
3019 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003020{
Jens Axboe9adbd452019-12-20 08:45:55 -07003021 void __user *buf = u64_to_user_ptr(req->rw.addr);
3022 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003023 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07003024 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07003025
Pavel Begunkov7d009162019-11-25 23:14:40 +03003026 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07003027 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07003028 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07003029 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003030
Jens Axboebcda7ba2020-02-23 16:42:51 -07003031 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003032 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07003033 return -EINVAL;
3034
Jens Axboe3a6820f2019-12-22 15:19:35 -07003035 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07003036 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07003037 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03003038 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07003039 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06003040 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07003041 }
3042
Jens Axboe3a6820f2019-12-22 15:19:35 -07003043 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
3044 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00003045 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07003046 }
3047
Jens Axboe4d954c22020-02-27 07:31:19 -07003048 if (req->flags & REQ_F_BUFFER_SELECT) {
3049 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00003050 if (!ret)
3051 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07003052 *iovec = NULL;
3053 return ret;
3054 }
3055
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02003056 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
3057 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003058}
3059
Jens Axboe0fef9482020-08-26 10:36:20 -06003060static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
3061{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03003062 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06003063}
3064
Jens Axboe32960612019-09-23 11:05:34 -06003065/*
3066 * For files that don't have ->read_iter() and ->write_iter(), handle them
3067 * by looping over ->read() or ->write() manually.
3068 */
Jens Axboe4017eb92020-10-22 14:14:12 -06003069static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06003070{
Jens Axboe4017eb92020-10-22 14:14:12 -06003071 struct kiocb *kiocb = &req->rw.kiocb;
3072 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003073 ssize_t ret = 0;
3074
3075 /*
3076 * Don't support polled IO through this interface, and we can't
3077 * support non-blocking either. For the latter, this just causes
3078 * the kiocb to be handled from an async context.
3079 */
3080 if (kiocb->ki_flags & IOCB_HIPRI)
3081 return -EOPNOTSUPP;
3082 if (kiocb->ki_flags & IOCB_NOWAIT)
3083 return -EAGAIN;
3084
3085 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003086 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003087 ssize_t nr;
3088
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003089 if (!iov_iter_is_bvec(iter)) {
3090 iovec = iov_iter_iovec(iter);
3091 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003092 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3093 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003094 }
3095
Jens Axboe32960612019-09-23 11:05:34 -06003096 if (rw == READ) {
3097 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003098 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003099 } else {
3100 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003101 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003102 }
3103
3104 if (nr < 0) {
3105 if (!ret)
3106 ret = nr;
3107 break;
3108 }
3109 ret += nr;
3110 if (nr != iovec.iov_len)
3111 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003112 req->rw.len -= nr;
3113 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003114 iov_iter_advance(iter, nr);
3115 }
3116
3117 return ret;
3118}
3119
Jens Axboeff6165b2020-08-13 09:47:43 -06003120static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3121 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003122{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003123 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003124
Jens Axboeff6165b2020-08-13 09:47:43 -06003125 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003126 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003127 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003128 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003129 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003130 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003131 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003132 unsigned iov_off = 0;
3133
3134 rw->iter.iov = rw->fast_iov;
3135 if (iter->iov != fast_iov) {
3136 iov_off = iter->iov - fast_iov;
3137 rw->iter.iov += iov_off;
3138 }
3139 if (rw->fast_iov != fast_iov)
3140 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003141 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003142 } else {
3143 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003144 }
3145}
3146
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003147static inline int io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003148{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003149 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3150 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3151 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003152}
3153
Jens Axboeff6165b2020-08-13 09:47:43 -06003154static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3155 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003156 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003157{
Pavel Begunkov26f05052021-02-28 22:35:18 +00003158 if (!force && !io_op_defs[req->opcode].needs_async_setup)
Jens Axboe74566df2020-01-13 19:23:24 -07003159 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003160 if (!req->async_data) {
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003161 if (io_alloc_async_data(req)) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003162 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003163 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003164 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003165
Jens Axboeff6165b2020-08-13 09:47:43 -06003166 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003167 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003168 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003169}
3170
Pavel Begunkov73debe62020-09-30 22:57:54 +03003171static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003172{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003173 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003174 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003175 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003176
Pavel Begunkov2846c482020-11-07 13:16:27 +00003177 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003178 if (unlikely(ret < 0))
3179 return ret;
3180
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003181 iorw->bytes_done = 0;
3182 iorw->free_iovec = iov;
3183 if (iov)
3184 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003185 return 0;
3186}
3187
Pavel Begunkov73debe62020-09-30 22:57:54 +03003188static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003189{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003190 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3191 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003192 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003193}
3194
Jens Axboec1dd91d2020-08-03 16:43:59 -06003195/*
3196 * This is our waitqueue callback handler, registered through lock_page_async()
3197 * when we initially tried to do the IO with the iocb armed our waitqueue.
3198 * This gets called when the page is unlocked, and we generally expect that to
3199 * happen when the page IO is completed and the page is now uptodate. This will
3200 * queue a task_work based retry of the operation, attempting to copy the data
3201 * again. If the latter fails because the page was NOT uptodate, then we will
3202 * do a thread based blocking retry of the operation. That's the unexpected
3203 * slow path.
3204 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003205static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3206 int sync, void *arg)
3207{
3208 struct wait_page_queue *wpq;
3209 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003210 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003211
3212 wpq = container_of(wait, struct wait_page_queue, wait);
3213
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003214 if (!wake_page_match(wpq, key))
3215 return 0;
3216
Hao Xuc8d317a2020-09-29 20:00:45 +08003217 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003218 list_del_init(&wait->entry);
3219
Jens Axboebcf5a062020-05-22 09:24:42 -06003220 /* submit ref gets dropped, acquire a new one */
Jens Axboede9b4cc2021-02-24 13:28:27 -07003221 req_ref_get(req);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003222 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003223 return 1;
3224}
3225
Jens Axboec1dd91d2020-08-03 16:43:59 -06003226/*
3227 * This controls whether a given IO request should be armed for async page
3228 * based retry. If we return false here, the request is handed to the async
3229 * worker threads for retry. If we're doing buffered reads on a regular file,
3230 * we prepare a private wait_page_queue entry and retry the operation. This
3231 * will either succeed because the page is now uptodate and unlocked, or it
3232 * will register a callback when the page is unlocked at IO completion. Through
3233 * that callback, io_uring uses task_work to setup a retry of the operation.
3234 * That retry will attempt the buffered read again. The retry will generally
3235 * succeed, or in rare cases where it fails, we then fall back to using the
3236 * async worker threads for a blocking retry.
3237 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003238static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003239{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003240 struct io_async_rw *rw = req->async_data;
3241 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003242 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003243
3244 /* never retry for NOWAIT, we just complete with -EAGAIN */
3245 if (req->flags & REQ_F_NOWAIT)
3246 return false;
3247
Jens Axboe227c0c92020-08-13 11:51:40 -06003248 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003249 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003250 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003251
Jens Axboebcf5a062020-05-22 09:24:42 -06003252 /*
3253 * just use poll if we can, and don't attempt if the fs doesn't
3254 * support callback based unlocks
3255 */
3256 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3257 return false;
3258
Jens Axboe3b2a4432020-08-16 10:58:43 -07003259 wait->wait.func = io_async_buf_func;
3260 wait->wait.private = req;
3261 wait->wait.flags = 0;
3262 INIT_LIST_HEAD(&wait->wait.entry);
3263 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003264 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003265 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003266 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003267}
3268
Pavel Begunkovaeab9502021-06-14 02:36:24 +01003269static inline int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
Jens Axboebcf5a062020-05-22 09:24:42 -06003270{
3271 if (req->file->f_op->read_iter)
3272 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003273 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003274 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003275 else
3276 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003277}
3278
Pavel Begunkov889fca72021-02-10 00:03:09 +00003279static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003280{
3281 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003282 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003283 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003284 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003285 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003286 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003287
Pavel Begunkov2846c482020-11-07 13:16:27 +00003288 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003289 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003290 iovec = NULL;
3291 } else {
3292 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3293 if (ret < 0)
3294 return ret;
3295 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003296 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003297 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003298
Jens Axboefd6c2e42019-12-18 12:19:41 -07003299 /* Ensure we clear previously set non-block flag */
3300 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003301 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003302 else
3303 kiocb->ki_flags |= IOCB_NOWAIT;
3304
Pavel Begunkov24c74672020-06-21 13:09:51 +03003305 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003306 if (force_nonblock && !io_file_supports_async(req, READ)) {
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003307 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003308 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003309 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003310
Pavel Begunkov632546c2020-11-07 13:16:26 +00003311 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003312 if (unlikely(ret)) {
3313 kfree(iovec);
3314 return ret;
3315 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003316
Jens Axboe227c0c92020-08-13 11:51:40 -06003317 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003318
Jens Axboe230d50d2021-04-01 20:41:15 -06003319 if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) {
Pavel Begunkov6ad7f232021-04-08 01:54:39 +01003320 req->flags &= ~REQ_F_REISSUE;
Jens Axboeeefdf302020-08-27 16:40:19 -06003321 /* IOPOLL retry should happen for io-wq threads */
3322 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003323 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003324 /* no retry on NONBLOCK nor RWF_NOWAIT */
3325 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003326 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003327 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003328 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003329 ret = 0;
Jens Axboe230d50d2021-04-01 20:41:15 -06003330 } else if (ret == -EIOCBQUEUED) {
3331 goto out_free;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003332 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003333 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003334 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003335 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003336 }
3337
Jens Axboe227c0c92020-08-13 11:51:40 -06003338 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003339 if (ret2)
3340 return ret2;
3341
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003342 iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003343 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003344 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003345 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003346
Pavel Begunkovb23df912021-02-04 13:52:04 +00003347 do {
3348 io_size -= ret;
3349 rw->bytes_done += ret;
3350 /* if we can retry, do so with the callbacks armed */
3351 if (!io_rw_should_retry(req)) {
3352 kiocb->ki_flags &= ~IOCB_WAITQ;
3353 return -EAGAIN;
3354 }
3355
3356 /*
3357 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3358 * we get -EIOCBQUEUED, then we'll get a notification when the
3359 * desired page gets unlocked. We can also get a partial read
3360 * here, and if we do, then just retry at the new offset.
3361 */
3362 ret = io_iter_do_read(req, iter);
3363 if (ret == -EIOCBQUEUED)
3364 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003365 /* we got some bytes, but not all. retry. */
Jens Axboeb5b0ecb2021-03-04 21:02:58 -07003366 kiocb->ki_flags &= ~IOCB_WAITQ;
Pavel Begunkovb23df912021-02-04 13:52:04 +00003367 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003368done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003369 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003370out_free:
3371 /* it's faster to check here then delegate to kfree */
3372 if (iovec)
3373 kfree(iovec);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003374 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003375}
3376
Pavel Begunkov73debe62020-09-30 22:57:54 +03003377static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003378{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003379 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3380 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003381 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003382}
3383
Pavel Begunkov889fca72021-02-10 00:03:09 +00003384static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003385{
3386 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003387 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003388 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003389 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003390 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003391 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003392
Pavel Begunkov2846c482020-11-07 13:16:27 +00003393 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003394 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003395 iovec = NULL;
3396 } else {
3397 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3398 if (ret < 0)
3399 return ret;
3400 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003401 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003402 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003403
Jens Axboefd6c2e42019-12-18 12:19:41 -07003404 /* Ensure we clear previously set non-block flag */
3405 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003406 kiocb->ki_flags &= ~IOCB_NOWAIT;
3407 else
3408 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003409
Pavel Begunkov24c74672020-06-21 13:09:51 +03003410 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003411 if (force_nonblock && !io_file_supports_async(req, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003412 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003413
Jens Axboe10d59342019-12-09 20:16:22 -07003414 /* file path doesn't support NOWAIT for non-direct_IO */
3415 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3416 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003417 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003418
Pavel Begunkov632546c2020-11-07 13:16:26 +00003419 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003420 if (unlikely(ret))
3421 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003422
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003423 /*
3424 * Open-code file_start_write here to grab freeze protection,
3425 * which will be released by another thread in
3426 * io_complete_rw(). Fool lockdep by telling it the lock got
3427 * released so that it doesn't complain about the held lock when
3428 * we return to userspace.
3429 */
3430 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003431 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003432 __sb_writers_release(file_inode(req->file)->i_sb,
3433 SB_FREEZE_WRITE);
3434 }
3435 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003436
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003437 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003438 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003439 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003440 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003441 else
3442 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003443
Pavel Begunkov6ad7f232021-04-08 01:54:39 +01003444 if (req->flags & REQ_F_REISSUE) {
3445 req->flags &= ~REQ_F_REISSUE;
Jens Axboe230d50d2021-04-01 20:41:15 -06003446 ret2 = -EAGAIN;
Pavel Begunkov6ad7f232021-04-08 01:54:39 +01003447 }
Jens Axboe230d50d2021-04-01 20:41:15 -06003448
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003449 /*
3450 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3451 * retry them without IOCB_NOWAIT.
3452 */
3453 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3454 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003455 /* no retry on NONBLOCK nor RWF_NOWAIT */
3456 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003457 goto done;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003458 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003459 /* IOPOLL retry should happen for io-wq threads */
3460 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3461 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003462done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003463 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003464 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003465copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003466 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003467 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003468 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003469 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003470 }
Jens Axboe31b51512019-01-18 22:56:34 -07003471out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003472 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003473 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003474 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003475 return ret;
3476}
3477
Jens Axboe80a261f2020-09-28 14:23:58 -06003478static int io_renameat_prep(struct io_kiocb *req,
3479 const struct io_uring_sqe *sqe)
3480{
3481 struct io_rename *ren = &req->rename;
3482 const char __user *oldf, *newf;
3483
Jens Axboeed7eb252021-06-23 09:04:13 -06003484 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3485 return -EINVAL;
3486 if (sqe->ioprio || sqe->buf_index)
3487 return -EINVAL;
Jens Axboe80a261f2020-09-28 14:23:58 -06003488 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3489 return -EBADF;
3490
3491 ren->old_dfd = READ_ONCE(sqe->fd);
3492 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3493 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3494 ren->new_dfd = READ_ONCE(sqe->len);
3495 ren->flags = READ_ONCE(sqe->rename_flags);
3496
3497 ren->oldpath = getname(oldf);
3498 if (IS_ERR(ren->oldpath))
3499 return PTR_ERR(ren->oldpath);
3500
3501 ren->newpath = getname(newf);
3502 if (IS_ERR(ren->newpath)) {
3503 putname(ren->oldpath);
3504 return PTR_ERR(ren->newpath);
3505 }
3506
3507 req->flags |= REQ_F_NEED_CLEANUP;
3508 return 0;
3509}
3510
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003511static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003512{
3513 struct io_rename *ren = &req->rename;
3514 int ret;
3515
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003516 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003517 return -EAGAIN;
3518
3519 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3520 ren->newpath, ren->flags);
3521
3522 req->flags &= ~REQ_F_NEED_CLEANUP;
3523 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003524 req_set_fail(req);
Jens Axboe80a261f2020-09-28 14:23:58 -06003525 io_req_complete(req, ret);
3526 return 0;
3527}
3528
Jens Axboe14a11432020-09-28 14:27:37 -06003529static int io_unlinkat_prep(struct io_kiocb *req,
3530 const struct io_uring_sqe *sqe)
3531{
3532 struct io_unlink *un = &req->unlink;
3533 const char __user *fname;
3534
Jens Axboe22634bc2021-06-23 09:07:45 -06003535 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3536 return -EINVAL;
3537 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
3538 return -EINVAL;
Jens Axboe14a11432020-09-28 14:27:37 -06003539 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3540 return -EBADF;
3541
3542 un->dfd = READ_ONCE(sqe->fd);
3543
3544 un->flags = READ_ONCE(sqe->unlink_flags);
3545 if (un->flags & ~AT_REMOVEDIR)
3546 return -EINVAL;
3547
3548 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3549 un->filename = getname(fname);
3550 if (IS_ERR(un->filename))
3551 return PTR_ERR(un->filename);
3552
3553 req->flags |= REQ_F_NEED_CLEANUP;
3554 return 0;
3555}
3556
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003557static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003558{
3559 struct io_unlink *un = &req->unlink;
3560 int ret;
3561
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003562 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003563 return -EAGAIN;
3564
3565 if (un->flags & AT_REMOVEDIR)
3566 ret = do_rmdir(un->dfd, un->filename);
3567 else
3568 ret = do_unlinkat(un->dfd, un->filename);
3569
3570 req->flags &= ~REQ_F_NEED_CLEANUP;
3571 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003572 req_set_fail(req);
Jens Axboe14a11432020-09-28 14:27:37 -06003573 io_req_complete(req, ret);
3574 return 0;
3575}
3576
Jens Axboe36f4fa62020-09-05 11:14:22 -06003577static int io_shutdown_prep(struct io_kiocb *req,
3578 const struct io_uring_sqe *sqe)
3579{
3580#if defined(CONFIG_NET)
3581 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3582 return -EINVAL;
3583 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3584 sqe->buf_index)
3585 return -EINVAL;
3586
3587 req->shutdown.how = READ_ONCE(sqe->len);
3588 return 0;
3589#else
3590 return -EOPNOTSUPP;
3591#endif
3592}
3593
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003594static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003595{
3596#if defined(CONFIG_NET)
3597 struct socket *sock;
3598 int ret;
3599
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003600 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003601 return -EAGAIN;
3602
Linus Torvalds48aba792020-12-16 12:44:05 -08003603 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003604 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003605 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003606
3607 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003608 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003609 req_set_fail(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003610 io_req_complete(req, ret);
3611 return 0;
3612#else
3613 return -EOPNOTSUPP;
3614#endif
3615}
3616
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003617static int __io_splice_prep(struct io_kiocb *req,
3618 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003619{
Pavel Begunkovfe7e3252021-06-24 15:09:57 +01003620 struct io_splice *sp = &req->splice;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003621 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003622
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003623 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3624 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003625
3626 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003627 sp->len = READ_ONCE(sqe->len);
3628 sp->flags = READ_ONCE(sqe->splice_flags);
3629
3630 if (unlikely(sp->flags & ~valid_flags))
3631 return -EINVAL;
3632
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003633 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3634 (sp->flags & SPLICE_F_FD_IN_FIXED));
3635 if (!sp->file_in)
3636 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003637 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003638 return 0;
3639}
3640
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003641static int io_tee_prep(struct io_kiocb *req,
3642 const struct io_uring_sqe *sqe)
3643{
3644 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3645 return -EINVAL;
3646 return __io_splice_prep(req, sqe);
3647}
3648
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003649static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003650{
3651 struct io_splice *sp = &req->splice;
3652 struct file *in = sp->file_in;
3653 struct file *out = sp->file_out;
3654 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3655 long ret = 0;
3656
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003657 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003658 return -EAGAIN;
3659 if (sp->len)
3660 ret = do_tee(in, out, sp->len, flags);
3661
Pavel Begunkove1d767f2021-03-19 17:22:43 +00003662 if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
3663 io_put_file(in);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003664 req->flags &= ~REQ_F_NEED_CLEANUP;
3665
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003666 if (ret != sp->len)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003667 req_set_fail(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003668 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003669 return 0;
3670}
3671
3672static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3673{
Pavel Begunkovfe7e3252021-06-24 15:09:57 +01003674 struct io_splice *sp = &req->splice;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003675
3676 sp->off_in = READ_ONCE(sqe->splice_off_in);
3677 sp->off_out = READ_ONCE(sqe->off);
3678 return __io_splice_prep(req, sqe);
3679}
3680
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003681static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003682{
3683 struct io_splice *sp = &req->splice;
3684 struct file *in = sp->file_in;
3685 struct file *out = sp->file_out;
3686 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3687 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003688 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003689
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003690 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003691 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003692
3693 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3694 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003695
Jens Axboe948a7742020-05-17 14:21:38 -06003696 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003697 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003698
Pavel Begunkove1d767f2021-03-19 17:22:43 +00003699 if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
3700 io_put_file(in);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003701 req->flags &= ~REQ_F_NEED_CLEANUP;
3702
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003703 if (ret != sp->len)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003704 req_set_fail(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003705 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003706 return 0;
3707}
3708
Jens Axboe2b188cc2019-01-07 10:46:33 -07003709/*
3710 * IORING_OP_NOP just posts a completion event, nothing else.
3711 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003712static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003713{
3714 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003715
Jens Axboedef596e2019-01-09 08:59:42 -07003716 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3717 return -EINVAL;
3718
Pavel Begunkov889fca72021-02-10 00:03:09 +00003719 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003720 return 0;
3721}
3722
Pavel Begunkov1155c762021-02-18 18:29:38 +00003723static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003724{
Jens Axboe6b063142019-01-10 22:13:58 -07003725 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003726
Jens Axboe09bb8392019-03-13 12:39:28 -06003727 if (!req->file)
3728 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003729
Jens Axboe6b063142019-01-10 22:13:58 -07003730 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003731 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003732 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003733 return -EINVAL;
3734
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003735 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3736 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3737 return -EINVAL;
3738
3739 req->sync.off = READ_ONCE(sqe->off);
3740 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003741 return 0;
3742}
3743
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003744static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07003745{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003746 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003747 int ret;
3748
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003749 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003750 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003751 return -EAGAIN;
3752
Jens Axboe9adbd452019-12-20 08:45:55 -07003753 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003754 end > 0 ? end : LLONG_MAX,
3755 req->sync.flags & IORING_FSYNC_DATASYNC);
3756 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003757 req_set_fail(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003758 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003759 return 0;
3760}
3761
Jens Axboed63d1b52019-12-10 10:38:56 -07003762static int io_fallocate_prep(struct io_kiocb *req,
3763 const struct io_uring_sqe *sqe)
3764{
3765 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3766 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003767 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3768 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003769
3770 req->sync.off = READ_ONCE(sqe->off);
3771 req->sync.len = READ_ONCE(sqe->addr);
3772 req->sync.mode = READ_ONCE(sqe->len);
3773 return 0;
3774}
3775
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003776static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07003777{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003778 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003779
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003780 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003781 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003782 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003783 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3784 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003785 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003786 req_set_fail(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003787 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003788 return 0;
3789}
3790
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003791static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003792{
Jens Axboef8748882020-01-08 17:47:02 -07003793 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003794 int ret;
3795
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003796 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003797 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003798 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003799 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003800
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003801 /* open.how should be already initialised */
3802 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003803 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003804
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003805 req->open.dfd = READ_ONCE(sqe->fd);
3806 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003807 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003808 if (IS_ERR(req->open.filename)) {
3809 ret = PTR_ERR(req->open.filename);
3810 req->open.filename = NULL;
3811 return ret;
3812 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003813 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003814 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003815 return 0;
3816}
3817
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003818static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3819{
3820 u64 flags, mode;
3821
Jens Axboe14587a462020-09-05 11:36:08 -06003822 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003823 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003824 mode = READ_ONCE(sqe->len);
3825 flags = READ_ONCE(sqe->open_flags);
3826 req->open.how = build_open_how(flags, mode);
3827 return __io_openat_prep(req, sqe);
3828}
3829
Jens Axboecebdb982020-01-08 17:59:24 -07003830static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3831{
3832 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003833 size_t len;
3834 int ret;
3835
Jens Axboe14587a462020-09-05 11:36:08 -06003836 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003837 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003838 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3839 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003840 if (len < OPEN_HOW_SIZE_VER0)
3841 return -EINVAL;
3842
3843 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3844 len);
3845 if (ret)
3846 return ret;
3847
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003848 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003849}
3850
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003851static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003852{
3853 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003854 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003855 bool nonblock_set;
3856 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003857 int ret;
3858
Jens Axboecebdb982020-01-08 17:59:24 -07003859 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003860 if (ret)
3861 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003862 nonblock_set = op.open_flag & O_NONBLOCK;
3863 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003864 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003865 /*
3866 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
3867 * it'll always -EAGAIN
3868 */
3869 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
3870 return -EAGAIN;
3871 op.lookup_flags |= LOOKUP_CACHED;
3872 op.open_flag |= O_NONBLOCK;
3873 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07003874
Jens Axboe4022e7a2020-03-19 19:23:18 -06003875 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003876 if (ret < 0)
3877 goto err;
3878
3879 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Pavel Begunkov12dcb58a2021-06-24 15:10:00 +01003880 if (IS_ERR(file)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003881 /*
Pavel Begunkov12dcb58a2021-06-24 15:10:00 +01003882 * We could hang on to this 'fd' on retrying, but seems like
3883 * marginal gain for something that is now known to be a slower
3884 * path. So just put it, and we'll get a new one when we retry.
