blob: 472eab7359f2a26c9d3c60e6be5e4653089f9c81 [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
14 * through a control-dependency in io_get_cqring (smp_store_release to
15 * 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 Axboee4b4a132021-03-01 18:36:25 -070081#include <linux/freezer.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070082
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020083#define CREATE_TRACE_POINTS
84#include <trace/events/io_uring.h>
85
Jens Axboe2b188cc2019-01-07 10:46:33 -070086#include <uapi/linux/io_uring.h>
87
88#include "internal.h"
Jens Axboe561fb042019-10-24 07:25:42 -060089#include "io-wq.h"
Jens Axboe2b188cc2019-01-07 10:46:33 -070090
Daniel Xu5277dea2019-09-14 14:23:45 -070091#define IORING_MAX_ENTRIES 32768
Jens Axboe33a107f2019-10-04 12:10:03 -060092#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
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 Begunkovb16fed662021-02-18 18:29:40 +0000104#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
105 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
106 IOSQE_BUFFER_SELECT)
107
Jens Axboe2b188cc2019-01-07 10:46:33 -0700108struct io_uring {
109 u32 head ____cacheline_aligned_in_smp;
110 u32 tail ____cacheline_aligned_in_smp;
111};
112
Stefan Bühler1e84b972019-04-24 23:54:16 +0200113/*
Hristo Venev75b28af2019-08-26 17:23:46 +0000114 * This data is shared with the application through the mmap at offsets
115 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200116 *
117 * The offsets to the member fields are published through struct
118 * io_sqring_offsets when calling io_uring_setup.
119 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000120struct io_rings {
Stefan Bühler1e84b972019-04-24 23:54:16 +0200121 /*
122 * Head and tail offsets into the ring; the offsets need to be
123 * masked to get valid indices.
124 *
Hristo Venev75b28af2019-08-26 17:23:46 +0000125 * The kernel controls head of the sq ring and the tail of the cq ring,
126 * and the application controls tail of the sq ring and the head of the
127 * cq ring.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200128 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000129 struct io_uring sq, cq;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200130 /*
Hristo Venev75b28af2019-08-26 17:23:46 +0000131 * Bitmasks to apply to head and tail offsets (constant, equals
Stefan Bühler1e84b972019-04-24 23:54:16 +0200132 * ring_entries - 1)
133 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000134 u32 sq_ring_mask, cq_ring_mask;
135 /* Ring sizes (constant, power of 2) */
136 u32 sq_ring_entries, cq_ring_entries;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200137 /*
138 * Number of invalid entries dropped by the kernel due to
139 * invalid index stored in array
140 *
141 * Written by the kernel, shouldn't be modified by the
142 * application (i.e. get number of "new events" by comparing to
143 * cached value).
144 *
145 * After a new SQ head value was read by the application this
146 * counter includes all submissions that were dropped reaching
147 * the new SQ head (and possibly more).
148 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000149 u32 sq_dropped;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200150 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200151 * Runtime SQ flags
Stefan Bühler1e84b972019-04-24 23:54:16 +0200152 *
153 * Written by the kernel, shouldn't be modified by the
154 * application.
155 *
156 * The application needs a full memory barrier before checking
157 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
158 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000159 u32 sq_flags;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200160 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200161 * Runtime CQ flags
162 *
163 * Written by the application, shouldn't be modified by the
164 * kernel.
165 */
166 u32 cq_flags;
167 /*
Stefan Bühler1e84b972019-04-24 23:54:16 +0200168 * Number of completion events lost because the queue was full;
169 * this should be avoided by the application by making sure
LimingWu0b4295b2019-12-05 20:18:18 +0800170 * there are not more requests pending than there is space in
Stefan Bühler1e84b972019-04-24 23:54:16 +0200171 * the completion queue.
172 *
173 * Written by the kernel, shouldn't be modified by the
174 * application (i.e. get number of "new events" by comparing to
175 * cached value).
176 *
177 * As completion events come in out of order this counter is not
178 * ordered with any other data.
179 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000180 u32 cq_overflow;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200181 /*
182 * Ring buffer of completion events.
183 *
184 * The kernel writes completion events fresh every time they are
185 * produced, so the application is allowed to modify pending
186 * entries.
187 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000188 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700189};
190
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000191enum io_uring_cmd_flags {
192 IO_URING_F_NONBLOCK = 1,
Pavel Begunkov889fca72021-02-10 00:03:09 +0000193 IO_URING_F_COMPLETE_DEFER = 2,
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000194};
195
Jens Axboeedafcce2019-01-09 09:16:05 -0700196struct io_mapped_ubuf {
197 u64 ubuf;
198 size_t len;
199 struct bio_vec *bvec;
200 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600201 unsigned long acct_pages;
Jens Axboeedafcce2019-01-09 09:16:05 -0700202};
203
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000204struct io_ring_ctx;
205
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000206struct io_rsrc_put {
207 struct list_head list;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000208 union {
209 void *rsrc;
210 struct file *file;
211 };
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000212};
213
214struct fixed_rsrc_table {
Jens Axboe65e19f52019-10-26 07:20:21 -0600215 struct file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700216};
217
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000218struct fixed_rsrc_ref_node {
Xiaoguang Wang05589552020-03-31 14:05:18 +0800219 struct percpu_ref refs;
220 struct list_head node;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000221 struct list_head rsrc_list;
222 struct fixed_rsrc_data *rsrc_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000223 void (*rsrc_put)(struct io_ring_ctx *ctx,
224 struct io_rsrc_put *prsrc);
Jens Axboe4a38aed22020-05-14 17:21:15 -0600225 struct llist_node llist;
Pavel Begunkove2978222020-11-18 14:56:26 +0000226 bool done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800227};
228
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000229struct fixed_rsrc_data {
230 struct fixed_rsrc_table *table;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700231 struct io_ring_ctx *ctx;
232
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000233 struct fixed_rsrc_ref_node *node;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700234 struct percpu_ref refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700235 struct completion done;
Hao Xu8bad28d2021-02-19 17:19:36 +0800236 bool quiesce;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700237};
238
Jens Axboe5a2e7452020-02-23 16:23:11 -0700239struct io_buffer {
240 struct list_head list;
241 __u64 addr;
242 __s32 len;
243 __u16 bid;
244};
245
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200246struct io_restriction {
247 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
248 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
249 u8 sqe_flags_allowed;
250 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200251 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200252};
253
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700254enum {
255 IO_SQ_THREAD_SHOULD_STOP = 0,
256 IO_SQ_THREAD_SHOULD_PARK,
257};
258
Jens Axboe534ca6d2020-09-02 13:52:19 -0600259struct io_sq_data {
260 refcount_t refs;
Jens Axboe05962f92021-03-06 13:58:48 -0700261 struct rw_semaphore rw_lock;
Jens Axboe69fb2132020-09-14 11:16:23 -0600262
263 /* ctx's that are using this sqd */
264 struct list_head ctx_list;
Jens Axboe69fb2132020-09-14 11:16:23 -0600265
Jens Axboe534ca6d2020-09-02 13:52:19 -0600266 struct task_struct *thread;
267 struct wait_queue_head wait;
Xiaoguang Wang08369242020-11-03 14:15:59 +0800268
269 unsigned sq_thread_idle;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700270 int sq_cpu;
271 pid_t task_pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -0700272 pid_t task_tgid;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700273
274 unsigned long state;
275 struct completion startup;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700276 struct completion exited;
Jens Axboe534ca6d2020-09-02 13:52:19 -0600277};
278
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000279#define IO_IOPOLL_BATCH 8
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000280#define IO_COMPL_BATCH 32
Pavel Begunkov6ff119a2021-02-10 00:03:18 +0000281#define IO_REQ_CACHE_SIZE 32
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000282#define IO_REQ_ALLOC_BATCH 8
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000283
284struct io_comp_state {
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000285 struct io_kiocb *reqs[IO_COMPL_BATCH];
Jens Axboe1b4c3512021-02-10 00:03:19 +0000286 unsigned int nr;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700287 unsigned int locked_free_nr;
288 /* inline/task_work completion list, under ->uring_lock */
Jens Axboe1b4c3512021-02-10 00:03:19 +0000289 struct list_head free_list;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700290 /* IRQ completion list, under ->completion_lock */
291 struct list_head locked_free_list;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000292};
293
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000294struct io_submit_link {
295 struct io_kiocb *head;
296 struct io_kiocb *last;
297};
298
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000299struct io_submit_state {
300 struct blk_plug plug;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000301 struct io_submit_link link;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000302
303 /*
304 * io_kiocb alloc cache
305 */
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000306 void *reqs[IO_REQ_CACHE_SIZE];
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000307 unsigned int free_reqs;
308
309 bool plug_started;
310
311 /*
312 * Batch completion logic
313 */
314 struct io_comp_state comp;
315
316 /*
317 * File reference cache
318 */
319 struct file *file;
320 unsigned int fd;
321 unsigned int file_refs;
322 unsigned int ios_left;
323};
324
Jens Axboe2b188cc2019-01-07 10:46:33 -0700325struct io_ring_ctx {
326 struct {
327 struct percpu_ref refs;
328 } ____cacheline_aligned_in_smp;
329
330 struct {
331 unsigned int flags;
Randy Dunlape1d85332020-02-05 20:57:10 -0800332 unsigned int compat: 1;
Randy Dunlape1d85332020-02-05 20:57:10 -0800333 unsigned int cq_overflow_flushed: 1;
334 unsigned int drain_next: 1;
335 unsigned int eventfd_async: 1;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200336 unsigned int restricted: 1;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700337
Hristo Venev75b28af2019-08-26 17:23:46 +0000338 /*
339 * Ring buffer of indices into array of io_uring_sqe, which is
340 * mmapped by the application using the IORING_OFF_SQES offset.
341 *
342 * This indirection could e.g. be used to assign fixed
343 * io_uring_sqe entries to operations and only submit them to
344 * the queue when needed.
345 *
346 * The kernel modifies neither the indices array nor the entries
347 * array.
348 */
349 u32 *sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700350 unsigned cached_sq_head;
351 unsigned sq_entries;
352 unsigned sq_mask;
Jens Axboe6c271ce2019-01-10 11:22:30 -0700353 unsigned sq_thread_idle;
Jens Axboe498ccd92019-10-25 10:04:25 -0600354 unsigned cached_sq_dropped;
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +0100355 unsigned cached_cq_overflow;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700356 unsigned long sq_check_overflow;
Jens Axboede0617e2019-04-06 21:51:27 -0600357
Jens Axboee9418942021-02-19 12:33:30 -0700358 /* hashed buffered write serialization */
359 struct io_wq_hash *hash_map;
360
Jens Axboede0617e2019-04-06 21:51:27 -0600361 struct list_head defer_list;
Jens Axboe5262f562019-09-17 12:26:57 -0600362 struct list_head timeout_list;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -0700363 struct list_head cq_overflow_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700364
Jens Axboead3eb2c2019-12-18 17:12:20 -0700365 struct io_uring_sqe *sq_sqes;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700366 } ____cacheline_aligned_in_smp;
367
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700368 struct {
369 struct mutex uring_lock;
370 wait_queue_head_t wait;
371 } ____cacheline_aligned_in_smp;
372
373 struct io_submit_state submit_state;
374
Hristo Venev75b28af2019-08-26 17:23:46 +0000375 struct io_rings *rings;
376
Jens Axboe2aede0e2020-09-14 10:45:53 -0600377 /* Only used for accounting purposes */
378 struct mm_struct *mm_account;
379
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +0100380 const struct cred *sq_creds; /* cred used for __io_sq_thread() */
Jens Axboe534ca6d2020-09-02 13:52:19 -0600381 struct io_sq_data *sq_data; /* if using sq thread polling */
382
Jens Axboe90554202020-09-03 12:12:41 -0600383 struct wait_queue_head sqo_sq_wait;
Jens Axboe69fb2132020-09-14 11:16:23 -0600384 struct list_head sqd_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700385
Jens Axboe6b063142019-01-10 22:13:58 -0700386 /*
387 * If used, fixed file set. Writers must ensure that ->refs is dead,
388 * readers must ensure that ->refs is alive as long as the file* is
389 * used. Only updated through io_uring_register(2).
390 */
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000391 struct fixed_rsrc_data *file_data;
Jens Axboe6b063142019-01-10 22:13:58 -0700392 unsigned nr_user_files;
393
Jens Axboeedafcce2019-01-09 09:16:05 -0700394 /* if used, fixed mapped user buffers */
395 unsigned nr_user_bufs;
396 struct io_mapped_ubuf *user_bufs;
397
Jens Axboe2b188cc2019-01-07 10:46:33 -0700398 struct user_struct *user;
399
Jens Axboe0f158b42020-05-14 17:18:39 -0600400 struct completion ref_comp;
Jens Axboe206aefd2019-11-07 18:27:42 -0700401
402#if defined(CONFIG_UNIX)
403 struct socket *ring_sock;
404#endif
405
Jens Axboe5a2e7452020-02-23 16:23:11 -0700406 struct idr io_buffer_idr;
407
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +0000408 struct xarray personalities;
409 u32 pers_next;
Jens Axboe071698e2020-01-28 10:04:42 -0700410
Jens Axboe206aefd2019-11-07 18:27:42 -0700411 struct {
412 unsigned cached_cq_tail;
413 unsigned cq_entries;
414 unsigned cq_mask;
415 atomic_t cq_timeouts;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -0500416 unsigned cq_last_tm_flush;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700417 unsigned long cq_check_overflow;
Jens Axboe206aefd2019-11-07 18:27:42 -0700418 struct wait_queue_head cq_wait;
419 struct fasync_struct *cq_fasync;
420 struct eventfd_ctx *cq_ev_fd;
421 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700422
423 struct {
Jens Axboe2b188cc2019-01-07 10:46:33 -0700424 spinlock_t completion_lock;
Jens Axboee94f1412019-12-19 12:06:02 -0700425
Jens Axboedef596e2019-01-09 08:59:42 -0700426 /*
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300427 * ->iopoll_list is protected by the ctx->uring_lock for
Jens Axboedef596e2019-01-09 08:59:42 -0700428 * io_uring instances that don't use IORING_SETUP_SQPOLL.
429 * For SQPOLL, only the single threaded io_sq_thread() will
430 * manipulate the list, hence no extra locking is needed there.
431 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300432 struct list_head iopoll_list;
Jens Axboe78076bb2019-12-04 19:56:40 -0700433 struct hlist_head *cancel_hash;
434 unsigned cancel_hash_bits;
Jens Axboee94f1412019-12-19 12:06:02 -0700435 bool poll_multi_file;
Jens Axboefcb323c2019-10-24 12:39:47 -0600436
437 spinlock_t inflight_lock;
438 struct list_head inflight_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700439 } ____cacheline_aligned_in_smp;
Jens Axboe85faa7b2020-04-09 18:14:00 -0600440
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000441 struct delayed_work rsrc_put_work;
442 struct llist_head rsrc_put_llist;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +0000443 struct list_head rsrc_ref_list;
444 spinlock_t rsrc_ref_lock;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600445
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200446 struct io_restriction restrictions;
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700447
Jens Axboe7c25c0d2021-02-16 07:17:00 -0700448 /* exit task_work */
449 struct callback_head *exit_task_work;
450
Jens Axboee9418942021-02-19 12:33:30 -0700451 struct wait_queue_head hash_wait;
452
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700453 /* Keep this last, we don't need it for the fast path */
454 struct work_struct exit_work;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000455 struct list_head tctx_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700456};
457
Jens Axboe09bb8392019-03-13 12:39:28 -0600458/*
459 * First field must be the file pointer in all the
460 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
461 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700462struct io_poll_iocb {
463 struct file *file;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000464 struct wait_queue_head *head;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700465 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600466 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700467 bool canceled;
Jens Axboe392edb42019-12-09 17:52:20 -0700468 struct wait_queue_entry wait;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700469};
470
Pavel Begunkov018043b2020-10-27 23:17:18 +0000471struct io_poll_remove {
472 struct file *file;
473 u64 addr;
474};
475
Jens Axboeb5dba592019-12-11 14:02:38 -0700476struct io_close {
477 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700478 int fd;
479};
480
Jens Axboead8a48a2019-11-15 08:49:11 -0700481struct io_timeout_data {
482 struct io_kiocb *req;
483 struct hrtimer timer;
484 struct timespec64 ts;
485 enum hrtimer_mode mode;
486};
487
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700488struct io_accept {
489 struct file *file;
490 struct sockaddr __user *addr;
491 int __user *addr_len;
492 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600493 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700494};
495
496struct io_sync {
497 struct file *file;
498 loff_t len;
499 loff_t off;
500 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700501 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700502};
503
Jens Axboefbf23842019-12-17 18:45:56 -0700504struct io_cancel {
505 struct file *file;
506 u64 addr;
507};
508
Jens Axboeb29472e2019-12-17 18:50:29 -0700509struct io_timeout {
510 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300511 u32 off;
512 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300513 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000514 /* head of the link, used by linked timeouts only */
515 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700516};
517
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100518struct io_timeout_rem {
519 struct file *file;
520 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000521
522 /* timeout update */
523 struct timespec64 ts;
524 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100525};
526
Jens Axboe9adbd452019-12-20 08:45:55 -0700527struct io_rw {
528 /* NOTE: kiocb has the file as the first member, so don't do it here */
529 struct kiocb kiocb;
530 u64 addr;
531 u64 len;
532};
533
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700534struct io_connect {
535 struct file *file;
536 struct sockaddr __user *addr;
537 int addr_len;
538};
539
Jens Axboee47293f2019-12-20 08:58:21 -0700540struct io_sr_msg {
541 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700542 union {
Pavel Begunkov270a5942020-07-12 20:41:04 +0300543 struct user_msghdr __user *umsg;
Jens Axboefddafac2020-01-04 20:19:44 -0700544 void __user *buf;
545 };
Jens Axboee47293f2019-12-20 08:58:21 -0700546 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700547 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700548 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700549 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700550};
551
Jens Axboe15b71ab2019-12-11 11:20:36 -0700552struct io_open {
553 struct file *file;
554 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700555 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700556 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600557 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700558};
559
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000560struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700561 struct file *file;
562 u64 arg;
563 u32 nr_args;
564 u32 offset;
565};
566
Jens Axboe4840e412019-12-25 22:03:45 -0700567struct io_fadvise {
568 struct file *file;
569 u64 offset;
570 u32 len;
571 u32 advice;
572};
573
Jens Axboec1ca7572019-12-25 22:18:28 -0700574struct io_madvise {
575 struct file *file;
576 u64 addr;
577 u32 len;
578 u32 advice;
579};
580
Jens Axboe3e4827b2020-01-08 15:18:09 -0700581struct io_epoll {
582 struct file *file;
583 int epfd;
584 int op;
585 int fd;
586 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700587};
588
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300589struct io_splice {
590 struct file *file_out;
591 struct file *file_in;
592 loff_t off_out;
593 loff_t off_in;
594 u64 len;
595 unsigned int flags;
596};
597
Jens Axboeddf0322d2020-02-23 16:41:33 -0700598struct io_provide_buf {
599 struct file *file;
600 __u64 addr;
601 __s32 len;
602 __u32 bgid;
603 __u16 nbufs;
604 __u16 bid;
605};
606
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700607struct io_statx {
608 struct file *file;
609 int dfd;
610 unsigned int mask;
611 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700612 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700613 struct statx __user *buffer;
614};
615
Jens Axboe36f4fa62020-09-05 11:14:22 -0600616struct io_shutdown {
617 struct file *file;
618 int how;
619};
620
Jens Axboe80a261f2020-09-28 14:23:58 -0600621struct io_rename {
622 struct file *file;
623 int old_dfd;
624 int new_dfd;
625 struct filename *oldpath;
626 struct filename *newpath;
627 int flags;
628};
629
Jens Axboe14a11432020-09-28 14:27:37 -0600630struct io_unlink {
631 struct file *file;
632 int dfd;
633 int flags;
634 struct filename *filename;
635};
636
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300637struct io_completion {
638 struct file *file;
639 struct list_head list;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +0300640 int cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300641};
642
Jens Axboef499a022019-12-02 16:28:46 -0700643struct io_async_connect {
644 struct sockaddr_storage address;
645};
646
Jens Axboe03b12302019-12-02 18:50:25 -0700647struct io_async_msghdr {
648 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000649 /* points to an allocated iov, if NULL we use fast_iov instead */
650 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700651 struct sockaddr __user *uaddr;
652 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700653 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700654};
655
Jens Axboef67676d2019-12-02 11:03:47 -0700656struct io_async_rw {
657 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600658 const struct iovec *free_iovec;
659 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600660 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600661 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700662};
663
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300664enum {
665 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
666 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
667 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
668 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
669 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700670 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300671
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300672 REQ_F_FAIL_LINK_BIT,
673 REQ_F_INFLIGHT_BIT,
674 REQ_F_CUR_POS_BIT,
675 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300676 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300677 REQ_F_ISREG_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300678 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700679 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700680 REQ_F_BUFFER_SELECTED_BIT,
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600681 REQ_F_NO_FILE_TABLE_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100682 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000683 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700684
685 /* not a real bit, just to check we're not overflowing the space */
686 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300687};
688
689enum {
690 /* ctx owns file */
691 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
692 /* drain existing IO first */
693 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
694 /* linked sqes */
695 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
696 /* doesn't sever on completion < 0 */
697 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
698 /* IOSQE_ASYNC */
699 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700700 /* IOSQE_BUFFER_SELECT */
701 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300702
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300703 /* fail rest of links */
704 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +0000705 /* on inflight list, should be cancelled and waited on exit reliably */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300706 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
707 /* read/write uses file position */
708 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
709 /* must not punt to workers */
710 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100711 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300712 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300713 /* regular file */
714 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300715 /* needs cleanup */
716 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700717 /* already went through poll handler */
718 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700719 /* buffer already selected */
720 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600721 /* doesn't need file table for this request */
722 REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100723 /* linked timeout is active, i.e. prepared by link's head */
724 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000725 /* completion is deferred through io_comp_state */
726 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700727};
728
729struct async_poll {
730 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600731 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300732};
733
Jens Axboe7cbf1722021-02-10 00:03:20 +0000734struct io_task_work {
735 struct io_wq_work_node node;
736 task_work_func_t func;
737};
738
Jens Axboe09bb8392019-03-13 12:39:28 -0600739/*
740 * NOTE! Each of the iocb union members has the file pointer
741 * as the first entry in their struct definition. So you can
742 * access the file pointer through any of the sub-structs,
743 * or directly as just 'ki_filp' in this struct.
744 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700745struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700746 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600747 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700748 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700749 struct io_poll_iocb poll;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000750 struct io_poll_remove poll_remove;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700751 struct io_accept accept;
752 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700753 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700754 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100755 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700756 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700757 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700758 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700759 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000760 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700761 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700762 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700763 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300764 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700765 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700766 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600767 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600768 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600769 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300770 /* use only after cleaning per-op data, see io_clean_op() */
771 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700772 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700773
Jens Axboee8c2bc12020-08-15 18:44:09 -0700774 /* opcode allocated if it needs to store data for async defer */
775 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700776 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800777 /* polled IO has completed */
778 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700779
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700780 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300781 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700782
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300783 struct io_ring_ctx *ctx;
784 unsigned int flags;
785 refcount_t refs;
786 struct task_struct *task;
787 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700788
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000789 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000790 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700791
Pavel Begunkovd21ffe72020-07-13 23:37:10 +0300792 /*
793 * 1. used with ctx->iopoll_list with reads/writes
794 * 2. to track reqs with ->files (see io_op_def::file_table)
795 */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300796 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000797 union {
798 struct io_task_work io_task_work;
799 struct callback_head task_work;
800 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300801 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
802 struct hlist_node hash_node;
803 struct async_poll *apoll;
804 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700805};
806
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000807struct io_tctx_node {
808 struct list_head ctx_node;
809 struct task_struct *task;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000810 struct io_ring_ctx *ctx;
811};
812
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300813struct io_defer_entry {
814 struct list_head list;
815 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300816 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300817};
818
Jens Axboed3656342019-12-18 09:50:26 -0700819struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700820 /* needs req->file assigned */
821 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700822 /* hash wq insertion if file is a regular file */
823 unsigned hash_reg_file : 1;
824 /* unbound wq insertion if file is a non-regular file */
825 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700826 /* opcode is not supported by this kernel */
827 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700828 /* set if opcode supports polled "wait" */
829 unsigned pollin : 1;
830 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700831 /* op supports buffer selection */
832 unsigned buffer_select : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700833 /* must always have async data allocated */
834 unsigned needs_async_data : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600835 /* should block plug */
836 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700837 /* size of async data needed, if any */
838 unsigned short async_size;
Jens Axboed3656342019-12-18 09:50:26 -0700839};
840
Jens Axboe09186822020-10-13 15:01:40 -0600841static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300842 [IORING_OP_NOP] = {},
843 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700844 .needs_file = 1,
845 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700846 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700847 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700848 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600849 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700850 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700851 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300852 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700853 .needs_file = 1,
854 .hash_reg_file = 1,
855 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700856 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700857 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600858 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700859 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700860 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300861 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700862 .needs_file = 1,
863 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300864 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700865 .needs_file = 1,
866 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700867 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600868 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700869 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700870 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300871 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700872 .needs_file = 1,
873 .hash_reg_file = 1,
874 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700875 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600876 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700877 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700878 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300879 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700880 .needs_file = 1,
881 .unbound_nonreg_file = 1,
882 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300883 [IORING_OP_POLL_REMOVE] = {},
884 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700885 .needs_file = 1,
886 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300887 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700888 .needs_file = 1,
889 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700890 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700891 .needs_async_data = 1,
892 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700893 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300894 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700895 .needs_file = 1,
896 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700897 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700898 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700899 .needs_async_data = 1,
900 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700901 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300902 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700903 .needs_async_data = 1,
904 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700905 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000906 [IORING_OP_TIMEOUT_REMOVE] = {
907 /* used by timeout updates' prep() */
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000908 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300909 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700910 .needs_file = 1,
911 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700912 .pollin = 1,
Jens Axboed3656342019-12-18 09:50:26 -0700913 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300914 [IORING_OP_ASYNC_CANCEL] = {},
915 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700916 .needs_async_data = 1,
917 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700918 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300919 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700920 .needs_file = 1,
921 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700922 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700923 .needs_async_data = 1,
924 .async_size = sizeof(struct io_async_connect),
Jens Axboed3656342019-12-18 09:50:26 -0700925 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300926 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700927 .needs_file = 1,
928 },
Jens Axboe44526be2021-02-15 13:32:18 -0700929 [IORING_OP_OPENAT] = {},
930 [IORING_OP_CLOSE] = {},
931 [IORING_OP_FILES_UPDATE] = {},
932 [IORING_OP_STATX] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300933 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700934 .needs_file = 1,
935 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700936 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700937 .buffer_select = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600938 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700939 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700940 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300941 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700942 .needs_file = 1,
943 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700944 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600945 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700946 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700947 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300948 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700949 .needs_file = 1,
950 },
Jens Axboe44526be2021-02-15 13:32:18 -0700951 [IORING_OP_MADVISE] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300952 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700953 .needs_file = 1,
954 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700955 .pollout = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700956 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300957 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700958 .needs_file = 1,
959 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700960 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700961 .buffer_select = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700962 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300963 [IORING_OP_OPENAT2] = {
Jens Axboecebdb982020-01-08 17:59:24 -0700964 },
Jens Axboe3e4827b2020-01-08 15:18:09 -0700965 [IORING_OP_EPOLL_CTL] = {
966 .unbound_nonreg_file = 1,
Jens Axboe3e4827b2020-01-08 15:18:09 -0700967 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300968 [IORING_OP_SPLICE] = {
969 .needs_file = 1,
970 .hash_reg_file = 1,
971 .unbound_nonreg_file = 1,
Jens Axboeddf0322d2020-02-23 16:41:33 -0700972 },
973 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -0700974 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +0300975 [IORING_OP_TEE] = {
976 .needs_file = 1,
977 .hash_reg_file = 1,
978 .unbound_nonreg_file = 1,
979 },
Jens Axboe36f4fa62020-09-05 11:14:22 -0600980 [IORING_OP_SHUTDOWN] = {
981 .needs_file = 1,
982 },
Jens Axboe44526be2021-02-15 13:32:18 -0700983 [IORING_OP_RENAMEAT] = {},
984 [IORING_OP_UNLINKAT] = {},
Jens Axboed3656342019-12-18 09:50:26 -0700985};
986
Pavel Begunkov7a612352021-03-09 00:37:59 +0000987static bool io_disarm_next(struct io_kiocb *req);
Pavel Begunkovd56d9382021-03-06 11:02:13 +0000988static void io_uring_del_task_file(unsigned long index);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +0000989static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
990 struct task_struct *task,
991 struct files_struct *files);
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700992static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000993static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node);
Pavel Begunkovbc9744c2021-01-15 17:37:49 +0000994static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Pavel Begunkov1ffc5422020-12-30 21:34:15 +0000995 struct io_ring_ctx *ctx);
Pavel Begunkovf2303b12021-02-20 18:03:49 +0000996static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +0000997
Pavel Begunkov23faba32021-02-11 18:28:22 +0000998static bool io_rw_reissue(struct io_kiocb *req);
Jens Axboe78e19bb2019-11-06 15:21:34 -0700999static void io_cqring_fill_event(struct io_kiocb *req, long res);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001000static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001001static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec40f6372020-06-25 15:39:59 -06001002static void io_double_put_req(struct io_kiocb *req);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001003static void io_dismantle_req(struct io_kiocb *req);
1004static void io_put_task(struct task_struct *task, int nr);
1005static void io_queue_next(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001006static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001007static void __io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001008static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -07001009static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001010 struct io_uring_rsrc_update *ip,
Jens Axboe05f3fb32019-12-09 11:22:50 -07001011 unsigned nr_args);
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001012static void __io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001013static struct file *io_file_get(struct io_submit_state *state,
1014 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001015static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001016static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001017
Pavel Begunkov847595d2021-02-04 13:52:06 +00001018static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
1019 struct iov_iter *iter, bool needs_lock);
Jens Axboeff6165b2020-08-13 09:47:43 -06001020static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
1021 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06001022 struct iov_iter *iter, bool force);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001023static void io_req_task_queue(struct io_kiocb *req);
Jens Axboe65453d12021-02-10 00:03:21 +00001024static void io_submit_flush_completions(struct io_comp_state *cs,
1025 struct io_ring_ctx *ctx);
Jens Axboe9a56a232019-01-09 09:06:50 -07001026
Jens Axboe2b188cc2019-01-07 10:46:33 -07001027static struct kmem_cache *req_cachep;
1028
Jens Axboe09186822020-10-13 15:01:40 -06001029static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001030
1031struct sock *io_uring_get_socket(struct file *file)
1032{
1033#if defined(CONFIG_UNIX)
1034 if (file->f_op == &io_uring_fops) {
1035 struct io_ring_ctx *ctx = file->private_data;
1036
1037 return ctx->ring_sock->sk;
1038 }
1039#endif
1040 return NULL;
1041}
1042EXPORT_SYMBOL(io_uring_get_socket);
1043
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001044#define io_for_each_link(pos, head) \
1045 for (pos = (head); pos; pos = pos->link)
1046
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001047static inline void io_clean_op(struct io_kiocb *req)
1048{
Pavel Begunkov9d5c8192021-01-24 15:08:14 +00001049 if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED))
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001050 __io_clean_op(req);
1051}
1052
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001053static inline void io_set_resource_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001054{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001055 struct io_ring_ctx *ctx = req->ctx;
1056
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001057 if (!req->fixed_rsrc_refs) {
1058 req->fixed_rsrc_refs = &ctx->file_data->node->refs;
1059 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001060 }
1061}
1062
Pavel Begunkov08d23632020-11-06 13:00:22 +00001063static bool io_match_task(struct io_kiocb *head,
1064 struct task_struct *task,
1065 struct files_struct *files)
1066{
1067 struct io_kiocb *req;
1068
Jens Axboe84965ff2021-01-23 15:51:11 -07001069 if (task && head->task != task) {
1070 /* in terms of cancelation, always match if req task is dead */
1071 if (head->task->flags & PF_EXITING)
1072 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001073 return false;
Jens Axboe84965ff2021-01-23 15:51:11 -07001074 }
Pavel Begunkov08d23632020-11-06 13:00:22 +00001075 if (!files)
1076 return true;
1077
1078 io_for_each_link(req, head) {
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +00001079 if (req->flags & REQ_F_INFLIGHT)
Jens Axboe02a13672021-01-23 15:49:31 -07001080 return true;
Jens Axboe4379bf82021-02-15 13:40:22 -07001081 if (req->task->files == files)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001082 return true;
1083 }
1084 return false;
1085}
1086
Jens Axboec40f6372020-06-25 15:39:59 -06001087static inline void req_set_fail_links(struct io_kiocb *req)
1088{
1089 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1090 req->flags |= REQ_F_FAIL_LINK;
1091}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001092
Jens Axboe2b188cc2019-01-07 10:46:33 -07001093static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1094{
1095 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1096
Jens Axboe0f158b42020-05-14 17:18:39 -06001097 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001098}
1099
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001100static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1101{
1102 return !req->timeout.off;
1103}
1104
Jens Axboe2b188cc2019-01-07 10:46:33 -07001105static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1106{
1107 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001108 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001109
1110 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1111 if (!ctx)
1112 return NULL;
1113
Jens Axboe78076bb2019-12-04 19:56:40 -07001114 /*
1115 * Use 5 bits less than the max cq entries, that should give us around
1116 * 32 entries per hash list if totally full and uniformly spread.
1117 */
1118 hash_bits = ilog2(p->cq_entries);
1119 hash_bits -= 5;
1120 if (hash_bits <= 0)
1121 hash_bits = 1;
1122 ctx->cancel_hash_bits = hash_bits;
1123 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1124 GFP_KERNEL);
1125 if (!ctx->cancel_hash)
1126 goto err;
1127 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1128
Roman Gushchin21482892019-05-07 10:01:48 -07001129 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001130 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1131 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001132
1133 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001134 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001135 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001136 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001137 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001138 init_completion(&ctx->ref_comp);
Jens Axboe5a2e7452020-02-23 16:23:11 -07001139 idr_init(&ctx->io_buffer_idr);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00001140 xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001141 mutex_init(&ctx->uring_lock);
1142 init_waitqueue_head(&ctx->wait);
1143 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001144 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001145 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001146 INIT_LIST_HEAD(&ctx->timeout_list);
Jens Axboefcb323c2019-10-24 12:39:47 -06001147 spin_lock_init(&ctx->inflight_lock);
1148 INIT_LIST_HEAD(&ctx->inflight_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001149 spin_lock_init(&ctx->rsrc_ref_lock);
1150 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001151 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1152 init_llist_head(&ctx->rsrc_put_llist);
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00001153 INIT_LIST_HEAD(&ctx->tctx_list);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001154 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001155 INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001156 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001157err:
Jens Axboe78076bb2019-12-04 19:56:40 -07001158 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001159 kfree(ctx);
1160 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001161}
1162
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001163static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001164{
Jens Axboe2bc99302020-07-09 09:43:27 -06001165 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1166 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001167
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001168 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001169 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001170 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001171
Bob Liu9d858b22019-11-13 18:06:25 +08001172 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001173}
1174
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001175static void io_req_track_inflight(struct io_kiocb *req)
1176{
1177 struct io_ring_ctx *ctx = req->ctx;
1178
1179 if (!(req->flags & REQ_F_INFLIGHT)) {
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001180 req->flags |= REQ_F_INFLIGHT;
1181
1182 spin_lock_irq(&ctx->inflight_lock);
1183 list_add(&req->inflight_entry, &ctx->inflight_list);
1184 spin_unlock_irq(&ctx->inflight_lock);
1185 }
1186}
1187
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001188static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001189{
Jens Axboed3656342019-12-18 09:50:26 -07001190 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001191 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001192
Jens Axboe003e8dc2021-03-06 09:22:27 -07001193 if (!req->work.creds)
1194 req->work.creds = get_current_cred();
1195
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001196 if (req->flags & REQ_F_FORCE_ASYNC)
1197 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1198
Jens Axboed3656342019-12-18 09:50:26 -07001199 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001200 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001201 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboed3656342019-12-18 09:50:26 -07001202 } else {
1203 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001204 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001205 }
Jens Axboe561fb042019-10-24 07:25:42 -06001206}
1207
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001208static void io_prep_async_link(struct io_kiocb *req)
1209{
1210 struct io_kiocb *cur;
1211
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001212 io_for_each_link(cur, req)
1213 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001214}
1215
Pavel Begunkovebf93662021-03-01 18:20:47 +00001216static void io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001217{
Jackie Liua197f662019-11-08 08:09:12 -07001218 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001219 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001220 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe561fb042019-10-24 07:25:42 -06001221
Jens Axboe3bfe6102021-02-16 14:15:30 -07001222 BUG_ON(!tctx);
1223 BUG_ON(!tctx->io_wq);
Jens Axboe561fb042019-10-24 07:25:42 -06001224
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001225 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1226 &req->work, req->flags);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001227 /* init ->work of the whole link before punting */
1228 io_prep_async_link(req);
Pavel Begunkovebf93662021-03-01 18:20:47 +00001229 io_wq_enqueue(tctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001230 if (link)
1231 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001232}
1233
Jens Axboe5262f562019-09-17 12:26:57 -06001234static void io_kill_timeout(struct io_kiocb *req)
1235{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001236 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001237 int ret;
1238
Jens Axboee8c2bc12020-08-15 18:44:09 -07001239 ret = hrtimer_try_to_cancel(&io->timer);
Jens Axboe5262f562019-09-17 12:26:57 -06001240 if (ret != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001241 atomic_set(&req->ctx->cq_timeouts,
1242 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001243 list_del_init(&req->timeout.list);
Jens Axboe78e19bb2019-11-06 15:21:34 -07001244 io_cqring_fill_event(req, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001245 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001246 }
1247}
1248
Jens Axboe76e1b642020-09-26 15:05:03 -06001249/*
1250 * Returns true if we found and killed one or more timeouts
1251 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00001252static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
1253 struct files_struct *files)
Jens Axboe5262f562019-09-17 12:26:57 -06001254{
1255 struct io_kiocb *req, *tmp;
Jens Axboe76e1b642020-09-26 15:05:03 -06001256 int canceled = 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001257
1258 spin_lock_irq(&ctx->completion_lock);
Jens Axboef3606e32020-09-22 08:18:24 -06001259 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00001260 if (io_match_task(req, tsk, files)) {
Jens Axboef3606e32020-09-22 08:18:24 -06001261 io_kill_timeout(req);
Jens Axboe76e1b642020-09-26 15:05:03 -06001262 canceled++;
1263 }
Jens Axboef3606e32020-09-22 08:18:24 -06001264 }
Jens Axboe5262f562019-09-17 12:26:57 -06001265 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe76e1b642020-09-26 15:05:03 -06001266 return canceled != 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001267}
1268
Pavel Begunkov04518942020-05-26 20:34:05 +03001269static void __io_queue_deferred(struct io_ring_ctx *ctx)
1270{
1271 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001272 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1273 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001274
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001275 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001276 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001277 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001278 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001279 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001280 } while (!list_empty(&ctx->defer_list));
1281}
1282
Pavel Begunkov360428f2020-05-30 14:54:17 +03001283static void io_flush_timeouts(struct io_ring_ctx *ctx)
1284{
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001285 u32 seq;
1286
1287 if (list_empty(&ctx->timeout_list))
1288 return;
1289
1290 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
1291
1292 do {
1293 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001294 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001295 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001296
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001297 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001298 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001299
1300 /*
1301 * Since seq can easily wrap around over time, subtract
1302 * the last seq at which timeouts were flushed before comparing.
