blob: 1c846d5f39d8e33097b5d25f15cd5796d424a31f [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 Axboe2b188cc2019-01-07 10:46:33 -070081
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020082#define CREATE_TRACE_POINTS
83#include <trace/events/io_uring.h>
84
Jens Axboe2b188cc2019-01-07 10:46:33 -070085#include <uapi/linux/io_uring.h>
86
87#include "internal.h"
Jens Axboe561fb042019-10-24 07:25:42 -060088#include "io-wq.h"
Jens Axboe2b188cc2019-01-07 10:46:33 -070089
Daniel Xu5277dea2019-09-14 14:23:45 -070090#define IORING_MAX_ENTRIES 32768
Jens Axboe33a107f2019-10-04 12:10:03 -060091#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
Jens Axboe65e19f52019-10-26 07:20:21 -060092
93/*
94 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
95 */
96#define IORING_FILE_TABLE_SHIFT 9
97#define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT)
98#define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1)
99#define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE)
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200100#define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \
101 IORING_REGISTER_LAST + IORING_OP_LAST)
Jens Axboe2b188cc2019-01-07 10:46:33 -0700102
Pavel Begunkovb16fed662021-02-18 18:29:40 +0000103#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
104 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
105 IOSQE_BUFFER_SELECT)
106
Jens Axboe2b188cc2019-01-07 10:46:33 -0700107struct io_uring {
108 u32 head ____cacheline_aligned_in_smp;
109 u32 tail ____cacheline_aligned_in_smp;
110};
111
Stefan Bühler1e84b972019-04-24 23:54:16 +0200112/*
Hristo Venev75b28af2019-08-26 17:23:46 +0000113 * This data is shared with the application through the mmap at offsets
114 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200115 *
116 * The offsets to the member fields are published through struct
117 * io_sqring_offsets when calling io_uring_setup.
118 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000119struct io_rings {
Stefan Bühler1e84b972019-04-24 23:54:16 +0200120 /*
121 * Head and tail offsets into the ring; the offsets need to be
122 * masked to get valid indices.
123 *
Hristo Venev75b28af2019-08-26 17:23:46 +0000124 * The kernel controls head of the sq ring and the tail of the cq ring,
125 * and the application controls tail of the sq ring and the head of the
126 * cq ring.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200127 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000128 struct io_uring sq, cq;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200129 /*
Hristo Venev75b28af2019-08-26 17:23:46 +0000130 * Bitmasks to apply to head and tail offsets (constant, equals
Stefan Bühler1e84b972019-04-24 23:54:16 +0200131 * ring_entries - 1)
132 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000133 u32 sq_ring_mask, cq_ring_mask;
134 /* Ring sizes (constant, power of 2) */
135 u32 sq_ring_entries, cq_ring_entries;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200136 /*
137 * Number of invalid entries dropped by the kernel due to
138 * invalid index stored in array
139 *
140 * Written by the kernel, shouldn't be modified by the
141 * application (i.e. get number of "new events" by comparing to
142 * cached value).
143 *
144 * After a new SQ head value was read by the application this
145 * counter includes all submissions that were dropped reaching
146 * the new SQ head (and possibly more).
147 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000148 u32 sq_dropped;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200149 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200150 * Runtime SQ flags
Stefan Bühler1e84b972019-04-24 23:54:16 +0200151 *
152 * Written by the kernel, shouldn't be modified by the
153 * application.
154 *
155 * The application needs a full memory barrier before checking
156 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
157 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000158 u32 sq_flags;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200159 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200160 * Runtime CQ flags
161 *
162 * Written by the application, shouldn't be modified by the
163 * kernel.
164 */
165 u32 cq_flags;
166 /*
Stefan Bühler1e84b972019-04-24 23:54:16 +0200167 * Number of completion events lost because the queue was full;
168 * this should be avoided by the application by making sure
LimingWu0b4295b2019-12-05 20:18:18 +0800169 * there are not more requests pending than there is space in
Stefan Bühler1e84b972019-04-24 23:54:16 +0200170 * the completion queue.
171 *
172 * Written by the kernel, shouldn't be modified by the
173 * application (i.e. get number of "new events" by comparing to
174 * cached value).
175 *
176 * As completion events come in out of order this counter is not
177 * ordered with any other data.
178 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000179 u32 cq_overflow;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200180 /*
181 * Ring buffer of completion events.
182 *
183 * The kernel writes completion events fresh every time they are
184 * produced, so the application is allowed to modify pending
185 * entries.
186 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000187 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700188};
189
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000190enum io_uring_cmd_flags {
191 IO_URING_F_NONBLOCK = 1,
Pavel Begunkov889fca72021-02-10 00:03:09 +0000192 IO_URING_F_COMPLETE_DEFER = 2,
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000193};
194
Jens Axboeedafcce2019-01-09 09:16:05 -0700195struct io_mapped_ubuf {
196 u64 ubuf;
197 size_t len;
198 struct bio_vec *bvec;
199 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600200 unsigned long acct_pages;
Jens Axboeedafcce2019-01-09 09:16:05 -0700201};
202
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000203struct io_ring_ctx;
204
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000205struct io_rsrc_put {
206 struct list_head list;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000207 union {
208 void *rsrc;
209 struct file *file;
210 };
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000211};
212
213struct fixed_rsrc_table {
Jens Axboe65e19f52019-10-26 07:20:21 -0600214 struct file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700215};
216
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000217struct fixed_rsrc_ref_node {
Xiaoguang Wang05589552020-03-31 14:05:18 +0800218 struct percpu_ref refs;
219 struct list_head node;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000220 struct list_head rsrc_list;
221 struct fixed_rsrc_data *rsrc_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000222 void (*rsrc_put)(struct io_ring_ctx *ctx,
223 struct io_rsrc_put *prsrc);
Jens Axboe4a38aed22020-05-14 17:21:15 -0600224 struct llist_node llist;
Pavel Begunkove2978222020-11-18 14:56:26 +0000225 bool done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800226};
227
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000228struct fixed_rsrc_data {
229 struct fixed_rsrc_table *table;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700230 struct io_ring_ctx *ctx;
231
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000232 struct fixed_rsrc_ref_node *node;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700233 struct percpu_ref refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700234 struct completion done;
Hao Xu8bad28d2021-02-19 17:19:36 +0800235 bool quiesce;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700236};
237
Jens Axboe5a2e7452020-02-23 16:23:11 -0700238struct io_buffer {
239 struct list_head list;
240 __u64 addr;
241 __s32 len;
242 __u16 bid;
243};
244
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200245struct io_restriction {
246 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
247 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
248 u8 sqe_flags_allowed;
249 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200250 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200251};
252
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700253enum {
254 IO_SQ_THREAD_SHOULD_STOP = 0,
255 IO_SQ_THREAD_SHOULD_PARK,
256};
257
Jens Axboe534ca6d2020-09-02 13:52:19 -0600258struct io_sq_data {
259 refcount_t refs;
Pavel Begunkov9e138a42021-03-14 20:57:12 +0000260 atomic_t park_pending;
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +0000261 struct mutex 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;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700275 struct completion exited;
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +0000276 struct callback_head *park_task_work;
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 Axboe9e15c3a2021-03-13 12:29:43 -0700406 struct xarray io_buffers;
Jens Axboe5a2e7452020-02-23 16:23:11 -0700407
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
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100458struct io_uring_task {
459 /* submission side */
460 struct xarray xa;
461 struct wait_queue_head wait;
Stefan Metzmacheree53fb22021-03-15 12:56:57 +0100462 const struct io_ring_ctx *last;
463 struct io_wq *io_wq;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100464 struct percpu_counter inflight;
465 atomic_t in_idle;
466 bool sqpoll;
467
468 spinlock_t task_lock;
469 struct io_wq_work_list task_list;
470 unsigned long task_state;
471 struct callback_head task_work;
472};
473
Jens Axboe09bb8392019-03-13 12:39:28 -0600474/*
475 * First field must be the file pointer in all the
476 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
477 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700478struct io_poll_iocb {
479 struct file *file;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000480 struct wait_queue_head *head;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700481 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600482 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700483 bool canceled;
Jens Axboe392edb42019-12-09 17:52:20 -0700484 struct wait_queue_entry wait;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700485};
486
Pavel Begunkov018043b2020-10-27 23:17:18 +0000487struct io_poll_remove {
488 struct file *file;
489 u64 addr;
490};
491
Jens Axboeb5dba592019-12-11 14:02:38 -0700492struct io_close {
493 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700494 int fd;
495};
496
Jens Axboead8a48a2019-11-15 08:49:11 -0700497struct io_timeout_data {
498 struct io_kiocb *req;
499 struct hrtimer timer;
500 struct timespec64 ts;
501 enum hrtimer_mode mode;
502};
503
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700504struct io_accept {
505 struct file *file;
506 struct sockaddr __user *addr;
507 int __user *addr_len;
508 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600509 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700510};
511
512struct io_sync {
513 struct file *file;
514 loff_t len;
515 loff_t off;
516 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700517 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700518};
519
Jens Axboefbf23842019-12-17 18:45:56 -0700520struct io_cancel {
521 struct file *file;
522 u64 addr;
523};
524
Jens Axboeb29472e2019-12-17 18:50:29 -0700525struct io_timeout {
526 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300527 u32 off;
528 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300529 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000530 /* head of the link, used by linked timeouts only */
531 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700532};
533
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100534struct io_timeout_rem {
535 struct file *file;
536 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000537
538 /* timeout update */
539 struct timespec64 ts;
540 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100541};
542
Jens Axboe9adbd452019-12-20 08:45:55 -0700543struct io_rw {
544 /* NOTE: kiocb has the file as the first member, so don't do it here */
545 struct kiocb kiocb;
546 u64 addr;
547 u64 len;
548};
549
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700550struct io_connect {
551 struct file *file;
552 struct sockaddr __user *addr;
553 int addr_len;
554};
555
Jens Axboee47293f2019-12-20 08:58:21 -0700556struct io_sr_msg {
557 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700558 union {
Pavel Begunkov270a5942020-07-12 20:41:04 +0300559 struct user_msghdr __user *umsg;
Jens Axboefddafac2020-01-04 20:19:44 -0700560 void __user *buf;
561 };
Jens Axboee47293f2019-12-20 08:58:21 -0700562 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700563 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700564 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700565 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700566};
567
Jens Axboe15b71ab2019-12-11 11:20:36 -0700568struct io_open {
569 struct file *file;
570 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700571 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700572 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600573 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700574};
575
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000576struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700577 struct file *file;
578 u64 arg;
579 u32 nr_args;
580 u32 offset;
581};
582
Jens Axboe4840e412019-12-25 22:03:45 -0700583struct io_fadvise {
584 struct file *file;
585 u64 offset;
586 u32 len;
587 u32 advice;
588};
589
Jens Axboec1ca7572019-12-25 22:18:28 -0700590struct io_madvise {
591 struct file *file;
592 u64 addr;
593 u32 len;
594 u32 advice;
595};
596
Jens Axboe3e4827b2020-01-08 15:18:09 -0700597struct io_epoll {
598 struct file *file;
599 int epfd;
600 int op;
601 int fd;
602 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700603};
604
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300605struct io_splice {
606 struct file *file_out;
607 struct file *file_in;
608 loff_t off_out;
609 loff_t off_in;
610 u64 len;
611 unsigned int flags;
612};
613
Jens Axboeddf0322d2020-02-23 16:41:33 -0700614struct io_provide_buf {
615 struct file *file;
616 __u64 addr;
617 __s32 len;
618 __u32 bgid;
619 __u16 nbufs;
620 __u16 bid;
621};
622
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700623struct io_statx {
624 struct file *file;
625 int dfd;
626 unsigned int mask;
627 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700628 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700629 struct statx __user *buffer;
630};
631
Jens Axboe36f4fa62020-09-05 11:14:22 -0600632struct io_shutdown {
633 struct file *file;
634 int how;
635};
636
Jens Axboe80a261f2020-09-28 14:23:58 -0600637struct io_rename {
638 struct file *file;
639 int old_dfd;
640 int new_dfd;
641 struct filename *oldpath;
642 struct filename *newpath;
643 int flags;
644};
645
Jens Axboe14a11432020-09-28 14:27:37 -0600646struct io_unlink {
647 struct file *file;
648 int dfd;
649 int flags;
650 struct filename *filename;
651};
652
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300653struct io_completion {
654 struct file *file;
655 struct list_head list;
Pavel Begunkov8c3f9cd2021-02-28 22:35:15 +0000656 u32 cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300657};
658
Jens Axboef499a022019-12-02 16:28:46 -0700659struct io_async_connect {
660 struct sockaddr_storage address;
661};
662
Jens Axboe03b12302019-12-02 18:50:25 -0700663struct io_async_msghdr {
664 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000665 /* points to an allocated iov, if NULL we use fast_iov instead */
666 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700667 struct sockaddr __user *uaddr;
668 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700669 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700670};
671
Jens Axboef67676d2019-12-02 11:03:47 -0700672struct io_async_rw {
673 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600674 const struct iovec *free_iovec;
675 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600676 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600677 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700678};
679
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300680enum {
681 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
682 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
683 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
684 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
685 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700686 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300687
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300688 REQ_F_FAIL_LINK_BIT,
689 REQ_F_INFLIGHT_BIT,
690 REQ_F_CUR_POS_BIT,
691 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300692 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300693 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700694 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700695 REQ_F_BUFFER_SELECTED_BIT,
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600696 REQ_F_NO_FILE_TABLE_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100697 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000698 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe230d50d2021-04-01 20:41:15 -0600699 REQ_F_REISSUE_BIT,
Jens Axboe7b29f922021-03-12 08:30:14 -0700700 /* keep async read/write and isreg together and in order */
701 REQ_F_ASYNC_READ_BIT,
702 REQ_F_ASYNC_WRITE_BIT,
703 REQ_F_ISREG_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700704
705 /* not a real bit, just to check we're not overflowing the space */
706 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300707};
708
709enum {
710 /* ctx owns file */
711 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
712 /* drain existing IO first */
713 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
714 /* linked sqes */
715 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
716 /* doesn't sever on completion < 0 */
717 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
718 /* IOSQE_ASYNC */
719 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700720 /* IOSQE_BUFFER_SELECT */
721 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300722
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300723 /* fail rest of links */
724 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +0000725 /* on inflight list, should be cancelled and waited on exit reliably */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300726 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
727 /* read/write uses file position */
728 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
729 /* must not punt to workers */
730 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100731 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300732 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300733 /* needs cleanup */
734 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700735 /* already went through poll handler */
736 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700737 /* buffer already selected */
738 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600739 /* doesn't need file table for this request */
740 REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100741 /* linked timeout is active, i.e. prepared by link's head */
742 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000743 /* completion is deferred through io_comp_state */
744 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboe230d50d2021-04-01 20:41:15 -0600745 /* caller should reissue async */
746 REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT),
Jens Axboe7b29f922021-03-12 08:30:14 -0700747 /* supports async reads */
748 REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT),
749 /* supports async writes */
750 REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT),
751 /* regular file */
752 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700753};
754
755struct async_poll {
756 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600757 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300758};
759
Jens Axboe7cbf1722021-02-10 00:03:20 +0000760struct io_task_work {
761 struct io_wq_work_node node;
762 task_work_func_t func;
763};
764
Jens Axboe09bb8392019-03-13 12:39:28 -0600765/*
766 * NOTE! Each of the iocb union members has the file pointer
767 * as the first entry in their struct definition. So you can
768 * access the file pointer through any of the sub-structs,
769 * or directly as just 'ki_filp' in this struct.
770 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700771struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700772 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600773 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700774 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700775 struct io_poll_iocb poll;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000776 struct io_poll_remove poll_remove;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700777 struct io_accept accept;
778 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700779 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700780 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100781 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700782 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700783 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700784 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700785 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000786 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700787 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700788 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700789 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300790 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700791 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700792 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600793 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600794 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600795 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300796 /* use only after cleaning per-op data, see io_clean_op() */
797 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700798 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700799
Jens Axboee8c2bc12020-08-15 18:44:09 -0700800 /* opcode allocated if it needs to store data for async defer */
801 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700802 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800803 /* polled IO has completed */
804 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700805
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700806 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300807 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700808
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300809 struct io_ring_ctx *ctx;
810 unsigned int flags;
Jens Axboeabc54d62021-02-24 13:32:30 -0700811 atomic_t refs;
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300812 struct task_struct *task;
813 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700814
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000815 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000816 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700817
Pavel Begunkovd21ffe72020-07-13 23:37:10 +0300818 /*
819 * 1. used with ctx->iopoll_list with reads/writes
820 * 2. to track reqs with ->files (see io_op_def::file_table)
821 */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300822 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000823 union {
824 struct io_task_work io_task_work;
825 struct callback_head task_work;
826 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300827 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
828 struct hlist_node hash_node;
829 struct async_poll *apoll;
830 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700831};
832
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000833struct io_tctx_node {
834 struct list_head ctx_node;
835 struct task_struct *task;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000836 struct io_ring_ctx *ctx;
837};
838
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300839struct io_defer_entry {
840 struct list_head list;
841 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300842 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300843};
844
Jens Axboed3656342019-12-18 09:50:26 -0700845struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700846 /* needs req->file assigned */
847 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700848 /* hash wq insertion if file is a regular file */
849 unsigned hash_reg_file : 1;
850 /* unbound wq insertion if file is a non-regular file */
851 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700852 /* opcode is not supported by this kernel */
853 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700854 /* set if opcode supports polled "wait" */
855 unsigned pollin : 1;
856 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700857 /* op supports buffer selection */
858 unsigned buffer_select : 1;
Pavel Begunkov26f05052021-02-28 22:35:18 +0000859 /* do prep async if is going to be punted */
860 unsigned needs_async_setup : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600861 /* should block plug */
862 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700863 /* size of async data needed, if any */
864 unsigned short async_size;
Jens Axboed3656342019-12-18 09:50:26 -0700865};
866
Jens Axboe09186822020-10-13 15:01:40 -0600867static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300868 [IORING_OP_NOP] = {},
869 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700870 .needs_file = 1,
871 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700872 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700873 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000874 .needs_async_setup = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600875 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700876 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700877 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300878 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700879 .needs_file = 1,
880 .hash_reg_file = 1,
881 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700882 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000883 .needs_async_setup = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600884 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700885 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700886 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300887 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700888 .needs_file = 1,
889 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300890 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700891 .needs_file = 1,
892 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700893 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600894 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700895 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700896 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300897 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700898 .needs_file = 1,
899 .hash_reg_file = 1,
900 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700901 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600902 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700903 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700904 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300905 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700906 .needs_file = 1,
907 .unbound_nonreg_file = 1,
908 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300909 [IORING_OP_POLL_REMOVE] = {},
910 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700911 .needs_file = 1,
912 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300913 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700914 .needs_file = 1,
915 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700916 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000917 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700918 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700919 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300920 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700921 .needs_file = 1,
922 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700923 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700924 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000925 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700926 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700927 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300928 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700929 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700930 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000931 [IORING_OP_TIMEOUT_REMOVE] = {
932 /* used by timeout updates' prep() */
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000933 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300934 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700935 .needs_file = 1,
936 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700937 .pollin = 1,
Jens Axboed3656342019-12-18 09:50:26 -0700938 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300939 [IORING_OP_ASYNC_CANCEL] = {},
940 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700941 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700942 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300943 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700944 .needs_file = 1,
945 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700946 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000947 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700948 .async_size = sizeof(struct io_async_connect),
Jens Axboed3656342019-12-18 09:50:26 -0700949 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300950 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700951 .needs_file = 1,
952 },
Jens Axboe44526be2021-02-15 13:32:18 -0700953 [IORING_OP_OPENAT] = {},
954 [IORING_OP_CLOSE] = {},
955 [IORING_OP_FILES_UPDATE] = {},
956 [IORING_OP_STATX] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300957 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -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 Axboe27926b62020-10-28 09:33:23 -0600962 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700963 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700964 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300965 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700966 .needs_file = 1,
967 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700968 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600969 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700970 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700971 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300972 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700973 .needs_file = 1,
974 },
Jens Axboe44526be2021-02-15 13:32:18 -0700975 [IORING_OP_MADVISE] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300976 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700977 .needs_file = 1,
978 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700979 .pollout = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700980 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300981 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700982 .needs_file = 1,
983 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700984 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700985 .buffer_select = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700986 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300987 [IORING_OP_OPENAT2] = {
Jens Axboecebdb982020-01-08 17:59:24 -0700988 },
Jens Axboe3e4827b2020-01-08 15:18:09 -0700989 [IORING_OP_EPOLL_CTL] = {
990 .unbound_nonreg_file = 1,
Jens Axboe3e4827b2020-01-08 15:18:09 -0700991 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300992 [IORING_OP_SPLICE] = {
993 .needs_file = 1,
994 .hash_reg_file = 1,
995 .unbound_nonreg_file = 1,
Jens Axboeddf0322d2020-02-23 16:41:33 -0700996 },
997 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -0700998 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +0300999 [IORING_OP_TEE] = {
1000 .needs_file = 1,
1001 .hash_reg_file = 1,
1002 .unbound_nonreg_file = 1,
1003 },
Jens Axboe36f4fa62020-09-05 11:14:22 -06001004 [IORING_OP_SHUTDOWN] = {
1005 .needs_file = 1,
1006 },
Jens Axboe44526be2021-02-15 13:32:18 -07001007 [IORING_OP_RENAMEAT] = {},
1008 [IORING_OP_UNLINKAT] = {},
Jens Axboed3656342019-12-18 09:50:26 -07001009};
1010
Pavel Begunkov7a612352021-03-09 00:37:59 +00001011static bool io_disarm_next(struct io_kiocb *req);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00001012static void io_uring_del_task_file(unsigned long index);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00001013static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
1014 struct task_struct *task,
1015 struct files_struct *files);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07001016static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001017static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node);
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00001018static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00001019 struct io_ring_ctx *ctx);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00001020static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00001021
Pavel Begunkov23faba32021-02-11 18:28:22 +00001022static bool io_rw_reissue(struct io_kiocb *req);
Jens Axboe78e19bb2019-11-06 15:21:34 -07001023static void io_cqring_fill_event(struct io_kiocb *req, long res);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001024static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001025static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec40f6372020-06-25 15:39:59 -06001026static void io_double_put_req(struct io_kiocb *req);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001027static void io_dismantle_req(struct io_kiocb *req);
1028static void io_put_task(struct task_struct *task, int nr);
1029static void io_queue_next(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001030static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001031static void __io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001032static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -07001033static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001034 struct io_uring_rsrc_update *ip,
Jens Axboe05f3fb32019-12-09 11:22:50 -07001035 unsigned nr_args);
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001036static void __io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001037static struct file *io_file_get(struct io_submit_state *state,
1038 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001039static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001040static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001041
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001042static void io_req_task_queue(struct io_kiocb *req);
Jens Axboe65453d12021-02-10 00:03:21 +00001043static void io_submit_flush_completions(struct io_comp_state *cs,
1044 struct io_ring_ctx *ctx);
Pavel Begunkov179ae0d2021-02-28 22:35:20 +00001045static int io_req_prep_async(struct io_kiocb *req);
Jens Axboe9a56a232019-01-09 09:06:50 -07001046
Jens Axboe2b188cc2019-01-07 10:46:33 -07001047static struct kmem_cache *req_cachep;
1048
Jens Axboe09186822020-10-13 15:01:40 -06001049static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001050
1051struct sock *io_uring_get_socket(struct file *file)
1052{
1053#if defined(CONFIG_UNIX)
1054 if (file->f_op == &io_uring_fops) {
1055 struct io_ring_ctx *ctx = file->private_data;
1056
1057 return ctx->ring_sock->sk;
1058 }
1059#endif
1060 return NULL;
1061}
1062EXPORT_SYMBOL(io_uring_get_socket);
1063
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001064#define io_for_each_link(pos, head) \
1065 for (pos = (head); pos; pos = pos->link)
1066
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001067static inline void io_clean_op(struct io_kiocb *req)
1068{
Pavel Begunkov9d5c8192021-01-24 15:08:14 +00001069 if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED))
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001070 __io_clean_op(req);
1071}
1072
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001073static inline void io_set_resource_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001074{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001075 struct io_ring_ctx *ctx = req->ctx;
1076
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001077 if (!req->fixed_rsrc_refs) {
1078 req->fixed_rsrc_refs = &ctx->file_data->node->refs;
1079 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001080 }
1081}
1082
Pavel Begunkov08d23632020-11-06 13:00:22 +00001083static bool io_match_task(struct io_kiocb *head,
1084 struct task_struct *task,
1085 struct files_struct *files)
1086{
1087 struct io_kiocb *req;
1088
Jens Axboe84965ff2021-01-23 15:51:11 -07001089 if (task && head->task != task) {
1090 /* in terms of cancelation, always match if req task is dead */
1091 if (head->task->flags & PF_EXITING)
1092 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001093 return false;
Jens Axboe84965ff2021-01-23 15:51:11 -07001094 }
Pavel Begunkov08d23632020-11-06 13:00:22 +00001095 if (!files)
1096 return true;
1097
1098 io_for_each_link(req, head) {
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +00001099 if (req->flags & REQ_F_INFLIGHT)
Jens Axboe02a13672021-01-23 15:49:31 -07001100 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001101 }
1102 return false;
1103}
1104
Jens Axboec40f6372020-06-25 15:39:59 -06001105static inline void req_set_fail_links(struct io_kiocb *req)
1106{
1107 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1108 req->flags |= REQ_F_FAIL_LINK;
1109}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001110
Jens Axboe2b188cc2019-01-07 10:46:33 -07001111static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1112{
1113 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1114
Jens Axboe0f158b42020-05-14 17:18:39 -06001115 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001116}
1117
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001118static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1119{
1120 return !req->timeout.off;
1121}
1122
Jens Axboe2b188cc2019-01-07 10:46:33 -07001123static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1124{
1125 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001126 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001127
1128 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1129 if (!ctx)
1130 return NULL;
1131
Jens Axboe78076bb2019-12-04 19:56:40 -07001132 /*
1133 * Use 5 bits less than the max cq entries, that should give us around
1134 * 32 entries per hash list if totally full and uniformly spread.
1135 */
1136 hash_bits = ilog2(p->cq_entries);
1137 hash_bits -= 5;
1138 if (hash_bits <= 0)
1139 hash_bits = 1;
1140 ctx->cancel_hash_bits = hash_bits;
1141 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1142 GFP_KERNEL);
1143 if (!ctx->cancel_hash)
1144 goto err;
1145 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1146
Roman Gushchin21482892019-05-07 10:01:48 -07001147 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001148 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1149 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001150
1151 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001152 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001153 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001154 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001155 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001156 init_completion(&ctx->ref_comp);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07001157 xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00001158 xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001159 mutex_init(&ctx->uring_lock);
1160 init_waitqueue_head(&ctx->wait);
1161 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001162 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001163 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001164 INIT_LIST_HEAD(&ctx->timeout_list);
Jens Axboefcb323c2019-10-24 12:39:47 -06001165 spin_lock_init(&ctx->inflight_lock);
1166 INIT_LIST_HEAD(&ctx->inflight_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001167 spin_lock_init(&ctx->rsrc_ref_lock);
1168 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001169 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1170 init_llist_head(&ctx->rsrc_put_llist);
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00001171 INIT_LIST_HEAD(&ctx->tctx_list);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001172 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001173 INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001174 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001175err:
Jens Axboe78076bb2019-12-04 19:56:40 -07001176 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001177 kfree(ctx);
1178 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001179}
1180
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001181static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001182{
Jens Axboe2bc99302020-07-09 09:43:27 -06001183 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1184 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001185
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001186 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001187 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001188 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001189
Bob Liu9d858b22019-11-13 18:06:25 +08001190 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001191}
1192
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001193static void io_req_track_inflight(struct io_kiocb *req)
1194{
1195 struct io_ring_ctx *ctx = req->ctx;
1196
1197 if (!(req->flags & REQ_F_INFLIGHT)) {
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001198 req->flags |= REQ_F_INFLIGHT;
1199
1200 spin_lock_irq(&ctx->inflight_lock);
1201 list_add(&req->inflight_entry, &ctx->inflight_list);
1202 spin_unlock_irq(&ctx->inflight_lock);
1203 }
1204}
1205
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001206static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001207{
Jens Axboed3656342019-12-18 09:50:26 -07001208 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001209 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001210
Jens Axboe003e8dc2021-03-06 09:22:27 -07001211 if (!req->work.creds)
1212 req->work.creds = get_current_cred();
1213
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001214 if (req->flags & REQ_F_FORCE_ASYNC)
1215 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1216
Jens Axboed3656342019-12-18 09:50:26 -07001217 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001218 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001219 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboe4b982bd2021-04-01 08:38:34 -06001220 } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) {
Jens Axboed3656342019-12-18 09:50:26 -07001221 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001222 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001223 }
Jens Axboe561fb042019-10-24 07:25:42 -06001224}
1225
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001226static void io_prep_async_link(struct io_kiocb *req)
1227{
1228 struct io_kiocb *cur;
1229
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001230 io_for_each_link(cur, req)
1231 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001232}
1233
Pavel Begunkovebf93662021-03-01 18:20:47 +00001234static void io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001235{
Jackie Liua197f662019-11-08 08:09:12 -07001236 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001237 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001238 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe561fb042019-10-24 07:25:42 -06001239
Jens Axboe3bfe6102021-02-16 14:15:30 -07001240 BUG_ON(!tctx);
1241 BUG_ON(!tctx->io_wq);
Jens Axboe561fb042019-10-24 07:25:42 -06001242
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001243 /* init ->work of the whole link before punting */
1244 io_prep_async_link(req);
Pavel Begunkovd07f1e8a2021-03-22 01:45:58 +00001245 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1246 &req->work, req->flags);
Pavel Begunkovebf93662021-03-01 18:20:47 +00001247 io_wq_enqueue(tctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001248 if (link)
1249 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001250}
1251
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001252static void io_kill_timeout(struct io_kiocb *req, int status)
Jens Axboe5262f562019-09-17 12:26:57 -06001253{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001254 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001255 int ret;
1256
Jens Axboee8c2bc12020-08-15 18:44:09 -07001257 ret = hrtimer_try_to_cancel(&io->timer);
Jens Axboe5262f562019-09-17 12:26:57 -06001258 if (ret != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001259 atomic_set(&req->ctx->cq_timeouts,
1260 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001261 list_del_init(&req->timeout.list);
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001262 io_cqring_fill_event(req, status);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001263 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001264 }
1265}
1266
Pavel Begunkov04518942020-05-26 20:34:05 +03001267static void __io_queue_deferred(struct io_ring_ctx *ctx)
1268{
1269 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001270 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1271 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001272
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001273 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001274 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001275 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001276 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001277 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001278 } while (!list_empty(&ctx->defer_list));
1279}
1280
Pavel Begunkov360428f2020-05-30 14:54:17 +03001281static void io_flush_timeouts(struct io_ring_ctx *ctx)
1282{
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001283 u32 seq;
1284
1285 if (list_empty(&ctx->timeout_list))
1286 return;
1287
1288 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
1289
1290 do {
1291 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001292 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001293 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001294
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001295 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001296 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001297
1298 /*
1299 * Since seq can easily wrap around over time, subtract
1300 * the last seq at which timeouts were flushed before comparing.
1301 * Assuming not more than 2^31-1 events have happened since,
1302 * these subtractions won't have wrapped, so we can check if
1303 * target is in [last_seq, current_seq] by comparing the two.