Jens Axboe3a81fd02020-12-10 12:25:36 -07003885 */
3886 put_unused_fd(ret);
Pavel Begunkov12dcb58a2021-06-24 15:10:00 +01003887
3888 ret = PTR_ERR(file);
3889 /* only retry if RESOLVE_CACHED wasn't already set by application */
3890 if (ret == -EAGAIN &&
3891 (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)))
3892 return -EAGAIN;
3893 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003894 }
3895
Pavel Begunkov12dcb58a2021-06-24 15:10:00 +01003896 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
3897 file->f_flags &= ~O_NONBLOCK;
3898 fsnotify_open(file);
3899 fd_install(ret, file);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003900err:
3901 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003902 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003903 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003904 req_set_fail(req);
Pavel Begunkov0bdf3392021-04-11 01:46:29 +01003905 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003906 return 0;
3907}
3908
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003909static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07003910{
Pavel Begunkove45cff52021-02-28 22:35:14 +00003911 return io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07003912}
3913
Jens Axboe067524e2020-03-02 16:32:28 -07003914static int io_remove_buffers_prep(struct io_kiocb *req,
3915 const struct io_uring_sqe *sqe)
3916{
3917 struct io_provide_buf *p = &req->pbuf;
3918 u64 tmp;
3919
3920 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3921 return -EINVAL;
3922
3923 tmp = READ_ONCE(sqe->fd);
3924 if (!tmp || tmp > USHRT_MAX)
3925 return -EINVAL;
3926
3927 memset(p, 0, sizeof(*p));
3928 p->nbufs = tmp;
3929 p->bgid = READ_ONCE(sqe->buf_group);
3930 return 0;
3931}
3932
3933static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3934 int bgid, unsigned nbufs)
3935{
3936 unsigned i = 0;
3937
3938 /* shouldn't happen */
3939 if (!nbufs)
3940 return 0;
3941
3942 /* the head kbuf is the list itself */
3943 while (!list_empty(&buf->list)) {
3944 struct io_buffer *nxt;
3945
3946 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3947 list_del(&nxt->list);
3948 kfree(nxt);
3949 if (++i == nbufs)
3950 return i;
3951 }
3952 i++;
3953 kfree(buf);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003954 xa_erase(&ctx->io_buffers, bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003955
3956 return i;
3957}
3958
Pavel Begunkov889fca72021-02-10 00:03:09 +00003959static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07003960{
3961 struct io_provide_buf *p = &req->pbuf;
3962 struct io_ring_ctx *ctx = req->ctx;
3963 struct io_buffer *head;
3964 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003965 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07003966
3967 io_ring_submit_lock(ctx, !force_nonblock);
3968
3969 lockdep_assert_held(&ctx->uring_lock);
3970
3971 ret = -ENOENT;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003972 head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003973 if (head)
3974 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07003975 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01003976 req_set_fail(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003977
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00003978 /* complete before unlock, IOPOLL may need the lock */
3979 __io_req_complete(req, issue_flags, ret, 0);
3980 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboe067524e2020-03-02 16:32:28 -07003981 return 0;
3982}
3983
Jens Axboeddf0322d2020-02-23 16:41:33 -07003984static int io_provide_buffers_prep(struct io_kiocb *req,
3985 const struct io_uring_sqe *sqe)
3986{
Pavel Begunkov38134ad2021-04-15 13:07:39 +01003987 unsigned long size, tmp_check;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003988 struct io_provide_buf *p = &req->pbuf;
3989 u64 tmp;
3990
3991 if (sqe->ioprio || sqe->rw_flags)
3992 return -EINVAL;
3993
3994 tmp = READ_ONCE(sqe->fd);
3995 if (!tmp || tmp > USHRT_MAX)
3996 return -E2BIG;
3997 p->nbufs = tmp;
3998 p->addr = READ_ONCE(sqe->addr);
3999 p->len = READ_ONCE(sqe->len);
4000
Pavel Begunkov38134ad2021-04-15 13:07:39 +01004001 if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs,
4002 &size))
4003 return -EOVERFLOW;
4004 if (check_add_overflow((unsigned long)p->addr, size, &tmp_check))
4005 return -EOVERFLOW;
4006
Pavel Begunkovd81269f2021-03-19 10:21:19 +00004007 size = (unsigned long)p->len * p->nbufs;
4008 if (!access_ok(u64_to_user_ptr(p->addr), size))
Jens Axboeddf0322d2020-02-23 16:41:33 -07004009 return -EFAULT;
4010
4011 p->bgid = READ_ONCE(sqe->buf_group);
4012 tmp = READ_ONCE(sqe->off);
4013 if (tmp > USHRT_MAX)
4014 return -E2BIG;
4015 p->bid = tmp;
4016 return 0;
4017}
4018
4019static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
4020{
4021 struct io_buffer *buf;
4022 u64 addr = pbuf->addr;
4023 int i, bid = pbuf->bid;
4024
4025 for (i = 0; i < pbuf->nbufs; i++) {
4026 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
4027 if (!buf)
4028 break;
4029
4030 buf->addr = addr;
Thadeu Lima de Souza Cascardod1f82802021-05-05 09:47:06 -03004031 buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004032 buf->bid = bid;
4033 addr += pbuf->len;
4034 bid++;
4035 if (!*head) {
4036 INIT_LIST_HEAD(&buf->list);
4037 *head = buf;
4038 } else {
4039 list_add_tail(&buf->list, &(*head)->list);
4040 }
4041 }
4042
4043 return i ? i : -ENOMEM;
4044}
4045
Pavel Begunkov889fca72021-02-10 00:03:09 +00004046static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07004047{
4048 struct io_provide_buf *p = &req->pbuf;
4049 struct io_ring_ctx *ctx = req->ctx;
4050 struct io_buffer *head, *list;
4051 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004052 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07004053
4054 io_ring_submit_lock(ctx, !force_nonblock);
4055
4056 lockdep_assert_held(&ctx->uring_lock);
4057
Jens Axboe9e15c3a2021-03-13 12:29:43 -07004058 list = head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004059
4060 ret = io_add_buffers(p, &head);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07004061 if (ret >= 0 && !list) {
4062 ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL);
4063 if (ret < 0)
Jens Axboe067524e2020-03-02 16:32:28 -07004064 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004065 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07004066 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004067 req_set_fail(req);
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00004068 /* complete before unlock, IOPOLL may need the lock */
4069 __io_req_complete(req, issue_flags, ret, 0);
4070 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004071 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004072}
4073
Jens Axboe3e4827b2020-01-08 15:18:09 -07004074static int io_epoll_ctl_prep(struct io_kiocb *req,
4075 const struct io_uring_sqe *sqe)
4076{
4077#if defined(CONFIG_EPOLL)
4078 if (sqe->ioprio || sqe->buf_index)
4079 return -EINVAL;
Pavel Begunkov2d74d042021-05-14 12:05:46 +01004080 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004081 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004082
4083 req->epoll.epfd = READ_ONCE(sqe->fd);
4084 req->epoll.op = READ_ONCE(sqe->len);
4085 req->epoll.fd = READ_ONCE(sqe->off);
4086
4087 if (ep_op_has_event(req->epoll.op)) {
4088 struct epoll_event __user *ev;
4089
4090 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4091 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4092 return -EFAULT;
4093 }
4094
4095 return 0;
4096#else
4097 return -EOPNOTSUPP;
4098#endif
4099}
4100
Pavel Begunkov889fca72021-02-10 00:03:09 +00004101static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004102{
4103#if defined(CONFIG_EPOLL)
4104 struct io_epoll *ie = &req->epoll;
4105 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004106 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004107
4108 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4109 if (force_nonblock && ret == -EAGAIN)
4110 return -EAGAIN;
4111
4112 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004113 req_set_fail(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004114 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004115 return 0;
4116#else
4117 return -EOPNOTSUPP;
4118#endif
4119}
4120
Jens Axboec1ca7572019-12-25 22:18:28 -07004121static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4122{
4123#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4124 if (sqe->ioprio || sqe->buf_index || sqe->off)
4125 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004126 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4127 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004128
4129 req->madvise.addr = READ_ONCE(sqe->addr);
4130 req->madvise.len = READ_ONCE(sqe->len);
4131 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4132 return 0;
4133#else
4134 return -EOPNOTSUPP;
4135#endif
4136}
4137
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004138static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004139{
4140#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4141 struct io_madvise *ma = &req->madvise;
4142 int ret;
4143
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004144 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004145 return -EAGAIN;
4146
Minchan Kim0726b012020-10-17 16:14:50 -07004147 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004148 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004149 req_set_fail(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004150 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004151 return 0;
4152#else
4153 return -EOPNOTSUPP;
4154#endif
4155}
4156
Jens Axboe4840e412019-12-25 22:03:45 -07004157static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4158{
4159 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4160 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004161 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4162 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004163
4164 req->fadvise.offset = READ_ONCE(sqe->off);
4165 req->fadvise.len = READ_ONCE(sqe->len);
4166 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4167 return 0;
4168}
4169
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004170static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004171{
4172 struct io_fadvise *fa = &req->fadvise;
4173 int ret;
4174
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004175 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004176 switch (fa->advice) {
4177 case POSIX_FADV_NORMAL:
4178 case POSIX_FADV_RANDOM:
4179 case POSIX_FADV_SEQUENTIAL:
4180 break;
4181 default:
4182 return -EAGAIN;
4183 }
4184 }
Jens Axboe4840e412019-12-25 22:03:45 -07004185
4186 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4187 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004188 req_set_fail(req);
Pavel Begunkov0bdf3392021-04-11 01:46:29 +01004189 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe4840e412019-12-25 22:03:45 -07004190 return 0;
4191}
4192
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004193static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4194{
Pavel Begunkov2d74d042021-05-14 12:05:46 +01004195 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004196 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004197 if (sqe->ioprio || sqe->buf_index)
4198 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004199 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004200 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004201
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004202 req->statx.dfd = READ_ONCE(sqe->fd);
4203 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004204 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004205 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4206 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004207
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004208 return 0;
4209}
4210
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004211static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004212{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004213 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004214 int ret;
4215
Pavel Begunkov59d70012021-03-22 01:58:30 +00004216 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004217 return -EAGAIN;
4218
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004219 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4220 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004221
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004222 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004223 req_set_fail(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004224 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004225 return 0;
4226}
4227
Jens Axboeb5dba592019-12-11 14:02:38 -07004228static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4229{
Jens Axboe14587a462020-09-05 11:36:08 -06004230 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004231 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004232 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4233 sqe->rw_flags || sqe->buf_index)
4234 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004235 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004236 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004237
4238 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004239 return 0;
4240}
4241
Pavel Begunkov889fca72021-02-10 00:03:09 +00004242static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004243{
Jens Axboe9eac1902021-01-19 15:50:37 -07004244 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004245 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004246 struct fdtable *fdt;
Pavel Begunkova1fde922021-04-11 01:46:28 +01004247 struct file *file = NULL;
4248 int ret = -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004249
Jens Axboe9eac1902021-01-19 15:50:37 -07004250 spin_lock(&files->file_lock);
4251 fdt = files_fdtable(files);
4252 if (close->fd >= fdt->max_fds) {
4253 spin_unlock(&files->file_lock);
4254 goto err;
4255 }
4256 file = fdt->fd[close->fd];
Pavel Begunkova1fde922021-04-11 01:46:28 +01004257 if (!file || file->f_op == &io_uring_fops) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004258 spin_unlock(&files->file_lock);
4259 file = NULL;
4260 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004261 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004262
4263 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004264 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004265 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004266 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004267 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004268
Jens Axboe9eac1902021-01-19 15:50:37 -07004269 ret = __close_fd_get_file(close->fd, &file);
4270 spin_unlock(&files->file_lock);
4271 if (ret < 0) {
4272 if (ret == -ENOENT)
4273 ret = -EBADF;
4274 goto err;
4275 }
4276
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004277 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004278 ret = filp_close(file, current->files);
4279err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004280 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004281 req_set_fail(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004282 if (file)
4283 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004284 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004285 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004286}
4287
Pavel Begunkov1155c762021-02-18 18:29:38 +00004288static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004289{
4290 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004291
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004292 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4293 return -EINVAL;
4294 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4295 return -EINVAL;
4296
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004297 req->sync.off = READ_ONCE(sqe->off);
4298 req->sync.len = READ_ONCE(sqe->len);
4299 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004300 return 0;
4301}
4302
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004303static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004304{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004305 int ret;
4306
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004307 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004308 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004309 return -EAGAIN;
4310
Jens Axboe9adbd452019-12-20 08:45:55 -07004311 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004312 req->sync.flags);
4313 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004314 req_set_fail(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004315 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004316 return 0;
4317}
4318
YueHaibing469956e2020-03-04 15:53:52 +08004319#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004320static int io_setup_async_msg(struct io_kiocb *req,
4321 struct io_async_msghdr *kmsg)
4322{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004323 struct io_async_msghdr *async_msg = req->async_data;
4324
4325 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004326 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004327 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004328 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004329 return -ENOMEM;
4330 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004331 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004332 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004333 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004334 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004335 /* if were using fast_iov, set it to the new one */
4336 if (!async_msg->free_iov)
4337 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4338
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004339 return -EAGAIN;
4340}
4341
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004342static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4343 struct io_async_msghdr *iomsg)
4344{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004345 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004346 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004347 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004348 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004349}
4350
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004351static int io_sendmsg_prep_async(struct io_kiocb *req)
4352{
4353 int ret;
4354
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004355 ret = io_sendmsg_copy_hdr(req, req->async_data);
4356 if (!ret)
4357 req->flags |= REQ_F_NEED_CLEANUP;
4358 return ret;
4359}
4360
Jens Axboe3529d8c2019-12-19 18:24:38 -07004361static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004362{
Jens Axboee47293f2019-12-20 08:58:21 -07004363 struct io_sr_msg *sr = &req->sr_msg;
Jens Axboe03b12302019-12-02 18:50:25 -07004364
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004365 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4366 return -EINVAL;
4367
Pavel Begunkov270a5942020-07-12 20:41:04 +03004368 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004369 sr->len = READ_ONCE(sqe->len);
Pavel Begunkov04411802021-04-01 15:44:00 +01004370 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
4371 if (sr->msg_flags & MSG_DONTWAIT)
4372 req->flags |= REQ_F_NOWAIT;
Jens Axboe3529d8c2019-12-19 18:24:38 -07004373
Jens Axboed8768362020-02-27 14:17:49 -07004374#ifdef CONFIG_COMPAT
4375 if (req->ctx->compat)
4376 sr->msg_flags |= MSG_CMSG_COMPAT;
4377#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004378 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004379}
4380
Pavel Begunkov889fca72021-02-10 00:03:09 +00004381static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004382{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004383 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004384 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004385 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004386 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004387 int ret;
4388
Florent Revestdba4a922020-12-04 12:36:04 +01004389 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004390 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004391 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004392
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004393 kmsg = req->async_data;
4394 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004395 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004396 if (ret)
4397 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004398 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004399 }
4400
Pavel Begunkov04411802021-04-01 15:44:00 +01004401 flags = req->sr_msg.msg_flags;
4402 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004403 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004404 if (flags & MSG_WAITALL)
4405 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4406
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004407 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004408 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004409 return io_setup_async_msg(req, kmsg);
4410 if (ret == -ERESTARTSYS)
4411 ret = -EINTR;
4412
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004413 /* fast path, check for non-NULL to avoid function call */
4414 if (kmsg->free_iov)
4415 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004416 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004417 if (ret < min_ret)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004418 req_set_fail(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004419 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004420 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004421}
4422
Pavel Begunkov889fca72021-02-10 00:03:09 +00004423static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004424{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004425 struct io_sr_msg *sr = &req->sr_msg;
4426 struct msghdr msg;
4427 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004428 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004429 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004430 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004431 int ret;
4432
Florent Revestdba4a922020-12-04 12:36:04 +01004433 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004434 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004435 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004436
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004437 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4438 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004439 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004440
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004441 msg.msg_name = NULL;
4442 msg.msg_control = NULL;
4443 msg.msg_controllen = 0;
4444 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004445
Pavel Begunkov04411802021-04-01 15:44:00 +01004446 flags = req->sr_msg.msg_flags;
4447 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004448 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004449 if (flags & MSG_WAITALL)
4450 min_ret = iov_iter_count(&msg.msg_iter);
4451
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004452 msg.msg_flags = flags;
4453 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004454 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004455 return -EAGAIN;
4456 if (ret == -ERESTARTSYS)
4457 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004458
Stefan Metzmacher00312752021-03-20 20:33:36 +01004459 if (ret < min_ret)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004460 req_set_fail(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004461 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004462 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004463}
4464
Pavel Begunkov1400e692020-07-12 20:41:05 +03004465static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4466 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004467{
4468 struct io_sr_msg *sr = &req->sr_msg;
4469 struct iovec __user *uiov;
4470 size_t iov_len;
4471 int ret;
4472
Pavel Begunkov1400e692020-07-12 20:41:05 +03004473 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4474 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004475 if (ret)
4476 return ret;
4477
4478 if (req->flags & REQ_F_BUFFER_SELECT) {
4479 if (iov_len > 1)
4480 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004481 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004482 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004483 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004484 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004485 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004486 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004487 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004488 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004489 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004490 if (ret > 0)
4491 ret = 0;
4492 }
4493
4494 return ret;
4495}
4496
4497#ifdef CONFIG_COMPAT
4498static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004499 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004500{
Jens Axboe52de1fe2020-02-27 10:15:42 -07004501 struct io_sr_msg *sr = &req->sr_msg;
4502 struct compat_iovec __user *uiov;
4503 compat_uptr_t ptr;
4504 compat_size_t len;
4505 int ret;
4506
Pavel Begunkov4af34172021-04-11 01:46:30 +01004507 ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr,
4508 &ptr, &len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004509 if (ret)
4510 return ret;
4511
4512 uiov = compat_ptr(ptr);
4513 if (req->flags & REQ_F_BUFFER_SELECT) {
4514 compat_ssize_t clen;
4515
4516 if (len > 1)
4517 return -EINVAL;
4518 if (!access_ok(uiov, sizeof(*uiov)))
4519 return -EFAULT;
4520 if (__get_user(clen, &uiov->iov_len))
4521 return -EFAULT;
4522 if (clen < 0)
4523 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004524 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004525 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004526 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004527 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004528 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004529 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004530 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004531 if (ret < 0)
4532 return ret;
4533 }
4534
4535 return 0;
4536}
Jens Axboe03b12302019-12-02 18:50:25 -07004537#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004538
Pavel Begunkov1400e692020-07-12 20:41:05 +03004539static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4540 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004541{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004542 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004543
4544#ifdef CONFIG_COMPAT
4545 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004546 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004547#endif
4548
Pavel Begunkov1400e692020-07-12 20:41:05 +03004549 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004550}
4551
Jens Axboebcda7ba2020-02-23 16:42:51 -07004552static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004553 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004554{
4555 struct io_sr_msg *sr = &req->sr_msg;
4556 struct io_buffer *kbuf;
4557
Jens Axboebcda7ba2020-02-23 16:42:51 -07004558 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4559 if (IS_ERR(kbuf))
4560 return kbuf;
4561
4562 sr->kbuf = kbuf;
4563 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004564 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004565}
4566
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004567static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4568{
4569 return io_put_kbuf(req, req->sr_msg.kbuf);
4570}
4571
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004572static int io_recvmsg_prep_async(struct io_kiocb *req)
Jens Axboe03b12302019-12-02 18:50:25 -07004573{
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004574 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004575
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004576 ret = io_recvmsg_copy_hdr(req, req->async_data);
4577 if (!ret)
4578 req->flags |= REQ_F_NEED_CLEANUP;
4579 return ret;
4580}
4581
4582static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4583{
4584 struct io_sr_msg *sr = &req->sr_msg;
4585
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004586 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4587 return -EINVAL;
4588
Pavel Begunkov270a5942020-07-12 20:41:04 +03004589 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004590 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004591 sr->bgid = READ_ONCE(sqe->buf_group);
Pavel Begunkov04411802021-04-01 15:44:00 +01004592 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
4593 if (sr->msg_flags & MSG_DONTWAIT)
4594 req->flags |= REQ_F_NOWAIT;
Jens Axboe3529d8c2019-12-19 18:24:38 -07004595
Jens Axboed8768362020-02-27 14:17:49 -07004596#ifdef CONFIG_COMPAT
4597 if (req->ctx->compat)
4598 sr->msg_flags |= MSG_CMSG_COMPAT;
4599#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004600 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004601}
4602
Pavel Begunkov889fca72021-02-10 00:03:09 +00004603static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004604{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004605 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004606 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004607 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004608 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004609 int min_ret = 0;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004610 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004611 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004612
Florent Revestdba4a922020-12-04 12:36:04 +01004613 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004614 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004615 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004616
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004617 kmsg = req->async_data;
4618 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004619 ret = io_recvmsg_copy_hdr(req, &iomsg);
4620 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004621 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004622 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004623 }
4624
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004625 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004626 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004627 if (IS_ERR(kbuf))
4628 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004629 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004630 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4631 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004632 1, req->sr_msg.len);
4633 }
4634
Pavel Begunkov04411802021-04-01 15:44:00 +01004635 flags = req->sr_msg.msg_flags;
4636 if (force_nonblock)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004637 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004638 if (flags & MSG_WAITALL)
4639 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4640
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004641 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4642 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004643 if (force_nonblock && ret == -EAGAIN)
4644 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004645 if (ret == -ERESTARTSYS)
4646 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004647
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004648 if (req->flags & REQ_F_BUFFER_SELECTED)
4649 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004650 /* fast path, check for non-NULL to avoid function call */
4651 if (kmsg->free_iov)
4652 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004653 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004654 if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004655 req_set_fail(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004656 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004657 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004658}
4659
Pavel Begunkov889fca72021-02-10 00:03:09 +00004660static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004661{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004662 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004663 struct io_sr_msg *sr = &req->sr_msg;
4664 struct msghdr msg;
4665 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004666 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004667 struct iovec iov;
4668 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004669 int min_ret = 0;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004670 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004671 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004672
Florent Revestdba4a922020-12-04 12:36:04 +01004673 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004674 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004675 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004676
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004677 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004678 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004679 if (IS_ERR(kbuf))
4680 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004681 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004682 }
4683
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004684 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004685 if (unlikely(ret))
4686 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004687
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004688 msg.msg_name = NULL;
4689 msg.msg_control = NULL;
4690 msg.msg_controllen = 0;
4691 msg.msg_namelen = 0;
4692 msg.msg_iocb = NULL;
4693 msg.msg_flags = 0;
4694
Pavel Begunkov04411802021-04-01 15:44:00 +01004695 flags = req->sr_msg.msg_flags;
4696 if (force_nonblock)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004697 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004698 if (flags & MSG_WAITALL)
4699 min_ret = iov_iter_count(&msg.msg_iter);
4700
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004701 ret = sock_recvmsg(sock, &msg, flags);
4702 if (force_nonblock && ret == -EAGAIN)
4703 return -EAGAIN;
4704 if (ret == -ERESTARTSYS)
4705 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004706out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004707 if (req->flags & REQ_F_BUFFER_SELECTED)
4708 cflags = io_put_recv_kbuf(req);
Stefan Metzmacher00312752021-03-20 20:33:36 +01004709 if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004710 req_set_fail(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004711 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07004712 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004713}
4714
Jens Axboe3529d8c2019-12-19 18:24:38 -07004715static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004716{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004717 struct io_accept *accept = &req->accept;
4718
Jens Axboe14587a462020-09-05 11:36:08 -06004719 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06004720 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004721 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004722 return -EINVAL;
4723
Jens Axboed55e5f52019-12-11 16:12:15 -07004724 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4725 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004726 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004727 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004728 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004729}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004730
Pavel Begunkov889fca72021-02-10 00:03:09 +00004731static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004732{
4733 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004734 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004735 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004736 int ret;
4737
Jiufei Xuee697dee2020-06-10 13:41:59 +08004738 if (req->file->f_flags & O_NONBLOCK)
4739 req->flags |= REQ_F_NOWAIT;
4740
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004741 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004742 accept->addr_len, accept->flags,
4743 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004744 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004745 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004746 if (ret < 0) {
4747 if (ret == -ERESTARTSYS)
4748 ret = -EINTR;
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004749 req_set_fail(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004750 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00004751 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004752 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004753}
4754
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004755static int io_connect_prep_async(struct io_kiocb *req)
4756{
4757 struct io_async_connect *io = req->async_data;
4758 struct io_connect *conn = &req->connect;
4759
4760 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
4761}
4762
Jens Axboe3529d8c2019-12-19 18:24:38 -07004763static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004764{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004765 struct io_connect *conn = &req->connect;
Jens Axboef499a022019-12-02 16:28:46 -07004766
Jens Axboe14587a462020-09-05 11:36:08 -06004767 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004768 return -EINVAL;
4769 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4770 return -EINVAL;
4771
Jens Axboe3529d8c2019-12-19 18:24:38 -07004772 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4773 conn->addr_len = READ_ONCE(sqe->addr2);
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004774 return 0;
Jens Axboef499a022019-12-02 16:28:46 -07004775}
4776
Pavel Begunkov889fca72021-02-10 00:03:09 +00004777static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004778{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004779 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004780 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004781 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004782 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004783
Jens Axboee8c2bc12020-08-15 18:44:09 -07004784 if (req->async_data) {
4785 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004786 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004787 ret = move_addr_to_kernel(req->connect.addr,
4788 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004789 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004790 if (ret)
4791 goto out;
4792 io = &__io;
4793 }
4794
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004795 file_flags = force_nonblock ? O_NONBLOCK : 0;
4796
Jens Axboee8c2bc12020-08-15 18:44:09 -07004797 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004798 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004799 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004800 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004801 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004802 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004803 ret = -ENOMEM;
4804 goto out;
4805 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004806 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004807 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004808 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004809 if (ret == -ERESTARTSYS)
4810 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004811out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004812 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01004813 req_set_fail(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004814 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004815 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004816}
YueHaibing469956e2020-03-04 15:53:52 +08004817#else /* !CONFIG_NET */
Jens Axboe99a10082021-02-19 09:35:19 -07004818#define IO_NETOP_FN(op) \
4819static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \
4820{ \
4821 return -EOPNOTSUPP; \
Jens Axboef8e85cf2019-11-23 14:24:24 -07004822}
4823
Jens Axboe99a10082021-02-19 09:35:19 -07004824#define IO_NETOP_PREP(op) \
4825IO_NETOP_FN(op) \
4826static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \
4827{ \
4828 return -EOPNOTSUPP; \
4829} \
4830
4831#define IO_NETOP_PREP_ASYNC(op) \
4832IO_NETOP_PREP(op) \
4833static int io_##op##_prep_async(struct io_kiocb *req) \
4834{ \
4835 return -EOPNOTSUPP; \
YueHaibing469956e2020-03-04 15:53:52 +08004836}
4837
Jens Axboe99a10082021-02-19 09:35:19 -07004838IO_NETOP_PREP_ASYNC(sendmsg);
4839IO_NETOP_PREP_ASYNC(recvmsg);
4840IO_NETOP_PREP_ASYNC(connect);
4841IO_NETOP_PREP(accept);
4842IO_NETOP_FN(send);
4843IO_NETOP_FN(recv);
YueHaibing469956e2020-03-04 15:53:52 +08004844#endif /* CONFIG_NET */
Jens Axboe17f2fe32019-10-17 14:42:58 -06004845
Jens Axboed7718a92020-02-14 22:23:12 -07004846struct io_poll_table {
4847 struct poll_table_struct pt;
4848 struct io_kiocb *req;
4849 int error;
4850};
4851
Jens Axboed7718a92020-02-14 22:23:12 -07004852static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4853 __poll_t mask, task_work_func_t func)
4854{
Jens Axboeaa96bf82020-04-03 11:26:26 -06004855 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004856
4857 /* for instances that support it check for an event match first: */
4858 if (mask && !(mask & poll->events))
4859 return 0;
4860
4861 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4862
4863 list_del_init(&poll->wait.entry);
4864
Jens Axboed7718a92020-02-14 22:23:12 -07004865 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00004866 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06004867
Jens Axboed7718a92020-02-14 22:23:12 -07004868 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004869 * If this fails, then the task is exiting. When a task exits, the
4870 * work gets canceled, so just cancel this request as well instead
4871 * of executing it. We can't safely execute it anyway, as we may not
4872 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004873 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06004874 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004875 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06004876 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00004877 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004878 }
Jens Axboed7718a92020-02-14 22:23:12 -07004879 return 1;
4880}
4881
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004882static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4883 __acquires(&req->ctx->completion_lock)
4884{
4885 struct io_ring_ctx *ctx = req->ctx;
4886
4887 if (!req->result && !READ_ONCE(poll->canceled)) {
4888 struct poll_table_struct pt = { ._key = poll->events };
4889
4890 req->result = vfs_poll(req->file, &pt) & poll->events;
4891 }
4892
4893 spin_lock_irq(&ctx->completion_lock);
4894 if (!req->result && !READ_ONCE(poll->canceled)) {
4895 add_wait_queue(poll->head, &poll->wait);
4896 return true;
4897 }
4898
4899 return false;
4900}
4901
Jens Axboed4e7cd32020-08-15 11:44:50 -07004902static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004903{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004904 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004905 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004906 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004907 return req->apoll->double_poll;
4908}
4909
4910static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4911{
4912 if (req->opcode == IORING_OP_POLL_ADD)
4913 return &req->poll;
4914 return &req->apoll->poll;
4915}
4916
4917static void io_poll_remove_double(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01004918 __must_hold(&req->ctx->completion_lock)
Jens Axboed4e7cd32020-08-15 11:44:50 -07004919{
4920 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004921
4922 lockdep_assert_held(&req->ctx->completion_lock);
4923
4924 if (poll && poll->head) {
4925 struct wait_queue_head *head = poll->head;
4926
4927 spin_lock(&head->lock);
4928 list_del_init(&poll->wait.entry);
4929 if (poll->wait.private)
Jens Axboede9b4cc2021-02-24 13:28:27 -07004930 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004931 poll->head = NULL;
4932 spin_unlock(&head->lock);
4933 }
4934}
4935
Pavel Begunkove27414b2021-04-09 09:13:20 +01004936static bool io_poll_complete(struct io_kiocb *req, __poll_t mask)
Pavel Begunkove07785b2021-04-01 15:43:57 +01004937 __must_hold(&req->ctx->completion_lock)
Jens Axboe18bceab2020-05-15 11:56:54 -06004938{
4939 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004940 unsigned flags = IORING_CQE_F_MORE;
Pavel Begunkove27414b2021-04-09 09:13:20 +01004941 int error;
Jens Axboe18bceab2020-05-15 11:56:54 -06004942
Pavel Begunkove27414b2021-04-09 09:13:20 +01004943 if (READ_ONCE(req->poll.canceled)) {
Jens Axboe45ab03b2021-02-23 08:19:33 -07004944 error = -ECANCELED;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004945 req->poll.events |= EPOLLONESHOT;
Pavel Begunkove27414b2021-04-09 09:13:20 +01004946 } else {
Jens Axboe50826202021-02-23 09:02:26 -07004947 error = mangle_poll(mask);
Pavel Begunkove27414b2021-04-09 09:13:20 +01004948 }
Jens Axboeb69de282021-03-17 08:37:41 -06004949 if (req->poll.events & EPOLLONESHOT)
4950 flags = 0;
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01004951 if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) {
Jens Axboe50826202021-02-23 09:02:26 -07004952 io_poll_remove_waitqs(req);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004953 req->poll.done = true;
4954 flags = 0;
4955 }
Hao Xu7b289c32021-04-13 15:20:39 +08004956 if (flags & IORING_CQE_F_MORE)
4957 ctx->cq_extra++;
4958
Jens Axboe18bceab2020-05-15 11:56:54 -06004959 io_commit_cqring(ctx);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004960 return !(flags & IORING_CQE_F_MORE);
Jens Axboe18bceab2020-05-15 11:56:54 -06004961}
4962
Jens Axboe18bceab2020-05-15 11:56:54 -06004963static void io_poll_task_func(struct callback_head *cb)
4964{
4965 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004966 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004967 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004968
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004969 if (io_poll_rewait(req, &req->poll)) {
4970 spin_unlock_irq(&ctx->completion_lock);
4971 } else {
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004972 bool done;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004973
Pavel Begunkove27414b2021-04-09 09:13:20 +01004974 done = io_poll_complete(req, req->result);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004975 if (done) {
4976 hash_del(&req->hash_node);
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004977 } else {
Jens Axboe88e41cf2021-02-22 22:08:01 -07004978 req->result = 0;
4979 add_wait_queue(req->poll.head, &req->poll.wait);
4980 }
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004981 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004982 io_cqring_ev_posted(ctx);
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004983
Jens Axboe88e41cf2021-02-22 22:08:01 -07004984 if (done) {
4985 nxt = io_put_req_find_next(req);
4986 if (nxt)
4987 __io_req_task_submit(nxt);
4988 }
Pavel Begunkovea1164e2020-06-30 15:20:41 +03004989 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004990}
4991
4992static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4993 int sync, void *key)
4994{
4995 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004996 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004997 __poll_t mask = key_to_poll(key);
4998
4999 /* for instances that support it check for an event match first: */
5000 if (mask && !(mask & poll->events))
5001 return 0;
Jens Axboe88e41cf2021-02-22 22:08:01 -07005002 if (!(poll->events & EPOLLONESHOT))
5003 return poll->wait.func(&poll->wait, mode, sync, key);
Jens Axboe18bceab2020-05-15 11:56:54 -06005004
Jens Axboe8706e042020-09-28 08:38:54 -06005005 list_del_init(&wait->entry);
5006
Jens Axboe807abcb2020-07-17 17:09:27 -06005007 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005008 bool done;
5009
Jens Axboe807abcb2020-07-17 17:09:27 -06005010 spin_lock(&poll->head->lock);
5011 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06005012 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06005013 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005014 /* make sure double remove sees this as being gone */
5015 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06005016 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06005017 if (!done) {
5018 /* use wait func handler, so it matches the rq type */
5019 poll->wait.func(&poll->wait, mode, sync, key);
5020 }
Jens Axboe18bceab2020-05-15 11:56:54 -06005021 }
Jens Axboede9b4cc2021-02-24 13:28:27 -07005022 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005023 return 1;
5024}
5025
5026static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
5027 wait_queue_func_t wake_func)
5028{
5029 poll->head = NULL;
5030 poll->done = false;
5031 poll->canceled = false;
Jens Axboe464dca62021-03-19 14:06:24 -06005032#define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP)
5033 /* mask in events that we always want/need */
5034 poll->events = events | IO_POLL_UNMASK;
Jens Axboe18bceab2020-05-15 11:56:54 -06005035 INIT_LIST_HEAD(&poll->wait.entry);
5036 init_waitqueue_func_entry(&poll->wait, wake_func);
5037}
5038
5039static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06005040 struct wait_queue_head *head,
5041 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06005042{
5043 struct io_kiocb *req = pt->req;
5044
5045 /*
5046 * If poll->head is already set, it's because the file being polled
5047 * uses multiple waitqueues for poll handling (eg one for read, one
5048 * for write). Setup a separate io_poll_iocb if this happens.
5049 */
5050 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01005051 struct io_poll_iocb *poll_one = poll;
5052
Jens Axboe18bceab2020-05-15 11:56:54 -06005053 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06005054 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005055 pt->error = -EINVAL;
5056 return;
5057 }
Jens Axboeea6a693d2021-04-15 09:47:13 -06005058 /*
5059 * Can't handle multishot for double wait for now, turn it
5060 * into one-shot mode.