1303 * Assuming not more than 2^31-1 events have happened since,
1304 * these subtractions won't have wrapped, so we can check if
1305 * target is in [last_seq, current_seq] by comparing the two.
1306 */
1307 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1308 events_got = seq - ctx->cq_last_tm_flush;
1309 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001310 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001311
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001312 list_del_init(&req->timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001313 io_kill_timeout(req);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001314 } while (!list_empty(&ctx->timeout_list));
1315
1316 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001317}
1318
Jens Axboede0617e2019-04-06 21:51:27 -06001319static void io_commit_cqring(struct io_ring_ctx *ctx)
1320{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001321 io_flush_timeouts(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001322
1323 /* order cqe stores with ring update */
1324 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001325
Pavel Begunkov04518942020-05-26 20:34:05 +03001326 if (unlikely(!list_empty(&ctx->defer_list)))
1327 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001328}
1329
Jens Axboe90554202020-09-03 12:12:41 -06001330static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1331{
1332 struct io_rings *r = ctx->rings;
1333
1334 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1335}
1336
Pavel Begunkov888aae22021-01-19 13:32:39 +00001337static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1338{
1339 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1340}
1341
Jens Axboe2b188cc2019-01-07 10:46:33 -07001342static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
1343{
Hristo Venev75b28af2019-08-26 17:23:46 +00001344 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001345 unsigned tail;
1346
Stefan Bühler115e12e2019-04-24 23:54:18 +02001347 /*
1348 * writes to the cq entry need to come after reading head; the
1349 * control dependency is enough as we're using WRITE_ONCE to
1350 * fill the cq entry
1351 */
Pavel Begunkov888aae22021-01-19 13:32:39 +00001352 if (__io_cqring_events(ctx) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001353 return NULL;
1354
Pavel Begunkov888aae22021-01-19 13:32:39 +00001355 tail = ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001356 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001357}
1358
Jens Axboef2842ab2020-01-08 11:04:00 -07001359static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1360{
Jens Axboef0b493e2020-02-01 21:30:11 -07001361 if (!ctx->cq_ev_fd)
1362 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001363 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1364 return false;
Jens Axboef2842ab2020-01-08 11:04:00 -07001365 if (!ctx->eventfd_async)
1366 return true;
Jens Axboeb41e9852020-02-17 09:52:41 -07001367 return io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001368}
1369
Jens Axboeb41e9852020-02-17 09:52:41 -07001370static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001371{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001372 /* see waitqueue_active() comment */
1373 smp_mb();
1374
Jens Axboe8c838782019-03-12 15:48:16 -06001375 if (waitqueue_active(&ctx->wait))
1376 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001377 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1378 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001379 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001380 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001381 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001382 wake_up_interruptible(&ctx->cq_wait);
1383 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1384 }
Jens Axboe8c838782019-03-12 15:48:16 -06001385}
1386
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001387static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1388{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001389 /* see waitqueue_active() comment */
1390 smp_mb();
1391
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001392 if (ctx->flags & IORING_SETUP_SQPOLL) {
1393 if (waitqueue_active(&ctx->wait))
1394 wake_up(&ctx->wait);
1395 }
1396 if (io_should_trigger_evfd(ctx))
1397 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001398 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001399 wake_up_interruptible(&ctx->cq_wait);
1400 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1401 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001402}
1403
Jens Axboec4a2ed72019-11-21 21:01:26 -07001404/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c503152021-01-04 20:36:36 +00001405static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1406 struct task_struct *tsk,
1407 struct files_struct *files)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001408{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001409 struct io_rings *rings = ctx->rings;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001410 struct io_kiocb *req, *tmp;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001411 struct io_uring_cqe *cqe;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001412 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001413 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001414 LIST_HEAD(list);
1415
Pavel Begunkove23de152020-12-17 00:24:37 +00001416 if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
1417 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001418
Jens Axboeb18032b2021-01-24 16:58:56 -07001419 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001420 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboee6c8aa92020-09-28 13:10:13 -06001421 list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00001422 if (!io_match_task(req, tsk, files))
Jens Axboee6c8aa92020-09-28 13:10:13 -06001423 continue;
1424
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001425 cqe = io_get_cqring(ctx);
1426 if (!cqe && !force)
1427 break;
1428
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001429 list_move(&req->compl.list, &list);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001430 if (cqe) {
1431 WRITE_ONCE(cqe->user_data, req->user_data);
1432 WRITE_ONCE(cqe->res, req->result);
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001433 WRITE_ONCE(cqe->flags, req->compl.cflags);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001434 } else {
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001435 ctx->cached_cq_overflow++;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001436 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001437 ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001438 }
Jens Axboeb18032b2021-01-24 16:58:56 -07001439 posted = true;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001440 }
1441
Pavel Begunkov09e88402020-12-17 00:24:38 +00001442 all_flushed = list_empty(&ctx->cq_overflow_list);
1443 if (all_flushed) {
1444 clear_bit(0, &ctx->sq_check_overflow);
1445 clear_bit(0, &ctx->cq_check_overflow);
1446 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1447 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001448
Jens Axboeb18032b2021-01-24 16:58:56 -07001449 if (posted)
1450 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001451 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001452 if (posted)
1453 io_cqring_ev_posted(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001454
1455 while (!list_empty(&list)) {
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001456 req = list_first_entry(&list, struct io_kiocb, compl.list);
1457 list_del(&req->compl.list);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001458 io_put_req(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001459 }
Jens Axboec4a2ed72019-11-21 21:01:26 -07001460
Pavel Begunkov09e88402020-12-17 00:24:38 +00001461 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001462}
1463
Jens Axboeca0a2652021-03-04 17:15:48 -07001464static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
Pavel Begunkov6c503152021-01-04 20:36:36 +00001465 struct task_struct *tsk,
1466 struct files_struct *files)
1467{
Jens Axboeca0a2652021-03-04 17:15:48 -07001468 bool ret = true;
1469
Pavel Begunkov6c503152021-01-04 20:36:36 +00001470 if (test_bit(0, &ctx->cq_check_overflow)) {
1471 /* iopoll syncs against uring_lock, not completion_lock */
1472 if (ctx->flags & IORING_SETUP_IOPOLL)
1473 mutex_lock(&ctx->uring_lock);
Jens Axboeca0a2652021-03-04 17:15:48 -07001474 ret = __io_cqring_overflow_flush(ctx, force, tsk, files);
Pavel Begunkov6c503152021-01-04 20:36:36 +00001475 if (ctx->flags & IORING_SETUP_IOPOLL)
1476 mutex_unlock(&ctx->uring_lock);
1477 }
Jens Axboeca0a2652021-03-04 17:15:48 -07001478
1479 return ret;
Pavel Begunkov6c503152021-01-04 20:36:36 +00001480}
1481
Jens Axboebcda7ba2020-02-23 16:42:51 -07001482static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001483{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001484 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001485 struct io_uring_cqe *cqe;
1486
Jens Axboe78e19bb2019-11-06 15:21:34 -07001487 trace_io_uring_complete(ctx, req->user_data, res);
Jens Axboe51c3ff62019-11-03 06:52:50 -07001488
Jens Axboe2b188cc2019-01-07 10:46:33 -07001489 /*
1490 * If we can't get a cq entry, userspace overflowed the
1491 * submission (by quite a lot). Increment the overflow count in
1492 * the ring.
1493 */
1494 cqe = io_get_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001495 if (likely(cqe)) {
Jens Axboe78e19bb2019-11-06 15:21:34 -07001496 WRITE_ONCE(cqe->user_data, req->user_data);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001497 WRITE_ONCE(cqe->res, res);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001498 WRITE_ONCE(cqe->flags, cflags);
Jens Axboefdaf0832020-10-30 09:37:30 -06001499 } else if (ctx->cq_overflow_flushed ||
1500 atomic_read(&req->task->io_uring->in_idle)) {
Jens Axboe0f212202020-09-13 13:09:39 -06001501 /*
1502 * If we're in ring overflow flush mode, or in task cancel mode,
1503 * then we cannot store the request for later flushing, we need
1504 * to drop it on the floor.
1505 */
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001506 ctx->cached_cq_overflow++;
1507 WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001508 } else {
Jens Axboead3eb2c2019-12-18 17:12:20 -07001509 if (list_empty(&ctx->cq_overflow_list)) {
1510 set_bit(0, &ctx->sq_check_overflow);
1511 set_bit(0, &ctx->cq_check_overflow);
Xiaoguang Wang6d5f9042020-07-09 09:15:29 +08001512 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
Jens Axboead3eb2c2019-12-18 17:12:20 -07001513 }
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001514 io_clean_op(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001515 req->result = res;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001516 req->compl.cflags = cflags;
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001517 refcount_inc(&req->refs);
1518 list_add_tail(&req->compl.list, &ctx->cq_overflow_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001519 }
1520}
1521
Jens Axboebcda7ba2020-02-23 16:42:51 -07001522static void io_cqring_fill_event(struct io_kiocb *req, long res)
1523{
1524 __io_cqring_fill_event(req, res, 0);
1525}
1526
Pavel Begunkov7a612352021-03-09 00:37:59 +00001527static void io_req_complete_post(struct io_kiocb *req, long res,
1528 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001529{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001530 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001531 unsigned long flags;
1532
1533 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001534 __io_cqring_fill_event(req, res, cflags);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001535 /*
1536 * If we're the last reference to this request, add to our locked
1537 * free_list cache.
1538 */
1539 if (refcount_dec_and_test(&req->refs)) {
1540 struct io_comp_state *cs = &ctx->submit_state.comp;
1541
Pavel Begunkov7a612352021-03-09 00:37:59 +00001542 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
1543 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK))
1544 io_disarm_next(req);
1545 if (req->link) {
1546 io_req_task_queue(req->link);
1547 req->link = NULL;
1548 }
1549 }
Jens Axboec7dae4b2021-02-09 19:53:37 -07001550 io_dismantle_req(req);
1551 io_put_task(req->task, 1);
1552 list_add(&req->compl.list, &cs->locked_free_list);
1553 cs->locked_free_nr++;
1554 } else
1555 req = NULL;
Pavel Begunkov7a612352021-03-09 00:37:59 +00001556 io_commit_cqring(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001557 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboe8c838782019-03-12 15:48:16 -06001558 io_cqring_ev_posted(ctx);
Pavel Begunkov7a612352021-03-09 00:37:59 +00001559
1560 if (req)
Jens Axboec7dae4b2021-02-09 19:53:37 -07001561 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001562}
1563
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001564static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001565 unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001566{
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001567 io_clean_op(req);
1568 req->result = res;
1569 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001570 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001571}
Jens Axboe2b188cc2019-01-07 10:46:33 -07001572
Pavel Begunkov889fca72021-02-10 00:03:09 +00001573static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1574 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001575{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001576 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1577 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001578 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001579 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001580}
Jens Axboebcda7ba2020-02-23 16:42:51 -07001581
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001582static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001583{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001584 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001585}
1586
Jens Axboec7dae4b2021-02-09 19:53:37 -07001587static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001588{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001589 struct io_submit_state *state = &ctx->submit_state;
1590 struct io_comp_state *cs = &state->comp;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001591 struct io_kiocb *req = NULL;
1592
Jens Axboec7dae4b2021-02-09 19:53:37 -07001593 /*
1594 * If we have more than a batch's worth of requests in our IRQ side
1595 * locked cache, grab the lock and move them over to our submission
1596 * side cache.
1597 */
1598 if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) {
1599 spin_lock_irq(&ctx->completion_lock);
1600 list_splice_init(&cs->locked_free_list, &cs->free_list);
1601 cs->locked_free_nr = 0;
1602 spin_unlock_irq(&ctx->completion_lock);
1603 }
1604
1605 while (!list_empty(&cs->free_list)) {
1606 req = list_first_entry(&cs->free_list, struct io_kiocb,
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001607 compl.list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001608 list_del(&req->compl.list);
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001609 state->reqs[state->free_reqs++] = req;
1610 if (state->free_reqs == ARRAY_SIZE(state->reqs))
1611 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001612 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001613
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001614 return req != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001615}
1616
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001617static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001618{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001619 struct io_submit_state *state = &ctx->submit_state;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001620
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001621 BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs));
Jens Axboe2b188cc2019-01-07 10:46:33 -07001622
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001623 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001624 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07001625 int ret;
1626
Jens Axboec7dae4b2021-02-09 19:53:37 -07001627 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001628 goto got_req;
1629
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001630 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
1631 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06001632
1633 /*
1634 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1635 * retry single alloc to be on the safe side.
1636 */
1637 if (unlikely(ret <= 0)) {
1638 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1639 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00001640 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06001641 ret = 1;
1642 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001643 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001644 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001645got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03001646 state->free_reqs--;
1647 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001648}
1649
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001650static inline void io_put_file(struct io_kiocb *req, struct file *file,
1651 bool fixed)
1652{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001653 if (!fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001654 fput(file);
1655}
1656
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001657static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001658{
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001659 io_clean_op(req);
Pavel Begunkov929a3af2020-02-19 00:19:09 +03001660
Jens Axboee8c2bc12020-08-15 18:44:09 -07001661 if (req->async_data)
1662 kfree(req->async_data);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001663 if (req->file)
1664 io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001665 if (req->fixed_rsrc_refs)
1666 percpu_ref_put(req->fixed_rsrc_refs);
Jens Axboe003e8dc2021-03-06 09:22:27 -07001667 if (req->work.creds) {
1668 put_cred(req->work.creds);
1669 req->work.creds = NULL;
1670 }
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001671
1672 if (req->flags & REQ_F_INFLIGHT) {
1673 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001674 unsigned long flags;
1675
1676 spin_lock_irqsave(&ctx->inflight_lock, flags);
1677 list_del(&req->inflight_entry);
1678 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1679 req->flags &= ~REQ_F_INFLIGHT;
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001680 }
Pavel Begunkove6543a82020-06-28 12:52:30 +03001681}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001682
Pavel Begunkovb23fcf42021-03-01 18:20:48 +00001683/* must to be called somewhat shortly after putting a request */
Pavel Begunkov7c660732021-01-25 11:42:21 +00001684static inline void io_put_task(struct task_struct *task, int nr)
1685{
1686 struct io_uring_task *tctx = task->io_uring;
1687
1688 percpu_counter_sub(&tctx->inflight, nr);
1689 if (unlikely(atomic_read(&tctx->in_idle)))
1690 wake_up(&tctx->wait);
1691 put_task_struct_many(task, nr);
1692}
1693
Pavel Begunkov216578e2020-10-13 09:44:00 +01001694static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001695{
Jens Axboe51a4cc12020-08-10 10:55:56 -06001696 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001697
Pavel Begunkov216578e2020-10-13 09:44:00 +01001698 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00001699 io_put_task(req->task, 1);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001700
Pavel Begunkov3893f392021-02-10 00:03:15 +00001701 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001702 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001703}
1704
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001705static inline void io_remove_next_linked(struct io_kiocb *req)
1706{
1707 struct io_kiocb *nxt = req->link;
1708
1709 req->link = nxt->link;
1710 nxt->link = NULL;
1711}
1712
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001713static bool io_kill_linked_timeout(struct io_kiocb *req)
1714 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001715{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001716 struct io_kiocb *link = req->link;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001717 bool cancelled = false;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001718
Pavel Begunkov900fad42020-10-19 16:39:16 +01001719 /*
1720 * Can happen if a linked timeout fired and link had been like
1721 * req -> link t-out -> link t-out [-> ...]
1722 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001723 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1724 struct io_timeout_data *io = link->async_data;
1725 int ret;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001726
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001727 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00001728 link->timeout.head = NULL;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001729 ret = hrtimer_try_to_cancel(&io->timer);
1730 if (ret != -1) {
1731 io_cqring_fill_event(link, -ECANCELED);
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001732 io_put_req_deferred(link, 1);
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001733 cancelled = true;
1734 }
1735 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001736 req->flags &= ~REQ_F_LINK_TIMEOUT;
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001737 return cancelled;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001738}
1739
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001740static void io_fail_links(struct io_kiocb *req)
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001741 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001742{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001743 struct io_kiocb *nxt, *link = req->link;
Jens Axboe9e645e112019-05-10 16:07:28 -06001744
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001745 req->link = NULL;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001746 while (link) {
1747 nxt = link->link;
1748 link->link = NULL;
1749
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001750 trace_io_uring_fail_link(req, link);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001751 io_cqring_fill_event(link, -ECANCELED);
Jens Axboe1575f212021-02-27 15:20:49 -07001752 io_put_req_deferred(link, 2);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001753 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001754 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001755}
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001756
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001757static bool io_disarm_next(struct io_kiocb *req)
1758 __must_hold(&req->ctx->completion_lock)
1759{
1760 bool posted = false;
1761
1762 if (likely(req->flags & REQ_F_LINK_TIMEOUT))
1763 posted = io_kill_linked_timeout(req);
1764 if (unlikely(req->flags & REQ_F_FAIL_LINK)) {
1765 posted |= (req->link != NULL);
1766 io_fail_links(req);
1767 }
1768 return posted;
Jens Axboe9e645e112019-05-10 16:07:28 -06001769}
1770
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001771static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001772{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001773 struct io_kiocb *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07001774
Jens Axboe9e645e112019-05-10 16:07:28 -06001775 /*
1776 * If LINK is set, we have dependent requests in this chain. If we
1777 * didn't fail this request, queue the first one up, moving any other
1778 * dependencies to the next request. In case of failure, fail the rest
1779 * of the chain.
1780 */
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001781 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) {
1782 struct io_ring_ctx *ctx = req->ctx;
1783 unsigned long flags;
1784 bool posted;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001785
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001786 spin_lock_irqsave(&ctx->completion_lock, flags);
1787 posted = io_disarm_next(req);
1788 if (posted)
1789 io_commit_cqring(req->ctx);
1790 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1791 if (posted)
1792 io_cqring_ev_posted(ctx);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001793 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001794 nxt = req->link;
1795 req->link = NULL;
1796 return nxt;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001797}
Jens Axboe2665abf2019-11-05 12:40:47 -07001798
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001799static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001800{
Pavel Begunkovcdbff982021-02-12 18:41:16 +00001801 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK))))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001802 return NULL;
1803 return __io_req_find_next(req);
1804}
1805
Pavel Begunkov2c323952021-02-28 22:04:53 +00001806static void ctx_flush_and_put(struct io_ring_ctx *ctx)
1807{
1808 if (!ctx)
1809 return;
1810 if (ctx->submit_state.comp.nr) {
1811 mutex_lock(&ctx->uring_lock);
1812 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
1813 mutex_unlock(&ctx->uring_lock);
1814 }
1815 percpu_ref_put(&ctx->refs);
1816}
1817
Jens Axboe7cbf1722021-02-10 00:03:20 +00001818static bool __tctx_task_work(struct io_uring_task *tctx)
1819{
Jens Axboe65453d12021-02-10 00:03:21 +00001820 struct io_ring_ctx *ctx = NULL;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001821 struct io_wq_work_list list;
1822 struct io_wq_work_node *node;
1823
1824 if (wq_list_empty(&tctx->task_list))
1825 return false;
1826
Jens Axboe0b81e802021-02-16 10:33:53 -07001827 spin_lock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001828 list = tctx->task_list;
1829 INIT_WQ_LIST(&tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001830 spin_unlock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001831
1832 node = list.first;
1833 while (node) {
1834 struct io_wq_work_node *next = node->next;
1835 struct io_kiocb *req;
1836
1837 req = container_of(node, struct io_kiocb, io_task_work.node);
Pavel Begunkov2c323952021-02-28 22:04:53 +00001838 if (req->ctx != ctx) {
1839 ctx_flush_and_put(ctx);
1840 ctx = req->ctx;
1841 percpu_ref_get(&ctx->refs);
1842 }
1843
Jens Axboe7cbf1722021-02-10 00:03:20 +00001844 req->task_work.func(&req->task_work);
1845 node = next;
Jens Axboe65453d12021-02-10 00:03:21 +00001846 }
1847
Pavel Begunkov2c323952021-02-28 22:04:53 +00001848 ctx_flush_and_put(ctx);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001849 return list.first != NULL;
1850}
1851
1852static void tctx_task_work(struct callback_head *cb)
1853{
1854 struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work);
1855
Jens Axboe1d5f3602021-02-26 14:54:16 -07001856 clear_bit(0, &tctx->task_state);
1857
Jens Axboe7cbf1722021-02-10 00:03:20 +00001858 while (__tctx_task_work(tctx))
1859 cond_resched();
Jens Axboe7cbf1722021-02-10 00:03:20 +00001860}
1861
1862static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req,
1863 enum task_work_notify_mode notify)
1864{
1865 struct io_uring_task *tctx = tsk->io_uring;
1866 struct io_wq_work_node *node, *prev;
Jens Axboe0b81e802021-02-16 10:33:53 -07001867 unsigned long flags;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001868 int ret;
1869
1870 WARN_ON_ONCE(!tctx);
1871
Jens Axboe0b81e802021-02-16 10:33:53 -07001872 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001873 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001874 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001875
1876 /* task_work already pending, we're done */
1877 if (test_bit(0, &tctx->task_state) ||
1878 test_and_set_bit(0, &tctx->task_state))
1879 return 0;
1880
1881 if (!task_work_add(tsk, &tctx->task_work, notify))
1882 return 0;
1883
1884 /*
1885 * Slow path - we failed, find and delete work. if the work is not
1886 * in the list, it got run and we're fine.
1887 */
1888 ret = 0;
Jens Axboe0b81e802021-02-16 10:33:53 -07001889 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001890 wq_list_for_each(node, prev, &tctx->task_list) {
1891 if (&req->io_task_work.node == node) {
1892 wq_list_del(&tctx->task_list, node, prev);
1893 ret = 1;
1894 break;
1895 }
1896 }
Jens Axboe0b81e802021-02-16 10:33:53 -07001897 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001898 clear_bit(0, &tctx->task_state);
1899 return ret;
1900}
1901
Jens Axboe355fb9e2020-10-22 20:19:35 -06001902static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboec2c4c832020-07-01 15:37:11 -06001903{
1904 struct task_struct *tsk = req->task;
1905 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe91989c72020-10-16 09:02:26 -06001906 enum task_work_notify_mode notify;
1907 int ret;
Jens Axboec2c4c832020-07-01 15:37:11 -06001908
Jens Axboe6200b0a2020-09-13 14:38:30 -06001909 if (tsk->flags & PF_EXITING)
1910 return -ESRCH;
1911
Jens Axboec2c4c832020-07-01 15:37:11 -06001912 /*
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001913 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
1914 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
1915 * processing task_work. There's no reliable way to tell if TWA_RESUME
1916 * will do the job.
Jens Axboec2c4c832020-07-01 15:37:11 -06001917 */
Jens Axboe91989c72020-10-16 09:02:26 -06001918 notify = TWA_NONE;
Jens Axboe355fb9e2020-10-22 20:19:35 -06001919 if (!(ctx->flags & IORING_SETUP_SQPOLL))
Jens Axboec2c4c832020-07-01 15:37:11 -06001920 notify = TWA_SIGNAL;
1921
Jens Axboe7cbf1722021-02-10 00:03:20 +00001922 ret = io_task_work_add(tsk, req, notify);
Jens Axboec2c4c832020-07-01 15:37:11 -06001923 if (!ret)
1924 wake_up_process(tsk);
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001925
Jens Axboec2c4c832020-07-01 15:37:11 -06001926 return ret;
1927}
1928
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001929static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00001930 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001931{
Jens Axboe7c25c0d2021-02-16 07:17:00 -07001932 struct io_ring_ctx *ctx = req->ctx;
1933 struct callback_head *head;
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001934
1935 init_task_work(&req->task_work, cb);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07001936 do {
1937 head = READ_ONCE(ctx->exit_task_work);
1938 req->task_work.next = head;
1939 } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001940}
1941
Jens Axboec40f6372020-06-25 15:39:59 -06001942static void __io_req_task_cancel(struct io_kiocb *req, int error)
1943{
1944 struct io_ring_ctx *ctx = req->ctx;
1945
1946 spin_lock_irq(&ctx->completion_lock);
1947 io_cqring_fill_event(req, error);
1948 io_commit_cqring(ctx);
1949 spin_unlock_irq(&ctx->completion_lock);
1950
1951 io_cqring_ev_posted(ctx);
1952 req_set_fail_links(req);
1953 io_double_put_req(req);
1954}
1955
1956static void io_req_task_cancel(struct callback_head *cb)
1957{
1958 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06001959 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06001960
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00001961 mutex_lock(&ctx->uring_lock);
Pavel Begunkova3df76982021-02-18 22:32:52 +00001962 __io_req_task_cancel(req, req->result);
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00001963 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06001964 percpu_ref_put(&ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06001965}
1966
1967static void __io_req_task_submit(struct io_kiocb *req)
1968{
1969 struct io_ring_ctx *ctx = req->ctx;
1970
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00001971 /* ctx stays valid until unlock, even if we drop all ours ctx->refs */
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001972 mutex_lock(&ctx->uring_lock);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00001973 if (!(current->flags & PF_EXITING) && !current->in_execve)
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001974 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001975 else
Jens Axboec40f6372020-06-25 15:39:59 -06001976 __io_req_task_cancel(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001977 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06001978}
1979
Jens Axboec40f6372020-06-25 15:39:59 -06001980static void io_req_task_submit(struct callback_head *cb)
1981{
1982 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
1983
1984 __io_req_task_submit(req);
1985}
1986
1987static void io_req_task_queue(struct io_kiocb *req)
1988{
Jens Axboec40f6372020-06-25 15:39:59 -06001989 int ret;
1990
Jens Axboe7cbf1722021-02-10 00:03:20 +00001991 req->task_work.func = io_req_task_submit;
Jens Axboe355fb9e2020-10-22 20:19:35 -06001992 ret = io_req_task_work_add(req);
Jens Axboec40f6372020-06-25 15:39:59 -06001993 if (unlikely(ret)) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00001994 req->result = -ECANCELED;
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00001995 percpu_ref_get(&req->ctx->refs);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001996 io_req_task_work_add_fallback(req, io_req_task_cancel);
Jens Axboec40f6372020-06-25 15:39:59 -06001997 }
Jens Axboec40f6372020-06-25 15:39:59 -06001998}
1999
Pavel Begunkova3df76982021-02-18 22:32:52 +00002000static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
2001{
2002 percpu_ref_get(&req->ctx->refs);
2003 req->result = ret;
2004 req->task_work.func = io_req_task_cancel;
2005
2006 if (unlikely(io_req_task_work_add(req)))
2007 io_req_task_work_add_fallback(req, io_req_task_cancel);
2008}
2009
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002010static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002011{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002012 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002013
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002014 if (nxt)
2015 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002016}
2017
Jens Axboe9e645e112019-05-10 16:07:28 -06002018static void io_free_req(struct io_kiocb *req)
2019{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002020 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002021 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002022}
2023
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002024struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002025 struct task_struct *task;
2026 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002027 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002028};
2029
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002030static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002031{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002032 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002033 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002034 rb->task = NULL;
2035}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002036
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002037static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2038 struct req_batch *rb)
2039{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002040 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002041 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002042 if (rb->ctx_refs)
2043 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002044}
2045
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002046static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2047 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002048{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002049 io_queue_next(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002050
Jens Axboee3bc8e92020-09-24 08:45:57 -06002051 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002052 if (rb->task)
2053 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002054 rb->task = req->task;
2055 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002056 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002057 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002058 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002059
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002060 io_dismantle_req(req);
Pavel Begunkovbd759042021-02-12 03:23:50 +00002061 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002062 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002063 else
2064 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002065}
2066
Pavel Begunkov905c1722021-02-10 00:03:14 +00002067static void io_submit_flush_completions(struct io_comp_state *cs,
2068 struct io_ring_ctx *ctx)
2069{
2070 int i, nr = cs->nr;
2071 struct io_kiocb *req;
2072 struct req_batch rb;
2073
2074 io_init_req_batch(&rb);
2075 spin_lock_irq(&ctx->completion_lock);
2076 for (i = 0; i < nr; i++) {
2077 req = cs->reqs[i];
2078 __io_cqring_fill_event(req, req->result, req->compl.cflags);
2079 }
2080 io_commit_cqring(ctx);
2081 spin_unlock_irq(&ctx->completion_lock);
2082
2083 io_cqring_ev_posted(ctx);
2084 for (i = 0; i < nr; i++) {
2085 req = cs->reqs[i];
2086
2087 /* submission and completion refs */
2088 if (refcount_sub_and_test(2, &req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002089 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002090 }
2091
2092 io_req_free_batch_finish(ctx, &rb);
2093 cs->nr = 0;
Jens Axboee65ef562019-03-12 10:16:44 -06002094}
2095
Jens Axboeba816ad2019-09-28 11:36:45 -06002096/*
2097 * Drop reference to request, return next in chain (if there is one) if this
2098 * was the last reference to this request.
2099 */
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002100static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002101{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002102 struct io_kiocb *nxt = NULL;
2103
Jens Axboe2a44f462020-02-25 13:25:41 -07002104 if (refcount_dec_and_test(&req->refs)) {
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002105 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002106 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002107 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002108 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002109}
2110
Jens Axboe2b188cc2019-01-07 10:46:33 -07002111static void io_put_req(struct io_kiocb *req)
2112{
Jens Axboedef596e2019-01-09 08:59:42 -07002113 if (refcount_dec_and_test(&req->refs))
2114 io_free_req(req);
2115}
2116
Pavel Begunkov216578e2020-10-13 09:44:00 +01002117static void io_put_req_deferred_cb(struct callback_head *cb)
2118{
2119 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2120
2121 io_free_req(req);
2122}
2123
2124static void io_free_req_deferred(struct io_kiocb *req)
2125{
2126 int ret;
2127
Jens Axboe7cbf1722021-02-10 00:03:20 +00002128 req->task_work.func = io_put_req_deferred_cb;
Jens Axboe355fb9e2020-10-22 20:19:35 -06002129 ret = io_req_task_work_add(req);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002130 if (unlikely(ret))
2131 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002132}
2133
2134static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2135{
2136 if (refcount_sub_and_test(refs, &req->refs))
2137 io_free_req_deferred(req);
2138}
2139
Jens Axboe978db572019-11-14 22:39:04 -07002140static void io_double_put_req(struct io_kiocb *req)
2141{
2142 /* drop both submit and complete references */
2143 if (refcount_sub_and_test(2, &req->refs))
2144 io_free_req(req);
2145}
2146
Pavel Begunkov6c503152021-01-04 20:36:36 +00002147static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002148{
2149 /* See comment at the top of this file */
2150 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002151 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002152}
2153
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002154static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2155{
2156 struct io_rings *rings = ctx->rings;
2157
2158 /* make sure SQ entry isn't read before tail */
2159 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2160}
2161
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002162static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002163{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002164 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002165
Jens Axboebcda7ba2020-02-23 16:42:51 -07002166 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2167 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002168 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002169 kfree(kbuf);
2170 return cflags;
2171}
2172
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002173static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2174{
2175 struct io_buffer *kbuf;
2176
2177 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2178 return io_put_kbuf(req, kbuf);
2179}
2180
Jens Axboe4c6e2772020-07-01 11:29:10 -06002181static inline bool io_run_task_work(void)
2182{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002183 /*
2184 * Not safe to run on exiting task, and the task_work handling will
2185 * not add work to such a task.
2186 */
2187 if (unlikely(current->flags & PF_EXITING))
2188 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002189 if (current->task_works) {
2190 __set_current_state(TASK_RUNNING);
2191 task_work_run();
2192 return true;
2193 }
2194
2195 return false;
2196}
2197
Jens Axboedef596e2019-01-09 08:59:42 -07002198/*
2199 * Find and free completed poll iocbs
2200 */
2201static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2202 struct list_head *done)
2203{
Jens Axboe8237e042019-12-28 10:48:22 -07002204 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002205 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002206
2207 /* order with ->result store in io_complete_rw_iopoll() */
2208 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002209
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002210 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002211 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002212 int cflags = 0;
2213
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002214 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002215 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002216
Pavel Begunkovf1613402021-02-11 18:28:21 +00002217 if (READ_ONCE(req->result) == -EAGAIN) {
2218 req->iopoll_completed = 0;
Pavel Begunkov23faba32021-02-11 18:28:22 +00002219 if (io_rw_reissue(req))
Pavel Begunkovf1613402021-02-11 18:28:21 +00002220 continue;
2221 }
2222
Jens Axboebcda7ba2020-02-23 16:42:51 -07002223 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002224 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002225
2226 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002227 (*nr_events)++;
2228
Pavel Begunkovc3524382020-06-28 12:52:32 +03002229 if (refcount_dec_and_test(&req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002230 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002231 }
Jens Axboedef596e2019-01-09 08:59:42 -07002232
Jens Axboe09bb8392019-03-13 12:39:28 -06002233 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002234 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002235 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002236}
2237
Jens Axboedef596e2019-01-09 08:59:42 -07002238static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2239 long min)
2240{
2241 struct io_kiocb *req, *tmp;
2242 LIST_HEAD(done);
2243 bool spin;
2244 int ret;
2245
2246 /*
2247 * Only spin for completions if we don't have multiple devices hanging
2248 * off our complete list, and we're under the requested amount.
2249 */
2250 spin = !ctx->poll_multi_file && *nr_events < min;
2251
2252 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002253 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002254 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002255
2256 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002257 * Move completed and retryable entries to our local lists.
2258 * If we find a request that requires polling, break out
2259 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002260 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002261 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002262 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002263 continue;
2264 }
2265 if (!list_empty(&done))
2266 break;
2267
2268 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2269 if (ret < 0)
2270 break;
2271
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002272 /* iopoll may have completed current req */
2273 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002274 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002275
Jens Axboedef596e2019-01-09 08:59:42 -07002276 if (ret && spin)
2277 spin = false;
2278 ret = 0;
2279 }
2280
2281 if (!list_empty(&done))
2282 io_iopoll_complete(ctx, nr_events, &done);
2283
2284 return ret;
2285}
2286
2287/*
Brian Gianforcarod195a662019-12-13 03:09:50 -08002288 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
Jens Axboedef596e2019-01-09 08:59:42 -07002289 * non-spinning poll check - we'll still enter the driver poll loop, but only
2290 * as a non-spinning completion check.
2291 */
2292static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
2293 long min)
2294{
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002295 while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
Jens Axboedef596e2019-01-09 08:59:42 -07002296 int ret;
2297
2298 ret = io_do_iopoll(ctx, nr_events, min);
2299 if (ret < 0)
2300 return ret;
Pavel Begunkoveba0a4d2020-07-06 17:59:30 +03002301 if (*nr_events >= min)
Jens Axboedef596e2019-01-09 08:59:42 -07002302 return 0;
2303 }
2304
2305 return 1;
2306}
2307
2308/*
2309 * We can't just wait for polled events to come to us, we have to actively
2310 * find and complete them.
2311 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002312static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002313{
2314 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2315 return;
2316
2317 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002318 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002319 unsigned int nr_events = 0;
2320
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002321 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002322
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002323 /* let it sleep and repeat later if can't complete a request */
2324 if (nr_events == 0)
2325 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002326 /*
2327 * Ensure we allow local-to-the-cpu processing to take place,
2328 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002329 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002330 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002331 if (need_resched()) {
2332 mutex_unlock(&ctx->uring_lock);
2333 cond_resched();
2334 mutex_lock(&ctx->uring_lock);
2335 }
Jens Axboedef596e2019-01-09 08:59:42 -07002336 }
2337 mutex_unlock(&ctx->uring_lock);
2338}
2339
Pavel Begunkov7668b922020-07-07 16:36:21 +03002340static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002341{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002342 unsigned int nr_events = 0;
Jens Axboe2b2ed972019-10-25 10:06:15 -06002343 int iters = 0, ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002344
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002345 /*
2346 * We disallow the app entering submit/complete with polling, but we
2347 * still need to lock the ring to prevent racing with polled issue
2348 * that got punted to a workqueue.
2349 */
2350 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002351 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002352 /*
Jens Axboea3a0e432019-08-20 11:03:11 -06002353 * Don't enter poll loop if we already have events pending.
2354 * If we do, we can potentially be spinning for commands that
2355 * already triggered a CQE (eg in error).
2356 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00002357 if (test_bit(0, &ctx->cq_check_overflow))
2358 __io_cqring_overflow_flush(ctx, false, NULL, NULL);
2359 if (io_cqring_events(ctx))
Jens Axboea3a0e432019-08-20 11:03:11 -06002360 break;
2361
2362 /*
Jens Axboe500f9fb2019-08-19 12:15:59 -06002363 * If a submit got punted to a workqueue, we can have the
2364 * application entering polling for a command before it gets
2365 * issued. That app will hold the uring_lock for the duration
2366 * of the poll right here, so we need to take a breather every
2367 * now and then to ensure that the issue has a chance to add
2368 * the poll to the issued list. Otherwise we can spin here
2369 * forever, while the workqueue is stuck trying to acquire the
2370 * very same mutex.
2371 */
2372 if (!(++iters & 7)) {
2373 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002374 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002375 mutex_lock(&ctx->uring_lock);
2376 }
2377
Pavel Begunkov7668b922020-07-07 16:36:21 +03002378 ret = io_iopoll_getevents(ctx, &nr_events, min);
Jens Axboedef596e2019-01-09 08:59:42 -07002379 if (ret <= 0)
2380 break;
2381 ret = 0;
Pavel Begunkov7668b922020-07-07 16:36:21 +03002382 } while (min && !nr_events && !need_resched());
Jens Axboedef596e2019-01-09 08:59:42 -07002383
Jens Axboe500f9fb2019-08-19 12:15:59 -06002384 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002385 return ret;
2386}
2387
Jens Axboe491381ce2019-10-17 09:20:46 -06002388static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002389{
Jens Axboe491381ce2019-10-17 09:20:46 -06002390 /*
2391 * Tell lockdep we inherited freeze protection from submission
2392 * thread.