1304 */
1305 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1306 events_got = seq - ctx->cq_last_tm_flush;
1307 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001308 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001309
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001310 list_del_init(&req->timeout.list);
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001311 io_kill_timeout(req, 0);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001312 } while (!list_empty(&ctx->timeout_list));
1313
1314 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001315}
1316
Jens Axboede0617e2019-04-06 21:51:27 -06001317static void io_commit_cqring(struct io_ring_ctx *ctx)
1318{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001319 io_flush_timeouts(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001320
1321 /* order cqe stores with ring update */
1322 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001323
Pavel Begunkov04518942020-05-26 20:34:05 +03001324 if (unlikely(!list_empty(&ctx->defer_list)))
1325 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001326}
1327
Jens Axboe90554202020-09-03 12:12:41 -06001328static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1329{
1330 struct io_rings *r = ctx->rings;
1331
1332 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1333}
1334
Pavel Begunkov888aae22021-01-19 13:32:39 +00001335static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1336{
1337 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1338}
1339
Jens Axboe2b188cc2019-01-07 10:46:33 -07001340static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
1341{
Hristo Venev75b28af2019-08-26 17:23:46 +00001342 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001343 unsigned tail;
1344
Stefan Bühler115e12e2019-04-24 23:54:18 +02001345 /*
1346 * writes to the cq entry need to come after reading head; the
1347 * control dependency is enough as we're using WRITE_ONCE to
1348 * fill the cq entry
1349 */
Pavel Begunkov888aae22021-01-19 13:32:39 +00001350 if (__io_cqring_events(ctx) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001351 return NULL;
1352
Pavel Begunkov888aae22021-01-19 13:32:39 +00001353 tail = ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001354 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001355}
1356
Jens Axboef2842ab2020-01-08 11:04:00 -07001357static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1358{
Jens Axboef0b493e2020-02-01 21:30:11 -07001359 if (!ctx->cq_ev_fd)
1360 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001361 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1362 return false;
Jens Axboef2842ab2020-01-08 11:04:00 -07001363 if (!ctx->eventfd_async)
1364 return true;
Jens Axboeb41e9852020-02-17 09:52:41 -07001365 return io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001366}
1367
Jens Axboeb41e9852020-02-17 09:52:41 -07001368static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001369{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001370 /* see waitqueue_active() comment */
1371 smp_mb();
1372
Jens Axboe8c838782019-03-12 15:48:16 -06001373 if (waitqueue_active(&ctx->wait))
1374 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001375 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1376 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001377 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001378 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001379 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001380 wake_up_interruptible(&ctx->cq_wait);
1381 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1382 }
Jens Axboe8c838782019-03-12 15:48:16 -06001383}
1384
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001385static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1386{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001387 /* see waitqueue_active() comment */
1388 smp_mb();
1389
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001390 if (ctx->flags & IORING_SETUP_SQPOLL) {
1391 if (waitqueue_active(&ctx->wait))
1392 wake_up(&ctx->wait);
1393 }
1394 if (io_should_trigger_evfd(ctx))
1395 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001396 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001397 wake_up_interruptible(&ctx->cq_wait);
1398 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1399 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001400}
1401
Jens Axboec4a2ed72019-11-21 21:01:26 -07001402/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c503152021-01-04 20:36:36 +00001403static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1404 struct task_struct *tsk,
1405 struct files_struct *files)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001406{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001407 struct io_rings *rings = ctx->rings;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001408 struct io_kiocb *req, *tmp;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001409 struct io_uring_cqe *cqe;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001410 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001411 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001412 LIST_HEAD(list);
1413
Pavel Begunkove23de152020-12-17 00:24:37 +00001414 if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
1415 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001416
Jens Axboeb18032b2021-01-24 16:58:56 -07001417 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001418 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboee6c8aa92020-09-28 13:10:13 -06001419 list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00001420 if (!io_match_task(req, tsk, files))
Jens Axboee6c8aa92020-09-28 13:10:13 -06001421 continue;
1422
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001423 cqe = io_get_cqring(ctx);
1424 if (!cqe && !force)
1425 break;
1426
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001427 list_move(&req->compl.list, &list);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001428 if (cqe) {
1429 WRITE_ONCE(cqe->user_data, req->user_data);
1430 WRITE_ONCE(cqe->res, req->result);
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001431 WRITE_ONCE(cqe->flags, req->compl.cflags);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001432 } else {
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001433 ctx->cached_cq_overflow++;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001434 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001435 ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001436 }
Jens Axboeb18032b2021-01-24 16:58:56 -07001437 posted = true;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001438 }
1439
Pavel Begunkov09e88402020-12-17 00:24:38 +00001440 all_flushed = list_empty(&ctx->cq_overflow_list);
1441 if (all_flushed) {
1442 clear_bit(0, &ctx->sq_check_overflow);
1443 clear_bit(0, &ctx->cq_check_overflow);
1444 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1445 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001446
Jens Axboeb18032b2021-01-24 16:58:56 -07001447 if (posted)
1448 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001449 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001450 if (posted)
1451 io_cqring_ev_posted(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001452
1453 while (!list_empty(&list)) {
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001454 req = list_first_entry(&list, struct io_kiocb, compl.list);
1455 list_del(&req->compl.list);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001456 io_put_req(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001457 }
Jens Axboec4a2ed72019-11-21 21:01:26 -07001458
Pavel Begunkov09e88402020-12-17 00:24:38 +00001459 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001460}
1461
Jens Axboeca0a2652021-03-04 17:15:48 -07001462static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
Pavel Begunkov6c503152021-01-04 20:36:36 +00001463 struct task_struct *tsk,
1464 struct files_struct *files)
1465{
Jens Axboeca0a2652021-03-04 17:15:48 -07001466 bool ret = true;
1467
Pavel Begunkov6c503152021-01-04 20:36:36 +00001468 if (test_bit(0, &ctx->cq_check_overflow)) {
1469 /* iopoll syncs against uring_lock, not completion_lock */
1470 if (ctx->flags & IORING_SETUP_IOPOLL)
1471 mutex_lock(&ctx->uring_lock);
Jens Axboeca0a2652021-03-04 17:15:48 -07001472 ret = __io_cqring_overflow_flush(ctx, force, tsk, files);
Pavel Begunkov6c503152021-01-04 20:36:36 +00001473 if (ctx->flags & IORING_SETUP_IOPOLL)
1474 mutex_unlock(&ctx->uring_lock);
1475 }
Jens Axboeca0a2652021-03-04 17:15:48 -07001476
1477 return ret;
Pavel Begunkov6c503152021-01-04 20:36:36 +00001478}
1479
Jens Axboeabc54d62021-02-24 13:32:30 -07001480/*
1481 * Shamelessly stolen from the mm implementation of page reference checking,
1482 * see commit f958d7b528b1 for details.
1483 */
1484#define req_ref_zero_or_close_to_overflow(req) \
1485 ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u)
1486
Jens Axboede9b4cc2021-02-24 13:28:27 -07001487static inline bool req_ref_inc_not_zero(struct io_kiocb *req)
1488{
Jens Axboeabc54d62021-02-24 13:32:30 -07001489 return atomic_inc_not_zero(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001490}
1491
1492static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs)
1493{
Jens Axboeabc54d62021-02-24 13:32:30 -07001494 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1495 return atomic_sub_and_test(refs, &req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001496}
1497
1498static inline bool req_ref_put_and_test(struct io_kiocb *req)
1499{
Jens Axboeabc54d62021-02-24 13:32:30 -07001500 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1501 return atomic_dec_and_test(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001502}
1503
1504static inline void req_ref_put(struct io_kiocb *req)
1505{
Jens Axboeabc54d62021-02-24 13:32:30 -07001506 WARN_ON_ONCE(req_ref_put_and_test(req));
Jens Axboede9b4cc2021-02-24 13:28:27 -07001507}
1508
1509static inline void req_ref_get(struct io_kiocb *req)
1510{
Jens Axboeabc54d62021-02-24 13:32:30 -07001511 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1512 atomic_inc(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001513}
1514
Pavel Begunkov8c3f9cd2021-02-28 22:35:15 +00001515static void __io_cqring_fill_event(struct io_kiocb *req, long res,
1516 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001517{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001518 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001519 struct io_uring_cqe *cqe;
1520
Jens Axboe78e19bb2019-11-06 15:21:34 -07001521 trace_io_uring_complete(ctx, req->user_data, res);
Jens Axboe51c3ff62019-11-03 06:52:50 -07001522
Jens Axboe2b188cc2019-01-07 10:46:33 -07001523 /*
1524 * If we can't get a cq entry, userspace overflowed the
1525 * submission (by quite a lot). Increment the overflow count in
1526 * the ring.
1527 */
1528 cqe = io_get_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001529 if (likely(cqe)) {
Jens Axboe78e19bb2019-11-06 15:21:34 -07001530 WRITE_ONCE(cqe->user_data, req->user_data);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001531 WRITE_ONCE(cqe->res, res);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001532 WRITE_ONCE(cqe->flags, cflags);
Jens Axboefdaf0832020-10-30 09:37:30 -06001533 } else if (ctx->cq_overflow_flushed ||
1534 atomic_read(&req->task->io_uring->in_idle)) {
Jens Axboe0f212202020-09-13 13:09:39 -06001535 /*
1536 * If we're in ring overflow flush mode, or in task cancel mode,
1537 * then we cannot store the request for later flushing, we need
1538 * to drop it on the floor.
1539 */
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001540 ctx->cached_cq_overflow++;
1541 WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001542 } else {
Jens Axboead3eb2c2019-12-18 17:12:20 -07001543 if (list_empty(&ctx->cq_overflow_list)) {
1544 set_bit(0, &ctx->sq_check_overflow);
1545 set_bit(0, &ctx->cq_check_overflow);
Xiaoguang Wang6d5f9042020-07-09 09:15:29 +08001546 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
Jens Axboead3eb2c2019-12-18 17:12:20 -07001547 }
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001548 io_clean_op(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001549 req->result = res;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001550 req->compl.cflags = cflags;
Jens Axboede9b4cc2021-02-24 13:28:27 -07001551 req_ref_get(req);
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001552 list_add_tail(&req->compl.list, &ctx->cq_overflow_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001553 }
1554}
1555
Jens Axboebcda7ba2020-02-23 16:42:51 -07001556static void io_cqring_fill_event(struct io_kiocb *req, long res)
1557{
1558 __io_cqring_fill_event(req, res, 0);
1559}
1560
Pavel Begunkov7a612352021-03-09 00:37:59 +00001561static void io_req_complete_post(struct io_kiocb *req, long res,
1562 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001563{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001564 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001565 unsigned long flags;
1566
1567 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001568 __io_cqring_fill_event(req, res, cflags);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001569 /*
1570 * If we're the last reference to this request, add to our locked
1571 * free_list cache.
1572 */
Jens Axboede9b4cc2021-02-24 13:28:27 -07001573 if (req_ref_put_and_test(req)) {
Jens Axboec7dae4b2021-02-09 19:53:37 -07001574 struct io_comp_state *cs = &ctx->submit_state.comp;
1575
Pavel Begunkov7a612352021-03-09 00:37:59 +00001576 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
1577 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK))
1578 io_disarm_next(req);
1579 if (req->link) {
1580 io_req_task_queue(req->link);
1581 req->link = NULL;
1582 }
1583 }
Jens Axboec7dae4b2021-02-09 19:53:37 -07001584 io_dismantle_req(req);
1585 io_put_task(req->task, 1);
1586 list_add(&req->compl.list, &cs->locked_free_list);
1587 cs->locked_free_nr++;
Pavel Begunkov180f8292021-03-14 20:57:09 +00001588 } else {
1589 if (!percpu_ref_tryget(&ctx->refs))
1590 req = NULL;
1591 }
Pavel Begunkov7a612352021-03-09 00:37:59 +00001592 io_commit_cqring(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001593 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7a612352021-03-09 00:37:59 +00001594
Pavel Begunkov180f8292021-03-14 20:57:09 +00001595 if (req) {
1596 io_cqring_ev_posted(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001597 percpu_ref_put(&ctx->refs);
Pavel Begunkov180f8292021-03-14 20:57:09 +00001598 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001599}
1600
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001601static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001602 unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001603{
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001604 io_clean_op(req);
1605 req->result = res;
1606 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001607 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001608}
Jens Axboe2b188cc2019-01-07 10:46:33 -07001609
Pavel Begunkov889fca72021-02-10 00:03:09 +00001610static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1611 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001612{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001613 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1614 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001615 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001616 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001617}
Jens Axboebcda7ba2020-02-23 16:42:51 -07001618
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001619static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001620{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001621 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001622}
1623
Pavel Begunkovf41db2732021-02-28 22:35:12 +00001624static void io_req_complete_failed(struct io_kiocb *req, long res)
1625{
1626 req_set_fail_links(req);
1627 io_put_req(req);
1628 io_req_complete_post(req, res, 0);
1629}
1630
Jens Axboec7dae4b2021-02-09 19:53:37 -07001631static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001632{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001633 struct io_submit_state *state = &ctx->submit_state;
1634 struct io_comp_state *cs = &state->comp;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001635 struct io_kiocb *req = NULL;
1636
Jens Axboec7dae4b2021-02-09 19:53:37 -07001637 /*
1638 * If we have more than a batch's worth of requests in our IRQ side
1639 * locked cache, grab the lock and move them over to our submission
1640 * side cache.
1641 */
1642 if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) {
1643 spin_lock_irq(&ctx->completion_lock);
1644 list_splice_init(&cs->locked_free_list, &cs->free_list);
1645 cs->locked_free_nr = 0;
1646 spin_unlock_irq(&ctx->completion_lock);
1647 }
1648
1649 while (!list_empty(&cs->free_list)) {
1650 req = list_first_entry(&cs->free_list, struct io_kiocb,
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001651 compl.list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001652 list_del(&req->compl.list);
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001653 state->reqs[state->free_reqs++] = req;
1654 if (state->free_reqs == ARRAY_SIZE(state->reqs))
1655 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001656 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001657
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001658 return req != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001659}
1660
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001661static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001662{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001663 struct io_submit_state *state = &ctx->submit_state;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001664
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001665 BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs));
Jens Axboe2b188cc2019-01-07 10:46:33 -07001666
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001667 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001668 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07001669 int ret;
1670
Jens Axboec7dae4b2021-02-09 19:53:37 -07001671 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001672 goto got_req;
1673
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001674 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
1675 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06001676
1677 /*
1678 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1679 * retry single alloc to be on the safe side.
1680 */
1681 if (unlikely(ret <= 0)) {
1682 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1683 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00001684 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06001685 ret = 1;
1686 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001687 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001688 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001689got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03001690 state->free_reqs--;
1691 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001692}
1693
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001694static inline void io_put_file(struct io_kiocb *req, struct file *file,
1695 bool fixed)
1696{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001697 if (!fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001698 fput(file);
1699}
1700
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001701static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001702{
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001703 io_clean_op(req);
Pavel Begunkov929a3af2020-02-19 00:19:09 +03001704
Jens Axboee8c2bc12020-08-15 18:44:09 -07001705 if (req->async_data)
1706 kfree(req->async_data);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001707 if (req->file)
1708 io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001709 if (req->fixed_rsrc_refs)
1710 percpu_ref_put(req->fixed_rsrc_refs);
Jens Axboe003e8dc2021-03-06 09:22:27 -07001711 if (req->work.creds) {
1712 put_cred(req->work.creds);
1713 req->work.creds = NULL;
1714 }
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001715
1716 if (req->flags & REQ_F_INFLIGHT) {
1717 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001718 unsigned long flags;
1719
1720 spin_lock_irqsave(&ctx->inflight_lock, flags);
1721 list_del(&req->inflight_entry);
1722 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1723 req->flags &= ~REQ_F_INFLIGHT;
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001724 }
Pavel Begunkove6543a82020-06-28 12:52:30 +03001725}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001726
Pavel Begunkovb23fcf42021-03-01 18:20:48 +00001727/* must to be called somewhat shortly after putting a request */
Pavel Begunkov7c660732021-01-25 11:42:21 +00001728static inline void io_put_task(struct task_struct *task, int nr)
1729{
1730 struct io_uring_task *tctx = task->io_uring;
1731
1732 percpu_counter_sub(&tctx->inflight, nr);
1733 if (unlikely(atomic_read(&tctx->in_idle)))
1734 wake_up(&tctx->wait);
1735 put_task_struct_many(task, nr);
1736}
1737
Pavel Begunkov216578e2020-10-13 09:44:00 +01001738static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001739{
Jens Axboe51a4cc12020-08-10 10:55:56 -06001740 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001741
Pavel Begunkov216578e2020-10-13 09:44:00 +01001742 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00001743 io_put_task(req->task, 1);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001744
Pavel Begunkov3893f392021-02-10 00:03:15 +00001745 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001746 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001747}
1748
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001749static inline void io_remove_next_linked(struct io_kiocb *req)
1750{
1751 struct io_kiocb *nxt = req->link;
1752
1753 req->link = nxt->link;
1754 nxt->link = NULL;
1755}
1756
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001757static bool io_kill_linked_timeout(struct io_kiocb *req)
1758 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001759{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001760 struct io_kiocb *link = req->link;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001761 bool cancelled = false;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001762
Pavel Begunkov900fad42020-10-19 16:39:16 +01001763 /*
1764 * Can happen if a linked timeout fired and link had been like
1765 * req -> link t-out -> link t-out [-> ...]
1766 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001767 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1768 struct io_timeout_data *io = link->async_data;
1769 int ret;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001770
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001771 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00001772 link->timeout.head = NULL;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001773 ret = hrtimer_try_to_cancel(&io->timer);
1774 if (ret != -1) {
1775 io_cqring_fill_event(link, -ECANCELED);
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001776 io_put_req_deferred(link, 1);
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001777 cancelled = true;
1778 }
1779 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001780 req->flags &= ~REQ_F_LINK_TIMEOUT;
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001781 return cancelled;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001782}
1783
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001784static void io_fail_links(struct io_kiocb *req)
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001785 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001786{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001787 struct io_kiocb *nxt, *link = req->link;
Jens Axboe9e645e112019-05-10 16:07:28 -06001788
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001789 req->link = NULL;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001790 while (link) {
1791 nxt = link->link;
1792 link->link = NULL;
1793
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001794 trace_io_uring_fail_link(req, link);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001795 io_cqring_fill_event(link, -ECANCELED);
Jens Axboe1575f212021-02-27 15:20:49 -07001796 io_put_req_deferred(link, 2);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001797 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001798 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001799}
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001800
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001801static bool io_disarm_next(struct io_kiocb *req)
1802 __must_hold(&req->ctx->completion_lock)
1803{
1804 bool posted = false;
1805
1806 if (likely(req->flags & REQ_F_LINK_TIMEOUT))
1807 posted = io_kill_linked_timeout(req);
1808 if (unlikely(req->flags & REQ_F_FAIL_LINK)) {
1809 posted |= (req->link != NULL);
1810 io_fail_links(req);
1811 }
1812 return posted;
Jens Axboe9e645e112019-05-10 16:07:28 -06001813}
1814
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001815static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001816{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001817 struct io_kiocb *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07001818
Jens Axboe9e645e112019-05-10 16:07:28 -06001819 /*
1820 * If LINK is set, we have dependent requests in this chain. If we
1821 * didn't fail this request, queue the first one up, moving any other
1822 * dependencies to the next request. In case of failure, fail the rest
1823 * of the chain.
1824 */
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001825 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) {
1826 struct io_ring_ctx *ctx = req->ctx;
1827 unsigned long flags;
1828 bool posted;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001829
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001830 spin_lock_irqsave(&ctx->completion_lock, flags);
1831 posted = io_disarm_next(req);
1832 if (posted)
1833 io_commit_cqring(req->ctx);
1834 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1835 if (posted)
1836 io_cqring_ev_posted(ctx);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001837 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001838 nxt = req->link;
1839 req->link = NULL;
1840 return nxt;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001841}
Jens Axboe2665abf2019-11-05 12:40:47 -07001842
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001843static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001844{
Pavel Begunkovcdbff982021-02-12 18:41:16 +00001845 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK))))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001846 return NULL;
1847 return __io_req_find_next(req);
1848}
1849
Pavel Begunkov2c323952021-02-28 22:04:53 +00001850static void ctx_flush_and_put(struct io_ring_ctx *ctx)
1851{
1852 if (!ctx)
1853 return;
1854 if (ctx->submit_state.comp.nr) {
1855 mutex_lock(&ctx->uring_lock);
1856 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
1857 mutex_unlock(&ctx->uring_lock);
1858 }
1859 percpu_ref_put(&ctx->refs);
1860}
1861
Jens Axboe7cbf1722021-02-10 00:03:20 +00001862static bool __tctx_task_work(struct io_uring_task *tctx)
1863{
Jens Axboe65453d12021-02-10 00:03:21 +00001864 struct io_ring_ctx *ctx = NULL;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001865 struct io_wq_work_list list;
1866 struct io_wq_work_node *node;
1867
1868 if (wq_list_empty(&tctx->task_list))
1869 return false;
1870
Jens Axboe0b81e802021-02-16 10:33:53 -07001871 spin_lock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001872 list = tctx->task_list;
1873 INIT_WQ_LIST(&tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001874 spin_unlock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001875
1876 node = list.first;
1877 while (node) {
1878 struct io_wq_work_node *next = node->next;
1879 struct io_kiocb *req;
1880
1881 req = container_of(node, struct io_kiocb, io_task_work.node);
Pavel Begunkov2c323952021-02-28 22:04:53 +00001882 if (req->ctx != ctx) {
1883 ctx_flush_and_put(ctx);
1884 ctx = req->ctx;
1885 percpu_ref_get(&ctx->refs);
1886 }
1887
Jens Axboe7cbf1722021-02-10 00:03:20 +00001888 req->task_work.func(&req->task_work);
1889 node = next;
Jens Axboe65453d12021-02-10 00:03:21 +00001890 }
1891
Pavel Begunkov2c323952021-02-28 22:04:53 +00001892 ctx_flush_and_put(ctx);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001893 return list.first != NULL;
1894}
1895
1896static void tctx_task_work(struct callback_head *cb)
1897{
1898 struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work);
1899
Jens Axboe1d5f3602021-02-26 14:54:16 -07001900 clear_bit(0, &tctx->task_state);
1901
Jens Axboe7cbf1722021-02-10 00:03:20 +00001902 while (__tctx_task_work(tctx))
1903 cond_resched();
Jens Axboe7cbf1722021-02-10 00:03:20 +00001904}
1905
1906static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req,
1907 enum task_work_notify_mode notify)
1908{
1909 struct io_uring_task *tctx = tsk->io_uring;
1910 struct io_wq_work_node *node, *prev;
Jens Axboe0b81e802021-02-16 10:33:53 -07001911 unsigned long flags;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001912 int ret;
1913
1914 WARN_ON_ONCE(!tctx);
1915
Jens Axboe0b81e802021-02-16 10:33:53 -07001916 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001917 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001918 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001919
1920 /* task_work already pending, we're done */
1921 if (test_bit(0, &tctx->task_state) ||
1922 test_and_set_bit(0, &tctx->task_state))
1923 return 0;
1924
1925 if (!task_work_add(tsk, &tctx->task_work, notify))
1926 return 0;
1927
1928 /*
1929 * Slow path - we failed, find and delete work. if the work is not
1930 * in the list, it got run and we're fine.
1931 */
1932 ret = 0;
Jens Axboe0b81e802021-02-16 10:33:53 -07001933 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001934 wq_list_for_each(node, prev, &tctx->task_list) {
1935 if (&req->io_task_work.node == node) {
1936 wq_list_del(&tctx->task_list, node, prev);
1937 ret = 1;
1938 break;
1939 }
1940 }
Jens Axboe0b81e802021-02-16 10:33:53 -07001941 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001942 clear_bit(0, &tctx->task_state);
1943 return ret;
1944}
1945
Jens Axboe355fb9e2020-10-22 20:19:35 -06001946static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboec2c4c832020-07-01 15:37:11 -06001947{
1948 struct task_struct *tsk = req->task;
1949 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe91989c72020-10-16 09:02:26 -06001950 enum task_work_notify_mode notify;
1951 int ret;
Jens Axboec2c4c832020-07-01 15:37:11 -06001952
Jens Axboe6200b0a2020-09-13 14:38:30 -06001953 if (tsk->flags & PF_EXITING)
1954 return -ESRCH;
1955
Jens Axboec2c4c832020-07-01 15:37:11 -06001956 /*
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001957 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
1958 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
1959 * processing task_work. There's no reliable way to tell if TWA_RESUME
1960 * will do the job.
Jens Axboec2c4c832020-07-01 15:37:11 -06001961 */
Jens Axboe91989c72020-10-16 09:02:26 -06001962 notify = TWA_NONE;
Jens Axboe355fb9e2020-10-22 20:19:35 -06001963 if (!(ctx->flags & IORING_SETUP_SQPOLL))
Jens Axboec2c4c832020-07-01 15:37:11 -06001964 notify = TWA_SIGNAL;
1965
Jens Axboe7cbf1722021-02-10 00:03:20 +00001966 ret = io_task_work_add(tsk, req, notify);
Jens Axboec2c4c832020-07-01 15:37:11 -06001967 if (!ret)
1968 wake_up_process(tsk);
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001969
Jens Axboec2c4c832020-07-01 15:37:11 -06001970 return ret;
1971}
1972
Pavel Begunkov9b465712021-03-15 14:23:07 +00001973static bool io_run_task_work_head(struct callback_head **work_head)
1974{
1975 struct callback_head *work, *next;
1976 bool executed = false;
1977
1978 do {
1979 work = xchg(work_head, NULL);
1980 if (!work)
1981 break;
1982
1983 do {
1984 next = work->next;
1985 work->func(work);
1986 work = next;
1987 cond_resched();
1988 } while (work);
1989 executed = true;
1990 } while (1);
1991
1992 return executed;
1993}
1994
1995static void io_task_work_add_head(struct callback_head **work_head,
1996 struct callback_head *task_work)
1997{
1998 struct callback_head *head;
1999
2000 do {
2001 head = READ_ONCE(*work_head);
2002 task_work->next = head;
2003 } while (cmpxchg(work_head, head, task_work) != head);
2004}
2005
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002006static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00002007 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002008{
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002009 init_task_work(&req->task_work, cb);
Pavel Begunkov9b465712021-03-15 14:23:07 +00002010 io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002011}
2012
Jens Axboec40f6372020-06-25 15:39:59 -06002013static void __io_req_task_cancel(struct io_kiocb *req, int error)
2014{
2015 struct io_ring_ctx *ctx = req->ctx;
2016
2017 spin_lock_irq(&ctx->completion_lock);
2018 io_cqring_fill_event(req, error);
2019 io_commit_cqring(ctx);
2020 spin_unlock_irq(&ctx->completion_lock);
2021
2022 io_cqring_ev_posted(ctx);
2023 req_set_fail_links(req);
2024 io_double_put_req(req);
2025}
2026
2027static void io_req_task_cancel(struct callback_head *cb)
2028{
2029 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002030 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002031
Pavel Begunkove83acd72021-02-28 22:35:09 +00002032 /* ctx is guaranteed to stay alive while we hold uring_lock */
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002033 mutex_lock(&ctx->uring_lock);
Pavel Begunkova3df76982021-02-18 22:32:52 +00002034 __io_req_task_cancel(req, req->result);
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002035 mutex_unlock(&ctx->uring_lock);
Jens Axboec40f6372020-06-25 15:39:59 -06002036}
2037
2038static void __io_req_task_submit(struct io_kiocb *req)
2039{
2040 struct io_ring_ctx *ctx = req->ctx;
2041
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00002042 /* ctx stays valid until unlock, even if we drop all ours ctx->refs */
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002043 mutex_lock(&ctx->uring_lock);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00002044 if (!(current->flags & PF_EXITING) && !current->in_execve)
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00002045 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002046 else
Jens Axboec40f6372020-06-25 15:39:59 -06002047 __io_req_task_cancel(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002048 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06002049}
2050
Jens Axboec40f6372020-06-25 15:39:59 -06002051static void io_req_task_submit(struct callback_head *cb)
2052{
2053 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2054
2055 __io_req_task_submit(req);
2056}
2057
Pavel Begunkova3df76982021-02-18 22:32:52 +00002058static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
2059{
Pavel Begunkova3df76982021-02-18 22:32:52 +00002060 req->result = ret;
2061 req->task_work.func = io_req_task_cancel;
2062
2063 if (unlikely(io_req_task_work_add(req)))
2064 io_req_task_work_add_fallback(req, io_req_task_cancel);
2065}
2066
Pavel Begunkov2c4b8eb2021-02-28 22:35:10 +00002067static void io_req_task_queue(struct io_kiocb *req)
2068{
2069 req->task_work.func = io_req_task_submit;
2070
2071 if (unlikely(io_req_task_work_add(req)))
2072 io_req_task_queue_fail(req, -ECANCELED);
2073}
2074
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002075static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002076{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002077 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002078
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002079 if (nxt)
2080 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002081}
2082
Jens Axboe9e645e112019-05-10 16:07:28 -06002083static void io_free_req(struct io_kiocb *req)
2084{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002085 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002086 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002087}
2088
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002089struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002090 struct task_struct *task;
2091 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002092 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002093};
2094
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002095static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002096{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002097 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002098 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002099 rb->task = NULL;
2100}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002101
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002102static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2103 struct req_batch *rb)
2104{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002105 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002106 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002107 if (rb->ctx_refs)
2108 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002109}
2110
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002111static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2112 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002113{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002114 io_queue_next(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002115
Jens Axboee3bc8e92020-09-24 08:45:57 -06002116 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002117 if (rb->task)
2118 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002119 rb->task = req->task;
2120 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002121 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002122 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002123 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002124
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002125 io_dismantle_req(req);
Pavel Begunkovbd759042021-02-12 03:23:50 +00002126 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002127 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002128 else
2129 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002130}
2131
Pavel Begunkov905c1722021-02-10 00:03:14 +00002132static void io_submit_flush_completions(struct io_comp_state *cs,
2133 struct io_ring_ctx *ctx)
2134{
2135 int i, nr = cs->nr;
2136 struct io_kiocb *req;
2137 struct req_batch rb;
2138
2139 io_init_req_batch(&rb);
2140 spin_lock_irq(&ctx->completion_lock);
2141 for (i = 0; i < nr; i++) {
2142 req = cs->reqs[i];
2143 __io_cqring_fill_event(req, req->result, req->compl.cflags);
2144 }
2145 io_commit_cqring(ctx);
2146 spin_unlock_irq(&ctx->completion_lock);
2147
2148 io_cqring_ev_posted(ctx);
2149 for (i = 0; i < nr; i++) {
2150 req = cs->reqs[i];
2151
2152 /* submission and completion refs */
Jens Axboede9b4cc2021-02-24 13:28:27 -07002153 if (req_ref_sub_and_test(req, 2))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002154 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002155 }
2156
2157 io_req_free_batch_finish(ctx, &rb);
2158 cs->nr = 0;
Jens Axboee65ef562019-03-12 10:16:44 -06002159}
2160
Jens Axboeba816ad2019-09-28 11:36:45 -06002161/*
2162 * Drop reference to request, return next in chain (if there is one) if this
2163 * was the last reference to this request.
2164 */
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002165static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002166{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002167 struct io_kiocb *nxt = NULL;
2168
Jens Axboede9b4cc2021-02-24 13:28:27 -07002169 if (req_ref_put_and_test(req)) {
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002170 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002171 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002172 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002173 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002174}
2175
Jens Axboe2b188cc2019-01-07 10:46:33 -07002176static void io_put_req(struct io_kiocb *req)
2177{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002178 if (req_ref_put_and_test(req))
Jens Axboedef596e2019-01-09 08:59:42 -07002179 io_free_req(req);
2180}
2181
Pavel Begunkov216578e2020-10-13 09:44:00 +01002182static void io_put_req_deferred_cb(struct callback_head *cb)
2183{
2184 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2185
2186 io_free_req(req);
2187}
2188
2189static void io_free_req_deferred(struct io_kiocb *req)
2190{
2191 int ret;
2192
Jens Axboe7cbf1722021-02-10 00:03:20 +00002193 req->task_work.func = io_put_req_deferred_cb;
Jens Axboe355fb9e2020-10-22 20:19:35 -06002194 ret = io_req_task_work_add(req);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002195 if (unlikely(ret))
2196 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002197}
2198
2199static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2200{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002201 if (req_ref_sub_and_test(req, refs))
Pavel Begunkov216578e2020-10-13 09:44:00 +01002202 io_free_req_deferred(req);
2203}
2204
Jens Axboe978db572019-11-14 22:39:04 -07002205static void io_double_put_req(struct io_kiocb *req)
2206{
2207 /* drop both submit and complete references */
Jens Axboede9b4cc2021-02-24 13:28:27 -07002208 if (req_ref_sub_and_test(req, 2))
Jens Axboe978db572019-11-14 22:39:04 -07002209 io_free_req(req);
2210}
2211
Pavel Begunkov6c503152021-01-04 20:36:36 +00002212static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002213{
2214 /* See comment at the top of this file */
2215 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002216 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002217}
2218
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002219static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2220{
2221 struct io_rings *rings = ctx->rings;
2222
2223 /* make sure SQ entry isn't read before tail */
2224 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2225}
2226
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002227static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002228{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002229 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002230
Jens Axboebcda7ba2020-02-23 16:42:51 -07002231 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2232 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002233 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002234 kfree(kbuf);
2235 return cflags;
2236}
2237
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002238static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2239{
2240 struct io_buffer *kbuf;
2241
2242 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2243 return io_put_kbuf(req, kbuf);
2244}
2245
Jens Axboe4c6e2772020-07-01 11:29:10 -06002246static inline bool io_run_task_work(void)
2247{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002248 /*
2249 * Not safe to run on exiting task, and the task_work handling will
2250 * not add work to such a task.