5061 */
Pavel Begunkov7a274722021-05-17 12:43:34 +01005062 if (!(poll_one->events & EPOLLONESHOT))
5063 poll_one->events |= EPOLLONESHOT;
Jens Axboe1c3b3e62021-02-28 16:07:30 -07005064 /* double add on the same waitqueue head, ignore */
Pavel Begunkov7a274722021-05-17 12:43:34 +01005065 if (poll_one->head == head)
Jens Axboe1c3b3e62021-02-28 16:07:30 -07005066 return;
Jens Axboe18bceab2020-05-15 11:56:54 -06005067 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
5068 if (!poll) {
5069 pt->error = -ENOMEM;
5070 return;
5071 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01005072 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboede9b4cc2021-02-24 13:28:27 -07005073 req_ref_get(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005074 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06005075 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005076 }
5077
5078 pt->error = 0;
5079 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005080
5081 if (poll->events & EPOLLEXCLUSIVE)
5082 add_wait_queue_exclusive(head, &poll->wait);
5083 else
5084 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005085}
5086
5087static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5088 struct poll_table_struct *p)
5089{
5090 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005091 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005092
Jens Axboe807abcb2020-07-17 17:09:27 -06005093 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005094}
5095
Jens Axboed7718a92020-02-14 22:23:12 -07005096static void io_async_task_func(struct callback_head *cb)
5097{
5098 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5099 struct async_poll *apoll = req->apoll;
5100 struct io_ring_ctx *ctx = req->ctx;
5101
Olivier Langlois236daeae2021-05-31 02:36:37 -04005102 trace_io_uring_task_run(req->ctx, req, req->opcode, req->user_data);
Jens Axboed7718a92020-02-14 22:23:12 -07005103
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005104 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005105 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005106 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005107 }
5108
Pavel Begunkov0ea13b42021-04-09 09:13:21 +01005109 hash_del(&req->hash_node);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005110 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005111 spin_unlock_irq(&ctx->completion_lock);
5112
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005113 if (!READ_ONCE(apoll->poll.canceled))
5114 __io_req_task_submit(req);
5115 else
Pavel Begunkov25935532021-03-19 17:22:40 +00005116 io_req_complete_failed(req, -ECANCELED);
Jens Axboed7718a92020-02-14 22:23:12 -07005117}
5118
5119static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5120 void *key)
5121{
5122 struct io_kiocb *req = wait->private;
5123 struct io_poll_iocb *poll = &req->apoll->poll;
5124
5125 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5126 key_to_poll(key));
5127
5128 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5129}
5130
5131static void io_poll_req_insert(struct io_kiocb *req)
5132{
5133 struct io_ring_ctx *ctx = req->ctx;
5134 struct hlist_head *list;
5135
5136 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5137 hlist_add_head(&req->hash_node, list);
5138}
5139
5140static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5141 struct io_poll_iocb *poll,
5142 struct io_poll_table *ipt, __poll_t mask,
5143 wait_queue_func_t wake_func)
5144 __acquires(&ctx->completion_lock)
5145{
5146 struct io_ring_ctx *ctx = req->ctx;
5147 bool cancel = false;
5148
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005149 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005150 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005151 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005152 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005153
5154 ipt->pt._key = mask;
5155 ipt->req = req;
5156 ipt->error = -EINVAL;
5157
Jens Axboed7718a92020-02-14 22:23:12 -07005158 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5159
5160 spin_lock_irq(&ctx->completion_lock);
5161 if (likely(poll->head)) {
5162 spin_lock(&poll->head->lock);
5163 if (unlikely(list_empty(&poll->wait.entry))) {
5164 if (ipt->error)
5165 cancel = true;
5166 ipt->error = 0;
5167 mask = 0;
5168 }
Jens Axboe88e41cf2021-02-22 22:08:01 -07005169 if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error)
Jens Axboed7718a92020-02-14 22:23:12 -07005170 list_del_init(&poll->wait.entry);
5171 else if (cancel)
5172 WRITE_ONCE(poll->canceled, true);
5173 else if (!poll->done) /* actually waiting for an event */
5174 io_poll_req_insert(req);
5175 spin_unlock(&poll->head->lock);
5176 }
5177
5178 return mask;
5179}
5180
Olivier Langlois59b735a2021-06-22 05:17:39 -07005181enum {
5182 IO_APOLL_OK,
5183 IO_APOLL_ABORTED,
5184 IO_APOLL_READY
5185};
5186
5187static int io_arm_poll_handler(struct io_kiocb *req)
Jens Axboed7718a92020-02-14 22:23:12 -07005188{
5189 const struct io_op_def *def = &io_op_defs[req->opcode];
5190 struct io_ring_ctx *ctx = req->ctx;
5191 struct async_poll *apoll;
5192 struct io_poll_table ipt;
Pavel Begunkovb2d9c3d2021-06-26 21:40:44 +01005193 __poll_t ret, mask = EPOLLONESHOT | POLLERR | POLLPRI;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005194 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005195
5196 if (!req->file || !file_can_poll(req->file))
Olivier Langlois59b735a2021-06-22 05:17:39 -07005197 return IO_APOLL_ABORTED;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005198 if (req->flags & REQ_F_POLLED)
Olivier Langlois59b735a2021-06-22 05:17:39 -07005199 return IO_APOLL_ABORTED;
Pavel Begunkovb2d9c3d2021-06-26 21:40:44 +01005200 if (!def->pollin && !def->pollout)
Olivier Langlois59b735a2021-06-22 05:17:39 -07005201 return IO_APOLL_ABORTED;
Pavel Begunkovb2d9c3d2021-06-26 21:40:44 +01005202
5203 if (def->pollin) {
5204 rw = READ;
5205 mask |= POLLIN | POLLRDNORM;
5206
5207 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5208 if ((req->opcode == IORING_OP_RECVMSG) &&
5209 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5210 mask &= ~POLLIN;
5211 } else {
5212 rw = WRITE;
5213 mask |= POLLOUT | POLLWRNORM;
5214 }
5215
Jens Axboe9dab14b2020-08-25 12:27:50 -06005216 /* if we can't nonblock try, then no point in arming a poll handler */
Jens Axboe7b29f922021-03-12 08:30:14 -07005217 if (!io_file_supports_async(req, rw))
Olivier Langlois59b735a2021-06-22 05:17:39 -07005218 return IO_APOLL_ABORTED;
Jens Axboed7718a92020-02-14 22:23:12 -07005219
5220 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5221 if (unlikely(!apoll))
Olivier Langlois59b735a2021-06-22 05:17:39 -07005222 return IO_APOLL_ABORTED;
Jens Axboe807abcb2020-07-17 17:09:27 -06005223 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005224 req->apoll = apoll;
Pavel Begunkovb2d9c3d2021-06-26 21:40:44 +01005225 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005226 ipt.pt._qproc = io_async_queue_proc;
5227
5228 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5229 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005230 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005231 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005232 spin_unlock_irq(&ctx->completion_lock);
Olivier Langlois59b735a2021-06-22 05:17:39 -07005233 if (ret)
5234 return IO_APOLL_READY;
5235 return IO_APOLL_ABORTED;
Jens Axboed7718a92020-02-14 22:23:12 -07005236 }
5237 spin_unlock_irq(&ctx->completion_lock);
Olivier Langlois236daeae2021-05-31 02:36:37 -04005238 trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data,
5239 mask, apoll->poll.events);
Olivier Langlois59b735a2021-06-22 05:17:39 -07005240 return IO_APOLL_OK;
Jens Axboed7718a92020-02-14 22:23:12 -07005241}
5242
5243static bool __io_poll_remove_one(struct io_kiocb *req,
Jens Axboeb2e720a2021-03-31 09:03:03 -06005244 struct io_poll_iocb *poll, bool do_cancel)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005245 __must_hold(&req->ctx->completion_lock)
Jens Axboed7718a92020-02-14 22:23:12 -07005246{
Jens Axboeb41e9852020-02-17 09:52:41 -07005247 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005248
Jens Axboe50826202021-02-23 09:02:26 -07005249 if (!poll->head)
5250 return false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005251 spin_lock(&poll->head->lock);
Jens Axboeb2e720a2021-03-31 09:03:03 -06005252 if (do_cancel)
5253 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005254 if (!list_empty(&poll->wait.entry)) {
5255 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005256 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005257 }
5258 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005259 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005260 return do_complete;
5261}
5262
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005263static bool io_poll_remove_waitqs(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005264 __must_hold(&req->ctx->completion_lock)
Jens Axboed7718a92020-02-14 22:23:12 -07005265{
5266 bool do_complete;
5267
Jens Axboed4e7cd32020-08-15 11:44:50 -07005268 io_poll_remove_double(req);
Pavel Begunkove31001a2021-04-13 02:58:43 +01005269 do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005270
Pavel Begunkove31001a2021-04-13 02:58:43 +01005271 if (req->opcode != IORING_OP_POLL_ADD && do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005272 /* non-poll requests have submit ref still */
Pavel Begunkove31001a2021-04-13 02:58:43 +01005273 req_ref_put(req);
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005274 }
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005275 return do_complete;
5276}
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005277
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005278static bool io_poll_remove_one(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005279 __must_hold(&req->ctx->completion_lock)
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005280{
5281 bool do_complete;
5282
5283 do_complete = io_poll_remove_waitqs(req);
Jens Axboeb41e9852020-02-17 09:52:41 -07005284 if (do_complete) {
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01005285 io_cqring_fill_event(req->ctx, req->user_data, -ECANCELED, 0);
Jens Axboeb41e9852020-02-17 09:52:41 -07005286 io_commit_cqring(req->ctx);
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005287 req_set_fail(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005288 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005289 }
5290
5291 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005292}
5293
Jens Axboe76e1b642020-09-26 15:05:03 -06005294/*
5295 * Returns true if we found and killed one or more poll requests
5296 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005297static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01005298 bool cancel_all)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005299{
Jens Axboe78076bb2019-12-04 19:56:40 -07005300 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005301 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005302 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005303
5304 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005305 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5306 struct hlist_head *list;
5307
5308 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005309 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01005310 if (io_match_task(req, tsk, cancel_all))
Jens Axboef3606e32020-09-22 08:18:24 -06005311 posted += io_poll_remove_one(req);
5312 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005313 }
5314 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005315
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005316 if (posted)
5317 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005318
5319 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005320}
5321
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005322static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr,
5323 bool poll_only)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005324 __must_hold(&ctx->completion_lock)
Jens Axboe47f46762019-11-09 17:43:02 -07005325{
Jens Axboe78076bb2019-12-04 19:56:40 -07005326 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005327 struct io_kiocb *req;
5328
Jens Axboe78076bb2019-12-04 19:56:40 -07005329 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5330 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005331 if (sqe_addr != req->user_data)
5332 continue;
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005333 if (poll_only && req->opcode != IORING_OP_POLL_ADD)
5334 continue;
Jens Axboeb2cb8052021-03-17 08:17:19 -06005335 return req;
Jens Axboe47f46762019-11-09 17:43:02 -07005336 }
Jens Axboeb2cb8052021-03-17 08:17:19 -06005337 return NULL;
Jens Axboe47f46762019-11-09 17:43:02 -07005338}
5339
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005340static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr,
5341 bool poll_only)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005342 __must_hold(&ctx->completion_lock)
Jens Axboeb2cb8052021-03-17 08:17:19 -06005343{
5344 struct io_kiocb *req;
5345
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005346 req = io_poll_find(ctx, sqe_addr, poll_only);
Jens Axboeb2cb8052021-03-17 08:17:19 -06005347 if (!req)
5348 return -ENOENT;
5349 if (io_poll_remove_one(req))
5350 return 0;
5351
5352 return -EALREADY;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005353}
5354
Pavel Begunkov9096af32021-04-14 13:38:36 +01005355static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe,
5356 unsigned int flags)
5357{
5358 u32 events;
5359
5360 events = READ_ONCE(sqe->poll32_events);
5361#ifdef __BIG_ENDIAN
5362 events = swahw32(events);
5363#endif
5364 if (!(flags & IORING_POLL_ADD_MULTI))
5365 events |= EPOLLONESHOT;
5366 return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT));
5367}
5368
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005369static int io_poll_update_prep(struct io_kiocb *req,
Jens Axboe3529d8c2019-12-19 18:24:38 -07005370 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005371{
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005372 struct io_poll_update *upd = &req->poll_update;
5373 u32 flags;
5374
Jens Axboe221c5eb2019-01-17 09:41:58 -07005375 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5376 return -EINVAL;
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005377 if (sqe->ioprio || sqe->buf_index)
5378 return -EINVAL;
5379 flags = READ_ONCE(sqe->len);
5380 if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA |
5381 IORING_POLL_ADD_MULTI))
5382 return -EINVAL;
5383 /* meaningless without update */
5384 if (flags == IORING_POLL_ADD_MULTI)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005385 return -EINVAL;
5386
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005387 upd->old_user_data = READ_ONCE(sqe->addr);
5388 upd->update_events = flags & IORING_POLL_UPDATE_EVENTS;
5389 upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA;
Jens Axboe0969e782019-12-17 18:40:57 -07005390
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005391 upd->new_user_data = READ_ONCE(sqe->off);
5392 if (!upd->update_user_data && upd->new_user_data)
5393 return -EINVAL;
5394 if (upd->update_events)
5395 upd->events = io_poll_parse_events(sqe, flags);
5396 else if (sqe->poll32_events)
5397 return -EINVAL;
Jens Axboe0969e782019-12-17 18:40:57 -07005398
Jens Axboe221c5eb2019-01-17 09:41:58 -07005399 return 0;
5400}
5401
Jens Axboe221c5eb2019-01-17 09:41:58 -07005402static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5403 void *key)
5404{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005405 struct io_kiocb *req = wait->private;
5406 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005407
Jens Axboed7718a92020-02-14 22:23:12 -07005408 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005409}
5410
Jens Axboe221c5eb2019-01-17 09:41:58 -07005411static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5412 struct poll_table_struct *p)
5413{
5414 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5415
Jens Axboee8c2bc12020-08-15 18:44:09 -07005416 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c2019-11-14 12:09:58 -07005417}
5418
Jens Axboe3529d8c2019-12-19 18:24:38 -07005419static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005420{
5421 struct io_poll_iocb *poll = &req->poll;
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005422 u32 flags;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005423
5424 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5425 return -EINVAL;
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005426 if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr)
Jens Axboe88e41cf2021-02-22 22:08:01 -07005427 return -EINVAL;
5428 flags = READ_ONCE(sqe->len);
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005429 if (flags & ~IORING_POLL_ADD_MULTI)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005430 return -EINVAL;
5431
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005432 poll->events = io_poll_parse_events(sqe, flags);
Jens Axboe0969e782019-12-17 18:40:57 -07005433 return 0;
5434}
5435
Pavel Begunkov61e98202021-02-10 00:03:08 +00005436static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005437{
5438 struct io_poll_iocb *poll = &req->poll;
5439 struct io_ring_ctx *ctx = req->ctx;
5440 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005441 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005442
Jens Axboed7718a92020-02-14 22:23:12 -07005443 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005444
Jens Axboed7718a92020-02-14 22:23:12 -07005445 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5446 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005447
Jens Axboe8c838782019-03-12 15:48:16 -06005448 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005449 ipt.error = 0;
Pavel Begunkove27414b2021-04-09 09:13:20 +01005450 io_poll_complete(req, mask);
Jens Axboe8c838782019-03-12 15:48:16 -06005451 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005452 spin_unlock_irq(&ctx->completion_lock);
5453
Jens Axboe8c838782019-03-12 15:48:16 -06005454 if (mask) {
5455 io_cqring_ev_posted(ctx);
Jens Axboe88e41cf2021-02-22 22:08:01 -07005456 if (poll->events & EPOLLONESHOT)
5457 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005458 }
Jens Axboe8c838782019-03-12 15:48:16 -06005459 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005460}
5461
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005462static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb69de282021-03-17 08:37:41 -06005463{
5464 struct io_ring_ctx *ctx = req->ctx;
5465 struct io_kiocb *preq;
Jens Axboecb3b200e2021-04-06 09:49:31 -06005466 bool completing;
Jens Axboeb69de282021-03-17 08:37:41 -06005467 int ret;
5468
5469 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005470 preq = io_poll_find(ctx, req->poll_update.old_user_data, true);
Jens Axboeb69de282021-03-17 08:37:41 -06005471 if (!preq) {
5472 ret = -ENOENT;
5473 goto err;
Jens Axboeb69de282021-03-17 08:37:41 -06005474 }
Jens Axboecb3b200e2021-04-06 09:49:31 -06005475
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005476 if (!req->poll_update.update_events && !req->poll_update.update_user_data) {
5477 completing = true;
5478 ret = io_poll_remove_one(preq) ? 0 : -EALREADY;
5479 goto err;
5480 }
5481
Jens Axboecb3b200e2021-04-06 09:49:31 -06005482 /*
5483 * Don't allow racy completion with singleshot, as we cannot safely
5484 * update those. For multishot, if we're racing with completion, just
5485 * let completion re-add it.
5486 */
5487 completing = !__io_poll_remove_one(preq, &preq->poll, false);
5488 if (completing && (preq->poll.events & EPOLLONESHOT)) {
5489 ret = -EALREADY;
5490 goto err;
Jens Axboeb69de282021-03-17 08:37:41 -06005491 }
5492 /* we now have a detached poll request. reissue. */
5493 ret = 0;
5494err:
Jens Axboeb69de282021-03-17 08:37:41 -06005495 if (ret < 0) {
Jens Axboecb3b200e2021-04-06 09:49:31 -06005496 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005497 req_set_fail(req);
Jens Axboeb69de282021-03-17 08:37:41 -06005498 io_req_complete(req, ret);
5499 return 0;
5500 }
5501 /* only mask one event flags, keep behavior flags */
Pavel Begunkov9d805892021-04-13 02:58:40 +01005502 if (req->poll_update.update_events) {
Jens Axboeb69de282021-03-17 08:37:41 -06005503 preq->poll.events &= ~0xffff;
Pavel Begunkov9d805892021-04-13 02:58:40 +01005504 preq->poll.events |= req->poll_update.events & 0xffff;
Jens Axboeb69de282021-03-17 08:37:41 -06005505 preq->poll.events |= IO_POLL_UNMASK;
5506 }
Pavel Begunkov9d805892021-04-13 02:58:40 +01005507 if (req->poll_update.update_user_data)
5508 preq->user_data = req->poll_update.new_user_data;
Jens Axboecb3b200e2021-04-06 09:49:31 -06005509 spin_unlock_irq(&ctx->completion_lock);
5510
Jens Axboeb69de282021-03-17 08:37:41 -06005511 /* complete update request, we're done with it */
5512 io_req_complete(req, ret);
5513
Jens Axboecb3b200e2021-04-06 09:49:31 -06005514 if (!completing) {
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005515 ret = io_poll_add(preq, issue_flags);
Jens Axboecb3b200e2021-04-06 09:49:31 -06005516 if (ret < 0) {
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005517 req_set_fail(preq);
Jens Axboecb3b200e2021-04-06 09:49:31 -06005518 io_req_complete(preq, ret);
5519 }
Jens Axboeb69de282021-03-17 08:37:41 -06005520 }
5521 return 0;
5522}
5523
Jens Axboe5262f562019-09-17 12:26:57 -06005524static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5525{
Jens Axboead8a48a2019-11-15 08:49:11 -07005526 struct io_timeout_data *data = container_of(timer,
5527 struct io_timeout_data, timer);
5528 struct io_kiocb *req = data->req;
5529 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005530 unsigned long flags;
5531
Jens Axboe5262f562019-09-17 12:26:57 -06005532 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005533 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005534 atomic_set(&req->ctx->cq_timeouts,
5535 atomic_read(&req->ctx->cq_timeouts) + 1);
5536
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01005537 io_cqring_fill_event(ctx, req->user_data, -ETIME, 0);
Jens Axboe5262f562019-09-17 12:26:57 -06005538 io_commit_cqring(ctx);
5539 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5540
5541 io_cqring_ev_posted(ctx);
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005542 req_set_fail(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005543 io_put_req(req);
5544 return HRTIMER_NORESTART;
5545}
5546
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005547static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5548 __u64 user_data)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005549 __must_hold(&ctx->completion_lock)
Jens Axboe47f46762019-11-09 17:43:02 -07005550{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005551 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005552 struct io_kiocb *req;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005553 bool found = false;
Jens Axboef254ac02020-08-12 17:33:30 -06005554
5555 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005556 found = user_data == req->user_data;
5557 if (found)
Jens Axboef254ac02020-08-12 17:33:30 -06005558 break;
Jens Axboef254ac02020-08-12 17:33:30 -06005559 }
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005560 if (!found)
5561 return ERR_PTR(-ENOENT);
Jens Axboef254ac02020-08-12 17:33:30 -06005562
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005563 io = req->async_data;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005564 if (hrtimer_try_to_cancel(&io->timer) == -1)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005565 return ERR_PTR(-EALREADY);
5566 list_del_init(&req->timeout.list);
5567 return req;
5568}
5569
5570static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005571 __must_hold(&ctx->completion_lock)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005572{
5573 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5574
5575 if (IS_ERR(req))
5576 return PTR_ERR(req);
5577
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005578 req_set_fail(req);
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01005579 io_cqring_fill_event(ctx, req->user_data, -ECANCELED, 0);
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005580 io_put_req_deferred(req, 1);
5581 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005582}
5583
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005584static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5585 struct timespec64 *ts, enum hrtimer_mode mode)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005586 __must_hold(&ctx->completion_lock)
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005587{
5588 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5589 struct io_timeout_data *data;
5590
5591 if (IS_ERR(req))
5592 return PTR_ERR(req);
5593
5594 req->timeout.off = 0; /* noseq */
5595 data = req->async_data;
5596 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5597 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5598 data->timer.function = io_timeout_fn;
5599 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5600 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005601}
5602
Jens Axboe3529d8c2019-12-19 18:24:38 -07005603static int io_timeout_remove_prep(struct io_kiocb *req,
5604 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005605{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005606 struct io_timeout_rem *tr = &req->timeout_rem;
5607
Jens Axboeb29472e2019-12-17 18:50:29 -07005608 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5609 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005610 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5611 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005612 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005613 return -EINVAL;
5614
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005615 tr->addr = READ_ONCE(sqe->addr);
5616 tr->flags = READ_ONCE(sqe->timeout_flags);
5617 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5618 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5619 return -EINVAL;
5620 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5621 return -EFAULT;
5622 } else if (tr->flags) {
5623 /* timeout removal doesn't support flags */
5624 return -EINVAL;
5625 }
5626
Jens Axboeb29472e2019-12-17 18:50:29 -07005627 return 0;
5628}
5629
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005630static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5631{
5632 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5633 : HRTIMER_MODE_REL;
5634}
5635
Jens Axboe11365042019-10-16 09:08:32 -06005636/*
5637 * Remove or update an existing timeout command
5638 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005639static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005640{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005641 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005642 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005643 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005644
Jens Axboe11365042019-10-16 09:08:32 -06005645 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005646 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005647 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005648 else
5649 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5650 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005651
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01005652 io_cqring_fill_event(ctx, req->user_data, ret, 0);
Jens Axboe11365042019-10-16 09:08:32 -06005653 io_commit_cqring(ctx);
5654 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005655 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005656 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005657 req_set_fail(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005658 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005659 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005660}
5661
Jens Axboe3529d8c2019-12-19 18:24:38 -07005662static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005663 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005664{
Jens Axboead8a48a2019-11-15 08:49:11 -07005665 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005666 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005667 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005668
Jens Axboead8a48a2019-11-15 08:49:11 -07005669 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005670 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005671 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005672 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005673 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005674 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005675 flags = READ_ONCE(sqe->timeout_flags);
5676 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005677 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005678
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005679 req->timeout.off = off;
Pavel Begunkovf18ee4c2021-06-14 23:37:25 +01005680 if (unlikely(off && !req->ctx->off_timeout_used))
5681 req->ctx->off_timeout_used = true;
Jens Axboe26a61672019-12-20 09:02:01 -07005682
Jens Axboee8c2bc12020-08-15 18:44:09 -07005683 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005684 return -ENOMEM;
5685
Jens Axboee8c2bc12020-08-15 18:44:09 -07005686 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005687 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005688
5689 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005690 return -EFAULT;
5691
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005692 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005693 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
Pavel Begunkov2482b582021-03-25 18:32:44 +00005694 if (is_timeout_link)
5695 io_req_track_inflight(req);
Jens Axboead8a48a2019-11-15 08:49:11 -07005696 return 0;
5697}
5698
Pavel Begunkov61e98202021-02-10 00:03:08 +00005699static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005700{
Jens Axboead8a48a2019-11-15 08:49:11 -07005701 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005702 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005703 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005704 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005705
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005706 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005707
Jens Axboe5262f562019-09-17 12:26:57 -06005708 /*
5709 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005710 * timeout event to be satisfied. If it isn't set, then this is
5711 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005712 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005713 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005714 entry = ctx->timeout_list.prev;
5715 goto add;
5716 }
Jens Axboe5262f562019-09-17 12:26:57 -06005717
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005718 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5719 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005720
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005721 /* Update the last seq here in case io_flush_timeouts() hasn't.
5722 * This is safe because ->completion_lock is held, and submissions
5723 * and completions are never mixed in the same ->completion_lock section.
5724 */
5725 ctx->cq_last_tm_flush = tail;
5726
Jens Axboe5262f562019-09-17 12:26:57 -06005727 /*
5728 * Insertion sort, ensuring the first entry in the list is always
5729 * the one we need first.
5730 */
Jens Axboe5262f562019-09-17 12:26:57 -06005731 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005732 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5733 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005734
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005735 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005736 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005737 /* nxt.seq is behind @tail, otherwise would've been completed */
5738 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005739 break;
5740 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005741add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005742 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005743 data->timer.function = io_timeout_fn;
5744 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005745 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005746 return 0;
5747}
5748
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005749struct io_cancel_data {
5750 struct io_ring_ctx *ctx;
5751 u64 user_data;
5752};
5753
Jens Axboe62755e32019-10-28 21:49:21 -06005754static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005755{
Jens Axboe62755e32019-10-28 21:49:21 -06005756 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005757 struct io_cancel_data *cd = data;
Jens Axboede0617e2019-04-06 21:51:27 -06005758
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005759 return req->ctx == cd->ctx && req->user_data == cd->user_data;
Jens Axboe62755e32019-10-28 21:49:21 -06005760}
5761
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005762static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data,
5763 struct io_ring_ctx *ctx)
Jens Axboe62755e32019-10-28 21:49:21 -06005764{
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005765 struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, };
Jens Axboe62755e32019-10-28 21:49:21 -06005766 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005767 int ret = 0;
5768
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005769 if (!tctx || !tctx->io_wq)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005770 return -ENOENT;
5771
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005772 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005773 switch (cancel_ret) {
5774 case IO_WQ_CANCEL_OK:
5775 ret = 0;
5776 break;
5777 case IO_WQ_CANCEL_RUNNING:
5778 ret = -EALREADY;
5779 break;
5780 case IO_WQ_CANCEL_NOTFOUND:
5781 ret = -ENOENT;
5782 break;
5783 }
5784
Jens Axboee977d6d2019-11-05 12:39:45 -07005785 return ret;
5786}
5787
Jens Axboe47f46762019-11-09 17:43:02 -07005788static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5789 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005790 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005791{
5792 unsigned long flags;
5793 int ret;
5794
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005795 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
Jens Axboe47f46762019-11-09 17:43:02 -07005796 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovdf9727a2021-04-01 15:43:59 +01005797 if (ret != -ENOENT)
5798 goto done;
Jens Axboe47f46762019-11-09 17:43:02 -07005799 ret = io_timeout_cancel(ctx, sqe_addr);
5800 if (ret != -ENOENT)
5801 goto done;
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005802 ret = io_poll_cancel(ctx, sqe_addr, false);
Jens Axboe47f46762019-11-09 17:43:02 -07005803done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005804 if (!ret)
5805 ret = success_ret;
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01005806 io_cqring_fill_event(ctx, req->user_data, ret, 0);
Jens Axboe47f46762019-11-09 17:43:02 -07005807 io_commit_cqring(ctx);
5808 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5809 io_cqring_ev_posted(ctx);
5810
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005811 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005812 req_set_fail(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005813}
5814
Jens Axboe3529d8c2019-12-19 18:24:38 -07005815static int io_async_cancel_prep(struct io_kiocb *req,
5816 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005817{
Jens Axboefbf23842019-12-17 18:45:56 -07005818 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005819 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005820 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5821 return -EINVAL;
5822 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005823 return -EINVAL;
5824
Jens Axboefbf23842019-12-17 18:45:56 -07005825 req->cancel.addr = READ_ONCE(sqe->addr);
5826 return 0;
5827}
5828
Pavel Begunkov61e98202021-02-10 00:03:08 +00005829static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07005830{
5831 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov58f99372021-03-12 16:25:55 +00005832 u64 sqe_addr = req->cancel.addr;
5833 struct io_tctx_node *node;
5834 int ret;
Jens Axboefbf23842019-12-17 18:45:56 -07005835
Pavel Begunkov58f99372021-03-12 16:25:55 +00005836 /* tasks should wait for their io-wq threads, so safe w/o sync */
5837 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
5838 spin_lock_irq(&ctx->completion_lock);
5839 if (ret != -ENOENT)
5840 goto done;
5841 ret = io_timeout_cancel(ctx, sqe_addr);
5842 if (ret != -ENOENT)
5843 goto done;
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005844 ret = io_poll_cancel(ctx, sqe_addr, false);
Pavel Begunkov58f99372021-03-12 16:25:55 +00005845 if (ret != -ENOENT)
5846 goto done;
5847 spin_unlock_irq(&ctx->completion_lock);
5848
5849 /* slow path, try all io-wq's */
5850 io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5851 ret = -ENOENT;
5852 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
5853 struct io_uring_task *tctx = node->task->io_uring;
5854
Pavel Begunkov58f99372021-03-12 16:25:55 +00005855 ret = io_async_cancel_one(tctx, req->cancel.addr, ctx);
5856 if (ret != -ENOENT)
5857 break;
5858 }
5859 io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5860
5861 spin_lock_irq(&ctx->completion_lock);
5862done:
Pavel Begunkovd4d19c12021-04-25 14:32:17 +01005863 io_cqring_fill_event(ctx, req->user_data, ret, 0);
Pavel Begunkov58f99372021-03-12 16:25:55 +00005864 io_commit_cqring(ctx);
5865 spin_unlock_irq(&ctx->completion_lock);
5866 io_cqring_ev_posted(ctx);
5867
5868 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005869 req_set_fail(req);
Pavel Begunkov58f99372021-03-12 16:25:55 +00005870 io_put_req(req);
Jens Axboe62755e32019-10-28 21:49:21 -06005871 return 0;
5872}
5873
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005874static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07005875 const struct io_uring_sqe *sqe)
5876{
Daniele Albano61710e42020-07-18 14:15:16 -06005877 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5878 return -EINVAL;
5879 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005880 return -EINVAL;
5881
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005882 req->rsrc_update.offset = READ_ONCE(sqe->off);
5883 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
5884 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005885 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005886 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005887 return 0;
5888}
5889
Pavel Begunkov889fca72021-02-10 00:03:09 +00005890static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005891{
5892 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01005893 struct io_uring_rsrc_update2 up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005894 int ret;
5895
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005896 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005897 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005898
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005899 up.offset = req->rsrc_update.offset;
5900 up.data = req->rsrc_update.arg;
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01005901 up.nr = 0;
5902 up.tags = 0;
Colin Ian King615cee42021-04-26 10:47:35 +01005903 up.resv = 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005904
5905 mutex_lock(&ctx->uring_lock);
Pavel Begunkovfdecb662021-04-25 14:32:20 +01005906 ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE,
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01005907 &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005908 mutex_unlock(&ctx->uring_lock);
5909
5910 if (ret < 0)
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01005911 req_set_fail(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005912 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005913 return 0;
5914}
5915
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005916static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005917{
Jens Axboed625c6e2019-12-17 19:53:05 -07005918 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005919 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005920 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005921 case IORING_OP_READV:
5922 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005923 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005924 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005925 case IORING_OP_WRITEV:
5926 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005927 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005928 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005929 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005930 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005931 case IORING_OP_POLL_REMOVE:
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005932 return io_poll_update_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005933 case IORING_OP_FSYNC:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005934 return io_fsync_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005935 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005936 return io_sfr_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005937 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005938 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005939 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005940 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005941 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005942 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005943 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005944 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005945 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005946 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005947 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005948 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005949 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005950 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005951 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005952 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005953 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005954 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005955 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005956 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005957 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005958 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005959 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005960 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005961 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005962 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005963 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005964 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005965 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005966 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005967 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005968 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005969 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005970 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005971 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005972 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005973 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005974 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005975 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005976 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005977 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005978 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005979 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005980 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06005981 case IORING_OP_SHUTDOWN:
5982 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06005983 case IORING_OP_RENAMEAT:
5984 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06005985 case IORING_OP_UNLINKAT:
5986 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005987 }
5988
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005989 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5990 req->opcode);
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01005991 return -EINVAL;
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005992}
5993
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005994static int io_req_prep_async(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005995{
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005996 if (!io_op_defs[req->opcode].needs_async_setup)
5997 return 0;
5998 if (WARN_ON_ONCE(req->async_data))
5999 return -EFAULT;
6000 if (io_alloc_async_data(req))
6001 return -EAGAIN;
6002
Pavel Begunkov93642ef2021-02-18 18:29:44 +00006003 switch (req->opcode) {
6004 case IORING_OP_READV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00006005 return io_rw_prep_async(req, READ);
6006 case IORING_OP_WRITEV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00006007 return io_rw_prep_async(req, WRITE);
6008 case IORING_OP_SENDMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00006009 return io_sendmsg_prep_async(req);
6010 case IORING_OP_RECVMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00006011 return io_recvmsg_prep_async(req);
6012 case IORING_OP_CONNECT:
6013 return io_connect_prep_async(req);
6014 }
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006015 printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n",
6016 req->opcode);
6017 return -EFAULT;
Jens Axboedef596e2019-01-09 08:59:42 -07006018}
6019
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006020static u32 io_get_sequence(struct io_kiocb *req)
6021{
Pavel Begunkova3dbdf52021-06-17 18:14:05 +01006022 u32 seq = req->ctx->cached_sq_head;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006023
Pavel Begunkova3dbdf52021-06-17 18:14:05 +01006024 /* need original cached_sq_head, but it was increased for each req */
6025 io_for_each_link(req, req)
6026 seq--;
6027 return seq;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006028}
6029
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006030static bool io_drain_req(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07006031{
Pavel Begunkov3c199662021-06-15 16:47:57 +01006032 struct io_kiocb *pos;
Jens Axboedef596e2019-01-09 08:59:42 -07006033 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006034 struct io_defer_entry *de;
Jens Axboedef596e2019-01-09 08:59:42 -07006035 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006036 u32 seq;
Jens Axboedef596e2019-01-09 08:59:42 -07006037
Pavel Begunkov3c199662021-06-15 16:47:57 +01006038 /*
6039 * If we need to drain a request in the middle of a link, drain the
6040 * head request and the next request/link after the current link.
6041 * Considering sequential execution of links, IOSQE_IO_DRAIN will be
6042 * maintained for every request of our link.