2393 */
2394 if (req->flags & REQ_F_ISREG) {
2395 struct inode *inode = file_inode(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002396
Jens Axboe491381ce2019-10-17 09:20:46 -06002397 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002398 }
Jens Axboe491381ce2019-10-17 09:20:46 -06002399 file_end_write(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002400}
2401
Jens Axboeb63534c2020-06-04 11:28:00 -06002402#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002403static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002404{
2405 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Colin Ian King4a245472021-02-10 20:00:07 +00002406 int rw, ret;
Jens Axboeb63534c2020-06-04 11:28:00 -06002407 struct iov_iter iter;
Jens Axboeb63534c2020-06-04 11:28:00 -06002408
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002409 /* already prepared */
2410 if (req->async_data)
2411 return true;
Jens Axboeb63534c2020-06-04 11:28:00 -06002412
2413 switch (req->opcode) {
2414 case IORING_OP_READV:
2415 case IORING_OP_READ_FIXED:
2416 case IORING_OP_READ:
2417 rw = READ;
2418 break;
2419 case IORING_OP_WRITEV:
2420 case IORING_OP_WRITE_FIXED:
2421 case IORING_OP_WRITE:
2422 rw = WRITE;
2423 break;
2424 default:
2425 printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n",
2426 req->opcode);
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002427 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002428 }
2429
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002430 ret = io_import_iovec(rw, req, &iovec, &iter, false);
2431 if (ret < 0)
2432 return false;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00002433 return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false);
Jens Axboeb63534c2020-06-04 11:28:00 -06002434}
Jens Axboeb63534c2020-06-04 11:28:00 -06002435
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002436static bool io_rw_should_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002437{
Jens Axboe355afae2020-09-02 09:30:31 -06002438 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002439 struct io_ring_ctx *ctx = req->ctx;
Jens Axboeb63534c2020-06-04 11:28:00 -06002440
Jens Axboe355afae2020-09-02 09:30:31 -06002441 if (!S_ISBLK(mode) && !S_ISREG(mode))
2442 return false;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002443 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() &&
2444 !(ctx->flags & IORING_SETUP_IOPOLL)))
Jens Axboeb63534c2020-06-04 11:28:00 -06002445 return false;
Jens Axboe7c977a52021-02-23 19:17:35 -07002446 /*
2447 * If ref is dying, we might be running poll reap from the exit work.
2448 * Don't attempt to reissue from that path, just let it fail with
2449 * -EAGAIN.
2450 */
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002451 if (percpu_ref_is_dying(&ctx->refs))
2452 return false;
2453 return true;
2454}
2455#endif
2456
2457static bool io_rw_reissue(struct io_kiocb *req)
2458{
2459#ifdef CONFIG_BLOCK
2460 if (!io_rw_should_reissue(req))
Jens Axboe7c977a52021-02-23 19:17:35 -07002461 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002462
Pavel Begunkov55e6ac12021-01-08 20:57:22 +00002463 lockdep_assert_held(&req->ctx->uring_lock);
2464
Jens Axboe37d1e2e2021-02-17 21:03:43 -07002465 if (io_resubmit_prep(req)) {
Jens Axboefdee9462020-08-27 16:46:24 -06002466 refcount_inc(&req->refs);
2467 io_queue_async_work(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002468 return true;
Jens Axboefdee9462020-08-27 16:46:24 -06002469 }
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002470 req_set_fail_links(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002471#endif
2472 return false;
2473}
2474
Jens Axboea1d7c392020-06-22 11:09:46 -06002475static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002476 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002477{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002478 int cflags = 0;
2479
Pavel Begunkov23faba32021-02-11 18:28:22 +00002480 if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req))
2481 return;
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002482 if (res != req->result)
2483 req_set_fail_links(req);
Pavel Begunkov23faba32021-02-11 18:28:22 +00002484
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002485 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2486 kiocb_end_write(req);
2487 if (req->flags & REQ_F_BUFFER_SELECTED)
2488 cflags = io_put_rw_kbuf(req);
2489 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002490}
2491
2492static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2493{
Jens Axboe9adbd452019-12-20 08:45:55 -07002494 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002495
Pavel Begunkov889fca72021-02-10 00:03:09 +00002496 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002497}
2498
Jens Axboedef596e2019-01-09 08:59:42 -07002499static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2500{
Jens Axboe9adbd452019-12-20 08:45:55 -07002501 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002502
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002503#ifdef CONFIG_BLOCK
2504 /* Rewind iter, if we have one. iopoll path resubmits as usual */
2505 if (res == -EAGAIN && io_rw_should_reissue(req)) {
2506 struct io_async_rw *rw = req->async_data;
2507
2508 if (rw)
2509 iov_iter_revert(&rw->iter,
2510 req->result - iov_iter_count(&rw->iter));
2511 else if (!io_resubmit_prep(req))
2512 res = -EIO;
2513 }
2514#endif
2515
Jens Axboe491381ce2019-10-17 09:20:46 -06002516 if (kiocb->ki_flags & IOCB_WRITE)
2517 kiocb_end_write(req);
Jens Axboedef596e2019-01-09 08:59:42 -07002518
Xiaoguang Wang2d7d6792020-06-16 02:06:37 +08002519 if (res != -EAGAIN && res != req->result)
Jens Axboe4e88d6e2019-12-07 20:59:47 -07002520 req_set_fail_links(req);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002521
2522 WRITE_ONCE(req->result, res);
2523 /* order with io_poll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002524 smp_wmb();
2525 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002526}
2527
2528/*
2529 * After the iocb has been issued, it's safe to be found on the poll list.
2530 * Adding the kiocb to the list AFTER submission ensures that we don't
2531 * find it from a io_iopoll_getevents() thread before the issuer is done
2532 * accessing the kiocb cookie.
2533 */
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002534static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
Jens Axboedef596e2019-01-09 08:59:42 -07002535{
2536 struct io_ring_ctx *ctx = req->ctx;
2537
2538 /*
2539 * Track whether we have multiple files in our lists. This will impact
2540 * how we do polling eventually, not spinning if we're on potentially
2541 * different devices.
2542 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002543 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002544 ctx->poll_multi_file = false;
2545 } else if (!ctx->poll_multi_file) {
2546 struct io_kiocb *list_req;
2547
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002548 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002549 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002550 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002551 ctx->poll_multi_file = true;
2552 }
2553
2554 /*
2555 * For fast devices, IO may have already completed. If it has, add
2556 * it to the front so we find it first.
2557 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002558 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002559 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002560 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002561 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002562
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002563 /*
2564 * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread
2565 * task context or in io worker task context. If current task context is
2566 * sq thread, we don't need to check whether should wake up sq thread.
2567 */
2568 if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002569 wq_has_sleeper(&ctx->sq_data->wait))
2570 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002571}
2572
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002573static inline void io_state_file_put(struct io_submit_state *state)
2574{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002575 if (state->file_refs) {
2576 fput_many(state->file, state->file_refs);
2577 state->file_refs = 0;
2578 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002579}
2580
2581/*
2582 * Get as many references to a file as we have IOs left in this submission,
2583 * assuming most submissions are for one file, or at least that each file
2584 * has more than one submission.
2585 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002586static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002587{
2588 if (!state)
2589 return fget(fd);
2590
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002591 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002592 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002593 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002594 return state->file;
2595 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002596 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002597 }
2598 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002599 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002600 return NULL;
2601
2602 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002603 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002604 return state->file;
2605}
2606
Jens Axboe4503b762020-06-01 10:00:27 -06002607static bool io_bdev_nowait(struct block_device *bdev)
2608{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002609 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002610}
2611
Jens Axboe2b188cc2019-01-07 10:46:33 -07002612/*
2613 * If we tracked the file through the SCM inflight mechanism, we could support
2614 * any file. For now, just ensure that anything potentially problematic is done
2615 * inline.
2616 */
Jens Axboeaf197f52020-04-28 13:15:06 -06002617static bool io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002618{
2619 umode_t mode = file_inode(file)->i_mode;
2620
Jens Axboe4503b762020-06-01 10:00:27 -06002621 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002622 if (IS_ENABLED(CONFIG_BLOCK) &&
2623 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002624 return true;
2625 return false;
2626 }
2627 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002628 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002629 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002630 if (IS_ENABLED(CONFIG_BLOCK) &&
2631 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002632 file->f_op != &io_uring_fops)
2633 return true;
2634 return false;
2635 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002636
Jens Axboec5b85622020-06-09 19:23:05 -06002637 /* any ->read/write should understand O_NONBLOCK */
2638 if (file->f_flags & O_NONBLOCK)
2639 return true;
2640
Jens Axboeaf197f52020-04-28 13:15:06 -06002641 if (!(file->f_mode & FMODE_NOWAIT))
2642 return false;
2643
2644 if (rw == READ)
2645 return file->f_op->read_iter != NULL;
2646
2647 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002648}
2649
Pavel Begunkova88fc402020-09-30 22:57:53 +03002650static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002651{
Jens Axboedef596e2019-01-09 08:59:42 -07002652 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002653 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002654 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002655 unsigned ioprio;
2656 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002657
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002658 if (S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002659 req->flags |= REQ_F_ISREG;
2660
Jens Axboe2b188cc2019-01-07 10:46:33 -07002661 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002662 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002663 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002664 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002665 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002666 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002667 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2668 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2669 if (unlikely(ret))
2670 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002671
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002672 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
2673 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
2674 req->flags |= REQ_F_NOWAIT;
2675
Jens Axboe2b188cc2019-01-07 10:46:33 -07002676 ioprio = READ_ONCE(sqe->ioprio);
2677 if (ioprio) {
2678 ret = ioprio_check_cap(ioprio);
2679 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002680 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002681
2682 kiocb->ki_ioprio = ioprio;
2683 } else
2684 kiocb->ki_ioprio = get_current_ioprio();
2685
Jens Axboedef596e2019-01-09 08:59:42 -07002686 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002687 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2688 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002689 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002690
Jens Axboedef596e2019-01-09 08:59:42 -07002691 kiocb->ki_flags |= IOCB_HIPRI;
2692 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002693 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002694 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002695 if (kiocb->ki_flags & IOCB_HIPRI)
2696 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002697 kiocb->ki_complete = io_complete_rw;
2698 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002699
Jens Axboe3529d8c2019-12-19 18:24:38 -07002700 req->rw.addr = READ_ONCE(sqe->addr);
2701 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002702 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002703 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002704}
2705
2706static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2707{
2708 switch (ret) {
2709 case -EIOCBQUEUED:
2710 break;
2711 case -ERESTARTSYS:
2712 case -ERESTARTNOINTR:
2713 case -ERESTARTNOHAND:
2714 case -ERESTART_RESTARTBLOCK:
2715 /*
2716 * We can't just restart the syscall, since previously
2717 * submitted sqes may already be in progress. Just fail this
2718 * IO with EINTR.
2719 */
2720 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002721 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002722 default:
2723 kiocb->ki_complete(kiocb, ret, 0);
2724 }
2725}
2726
Jens Axboea1d7c392020-06-22 11:09:46 -06002727static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002728 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06002729{
Jens Axboeba042912019-12-25 16:33:42 -07002730 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002731 struct io_async_rw *io = req->async_data;
Jens Axboeba042912019-12-25 16:33:42 -07002732
Jens Axboe227c0c92020-08-13 11:51:40 -06002733 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002734 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002735 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002736 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002737 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002738 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002739 }
2740
Jens Axboeba042912019-12-25 16:33:42 -07002741 if (req->flags & REQ_F_CUR_POS)
2742 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002743 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00002744 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002745 else
2746 io_rw_done(kiocb, ret);
2747}
2748
Pavel Begunkov847595d2021-02-04 13:52:06 +00002749static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07002750{
Jens Axboe9adbd452019-12-20 08:45:55 -07002751 struct io_ring_ctx *ctx = req->ctx;
2752 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002753 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03002754 u16 index, buf_index = req->buf_index;
Jens Axboeedafcce2019-01-09 09:16:05 -07002755 size_t offset;
2756 u64 buf_addr;
2757
Jens Axboeedafcce2019-01-09 09:16:05 -07002758 if (unlikely(buf_index >= ctx->nr_user_bufs))
2759 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07002760 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2761 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07002762 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002763
2764 /* overflow */
2765 if (buf_addr + len < buf_addr)
2766 return -EFAULT;
2767 /* not inside the mapped region */
2768 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
2769 return -EFAULT;
2770
2771 /*
2772 * May not be a start of buffer, set size appropriately
2773 * and advance us to the beginning.
2774 */
2775 offset = buf_addr - imu->ubuf;
2776 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002777
2778 if (offset) {
2779 /*
2780 * Don't use iov_iter_advance() here, as it's really slow for
2781 * using the latter parts of a big fixed buffer - it iterates
2782 * over each segment manually. We can cheat a bit here, because
2783 * we know that:
2784 *
2785 * 1) it's a BVEC iter, we set it up
2786 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2787 * first and last bvec
2788 *
2789 * So just find our index, and adjust the iterator afterwards.
2790 * If the offset is within the first bvec (or the whole first
2791 * bvec, just use iov_iter_advance(). This makes it easier
2792 * since we can just skip the first segment, which may not
2793 * be PAGE_SIZE aligned.
2794 */
2795 const struct bio_vec *bvec = imu->bvec;
2796
2797 if (offset <= bvec->bv_len) {
2798 iov_iter_advance(iter, offset);
2799 } else {
2800 unsigned long seg_skip;
2801
2802 /* skip first vec */
2803 offset -= bvec->bv_len;
2804 seg_skip = 1 + (offset >> PAGE_SHIFT);
2805
2806 iter->bvec = bvec + seg_skip;
2807 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002808 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002809 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002810 }
2811 }
2812
Pavel Begunkov847595d2021-02-04 13:52:06 +00002813 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07002814}
2815
Jens Axboebcda7ba2020-02-23 16:42:51 -07002816static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2817{
2818 if (needs_lock)
2819 mutex_unlock(&ctx->uring_lock);
2820}
2821
2822static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2823{
2824 /*
2825 * "Normal" inline submissions always hold the uring_lock, since we
2826 * grab it from the system call. Same is true for the SQPOLL offload.
2827 * The only exception is when we've detached the request and issue it
2828 * from an async worker thread, grab the lock for that case.
2829 */
2830 if (needs_lock)
2831 mutex_lock(&ctx->uring_lock);
2832}
2833
2834static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2835 int bgid, struct io_buffer *kbuf,
2836 bool needs_lock)
2837{
2838 struct io_buffer *head;
2839
2840 if (req->flags & REQ_F_BUFFER_SELECTED)
2841 return kbuf;
2842
2843 io_ring_submit_lock(req->ctx, needs_lock);
2844
2845 lockdep_assert_held(&req->ctx->uring_lock);
2846
2847 head = idr_find(&req->ctx->io_buffer_idr, bgid);
2848 if (head) {
2849 if (!list_empty(&head->list)) {
2850 kbuf = list_last_entry(&head->list, struct io_buffer,
2851 list);
2852 list_del(&kbuf->list);
2853 } else {
2854 kbuf = head;
2855 idr_remove(&req->ctx->io_buffer_idr, bgid);
2856 }
2857 if (*len > kbuf->len)
2858 *len = kbuf->len;
2859 } else {
2860 kbuf = ERR_PTR(-ENOBUFS);
2861 }
2862
2863 io_ring_submit_unlock(req->ctx, needs_lock);
2864
2865 return kbuf;
2866}
2867
Jens Axboe4d954c22020-02-27 07:31:19 -07002868static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2869 bool needs_lock)
2870{
2871 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002872 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002873
2874 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002875 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002876 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2877 if (IS_ERR(kbuf))
2878 return kbuf;
2879 req->rw.addr = (u64) (unsigned long) kbuf;
2880 req->flags |= REQ_F_BUFFER_SELECTED;
2881 return u64_to_user_ptr(kbuf->addr);
2882}
2883
2884#ifdef CONFIG_COMPAT
2885static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2886 bool needs_lock)
2887{
2888 struct compat_iovec __user *uiov;
2889 compat_ssize_t clen;
2890 void __user *buf;
2891 ssize_t len;
2892
2893 uiov = u64_to_user_ptr(req->rw.addr);
2894 if (!access_ok(uiov, sizeof(*uiov)))
2895 return -EFAULT;
2896 if (__get_user(clen, &uiov->iov_len))
2897 return -EFAULT;
2898 if (clen < 0)
2899 return -EINVAL;
2900
2901 len = clen;
2902 buf = io_rw_buffer_select(req, &len, needs_lock);
2903 if (IS_ERR(buf))
2904 return PTR_ERR(buf);
2905 iov[0].iov_base = buf;
2906 iov[0].iov_len = (compat_size_t) len;
2907 return 0;
2908}
2909#endif
2910
2911static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2912 bool needs_lock)
2913{
2914 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
2915 void __user *buf;
2916 ssize_t len;
2917
2918 if (copy_from_user(iov, uiov, sizeof(*uiov)))
2919 return -EFAULT;
2920
2921 len = iov[0].iov_len;
2922 if (len < 0)
2923 return -EINVAL;
2924 buf = io_rw_buffer_select(req, &len, needs_lock);
2925 if (IS_ERR(buf))
2926 return PTR_ERR(buf);
2927 iov[0].iov_base = buf;
2928 iov[0].iov_len = len;
2929 return 0;
2930}
2931
2932static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2933 bool needs_lock)
2934{
Jens Axboedddb3e22020-06-04 11:27:01 -06002935 if (req->flags & REQ_F_BUFFER_SELECTED) {
2936 struct io_buffer *kbuf;
2937
2938 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2939 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
2940 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07002941 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06002942 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00002943 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07002944 return -EINVAL;
2945
2946#ifdef CONFIG_COMPAT
2947 if (req->ctx->compat)
2948 return io_compat_import(req, iov, needs_lock);
2949#endif
2950
2951 return __io_iov_buffer_select(req, iov, needs_lock);
2952}
2953
Pavel Begunkov847595d2021-02-04 13:52:06 +00002954static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
2955 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002956{
Jens Axboe9adbd452019-12-20 08:45:55 -07002957 void __user *buf = u64_to_user_ptr(req->rw.addr);
2958 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00002959 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07002960 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07002961
Pavel Begunkov7d009162019-11-25 23:14:40 +03002962 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07002963 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07002964 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07002965 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002966
Jens Axboebcda7ba2020-02-23 16:42:51 -07002967 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002968 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07002969 return -EINVAL;
2970
Jens Axboe3a6820f2019-12-22 15:19:35 -07002971 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002972 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07002973 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03002974 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07002975 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06002976 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002977 }
2978
Jens Axboe3a6820f2019-12-22 15:19:35 -07002979 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
2980 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00002981 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07002982 }
2983
Jens Axboe4d954c22020-02-27 07:31:19 -07002984 if (req->flags & REQ_F_BUFFER_SELECT) {
2985 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00002986 if (!ret)
2987 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07002988 *iovec = NULL;
2989 return ret;
2990 }
2991
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02002992 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
2993 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002994}
2995
Jens Axboe0fef9482020-08-26 10:36:20 -06002996static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
2997{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03002998 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06002999}
3000
Jens Axboe32960612019-09-23 11:05:34 -06003001/*
3002 * For files that don't have ->read_iter() and ->write_iter(), handle them
3003 * by looping over ->read() or ->write() manually.
3004 */
Jens Axboe4017eb92020-10-22 14:14:12 -06003005static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06003006{
Jens Axboe4017eb92020-10-22 14:14:12 -06003007 struct kiocb *kiocb = &req->rw.kiocb;
3008 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003009 ssize_t ret = 0;
3010
3011 /*
3012 * Don't support polled IO through this interface, and we can't
3013 * support non-blocking either. For the latter, this just causes
3014 * the kiocb to be handled from an async context.
3015 */
3016 if (kiocb->ki_flags & IOCB_HIPRI)
3017 return -EOPNOTSUPP;
3018 if (kiocb->ki_flags & IOCB_NOWAIT)
3019 return -EAGAIN;
3020
3021 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003022 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003023 ssize_t nr;
3024
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003025 if (!iov_iter_is_bvec(iter)) {
3026 iovec = iov_iter_iovec(iter);
3027 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003028 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3029 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003030 }
3031
Jens Axboe32960612019-09-23 11:05:34 -06003032 if (rw == READ) {
3033 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003034 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003035 } else {
3036 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003037 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003038 }
3039
3040 if (nr < 0) {
3041 if (!ret)
3042 ret = nr;
3043 break;
3044 }
3045 ret += nr;
3046 if (nr != iovec.iov_len)
3047 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003048 req->rw.len -= nr;
3049 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003050 iov_iter_advance(iter, nr);
3051 }
3052
3053 return ret;
3054}
3055
Jens Axboeff6165b2020-08-13 09:47:43 -06003056static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3057 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003058{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003059 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003060
Jens Axboeff6165b2020-08-13 09:47:43 -06003061 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003062 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003063 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003064 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003065 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003066 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003067 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003068 unsigned iov_off = 0;
3069
3070 rw->iter.iov = rw->fast_iov;
3071 if (iter->iov != fast_iov) {
3072 iov_off = iter->iov - fast_iov;
3073 rw->iter.iov += iov_off;
3074 }
3075 if (rw->fast_iov != fast_iov)
3076 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003077 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003078 } else {
3079 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003080 }
3081}
3082
Jens Axboee8c2bc12020-08-15 18:44:09 -07003083static inline int __io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003084{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003085 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3086 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3087 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003088}
3089
Jens Axboee8c2bc12020-08-15 18:44:09 -07003090static int io_alloc_async_data(struct io_kiocb *req)
Jens Axboef67676d2019-12-02 11:03:47 -07003091{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003092 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboed3656342019-12-18 09:50:26 -07003093 return 0;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003094
Jens Axboee8c2bc12020-08-15 18:44:09 -07003095 return __io_alloc_async_data(req);
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003096}
3097
Jens Axboeff6165b2020-08-13 09:47:43 -06003098static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3099 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003100 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003101{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003102 if (!force && !io_op_defs[req->opcode].needs_async_data)
Jens Axboe74566df2020-01-13 19:23:24 -07003103 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003104 if (!req->async_data) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003105 if (__io_alloc_async_data(req)) {
3106 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003107 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003108 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003109
Jens Axboeff6165b2020-08-13 09:47:43 -06003110 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003111 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003112 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003113}
3114
Pavel Begunkov73debe62020-09-30 22:57:54 +03003115static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003116{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003117 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003118 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003119 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003120
Pavel Begunkov2846c482020-11-07 13:16:27 +00003121 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003122 if (unlikely(ret < 0))
3123 return ret;
3124
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003125 iorw->bytes_done = 0;
3126 iorw->free_iovec = iov;
3127 if (iov)
3128 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003129 return 0;
3130}
3131
Pavel Begunkov73debe62020-09-30 22:57:54 +03003132static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003133{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003134 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3135 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003136 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003137}
3138
Jens Axboec1dd91d2020-08-03 16:43:59 -06003139/*
3140 * This is our waitqueue callback handler, registered through lock_page_async()
3141 * when we initially tried to do the IO with the iocb armed our waitqueue.
3142 * This gets called when the page is unlocked, and we generally expect that to
3143 * happen when the page IO is completed and the page is now uptodate. This will
3144 * queue a task_work based retry of the operation, attempting to copy the data
3145 * again. If the latter fails because the page was NOT uptodate, then we will
3146 * do a thread based blocking retry of the operation. That's the unexpected
3147 * slow path.
3148 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003149static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3150 int sync, void *arg)
3151{
3152 struct wait_page_queue *wpq;
3153 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003154 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003155
3156 wpq = container_of(wait, struct wait_page_queue, wait);
3157
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003158 if (!wake_page_match(wpq, key))
3159 return 0;
3160
Hao Xuc8d317a2020-09-29 20:00:45 +08003161 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003162 list_del_init(&wait->entry);
3163
Jens Axboebcf5a062020-05-22 09:24:42 -06003164 /* submit ref gets dropped, acquire a new one */
3165 refcount_inc(&req->refs);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003166 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003167 return 1;
3168}
3169
Jens Axboec1dd91d2020-08-03 16:43:59 -06003170/*
3171 * This controls whether a given IO request should be armed for async page
3172 * based retry. If we return false here, the request is handed to the async
3173 * worker threads for retry. If we're doing buffered reads on a regular file,
3174 * we prepare a private wait_page_queue entry and retry the operation. This
3175 * will either succeed because the page is now uptodate and unlocked, or it
3176 * will register a callback when the page is unlocked at IO completion. Through
3177 * that callback, io_uring uses task_work to setup a retry of the operation.
3178 * That retry will attempt the buffered read again. The retry will generally
3179 * succeed, or in rare cases where it fails, we then fall back to using the
3180 * async worker threads for a blocking retry.
3181 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003182static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003183{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003184 struct io_async_rw *rw = req->async_data;
3185 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003186 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003187
3188 /* never retry for NOWAIT, we just complete with -EAGAIN */
3189 if (req->flags & REQ_F_NOWAIT)
3190 return false;
3191
Jens Axboe227c0c92020-08-13 11:51:40 -06003192 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003193 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003194 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003195
Jens Axboebcf5a062020-05-22 09:24:42 -06003196 /*
3197 * just use poll if we can, and don't attempt if the fs doesn't
3198 * support callback based unlocks
3199 */
3200 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3201 return false;
3202
Jens Axboe3b2a4432020-08-16 10:58:43 -07003203 wait->wait.func = io_async_buf_func;
3204 wait->wait.private = req;
3205 wait->wait.flags = 0;
3206 INIT_LIST_HEAD(&wait->wait.entry);
3207 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003208 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003209 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003210 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003211}
3212
3213static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3214{
3215 if (req->file->f_op->read_iter)
3216 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003217 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003218 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003219 else
3220 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003221}
3222
Pavel Begunkov889fca72021-02-10 00:03:09 +00003223static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003224{
3225 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003226 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003227 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003228 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003229 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003230 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003231
Pavel Begunkov2846c482020-11-07 13:16:27 +00003232 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003233 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003234 iovec = NULL;
3235 } else {
3236 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3237 if (ret < 0)
3238 return ret;
3239 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003240 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003241 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003242
Jens Axboefd6c2e42019-12-18 12:19:41 -07003243 /* Ensure we clear previously set non-block flag */
3244 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003245 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003246 else
3247 kiocb->ki_flags |= IOCB_NOWAIT;
3248
Pavel Begunkov24c74672020-06-21 13:09:51 +03003249 /* If the file doesn't support async, just async punt */
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003250 if (force_nonblock && !io_file_supports_async(req->file, READ)) {
3251 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003252 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003253 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003254
Pavel Begunkov632546c2020-11-07 13:16:26 +00003255 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003256 if (unlikely(ret)) {
3257 kfree(iovec);
3258 return ret;
3259 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003260
Jens Axboe227c0c92020-08-13 11:51:40 -06003261 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003262
Pavel Begunkov57cd6572021-02-01 18:59:56 +00003263 if (ret == -EIOCBQUEUED) {
Jens Axboe3e6a0d32021-03-01 13:56:00 -07003264 if (req->async_data)
3265 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003266 goto out_free;
Jens Axboe227c0c92020-08-13 11:51:40 -06003267 } else if (ret == -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003268 /* IOPOLL retry should happen for io-wq threads */
3269 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003270 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003271 /* no retry on NONBLOCK nor RWF_NOWAIT */
3272 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003273 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003274 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003275 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003276 ret = 0;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003277 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003278 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003279 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003280 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003281 }
3282
Jens Axboe227c0c92020-08-13 11:51:40 -06003283 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003284 if (ret2)
3285 return ret2;
3286
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003287 iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003288 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003289 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003290 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003291
Pavel Begunkovb23df912021-02-04 13:52:04 +00003292 do {
3293 io_size -= ret;
3294 rw->bytes_done += ret;
3295 /* if we can retry, do so with the callbacks armed */
3296 if (!io_rw_should_retry(req)) {
3297 kiocb->ki_flags &= ~IOCB_WAITQ;
3298 return -EAGAIN;
3299 }
3300
3301 /*
3302 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3303 * we get -EIOCBQUEUED, then we'll get a notification when the
3304 * desired page gets unlocked. We can also get a partial read
3305 * here, and if we do, then just retry at the new offset.
3306 */
3307 ret = io_iter_do_read(req, iter);
3308 if (ret == -EIOCBQUEUED)
3309 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003310 /* we got some bytes, but not all. retry. */
Jens Axboeb5b0ecb2021-03-04 21:02:58 -07003311 kiocb->ki_flags &= ~IOCB_WAITQ;
Pavel Begunkovb23df912021-02-04 13:52:04 +00003312 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003313done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003314 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003315out_free:
3316 /* it's faster to check here then delegate to kfree */
3317 if (iovec)
3318 kfree(iovec);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003319 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003320}
3321
Pavel Begunkov73debe62020-09-30 22:57:54 +03003322static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003323{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003324 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3325 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003326 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003327}
3328
Pavel Begunkov889fca72021-02-10 00:03:09 +00003329static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003330{
3331 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003332 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003333 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003334 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003335 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003336 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003337
Pavel Begunkov2846c482020-11-07 13:16:27 +00003338 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003339 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003340 iovec = NULL;
3341 } else {
3342 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3343 if (ret < 0)
3344 return ret;
3345 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003346 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003347 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003348
Jens Axboefd6c2e42019-12-18 12:19:41 -07003349 /* Ensure we clear previously set non-block flag */
3350 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003351 kiocb->ki_flags &= ~IOCB_NOWAIT;
3352 else
3353 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003354
Pavel Begunkov24c74672020-06-21 13:09:51 +03003355 /* If the file doesn't support async, just async punt */
Jens Axboeaf197f52020-04-28 13:15:06 -06003356 if (force_nonblock && !io_file_supports_async(req->file, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003357 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003358
Jens Axboe10d59342019-12-09 20:16:22 -07003359 /* file path doesn't support NOWAIT for non-direct_IO */
3360 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3361 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003362 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003363
Pavel Begunkov632546c2020-11-07 13:16:26 +00003364 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003365 if (unlikely(ret))
3366 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003367
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003368 /*
3369 * Open-code file_start_write here to grab freeze protection,
3370 * which will be released by another thread in
3371 * io_complete_rw(). Fool lockdep by telling it the lock got
3372 * released so that it doesn't complain about the held lock when
3373 * we return to userspace.
3374 */
3375 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003376 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003377 __sb_writers_release(file_inode(req->file)->i_sb,
3378 SB_FREEZE_WRITE);
3379 }
3380 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003381
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003382 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003383 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003384 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003385 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003386 else
3387 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003388
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003389 /*
3390 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3391 * retry them without IOCB_NOWAIT.
3392 */
3393 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3394 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003395 /* no retry on NONBLOCK nor RWF_NOWAIT */
3396 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003397 goto done;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07003398 if (ret2 == -EIOCBQUEUED && req->async_data)
3399 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003400 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003401 /* IOPOLL retry should happen for io-wq threads */
3402 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3403 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003404done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003405 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003406 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003407copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003408 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003409 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003410 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003411 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003412 }
Jens Axboe31b51512019-01-18 22:56:34 -07003413out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003414 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003415 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003416 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003417 return ret;
3418}
3419
Jens Axboe80a261f2020-09-28 14:23:58 -06003420static int io_renameat_prep(struct io_kiocb *req,
3421 const struct io_uring_sqe *sqe)
3422{
3423 struct io_rename *ren = &req->rename;
3424 const char __user *oldf, *newf;
3425
3426 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3427 return -EBADF;
3428
3429 ren->old_dfd = READ_ONCE(sqe->fd);
3430 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3431 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3432 ren->new_dfd = READ_ONCE(sqe->len);
3433 ren->flags = READ_ONCE(sqe->rename_flags);
3434
3435 ren->oldpath = getname(oldf);
3436 if (IS_ERR(ren->oldpath))
3437 return PTR_ERR(ren->oldpath);
3438
3439 ren->newpath = getname(newf);
3440 if (IS_ERR(ren->newpath)) {
3441 putname(ren->oldpath);
3442 return PTR_ERR(ren->newpath);
3443 }
3444
3445 req->flags |= REQ_F_NEED_CLEANUP;
3446 return 0;
3447}
3448
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003449static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003450{
3451 struct io_rename *ren = &req->rename;
3452 int ret;
3453
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003454 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003455 return -EAGAIN;
3456
3457 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3458 ren->newpath, ren->flags);
3459
3460 req->flags &= ~REQ_F_NEED_CLEANUP;
3461 if (ret < 0)
3462 req_set_fail_links(req);
3463 io_req_complete(req, ret);
3464 return 0;
3465}
3466
Jens Axboe14a11432020-09-28 14:27:37 -06003467static int io_unlinkat_prep(struct io_kiocb *req,
3468 const struct io_uring_sqe *sqe)
3469{
3470 struct io_unlink *un = &req->unlink;
3471 const char __user *fname;
3472
3473 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3474 return -EBADF;
3475
3476 un->dfd = READ_ONCE(sqe->fd);
3477
3478 un->flags = READ_ONCE(sqe->unlink_flags);
3479 if (un->flags & ~AT_REMOVEDIR)
3480 return -EINVAL;
3481
3482 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3483 un->filename = getname(fname);
3484 if (IS_ERR(un->filename))
3485 return PTR_ERR(un->filename);
3486
3487 req->flags |= REQ_F_NEED_CLEANUP;
3488 return 0;
3489}
3490
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003491static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003492{
3493 struct io_unlink *un = &req->unlink;
3494 int ret;
3495
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003496 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003497 return -EAGAIN;
3498
3499 if (un->flags & AT_REMOVEDIR)
3500 ret = do_rmdir(un->dfd, un->filename);
3501 else
3502 ret = do_unlinkat(un->dfd, un->filename);
3503
3504 req->flags &= ~REQ_F_NEED_CLEANUP;
3505 if (ret < 0)
3506 req_set_fail_links(req);
3507 io_req_complete(req, ret);
3508 return 0;
3509}
3510
Jens Axboe36f4fa62020-09-05 11:14:22 -06003511static int io_shutdown_prep(struct io_kiocb *req,
3512 const struct io_uring_sqe *sqe)
3513{
3514#if defined(CONFIG_NET)
3515 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3516 return -EINVAL;
3517 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3518 sqe->buf_index)
3519 return -EINVAL;
3520
3521 req->shutdown.how = READ_ONCE(sqe->len);
3522 return 0;
3523#else
3524 return -EOPNOTSUPP;
3525#endif
3526}
3527
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003528static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003529{
3530#if defined(CONFIG_NET)
3531 struct socket *sock;
3532 int ret;
3533
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003534 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003535 return -EAGAIN;
3536
Linus Torvalds48aba792020-12-16 12:44:05 -08003537 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003538 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003539 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003540
3541 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003542 if (ret < 0)
3543 req_set_fail_links(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003544 io_req_complete(req, ret);
3545 return 0;
3546#else
3547 return -EOPNOTSUPP;
3548#endif
3549}
3550
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003551static int __io_splice_prep(struct io_kiocb *req,
3552 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003553{
3554 struct io_splice* sp = &req->splice;
3555 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003556
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003557 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3558 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003559
3560 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003561 sp->len = READ_ONCE(sqe->len);
3562 sp->flags = READ_ONCE(sqe->splice_flags);
3563
3564 if (unlikely(sp->flags & ~valid_flags))
3565 return -EINVAL;
3566
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003567 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3568 (sp->flags & SPLICE_F_FD_IN_FIXED));
3569 if (!sp->file_in)
3570 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003571 req->flags |= REQ_F_NEED_CLEANUP;
3572
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003573 if (!S_ISREG(file_inode(sp->file_in)->i_mode)) {
3574 /*
3575 * Splice operation will be punted aync, and here need to
3576 * modify io_wq_work.flags, so initialize io_wq_work firstly.
3577 */
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003578 req->work.flags |= IO_WQ_WORK_UNBOUND;
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003579 }
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003580
3581 return 0;
3582}
3583
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003584static int io_tee_prep(struct io_kiocb *req,
3585 const struct io_uring_sqe *sqe)
3586{
3587 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3588 return -EINVAL;
3589 return __io_splice_prep(req, sqe);
3590}
3591
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003592static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003593{
3594 struct io_splice *sp = &req->splice;
3595 struct file *in = sp->file_in;
3596 struct file *out = sp->file_out;
3597 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3598 long ret = 0;
3599
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003600 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003601 return -EAGAIN;
3602 if (sp->len)
3603 ret = do_tee(in, out, sp->len, flags);
3604
3605 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3606 req->flags &= ~REQ_F_NEED_CLEANUP;
3607
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003608 if (ret != sp->len)
3609 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003610 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003611 return 0;
3612}
3613
3614static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3615{
3616 struct io_splice* sp = &req->splice;
3617
3618 sp->off_in = READ_ONCE(sqe->splice_off_in);
3619 sp->off_out = READ_ONCE(sqe->off);
3620 return __io_splice_prep(req, sqe);
3621}
3622
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003623static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003624{
3625 struct io_splice *sp = &req->splice;
3626 struct file *in = sp->file_in;
3627 struct file *out = sp->file_out;
3628 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3629 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003630 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003631
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003632 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003633 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003634
3635 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3636 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003637
Jens Axboe948a7742020-05-17 14:21:38 -06003638 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003639 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003640
3641 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3642 req->flags &= ~REQ_F_NEED_CLEANUP;
3643
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003644 if (ret != sp->len)
3645 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003646 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003647 return 0;
3648}
3649
Jens Axboe2b188cc2019-01-07 10:46:33 -07003650/*
3651 * IORING_OP_NOP just posts a completion event, nothing else.