2251 */
2252 if (unlikely(current->flags & PF_EXITING))
2253 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002254 if (current->task_works) {
2255 __set_current_state(TASK_RUNNING);
2256 task_work_run();
2257 return true;
2258 }
2259
2260 return false;
2261}
2262
Jens Axboedef596e2019-01-09 08:59:42 -07002263/*
2264 * Find and free completed poll iocbs
2265 */
2266static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2267 struct list_head *done)
2268{
Jens Axboe8237e042019-12-28 10:48:22 -07002269 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002270 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002271
2272 /* order with ->result store in io_complete_rw_iopoll() */
2273 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002274
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002275 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002276 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002277 int cflags = 0;
2278
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002279 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002280 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002281
Pavel Begunkovf1613402021-02-11 18:28:21 +00002282 if (READ_ONCE(req->result) == -EAGAIN) {
2283 req->iopoll_completed = 0;
Pavel Begunkov23faba32021-02-11 18:28:22 +00002284 if (io_rw_reissue(req))
Pavel Begunkovf1613402021-02-11 18:28:21 +00002285 continue;
2286 }
2287
Jens Axboebcda7ba2020-02-23 16:42:51 -07002288 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002289 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002290
2291 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002292 (*nr_events)++;
2293
Jens Axboede9b4cc2021-02-24 13:28:27 -07002294 if (req_ref_put_and_test(req))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002295 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002296 }
Jens Axboedef596e2019-01-09 08:59:42 -07002297
Jens Axboe09bb8392019-03-13 12:39:28 -06002298 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002299 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002300 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002301}
2302
Jens Axboedef596e2019-01-09 08:59:42 -07002303static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2304 long min)
2305{
2306 struct io_kiocb *req, *tmp;
2307 LIST_HEAD(done);
2308 bool spin;
2309 int ret;
2310
2311 /*
2312 * Only spin for completions if we don't have multiple devices hanging
2313 * off our complete list, and we're under the requested amount.
2314 */
2315 spin = !ctx->poll_multi_file && *nr_events < min;
2316
2317 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002318 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002319 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002320
2321 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002322 * Move completed and retryable entries to our local lists.
2323 * If we find a request that requires polling, break out
2324 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002325 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002326 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002327 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002328 continue;
2329 }
2330 if (!list_empty(&done))
2331 break;
2332
2333 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2334 if (ret < 0)
2335 break;
2336
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002337 /* iopoll may have completed current req */
2338 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002339 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002340
Jens Axboedef596e2019-01-09 08:59:42 -07002341 if (ret && spin)
2342 spin = false;
2343 ret = 0;
2344 }
2345
2346 if (!list_empty(&done))
2347 io_iopoll_complete(ctx, nr_events, &done);
2348
2349 return ret;
2350}
2351
2352/*
Brian Gianforcarod195a662019-12-13 03:09:50 -08002353 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
Jens Axboedef596e2019-01-09 08:59:42 -07002354 * non-spinning poll check - we'll still enter the driver poll loop, but only
2355 * as a non-spinning completion check.
2356 */
2357static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
2358 long min)
2359{
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002360 while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
Jens Axboedef596e2019-01-09 08:59:42 -07002361 int ret;
2362
2363 ret = io_do_iopoll(ctx, nr_events, min);
2364 if (ret < 0)
2365 return ret;
Pavel Begunkoveba0a4d2020-07-06 17:59:30 +03002366 if (*nr_events >= min)
Jens Axboedef596e2019-01-09 08:59:42 -07002367 return 0;
2368 }
2369
2370 return 1;
2371}
2372
2373/*
2374 * We can't just wait for polled events to come to us, we have to actively
2375 * find and complete them.
2376 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002377static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002378{
2379 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2380 return;
2381
2382 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002383 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002384 unsigned int nr_events = 0;
2385
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002386 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002387
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002388 /* let it sleep and repeat later if can't complete a request */
2389 if (nr_events == 0)
2390 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002391 /*
2392 * Ensure we allow local-to-the-cpu processing to take place,
2393 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002394 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002395 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002396 if (need_resched()) {
2397 mutex_unlock(&ctx->uring_lock);
2398 cond_resched();
2399 mutex_lock(&ctx->uring_lock);
2400 }
Jens Axboedef596e2019-01-09 08:59:42 -07002401 }
2402 mutex_unlock(&ctx->uring_lock);
2403}
2404
Pavel Begunkov7668b922020-07-07 16:36:21 +03002405static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002406{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002407 unsigned int nr_events = 0;
Jens Axboe2b2ed972019-10-25 10:06:15 -06002408 int iters = 0, ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002409
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002410 /*
2411 * We disallow the app entering submit/complete with polling, but we
2412 * still need to lock the ring to prevent racing with polled issue
2413 * that got punted to a workqueue.
2414 */
2415 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002416 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002417 /*
Jens Axboea3a0e432019-08-20 11:03:11 -06002418 * Don't enter poll loop if we already have events pending.
2419 * If we do, we can potentially be spinning for commands that
2420 * already triggered a CQE (eg in error).
2421 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00002422 if (test_bit(0, &ctx->cq_check_overflow))
2423 __io_cqring_overflow_flush(ctx, false, NULL, NULL);
2424 if (io_cqring_events(ctx))
Jens Axboea3a0e432019-08-20 11:03:11 -06002425 break;
2426
2427 /*
Jens Axboe500f9fb2019-08-19 12:15:59 -06002428 * If a submit got punted to a workqueue, we can have the
2429 * application entering polling for a command before it gets
2430 * issued. That app will hold the uring_lock for the duration
2431 * of the poll right here, so we need to take a breather every
2432 * now and then to ensure that the issue has a chance to add
2433 * the poll to the issued list. Otherwise we can spin here
2434 * forever, while the workqueue is stuck trying to acquire the
2435 * very same mutex.
2436 */
2437 if (!(++iters & 7)) {
2438 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002439 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002440 mutex_lock(&ctx->uring_lock);
2441 }
2442
Pavel Begunkov7668b922020-07-07 16:36:21 +03002443 ret = io_iopoll_getevents(ctx, &nr_events, min);
Jens Axboedef596e2019-01-09 08:59:42 -07002444 if (ret <= 0)
2445 break;
2446 ret = 0;
Pavel Begunkov7668b922020-07-07 16:36:21 +03002447 } while (min && !nr_events && !need_resched());
Jens Axboedef596e2019-01-09 08:59:42 -07002448
Jens Axboe500f9fb2019-08-19 12:15:59 -06002449 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002450 return ret;
2451}
2452
Jens Axboe491381ce2019-10-17 09:20:46 -06002453static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002454{
Jens Axboe491381ce2019-10-17 09:20:46 -06002455 /*
2456 * Tell lockdep we inherited freeze protection from submission
2457 * thread.
2458 */
2459 if (req->flags & REQ_F_ISREG) {
2460 struct inode *inode = file_inode(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002461
Jens Axboe491381ce2019-10-17 09:20:46 -06002462 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002463 }
Jens Axboe491381ce2019-10-17 09:20:46 -06002464 file_end_write(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002465}
2466
Jens Axboeb63534c2020-06-04 11:28:00 -06002467#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002468static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002469{
Pavel Begunkov179ae0d2021-02-28 22:35:20 +00002470 /* either already prepared or successfully done */
2471 return req->async_data || !io_req_prep_async(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002472}
Jens Axboeb63534c2020-06-04 11:28:00 -06002473
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002474static bool io_rw_should_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002475{
Jens Axboe355afae2020-09-02 09:30:31 -06002476 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002477 struct io_ring_ctx *ctx = req->ctx;
Jens Axboeb63534c2020-06-04 11:28:00 -06002478
Jens Axboe355afae2020-09-02 09:30:31 -06002479 if (!S_ISBLK(mode) && !S_ISREG(mode))
2480 return false;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002481 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() &&
2482 !(ctx->flags & IORING_SETUP_IOPOLL)))
Jens Axboeb63534c2020-06-04 11:28:00 -06002483 return false;
Jens Axboe7c977a52021-02-23 19:17:35 -07002484 /*
2485 * If ref is dying, we might be running poll reap from the exit work.
2486 * Don't attempt to reissue from that path, just let it fail with
2487 * -EAGAIN.
2488 */
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002489 if (percpu_ref_is_dying(&ctx->refs))
2490 return false;
2491 return true;
2492}
Jens Axboee82ad482021-04-02 19:45:34 -06002493#else
2494static bool io_rw_should_reissue(struct io_kiocb *req)
2495{
2496 return false;
2497}
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002498#endif
2499
2500static bool io_rw_reissue(struct io_kiocb *req)
2501{
2502#ifdef CONFIG_BLOCK
2503 if (!io_rw_should_reissue(req))
Jens Axboe7c977a52021-02-23 19:17:35 -07002504 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002505
Pavel Begunkov55e6ac12021-01-08 20:57:22 +00002506 lockdep_assert_held(&req->ctx->uring_lock);
2507
Jens Axboe37d1e2e2021-02-17 21:03:43 -07002508 if (io_resubmit_prep(req)) {
Jens Axboede9b4cc2021-02-24 13:28:27 -07002509 req_ref_get(req);
Jens Axboefdee9462020-08-27 16:46:24 -06002510 io_queue_async_work(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002511 return true;
Jens Axboefdee9462020-08-27 16:46:24 -06002512 }
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002513 req_set_fail_links(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002514#endif
2515 return false;
2516}
2517
Jens Axboea1d7c392020-06-22 11:09:46 -06002518static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002519 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002520{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002521 int cflags = 0;
2522
Pavel Begunkovb65c1282021-03-22 01:45:59 +00002523 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2524 kiocb_end_write(req);
Jens Axboe230d50d2021-04-01 20:41:15 -06002525 if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_should_reissue(req)) {
2526 req->flags |= REQ_F_REISSUE;
Pavel Begunkov23faba32021-02-11 18:28:22 +00002527 return;
Jens Axboe230d50d2021-04-01 20:41:15 -06002528 }
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002529 if (res != req->result)
2530 req_set_fail_links(req);
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002531 if (req->flags & REQ_F_BUFFER_SELECTED)
2532 cflags = io_put_rw_kbuf(req);
2533 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002534}
2535
2536static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2537{
Jens Axboe9adbd452019-12-20 08:45:55 -07002538 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002539
Pavel Begunkov889fca72021-02-10 00:03:09 +00002540 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002541}
2542
Jens Axboedef596e2019-01-09 08:59:42 -07002543static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2544{
Jens Axboe9adbd452019-12-20 08:45:55 -07002545 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002546
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002547#ifdef CONFIG_BLOCK
2548 /* Rewind iter, if we have one. iopoll path resubmits as usual */
2549 if (res == -EAGAIN && io_rw_should_reissue(req)) {
2550 struct io_async_rw *rw = req->async_data;
2551
2552 if (rw)
2553 iov_iter_revert(&rw->iter,
2554 req->result - iov_iter_count(&rw->iter));
2555 else if (!io_resubmit_prep(req))
2556 res = -EIO;
2557 }
2558#endif
2559
Jens Axboe491381ce2019-10-17 09:20:46 -06002560 if (kiocb->ki_flags & IOCB_WRITE)
2561 kiocb_end_write(req);
Jens Axboedef596e2019-01-09 08:59:42 -07002562
Xiaoguang Wang2d7d6792020-06-16 02:06:37 +08002563 if (res != -EAGAIN && res != req->result)
Jens Axboe4e88d6e2019-12-07 20:59:47 -07002564 req_set_fail_links(req);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002565
2566 WRITE_ONCE(req->result, res);
Jens Axboeb9b0e0d2021-02-23 08:18:36 -07002567 /* order with io_iopoll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002568 smp_wmb();
2569 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002570}
2571
2572/*
2573 * After the iocb has been issued, it's safe to be found on the poll list.
2574 * Adding the kiocb to the list AFTER submission ensures that we don't
2575 * find it from a io_iopoll_getevents() thread before the issuer is done
2576 * accessing the kiocb cookie.
2577 */
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002578static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
Jens Axboedef596e2019-01-09 08:59:42 -07002579{
2580 struct io_ring_ctx *ctx = req->ctx;
2581
2582 /*
2583 * Track whether we have multiple files in our lists. This will impact
2584 * how we do polling eventually, not spinning if we're on potentially
2585 * different devices.
2586 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002587 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002588 ctx->poll_multi_file = false;
2589 } else if (!ctx->poll_multi_file) {
2590 struct io_kiocb *list_req;
2591
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002592 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002593 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002594 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002595 ctx->poll_multi_file = true;
2596 }
2597
2598 /*
2599 * For fast devices, IO may have already completed. If it has, add
2600 * it to the front so we find it first.
2601 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002602 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002603 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002604 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002605 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002606
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002607 /*
2608 * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread
2609 * task context or in io worker task context. If current task context is
2610 * sq thread, we don't need to check whether should wake up sq thread.
2611 */
2612 if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002613 wq_has_sleeper(&ctx->sq_data->wait))
2614 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002615}
2616
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002617static inline void io_state_file_put(struct io_submit_state *state)
2618{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002619 if (state->file_refs) {
2620 fput_many(state->file, state->file_refs);
2621 state->file_refs = 0;
2622 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002623}
2624
2625/*
2626 * Get as many references to a file as we have IOs left in this submission,
2627 * assuming most submissions are for one file, or at least that each file
2628 * has more than one submission.
2629 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002630static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002631{
2632 if (!state)
2633 return fget(fd);
2634
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002635 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002636 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002637 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002638 return state->file;
2639 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002640 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002641 }
2642 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002643 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002644 return NULL;
2645
2646 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002647 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002648 return state->file;
2649}
2650
Jens Axboe4503b762020-06-01 10:00:27 -06002651static bool io_bdev_nowait(struct block_device *bdev)
2652{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002653 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002654}
2655
Jens Axboe2b188cc2019-01-07 10:46:33 -07002656/*
2657 * If we tracked the file through the SCM inflight mechanism, we could support
2658 * any file. For now, just ensure that anything potentially problematic is done
2659 * inline.
2660 */
Jens Axboe7b29f922021-03-12 08:30:14 -07002661static bool __io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002662{
2663 umode_t mode = file_inode(file)->i_mode;
2664
Jens Axboe4503b762020-06-01 10:00:27 -06002665 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002666 if (IS_ENABLED(CONFIG_BLOCK) &&
2667 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002668 return true;
2669 return false;
2670 }
2671 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002672 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002673 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002674 if (IS_ENABLED(CONFIG_BLOCK) &&
2675 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002676 file->f_op != &io_uring_fops)
2677 return true;
2678 return false;
2679 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002680
Jens Axboec5b85622020-06-09 19:23:05 -06002681 /* any ->read/write should understand O_NONBLOCK */
2682 if (file->f_flags & O_NONBLOCK)
2683 return true;
2684
Jens Axboeaf197f52020-04-28 13:15:06 -06002685 if (!(file->f_mode & FMODE_NOWAIT))
2686 return false;
2687
2688 if (rw == READ)
2689 return file->f_op->read_iter != NULL;
2690
2691 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002692}
2693
Jens Axboe7b29f922021-03-12 08:30:14 -07002694static bool io_file_supports_async(struct io_kiocb *req, int rw)
2695{
2696 if (rw == READ && (req->flags & REQ_F_ASYNC_READ))
2697 return true;
2698 else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE))
2699 return true;
2700
2701 return __io_file_supports_async(req->file, rw);
2702}
2703
Pavel Begunkova88fc402020-09-30 22:57:53 +03002704static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002705{
Jens Axboedef596e2019-01-09 08:59:42 -07002706 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002707 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002708 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002709 unsigned ioprio;
2710 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002711
Jens Axboe7b29f922021-03-12 08:30:14 -07002712 if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002713 req->flags |= REQ_F_ISREG;
2714
Jens Axboe2b188cc2019-01-07 10:46:33 -07002715 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002716 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002717 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002718 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002719 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002720 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002721 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2722 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2723 if (unlikely(ret))
2724 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002725
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002726 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
2727 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
2728 req->flags |= REQ_F_NOWAIT;
2729
Jens Axboe2b188cc2019-01-07 10:46:33 -07002730 ioprio = READ_ONCE(sqe->ioprio);
2731 if (ioprio) {
2732 ret = ioprio_check_cap(ioprio);
2733 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002734 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002735
2736 kiocb->ki_ioprio = ioprio;
2737 } else
2738 kiocb->ki_ioprio = get_current_ioprio();
2739
Jens Axboedef596e2019-01-09 08:59:42 -07002740 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002741 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2742 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002743 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002744
Jens Axboedef596e2019-01-09 08:59:42 -07002745 kiocb->ki_flags |= IOCB_HIPRI;
2746 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002747 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002748 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002749 if (kiocb->ki_flags & IOCB_HIPRI)
2750 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002751 kiocb->ki_complete = io_complete_rw;
2752 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002753
Jens Axboe3529d8c2019-12-19 18:24:38 -07002754 req->rw.addr = READ_ONCE(sqe->addr);
2755 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002756 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002757 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002758}
2759
2760static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2761{
2762 switch (ret) {
2763 case -EIOCBQUEUED:
2764 break;
2765 case -ERESTARTSYS:
2766 case -ERESTARTNOINTR:
2767 case -ERESTARTNOHAND:
2768 case -ERESTART_RESTARTBLOCK:
2769 /*
2770 * We can't just restart the syscall, since previously
2771 * submitted sqes may already be in progress. Just fail this
2772 * IO with EINTR.
2773 */
2774 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002775 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002776 default:
2777 kiocb->ki_complete(kiocb, ret, 0);
2778 }
2779}
2780
Jens Axboea1d7c392020-06-22 11:09:46 -06002781static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002782 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06002783{
Jens Axboeba042912019-12-25 16:33:42 -07002784 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002785 struct io_async_rw *io = req->async_data;
Pavel Begunkov97284632021-04-08 19:28:03 +01002786 bool check_reissue = kiocb->ki_complete == io_complete_rw;
Jens Axboeba042912019-12-25 16:33:42 -07002787
Jens Axboe227c0c92020-08-13 11:51:40 -06002788 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002789 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002790 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002791 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002792 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002793 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002794 }
2795
Jens Axboeba042912019-12-25 16:33:42 -07002796 if (req->flags & REQ_F_CUR_POS)
2797 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002798 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00002799 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002800 else
2801 io_rw_done(kiocb, ret);
Pavel Begunkov97284632021-04-08 19:28:03 +01002802
2803 if (check_reissue && req->flags & REQ_F_REISSUE) {
2804 req->flags &= ~REQ_F_REISSUE;
2805 if (!io_rw_reissue(req)) {
2806 int cflags = 0;
2807
2808 req_set_fail_links(req);
2809 if (req->flags & REQ_F_BUFFER_SELECTED)
2810 cflags = io_put_rw_kbuf(req);
2811 __io_req_complete(req, issue_flags, ret, cflags);
2812 }
2813 }
Jens Axboeba816ad2019-09-28 11:36:45 -06002814}
2815
Pavel Begunkov847595d2021-02-04 13:52:06 +00002816static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07002817{
Jens Axboe9adbd452019-12-20 08:45:55 -07002818 struct io_ring_ctx *ctx = req->ctx;
2819 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002820 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03002821 u16 index, buf_index = req->buf_index;
Jens Axboeedafcce2019-01-09 09:16:05 -07002822 size_t offset;
2823 u64 buf_addr;
2824
Jens Axboeedafcce2019-01-09 09:16:05 -07002825 if (unlikely(buf_index >= ctx->nr_user_bufs))
2826 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07002827 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2828 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07002829 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002830
2831 /* overflow */
2832 if (buf_addr + len < buf_addr)
2833 return -EFAULT;
2834 /* not inside the mapped region */
2835 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
2836 return -EFAULT;
2837
2838 /*
2839 * May not be a start of buffer, set size appropriately
2840 * and advance us to the beginning.
2841 */
2842 offset = buf_addr - imu->ubuf;
2843 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002844
2845 if (offset) {
2846 /*
2847 * Don't use iov_iter_advance() here, as it's really slow for
2848 * using the latter parts of a big fixed buffer - it iterates
2849 * over each segment manually. We can cheat a bit here, because
2850 * we know that:
2851 *
2852 * 1) it's a BVEC iter, we set it up
2853 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2854 * first and last bvec
2855 *
2856 * So just find our index, and adjust the iterator afterwards.
2857 * If the offset is within the first bvec (or the whole first
2858 * bvec, just use iov_iter_advance(). This makes it easier
2859 * since we can just skip the first segment, which may not
2860 * be PAGE_SIZE aligned.
2861 */
2862 const struct bio_vec *bvec = imu->bvec;
2863
2864 if (offset <= bvec->bv_len) {
2865 iov_iter_advance(iter, offset);
2866 } else {
2867 unsigned long seg_skip;
2868
2869 /* skip first vec */
2870 offset -= bvec->bv_len;
2871 seg_skip = 1 + (offset >> PAGE_SHIFT);
2872
2873 iter->bvec = bvec + seg_skip;
2874 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002875 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002876 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002877 }
2878 }
2879
Pavel Begunkov847595d2021-02-04 13:52:06 +00002880 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07002881}
2882
Jens Axboebcda7ba2020-02-23 16:42:51 -07002883static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2884{
2885 if (needs_lock)
2886 mutex_unlock(&ctx->uring_lock);
2887}
2888
2889static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2890{
2891 /*
2892 * "Normal" inline submissions always hold the uring_lock, since we
2893 * grab it from the system call. Same is true for the SQPOLL offload.
2894 * The only exception is when we've detached the request and issue it
2895 * from an async worker thread, grab the lock for that case.
2896 */
2897 if (needs_lock)
2898 mutex_lock(&ctx->uring_lock);
2899}
2900
2901static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2902 int bgid, struct io_buffer *kbuf,
2903 bool needs_lock)
2904{
2905 struct io_buffer *head;
2906
2907 if (req->flags & REQ_F_BUFFER_SELECTED)
2908 return kbuf;
2909
2910 io_ring_submit_lock(req->ctx, needs_lock);
2911
2912 lockdep_assert_held(&req->ctx->uring_lock);
2913
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002914 head = xa_load(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002915 if (head) {
2916 if (!list_empty(&head->list)) {
2917 kbuf = list_last_entry(&head->list, struct io_buffer,
2918 list);
2919 list_del(&kbuf->list);
2920 } else {
2921 kbuf = head;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002922 xa_erase(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002923 }
2924 if (*len > kbuf->len)
2925 *len = kbuf->len;
2926 } else {
2927 kbuf = ERR_PTR(-ENOBUFS);
2928 }
2929
2930 io_ring_submit_unlock(req->ctx, needs_lock);
2931
2932 return kbuf;
2933}
2934
Jens Axboe4d954c22020-02-27 07:31:19 -07002935static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2936 bool needs_lock)
2937{
2938 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002939 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002940
2941 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002942 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002943 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2944 if (IS_ERR(kbuf))
2945 return kbuf;
2946 req->rw.addr = (u64) (unsigned long) kbuf;
2947 req->flags |= REQ_F_BUFFER_SELECTED;
2948 return u64_to_user_ptr(kbuf->addr);
2949}
2950
2951#ifdef CONFIG_COMPAT
2952static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2953 bool needs_lock)
2954{
2955 struct compat_iovec __user *uiov;
2956 compat_ssize_t clen;
2957 void __user *buf;
2958 ssize_t len;
2959
2960 uiov = u64_to_user_ptr(req->rw.addr);
2961 if (!access_ok(uiov, sizeof(*uiov)))
2962 return -EFAULT;
2963 if (__get_user(clen, &uiov->iov_len))
2964 return -EFAULT;
2965 if (clen < 0)
2966 return -EINVAL;
2967
2968 len = clen;
2969 buf = io_rw_buffer_select(req, &len, needs_lock);
2970 if (IS_ERR(buf))
2971 return PTR_ERR(buf);
2972 iov[0].iov_base = buf;
2973 iov[0].iov_len = (compat_size_t) len;
2974 return 0;
2975}
2976#endif
2977
2978static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2979 bool needs_lock)
2980{
2981 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
2982 void __user *buf;
2983 ssize_t len;
2984
2985 if (copy_from_user(iov, uiov, sizeof(*uiov)))
2986 return -EFAULT;
2987
2988 len = iov[0].iov_len;
2989 if (len < 0)
2990 return -EINVAL;
2991 buf = io_rw_buffer_select(req, &len, needs_lock);
2992 if (IS_ERR(buf))
2993 return PTR_ERR(buf);
2994 iov[0].iov_base = buf;
2995 iov[0].iov_len = len;
2996 return 0;
2997}
2998
2999static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
3000 bool needs_lock)
3001{
Jens Axboedddb3e22020-06-04 11:27:01 -06003002 if (req->flags & REQ_F_BUFFER_SELECTED) {
3003 struct io_buffer *kbuf;
3004
3005 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
3006 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
3007 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07003008 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06003009 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00003010 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07003011 return -EINVAL;
3012
3013#ifdef CONFIG_COMPAT
3014 if (req->ctx->compat)
3015 return io_compat_import(req, iov, needs_lock);
3016#endif
3017
3018 return __io_iov_buffer_select(req, iov, needs_lock);
3019}
3020
Pavel Begunkov847595d2021-02-04 13:52:06 +00003021static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
3022 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003023{
Jens Axboe9adbd452019-12-20 08:45:55 -07003024 void __user *buf = u64_to_user_ptr(req->rw.addr);
3025 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003026 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07003027 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07003028
Pavel Begunkov7d009162019-11-25 23:14:40 +03003029 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07003030 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07003031 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07003032 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003033
Jens Axboebcda7ba2020-02-23 16:42:51 -07003034 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003035 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07003036 return -EINVAL;
3037
Jens Axboe3a6820f2019-12-22 15:19:35 -07003038 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07003039 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07003040 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03003041 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07003042 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06003043 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07003044 }
3045
Jens Axboe3a6820f2019-12-22 15:19:35 -07003046 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
3047 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00003048 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07003049 }
3050
Jens Axboe4d954c22020-02-27 07:31:19 -07003051 if (req->flags & REQ_F_BUFFER_SELECT) {
3052 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00003053 if (!ret)
3054 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07003055 *iovec = NULL;
3056 return ret;
3057 }
3058
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02003059 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
3060 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003061}
3062
Jens Axboe0fef9482020-08-26 10:36:20 -06003063static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
3064{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03003065 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06003066}
3067
Jens Axboe32960612019-09-23 11:05:34 -06003068/*
3069 * For files that don't have ->read_iter() and ->write_iter(), handle them
3070 * by looping over ->read() or ->write() manually.
3071 */
Jens Axboe4017eb92020-10-22 14:14:12 -06003072static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06003073{
Jens Axboe4017eb92020-10-22 14:14:12 -06003074 struct kiocb *kiocb = &req->rw.kiocb;
3075 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003076 ssize_t ret = 0;
3077
3078 /*
3079 * Don't support polled IO through this interface, and we can't
3080 * support non-blocking either. For the latter, this just causes
3081 * the kiocb to be handled from an async context.
3082 */
3083 if (kiocb->ki_flags & IOCB_HIPRI)
3084 return -EOPNOTSUPP;
3085 if (kiocb->ki_flags & IOCB_NOWAIT)
3086 return -EAGAIN;
3087
3088 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003089 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003090 ssize_t nr;
3091
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003092 if (!iov_iter_is_bvec(iter)) {
3093 iovec = iov_iter_iovec(iter);
3094 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003095 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3096 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003097 }
3098
Jens Axboe32960612019-09-23 11:05:34 -06003099 if (rw == READ) {
3100 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003101 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003102 } else {
3103 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003104 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003105 }
3106
3107 if (nr < 0) {
3108 if (!ret)
3109 ret = nr;
3110 break;
3111 }
3112 ret += nr;
3113 if (nr != iovec.iov_len)
3114 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003115 req->rw.len -= nr;
3116 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003117 iov_iter_advance(iter, nr);
3118 }
3119
3120 return ret;
3121}
3122
Jens Axboeff6165b2020-08-13 09:47:43 -06003123static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3124 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003125{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003126 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003127
Jens Axboeff6165b2020-08-13 09:47:43 -06003128 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003129 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003130 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003131 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003132 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003133 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003134 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003135 unsigned iov_off = 0;
3136
3137 rw->iter.iov = rw->fast_iov;
3138 if (iter->iov != fast_iov) {
3139 iov_off = iter->iov - fast_iov;
3140 rw->iter.iov += iov_off;
3141 }
3142 if (rw->fast_iov != fast_iov)
3143 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003144 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003145 } else {
3146 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003147 }
3148}
3149
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003150static inline int io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003151{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003152 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3153 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3154 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003155}
3156
Jens Axboeff6165b2020-08-13 09:47:43 -06003157static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3158 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003159 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003160{
Pavel Begunkov26f05052021-02-28 22:35:18 +00003161 if (!force && !io_op_defs[req->opcode].needs_async_setup)
Jens Axboe74566df2020-01-13 19:23:24 -07003162 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003163 if (!req->async_data) {
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003164 if (io_alloc_async_data(req)) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003165 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003166 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003167 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003168
Jens Axboeff6165b2020-08-13 09:47:43 -06003169 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003170 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003171 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003172}
3173
Pavel Begunkov73debe62020-09-30 22:57:54 +03003174static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003175{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003176 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003177 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003178 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003179
Pavel Begunkov2846c482020-11-07 13:16:27 +00003180 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003181 if (unlikely(ret < 0))
3182 return ret;
3183
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003184 iorw->bytes_done = 0;
3185 iorw->free_iovec = iov;
3186 if (iov)
3187 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003188 return 0;
3189}
3190
Pavel Begunkov73debe62020-09-30 22:57:54 +03003191static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003192{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003193 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3194 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003195 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003196}
3197
Jens Axboec1dd91d2020-08-03 16:43:59 -06003198/*
3199 * This is our waitqueue callback handler, registered through lock_page_async()
3200 * when we initially tried to do the IO with the iocb armed our waitqueue.
3201 * This gets called when the page is unlocked, and we generally expect that to
3202 * happen when the page IO is completed and the page is now uptodate. This will
3203 * queue a task_work based retry of the operation, attempting to copy the data
3204 * again. If the latter fails because the page was NOT uptodate, then we will
3205 * do a thread based blocking retry of the operation. That's the unexpected
3206 * slow path.
3207 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003208static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3209 int sync, void *arg)
3210{
3211 struct wait_page_queue *wpq;
3212 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003213 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003214
3215 wpq = container_of(wait, struct wait_page_queue, wait);
3216
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003217 if (!wake_page_match(wpq, key))
3218 return 0;
3219
Hao Xuc8d317a2020-09-29 20:00:45 +08003220 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003221 list_del_init(&wait->entry);
3222
Jens Axboebcf5a062020-05-22 09:24:42 -06003223 /* submit ref gets dropped, acquire a new one */
Jens Axboede9b4cc2021-02-24 13:28:27 -07003224 req_ref_get(req);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003225 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003226 return 1;
3227}
3228
Jens Axboec1dd91d2020-08-03 16:43:59 -06003229/*
3230 * This controls whether a given IO request should be armed for async page
3231 * based retry. If we return false here, the request is handed to the async
3232 * worker threads for retry. If we're doing buffered reads on a regular file,
3233 * we prepare a private wait_page_queue entry and retry the operation. This
3234 * will either succeed because the page is now uptodate and unlocked, or it
3235 * will register a callback when the page is unlocked at IO completion. Through
3236 * that callback, io_uring uses task_work to setup a retry of the operation.