6043 */
6044 if (ctx->drain_next) {
6045 req->flags |= REQ_F_IO_DRAIN;
6046 ctx->drain_next = false;
6047 }
6048 /* not interested in head, start from the first linked */
6049 io_for_each_link(pos, req->link) {
6050 if (pos->flags & REQ_F_IO_DRAIN) {
6051 ctx->drain_next = true;
6052 req->flags |= REQ_F_IO_DRAIN;
6053 break;
6054 }
6055 }
6056
Jens Axboedef596e2019-01-09 08:59:42 -07006057 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006058 if (likely(list_empty_careful(&ctx->defer_list) &&
Pavel Begunkov10c66902021-06-15 16:47:56 +01006059 !(req->flags & REQ_F_IO_DRAIN))) {
6060 ctx->drain_active = false;
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006061 return false;
Pavel Begunkov10c66902021-06-15 16:47:56 +01006062 }
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006063
6064 seq = io_get_sequence(req);
6065 /* Still a chance to pass the sequence check */
6066 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006067 return false;
Jens Axboedef596e2019-01-09 08:59:42 -07006068
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006069 ret = io_req_prep_async(req);
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006070 if (ret)
6071 return ret;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03006072 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006073 de = kmalloc(sizeof(*de), GFP_KERNEL);
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006074 if (!de) {
6075 io_req_complete_failed(req, ret);
6076 return true;
6077 }
Jens Axboe31b51512019-01-18 22:56:34 -07006078
6079 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006080 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07006081 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006082 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03006083 io_queue_async_work(req);
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006084 return true;
Jens Axboe31b51512019-01-18 22:56:34 -07006085 }
6086
6087 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006088 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006089 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006090 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07006091 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006092 return true;
Jens Axboe31b51512019-01-18 22:56:34 -07006093}
6094
Pavel Begunkov68fb8972021-03-19 17:22:41 +00006095static void io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006096{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006097 if (req->flags & REQ_F_BUFFER_SELECTED) {
6098 switch (req->opcode) {
6099 case IORING_OP_READV:
6100 case IORING_OP_READ_FIXED:
6101 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07006102 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006103 break;
6104 case IORING_OP_RECVMSG:
6105 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07006106 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006107 break;
6108 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006109 }
6110
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006111 if (req->flags & REQ_F_NEED_CLEANUP) {
6112 switch (req->opcode) {
6113 case IORING_OP_READV:
6114 case IORING_OP_READ_FIXED:
6115 case IORING_OP_READ:
6116 case IORING_OP_WRITEV:
6117 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006118 case IORING_OP_WRITE: {
6119 struct io_async_rw *io = req->async_data;
Pavel Begunkov1dacb4d2021-06-17 18:14:03 +01006120
6121 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006122 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006123 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006124 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006125 case IORING_OP_SENDMSG: {
6126 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00006127
6128 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006129 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006130 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006131 case IORING_OP_SPLICE:
6132 case IORING_OP_TEE:
Pavel Begunkove1d767f2021-03-19 17:22:43 +00006133 if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED))
6134 io_put_file(req->splice.file_in);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006135 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06006136 case IORING_OP_OPENAT:
6137 case IORING_OP_OPENAT2:
6138 if (req->open.filename)
6139 putname(req->open.filename);
6140 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006141 case IORING_OP_RENAMEAT:
6142 putname(req->rename.oldpath);
6143 putname(req->rename.newpath);
6144 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006145 case IORING_OP_UNLINKAT:
6146 putname(req->unlink.filename);
6147 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006148 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006149 }
Jens Axboe75652a302021-04-15 09:52:40 -06006150 if ((req->flags & REQ_F_POLLED) && req->apoll) {
6151 kfree(req->apoll->double_poll);
6152 kfree(req->apoll);
6153 req->apoll = NULL;
6154 }
Pavel Begunkov3a0a6902021-04-20 12:03:31 +01006155 if (req->flags & REQ_F_INFLIGHT) {
6156 struct io_uring_task *tctx = req->task->io_uring;
6157
6158 atomic_dec(&tctx->inflight_tracked);
Pavel Begunkov3a0a6902021-04-20 12:03:31 +01006159 }
Pavel Begunkovc8543572021-06-17 18:14:04 +01006160 if (req->flags & REQ_F_CREDS)
Pavel Begunkovb8e64b52021-06-17 18:14:02 +01006161 put_cred(req->creds);
Pavel Begunkovc8543572021-06-17 18:14:04 +01006162
6163 req->flags &= ~IO_REQ_CLEAN_FLAGS;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006164}
6165
Pavel Begunkov889fca72021-02-10 00:03:09 +00006166static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07006167{
Jens Axboeedafcce2019-01-09 09:16:05 -07006168 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5730b272021-02-27 15:57:30 -07006169 const struct cred *creds = NULL;
Jens Axboed625c6e2019-12-17 19:53:05 -07006170 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07006171
Pavel Begunkovb8e64b52021-06-17 18:14:02 +01006172 if ((req->flags & REQ_F_CREDS) && req->creds != current_cred())
Pavel Begunkovc10d1f92021-06-17 18:14:01 +01006173 creds = override_creds(req->creds);
Jens Axboe5730b272021-02-27 15:57:30 -07006174
Jens Axboed625c6e2019-12-17 19:53:05 -07006175 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07006176 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006177 ret = io_nop(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006178 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006179 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07006180 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006181 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006182 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006183 break;
6184 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07006185 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006186 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006187 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006188 break;
6189 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006190 ret = io_fsync(req, issue_flags);
Jackie Liuba5290c2019-10-09 09:19:59 +08006191 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006192 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006193 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006194 break;
6195 case IORING_OP_POLL_REMOVE:
Pavel Begunkovc5de0032021-04-14 13:38:37 +01006196 ret = io_poll_update(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006197 break;
Jens Axboeb76da702019-11-20 13:05:32 -07006198 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006199 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006200 break;
6201 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006202 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006203 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006204 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006205 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006206 break;
6207 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006208 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006209 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006210 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006211 ret = io_recv(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006212 break;
Jens Axboe561fb042019-10-24 07:25:42 -06006213 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006214 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006215 break;
6216 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006217 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006218 break;
6219 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006220 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006221 break;
6222 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006223 ret = io_connect(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07006224 break;
6225 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006226 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006227 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07006228 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006229 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07006230 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07006231 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006232 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006233 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07006234 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006235 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07006236 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006237 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006238 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006239 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006240 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006241 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006242 break;
Jens Axboe4840e412019-12-25 22:03:45 -07006243 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006244 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006245 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006246 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006247 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006248 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006249 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006250 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006251 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006252 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006253 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006254 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006255 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006256 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006257 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006258 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006259 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006260 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006261 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006262 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006263 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006264 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006265 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006266 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006267 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006268 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006269 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006270 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006271 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006272 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006273 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006274 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006275 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006276 default:
6277 ret = -EINVAL;
6278 break;
6279 }
Jens Axboe31b51512019-01-18 22:56:34 -07006280
Jens Axboe5730b272021-02-27 15:57:30 -07006281 if (creds)
6282 revert_creds(creds);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006283 if (ret)
6284 return ret;
Jens Axboeb5325762020-05-19 21:20:27 -06006285 /* If the op doesn't have a file, we're not polling for it */
Pavel Begunkovcb3d8972021-06-14 02:36:14 +01006286 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file)
6287 io_iopoll_req_issued(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006288
6289 return 0;
6290}
6291
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006292static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006293{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006294 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006295 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006296 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006297
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006298 timeout = io_prep_linked_timeout(req);
6299 if (timeout)
6300 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006301
Jens Axboe4014d942021-01-19 15:53:54 -07006302 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006303 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006304
Jens Axboe561fb042019-10-24 07:25:42 -06006305 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006306 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006307 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006308 /*
6309 * We can get EAGAIN for polled IO even though we're
6310 * forcing a sync submission from here, since we can't
6311 * wait for request slots on the block side.
6312 */
6313 if (ret != -EAGAIN)
6314 break;
6315 cond_resched();
6316 } while (1);
6317 }
Jens Axboe31b51512019-01-18 22:56:34 -07006318
Pavel Begunkova3df76982021-02-18 22:32:52 +00006319 /* avoid locking problems by failing it from a clean context */
Jens Axboe561fb042019-10-24 07:25:42 -06006320 if (ret) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00006321 /* io-wq is going to take one down */
Jens Axboede9b4cc2021-02-24 13:28:27 -07006322 req_ref_get(req);
Pavel Begunkova3df76982021-02-18 22:32:52 +00006323 io_req_task_queue_fail(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006324 }
Jens Axboe31b51512019-01-18 22:56:34 -07006325}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006326
Jens Axboe7b29f922021-03-12 08:30:14 -07006327#define FFS_ASYNC_READ 0x1UL
6328#define FFS_ASYNC_WRITE 0x2UL
6329#ifdef CONFIG_64BIT
6330#define FFS_ISREG 0x4UL
6331#else
6332#define FFS_ISREG 0x0UL
6333#endif
6334#define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG)
6335
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006336static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table,
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006337 unsigned i)
Jens Axboe09bb8392019-03-13 12:39:28 -06006338{
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006339 struct io_fixed_file *table_l2;
Jens Axboe65e19f52019-10-26 07:20:21 -06006340
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006341 table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT];
6342 return &table_l2[i & IORING_FILE_TABLE_MASK];
Pavel Begunkovdafecf12021-02-28 22:35:11 +00006343}
6344
Jens Axboe09bb8392019-03-13 12:39:28 -06006345static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6346 int index)
6347{
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006348 struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index);
Jens Axboe65e19f52019-10-26 07:20:21 -06006349
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006350 return (struct file *) (slot->file_ptr & FFS_MASK);
Jens Axboe65e19f52019-10-26 07:20:21 -06006351}
6352
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006353static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file)
Pavel Begunkov9a321c92021-04-01 15:44:01 +01006354{
6355 unsigned long file_ptr = (unsigned long) file;
6356
6357 if (__io_file_supports_async(file, READ))
6358 file_ptr |= FFS_ASYNC_READ;
6359 if (__io_file_supports_async(file, WRITE))
6360 file_ptr |= FFS_ASYNC_WRITE;
6361 if (S_ISREG(file_inode(file)->i_mode))
6362 file_ptr |= FFS_ISREG;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006363 file_slot->file_ptr = file_ptr;
Jens Axboe09bb8392019-03-13 12:39:28 -06006364}
6365
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006366static struct file *io_file_get(struct io_submit_state *state,
6367 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006368{
6369 struct io_ring_ctx *ctx = req->ctx;
6370 struct file *file;
6371
6372 if (fixed) {
Jens Axboe7b29f922021-03-12 08:30:14 -07006373 unsigned long file_ptr;
6374
Pavel Begunkov479f5172020-10-10 18:34:07 +01006375 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006376 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006377 fd = array_index_nospec(fd, ctx->nr_user_files);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006378 file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr;
Jens Axboe7b29f922021-03-12 08:30:14 -07006379 file = (struct file *) (file_ptr & FFS_MASK);
6380 file_ptr &= ~FFS_MASK;
6381 /* mask in overlapping REQ_F and FFS bits */
6382 req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01006383 io_req_set_rsrc_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006384 } else {
6385 trace_io_uring_file_get(ctx, fd);
6386 file = __io_file_get(state, fd);
Jens Axboed44f5542021-03-12 08:27:05 -07006387
6388 /* we don't allow fixed io_uring files */
6389 if (file && unlikely(file->f_op == &io_uring_fops))
6390 io_req_track_inflight(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006391 }
6392
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006393 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006394}
6395
Jens Axboe2665abf2019-11-05 12:40:47 -07006396static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6397{
Jens Axboead8a48a2019-11-15 08:49:11 -07006398 struct io_timeout_data *data = container_of(timer,
6399 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006400 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006401 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006402 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006403
6404 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006405 prev = req->timeout.head;
6406 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006407
6408 /*
6409 * We don't expect the list to be empty, that will only happen if we
6410 * race with the completion of the linked work.
6411 */
Pavel Begunkov447c19f2021-05-14 12:02:50 +01006412 if (prev) {
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006413 io_remove_next_linked(prev);
Pavel Begunkov447c19f2021-05-14 12:02:50 +01006414 if (!req_ref_inc_not_zero(prev))
6415 prev = NULL;
6416 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006417 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6418
6419 if (prev) {
Pavel Begunkov014db002020-03-03 21:33:12 +03006420 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006421 io_put_req_deferred(prev, 1);
Pavel Begunkova2982322021-05-07 21:06:38 +01006422 io_put_req_deferred(req, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006423 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006424 io_req_complete_post(req, -ETIME, 0);
Jens Axboe2665abf2019-11-05 12:40:47 -07006425 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006426 return HRTIMER_NORESTART;
6427}
6428
Pavel Begunkovde968c12021-03-19 17:22:33 +00006429static void io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006430{
Pavel Begunkovde968c12021-03-19 17:22:33 +00006431 struct io_ring_ctx *ctx = req->ctx;
6432
6433 spin_lock_irq(&ctx->completion_lock);
Jens Axboe76a46e02019-11-10 23:34:16 -07006434 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006435 * If the back reference is NULL, then our linked request finished
6436 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006437 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006438 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006439 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006440
Jens Axboead8a48a2019-11-15 08:49:11 -07006441 data->timer.function = io_link_timeout_fn;
6442 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6443 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006444 }
Jens Axboe76a46e02019-11-10 23:34:16 -07006445 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006446 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006447 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006448}
6449
Jens Axboead8a48a2019-11-15 08:49:11 -07006450static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006451{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006452 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006453
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006454 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6455 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006456 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006457
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006458 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006459 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006460 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006461 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006462}
6463
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006464static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006465{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006466 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006467 int ret;
6468
Olivier Langlois59b735a2021-06-22 05:17:39 -07006469issue_sqe:
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006470 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006471
6472 /*
6473 * We async punt it if the file wasn't marked NOWAIT, or if the file
6474 * doesn't support non-blocking read/write attempts
6475 */
Pavel Begunkov18400382021-03-19 17:22:34 +00006476 if (likely(!ret)) {
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006477 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006478 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006479 struct io_ring_ctx *ctx = req->ctx;
6480 struct io_comp_state *cs = &ctx->submit_state.comp;
Jens Axboee65ef562019-03-12 10:16:44 -06006481
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006482 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006483 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkov2a2758f2021-06-17 18:14:00 +01006484 io_submit_flush_completions(ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006485 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006486 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006487 }
Pavel Begunkov18400382021-03-19 17:22:34 +00006488 } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
Olivier Langlois59b735a2021-06-22 05:17:39 -07006489 switch (io_arm_poll_handler(req)) {
6490 case IO_APOLL_READY:
6491 goto issue_sqe;
6492 case IO_APOLL_ABORTED:
Pavel Begunkov18400382021-03-19 17:22:34 +00006493 /*
6494 * Queued up for async execution, worker will release
6495 * submit reference when the iocb is actually submitted.
6496 */
6497 io_queue_async_work(req);
Olivier Langlois59b735a2021-06-22 05:17:39 -07006498 break;
Pavel Begunkov18400382021-03-19 17:22:34 +00006499 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006500 } else {
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006501 io_req_complete_failed(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006502 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006503 if (linked_timeout)
6504 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006505}
6506
Pavel Begunkov441b8a72021-06-14 23:37:31 +01006507static inline void io_queue_sqe(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006508{
Pavel Begunkov10c66902021-06-15 16:47:56 +01006509 if (unlikely(req->ctx->drain_active) && io_drain_req(req))
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006510 return;
Jackie Liu4fe2c962019-09-09 20:50:40 +08006511
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006512 if (likely(!(req->flags & REQ_F_FORCE_ASYNC))) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006513 __io_queue_sqe(req);
Pavel Begunkov76cc33d2021-06-14 23:37:30 +01006514 } else {
6515 int ret = io_req_prep_async(req);
6516
6517 if (unlikely(ret))
6518 io_req_complete_failed(req, ret);
6519 else
6520 io_queue_async_work(req);
Jens Axboece35a472019-12-17 08:04:44 -07006521 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006522}
6523
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006524/*
6525 * Check SQE restrictions (opcode and flags).
6526 *
6527 * Returns 'true' if SQE is allowed, 'false' otherwise.
6528 */
6529static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6530 struct io_kiocb *req,
6531 unsigned int sqe_flags)
6532{
Pavel Begunkov4cfb25b2021-06-26 21:40:47 +01006533 if (likely(!ctx->restricted))
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006534 return true;
6535
6536 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6537 return false;
6538
6539 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6540 ctx->restrictions.sqe_flags_required)
6541 return false;
6542
6543 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6544 ctx->restrictions.sqe_flags_required))
6545 return false;
6546
6547 return true;
6548}
6549
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006550static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006551 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006552{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006553 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006554 unsigned int sqe_flags;
Jens Axboe003e8dc2021-03-06 09:22:27 -07006555 int personality, ret = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006556
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006557 req->opcode = READ_ONCE(sqe->opcode);
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006558 /* same numerical values with corresponding REQ_F_*, safe to copy */
6559 req->flags = sqe_flags = READ_ONCE(sqe->flags);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006560 req->user_data = READ_ONCE(sqe->user_data);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006561 req->file = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006562 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006563 /* one is dropped after submission, the other at completion */
Jens Axboeabc54d62021-02-24 13:32:30 -07006564 atomic_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006565 req->task = current;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006566
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006567 /* enforce forwards compatibility on users */
Pavel Begunkovdddca222021-04-27 16:13:52 +01006568 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS))
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006569 return -EINVAL;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006570 if (unlikely(req->opcode >= IORING_OP_LAST))
6571 return -EINVAL;
Pavel Begunkov4cfb25b2021-06-26 21:40:47 +01006572 if (!io_check_restriction(ctx, req, sqe_flags))
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006573 return -EACCES;
6574
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006575 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6576 !io_op_defs[req->opcode].buffer_select)
6577 return -EOPNOTSUPP;
Pavel Begunkov3c199662021-06-15 16:47:57 +01006578 if (unlikely(sqe_flags & IOSQE_IO_DRAIN))
6579 ctx->drain_active = true;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006580
Jens Axboe003e8dc2021-03-06 09:22:27 -07006581 personality = READ_ONCE(sqe->personality);
6582 if (personality) {
Pavel Begunkovc10d1f92021-06-17 18:14:01 +01006583 req->creds = xa_load(&ctx->personalities, personality);
6584 if (!req->creds)
Jens Axboe003e8dc2021-03-06 09:22:27 -07006585 return -EINVAL;
Pavel Begunkovc10d1f92021-06-17 18:14:01 +01006586 get_cred(req->creds);
Pavel Begunkovb8e64b52021-06-17 18:14:02 +01006587 req->flags |= REQ_F_CREDS;
Jens Axboe003e8dc2021-03-06 09:22:27 -07006588 }
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006589 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006590
Jens Axboe27926b62020-10-28 09:33:23 -06006591 /*
6592 * Plug now if we have more than 1 IO left after this, and the target
6593 * is potentially a read/write to block based storage.
6594 */
6595 if (!state->plug_started && state->ios_left > 1 &&
6596 io_op_defs[req->opcode].plug) {
6597 blk_start_plug(&state->plug);
6598 state->plug_started = true;
6599 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006600
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006601 if (io_op_defs[req->opcode].needs_file) {
6602 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006603
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006604 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006605 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006606 ret = -EBADF;
6607 }
6608
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006609 state->ios_left--;
6610 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006611}
6612
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006613static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006614 const struct io_uring_sqe *sqe)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006615{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006616 struct io_submit_link *link = &ctx->submit_state.link;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006617 int ret;
6618
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006619 ret = io_init_req(ctx, req, sqe);
6620 if (unlikely(ret)) {
6621fail_req:
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006622 if (link->head) {
6623 /* fail even hard links since we don't submit */
Pavel Begunkov93d2bcd2021-05-16 22:58:05 +01006624 req_set_fail(link->head);
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006625 io_req_complete_failed(link->head, -ECANCELED);
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006626 link->head = NULL;
6627 }
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006628 io_req_complete_failed(req, ret);
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006629 return ret;
6630 }
Pavel Begunkov441b8a72021-06-14 23:37:31 +01006631
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006632 ret = io_req_prep(req, sqe);
6633 if (unlikely(ret))
6634 goto fail_req;
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006635
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006636 /* don't need @sqe from now on */
Olivier Langlois236daeae2021-05-31 02:36:37 -04006637 trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data,
6638 req->flags, true,
6639 ctx->flags & IORING_SETUP_SQPOLL);
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006640
Jens Axboe6c271ce2019-01-10 11:22:30 -07006641 /*
6642 * If we already have a head request, queue this one for async
6643 * submittal once the head completes. If we don't have a head but
6644 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6645 * submitted sync once the chain is complete. If none of those
6646 * conditions are true (normal request), then just queue it.
6647 */
6648 if (link->head) {
6649 struct io_kiocb *head = link->head;
6650
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006651 ret = io_req_prep_async(req);
Pavel Begunkovcf109602021-02-18 18:29:43 +00006652 if (unlikely(ret))
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006653 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006654 trace_io_uring_link(ctx, req, head);
6655 link->last->link = req;
6656 link->last = req;
6657
6658 /* last request of a link, enqueue the link */
6659 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
6660 link->head = NULL;
Pavel Begunkov5e159202021-06-14 23:37:26 +01006661 io_queue_sqe(head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006662 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006663 } else {
Jens Axboe2b188cc2019-01-07 10:46:33 -07006664 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Jackie Liu4fe2c962019-09-09 20:50:40 +08006665 link->head = req;
6666 link->last = req;
6667 } else {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006668 io_queue_sqe(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006669 }
6670 }
6671
6672 return 0;
6673}
6674
6675/*
6676 * Batched submission is done, ensure local IO is flushed out.
6677 */
6678static void io_submit_state_end(struct io_submit_state *state,
6679 struct io_ring_ctx *ctx)
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006680{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006681 if (state->link.head)
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006682 io_queue_sqe(state->link.head);
Jens Axboe3529d8c2019-12-19 18:24:38 -07006683 if (state->comp.nr)
Pavel Begunkov2a2758f2021-06-17 18:14:00 +01006684 io_submit_flush_completions(ctx);
Jackie Liua197f662019-11-08 08:09:12 -07006685 if (state->plug_started)
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006686 blk_finish_plug(&state->plug);
Jens Axboe75c6a032020-01-28 10:15:23 -07006687 io_state_file_put(state);
Jens Axboe9e645e112019-05-10 16:07:28 -06006688}
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006689
Jens Axboe9e645e112019-05-10 16:07:28 -06006690/*
6691 * Start submission side cache.
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006692 */
Jens Axboe9e645e112019-05-10 16:07:28 -06006693static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkov196be952019-11-07 01:41:06 +03006694 unsigned int max_ios)
Jens Axboe9e645e112019-05-10 16:07:28 -06006695{
6696 state->plug_started = false;
Jens Axboebcda7ba2020-02-23 16:42:51 -07006697 state->ios_left = max_ios;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006698 /* set only head, no need to init link_last in advance */
6699 state->link.head = NULL;
Jens Axboe75c6a032020-01-28 10:15:23 -07006700}
6701
Jens Axboe193155c2020-02-22 23:22:19 -07006702static void io_commit_sqring(struct io_ring_ctx *ctx)
6703{
Jens Axboe75c6a032020-01-28 10:15:23 -07006704 struct io_rings *rings = ctx->rings;
6705
6706 /*
Jens Axboe193155c2020-02-22 23:22:19 -07006707 * Ensure any loads from the SQEs are done at this point,
Jens Axboe75c6a032020-01-28 10:15:23 -07006708 * since once we write the new head, the application could
6709 * write new data to them.
Pavel Begunkov6b47ee62020-01-18 20:22:41 +03006710 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006711 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboebcda7ba2020-02-23 16:42:51 -07006712}
6713
Jens Axboe9e645e112019-05-10 16:07:28 -06006714/*
Fam Zhengdd9ae8a2021-06-04 17:42:56 +01006715 * Fetch an sqe, if one is available. Note this returns a pointer to memory
Jens Axboe9e645e112019-05-10 16:07:28 -06006716 * that is mapped by userspace. This means that care needs to be taken to
6717 * ensure that reads are stable, as we cannot rely on userspace always
Jens Axboe78e19bb2019-11-06 15:21:34 -07006718 * being a good citizen. If members of the sqe are validated and then later
6719 * used, it's important that those reads are done through READ_ONCE() to
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006720 * prevent a re-load down the line.
Jens Axboe9e645e112019-05-10 16:07:28 -06006721 */
6722static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe9e645e112019-05-10 16:07:28 -06006723{
Pavel Begunkovea5ab3b2021-05-16 22:58:09 +01006724 unsigned head, mask = ctx->sq_entries - 1;
Pavel Begunkov17d3aeb2021-06-14 23:37:23 +01006725 unsigned sq_idx = ctx->cached_sq_head++ & mask;
Jens Axboe9e645e112019-05-10 16:07:28 -06006726
6727 /*
6728 * The cached sq head (or cq tail) serves two purposes:
6729 *
6730 * 1) allows us to batch the cost of updating the user visible
Pavel Begunkov9d763772019-12-17 02:22:07 +03006731 * head updates.
Jens Axboe9e645e112019-05-10 16:07:28 -06006732 * 2) allows the kernel side to track the head on its own, even
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006733 * though the application is the one updating it.
6734 */
Pavel Begunkov17d3aeb2021-06-14 23:37:23 +01006735 head = READ_ONCE(ctx->sq_array[sq_idx]);
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006736 if (likely(head < ctx->sq_entries))
6737 return &ctx->sq_sqes[head];
6738
6739 /* drop invalid entries */
Pavel Begunkov15641e42021-06-14 23:37:24 +01006740 ctx->cq_extra--;
6741 WRITE_ONCE(ctx->rings->sq_dropped,
6742 READ_ONCE(ctx->rings->sq_dropped) + 1);
Pavel Begunkov711be032020-01-17 03:57:59 +03006743 return NULL;
6744}
Jens Axboeb7bb4f72019-12-15 22:13:43 -07006745
Jens Axboe0f212202020-09-13 13:09:39 -06006746static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006747{
Pavel Begunkov09899b12021-06-14 02:36:22 +01006748 struct io_uring_task *tctx;
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006749 int submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006750
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006751 /* make sure SQ entry isn't read before tail */
6752 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006753 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6754 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006755
Pavel Begunkov09899b12021-06-14 02:36:22 +01006756 tctx = current->io_uring;
6757 tctx->cached_refs -= nr;
6758 if (unlikely(tctx->cached_refs < 0)) {
6759 unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR;
6760
6761 percpu_counter_add(&tctx->inflight, refill);
6762 refcount_add(refill, &current->usage);
6763 tctx->cached_refs += refill;
6764 }
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006765 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006766
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006767 while (submitted < nr) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006768 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006769 struct io_kiocb *req;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006770
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006771 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006772 if (unlikely(!req)) {
6773 if (!submitted)
6774 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006775 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006776 }
Pavel Begunkov4fccfcb2021-02-12 11:55:17 +00006777 sqe = io_get_sqe(ctx);
6778 if (unlikely(!sqe)) {
6779 kmem_cache_free(req_cachep, req);
6780 break;
6781 }
Jens Axboed3656342019-12-18 09:50:26 -07006782 /* will complete beyond this point, count as submitted */
6783 submitted++;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006784 if (io_submit_sqe(ctx, req, sqe))
Jens Axboed3656342019-12-18 09:50:26 -07006785 break;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006786 }
6787
Pavel Begunkov9466f432020-01-25 22:34:01 +03006788 if (unlikely(submitted != nr)) {
6789 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006790 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006791
Pavel Begunkov09899b12021-06-14 02:36:22 +01006792 current->io_uring->cached_refs += unused;
Jens Axboed8a6df12020-10-15 16:24:45 -06006793 percpu_ref_put_many(&ctx->refs, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006794 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006795
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006796 io_submit_state_end(&ctx->submit_state, ctx);
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006797 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6798 io_commit_sqring(ctx);
6799
Jens Axboe6c271ce2019-01-10 11:22:30 -07006800 return submitted;
6801}
6802
Pavel Begunkove4b6d902021-05-16 22:58:00 +01006803static inline bool io_sqd_events_pending(struct io_sq_data *sqd)
6804{
6805 return READ_ONCE(sqd->state);
6806}
6807
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006808static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6809{
6810 /* Tell userspace we may need a wakeup call */
6811 spin_lock_irq(&ctx->completion_lock);
6812 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6813 spin_unlock_irq(&ctx->completion_lock);
6814}
6815
6816static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6817{
6818 spin_lock_irq(&ctx->completion_lock);
6819 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6820 spin_unlock_irq(&ctx->completion_lock);
6821}
6822
Xiaoguang Wang08369242020-11-03 14:15:59 +08006823static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006824{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006825 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006826 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006827
Jens Axboec8d1ba52020-09-14 11:07:26 -06006828 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006829 /* if we're handling multiple rings, cap submit size for fairness */
Olivier Langlois4ce8ad92021-06-23 11:50:18 -07006830 if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE)
6831 to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE;
Jens Axboee95eee22020-09-08 09:11:32 -06006832
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006833 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6834 unsigned nr_events = 0;
Pavel Begunkov948e1942021-06-24 15:09:55 +01006835 const struct cred *creds = NULL;
6836
6837 if (ctx->sq_creds != current_cred())
6838 creds = override_creds(ctx->sq_creds);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006839
Xiaoguang Wang08369242020-11-03 14:15:59 +08006840 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006841 if (!list_empty(&ctx->iopoll_list))
6842 io_do_iopoll(ctx, &nr_events, 0);
6843
Pavel Begunkov3b763ba2021-04-18 14:52:08 +01006844 /*
6845 * Don't submit if refs are dying, good for io_uring_register(),
6846 * but also it is relied upon by io_ring_exit_work()
6847 */
Pavel Begunkov0298ef92021-03-08 13:20:57 +00006848 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
6849 !(ctx->flags & IORING_SETUP_R_DISABLED))
Xiaoguang Wang08369242020-11-03 14:15:59 +08006850 ret = io_submit_sqes(ctx, to_submit);
6851 mutex_unlock(&ctx->uring_lock);
Jens Axboe90554202020-09-03 12:12:41 -06006852
Pavel Begunkovacfb3812021-05-16 22:58:03 +01006853 if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait))
6854 wake_up(&ctx->sqo_sq_wait);
Pavel Begunkov948e1942021-06-24 15:09:55 +01006855 if (creds)
6856 revert_creds(creds);
Pavel Begunkovacfb3812021-05-16 22:58:03 +01006857 }
Jens Axboe90554202020-09-03 12:12:41 -06006858
Xiaoguang Wang08369242020-11-03 14:15:59 +08006859 return ret;
6860}
6861
6862static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
6863{
6864 struct io_ring_ctx *ctx;
6865 unsigned sq_thread_idle = 0;
6866
Pavel Begunkovc9dca272021-03-10 13:13:55 +00006867 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6868 sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006869 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006870}
6871
Pavel Begunkove4b6d902021-05-16 22:58:00 +01006872static bool io_sqd_handle_event(struct io_sq_data *sqd)
6873{
6874 bool did_sig = false;
6875 struct ksignal ksig;
6876
6877 if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) ||
6878 signal_pending(current)) {
6879 mutex_unlock(&sqd->lock);
6880 if (signal_pending(current))
6881 did_sig = get_signal(&ksig);
6882 cond_resched();
6883 mutex_lock(&sqd->lock);
6884 }
Pavel Begunkove4b6d902021-05-16 22:58:00 +01006885 return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
6886}
6887
Jens Axboe6c271ce2019-01-10 11:22:30 -07006888static int io_sq_thread(void *data)
6889{
Jens Axboe69fb2132020-09-14 11:16:23 -06006890 struct io_sq_data *sqd = data;
6891 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d92052020-11-12 14:55:59 +08006892 unsigned long timeout = 0;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006893 char buf[TASK_COMM_LEN];
Xiaoguang Wang08369242020-11-03 14:15:59 +08006894 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006895
Pavel Begunkov696ee882021-04-01 09:55:04 +01006896 snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006897 set_task_comm(current, buf);
Jens Axboe28cea78a2020-09-14 10:51:17 -06006898
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006899 if (sqd->sq_cpu != -1)
6900 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
6901 else
6902 set_cpus_allowed_ptr(current, cpu_online_mask);
6903 current->flags |= PF_NO_SETAFFINITY;
6904
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006905 mutex_lock(&sqd->lock);
Pavel Begunkove4b6d902021-05-16 22:58:00 +01006906 while (1) {
Pavel Begunkov1a924a82021-06-24 15:09:56 +01006907 bool cap_entries, sqt_spin = false;
Jens Axboec1edbf52019-11-10 16:56:04 -07006908
Pavel Begunkove4b6d902021-05-16 22:58:00 +01006909 if (io_sqd_events_pending(sqd) || signal_pending(current)) {
6910 if (io_sqd_handle_event(sqd))
Pavel Begunkovc7d95612021-04-13 11:43:00 +01006911 break;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006912 timeout = jiffies + sqd->sq_thread_idle;
6913 }
Pavel Begunkove4b6d902021-05-16 22:58:00 +01006914
Jens Axboee95eee22020-09-08 09:11:32 -06006915 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006916 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Pavel Begunkov948e1942021-06-24 15:09:55 +01006917 int ret = __io_sq_thread(ctx, cap_entries);
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006918
Xiaoguang Wang08369242020-11-03 14:15:59 +08006919 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
6920 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006921 }
Pavel Begunkovdd432ea52021-06-26 21:40:45 +01006922 if (io_run_task_work())
6923 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006924
Xiaoguang Wang08369242020-11-03 14:15:59 +08006925 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006926 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08006927 if (sqt_spin)
6928 timeout = jiffies + sqd->sq_thread_idle;
6929 continue;
6930 }
6931
Xiaoguang Wang08369242020-11-03 14:15:59 +08006932 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
Pavel Begunkovdd432ea52021-06-26 21:40:45 +01006933 if (!io_sqd_events_pending(sqd) && !current->task_works) {
Pavel Begunkov1a924a82021-06-24 15:09:56 +01006934 bool needs_sched = true;
6935
Hao Xu724cb4f2021-04-21 23:19:11 +08006936 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Pavel Begunkovaaa9f0f2021-05-16 22:58:01 +01006937 io_ring_set_wakeup_flag(ctx);
6938
Hao Xu724cb4f2021-04-21 23:19:11 +08006939 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6940 !list_empty_careful(&ctx->iopoll_list)) {
6941 needs_sched = false;
6942 break;
6943 }
6944 if (io_sqring_entries(ctx)) {
6945 needs_sched = false;
6946 break;
6947 }
6948 }
6949
6950 if (needs_sched) {
6951 mutex_unlock(&sqd->lock);
6952 schedule();
6953 mutex_lock(&sqd->lock);
6954 }
Jens Axboe69fb2132020-09-14 11:16:23 -06006955 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6956 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006957 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006958
6959 finish_wait(&sqd->wait, &wait);
6960 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006961 }
6962
Pavel Begunkov78cc6872021-06-14 02:36:23 +01006963 io_uring_cancel_generic(true, sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006964 sqd->thread = NULL;
Jens Axboe05962f92021-03-06 13:58:48 -07006965 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
Jens Axboe5f3f26f2021-02-25 10:17:46 -07006966 io_ring_set_wakeup_flag(ctx);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00006967 io_run_task_work();
Pavel Begunkov734551d2021-04-18 14:52:09 +01006968 mutex_unlock(&sqd->lock);
6969
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006970 complete(&sqd->exited);
6971 do_exit(0);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006972}
6973
Jens Axboebda52162019-09-24 13:47:15 -06006974struct io_wait_queue {
6975 struct wait_queue_entry wq;
6976 struct io_ring_ctx *ctx;
6977 unsigned to_wait;
6978 unsigned nr_timeouts;
6979};
6980
Pavel Begunkov6c503152021-01-04 20:36:36 +00006981static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06006982{
6983 struct io_ring_ctx *ctx = iowq->ctx;
6984
6985 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006986 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006987 * started waiting. For timeouts, we always want to return to userspace,
6988 * regardless of event count.