3652 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003653static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003654{
3655 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003656
Jens Axboedef596e2019-01-09 08:59:42 -07003657 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3658 return -EINVAL;
3659
Pavel Begunkov889fca72021-02-10 00:03:09 +00003660 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003661 return 0;
3662}
3663
Pavel Begunkov1155c762021-02-18 18:29:38 +00003664static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003665{
Jens Axboe6b063142019-01-10 22:13:58 -07003666 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003667
Jens Axboe09bb8392019-03-13 12:39:28 -06003668 if (!req->file)
3669 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003670
Jens Axboe6b063142019-01-10 22:13:58 -07003671 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003672 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003673 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003674 return -EINVAL;
3675
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003676 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3677 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3678 return -EINVAL;
3679
3680 req->sync.off = READ_ONCE(sqe->off);
3681 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003682 return 0;
3683}
3684
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003685static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07003686{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003687 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003688 int ret;
3689
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003690 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003691 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003692 return -EAGAIN;
3693
Jens Axboe9adbd452019-12-20 08:45:55 -07003694 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003695 end > 0 ? end : LLONG_MAX,
3696 req->sync.flags & IORING_FSYNC_DATASYNC);
3697 if (ret < 0)
3698 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003699 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003700 return 0;
3701}
3702
Jens Axboed63d1b52019-12-10 10:38:56 -07003703static int io_fallocate_prep(struct io_kiocb *req,
3704 const struct io_uring_sqe *sqe)
3705{
3706 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3707 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003708 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3709 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003710
3711 req->sync.off = READ_ONCE(sqe->off);
3712 req->sync.len = READ_ONCE(sqe->addr);
3713 req->sync.mode = READ_ONCE(sqe->len);
3714 return 0;
3715}
3716
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003717static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07003718{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003719 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003720
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003721 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003722 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003723 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003724 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3725 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003726 if (ret < 0)
3727 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003728 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003729 return 0;
3730}
3731
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003732static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003733{
Jens Axboef8748882020-01-08 17:47:02 -07003734 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003735 int ret;
3736
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003737 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003738 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003739 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003740 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003741
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003742 /* open.how should be already initialised */
3743 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003744 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003745
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003746 req->open.dfd = READ_ONCE(sqe->fd);
3747 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003748 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003749 if (IS_ERR(req->open.filename)) {
3750 ret = PTR_ERR(req->open.filename);
3751 req->open.filename = NULL;
3752 return ret;
3753 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003754 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003755 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003756 return 0;
3757}
3758
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003759static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3760{
3761 u64 flags, mode;
3762
Jens Axboe14587a462020-09-05 11:36:08 -06003763 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003764 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003765 mode = READ_ONCE(sqe->len);
3766 flags = READ_ONCE(sqe->open_flags);
3767 req->open.how = build_open_how(flags, mode);
3768 return __io_openat_prep(req, sqe);
3769}
3770
Jens Axboecebdb982020-01-08 17:59:24 -07003771static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3772{
3773 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003774 size_t len;
3775 int ret;
3776
Jens Axboe14587a462020-09-05 11:36:08 -06003777 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003778 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003779 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3780 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003781 if (len < OPEN_HOW_SIZE_VER0)
3782 return -EINVAL;
3783
3784 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3785 len);
3786 if (ret)
3787 return ret;
3788
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003789 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003790}
3791
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003792static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003793{
3794 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003795 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003796 bool nonblock_set;
3797 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003798 int ret;
3799
Jens Axboecebdb982020-01-08 17:59:24 -07003800 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003801 if (ret)
3802 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003803 nonblock_set = op.open_flag & O_NONBLOCK;
3804 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003805 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003806 /*
3807 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
3808 * it'll always -EAGAIN
3809 */
3810 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
3811 return -EAGAIN;
3812 op.lookup_flags |= LOOKUP_CACHED;
3813 op.open_flag |= O_NONBLOCK;
3814 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07003815
Jens Axboe4022e7a2020-03-19 19:23:18 -06003816 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003817 if (ret < 0)
3818 goto err;
3819
3820 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Jens Axboe3a81fd02020-12-10 12:25:36 -07003821 /* only retry if RESOLVE_CACHED wasn't already set by application */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003822 if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) &&
3823 file == ERR_PTR(-EAGAIN)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003824 /*
3825 * We could hang on to this 'fd', but seems like marginal
3826 * gain for something that is now known to be a slower path.
3827 * So just put it, and we'll get a new one when we retry.
3828 */
3829 put_unused_fd(ret);
3830 return -EAGAIN;
3831 }
3832
Jens Axboe15b71ab2019-12-11 11:20:36 -07003833 if (IS_ERR(file)) {
3834 put_unused_fd(ret);
3835 ret = PTR_ERR(file);
3836 } else {
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003837 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
Jens Axboe3a81fd02020-12-10 12:25:36 -07003838 file->f_flags &= ~O_NONBLOCK;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003839 fsnotify_open(file);
3840 fd_install(ret, file);
3841 }
3842err:
3843 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003844 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003845 if (ret < 0)
3846 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003847 io_req_complete(req, ret);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003848 return 0;
3849}
3850
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003851static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07003852{
Pavel Begunkove45cff52021-02-28 22:35:14 +00003853 return io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07003854}
3855
Jens Axboe067524e2020-03-02 16:32:28 -07003856static int io_remove_buffers_prep(struct io_kiocb *req,
3857 const struct io_uring_sqe *sqe)
3858{
3859 struct io_provide_buf *p = &req->pbuf;
3860 u64 tmp;
3861
3862 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3863 return -EINVAL;
3864
3865 tmp = READ_ONCE(sqe->fd);
3866 if (!tmp || tmp > USHRT_MAX)
3867 return -EINVAL;
3868
3869 memset(p, 0, sizeof(*p));
3870 p->nbufs = tmp;
3871 p->bgid = READ_ONCE(sqe->buf_group);
3872 return 0;
3873}
3874
3875static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3876 int bgid, unsigned nbufs)
3877{
3878 unsigned i = 0;
3879
3880 /* shouldn't happen */
3881 if (!nbufs)
3882 return 0;
3883
3884 /* the head kbuf is the list itself */
3885 while (!list_empty(&buf->list)) {
3886 struct io_buffer *nxt;
3887
3888 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3889 list_del(&nxt->list);
3890 kfree(nxt);
3891 if (++i == nbufs)
3892 return i;
3893 }
3894 i++;
3895 kfree(buf);
3896 idr_remove(&ctx->io_buffer_idr, bgid);
3897
3898 return i;
3899}
3900
Pavel Begunkov889fca72021-02-10 00:03:09 +00003901static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07003902{
3903 struct io_provide_buf *p = &req->pbuf;
3904 struct io_ring_ctx *ctx = req->ctx;
3905 struct io_buffer *head;
3906 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003907 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07003908
3909 io_ring_submit_lock(ctx, !force_nonblock);
3910
3911 lockdep_assert_held(&ctx->uring_lock);
3912
3913 ret = -ENOENT;
3914 head = idr_find(&ctx->io_buffer_idr, p->bgid);
3915 if (head)
3916 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07003917 if (ret < 0)
3918 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003919
3920 /* need to hold the lock to complete IOPOLL requests */
3921 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00003922 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003923 io_ring_submit_unlock(ctx, !force_nonblock);
3924 } else {
3925 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00003926 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003927 }
Jens Axboe067524e2020-03-02 16:32:28 -07003928 return 0;
3929}
3930
Jens Axboeddf0322d2020-02-23 16:41:33 -07003931static int io_provide_buffers_prep(struct io_kiocb *req,
3932 const struct io_uring_sqe *sqe)
3933{
3934 struct io_provide_buf *p = &req->pbuf;
3935 u64 tmp;
3936
3937 if (sqe->ioprio || sqe->rw_flags)
3938 return -EINVAL;
3939
3940 tmp = READ_ONCE(sqe->fd);
3941 if (!tmp || tmp > USHRT_MAX)
3942 return -E2BIG;
3943 p->nbufs = tmp;
3944 p->addr = READ_ONCE(sqe->addr);
3945 p->len = READ_ONCE(sqe->len);
3946
Bijan Mottahedehefe68c12020-06-04 18:01:52 -07003947 if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs)))
Jens Axboeddf0322d2020-02-23 16:41:33 -07003948 return -EFAULT;
3949
3950 p->bgid = READ_ONCE(sqe->buf_group);
3951 tmp = READ_ONCE(sqe->off);
3952 if (tmp > USHRT_MAX)
3953 return -E2BIG;
3954 p->bid = tmp;
3955 return 0;
3956}
3957
3958static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
3959{
3960 struct io_buffer *buf;
3961 u64 addr = pbuf->addr;
3962 int i, bid = pbuf->bid;
3963
3964 for (i = 0; i < pbuf->nbufs; i++) {
3965 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
3966 if (!buf)
3967 break;
3968
3969 buf->addr = addr;
3970 buf->len = pbuf->len;
3971 buf->bid = bid;
3972 addr += pbuf->len;
3973 bid++;
3974 if (!*head) {
3975 INIT_LIST_HEAD(&buf->list);
3976 *head = buf;
3977 } else {
3978 list_add_tail(&buf->list, &(*head)->list);
3979 }
3980 }
3981
3982 return i ? i : -ENOMEM;
3983}
3984
Pavel Begunkov889fca72021-02-10 00:03:09 +00003985static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07003986{
3987 struct io_provide_buf *p = &req->pbuf;
3988 struct io_ring_ctx *ctx = req->ctx;
3989 struct io_buffer *head, *list;
3990 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003991 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003992
3993 io_ring_submit_lock(ctx, !force_nonblock);
3994
3995 lockdep_assert_held(&ctx->uring_lock);
3996
3997 list = head = idr_find(&ctx->io_buffer_idr, p->bgid);
3998
3999 ret = io_add_buffers(p, &head);
4000 if (ret < 0)
4001 goto out;
4002
4003 if (!list) {
4004 ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1,
4005 GFP_KERNEL);
4006 if (ret < 0) {
Jens Axboe067524e2020-03-02 16:32:28 -07004007 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004008 goto out;
4009 }
4010 }
4011out:
Jens Axboeddf0322d2020-02-23 16:41:33 -07004012 if (ret < 0)
4013 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004014
4015 /* need to hold the lock to complete IOPOLL requests */
4016 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00004017 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004018 io_ring_submit_unlock(ctx, !force_nonblock);
4019 } else {
4020 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004021 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004022 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07004023 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004024}
4025
Jens Axboe3e4827b2020-01-08 15:18:09 -07004026static int io_epoll_ctl_prep(struct io_kiocb *req,
4027 const struct io_uring_sqe *sqe)
4028{
4029#if defined(CONFIG_EPOLL)
4030 if (sqe->ioprio || sqe->buf_index)
4031 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004032 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004033 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004034
4035 req->epoll.epfd = READ_ONCE(sqe->fd);
4036 req->epoll.op = READ_ONCE(sqe->len);
4037 req->epoll.fd = READ_ONCE(sqe->off);
4038
4039 if (ep_op_has_event(req->epoll.op)) {
4040 struct epoll_event __user *ev;
4041
4042 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4043 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4044 return -EFAULT;
4045 }
4046
4047 return 0;
4048#else
4049 return -EOPNOTSUPP;
4050#endif
4051}
4052
Pavel Begunkov889fca72021-02-10 00:03:09 +00004053static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004054{
4055#if defined(CONFIG_EPOLL)
4056 struct io_epoll *ie = &req->epoll;
4057 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004058 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004059
4060 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4061 if (force_nonblock && ret == -EAGAIN)
4062 return -EAGAIN;
4063
4064 if (ret < 0)
4065 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004066 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004067 return 0;
4068#else
4069 return -EOPNOTSUPP;
4070#endif
4071}
4072
Jens Axboec1ca7572019-12-25 22:18:28 -07004073static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4074{
4075#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4076 if (sqe->ioprio || sqe->buf_index || sqe->off)
4077 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004078 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4079 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004080
4081 req->madvise.addr = READ_ONCE(sqe->addr);
4082 req->madvise.len = READ_ONCE(sqe->len);
4083 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4084 return 0;
4085#else
4086 return -EOPNOTSUPP;
4087#endif
4088}
4089
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004090static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004091{
4092#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4093 struct io_madvise *ma = &req->madvise;
4094 int ret;
4095
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004096 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004097 return -EAGAIN;
4098
Minchan Kim0726b012020-10-17 16:14:50 -07004099 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004100 if (ret < 0)
4101 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004102 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004103 return 0;
4104#else
4105 return -EOPNOTSUPP;
4106#endif
4107}
4108
Jens Axboe4840e412019-12-25 22:03:45 -07004109static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4110{
4111 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4112 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004113 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4114 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004115
4116 req->fadvise.offset = READ_ONCE(sqe->off);
4117 req->fadvise.len = READ_ONCE(sqe->len);
4118 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4119 return 0;
4120}
4121
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004122static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004123{
4124 struct io_fadvise *fa = &req->fadvise;
4125 int ret;
4126
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004127 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004128 switch (fa->advice) {
4129 case POSIX_FADV_NORMAL:
4130 case POSIX_FADV_RANDOM:
4131 case POSIX_FADV_SEQUENTIAL:
4132 break;
4133 default:
4134 return -EAGAIN;
4135 }
4136 }
Jens Axboe4840e412019-12-25 22:03:45 -07004137
4138 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4139 if (ret < 0)
4140 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004141 io_req_complete(req, ret);
Jens Axboe4840e412019-12-25 22:03:45 -07004142 return 0;
4143}
4144
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004145static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4146{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004147 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004148 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004149 if (sqe->ioprio || sqe->buf_index)
4150 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004151 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004152 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004153
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004154 req->statx.dfd = READ_ONCE(sqe->fd);
4155 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004156 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004157 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4158 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004159
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004160 return 0;
4161}
4162
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004163static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004164{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004165 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004166 int ret;
4167
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004168 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004169 /* only need file table for an actual valid fd */
4170 if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD)
4171 req->flags |= REQ_F_NO_FILE_TABLE;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004172 return -EAGAIN;
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004173 }
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004174
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004175 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4176 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004177
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004178 if (ret < 0)
4179 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004180 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004181 return 0;
4182}
4183
Jens Axboeb5dba592019-12-11 14:02:38 -07004184static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4185{
Jens Axboe14587a462020-09-05 11:36:08 -06004186 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004187 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004188 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4189 sqe->rw_flags || sqe->buf_index)
4190 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004191 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004192 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004193
4194 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004195 return 0;
4196}
4197
Pavel Begunkov889fca72021-02-10 00:03:09 +00004198static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004199{
Jens Axboe9eac1902021-01-19 15:50:37 -07004200 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004201 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004202 struct fdtable *fdt;
4203 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -07004204 int ret;
4205
Jens Axboe9eac1902021-01-19 15:50:37 -07004206 file = NULL;
4207 ret = -EBADF;
4208 spin_lock(&files->file_lock);
4209 fdt = files_fdtable(files);
4210 if (close->fd >= fdt->max_fds) {
4211 spin_unlock(&files->file_lock);
4212 goto err;
4213 }
4214 file = fdt->fd[close->fd];
4215 if (!file) {
4216 spin_unlock(&files->file_lock);
4217 goto err;
4218 }
4219
4220 if (file->f_op == &io_uring_fops) {
4221 spin_unlock(&files->file_lock);
4222 file = NULL;
4223 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004224 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004225
4226 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004227 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004228 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004229 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004230 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004231
Jens Axboe9eac1902021-01-19 15:50:37 -07004232 ret = __close_fd_get_file(close->fd, &file);
4233 spin_unlock(&files->file_lock);
4234 if (ret < 0) {
4235 if (ret == -ENOENT)
4236 ret = -EBADF;
4237 goto err;
4238 }
4239
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004240 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004241 ret = filp_close(file, current->files);
4242err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004243 if (ret < 0)
4244 req_set_fail_links(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004245 if (file)
4246 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004247 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004248 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004249}
4250
Pavel Begunkov1155c762021-02-18 18:29:38 +00004251static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004252{
4253 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004254
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004255 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4256 return -EINVAL;
4257 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4258 return -EINVAL;
4259
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004260 req->sync.off = READ_ONCE(sqe->off);
4261 req->sync.len = READ_ONCE(sqe->len);
4262 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004263 return 0;
4264}
4265
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004266static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004267{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004268 int ret;
4269
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004270 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004271 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004272 return -EAGAIN;
4273
Jens Axboe9adbd452019-12-20 08:45:55 -07004274 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004275 req->sync.flags);
4276 if (ret < 0)
4277 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004278 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004279 return 0;
4280}
4281
YueHaibing469956e2020-03-04 15:53:52 +08004282#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004283static int io_setup_async_msg(struct io_kiocb *req,
4284 struct io_async_msghdr *kmsg)
4285{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004286 struct io_async_msghdr *async_msg = req->async_data;
4287
4288 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004289 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004290 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004291 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004292 return -ENOMEM;
4293 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004294 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004295 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004296 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004297 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004298 /* if were using fast_iov, set it to the new one */
4299 if (!async_msg->free_iov)
4300 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4301
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004302 return -EAGAIN;
4303}
4304
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004305static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4306 struct io_async_msghdr *iomsg)
4307{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004308 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004309 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004310 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004311 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004312}
4313
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004314static int io_sendmsg_prep_async(struct io_kiocb *req)
4315{
4316 int ret;
4317
4318 if (!io_op_defs[req->opcode].needs_async_data)
4319 return 0;
4320 ret = io_sendmsg_copy_hdr(req, req->async_data);
4321 if (!ret)
4322 req->flags |= REQ_F_NEED_CLEANUP;
4323 return ret;
4324}
4325
Jens Axboe3529d8c2019-12-19 18:24:38 -07004326static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004327{
Jens Axboee47293f2019-12-20 08:58:21 -07004328 struct io_sr_msg *sr = &req->sr_msg;
Jens Axboe03b12302019-12-02 18:50:25 -07004329
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004330 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4331 return -EINVAL;
4332
Jens Axboee47293f2019-12-20 08:58:21 -07004333 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004334 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004335 sr->len = READ_ONCE(sqe->len);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004336
Jens Axboed8768362020-02-27 14:17:49 -07004337#ifdef CONFIG_COMPAT
4338 if (req->ctx->compat)
4339 sr->msg_flags |= MSG_CMSG_COMPAT;
4340#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004341 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004342}
4343
Pavel Begunkov889fca72021-02-10 00:03:09 +00004344static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004345{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004346 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004347 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004348 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004349 int ret;
4350
Florent Revestdba4a922020-12-04 12:36:04 +01004351 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004352 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004353 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004354
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004355 kmsg = req->async_data;
4356 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004357 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004358 if (ret)
4359 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004360 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004361 }
4362
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004363 flags = req->sr_msg.msg_flags;
4364 if (flags & MSG_DONTWAIT)
4365 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004366 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004367 flags |= MSG_DONTWAIT;
4368
4369 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004370 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004371 return io_setup_async_msg(req, kmsg);
4372 if (ret == -ERESTARTSYS)
4373 ret = -EINTR;
4374
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004375 /* fast path, check for non-NULL to avoid function call */
4376 if (kmsg->free_iov)
4377 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004378 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboefddafac2020-01-04 20:19:44 -07004379 if (ret < 0)
4380 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004381 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004382 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004383}
4384
Pavel Begunkov889fca72021-02-10 00:03:09 +00004385static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004386{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004387 struct io_sr_msg *sr = &req->sr_msg;
4388 struct msghdr msg;
4389 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004390 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004391 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004392 int ret;
4393
Florent Revestdba4a922020-12-04 12:36:04 +01004394 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004395 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004396 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004397
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004398 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4399 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004400 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004401
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004402 msg.msg_name = NULL;
4403 msg.msg_control = NULL;
4404 msg.msg_controllen = 0;
4405 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004406
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004407 flags = req->sr_msg.msg_flags;
4408 if (flags & MSG_DONTWAIT)
4409 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004410 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004411 flags |= MSG_DONTWAIT;
Jens Axboe03b12302019-12-02 18:50:25 -07004412
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004413 msg.msg_flags = flags;
4414 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004415 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004416 return -EAGAIN;
4417 if (ret == -ERESTARTSYS)
4418 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004419
Jens Axboe03b12302019-12-02 18:50:25 -07004420 if (ret < 0)
4421 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004422 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004423 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004424}
4425
Pavel Begunkov1400e692020-07-12 20:41:05 +03004426static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4427 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004428{
4429 struct io_sr_msg *sr = &req->sr_msg;
4430 struct iovec __user *uiov;
4431 size_t iov_len;
4432 int ret;
4433
Pavel Begunkov1400e692020-07-12 20:41:05 +03004434 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4435 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004436 if (ret)
4437 return ret;
4438
4439 if (req->flags & REQ_F_BUFFER_SELECT) {
4440 if (iov_len > 1)
4441 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004442 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004443 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004444 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004445 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004446 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004447 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004448 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004449 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004450 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004451 if (ret > 0)
4452 ret = 0;
4453 }
4454
4455 return ret;
4456}
4457
4458#ifdef CONFIG_COMPAT
4459static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004460 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004461{
4462 struct compat_msghdr __user *msg_compat;
4463 struct io_sr_msg *sr = &req->sr_msg;
4464 struct compat_iovec __user *uiov;
4465 compat_uptr_t ptr;
4466 compat_size_t len;
4467 int ret;
4468
Pavel Begunkov270a5942020-07-12 20:41:04 +03004469 msg_compat = (struct compat_msghdr __user *) sr->umsg;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004470 ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr,
Jens Axboe52de1fe2020-02-27 10:15:42 -07004471 &ptr, &len);
4472 if (ret)
4473 return ret;
4474
4475 uiov = compat_ptr(ptr);
4476 if (req->flags & REQ_F_BUFFER_SELECT) {
4477 compat_ssize_t clen;
4478
4479 if (len > 1)
4480 return -EINVAL;
4481 if (!access_ok(uiov, sizeof(*uiov)))
4482 return -EFAULT;
4483 if (__get_user(clen, &uiov->iov_len))
4484 return -EFAULT;
4485 if (clen < 0)
4486 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004487 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004488 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004489 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004490 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004491 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004492 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004493 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004494 if (ret < 0)
4495 return ret;
4496 }
4497
4498 return 0;
4499}
Jens Axboe03b12302019-12-02 18:50:25 -07004500#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004501
Pavel Begunkov1400e692020-07-12 20:41:05 +03004502static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4503 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004504{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004505 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004506
4507#ifdef CONFIG_COMPAT
4508 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004509 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004510#endif
4511
Pavel Begunkov1400e692020-07-12 20:41:05 +03004512 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004513}
4514
Jens Axboebcda7ba2020-02-23 16:42:51 -07004515static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004516 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004517{
4518 struct io_sr_msg *sr = &req->sr_msg;
4519 struct io_buffer *kbuf;
4520
Jens Axboebcda7ba2020-02-23 16:42:51 -07004521 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4522 if (IS_ERR(kbuf))
4523 return kbuf;
4524
4525 sr->kbuf = kbuf;
4526 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004527 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004528}
4529
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004530static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4531{
4532 return io_put_kbuf(req, req->sr_msg.kbuf);
4533}
4534
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004535static int io_recvmsg_prep_async(struct io_kiocb *req)
Jens Axboe03b12302019-12-02 18:50:25 -07004536{
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004537 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004538
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004539 if (!io_op_defs[req->opcode].needs_async_data)
4540 return 0;
4541 ret = io_recvmsg_copy_hdr(req, req->async_data);
4542 if (!ret)
4543 req->flags |= REQ_F_NEED_CLEANUP;
4544 return ret;
4545}
4546
4547static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4548{
4549 struct io_sr_msg *sr = &req->sr_msg;
4550
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004551 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4552 return -EINVAL;
4553
Jens Axboe3529d8c2019-12-19 18:24:38 -07004554 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004555 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004556 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004557 sr->bgid = READ_ONCE(sqe->buf_group);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004558
Jens Axboed8768362020-02-27 14:17:49 -07004559#ifdef CONFIG_COMPAT
4560 if (req->ctx->compat)
4561 sr->msg_flags |= MSG_CMSG_COMPAT;
4562#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004563 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004564}
4565
Pavel Begunkov889fca72021-02-10 00:03:09 +00004566static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004567{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004568 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004569 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004570 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004571 unsigned flags;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004572 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004573 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004574
Florent Revestdba4a922020-12-04 12:36:04 +01004575 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004576 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004577 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004578
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004579 kmsg = req->async_data;
4580 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004581 ret = io_recvmsg_copy_hdr(req, &iomsg);
4582 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004583 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004584 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004585 }
4586
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004587 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004588 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004589 if (IS_ERR(kbuf))
4590 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004591 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004592 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4593 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004594 1, req->sr_msg.len);
4595 }
4596
4597 flags = req->sr_msg.msg_flags;
4598 if (flags & MSG_DONTWAIT)
4599 req->flags |= REQ_F_NOWAIT;
4600 else if (force_nonblock)
4601 flags |= MSG_DONTWAIT;
4602
4603 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4604 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004605 if (force_nonblock && ret == -EAGAIN)
4606 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004607 if (ret == -ERESTARTSYS)
4608 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004609
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004610 if (req->flags & REQ_F_BUFFER_SELECTED)
4611 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004612 /* fast path, check for non-NULL to avoid function call */
4613 if (kmsg->free_iov)
4614 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004615 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004616 if (ret < 0)
4617 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004618 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004619 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004620}
4621
Pavel Begunkov889fca72021-02-10 00:03:09 +00004622static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004623{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004624 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004625 struct io_sr_msg *sr = &req->sr_msg;
4626 struct msghdr msg;
4627 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004628 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004629 struct iovec iov;
4630 unsigned flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004631 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004632 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004633
Florent Revestdba4a922020-12-04 12:36:04 +01004634 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004635 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004636 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004637
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004638 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004639 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004640 if (IS_ERR(kbuf))
4641 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004642 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004643 }
4644
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004645 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004646 if (unlikely(ret))
4647 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004648
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004649 msg.msg_name = NULL;
4650 msg.msg_control = NULL;
4651 msg.msg_controllen = 0;
4652 msg.msg_namelen = 0;
4653 msg.msg_iocb = NULL;
4654 msg.msg_flags = 0;
4655
4656 flags = req->sr_msg.msg_flags;
4657 if (flags & MSG_DONTWAIT)
4658 req->flags |= REQ_F_NOWAIT;
4659 else if (force_nonblock)
4660 flags |= MSG_DONTWAIT;
4661
4662 ret = sock_recvmsg(sock, &msg, flags);
4663 if (force_nonblock && ret == -EAGAIN)
4664 return -EAGAIN;
4665 if (ret == -ERESTARTSYS)
4666 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004667out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004668 if (req->flags & REQ_F_BUFFER_SELECTED)
4669 cflags = io_put_recv_kbuf(req);
Jens Axboefddafac2020-01-04 20:19:44 -07004670 if (ret < 0)
4671 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004672 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07004673 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004674}
4675
Jens Axboe3529d8c2019-12-19 18:24:38 -07004676static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004677{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004678 struct io_accept *accept = &req->accept;
4679
Jens Axboe14587a462020-09-05 11:36:08 -06004680 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06004681 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004682 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004683 return -EINVAL;
4684
Jens Axboed55e5f52019-12-11 16:12:15 -07004685 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4686 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004687 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004688 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004689 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004690}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004691
Pavel Begunkov889fca72021-02-10 00:03:09 +00004692static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004693{
4694 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004695 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004696 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004697 int ret;
4698
Jiufei Xuee697dee2020-06-10 13:41:59 +08004699 if (req->file->f_flags & O_NONBLOCK)
4700 req->flags |= REQ_F_NOWAIT;
4701
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004702 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004703 accept->addr_len, accept->flags,
4704 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004705 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004706 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004707 if (ret < 0) {
4708 if (ret == -ERESTARTSYS)
4709 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004710 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004711 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00004712 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004713 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004714}
4715
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004716static int io_connect_prep_async(struct io_kiocb *req)
4717{
4718 struct io_async_connect *io = req->async_data;
4719 struct io_connect *conn = &req->connect;
4720
4721 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
4722}
4723
Jens Axboe3529d8c2019-12-19 18:24:38 -07004724static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004725{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004726 struct io_connect *conn = &req->connect;
Jens Axboef499a022019-12-02 16:28:46 -07004727
Jens Axboe14587a462020-09-05 11:36:08 -06004728 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004729 return -EINVAL;
4730 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4731 return -EINVAL;
4732
Jens Axboe3529d8c2019-12-19 18:24:38 -07004733 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4734 conn->addr_len = READ_ONCE(sqe->addr2);
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004735 return 0;
Jens Axboef499a022019-12-02 16:28:46 -07004736}
4737
Pavel Begunkov889fca72021-02-10 00:03:09 +00004738static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004739{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004740 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004741 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004742 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004743 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004744
Jens Axboee8c2bc12020-08-15 18:44:09 -07004745 if (req->async_data) {
4746 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004747 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004748 ret = move_addr_to_kernel(req->connect.addr,
4749 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004750 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004751 if (ret)
4752 goto out;
4753 io = &__io;
4754 }
4755
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004756 file_flags = force_nonblock ? O_NONBLOCK : 0;
4757
Jens Axboee8c2bc12020-08-15 18:44:09 -07004758 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004759 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004760 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004761 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004762 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004763 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004764 ret = -ENOMEM;
4765 goto out;
4766 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004767 io = req->async_data;
4768 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004769 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004770 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004771 if (ret == -ERESTARTSYS)
4772 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004773out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004774 if (ret < 0)
4775 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004776 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004777 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004778}
YueHaibing469956e2020-03-04 15:53:52 +08004779#else /* !CONFIG_NET */
Jens Axboe99a10082021-02-19 09:35:19 -07004780#define IO_NETOP_FN(op) \
4781static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \
4782{ \
4783 return -EOPNOTSUPP; \
Jens Axboef8e85cf2019-11-23 14:24:24 -07004784}
4785
Jens Axboe99a10082021-02-19 09:35:19 -07004786#define IO_NETOP_PREP(op) \
4787IO_NETOP_FN(op) \
4788static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \
4789{ \
4790 return -EOPNOTSUPP; \
4791} \
4792
4793#define IO_NETOP_PREP_ASYNC(op) \
4794IO_NETOP_PREP(op) \
4795static int io_##op##_prep_async(struct io_kiocb *req) \
4796{ \
4797 return -EOPNOTSUPP; \
YueHaibing469956e2020-03-04 15:53:52 +08004798}
4799
Jens Axboe99a10082021-02-19 09:35:19 -07004800IO_NETOP_PREP_ASYNC(sendmsg);
4801IO_NETOP_PREP_ASYNC(recvmsg);
4802IO_NETOP_PREP_ASYNC(connect);
4803IO_NETOP_PREP(accept);
4804IO_NETOP_FN(send);
4805IO_NETOP_FN(recv);
YueHaibing469956e2020-03-04 15:53:52 +08004806#endif /* CONFIG_NET */
Jens Axboe17f2fe32019-10-17 14:42:58 -06004807
Jens Axboed7718a92020-02-14 22:23:12 -07004808struct io_poll_table {
4809 struct poll_table_struct pt;
4810 struct io_kiocb *req;
4811 int error;
4812};
4813
Jens Axboed7718a92020-02-14 22:23:12 -07004814static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4815 __poll_t mask, task_work_func_t func)
4816{
Jens Axboeaa96bf82020-04-03 11:26:26 -06004817 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004818
4819 /* for instances that support it check for an event match first: */
4820 if (mask && !(mask & poll->events))
4821 return 0;
4822
4823 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4824
4825 list_del_init(&poll->wait.entry);
4826
Jens Axboed7718a92020-02-14 22:23:12 -07004827 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00004828 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06004829 percpu_ref_get(&req->ctx->refs);
4830
Jens Axboed7718a92020-02-14 22:23:12 -07004831 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004832 * If this fails, then the task is exiting. When a task exits, the
4833 * work gets canceled, so just cancel this request as well instead
4834 * of executing it. We can't safely execute it anyway, as we may not
4835 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004836 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06004837 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004838 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06004839 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00004840 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004841 }
Jens Axboed7718a92020-02-14 22:23:12 -07004842 return 1;
4843}
4844
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004845static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4846 __acquires(&req->ctx->completion_lock)
4847{
4848 struct io_ring_ctx *ctx = req->ctx;
4849
4850 if (!req->result && !READ_ONCE(poll->canceled)) {
4851 struct poll_table_struct pt = { ._key = poll->events };
4852
4853 req->result = vfs_poll(req->file, &pt) & poll->events;
4854 }
4855
4856 spin_lock_irq(&ctx->completion_lock);
4857 if (!req->result && !READ_ONCE(poll->canceled)) {
4858 add_wait_queue(poll->head, &poll->wait);
4859 return true;
4860 }
4861
4862 return false;
4863}
4864
Jens Axboed4e7cd32020-08-15 11:44:50 -07004865static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004866{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004867 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004868 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004869 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004870 return req->apoll->double_poll;
4871}
4872
4873static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4874{
4875 if (req->opcode == IORING_OP_POLL_ADD)
4876 return &req->poll;
4877 return &req->apoll->poll;
4878}
4879
4880static void io_poll_remove_double(struct io_kiocb *req)
4881{
4882 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004883
4884 lockdep_assert_held(&req->ctx->completion_lock);
4885
4886 if (poll && poll->head) {
4887 struct wait_queue_head *head = poll->head;
4888
4889 spin_lock(&head->lock);
4890 list_del_init(&poll->wait.entry);
4891 if (poll->wait.private)
4892 refcount_dec(&req->refs);
4893 poll->head = NULL;
4894 spin_unlock(&head->lock);
4895 }
4896}
4897
4898static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
4899{
4900 struct io_ring_ctx *ctx = req->ctx;
4901
Jens Axboed4e7cd32020-08-15 11:44:50 -07004902 io_poll_remove_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004903 req->poll.done = true;
4904 io_cqring_fill_event(req, error ? error : mangle_poll(mask));
4905 io_commit_cqring(ctx);
4906}
4907
Jens Axboe18bceab2020-05-15 11:56:54 -06004908static void io_poll_task_func(struct callback_head *cb)
4909{
4910 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004911 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004912 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004913
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004914 if (io_poll_rewait(req, &req->poll)) {
4915 spin_unlock_irq(&ctx->completion_lock);
4916 } else {
4917 hash_del(&req->hash_node);
4918 io_poll_complete(req, req->result, 0);
4919 spin_unlock_irq(&ctx->completion_lock);
4920
4921 nxt = io_put_req_find_next(req);
4922 io_cqring_ev_posted(ctx);
4923 if (nxt)
4924 __io_req_task_submit(nxt);
4925 }
4926
Jens Axboe6d816e02020-08-11 08:04:14 -06004927 percpu_ref_put(&ctx->refs);
Jens Axboe18bceab2020-05-15 11:56:54 -06004928}
4929
4930static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4931 int sync, void *key)
4932{
4933 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004934 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004935 __poll_t mask = key_to_poll(key);
4936
4937 /* for instances that support it check for an event match first: */
4938 if (mask && !(mask & poll->events))
4939 return 0;
4940
Jens Axboe8706e042020-09-28 08:38:54 -06004941 list_del_init(&wait->entry);
4942
Jens Axboe807abcb2020-07-17 17:09:27 -06004943 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004944 bool done;
4945
Jens Axboe807abcb2020-07-17 17:09:27 -06004946 spin_lock(&poll->head->lock);
4947 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06004948 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06004949 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07004950 /* make sure double remove sees this as being gone */
4951 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06004952 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06004953 if (!done) {
4954 /* use wait func handler, so it matches the rq type */
4955 poll->wait.func(&poll->wait, mode, sync, key);
4956 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004957 }
4958 refcount_dec(&req->refs);
4959 return 1;
4960}
4961
4962static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
4963 wait_queue_func_t wake_func)
4964{
4965 poll->head = NULL;
4966 poll->done = false;
4967 poll->canceled = false;
4968 poll->events = events;
4969 INIT_LIST_HEAD(&poll->wait.entry);
4970 init_waitqueue_func_entry(&poll->wait, wake_func);
4971}
4972
4973static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06004974 struct wait_queue_head *head,
4975 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06004976{
4977 struct io_kiocb *req = pt->req;
4978
4979 /*
4980 * If poll->head is already set, it's because the file being polled
4981 * uses multiple waitqueues for poll handling (eg one for read, one
4982 * for write). Setup a separate io_poll_iocb if this happens.