3237 * That retry will attempt the buffered read again. The retry will generally
3238 * succeed, or in rare cases where it fails, we then fall back to using the
3239 * async worker threads for a blocking retry.
3240 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003241static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003242{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003243 struct io_async_rw *rw = req->async_data;
3244 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003245 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003246
3247 /* never retry for NOWAIT, we just complete with -EAGAIN */
3248 if (req->flags & REQ_F_NOWAIT)
3249 return false;
3250
Jens Axboe227c0c92020-08-13 11:51:40 -06003251 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003252 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003253 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003254
Jens Axboebcf5a062020-05-22 09:24:42 -06003255 /*
3256 * just use poll if we can, and don't attempt if the fs doesn't
3257 * support callback based unlocks
3258 */
3259 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3260 return false;
3261
Jens Axboe3b2a4432020-08-16 10:58:43 -07003262 wait->wait.func = io_async_buf_func;
3263 wait->wait.private = req;
3264 wait->wait.flags = 0;
3265 INIT_LIST_HEAD(&wait->wait.entry);
3266 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003267 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003268 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003269 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003270}
3271
3272static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3273{
3274 if (req->file->f_op->read_iter)
3275 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003276 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003277 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003278 else
3279 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003280}
3281
Pavel Begunkov889fca72021-02-10 00:03:09 +00003282static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003283{
3284 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003285 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003286 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003287 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003288 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003289 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003290
Pavel Begunkov2846c482020-11-07 13:16:27 +00003291 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003292 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003293 iovec = NULL;
3294 } else {
3295 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3296 if (ret < 0)
3297 return ret;
3298 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003299 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003300 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003301
Jens Axboefd6c2e42019-12-18 12:19:41 -07003302 /* Ensure we clear previously set non-block flag */
3303 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003304 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003305 else
3306 kiocb->ki_flags |= IOCB_NOWAIT;
3307
Pavel Begunkov24c74672020-06-21 13:09:51 +03003308 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003309 if (force_nonblock && !io_file_supports_async(req, READ)) {
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003310 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003311 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003312 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003313
Pavel Begunkov632546c2020-11-07 13:16:26 +00003314 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003315 if (unlikely(ret)) {
3316 kfree(iovec);
3317 return ret;
3318 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003319
Jens Axboe227c0c92020-08-13 11:51:40 -06003320 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003321
Jens Axboe230d50d2021-04-01 20:41:15 -06003322 if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) {
Pavel Begunkov6ad7f232021-04-08 01:54:39 +01003323 req->flags &= ~REQ_F_REISSUE;
Jens Axboeeefdf302020-08-27 16:40:19 -06003324 /* IOPOLL retry should happen for io-wq threads */
3325 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003326 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003327 /* no retry on NONBLOCK nor RWF_NOWAIT */
3328 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003329 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003330 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003331 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003332 ret = 0;
Jens Axboe230d50d2021-04-01 20:41:15 -06003333 } else if (ret == -EIOCBQUEUED) {
3334 goto out_free;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003335 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003336 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003337 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003338 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003339 }
3340
Jens Axboe227c0c92020-08-13 11:51:40 -06003341 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003342 if (ret2)
3343 return ret2;
3344
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003345 iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003346 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003347 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003348 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003349
Pavel Begunkovb23df912021-02-04 13:52:04 +00003350 do {
3351 io_size -= ret;
3352 rw->bytes_done += ret;
3353 /* if we can retry, do so with the callbacks armed */
3354 if (!io_rw_should_retry(req)) {
3355 kiocb->ki_flags &= ~IOCB_WAITQ;
3356 return -EAGAIN;
3357 }
3358
3359 /*
3360 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3361 * we get -EIOCBQUEUED, then we'll get a notification when the
3362 * desired page gets unlocked. We can also get a partial read
3363 * here, and if we do, then just retry at the new offset.
3364 */
3365 ret = io_iter_do_read(req, iter);
3366 if (ret == -EIOCBQUEUED)
3367 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003368 /* we got some bytes, but not all. retry. */
Jens Axboeb5b0ecb2021-03-04 21:02:58 -07003369 kiocb->ki_flags &= ~IOCB_WAITQ;
Pavel Begunkovb23df912021-02-04 13:52:04 +00003370 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003371done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003372 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003373out_free:
3374 /* it's faster to check here then delegate to kfree */
3375 if (iovec)
3376 kfree(iovec);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003377 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003378}
3379
Pavel Begunkov73debe62020-09-30 22:57:54 +03003380static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003381{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003382 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3383 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003384 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003385}
3386
Pavel Begunkov889fca72021-02-10 00:03:09 +00003387static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003388{
3389 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003390 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003391 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003392 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003393 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003394 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003395
Pavel Begunkov2846c482020-11-07 13:16:27 +00003396 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003397 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003398 iovec = NULL;
3399 } else {
3400 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3401 if (ret < 0)
3402 return ret;
3403 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003404 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003405 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003406
Jens Axboefd6c2e42019-12-18 12:19:41 -07003407 /* Ensure we clear previously set non-block flag */
3408 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003409 kiocb->ki_flags &= ~IOCB_NOWAIT;
3410 else
3411 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003412
Pavel Begunkov24c74672020-06-21 13:09:51 +03003413 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003414 if (force_nonblock && !io_file_supports_async(req, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003415 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003416
Jens Axboe10d59342019-12-09 20:16:22 -07003417 /* file path doesn't support NOWAIT for non-direct_IO */
3418 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3419 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003420 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003421
Pavel Begunkov632546c2020-11-07 13:16:26 +00003422 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003423 if (unlikely(ret))
3424 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003425
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003426 /*
3427 * Open-code file_start_write here to grab freeze protection,
3428 * which will be released by another thread in
3429 * io_complete_rw(). Fool lockdep by telling it the lock got
3430 * released so that it doesn't complain about the held lock when
3431 * we return to userspace.
3432 */
3433 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003434 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003435 __sb_writers_release(file_inode(req->file)->i_sb,
3436 SB_FREEZE_WRITE);
3437 }
3438 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003439
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003440 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003441 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003442 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003443 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003444 else
3445 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003446
Pavel Begunkov6ad7f232021-04-08 01:54:39 +01003447 if (req->flags & REQ_F_REISSUE) {
3448 req->flags &= ~REQ_F_REISSUE;
Jens Axboe230d50d2021-04-01 20:41:15 -06003449 ret2 = -EAGAIN;
Pavel Begunkov6ad7f232021-04-08 01:54:39 +01003450 }
Jens Axboe230d50d2021-04-01 20:41:15 -06003451
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003452 /*
3453 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3454 * retry them without IOCB_NOWAIT.
3455 */
3456 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3457 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003458 /* no retry on NONBLOCK nor RWF_NOWAIT */
3459 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003460 goto done;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003461 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003462 /* IOPOLL retry should happen for io-wq threads */
3463 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3464 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003465done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003466 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003467 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003468copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003469 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003470 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003471 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003472 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003473 }
Jens Axboe31b51512019-01-18 22:56:34 -07003474out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003475 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003476 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003477 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003478 return ret;
3479}
3480
Jens Axboe80a261f2020-09-28 14:23:58 -06003481static int io_renameat_prep(struct io_kiocb *req,
3482 const struct io_uring_sqe *sqe)
3483{
3484 struct io_rename *ren = &req->rename;
3485 const char __user *oldf, *newf;
3486
3487 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3488 return -EBADF;
3489
3490 ren->old_dfd = READ_ONCE(sqe->fd);
3491 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3492 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3493 ren->new_dfd = READ_ONCE(sqe->len);
3494 ren->flags = READ_ONCE(sqe->rename_flags);
3495
3496 ren->oldpath = getname(oldf);
3497 if (IS_ERR(ren->oldpath))
3498 return PTR_ERR(ren->oldpath);
3499
3500 ren->newpath = getname(newf);
3501 if (IS_ERR(ren->newpath)) {
3502 putname(ren->oldpath);
3503 return PTR_ERR(ren->newpath);
3504 }
3505
3506 req->flags |= REQ_F_NEED_CLEANUP;
3507 return 0;
3508}
3509
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003510static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003511{
3512 struct io_rename *ren = &req->rename;
3513 int ret;
3514
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003515 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003516 return -EAGAIN;
3517
3518 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3519 ren->newpath, ren->flags);
3520
3521 req->flags &= ~REQ_F_NEED_CLEANUP;
3522 if (ret < 0)
3523 req_set_fail_links(req);
3524 io_req_complete(req, ret);
3525 return 0;
3526}
3527
Jens Axboe14a11432020-09-28 14:27:37 -06003528static int io_unlinkat_prep(struct io_kiocb *req,
3529 const struct io_uring_sqe *sqe)
3530{
3531 struct io_unlink *un = &req->unlink;
3532 const char __user *fname;
3533
3534 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3535 return -EBADF;
3536
3537 un->dfd = READ_ONCE(sqe->fd);
3538
3539 un->flags = READ_ONCE(sqe->unlink_flags);
3540 if (un->flags & ~AT_REMOVEDIR)
3541 return -EINVAL;
3542
3543 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3544 un->filename = getname(fname);
3545 if (IS_ERR(un->filename))
3546 return PTR_ERR(un->filename);
3547
3548 req->flags |= REQ_F_NEED_CLEANUP;
3549 return 0;
3550}
3551
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003552static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003553{
3554 struct io_unlink *un = &req->unlink;
3555 int ret;
3556
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003557 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003558 return -EAGAIN;
3559
3560 if (un->flags & AT_REMOVEDIR)
3561 ret = do_rmdir(un->dfd, un->filename);
3562 else
3563 ret = do_unlinkat(un->dfd, un->filename);
3564
3565 req->flags &= ~REQ_F_NEED_CLEANUP;
3566 if (ret < 0)
3567 req_set_fail_links(req);
3568 io_req_complete(req, ret);
3569 return 0;
3570}
3571
Jens Axboe36f4fa62020-09-05 11:14:22 -06003572static int io_shutdown_prep(struct io_kiocb *req,
3573 const struct io_uring_sqe *sqe)
3574{
3575#if defined(CONFIG_NET)
3576 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3577 return -EINVAL;
3578 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3579 sqe->buf_index)
3580 return -EINVAL;
3581
3582 req->shutdown.how = READ_ONCE(sqe->len);
3583 return 0;
3584#else
3585 return -EOPNOTSUPP;
3586#endif
3587}
3588
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003589static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003590{
3591#if defined(CONFIG_NET)
3592 struct socket *sock;
3593 int ret;
3594
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003595 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003596 return -EAGAIN;
3597
Linus Torvalds48aba792020-12-16 12:44:05 -08003598 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003599 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003600 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003601
3602 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003603 if (ret < 0)
3604 req_set_fail_links(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003605 io_req_complete(req, ret);
3606 return 0;
3607#else
3608 return -EOPNOTSUPP;
3609#endif
3610}
3611
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003612static int __io_splice_prep(struct io_kiocb *req,
3613 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003614{
3615 struct io_splice* sp = &req->splice;
3616 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003617
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003618 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3619 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003620
3621 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003622 sp->len = READ_ONCE(sqe->len);
3623 sp->flags = READ_ONCE(sqe->splice_flags);
3624
3625 if (unlikely(sp->flags & ~valid_flags))
3626 return -EINVAL;
3627
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003628 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3629 (sp->flags & SPLICE_F_FD_IN_FIXED));
3630 if (!sp->file_in)
3631 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003632 req->flags |= REQ_F_NEED_CLEANUP;
3633
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003634 if (!S_ISREG(file_inode(sp->file_in)->i_mode)) {
3635 /*
3636 * Splice operation will be punted aync, and here need to
3637 * modify io_wq_work.flags, so initialize io_wq_work firstly.
3638 */
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003639 req->work.flags |= IO_WQ_WORK_UNBOUND;
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003640 }
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003641
3642 return 0;
3643}
3644
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003645static int io_tee_prep(struct io_kiocb *req,
3646 const struct io_uring_sqe *sqe)
3647{
3648 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3649 return -EINVAL;
3650 return __io_splice_prep(req, sqe);
3651}
3652
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003653static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003654{
3655 struct io_splice *sp = &req->splice;
3656 struct file *in = sp->file_in;
3657 struct file *out = sp->file_out;
3658 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3659 long ret = 0;
3660
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003661 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003662 return -EAGAIN;
3663 if (sp->len)
3664 ret = do_tee(in, out, sp->len, flags);
3665
3666 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3667 req->flags &= ~REQ_F_NEED_CLEANUP;
3668
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003669 if (ret != sp->len)
3670 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003671 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003672 return 0;
3673}
3674
3675static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3676{
3677 struct io_splice* sp = &req->splice;
3678
3679 sp->off_in = READ_ONCE(sqe->splice_off_in);
3680 sp->off_out = READ_ONCE(sqe->off);
3681 return __io_splice_prep(req, sqe);
3682}
3683
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003684static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003685{
3686 struct io_splice *sp = &req->splice;
3687 struct file *in = sp->file_in;
3688 struct file *out = sp->file_out;
3689 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3690 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003691 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003692
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003693 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003694 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003695
3696 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3697 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003698
Jens Axboe948a7742020-05-17 14:21:38 -06003699 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003700 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003701
3702 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3703 req->flags &= ~REQ_F_NEED_CLEANUP;
3704
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003705 if (ret != sp->len)
3706 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003707 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003708 return 0;
3709}
3710
Jens Axboe2b188cc2019-01-07 10:46:33 -07003711/*
3712 * IORING_OP_NOP just posts a completion event, nothing else.
3713 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003714static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003715{
3716 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003717
Jens Axboedef596e2019-01-09 08:59:42 -07003718 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3719 return -EINVAL;
3720
Pavel Begunkov889fca72021-02-10 00:03:09 +00003721 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003722 return 0;
3723}
3724
Pavel Begunkov1155c762021-02-18 18:29:38 +00003725static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003726{
Jens Axboe6b063142019-01-10 22:13:58 -07003727 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003728
Jens Axboe09bb8392019-03-13 12:39:28 -06003729 if (!req->file)
3730 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003731
Jens Axboe6b063142019-01-10 22:13:58 -07003732 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003733 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003734 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003735 return -EINVAL;
3736
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003737 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3738 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3739 return -EINVAL;
3740
3741 req->sync.off = READ_ONCE(sqe->off);
3742 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003743 return 0;
3744}
3745
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003746static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07003747{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003748 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003749 int ret;
3750
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003751 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003752 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003753 return -EAGAIN;
3754
Jens Axboe9adbd452019-12-20 08:45:55 -07003755 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003756 end > 0 ? end : LLONG_MAX,
3757 req->sync.flags & IORING_FSYNC_DATASYNC);
3758 if (ret < 0)
3759 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003760 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003761 return 0;
3762}
3763
Jens Axboed63d1b52019-12-10 10:38:56 -07003764static int io_fallocate_prep(struct io_kiocb *req,
3765 const struct io_uring_sqe *sqe)
3766{
3767 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3768 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003769 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3770 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003771
3772 req->sync.off = READ_ONCE(sqe->off);
3773 req->sync.len = READ_ONCE(sqe->addr);
3774 req->sync.mode = READ_ONCE(sqe->len);
3775 return 0;
3776}
3777
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003778static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07003779{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003780 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003781
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003782 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003783 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003784 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003785 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3786 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003787 if (ret < 0)
3788 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003789 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003790 return 0;
3791}
3792
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003793static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003794{
Jens Axboef8748882020-01-08 17:47:02 -07003795 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003796 int ret;
3797
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003798 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003799 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003800 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003801 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003802
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003803 /* open.how should be already initialised */
3804 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003805 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003806
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003807 req->open.dfd = READ_ONCE(sqe->fd);
3808 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003809 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003810 if (IS_ERR(req->open.filename)) {
3811 ret = PTR_ERR(req->open.filename);
3812 req->open.filename = NULL;
3813 return ret;
3814 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003815 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003816 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003817 return 0;
3818}
3819
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003820static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3821{
3822 u64 flags, mode;
3823
Jens Axboe14587a462020-09-05 11:36:08 -06003824 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003825 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003826 mode = READ_ONCE(sqe->len);
3827 flags = READ_ONCE(sqe->open_flags);
3828 req->open.how = build_open_how(flags, mode);
3829 return __io_openat_prep(req, sqe);
3830}
3831
Jens Axboecebdb982020-01-08 17:59:24 -07003832static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3833{
3834 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003835 size_t len;
3836 int ret;
3837
Jens Axboe14587a462020-09-05 11:36:08 -06003838 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003839 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003840 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3841 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003842 if (len < OPEN_HOW_SIZE_VER0)
3843 return -EINVAL;
3844
3845 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3846 len);
3847 if (ret)
3848 return ret;
3849
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003850 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003851}
3852
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003853static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003854{
3855 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003856 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003857 bool nonblock_set;
3858 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003859 int ret;
3860
Jens Axboecebdb982020-01-08 17:59:24 -07003861 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003862 if (ret)
3863 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003864 nonblock_set = op.open_flag & O_NONBLOCK;
3865 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003866 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003867 /*
3868 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
3869 * it'll always -EAGAIN
3870 */
3871 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
3872 return -EAGAIN;
3873 op.lookup_flags |= LOOKUP_CACHED;
3874 op.open_flag |= O_NONBLOCK;
3875 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07003876
Jens Axboe4022e7a2020-03-19 19:23:18 -06003877 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003878 if (ret < 0)
3879 goto err;
3880
3881 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Jens Axboe3a81fd02020-12-10 12:25:36 -07003882 /* only retry if RESOLVE_CACHED wasn't already set by application */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003883 if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) &&
3884 file == ERR_PTR(-EAGAIN)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003885 /*
3886 * We could hang on to this 'fd', but seems like marginal
3887 * gain for something that is now known to be a slower path.
3888 * So just put it, and we'll get a new one when we retry.
3889 */
3890 put_unused_fd(ret);
3891 return -EAGAIN;
3892 }
3893
Jens Axboe15b71ab2019-12-11 11:20:36 -07003894 if (IS_ERR(file)) {
3895 put_unused_fd(ret);
3896 ret = PTR_ERR(file);
3897 } else {
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003898 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
Jens Axboe3a81fd02020-12-10 12:25:36 -07003899 file->f_flags &= ~O_NONBLOCK;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003900 fsnotify_open(file);
3901 fd_install(ret, file);
3902 }
3903err:
3904 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003905 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003906 if (ret < 0)
3907 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003908 io_req_complete(req, ret);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003909 return 0;
3910}
3911
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003912static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07003913{
Pavel Begunkove45cff52021-02-28 22:35:14 +00003914 return io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07003915}
3916
Jens Axboe067524e2020-03-02 16:32:28 -07003917static int io_remove_buffers_prep(struct io_kiocb *req,
3918 const struct io_uring_sqe *sqe)
3919{
3920 struct io_provide_buf *p = &req->pbuf;
3921 u64 tmp;
3922
3923 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3924 return -EINVAL;
3925
3926 tmp = READ_ONCE(sqe->fd);
3927 if (!tmp || tmp > USHRT_MAX)
3928 return -EINVAL;
3929
3930 memset(p, 0, sizeof(*p));
3931 p->nbufs = tmp;
3932 p->bgid = READ_ONCE(sqe->buf_group);
3933 return 0;
3934}
3935
3936static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3937 int bgid, unsigned nbufs)
3938{
3939 unsigned i = 0;
3940
3941 /* shouldn't happen */
3942 if (!nbufs)
3943 return 0;
3944
3945 /* the head kbuf is the list itself */
3946 while (!list_empty(&buf->list)) {
3947 struct io_buffer *nxt;
3948
3949 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3950 list_del(&nxt->list);
3951 kfree(nxt);
3952 if (++i == nbufs)
3953 return i;
3954 }
3955 i++;
3956 kfree(buf);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003957 xa_erase(&ctx->io_buffers, bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003958
3959 return i;
3960}
3961
Pavel Begunkov889fca72021-02-10 00:03:09 +00003962static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07003963{
3964 struct io_provide_buf *p = &req->pbuf;
3965 struct io_ring_ctx *ctx = req->ctx;
3966 struct io_buffer *head;
3967 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003968 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07003969
3970 io_ring_submit_lock(ctx, !force_nonblock);
3971
3972 lockdep_assert_held(&ctx->uring_lock);
3973
3974 ret = -ENOENT;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003975 head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003976 if (head)
3977 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07003978 if (ret < 0)
3979 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003980
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00003981 /* complete before unlock, IOPOLL may need the lock */
3982 __io_req_complete(req, issue_flags, ret, 0);
3983 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboe067524e2020-03-02 16:32:28 -07003984 return 0;
3985}
3986
Jens Axboeddf0322d2020-02-23 16:41:33 -07003987static int io_provide_buffers_prep(struct io_kiocb *req,
3988 const struct io_uring_sqe *sqe)
3989{
Pavel Begunkovd81269f2021-03-19 10:21:19 +00003990 unsigned long size;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003991 struct io_provide_buf *p = &req->pbuf;
3992 u64 tmp;
3993
3994 if (sqe->ioprio || sqe->rw_flags)
3995 return -EINVAL;
3996
3997 tmp = READ_ONCE(sqe->fd);
3998 if (!tmp || tmp > USHRT_MAX)
3999 return -E2BIG;
4000 p->nbufs = tmp;
4001 p->addr = READ_ONCE(sqe->addr);
4002 p->len = READ_ONCE(sqe->len);
4003
Pavel Begunkovd81269f2021-03-19 10:21:19 +00004004 size = (unsigned long)p->len * p->nbufs;
4005 if (!access_ok(u64_to_user_ptr(p->addr), size))
Jens Axboeddf0322d2020-02-23 16:41:33 -07004006 return -EFAULT;
4007
4008 p->bgid = READ_ONCE(sqe->buf_group);
4009 tmp = READ_ONCE(sqe->off);
4010 if (tmp > USHRT_MAX)
4011 return -E2BIG;
4012 p->bid = tmp;
4013 return 0;
4014}
4015
4016static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
4017{
4018 struct io_buffer *buf;
4019 u64 addr = pbuf->addr;
4020 int i, bid = pbuf->bid;
4021
4022 for (i = 0; i < pbuf->nbufs; i++) {
4023 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
4024 if (!buf)
4025 break;
4026
4027 buf->addr = addr;
4028 buf->len = pbuf->len;
4029 buf->bid = bid;
4030 addr += pbuf->len;
4031 bid++;
4032 if (!*head) {
4033 INIT_LIST_HEAD(&buf->list);
4034 *head = buf;
4035 } else {
4036 list_add_tail(&buf->list, &(*head)->list);
4037 }
4038 }
4039
4040 return i ? i : -ENOMEM;
4041}
4042
Pavel Begunkov889fca72021-02-10 00:03:09 +00004043static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07004044{
4045 struct io_provide_buf *p = &req->pbuf;
4046 struct io_ring_ctx *ctx = req->ctx;
4047 struct io_buffer *head, *list;
4048 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004049 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07004050
4051 io_ring_submit_lock(ctx, !force_nonblock);
4052
4053 lockdep_assert_held(&ctx->uring_lock);
4054
Jens Axboe9e15c3a2021-03-13 12:29:43 -07004055 list = head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004056
4057 ret = io_add_buffers(p, &head);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07004058 if (ret >= 0 && !list) {
4059 ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL);
4060 if (ret < 0)
Jens Axboe067524e2020-03-02 16:32:28 -07004061 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004062 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07004063 if (ret < 0)
4064 req_set_fail_links(req);
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00004065 /* complete before unlock, IOPOLL may need the lock */
4066 __io_req_complete(req, issue_flags, ret, 0);
4067 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004068 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004069}
4070
Jens Axboe3e4827b2020-01-08 15:18:09 -07004071static int io_epoll_ctl_prep(struct io_kiocb *req,
4072 const struct io_uring_sqe *sqe)
4073{
4074#if defined(CONFIG_EPOLL)
4075 if (sqe->ioprio || sqe->buf_index)
4076 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004077 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004078 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004079
4080 req->epoll.epfd = READ_ONCE(sqe->fd);
4081 req->epoll.op = READ_ONCE(sqe->len);
4082 req->epoll.fd = READ_ONCE(sqe->off);
4083
4084 if (ep_op_has_event(req->epoll.op)) {
4085 struct epoll_event __user *ev;
4086
4087 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4088 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4089 return -EFAULT;
4090 }
4091
4092 return 0;
4093#else
4094 return -EOPNOTSUPP;
4095#endif
4096}
4097
Pavel Begunkov889fca72021-02-10 00:03:09 +00004098static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004099{
4100#if defined(CONFIG_EPOLL)
4101 struct io_epoll *ie = &req->epoll;
4102 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004103 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004104
4105 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4106 if (force_nonblock && ret == -EAGAIN)
4107 return -EAGAIN;
4108
4109 if (ret < 0)
4110 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004111 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004112 return 0;
4113#else
4114 return -EOPNOTSUPP;
4115#endif
4116}
4117
Jens Axboec1ca7572019-12-25 22:18:28 -07004118static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4119{
4120#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4121 if (sqe->ioprio || sqe->buf_index || sqe->off)
4122 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004123 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4124 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004125
4126 req->madvise.addr = READ_ONCE(sqe->addr);
4127 req->madvise.len = READ_ONCE(sqe->len);
4128 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4129 return 0;
4130#else
4131 return -EOPNOTSUPP;
4132#endif
4133}
4134
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004135static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004136{
4137#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4138 struct io_madvise *ma = &req->madvise;
4139 int ret;
4140
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004141 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004142 return -EAGAIN;
4143
Minchan Kim0726b012020-10-17 16:14:50 -07004144 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004145 if (ret < 0)
4146 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004147 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004148 return 0;
4149#else
4150 return -EOPNOTSUPP;
4151#endif
4152}
4153
Jens Axboe4840e412019-12-25 22:03:45 -07004154static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4155{
4156 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4157 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004158 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4159 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004160
4161 req->fadvise.offset = READ_ONCE(sqe->off);
4162 req->fadvise.len = READ_ONCE(sqe->len);
4163 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4164 return 0;
4165}
4166
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004167static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004168{
4169 struct io_fadvise *fa = &req->fadvise;
4170 int ret;
4171
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004172 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004173 switch (fa->advice) {
4174 case POSIX_FADV_NORMAL:
4175 case POSIX_FADV_RANDOM:
4176 case POSIX_FADV_SEQUENTIAL:
4177 break;
4178 default:
4179 return -EAGAIN;
4180 }
4181 }
Jens Axboe4840e412019-12-25 22:03:45 -07004182
4183 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4184 if (ret < 0)
4185 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004186 io_req_complete(req, ret);
Jens Axboe4840e412019-12-25 22:03:45 -07004187 return 0;
4188}
4189
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004190static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4191{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004192 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004193 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004194 if (sqe->ioprio || sqe->buf_index)
4195 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004196 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004197 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004198
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004199 req->statx.dfd = READ_ONCE(sqe->fd);
4200 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004201 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004202 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4203 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004204
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004205 return 0;
4206}
4207
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004208static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004209{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004210 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004211 int ret;
4212
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004213 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004214 /* only need file table for an actual valid fd */
4215 if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD)
4216 req->flags |= REQ_F_NO_FILE_TABLE;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004217 return -EAGAIN;
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004218 }
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004219
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004220 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4221 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004222
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004223 if (ret < 0)
4224 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004225 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004226 return 0;
4227}
4228
Jens Axboeb5dba592019-12-11 14:02:38 -07004229static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4230{
Jens Axboe14587a462020-09-05 11:36:08 -06004231 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004232 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004233 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4234 sqe->rw_flags || sqe->buf_index)
4235 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004236 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004237 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004238
4239 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004240 return 0;
4241}
4242
Pavel Begunkov889fca72021-02-10 00:03:09 +00004243static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004244{
Jens Axboe9eac1902021-01-19 15:50:37 -07004245 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004246 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004247 struct fdtable *fdt;
4248 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -07004249 int ret;
4250
Jens Axboe9eac1902021-01-19 15:50:37 -07004251 file = NULL;
4252 ret = -EBADF;
4253 spin_lock(&files->file_lock);
4254 fdt = files_fdtable(files);
4255 if (close->fd >= fdt->max_fds) {
4256 spin_unlock(&files->file_lock);
4257 goto err;
4258 }
4259 file = fdt->fd[close->fd];
4260 if (!file) {
4261 spin_unlock(&files->file_lock);
4262 goto err;
4263 }
4264
4265 if (file->f_op == &io_uring_fops) {
4266 spin_unlock(&files->file_lock);
4267 file = NULL;
4268 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004269 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004270
4271 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004272 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004273 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004274 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004275 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004276
Jens Axboe9eac1902021-01-19 15:50:37 -07004277 ret = __close_fd_get_file(close->fd, &file);
4278 spin_unlock(&files->file_lock);
4279 if (ret < 0) {
4280 if (ret == -ENOENT)
4281 ret = -EBADF;
4282 goto err;
4283 }
4284
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004285 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004286 ret = filp_close(file, current->files);
4287err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004288 if (ret < 0)
4289 req_set_fail_links(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004290 if (file)
4291 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004292 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004293 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004294}
4295
Pavel Begunkov1155c762021-02-18 18:29:38 +00004296static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004297{
4298 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004299
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004300 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4301 return -EINVAL;
4302 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4303 return -EINVAL;
4304
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004305 req->sync.off = READ_ONCE(sqe->off);
4306 req->sync.len = READ_ONCE(sqe->len);
4307 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004308 return 0;
4309}
4310
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004311static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004312{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004313 int ret;
4314
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004315 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004316 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004317 return -EAGAIN;
4318
Jens Axboe9adbd452019-12-20 08:45:55 -07004319 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004320 req->sync.flags);
4321 if (ret < 0)
4322 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004323 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004324 return 0;
4325}
4326
YueHaibing469956e2020-03-04 15:53:52 +08004327#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004328static int io_setup_async_msg(struct io_kiocb *req,
4329 struct io_async_msghdr *kmsg)
4330{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004331 struct io_async_msghdr *async_msg = req->async_data;
4332
4333 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004334 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004335 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004336 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004337 return -ENOMEM;
4338 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004339 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004340 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004341 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004342 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004343 /* if were using fast_iov, set it to the new one */
4344 if (!async_msg->free_iov)
4345 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4346
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004347 return -EAGAIN;
4348}
4349
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004350static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4351 struct io_async_msghdr *iomsg)
4352{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004353 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004354 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004355 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004356 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004357}
4358
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004359static int io_sendmsg_prep_async(struct io_kiocb *req)
4360{
4361 int ret;
4362
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004363 ret = io_sendmsg_copy_hdr(req, req->async_data);
4364 if (!ret)
4365 req->flags |= REQ_F_NEED_CLEANUP;
4366 return ret;
4367}
4368
Jens Axboe3529d8c2019-12-19 18:24:38 -07004369static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004370{
Jens Axboee47293f2019-12-20 08:58:21 -07004371 struct io_sr_msg *sr = &req->sr_msg;
Jens Axboe03b12302019-12-02 18:50:25 -07004372
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004373 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4374 return -EINVAL;
4375
Jens Axboee47293f2019-12-20 08:58:21 -07004376 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004377 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004378 sr->len = READ_ONCE(sqe->len);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004379
Jens Axboed8768362020-02-27 14:17:49 -07004380#ifdef CONFIG_COMPAT
4381 if (req->ctx->compat)
4382 sr->msg_flags |= MSG_CMSG_COMPAT;
4383#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004384 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004385}
4386
Pavel Begunkov889fca72021-02-10 00:03:09 +00004387static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004388{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004389 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004390 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004391 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004392 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004393 int ret;
4394
Florent Revestdba4a922020-12-04 12:36:04 +01004395 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004396 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004397 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004398
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004399 kmsg = req->async_data;
4400 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004401 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004402 if (ret)
4403 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004404 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004405 }
4406
Stefan Metzmacher76cd9792021-03-16 16:33:27 +01004407 flags = req->sr_msg.msg_flags | MSG_NOSIGNAL;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004408 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;
4412
Stefan Metzmacher00312752021-03-20 20:33:36 +01004413 if (flags & MSG_WAITALL)
4414 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4415
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004416 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004417 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004418 return io_setup_async_msg(req, kmsg);
4419 if (ret == -ERESTARTSYS)
4420 ret = -EINTR;
4421
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004422 /* fast path, check for non-NULL to avoid function call */
4423 if (kmsg->free_iov)
4424 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004425 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004426 if (ret < min_ret)
Jens Axboefddafac2020-01-04 20:19:44 -07004427 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004428 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004429 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004430}
4431
Pavel Begunkov889fca72021-02-10 00:03:09 +00004432static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004433{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004434 struct io_sr_msg *sr = &req->sr_msg;
4435 struct msghdr msg;
4436 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004437 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004438 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004439 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004440 int ret;
4441
Florent Revestdba4a922020-12-04 12:36:04 +01004442 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004443 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004444 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004445
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004446 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4447 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004448 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004449