6989 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00006990 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006991 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6992}
6993
6994static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6995 int wake_flags, void *key)
6996{
6997 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6998 wq);
6999
Pavel Begunkov6c503152021-01-04 20:36:36 +00007000 /*
7001 * Cannot safely flush overflowed CQEs from here, ensure we wake up
7002 * the task, and the next invocation will do it.
7003 */
Pavel Begunkov5ed7a372021-06-14 23:37:27 +01007004 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow))
Pavel Begunkov6c503152021-01-04 20:36:36 +00007005 return autoremove_wake_function(curr, mode, wake_flags, key);
7006 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06007007}
7008
Jens Axboeaf9c1a42020-09-24 13:32:18 -06007009static int io_run_task_work_sig(void)
7010{
7011 if (io_run_task_work())
7012 return 1;
7013 if (!signal_pending(current))
7014 return 0;
Jens Axboe0b8cfa92021-03-21 14:16:08 -06007015 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
Jens Axboe792ee0f62020-10-22 20:17:18 -06007016 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06007017 return -EINTR;
7018}
7019
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007020/* when returns >0, the caller should retry */
7021static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
7022 struct io_wait_queue *iowq,
7023 signed long *timeout)
7024{
7025 int ret;
7026
7027 /* make sure we run task_work before checking for signals */
7028 ret = io_run_task_work_sig();
7029 if (ret || io_should_wake(iowq))
7030 return ret;
7031 /* let the caller flush overflows, retry */
Pavel Begunkov5ed7a372021-06-14 23:37:27 +01007032 if (test_bit(0, &ctx->check_cq_overflow))
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007033 return 1;
7034
7035 *timeout = schedule_timeout(*timeout);
7036 return !*timeout ? -ETIME : 1;
7037}
7038
Jens Axboe2b188cc2019-01-07 10:46:33 -07007039/*
7040 * Wait until events become available, if we don't already have some. The
7041 * application must reap them itself, as they reside on the shared cq ring.
7042 */
7043static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08007044 const sigset_t __user *sig, size_t sigsz,
7045 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007046{
Jens Axboebda52162019-09-24 13:47:15 -06007047 struct io_wait_queue iowq = {
7048 .wq = {
7049 .private = current,
7050 .func = io_wake_function,
7051 .entry = LIST_HEAD_INIT(iowq.wq.entry),
7052 },
7053 .ctx = ctx,
7054 .to_wait = min_events,
7055 };
Hristo Venev75b28af2019-08-26 17:23:46 +00007056 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00007057 signed long timeout = MAX_SCHEDULE_TIMEOUT;
7058 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007059
Jens Axboeb41e9852020-02-17 09:52:41 -07007060 do {
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00007061 io_cqring_overflow_flush(ctx, false);
Pavel Begunkov6c503152021-01-04 20:36:36 +00007062 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07007063 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06007064 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07007065 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07007066 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007067
7068 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007069#ifdef CONFIG_COMPAT
7070 if (in_compat_syscall())
7071 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07007072 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007073 else
7074#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07007075 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007076
Jens Axboe2b188cc2019-01-07 10:46:33 -07007077 if (ret)
7078 return ret;
7079 }
7080
Hao Xuc73ebb62020-11-03 10:54:37 +08007081 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00007082 struct timespec64 ts;
7083
Hao Xuc73ebb62020-11-03 10:54:37 +08007084 if (get_timespec64(&ts, uts))
7085 return -EFAULT;
7086 timeout = timespec64_to_jiffies(&ts);
7087 }
7088
Jens Axboebda52162019-09-24 13:47:15 -06007089 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02007090 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06007091 do {
Jens Axboeca0a2652021-03-04 17:15:48 -07007092 /* if we can't even flush overflow, don't wait for more */
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00007093 if (!io_cqring_overflow_flush(ctx, false)) {
Jens Axboeca0a2652021-03-04 17:15:48 -07007094 ret = -EBUSY;
7095 break;
7096 }
Pavel Begunkov311997b2021-06-14 23:37:28 +01007097 prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq,
Jens Axboebda52162019-09-24 13:47:15 -06007098 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007099 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
Pavel Begunkov311997b2021-06-14 23:37:28 +01007100 finish_wait(&ctx->cq_wait, &iowq.wq);
Jens Axboeca0a2652021-03-04 17:15:48 -07007101 cond_resched();
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007102 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06007103
Jens Axboeb7db41c2020-07-04 08:55:50 -06007104 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007105
Hristo Venev75b28af2019-08-26 17:23:46 +00007106 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007107}
7108
Pavel Begunkov9123c8f2021-06-14 02:36:20 +01007109static void io_free_page_table(void **table, size_t size)
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007110{
Pavel Begunkov9123c8f2021-06-14 02:36:20 +01007111 unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE);
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007112
7113 for (i = 0; i < nr_tables; i++)
Pavel Begunkov9123c8f2021-06-14 02:36:20 +01007114 kfree(table[i]);
7115 kfree(table);
7116}
7117
7118static void **io_alloc_page_table(size_t size)
7119{
7120 unsigned i, nr_tables = DIV_ROUND_UP(size, PAGE_SIZE);
7121 size_t init_size = size;
7122 void **table;
7123
7124 table = kcalloc(nr_tables, sizeof(*table), GFP_KERNEL);
7125 if (!table)
7126 return NULL;
7127
7128 for (i = 0; i < nr_tables; i++) {
Pavel Begunkov27f6b312021-06-15 13:20:13 +01007129 unsigned int this_size = min_t(size_t, size, PAGE_SIZE);
Pavel Begunkov9123c8f2021-06-14 02:36:20 +01007130
7131 table[i] = kzalloc(this_size, GFP_KERNEL);
7132 if (!table[i]) {
7133 io_free_page_table(table, init_size);
7134 return NULL;
7135 }
7136 size -= this_size;
7137 }
7138 return table;
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007139}
7140
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007141static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
Pavel Begunkov1642b442020-12-30 21:34:14 +00007142{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007143 spin_lock_bh(&ctx->rsrc_ref_lock);
Pavel Begunkov1642b442020-12-30 21:34:14 +00007144}
7145
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007146static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
Jens Axboe6b063142019-01-10 22:13:58 -07007147{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007148 spin_unlock_bh(&ctx->rsrc_ref_lock);
7149}
7150
Pavel Begunkov28a9fe22021-04-01 15:43:47 +01007151static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node)
7152{
7153 percpu_ref_exit(&ref_node->refs);
7154 kfree(ref_node);
7155}
7156
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007157static void io_rsrc_node_switch(struct io_ring_ctx *ctx,
7158 struct io_rsrc_data *data_to_kill)
Jens Axboe6b063142019-01-10 22:13:58 -07007159{
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007160 WARN_ON_ONCE(!ctx->rsrc_backup_node);
7161 WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node);
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007162
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007163 if (data_to_kill) {
7164 struct io_rsrc_node *rsrc_node = ctx->rsrc_node;
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007165
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007166 rsrc_node->rsrc_data = data_to_kill;
7167 io_rsrc_ref_lock(ctx);
7168 list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list);
7169 io_rsrc_ref_unlock(ctx);
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007170
Pavel Begunkov3e942492021-04-11 01:46:34 +01007171 atomic_inc(&data_to_kill->refs);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007172 percpu_ref_kill(&rsrc_node->refs);
7173 ctx->rsrc_node = NULL;
7174 }
7175
7176 if (!ctx->rsrc_node) {
7177 ctx->rsrc_node = ctx->rsrc_backup_node;
7178 ctx->rsrc_backup_node = NULL;
7179 }
Jens Axboe6b063142019-01-10 22:13:58 -07007180}
7181
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007182static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx)
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007183{
7184 if (ctx->rsrc_backup_node)
7185 return 0;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007186 ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007187 return ctx->rsrc_backup_node ? 0 : -ENOMEM;
7188}
7189
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007190static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx)
Hao Xu8bad28d2021-02-19 17:19:36 +08007191{
7192 int ret;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007193
Pavel Begunkov215c3902021-04-01 15:43:48 +01007194 /* As we may drop ->uring_lock, other task may have started quiesce */
Hao Xu8bad28d2021-02-19 17:19:36 +08007195 if (data->quiesce)
7196 return -ENXIO;
7197
7198 data->quiesce = true;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007199 do {
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007200 ret = io_rsrc_node_switch_start(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007201 if (ret)
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007202 break;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007203 io_rsrc_node_switch(ctx, data);
7204
Pavel Begunkov3e942492021-04-11 01:46:34 +01007205 /* kill initial ref, already quiesced if zero */
7206 if (atomic_dec_and_test(&data->refs))
7207 break;
Hao Xu8bad28d2021-02-19 17:19:36 +08007208 flush_delayed_work(&ctx->rsrc_put_work);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007209 ret = wait_for_completion_interruptible(&data->done);
7210 if (!ret)
7211 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007212
Pavel Begunkov3e942492021-04-11 01:46:34 +01007213 atomic_inc(&data->refs);
7214 /* wait for all works potentially completing data->done */
7215 flush_delayed_work(&ctx->rsrc_put_work);
Jens Axboecb5e1b82021-02-25 07:37:35 -07007216 reinit_completion(&data->done);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007217
Hao Xu8bad28d2021-02-19 17:19:36 +08007218 mutex_unlock(&ctx->uring_lock);
7219 ret = io_run_task_work_sig();
7220 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007221 } while (ret >= 0);
Hao Xu8bad28d2021-02-19 17:19:36 +08007222 data->quiesce = false;
7223
Hao Xu8bad28d2021-02-19 17:19:36 +08007224 return ret;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007225}
7226
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007227static u64 *io_get_tag_slot(struct io_rsrc_data *data, unsigned int idx)
7228{
7229 unsigned int off = idx & IO_RSRC_TAG_TABLE_MASK;
7230 unsigned int table_idx = idx >> IO_RSRC_TAG_TABLE_SHIFT;
7231
7232 return &data->tags[table_idx][off];
7233}
7234
Pavel Begunkov44b31f22021-04-25 14:32:16 +01007235static void io_rsrc_data_free(struct io_rsrc_data *data)
7236{
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007237 size_t size = data->nr * sizeof(data->tags[0][0]);
7238
7239 if (data->tags)
7240 io_free_page_table((void **)data->tags, size);
Pavel Begunkov44b31f22021-04-25 14:32:16 +01007241 kfree(data);
7242}
7243
Pavel Begunkovd878c812021-06-14 02:36:18 +01007244static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put,
7245 u64 __user *utags, unsigned nr,
7246 struct io_rsrc_data **pdata)
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007247{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007248 struct io_rsrc_data *data;
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007249 int ret = -ENOMEM;
Pavel Begunkovd878c812021-06-14 02:36:18 +01007250 unsigned i;
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007251
7252 data = kzalloc(sizeof(*data), GFP_KERNEL);
7253 if (!data)
Pavel Begunkovd878c812021-06-14 02:36:18 +01007254 return -ENOMEM;
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007255 data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0]));
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007256 if (!data->tags) {
7257 kfree(data);
Pavel Begunkovd878c812021-06-14 02:36:18 +01007258 return -ENOMEM;
7259 }
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007260
7261 data->nr = nr;
7262 data->ctx = ctx;
7263 data->do_put = do_put;
Pavel Begunkovd878c812021-06-14 02:36:18 +01007264 if (utags) {
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007265 ret = -EFAULT;
Pavel Begunkovd878c812021-06-14 02:36:18 +01007266 for (i = 0; i < nr; i++) {
Colin Ian Kingfdd1dc32021-06-15 14:00:11 +01007267 u64 *tag_slot = io_get_tag_slot(data, i);
7268
7269 if (copy_from_user(tag_slot, &utags[i],
7270 sizeof(*tag_slot)))
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007271 goto fail;
Pavel Begunkovd878c812021-06-14 02:36:18 +01007272 }
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007273 }
7274
Pavel Begunkov3e942492021-04-11 01:46:34 +01007275 atomic_set(&data->refs, 1);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007276 init_completion(&data->done);
Pavel Begunkovd878c812021-06-14 02:36:18 +01007277 *pdata = data;
7278 return 0;
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007279fail:
7280 io_rsrc_data_free(data);
7281 return ret;
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007282}
7283
Pavel Begunkov9123c8f2021-06-14 02:36:20 +01007284static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files)
7285{
7286 size_t size = nr_files * sizeof(struct io_fixed_file);
7287
7288 table->files = (struct io_fixed_file **)io_alloc_page_table(size);
7289 return !!table->files;
7290}
7291
7292static void io_free_file_tables(struct io_file_table *table, unsigned nr_files)
7293{
7294 size_t size = nr_files * sizeof(struct io_fixed_file);
7295
7296 io_free_page_table((void **)table->files, size);
7297 table->files = NULL;
7298}
7299
Roman Penyaev2bbcd6d2019-05-16 10:53:57 +02007300static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
7301{
Jens Axboe06058632019-04-13 09:26:03 -06007302#if defined(CONFIG_UNIX)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007303 if (ctx->ring_sock) {
7304 struct sock *sock = ctx->ring_sock->sk;
7305 struct sk_buff *skb;
7306
7307 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
Jens Axboe6b063142019-01-10 22:13:58 -07007308 kfree_skb(skb);
7309 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07007310#else
7311 int i;
Jens Axboe6b063142019-01-10 22:13:58 -07007312
7313 for (i = 0; i < ctx->nr_user_files; i++) {
7314 struct file *file;
7315
7316 file = io_file_from_index(ctx, i);
7317 if (file)
7318 fput(file);
7319 }
7320#endif
Pavel Begunkovfff4db72021-04-25 14:32:15 +01007321 io_free_file_tables(&ctx->file_table, ctx->nr_user_files);
Pavel Begunkov44b31f22021-04-25 14:32:16 +01007322 io_rsrc_data_free(ctx->file_data);
Pavel Begunkovfff4db72021-04-25 14:32:15 +01007323 ctx->file_data = NULL;
7324 ctx->nr_user_files = 0;
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007325}
7326
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007327static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7328{
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007329 int ret;
7330
Pavel Begunkov08480402021-04-13 02:58:38 +01007331 if (!ctx->file_data)
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007332 return -ENXIO;
Pavel Begunkov08480402021-04-13 02:58:38 +01007333 ret = io_rsrc_ref_quiesce(ctx->file_data, ctx);
7334 if (!ret)
7335 __io_sqe_files_unregister(ctx);
7336 return ret;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007337}
7338
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007339static void io_sq_thread_unpark(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007340 __releases(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007341{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007342 WARN_ON_ONCE(sqd->thread == current);
7343
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007344 /*
7345 * Do the dance but not conditional clear_bit() because it'd race with
7346 * other threads incrementing park_pending and setting the bit.
7347 */
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007348 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007349 if (atomic_dec_return(&sqd->park_pending))
7350 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007351 mutex_unlock(&sqd->lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007352}
7353
Jens Axboe86e0d672021-03-05 08:44:39 -07007354static void io_sq_thread_park(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007355 __acquires(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007356{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007357 WARN_ON_ONCE(sqd->thread == current);
7358
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007359 atomic_inc(&sqd->park_pending);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007360 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007361 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007362 if (sqd->thread)
Jens Axboe86e0d672021-03-05 08:44:39 -07007363 wake_up_process(sqd->thread);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007364}
7365
7366static void io_sq_thread_stop(struct io_sq_data *sqd)
7367{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007368 WARN_ON_ONCE(sqd->thread == current);
Pavel Begunkov88885f62021-04-11 01:46:38 +01007369 WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state));
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007370
Jens Axboe05962f92021-03-06 13:58:48 -07007371 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
Pavel Begunkov88885f62021-04-11 01:46:38 +01007372 mutex_lock(&sqd->lock);
Jens Axboee8f98f242021-03-09 16:32:13 -07007373 if (sqd->thread)
7374 wake_up_process(sqd->thread);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007375 mutex_unlock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007376 wait_for_completion(&sqd->exited);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007377}
7378
Jens Axboe534ca6d2020-09-02 13:52:19 -06007379static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007380{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007381 if (refcount_dec_and_test(&sqd->refs)) {
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007382 WARN_ON_ONCE(atomic_read(&sqd->park_pending));
7383
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007384 io_sq_thread_stop(sqd);
7385 kfree(sqd);
7386 }
7387}
7388
7389static void io_sq_thread_finish(struct io_ring_ctx *ctx)
7390{
7391 struct io_sq_data *sqd = ctx->sq_data;
7392
7393 if (sqd) {
Jens Axboe05962f92021-03-06 13:58:48 -07007394 io_sq_thread_park(sqd);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007395 list_del_init(&ctx->sqd_list);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007396 io_sqd_update_thread_idle(sqd);
Jens Axboe05962f92021-03-06 13:58:48 -07007397 io_sq_thread_unpark(sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007398
7399 io_put_sq_data(sqd);
7400 ctx->sq_data = NULL;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007401 }
7402}
7403
Jens Axboeaa061652020-09-02 14:50:27 -06007404static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7405{
7406 struct io_ring_ctx *ctx_attach;
7407 struct io_sq_data *sqd;
7408 struct fd f;
7409
7410 f = fdget(p->wq_fd);
7411 if (!f.file)
7412 return ERR_PTR(-ENXIO);
7413 if (f.file->f_op != &io_uring_fops) {
7414 fdput(f);
7415 return ERR_PTR(-EINVAL);
7416 }
7417
7418 ctx_attach = f.file->private_data;
7419 sqd = ctx_attach->sq_data;
7420 if (!sqd) {
7421 fdput(f);
7422 return ERR_PTR(-EINVAL);
7423 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007424 if (sqd->task_tgid != current->tgid) {
7425 fdput(f);
7426 return ERR_PTR(-EPERM);
7427 }
Jens Axboeaa061652020-09-02 14:50:27 -06007428
7429 refcount_inc(&sqd->refs);
7430 fdput(f);
7431 return sqd;
7432}
7433
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007434static struct io_sq_data *io_get_sq_data(struct io_uring_params *p,
7435 bool *attached)
Jens Axboe534ca6d2020-09-02 13:52:19 -06007436{
7437 struct io_sq_data *sqd;
7438
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007439 *attached = false;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007440 if (p->flags & IORING_SETUP_ATTACH_WQ) {
7441 sqd = io_attach_sq_data(p);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007442 if (!IS_ERR(sqd)) {
7443 *attached = true;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007444 return sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007445 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007446 /* fall through for EPERM case, setup new sqd/task */
7447 if (PTR_ERR(sqd) != -EPERM)
7448 return sqd;
7449 }
Jens Axboeaa061652020-09-02 14:50:27 -06007450
Jens Axboe534ca6d2020-09-02 13:52:19 -06007451 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7452 if (!sqd)
7453 return ERR_PTR(-ENOMEM);
7454
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007455 atomic_set(&sqd->park_pending, 0);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007456 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007457 INIT_LIST_HEAD(&sqd->ctx_list);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007458 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007459 init_waitqueue_head(&sqd->wait);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007460 init_completion(&sqd->exited);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007461 return sqd;
7462}
7463
Jens Axboe6b063142019-01-10 22:13:58 -07007464#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007465/*
7466 * Ensure the UNIX gc is aware of our file set, so we are certain that
7467 * the io_uring can be safely unregistered on process exit, even if we have
7468 * loops in the file referencing.
7469 */
7470static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7471{
7472 struct sock *sk = ctx->ring_sock->sk;
7473 struct scm_fp_list *fpl;
7474 struct sk_buff *skb;
Jens Axboe08a45172019-10-03 08:11:03 -06007475 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007476
Jens Axboe6b063142019-01-10 22:13:58 -07007477 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7478 if (!fpl)
7479 return -ENOMEM;
7480
7481 skb = alloc_skb(0, GFP_KERNEL);
7482 if (!skb) {
7483 kfree(fpl);
7484 return -ENOMEM;
7485 }
7486
7487 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007488
Jens Axboe08a45172019-10-03 08:11:03 -06007489 nr_files = 0;
Jens Axboe62e398b2021-02-21 16:19:37 -07007490 fpl->user = get_uid(current_user());
Jens Axboe6b063142019-01-10 22:13:58 -07007491 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007492 struct file *file = io_file_from_index(ctx, i + offset);
7493
7494 if (!file)
Jens Axboe08a45172019-10-03 08:11:03 -06007495 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007496 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a45172019-10-03 08:11:03 -06007497 unix_inflight(fpl->user, fpl->fp[nr_files]);
7498 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007499 }
7500
Jens Axboe08a45172019-10-03 08:11:03 -06007501 if (nr_files) {
7502 fpl->max = SCM_MAX_FD;
7503 fpl->count = nr_files;
7504 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007505 skb->destructor = unix_destruct_scm;
Jens Axboe08a45172019-10-03 08:11:03 -06007506 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7507 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007508
Jens Axboe08a45172019-10-03 08:11:03 -06007509 for (i = 0; i < nr_files; i++)
7510 fput(fpl->fp[i]);
7511 } else {
7512 kfree_skb(skb);
7513 kfree(fpl);
7514 }
Jens Axboe6b063142019-01-10 22:13:58 -07007515
7516 return 0;
7517}
7518
7519/*
7520 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7521 * causes regular reference counting to break down. We rely on the UNIX
7522 * garbage collection to take care of this problem for us.
7523 */
7524static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7525{
7526 unsigned left, total;
7527 int ret = 0;
7528
7529 total = 0;
7530 left = ctx->nr_user_files;
7531 while (left) {
7532 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
7533
7534 ret = __io_sqe_files_scm(ctx, this_files, total);
7535 if (ret)
7536 break;
7537 left -= this_files;
7538 total += this_files;
7539 }
7540
7541 if (!ret)
7542 return 0;
7543
7544 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007545 struct file *file = io_file_from_index(ctx, total);
7546
7547 if (file)
7548 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007549 total++;
7550 }
7551
7552 return ret;
7553}
7554#else
7555static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7556{
7557 return 0;
7558}
7559#endif
7560
Pavel Begunkov47e90392021-04-01 15:43:56 +01007561static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007562{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007563 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007564#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007565 struct sock *sock = ctx->ring_sock->sk;
7566 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7567 struct sk_buff *skb;
7568 int i;
7569
7570 __skb_queue_head_init(&list);
7571
7572 /*
7573 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7574 * remove this entry and rearrange the file array.
7575 */
7576 skb = skb_dequeue(head);
7577 while (skb) {
7578 struct scm_fp_list *fp;
7579
7580 fp = UNIXCB(skb).fp;
7581 for (i = 0; i < fp->count; i++) {
7582 int left;
7583
7584 if (fp->fp[i] != file)
7585 continue;
7586
7587 unix_notinflight(fp->user, fp->fp[i]);
7588 left = fp->count - 1 - i;
7589 if (left) {
7590 memmove(&fp->fp[i], &fp->fp[i + 1],
7591 left * sizeof(struct file *));
7592 }
7593 fp->count--;
7594 if (!fp->count) {
7595 kfree_skb(skb);
7596 skb = NULL;
7597 } else {
7598 __skb_queue_tail(&list, skb);
7599 }
7600 fput(file);
7601 file = NULL;
7602 break;
7603 }
7604
7605 if (!file)
7606 break;
7607
7608 __skb_queue_tail(&list, skb);
7609
7610 skb = skb_dequeue(head);
7611 }
7612
7613 if (skb_peek(&list)) {
7614 spin_lock_irq(&head->lock);
7615 while ((skb = __skb_dequeue(&list)) != NULL)
7616 __skb_queue_tail(head, skb);
7617 spin_unlock_irq(&head->lock);
7618 }
7619#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007620 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007621#endif
7622}
7623
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007624static void __io_rsrc_put_work(struct io_rsrc_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007625{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007626 struct io_rsrc_data *rsrc_data = ref_node->rsrc_data;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007627 struct io_ring_ctx *ctx = rsrc_data->ctx;
7628 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007629
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007630 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7631 list_del(&prsrc->list);
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007632
7633 if (prsrc->tag) {
7634 bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL;
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007635
7636 io_ring_submit_lock(ctx, lock_ring);
Pavel Begunkov157d2572021-06-14 02:36:19 +01007637 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007638 io_cqring_fill_event(ctx, prsrc->tag, 0, 0);
Pavel Begunkov2840f712021-04-27 16:13:51 +01007639 ctx->cq_extra++;
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007640 io_commit_cqring(ctx);
Pavel Begunkov157d2572021-06-14 02:36:19 +01007641 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007642 io_cqring_ev_posted(ctx);
7643 io_ring_submit_unlock(ctx, lock_ring);
7644 }
7645
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007646 rsrc_data->do_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007647 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007648 }
7649
Pavel Begunkov28a9fe22021-04-01 15:43:47 +01007650 io_rsrc_node_destroy(ref_node);
Pavel Begunkov3e942492021-04-11 01:46:34 +01007651 if (atomic_dec_and_test(&rsrc_data->refs))
7652 complete(&rsrc_data->done);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007653}
7654
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007655static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007656{
7657 struct io_ring_ctx *ctx;
7658 struct llist_node *node;
7659
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007660 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7661 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007662
7663 while (node) {
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007664 struct io_rsrc_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007665 struct llist_node *next = node->next;
7666
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007667 ref_node = llist_entry(node, struct io_rsrc_node, llist);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007668 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007669 node = next;
7670 }
7671}
7672
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007673static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007674{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007675 struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs);
Pavel Begunkov3e942492021-04-11 01:46:34 +01007676 struct io_ring_ctx *ctx = node->rsrc_data->ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007677 bool first_add = false;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007678
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007679 io_rsrc_ref_lock(ctx);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007680 node->done = true;
Pavel Begunkove2978222020-11-18 14:56:26 +00007681
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007682 while (!list_empty(&ctx->rsrc_ref_list)) {
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007683 node = list_first_entry(&ctx->rsrc_ref_list,
7684 struct io_rsrc_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007685 /* recycle ref nodes in order */
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007686 if (!node->done)
Pavel Begunkove2978222020-11-18 14:56:26 +00007687 break;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007688 list_del(&node->node);
7689 first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007690 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007691 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007692
Pavel Begunkov3e942492021-04-11 01:46:34 +01007693 if (first_add)
7694 mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007695}
7696
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007697static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007698{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007699 struct io_rsrc_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007700
7701 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7702 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007703 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007704
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007705 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007706 0, GFP_KERNEL)) {
7707 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007708 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007709 }
7710 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007711 INIT_LIST_HEAD(&ref_node->rsrc_list);
Pavel Begunkove2978222020-11-18 14:56:26 +00007712 ref_node->done = false;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007713 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007714}
7715
Jens Axboe05f3fb32019-12-09 11:22:50 -07007716static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
Pavel Begunkov792e3582021-04-25 14:32:21 +01007717 unsigned nr_args, u64 __user *tags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007718{
7719 __s32 __user *fds = (__s32 __user *) arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007720 struct file *file;
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007721 int fd, ret;
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007722 unsigned i;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007723
7724 if (ctx->file_data)
7725 return -EBUSY;
7726 if (!nr_args)
7727 return -EINVAL;
7728 if (nr_args > IORING_MAX_FIXED_FILES)
7729 return -EMFILE;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007730 ret = io_rsrc_node_switch_start(ctx);
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007731 if (ret)
7732 return ret;
Pavel Begunkovd878c812021-06-14 02:36:18 +01007733 ret = io_rsrc_data_alloc(ctx, io_rsrc_file_put, tags, nr_args,
7734 &ctx->file_data);
7735 if (ret)
7736 return ret;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007737
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007738 ret = -ENOMEM;
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007739 if (!io_alloc_file_tables(&ctx->file_table, nr_args))
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007740 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007741
Jens Axboe05f3fb32019-12-09 11:22:50 -07007742 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Pavel Begunkovd878c812021-06-14 02:36:18 +01007743 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007744 ret = -EFAULT;
7745 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007746 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007747 /* allow sparse sets */
Pavel Begunkov792e3582021-04-25 14:32:21 +01007748 if (fd == -1) {
7749 ret = -EINVAL;
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007750 if (unlikely(*io_get_tag_slot(ctx->file_data, i)))
Pavel Begunkov792e3582021-04-25 14:32:21 +01007751 goto out_fput;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007752 continue;
Pavel Begunkov792e3582021-04-25 14:32:21 +01007753 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007754
Jens Axboe05f3fb32019-12-09 11:22:50 -07007755 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007756 ret = -EBADF;
Pavel Begunkov792e3582021-04-25 14:32:21 +01007757 if (unlikely(!file))
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007758 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007759
7760 /*
7761 * Don't allow io_uring instances to be registered. If UNIX
7762 * isn't enabled, then this causes a reference cycle and this
7763 * instance can never get freed. If UNIX is enabled we'll
7764 * handle it just fine, but there's still no point in allowing
7765 * a ring fd as it doesn't support regular read/write anyway.