4983 */
4984 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01004985 struct io_poll_iocb *poll_one = poll;
4986
Jens Axboe18bceab2020-05-15 11:56:54 -06004987 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06004988 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004989 pt->error = -EINVAL;
4990 return;
4991 }
Jens Axboe1c3b3e62021-02-28 16:07:30 -07004992 /* double add on the same waitqueue head, ignore */
4993 if (poll->head == head)
4994 return;
Jens Axboe18bceab2020-05-15 11:56:54 -06004995 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
4996 if (!poll) {
4997 pt->error = -ENOMEM;
4998 return;
4999 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01005000 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboe18bceab2020-05-15 11:56:54 -06005001 refcount_inc(&req->refs);
5002 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06005003 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005004 }
5005
5006 pt->error = 0;
5007 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005008
5009 if (poll->events & EPOLLEXCLUSIVE)
5010 add_wait_queue_exclusive(head, &poll->wait);
5011 else
5012 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005013}
5014
5015static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5016 struct poll_table_struct *p)
5017{
5018 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005019 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005020
Jens Axboe807abcb2020-07-17 17:09:27 -06005021 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005022}
5023
Jens Axboed7718a92020-02-14 22:23:12 -07005024static void io_async_task_func(struct callback_head *cb)
5025{
5026 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5027 struct async_poll *apoll = req->apoll;
5028 struct io_ring_ctx *ctx = req->ctx;
5029
5030 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5031
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005032 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005033 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe6d816e02020-08-11 08:04:14 -06005034 percpu_ref_put(&ctx->refs);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005035 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005036 }
5037
Jens Axboe31067252020-05-17 17:43:31 -06005038 /* If req is still hashed, it cannot have been canceled. Don't check. */
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005039 if (hash_hashed(&req->hash_node))
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005040 hash_del(&req->hash_node);
Jens Axboe2bae0472020-04-13 11:16:34 -06005041
Jens Axboed4e7cd32020-08-15 11:44:50 -07005042 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005043 spin_unlock_irq(&ctx->completion_lock);
5044
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005045 if (!READ_ONCE(apoll->poll.canceled))
5046 __io_req_task_submit(req);
5047 else
5048 __io_req_task_cancel(req, -ECANCELED);
Dan Carpenteraa340842020-07-08 21:47:11 +03005049
Jens Axboe6d816e02020-08-11 08:04:14 -06005050 percpu_ref_put(&ctx->refs);
Jens Axboe807abcb2020-07-17 17:09:27 -06005051 kfree(apoll->double_poll);
Jens Axboe31067252020-05-17 17:43:31 -06005052 kfree(apoll);
Jens Axboed7718a92020-02-14 22:23:12 -07005053}
5054
5055static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5056 void *key)
5057{
5058 struct io_kiocb *req = wait->private;
5059 struct io_poll_iocb *poll = &req->apoll->poll;
5060
5061 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5062 key_to_poll(key));
5063
5064 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5065}
5066
5067static void io_poll_req_insert(struct io_kiocb *req)
5068{
5069 struct io_ring_ctx *ctx = req->ctx;
5070 struct hlist_head *list;
5071
5072 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5073 hlist_add_head(&req->hash_node, list);
5074}
5075
5076static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5077 struct io_poll_iocb *poll,
5078 struct io_poll_table *ipt, __poll_t mask,
5079 wait_queue_func_t wake_func)
5080 __acquires(&ctx->completion_lock)
5081{
5082 struct io_ring_ctx *ctx = req->ctx;
5083 bool cancel = false;
5084
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005085 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005086 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005087 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005088 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005089
5090 ipt->pt._key = mask;
5091 ipt->req = req;
5092 ipt->error = -EINVAL;
5093
Jens Axboed7718a92020-02-14 22:23:12 -07005094 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5095
5096 spin_lock_irq(&ctx->completion_lock);
5097 if (likely(poll->head)) {
5098 spin_lock(&poll->head->lock);
5099 if (unlikely(list_empty(&poll->wait.entry))) {
5100 if (ipt->error)
5101 cancel = true;
5102 ipt->error = 0;
5103 mask = 0;
5104 }
5105 if (mask || ipt->error)
5106 list_del_init(&poll->wait.entry);
5107 else if (cancel)
5108 WRITE_ONCE(poll->canceled, true);
5109 else if (!poll->done) /* actually waiting for an event */
5110 io_poll_req_insert(req);
5111 spin_unlock(&poll->head->lock);
5112 }
5113
5114 return mask;
5115}
5116
5117static bool io_arm_poll_handler(struct io_kiocb *req)
5118{
5119 const struct io_op_def *def = &io_op_defs[req->opcode];
5120 struct io_ring_ctx *ctx = req->ctx;
5121 struct async_poll *apoll;
5122 struct io_poll_table ipt;
5123 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005124 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005125
5126 if (!req->file || !file_can_poll(req->file))
5127 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005128 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005129 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005130 if (def->pollin)
5131 rw = READ;
5132 else if (def->pollout)
5133 rw = WRITE;
5134 else
5135 return false;
5136 /* if we can't nonblock try, then no point in arming a poll handler */
5137 if (!io_file_supports_async(req->file, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005138 return false;
5139
5140 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5141 if (unlikely(!apoll))
5142 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005143 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005144
5145 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005146 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005147
Nathan Chancellor8755d972020-03-02 16:01:19 -07005148 mask = 0;
Jens Axboed7718a92020-02-14 22:23:12 -07005149 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005150 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005151 if (def->pollout)
5152 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005153
5154 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5155 if ((req->opcode == IORING_OP_RECVMSG) &&
5156 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5157 mask &= ~POLLIN;
5158
Jens Axboed7718a92020-02-14 22:23:12 -07005159 mask |= POLLERR | POLLPRI;
5160
5161 ipt.pt._qproc = io_async_queue_proc;
5162
5163 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5164 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005165 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005166 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005167 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe807abcb2020-07-17 17:09:27 -06005168 kfree(apoll->double_poll);
Jens Axboed7718a92020-02-14 22:23:12 -07005169 kfree(apoll);
5170 return false;
5171 }
5172 spin_unlock_irq(&ctx->completion_lock);
5173 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5174 apoll->poll.events);
5175 return true;
5176}
5177
5178static bool __io_poll_remove_one(struct io_kiocb *req,
5179 struct io_poll_iocb *poll)
5180{
Jens Axboeb41e9852020-02-17 09:52:41 -07005181 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005182
5183 spin_lock(&poll->head->lock);
5184 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005185 if (!list_empty(&poll->wait.entry)) {
5186 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005187 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005188 }
5189 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005190 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005191 return do_complete;
5192}
5193
5194static bool io_poll_remove_one(struct io_kiocb *req)
5195{
5196 bool do_complete;
5197
Jens Axboed4e7cd32020-08-15 11:44:50 -07005198 io_poll_remove_double(req);
5199
Jens Axboed7718a92020-02-14 22:23:12 -07005200 if (req->opcode == IORING_OP_POLL_ADD) {
5201 do_complete = __io_poll_remove_one(req, &req->poll);
5202 } else {
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005203 struct async_poll *apoll = req->apoll;
5204
Jens Axboed7718a92020-02-14 22:23:12 -07005205 /* non-poll requests have submit ref still */
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005206 do_complete = __io_poll_remove_one(req, &apoll->poll);
5207 if (do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005208 io_put_req(req);
Jens Axboe807abcb2020-07-17 17:09:27 -06005209 kfree(apoll->double_poll);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005210 kfree(apoll);
5211 }
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005212 }
5213
Jens Axboeb41e9852020-02-17 09:52:41 -07005214 if (do_complete) {
5215 io_cqring_fill_event(req, -ECANCELED);
5216 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005217 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005218 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005219 }
5220
5221 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005222}
5223
Jens Axboe76e1b642020-09-26 15:05:03 -06005224/*
5225 * Returns true if we found and killed one or more poll requests
5226 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005227static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
5228 struct files_struct *files)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005229{
Jens Axboe78076bb2019-12-04 19:56:40 -07005230 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005231 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005232 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005233
5234 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005235 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5236 struct hlist_head *list;
5237
5238 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005239 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00005240 if (io_match_task(req, tsk, files))
Jens Axboef3606e32020-09-22 08:18:24 -06005241 posted += io_poll_remove_one(req);
5242 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005243 }
5244 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005245
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005246 if (posted)
5247 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005248
5249 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005250}
5251
Jens Axboe47f46762019-11-09 17:43:02 -07005252static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
5253{
Jens Axboe78076bb2019-12-04 19:56:40 -07005254 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005255 struct io_kiocb *req;
5256
Jens Axboe78076bb2019-12-04 19:56:40 -07005257 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5258 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005259 if (sqe_addr != req->user_data)
5260 continue;
5261 if (io_poll_remove_one(req))
Jens Axboeeac406c2019-11-14 12:09:58 -07005262 return 0;
Jens Axboeb41e9852020-02-17 09:52:41 -07005263 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005264 }
5265
5266 return -ENOENT;
5267}
5268
Jens Axboe3529d8c2019-12-19 18:24:38 -07005269static int io_poll_remove_prep(struct io_kiocb *req,
5270 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005271{
Jens Axboe221c5eb2019-01-17 09:41:58 -07005272 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5273 return -EINVAL;
5274 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
5275 sqe->poll_events)
5276 return -EINVAL;
5277
Pavel Begunkov018043b2020-10-27 23:17:18 +00005278 req->poll_remove.addr = READ_ONCE(sqe->addr);
Jens Axboe0969e782019-12-17 18:40:57 -07005279 return 0;
5280}
5281
5282/*
5283 * Find a running poll command that matches one specified in sqe->addr,
5284 * and remove it if found.
5285 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005286static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005287{
5288 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe0969e782019-12-17 18:40:57 -07005289 int ret;
5290
Jens Axboe221c5eb2019-01-17 09:41:58 -07005291 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov018043b2020-10-27 23:17:18 +00005292 ret = io_poll_cancel(ctx, req->poll_remove.addr);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005293 spin_unlock_irq(&ctx->completion_lock);
5294
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005295 if (ret < 0)
5296 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06005297 io_req_complete(req, ret);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005298 return 0;
5299}
5300
Jens Axboe221c5eb2019-01-17 09:41:58 -07005301static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5302 void *key)
5303{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005304 struct io_kiocb *req = wait->private;
5305 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005306
Jens Axboed7718a92020-02-14 22:23:12 -07005307 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005308}
5309
Jens Axboe221c5eb2019-01-17 09:41:58 -07005310static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5311 struct poll_table_struct *p)
5312{
5313 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5314
Jens Axboee8c2bc12020-08-15 18:44:09 -07005315 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c2019-11-14 12:09:58 -07005316}
5317
Jens Axboe3529d8c2019-12-19 18:24:38 -07005318static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005319{
5320 struct io_poll_iocb *poll = &req->poll;
Jiufei Xue5769a352020-06-17 17:53:55 +08005321 u32 events;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005322
5323 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5324 return -EINVAL;
5325 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
5326 return -EINVAL;
5327
Jiufei Xue5769a352020-06-17 17:53:55 +08005328 events = READ_ONCE(sqe->poll32_events);
5329#ifdef __BIG_ENDIAN
5330 events = swahw32(events);
5331#endif
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005332 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP |
5333 (events & EPOLLEXCLUSIVE);
Jens Axboe0969e782019-12-17 18:40:57 -07005334 return 0;
5335}
5336
Pavel Begunkov61e98202021-02-10 00:03:08 +00005337static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005338{
5339 struct io_poll_iocb *poll = &req->poll;
5340 struct io_ring_ctx *ctx = req->ctx;
5341 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005342 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005343
Jens Axboed7718a92020-02-14 22:23:12 -07005344 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005345
Jens Axboed7718a92020-02-14 22:23:12 -07005346 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5347 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005348
Jens Axboe8c838782019-03-12 15:48:16 -06005349 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005350 ipt.error = 0;
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005351 io_poll_complete(req, mask, 0);
Jens Axboe8c838782019-03-12 15:48:16 -06005352 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005353 spin_unlock_irq(&ctx->completion_lock);
5354
Jens Axboe8c838782019-03-12 15:48:16 -06005355 if (mask) {
5356 io_cqring_ev_posted(ctx);
Pavel Begunkov014db002020-03-03 21:33:12 +03005357 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005358 }
Jens Axboe8c838782019-03-12 15:48:16 -06005359 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005360}
5361
Jens Axboe5262f562019-09-17 12:26:57 -06005362static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5363{
Jens Axboead8a48a2019-11-15 08:49:11 -07005364 struct io_timeout_data *data = container_of(timer,
5365 struct io_timeout_data, timer);
5366 struct io_kiocb *req = data->req;
5367 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005368 unsigned long flags;
5369
Jens Axboe5262f562019-09-17 12:26:57 -06005370 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005371 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005372 atomic_set(&req->ctx->cq_timeouts,
5373 atomic_read(&req->ctx->cq_timeouts) + 1);
5374
Jens Axboe78e19bb2019-11-06 15:21:34 -07005375 io_cqring_fill_event(req, -ETIME);
Jens Axboe5262f562019-09-17 12:26:57 -06005376 io_commit_cqring(ctx);
5377 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5378
5379 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005380 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005381 io_put_req(req);
5382 return HRTIMER_NORESTART;
5383}
5384
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005385static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5386 __u64 user_data)
Jens Axboe47f46762019-11-09 17:43:02 -07005387{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005388 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005389 struct io_kiocb *req;
5390 int ret = -ENOENT;
5391
5392 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
5393 if (user_data == req->user_data) {
5394 ret = 0;
5395 break;
5396 }
5397 }
5398
5399 if (ret == -ENOENT)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005400 return ERR_PTR(ret);
Jens Axboef254ac02020-08-12 17:33:30 -06005401
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005402 io = req->async_data;
5403 ret = hrtimer_try_to_cancel(&io->timer);
5404 if (ret == -1)
5405 return ERR_PTR(-EALREADY);
5406 list_del_init(&req->timeout.list);
5407 return req;
5408}
5409
5410static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
5411{
5412 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5413
5414 if (IS_ERR(req))
5415 return PTR_ERR(req);
5416
5417 req_set_fail_links(req);
5418 io_cqring_fill_event(req, -ECANCELED);
5419 io_put_req_deferred(req, 1);
5420 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005421}
5422
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005423static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5424 struct timespec64 *ts, enum hrtimer_mode mode)
5425{
5426 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5427 struct io_timeout_data *data;
5428
5429 if (IS_ERR(req))
5430 return PTR_ERR(req);
5431
5432 req->timeout.off = 0; /* noseq */
5433 data = req->async_data;
5434 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5435 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5436 data->timer.function = io_timeout_fn;
5437 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5438 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005439}
5440
Jens Axboe3529d8c2019-12-19 18:24:38 -07005441static int io_timeout_remove_prep(struct io_kiocb *req,
5442 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005443{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005444 struct io_timeout_rem *tr = &req->timeout_rem;
5445
Jens Axboeb29472e2019-12-17 18:50:29 -07005446 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5447 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005448 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5449 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005450 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005451 return -EINVAL;
5452
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005453 tr->addr = READ_ONCE(sqe->addr);
5454 tr->flags = READ_ONCE(sqe->timeout_flags);
5455 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5456 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5457 return -EINVAL;
5458 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5459 return -EFAULT;
5460 } else if (tr->flags) {
5461 /* timeout removal doesn't support flags */
5462 return -EINVAL;
5463 }
5464
Jens Axboeb29472e2019-12-17 18:50:29 -07005465 return 0;
5466}
5467
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005468static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5469{
5470 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5471 : HRTIMER_MODE_REL;
5472}
5473
Jens Axboe11365042019-10-16 09:08:32 -06005474/*
5475 * Remove or update an existing timeout command
5476 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005477static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005478{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005479 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005480 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005481 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005482
Jens Axboe11365042019-10-16 09:08:32 -06005483 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005484 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005485 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005486 else
5487 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5488 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005489
Jens Axboe47f46762019-11-09 17:43:02 -07005490 io_cqring_fill_event(req, ret);
Jens Axboe11365042019-10-16 09:08:32 -06005491 io_commit_cqring(ctx);
5492 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005493 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005494 if (ret < 0)
5495 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005496 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005497 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005498}
5499
Jens Axboe3529d8c2019-12-19 18:24:38 -07005500static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005501 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005502{
Jens Axboead8a48a2019-11-15 08:49:11 -07005503 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005504 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005505 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005506
Jens Axboead8a48a2019-11-15 08:49:11 -07005507 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005508 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005509 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005510 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005511 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005512 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005513 flags = READ_ONCE(sqe->timeout_flags);
5514 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005515 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005516
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005517 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005518
Jens Axboee8c2bc12020-08-15 18:44:09 -07005519 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005520 return -ENOMEM;
5521
Jens Axboee8c2bc12020-08-15 18:44:09 -07005522 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005523 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005524
5525 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005526 return -EFAULT;
5527
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005528 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005529 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
Pavel Begunkovdd59a3d2021-03-04 13:59:25 +00005530 io_req_track_inflight(req);
Jens Axboead8a48a2019-11-15 08:49:11 -07005531 return 0;
5532}
5533
Pavel Begunkov61e98202021-02-10 00:03:08 +00005534static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005535{
Jens Axboead8a48a2019-11-15 08:49:11 -07005536 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005537 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005538 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005539 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005540
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005541 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005542
Jens Axboe5262f562019-09-17 12:26:57 -06005543 /*
5544 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005545 * timeout event to be satisfied. If it isn't set, then this is
5546 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005547 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005548 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005549 entry = ctx->timeout_list.prev;
5550 goto add;
5551 }
Jens Axboe5262f562019-09-17 12:26:57 -06005552
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005553 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5554 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005555
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005556 /* Update the last seq here in case io_flush_timeouts() hasn't.
5557 * This is safe because ->completion_lock is held, and submissions
5558 * and completions are never mixed in the same ->completion_lock section.
5559 */
5560 ctx->cq_last_tm_flush = tail;
5561
Jens Axboe5262f562019-09-17 12:26:57 -06005562 /*
5563 * Insertion sort, ensuring the first entry in the list is always
5564 * the one we need first.
5565 */
Jens Axboe5262f562019-09-17 12:26:57 -06005566 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005567 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5568 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005569
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005570 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005571 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005572 /* nxt.seq is behind @tail, otherwise would've been completed */
5573 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005574 break;
5575 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005576add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005577 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005578 data->timer.function = io_timeout_fn;
5579 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005580 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005581 return 0;
5582}
5583
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005584struct io_cancel_data {
5585 struct io_ring_ctx *ctx;
5586 u64 user_data;
5587};
5588
Jens Axboe62755e32019-10-28 21:49:21 -06005589static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005590{
Jens Axboe62755e32019-10-28 21:49:21 -06005591 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005592 struct io_cancel_data *cd = data;
Jens Axboede0617e2019-04-06 21:51:27 -06005593
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005594 return req->ctx == cd->ctx && req->user_data == cd->user_data;
Jens Axboe62755e32019-10-28 21:49:21 -06005595}
5596
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005597static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data,
5598 struct io_ring_ctx *ctx)
Jens Axboe62755e32019-10-28 21:49:21 -06005599{
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005600 struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, };
Jens Axboe62755e32019-10-28 21:49:21 -06005601 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005602 int ret = 0;
5603
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005604 if (!tctx || !tctx->io_wq)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005605 return -ENOENT;
5606
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005607 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005608 switch (cancel_ret) {
5609 case IO_WQ_CANCEL_OK:
5610 ret = 0;
5611 break;
5612 case IO_WQ_CANCEL_RUNNING:
5613 ret = -EALREADY;
5614 break;
5615 case IO_WQ_CANCEL_NOTFOUND:
5616 ret = -ENOENT;
5617 break;
5618 }
5619
Jens Axboee977d6d2019-11-05 12:39:45 -07005620 return ret;
5621}
5622
Jens Axboe47f46762019-11-09 17:43:02 -07005623static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5624 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005625 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005626{
5627 unsigned long flags;
5628 int ret;
5629
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005630 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
Jens Axboe47f46762019-11-09 17:43:02 -07005631 if (ret != -ENOENT) {
5632 spin_lock_irqsave(&ctx->completion_lock, flags);
5633 goto done;
5634 }
5635
5636 spin_lock_irqsave(&ctx->completion_lock, flags);
5637 ret = io_timeout_cancel(ctx, sqe_addr);
5638 if (ret != -ENOENT)
5639 goto done;
5640 ret = io_poll_cancel(ctx, sqe_addr);
5641done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005642 if (!ret)
5643 ret = success_ret;
Jens Axboe47f46762019-11-09 17:43:02 -07005644 io_cqring_fill_event(req, ret);
5645 io_commit_cqring(ctx);
5646 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5647 io_cqring_ev_posted(ctx);
5648
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005649 if (ret < 0)
5650 req_set_fail_links(req);
Pavel Begunkov014db002020-03-03 21:33:12 +03005651 io_put_req(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005652}
5653
Jens Axboe3529d8c2019-12-19 18:24:38 -07005654static int io_async_cancel_prep(struct io_kiocb *req,
5655 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005656{
Jens Axboefbf23842019-12-17 18:45:56 -07005657 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005658 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005659 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5660 return -EINVAL;
5661 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005662 return -EINVAL;
5663
Jens Axboefbf23842019-12-17 18:45:56 -07005664 req->cancel.addr = READ_ONCE(sqe->addr);
5665 return 0;
5666}
5667
Pavel Begunkov61e98202021-02-10 00:03:08 +00005668static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07005669{
5670 struct io_ring_ctx *ctx = req->ctx;
Jens Axboefbf23842019-12-17 18:45:56 -07005671
Pavel Begunkov014db002020-03-03 21:33:12 +03005672 io_async_find_and_cancel(ctx, req, req->cancel.addr, 0);
Jens Axboe62755e32019-10-28 21:49:21 -06005673 return 0;
5674}
5675
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005676static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07005677 const struct io_uring_sqe *sqe)
5678{
Jens Axboe6ca56f82020-09-18 16:51:19 -06005679 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
5680 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005681 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5682 return -EINVAL;
5683 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005684 return -EINVAL;
5685
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005686 req->rsrc_update.offset = READ_ONCE(sqe->off);
5687 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
5688 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005689 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005690 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005691 return 0;
5692}
5693
Pavel Begunkov889fca72021-02-10 00:03:09 +00005694static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005695{
5696 struct io_ring_ctx *ctx = req->ctx;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005697 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005698 int ret;
5699
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005700 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005701 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005702
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005703 up.offset = req->rsrc_update.offset;
5704 up.data = req->rsrc_update.arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005705
5706 mutex_lock(&ctx->uring_lock);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005707 ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005708 mutex_unlock(&ctx->uring_lock);
5709
5710 if (ret < 0)
5711 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005712 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005713 return 0;
5714}
5715
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005716static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005717{
Jens Axboed625c6e2019-12-17 19:53:05 -07005718 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005719 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005720 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005721 case IORING_OP_READV:
5722 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005723 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005724 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005725 case IORING_OP_WRITEV:
5726 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005727 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005728 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005729 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005730 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005731 case IORING_OP_POLL_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005732 return io_poll_remove_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005733 case IORING_OP_FSYNC:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005734 return io_fsync_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005735 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005736 return io_sfr_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005737 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005738 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005739 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005740 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005741 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005742 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005743 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005744 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005745 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005746 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005747 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005748 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005749 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005750 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005751 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005752 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005753 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005754 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005755 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005756 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005757 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005758 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005759 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005760 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005761 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005762 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005763 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005764 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005765 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005766 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005767 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005768 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005769 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005770 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005771 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005772 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005773 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005774 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005775 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005776 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005777 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005778 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005779 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005780 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06005781 case IORING_OP_SHUTDOWN:
5782 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06005783 case IORING_OP_RENAMEAT:
5784 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06005785 case IORING_OP_UNLINKAT:
5786 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005787 }
5788
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005789 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5790 req->opcode);
5791 return-EINVAL;
5792}
5793
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005794static int io_req_prep_async(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005795{
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005796 switch (req->opcode) {
5797 case IORING_OP_READV:
5798 case IORING_OP_READ_FIXED:
5799 case IORING_OP_READ:
5800 return io_rw_prep_async(req, READ);
5801 case IORING_OP_WRITEV:
5802 case IORING_OP_WRITE_FIXED:
5803 case IORING_OP_WRITE:
5804 return io_rw_prep_async(req, WRITE);
5805 case IORING_OP_SENDMSG:
5806 case IORING_OP_SEND:
5807 return io_sendmsg_prep_async(req);
5808 case IORING_OP_RECVMSG:
5809 case IORING_OP_RECV:
5810 return io_recvmsg_prep_async(req);
5811 case IORING_OP_CONNECT:
5812 return io_connect_prep_async(req);
5813 }
5814 return 0;
5815}
5816
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005817static int io_req_defer_prep(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005818{
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005819 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005820 return 0;
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005821 /* some opcodes init it during the inital prep */
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005822 if (req->async_data)
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005823 return 0;
5824 if (__io_alloc_async_data(req))
Jens Axboeb76da702019-11-20 13:05:32 -07005825 return -EAGAIN;
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005826 return io_req_prep_async(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005827}
5828
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005829static u32 io_get_sequence(struct io_kiocb *req)
5830{
5831 struct io_kiocb *pos;
5832 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005833 u32 total_submitted, nr_reqs = 0;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005834
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005835 io_for_each_link(pos, req)
5836 nr_reqs++;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005837
5838 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
5839 return total_submitted - nr_reqs;
5840}
5841
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005842static int io_req_defer(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005843{
5844 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005845 struct io_defer_entry *de;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005846 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005847 u32 seq;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005848
5849 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005850 if (likely(list_empty_careful(&ctx->defer_list) &&
5851 !(req->flags & REQ_F_IO_DRAIN)))
5852 return 0;
5853
5854 seq = io_get_sequence(req);
5855 /* Still a chance to pass the sequence check */
5856 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboe2b188cc2019-01-07 10:46:33 -07005857 return 0;
5858
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005859 ret = io_req_defer_prep(req);
5860 if (ret)
5861 return ret;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03005862 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005863 de = kmalloc(sizeof(*de), GFP_KERNEL);
5864 if (!de)
5865 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07005866
5867 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005868 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07005869 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005870 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03005871 io_queue_async_work(req);
5872 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07005873 }
5874
5875 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005876 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005877 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005878 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07005879 spin_unlock_irq(&ctx->completion_lock);
5880 return -EIOCBQUEUED;
5881}
Jens Axboeedafcce2019-01-09 09:16:05 -07005882
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03005883static void __io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005884{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005885 if (req->flags & REQ_F_BUFFER_SELECTED) {
5886 switch (req->opcode) {
5887 case IORING_OP_READV:
5888 case IORING_OP_READ_FIXED:
5889 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07005890 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005891 break;
5892 case IORING_OP_RECVMSG:
5893 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07005894 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005895 break;
5896 }
5897 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005898 }
5899
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005900 if (req->flags & REQ_F_NEED_CLEANUP) {
5901 switch (req->opcode) {
5902 case IORING_OP_READV:
5903 case IORING_OP_READ_FIXED:
5904 case IORING_OP_READ:
5905 case IORING_OP_WRITEV:
5906 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005907 case IORING_OP_WRITE: {
5908 struct io_async_rw *io = req->async_data;
5909 if (io->free_iovec)
5910 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005911 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005912 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005913 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005914 case IORING_OP_SENDMSG: {
5915 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00005916
5917 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005918 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005919 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005920 case IORING_OP_SPLICE:
5921 case IORING_OP_TEE:
5922 io_put_file(req, req->splice.file_in,
5923 (req->splice.flags & SPLICE_F_FD_IN_FIXED));
5924 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06005925 case IORING_OP_OPENAT:
5926 case IORING_OP_OPENAT2:
5927 if (req->open.filename)
5928 putname(req->open.filename);
5929 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06005930 case IORING_OP_RENAMEAT:
5931 putname(req->rename.oldpath);
5932 putname(req->rename.newpath);
5933 break;
Jens Axboe14a11432020-09-28 14:27:37 -06005934 case IORING_OP_UNLINKAT:
5935 putname(req->unlink.filename);
5936 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005937 }
5938 req->flags &= ~REQ_F_NEED_CLEANUP;
5939 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005940}
5941
Pavel Begunkov889fca72021-02-10 00:03:09 +00005942static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07005943{
Jens Axboeedafcce2019-01-09 09:16:05 -07005944 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5730b272021-02-27 15:57:30 -07005945 const struct cred *creds = NULL;
Jens Axboed625c6e2019-12-17 19:53:05 -07005946 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07005947
Jens Axboe003e8dc2021-03-06 09:22:27 -07005948 if (req->work.creds && req->work.creds != current_cred())
5949 creds = override_creds(req->work.creds);
Jens Axboe5730b272021-02-27 15:57:30 -07005950
Jens Axboed625c6e2019-12-17 19:53:05 -07005951 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07005952 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005953 ret = io_nop(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07005954 break;
5955 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07005956 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005957 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005958 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005959 break;
5960 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07005961 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005962 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005963 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005964 break;
5965 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005966 ret = io_fsync(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005967 break;
5968 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005969 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005970 break;
5971 case IORING_OP_POLL_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005972 ret = io_poll_remove(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005973 break;
5974 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005975 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005976 break;
5977 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005978 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005979 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005980 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005981 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005982 break;
5983 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005984 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005985 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005986 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005987 ret = io_recv(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005988 break;
5989 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005990 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005991 break;
5992 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005993 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005994 break;
5995 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005996 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005997 break;
5998 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005999 ret = io_connect(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006000 break;
6001 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006002 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006003 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07006004 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006005 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07006006 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07006007 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006008 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006009 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07006010 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006011 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07006012 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006013 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006014 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006015 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006016 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006017 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006018 break;
Jens Axboe4840e412019-12-25 22:03:45 -07006019 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006020 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006021 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006022 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006023 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006024 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006025 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006026 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006027 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006028 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006029 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006030 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006031 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006032 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006033 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006034 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006035 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006036 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006037 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006038 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006039 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006040 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006041 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006042 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006043 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006044 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006045 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006046 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006047 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006048 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006049 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006050 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006051 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006052 default:
6053 ret = -EINVAL;
6054 break;
Jens Axboe31b51512019-01-18 22:56:34 -07006055 }
6056
Jens Axboe5730b272021-02-27 15:57:30 -07006057 if (creds)
6058 revert_creds(creds);
6059
Jens Axboe2b188cc2019-01-07 10:46:33 -07006060 if (ret)
6061 return ret;
6062
Jens Axboeb5325762020-05-19 21:20:27 -06006063 /* If the op doesn't have a file, we're not polling for it */
6064 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006065 const bool in_async = io_wq_current_is_worker();
6066
Jens Axboe11ba8202020-01-15 21:51:17 -07006067 /* workqueue context doesn't hold uring_lock, grab it now */
6068 if (in_async)
6069 mutex_lock(&ctx->uring_lock);
6070
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08006071 io_iopoll_req_issued(req, in_async);
Jens Axboe11ba8202020-01-15 21:51:17 -07006072
6073 if (in_async)
6074 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006075 }
6076
6077 return 0;
6078}
6079
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006080static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006081{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006082 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006083 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006084 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006085
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006086 timeout = io_prep_linked_timeout(req);
6087 if (timeout)
6088 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006089
Jens Axboe4014d942021-01-19 15:53:54 -07006090 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006091 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006092
Jens Axboe561fb042019-10-24 07:25:42 -06006093 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006094 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006095 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006096 /*
6097 * We can get EAGAIN for polled IO even though we're
6098 * forcing a sync submission from here, since we can't
6099 * wait for request slots on the block side.
6100 */
6101 if (ret != -EAGAIN)
6102 break;
6103 cond_resched();
6104 } while (1);
6105 }
Jens Axboe31b51512019-01-18 22:56:34 -07006106
Pavel Begunkova3df76982021-02-18 22:32:52 +00006107 /* avoid locking problems by failing it from a clean context */
Jens Axboe561fb042019-10-24 07:25:42 -06006108 if (ret) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00006109 /* io-wq is going to take one down */
6110 refcount_inc(&req->refs);
6111 io_req_task_queue_fail(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006112 }
Jens Axboe31b51512019-01-18 22:56:34 -07006113}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006114
Jens Axboe65e19f52019-10-26 07:20:21 -06006115static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6116 int index)
Jens Axboe09bb8392019-03-13 12:39:28 -06006117{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006118 struct fixed_rsrc_table *table;
Jens Axboe65e19f52019-10-26 07:20:21 -06006119
Jens Axboe05f3fb32019-12-09 11:22:50 -07006120 table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT];
Xiaoming Ni84695082020-05-11 19:25:43 +08006121 return table->files[index & IORING_FILE_TABLE_MASK];
Jens Axboe65e19f52019-10-26 07:20:21 -06006122}
6123
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006124static struct file *io_file_get(struct io_submit_state *state,
6125 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006126{
6127 struct io_ring_ctx *ctx = req->ctx;
6128 struct file *file;
6129
6130 if (fixed) {
Pavel Begunkov479f5172020-10-10 18:34:07 +01006131 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006132 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006133 fd = array_index_nospec(fd, ctx->nr_user_files);
6134 file = io_file_from_index(ctx, fd);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00006135 io_set_resource_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006136 } else {
6137 trace_io_uring_file_get(ctx, fd);
6138 file = __io_file_get(state, fd);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006139 }
6140
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00006141 if (file && unlikely(file->f_op == &io_uring_fops))
6142 io_req_track_inflight(req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006143 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006144}
6145
Jens Axboe2665abf2019-11-05 12:40:47 -07006146static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6147{
Jens Axboead8a48a2019-11-15 08:49:11 -07006148 struct io_timeout_data *data = container_of(timer,
6149 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006150 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006151 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006152 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006153
6154 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006155 prev = req->timeout.head;
6156 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006157
6158 /*
6159 * We don't expect the list to be empty, that will only happen if we
6160 * race with the completion of the linked work.
6161 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006162 if (prev && refcount_inc_not_zero(&prev->refs))
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006163 io_remove_next_linked(prev);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006164 else
6165 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006166 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6167
6168 if (prev) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006169 req_set_fail_links(prev);
Pavel Begunkov014db002020-03-03 21:33:12 +03006170 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006171 io_put_req_deferred(prev, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006172 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006173 io_req_complete_post(req, -ETIME, 0);
6174 io_put_req_deferred(req, 1);
Jens Axboe2665abf2019-11-05 12:40:47 -07006175 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006176 return HRTIMER_NORESTART;
6177}
6178
Jens Axboe7271ef32020-08-10 09:55:22 -06006179static void __io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006180{
Jens Axboe76a46e02019-11-10 23:34:16 -07006181 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006182 * If the back reference is NULL, then our linked request finished
6183 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006184 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006185 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006186 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006187
Jens Axboead8a48a2019-11-15 08:49:11 -07006188 data->timer.function = io_link_timeout_fn;
6189 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6190 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006191 }
Jens Axboe7271ef32020-08-10 09:55:22 -06006192}
6193
6194static void io_queue_linked_timeout(struct io_kiocb *req)
6195{
6196 struct io_ring_ctx *ctx = req->ctx;
6197
6198 spin_lock_irq(&ctx->completion_lock);
6199 __io_queue_linked_timeout(req);
Jens Axboe76a46e02019-11-10 23:34:16 -07006200 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006201
Jens Axboe2665abf2019-11-05 12:40:47 -07006202 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006203 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006204}
6205
Jens Axboead8a48a2019-11-15 08:49:11 -07006206static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006207{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006208 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006209
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006210 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6211 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006212 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006213
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006214 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006215 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006216 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006217 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006218}
6219
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006220static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006221{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006222 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006223 int ret;
6224
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006225 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006226
6227 /*
6228 * We async punt it if the file wasn't marked NOWAIT, or if the file
6229 * doesn't support non-blocking read/write attempts
6230 */
Pavel Begunkov24c74672020-06-21 13:09:51 +03006231 if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006232 if (!io_arm_poll_handler(req)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006233 /*
6234 * Queued up for async execution, worker will release
6235 * submit reference when the iocb is actually submitted.
6236 */
6237 io_queue_async_work(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006238 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006239 } else if (likely(!ret)) {
6240 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006241 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006242 struct io_ring_ctx *ctx = req->ctx;
6243 struct io_comp_state *cs = &ctx->submit_state.comp;
Jens Axboee65ef562019-03-12 10:16:44 -06006244
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006245 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006246 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006247 io_submit_flush_completions(cs, ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006248 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006249 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006250 }
6251 } else {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006252 req_set_fail_links(req);
Jens Axboee65ef562019-03-12 10:16:44 -06006253 io_put_req(req);
Pavel Begunkov652532a2020-07-03 22:15:07 +03006254 io_req_complete(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006255 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006256 if (linked_timeout)
6257 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006258}
6259
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006260static void io_queue_sqe(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006261{
6262 int ret;
6263
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006264 ret = io_req_defer(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006265 if (ret) {
6266 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006267fail_req:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006268 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06006269 io_put_req(req);
6270 io_req_complete(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006271 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006272 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006273 ret = io_req_defer_prep(req);
6274 if (unlikely(ret))
6275 goto fail_req;
Jens Axboece35a472019-12-17 08:04:44 -07006276 io_queue_async_work(req);
6277 } else {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006278 __io_queue_sqe(req);
Jens Axboece35a472019-12-17 08:04:44 -07006279 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006280}
6281
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006282/*
6283 * Check SQE restrictions (opcode and flags).
6284 *
6285 * Returns 'true' if SQE is allowed, 'false' otherwise.
6286 */
6287static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6288 struct io_kiocb *req,
6289 unsigned int sqe_flags)
6290{
6291 if (!ctx->restricted)
6292 return true;
6293
6294 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6295 return false;
6296
6297 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6298 ctx->restrictions.sqe_flags_required)
6299 return false;
6300
6301 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6302 ctx->restrictions.sqe_flags_required))
6303 return false;
6304
6305 return true;
6306}
6307
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006308static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006309 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006310{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006311 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006312 unsigned int sqe_flags;
Jens Axboe003e8dc2021-03-06 09:22:27 -07006313 int personality, ret = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006314
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006315 req->opcode = READ_ONCE(sqe->opcode);
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006316 /* same numerical values with corresponding REQ_F_*, safe to copy */
6317 req->flags = sqe_flags = READ_ONCE(sqe->flags);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006318 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006319 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006320 req->file = NULL;
6321 req->ctx = ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006322 req->link = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006323 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006324 /* one is dropped after submission, the other at completion */
6325 refcount_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006326 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006327 req->result = 0;
Jens Axboe93e68e02021-03-09 07:02:21 -07006328 req->work.list.next = NULL;
6329 req->work.creds = NULL;
6330 req->work.flags = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006331
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006332 /* enforce forwards compatibility on users */
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006333 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
6334 req->flags = 0;
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006335 return -EINVAL;
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006336 }
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006337
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006338 if (unlikely(req->opcode >= IORING_OP_LAST))
6339 return -EINVAL;
6340
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006341 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6342 return -EACCES;
6343
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006344 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6345 !io_op_defs[req->opcode].buffer_select)
6346 return -EOPNOTSUPP;
6347
Jens Axboe003e8dc2021-03-06 09:22:27 -07006348 personality = READ_ONCE(sqe->personality);
6349 if (personality) {
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00006350 req->work.creds = xa_load(&ctx->personalities, personality);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006351 if (!req->work.creds)
6352 return -EINVAL;
6353 get_cred(req->work.creds);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006354 }
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006355 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006356
Jens Axboe27926b62020-10-28 09:33:23 -06006357 /*
6358 * Plug now if we have more than 1 IO left after this, and the target
6359 * is potentially a read/write to block based storage.
6360 */
6361 if (!state->plug_started && state->ios_left > 1 &&
6362 io_op_defs[req->opcode].plug) {
6363 blk_start_plug(&state->plug);
6364 state->plug_started = true;
6365 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006366
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006367 if (io_op_defs[req->opcode].needs_file) {
6368 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006369
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006370 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006371 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006372 ret = -EBADF;
6373 }
6374
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006375 state->ios_left--;
6376 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006377}
6378
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006379static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006380 const struct io_uring_sqe *sqe)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006381{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006382 struct io_submit_link *link = &ctx->submit_state.link;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006383 int ret;
6384
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006385 ret = io_init_req(ctx, req, sqe);
6386 if (unlikely(ret)) {
6387fail_req:
6388 io_put_req(req);
6389 io_req_complete(req, ret);
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006390 if (link->head) {
6391 /* fail even hard links since we don't submit */
Pavel Begunkovcf109602021-02-18 18:29:43 +00006392 link->head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006393 io_put_req(link->head);
6394 io_req_complete(link->head, -ECANCELED);
6395 link->head = NULL;
6396 }
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006397 return ret;
6398 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006399 ret = io_req_prep(req, sqe);
6400 if (unlikely(ret))
6401 goto fail_req;
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006402
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006403 /* don't need @sqe from now on */
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006404 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
6405 true, ctx->flags & IORING_SETUP_SQPOLL);
6406
Jens Axboe6c271ce2019-01-10 11:22:30 -07006407 /*
6408 * If we already have a head request, queue this one for async
6409 * submittal once the head completes. If we don't have a head but
6410 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6411 * submitted sync once the chain is complete. If none of those
6412 * conditions are true (normal request), then just queue it.
6413 */
6414 if (link->head) {
6415 struct io_kiocb *head = link->head;
6416
6417 /*
6418 * Taking sequential execution of a link, draining both sides
6419 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6420 * requests in the link. So, it drains the head and the
6421 * next after the link request. The last one is done via
6422 * drain_next flag to persist the effect across calls.
6423 */
6424 if (req->flags & REQ_F_IO_DRAIN) {
6425 head->flags |= REQ_F_IO_DRAIN;
6426 ctx->drain_next = 1;
6427 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006428 ret = io_req_defer_prep(req);
Pavel Begunkovcf109602021-02-18 18:29:43 +00006429 if (unlikely(ret))
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006430 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006431 trace_io_uring_link(ctx, req, head);
6432 link->last->link = req;
6433 link->last = req;
6434
6435 /* last request of a link, enqueue the link */
6436 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006437 io_queue_sqe(head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006438 link->head = NULL;
6439 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006440 } else {
6441 if (unlikely(ctx->drain_next)) {
6442 req->flags |= REQ_F_IO_DRAIN;
6443 ctx->drain_next = 0;
6444 }
6445 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Jackie Liu4fe2c962019-09-09 20:50:40 +08006446 link->head = req;
6447 link->last = req;
6448 } else {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006449 io_queue_sqe(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006450 }
6451 }
6452
6453 return 0;
6454}
6455
6456/*
6457 * Batched submission is done, ensure local IO is flushed out.