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004450 msg.msg_name = NULL;
4451 msg.msg_control = NULL;
4452 msg.msg_controllen = 0;
4453 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004454
Stefan Metzmacher76cd9792021-03-16 16:33:27 +01004455 flags = req->sr_msg.msg_flags | MSG_NOSIGNAL;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004456 if (flags & MSG_DONTWAIT)
4457 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004458 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004459 flags |= MSG_DONTWAIT;
Jens Axboe03b12302019-12-02 18:50:25 -07004460
Stefan Metzmacher00312752021-03-20 20:33:36 +01004461 if (flags & MSG_WAITALL)
4462 min_ret = iov_iter_count(&msg.msg_iter);
4463
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004464 msg.msg_flags = flags;
4465 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004466 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004467 return -EAGAIN;
4468 if (ret == -ERESTARTSYS)
4469 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004470
Stefan Metzmacher00312752021-03-20 20:33:36 +01004471 if (ret < min_ret)
Jens Axboe03b12302019-12-02 18:50:25 -07004472 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004473 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004474 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004475}
4476
Pavel Begunkov1400e692020-07-12 20:41:05 +03004477static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4478 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004479{
4480 struct io_sr_msg *sr = &req->sr_msg;
4481 struct iovec __user *uiov;
4482 size_t iov_len;
4483 int ret;
4484
Pavel Begunkov1400e692020-07-12 20:41:05 +03004485 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4486 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004487 if (ret)
4488 return ret;
4489
4490 if (req->flags & REQ_F_BUFFER_SELECT) {
4491 if (iov_len > 1)
4492 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004493 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004494 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004495 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004496 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004497 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004498 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004499 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004500 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004501 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004502 if (ret > 0)
4503 ret = 0;
4504 }
4505
4506 return ret;
4507}
4508
4509#ifdef CONFIG_COMPAT
4510static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004511 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004512{
4513 struct compat_msghdr __user *msg_compat;
4514 struct io_sr_msg *sr = &req->sr_msg;
4515 struct compat_iovec __user *uiov;
4516 compat_uptr_t ptr;
4517 compat_size_t len;
4518 int ret;
4519
Pavel Begunkov270a5942020-07-12 20:41:04 +03004520 msg_compat = (struct compat_msghdr __user *) sr->umsg;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004521 ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr,
Jens Axboe52de1fe2020-02-27 10:15:42 -07004522 &ptr, &len);
4523 if (ret)
4524 return ret;
4525
4526 uiov = compat_ptr(ptr);
4527 if (req->flags & REQ_F_BUFFER_SELECT) {
4528 compat_ssize_t clen;
4529
4530 if (len > 1)
4531 return -EINVAL;
4532 if (!access_ok(uiov, sizeof(*uiov)))
4533 return -EFAULT;
4534 if (__get_user(clen, &uiov->iov_len))
4535 return -EFAULT;
4536 if (clen < 0)
4537 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004538 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004539 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004540 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004541 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004542 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004543 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004544 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004545 if (ret < 0)
4546 return ret;
4547 }
4548
4549 return 0;
4550}
Jens Axboe03b12302019-12-02 18:50:25 -07004551#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004552
Pavel Begunkov1400e692020-07-12 20:41:05 +03004553static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4554 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004555{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004556 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004557
4558#ifdef CONFIG_COMPAT
4559 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004560 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004561#endif
4562
Pavel Begunkov1400e692020-07-12 20:41:05 +03004563 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004564}
4565
Jens Axboebcda7ba2020-02-23 16:42:51 -07004566static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004567 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004568{
4569 struct io_sr_msg *sr = &req->sr_msg;
4570 struct io_buffer *kbuf;
4571
Jens Axboebcda7ba2020-02-23 16:42:51 -07004572 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4573 if (IS_ERR(kbuf))
4574 return kbuf;
4575
4576 sr->kbuf = kbuf;
4577 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004578 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004579}
4580
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004581static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4582{
4583 return io_put_kbuf(req, req->sr_msg.kbuf);
4584}
4585
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004586static int io_recvmsg_prep_async(struct io_kiocb *req)
Jens Axboe03b12302019-12-02 18:50:25 -07004587{
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004588 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004589
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004590 ret = io_recvmsg_copy_hdr(req, req->async_data);
4591 if (!ret)
4592 req->flags |= REQ_F_NEED_CLEANUP;
4593 return ret;
4594}
4595
4596static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4597{
4598 struct io_sr_msg *sr = &req->sr_msg;
4599
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004600 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4601 return -EINVAL;
4602
Jens Axboe3529d8c2019-12-19 18:24:38 -07004603 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004604 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004605 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004606 sr->bgid = READ_ONCE(sqe->buf_group);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004607
Jens Axboed8768362020-02-27 14:17:49 -07004608#ifdef CONFIG_COMPAT
4609 if (req->ctx->compat)
4610 sr->msg_flags |= MSG_CMSG_COMPAT;
4611#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004612 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004613}
4614
Pavel Begunkov889fca72021-02-10 00:03:09 +00004615static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004616{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004617 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004618 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004619 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004620 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004621 int min_ret = 0;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004622 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004623 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004624
Florent Revestdba4a922020-12-04 12:36:04 +01004625 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004626 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004627 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004628
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004629 kmsg = req->async_data;
4630 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004631 ret = io_recvmsg_copy_hdr(req, &iomsg);
4632 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004633 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004634 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004635 }
4636
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004637 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004638 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004639 if (IS_ERR(kbuf))
4640 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004641 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004642 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4643 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004644 1, req->sr_msg.len);
4645 }
4646
Stefan Metzmacher76cd9792021-03-16 16:33:27 +01004647 flags = req->sr_msg.msg_flags | MSG_NOSIGNAL;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004648 if (flags & MSG_DONTWAIT)
4649 req->flags |= REQ_F_NOWAIT;
4650 else if (force_nonblock)
4651 flags |= MSG_DONTWAIT;
4652
Stefan Metzmacher00312752021-03-20 20:33:36 +01004653 if (flags & MSG_WAITALL)
4654 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4655
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004656 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4657 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004658 if (force_nonblock && ret == -EAGAIN)
4659 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004660 if (ret == -ERESTARTSYS)
4661 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004662
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004663 if (req->flags & REQ_F_BUFFER_SELECTED)
4664 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004665 /* fast path, check for non-NULL to avoid function call */
4666 if (kmsg->free_iov)
4667 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004668 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004669 if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004670 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004671 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004672 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004673}
4674
Pavel Begunkov889fca72021-02-10 00:03:09 +00004675static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004676{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004677 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004678 struct io_sr_msg *sr = &req->sr_msg;
4679 struct msghdr msg;
4680 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004681 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004682 struct iovec iov;
4683 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004684 int min_ret = 0;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004685 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004686 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004687
Florent Revestdba4a922020-12-04 12:36:04 +01004688 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004689 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004690 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004691
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004692 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004693 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004694 if (IS_ERR(kbuf))
4695 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004696 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004697 }
4698
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004699 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004700 if (unlikely(ret))
4701 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004702
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004703 msg.msg_name = NULL;
4704 msg.msg_control = NULL;
4705 msg.msg_controllen = 0;
4706 msg.msg_namelen = 0;
4707 msg.msg_iocb = NULL;
4708 msg.msg_flags = 0;
4709
Stefan Metzmacher76cd9792021-03-16 16:33:27 +01004710 flags = req->sr_msg.msg_flags | MSG_NOSIGNAL;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004711 if (flags & MSG_DONTWAIT)
4712 req->flags |= REQ_F_NOWAIT;
4713 else if (force_nonblock)
4714 flags |= MSG_DONTWAIT;
4715
Stefan Metzmacher00312752021-03-20 20:33:36 +01004716 if (flags & MSG_WAITALL)
4717 min_ret = iov_iter_count(&msg.msg_iter);
4718
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004719 ret = sock_recvmsg(sock, &msg, flags);
4720 if (force_nonblock && ret == -EAGAIN)
4721 return -EAGAIN;
4722 if (ret == -ERESTARTSYS)
4723 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004724out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004725 if (req->flags & REQ_F_BUFFER_SELECTED)
4726 cflags = io_put_recv_kbuf(req);
Stefan Metzmacher00312752021-03-20 20:33:36 +01004727 if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Jens Axboefddafac2020-01-04 20:19:44 -07004728 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004729 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07004730 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004731}
4732
Jens Axboe3529d8c2019-12-19 18:24:38 -07004733static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004734{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004735 struct io_accept *accept = &req->accept;
4736
Jens Axboe14587a462020-09-05 11:36:08 -06004737 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06004738 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004739 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004740 return -EINVAL;
4741
Jens Axboed55e5f52019-12-11 16:12:15 -07004742 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4743 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004744 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004745 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004746 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004747}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004748
Pavel Begunkov889fca72021-02-10 00:03:09 +00004749static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004750{
4751 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004752 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004753 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004754 int ret;
4755
Jiufei Xuee697dee2020-06-10 13:41:59 +08004756 if (req->file->f_flags & O_NONBLOCK)
4757 req->flags |= REQ_F_NOWAIT;
4758
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004759 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004760 accept->addr_len, accept->flags,
4761 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004762 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004763 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004764 if (ret < 0) {
4765 if (ret == -ERESTARTSYS)
4766 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004767 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004768 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00004769 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004770 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004771}
4772
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004773static int io_connect_prep_async(struct io_kiocb *req)
4774{
4775 struct io_async_connect *io = req->async_data;
4776 struct io_connect *conn = &req->connect;
4777
4778 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
4779}
4780
Jens Axboe3529d8c2019-12-19 18:24:38 -07004781static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004782{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004783 struct io_connect *conn = &req->connect;
Jens Axboef499a022019-12-02 16:28:46 -07004784
Jens Axboe14587a462020-09-05 11:36:08 -06004785 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004786 return -EINVAL;
4787 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4788 return -EINVAL;
4789
Jens Axboe3529d8c2019-12-19 18:24:38 -07004790 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4791 conn->addr_len = READ_ONCE(sqe->addr2);
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004792 return 0;
Jens Axboef499a022019-12-02 16:28:46 -07004793}
4794
Pavel Begunkov889fca72021-02-10 00:03:09 +00004795static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004796{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004797 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004798 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004799 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004800 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004801
Jens Axboee8c2bc12020-08-15 18:44:09 -07004802 if (req->async_data) {
4803 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004804 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004805 ret = move_addr_to_kernel(req->connect.addr,
4806 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004807 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004808 if (ret)
4809 goto out;
4810 io = &__io;
4811 }
4812
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004813 file_flags = force_nonblock ? O_NONBLOCK : 0;
4814
Jens Axboee8c2bc12020-08-15 18:44:09 -07004815 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004816 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004817 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004818 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004819 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004820 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004821 ret = -ENOMEM;
4822 goto out;
4823 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004824 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004825 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004826 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004827 if (ret == -ERESTARTSYS)
4828 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004829out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004830 if (ret < 0)
4831 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004832 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004833 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004834}
YueHaibing469956e2020-03-04 15:53:52 +08004835#else /* !CONFIG_NET */
Jens Axboe99a10082021-02-19 09:35:19 -07004836#define IO_NETOP_FN(op) \
4837static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \
4838{ \
4839 return -EOPNOTSUPP; \
Jens Axboef8e85cf2019-11-23 14:24:24 -07004840}
4841
Jens Axboe99a10082021-02-19 09:35:19 -07004842#define IO_NETOP_PREP(op) \
4843IO_NETOP_FN(op) \
4844static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \
4845{ \
4846 return -EOPNOTSUPP; \
4847} \
4848
4849#define IO_NETOP_PREP_ASYNC(op) \
4850IO_NETOP_PREP(op) \
4851static int io_##op##_prep_async(struct io_kiocb *req) \
4852{ \
4853 return -EOPNOTSUPP; \
YueHaibing469956e2020-03-04 15:53:52 +08004854}
4855
Jens Axboe99a10082021-02-19 09:35:19 -07004856IO_NETOP_PREP_ASYNC(sendmsg);
4857IO_NETOP_PREP_ASYNC(recvmsg);
4858IO_NETOP_PREP_ASYNC(connect);
4859IO_NETOP_PREP(accept);
4860IO_NETOP_FN(send);
4861IO_NETOP_FN(recv);
YueHaibing469956e2020-03-04 15:53:52 +08004862#endif /* CONFIG_NET */
Jens Axboe17f2fe32019-10-17 14:42:58 -06004863
Jens Axboed7718a92020-02-14 22:23:12 -07004864struct io_poll_table {
4865 struct poll_table_struct pt;
4866 struct io_kiocb *req;
4867 int error;
4868};
4869
Jens Axboed7718a92020-02-14 22:23:12 -07004870static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4871 __poll_t mask, task_work_func_t func)
4872{
Jens Axboeaa96bf82020-04-03 11:26:26 -06004873 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004874
4875 /* for instances that support it check for an event match first: */
4876 if (mask && !(mask & poll->events))
4877 return 0;
4878
4879 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4880
4881 list_del_init(&poll->wait.entry);
4882
Jens Axboed7718a92020-02-14 22:23:12 -07004883 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00004884 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06004885
Jens Axboed7718a92020-02-14 22:23:12 -07004886 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004887 * If this fails, then the task is exiting. When a task exits, the
4888 * work gets canceled, so just cancel this request as well instead
4889 * of executing it. We can't safely execute it anyway, as we may not
4890 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004891 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06004892 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004893 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06004894 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00004895 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004896 }
Jens Axboed7718a92020-02-14 22:23:12 -07004897 return 1;
4898}
4899
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004900static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4901 __acquires(&req->ctx->completion_lock)
4902{
4903 struct io_ring_ctx *ctx = req->ctx;
4904
4905 if (!req->result && !READ_ONCE(poll->canceled)) {
4906 struct poll_table_struct pt = { ._key = poll->events };
4907
4908 req->result = vfs_poll(req->file, &pt) & poll->events;
4909 }
4910
4911 spin_lock_irq(&ctx->completion_lock);
4912 if (!req->result && !READ_ONCE(poll->canceled)) {
4913 add_wait_queue(poll->head, &poll->wait);
4914 return true;
4915 }
4916
4917 return false;
4918}
4919
Jens Axboed4e7cd32020-08-15 11:44:50 -07004920static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004921{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004922 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004923 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004924 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004925 return req->apoll->double_poll;
4926}
4927
4928static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4929{
4930 if (req->opcode == IORING_OP_POLL_ADD)
4931 return &req->poll;
4932 return &req->apoll->poll;
4933}
4934
4935static void io_poll_remove_double(struct io_kiocb *req)
4936{
4937 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004938
4939 lockdep_assert_held(&req->ctx->completion_lock);
4940
4941 if (poll && poll->head) {
4942 struct wait_queue_head *head = poll->head;
4943
4944 spin_lock(&head->lock);
4945 list_del_init(&poll->wait.entry);
4946 if (poll->wait.private)
Jens Axboede9b4cc2021-02-24 13:28:27 -07004947 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004948 poll->head = NULL;
4949 spin_unlock(&head->lock);
4950 }
4951}
4952
4953static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
4954{
4955 struct io_ring_ctx *ctx = req->ctx;
4956
Jens Axboe45ab03b2021-02-23 08:19:33 -07004957 if (!error && req->poll.canceled)
4958 error = -ECANCELED;
4959
Jens Axboed4e7cd32020-08-15 11:44:50 -07004960 io_poll_remove_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004961 req->poll.done = true;
4962 io_cqring_fill_event(req, error ? error : mangle_poll(mask));
4963 io_commit_cqring(ctx);
4964}
4965
Jens Axboe18bceab2020-05-15 11:56:54 -06004966static void io_poll_task_func(struct callback_head *cb)
4967{
4968 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004969 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004970 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004971
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004972 if (io_poll_rewait(req, &req->poll)) {
4973 spin_unlock_irq(&ctx->completion_lock);
4974 } else {
4975 hash_del(&req->hash_node);
4976 io_poll_complete(req, req->result, 0);
4977 spin_unlock_irq(&ctx->completion_lock);
4978
4979 nxt = io_put_req_find_next(req);
4980 io_cqring_ev_posted(ctx);
4981 if (nxt)
4982 __io_req_task_submit(nxt);
4983 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004984}
4985
4986static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4987 int sync, void *key)
4988{
4989 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004990 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004991 __poll_t mask = key_to_poll(key);
4992
4993 /* for instances that support it check for an event match first: */
4994 if (mask && !(mask & poll->events))
4995 return 0;
4996
Jens Axboe8706e042020-09-28 08:38:54 -06004997 list_del_init(&wait->entry);
4998
Jens Axboe807abcb2020-07-17 17:09:27 -06004999 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005000 bool done;
5001
Jens Axboe807abcb2020-07-17 17:09:27 -06005002 spin_lock(&poll->head->lock);
5003 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06005004 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06005005 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005006 /* make sure double remove sees this as being gone */
5007 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06005008 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06005009 if (!done) {
5010 /* use wait func handler, so it matches the rq type */
5011 poll->wait.func(&poll->wait, mode, sync, key);
5012 }
Jens Axboe18bceab2020-05-15 11:56:54 -06005013 }
Jens Axboede9b4cc2021-02-24 13:28:27 -07005014 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005015 return 1;
5016}
5017
5018static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
5019 wait_queue_func_t wake_func)
5020{
5021 poll->head = NULL;
5022 poll->done = false;
5023 poll->canceled = false;
5024 poll->events = events;
5025 INIT_LIST_HEAD(&poll->wait.entry);
5026 init_waitqueue_func_entry(&poll->wait, wake_func);
5027}
5028
5029static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06005030 struct wait_queue_head *head,
5031 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06005032{
5033 struct io_kiocb *req = pt->req;
5034
5035 /*
5036 * If poll->head is already set, it's because the file being polled
5037 * uses multiple waitqueues for poll handling (eg one for read, one
5038 * for write). Setup a separate io_poll_iocb if this happens.
5039 */
5040 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01005041 struct io_poll_iocb *poll_one = poll;
5042
Jens Axboe18bceab2020-05-15 11:56:54 -06005043 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06005044 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005045 pt->error = -EINVAL;
5046 return;
5047 }
Jens Axboe1c3b3e62021-02-28 16:07:30 -07005048 /* double add on the same waitqueue head, ignore */
5049 if (poll->head == head)
5050 return;
Jens Axboe18bceab2020-05-15 11:56:54 -06005051 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
5052 if (!poll) {
5053 pt->error = -ENOMEM;
5054 return;
5055 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01005056 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboede9b4cc2021-02-24 13:28:27 -07005057 req_ref_get(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005058 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06005059 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005060 }
5061
5062 pt->error = 0;
5063 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005064
5065 if (poll->events & EPOLLEXCLUSIVE)
5066 add_wait_queue_exclusive(head, &poll->wait);
5067 else
5068 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005069}
5070
5071static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5072 struct poll_table_struct *p)
5073{
5074 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005075 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005076
Jens Axboe807abcb2020-07-17 17:09:27 -06005077 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005078}
5079
Jens Axboed7718a92020-02-14 22:23:12 -07005080static void io_async_task_func(struct callback_head *cb)
5081{
5082 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5083 struct async_poll *apoll = req->apoll;
5084 struct io_ring_ctx *ctx = req->ctx;
5085
5086 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5087
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005088 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005089 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005090 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005091 }
5092
Jens Axboe31067252020-05-17 17:43:31 -06005093 /* If req is still hashed, it cannot have been canceled. Don't check. */
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005094 if (hash_hashed(&req->hash_node))
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005095 hash_del(&req->hash_node);
Jens Axboe2bae0472020-04-13 11:16:34 -06005096
Jens Axboed4e7cd32020-08-15 11:44:50 -07005097 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005098 spin_unlock_irq(&ctx->completion_lock);
5099
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005100 if (!READ_ONCE(apoll->poll.canceled))
5101 __io_req_task_submit(req);
5102 else
5103 __io_req_task_cancel(req, -ECANCELED);
Dan Carpenteraa340842020-07-08 21:47:11 +03005104
Jens Axboe807abcb2020-07-17 17:09:27 -06005105 kfree(apoll->double_poll);
Jens Axboe31067252020-05-17 17:43:31 -06005106 kfree(apoll);
Jens Axboed7718a92020-02-14 22:23:12 -07005107}
5108
5109static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5110 void *key)
5111{
5112 struct io_kiocb *req = wait->private;
5113 struct io_poll_iocb *poll = &req->apoll->poll;
5114
5115 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5116 key_to_poll(key));
5117
5118 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5119}
5120
5121static void io_poll_req_insert(struct io_kiocb *req)
5122{
5123 struct io_ring_ctx *ctx = req->ctx;
5124 struct hlist_head *list;
5125
5126 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5127 hlist_add_head(&req->hash_node, list);
5128}
5129
5130static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5131 struct io_poll_iocb *poll,
5132 struct io_poll_table *ipt, __poll_t mask,
5133 wait_queue_func_t wake_func)
5134 __acquires(&ctx->completion_lock)
5135{
5136 struct io_ring_ctx *ctx = req->ctx;
5137 bool cancel = false;
5138
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005139 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005140 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005141 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005142 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005143
5144 ipt->pt._key = mask;
5145 ipt->req = req;
5146 ipt->error = -EINVAL;
5147
Jens Axboed7718a92020-02-14 22:23:12 -07005148 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5149
5150 spin_lock_irq(&ctx->completion_lock);
5151 if (likely(poll->head)) {
5152 spin_lock(&poll->head->lock);
5153 if (unlikely(list_empty(&poll->wait.entry))) {
5154 if (ipt->error)
5155 cancel = true;
5156 ipt->error = 0;
5157 mask = 0;
5158 }
5159 if (mask || ipt->error)
5160 list_del_init(&poll->wait.entry);
5161 else if (cancel)
5162 WRITE_ONCE(poll->canceled, true);
5163 else if (!poll->done) /* actually waiting for an event */
5164 io_poll_req_insert(req);
5165 spin_unlock(&poll->head->lock);
5166 }
5167
5168 return mask;
5169}
5170
5171static bool io_arm_poll_handler(struct io_kiocb *req)
5172{
5173 const struct io_op_def *def = &io_op_defs[req->opcode];
5174 struct io_ring_ctx *ctx = req->ctx;
5175 struct async_poll *apoll;
5176 struct io_poll_table ipt;
5177 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005178 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005179
5180 if (!req->file || !file_can_poll(req->file))
5181 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005182 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005183 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005184 if (def->pollin)
5185 rw = READ;
5186 else if (def->pollout)
5187 rw = WRITE;
5188 else
5189 return false;
5190 /* if we can't nonblock try, then no point in arming a poll handler */
Jens Axboe7b29f922021-03-12 08:30:14 -07005191 if (!io_file_supports_async(req, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005192 return false;
5193
5194 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5195 if (unlikely(!apoll))
5196 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005197 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005198
5199 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005200 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005201
Nathan Chancellor8755d972020-03-02 16:01:19 -07005202 mask = 0;
Jens Axboed7718a92020-02-14 22:23:12 -07005203 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005204 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005205 if (def->pollout)
5206 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005207
5208 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5209 if ((req->opcode == IORING_OP_RECVMSG) &&
5210 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5211 mask &= ~POLLIN;
5212
Jens Axboed7718a92020-02-14 22:23:12 -07005213 mask |= POLLERR | POLLPRI;
5214
5215 ipt.pt._qproc = io_async_queue_proc;
5216
5217 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5218 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005219 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005220 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005221 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe807abcb2020-07-17 17:09:27 -06005222 kfree(apoll->double_poll);
Jens Axboed7718a92020-02-14 22:23:12 -07005223 kfree(apoll);
5224 return false;
5225 }
5226 spin_unlock_irq(&ctx->completion_lock);
5227 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5228 apoll->poll.events);
5229 return true;
5230}
5231
5232static bool __io_poll_remove_one(struct io_kiocb *req,
5233 struct io_poll_iocb *poll)
5234{
Jens Axboeb41e9852020-02-17 09:52:41 -07005235 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005236
5237 spin_lock(&poll->head->lock);
5238 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005239 if (!list_empty(&poll->wait.entry)) {
5240 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005241 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005242 }
5243 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005244 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005245 return do_complete;
5246}
5247
5248static bool io_poll_remove_one(struct io_kiocb *req)
5249{
5250 bool do_complete;
5251
Jens Axboed4e7cd32020-08-15 11:44:50 -07005252 io_poll_remove_double(req);
5253
Jens Axboed7718a92020-02-14 22:23:12 -07005254 if (req->opcode == IORING_OP_POLL_ADD) {
5255 do_complete = __io_poll_remove_one(req, &req->poll);
5256 } else {
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005257 struct async_poll *apoll = req->apoll;
5258
Jens Axboed7718a92020-02-14 22:23:12 -07005259 /* non-poll requests have submit ref still */
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005260 do_complete = __io_poll_remove_one(req, &apoll->poll);
5261 if (do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005262 io_put_req(req);
Jens Axboe807abcb2020-07-17 17:09:27 -06005263 kfree(apoll->double_poll);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005264 kfree(apoll);
5265 }
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005266 }
5267
Jens Axboeb41e9852020-02-17 09:52:41 -07005268 if (do_complete) {
5269 io_cqring_fill_event(req, -ECANCELED);
5270 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005271 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005272 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005273 }
5274
5275 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005276}
5277
Jens Axboe76e1b642020-09-26 15:05:03 -06005278/*
5279 * Returns true if we found and killed one or more poll requests
5280 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005281static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
5282 struct files_struct *files)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005283{
Jens Axboe78076bb2019-12-04 19:56:40 -07005284 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005285 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005286 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005287
5288 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005289 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5290 struct hlist_head *list;
5291
5292 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005293 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00005294 if (io_match_task(req, tsk, files))
Jens Axboef3606e32020-09-22 08:18:24 -06005295 posted += io_poll_remove_one(req);
5296 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005297 }
5298 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005299
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005300 if (posted)
5301 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005302
5303 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005304}
5305
Jens Axboe47f46762019-11-09 17:43:02 -07005306static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
5307{
Jens Axboe78076bb2019-12-04 19:56:40 -07005308 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005309 struct io_kiocb *req;
5310
Jens Axboe78076bb2019-12-04 19:56:40 -07005311 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5312 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005313 if (sqe_addr != req->user_data)
5314 continue;
5315 if (io_poll_remove_one(req))
Jens Axboeeac406c2019-11-14 12:09:58 -07005316 return 0;
Jens Axboeb41e9852020-02-17 09:52:41 -07005317 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005318 }
5319
5320 return -ENOENT;
5321}
5322
Jens Axboe3529d8c2019-12-19 18:24:38 -07005323static int io_poll_remove_prep(struct io_kiocb *req,
5324 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005325{
Jens Axboe221c5eb2019-01-17 09:41:58 -07005326 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5327 return -EINVAL;
5328 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
5329 sqe->poll_events)
5330 return -EINVAL;
5331
Pavel Begunkov018043b2020-10-27 23:17:18 +00005332 req->poll_remove.addr = READ_ONCE(sqe->addr);
Jens Axboe0969e782019-12-17 18:40:57 -07005333 return 0;
5334}
5335
5336/*
5337 * Find a running poll command that matches one specified in sqe->addr,
5338 * and remove it if found.
5339 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005340static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005341{
5342 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe0969e782019-12-17 18:40:57 -07005343 int ret;
5344
Jens Axboe221c5eb2019-01-17 09:41:58 -07005345 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov018043b2020-10-27 23:17:18 +00005346 ret = io_poll_cancel(ctx, req->poll_remove.addr);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005347 spin_unlock_irq(&ctx->completion_lock);
5348
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005349 if (ret < 0)
5350 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06005351 io_req_complete(req, ret);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005352 return 0;
5353}
5354
Jens Axboe221c5eb2019-01-17 09:41:58 -07005355static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5356 void *key)
5357{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005358 struct io_kiocb *req = wait->private;
5359 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005360
Jens Axboed7718a92020-02-14 22:23:12 -07005361 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005362}
5363
Jens Axboe221c5eb2019-01-17 09:41:58 -07005364static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5365 struct poll_table_struct *p)
5366{
5367 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5368
Jens Axboee8c2bc12020-08-15 18:44:09 -07005369 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c2019-11-14 12:09:58 -07005370}
5371
Jens Axboe3529d8c2019-12-19 18:24:38 -07005372static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005373{
5374 struct io_poll_iocb *poll = &req->poll;
Jiufei Xue5769a352020-06-17 17:53:55 +08005375 u32 events;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005376
5377 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5378 return -EINVAL;
5379 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
5380 return -EINVAL;
5381
Jiufei Xue5769a352020-06-17 17:53:55 +08005382 events = READ_ONCE(sqe->poll32_events);
5383#ifdef __BIG_ENDIAN
5384 events = swahw32(events);
5385#endif
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005386 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP |
5387 (events & EPOLLEXCLUSIVE);
Jens Axboe0969e782019-12-17 18:40:57 -07005388 return 0;
5389}
5390
Pavel Begunkov61e98202021-02-10 00:03:08 +00005391static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005392{
5393 struct io_poll_iocb *poll = &req->poll;
5394 struct io_ring_ctx *ctx = req->ctx;
5395 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005396 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005397
Jens Axboed7718a92020-02-14 22:23:12 -07005398 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005399
Jens Axboed7718a92020-02-14 22:23:12 -07005400 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5401 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005402
Jens Axboe8c838782019-03-12 15:48:16 -06005403 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005404 ipt.error = 0;
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005405 io_poll_complete(req, mask, 0);
Jens Axboe8c838782019-03-12 15:48:16 -06005406 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005407 spin_unlock_irq(&ctx->completion_lock);
5408
Jens Axboe8c838782019-03-12 15:48:16 -06005409 if (mask) {
5410 io_cqring_ev_posted(ctx);
Pavel Begunkov014db002020-03-03 21:33:12 +03005411 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005412 }
Jens Axboe8c838782019-03-12 15:48:16 -06005413 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005414}
5415
Jens Axboe5262f562019-09-17 12:26:57 -06005416static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5417{
Jens Axboead8a48a2019-11-15 08:49:11 -07005418 struct io_timeout_data *data = container_of(timer,
5419 struct io_timeout_data, timer);
5420 struct io_kiocb *req = data->req;
5421 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005422 unsigned long flags;
5423
Jens Axboe5262f562019-09-17 12:26:57 -06005424 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005425 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005426 atomic_set(&req->ctx->cq_timeouts,
5427 atomic_read(&req->ctx->cq_timeouts) + 1);
5428
Jens Axboe78e19bb2019-11-06 15:21:34 -07005429 io_cqring_fill_event(req, -ETIME);
Jens Axboe5262f562019-09-17 12:26:57 -06005430 io_commit_cqring(ctx);
5431 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5432
5433 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005434 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005435 io_put_req(req);
5436 return HRTIMER_NORESTART;
5437}
5438
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005439static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5440 __u64 user_data)
Jens Axboe47f46762019-11-09 17:43:02 -07005441{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005442 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005443 struct io_kiocb *req;
5444 int ret = -ENOENT;
5445
5446 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
5447 if (user_data == req->user_data) {
5448 ret = 0;
5449 break;
5450 }
5451 }
5452
5453 if (ret == -ENOENT)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005454 return ERR_PTR(ret);
Jens Axboef254ac02020-08-12 17:33:30 -06005455
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005456 io = req->async_data;
5457 ret = hrtimer_try_to_cancel(&io->timer);
5458 if (ret == -1)
5459 return ERR_PTR(-EALREADY);
5460 list_del_init(&req->timeout.list);
5461 return req;
5462}
5463
5464static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
5465{
5466 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5467
5468 if (IS_ERR(req))
5469 return PTR_ERR(req);
5470
5471 req_set_fail_links(req);
5472 io_cqring_fill_event(req, -ECANCELED);
5473 io_put_req_deferred(req, 1);
5474 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005475}
5476
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005477static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5478 struct timespec64 *ts, enum hrtimer_mode mode)
5479{
5480 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5481 struct io_timeout_data *data;
5482
5483 if (IS_ERR(req))
5484 return PTR_ERR(req);
5485
5486 req->timeout.off = 0; /* noseq */
5487 data = req->async_data;
5488 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5489 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5490 data->timer.function = io_timeout_fn;
5491 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5492 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005493}
5494
Jens Axboe3529d8c2019-12-19 18:24:38 -07005495static int io_timeout_remove_prep(struct io_kiocb *req,
5496 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005497{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005498 struct io_timeout_rem *tr = &req->timeout_rem;
5499
Jens Axboeb29472e2019-12-17 18:50:29 -07005500 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5501 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005502 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5503 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005504 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005505 return -EINVAL;
5506
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005507 tr->addr = READ_ONCE(sqe->addr);
5508 tr->flags = READ_ONCE(sqe->timeout_flags);
5509 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5510 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5511 return -EINVAL;
5512 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5513 return -EFAULT;
5514 } else if (tr->flags) {
5515 /* timeout removal doesn't support flags */
5516 return -EINVAL;
5517 }
5518
Jens Axboeb29472e2019-12-17 18:50:29 -07005519 return 0;
5520}
5521
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005522static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5523{
5524 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5525 : HRTIMER_MODE_REL;
5526}
5527
Jens Axboe11365042019-10-16 09:08:32 -06005528/*
5529 * Remove or update an existing timeout command
5530 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005531static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005532{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005533 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005534 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005535 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005536
Jens Axboe11365042019-10-16 09:08:32 -06005537 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005538 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005539 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005540 else
5541 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5542 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005543
Jens Axboe47f46762019-11-09 17:43:02 -07005544 io_cqring_fill_event(req, ret);
Jens Axboe11365042019-10-16 09:08:32 -06005545 io_commit_cqring(ctx);
5546 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005547 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005548 if (ret < 0)
5549 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005550 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005551 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005552}
5553
Jens Axboe3529d8c2019-12-19 18:24:38 -07005554static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005555 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005556{
Jens Axboead8a48a2019-11-15 08:49:11 -07005557 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005558 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005559 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005560
Jens Axboead8a48a2019-11-15 08:49:11 -07005561 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005562 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005563 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005564 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005565 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005566 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005567 flags = READ_ONCE(sqe->timeout_flags);
5568 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005569 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005570
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005571 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005572
Jens Axboee8c2bc12020-08-15 18:44:09 -07005573 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005574 return -ENOMEM;
5575
Jens Axboee8c2bc12020-08-15 18:44:09 -07005576 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005577 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005578
5579 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005580 return -EFAULT;
5581
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005582 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005583 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
Pavel Begunkov2482b582021-03-25 18:32:44 +00005584 if (is_timeout_link)
5585 io_req_track_inflight(req);
Jens Axboead8a48a2019-11-15 08:49:11 -07005586 return 0;
5587}
5588
Pavel Begunkov61e98202021-02-10 00:03:08 +00005589static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005590{
Jens Axboead8a48a2019-11-15 08:49:11 -07005591 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005592 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005593 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005594 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005595
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005596 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005597
Jens Axboe5262f562019-09-17 12:26:57 -06005598 /*
5599 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005600 * timeout event to be satisfied. If it isn't set, then this is
5601 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005602 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005603 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005604 entry = ctx->timeout_list.prev;
5605 goto add;
5606 }
Jens Axboe5262f562019-09-17 12:26:57 -06005607
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005608 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5609 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005610
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005611 /* Update the last seq here in case io_flush_timeouts() hasn't.