7766 */
7767 if (file->f_op == &io_uring_fops) {
7768 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007769 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007770 }
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007771 io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007772 }
7773
Jens Axboe05f3fb32019-12-09 11:22:50 -07007774 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007775 if (ret) {
Pavel Begunkov08480402021-04-13 02:58:38 +01007776 __io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007777 return ret;
7778 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007779
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007780 io_rsrc_node_switch(ctx, NULL);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007781 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007782out_fput:
7783 for (i = 0; i < ctx->nr_user_files; i++) {
7784 file = io_file_from_index(ctx, i);
7785 if (file)
7786 fput(file);
7787 }
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007788 io_free_file_tables(&ctx->file_table, nr_args);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007789 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007790out_free:
Pavel Begunkov44b31f22021-04-25 14:32:16 +01007791 io_rsrc_data_free(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007792 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007793 return ret;
7794}
7795
Jens Axboec3a31e62019-10-03 13:59:56 -06007796static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7797 int index)
7798{
7799#if defined(CONFIG_UNIX)
7800 struct sock *sock = ctx->ring_sock->sk;
7801 struct sk_buff_head *head = &sock->sk_receive_queue;
7802 struct sk_buff *skb;
7803
7804 /*
7805 * See if we can merge this file into an existing skb SCM_RIGHTS
7806 * file set. If there's no room, fall back to allocating a new skb
7807 * and filling it in.
7808 */
7809 spin_lock_irq(&head->lock);
7810 skb = skb_peek(head);
7811 if (skb) {
7812 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7813
7814 if (fpl->count < SCM_MAX_FD) {
7815 __skb_unlink(skb, head);
7816 spin_unlock_irq(&head->lock);
7817 fpl->fp[fpl->count] = get_file(file);
7818 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7819 fpl->count++;
7820 spin_lock_irq(&head->lock);
7821 __skb_queue_head(head, skb);
7822 } else {
7823 skb = NULL;
7824 }
7825 }
7826 spin_unlock_irq(&head->lock);
7827
7828 if (skb) {
7829 fput(file);
7830 return 0;
7831 }
7832
7833 return __io_sqe_files_scm(ctx, 1, index);
7834#else
7835 return 0;
7836#endif
7837}
7838
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007839static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx,
Pavel Begunkove7c78372021-04-01 15:43:45 +01007840 struct io_rsrc_node *node, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007841{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007842 struct io_rsrc_put *prsrc;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007843
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007844 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7845 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007846 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007847
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007848 prsrc->tag = *io_get_tag_slot(data, idx);
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007849 prsrc->rsrc = rsrc;
Pavel Begunkove7c78372021-04-01 15:43:45 +01007850 list_add(&prsrc->list, &node->rsrc_list);
Hillf Dantona5318d32020-03-23 17:47:15 +08007851 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007852}
7853
7854static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01007855 struct io_uring_rsrc_update2 *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07007856 unsigned nr_args)
7857{
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01007858 u64 __user *tags = u64_to_user_ptr(up->tags);
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01007859 __s32 __user *fds = u64_to_user_ptr(up->data);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007860 struct io_rsrc_data *data = ctx->file_data;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007861 struct io_fixed_file *file_slot;
7862 struct file *file;
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01007863 int fd, i, err = 0;
7864 unsigned int done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007865 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007866
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01007867 if (!ctx->file_data)
7868 return -ENXIO;
7869 if (up->offset + nr_args > ctx->nr_user_files)
Jens Axboec3a31e62019-10-03 13:59:56 -06007870 return -EINVAL;
7871
Pavel Begunkov67973b92021-01-26 13:51:09 +00007872 for (done = 0; done < nr_args; done++) {
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01007873 u64 tag = 0;
7874
7875 if ((tags && copy_from_user(&tag, &tags[done], sizeof(tag))) ||
7876 copy_from_user(&fd, &fds[done], sizeof(fd))) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007877 err = -EFAULT;
7878 break;
7879 }
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01007880 if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) {
7881 err = -EINVAL;
7882 break;
7883 }
noah4e0377a2021-01-26 15:23:28 -05007884 if (fd == IORING_REGISTER_FILES_SKIP)
7885 continue;
7886
Pavel Begunkov67973b92021-01-26 13:51:09 +00007887 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007888 file_slot = io_fixed_file_slot(&ctx->file_table, i);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007889
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007890 if (file_slot->file_ptr) {
7891 file = (struct file *)(file_slot->file_ptr & FFS_MASK);
Pavel Begunkovb60c8dc2021-04-25 14:32:18 +01007892 err = io_queue_rsrc_removal(data, up->offset + done,
7893 ctx->rsrc_node, file);
Hillf Dantona5318d32020-03-23 17:47:15 +08007894 if (err)
7895 break;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007896 file_slot->file_ptr = 0;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007897 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007898 }
7899 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007900 file = fget(fd);
7901 if (!file) {
7902 err = -EBADF;
7903 break;
7904 }
7905 /*
7906 * Don't allow io_uring instances to be registered. If
7907 * UNIX isn't enabled, then this causes a reference
7908 * cycle and this instance can never get freed. If UNIX
7909 * is enabled we'll handle it just fine, but there's
7910 * still no point in allowing a ring fd as it doesn't
7911 * support regular read/write anyway.
7912 */
7913 if (file->f_op == &io_uring_fops) {
7914 fput(file);
7915 err = -EBADF;
7916 break;
7917 }
Pavel Begunkov2d091d62021-06-14 02:36:21 +01007918 *io_get_tag_slot(data, up->offset + done) = tag;
Pavel Begunkov9a321c92021-04-01 15:44:01 +01007919 io_fixed_file_set(file_slot, file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007920 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007921 if (err) {
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007922 file_slot->file_ptr = 0;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007923 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007924 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007925 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007926 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007927 }
7928
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007929 if (needs_switch)
7930 io_rsrc_node_switch(ctx, data);
Jens Axboec3a31e62019-10-03 13:59:56 -06007931 return done ? done : err;
7932}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007933
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007934static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007935{
7936 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7937
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007938 req = io_put_req_find_next(req);
7939 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07007940}
7941
Jens Axboe685fe7f2021-03-08 09:37:51 -07007942static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx,
7943 struct task_struct *task)
Pavel Begunkov24369c22020-01-28 03:15:48 +03007944{
Jens Axboee9418942021-02-19 12:33:30 -07007945 struct io_wq_hash *hash;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007946 struct io_wq_data data;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007947 unsigned int concurrency;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007948
Jens Axboee9418942021-02-19 12:33:30 -07007949 hash = ctx->hash_map;
7950 if (!hash) {
7951 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
7952 if (!hash)
7953 return ERR_PTR(-ENOMEM);
7954 refcount_set(&hash->refs, 1);
7955 init_waitqueue_head(&hash->wait);
7956 ctx->hash_map = hash;
7957 }
7958
7959 data.hash = hash;
Jens Axboe685fe7f2021-03-08 09:37:51 -07007960 data.task = task;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007961 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007962 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007963
Jens Axboed25e3a32021-02-16 11:41:41 -07007964 /* Do QD, or 4 * CPUS, whatever is smallest */
7965 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
Pavel Begunkov24369c22020-01-28 03:15:48 +03007966
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007967 return io_wq_create(concurrency, &data);
Pavel Begunkov24369c22020-01-28 03:15:48 +03007968}
7969
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007970static int io_uring_alloc_task_context(struct task_struct *task,
7971 struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06007972{
7973 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007974 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007975
Pavel Begunkov09899b12021-06-14 02:36:22 +01007976 tctx = kzalloc(sizeof(*tctx), GFP_KERNEL);
Jens Axboe0f212202020-09-13 13:09:39 -06007977 if (unlikely(!tctx))
7978 return -ENOMEM;
7979
Jens Axboed8a6df12020-10-15 16:24:45 -06007980 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7981 if (unlikely(ret)) {
7982 kfree(tctx);
7983 return ret;
7984 }
7985
Jens Axboe685fe7f2021-03-08 09:37:51 -07007986 tctx->io_wq = io_init_wq_offload(ctx, task);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007987 if (IS_ERR(tctx->io_wq)) {
7988 ret = PTR_ERR(tctx->io_wq);
7989 percpu_counter_destroy(&tctx->inflight);
7990 kfree(tctx);
7991 return ret;
7992 }
7993
Jens Axboe0f212202020-09-13 13:09:39 -06007994 xa_init(&tctx->xa);
7995 init_waitqueue_head(&tctx->wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06007996 atomic_set(&tctx->in_idle, 0);
Pavel Begunkovb303fe22021-04-11 01:46:26 +01007997 atomic_set(&tctx->inflight_tracked, 0);
Jens Axboe0f212202020-09-13 13:09:39 -06007998 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00007999 spin_lock_init(&tctx->task_lock);
8000 INIT_WQ_LIST(&tctx->task_list);
Jens Axboe7cbf1722021-02-10 00:03:20 +00008001 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06008002 return 0;
8003}
8004
8005void __io_uring_free(struct task_struct *tsk)
8006{
8007 struct io_uring_task *tctx = tsk->io_uring;
8008
8009 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00008010 WARN_ON_ONCE(tctx->io_wq);
Pavel Begunkov09899b12021-06-14 02:36:22 +01008011 WARN_ON_ONCE(tctx->cached_refs);
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00008012
Jens Axboed8a6df12020-10-15 16:24:45 -06008013 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06008014 kfree(tctx);
8015 tsk->io_uring = NULL;
8016}
8017
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02008018static int io_sq_offload_create(struct io_ring_ctx *ctx,
8019 struct io_uring_params *p)
Jens Axboe6b063142019-01-10 22:13:58 -07008020{
8021 int ret;
8022
Jens Axboed25e3a32021-02-16 11:41:41 -07008023 /* Retain compatibility with failing for an invalid attach attempt */
8024 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
8025 IORING_SETUP_ATTACH_WQ) {
8026 struct fd f;
8027
8028 f = fdget(p->wq_fd);
8029 if (!f.file)
8030 return -ENXIO;
Jens Axboed25e3a32021-02-16 11:41:41 -07008031 fdput(f);
Pavel Begunkovf2a48dd2021-04-20 12:03:33 +01008032 if (f.file->f_op != &io_uring_fops)
8033 return -EINVAL;
Jens Axboed25e3a32021-02-16 11:41:41 -07008034 }
Jens Axboe6b063142019-01-10 22:13:58 -07008035 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe46fe18b2021-03-04 12:39:36 -07008036 struct task_struct *tsk;
Jens Axboe534ca6d2020-09-02 13:52:19 -06008037 struct io_sq_data *sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00008038 bool attached;
Jens Axboe534ca6d2020-09-02 13:52:19 -06008039
Pavel Begunkov26984fb2021-03-11 23:29:37 +00008040 sqd = io_get_sq_data(p, &attached);
Jens Axboe534ca6d2020-09-02 13:52:19 -06008041 if (IS_ERR(sqd)) {
8042 ret = PTR_ERR(sqd);
8043 goto err;
8044 }
Jens Axboe69fb2132020-09-14 11:16:23 -06008045
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01008046 ctx->sq_creds = get_current_cred();
Jens Axboe534ca6d2020-09-02 13:52:19 -06008047 ctx->sq_data = sqd;
Jens Axboe6b063142019-01-10 22:13:58 -07008048 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
8049 if (!ctx->sq_thread_idle)
8050 ctx->sq_thread_idle = HZ;
8051
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00008052 io_sq_thread_park(sqd);
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00008053 list_add(&ctx->sqd_list, &sqd->ctx_list);
8054 io_sqd_update_thread_idle(sqd);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00008055 /* don't attach to a dying SQPOLL thread, would be racy */
Pavel Begunkovf2a48dd2021-04-20 12:03:33 +01008056 ret = (attached && !sqd->thread) ? -ENXIO : 0;
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00008057 io_sq_thread_unpark(sqd);
8058
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00008059 if (ret < 0)
8060 goto err;
8061 if (attached)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008062 return 0;
Jens Axboeaa061652020-09-02 14:50:27 -06008063
Jens Axboe6b063142019-01-10 22:13:58 -07008064 if (p->flags & IORING_SETUP_SQ_AFF) {
8065 int cpu = p->sq_thread_cpu;
8066
8067 ret = -EINVAL;
Pavel Begunkovf2a48dd2021-04-20 12:03:33 +01008068 if (cpu >= nr_cpu_ids || !cpu_online(cpu))
Jens Axboee8f98f242021-03-09 16:32:13 -07008069 goto err_sqpoll;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008070 sqd->sq_cpu = cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008071 } else {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008072 sqd->sq_cpu = -1;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008073 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008074
8075 sqd->task_pid = current->pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -07008076 sqd->task_tgid = current->tgid;
Jens Axboe46fe18b2021-03-04 12:39:36 -07008077 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
8078 if (IS_ERR(tsk)) {
8079 ret = PTR_ERR(tsk);
Jens Axboee8f98f242021-03-09 16:32:13 -07008080 goto err_sqpoll;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008081 }
Pavel Begunkov97a73a02021-03-08 17:30:54 +00008082
Jens Axboe46fe18b2021-03-04 12:39:36 -07008083 sqd->thread = tsk;
Pavel Begunkov97a73a02021-03-08 17:30:54 +00008084 ret = io_uring_alloc_task_context(tsk, ctx);
Jens Axboe46fe18b2021-03-04 12:39:36 -07008085 wake_up_new_task(tsk);
Jens Axboe0f212202020-09-13 13:09:39 -06008086 if (ret)
8087 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008088 } else if (p->flags & IORING_SETUP_SQ_AFF) {
8089 /* Can't have SQ_AFF without SQPOLL */
8090 ret = -EINVAL;
8091 goto err;
8092 }
8093
Jens Axboe2b188cc2019-01-07 10:46:33 -07008094 return 0;
Pavel Begunkovf2a48dd2021-04-20 12:03:33 +01008095err_sqpoll:
8096 complete(&ctx->sq_data->exited);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008097err:
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008098 io_sq_thread_finish(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008099 return ret;
8100}
8101
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008102static inline void __io_unaccount_mem(struct user_struct *user,
8103 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008104{
8105 atomic_long_sub(nr_pages, &user->locked_vm);
8106}
8107
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008108static inline int __io_account_mem(struct user_struct *user,
8109 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008110{
8111 unsigned long page_limit, cur_pages, new_pages;
8112
8113 /* Don't allow more pages than we can safely lock */
8114 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
8115
8116 do {
8117 cur_pages = atomic_long_read(&user->locked_vm);
8118 new_pages = cur_pages + nr_pages;
8119 if (new_pages > page_limit)
8120 return -ENOMEM;
8121 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
8122 new_pages) != cur_pages);
8123
8124 return 0;
8125}
8126
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008127static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008128{
Jens Axboe62e398b2021-02-21 16:19:37 -07008129 if (ctx->user)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008130 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008131
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008132 if (ctx->mm_account)
8133 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008134}
8135
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008136static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008137{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008138 int ret;
8139
Jens Axboe62e398b2021-02-21 16:19:37 -07008140 if (ctx->user) {
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008141 ret = __io_account_mem(ctx->user, nr_pages);
8142 if (ret)
8143 return ret;
8144 }
8145
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008146 if (ctx->mm_account)
8147 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008148
8149 return 0;
8150}
8151
Jens Axboe2b188cc2019-01-07 10:46:33 -07008152static void io_mem_free(void *ptr)
8153{
Mark Rutland52e04ef2019-04-30 17:30:21 +01008154 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008155
Mark Rutland52e04ef2019-04-30 17:30:21 +01008156 if (!ptr)
8157 return;
8158
8159 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008160 if (put_page_testzero(page))
8161 free_compound_page(page);
8162}
8163
8164static void *io_mem_alloc(size_t size)
8165{
8166 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008167 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008168
8169 return (void *) __get_free_pages(gfp_flags, get_order(size));
8170}
8171
Hristo Venev75b28af2019-08-26 17:23:46 +00008172static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
8173 size_t *sq_offset)
8174{
8175 struct io_rings *rings;
8176 size_t off, sq_array_size;
8177
8178 off = struct_size(rings, cqes, cq_entries);
8179 if (off == SIZE_MAX)
8180 return SIZE_MAX;
8181
8182#ifdef CONFIG_SMP
8183 off = ALIGN(off, SMP_CACHE_BYTES);
8184 if (off == 0)
8185 return SIZE_MAX;
8186#endif
8187
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02008188 if (sq_offset)
8189 *sq_offset = off;
8190
Hristo Venev75b28af2019-08-26 17:23:46 +00008191 sq_array_size = array_size(sizeof(u32), sq_entries);
8192 if (sq_array_size == SIZE_MAX)
8193 return SIZE_MAX;
8194
8195 if (check_add_overflow(off, sq_array_size, &off))
8196 return SIZE_MAX;
8197
Hristo Venev75b28af2019-08-26 17:23:46 +00008198 return off;
8199}
8200
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008201static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slot)
Pavel Begunkov7f61a1e2021-04-11 01:46:35 +01008202{
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008203 struct io_mapped_ubuf *imu = *slot;
Pavel Begunkov7f61a1e2021-04-11 01:46:35 +01008204 unsigned int i;
8205
Pavel Begunkov62248432021-04-28 13:11:29 +01008206 if (imu != ctx->dummy_ubuf) {
8207 for (i = 0; i < imu->nr_bvecs; i++)
8208 unpin_user_page(imu->bvec[i].bv_page);
8209 if (imu->acct_pages)
8210 io_unaccount_mem(ctx, imu->acct_pages);
8211 kvfree(imu);
8212 }
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008213 *slot = NULL;
Pavel Begunkov7f61a1e2021-04-11 01:46:35 +01008214}
8215
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008216static void io_rsrc_buf_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
8217{
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008218 io_buffer_unmap(ctx, &prsrc->buf);
8219 prsrc->buf = NULL;
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008220}
8221
8222static void __io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07008223{
Pavel Begunkov7f61a1e2021-04-11 01:46:35 +01008224 unsigned int i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008225
Pavel Begunkov7f61a1e2021-04-11 01:46:35 +01008226 for (i = 0; i < ctx->nr_user_bufs; i++)
8227 io_buffer_unmap(ctx, &ctx->user_bufs[i]);
Jens Axboeedafcce2019-01-09 09:16:05 -07008228 kfree(ctx->user_bufs);
Zqiangbb6659c2021-04-30 16:25:15 +08008229 io_rsrc_data_free(ctx->buf_data);
Jens Axboeedafcce2019-01-09 09:16:05 -07008230 ctx->user_bufs = NULL;
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008231 ctx->buf_data = NULL;
Jens Axboeedafcce2019-01-09 09:16:05 -07008232 ctx->nr_user_bufs = 0;
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008233}
8234
Jens Axboeedafcce2019-01-09 09:16:05 -07008235static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
8236{
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008237 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07008238
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008239 if (!ctx->buf_data)
Jens Axboeedafcce2019-01-09 09:16:05 -07008240 return -ENXIO;
8241
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008242 ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx);
8243 if (!ret)
8244 __io_sqe_buffers_unregister(ctx);
8245 return ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07008246}
8247
8248static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8249 void __user *arg, unsigned index)
8250{
8251 struct iovec __user *src;
8252
8253#ifdef CONFIG_COMPAT
8254 if (ctx->compat) {
8255 struct compat_iovec __user *ciovs;
8256 struct compat_iovec ciov;
8257
8258 ciovs = (struct compat_iovec __user *) arg;
8259 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8260 return -EFAULT;
8261
Jens Axboed55e5f52019-12-11 16:12:15 -07008262 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008263 dst->iov_len = ciov.iov_len;
8264 return 0;
8265 }
8266#endif
8267 src = (struct iovec __user *) arg;
8268 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8269 return -EFAULT;
8270 return 0;
8271}
8272
Jens Axboede293932020-09-17 16:19:16 -06008273/*
8274 * Not super efficient, but this is just a registration time. And we do cache
8275 * the last compound head, so generally we'll only do a full search if we don't
8276 * match that one.
8277 *
8278 * We check if the given compound head page has already been accounted, to
8279 * avoid double accounting it. This allows us to account the full size of the
8280 * page, not just the constituent pages of a huge page.
8281 */
8282static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8283 int nr_pages, struct page *hpage)
8284{
8285 int i, j;
8286
8287 /* check current page array */
8288 for (i = 0; i < nr_pages; i++) {
8289 if (!PageCompound(pages[i]))
8290 continue;
8291 if (compound_head(pages[i]) == hpage)
8292 return true;
8293 }
8294
8295 /* check previously registered pages */
8296 for (i = 0; i < ctx->nr_user_bufs; i++) {
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008297 struct io_mapped_ubuf *imu = ctx->user_bufs[i];
Jens Axboede293932020-09-17 16:19:16 -06008298
8299 for (j = 0; j < imu->nr_bvecs; j++) {
8300 if (!PageCompound(imu->bvec[j].bv_page))
8301 continue;
8302 if (compound_head(imu->bvec[j].bv_page) == hpage)
8303 return true;
8304 }
8305 }
8306
8307 return false;
8308}
8309
8310static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8311 int nr_pages, struct io_mapped_ubuf *imu,
8312 struct page **last_hpage)
8313{
8314 int i, ret;
8315
Pavel Begunkov216e5832021-05-29 12:01:02 +01008316 imu->acct_pages = 0;
Jens Axboede293932020-09-17 16:19:16 -06008317 for (i = 0; i < nr_pages; i++) {
8318 if (!PageCompound(pages[i])) {
8319 imu->acct_pages++;
8320 } else {
8321 struct page *hpage;
8322
8323 hpage = compound_head(pages[i]);
8324 if (hpage == *last_hpage)
8325 continue;
8326 *last_hpage = hpage;
8327 if (headpage_already_acct(ctx, pages, i, hpage))
8328 continue;
8329 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8330 }
8331 }
8332
8333 if (!imu->acct_pages)
8334 return 0;
8335
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008336 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008337 if (ret)
8338 imu->acct_pages = 0;
8339 return ret;
8340}
8341
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008342static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008343 struct io_mapped_ubuf **pimu,
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008344 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008345{
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008346 struct io_mapped_ubuf *imu = NULL;
Jens Axboeedafcce2019-01-09 09:16:05 -07008347 struct vm_area_struct **vmas = NULL;
8348 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008349 unsigned long off, start, end, ubuf;
8350 size_t size;
8351 int ret, pret, nr_pages, i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008352
Pavel Begunkov62248432021-04-28 13:11:29 +01008353 if (!iov->iov_base) {
8354 *pimu = ctx->dummy_ubuf;
8355 return 0;
8356 }
8357
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008358 ubuf = (unsigned long) iov->iov_base;
8359 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8360 start = ubuf >> PAGE_SHIFT;
8361 nr_pages = end - start;
8362
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008363 *pimu = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008364 ret = -ENOMEM;
8365
8366 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8367 if (!pages)
8368 goto done;
8369
8370 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8371 GFP_KERNEL);
8372 if (!vmas)
8373 goto done;
8374
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008375 imu = kvmalloc(struct_size(imu, bvec, nr_pages), GFP_KERNEL);
Pavel Begunkova2b41982021-04-26 00:16:31 +01008376 if (!imu)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008377 goto done;
8378
8379 ret = 0;
8380 mmap_read_lock(current->mm);
8381 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8382 pages, vmas);
8383 if (pret == nr_pages) {
8384 /* don't support file backed memory */
8385 for (i = 0; i < nr_pages; i++) {
8386 struct vm_area_struct *vma = vmas[i];
8387
Pavel Begunkov40dad762021-06-09 15:26:54 +01008388 if (vma_is_shmem(vma))
8389 continue;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008390 if (vma->vm_file &&
8391 !is_file_hugepages(vma->vm_file)) {
8392 ret = -EOPNOTSUPP;
8393 break;
8394 }
8395 }
8396 } else {
8397 ret = pret < 0 ? pret : -EFAULT;
8398 }
8399 mmap_read_unlock(current->mm);
8400 if (ret) {
8401 /*
8402 * if we did partial map, or found file backed vmas,
8403 * release any pages we did get
8404 */
8405 if (pret > 0)
8406 unpin_user_pages(pages, pret);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008407 goto done;
8408 }
8409
8410 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8411 if (ret) {
8412 unpin_user_pages(pages, pret);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008413 goto done;
8414 }
8415
8416 off = ubuf & ~PAGE_MASK;
8417 size = iov->iov_len;
8418 for (i = 0; i < nr_pages; i++) {
8419 size_t vec_len;
8420
8421 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8422 imu->bvec[i].bv_page = pages[i];
8423 imu->bvec[i].bv_len = vec_len;
8424 imu->bvec[i].bv_offset = off;
8425 off = 0;
8426 size -= vec_len;
8427 }
8428 /* store original address for later verification */
8429 imu->ubuf = ubuf;
Pavel Begunkov4751f532021-04-01 15:43:55 +01008430 imu->ubuf_end = ubuf + iov->iov_len;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008431 imu->nr_bvecs = nr_pages;
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008432 *pimu = imu;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008433 ret = 0;
8434done:
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008435 if (ret)
8436 kvfree(imu);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008437 kvfree(pages);
8438 kvfree(vmas);
8439 return ret;
8440}
8441
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008442static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008443{
Pavel Begunkov87094462021-04-11 01:46:36 +01008444 ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL);
8445 return ctx->user_bufs ? 0 : -ENOMEM;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008446}
8447
8448static int io_buffer_validate(struct iovec *iov)
8449{
Pavel Begunkov50e96982021-03-24 22:59:01 +00008450 unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1);
8451
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008452 /*
8453 * Don't impose further limits on the size and buffer
8454 * constraints here, we'll -EINVAL later when IO is
8455 * submitted if they are wrong.