6458 */
6459static void io_submit_state_end(struct io_submit_state *state,
6460 struct io_ring_ctx *ctx)
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006461{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006462 if (state->link.head)
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006463 io_queue_sqe(state->link.head);
Jens Axboe3529d8c2019-12-19 18:24:38 -07006464 if (state->comp.nr)
Jens Axboe9e645e112019-05-10 16:07:28 -06006465 io_submit_flush_completions(&state->comp, ctx);
Jackie Liua197f662019-11-08 08:09:12 -07006466 if (state->plug_started)
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006467 blk_finish_plug(&state->plug);
Jens Axboe75c6a032020-01-28 10:15:23 -07006468 io_state_file_put(state);
Jens Axboe9e645e112019-05-10 16:07:28 -06006469}
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006470
Jens Axboe9e645e112019-05-10 16:07:28 -06006471/*
6472 * Start submission side cache.
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006473 */
Jens Axboe9e645e112019-05-10 16:07:28 -06006474static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkov196be952019-11-07 01:41:06 +03006475 unsigned int max_ios)
Jens Axboe9e645e112019-05-10 16:07:28 -06006476{
6477 state->plug_started = false;
Jens Axboebcda7ba2020-02-23 16:42:51 -07006478 state->ios_left = max_ios;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006479 /* set only head, no need to init link_last in advance */
6480 state->link.head = NULL;
Jens Axboe75c6a032020-01-28 10:15:23 -07006481}
6482
Jens Axboe193155c2020-02-22 23:22:19 -07006483static void io_commit_sqring(struct io_ring_ctx *ctx)
6484{
Jens Axboe75c6a032020-01-28 10:15:23 -07006485 struct io_rings *rings = ctx->rings;
6486
6487 /*
Jens Axboe193155c2020-02-22 23:22:19 -07006488 * Ensure any loads from the SQEs are done at this point,
Jens Axboe75c6a032020-01-28 10:15:23 -07006489 * since once we write the new head, the application could
6490 * write new data to them.
Pavel Begunkov6b47ee62020-01-18 20:22:41 +03006491 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006492 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboebcda7ba2020-02-23 16:42:51 -07006493}
6494
Jens Axboe9e645e112019-05-10 16:07:28 -06006495/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006496 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe9e645e112019-05-10 16:07:28 -06006497 * that is mapped by userspace. This means that care needs to be taken to
6498 * ensure that reads are stable, as we cannot rely on userspace always
Jens Axboe78e19bb2019-11-06 15:21:34 -07006499 * being a good citizen. If members of the sqe are validated and then later
6500 * used, it's important that those reads are done through READ_ONCE() to
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006501 * prevent a re-load down the line.
Jens Axboe9e645e112019-05-10 16:07:28 -06006502 */
6503static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe9e645e112019-05-10 16:07:28 -06006504{
6505 u32 *sq_array = ctx->sq_array;
6506 unsigned head;
6507
6508 /*
6509 * The cached sq head (or cq tail) serves two purposes:
6510 *
6511 * 1) allows us to batch the cost of updating the user visible
Pavel Begunkov9d763772019-12-17 02:22:07 +03006512 * head updates.
Jens Axboe9e645e112019-05-10 16:07:28 -06006513 * 2) allows the kernel side to track the head on its own, even
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006514 * though the application is the one updating it.
6515 */
6516 head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]);
6517 if (likely(head < ctx->sq_entries))
6518 return &ctx->sq_sqes[head];
6519
6520 /* drop invalid entries */
Pavel Begunkov711be032020-01-17 03:57:59 +03006521 ctx->cached_sq_dropped++;
6522 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
6523 return NULL;
6524}
Jens Axboeb7bb4f72019-12-15 22:13:43 -07006525
Jens Axboe0f212202020-09-13 13:09:39 -06006526static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006527{
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006528 int submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006529
Jens Axboec4a2ed72019-11-21 21:01:26 -07006530 /* if we have a backlog and couldn't flush it all, return BUSY */
Jens Axboead3eb2c2019-12-18 17:12:20 -07006531 if (test_bit(0, &ctx->sq_check_overflow)) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006532 if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
Jens Axboead3eb2c2019-12-18 17:12:20 -07006533 return -EBUSY;
6534 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006535
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006536 /* make sure SQ entry isn't read before tail */
6537 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006538
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006539 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6540 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006541
Jens Axboed8a6df12020-10-15 16:24:45 -06006542 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006543 refcount_add(nr, &current->usage);
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006544 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006545
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006546 while (submitted < nr) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006547 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006548 struct io_kiocb *req;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006549
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006550 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006551 if (unlikely(!req)) {
6552 if (!submitted)
6553 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006554 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006555 }
Pavel Begunkov4fccfcb2021-02-12 11:55:17 +00006556 sqe = io_get_sqe(ctx);
6557 if (unlikely(!sqe)) {
6558 kmem_cache_free(req_cachep, req);
6559 break;
6560 }
Jens Axboed3656342019-12-18 09:50:26 -07006561 /* will complete beyond this point, count as submitted */
6562 submitted++;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006563 if (io_submit_sqe(ctx, req, sqe))
Jens Axboed3656342019-12-18 09:50:26 -07006564 break;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006565 }
6566
Pavel Begunkov9466f432020-01-25 22:34:01 +03006567 if (unlikely(submitted != nr)) {
6568 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006569 struct io_uring_task *tctx = current->io_uring;
6570 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006571
Jens Axboed8a6df12020-10-15 16:24:45 -06006572 percpu_ref_put_many(&ctx->refs, unused);
6573 percpu_counter_sub(&tctx->inflight, unused);
6574 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006575 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006576
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006577 io_submit_state_end(&ctx->submit_state, ctx);
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006578 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6579 io_commit_sqring(ctx);
6580
Jens Axboe6c271ce2019-01-10 11:22:30 -07006581 return submitted;
6582}
6583
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006584static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6585{
6586 /* Tell userspace we may need a wakeup call */
6587 spin_lock_irq(&ctx->completion_lock);
6588 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6589 spin_unlock_irq(&ctx->completion_lock);
6590}
6591
6592static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6593{
6594 spin_lock_irq(&ctx->completion_lock);
6595 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6596 spin_unlock_irq(&ctx->completion_lock);
6597}
6598
Xiaoguang Wang08369242020-11-03 14:15:59 +08006599static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006600{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006601 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006602 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006603
Jens Axboec8d1ba52020-09-14 11:07:26 -06006604 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006605 /* if we're handling multiple rings, cap submit size for fairness */
6606 if (cap_entries && to_submit > 8)
6607 to_submit = 8;
6608
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006609 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6610 unsigned nr_events = 0;
6611
Xiaoguang Wang08369242020-11-03 14:15:59 +08006612 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006613 if (!list_empty(&ctx->iopoll_list))
6614 io_do_iopoll(ctx, &nr_events, 0);
6615
Pavel Begunkov0298ef92021-03-08 13:20:57 +00006616 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
6617 !(ctx->flags & IORING_SETUP_R_DISABLED))
Xiaoguang Wang08369242020-11-03 14:15:59 +08006618 ret = io_submit_sqes(ctx, to_submit);
6619 mutex_unlock(&ctx->uring_lock);
6620 }
Jens Axboe90554202020-09-03 12:12:41 -06006621
6622 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
6623 wake_up(&ctx->sqo_sq_wait);
6624
Xiaoguang Wang08369242020-11-03 14:15:59 +08006625 return ret;
6626}
6627
6628static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
6629{
6630 struct io_ring_ctx *ctx;
6631 unsigned sq_thread_idle = 0;
6632
6633 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6634 if (sq_thread_idle < ctx->sq_thread_idle)
6635 sq_thread_idle = ctx->sq_thread_idle;
6636 }
6637
6638 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006639}
6640
Jens Axboe6c271ce2019-01-10 11:22:30 -07006641static int io_sq_thread(void *data)
6642{
Jens Axboe69fb2132020-09-14 11:16:23 -06006643 struct io_sq_data *sqd = data;
6644 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d92052020-11-12 14:55:59 +08006645 unsigned long timeout = 0;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006646 char buf[TASK_COMM_LEN];
Xiaoguang Wang08369242020-11-03 14:15:59 +08006647 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006648
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006649 sprintf(buf, "iou-sqp-%d", sqd->task_pid);
6650 set_task_comm(current, buf);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006651 current->pf_io_worker = NULL;
Jens Axboe28cea78a2020-09-14 10:51:17 -06006652
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006653 if (sqd->sq_cpu != -1)
6654 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
6655 else
6656 set_cpus_allowed_ptr(current, cpu_online_mask);
6657 current->flags |= PF_NO_SETAFFINITY;
6658
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006659 wait_for_completion(&sqd->startup);
6660
Jens Axboe05962f92021-03-06 13:58:48 -07006661 down_read(&sqd->rw_lock);
6662
6663 while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006664 int ret;
6665 bool cap_entries, sqt_spin, needs_sched;
Jens Axboec1edbf52019-11-10 16:56:04 -07006666
Jens Axboe05962f92021-03-06 13:58:48 -07006667 if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) {
6668 up_read(&sqd->rw_lock);
6669 cond_resched();
6670 down_read(&sqd->rw_lock);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006671 timeout = jiffies + sqd->sq_thread_idle;
Pavel Begunkov7d41e852021-03-10 13:13:54 +00006672 continue;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006673 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006674 if (fatal_signal_pending(current))
6675 break;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006676 sqt_spin = false;
Jens Axboee95eee22020-09-08 09:11:32 -06006677 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006678 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006679 const struct cred *creds = NULL;
6680
6681 if (ctx->sq_creds != current_cred())
6682 creds = override_creds(ctx->sq_creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006683 ret = __io_sq_thread(ctx, cap_entries);
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006684 if (creds)
6685 revert_creds(creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006686 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
6687 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006688 }
6689
Xiaoguang Wang08369242020-11-03 14:15:59 +08006690 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006691 io_run_task_work();
6692 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08006693 if (sqt_spin)
6694 timeout = jiffies + sqd->sq_thread_idle;
6695 continue;
6696 }
6697
Xiaoguang Wang08369242020-11-03 14:15:59 +08006698 needs_sched = true;
6699 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
6700 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6701 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6702 !list_empty_careful(&ctx->iopoll_list)) {
6703 needs_sched = false;
6704 break;
6705 }
6706 if (io_sqring_entries(ctx)) {
6707 needs_sched = false;
6708 break;
6709 }
6710 }
6711
Jens Axboe05962f92021-03-06 13:58:48 -07006712 if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006713 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6714 io_ring_set_wakeup_flag(ctx);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006715
Jens Axboe05962f92021-03-06 13:58:48 -07006716 up_read(&sqd->rw_lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006717 schedule();
Jens Axboe05962f92021-03-06 13:58:48 -07006718 down_read(&sqd->rw_lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006719 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6720 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006721 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006722
6723 finish_wait(&sqd->wait, &wait);
6724 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006725 }
6726
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006727 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6728 io_uring_cancel_sqpoll(ctx);
Jens Axboe05962f92021-03-06 13:58:48 -07006729 up_read(&sqd->rw_lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006730
Jens Axboe4c6e2772020-07-01 11:29:10 -06006731 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07006732
Jens Axboe05962f92021-03-06 13:58:48 -07006733 down_write(&sqd->rw_lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006734 sqd->thread = NULL;
Jens Axboe05962f92021-03-06 13:58:48 -07006735 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
Jens Axboe5f3f26f2021-02-25 10:17:46 -07006736 io_ring_set_wakeup_flag(ctx);
Jens Axboe05962f92021-03-06 13:58:48 -07006737 up_write(&sqd->rw_lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006738 complete(&sqd->exited);
6739 do_exit(0);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006740}
6741
Jens Axboebda52162019-09-24 13:47:15 -06006742struct io_wait_queue {
6743 struct wait_queue_entry wq;
6744 struct io_ring_ctx *ctx;
6745 unsigned to_wait;
6746 unsigned nr_timeouts;
6747};
6748
Pavel Begunkov6c503152021-01-04 20:36:36 +00006749static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06006750{
6751 struct io_ring_ctx *ctx = iowq->ctx;
6752
6753 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006754 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006755 * started waiting. For timeouts, we always want to return to userspace,
6756 * regardless of event count.
6757 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00006758 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006759 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6760}
6761
6762static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6763 int wake_flags, void *key)
6764{
6765 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6766 wq);
6767
Pavel Begunkov6c503152021-01-04 20:36:36 +00006768 /*
6769 * Cannot safely flush overflowed CQEs from here, ensure we wake up
6770 * the task, and the next invocation will do it.
6771 */
6772 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
6773 return autoremove_wake_function(curr, mode, wake_flags, key);
6774 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06006775}
6776
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006777static int io_run_task_work_sig(void)
6778{
6779 if (io_run_task_work())
6780 return 1;
6781 if (!signal_pending(current))
6782 return 0;
Jens Axboe792ee0f62020-10-22 20:17:18 -06006783 if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL))
6784 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006785 return -EINTR;
6786}
6787
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006788/* when returns >0, the caller should retry */
6789static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
6790 struct io_wait_queue *iowq,
6791 signed long *timeout)
6792{
6793 int ret;
6794
6795 /* make sure we run task_work before checking for signals */
6796 ret = io_run_task_work_sig();
6797 if (ret || io_should_wake(iowq))
6798 return ret;
6799 /* let the caller flush overflows, retry */
6800 if (test_bit(0, &ctx->cq_check_overflow))
6801 return 1;
6802
6803 *timeout = schedule_timeout(*timeout);
6804 return !*timeout ? -ETIME : 1;
6805}
6806
Jens Axboe2b188cc2019-01-07 10:46:33 -07006807/*
6808 * Wait until events become available, if we don't already have some. The
6809 * application must reap them itself, as they reside on the shared cq ring.
6810 */
6811static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08006812 const sigset_t __user *sig, size_t sigsz,
6813 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006814{
Jens Axboebda52162019-09-24 13:47:15 -06006815 struct io_wait_queue iowq = {
6816 .wq = {
6817 .private = current,
6818 .func = io_wake_function,
6819 .entry = LIST_HEAD_INIT(iowq.wq.entry),
6820 },
6821 .ctx = ctx,
6822 .to_wait = min_events,
6823 };
Hristo Venev75b28af2019-08-26 17:23:46 +00006824 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006825 signed long timeout = MAX_SCHEDULE_TIMEOUT;
6826 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006827
Jens Axboeb41e9852020-02-17 09:52:41 -07006828 do {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006829 io_cqring_overflow_flush(ctx, false, NULL, NULL);
6830 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07006831 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06006832 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07006833 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07006834 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006835
6836 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006837#ifdef CONFIG_COMPAT
6838 if (in_compat_syscall())
6839 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07006840 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006841 else
6842#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07006843 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006844
Jens Axboe2b188cc2019-01-07 10:46:33 -07006845 if (ret)
6846 return ret;
6847 }
6848
Hao Xuc73ebb62020-11-03 10:54:37 +08006849 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006850 struct timespec64 ts;
6851
Hao Xuc73ebb62020-11-03 10:54:37 +08006852 if (get_timespec64(&ts, uts))
6853 return -EFAULT;
6854 timeout = timespec64_to_jiffies(&ts);
6855 }
6856
Jens Axboebda52162019-09-24 13:47:15 -06006857 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02006858 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06006859 do {
Jens Axboeca0a2652021-03-04 17:15:48 -07006860 /* if we can't even flush overflow, don't wait for more */
6861 if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) {
6862 ret = -EBUSY;
6863 break;
6864 }
Jens Axboebda52162019-09-24 13:47:15 -06006865 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
6866 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006867 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
6868 finish_wait(&ctx->wait, &iowq.wq);
Jens Axboeca0a2652021-03-04 17:15:48 -07006869 cond_resched();
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006870 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06006871
Jens Axboeb7db41c2020-07-04 08:55:50 -06006872 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006873
Hristo Venev75b28af2019-08-26 17:23:46 +00006874 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006875}
6876
Jens Axboe6b063142019-01-10 22:13:58 -07006877static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
6878{
6879#if defined(CONFIG_UNIX)
6880 if (ctx->ring_sock) {
6881 struct sock *sock = ctx->ring_sock->sk;
6882 struct sk_buff *skb;
6883
6884 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
6885 kfree_skb(skb);
6886 }
6887#else
6888 int i;
6889
Jens Axboe65e19f52019-10-26 07:20:21 -06006890 for (i = 0; i < ctx->nr_user_files; i++) {
6891 struct file *file;
6892
6893 file = io_file_from_index(ctx, i);
6894 if (file)
6895 fput(file);
6896 }
Jens Axboe6b063142019-01-10 22:13:58 -07006897#endif
6898}
6899
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00006900static void io_rsrc_data_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006901{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006902 struct fixed_rsrc_data *data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006903
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006904 data = container_of(ref, struct fixed_rsrc_data, refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006905 complete(&data->done);
6906}
6907
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006908static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
Pavel Begunkov1642b442020-12-30 21:34:14 +00006909{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006910 spin_lock_bh(&ctx->rsrc_ref_lock);
Pavel Begunkov1642b442020-12-30 21:34:14 +00006911}
6912
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006913static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
Jens Axboe6b063142019-01-10 22:13:58 -07006914{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006915 spin_unlock_bh(&ctx->rsrc_ref_lock);
6916}
6917
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00006918static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx,
6919 struct fixed_rsrc_data *rsrc_data,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006920 struct fixed_rsrc_ref_node *ref_node)
Jens Axboe6b063142019-01-10 22:13:58 -07006921{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006922 io_rsrc_ref_lock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006923 rsrc_data->node = ref_node;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00006924 list_add_tail(&ref_node->node, &ctx->rsrc_ref_list);
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006925 io_rsrc_ref_unlock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006926 percpu_ref_get(&rsrc_data->refs);
Jens Axboe6b063142019-01-10 22:13:58 -07006927}
6928
Hao Xu8bad28d2021-02-19 17:19:36 +08006929static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data)
Jens Axboe6b063142019-01-10 22:13:58 -07006930{
Hao Xu8bad28d2021-02-19 17:19:36 +08006931 struct fixed_rsrc_ref_node *ref_node = NULL;
Jens Axboe65e19f52019-10-26 07:20:21 -06006932
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006933 io_rsrc_ref_lock(ctx);
Pavel Begunkov1e5d7702020-11-18 14:56:25 +00006934 ref_node = data->node;
Pavel Begunkove6cb0072021-02-20 18:03:47 +00006935 data->node = NULL;
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006936 io_rsrc_ref_unlock(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08006937 if (ref_node)
6938 percpu_ref_kill(&ref_node->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08006939}
Xiaoguang Wang05589552020-03-31 14:05:18 +08006940
Hao Xu8bad28d2021-02-19 17:19:36 +08006941static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data,
6942 struct io_ring_ctx *ctx,
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006943 void (*rsrc_put)(struct io_ring_ctx *ctx,
6944 struct io_rsrc_put *prsrc))
Hao Xu8bad28d2021-02-19 17:19:36 +08006945{
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006946 struct fixed_rsrc_ref_node *backup_node;
Hao Xu8bad28d2021-02-19 17:19:36 +08006947 int ret;
Xiaoguang Wang05589552020-03-31 14:05:18 +08006948
Hao Xu8bad28d2021-02-19 17:19:36 +08006949 if (data->quiesce)
6950 return -ENXIO;
6951
6952 data->quiesce = true;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00006953 do {
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006954 ret = -ENOMEM;
6955 backup_node = alloc_fixed_rsrc_ref_node(ctx);
6956 if (!backup_node)
6957 break;
6958 backup_node->rsrc_data = data;
6959 backup_node->rsrc_put = rsrc_put;
6960
Hao Xu8bad28d2021-02-19 17:19:36 +08006961 io_sqe_rsrc_kill_node(ctx, data);
6962 percpu_ref_kill(&data->refs);
6963 flush_delayed_work(&ctx->rsrc_put_work);
6964
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00006965 ret = wait_for_completion_interruptible(&data->done);
6966 if (!ret)
6967 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006968
Jens Axboecb5e1b82021-02-25 07:37:35 -07006969 percpu_ref_resurrect(&data->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08006970 io_sqe_rsrc_set_node(ctx, data, backup_node);
6971 backup_node = NULL;
Jens Axboecb5e1b82021-02-25 07:37:35 -07006972 reinit_completion(&data->done);
Hao Xu8bad28d2021-02-19 17:19:36 +08006973 mutex_unlock(&ctx->uring_lock);
6974 ret = io_run_task_work_sig();
6975 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006976 } while (ret >= 0);
Hao Xu8bad28d2021-02-19 17:19:36 +08006977 data->quiesce = false;
6978
6979 if (backup_node)
6980 destroy_fixed_rsrc_ref_node(backup_node);
6981 return ret;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00006982}
6983
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00006984static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx)
6985{
6986 struct fixed_rsrc_data *data;
6987
6988 data = kzalloc(sizeof(*data), GFP_KERNEL);
6989 if (!data)
6990 return NULL;
6991
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00006992 if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero,
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00006993 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
6994 kfree(data);
6995 return NULL;
6996 }
6997 data->ctx = ctx;
6998 init_completion(&data->done);
6999 return data;
7000}
7001
7002static void free_fixed_rsrc_data(struct fixed_rsrc_data *data)
7003{
7004 percpu_ref_exit(&data->refs);
7005 kfree(data->table);
7006 kfree(data);
7007}
7008
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007009static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7010{
7011 struct fixed_rsrc_data *data = ctx->file_data;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007012 unsigned nr_tables, i;
7013 int ret;
7014
Hao Xu8bad28d2021-02-19 17:19:36 +08007015 /*
7016 * percpu_ref_is_dying() is to stop parallel files unregister
7017 * Since we possibly drop uring lock later in this function to
7018 * run task work.
7019 */
7020 if (!data || percpu_ref_is_dying(&data->refs))
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007021 return -ENXIO;
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007022 ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put);
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007023 if (ret)
7024 return ret;
7025
Jens Axboe6b063142019-01-10 22:13:58 -07007026 __io_sqe_files_unregister(ctx);
Jens Axboe65e19f52019-10-26 07:20:21 -06007027 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
7028 for (i = 0; i < nr_tables; i++)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007029 kfree(data->table[i].files);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007030 free_fixed_rsrc_data(data);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007031 ctx->file_data = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07007032 ctx->nr_user_files = 0;
7033 return 0;
7034}
7035
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007036static void io_sq_thread_unpark(struct io_sq_data *sqd)
Jens Axboe05962f92021-03-06 13:58:48 -07007037 __releases(&sqd->rw_lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007038{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007039 if (sqd->thread == current)
7040 return;
7041 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Jens Axboe05962f92021-03-06 13:58:48 -07007042 up_write(&sqd->rw_lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007043}
7044
Jens Axboe86e0d672021-03-05 08:44:39 -07007045static void io_sq_thread_park(struct io_sq_data *sqd)
Jens Axboe05962f92021-03-06 13:58:48 -07007046 __acquires(&sqd->rw_lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007047{
7048 if (sqd->thread == current)
Jens Axboe86e0d672021-03-05 08:44:39 -07007049 return;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007050 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Jens Axboe05962f92021-03-06 13:58:48 -07007051 down_write(&sqd->rw_lock);
7052 /* set again for consistency, in case concurrent parks are happening */
7053 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
7054 if (sqd->thread)
Jens Axboe86e0d672021-03-05 08:44:39 -07007055 wake_up_process(sqd->thread);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007056}
7057
7058static void io_sq_thread_stop(struct io_sq_data *sqd)
7059{
Jens Axboee54945a2021-02-26 11:27:15 -07007060 if (test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state))
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007061 return;
Jens Axboe05962f92021-03-06 13:58:48 -07007062 down_write(&sqd->rw_lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007063 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
Jens Axboee8f98f242021-03-09 16:32:13 -07007064 if (sqd->thread)
7065 wake_up_process(sqd->thread);
Jens Axboe05962f92021-03-06 13:58:48 -07007066 up_write(&sqd->rw_lock);
7067 wait_for_completion(&sqd->exited);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007068}
7069
Jens Axboe534ca6d2020-09-02 13:52:19 -06007070static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007071{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007072 if (refcount_dec_and_test(&sqd->refs)) {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007073 io_sq_thread_stop(sqd);
7074 kfree(sqd);
7075 }
7076}
7077
7078static void io_sq_thread_finish(struct io_ring_ctx *ctx)
7079{
7080 struct io_sq_data *sqd = ctx->sq_data;
7081
7082 if (sqd) {
Jens Axboeeb858902021-02-25 10:13:29 -07007083 complete(&sqd->startup);
Jens Axboe05962f92021-03-06 13:58:48 -07007084 io_sq_thread_park(sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007085 list_del(&ctx->sqd_list);
7086 io_sqd_update_thread_idle(sqd);
Jens Axboe05962f92021-03-06 13:58:48 -07007087 io_sq_thread_unpark(sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007088
7089 io_put_sq_data(sqd);
7090 ctx->sq_data = NULL;
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01007091 if (ctx->sq_creds)
7092 put_cred(ctx->sq_creds);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007093 }
7094}
7095
Jens Axboeaa061652020-09-02 14:50:27 -06007096static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7097{
7098 struct io_ring_ctx *ctx_attach;
7099 struct io_sq_data *sqd;
7100 struct fd f;
7101
7102 f = fdget(p->wq_fd);
7103 if (!f.file)
7104 return ERR_PTR(-ENXIO);
7105 if (f.file->f_op != &io_uring_fops) {
7106 fdput(f);
7107 return ERR_PTR(-EINVAL);
7108 }
7109
7110 ctx_attach = f.file->private_data;
7111 sqd = ctx_attach->sq_data;
7112 if (!sqd) {
7113 fdput(f);
7114 return ERR_PTR(-EINVAL);
7115 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007116 if (sqd->task_tgid != current->tgid) {
7117 fdput(f);
7118 return ERR_PTR(-EPERM);
7119 }
Jens Axboeaa061652020-09-02 14:50:27 -06007120
7121 refcount_inc(&sqd->refs);
7122 fdput(f);
7123 return sqd;
7124}
7125
Jens Axboe534ca6d2020-09-02 13:52:19 -06007126static struct io_sq_data *io_get_sq_data(struct io_uring_params *p)
7127{
7128 struct io_sq_data *sqd;
7129
Jens Axboe5c2469e2021-03-11 10:17:56 -07007130 if (p->flags & IORING_SETUP_ATTACH_WQ) {
7131 sqd = io_attach_sq_data(p);
7132 if (!IS_ERR(sqd))
7133 return sqd;
7134 /* fall through for EPERM case, setup new sqd/task */
7135 if (PTR_ERR(sqd) != -EPERM)
7136 return sqd;
7137 }
Jens Axboeaa061652020-09-02 14:50:27 -06007138
Jens Axboe534ca6d2020-09-02 13:52:19 -06007139 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7140 if (!sqd)
7141 return ERR_PTR(-ENOMEM);
7142
7143 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007144 INIT_LIST_HEAD(&sqd->ctx_list);
Jens Axboe05962f92021-03-06 13:58:48 -07007145 init_rwsem(&sqd->rw_lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007146 init_waitqueue_head(&sqd->wait);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007147 init_completion(&sqd->startup);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007148 init_completion(&sqd->exited);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007149 return sqd;
7150}
7151
Jens Axboe6b063142019-01-10 22:13:58 -07007152#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007153/*
7154 * Ensure the UNIX gc is aware of our file set, so we are certain that
7155 * the io_uring can be safely unregistered on process exit, even if we have
7156 * loops in the file referencing.
7157 */
7158static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7159{
7160 struct sock *sk = ctx->ring_sock->sk;
7161 struct scm_fp_list *fpl;
7162 struct sk_buff *skb;
Jens Axboe08a45172019-10-03 08:11:03 -06007163 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007164
Jens Axboe6b063142019-01-10 22:13:58 -07007165 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7166 if (!fpl)
7167 return -ENOMEM;
7168
7169 skb = alloc_skb(0, GFP_KERNEL);
7170 if (!skb) {
7171 kfree(fpl);
7172 return -ENOMEM;
7173 }
7174
7175 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007176
Jens Axboe08a45172019-10-03 08:11:03 -06007177 nr_files = 0;
Jens Axboe62e398b2021-02-21 16:19:37 -07007178 fpl->user = get_uid(current_user());
Jens Axboe6b063142019-01-10 22:13:58 -07007179 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007180 struct file *file = io_file_from_index(ctx, i + offset);
7181
7182 if (!file)
Jens Axboe08a45172019-10-03 08:11:03 -06007183 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007184 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a45172019-10-03 08:11:03 -06007185 unix_inflight(fpl->user, fpl->fp[nr_files]);
7186 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007187 }
7188
Jens Axboe08a45172019-10-03 08:11:03 -06007189 if (nr_files) {
7190 fpl->max = SCM_MAX_FD;
7191 fpl->count = nr_files;
7192 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007193 skb->destructor = unix_destruct_scm;
Jens Axboe08a45172019-10-03 08:11:03 -06007194 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7195 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007196
Jens Axboe08a45172019-10-03 08:11:03 -06007197 for (i = 0; i < nr_files; i++)
7198 fput(fpl->fp[i]);
7199 } else {
7200 kfree_skb(skb);
7201 kfree(fpl);
7202 }
Jens Axboe6b063142019-01-10 22:13:58 -07007203
7204 return 0;
7205}
7206
7207/*
7208 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7209 * causes regular reference counting to break down. We rely on the UNIX
7210 * garbage collection to take care of this problem for us.
7211 */
7212static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7213{
7214 unsigned left, total;
7215 int ret = 0;
7216
7217 total = 0;
7218 left = ctx->nr_user_files;
7219 while (left) {
7220 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007221
7222 ret = __io_sqe_files_scm(ctx, this_files, total);
7223 if (ret)
7224 break;
7225 left -= this_files;
7226 total += this_files;
7227 }
7228
7229 if (!ret)
7230 return 0;
7231
7232 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007233 struct file *file = io_file_from_index(ctx, total);
7234
7235 if (file)
7236 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007237 total++;
7238 }
7239
7240 return ret;
7241}
7242#else
7243static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7244{
7245 return 0;
7246}
7247#endif
7248
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007249static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data,
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007250 unsigned nr_tables, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007251{
7252 int i;
7253
7254 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007255 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007256 unsigned this_files;
7257
7258 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
7259 table->files = kcalloc(this_files, sizeof(struct file *),
7260 GFP_KERNEL);
7261 if (!table->files)
7262 break;
7263 nr_files -= this_files;
7264 }
7265
7266 if (i == nr_tables)
7267 return 0;
7268
7269 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007270 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007271 kfree(table->files);
7272 }
7273 return 1;
7274}
7275
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007276static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007277{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007278 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007279#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007280 struct sock *sock = ctx->ring_sock->sk;
7281 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7282 struct sk_buff *skb;
7283 int i;
7284
7285 __skb_queue_head_init(&list);
7286
7287 /*
7288 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7289 * remove this entry and rearrange the file array.
7290 */
7291 skb = skb_dequeue(head);
7292 while (skb) {
7293 struct scm_fp_list *fp;
7294
7295 fp = UNIXCB(skb).fp;
7296 for (i = 0; i < fp->count; i++) {
7297 int left;
7298
7299 if (fp->fp[i] != file)
7300 continue;
7301
7302 unix_notinflight(fp->user, fp->fp[i]);
7303 left = fp->count - 1 - i;
7304 if (left) {
7305 memmove(&fp->fp[i], &fp->fp[i + 1],
7306 left * sizeof(struct file *));
7307 }
7308 fp->count--;
7309 if (!fp->count) {
7310 kfree_skb(skb);
7311 skb = NULL;
7312 } else {
7313 __skb_queue_tail(&list, skb);
7314 }
7315 fput(file);
7316 file = NULL;
7317 break;
7318 }
7319
7320 if (!file)
7321 break;
7322
7323 __skb_queue_tail(&list, skb);
7324
7325 skb = skb_dequeue(head);
7326 }
7327
7328 if (skb_peek(&list)) {
7329 spin_lock_irq(&head->lock);
7330 while ((skb = __skb_dequeue(&list)) != NULL)
7331 __skb_queue_tail(head, skb);
7332 spin_unlock_irq(&head->lock);
7333 }
7334#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007335 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007336#endif
7337}
7338
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007339static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007340{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007341 struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data;
7342 struct io_ring_ctx *ctx = rsrc_data->ctx;
7343 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007344
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007345 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7346 list_del(&prsrc->list);
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007347 ref_node->rsrc_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007348 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007349 }
7350
Xiaoguang Wang05589552020-03-31 14:05:18 +08007351 percpu_ref_exit(&ref_node->refs);
7352 kfree(ref_node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007353 percpu_ref_put(&rsrc_data->refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007354}
7355
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007356static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007357{
7358 struct io_ring_ctx *ctx;
7359 struct llist_node *node;
7360
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007361 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7362 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007363
7364 while (node) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007365 struct fixed_rsrc_ref_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007366 struct llist_node *next = node->next;
7367
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007368 ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist);
7369 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007370 node = next;
7371 }
7372}
7373
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007374static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data,
7375 unsigned i)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007376{
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007377 struct fixed_rsrc_table *table;
7378
7379 table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT];
7380 return &table->files[i & IORING_FILE_TABLE_MASK];
7381}
7382
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007383static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007384{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007385 struct fixed_rsrc_ref_node *ref_node;
7386 struct fixed_rsrc_data *data;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007387 struct io_ring_ctx *ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007388 bool first_add = false;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007389 int delay = HZ;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007390
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007391 ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs);
7392 data = ref_node->rsrc_data;
Pavel Begunkove2978222020-11-18 14:56:26 +00007393 ctx = data->ctx;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007394
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007395 io_rsrc_ref_lock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007396 ref_node->done = true;
7397
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007398 while (!list_empty(&ctx->rsrc_ref_list)) {
7399 ref_node = list_first_entry(&ctx->rsrc_ref_list,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007400 struct fixed_rsrc_ref_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007401 /* recycle ref nodes in order */
7402 if (!ref_node->done)
7403 break;
7404 list_del(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007405 first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007406 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007407 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007408
7409 if (percpu_ref_is_dying(&data->refs))
Jens Axboe4a38aed22020-05-14 17:21:15 -06007410 delay = 0;
7411
Jens Axboe4a38aed22020-05-14 17:21:15 -06007412 if (!delay)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007413 mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007414 else if (first_add)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007415 queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007416}
7417
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007418static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Xiaoguang Wang05589552020-03-31 14:05:18 +08007419 struct io_ring_ctx *ctx)
7420{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007421 struct fixed_rsrc_ref_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007422
7423 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7424 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007425 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007426
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007427 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007428 0, GFP_KERNEL)) {
7429 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007430 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007431 }
7432 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007433 INIT_LIST_HEAD(&ref_node->rsrc_list);
Pavel Begunkove2978222020-11-18 14:56:26 +00007434 ref_node->done = false;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007435 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007436}
7437
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007438static void init_fixed_file_ref_node(struct io_ring_ctx *ctx,
7439 struct fixed_rsrc_ref_node *ref_node)
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007440{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007441 ref_node->rsrc_data = ctx->file_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007442 ref_node->rsrc_put = io_ring_file_put;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007443}
7444
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007445static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007446{
7447 percpu_ref_exit(&ref_node->refs);
7448 kfree(ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007449}
7450
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007451
Jens Axboe05f3fb32019-12-09 11:22:50 -07007452static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7453 unsigned nr_args)
7454{
7455 __s32 __user *fds = (__s32 __user *) arg;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007456 unsigned nr_tables, i;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007457 struct file *file;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007458 int fd, ret = -ENOMEM;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007459 struct fixed_rsrc_ref_node *ref_node;
7460 struct fixed_rsrc_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007461
7462 if (ctx->file_data)
7463 return -EBUSY;
7464 if (!nr_args)
7465 return -EINVAL;
7466 if (nr_args > IORING_MAX_FIXED_FILES)
7467 return -EMFILE;
7468
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007469 file_data = alloc_fixed_rsrc_data(ctx);
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007470 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007471 return -ENOMEM;
Dan Carpenter13770a72021-02-01 15:23:42 +03007472 ctx->file_data = file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007473
7474 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
Colin Ian King035fbaf2020-10-12 15:03:41 +01007475 file_data->table = kcalloc(nr_tables, sizeof(*file_data->table),
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007476 GFP_KERNEL);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007477 if (!file_data->table)
7478 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007479
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007480 if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args))
Jens Axboe05f3fb32019-12-09 11:22:50 -07007481 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007482
Jens Axboe05f3fb32019-12-09 11:22:50 -07007483 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007484 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7485 ret = -EFAULT;
7486 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007487 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007488 /* allow sparse sets */
7489 if (fd == -1)
7490 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007491
Jens Axboe05f3fb32019-12-09 11:22:50 -07007492 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007493 ret = -EBADF;
7494 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007495 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007496
7497 /*
7498 * Don't allow io_uring instances to be registered. If UNIX
7499 * isn't enabled, then this causes a reference cycle and this
7500 * instance can never get freed. If UNIX is enabled we'll
7501 * handle it just fine, but there's still no point in allowing
7502 * a ring fd as it doesn't support regular read/write anyway.
7503 */
7504 if (file->f_op == &io_uring_fops) {
7505 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007506 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007507 }
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007508 *io_fixed_file_slot(file_data, i) = file;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007509 }
7510
Jens Axboe05f3fb32019-12-09 11:22:50 -07007511 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007512 if (ret) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07007513 io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007514 return ret;
7515 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007516
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007517 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007518 if (!ref_node) {
Xiaoguang Wang05589552020-03-31 14:05:18 +08007519 io_sqe_files_unregister(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007520 return -ENOMEM;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007521 }
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007522 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007523
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007524 io_sqe_rsrc_set_node(ctx, file_data, ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007525 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007526out_fput:
7527 for (i = 0; i < ctx->nr_user_files; i++) {
7528 file = io_file_from_index(ctx, i);
7529 if (file)
7530 fput(file);
7531 }
7532 for (i = 0; i < nr_tables; i++)
7533 kfree(file_data->table[i].files);
7534 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007535out_free:
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007536 free_fixed_rsrc_data(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007537 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007538 return ret;
7539}
7540
Jens Axboec3a31e62019-10-03 13:59:56 -06007541static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7542 int index)
7543{
7544#if defined(CONFIG_UNIX)
7545 struct sock *sock = ctx->ring_sock->sk;
7546 struct sk_buff_head *head = &sock->sk_receive_queue;
7547 struct sk_buff *skb;
7548
7549 /*
7550 * See if we can merge this file into an existing skb SCM_RIGHTS
7551 * file set. If there's no room, fall back to allocating a new skb
7552 * and filling it in.