5612 * This is safe because ->completion_lock is held, and submissions
5613 * and completions are never mixed in the same ->completion_lock section.
5614 */
5615 ctx->cq_last_tm_flush = tail;
5616
Jens Axboe5262f562019-09-17 12:26:57 -06005617 /*
5618 * Insertion sort, ensuring the first entry in the list is always
5619 * the one we need first.
5620 */
Jens Axboe5262f562019-09-17 12:26:57 -06005621 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005622 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5623 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005624
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005625 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005626 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005627 /* nxt.seq is behind @tail, otherwise would've been completed */
5628 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005629 break;
5630 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005631add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005632 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005633 data->timer.function = io_timeout_fn;
5634 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005635 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005636 return 0;
5637}
5638
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005639struct io_cancel_data {
5640 struct io_ring_ctx *ctx;
5641 u64 user_data;
5642};
5643
Jens Axboe62755e32019-10-28 21:49:21 -06005644static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005645{
Jens Axboe62755e32019-10-28 21:49:21 -06005646 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005647 struct io_cancel_data *cd = data;
Jens Axboede0617e2019-04-06 21:51:27 -06005648
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005649 return req->ctx == cd->ctx && req->user_data == cd->user_data;
Jens Axboe62755e32019-10-28 21:49:21 -06005650}
5651
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005652static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data,
5653 struct io_ring_ctx *ctx)
Jens Axboe62755e32019-10-28 21:49:21 -06005654{
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005655 struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, };
Jens Axboe62755e32019-10-28 21:49:21 -06005656 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005657 int ret = 0;
5658
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005659 if (!tctx || !tctx->io_wq)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005660 return -ENOENT;
5661
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005662 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005663 switch (cancel_ret) {
5664 case IO_WQ_CANCEL_OK:
5665 ret = 0;
5666 break;
5667 case IO_WQ_CANCEL_RUNNING:
5668 ret = -EALREADY;
5669 break;
5670 case IO_WQ_CANCEL_NOTFOUND:
5671 ret = -ENOENT;
5672 break;
5673 }
5674
Jens Axboee977d6d2019-11-05 12:39:45 -07005675 return ret;
5676}
5677
Jens Axboe47f46762019-11-09 17:43:02 -07005678static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5679 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005680 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005681{
5682 unsigned long flags;
5683 int ret;
5684
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005685 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
Jens Axboe47f46762019-11-09 17:43:02 -07005686 if (ret != -ENOENT) {
5687 spin_lock_irqsave(&ctx->completion_lock, flags);
5688 goto done;
5689 }
5690
5691 spin_lock_irqsave(&ctx->completion_lock, flags);
5692 ret = io_timeout_cancel(ctx, sqe_addr);
5693 if (ret != -ENOENT)
5694 goto done;
5695 ret = io_poll_cancel(ctx, sqe_addr);
5696done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005697 if (!ret)
5698 ret = success_ret;
Jens Axboe47f46762019-11-09 17:43:02 -07005699 io_cqring_fill_event(req, ret);
5700 io_commit_cqring(ctx);
5701 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5702 io_cqring_ev_posted(ctx);
5703
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005704 if (ret < 0)
5705 req_set_fail_links(req);
Pavel Begunkov014db002020-03-03 21:33:12 +03005706 io_put_req(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005707}
5708
Jens Axboe3529d8c2019-12-19 18:24:38 -07005709static int io_async_cancel_prep(struct io_kiocb *req,
5710 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005711{
Jens Axboefbf23842019-12-17 18:45:56 -07005712 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005713 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005714 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5715 return -EINVAL;
5716 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005717 return -EINVAL;
5718
Jens Axboefbf23842019-12-17 18:45:56 -07005719 req->cancel.addr = READ_ONCE(sqe->addr);
5720 return 0;
5721}
5722
Pavel Begunkov61e98202021-02-10 00:03:08 +00005723static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07005724{
5725 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov58f99372021-03-12 16:25:55 +00005726 u64 sqe_addr = req->cancel.addr;
5727 struct io_tctx_node *node;
5728 int ret;
Jens Axboefbf23842019-12-17 18:45:56 -07005729
Pavel Begunkov58f99372021-03-12 16:25:55 +00005730 /* tasks should wait for their io-wq threads, so safe w/o sync */
5731 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
5732 spin_lock_irq(&ctx->completion_lock);
5733 if (ret != -ENOENT)
5734 goto done;
5735 ret = io_timeout_cancel(ctx, sqe_addr);
5736 if (ret != -ENOENT)
5737 goto done;
5738 ret = io_poll_cancel(ctx, sqe_addr);
5739 if (ret != -ENOENT)
5740 goto done;
5741 spin_unlock_irq(&ctx->completion_lock);
5742
5743 /* slow path, try all io-wq's */
5744 io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5745 ret = -ENOENT;
5746 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
5747 struct io_uring_task *tctx = node->task->io_uring;
5748
5749 if (!tctx || !tctx->io_wq)
5750 continue;
5751 ret = io_async_cancel_one(tctx, req->cancel.addr, ctx);
5752 if (ret != -ENOENT)
5753 break;
5754 }
5755 io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5756
5757 spin_lock_irq(&ctx->completion_lock);
5758done:
5759 io_cqring_fill_event(req, ret);
5760 io_commit_cqring(ctx);
5761 spin_unlock_irq(&ctx->completion_lock);
5762 io_cqring_ev_posted(ctx);
5763
5764 if (ret < 0)
5765 req_set_fail_links(req);
5766 io_put_req(req);
Jens Axboe62755e32019-10-28 21:49:21 -06005767 return 0;
5768}
5769
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005770static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07005771 const struct io_uring_sqe *sqe)
5772{
Jens Axboe6ca56f82020-09-18 16:51:19 -06005773 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
5774 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005775 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5776 return -EINVAL;
5777 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005778 return -EINVAL;
5779
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005780 req->rsrc_update.offset = READ_ONCE(sqe->off);
5781 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
5782 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005783 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005784 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005785 return 0;
5786}
5787
Pavel Begunkov889fca72021-02-10 00:03:09 +00005788static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005789{
5790 struct io_ring_ctx *ctx = req->ctx;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005791 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005792 int ret;
5793
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005794 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005795 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005796
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005797 up.offset = req->rsrc_update.offset;
5798 up.data = req->rsrc_update.arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005799
5800 mutex_lock(&ctx->uring_lock);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005801 ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005802 mutex_unlock(&ctx->uring_lock);
5803
5804 if (ret < 0)
5805 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005806 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005807 return 0;
5808}
5809
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005810static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005811{
Jens Axboed625c6e2019-12-17 19:53:05 -07005812 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005813 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005814 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005815 case IORING_OP_READV:
5816 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005817 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005818 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005819 case IORING_OP_WRITEV:
5820 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005821 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005822 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005823 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005824 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005825 case IORING_OP_POLL_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005826 return io_poll_remove_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005827 case IORING_OP_FSYNC:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005828 return io_fsync_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005829 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005830 return io_sfr_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005831 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005832 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005833 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005834 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005835 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005836 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005837 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005838 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005839 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005840 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005841 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005842 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005843 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005844 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005845 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005846 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005847 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005848 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005849 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005850 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005851 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005852 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005853 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005854 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005855 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005856 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005857 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005858 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005859 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005860 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005861 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005862 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005863 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005864 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005865 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005866 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005867 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005868 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005869 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005870 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005871 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005872 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005873 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005874 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06005875 case IORING_OP_SHUTDOWN:
5876 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06005877 case IORING_OP_RENAMEAT:
5878 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06005879 case IORING_OP_UNLINKAT:
5880 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005881 }
5882
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005883 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5884 req->opcode);
5885 return-EINVAL;
5886}
5887
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005888static int io_req_prep_async(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005889{
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005890 if (!io_op_defs[req->opcode].needs_async_setup)
5891 return 0;
5892 if (WARN_ON_ONCE(req->async_data))
5893 return -EFAULT;
5894 if (io_alloc_async_data(req))
5895 return -EAGAIN;
5896
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005897 switch (req->opcode) {
5898 case IORING_OP_READV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005899 return io_rw_prep_async(req, READ);
5900 case IORING_OP_WRITEV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005901 return io_rw_prep_async(req, WRITE);
5902 case IORING_OP_SENDMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005903 return io_sendmsg_prep_async(req);
5904 case IORING_OP_RECVMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005905 return io_recvmsg_prep_async(req);
5906 case IORING_OP_CONNECT:
5907 return io_connect_prep_async(req);
5908 }
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005909 printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n",
5910 req->opcode);
5911 return -EFAULT;
Jens Axboedef596e2019-01-09 08:59:42 -07005912}
5913
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005914static u32 io_get_sequence(struct io_kiocb *req)
5915{
5916 struct io_kiocb *pos;
5917 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005918 u32 total_submitted, nr_reqs = 0;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005919
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005920 io_for_each_link(pos, req)
5921 nr_reqs++;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005922
5923 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
5924 return total_submitted - nr_reqs;
5925}
5926
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005927static int io_req_defer(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005928{
5929 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005930 struct io_defer_entry *de;
Jens Axboedef596e2019-01-09 08:59:42 -07005931 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005932 u32 seq;
Jens Axboedef596e2019-01-09 08:59:42 -07005933
5934 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005935 if (likely(list_empty_careful(&ctx->defer_list) &&
5936 !(req->flags & REQ_F_IO_DRAIN)))
5937 return 0;
5938
5939 seq = io_get_sequence(req);
5940 /* Still a chance to pass the sequence check */
5941 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboedef596e2019-01-09 08:59:42 -07005942 return 0;
5943
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005944 ret = io_req_prep_async(req);
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005945 if (ret)
5946 return ret;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03005947 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005948 de = kmalloc(sizeof(*de), GFP_KERNEL);
5949 if (!de)
5950 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07005951
5952 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005953 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07005954 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005955 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03005956 io_queue_async_work(req);
5957 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07005958 }
5959
5960 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005961 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005962 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005963 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07005964 spin_unlock_irq(&ctx->completion_lock);
5965 return -EIOCBQUEUED;
5966}
5967
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03005968static void __io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005969{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005970 if (req->flags & REQ_F_BUFFER_SELECTED) {
5971 switch (req->opcode) {
5972 case IORING_OP_READV:
5973 case IORING_OP_READ_FIXED:
5974 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07005975 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005976 break;
5977 case IORING_OP_RECVMSG:
5978 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07005979 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005980 break;
5981 }
5982 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005983 }
5984
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005985 if (req->flags & REQ_F_NEED_CLEANUP) {
5986 switch (req->opcode) {
5987 case IORING_OP_READV:
5988 case IORING_OP_READ_FIXED:
5989 case IORING_OP_READ:
5990 case IORING_OP_WRITEV:
5991 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005992 case IORING_OP_WRITE: {
5993 struct io_async_rw *io = req->async_data;
5994 if (io->free_iovec)
5995 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005996 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005997 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005998 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005999 case IORING_OP_SENDMSG: {
6000 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00006001
6002 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006003 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006004 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006005 case IORING_OP_SPLICE:
6006 case IORING_OP_TEE:
6007 io_put_file(req, req->splice.file_in,
6008 (req->splice.flags & SPLICE_F_FD_IN_FIXED));
6009 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06006010 case IORING_OP_OPENAT:
6011 case IORING_OP_OPENAT2:
6012 if (req->open.filename)
6013 putname(req->open.filename);
6014 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006015 case IORING_OP_RENAMEAT:
6016 putname(req->rename.oldpath);
6017 putname(req->rename.newpath);
6018 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006019 case IORING_OP_UNLINKAT:
6020 putname(req->unlink.filename);
6021 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006022 }
6023 req->flags &= ~REQ_F_NEED_CLEANUP;
6024 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006025}
6026
Pavel Begunkov889fca72021-02-10 00:03:09 +00006027static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07006028{
Jens Axboeedafcce2019-01-09 09:16:05 -07006029 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5730b272021-02-27 15:57:30 -07006030 const struct cred *creds = NULL;
Jens Axboed625c6e2019-12-17 19:53:05 -07006031 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07006032
Jens Axboe003e8dc2021-03-06 09:22:27 -07006033 if (req->work.creds && req->work.creds != current_cred())
6034 creds = override_creds(req->work.creds);
Jens Axboe5730b272021-02-27 15:57:30 -07006035
Jens Axboed625c6e2019-12-17 19:53:05 -07006036 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07006037 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006038 ret = io_nop(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006039 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006040 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07006041 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006042 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006043 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006044 break;
6045 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07006046 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006047 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006048 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006049 break;
6050 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006051 ret = io_fsync(req, issue_flags);
Jackie Liuba5290c2019-10-09 09:19:59 +08006052 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006053 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006054 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006055 break;
6056 case IORING_OP_POLL_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006057 ret = io_poll_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006058 break;
Jens Axboeb76da702019-11-20 13:05:32 -07006059 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006060 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006061 break;
6062 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006063 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006064 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006065 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006066 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006067 break;
6068 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006069 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006070 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006071 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006072 ret = io_recv(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006073 break;
Jens Axboe561fb042019-10-24 07:25:42 -06006074 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006075 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006076 break;
6077 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006078 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006079 break;
6080 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006081 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006082 break;
6083 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006084 ret = io_connect(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07006085 break;
6086 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006087 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006088 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07006089 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006090 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07006091 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07006092 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006093 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006094 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07006095 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006096 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07006097 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006098 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006099 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006100 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006101 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006102 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006103 break;
Jens Axboe4840e412019-12-25 22:03:45 -07006104 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006105 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006106 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006107 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006108 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006109 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006110 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006111 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006112 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006113 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006114 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006115 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006116 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006117 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006118 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006119 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006120 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006121 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006122 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006123 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006124 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006125 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006126 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006127 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006128 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006129 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006130 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006131 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006132 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006133 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006134 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006135 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006136 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006137 default:
6138 ret = -EINVAL;
6139 break;
6140 }
Jens Axboe31b51512019-01-18 22:56:34 -07006141
Jens Axboe5730b272021-02-27 15:57:30 -07006142 if (creds)
6143 revert_creds(creds);
6144
Jens Axboe2b188cc2019-01-07 10:46:33 -07006145 if (ret)
6146 return ret;
6147
Jens Axboeb5325762020-05-19 21:20:27 -06006148 /* If the op doesn't have a file, we're not polling for it */
6149 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006150 const bool in_async = io_wq_current_is_worker();
6151
Jens Axboe11ba8202020-01-15 21:51:17 -07006152 /* workqueue context doesn't hold uring_lock, grab it now */
6153 if (in_async)
6154 mutex_lock(&ctx->uring_lock);
6155
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08006156 io_iopoll_req_issued(req, in_async);
Jens Axboe11ba8202020-01-15 21:51:17 -07006157
6158 if (in_async)
6159 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006160 }
6161
6162 return 0;
6163}
6164
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006165static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006166{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006167 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006168 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006169 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006170
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006171 timeout = io_prep_linked_timeout(req);
6172 if (timeout)
6173 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006174
Jens Axboe4014d942021-01-19 15:53:54 -07006175 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006176 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006177
Jens Axboe561fb042019-10-24 07:25:42 -06006178 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006179 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006180 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006181 /*
6182 * We can get EAGAIN for polled IO even though we're
6183 * forcing a sync submission from here, since we can't
6184 * wait for request slots on the block side.
6185 */
6186 if (ret != -EAGAIN)
6187 break;
6188 cond_resched();
6189 } while (1);
6190 }
Jens Axboe31b51512019-01-18 22:56:34 -07006191
Pavel Begunkova3df76982021-02-18 22:32:52 +00006192 /* avoid locking problems by failing it from a clean context */
Jens Axboe561fb042019-10-24 07:25:42 -06006193 if (ret) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00006194 /* io-wq is going to take one down */
Jens Axboede9b4cc2021-02-24 13:28:27 -07006195 req_ref_get(req);
Pavel Begunkova3df76982021-02-18 22:32:52 +00006196 io_req_task_queue_fail(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006197 }
Jens Axboe31b51512019-01-18 22:56:34 -07006198}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006199
Jens Axboe7b29f922021-03-12 08:30:14 -07006200#define FFS_ASYNC_READ 0x1UL
6201#define FFS_ASYNC_WRITE 0x2UL
6202#ifdef CONFIG_64BIT
6203#define FFS_ISREG 0x4UL
6204#else
6205#define FFS_ISREG 0x0UL
6206#endif
6207#define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG)
6208
Pavel Begunkovdafecf12021-02-28 22:35:11 +00006209static inline struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data,
6210 unsigned i)
Jens Axboe09bb8392019-03-13 12:39:28 -06006211{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006212 struct fixed_rsrc_table *table;
Jens Axboe65e19f52019-10-26 07:20:21 -06006213
Pavel Begunkovdafecf12021-02-28 22:35:11 +00006214 table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT];
6215 return &table->files[i & IORING_FILE_TABLE_MASK];
6216}
6217
6218static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6219 int index)
6220{
Jens Axboe7b29f922021-03-12 08:30:14 -07006221 struct file **file_slot = io_fixed_file_slot(ctx->file_data, index);
6222
6223 return (struct file *) ((unsigned long) *file_slot & FFS_MASK);
Jens Axboe65e19f52019-10-26 07:20:21 -06006224}
6225
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006226static struct file *io_file_get(struct io_submit_state *state,
6227 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006228{
6229 struct io_ring_ctx *ctx = req->ctx;
6230 struct file *file;
6231
6232 if (fixed) {
Jens Axboe7b29f922021-03-12 08:30:14 -07006233 unsigned long file_ptr;
6234
Pavel Begunkov479f5172020-10-10 18:34:07 +01006235 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006236 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006237 fd = array_index_nospec(fd, ctx->nr_user_files);
Jens Axboe7b29f922021-03-12 08:30:14 -07006238 file_ptr = (unsigned long) *io_fixed_file_slot(ctx->file_data, fd);
6239 file = (struct file *) (file_ptr & FFS_MASK);
6240 file_ptr &= ~FFS_MASK;
6241 /* mask in overlapping REQ_F and FFS bits */
6242 req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00006243 io_set_resource_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006244 } else {
6245 trace_io_uring_file_get(ctx, fd);
6246 file = __io_file_get(state, fd);
Jens Axboed44f5542021-03-12 08:27:05 -07006247
6248 /* we don't allow fixed io_uring files */
6249 if (file && unlikely(file->f_op == &io_uring_fops))
6250 io_req_track_inflight(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006251 }
6252
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006253 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006254}
6255
Jens Axboe2665abf2019-11-05 12:40:47 -07006256static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6257{
Jens Axboead8a48a2019-11-15 08:49:11 -07006258 struct io_timeout_data *data = container_of(timer,
6259 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006260 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006261 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006262 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006263
6264 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006265 prev = req->timeout.head;
6266 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006267
6268 /*
6269 * We don't expect the list to be empty, that will only happen if we
6270 * race with the completion of the linked work.
6271 */
Jens Axboede9b4cc2021-02-24 13:28:27 -07006272 if (prev && req_ref_inc_not_zero(prev))
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006273 io_remove_next_linked(prev);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006274 else
6275 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006276 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6277
6278 if (prev) {
Pavel Begunkov014db002020-03-03 21:33:12 +03006279 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006280 io_put_req_deferred(prev, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006281 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006282 io_req_complete_post(req, -ETIME, 0);
6283 io_put_req_deferred(req, 1);
Jens Axboe2665abf2019-11-05 12:40:47 -07006284 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006285 return HRTIMER_NORESTART;
6286}
6287
Jens Axboe7271ef32020-08-10 09:55:22 -06006288static void __io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006289{
Jens Axboe76a46e02019-11-10 23:34:16 -07006290 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006291 * If the back reference is NULL, then our linked request finished
6292 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006293 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006294 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006295 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006296
Jens Axboead8a48a2019-11-15 08:49:11 -07006297 data->timer.function = io_link_timeout_fn;
6298 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6299 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006300 }
Jens Axboe7271ef32020-08-10 09:55:22 -06006301}
6302
6303static void io_queue_linked_timeout(struct io_kiocb *req)
6304{
6305 struct io_ring_ctx *ctx = req->ctx;
6306
6307 spin_lock_irq(&ctx->completion_lock);
6308 __io_queue_linked_timeout(req);
Jens Axboe76a46e02019-11-10 23:34:16 -07006309 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006310
Jens Axboe2665abf2019-11-05 12:40:47 -07006311 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006312 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006313}
6314
Jens Axboead8a48a2019-11-15 08:49:11 -07006315static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006316{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006317 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006318
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006319 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6320 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006321 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006322
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006323 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006324 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006325 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006326 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006327}
6328
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006329static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006330{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006331 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006332 int ret;
6333
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006334 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006335
6336 /*
6337 * We async punt it if the file wasn't marked NOWAIT, or if the file
6338 * doesn't support non-blocking read/write attempts
6339 */
Pavel Begunkov24c74672020-06-21 13:09:51 +03006340 if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006341 if (!io_arm_poll_handler(req)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006342 /*
6343 * Queued up for async execution, worker will release
6344 * submit reference when the iocb is actually submitted.
6345 */
6346 io_queue_async_work(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006347 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006348 } else if (likely(!ret)) {
6349 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006350 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006351 struct io_ring_ctx *ctx = req->ctx;
6352 struct io_comp_state *cs = &ctx->submit_state.comp;
Jens Axboee65ef562019-03-12 10:16:44 -06006353
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006354 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006355 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006356 io_submit_flush_completions(cs, ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006357 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006358 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006359 }
6360 } else {
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006361 io_req_complete_failed(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006362 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006363 if (linked_timeout)
6364 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006365}
6366
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006367static void io_queue_sqe(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006368{
6369 int ret;
6370
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006371 ret = io_req_defer(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006372 if (ret) {
6373 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006374fail_req:
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006375 io_req_complete_failed(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006376 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006377 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006378 ret = io_req_prep_async(req);
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006379 if (unlikely(ret))
6380 goto fail_req;
Jens Axboece35a472019-12-17 08:04:44 -07006381 io_queue_async_work(req);
6382 } else {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006383 __io_queue_sqe(req);
Jens Axboece35a472019-12-17 08:04:44 -07006384 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006385}
6386
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006387/*
6388 * Check SQE restrictions (opcode and flags).
6389 *
6390 * Returns 'true' if SQE is allowed, 'false' otherwise.
6391 */
6392static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6393 struct io_kiocb *req,
6394 unsigned int sqe_flags)
6395{
6396 if (!ctx->restricted)
6397 return true;
6398
6399 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6400 return false;
6401
6402 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6403 ctx->restrictions.sqe_flags_required)
6404 return false;
6405
6406 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6407 ctx->restrictions.sqe_flags_required))
6408 return false;
6409
6410 return true;
6411}
6412
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006413static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006414 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006415{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006416 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006417 unsigned int sqe_flags;
Jens Axboe003e8dc2021-03-06 09:22:27 -07006418 int personality, ret = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006419
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006420 req->opcode = READ_ONCE(sqe->opcode);
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006421 /* same numerical values with corresponding REQ_F_*, safe to copy */
6422 req->flags = sqe_flags = READ_ONCE(sqe->flags);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006423 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006424 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006425 req->file = NULL;
6426 req->ctx = ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006427 req->link = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006428 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006429 /* one is dropped after submission, the other at completion */
Jens Axboeabc54d62021-02-24 13:32:30 -07006430 atomic_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006431 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006432 req->result = 0;
Jens Axboe93e68e02021-03-09 07:02:21 -07006433 req->work.list.next = NULL;
6434 req->work.creds = NULL;
6435 req->work.flags = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006436
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006437 /* enforce forwards compatibility on users */
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006438 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
6439 req->flags = 0;
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006440 return -EINVAL;
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006441 }
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006442
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006443 if (unlikely(req->opcode >= IORING_OP_LAST))
6444 return -EINVAL;
6445
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006446 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6447 return -EACCES;
6448
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006449 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6450 !io_op_defs[req->opcode].buffer_select)
6451 return -EOPNOTSUPP;
6452
Jens Axboe003e8dc2021-03-06 09:22:27 -07006453 personality = READ_ONCE(sqe->personality);
6454 if (personality) {
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00006455 req->work.creds = xa_load(&ctx->personalities, personality);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006456 if (!req->work.creds)
6457 return -EINVAL;
6458 get_cred(req->work.creds);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006459 }
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006460 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006461
Jens Axboe27926b62020-10-28 09:33:23 -06006462 /*
6463 * Plug now if we have more than 1 IO left after this, and the target
6464 * is potentially a read/write to block based storage.
6465 */
6466 if (!state->plug_started && state->ios_left > 1 &&
6467 io_op_defs[req->opcode].plug) {
6468 blk_start_plug(&state->plug);
6469 state->plug_started = true;
6470 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006471
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006472 if (io_op_defs[req->opcode].needs_file) {
6473 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006474
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006475 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006476 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006477 ret = -EBADF;
6478 }
6479
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006480 state->ios_left--;
6481 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006482}
6483
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006484static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006485 const struct io_uring_sqe *sqe)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006486{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006487 struct io_submit_link *link = &ctx->submit_state.link;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006488 int ret;
6489
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006490 ret = io_init_req(ctx, req, sqe);
6491 if (unlikely(ret)) {
6492fail_req:
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006493 if (link->head) {
6494 /* fail even hard links since we don't submit */
Pavel Begunkovcf109602021-02-18 18:29:43 +00006495 link->head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006496 io_req_complete_failed(link->head, -ECANCELED);
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006497 link->head = NULL;
6498 }
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006499 io_req_complete_failed(req, ret);
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006500 return ret;
6501 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006502 ret = io_req_prep(req, sqe);
6503 if (unlikely(ret))
6504 goto fail_req;
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006505
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006506 /* don't need @sqe from now on */
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006507 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
6508 true, ctx->flags & IORING_SETUP_SQPOLL);
6509
Jens Axboe6c271ce2019-01-10 11:22:30 -07006510 /*
6511 * If we already have a head request, queue this one for async
6512 * submittal once the head completes. If we don't have a head but
6513 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6514 * submitted sync once the chain is complete. If none of those
6515 * conditions are true (normal request), then just queue it.
6516 */
6517 if (link->head) {
6518 struct io_kiocb *head = link->head;
6519
6520 /*
6521 * Taking sequential execution of a link, draining both sides
6522 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6523 * requests in the link. So, it drains the head and the
6524 * next after the link request. The last one is done via
6525 * drain_next flag to persist the effect across calls.
6526 */
6527 if (req->flags & REQ_F_IO_DRAIN) {
6528 head->flags |= REQ_F_IO_DRAIN;
6529 ctx->drain_next = 1;
6530 }
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006531 ret = io_req_prep_async(req);
Pavel Begunkovcf109602021-02-18 18:29:43 +00006532 if (unlikely(ret))
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006533 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006534 trace_io_uring_link(ctx, req, head);
6535 link->last->link = req;
6536 link->last = req;
6537
6538 /* last request of a link, enqueue the link */
6539 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006540 io_queue_sqe(head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006541 link->head = NULL;
6542 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006543 } else {
6544 if (unlikely(ctx->drain_next)) {
6545 req->flags |= REQ_F_IO_DRAIN;
6546 ctx->drain_next = 0;
6547 }
6548 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Jackie Liu4fe2c962019-09-09 20:50:40 +08006549 link->head = req;
6550 link->last = req;
6551 } else {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006552 io_queue_sqe(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006553 }
6554 }
6555
6556 return 0;
6557}
6558
6559/*
6560 * Batched submission is done, ensure local IO is flushed out.
6561 */
6562static void io_submit_state_end(struct io_submit_state *state,
6563 struct io_ring_ctx *ctx)
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006564{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006565 if (state->link.head)
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006566 io_queue_sqe(state->link.head);
Jens Axboe3529d8c2019-12-19 18:24:38 -07006567 if (state->comp.nr)
Jens Axboe9e645e112019-05-10 16:07:28 -06006568 io_submit_flush_completions(&state->comp, ctx);
Jackie Liua197f662019-11-08 08:09:12 -07006569 if (state->plug_started)
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006570 blk_finish_plug(&state->plug);
Jens Axboe75c6a032020-01-28 10:15:23 -07006571 io_state_file_put(state);
Jens Axboe9e645e112019-05-10 16:07:28 -06006572}
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006573
Jens Axboe9e645e112019-05-10 16:07:28 -06006574/*
6575 * Start submission side cache.