8456 */
Pavel Begunkov62248432021-04-28 13:11:29 +01008457 if (!iov->iov_base)
8458 return iov->iov_len ? -EFAULT : 0;
8459 if (!iov->iov_len)
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008460 return -EFAULT;
8461
8462 /* arbitrary limit, but we need something */
8463 if (iov->iov_len > SZ_1G)
8464 return -EFAULT;
8465
Pavel Begunkov50e96982021-03-24 22:59:01 +00008466 if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp))
8467 return -EOVERFLOW;
8468
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008469 return 0;
8470}
8471
8472static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008473 unsigned int nr_args, u64 __user *tags)
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008474{
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008475 struct page *last_hpage = NULL;
8476 struct io_rsrc_data *data;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008477 int i, ret;
8478 struct iovec iov;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008479
Pavel Begunkov87094462021-04-11 01:46:36 +01008480 if (ctx->user_bufs)
8481 return -EBUSY;
Pavel Begunkov489809e2021-05-14 12:06:44 +01008482 if (!nr_args || nr_args > IORING_MAX_REG_BUFFERS)
Pavel Begunkov87094462021-04-11 01:46:36 +01008483 return -EINVAL;
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008484 ret = io_rsrc_node_switch_start(ctx);
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008485 if (ret)
8486 return ret;
Pavel Begunkovd878c812021-06-14 02:36:18 +01008487 ret = io_rsrc_data_alloc(ctx, io_rsrc_buf_put, tags, nr_args, &data);
8488 if (ret)
8489 return ret;
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008490 ret = io_buffers_map_alloc(ctx, nr_args);
8491 if (ret) {
Zqiangbb6659c2021-04-30 16:25:15 +08008492 io_rsrc_data_free(data);
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008493 return ret;
8494 }
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008495
Pavel Begunkov87094462021-04-11 01:46:36 +01008496 for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) {
Jens Axboeedafcce2019-01-09 09:16:05 -07008497 ret = io_copy_iov(ctx, &iov, arg, i);
8498 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008499 break;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008500 ret = io_buffer_validate(&iov);
8501 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008502 break;
Pavel Begunkov2d091d62021-06-14 02:36:21 +01008503 if (!iov.iov_base && *io_get_tag_slot(data, i)) {
Colin Ian Kingcf3770e2021-04-29 11:46:02 +01008504 ret = -EINVAL;
8505 break;
8506 }
Jens Axboeedafcce2019-01-09 09:16:05 -07008507
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01008508 ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i],
8509 &last_hpage);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008510 if (ret)
8511 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008512 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008513
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008514 WARN_ON_ONCE(ctx->buf_data);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008515
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008516 ctx->buf_data = data;
8517 if (ret)
8518 __io_sqe_buffers_unregister(ctx);
8519 else
8520 io_rsrc_node_switch(ctx, NULL);
Jens Axboeedafcce2019-01-09 09:16:05 -07008521 return ret;
8522}
8523
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008524static int __io_sqe_buffers_update(struct io_ring_ctx *ctx,
8525 struct io_uring_rsrc_update2 *up,
8526 unsigned int nr_args)
8527{
8528 u64 __user *tags = u64_to_user_ptr(up->tags);
8529 struct iovec iov, __user *iovs = u64_to_user_ptr(up->data);
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008530 struct page *last_hpage = NULL;
8531 bool needs_switch = false;
8532 __u32 done;
8533 int i, err;
8534
8535 if (!ctx->buf_data)
8536 return -ENXIO;
8537 if (up->offset + nr_args > ctx->nr_user_bufs)
8538 return -EINVAL;
8539
8540 for (done = 0; done < nr_args; done++) {
Pavel Begunkov0b8c0e72021-04-26 15:17:38 +01008541 struct io_mapped_ubuf *imu;
8542 int offset = up->offset + done;
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008543 u64 tag = 0;
8544
8545 err = io_copy_iov(ctx, &iov, iovs, done);
8546 if (err)
8547 break;
8548 if (tags && copy_from_user(&tag, &tags[done], sizeof(tag))) {
8549 err = -EFAULT;
8550 break;
8551 }
Pavel Begunkov0b8c0e72021-04-26 15:17:38 +01008552 err = io_buffer_validate(&iov);
8553 if (err)
8554 break;
Colin Ian Kingcf3770e2021-04-29 11:46:02 +01008555 if (!iov.iov_base && tag) {
8556 err = -EINVAL;
8557 break;
8558 }
Pavel Begunkov0b8c0e72021-04-26 15:17:38 +01008559 err = io_sqe_buffer_register(ctx, &iov, &imu, &last_hpage);
8560 if (err)
8561 break;
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008562
Pavel Begunkov0b8c0e72021-04-26 15:17:38 +01008563 i = array_index_nospec(offset, ctx->nr_user_bufs);
Pavel Begunkov62248432021-04-28 13:11:29 +01008564 if (ctx->user_bufs[i] != ctx->dummy_ubuf) {
Pavel Begunkov0b8c0e72021-04-26 15:17:38 +01008565 err = io_queue_rsrc_removal(ctx->buf_data, offset,
8566 ctx->rsrc_node, ctx->user_bufs[i]);
8567 if (unlikely(err)) {
8568 io_buffer_unmap(ctx, &imu);
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008569 break;
Pavel Begunkov0b8c0e72021-04-26 15:17:38 +01008570 }
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008571 ctx->user_bufs[i] = NULL;
8572 needs_switch = true;
8573 }
8574
Pavel Begunkov0b8c0e72021-04-26 15:17:38 +01008575 ctx->user_bufs[i] = imu;
Pavel Begunkov2d091d62021-06-14 02:36:21 +01008576 *io_get_tag_slot(ctx->buf_data, offset) = tag;
Pavel Begunkov634d00d2021-04-25 14:32:26 +01008577 }
8578
8579 if (needs_switch)
8580 io_rsrc_node_switch(ctx, ctx->buf_data);
8581 return done ? done : err;
8582}
8583
Jens Axboe9b402842019-04-11 11:45:41 -06008584static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8585{
8586 __s32 __user *fds = arg;
8587 int fd;
8588
8589 if (ctx->cq_ev_fd)
8590 return -EBUSY;
8591
8592 if (copy_from_user(&fd, fds, sizeof(*fds)))
8593 return -EFAULT;
8594
8595 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8596 if (IS_ERR(ctx->cq_ev_fd)) {
8597 int ret = PTR_ERR(ctx->cq_ev_fd);
Pavel Begunkovfe7e3252021-06-24 15:09:57 +01008598
Jens Axboe9b402842019-04-11 11:45:41 -06008599 ctx->cq_ev_fd = NULL;
8600 return ret;
8601 }
8602
8603 return 0;
8604}
8605
8606static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8607{
8608 if (ctx->cq_ev_fd) {
8609 eventfd_ctx_put(ctx->cq_ev_fd);
8610 ctx->cq_ev_fd = NULL;
8611 return 0;
8612 }
8613
8614 return -ENXIO;
8615}
8616
Jens Axboe5a2e7452020-02-23 16:23:11 -07008617static void io_destroy_buffers(struct io_ring_ctx *ctx)
8618{
Jens Axboe9e15c3a2021-03-13 12:29:43 -07008619 struct io_buffer *buf;
8620 unsigned long index;
8621
8622 xa_for_each(&ctx->io_buffers, index, buf)
8623 __io_remove_buffers(ctx, buf, index, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008624}
8625
Jens Axboe68e68ee2021-02-13 09:00:02 -07008626static void io_req_cache_free(struct list_head *list, struct task_struct *tsk)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008627{
Jens Axboe68e68ee2021-02-13 09:00:02 -07008628 struct io_kiocb *req, *nxt;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008629
Jens Axboe68e68ee2021-02-13 09:00:02 -07008630 list_for_each_entry_safe(req, nxt, list, compl.list) {
8631 if (tsk && req->task != tsk)
8632 continue;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008633 list_del(&req->compl.list);
8634 kmem_cache_free(req_cachep, req);
8635 }
8636}
8637
Jens Axboe4010fec2021-02-27 15:04:18 -07008638static void io_req_caches_free(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008639{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008640 struct io_submit_state *submit_state = &ctx->submit_state;
Pavel Begunkove5547d22021-02-23 22:17:20 +00008641 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008642
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008643 mutex_lock(&ctx->uring_lock);
8644
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008645 if (submit_state->free_reqs) {
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008646 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8647 submit_state->reqs);
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008648 submit_state->free_reqs = 0;
8649 }
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008650
Pavel Begunkovdac7a092021-03-19 17:22:39 +00008651 io_flush_cached_locked_reqs(ctx, cs);
Pavel Begunkove5547d22021-02-23 22:17:20 +00008652 io_req_cache_free(&cs->free_list, NULL);
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008653 mutex_unlock(&ctx->uring_lock);
8654}
8655
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008656static bool io_wait_rsrc_data(struct io_rsrc_data *data)
8657{
8658 if (!data)
8659 return false;
8660 if (!atomic_dec_and_test(&data->refs))
8661 wait_for_completion(&data->done);
8662 return true;
8663}
8664
Jens Axboe2b188cc2019-01-07 10:46:33 -07008665static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8666{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008667 io_sq_thread_finish(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008668
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008669 if (ctx->mm_account) {
Jens Axboe2aede0e2020-09-14 10:45:53 -06008670 mmdrop(ctx->mm_account);
8671 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008672 }
Jens Axboedef596e2019-01-09 08:59:42 -07008673
Hao Xu8bad28d2021-02-19 17:19:36 +08008674 mutex_lock(&ctx->uring_lock);
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +01008675 if (io_wait_rsrc_data(ctx->buf_data))
8676 __io_sqe_buffers_unregister(ctx);
8677 if (io_wait_rsrc_data(ctx->file_data))
Pavel Begunkov08480402021-04-13 02:58:38 +01008678 __io_sqe_files_unregister(ctx);
Pavel Begunkovc4ea0602021-04-01 15:43:58 +01008679 if (ctx->rings)
8680 __io_cqring_overflow_flush(ctx, true);
Hao Xu8bad28d2021-02-19 17:19:36 +08008681 mutex_unlock(&ctx->uring_lock);
Jens Axboe9b402842019-04-11 11:45:41 -06008682 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008683 io_destroy_buffers(ctx);
Pavel Begunkov07db2982021-04-20 12:03:32 +01008684 if (ctx->sq_creds)
8685 put_cred(ctx->sq_creds);
Jens Axboedef596e2019-01-09 08:59:42 -07008686
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01008687 /* there are no registered resources left, nobody uses it */
8688 if (ctx->rsrc_node)
8689 io_rsrc_node_destroy(ctx->rsrc_node);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00008690 if (ctx->rsrc_backup_node)
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01008691 io_rsrc_node_destroy(ctx->rsrc_backup_node);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01008692 flush_delayed_work(&ctx->rsrc_put_work);
8693
8694 WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list));
8695 WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist));
Jens Axboe2b188cc2019-01-07 10:46:33 -07008696
8697#if defined(CONFIG_UNIX)
Eric Biggers355e8d22019-06-12 14:58:43 -07008698 if (ctx->ring_sock) {
8699 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008700 sock_release(ctx->ring_sock);
Eric Biggers355e8d22019-06-12 14:58:43 -07008701 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008702#endif
8703
Hristo Venev75b28af2019-08-26 17:23:46 +00008704 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008705 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008706
8707 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008708 free_uid(ctx->user);
Jens Axboe4010fec2021-02-27 15:04:18 -07008709 io_req_caches_free(ctx);
Jens Axboee9418942021-02-19 12:33:30 -07008710 if (ctx->hash_map)
8711 io_wq_put_hash(ctx->hash_map);
Jens Axboe78076bb2019-12-04 19:56:40 -07008712 kfree(ctx->cancel_hash);
Pavel Begunkov62248432021-04-28 13:11:29 +01008713 kfree(ctx->dummy_ubuf);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008714 kfree(ctx);
8715}
8716
8717static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8718{
8719 struct io_ring_ctx *ctx = file->private_data;
8720 __poll_t mask = 0;
8721
Pavel Begunkov311997b2021-06-14 23:37:28 +01008722 poll_wait(file, &ctx->poll_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008723 /*
8724 * synchronizes with barrier from wq_has_sleeper call in
8725 * io_commit_cqring
8726 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008727 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008728 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008729 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008730
8731 /*
8732 * Don't flush cqring overflow list here, just do a simple check.
8733 * Otherwise there could possible be ABBA deadlock:
8734 * CPU0 CPU1
8735 * ---- ----
8736 * lock(&ctx->uring_lock);
8737 * lock(&ep->mtx);
8738 * lock(&ctx->uring_lock);
8739 * lock(&ep->mtx);
8740 *
8741 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8742 * pushs them to do the flush.
8743 */
Pavel Begunkov5ed7a372021-06-14 23:37:27 +01008744 if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008745 mask |= EPOLLIN | EPOLLRDNORM;
8746
8747 return mask;
8748}
8749
8750static int io_uring_fasync(int fd, struct file *file, int on)
8751{
8752 struct io_ring_ctx *ctx = file->private_data;
8753
8754 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8755}
8756
Yejune Deng0bead8c2020-12-24 11:02:20 +08008757static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008758{
Jens Axboe4379bf82021-02-15 13:40:22 -07008759 const struct cred *creds;
Jens Axboe071698e2020-01-28 10:04:42 -07008760
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008761 creds = xa_erase(&ctx->personalities, id);
Jens Axboe4379bf82021-02-15 13:40:22 -07008762 if (creds) {
8763 put_cred(creds);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008764 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008765 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008766
8767 return -EINVAL;
8768}
8769
Pavel Begunkov9b465712021-03-15 14:23:07 +00008770static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008771{
Pavel Begunkov9b465712021-03-15 14:23:07 +00008772 return io_run_task_work_head(&ctx->exit_task_work);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008773}
8774
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008775struct io_tctx_exit {
8776 struct callback_head task_work;
8777 struct completion completion;
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008778 struct io_ring_ctx *ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008779};
8780
8781static void io_tctx_exit_cb(struct callback_head *cb)
8782{
8783 struct io_uring_task *tctx = current->io_uring;
8784 struct io_tctx_exit *work;
8785
8786 work = container_of(cb, struct io_tctx_exit, task_work);
8787 /*
8788 * When @in_idle, we're in cancellation and it's racy to remove the
8789 * node. It'll be removed by the end of cancellation, just ignore it.
8790 */
8791 if (!atomic_read(&tctx->in_idle))
Pavel Begunkoveef51da2021-06-14 02:36:15 +01008792 io_uring_del_tctx_node((unsigned long)work->ctx);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008793 complete(&work->completion);
8794}
8795
Pavel Begunkov28090c12021-04-25 23:34:45 +01008796static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data)
8797{
8798 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8799
8800 return req->ctx == data;
8801}
8802
Jens Axboe85faa7b2020-04-09 18:14:00 -06008803static void io_ring_exit_work(struct work_struct *work)
8804{
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008805 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008806 unsigned long timeout = jiffies + HZ * 60 * 5;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008807 struct io_tctx_exit exit;
8808 struct io_tctx_node *node;
8809 int ret;
Jens Axboe85faa7b2020-04-09 18:14:00 -06008810
Jens Axboe56952e92020-06-17 15:00:04 -06008811 /*
8812 * If we're doing polled IO and end up having requests being
8813 * submitted async (out-of-line), then completions can come in while
8814 * we're waiting for refs to drop. We need to reap these manually,
8815 * as nobody else will be looking for them.
8816 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008817 do {
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008818 io_uring_try_cancel_requests(ctx, NULL, true);
Pavel Begunkov28090c12021-04-25 23:34:45 +01008819 if (ctx->sq_data) {
8820 struct io_sq_data *sqd = ctx->sq_data;
8821 struct task_struct *tsk;
8822
8823 io_sq_thread_park(sqd);
8824 tsk = sqd->thread;
8825 if (tsk && tsk->io_uring && tsk->io_uring->io_wq)
8826 io_wq_cancel_cb(tsk->io_uring->io_wq,
8827 io_cancel_ctx_cb, ctx, true);
8828 io_sq_thread_unpark(sqd);
8829 }
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008830
8831 WARN_ON_ONCE(time_after(jiffies, timeout));
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008832 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008833
Pavel Begunkov7f006512021-04-14 13:38:34 +01008834 init_completion(&exit.completion);
8835 init_task_work(&exit.task_work, io_tctx_exit_cb);
8836 exit.ctx = ctx;
Pavel Begunkov89b50662021-04-01 15:43:50 +01008837 /*
8838 * Some may use context even when all refs and requests have been put,
8839 * and they are free to do so while still holding uring_lock or
8840 * completion_lock, see __io_req_task_submit(). Apart from other work,
8841 * this lock/unlock section also waits them to finish.
8842 */
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008843 mutex_lock(&ctx->uring_lock);
8844 while (!list_empty(&ctx->tctx_list)) {
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008845 WARN_ON_ONCE(time_after(jiffies, timeout));
8846
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008847 node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
8848 ctx_node);
Pavel Begunkov7f006512021-04-14 13:38:34 +01008849 /* don't spin on a single task if cancellation failed */
8850 list_rotate_left(&ctx->tctx_list);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008851 ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
8852 if (WARN_ON_ONCE(ret))
8853 continue;
8854 wake_up_process(node->task);
8855
8856 mutex_unlock(&ctx->uring_lock);
8857 wait_for_completion(&exit.completion);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008858 mutex_lock(&ctx->uring_lock);
8859 }
8860 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov89b50662021-04-01 15:43:50 +01008861 spin_lock_irq(&ctx->completion_lock);
8862 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008863
Jens Axboe85faa7b2020-04-09 18:14:00 -06008864 io_ring_ctx_free(ctx);
8865}
8866
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008867/* Returns true if we found and killed one or more timeouts */
8868static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008869 bool cancel_all)
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008870{
8871 struct io_kiocb *req, *tmp;
8872 int canceled = 0;
8873
8874 spin_lock_irq(&ctx->completion_lock);
8875 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008876 if (io_match_task(req, tsk, cancel_all)) {
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008877 io_kill_timeout(req, -ECANCELED);
8878 canceled++;
8879 }
8880 }
Pavel Begunkov51520422021-03-29 11:39:29 +01008881 if (canceled != 0)
8882 io_commit_cqring(ctx);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008883 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008884 if (canceled != 0)
8885 io_cqring_ev_posted(ctx);
8886 return canceled != 0;
8887}
8888
Jens Axboe2b188cc2019-01-07 10:46:33 -07008889static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8890{
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008891 unsigned long index;
8892 struct creds *creds;
8893
Jens Axboe2b188cc2019-01-07 10:46:33 -07008894 mutex_lock(&ctx->uring_lock);
8895 percpu_ref_kill(&ctx->refs);
Pavel Begunkov634578f2020-12-06 22:22:44 +00008896 if (ctx->rings)
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00008897 __io_cqring_overflow_flush(ctx, true);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008898 xa_for_each(&ctx->personalities, index, creds)
8899 io_unregister_personality(ctx, index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008900 mutex_unlock(&ctx->uring_lock);
8901
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008902 io_kill_timeouts(ctx, NULL, true);
8903 io_poll_remove_all(ctx, NULL, true);
Jens Axboe561fb042019-10-24 07:25:42 -06008904
Jens Axboe15dff282019-11-13 09:09:23 -07008905 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008906 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008907
Jens Axboe85faa7b2020-04-09 18:14:00 -06008908 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008909 /*
8910 * Use system_unbound_wq to avoid spawning tons of event kworkers
8911 * if we're exiting a ton of rings at the same time. It just adds
8912 * noise and overhead, there's no discernable change in runtime
8913 * over using system_wq.
8914 */
8915 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008916}
8917
8918static int io_uring_release(struct inode *inode, struct file *file)
8919{
8920 struct io_ring_ctx *ctx = file->private_data;
8921
8922 file->private_data = NULL;
8923 io_ring_ctx_wait_and_kill(ctx);
8924 return 0;
8925}
8926
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008927struct io_task_cancel {
8928 struct task_struct *task;
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008929 bool all;
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008930};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008931
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008932static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008933{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008934 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008935 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008936 bool ret;
8937
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008938 if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008939 unsigned long flags;
8940 struct io_ring_ctx *ctx = req->ctx;
8941
8942 /* protect against races with linked timeouts */
8943 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008944 ret = io_match_task(req, cancel->task, cancel->all);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008945 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8946 } else {
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008947 ret = io_match_task(req, cancel->task, cancel->all);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008948 }
8949 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008950}
8951
Pavel Begunkove1915f72021-03-11 23:29:35 +00008952static bool io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008953 struct task_struct *task, bool cancel_all)
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008954{
Pavel Begunkove1915f72021-03-11 23:29:35 +00008955 struct io_defer_entry *de;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008956 LIST_HEAD(list);
8957
8958 spin_lock_irq(&ctx->completion_lock);
8959 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01008960 if (io_match_task(de->req, task, cancel_all)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008961 list_cut_position(&list, &ctx->defer_list, &de->list);
8962 break;
8963 }
8964 }
8965 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkove1915f72021-03-11 23:29:35 +00008966 if (list_empty(&list))
8967 return false;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008968
8969 while (!list_empty(&list)) {
8970 de = list_first_entry(&list, struct io_defer_entry, list);
8971 list_del_init(&de->list);
Pavel Begunkovf41db2732021-02-28 22:35:12 +00008972 io_req_complete_failed(de->req, -ECANCELED);
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008973 kfree(de);
8974 }
Pavel Begunkove1915f72021-03-11 23:29:35 +00008975 return true;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008976}
8977
Pavel Begunkov1b007642021-03-06 11:02:17 +00008978static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
8979{
8980 struct io_tctx_node *node;
8981 enum io_wq_cancel cret;
8982 bool ret = false;
8983
8984 mutex_lock(&ctx->uring_lock);
8985 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
8986 struct io_uring_task *tctx = node->task->io_uring;
8987
8988 /*
8989 * io_wq will stay alive while we hold uring_lock, because it's
8990 * killed after ctx nodes, which requires to take the lock.
8991 */
8992 if (!tctx || !tctx->io_wq)
8993 continue;
8994 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true);
8995 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8996 }
8997 mutex_unlock(&ctx->uring_lock);
8998
8999 return ret;
9000}
9001
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00009002static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
9003 struct task_struct *task,
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01009004 bool cancel_all)
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00009005{
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01009006 struct io_task_cancel cancel = { .task = task, .all = cancel_all, };
Pavel Begunkov1b007642021-03-06 11:02:17 +00009007 struct io_uring_task *tctx = task ? task->io_uring : NULL;
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00009008
9009 while (1) {
9010 enum io_wq_cancel cret;
9011 bool ret = false;
9012
Pavel Begunkov1b007642021-03-06 11:02:17 +00009013 if (!task) {
9014 ret |= io_uring_try_cancel_iowq(ctx);
9015 } else if (tctx && tctx->io_wq) {
9016 /*
9017 * Cancels requests of all rings, not only @ctx, but
9018 * it's fine as the task is in exit/exec.
9019 */
Jens Axboe5aa75ed2021-02-16 12:56:50 -07009020 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00009021 &cancel, true);
9022 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
9023 }
9024
9025 /* SQPOLL thread does its own polling */
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01009026 if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) ||
Jens Axboed052d1d2021-03-11 10:49:20 -07009027 (ctx->sq_data && ctx->sq_data->thread == current)) {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00009028 while (!list_empty_careful(&ctx->iopoll_list)) {
9029 io_iopoll_try_reap_events(ctx);
9030 ret = true;
9031 }
9032 }
9033
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01009034 ret |= io_cancel_defer_files(ctx, task, cancel_all);
9035 ret |= io_poll_remove_all(ctx, task, cancel_all);
9036 ret |= io_kill_timeouts(ctx, task, cancel_all);
Pavel Begunkove5dc4802021-06-26 21:40:46 +01009037 if (task)
9038 ret |= io_run_task_work();
Pavel Begunkovba50a032021-02-26 15:47:56 +00009039 ret |= io_run_ctx_fallback(ctx);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00009040 if (!ret)
9041 break;
9042 cond_resched();
9043 }
9044}
9045
Pavel Begunkoveef51da2021-06-14 02:36:15 +01009046static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06009047{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009048 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00009049 struct io_tctx_node *node;
Pavel Begunkova528b042020-12-21 18:34:04 +00009050 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009051
9052 if (unlikely(!tctx)) {
Jens Axboe5aa75ed2021-02-16 12:56:50 -07009053 ret = io_uring_alloc_task_context(current, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06009054 if (unlikely(ret))
9055 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009056 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06009057 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00009058 if (!xa_load(&tctx->xa, (unsigned long)ctx)) {
9059 node = kmalloc(sizeof(*node), GFP_KERNEL);
9060 if (!node)
9061 return -ENOMEM;
9062 node->ctx = ctx;
9063 node->task = current;
Jens Axboe0f212202020-09-13 13:09:39 -06009064
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00009065 ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx,
9066 node, GFP_KERNEL));
9067 if (ret) {
9068 kfree(node);
9069 return ret;
Jens Axboe0f212202020-09-13 13:09:39 -06009070 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00009071
9072 mutex_lock(&ctx->uring_lock);
9073 list_add(&node->ctx_node, &ctx->tctx_list);
9074 mutex_unlock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009075 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00009076 tctx->last = ctx;
Jens Axboe0f212202020-09-13 13:09:39 -06009077 return 0;
9078}
9079
9080/*
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00009081 * Note that this task has used io_uring. We use it for cancelation purposes.
9082 */
Pavel Begunkoveef51da2021-06-14 02:36:15 +01009083static inline int io_uring_add_tctx_node(struct io_ring_ctx *ctx)
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00009084{
9085 struct io_uring_task *tctx = current->io_uring;
9086
9087 if (likely(tctx && tctx->last == ctx))
9088 return 0;
Pavel Begunkoveef51da2021-06-14 02:36:15 +01009089 return __io_uring_add_tctx_node(ctx);
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00009090}
9091
9092/*
Jens Axboe0f212202020-09-13 13:09:39 -06009093 * Remove this io_uring_file -> task mapping.
9094 */
Pavel Begunkoveef51da2021-06-14 02:36:15 +01009095static void io_uring_del_tctx_node(unsigned long index)
Jens Axboe0f212202020-09-13 13:09:39 -06009096{
9097 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00009098 struct io_tctx_node *node;
Pavel Begunkov29412672021-03-06 11:02:11 +00009099
Pavel Begunkoveebd2e32021-03-06 11:02:14 +00009100 if (!tctx)
9101 return;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00009102 node = xa_erase(&tctx->xa, index);
9103 if (!node)
Pavel Begunkov29412672021-03-06 11:02:11 +00009104 return;
Jens Axboe0f212202020-09-13 13:09:39 -06009105
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00009106 WARN_ON_ONCE(current != node->task);
9107 WARN_ON_ONCE(list_empty(&node->ctx_node));
9108
9109 mutex_lock(&node->ctx->uring_lock);
9110 list_del(&node->ctx_node);
9111 mutex_unlock(&node->ctx->uring_lock);
9112
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009113 if (tctx->last == node->ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06009114 tctx->last = NULL;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00009115 kfree(node);
Jens Axboe0f212202020-09-13 13:09:39 -06009116}
9117
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00009118static void io_uring_clean_tctx(struct io_uring_task *tctx)
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009119{
Pavel Begunkovba5ef6d2021-05-20 13:21:20 +01009120 struct io_wq *wq = tctx->io_wq;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00009121 struct io_tctx_node *node;
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009122 unsigned long index;
9123
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00009124 xa_for_each(&tctx->xa, index, node)
Pavel Begunkoveef51da2021-06-14 02:36:15 +01009125 io_uring_del_tctx_node(index);
Marco Elverb16ef422021-05-27 11:25:48 +02009126 if (wq) {
9127 /*
9128 * Must be after io_uring_del_task_file() (removes nodes under
9129 * uring_lock) to avoid race with io_uring_try_cancel_iowq().
9130 */
9131 tctx->io_wq = NULL;
Pavel Begunkovba5ef6d2021-05-20 13:21:20 +01009132 io_wq_put_and_exit(wq);
Marco Elverb16ef422021-05-27 11:25:48 +02009133 }
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009134}
9135
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009136static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked)
Pavel Begunkov521d6a72021-03-11 23:29:38 +00009137{
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009138 if (tracked)
9139 return atomic_read(&tctx->inflight_tracked);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00009140 return percpu_counter_sum(&tctx->inflight);
9141}
9142
Pavel Begunkov09899b12021-06-14 02:36:22 +01009143static void io_uring_drop_tctx_refs(struct task_struct *task)
9144{
9145 struct io_uring_task *tctx = task->io_uring;
9146 unsigned int refs = tctx->cached_refs;
9147
9148 tctx->cached_refs = 0;
9149 percpu_counter_sub(&tctx->inflight, refs);
9150 put_task_struct_many(task, refs);
9151}
9152
Pavel Begunkov78cc6872021-06-14 02:36:23 +01009153/*
9154 * Find any io_uring ctx that this task has registered or done IO on, and cancel
9155 * requests. @sqd should be not-null IIF it's an SQPOLL thread cancellation.
9156 */
9157static void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd)
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009158{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00009159 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov734551d2021-04-18 14:52:09 +01009160 struct io_ring_ctx *ctx;
Jens Axboefdaf0832020-10-30 09:37:30 -06009161 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009162 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06009163
Pavel Begunkov78cc6872021-06-14 02:36:23 +01009164 WARN_ON_ONCE(sqd && sqd->thread != current);
9165
Palash Oswal6d042ff2021-04-27 18:21:49 +05309166 if (!current->io_uring)
9167 return;
Pavel Begunkov17a91052021-05-23 15:48:39 +01009168 if (tctx->io_wq)
9169 io_wq_exit_start(tctx->io_wq);
9170
Pavel Begunkov09899b12021-06-14 02:36:22 +01009171 io_uring_drop_tctx_refs(current);
Jens Axboefdaf0832020-10-30 09:37:30 -06009172 atomic_inc(&tctx->in_idle);
Jens Axboed8a6df12020-10-15 16:24:45 -06009173 do {
Jens Axboe0f212202020-09-13 13:09:39 -06009174 /* read completions before cancelations */
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01009175 inflight = tctx_inflight(tctx, !cancel_all);
Jens Axboed8a6df12020-10-15 16:24:45 -06009176 if (!inflight)
9177 break;
Jens Axboe0f212202020-09-13 13:09:39 -06009178
Pavel Begunkov78cc6872021-06-14 02:36:23 +01009179 if (!sqd) {
9180 struct io_tctx_node *node;
9181 unsigned long index;
9182
9183 xa_for_each(&tctx->xa, index, node) {
9184 /* sqpoll task will cancel all its requests */
9185 if (node->ctx->sq_data)
9186 continue;
9187 io_uring_try_cancel_requests(node->ctx, current,
9188 cancel_all);
9189 }
9190 } else {
9191 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
9192 io_uring_try_cancel_requests(ctx, current,
9193 cancel_all);
9194 }
9195
9196 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
Jens Axboe0f212202020-09-13 13:09:39 -06009197 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009198 * If we've seen completions, retry without waiting. This
9199 * avoids a race where a completion comes in before we did
9200 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06009201 */
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01009202 if (inflight == tctx_inflight(tctx, !cancel_all))
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009203 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00009204 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06009205 } while (1);
Jens Axboefdaf0832020-10-30 09:37:30 -06009206 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009207
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00009208 io_uring_clean_tctx(tctx);
Pavel Begunkov3dd0c972021-05-16 22:58:04 +01009209 if (cancel_all) {
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009210 /* for exec all current's requests should be gone, kill tctx */
9211 __io_uring_free(current);
9212 }
Pavel Begunkov44e728b2020-06-15 10:24:04 +03009213}
9214
Pavel Begunkov78cc6872021-06-14 02:36:23 +01009215void __io_uring_cancel(struct files_struct *files)
9216{
9217 io_uring_cancel_generic(!files, NULL);
9218}
9219
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009220static void *io_uring_validate_mmap_request(struct file *file,
9221 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009222{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009223 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009224 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009225 struct page *page;
9226 void *ptr;
9227
9228 switch (offset) {
9229 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00009230 case IORING_OFF_CQ_RING:
9231 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009232 break;
9233 case IORING_OFF_SQES:
9234 ptr = ctx->sq_sqes;
9235 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009236 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009237 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009238 }
9239
9240 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07009241 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009242 return ERR_PTR(-EINVAL);
9243
9244 return ptr;
9245}
9246
9247#ifdef CONFIG_MMU
9248
9249static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9250{
9251 size_t sz = vma->vm_end - vma->vm_start;
9252 unsigned long pfn;
9253 void *ptr;
9254
9255 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
9256 if (IS_ERR(ptr))
9257 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009258
9259 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
9260 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
9261}
9262
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009263#else /* !CONFIG_MMU */
9264
9265static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9266{
9267 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
9268}
9269
9270static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
9271{
9272 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
9273}
9274
9275static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
9276 unsigned long addr, unsigned long len,
9277 unsigned long pgoff, unsigned long flags)
9278{
9279 void *ptr;
9280
9281 ptr = io_uring_validate_mmap_request(file, pgoff, len);
9282 if (IS_ERR(ptr))
9283 return PTR_ERR(ptr);
9284
9285 return (unsigned long) ptr;
9286}
9287
9288#endif /* !CONFIG_MMU */
9289
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009290static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06009291{
9292 DEFINE_WAIT(wait);
9293
9294 do {
9295 if (!io_sqring_full(ctx))
9296 break;
Jens Axboe90554202020-09-03 12:12:41 -06009297 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
9298
9299 if (!io_sqring_full(ctx))
9300 break;
Jens Axboe90554202020-09-03 12:12:41 -06009301 schedule();
9302 } while (!signal_pending(current));
9303
9304 finish_wait(&ctx->sqo_sq_wait, &wait);
Yang Li51993282021-03-09 14:30:41 +08009305 return 0;
Jens Axboe90554202020-09-03 12:12:41 -06009306}
9307
Hao Xuc73ebb62020-11-03 10:54:37 +08009308static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9309 struct __kernel_timespec __user **ts,
9310 const sigset_t __user **sig)
9311{
9312 struct io_uring_getevents_arg arg;
9313
9314 /*
9315 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9316 * is just a pointer to the sigset_t.
9317 */
9318 if (!(flags & IORING_ENTER_EXT_ARG)) {
9319 *sig = (const sigset_t __user *) argp;
9320 *ts = NULL;
9321 return 0;
9322 }
9323
9324 /*
9325 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9326 * timespec and sigset_t pointers if good.
9327 */
9328 if (*argsz != sizeof(arg))
9329 return -EINVAL;
9330 if (copy_from_user(&arg, argp, sizeof(arg)))
9331 return -EFAULT;
9332 *sig = u64_to_user_ptr(arg.sigmask);
9333 *argsz = arg.sigmask_sz;
9334 *ts = u64_to_user_ptr(arg.ts);
9335 return 0;
9336}
9337
Jens Axboe2b188cc2019-01-07 10:46:33 -07009338SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009339 u32, min_complete, u32, flags, const void __user *, argp,
9340 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009341{
9342 struct io_ring_ctx *ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009343 int submitted = 0;
9344 struct fd f;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009345 long ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009346
Jens Axboe4c6e2772020-07-01 11:29:10 -06009347 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009348
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009349 if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
9350 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009351 return -EINVAL;
9352
9353 f = fdget(fd);
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009354 if (unlikely(!f.file))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009355 return -EBADF;
9356
9357 ret = -EOPNOTSUPP;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009358 if (unlikely(f.file->f_op != &io_uring_fops))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009359 goto out_fput;
9360
9361 ret = -ENXIO;
9362 ctx = f.file->private_data;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009363 if (unlikely(!percpu_ref_tryget(&ctx->refs)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009364 goto out_fput;
9365
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009366 ret = -EBADFD;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009367 if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED))
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009368 goto out;
9369
Jens Axboe6c271ce2019-01-10 11:22:30 -07009370 /*
9371 * For SQ polling, the thread will do all submissions and completions.
9372 * Just return the requested submit count, and wake the thread if
9373 * we were asked to.
9374 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009375 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009376 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00009377 io_cqring_overflow_flush(ctx, false);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009378
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009379 ret = -EOWNERDEAD;
Pavel Begunkovfe7e3252021-06-24 15:09:57 +01009380 if (unlikely(ctx->sq_data->thread == NULL))
Stefan Metzmacher04147482021-03-07 11:54:29 +01009381 goto out;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009382 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009383 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009384 if (flags & IORING_ENTER_SQ_WAIT) {
9385 ret = io_sqpoll_wait_sq(ctx);
9386 if (ret)
9387 goto out;
9388 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009389 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009390 } else if (to_submit) {
Pavel Begunkoveef51da2021-06-14 02:36:15 +01009391 ret = io_uring_add_tctx_node(ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06009392 if (unlikely(ret))
9393 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009394 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009395 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009396 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009397
9398 if (submitted != to_submit)
9399 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009400 }
9401 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009402 const sigset_t __user *sig;
9403 struct __kernel_timespec __user *ts;
9404
9405 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9406 if (unlikely(ret))
9407 goto out;
9408
Jens Axboe2b188cc2019-01-07 10:46:33 -07009409 min_complete = min(min_complete, ctx->cq_entries);
9410
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009411 /*
9412 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9413 * space applications don't need to do io completion events
9414 * polling again, they can rely on io_sq_thread to do polling
9415 * work, which can reduce cpu usage and uring_lock contention.