7553 */
7554 spin_lock_irq(&head->lock);
7555 skb = skb_peek(head);
7556 if (skb) {
7557 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7558
7559 if (fpl->count < SCM_MAX_FD) {
7560 __skb_unlink(skb, head);
7561 spin_unlock_irq(&head->lock);
7562 fpl->fp[fpl->count] = get_file(file);
7563 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7564 fpl->count++;
7565 spin_lock_irq(&head->lock);
7566 __skb_queue_head(head, skb);
7567 } else {
7568 skb = NULL;
7569 }
7570 }
7571 spin_unlock_irq(&head->lock);
7572
7573 if (skb) {
7574 fput(file);
7575 return 0;
7576 }
7577
7578 return __io_sqe_files_scm(ctx, 1, index);
7579#else
7580 return 0;
7581#endif
7582}
7583
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007584static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007585{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007586 struct io_rsrc_put *prsrc;
7587 struct fixed_rsrc_ref_node *ref_node = data->node;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007588
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007589 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7590 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007591 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007592
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007593 prsrc->rsrc = rsrc;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007594 list_add(&prsrc->list, &ref_node->rsrc_list);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007595
Hillf Dantona5318d32020-03-23 17:47:15 +08007596 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007597}
7598
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007599static inline int io_queue_file_removal(struct fixed_rsrc_data *data,
7600 struct file *file)
7601{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007602 return io_queue_rsrc_removal(data, (void *)file);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007603}
7604
Jens Axboe05f3fb32019-12-09 11:22:50 -07007605static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007606 struct io_uring_rsrc_update *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07007607 unsigned nr_args)
7608{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007609 struct fixed_rsrc_data *data = ctx->file_data;
7610 struct fixed_rsrc_ref_node *ref_node;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007611 struct file *file, **file_slot;
Jens Axboec3a31e62019-10-03 13:59:56 -06007612 __s32 __user *fds;
7613 int fd, i, err;
7614 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007615 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007616
Jens Axboe05f3fb32019-12-09 11:22:50 -07007617 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06007618 return -EOVERFLOW;
7619 if (done > ctx->nr_user_files)
7620 return -EINVAL;
7621
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007622 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007623 if (!ref_node)
7624 return -ENOMEM;
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007625 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007626
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007627 fds = u64_to_user_ptr(up->data);
Pavel Begunkov67973b92021-01-26 13:51:09 +00007628 for (done = 0; done < nr_args; done++) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007629 err = 0;
7630 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
7631 err = -EFAULT;
7632 break;
7633 }
noah4e0377a2021-01-26 15:23:28 -05007634 if (fd == IORING_REGISTER_FILES_SKIP)
7635 continue;
7636
Pavel Begunkov67973b92021-01-26 13:51:09 +00007637 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007638 file_slot = io_fixed_file_slot(ctx->file_data, i);
7639
7640 if (*file_slot) {
7641 err = io_queue_file_removal(data, *file_slot);
Hillf Dantona5318d32020-03-23 17:47:15 +08007642 if (err)
7643 break;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007644 *file_slot = NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007645 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007646 }
7647 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007648 file = fget(fd);
7649 if (!file) {
7650 err = -EBADF;
7651 break;
7652 }
7653 /*
7654 * Don't allow io_uring instances to be registered. If
7655 * UNIX isn't enabled, then this causes a reference
7656 * cycle and this instance can never get freed. If UNIX
7657 * is enabled we'll handle it just fine, but there's
7658 * still no point in allowing a ring fd as it doesn't
7659 * support regular read/write anyway.
7660 */
7661 if (file->f_op == &io_uring_fops) {
7662 fput(file);
7663 err = -EBADF;
7664 break;
7665 }
Jens Axboee68a3ff2021-02-11 07:45:08 -07007666 *file_slot = file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007667 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007668 if (err) {
Jens Axboee68a3ff2021-02-11 07:45:08 -07007669 *file_slot = NULL;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007670 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007671 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007672 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007673 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007674 }
7675
Xiaoguang Wang05589552020-03-31 14:05:18 +08007676 if (needs_switch) {
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007677 percpu_ref_kill(&data->node->refs);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007678 io_sqe_rsrc_set_node(ctx, data, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007679 } else
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007680 destroy_fixed_rsrc_ref_node(ref_node);
Jens Axboec3a31e62019-10-03 13:59:56 -06007681
7682 return done ? done : err;
7683}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007684
Jens Axboe05f3fb32019-12-09 11:22:50 -07007685static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
7686 unsigned nr_args)
7687{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007688 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007689
7690 if (!ctx->file_data)
7691 return -ENXIO;
7692 if (!nr_args)
7693 return -EINVAL;
7694 if (copy_from_user(&up, arg, sizeof(up)))
7695 return -EFAULT;
7696 if (up.resv)
7697 return -EINVAL;
7698
7699 return __io_sqe_files_update(ctx, &up, nr_args);
7700}
Jens Axboec3a31e62019-10-03 13:59:56 -06007701
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007702static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007703{
7704 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7705
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007706 req = io_put_req_find_next(req);
7707 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07007708}
7709
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007710static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx)
Pavel Begunkov24369c22020-01-28 03:15:48 +03007711{
Jens Axboee9418942021-02-19 12:33:30 -07007712 struct io_wq_hash *hash;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007713 struct io_wq_data data;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007714 unsigned int concurrency;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007715
Jens Axboee9418942021-02-19 12:33:30 -07007716 hash = ctx->hash_map;
7717 if (!hash) {
7718 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
7719 if (!hash)
7720 return ERR_PTR(-ENOMEM);
7721 refcount_set(&hash->refs, 1);
7722 init_waitqueue_head(&hash->wait);
7723 ctx->hash_map = hash;
7724 }
7725
7726 data.hash = hash;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007727 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007728 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007729
Jens Axboed25e3a32021-02-16 11:41:41 -07007730 /* Do QD, or 4 * CPUS, whatever is smallest */
7731 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
Pavel Begunkov24369c22020-01-28 03:15:48 +03007732
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007733 return io_wq_create(concurrency, &data);
Pavel Begunkov24369c22020-01-28 03:15:48 +03007734}
7735
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007736static int io_uring_alloc_task_context(struct task_struct *task,
7737 struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06007738{
7739 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007740 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007741
7742 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
7743 if (unlikely(!tctx))
7744 return -ENOMEM;
7745
Jens Axboed8a6df12020-10-15 16:24:45 -06007746 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7747 if (unlikely(ret)) {
7748 kfree(tctx);
7749 return ret;
7750 }
7751
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007752 tctx->io_wq = io_init_wq_offload(ctx);
7753 if (IS_ERR(tctx->io_wq)) {
7754 ret = PTR_ERR(tctx->io_wq);
7755 percpu_counter_destroy(&tctx->inflight);
7756 kfree(tctx);
7757 return ret;
7758 }
7759
Jens Axboe0f212202020-09-13 13:09:39 -06007760 xa_init(&tctx->xa);
7761 init_waitqueue_head(&tctx->wait);
7762 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06007763 atomic_set(&tctx->in_idle, 0);
7764 tctx->sqpoll = false;
Jens Axboe0f212202020-09-13 13:09:39 -06007765 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00007766 spin_lock_init(&tctx->task_lock);
7767 INIT_WQ_LIST(&tctx->task_list);
7768 tctx->task_state = 0;
7769 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06007770 return 0;
7771}
7772
7773void __io_uring_free(struct task_struct *tsk)
7774{
7775 struct io_uring_task *tctx = tsk->io_uring;
7776
7777 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00007778 WARN_ON_ONCE(tctx->io_wq);
7779
Jens Axboed8a6df12020-10-15 16:24:45 -06007780 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06007781 kfree(tctx);
7782 tsk->io_uring = NULL;
7783}
7784
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007785static int io_sq_offload_create(struct io_ring_ctx *ctx,
7786 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007787{
7788 int ret;
7789
Jens Axboed25e3a32021-02-16 11:41:41 -07007790 /* Retain compatibility with failing for an invalid attach attempt */
7791 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
7792 IORING_SETUP_ATTACH_WQ) {
7793 struct fd f;
7794
7795 f = fdget(p->wq_fd);
7796 if (!f.file)
7797 return -ENXIO;
7798 if (f.file->f_op != &io_uring_fops) {
7799 fdput(f);
7800 return -EINVAL;
7801 }
7802 fdput(f);
7803 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07007804 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe46fe18b2021-03-04 12:39:36 -07007805 struct task_struct *tsk;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007806 struct io_sq_data *sqd;
7807
Jens Axboe3ec482d2019-04-08 10:51:01 -06007808 ret = -EPERM;
Jens Axboece59fc62020-09-02 13:28:09 -06007809 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE))
Jens Axboe3ec482d2019-04-08 10:51:01 -06007810 goto err;
7811
Jens Axboe534ca6d2020-09-02 13:52:19 -06007812 sqd = io_get_sq_data(p);
7813 if (IS_ERR(sqd)) {
7814 ret = PTR_ERR(sqd);
7815 goto err;
7816 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007817
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01007818 ctx->sq_creds = get_current_cred();
Jens Axboe534ca6d2020-09-02 13:52:19 -06007819 ctx->sq_data = sqd;
Jens Axboe917257d2019-04-13 09:28:55 -06007820 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
7821 if (!ctx->sq_thread_idle)
7822 ctx->sq_thread_idle = HZ;
7823
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007824 io_sq_thread_park(sqd);
Pavel Begunkov7d41e852021-03-10 13:13:54 +00007825 list_add(&ctx->sqd_list, &sqd->ctx_list);
7826 io_sqd_update_thread_idle(sqd);
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007827 io_sq_thread_unpark(sqd);
7828
Jens Axboeaa061652020-09-02 14:50:27 -06007829 if (sqd->thread)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007830 return 0;
Jens Axboeaa061652020-09-02 14:50:27 -06007831
Jens Axboe6c271ce2019-01-10 11:22:30 -07007832 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06007833 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007834
Jens Axboe917257d2019-04-13 09:28:55 -06007835 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06007836 if (cpu >= nr_cpu_ids)
Jens Axboee8f98f242021-03-09 16:32:13 -07007837 goto err_sqpoll;
Shenghui Wang7889f442019-05-07 16:03:19 +08007838 if (!cpu_online(cpu))
Jens Axboee8f98f242021-03-09 16:32:13 -07007839 goto err_sqpoll;
Jens Axboe917257d2019-04-13 09:28:55 -06007840
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007841 sqd->sq_cpu = cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007842 } else {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007843 sqd->sq_cpu = -1;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007844 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007845
7846 sqd->task_pid = current->pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007847 sqd->task_tgid = current->tgid;
Jens Axboe46fe18b2021-03-04 12:39:36 -07007848 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
7849 if (IS_ERR(tsk)) {
7850 ret = PTR_ERR(tsk);
Jens Axboee8f98f242021-03-09 16:32:13 -07007851 goto err_sqpoll;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007852 }
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007853
Jens Axboe46fe18b2021-03-04 12:39:36 -07007854 sqd->thread = tsk;
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007855 ret = io_uring_alloc_task_context(tsk, ctx);
Jens Axboe46fe18b2021-03-04 12:39:36 -07007856 wake_up_new_task(tsk);
Jens Axboe0f212202020-09-13 13:09:39 -06007857 if (ret)
7858 goto err;
Pavel Begunkov0298ef92021-03-08 13:20:57 +00007859 complete(&sqd->startup);
Jens Axboe6c271ce2019-01-10 11:22:30 -07007860 } else if (p->flags & IORING_SETUP_SQ_AFF) {
7861 /* Can't have SQ_AFF without SQPOLL */
7862 ret = -EINVAL;
7863 goto err;
7864 }
7865
Jens Axboe2b188cc2019-01-07 10:46:33 -07007866 return 0;
7867err:
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007868 io_sq_thread_finish(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007869 return ret;
Jens Axboee8f98f242021-03-09 16:32:13 -07007870err_sqpoll:
7871 complete(&ctx->sq_data->exited);
7872 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007873}
7874
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007875static inline void __io_unaccount_mem(struct user_struct *user,
7876 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007877{
7878 atomic_long_sub(nr_pages, &user->locked_vm);
7879}
7880
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007881static inline int __io_account_mem(struct user_struct *user,
7882 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007883{
7884 unsigned long page_limit, cur_pages, new_pages;
7885
7886 /* Don't allow more pages than we can safely lock */
7887 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
7888
7889 do {
7890 cur_pages = atomic_long_read(&user->locked_vm);
7891 new_pages = cur_pages + nr_pages;
7892 if (new_pages > page_limit)
7893 return -ENOMEM;
7894 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
7895 new_pages) != cur_pages);
7896
7897 return 0;
7898}
7899
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007900static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007901{
Jens Axboe62e398b2021-02-21 16:19:37 -07007902 if (ctx->user)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007903 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007904
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007905 if (ctx->mm_account)
7906 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007907}
7908
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007909static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007910{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007911 int ret;
7912
Jens Axboe62e398b2021-02-21 16:19:37 -07007913 if (ctx->user) {
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007914 ret = __io_account_mem(ctx->user, nr_pages);
7915 if (ret)
7916 return ret;
7917 }
7918
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007919 if (ctx->mm_account)
7920 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007921
7922 return 0;
7923}
7924
Jens Axboe2b188cc2019-01-07 10:46:33 -07007925static void io_mem_free(void *ptr)
7926{
Mark Rutland52e04ef2019-04-30 17:30:21 +01007927 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007928
Mark Rutland52e04ef2019-04-30 17:30:21 +01007929 if (!ptr)
7930 return;
7931
7932 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007933 if (put_page_testzero(page))
7934 free_compound_page(page);
7935}
7936
7937static void *io_mem_alloc(size_t size)
7938{
7939 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007940 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007941
7942 return (void *) __get_free_pages(gfp_flags, get_order(size));
7943}
7944
Hristo Venev75b28af2019-08-26 17:23:46 +00007945static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
7946 size_t *sq_offset)
7947{
7948 struct io_rings *rings;
7949 size_t off, sq_array_size;
7950
7951 off = struct_size(rings, cqes, cq_entries);
7952 if (off == SIZE_MAX)
7953 return SIZE_MAX;
7954
7955#ifdef CONFIG_SMP
7956 off = ALIGN(off, SMP_CACHE_BYTES);
7957 if (off == 0)
7958 return SIZE_MAX;
7959#endif
7960
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02007961 if (sq_offset)
7962 *sq_offset = off;
7963
Hristo Venev75b28af2019-08-26 17:23:46 +00007964 sq_array_size = array_size(sizeof(u32), sq_entries);
7965 if (sq_array_size == SIZE_MAX)
7966 return SIZE_MAX;
7967
7968 if (check_add_overflow(off, sq_array_size, &off))
7969 return SIZE_MAX;
7970
Hristo Venev75b28af2019-08-26 17:23:46 +00007971 return off;
7972}
7973
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08007974static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07007975{
7976 int i, j;
7977
7978 if (!ctx->user_bufs)
7979 return -ENXIO;
7980
7981 for (i = 0; i < ctx->nr_user_bufs; i++) {
7982 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
7983
7984 for (j = 0; j < imu->nr_bvecs; j++)
John Hubbardf1f6a7d2020-01-30 22:13:35 -08007985 unpin_user_page(imu->bvec[j].bv_page);
Jens Axboeedafcce2019-01-09 09:16:05 -07007986
Jens Axboede293932020-09-17 16:19:16 -06007987 if (imu->acct_pages)
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007988 io_unaccount_mem(ctx, imu->acct_pages);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01007989 kvfree(imu->bvec);
Jens Axboeedafcce2019-01-09 09:16:05 -07007990 imu->nr_bvecs = 0;
7991 }
7992
7993 kfree(ctx->user_bufs);
7994 ctx->user_bufs = NULL;
7995 ctx->nr_user_bufs = 0;
7996 return 0;
7997}
7998
7999static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8000 void __user *arg, unsigned index)
8001{
8002 struct iovec __user *src;
8003
8004#ifdef CONFIG_COMPAT
8005 if (ctx->compat) {
8006 struct compat_iovec __user *ciovs;
8007 struct compat_iovec ciov;
8008
8009 ciovs = (struct compat_iovec __user *) arg;
8010 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8011 return -EFAULT;
8012
Jens Axboed55e5f52019-12-11 16:12:15 -07008013 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008014 dst->iov_len = ciov.iov_len;
8015 return 0;
8016 }
8017#endif
8018 src = (struct iovec __user *) arg;
8019 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8020 return -EFAULT;
8021 return 0;
8022}
8023
Jens Axboede293932020-09-17 16:19:16 -06008024/*
8025 * Not super efficient, but this is just a registration time. And we do cache
8026 * the last compound head, so generally we'll only do a full search if we don't
8027 * match that one.
8028 *
8029 * We check if the given compound head page has already been accounted, to
8030 * avoid double accounting it. This allows us to account the full size of the
8031 * page, not just the constituent pages of a huge page.
8032 */
8033static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8034 int nr_pages, struct page *hpage)
8035{
8036 int i, j;
8037
8038 /* check current page array */
8039 for (i = 0; i < nr_pages; i++) {
8040 if (!PageCompound(pages[i]))
8041 continue;
8042 if (compound_head(pages[i]) == hpage)
8043 return true;
8044 }
8045
8046 /* check previously registered pages */
8047 for (i = 0; i < ctx->nr_user_bufs; i++) {
8048 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8049
8050 for (j = 0; j < imu->nr_bvecs; j++) {
8051 if (!PageCompound(imu->bvec[j].bv_page))
8052 continue;
8053 if (compound_head(imu->bvec[j].bv_page) == hpage)
8054 return true;
8055 }
8056 }
8057
8058 return false;
8059}
8060
8061static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8062 int nr_pages, struct io_mapped_ubuf *imu,
8063 struct page **last_hpage)
8064{
8065 int i, ret;
8066
8067 for (i = 0; i < nr_pages; i++) {
8068 if (!PageCompound(pages[i])) {
8069 imu->acct_pages++;
8070 } else {
8071 struct page *hpage;
8072
8073 hpage = compound_head(pages[i]);
8074 if (hpage == *last_hpage)
8075 continue;
8076 *last_hpage = hpage;
8077 if (headpage_already_acct(ctx, pages, i, hpage))
8078 continue;
8079 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8080 }
8081 }
8082
8083 if (!imu->acct_pages)
8084 return 0;
8085
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008086 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008087 if (ret)
8088 imu->acct_pages = 0;
8089 return ret;
8090}
8091
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008092static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
8093 struct io_mapped_ubuf *imu,
8094 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008095{
8096 struct vm_area_struct **vmas = NULL;
8097 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008098 unsigned long off, start, end, ubuf;
8099 size_t size;
8100 int ret, pret, nr_pages, i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008101
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008102 ubuf = (unsigned long) iov->iov_base;
8103 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8104 start = ubuf >> PAGE_SHIFT;
8105 nr_pages = end - start;
8106
8107 ret = -ENOMEM;
8108
8109 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8110 if (!pages)
8111 goto done;
8112
8113 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8114 GFP_KERNEL);
8115 if (!vmas)
8116 goto done;
8117
8118 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
8119 GFP_KERNEL);
8120 if (!imu->bvec)
8121 goto done;
8122
8123 ret = 0;
8124 mmap_read_lock(current->mm);
8125 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8126 pages, vmas);
8127 if (pret == nr_pages) {
8128 /* don't support file backed memory */
8129 for (i = 0; i < nr_pages; i++) {
8130 struct vm_area_struct *vma = vmas[i];
8131
8132 if (vma->vm_file &&
8133 !is_file_hugepages(vma->vm_file)) {
8134 ret = -EOPNOTSUPP;
8135 break;
8136 }
8137 }
8138 } else {
8139 ret = pret < 0 ? pret : -EFAULT;
8140 }
8141 mmap_read_unlock(current->mm);
8142 if (ret) {
8143 /*
8144 * if we did partial map, or found file backed vmas,
8145 * release any pages we did get
8146 */
8147 if (pret > 0)
8148 unpin_user_pages(pages, pret);
8149 kvfree(imu->bvec);
8150 goto done;
8151 }
8152
8153 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8154 if (ret) {
8155 unpin_user_pages(pages, pret);
8156 kvfree(imu->bvec);
8157 goto done;
8158 }
8159
8160 off = ubuf & ~PAGE_MASK;
8161 size = iov->iov_len;
8162 for (i = 0; i < nr_pages; i++) {
8163 size_t vec_len;
8164
8165 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8166 imu->bvec[i].bv_page = pages[i];
8167 imu->bvec[i].bv_len = vec_len;
8168 imu->bvec[i].bv_offset = off;
8169 off = 0;
8170 size -= vec_len;
8171 }
8172 /* store original address for later verification */
8173 imu->ubuf = ubuf;
8174 imu->len = iov->iov_len;
8175 imu->nr_bvecs = nr_pages;
8176 ret = 0;
8177done:
8178 kvfree(pages);
8179 kvfree(vmas);
8180 return ret;
8181}
8182
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008183static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008184{
Jens Axboeedafcce2019-01-09 09:16:05 -07008185 if (ctx->user_bufs)
8186 return -EBUSY;
8187 if (!nr_args || nr_args > UIO_MAXIOV)
8188 return -EINVAL;
8189
8190 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
8191 GFP_KERNEL);
8192 if (!ctx->user_bufs)
8193 return -ENOMEM;
8194
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008195 return 0;
8196}
8197
8198static int io_buffer_validate(struct iovec *iov)
8199{
8200 /*
8201 * Don't impose further limits on the size and buffer
8202 * constraints here, we'll -EINVAL later when IO is
8203 * submitted if they are wrong.
8204 */
8205 if (!iov->iov_base || !iov->iov_len)
8206 return -EFAULT;
8207
8208 /* arbitrary limit, but we need something */
8209 if (iov->iov_len > SZ_1G)
8210 return -EFAULT;
8211
8212 return 0;
8213}
8214
8215static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
8216 unsigned int nr_args)
8217{
8218 int i, ret;
8219 struct iovec iov;
8220 struct page *last_hpage = NULL;
8221
8222 ret = io_buffers_map_alloc(ctx, nr_args);
8223 if (ret)
8224 return ret;
8225
Jens Axboeedafcce2019-01-09 09:16:05 -07008226 for (i = 0; i < nr_args; i++) {
8227 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
Jens Axboeedafcce2019-01-09 09:16:05 -07008228
8229 ret = io_copy_iov(ctx, &iov, arg, i);
8230 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008231 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008232
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008233 ret = io_buffer_validate(&iov);
8234 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008235 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008236
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008237 ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage);
8238 if (ret)
8239 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008240
8241 ctx->nr_user_bufs++;
8242 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008243
8244 if (ret)
8245 io_sqe_buffers_unregister(ctx);
8246
Jens Axboeedafcce2019-01-09 09:16:05 -07008247 return ret;
8248}
8249
Jens Axboe9b402842019-04-11 11:45:41 -06008250static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8251{
8252 __s32 __user *fds = arg;
8253 int fd;
8254
8255 if (ctx->cq_ev_fd)
8256 return -EBUSY;
8257
8258 if (copy_from_user(&fd, fds, sizeof(*fds)))
8259 return -EFAULT;
8260
8261 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8262 if (IS_ERR(ctx->cq_ev_fd)) {
8263 int ret = PTR_ERR(ctx->cq_ev_fd);
8264 ctx->cq_ev_fd = NULL;
8265 return ret;
8266 }
8267
8268 return 0;
8269}
8270
8271static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8272{
8273 if (ctx->cq_ev_fd) {
8274 eventfd_ctx_put(ctx->cq_ev_fd);
8275 ctx->cq_ev_fd = NULL;
8276 return 0;
8277 }
8278
8279 return -ENXIO;
8280}
8281
Jens Axboe5a2e7452020-02-23 16:23:11 -07008282static int __io_destroy_buffers(int id, void *p, void *data)
8283{
8284 struct io_ring_ctx *ctx = data;
8285 struct io_buffer *buf = p;
8286
Jens Axboe067524e2020-03-02 16:32:28 -07008287 __io_remove_buffers(ctx, buf, id, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008288 return 0;
8289}
8290
8291static void io_destroy_buffers(struct io_ring_ctx *ctx)
8292{
8293 idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx);
8294 idr_destroy(&ctx->io_buffer_idr);
8295}
8296
Jens Axboe68e68ee2021-02-13 09:00:02 -07008297static void io_req_cache_free(struct list_head *list, struct task_struct *tsk)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008298{
Jens Axboe68e68ee2021-02-13 09:00:02 -07008299 struct io_kiocb *req, *nxt;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008300
Jens Axboe68e68ee2021-02-13 09:00:02 -07008301 list_for_each_entry_safe(req, nxt, list, compl.list) {
8302 if (tsk && req->task != tsk)
8303 continue;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008304 list_del(&req->compl.list);
8305 kmem_cache_free(req_cachep, req);
8306 }
8307}
8308
Jens Axboe4010fec2021-02-27 15:04:18 -07008309static void io_req_caches_free(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008310{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008311 struct io_submit_state *submit_state = &ctx->submit_state;
Pavel Begunkove5547d22021-02-23 22:17:20 +00008312 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008313
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008314 mutex_lock(&ctx->uring_lock);
8315
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008316 if (submit_state->free_reqs) {
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008317 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8318 submit_state->reqs);
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008319 submit_state->free_reqs = 0;
8320 }
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008321
8322 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkove5547d22021-02-23 22:17:20 +00008323 list_splice_init(&cs->locked_free_list, &cs->free_list);
8324 cs->locked_free_nr = 0;
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008325 spin_unlock_irq(&ctx->completion_lock);
8326
Pavel Begunkove5547d22021-02-23 22:17:20 +00008327 io_req_cache_free(&cs->free_list, NULL);
8328
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008329 mutex_unlock(&ctx->uring_lock);
8330}
8331
Jens Axboe2b188cc2019-01-07 10:46:33 -07008332static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8333{
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00008334 /*
8335 * Some may use context even when all refs and requests have been put,
8336 * and they are free to do so while still holding uring_lock, see
8337 * __io_req_task_submit(). Wait for them to finish.
8338 */
8339 mutex_lock(&ctx->uring_lock);
8340 mutex_unlock(&ctx->uring_lock);
8341
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008342 io_sq_thread_finish(ctx);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008343 io_sqe_buffers_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008344
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008345 if (ctx->mm_account) {
Jens Axboe2aede0e2020-09-14 10:45:53 -06008346 mmdrop(ctx->mm_account);
8347 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008348 }
Jens Axboedef596e2019-01-09 08:59:42 -07008349
Hao Xu8bad28d2021-02-19 17:19:36 +08008350 mutex_lock(&ctx->uring_lock);
Jens Axboe6b063142019-01-10 22:13:58 -07008351 io_sqe_files_unregister(ctx);
Hao Xu8bad28d2021-02-19 17:19:36 +08008352 mutex_unlock(&ctx->uring_lock);
Jens Axboe9b402842019-04-11 11:45:41 -06008353 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008354 io_destroy_buffers(ctx);
Jens Axboedef596e2019-01-09 08:59:42 -07008355
Jens Axboe2b188cc2019-01-07 10:46:33 -07008356#if defined(CONFIG_UNIX)
Eric Biggers355e8d22019-06-12 14:58:43 -07008357 if (ctx->ring_sock) {
8358 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008359 sock_release(ctx->ring_sock);
Eric Biggers355e8d22019-06-12 14:58:43 -07008360 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008361#endif
8362
Hristo Venev75b28af2019-08-26 17:23:46 +00008363 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008364 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008365
8366 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008367 free_uid(ctx->user);
Jens Axboe4010fec2021-02-27 15:04:18 -07008368 io_req_caches_free(ctx);
Jens Axboee9418942021-02-19 12:33:30 -07008369 if (ctx->hash_map)
8370 io_wq_put_hash(ctx->hash_map);
Jens Axboe78076bb2019-12-04 19:56:40 -07008371 kfree(ctx->cancel_hash);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008372 kfree(ctx);
8373}
8374
8375static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8376{
8377 struct io_ring_ctx *ctx = file->private_data;
8378 __poll_t mask = 0;
8379
8380 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008381 /*
8382 * synchronizes with barrier from wq_has_sleeper call in
8383 * io_commit_cqring
8384 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008385 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008386 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008387 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008388
8389 /*
8390 * Don't flush cqring overflow list here, just do a simple check.
8391 * Otherwise there could possible be ABBA deadlock:
8392 * CPU0 CPU1
8393 * ---- ----
8394 * lock(&ctx->uring_lock);
8395 * lock(&ep->mtx);
8396 * lock(&ctx->uring_lock);
8397 * lock(&ep->mtx);
8398 *
8399 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8400 * pushs them to do the flush.
8401 */
8402 if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008403 mask |= EPOLLIN | EPOLLRDNORM;
8404
8405 return mask;
8406}
8407
8408static int io_uring_fasync(int fd, struct file *file, int on)
8409{
8410 struct io_ring_ctx *ctx = file->private_data;
8411
8412 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8413}
8414
Yejune Deng0bead8c2020-12-24 11:02:20 +08008415static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008416{
Jens Axboe4379bf82021-02-15 13:40:22 -07008417 const struct cred *creds;
Jens Axboe071698e2020-01-28 10:04:42 -07008418
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008419 creds = xa_erase(&ctx->personalities, id);
Jens Axboe4379bf82021-02-15 13:40:22 -07008420 if (creds) {
8421 put_cred(creds);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008422 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008423 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008424
8425 return -EINVAL;
8426}
8427
Pavel Begunkovba50a032021-02-26 15:47:56 +00008428static bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008429{
Pavel Begunkov28c47212021-02-28 22:04:54 +00008430 struct callback_head *work, *next;
Pavel Begunkovba50a032021-02-26 15:47:56 +00008431 bool executed = false;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008432
8433 do {
Pavel Begunkov28c47212021-02-28 22:04:54 +00008434 work = xchg(&ctx->exit_task_work, NULL);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008435 if (!work)
8436 break;
8437
8438 do {
8439 next = work->next;
8440 work->func(work);
8441 work = next;
8442 cond_resched();
8443 } while (work);
Pavel Begunkovba50a032021-02-26 15:47:56 +00008444 executed = true;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008445 } while (1);
Pavel Begunkovba50a032021-02-26 15:47:56 +00008446
8447 return executed;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008448}
8449
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008450struct io_tctx_exit {
8451 struct callback_head task_work;
8452 struct completion completion;
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008453 struct io_ring_ctx *ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008454};
8455
8456static void io_tctx_exit_cb(struct callback_head *cb)
8457{
8458 struct io_uring_task *tctx = current->io_uring;
8459 struct io_tctx_exit *work;
8460
8461 work = container_of(cb, struct io_tctx_exit, task_work);
8462 /*
8463 * When @in_idle, we're in cancellation and it's racy to remove the
8464 * node. It'll be removed by the end of cancellation, just ignore it.
8465 */
8466 if (!atomic_read(&tctx->in_idle))
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008467 io_uring_del_task_file((unsigned long)work->ctx);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008468 complete(&work->completion);
8469}
8470
Jens Axboe85faa7b2020-04-09 18:14:00 -06008471static void io_ring_exit_work(struct work_struct *work)
8472{
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008473 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008474 unsigned long timeout = jiffies + HZ * 60 * 5;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008475 struct io_tctx_exit exit;
8476 struct io_tctx_node *node;
8477 int ret;
Jens Axboe85faa7b2020-04-09 18:14:00 -06008478
Jens Axboe56952e92020-06-17 15:00:04 -06008479 /*
8480 * If we're doing polled IO and end up having requests being
8481 * submitted async (out-of-line), then completions can come in while
8482 * we're waiting for refs to drop. We need to reap these manually,
8483 * as nobody else will be looking for them.
8484 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008485 do {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008486 io_uring_try_cancel_requests(ctx, NULL, NULL);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008487
8488 WARN_ON_ONCE(time_after(jiffies, timeout));
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008489 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008490
8491 mutex_lock(&ctx->uring_lock);
8492 while (!list_empty(&ctx->tctx_list)) {
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008493 WARN_ON_ONCE(time_after(jiffies, timeout));
8494
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008495 node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
8496 ctx_node);
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008497 exit.ctx = ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008498 init_completion(&exit.completion);
8499 init_task_work(&exit.task_work, io_tctx_exit_cb);
8500 ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
8501 if (WARN_ON_ONCE(ret))
8502 continue;
8503 wake_up_process(node->task);
8504
8505 mutex_unlock(&ctx->uring_lock);
8506 wait_for_completion(&exit.completion);
8507 cond_resched();
8508 mutex_lock(&ctx->uring_lock);
8509 }
8510 mutex_unlock(&ctx->uring_lock);
8511
Jens Axboe85faa7b2020-04-09 18:14:00 -06008512 io_ring_ctx_free(ctx);
8513}
8514
Jens Axboe2b188cc2019-01-07 10:46:33 -07008515static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8516{
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008517 unsigned long index;
8518 struct creds *creds;
8519
Jens Axboe2b188cc2019-01-07 10:46:33 -07008520 mutex_lock(&ctx->uring_lock);
8521 percpu_ref_kill(&ctx->refs);
Pavel Begunkovcda286f2020-12-17 00:24:35 +00008522 /* if force is set, the ring is going away. always drop after that */
8523 ctx->cq_overflow_flushed = 1;
Pavel Begunkov634578f2020-12-06 22:22:44 +00008524 if (ctx->rings)
Pavel Begunkov6c503152021-01-04 20:36:36 +00008525 __io_cqring_overflow_flush(ctx, true, NULL, NULL);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008526 xa_for_each(&ctx->personalities, index, creds)
8527 io_unregister_personality(ctx, index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008528 mutex_unlock(&ctx->uring_lock);
8529
Pavel Begunkov6b819282020-11-06 13:00:25 +00008530 io_kill_timeouts(ctx, NULL, NULL);
8531 io_poll_remove_all(ctx, NULL, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008532
Jens Axboe15dff282019-11-13 09:09:23 -07008533 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008534 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008535
Jens Axboe85faa7b2020-04-09 18:14:00 -06008536 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008537 /*
8538 * Use system_unbound_wq to avoid spawning tons of event kworkers
8539 * if we're exiting a ton of rings at the same time. It just adds
8540 * noise and overhead, there's no discernable change in runtime
8541 * over using system_wq.
8542 */
8543 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008544}
8545
8546static int io_uring_release(struct inode *inode, struct file *file)
8547{
8548 struct io_ring_ctx *ctx = file->private_data;
8549
8550 file->private_data = NULL;
8551 io_ring_ctx_wait_and_kill(ctx);
8552 return 0;
8553}
8554
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008555struct io_task_cancel {
8556 struct task_struct *task;
8557 struct files_struct *files;
8558};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008559
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008560static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008561{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008562 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008563 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008564 bool ret;
8565
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008566 if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008567 unsigned long flags;
8568 struct io_ring_ctx *ctx = req->ctx;
8569
8570 /* protect against races with linked timeouts */
8571 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008572 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008573 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8574 } else {
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008575 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008576 }
8577 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008578}
8579
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008580static void io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008581 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008582 struct files_struct *files)
8583{
8584 struct io_defer_entry *de = NULL;
8585 LIST_HEAD(list);
8586
8587 spin_lock_irq(&ctx->completion_lock);
8588 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00008589 if (io_match_task(de->req, task, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008590 list_cut_position(&list, &ctx->defer_list, &de->list);
8591 break;
8592 }
8593 }
8594 spin_unlock_irq(&ctx->completion_lock);
8595
8596 while (!list_empty(&list)) {
8597 de = list_first_entry(&list, struct io_defer_entry, list);
8598 list_del_init(&de->list);
8599 req_set_fail_links(de->req);
8600 io_put_req(de->req);
8601 io_req_complete(de->req, -ECANCELED);
8602 kfree(de);
8603 }
8604}
8605
Pavel Begunkov1b007642021-03-06 11:02:17 +00008606static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data)
8607{
8608 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8609
8610 return req->ctx == data;
8611}
8612
8613static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
8614{
8615 struct io_tctx_node *node;
8616 enum io_wq_cancel cret;
8617 bool ret = false;
8618
8619 mutex_lock(&ctx->uring_lock);
8620 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
8621 struct io_uring_task *tctx = node->task->io_uring;
8622
8623 /*
8624 * io_wq will stay alive while we hold uring_lock, because it's
8625 * killed after ctx nodes, which requires to take the lock.
8626 */
8627 if (!tctx || !tctx->io_wq)
8628 continue;
8629 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true);
8630 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8631 }
8632 mutex_unlock(&ctx->uring_lock);
8633
8634 return ret;
8635}
8636
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008637static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
8638 struct task_struct *task,
8639 struct files_struct *files)
8640{
8641 struct io_task_cancel cancel = { .task = task, .files = files, };
Pavel Begunkov1b007642021-03-06 11:02:17 +00008642 struct io_uring_task *tctx = task ? task->io_uring : NULL;
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008643
8644 while (1) {
8645 enum io_wq_cancel cret;
8646 bool ret = false;
8647
Pavel Begunkov1b007642021-03-06 11:02:17 +00008648 if (!task) {
8649 ret |= io_uring_try_cancel_iowq(ctx);
8650 } else if (tctx && tctx->io_wq) {
8651 /*
8652 * Cancels requests of all rings, not only @ctx, but
8653 * it's fine as the task is in exit/exec.