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006576 */
Jens Axboe9e645e112019-05-10 16:07:28 -06006577static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkov196be952019-11-07 01:41:06 +03006578 unsigned int max_ios)
Jens Axboe9e645e112019-05-10 16:07:28 -06006579{
6580 state->plug_started = false;
Jens Axboebcda7ba2020-02-23 16:42:51 -07006581 state->ios_left = max_ios;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006582 /* set only head, no need to init link_last in advance */
6583 state->link.head = NULL;
Jens Axboe75c6a032020-01-28 10:15:23 -07006584}
6585
Jens Axboe193155c2020-02-22 23:22:19 -07006586static void io_commit_sqring(struct io_ring_ctx *ctx)
6587{
Jens Axboe75c6a032020-01-28 10:15:23 -07006588 struct io_rings *rings = ctx->rings;
6589
6590 /*
Jens Axboe193155c2020-02-22 23:22:19 -07006591 * Ensure any loads from the SQEs are done at this point,
Jens Axboe75c6a032020-01-28 10:15:23 -07006592 * since once we write the new head, the application could
6593 * write new data to them.
Pavel Begunkov6b47ee62020-01-18 20:22:41 +03006594 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006595 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboebcda7ba2020-02-23 16:42:51 -07006596}
6597
Jens Axboe9e645e112019-05-10 16:07:28 -06006598/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006599 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe9e645e112019-05-10 16:07:28 -06006600 * that is mapped by userspace. This means that care needs to be taken to
6601 * ensure that reads are stable, as we cannot rely on userspace always
Jens Axboe78e19bb2019-11-06 15:21:34 -07006602 * being a good citizen. If members of the sqe are validated and then later
6603 * used, it's important that those reads are done through READ_ONCE() to
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006604 * prevent a re-load down the line.
Jens Axboe9e645e112019-05-10 16:07:28 -06006605 */
6606static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe9e645e112019-05-10 16:07:28 -06006607{
6608 u32 *sq_array = ctx->sq_array;
6609 unsigned head;
6610
6611 /*
6612 * The cached sq head (or cq tail) serves two purposes:
6613 *
6614 * 1) allows us to batch the cost of updating the user visible
Pavel Begunkov9d763772019-12-17 02:22:07 +03006615 * head updates.
Jens Axboe9e645e112019-05-10 16:07:28 -06006616 * 2) allows the kernel side to track the head on its own, even
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006617 * though the application is the one updating it.
6618 */
6619 head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]);
6620 if (likely(head < ctx->sq_entries))
6621 return &ctx->sq_sqes[head];
6622
6623 /* drop invalid entries */
Pavel Begunkov711be032020-01-17 03:57:59 +03006624 ctx->cached_sq_dropped++;
6625 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
6626 return NULL;
6627}
Jens Axboeb7bb4f72019-12-15 22:13:43 -07006628
Jens Axboe0f212202020-09-13 13:09:39 -06006629static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006630{
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006631 int submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006632
Jens Axboec4a2ed72019-11-21 21:01:26 -07006633 /* if we have a backlog and couldn't flush it all, return BUSY */
Jens Axboead3eb2c2019-12-18 17:12:20 -07006634 if (test_bit(0, &ctx->sq_check_overflow)) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006635 if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
Jens Axboead3eb2c2019-12-18 17:12:20 -07006636 return -EBUSY;
6637 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006638
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006639 /* make sure SQ entry isn't read before tail */
6640 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006641
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006642 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6643 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006644
Jens Axboed8a6df12020-10-15 16:24:45 -06006645 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006646 refcount_add(nr, &current->usage);
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006647 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006648
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006649 while (submitted < nr) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006650 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006651 struct io_kiocb *req;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006652
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006653 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006654 if (unlikely(!req)) {
6655 if (!submitted)
6656 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006657 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006658 }
Pavel Begunkov4fccfcb2021-02-12 11:55:17 +00006659 sqe = io_get_sqe(ctx);
6660 if (unlikely(!sqe)) {
6661 kmem_cache_free(req_cachep, req);
6662 break;
6663 }
Jens Axboed3656342019-12-18 09:50:26 -07006664 /* will complete beyond this point, count as submitted */
6665 submitted++;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006666 if (io_submit_sqe(ctx, req, sqe))
Jens Axboed3656342019-12-18 09:50:26 -07006667 break;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006668 }
6669
Pavel Begunkov9466f432020-01-25 22:34:01 +03006670 if (unlikely(submitted != nr)) {
6671 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006672 struct io_uring_task *tctx = current->io_uring;
6673 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006674
Jens Axboed8a6df12020-10-15 16:24:45 -06006675 percpu_ref_put_many(&ctx->refs, unused);
6676 percpu_counter_sub(&tctx->inflight, unused);
6677 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006678 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006679
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006680 io_submit_state_end(&ctx->submit_state, ctx);
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006681 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6682 io_commit_sqring(ctx);
6683
Jens Axboe6c271ce2019-01-10 11:22:30 -07006684 return submitted;
6685}
6686
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006687static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6688{
6689 /* Tell userspace we may need a wakeup call */
6690 spin_lock_irq(&ctx->completion_lock);
6691 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6692 spin_unlock_irq(&ctx->completion_lock);
6693}
6694
6695static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6696{
6697 spin_lock_irq(&ctx->completion_lock);
6698 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6699 spin_unlock_irq(&ctx->completion_lock);
6700}
6701
Xiaoguang Wang08369242020-11-03 14:15:59 +08006702static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006703{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006704 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006705 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006706
Jens Axboec8d1ba52020-09-14 11:07:26 -06006707 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006708 /* if we're handling multiple rings, cap submit size for fairness */
6709 if (cap_entries && to_submit > 8)
6710 to_submit = 8;
6711
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006712 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6713 unsigned nr_events = 0;
6714
Xiaoguang Wang08369242020-11-03 14:15:59 +08006715 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006716 if (!list_empty(&ctx->iopoll_list))
6717 io_do_iopoll(ctx, &nr_events, 0);
6718
Pavel Begunkov0298ef92021-03-08 13:20:57 +00006719 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
6720 !(ctx->flags & IORING_SETUP_R_DISABLED))
Xiaoguang Wang08369242020-11-03 14:15:59 +08006721 ret = io_submit_sqes(ctx, to_submit);
6722 mutex_unlock(&ctx->uring_lock);
6723 }
Jens Axboe90554202020-09-03 12:12:41 -06006724
6725 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
6726 wake_up(&ctx->sqo_sq_wait);
6727
Xiaoguang Wang08369242020-11-03 14:15:59 +08006728 return ret;
6729}
6730
6731static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
6732{
6733 struct io_ring_ctx *ctx;
6734 unsigned sq_thread_idle = 0;
6735
Pavel Begunkovc9dca272021-03-10 13:13:55 +00006736 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6737 sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006738 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006739}
6740
Jens Axboe6c271ce2019-01-10 11:22:30 -07006741static int io_sq_thread(void *data)
6742{
Jens Axboe69fb2132020-09-14 11:16:23 -06006743 struct io_sq_data *sqd = data;
6744 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d92052020-11-12 14:55:59 +08006745 unsigned long timeout = 0;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006746 char buf[TASK_COMM_LEN];
Xiaoguang Wang08369242020-11-03 14:15:59 +08006747 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006748
Pavel Begunkov696ee882021-04-01 09:55:04 +01006749 snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006750 set_task_comm(current, buf);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006751 current->pf_io_worker = NULL;
Jens Axboe28cea78a2020-09-14 10:51:17 -06006752
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006753 if (sqd->sq_cpu != -1)
6754 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
6755 else
6756 set_cpus_allowed_ptr(current, cpu_online_mask);
6757 current->flags |= PF_NO_SETAFFINITY;
6758
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006759 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07006760 while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006761 int ret;
6762 bool cap_entries, sqt_spin, needs_sched;
Jens Axboec1edbf52019-11-10 16:56:04 -07006763
Jens Axboe82734c52021-03-29 06:52:44 -06006764 if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) ||
6765 signal_pending(current)) {
6766 bool did_sig = false;
6767
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006768 mutex_unlock(&sqd->lock);
Jens Axboe82734c52021-03-29 06:52:44 -06006769 if (signal_pending(current)) {
6770 struct ksignal ksig;
6771
6772 did_sig = get_signal(&ksig);
6773 }
Jens Axboe05962f92021-03-06 13:58:48 -07006774 cond_resched();
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006775 mutex_lock(&sqd->lock);
Jens Axboe82734c52021-03-29 06:52:44 -06006776 if (did_sig)
6777 break;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00006778 io_run_task_work();
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006779 io_run_task_work_head(&sqd->park_task_work);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006780 timeout = jiffies + sqd->sq_thread_idle;
Pavel Begunkov7d41e852021-03-10 13:13:54 +00006781 continue;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006782 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006783 sqt_spin = false;
Jens Axboee95eee22020-09-08 09:11:32 -06006784 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006785 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006786 const struct cred *creds = NULL;
6787
6788 if (ctx->sq_creds != current_cred())
6789 creds = override_creds(ctx->sq_creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006790 ret = __io_sq_thread(ctx, cap_entries);
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006791 if (creds)
6792 revert_creds(creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006793 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
6794 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006795 }
6796
Xiaoguang Wang08369242020-11-03 14:15:59 +08006797 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006798 io_run_task_work();
6799 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08006800 if (sqt_spin)
6801 timeout = jiffies + sqd->sq_thread_idle;
6802 continue;
6803 }
6804
Xiaoguang Wang08369242020-11-03 14:15:59 +08006805 needs_sched = true;
6806 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
6807 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6808 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6809 !list_empty_careful(&ctx->iopoll_list)) {
6810 needs_sched = false;
6811 break;
6812 }
6813 if (io_sqring_entries(ctx)) {
6814 needs_sched = false;
6815 break;
6816 }
6817 }
6818
Jens Axboe05962f92021-03-06 13:58:48 -07006819 if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006820 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6821 io_ring_set_wakeup_flag(ctx);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006822
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006823 mutex_unlock(&sqd->lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006824 schedule();
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006825 mutex_lock(&sqd->lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006826 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6827 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006828 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006829
6830 finish_wait(&sqd->wait, &wait);
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006831 io_run_task_work_head(&sqd->park_task_work);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006832 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006833 }
6834
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006835 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6836 io_uring_cancel_sqpoll(ctx);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006837 sqd->thread = NULL;
Jens Axboe05962f92021-03-06 13:58:48 -07006838 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
Jens Axboe5f3f26f2021-02-25 10:17:46 -07006839 io_ring_set_wakeup_flag(ctx);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006840 mutex_unlock(&sqd->lock);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00006841
6842 io_run_task_work();
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006843 io_run_task_work_head(&sqd->park_task_work);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006844 complete(&sqd->exited);
6845 do_exit(0);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006846}
6847
Jens Axboebda52162019-09-24 13:47:15 -06006848struct io_wait_queue {
6849 struct wait_queue_entry wq;
6850 struct io_ring_ctx *ctx;
6851 unsigned to_wait;
6852 unsigned nr_timeouts;
6853};
6854
Pavel Begunkov6c503152021-01-04 20:36:36 +00006855static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06006856{
6857 struct io_ring_ctx *ctx = iowq->ctx;
6858
6859 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006860 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006861 * started waiting. For timeouts, we always want to return to userspace,
6862 * regardless of event count.
6863 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00006864 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006865 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6866}
6867
6868static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6869 int wake_flags, void *key)
6870{
6871 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6872 wq);
6873
Pavel Begunkov6c503152021-01-04 20:36:36 +00006874 /*
6875 * Cannot safely flush overflowed CQEs from here, ensure we wake up
6876 * the task, and the next invocation will do it.
6877 */
6878 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
6879 return autoremove_wake_function(curr, mode, wake_flags, key);
6880 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06006881}
6882
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006883static int io_run_task_work_sig(void)
6884{
6885 if (io_run_task_work())
6886 return 1;
6887 if (!signal_pending(current))
6888 return 0;
Jens Axboe0b8cfa92021-03-21 14:16:08 -06006889 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
Jens Axboe792ee0f62020-10-22 20:17:18 -06006890 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006891 return -EINTR;
6892}
6893
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006894/* when returns >0, the caller should retry */
6895static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
6896 struct io_wait_queue *iowq,
6897 signed long *timeout)
6898{
6899 int ret;
6900
6901 /* make sure we run task_work before checking for signals */
6902 ret = io_run_task_work_sig();
6903 if (ret || io_should_wake(iowq))
6904 return ret;
6905 /* let the caller flush overflows, retry */
6906 if (test_bit(0, &ctx->cq_check_overflow))
6907 return 1;
6908
6909 *timeout = schedule_timeout(*timeout);
6910 return !*timeout ? -ETIME : 1;
6911}
6912
Jens Axboe2b188cc2019-01-07 10:46:33 -07006913/*
6914 * Wait until events become available, if we don't already have some. The
6915 * application must reap them itself, as they reside on the shared cq ring.
6916 */
6917static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08006918 const sigset_t __user *sig, size_t sigsz,
6919 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006920{
Jens Axboebda52162019-09-24 13:47:15 -06006921 struct io_wait_queue iowq = {
6922 .wq = {
6923 .private = current,
6924 .func = io_wake_function,
6925 .entry = LIST_HEAD_INIT(iowq.wq.entry),
6926 },
6927 .ctx = ctx,
6928 .to_wait = min_events,
6929 };
Hristo Venev75b28af2019-08-26 17:23:46 +00006930 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006931 signed long timeout = MAX_SCHEDULE_TIMEOUT;
6932 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006933
Jens Axboeb41e9852020-02-17 09:52:41 -07006934 do {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006935 io_cqring_overflow_flush(ctx, false, NULL, NULL);
6936 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07006937 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06006938 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07006939 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07006940 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006941
6942 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006943#ifdef CONFIG_COMPAT
6944 if (in_compat_syscall())
6945 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07006946 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006947 else
6948#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07006949 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006950
Jens Axboe2b188cc2019-01-07 10:46:33 -07006951 if (ret)
6952 return ret;
6953 }
6954
Hao Xuc73ebb62020-11-03 10:54:37 +08006955 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006956 struct timespec64 ts;
6957
Hao Xuc73ebb62020-11-03 10:54:37 +08006958 if (get_timespec64(&ts, uts))
6959 return -EFAULT;
6960 timeout = timespec64_to_jiffies(&ts);
6961 }
6962
Jens Axboebda52162019-09-24 13:47:15 -06006963 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02006964 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06006965 do {
Jens Axboeca0a2652021-03-04 17:15:48 -07006966 /* if we can't even flush overflow, don't wait for more */
6967 if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) {
6968 ret = -EBUSY;
6969 break;
6970 }
Jens Axboebda52162019-09-24 13:47:15 -06006971 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
6972 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006973 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
6974 finish_wait(&ctx->wait, &iowq.wq);
Jens Axboeca0a2652021-03-04 17:15:48 -07006975 cond_resched();
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006976 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06006977
Jens Axboeb7db41c2020-07-04 08:55:50 -06006978 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006979
Hristo Venev75b28af2019-08-26 17:23:46 +00006980 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006981}
6982
Jens Axboe6b063142019-01-10 22:13:58 -07006983static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
6984{
6985#if defined(CONFIG_UNIX)
6986 if (ctx->ring_sock) {
6987 struct sock *sock = ctx->ring_sock->sk;
6988 struct sk_buff *skb;
6989
6990 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
6991 kfree_skb(skb);
6992 }
6993#else
6994 int i;
6995
Jens Axboe65e19f52019-10-26 07:20:21 -06006996 for (i = 0; i < ctx->nr_user_files; i++) {
6997 struct file *file;
6998
6999 file = io_file_from_index(ctx, i);
7000 if (file)
7001 fput(file);
7002 }
Jens Axboe6b063142019-01-10 22:13:58 -07007003#endif
7004}
7005
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007006static void io_rsrc_data_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007007{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007008 struct fixed_rsrc_data *data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007009
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007010 data = container_of(ref, struct fixed_rsrc_data, refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007011 complete(&data->done);
7012}
7013
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007014static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
Pavel Begunkov1642b442020-12-30 21:34:14 +00007015{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007016 spin_lock_bh(&ctx->rsrc_ref_lock);
Pavel Begunkov1642b442020-12-30 21:34:14 +00007017}
7018
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007019static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
Jens Axboe6b063142019-01-10 22:13:58 -07007020{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007021 spin_unlock_bh(&ctx->rsrc_ref_lock);
7022}
7023
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007024static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx,
7025 struct fixed_rsrc_data *rsrc_data,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007026 struct fixed_rsrc_ref_node *ref_node)
Jens Axboe6b063142019-01-10 22:13:58 -07007027{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007028 io_rsrc_ref_lock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007029 rsrc_data->node = ref_node;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007030 list_add_tail(&ref_node->node, &ctx->rsrc_ref_list);
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007031 io_rsrc_ref_unlock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007032 percpu_ref_get(&rsrc_data->refs);
Jens Axboe6b063142019-01-10 22:13:58 -07007033}
7034
Hao Xu8bad28d2021-02-19 17:19:36 +08007035static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data)
Jens Axboe6b063142019-01-10 22:13:58 -07007036{
Hao Xu8bad28d2021-02-19 17:19:36 +08007037 struct fixed_rsrc_ref_node *ref_node = NULL;
Jens Axboe65e19f52019-10-26 07:20:21 -06007038
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007039 io_rsrc_ref_lock(ctx);
Pavel Begunkov1e5d7702020-11-18 14:56:25 +00007040 ref_node = data->node;
Pavel Begunkove6cb0072021-02-20 18:03:47 +00007041 data->node = NULL;
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007042 io_rsrc_ref_unlock(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007043 if (ref_node)
7044 percpu_ref_kill(&ref_node->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08007045}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007046
Hao Xu8bad28d2021-02-19 17:19:36 +08007047static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data,
7048 struct io_ring_ctx *ctx,
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007049 void (*rsrc_put)(struct io_ring_ctx *ctx,
7050 struct io_rsrc_put *prsrc))
Hao Xu8bad28d2021-02-19 17:19:36 +08007051{
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007052 struct fixed_rsrc_ref_node *backup_node;
Hao Xu8bad28d2021-02-19 17:19:36 +08007053 int ret;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007054
Hao Xu8bad28d2021-02-19 17:19:36 +08007055 if (data->quiesce)
7056 return -ENXIO;
7057
7058 data->quiesce = true;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007059 do {
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007060 ret = -ENOMEM;
7061 backup_node = alloc_fixed_rsrc_ref_node(ctx);
7062 if (!backup_node)
7063 break;
7064 backup_node->rsrc_data = data;
7065 backup_node->rsrc_put = rsrc_put;
7066
Hao Xu8bad28d2021-02-19 17:19:36 +08007067 io_sqe_rsrc_kill_node(ctx, data);
7068 percpu_ref_kill(&data->refs);
7069 flush_delayed_work(&ctx->rsrc_put_work);
7070
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007071 ret = wait_for_completion_interruptible(&data->done);
7072 if (!ret)
7073 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007074
Jens Axboecb5e1b82021-02-25 07:37:35 -07007075 percpu_ref_resurrect(&data->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08007076 io_sqe_rsrc_set_node(ctx, data, backup_node);
7077 backup_node = NULL;
Jens Axboecb5e1b82021-02-25 07:37:35 -07007078 reinit_completion(&data->done);
Hao Xu8bad28d2021-02-19 17:19:36 +08007079 mutex_unlock(&ctx->uring_lock);
7080 ret = io_run_task_work_sig();
7081 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007082 } while (ret >= 0);
Hao Xu8bad28d2021-02-19 17:19:36 +08007083 data->quiesce = false;
7084
7085 if (backup_node)
7086 destroy_fixed_rsrc_ref_node(backup_node);
7087 return ret;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007088}
7089
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007090static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx)
7091{
7092 struct fixed_rsrc_data *data;
7093
7094 data = kzalloc(sizeof(*data), GFP_KERNEL);
7095 if (!data)
7096 return NULL;
7097
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007098 if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero,
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007099 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
7100 kfree(data);
7101 return NULL;
7102 }
7103 data->ctx = ctx;
7104 init_completion(&data->done);
7105 return data;
7106}
7107
7108static void free_fixed_rsrc_data(struct fixed_rsrc_data *data)
7109{
7110 percpu_ref_exit(&data->refs);
7111 kfree(data->table);
7112 kfree(data);
7113}
7114
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007115static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7116{
7117 struct fixed_rsrc_data *data = ctx->file_data;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007118 unsigned nr_tables, i;
7119 int ret;
7120
Hao Xu8bad28d2021-02-19 17:19:36 +08007121 /*
7122 * percpu_ref_is_dying() is to stop parallel files unregister
7123 * Since we possibly drop uring lock later in this function to
7124 * run task work.
7125 */
7126 if (!data || percpu_ref_is_dying(&data->refs))
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007127 return -ENXIO;
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007128 ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put);
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007129 if (ret)
7130 return ret;
7131
Jens Axboe6b063142019-01-10 22:13:58 -07007132 __io_sqe_files_unregister(ctx);
Jens Axboe65e19f52019-10-26 07:20:21 -06007133 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
7134 for (i = 0; i < nr_tables; i++)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007135 kfree(data->table[i].files);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007136 free_fixed_rsrc_data(data);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007137 ctx->file_data = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07007138 ctx->nr_user_files = 0;
7139 return 0;
7140}
7141
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007142static void io_sq_thread_unpark(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007143 __releases(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007144{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007145 WARN_ON_ONCE(sqd->thread == current);
7146
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007147 /*
7148 * Do the dance but not conditional clear_bit() because it'd race with
7149 * other threads incrementing park_pending and setting the bit.
7150 */
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007151 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007152 if (atomic_dec_return(&sqd->park_pending))
7153 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007154 mutex_unlock(&sqd->lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007155}
7156
Jens Axboe86e0d672021-03-05 08:44:39 -07007157static void io_sq_thread_park(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007158 __acquires(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007159{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007160 WARN_ON_ONCE(sqd->thread == current);
7161
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007162 atomic_inc(&sqd->park_pending);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007163 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007164 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007165 if (sqd->thread)
Jens Axboe86e0d672021-03-05 08:44:39 -07007166 wake_up_process(sqd->thread);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007167}
7168
7169static void io_sq_thread_stop(struct io_sq_data *sqd)
7170{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007171 WARN_ON_ONCE(sqd->thread == current);
7172
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007173 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007174 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
Jens Axboee8f98f242021-03-09 16:32:13 -07007175 if (sqd->thread)
7176 wake_up_process(sqd->thread);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007177 mutex_unlock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007178 wait_for_completion(&sqd->exited);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007179}
7180
Jens Axboe534ca6d2020-09-02 13:52:19 -06007181static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007182{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007183 if (refcount_dec_and_test(&sqd->refs)) {
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007184 WARN_ON_ONCE(atomic_read(&sqd->park_pending));
7185
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007186 io_sq_thread_stop(sqd);
7187 kfree(sqd);
7188 }
7189}
7190
7191static void io_sq_thread_finish(struct io_ring_ctx *ctx)
7192{
7193 struct io_sq_data *sqd = ctx->sq_data;
7194
7195 if (sqd) {
Jens Axboe05962f92021-03-06 13:58:48 -07007196 io_sq_thread_park(sqd);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007197 list_del_init(&ctx->sqd_list);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007198 io_sqd_update_thread_idle(sqd);
Jens Axboe05962f92021-03-06 13:58:48 -07007199 io_sq_thread_unpark(sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007200
7201 io_put_sq_data(sqd);
7202 ctx->sq_data = NULL;
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01007203 if (ctx->sq_creds)
7204 put_cred(ctx->sq_creds);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007205 }
7206}
7207
Jens Axboeaa061652020-09-02 14:50:27 -06007208static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7209{
7210 struct io_ring_ctx *ctx_attach;
7211 struct io_sq_data *sqd;
7212 struct fd f;
7213
7214 f = fdget(p->wq_fd);
7215 if (!f.file)
7216 return ERR_PTR(-ENXIO);
7217 if (f.file->f_op != &io_uring_fops) {
7218 fdput(f);
7219 return ERR_PTR(-EINVAL);
7220 }
7221
7222 ctx_attach = f.file->private_data;
7223 sqd = ctx_attach->sq_data;
7224 if (!sqd) {
7225 fdput(f);
7226 return ERR_PTR(-EINVAL);
7227 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007228 if (sqd->task_tgid != current->tgid) {
7229 fdput(f);
7230 return ERR_PTR(-EPERM);
7231 }
Jens Axboeaa061652020-09-02 14:50:27 -06007232
7233 refcount_inc(&sqd->refs);
7234 fdput(f);
7235 return sqd;
7236}
7237
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007238static struct io_sq_data *io_get_sq_data(struct io_uring_params *p,
7239 bool *attached)
Jens Axboe534ca6d2020-09-02 13:52:19 -06007240{
7241 struct io_sq_data *sqd;
7242
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007243 *attached = false;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007244 if (p->flags & IORING_SETUP_ATTACH_WQ) {
7245 sqd = io_attach_sq_data(p);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007246 if (!IS_ERR(sqd)) {
7247 *attached = true;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007248 return sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007249 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007250 /* fall through for EPERM case, setup new sqd/task */
7251 if (PTR_ERR(sqd) != -EPERM)
7252 return sqd;
7253 }
Jens Axboeaa061652020-09-02 14:50:27 -06007254
Jens Axboe534ca6d2020-09-02 13:52:19 -06007255 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7256 if (!sqd)
7257 return ERR_PTR(-ENOMEM);
7258
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007259 atomic_set(&sqd->park_pending, 0);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007260 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007261 INIT_LIST_HEAD(&sqd->ctx_list);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007262 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007263 init_waitqueue_head(&sqd->wait);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007264 init_completion(&sqd->exited);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007265 return sqd;
7266}
7267
Jens Axboe6b063142019-01-10 22:13:58 -07007268#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007269/*
7270 * Ensure the UNIX gc is aware of our file set, so we are certain that
7271 * the io_uring can be safely unregistered on process exit, even if we have
7272 * loops in the file referencing.
7273 */
7274static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7275{
7276 struct sock *sk = ctx->ring_sock->sk;
7277 struct scm_fp_list *fpl;
7278 struct sk_buff *skb;
Jens Axboe08a45172019-10-03 08:11:03 -06007279 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007280
Jens Axboe6b063142019-01-10 22:13:58 -07007281 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7282 if (!fpl)
7283 return -ENOMEM;
7284
7285 skb = alloc_skb(0, GFP_KERNEL);
7286 if (!skb) {
7287 kfree(fpl);
7288 return -ENOMEM;
7289 }
7290
7291 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007292
Jens Axboe08a45172019-10-03 08:11:03 -06007293 nr_files = 0;
Jens Axboe62e398b2021-02-21 16:19:37 -07007294 fpl->user = get_uid(current_user());
Jens Axboe6b063142019-01-10 22:13:58 -07007295 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007296 struct file *file = io_file_from_index(ctx, i + offset);
7297
7298 if (!file)
Jens Axboe08a45172019-10-03 08:11:03 -06007299 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007300 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a45172019-10-03 08:11:03 -06007301 unix_inflight(fpl->user, fpl->fp[nr_files]);
7302 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007303 }
7304
Jens Axboe08a45172019-10-03 08:11:03 -06007305 if (nr_files) {
7306 fpl->max = SCM_MAX_FD;
7307 fpl->count = nr_files;
7308 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007309 skb->destructor = unix_destruct_scm;
Jens Axboe08a45172019-10-03 08:11:03 -06007310 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7311 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007312
Jens Axboe08a45172019-10-03 08:11:03 -06007313 for (i = 0; i < nr_files; i++)
7314 fput(fpl->fp[i]);
7315 } else {
7316 kfree_skb(skb);
7317 kfree(fpl);
7318 }
Jens Axboe6b063142019-01-10 22:13:58 -07007319
7320 return 0;
7321}
7322
7323/*
7324 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7325 * causes regular reference counting to break down. We rely on the UNIX
7326 * garbage collection to take care of this problem for us.
7327 */
7328static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7329{
7330 unsigned left, total;
7331 int ret = 0;
7332
7333 total = 0;
7334 left = ctx->nr_user_files;
7335 while (left) {
7336 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007337
7338 ret = __io_sqe_files_scm(ctx, this_files, total);
7339 if (ret)
7340 break;
7341 left -= this_files;
7342 total += this_files;
7343 }
7344
7345 if (!ret)
7346 return 0;
7347
7348 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007349 struct file *file = io_file_from_index(ctx, total);
7350
7351 if (file)
7352 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007353 total++;
7354 }
7355
7356 return ret;
7357}
7358#else
7359static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7360{
7361 return 0;
7362}
7363#endif
7364
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007365static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data,
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007366 unsigned nr_tables, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007367{
7368 int i;
7369
7370 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007371 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007372 unsigned this_files;
7373
7374 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
7375 table->files = kcalloc(this_files, sizeof(struct file *),
7376 GFP_KERNEL);
7377 if (!table->files)
7378 break;
7379 nr_files -= this_files;
7380 }
7381
7382 if (i == nr_tables)
7383 return 0;
7384
7385 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007386 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007387 kfree(table->files);
7388 }
7389 return 1;
7390}
7391
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007392static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007393{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007394 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007395#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007396 struct sock *sock = ctx->ring_sock->sk;
7397 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7398 struct sk_buff *skb;
7399 int i;
7400
7401 __skb_queue_head_init(&list);
7402
7403 /*
7404 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7405 * remove this entry and rearrange the file array.