9416 */
9417 if (ctx->flags & IORING_SETUP_IOPOLL &&
9418 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009419 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009420 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009421 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009422 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009423 }
9424
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009425out:
Pavel Begunkov6805b322019-10-08 02:18:42 +03009426 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009427out_fput:
9428 fdput(f);
9429 return submitted ? submitted : ret;
9430}
9431
Tobias Klauserbebdb652020-02-26 18:38:32 +01009432#ifdef CONFIG_PROC_FS
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009433static int io_uring_show_cred(struct seq_file *m, unsigned int id,
9434 const struct cred *cred)
Jens Axboe87ce9552020-01-30 08:25:34 -07009435{
Jens Axboe87ce9552020-01-30 08:25:34 -07009436 struct user_namespace *uns = seq_user_ns(m);
9437 struct group_info *gi;
9438 kernel_cap_t cap;
9439 unsigned __capi;
9440 int g;
9441
9442 seq_printf(m, "%5d\n", id);
9443 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9444 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9445 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9446 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9447 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9448 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9449 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9450 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9451 seq_puts(m, "\n\tGroups:\t");
9452 gi = cred->group_info;
9453 for (g = 0; g < gi->ngroups; g++) {
9454 seq_put_decimal_ull(m, g ? " " : "",
9455 from_kgid_munged(uns, gi->gid[g]));
9456 }
9457 seq_puts(m, "\n\tCapEff:\t");
9458 cap = cred->cap_effective;
9459 CAP_FOR_EACH_U32(__capi)
9460 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9461 seq_putc(m, '\n');
9462 return 0;
9463}
9464
9465static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9466{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009467 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009468 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009469 int i;
9470
Jens Axboefad8e0d2020-09-28 08:57:48 -06009471 /*
9472 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9473 * since fdinfo case grabs it in the opposite direction of normal use
9474 * cases. If we fail to get the lock, we just don't iterate any
9475 * structures that could be going away outside the io_uring mutex.
9476 */
9477 has_lock = mutex_trylock(&ctx->uring_lock);
9478
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009479 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
Joseph Qidbbe9c62020-09-29 09:01:22 -06009480 sq = ctx->sq_data;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009481 if (!sq->thread)
9482 sq = NULL;
9483 }
Joseph Qidbbe9c62020-09-29 09:01:22 -06009484
9485 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9486 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009487 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009488 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Jens Axboe7b29f922021-03-12 08:30:14 -07009489 struct file *f = io_file_from_index(ctx, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009490
Jens Axboe87ce9552020-01-30 08:25:34 -07009491 if (f)
9492 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9493 else
9494 seq_printf(m, "%5u: <none>\n", i);
9495 }
9496 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009497 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Pavel Begunkov41edf1a2021-04-25 14:32:23 +01009498 struct io_mapped_ubuf *buf = ctx->user_bufs[i];
Pavel Begunkov4751f532021-04-01 15:43:55 +01009499 unsigned int len = buf->ubuf_end - buf->ubuf;
Jens Axboe87ce9552020-01-30 08:25:34 -07009500
Pavel Begunkov4751f532021-04-01 15:43:55 +01009501 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len);
Jens Axboe87ce9552020-01-30 08:25:34 -07009502 }
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009503 if (has_lock && !xa_empty(&ctx->personalities)) {
9504 unsigned long index;
9505 const struct cred *cred;
9506
Jens Axboe87ce9552020-01-30 08:25:34 -07009507 seq_printf(m, "Personalities:\n");
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009508 xa_for_each(&ctx->personalities, index, cred)
9509 io_uring_show_cred(m, index, cred);
Jens Axboe87ce9552020-01-30 08:25:34 -07009510 }
Jens Axboed7718a92020-02-14 22:23:12 -07009511 seq_printf(m, "PollList:\n");
9512 spin_lock_irq(&ctx->completion_lock);
9513 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9514 struct hlist_head *list = &ctx->cancel_hash[i];
9515 struct io_kiocb *req;
9516
9517 hlist_for_each_entry(req, list, hash_node)
9518 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9519 req->task->task_works != NULL);
9520 }
9521 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009522 if (has_lock)
9523 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009524}
9525
9526static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9527{
9528 struct io_ring_ctx *ctx = f->private_data;
9529
9530 if (percpu_ref_tryget(&ctx->refs)) {
9531 __io_uring_show_fdinfo(ctx, m);
9532 percpu_ref_put(&ctx->refs);
9533 }
9534}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009535#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009536
Jens Axboe2b188cc2019-01-07 10:46:33 -07009537static const struct file_operations io_uring_fops = {
9538 .release = io_uring_release,
9539 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009540#ifndef CONFIG_MMU
9541 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9542 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9543#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009544 .poll = io_uring_poll,
9545 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009546#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009547 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009548#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009549};
9550
9551static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9552 struct io_uring_params *p)
9553{
Hristo Venev75b28af2019-08-26 17:23:46 +00009554 struct io_rings *rings;
9555 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009556
Jens Axboebd740482020-08-05 12:58:23 -06009557 /* make sure these are sane, as we already accounted them */
9558 ctx->sq_entries = p->sq_entries;
9559 ctx->cq_entries = p->cq_entries;
9560
Hristo Venev75b28af2019-08-26 17:23:46 +00009561 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9562 if (size == SIZE_MAX)
9563 return -EOVERFLOW;
9564
9565 rings = io_mem_alloc(size);
9566 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009567 return -ENOMEM;
9568
Hristo Venev75b28af2019-08-26 17:23:46 +00009569 ctx->rings = rings;
9570 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9571 rings->sq_ring_mask = p->sq_entries - 1;
9572 rings->cq_ring_mask = p->cq_entries - 1;
9573 rings->sq_ring_entries = p->sq_entries;
9574 rings->cq_ring_entries = p->cq_entries;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009575
9576 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009577 if (size == SIZE_MAX) {
9578 io_mem_free(ctx->rings);
9579 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009580 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009581 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009582
9583 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009584 if (!ctx->sq_sqes) {
9585 io_mem_free(ctx->rings);
9586 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009587 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009588 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009589
Jens Axboe2b188cc2019-01-07 10:46:33 -07009590 return 0;
9591}
9592
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009593static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9594{
9595 int ret, fd;
9596
9597 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9598 if (fd < 0)
9599 return fd;
9600
Pavel Begunkoveef51da2021-06-14 02:36:15 +01009601 ret = io_uring_add_tctx_node(ctx);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009602 if (ret) {
9603 put_unused_fd(fd);
9604 return ret;
9605 }
9606 fd_install(fd, file);
9607 return fd;
9608}
9609
Jens Axboe2b188cc2019-01-07 10:46:33 -07009610/*
9611 * Allocate an anonymous fd, this is what constitutes the application
9612 * visible backing of an io_uring instance. The application mmaps this
9613 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9614 * we have to tie this fd to a socket for file garbage collection purposes.
9615 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009616static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009617{
9618 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009619#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009620 int ret;
9621
Jens Axboe2b188cc2019-01-07 10:46:33 -07009622 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9623 &ctx->ring_sock);
9624 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009625 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009626#endif
9627
Jens Axboe2b188cc2019-01-07 10:46:33 -07009628 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9629 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009630#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009631 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009632 sock_release(ctx->ring_sock);
9633 ctx->ring_sock = NULL;
9634 } else {
9635 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009636 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009637#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009638 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009639}
9640
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009641static int io_uring_create(unsigned entries, struct io_uring_params *p,
9642 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009643{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009644 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009645 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009646 int ret;
9647
Jens Axboe8110c1a2019-12-28 15:39:54 -07009648 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009649 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009650 if (entries > IORING_MAX_ENTRIES) {
9651 if (!(p->flags & IORING_SETUP_CLAMP))
9652 return -EINVAL;
9653 entries = IORING_MAX_ENTRIES;
9654 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009655
9656 /*
9657 * Use twice as many entries for the CQ ring. It's possible for the
9658 * application to drive a higher depth than the size of the SQ ring,
9659 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009660 * some flexibility in overcommitting a bit. If the application has
9661 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9662 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009663 */
9664 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009665 if (p->flags & IORING_SETUP_CQSIZE) {
9666 /*
9667 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9668 * to a power-of-two, if it isn't already. We do NOT impose
9669 * any cq vs sq ring sizing.
9670 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009671 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009672 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009673 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9674 if (!(p->flags & IORING_SETUP_CLAMP))
9675 return -EINVAL;
9676 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9677 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009678 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9679 if (p->cq_entries < p->sq_entries)
9680 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009681 } else {
9682 p->cq_entries = 2 * p->sq_entries;
9683 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009684
Jens Axboe2b188cc2019-01-07 10:46:33 -07009685 ctx = io_ring_ctx_alloc(p);
Jens Axboe62e398b2021-02-21 16:19:37 -07009686 if (!ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009687 return -ENOMEM;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009688 ctx->compat = in_compat_syscall();
Jens Axboe62e398b2021-02-21 16:19:37 -07009689 if (!capable(CAP_IPC_LOCK))
9690 ctx->user = get_uid(current_user());
Jens Axboe2aede0e2020-09-14 10:45:53 -06009691
9692 /*
9693 * This is just grabbed for accounting purposes. When a process exits,
9694 * the mm is exited and dropped before the files, hence we need to hang
9695 * on to this mm purely for the purposes of being able to unaccount
9696 * memory (locked/pinned vm). It's not used for anything else.
9697 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009698 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009699 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009700
Jens Axboe2b188cc2019-01-07 10:46:33 -07009701 ret = io_allocate_scq_urings(ctx, p);
9702 if (ret)
9703 goto err;
9704
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009705 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009706 if (ret)
9707 goto err;
Pavel Begunkoveae071c2021-04-25 14:32:24 +01009708 /* always set a rsrc node */
Pavel Begunkov47b228c2021-04-29 11:46:48 +01009709 ret = io_rsrc_node_switch_start(ctx);
9710 if (ret)
9711 goto err;
Pavel Begunkoveae071c2021-04-25 14:32:24 +01009712 io_rsrc_node_switch(ctx, NULL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009713
Jens Axboe2b188cc2019-01-07 10:46:33 -07009714 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009715 p->sq_off.head = offsetof(struct io_rings, sq.head);
9716 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9717 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9718 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9719 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9720 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9721 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009722
9723 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009724 p->cq_off.head = offsetof(struct io_rings, cq.head);
9725 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9726 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9727 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9728 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9729 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009730 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009731
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009732 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9733 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009734 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009735 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
Pavel Begunkov96905572021-06-10 16:37:38 +01009736 IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS |
9737 IORING_FEAT_RSRC_TAGS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009738
9739 if (copy_to_user(params, p, sizeof(*p))) {
9740 ret = -EFAULT;
9741 goto err;
9742 }
Jens Axboed1719f72020-07-30 13:43:53 -06009743
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009744 file = io_uring_get_file(ctx);
9745 if (IS_ERR(file)) {
9746 ret = PTR_ERR(file);
9747 goto err;
9748 }
9749
Jens Axboed1719f72020-07-30 13:43:53 -06009750 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009751 * Install ring fd as the very last thing, so we don't risk someone
9752 * having closed it before we finish setup
9753 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009754 ret = io_uring_install_fd(ctx, file);
9755 if (ret < 0) {
9756 /* fput will clean it up */
9757 fput(file);
9758 return ret;
9759 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009760
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009761 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009762 return ret;
9763err:
9764 io_ring_ctx_wait_and_kill(ctx);
9765 return ret;
9766}
9767
9768/*
9769 * Sets up an aio uring context, and returns the fd. Applications asks for a
9770 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9771 * params structure passed in.
9772 */
9773static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9774{
9775 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009776 int i;
9777
9778 if (copy_from_user(&p, params, sizeof(p)))
9779 return -EFAULT;
9780 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9781 if (p.resv[i])
9782 return -EINVAL;
9783 }
9784
Jens Axboe6c271ce2019-01-10 11:22:30 -07009785 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009786 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009787 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9788 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009789 return -EINVAL;
9790
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009791 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009792}
9793
9794SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9795 struct io_uring_params __user *, params)
9796{
9797 return io_uring_setup(entries, params);
9798}
9799
Jens Axboe66f4af92020-01-16 15:36:52 -07009800static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9801{
9802 struct io_uring_probe *p;
9803 size_t size;
9804 int i, ret;
9805
9806 size = struct_size(p, ops, nr_args);
9807 if (size == SIZE_MAX)
9808 return -EOVERFLOW;
9809 p = kzalloc(size, GFP_KERNEL);
9810 if (!p)
9811 return -ENOMEM;
9812
9813 ret = -EFAULT;
9814 if (copy_from_user(p, arg, size))
9815 goto out;
9816 ret = -EINVAL;
9817 if (memchr_inv(p, 0, size))
9818 goto out;
9819
9820 p->last_op = IORING_OP_LAST - 1;
9821 if (nr_args > IORING_OP_LAST)
9822 nr_args = IORING_OP_LAST;
9823
9824 for (i = 0; i < nr_args; i++) {
9825 p->ops[i].op = i;
9826 if (!io_op_defs[i].not_supported)
9827 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9828 }
9829 p->ops_len = i;
9830
9831 ret = 0;
9832 if (copy_to_user(arg, p, size))
9833 ret = -EFAULT;
9834out:
9835 kfree(p);
9836 return ret;
9837}
9838
Jens Axboe071698e2020-01-28 10:04:42 -07009839static int io_register_personality(struct io_ring_ctx *ctx)
9840{
Jens Axboe4379bf82021-02-15 13:40:22 -07009841 const struct cred *creds;
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009842 u32 id;
Jens Axboe1e6fa522020-10-15 08:46:24 -06009843 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009844
Jens Axboe4379bf82021-02-15 13:40:22 -07009845 creds = get_current_cred();
Jens Axboe1e6fa522020-10-15 08:46:24 -06009846
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009847 ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds,
9848 XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL);
9849 if (!ret)
9850 return id;
9851 put_cred(creds);
Jens Axboe1e6fa522020-10-15 08:46:24 -06009852 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009853}
9854
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009855static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9856 unsigned int nr_args)
9857{
9858 struct io_uring_restriction *res;
9859 size_t size;
9860 int i, ret;
9861
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009862 /* Restrictions allowed only if rings started disabled */
9863 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9864 return -EBADFD;
9865
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009866 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009867 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009868 return -EBUSY;
9869
9870 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9871 return -EINVAL;
9872
9873 size = array_size(nr_args, sizeof(*res));
9874 if (size == SIZE_MAX)
9875 return -EOVERFLOW;
9876
9877 res = memdup_user(arg, size);
9878 if (IS_ERR(res))
9879 return PTR_ERR(res);
9880
9881 ret = 0;
9882
9883 for (i = 0; i < nr_args; i++) {
9884 switch (res[i].opcode) {
9885 case IORING_RESTRICTION_REGISTER_OP:
9886 if (res[i].register_op >= IORING_REGISTER_LAST) {
9887 ret = -EINVAL;
9888 goto out;
9889 }
9890
9891 __set_bit(res[i].register_op,
9892 ctx->restrictions.register_op);
9893 break;
9894 case IORING_RESTRICTION_SQE_OP:
9895 if (res[i].sqe_op >= IORING_OP_LAST) {
9896 ret = -EINVAL;
9897 goto out;
9898 }
9899
9900 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9901 break;
9902 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9903 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9904 break;
9905 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9906 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9907 break;
9908 default:
9909 ret = -EINVAL;
9910 goto out;
9911 }
9912 }
9913
9914out:
9915 /* Reset all restrictions if an error happened */
9916 if (ret != 0)
9917 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9918 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009919 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009920
9921 kfree(res);
9922 return ret;
9923}
9924
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009925static int io_register_enable_rings(struct io_ring_ctx *ctx)
9926{
9927 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9928 return -EBADFD;
9929
9930 if (ctx->restrictions.registered)
9931 ctx->restricted = 1;
9932
Pavel Begunkov0298ef92021-03-08 13:20:57 +00009933 ctx->flags &= ~IORING_SETUP_R_DISABLED;
9934 if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait))
9935 wake_up(&ctx->sq_data->wait);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009936 return 0;
9937}
9938
Pavel Begunkovfdecb662021-04-25 14:32:20 +01009939static int __io_register_rsrc_update(struct io_ring_ctx *ctx, unsigned type,
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01009940 struct io_uring_rsrc_update2 *up,
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009941 unsigned nr_args)
9942{
9943 __u32 tmp;
9944 int err;
9945
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01009946 if (up->resv)
9947 return -EINVAL;
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009948 if (check_add_overflow(up->offset, nr_args, &tmp))
9949 return -EOVERFLOW;
9950 err = io_rsrc_node_switch_start(ctx);
9951 if (err)
9952 return err;
9953
Pavel Begunkovfdecb662021-04-25 14:32:20 +01009954 switch (type) {
9955 case IORING_RSRC_FILE:
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009956 return __io_sqe_files_update(ctx, up, nr_args);
Pavel Begunkov634d00d2021-04-25 14:32:26 +01009957 case IORING_RSRC_BUFFER:
9958 return __io_sqe_buffers_update(ctx, up, nr_args);
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009959 }
9960 return -EINVAL;
9961}
9962
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01009963static int io_register_files_update(struct io_ring_ctx *ctx, void __user *arg,
9964 unsigned nr_args)
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009965{
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01009966 struct io_uring_rsrc_update2 up;
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009967
9968 if (!nr_args)
9969 return -EINVAL;
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01009970 memset(&up, 0, sizeof(up));
9971 if (copy_from_user(&up, arg, sizeof(struct io_uring_rsrc_update)))
9972 return -EFAULT;
9973 return __io_register_rsrc_update(ctx, IORING_RSRC_FILE, &up, nr_args);
9974}
9975
9976static int io_register_rsrc_update(struct io_ring_ctx *ctx, void __user *arg,
Pavel Begunkov992da012021-06-10 16:37:37 +01009977 unsigned size, unsigned type)
Pavel Begunkovc3bdad02021-04-25 14:32:22 +01009978{
9979 struct io_uring_rsrc_update2 up;
9980
9981 if (size != sizeof(up))
9982 return -EINVAL;
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009983 if (copy_from_user(&up, arg, sizeof(up)))
9984 return -EFAULT;
Pavel Begunkov992da012021-06-10 16:37:37 +01009985 if (!up.nr || up.resv)
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009986 return -EINVAL;
Pavel Begunkov992da012021-06-10 16:37:37 +01009987 return __io_register_rsrc_update(ctx, type, &up, up.nr);
Pavel Begunkov98f0b3b2021-04-25 14:32:19 +01009988}
9989
Pavel Begunkov792e3582021-04-25 14:32:21 +01009990static int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg,
Pavel Begunkov992da012021-06-10 16:37:37 +01009991 unsigned int size, unsigned int type)
Pavel Begunkov792e3582021-04-25 14:32:21 +01009992{
9993 struct io_uring_rsrc_register rr;
9994
9995 /* keep it extendible */
9996 if (size != sizeof(rr))
9997 return -EINVAL;
9998
9999 memset(&rr, 0, sizeof(rr));
10000 if (copy_from_user(&rr, arg, size))
10001 return -EFAULT;
Pavel Begunkov992da012021-06-10 16:37:37 +010010002 if (!rr.nr || rr.resv || rr.resv2)
Pavel Begunkov792e3582021-04-25 14:32:21 +010010003 return -EINVAL;
10004
Pavel Begunkov992da012021-06-10 16:37:37 +010010005 switch (type) {
Pavel Begunkov792e3582021-04-25 14:32:21 +010010006 case IORING_RSRC_FILE:
10007 return io_sqe_files_register(ctx, u64_to_user_ptr(rr.data),
10008 rr.nr, u64_to_user_ptr(rr.tags));
Pavel Begunkov634d00d2021-04-25 14:32:26 +010010009 case IORING_RSRC_BUFFER:
10010 return io_sqe_buffers_register(ctx, u64_to_user_ptr(rr.data),
10011 rr.nr, u64_to_user_ptr(rr.tags));
Pavel Begunkov792e3582021-04-25 14:32:21 +010010012 }
10013 return -EINVAL;
10014}
10015
Jens Axboefe764212021-06-17 10:19:54 -060010016static int io_register_iowq_aff(struct io_ring_ctx *ctx, void __user *arg,
10017 unsigned len)
10018{
10019 struct io_uring_task *tctx = current->io_uring;
10020 cpumask_var_t new_mask;
10021 int ret;
10022
10023 if (!tctx || !tctx->io_wq)
10024 return -EINVAL;
10025
10026 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
10027 return -ENOMEM;
10028
10029 cpumask_clear(new_mask);
10030 if (len > cpumask_size())
10031 len = cpumask_size();
10032
10033 if (copy_from_user(new_mask, arg, len)) {
10034 free_cpumask_var(new_mask);
10035 return -EFAULT;
10036 }
10037
10038 ret = io_wq_cpu_affinity(tctx->io_wq, new_mask);
10039 free_cpumask_var(new_mask);
10040 return ret;
10041}
10042
10043static int io_unregister_iowq_aff(struct io_ring_ctx *ctx)
10044{
10045 struct io_uring_task *tctx = current->io_uring;
10046
10047 if (!tctx || !tctx->io_wq)
10048 return -EINVAL;
10049
10050 return io_wq_cpu_affinity(tctx->io_wq, NULL);
10051}
10052
Jens Axboe071698e2020-01-28 10:04:42 -070010053static bool io_register_op_must_quiesce(int op)
10054{
10055 switch (op) {
Bijan Mottahedehbd54b6f2021-04-25 14:32:25 +010010056 case IORING_REGISTER_BUFFERS:
10057 case IORING_UNREGISTER_BUFFERS:
Pavel Begunkovf4f7d212021-04-01 15:44:02 +010010058 case IORING_REGISTER_FILES:
Jens Axboe071698e2020-01-28 10:04:42 -070010059 case IORING_UNREGISTER_FILES:
10060 case IORING_REGISTER_FILES_UPDATE:
10061 case IORING_REGISTER_PROBE:
10062 case IORING_REGISTER_PERSONALITY:
10063 case IORING_UNREGISTER_PERSONALITY:
Pavel Begunkov992da012021-06-10 16:37:37 +010010064 case IORING_REGISTER_FILES2:
10065 case IORING_REGISTER_FILES_UPDATE2:
10066 case IORING_REGISTER_BUFFERS2:
10067 case IORING_REGISTER_BUFFERS_UPDATE:
Jens Axboefe764212021-06-17 10:19:54 -060010068 case IORING_REGISTER_IOWQ_AFF:
10069 case IORING_UNREGISTER_IOWQ_AFF:
Jens Axboe071698e2020-01-28 10:04:42 -070010070 return false;
10071 default:
10072 return true;
10073 }
10074}
10075
Jens Axboeedafcce2019-01-09 09:16:05 -070010076static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
10077 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -060010078 __releases(ctx->uring_lock)
10079 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -070010080{
10081 int ret;
10082
Jens Axboe35fa71a2019-04-22 10:23:23 -060010083 /*
10084 * We're inside the ring mutex, if the ref is already dying, then
10085 * someone else killed the ctx or is already going through
10086 * io_uring_register().
10087 */
10088 if (percpu_ref_is_dying(&ctx->refs))
10089 return -ENXIO;
10090
Pavel Begunkov75c40212021-04-15 13:07:40 +010010091 if (ctx->restricted) {
10092 if (opcode >= IORING_REGISTER_LAST)
10093 return -EINVAL;
10094 opcode = array_index_nospec(opcode, IORING_REGISTER_LAST);
10095 if (!test_bit(opcode, ctx->restrictions.register_op))
10096 return -EACCES;
10097 }
10098
Jens Axboe071698e2020-01-28 10:04:42 -070010099 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -070010100 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -060010101
Jens Axboe05f3fb32019-12-09 11:22:50 -070010102 /*
10103 * Drop uring mutex before waiting for references to exit. If
10104 * another thread is currently inside io_uring_enter() it might
10105 * need to grab the uring_lock to make progress. If we hold it
10106 * here across the drain wait, then we can deadlock. It's safe
10107 * to drop the mutex here, since no new references will come in
10108 * after we've killed the percpu ref.
10109 */
10110 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -060010111 do {
10112 ret = wait_for_completion_interruptible(&ctx->ref_comp);
10113 if (!ret)
10114 break;
Jens Axboeed6930c2020-10-08 19:09:46 -060010115 ret = io_run_task_work_sig();
10116 if (ret < 0)
10117 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -060010118 } while (1);
Jens Axboe05f3fb32019-12-09 11:22:50 -070010119 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -060010120
Jens Axboec1503682020-01-08 08:26:07 -070010121 if (ret) {
Pavel Begunkovf70865d2021-04-11 01:46:40 +010010122 io_refs_resurrect(&ctx->refs, &ctx->ref_comp);
10123 return ret;
Jens Axboec1503682020-01-08 08:26:07 -070010124 }
Jens Axboe05f3fb32019-12-09 11:22:50 -070010125 }
Jens Axboeedafcce2019-01-09 09:16:05 -070010126
10127 switch (opcode) {
10128 case IORING_REGISTER_BUFFERS:
Pavel Begunkov634d00d2021-04-25 14:32:26 +010010129 ret = io_sqe_buffers_register(ctx, arg, nr_args, NULL);
Jens Axboeedafcce2019-01-09 09:16:05 -070010130 break;
10131 case IORING_UNREGISTER_BUFFERS:
10132 ret = -EINVAL;
10133 if (arg || nr_args)
10134 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -080010135 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -070010136 break;
Jens Axboe6b063142019-01-10 22:13:58 -070010137 case IORING_REGISTER_FILES:
Pavel Begunkov792e3582021-04-25 14:32:21 +010010138 ret = io_sqe_files_register(ctx, arg, nr_args, NULL);
Jens Axboe6b063142019-01-10 22:13:58 -070010139 break;
10140 case IORING_UNREGISTER_FILES:
10141 ret = -EINVAL;
10142 if (arg || nr_args)
10143 break;
10144 ret = io_sqe_files_unregister(ctx);
10145 break;
Jens Axboec3a31e62019-10-03 13:59:56 -060010146 case IORING_REGISTER_FILES_UPDATE:
Pavel Begunkovc3bdad02021-04-25 14:32:22 +010010147 ret = io_register_files_update(ctx, arg, nr_args);
Jens Axboec3a31e62019-10-03 13:59:56 -060010148 break;
Jens Axboe9b402842019-04-11 11:45:41 -060010149 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -070010150 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -060010151 ret = -EINVAL;
10152 if (nr_args != 1)
10153 break;
10154 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -070010155 if (ret)
10156 break;
10157 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
10158 ctx->eventfd_async = 1;
10159 else
10160 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -060010161 break;
10162 case IORING_UNREGISTER_EVENTFD:
10163 ret = -EINVAL;
10164 if (arg || nr_args)
10165 break;
10166 ret = io_eventfd_unregister(ctx);
10167 break;
Jens Axboe66f4af92020-01-16 15:36:52 -070010168 case IORING_REGISTER_PROBE:
10169 ret = -EINVAL;
10170 if (!arg || nr_args > 256)
10171 break;
10172 ret = io_probe(ctx, arg, nr_args);
10173 break;
Jens Axboe071698e2020-01-28 10:04:42 -070010174 case IORING_REGISTER_PERSONALITY:
10175 ret = -EINVAL;
10176 if (arg || nr_args)
10177 break;
10178 ret = io_register_personality(ctx);
10179 break;
10180 case IORING_UNREGISTER_PERSONALITY:
10181 ret = -EINVAL;
10182 if (arg)
10183 break;
10184 ret = io_unregister_personality(ctx, nr_args);
10185 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +020010186 case IORING_REGISTER_ENABLE_RINGS:
10187 ret = -EINVAL;
10188 if (arg || nr_args)
10189 break;
10190 ret = io_register_enable_rings(ctx);
10191 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +020010192 case IORING_REGISTER_RESTRICTIONS:
10193 ret = io_register_restrictions(ctx, arg, nr_args);
10194 break;
Pavel Begunkov992da012021-06-10 16:37:37 +010010195 case IORING_REGISTER_FILES2:
10196 ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_FILE);
Pavel Begunkov792e3582021-04-25 14:32:21 +010010197 break;
Pavel Begunkov992da012021-06-10 16:37:37 +010010198 case IORING_REGISTER_FILES_UPDATE2:
10199 ret = io_register_rsrc_update(ctx, arg, nr_args,
10200 IORING_RSRC_FILE);
10201 break;
10202 case IORING_REGISTER_BUFFERS2:
10203 ret = io_register_rsrc(ctx, arg, nr_args, IORING_RSRC_BUFFER);
10204 break;
10205 case IORING_REGISTER_BUFFERS_UPDATE:
10206 ret = io_register_rsrc_update(ctx, arg, nr_args,
10207 IORING_RSRC_BUFFER);
Pavel Begunkovc3bdad02021-04-25 14:32:22 +010010208 break;
Jens Axboefe764212021-06-17 10:19:54 -060010209 case IORING_REGISTER_IOWQ_AFF:
10210 ret = -EINVAL;
10211 if (!arg || !nr_args)
10212 break;
10213 ret = io_register_iowq_aff(ctx, arg, nr_args);
10214 break;
10215 case IORING_UNREGISTER_IOWQ_AFF:
10216 ret = -EINVAL;
10217 if (arg || nr_args)
10218 break;
10219 ret = io_unregister_iowq_aff(ctx);
10220 break;
Jens Axboeedafcce2019-01-09 09:16:05 -070010221 default:
10222 ret = -EINVAL;
10223 break;
10224 }
10225
Jens Axboe071698e2020-01-28 10:04:42 -070010226 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -070010227 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -070010228 percpu_ref_reinit(&ctx->refs);
Jens Axboe0f158b42020-05-14 17:18:39 -060010229 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -070010230 }
Jens Axboeedafcce2019-01-09 09:16:05 -070010231 return ret;
10232}
10233
10234SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
10235 void __user *, arg, unsigned int, nr_args)
10236{
10237 struct io_ring_ctx *ctx;
10238 long ret = -EBADF;
10239 struct fd f;
10240
10241 f = fdget(fd);
10242 if (!f.file)
10243 return -EBADF;
10244
10245 ret = -EOPNOTSUPP;
10246 if (f.file->f_op != &io_uring_fops)
10247 goto out_fput;
10248
10249 ctx = f.file->private_data;
10250
Pavel Begunkovb6c23dd2021-02-20 15:17:18 +000010251 io_run_task_work();
10252
Jens Axboeedafcce2019-01-09 09:16:05 -070010253 mutex_lock(&ctx->uring_lock);
10254 ret = __io_uring_register(ctx, opcode, arg, nr_args);
10255 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020010256 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
10257 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -070010258out_fput:
10259 fdput(f);
10260 return ret;
10261}
10262
Jens Axboe2b188cc2019-01-07 10:46:33 -070010263static int __init io_uring_init(void)
10264{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010265#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
10266 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
10267 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
10268} while (0)
10269
10270#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
10271 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
10272 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
10273 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
10274 BUILD_BUG_SQE_ELEM(1, __u8, flags);
10275 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
10276 BUILD_BUG_SQE_ELEM(4, __s32, fd);
10277 BUILD_BUG_SQE_ELEM(8, __u64, off);
10278 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
10279 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +030010280 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010281 BUILD_BUG_SQE_ELEM(24, __u32, len);
10282 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
10283 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
10284 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
10285 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +080010286 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
10287 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010288 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
10289 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
10290 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
10291 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
10292 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
10293 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
10294 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
10295 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +030010296 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010297 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
10298 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
Pavel Begunkov16340ea2021-06-24 15:09:58 +010010299 BUILD_BUG_SQE_ELEM(40, __u16, buf_group);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010300 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +030010301 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010302
Pavel Begunkovb0d658ec2021-04-27 16:13:53 +010010303 BUILD_BUG_ON(sizeof(struct io_uring_files_update) !=
10304 sizeof(struct io_uring_rsrc_update));
10305 BUILD_BUG_ON(sizeof(struct io_uring_rsrc_update) >
10306 sizeof(struct io_uring_rsrc_update2));
10307 /* should fit into one byte */
10308 BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8));
10309
Jens Axboed3656342019-12-18 09:50:26 -070010310 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -070010311 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Pavel Begunkov16340ea2021-06-24 15:09:58 +010010312
Jens Axboe91f245d2021-02-09 13:48:50 -070010313 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
10314 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -070010315 return 0;
10316};
10317__initcall(io_uring_init);