8654 */
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008655 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008656 &cancel, true);
8657 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8658 }
8659
8660 /* SQPOLL thread does its own polling */
8661 if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) {
8662 while (!list_empty_careful(&ctx->iopoll_list)) {
8663 io_iopoll_try_reap_events(ctx);
8664 ret = true;
8665 }
8666 }
8667
8668 ret |= io_poll_remove_all(ctx, task, files);
8669 ret |= io_kill_timeouts(ctx, task, files);
8670 ret |= io_run_task_work();
Pavel Begunkovba50a032021-02-26 15:47:56 +00008671 ret |= io_run_ctx_fallback(ctx);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008672 io_cqring_overflow_flush(ctx, true, task, files);
8673 if (!ret)
8674 break;
8675 cond_resched();
8676 }
8677}
8678
Pavel Begunkovca70f002021-01-26 15:28:27 +00008679static int io_uring_count_inflight(struct io_ring_ctx *ctx,
8680 struct task_struct *task,
8681 struct files_struct *files)
8682{
8683 struct io_kiocb *req;
8684 int cnt = 0;
8685
8686 spin_lock_irq(&ctx->inflight_lock);
8687 list_for_each_entry(req, &ctx->inflight_list, inflight_entry)
8688 cnt += io_match_task(req, task, files);
8689 spin_unlock_irq(&ctx->inflight_lock);
8690 return cnt;
8691}
8692
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008693static void io_uring_cancel_files(struct io_ring_ctx *ctx,
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008694 struct task_struct *task,
Jens Axboefcb323c2019-10-24 12:39:47 -06008695 struct files_struct *files)
8696{
Jens Axboefcb323c2019-10-24 12:39:47 -06008697 while (!list_empty_careful(&ctx->inflight_list)) {
Xiaoguang Wangd8f1b972020-04-26 15:54:43 +08008698 DEFINE_WAIT(wait);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008699 int inflight;
Jens Axboefcb323c2019-10-24 12:39:47 -06008700
Pavel Begunkovca70f002021-01-26 15:28:27 +00008701 inflight = io_uring_count_inflight(ctx, task, files);
8702 if (!inflight)
Jens Axboefcb323c2019-10-24 12:39:47 -06008703 break;
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008704
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008705 io_uring_try_cancel_requests(ctx, task, files);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008706
Pavel Begunkov34343782021-02-10 11:45:42 +00008707 if (ctx->sq_data)
8708 io_sq_thread_unpark(ctx->sq_data);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008709 prepare_to_wait(&task->io_uring->wait, &wait,
8710 TASK_UNINTERRUPTIBLE);
8711 if (inflight == io_uring_count_inflight(ctx, task, files))
8712 schedule();
Pavel Begunkovc98de082020-11-15 12:56:32 +00008713 finish_wait(&task->io_uring->wait, &wait);
Pavel Begunkov34343782021-02-10 11:45:42 +00008714 if (ctx->sq_data)
8715 io_sq_thread_park(ctx->sq_data);
Jens Axboe0f212202020-09-13 13:09:39 -06008716 }
Jens Axboe0f212202020-09-13 13:09:39 -06008717}
8718
8719/*
8720 * We need to iteratively cancel requests, in case a request has dependent
8721 * hard links. These persist even for failure of cancelations, hence keep
8722 * looping until none are found.
8723 */
8724static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
8725 struct files_struct *files)
8726{
8727 struct task_struct *task = current;
8728
Jens Axboefdaf0832020-10-30 09:37:30 -06008729 if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
Jens Axboe86e0d672021-03-05 08:44:39 -07008730 io_sq_thread_park(ctx->sq_data);
8731 task = ctx->sq_data->thread;
8732 if (task)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008733 atomic_inc(&task->io_uring->in_idle);
Jens Axboefdaf0832020-10-30 09:37:30 -06008734 }
Jens Axboe0f212202020-09-13 13:09:39 -06008735
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008736 io_cancel_defer_files(ctx, task, files);
Jens Axboe0f212202020-09-13 13:09:39 -06008737
Pavel Begunkov3a7efd12021-01-28 23:23:42 +00008738 io_uring_cancel_files(ctx, task, files);
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008739 if (!files)
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008740 io_uring_try_cancel_requests(ctx, task, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008741
Jens Axboe86e0d672021-03-05 08:44:39 -07008742 if (task)
Jens Axboefdaf0832020-10-30 09:37:30 -06008743 atomic_dec(&task->io_uring->in_idle);
Jens Axboe86e0d672021-03-05 08:44:39 -07008744 if (ctx->sq_data)
Jens Axboefdaf0832020-10-30 09:37:30 -06008745 io_sq_thread_unpark(ctx->sq_data);
Jens Axboe0f212202020-09-13 13:09:39 -06008746}
8747
8748/*
8749 * Note that this task has used io_uring. We use it for cancelation purposes.
8750 */
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008751static int io_uring_add_task_file(struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008752{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008753 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008754 struct io_tctx_node *node;
Pavel Begunkova528b042020-12-21 18:34:04 +00008755 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008756
8757 if (unlikely(!tctx)) {
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008758 ret = io_uring_alloc_task_context(current, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06008759 if (unlikely(ret))
8760 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008761 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008762 }
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008763 if (tctx->last != ctx) {
8764 void *old = xa_load(&tctx->xa, (unsigned long)ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06008765
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008766 if (!old) {
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008767 node = kmalloc(sizeof(*node), GFP_KERNEL);
8768 if (!node)
8769 return -ENOMEM;
8770 node->ctx = ctx;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008771 node->task = current;
8772
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008773 ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx,
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008774 node, GFP_KERNEL));
Pavel Begunkova528b042020-12-21 18:34:04 +00008775 if (ret) {
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008776 kfree(node);
Pavel Begunkova528b042020-12-21 18:34:04 +00008777 return ret;
8778 }
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008779
8780 mutex_lock(&ctx->uring_lock);
8781 list_add(&node->ctx_node, &ctx->tctx_list);
8782 mutex_unlock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06008783 }
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008784 tctx->last = ctx;
Jens Axboe0f212202020-09-13 13:09:39 -06008785 }
8786
Jens Axboefdaf0832020-10-30 09:37:30 -06008787 /*
8788 * This is race safe in that the task itself is doing this, hence it
8789 * cannot be going through the exit/cancel paths at the same time.
8790 * This cannot be modified while exit/cancel is running.
8791 */
8792 if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL))
8793 tctx->sqpoll = true;
8794
Jens Axboe0f212202020-09-13 13:09:39 -06008795 return 0;
8796}
8797
8798/*
8799 * Remove this io_uring_file -> task mapping.
8800 */
Pavel Begunkov29412672021-03-06 11:02:11 +00008801static void io_uring_del_task_file(unsigned long index)
Jens Axboe0f212202020-09-13 13:09:39 -06008802{
8803 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008804 struct io_tctx_node *node;
Pavel Begunkov29412672021-03-06 11:02:11 +00008805
Pavel Begunkoveebd2e32021-03-06 11:02:14 +00008806 if (!tctx)
8807 return;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008808 node = xa_erase(&tctx->xa, index);
8809 if (!node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008810 return;
Jens Axboe0f212202020-09-13 13:09:39 -06008811
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008812 WARN_ON_ONCE(current != node->task);
8813 WARN_ON_ONCE(list_empty(&node->ctx_node));
8814
8815 mutex_lock(&node->ctx->uring_lock);
8816 list_del(&node->ctx_node);
8817 mutex_unlock(&node->ctx->uring_lock);
8818
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008819 if (tctx->last == node->ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008820 tctx->last = NULL;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008821 kfree(node);
Jens Axboe0f212202020-09-13 13:09:39 -06008822}
8823
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008824static void io_uring_clean_tctx(struct io_uring_task *tctx)
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008825{
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008826 struct io_tctx_node *node;
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008827 unsigned long index;
8828
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008829 xa_for_each(&tctx->xa, index, node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008830 io_uring_del_task_file(index);
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008831 if (tctx->io_wq) {
8832 io_wq_put_and_exit(tctx->io_wq);
8833 tctx->io_wq = NULL;
8834 }
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008835}
8836
Jens Axboe0f212202020-09-13 13:09:39 -06008837void __io_uring_files_cancel(struct files_struct *files)
8838{
8839 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008840 struct io_tctx_node *node;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008841 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06008842
8843 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008844 atomic_inc(&tctx->in_idle);
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008845 xa_for_each(&tctx->xa, index, node)
8846 io_uring_cancel_task_requests(node->ctx, files);
Jens Axboefdaf0832020-10-30 09:37:30 -06008847 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008848
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008849 if (files)
8850 io_uring_clean_tctx(tctx);
Jens Axboefdaf0832020-10-30 09:37:30 -06008851}
8852
8853static s64 tctx_inflight(struct io_uring_task *tctx)
8854{
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008855 return percpu_counter_sum(&tctx->inflight);
8856}
8857
8858static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx)
8859{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008860 struct io_sq_data *sqd = ctx->sq_data;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008861 struct io_uring_task *tctx;
Jens Axboefdaf0832020-10-30 09:37:30 -06008862 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008863 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06008864
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008865 if (!sqd)
8866 return;
Jens Axboe86e0d672021-03-05 08:44:39 -07008867 io_sq_thread_park(sqd);
8868 if (!sqd->thread || !sqd->thread->io_uring) {
Jens Axboee54945a2021-02-26 11:27:15 -07008869 io_sq_thread_unpark(sqd);
8870 return;
8871 }
Jens Axboe86e0d672021-03-05 08:44:39 -07008872 tctx = ctx->sq_data->thread->io_uring;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008873 atomic_inc(&tctx->in_idle);
8874 do {
8875 /* read completions before cancelations */
8876 inflight = tctx_inflight(tctx);
8877 if (!inflight)
8878 break;
8879 io_uring_cancel_task_requests(ctx, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008880
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008881 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8882 /*
8883 * If we've seen completions, retry without waiting. This
8884 * avoids a race where a completion comes in before we did
8885 * prepare_to_wait().
8886 */
8887 if (inflight == tctx_inflight(tctx))
8888 schedule();
8889 finish_wait(&tctx->wait, &wait);
8890 } while (1);
8891 atomic_dec(&tctx->in_idle);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008892 io_sq_thread_unpark(sqd);
Jens Axboe0f212202020-09-13 13:09:39 -06008893}
8894
Jens Axboe0f212202020-09-13 13:09:39 -06008895/*
8896 * Find any io_uring fd that this task has registered or done IO on, and cancel
8897 * requests.
8898 */
8899void __io_uring_task_cancel(void)
8900{
8901 struct io_uring_task *tctx = current->io_uring;
8902 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008903 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06008904
8905 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008906 atomic_inc(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06008907
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008908 if (tctx->sqpoll) {
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008909 struct io_tctx_node *node;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008910 unsigned long index;
8911
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008912 xa_for_each(&tctx->xa, index, node)
8913 io_uring_cancel_sqpoll(node->ctx);
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008914 }
Pavel Begunkov0b5cd6c2021-01-17 02:29:56 +00008915
Jens Axboed8a6df12020-10-15 16:24:45 -06008916 do {
Jens Axboe0f212202020-09-13 13:09:39 -06008917 /* read completions before cancelations */
Jens Axboefdaf0832020-10-30 09:37:30 -06008918 inflight = tctx_inflight(tctx);
Jens Axboed8a6df12020-10-15 16:24:45 -06008919 if (!inflight)
8920 break;
Jens Axboe0f212202020-09-13 13:09:39 -06008921 __io_uring_files_cancel(NULL);
8922
8923 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8924
8925 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008926 * If we've seen completions, retry without waiting. This
8927 * avoids a race where a completion comes in before we did
8928 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06008929 */
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008930 if (inflight == tctx_inflight(tctx))
8931 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00008932 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008933 } while (1);
Jens Axboe0f212202020-09-13 13:09:39 -06008934
Jens Axboefdaf0832020-10-30 09:37:30 -06008935 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008936
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008937 io_uring_clean_tctx(tctx);
8938 /* all current's requests should be gone, we can kill tctx */
8939 __io_uring_free(current);
Pavel Begunkov44e728b2020-06-15 10:24:04 +03008940}
8941
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008942static void *io_uring_validate_mmap_request(struct file *file,
8943 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008944{
Jens Axboe2b188cc2019-01-07 10:46:33 -07008945 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008946 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008947 struct page *page;
8948 void *ptr;
8949
8950 switch (offset) {
8951 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00008952 case IORING_OFF_CQ_RING:
8953 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008954 break;
8955 case IORING_OFF_SQES:
8956 ptr = ctx->sq_sqes;
8957 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008958 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008959 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008960 }
8961
8962 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07008963 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008964 return ERR_PTR(-EINVAL);
8965
8966 return ptr;
8967}
8968
8969#ifdef CONFIG_MMU
8970
8971static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8972{
8973 size_t sz = vma->vm_end - vma->vm_start;
8974 unsigned long pfn;
8975 void *ptr;
8976
8977 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
8978 if (IS_ERR(ptr))
8979 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008980
8981 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
8982 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
8983}
8984
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008985#else /* !CONFIG_MMU */
8986
8987static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8988{
8989 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
8990}
8991
8992static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
8993{
8994 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
8995}
8996
8997static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
8998 unsigned long addr, unsigned long len,
8999 unsigned long pgoff, unsigned long flags)
9000{
9001 void *ptr;
9002
9003 ptr = io_uring_validate_mmap_request(file, pgoff, len);
9004 if (IS_ERR(ptr))
9005 return PTR_ERR(ptr);
9006
9007 return (unsigned long) ptr;
9008}
9009
9010#endif /* !CONFIG_MMU */
9011
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009012static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06009013{
9014 DEFINE_WAIT(wait);
9015
9016 do {
9017 if (!io_sqring_full(ctx))
9018 break;
Jens Axboe90554202020-09-03 12:12:41 -06009019 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
9020
9021 if (!io_sqring_full(ctx))
9022 break;
Jens Axboe90554202020-09-03 12:12:41 -06009023 schedule();
9024 } while (!signal_pending(current));
9025
9026 finish_wait(&ctx->sqo_sq_wait, &wait);
Yang Li51993282021-03-09 14:30:41 +08009027 return 0;
Jens Axboe90554202020-09-03 12:12:41 -06009028}
9029
Hao Xuc73ebb62020-11-03 10:54:37 +08009030static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9031 struct __kernel_timespec __user **ts,
9032 const sigset_t __user **sig)
9033{
9034 struct io_uring_getevents_arg arg;
9035
9036 /*
9037 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9038 * is just a pointer to the sigset_t.
9039 */
9040 if (!(flags & IORING_ENTER_EXT_ARG)) {
9041 *sig = (const sigset_t __user *) argp;
9042 *ts = NULL;
9043 return 0;
9044 }
9045
9046 /*
9047 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9048 * timespec and sigset_t pointers if good.
9049 */
9050 if (*argsz != sizeof(arg))
9051 return -EINVAL;
9052 if (copy_from_user(&arg, argp, sizeof(arg)))
9053 return -EFAULT;
9054 *sig = u64_to_user_ptr(arg.sigmask);
9055 *argsz = arg.sigmask_sz;
9056 *ts = u64_to_user_ptr(arg.ts);
9057 return 0;
9058}
9059
Jens Axboe2b188cc2019-01-07 10:46:33 -07009060SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009061 u32, min_complete, u32, flags, const void __user *, argp,
9062 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009063{
9064 struct io_ring_ctx *ctx;
9065 long ret = -EBADF;
9066 int submitted = 0;
9067 struct fd f;
9068
Jens Axboe4c6e2772020-07-01 11:29:10 -06009069 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009070
Jens Axboe90554202020-09-03 12:12:41 -06009071 if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
Hao Xuc73ebb62020-11-03 10:54:37 +08009072 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009073 return -EINVAL;
9074
9075 f = fdget(fd);
9076 if (!f.file)
9077 return -EBADF;
9078
9079 ret = -EOPNOTSUPP;
9080 if (f.file->f_op != &io_uring_fops)
9081 goto out_fput;
9082
9083 ret = -ENXIO;
9084 ctx = f.file->private_data;
9085 if (!percpu_ref_tryget(&ctx->refs))
9086 goto out_fput;
9087
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009088 ret = -EBADFD;
9089 if (ctx->flags & IORING_SETUP_R_DISABLED)
9090 goto out;
9091
Jens Axboe6c271ce2019-01-10 11:22:30 -07009092 /*
9093 * For SQ polling, the thread will do all submissions and completions.
9094 * Just return the requested submit count, and wake the thread if
9095 * we were asked to.
9096 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009097 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009098 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00009099 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009100
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009101 ret = -EOWNERDEAD;
Stefan Metzmacher04147482021-03-07 11:54:29 +01009102 if (unlikely(ctx->sq_data->thread == NULL)) {
9103 goto out;
9104 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009105 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009106 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009107 if (flags & IORING_ENTER_SQ_WAIT) {
9108 ret = io_sqpoll_wait_sq(ctx);
9109 if (ret)
9110 goto out;
9111 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009112 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009113 } else if (to_submit) {
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009114 ret = io_uring_add_task_file(ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06009115 if (unlikely(ret))
9116 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009117 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009118 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009119 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009120
9121 if (submitted != to_submit)
9122 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009123 }
9124 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009125 const sigset_t __user *sig;
9126 struct __kernel_timespec __user *ts;
9127
9128 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9129 if (unlikely(ret))
9130 goto out;
9131
Jens Axboe2b188cc2019-01-07 10:46:33 -07009132 min_complete = min(min_complete, ctx->cq_entries);
9133
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009134 /*
9135 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9136 * space applications don't need to do io completion events
9137 * polling again, they can rely on io_sq_thread to do polling
9138 * work, which can reduce cpu usage and uring_lock contention.
9139 */
9140 if (ctx->flags & IORING_SETUP_IOPOLL &&
9141 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009142 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009143 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009144 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009145 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009146 }
9147
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009148out:
Pavel Begunkov6805b322019-10-08 02:18:42 +03009149 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009150out_fput:
9151 fdput(f);
9152 return submitted ? submitted : ret;
9153}
9154
Tobias Klauserbebdb652020-02-26 18:38:32 +01009155#ifdef CONFIG_PROC_FS
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009156static int io_uring_show_cred(struct seq_file *m, unsigned int id,
9157 const struct cred *cred)
Jens Axboe87ce9552020-01-30 08:25:34 -07009158{
Jens Axboe87ce9552020-01-30 08:25:34 -07009159 struct user_namespace *uns = seq_user_ns(m);
9160 struct group_info *gi;
9161 kernel_cap_t cap;
9162 unsigned __capi;
9163 int g;
9164
9165 seq_printf(m, "%5d\n", id);
9166 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9167 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9168 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9169 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9170 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9171 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9172 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9173 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9174 seq_puts(m, "\n\tGroups:\t");
9175 gi = cred->group_info;
9176 for (g = 0; g < gi->ngroups; g++) {
9177 seq_put_decimal_ull(m, g ? " " : "",
9178 from_kgid_munged(uns, gi->gid[g]));
9179 }
9180 seq_puts(m, "\n\tCapEff:\t");
9181 cap = cred->cap_effective;
9182 CAP_FOR_EACH_U32(__capi)
9183 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9184 seq_putc(m, '\n');
9185 return 0;
9186}
9187
9188static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9189{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009190 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009191 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009192 int i;
9193
Jens Axboefad8e0d2020-09-28 08:57:48 -06009194 /*
9195 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9196 * since fdinfo case grabs it in the opposite direction of normal use
9197 * cases. If we fail to get the lock, we just don't iterate any
9198 * structures that could be going away outside the io_uring mutex.
9199 */
9200 has_lock = mutex_trylock(&ctx->uring_lock);
9201
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009202 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
Joseph Qidbbe9c62020-09-29 09:01:22 -06009203 sq = ctx->sq_data;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009204 if (!sq->thread)
9205 sq = NULL;
9206 }
Joseph Qidbbe9c62020-09-29 09:01:22 -06009207
9208 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9209 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009210 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009211 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Pavel Begunkovea64ec022021-02-04 13:52:07 +00009212 struct file *f = *io_fixed_file_slot(ctx->file_data, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009213
Jens Axboe87ce9552020-01-30 08:25:34 -07009214 if (f)
9215 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9216 else
9217 seq_printf(m, "%5u: <none>\n", i);
9218 }
9219 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009220 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009221 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
9222
9223 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf,
9224 (unsigned int) buf->len);
9225 }
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009226 if (has_lock && !xa_empty(&ctx->personalities)) {
9227 unsigned long index;
9228 const struct cred *cred;
9229
Jens Axboe87ce9552020-01-30 08:25:34 -07009230 seq_printf(m, "Personalities:\n");
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009231 xa_for_each(&ctx->personalities, index, cred)
9232 io_uring_show_cred(m, index, cred);
Jens Axboe87ce9552020-01-30 08:25:34 -07009233 }
Jens Axboed7718a92020-02-14 22:23:12 -07009234 seq_printf(m, "PollList:\n");
9235 spin_lock_irq(&ctx->completion_lock);
9236 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9237 struct hlist_head *list = &ctx->cancel_hash[i];
9238 struct io_kiocb *req;
9239
9240 hlist_for_each_entry(req, list, hash_node)
9241 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9242 req->task->task_works != NULL);
9243 }
9244 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009245 if (has_lock)
9246 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009247}
9248
9249static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9250{
9251 struct io_ring_ctx *ctx = f->private_data;
9252
9253 if (percpu_ref_tryget(&ctx->refs)) {
9254 __io_uring_show_fdinfo(ctx, m);
9255 percpu_ref_put(&ctx->refs);
9256 }
9257}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009258#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009259
Jens Axboe2b188cc2019-01-07 10:46:33 -07009260static const struct file_operations io_uring_fops = {
9261 .release = io_uring_release,
9262 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009263#ifndef CONFIG_MMU
9264 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9265 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9266#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009267 .poll = io_uring_poll,
9268 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009269#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009270 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009271#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009272};
9273
9274static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9275 struct io_uring_params *p)
9276{
Hristo Venev75b28af2019-08-26 17:23:46 +00009277 struct io_rings *rings;
9278 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009279
Jens Axboebd740482020-08-05 12:58:23 -06009280 /* make sure these are sane, as we already accounted them */
9281 ctx->sq_entries = p->sq_entries;
9282 ctx->cq_entries = p->cq_entries;
9283
Hristo Venev75b28af2019-08-26 17:23:46 +00009284 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9285 if (size == SIZE_MAX)
9286 return -EOVERFLOW;
9287
9288 rings = io_mem_alloc(size);
9289 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009290 return -ENOMEM;
9291
Hristo Venev75b28af2019-08-26 17:23:46 +00009292 ctx->rings = rings;
9293 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9294 rings->sq_ring_mask = p->sq_entries - 1;
9295 rings->cq_ring_mask = p->cq_entries - 1;
9296 rings->sq_ring_entries = p->sq_entries;
9297 rings->cq_ring_entries = p->cq_entries;
9298 ctx->sq_mask = rings->sq_ring_mask;
9299 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009300
9301 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009302 if (size == SIZE_MAX) {
9303 io_mem_free(ctx->rings);
9304 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009305 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009306 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009307
9308 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009309 if (!ctx->sq_sqes) {
9310 io_mem_free(ctx->rings);
9311 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009312 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009313 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009314
Jens Axboe2b188cc2019-01-07 10:46:33 -07009315 return 0;
9316}
9317
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009318static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9319{
9320 int ret, fd;
9321
9322 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9323 if (fd < 0)
9324 return fd;
9325
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009326 ret = io_uring_add_task_file(ctx);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009327 if (ret) {
9328 put_unused_fd(fd);
9329 return ret;
9330 }
9331 fd_install(fd, file);
9332 return fd;
9333}
9334
Jens Axboe2b188cc2019-01-07 10:46:33 -07009335/*
9336 * Allocate an anonymous fd, this is what constitutes the application
9337 * visible backing of an io_uring instance. The application mmaps this
9338 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9339 * we have to tie this fd to a socket for file garbage collection purposes.
9340 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009341static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009342{
9343 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009344#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009345 int ret;
9346
Jens Axboe2b188cc2019-01-07 10:46:33 -07009347 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9348 &ctx->ring_sock);
9349 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009350 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009351#endif
9352
Jens Axboe2b188cc2019-01-07 10:46:33 -07009353 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9354 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009355#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009356 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009357 sock_release(ctx->ring_sock);
9358 ctx->ring_sock = NULL;
9359 } else {
9360 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009361 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009362#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009363 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009364}
9365
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009366static int io_uring_create(unsigned entries, struct io_uring_params *p,
9367 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009368{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009369 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009370 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009371 int ret;
9372
Jens Axboe8110c1a2019-12-28 15:39:54 -07009373 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009374 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009375 if (entries > IORING_MAX_ENTRIES) {
9376 if (!(p->flags & IORING_SETUP_CLAMP))
9377 return -EINVAL;
9378 entries = IORING_MAX_ENTRIES;
9379 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009380
9381 /*
9382 * Use twice as many entries for the CQ ring. It's possible for the
9383 * application to drive a higher depth than the size of the SQ ring,
9384 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009385 * some flexibility in overcommitting a bit. If the application has
9386 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9387 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009388 */
9389 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009390 if (p->flags & IORING_SETUP_CQSIZE) {
9391 /*
9392 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9393 * to a power-of-two, if it isn't already. We do NOT impose
9394 * any cq vs sq ring sizing.
9395 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009396 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009397 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009398 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9399 if (!(p->flags & IORING_SETUP_CLAMP))
9400 return -EINVAL;
9401 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9402 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009403 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9404 if (p->cq_entries < p->sq_entries)
9405 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009406 } else {
9407 p->cq_entries = 2 * p->sq_entries;
9408 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009409
Jens Axboe2b188cc2019-01-07 10:46:33 -07009410 ctx = io_ring_ctx_alloc(p);
Jens Axboe62e398b2021-02-21 16:19:37 -07009411 if (!ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009412 return -ENOMEM;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009413 ctx->compat = in_compat_syscall();
Jens Axboe62e398b2021-02-21 16:19:37 -07009414 if (!capable(CAP_IPC_LOCK))
9415 ctx->user = get_uid(current_user());
Jens Axboe2aede0e2020-09-14 10:45:53 -06009416
9417 /*
9418 * This is just grabbed for accounting purposes. When a process exits,
9419 * the mm is exited and dropped before the files, hence we need to hang
9420 * on to this mm purely for the purposes of being able to unaccount
9421 * memory (locked/pinned vm). It's not used for anything else.
9422 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009423 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009424 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009425
Jens Axboe2b188cc2019-01-07 10:46:33 -07009426 ret = io_allocate_scq_urings(ctx, p);
9427 if (ret)
9428 goto err;
9429
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009430 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009431 if (ret)
9432 goto err;
9433
Jens Axboe2b188cc2019-01-07 10:46:33 -07009434 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009435 p->sq_off.head = offsetof(struct io_rings, sq.head);
9436 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9437 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9438 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9439 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9440 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9441 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009442
9443 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009444 p->cq_off.head = offsetof(struct io_rings, cq.head);
9445 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9446 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9447 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9448 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9449 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009450 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009451
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009452 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9453 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009454 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009455 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
Jens Axboe1c0aa1f2021-02-20 11:55:28 -07009456 IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009457
9458 if (copy_to_user(params, p, sizeof(*p))) {
9459 ret = -EFAULT;
9460 goto err;
9461 }
Jens Axboed1719f72020-07-30 13:43:53 -06009462
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009463 file = io_uring_get_file(ctx);
9464 if (IS_ERR(file)) {
9465 ret = PTR_ERR(file);
9466 goto err;
9467 }
9468
Jens Axboed1719f72020-07-30 13:43:53 -06009469 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009470 * Install ring fd as the very last thing, so we don't risk someone
9471 * having closed it before we finish setup
9472 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009473 ret = io_uring_install_fd(ctx, file);
9474 if (ret < 0) {
9475 /* fput will clean it up */
9476 fput(file);
9477 return ret;
9478 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009479
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009480 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009481 return ret;
9482err:
9483 io_ring_ctx_wait_and_kill(ctx);
9484 return ret;
9485}
9486
9487/*
9488 * Sets up an aio uring context, and returns the fd. Applications asks for a
9489 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9490 * params structure passed in.
9491 */
9492static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9493{
9494 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009495 int i;
9496
9497 if (copy_from_user(&p, params, sizeof(p)))
9498 return -EFAULT;
9499 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9500 if (p.resv[i])
9501 return -EINVAL;
9502 }
9503
Jens Axboe6c271ce2019-01-10 11:22:30 -07009504 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009505 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009506 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9507 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009508 return -EINVAL;
9509
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009510 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009511}
9512
9513SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9514 struct io_uring_params __user *, params)
9515{
9516 return io_uring_setup(entries, params);
9517}
9518
Jens Axboe66f4af92020-01-16 15:36:52 -07009519static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9520{
9521 struct io_uring_probe *p;
9522 size_t size;
9523 int i, ret;
9524
9525 size = struct_size(p, ops, nr_args);
9526 if (size == SIZE_MAX)
9527 return -EOVERFLOW;
9528 p = kzalloc(size, GFP_KERNEL);
9529 if (!p)
9530 return -ENOMEM;
9531
9532 ret = -EFAULT;
9533 if (copy_from_user(p, arg, size))
9534 goto out;
9535 ret = -EINVAL;
9536 if (memchr_inv(p, 0, size))
9537 goto out;
9538
9539 p->last_op = IORING_OP_LAST - 1;
9540 if (nr_args > IORING_OP_LAST)
9541 nr_args = IORING_OP_LAST;
9542
9543 for (i = 0; i < nr_args; i++) {
9544 p->ops[i].op = i;
9545 if (!io_op_defs[i].not_supported)
9546 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9547 }
9548 p->ops_len = i;
9549
9550 ret = 0;
9551 if (copy_to_user(arg, p, size))
9552 ret = -EFAULT;
9553out:
9554 kfree(p);
9555 return ret;
9556}
9557
Jens Axboe071698e2020-01-28 10:04:42 -07009558static int io_register_personality(struct io_ring_ctx *ctx)
9559{
Jens Axboe4379bf82021-02-15 13:40:22 -07009560 const struct cred *creds;
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009561 u32 id;
Jens Axboe1e6fa522020-10-15 08:46:24 -06009562 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009563
Jens Axboe4379bf82021-02-15 13:40:22 -07009564 creds = get_current_cred();
Jens Axboe1e6fa522020-10-15 08:46:24 -06009565
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009566 ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds,
9567 XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL);
9568 if (!ret)
9569 return id;
9570 put_cred(creds);
Jens Axboe1e6fa522020-10-15 08:46:24 -06009571 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009572}
9573
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009574static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9575 unsigned int nr_args)
9576{
9577 struct io_uring_restriction *res;
9578 size_t size;
9579 int i, ret;
9580
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009581 /* Restrictions allowed only if rings started disabled */
9582 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9583 return -EBADFD;
9584
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009585 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009586 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009587 return -EBUSY;
9588
9589 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9590 return -EINVAL;
9591
9592 size = array_size(nr_args, sizeof(*res));
9593 if (size == SIZE_MAX)
9594 return -EOVERFLOW;
9595
9596 res = memdup_user(arg, size);
9597 if (IS_ERR(res))
9598 return PTR_ERR(res);
9599
9600 ret = 0;
9601
9602 for (i = 0; i < nr_args; i++) {
9603 switch (res[i].opcode) {
9604 case IORING_RESTRICTION_REGISTER_OP:
9605 if (res[i].register_op >= IORING_REGISTER_LAST) {
9606 ret = -EINVAL;
9607 goto out;
9608 }
9609
9610 __set_bit(res[i].register_op,
9611 ctx->restrictions.register_op);
9612 break;
9613 case IORING_RESTRICTION_SQE_OP:
9614 if (res[i].sqe_op >= IORING_OP_LAST) {
9615 ret = -EINVAL;
9616 goto out;
9617 }
9618
9619 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9620 break;
9621 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9622 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9623 break;
9624 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9625 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9626 break;
9627 default:
9628 ret = -EINVAL;
9629 goto out;
9630 }
9631 }
9632
9633out:
9634 /* Reset all restrictions if an error happened */
9635 if (ret != 0)
9636 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9637 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009638 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009639
9640 kfree(res);
9641 return ret;
9642}
9643
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009644static int io_register_enable_rings(struct io_ring_ctx *ctx)
9645{
9646 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9647 return -EBADFD;
9648
9649 if (ctx->restrictions.registered)
9650 ctx->restricted = 1;
9651
Pavel Begunkov0298ef92021-03-08 13:20:57 +00009652 ctx->flags &= ~IORING_SETUP_R_DISABLED;
9653 if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait))
9654 wake_up(&ctx->sq_data->wait);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009655 return 0;
9656}
9657
Jens Axboe071698e2020-01-28 10:04:42 -07009658static bool io_register_op_must_quiesce(int op)
9659{
9660 switch (op) {
9661 case IORING_UNREGISTER_FILES:
9662 case IORING_REGISTER_FILES_UPDATE:
9663 case IORING_REGISTER_PROBE:
9664 case IORING_REGISTER_PERSONALITY:
9665 case IORING_UNREGISTER_PERSONALITY:
9666 return false;
9667 default:
9668 return true;
9669 }
9670}
9671
Jens Axboeedafcce2019-01-09 09:16:05 -07009672static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
9673 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -06009674 __releases(ctx->uring_lock)
9675 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -07009676{
9677 int ret;
9678
Jens Axboe35fa71a2019-04-22 10:23:23 -06009679 /*
9680 * We're inside the ring mutex, if the ref is already dying, then
9681 * someone else killed the ctx or is already going through
9682 * io_uring_register().
9683 */
9684 if (percpu_ref_is_dying(&ctx->refs))
9685 return -ENXIO;
9686
Jens Axboe071698e2020-01-28 10:04:42 -07009687 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009688 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -06009689
Jens Axboe05f3fb32019-12-09 11:22:50 -07009690 /*
9691 * Drop uring mutex before waiting for references to exit. If
9692 * another thread is currently inside io_uring_enter() it might
9693 * need to grab the uring_lock to make progress. If we hold it
9694 * here across the drain wait, then we can deadlock. It's safe
9695 * to drop the mutex here, since no new references will come in
9696 * after we've killed the percpu ref.
9697 */
9698 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009699 do {
9700 ret = wait_for_completion_interruptible(&ctx->ref_comp);
9701 if (!ret)
9702 break;
Jens Axboeed6930c2020-10-08 19:09:46 -06009703 ret = io_run_task_work_sig();
9704 if (ret < 0)
9705 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009706 } while (1);
9707
Jens Axboe05f3fb32019-12-09 11:22:50 -07009708 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009709
Jens Axboec1503682020-01-08 08:26:07 -07009710 if (ret) {
9711 percpu_ref_resurrect(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009712 goto out_quiesce;
9713 }
9714 }
9715
9716 if (ctx->restricted) {
9717 if (opcode >= IORING_REGISTER_LAST) {
9718 ret = -EINVAL;
9719 goto out;
9720 }
9721
9722 if (!test_bit(opcode, ctx->restrictions.register_op)) {
9723 ret = -EACCES;
Jens Axboec1503682020-01-08 08:26:07 -07009724 goto out;
9725 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07009726 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009727
9728 switch (opcode) {
9729 case IORING_REGISTER_BUFFERS:
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009730 ret = io_sqe_buffers_register(ctx, arg, nr_args);
Jens Axboeedafcce2019-01-09 09:16:05 -07009731 break;
9732 case IORING_UNREGISTER_BUFFERS:
9733 ret = -EINVAL;
9734 if (arg || nr_args)
9735 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009736 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -07009737 break;
Jens Axboe6b063142019-01-10 22:13:58 -07009738 case IORING_REGISTER_FILES:
9739 ret = io_sqe_files_register(ctx, arg, nr_args);
9740 break;
9741 case IORING_UNREGISTER_FILES:
9742 ret = -EINVAL;
9743 if (arg || nr_args)
9744 break;
9745 ret = io_sqe_files_unregister(ctx);
9746 break;
Jens Axboec3a31e62019-10-03 13:59:56 -06009747 case IORING_REGISTER_FILES_UPDATE:
9748 ret = io_sqe_files_update(ctx, arg, nr_args);
9749 break;
Jens Axboe9b402842019-04-11 11:45:41 -06009750 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -07009751 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -06009752 ret = -EINVAL;
9753 if (nr_args != 1)
9754 break;
9755 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -07009756 if (ret)
9757 break;
9758 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
9759 ctx->eventfd_async = 1;
9760 else
9761 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -06009762 break;
9763 case IORING_UNREGISTER_EVENTFD:
9764 ret = -EINVAL;
9765 if (arg || nr_args)
9766 break;
9767 ret = io_eventfd_unregister(ctx);
9768 break;
Jens Axboe66f4af92020-01-16 15:36:52 -07009769 case IORING_REGISTER_PROBE:
9770 ret = -EINVAL;
9771 if (!arg || nr_args > 256)
9772 break;
9773 ret = io_probe(ctx, arg, nr_args);
9774 break;
Jens Axboe071698e2020-01-28 10:04:42 -07009775 case IORING_REGISTER_PERSONALITY:
9776 ret = -EINVAL;
9777 if (arg || nr_args)
9778 break;
9779 ret = io_register_personality(ctx);
9780 break;
9781 case IORING_UNREGISTER_PERSONALITY:
9782 ret = -EINVAL;
9783 if (arg)
9784 break;
9785 ret = io_unregister_personality(ctx, nr_args);
9786 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009787 case IORING_REGISTER_ENABLE_RINGS:
9788 ret = -EINVAL;
9789 if (arg || nr_args)
9790 break;
9791 ret = io_register_enable_rings(ctx);
9792 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009793 case IORING_REGISTER_RESTRICTIONS:
9794 ret = io_register_restrictions(ctx, arg, nr_args);
9795 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07009796 default:
9797 ret = -EINVAL;
9798 break;
9799 }
9800
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009801out:
Jens Axboe071698e2020-01-28 10:04:42 -07009802 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009803 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -07009804 percpu_ref_reinit(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009805out_quiesce:
Jens Axboe0f158b42020-05-14 17:18:39 -06009806 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009807 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009808 return ret;
9809}
9810
9811SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
9812 void __user *, arg, unsigned int, nr_args)
9813{
9814 struct io_ring_ctx *ctx;
9815 long ret = -EBADF;
9816 struct fd f;
9817
9818 f = fdget(fd);
9819 if (!f.file)
9820 return -EBADF;
9821
9822 ret = -EOPNOTSUPP;
9823 if (f.file->f_op != &io_uring_fops)
9824 goto out_fput;
9825
9826 ctx = f.file->private_data;
9827
Pavel Begunkovb6c23dd2021-02-20 15:17:18 +00009828 io_run_task_work();
9829
Jens Axboeedafcce2019-01-09 09:16:05 -07009830 mutex_lock(&ctx->uring_lock);
9831 ret = __io_uring_register(ctx, opcode, arg, nr_args);
9832 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009833 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
9834 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07009835out_fput:
9836 fdput(f);
9837 return ret;
9838}
9839
Jens Axboe2b188cc2019-01-07 10:46:33 -07009840static int __init io_uring_init(void)
9841{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009842#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
9843 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
9844 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
9845} while (0)
9846
9847#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
9848 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
9849 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
9850 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
9851 BUILD_BUG_SQE_ELEM(1, __u8, flags);
9852 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
9853 BUILD_BUG_SQE_ELEM(4, __s32, fd);
9854 BUILD_BUG_SQE_ELEM(8, __u64, off);
9855 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
9856 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009857 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009858 BUILD_BUG_SQE_ELEM(24, __u32, len);
9859 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
9860 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
9861 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
9862 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +08009863 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
9864 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009865 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
9866 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
9867 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
9868 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
9869 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
9870 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
9871 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
9872 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009873 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009874 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
9875 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
9876 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009877 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009878
Jens Axboed3656342019-12-18 09:50:26 -07009879 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -07009880 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe91f245d2021-02-09 13:48:50 -07009881 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
9882 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009883 return 0;
9884};
9885__initcall(io_uring_init);