7406 */
7407 skb = skb_dequeue(head);
7408 while (skb) {
7409 struct scm_fp_list *fp;
7410
7411 fp = UNIXCB(skb).fp;
7412 for (i = 0; i < fp->count; i++) {
7413 int left;
7414
7415 if (fp->fp[i] != file)
7416 continue;
7417
7418 unix_notinflight(fp->user, fp->fp[i]);
7419 left = fp->count - 1 - i;
7420 if (left) {
7421 memmove(&fp->fp[i], &fp->fp[i + 1],
7422 left * sizeof(struct file *));
7423 }
7424 fp->count--;
7425 if (!fp->count) {
7426 kfree_skb(skb);
7427 skb = NULL;
7428 } else {
7429 __skb_queue_tail(&list, skb);
7430 }
7431 fput(file);
7432 file = NULL;
7433 break;
7434 }
7435
7436 if (!file)
7437 break;
7438
7439 __skb_queue_tail(&list, skb);
7440
7441 skb = skb_dequeue(head);
7442 }
7443
7444 if (skb_peek(&list)) {
7445 spin_lock_irq(&head->lock);
7446 while ((skb = __skb_dequeue(&list)) != NULL)
7447 __skb_queue_tail(head, skb);
7448 spin_unlock_irq(&head->lock);
7449 }
7450#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007451 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007452#endif
7453}
7454
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007455static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007456{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007457 struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data;
7458 struct io_ring_ctx *ctx = rsrc_data->ctx;
7459 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007460
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007461 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7462 list_del(&prsrc->list);
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007463 ref_node->rsrc_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007464 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007465 }
7466
Xiaoguang Wang05589552020-03-31 14:05:18 +08007467 percpu_ref_exit(&ref_node->refs);
7468 kfree(ref_node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007469 percpu_ref_put(&rsrc_data->refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007470}
7471
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007472static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007473{
7474 struct io_ring_ctx *ctx;
7475 struct llist_node *node;
7476
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007477 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7478 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007479
7480 while (node) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007481 struct fixed_rsrc_ref_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007482 struct llist_node *next = node->next;
7483
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007484 ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist);
7485 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007486 node = next;
7487 }
7488}
7489
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007490static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007491{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007492 struct fixed_rsrc_ref_node *ref_node;
7493 struct fixed_rsrc_data *data;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007494 struct io_ring_ctx *ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007495 bool first_add = false;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007496 int delay = HZ;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007497
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007498 ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs);
7499 data = ref_node->rsrc_data;
Pavel Begunkove2978222020-11-18 14:56:26 +00007500 ctx = data->ctx;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007501
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007502 io_rsrc_ref_lock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007503 ref_node->done = true;
7504
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007505 while (!list_empty(&ctx->rsrc_ref_list)) {
7506 ref_node = list_first_entry(&ctx->rsrc_ref_list,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007507 struct fixed_rsrc_ref_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007508 /* recycle ref nodes in order */
7509 if (!ref_node->done)
7510 break;
7511 list_del(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007512 first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007513 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007514 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007515
7516 if (percpu_ref_is_dying(&data->refs))
Jens Axboe4a38aed22020-05-14 17:21:15 -06007517 delay = 0;
7518
Jens Axboe4a38aed22020-05-14 17:21:15 -06007519 if (!delay)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007520 mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007521 else if (first_add)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007522 queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007523}
7524
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007525static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Xiaoguang Wang05589552020-03-31 14:05:18 +08007526 struct io_ring_ctx *ctx)
7527{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007528 struct fixed_rsrc_ref_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007529
7530 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7531 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007532 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007533
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007534 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007535 0, GFP_KERNEL)) {
7536 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007537 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007538 }
7539 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007540 INIT_LIST_HEAD(&ref_node->rsrc_list);
Pavel Begunkove2978222020-11-18 14:56:26 +00007541 ref_node->done = false;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007542 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007543}
7544
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007545static void init_fixed_file_ref_node(struct io_ring_ctx *ctx,
7546 struct fixed_rsrc_ref_node *ref_node)
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007547{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007548 ref_node->rsrc_data = ctx->file_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007549 ref_node->rsrc_put = io_ring_file_put;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007550}
7551
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007552static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007553{
7554 percpu_ref_exit(&ref_node->refs);
7555 kfree(ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007556}
7557
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007558
Jens Axboe05f3fb32019-12-09 11:22:50 -07007559static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7560 unsigned nr_args)
7561{
7562 __s32 __user *fds = (__s32 __user *) arg;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007563 unsigned nr_tables, i;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007564 struct file *file;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007565 int fd, ret = -ENOMEM;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007566 struct fixed_rsrc_ref_node *ref_node;
7567 struct fixed_rsrc_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007568
7569 if (ctx->file_data)
7570 return -EBUSY;
7571 if (!nr_args)
7572 return -EINVAL;
7573 if (nr_args > IORING_MAX_FIXED_FILES)
7574 return -EMFILE;
7575
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007576 file_data = alloc_fixed_rsrc_data(ctx);
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007577 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007578 return -ENOMEM;
Dan Carpenter13770a72021-02-01 15:23:42 +03007579 ctx->file_data = file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007580
7581 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
Colin Ian King035fbaf2020-10-12 15:03:41 +01007582 file_data->table = kcalloc(nr_tables, sizeof(*file_data->table),
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007583 GFP_KERNEL);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007584 if (!file_data->table)
7585 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007586
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007587 if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args))
Jens Axboe05f3fb32019-12-09 11:22:50 -07007588 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007589
Jens Axboe05f3fb32019-12-09 11:22:50 -07007590 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Jens Axboe7b29f922021-03-12 08:30:14 -07007591 unsigned long file_ptr;
7592
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007593 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7594 ret = -EFAULT;
7595 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007596 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007597 /* allow sparse sets */
7598 if (fd == -1)
7599 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007600
Jens Axboe05f3fb32019-12-09 11:22:50 -07007601 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007602 ret = -EBADF;
7603 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007604 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007605
7606 /*
7607 * Don't allow io_uring instances to be registered. If UNIX
7608 * isn't enabled, then this causes a reference cycle and this
7609 * instance can never get freed. If UNIX is enabled we'll
7610 * handle it just fine, but there's still no point in allowing
7611 * a ring fd as it doesn't support regular read/write anyway.
7612 */
7613 if (file->f_op == &io_uring_fops) {
7614 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007615 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007616 }
Jens Axboe7b29f922021-03-12 08:30:14 -07007617 file_ptr = (unsigned long) file;
7618 if (__io_file_supports_async(file, READ))
7619 file_ptr |= FFS_ASYNC_READ;
7620 if (__io_file_supports_async(file, WRITE))
7621 file_ptr |= FFS_ASYNC_WRITE;
7622 if (S_ISREG(file_inode(file)->i_mode))
7623 file_ptr |= FFS_ISREG;
7624 *io_fixed_file_slot(file_data, i) = (struct file *) file_ptr;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007625 }
7626
Jens Axboe05f3fb32019-12-09 11:22:50 -07007627 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007628 if (ret) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07007629 io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007630 return ret;
7631 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007632
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007633 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007634 if (!ref_node) {
Xiaoguang Wang05589552020-03-31 14:05:18 +08007635 io_sqe_files_unregister(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007636 return -ENOMEM;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007637 }
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007638 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007639
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007640 io_sqe_rsrc_set_node(ctx, file_data, ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007641 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007642out_fput:
7643 for (i = 0; i < ctx->nr_user_files; i++) {
7644 file = io_file_from_index(ctx, i);
7645 if (file)
7646 fput(file);
7647 }
7648 for (i = 0; i < nr_tables; i++)
7649 kfree(file_data->table[i].files);
7650 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007651out_free:
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007652 free_fixed_rsrc_data(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007653 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007654 return ret;
7655}
7656
Jens Axboec3a31e62019-10-03 13:59:56 -06007657static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7658 int index)
7659{
7660#if defined(CONFIG_UNIX)
7661 struct sock *sock = ctx->ring_sock->sk;
7662 struct sk_buff_head *head = &sock->sk_receive_queue;
7663 struct sk_buff *skb;
7664
7665 /*
7666 * See if we can merge this file into an existing skb SCM_RIGHTS
7667 * file set. If there's no room, fall back to allocating a new skb
7668 * and filling it in.
7669 */
7670 spin_lock_irq(&head->lock);
7671 skb = skb_peek(head);
7672 if (skb) {
7673 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7674
7675 if (fpl->count < SCM_MAX_FD) {
7676 __skb_unlink(skb, head);
7677 spin_unlock_irq(&head->lock);
7678 fpl->fp[fpl->count] = get_file(file);
7679 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7680 fpl->count++;
7681 spin_lock_irq(&head->lock);
7682 __skb_queue_head(head, skb);
7683 } else {
7684 skb = NULL;
7685 }
7686 }
7687 spin_unlock_irq(&head->lock);
7688
7689 if (skb) {
7690 fput(file);
7691 return 0;
7692 }
7693
7694 return __io_sqe_files_scm(ctx, 1, index);
7695#else
7696 return 0;
7697#endif
7698}
7699
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007700static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007701{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007702 struct io_rsrc_put *prsrc;
7703 struct fixed_rsrc_ref_node *ref_node = data->node;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007704
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007705 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7706 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007707 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007708
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007709 prsrc->rsrc = rsrc;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007710 list_add(&prsrc->list, &ref_node->rsrc_list);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007711
Hillf Dantona5318d32020-03-23 17:47:15 +08007712 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007713}
7714
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007715static inline int io_queue_file_removal(struct fixed_rsrc_data *data,
7716 struct file *file)
7717{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007718 return io_queue_rsrc_removal(data, (void *)file);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007719}
7720
Jens Axboe05f3fb32019-12-09 11:22:50 -07007721static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007722 struct io_uring_rsrc_update *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07007723 unsigned nr_args)
7724{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007725 struct fixed_rsrc_data *data = ctx->file_data;
7726 struct fixed_rsrc_ref_node *ref_node;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007727 struct file *file, **file_slot;
Jens Axboec3a31e62019-10-03 13:59:56 -06007728 __s32 __user *fds;
7729 int fd, i, err;
7730 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007731 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007732
Jens Axboe05f3fb32019-12-09 11:22:50 -07007733 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06007734 return -EOVERFLOW;
7735 if (done > ctx->nr_user_files)
7736 return -EINVAL;
7737
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007738 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007739 if (!ref_node)
7740 return -ENOMEM;
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007741 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007742
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007743 fds = u64_to_user_ptr(up->data);
Pavel Begunkov67973b92021-01-26 13:51:09 +00007744 for (done = 0; done < nr_args; done++) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007745 err = 0;
7746 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
7747 err = -EFAULT;
7748 break;
7749 }
noah4e0377a2021-01-26 15:23:28 -05007750 if (fd == IORING_REGISTER_FILES_SKIP)
7751 continue;
7752
Pavel Begunkov67973b92021-01-26 13:51:09 +00007753 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007754 file_slot = io_fixed_file_slot(ctx->file_data, i);
7755
7756 if (*file_slot) {
Jens Axboe7b29f922021-03-12 08:30:14 -07007757 file = (struct file *) ((unsigned long) *file_slot & FFS_MASK);
7758 err = io_queue_file_removal(data, file);
Hillf Dantona5318d32020-03-23 17:47:15 +08007759 if (err)
7760 break;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007761 *file_slot = NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007762 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007763 }
7764 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007765 file = fget(fd);
7766 if (!file) {
7767 err = -EBADF;
7768 break;
7769 }
7770 /*
7771 * Don't allow io_uring instances to be registered. If
7772 * UNIX isn't enabled, then this causes a reference
7773 * cycle and this instance can never get freed. If UNIX
7774 * is enabled we'll handle it just fine, but there's
7775 * still no point in allowing a ring fd as it doesn't
7776 * support regular read/write anyway.
7777 */
7778 if (file->f_op == &io_uring_fops) {
7779 fput(file);
7780 err = -EBADF;
7781 break;
7782 }
Jens Axboee68a3ff2021-02-11 07:45:08 -07007783 *file_slot = file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007784 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007785 if (err) {
Jens Axboee68a3ff2021-02-11 07:45:08 -07007786 *file_slot = NULL;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007787 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007788 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007789 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007790 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007791 }
7792
Xiaoguang Wang05589552020-03-31 14:05:18 +08007793 if (needs_switch) {
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007794 percpu_ref_kill(&data->node->refs);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007795 io_sqe_rsrc_set_node(ctx, data, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007796 } else
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007797 destroy_fixed_rsrc_ref_node(ref_node);
Jens Axboec3a31e62019-10-03 13:59:56 -06007798
7799 return done ? done : err;
7800}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007801
Jens Axboe05f3fb32019-12-09 11:22:50 -07007802static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
7803 unsigned nr_args)
7804{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007805 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007806
7807 if (!ctx->file_data)
7808 return -ENXIO;
7809 if (!nr_args)
7810 return -EINVAL;
7811 if (copy_from_user(&up, arg, sizeof(up)))
7812 return -EFAULT;
7813 if (up.resv)
7814 return -EINVAL;
7815
7816 return __io_sqe_files_update(ctx, &up, nr_args);
7817}
Jens Axboec3a31e62019-10-03 13:59:56 -06007818
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007819static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007820{
7821 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7822
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007823 req = io_put_req_find_next(req);
7824 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07007825}
7826
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007827static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx)
Pavel Begunkov24369c22020-01-28 03:15:48 +03007828{
Jens Axboee9418942021-02-19 12:33:30 -07007829 struct io_wq_hash *hash;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007830 struct io_wq_data data;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007831 unsigned int concurrency;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007832
Jens Axboee9418942021-02-19 12:33:30 -07007833 hash = ctx->hash_map;
7834 if (!hash) {
7835 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
7836 if (!hash)
7837 return ERR_PTR(-ENOMEM);
7838 refcount_set(&hash->refs, 1);
7839 init_waitqueue_head(&hash->wait);
7840 ctx->hash_map = hash;
7841 }
7842
7843 data.hash = hash;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007844 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007845 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007846
Jens Axboed25e3a32021-02-16 11:41:41 -07007847 /* Do QD, or 4 * CPUS, whatever is smallest */
7848 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
Pavel Begunkov24369c22020-01-28 03:15:48 +03007849
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007850 return io_wq_create(concurrency, &data);
Pavel Begunkov24369c22020-01-28 03:15:48 +03007851}
7852
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007853static int io_uring_alloc_task_context(struct task_struct *task,
7854 struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06007855{
7856 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007857 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007858
7859 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
7860 if (unlikely(!tctx))
7861 return -ENOMEM;
7862
Jens Axboed8a6df12020-10-15 16:24:45 -06007863 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7864 if (unlikely(ret)) {
7865 kfree(tctx);
7866 return ret;
7867 }
7868
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007869 tctx->io_wq = io_init_wq_offload(ctx);
7870 if (IS_ERR(tctx->io_wq)) {
7871 ret = PTR_ERR(tctx->io_wq);
7872 percpu_counter_destroy(&tctx->inflight);
7873 kfree(tctx);
7874 return ret;
7875 }
7876
Jens Axboe0f212202020-09-13 13:09:39 -06007877 xa_init(&tctx->xa);
7878 init_waitqueue_head(&tctx->wait);
7879 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06007880 atomic_set(&tctx->in_idle, 0);
Jens Axboe0f212202020-09-13 13:09:39 -06007881 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00007882 spin_lock_init(&tctx->task_lock);
7883 INIT_WQ_LIST(&tctx->task_list);
7884 tctx->task_state = 0;
7885 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06007886 return 0;
7887}
7888
7889void __io_uring_free(struct task_struct *tsk)
7890{
7891 struct io_uring_task *tctx = tsk->io_uring;
7892
7893 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00007894 WARN_ON_ONCE(tctx->io_wq);
7895
Jens Axboed8a6df12020-10-15 16:24:45 -06007896 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06007897 kfree(tctx);
7898 tsk->io_uring = NULL;
7899}
7900
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007901static int io_sq_offload_create(struct io_ring_ctx *ctx,
7902 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007903{
7904 int ret;
7905
Jens Axboed25e3a32021-02-16 11:41:41 -07007906 /* Retain compatibility with failing for an invalid attach attempt */
7907 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
7908 IORING_SETUP_ATTACH_WQ) {
7909 struct fd f;
7910
7911 f = fdget(p->wq_fd);
7912 if (!f.file)
7913 return -ENXIO;
7914 if (f.file->f_op != &io_uring_fops) {
7915 fdput(f);
7916 return -EINVAL;
7917 }
7918 fdput(f);
7919 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07007920 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe46fe18b2021-03-04 12:39:36 -07007921 struct task_struct *tsk;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007922 struct io_sq_data *sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007923 bool attached;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007924
Jens Axboe3ec482d2019-04-08 10:51:01 -06007925 ret = -EPERM;
Jens Axboece59fc62020-09-02 13:28:09 -06007926 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE))
Jens Axboe3ec482d2019-04-08 10:51:01 -06007927 goto err;
7928
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007929 sqd = io_get_sq_data(p, &attached);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007930 if (IS_ERR(sqd)) {
7931 ret = PTR_ERR(sqd);
7932 goto err;
7933 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007934
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01007935 ctx->sq_creds = get_current_cred();
Jens Axboe534ca6d2020-09-02 13:52:19 -06007936 ctx->sq_data = sqd;
Jens Axboe917257d2019-04-13 09:28:55 -06007937 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
7938 if (!ctx->sq_thread_idle)
7939 ctx->sq_thread_idle = HZ;
7940
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007941 ret = 0;
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007942 io_sq_thread_park(sqd);
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007943 list_add(&ctx->sqd_list, &sqd->ctx_list);
7944 io_sqd_update_thread_idle(sqd);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007945 /* don't attach to a dying SQPOLL thread, would be racy */
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007946 if (attached && !sqd->thread)
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007947 ret = -ENXIO;
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007948 io_sq_thread_unpark(sqd);
7949
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007950 if (ret < 0)
7951 goto err;
7952 if (attached)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007953 return 0;
Jens Axboeaa061652020-09-02 14:50:27 -06007954
Jens Axboe6c271ce2019-01-10 11:22:30 -07007955 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06007956 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007957
Jens Axboe917257d2019-04-13 09:28:55 -06007958 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06007959 if (cpu >= nr_cpu_ids)
Jens Axboee8f98f242021-03-09 16:32:13 -07007960 goto err_sqpoll;
Shenghui Wang7889f442019-05-07 16:03:19 +08007961 if (!cpu_online(cpu))
Jens Axboee8f98f242021-03-09 16:32:13 -07007962 goto err_sqpoll;
Jens Axboe917257d2019-04-13 09:28:55 -06007963
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007964 sqd->sq_cpu = cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007965 } else {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007966 sqd->sq_cpu = -1;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007967 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007968
7969 sqd->task_pid = current->pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007970 sqd->task_tgid = current->tgid;
Jens Axboe46fe18b2021-03-04 12:39:36 -07007971 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
7972 if (IS_ERR(tsk)) {
7973 ret = PTR_ERR(tsk);
Jens Axboee8f98f242021-03-09 16:32:13 -07007974 goto err_sqpoll;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007975 }
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007976
Jens Axboe46fe18b2021-03-04 12:39:36 -07007977 sqd->thread = tsk;
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007978 ret = io_uring_alloc_task_context(tsk, ctx);
Jens Axboe46fe18b2021-03-04 12:39:36 -07007979 wake_up_new_task(tsk);
Jens Axboe0f212202020-09-13 13:09:39 -06007980 if (ret)
7981 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007982 } else if (p->flags & IORING_SETUP_SQ_AFF) {
7983 /* Can't have SQ_AFF without SQPOLL */
7984 ret = -EINVAL;
7985 goto err;
7986 }
7987
Jens Axboe2b188cc2019-01-07 10:46:33 -07007988 return 0;
7989err:
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007990 io_sq_thread_finish(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007991 return ret;
Jens Axboee8f98f242021-03-09 16:32:13 -07007992err_sqpoll:
7993 complete(&ctx->sq_data->exited);
7994 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007995}
7996
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007997static inline void __io_unaccount_mem(struct user_struct *user,
7998 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007999{
8000 atomic_long_sub(nr_pages, &user->locked_vm);
8001}
8002
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008003static inline int __io_account_mem(struct user_struct *user,
8004 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008005{
8006 unsigned long page_limit, cur_pages, new_pages;
8007
8008 /* Don't allow more pages than we can safely lock */
8009 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
8010
8011 do {
8012 cur_pages = atomic_long_read(&user->locked_vm);
8013 new_pages = cur_pages + nr_pages;
8014 if (new_pages > page_limit)
8015 return -ENOMEM;
8016 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
8017 new_pages) != cur_pages);
8018
8019 return 0;
8020}
8021
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008022static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008023{
Jens Axboe62e398b2021-02-21 16:19:37 -07008024 if (ctx->user)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008025 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008026
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008027 if (ctx->mm_account)
8028 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008029}
8030
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008031static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008032{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008033 int ret;
8034
Jens Axboe62e398b2021-02-21 16:19:37 -07008035 if (ctx->user) {
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008036 ret = __io_account_mem(ctx->user, nr_pages);
8037 if (ret)
8038 return ret;
8039 }
8040
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008041 if (ctx->mm_account)
8042 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008043
8044 return 0;
8045}
8046
Jens Axboe2b188cc2019-01-07 10:46:33 -07008047static void io_mem_free(void *ptr)
8048{
Mark Rutland52e04ef2019-04-30 17:30:21 +01008049 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008050
Mark Rutland52e04ef2019-04-30 17:30:21 +01008051 if (!ptr)
8052 return;
8053
8054 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008055 if (put_page_testzero(page))
8056 free_compound_page(page);
8057}
8058
8059static void *io_mem_alloc(size_t size)
8060{
8061 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008062 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008063
8064 return (void *) __get_free_pages(gfp_flags, get_order(size));
8065}
8066
Hristo Venev75b28af2019-08-26 17:23:46 +00008067static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
8068 size_t *sq_offset)
8069{
8070 struct io_rings *rings;
8071 size_t off, sq_array_size;
8072
8073 off = struct_size(rings, cqes, cq_entries);
8074 if (off == SIZE_MAX)
8075 return SIZE_MAX;
8076
8077#ifdef CONFIG_SMP
8078 off = ALIGN(off, SMP_CACHE_BYTES);
8079 if (off == 0)
8080 return SIZE_MAX;
8081#endif
8082
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02008083 if (sq_offset)
8084 *sq_offset = off;
8085
Hristo Venev75b28af2019-08-26 17:23:46 +00008086 sq_array_size = array_size(sizeof(u32), sq_entries);
8087 if (sq_array_size == SIZE_MAX)
8088 return SIZE_MAX;
8089
8090 if (check_add_overflow(off, sq_array_size, &off))
8091 return SIZE_MAX;
8092
Hristo Venev75b28af2019-08-26 17:23:46 +00008093 return off;
8094}
8095
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008096static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07008097{
8098 int i, j;
8099
8100 if (!ctx->user_bufs)
8101 return -ENXIO;
8102
8103 for (i = 0; i < ctx->nr_user_bufs; i++) {
8104 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8105
8106 for (j = 0; j < imu->nr_bvecs; j++)
John Hubbardf1f6a7d2020-01-30 22:13:35 -08008107 unpin_user_page(imu->bvec[j].bv_page);
Jens Axboeedafcce2019-01-09 09:16:05 -07008108
Jens Axboede293932020-09-17 16:19:16 -06008109 if (imu->acct_pages)
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008110 io_unaccount_mem(ctx, imu->acct_pages);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008111 kvfree(imu->bvec);
Jens Axboeedafcce2019-01-09 09:16:05 -07008112 imu->nr_bvecs = 0;
8113 }
8114
8115 kfree(ctx->user_bufs);
8116 ctx->user_bufs = NULL;
8117 ctx->nr_user_bufs = 0;
8118 return 0;
8119}
8120
8121static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8122 void __user *arg, unsigned index)
8123{
8124 struct iovec __user *src;
8125
8126#ifdef CONFIG_COMPAT
8127 if (ctx->compat) {
8128 struct compat_iovec __user *ciovs;
8129 struct compat_iovec ciov;
8130
8131 ciovs = (struct compat_iovec __user *) arg;
8132 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8133 return -EFAULT;
8134
Jens Axboed55e5f52019-12-11 16:12:15 -07008135 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008136 dst->iov_len = ciov.iov_len;
8137 return 0;
8138 }
8139#endif
8140 src = (struct iovec __user *) arg;
8141 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8142 return -EFAULT;
8143 return 0;
8144}
8145
Jens Axboede293932020-09-17 16:19:16 -06008146/*
8147 * Not super efficient, but this is just a registration time. And we do cache
8148 * the last compound head, so generally we'll only do a full search if we don't
8149 * match that one.
8150 *
8151 * We check if the given compound head page has already been accounted, to
8152 * avoid double accounting it. This allows us to account the full size of the
8153 * page, not just the constituent pages of a huge page.
8154 */
8155static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8156 int nr_pages, struct page *hpage)
8157{
8158 int i, j;
8159
8160 /* check current page array */
8161 for (i = 0; i < nr_pages; i++) {
8162 if (!PageCompound(pages[i]))
8163 continue;
8164 if (compound_head(pages[i]) == hpage)
8165 return true;
8166 }
8167
8168 /* check previously registered pages */
8169 for (i = 0; i < ctx->nr_user_bufs; i++) {
8170 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8171
8172 for (j = 0; j < imu->nr_bvecs; j++) {
8173 if (!PageCompound(imu->bvec[j].bv_page))
8174 continue;
8175 if (compound_head(imu->bvec[j].bv_page) == hpage)
8176 return true;
8177 }
8178 }
8179
8180 return false;
8181}
8182
8183static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8184 int nr_pages, struct io_mapped_ubuf *imu,
8185 struct page **last_hpage)
8186{
8187 int i, ret;
8188
8189 for (i = 0; i < nr_pages; i++) {
8190 if (!PageCompound(pages[i])) {
8191 imu->acct_pages++;
8192 } else {
8193 struct page *hpage;
8194
8195 hpage = compound_head(pages[i]);
8196 if (hpage == *last_hpage)
8197 continue;
8198 *last_hpage = hpage;
8199 if (headpage_already_acct(ctx, pages, i, hpage))
8200 continue;
8201 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8202 }
8203 }
8204
8205 if (!imu->acct_pages)
8206 return 0;
8207
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008208 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008209 if (ret)
8210 imu->acct_pages = 0;
8211 return ret;
8212}
8213
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008214static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
8215 struct io_mapped_ubuf *imu,
8216 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008217{
8218 struct vm_area_struct **vmas = NULL;
8219 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008220 unsigned long off, start, end, ubuf;
8221 size_t size;
8222 int ret, pret, nr_pages, i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008223
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008224 ubuf = (unsigned long) iov->iov_base;
8225 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8226 start = ubuf >> PAGE_SHIFT;
8227 nr_pages = end - start;
8228
8229 ret = -ENOMEM;
8230
8231 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8232 if (!pages)
8233 goto done;
8234
8235 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8236 GFP_KERNEL);
8237 if (!vmas)
8238 goto done;
8239
8240 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
8241 GFP_KERNEL);
8242 if (!imu->bvec)
8243 goto done;
8244
8245 ret = 0;
8246 mmap_read_lock(current->mm);
8247 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8248 pages, vmas);
8249 if (pret == nr_pages) {
8250 /* don't support file backed memory */
8251 for (i = 0; i < nr_pages; i++) {
8252 struct vm_area_struct *vma = vmas[i];
8253
8254 if (vma->vm_file &&
8255 !is_file_hugepages(vma->vm_file)) {
8256 ret = -EOPNOTSUPP;
8257 break;
8258 }
8259 }
8260 } else {
8261 ret = pret < 0 ? pret : -EFAULT;
8262 }
8263 mmap_read_unlock(current->mm);
8264 if (ret) {
8265 /*
8266 * if we did partial map, or found file backed vmas,
8267 * release any pages we did get
8268 */
8269 if (pret > 0)
8270 unpin_user_pages(pages, pret);
8271 kvfree(imu->bvec);
8272 goto done;
8273 }
8274
8275 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8276 if (ret) {
8277 unpin_user_pages(pages, pret);
8278 kvfree(imu->bvec);
8279 goto done;
8280 }
8281
8282 off = ubuf & ~PAGE_MASK;
8283 size = iov->iov_len;
8284 for (i = 0; i < nr_pages; i++) {
8285 size_t vec_len;
8286
8287 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8288 imu->bvec[i].bv_page = pages[i];
8289 imu->bvec[i].bv_len = vec_len;
8290 imu->bvec[i].bv_offset = off;
8291 off = 0;
8292 size -= vec_len;
8293 }
8294 /* store original address for later verification */
8295 imu->ubuf = ubuf;
8296 imu->len = iov->iov_len;
8297 imu->nr_bvecs = nr_pages;
8298 ret = 0;
8299done:
8300 kvfree(pages);
8301 kvfree(vmas);
8302 return ret;
8303}
8304
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008305static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008306{
Jens Axboeedafcce2019-01-09 09:16:05 -07008307 if (ctx->user_bufs)
8308 return -EBUSY;
8309 if (!nr_args || nr_args > UIO_MAXIOV)
8310 return -EINVAL;
8311
8312 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
8313 GFP_KERNEL);
8314 if (!ctx->user_bufs)
8315 return -ENOMEM;
8316
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008317 return 0;
8318}
8319
8320static int io_buffer_validate(struct iovec *iov)
8321{
8322 /*
8323 * Don't impose further limits on the size and buffer
8324 * constraints here, we'll -EINVAL later when IO is
8325 * submitted if they are wrong.
8326 */
8327 if (!iov->iov_base || !iov->iov_len)
8328 return -EFAULT;
8329
8330 /* arbitrary limit, but we need something */
8331 if (iov->iov_len > SZ_1G)
8332 return -EFAULT;
8333
8334 return 0;
8335}
8336
8337static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
8338 unsigned int nr_args)
8339{
8340 int i, ret;
8341 struct iovec iov;
8342 struct page *last_hpage = NULL;
8343
8344 ret = io_buffers_map_alloc(ctx, nr_args);
8345 if (ret)
8346 return ret;
8347
Jens Axboeedafcce2019-01-09 09:16:05 -07008348 for (i = 0; i < nr_args; i++) {
8349 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
Jens Axboeedafcce2019-01-09 09:16:05 -07008350
8351 ret = io_copy_iov(ctx, &iov, arg, i);
8352 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008353 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008354
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008355 ret = io_buffer_validate(&iov);
8356 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008357 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008358
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008359 ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage);
8360 if (ret)
8361 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008362
8363 ctx->nr_user_bufs++;
8364 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008365
8366 if (ret)
8367 io_sqe_buffers_unregister(ctx);
8368
Jens Axboeedafcce2019-01-09 09:16:05 -07008369 return ret;
8370}
8371
Jens Axboe9b402842019-04-11 11:45:41 -06008372static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8373{
8374 __s32 __user *fds = arg;
8375 int fd;
8376
8377 if (ctx->cq_ev_fd)
8378 return -EBUSY;
8379
8380 if (copy_from_user(&fd, fds, sizeof(*fds)))
8381 return -EFAULT;
8382
8383 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8384 if (IS_ERR(ctx->cq_ev_fd)) {
8385 int ret = PTR_ERR(ctx->cq_ev_fd);
8386 ctx->cq_ev_fd = NULL;
8387 return ret;
8388 }
8389
8390 return 0;
8391}
8392
8393static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8394{
8395 if (ctx->cq_ev_fd) {
8396 eventfd_ctx_put(ctx->cq_ev_fd);
8397 ctx->cq_ev_fd = NULL;
8398 return 0;
8399 }
8400
8401 return -ENXIO;
8402}
8403
Jens Axboe5a2e7452020-02-23 16:23:11 -07008404static void io_destroy_buffers(struct io_ring_ctx *ctx)
8405{
Jens Axboe9e15c3a2021-03-13 12:29:43 -07008406 struct io_buffer *buf;
8407 unsigned long index;
8408
8409 xa_for_each(&ctx->io_buffers, index, buf)
8410 __io_remove_buffers(ctx, buf, index, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008411}
8412
Jens Axboe68e68ee2021-02-13 09:00:02 -07008413static void io_req_cache_free(struct list_head *list, struct task_struct *tsk)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008414{
Jens Axboe68e68ee2021-02-13 09:00:02 -07008415 struct io_kiocb *req, *nxt;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008416
Jens Axboe68e68ee2021-02-13 09:00:02 -07008417 list_for_each_entry_safe(req, nxt, list, compl.list) {
8418 if (tsk && req->task != tsk)
8419 continue;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008420 list_del(&req->compl.list);
8421 kmem_cache_free(req_cachep, req);
8422 }
8423}
8424
Jens Axboe4010fec2021-02-27 15:04:18 -07008425static void io_req_caches_free(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008426{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008427 struct io_submit_state *submit_state = &ctx->submit_state;
Pavel Begunkove5547d22021-02-23 22:17:20 +00008428 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008429
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008430 mutex_lock(&ctx->uring_lock);
8431
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008432 if (submit_state->free_reqs) {
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008433 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8434 submit_state->reqs);
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008435 submit_state->free_reqs = 0;
8436 }
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008437
8438 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkove5547d22021-02-23 22:17:20 +00008439 list_splice_init(&cs->locked_free_list, &cs->free_list);
8440 cs->locked_free_nr = 0;
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008441 spin_unlock_irq(&ctx->completion_lock);
8442
Pavel Begunkove5547d22021-02-23 22:17:20 +00008443 io_req_cache_free(&cs->free_list, NULL);
8444
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008445 mutex_unlock(&ctx->uring_lock);
8446}
8447
Jens Axboe2b188cc2019-01-07 10:46:33 -07008448static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8449{
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00008450 /*
8451 * Some may use context even when all refs and requests have been put,
Pavel Begunkov180f8292021-03-14 20:57:09 +00008452 * and they are free to do so while still holding uring_lock or
8453 * completion_lock, see __io_req_task_submit(). Wait for them to finish.
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00008454 */
8455 mutex_lock(&ctx->uring_lock);
8456 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov180f8292021-03-14 20:57:09 +00008457 spin_lock_irq(&ctx->completion_lock);
8458 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00008459
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008460 io_sq_thread_finish(ctx);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008461 io_sqe_buffers_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008462
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008463 if (ctx->mm_account) {
Jens Axboe2aede0e2020-09-14 10:45:53 -06008464 mmdrop(ctx->mm_account);
8465 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008466 }
Jens Axboedef596e2019-01-09 08:59:42 -07008467
Hao Xu8bad28d2021-02-19 17:19:36 +08008468 mutex_lock(&ctx->uring_lock);
Jens Axboe6b063142019-01-10 22:13:58 -07008469 io_sqe_files_unregister(ctx);
Hao Xu8bad28d2021-02-19 17:19:36 +08008470 mutex_unlock(&ctx->uring_lock);
Jens Axboe9b402842019-04-11 11:45:41 -06008471 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008472 io_destroy_buffers(ctx);
Jens Axboedef596e2019-01-09 08:59:42 -07008473
Jens Axboe2b188cc2019-01-07 10:46:33 -07008474#if defined(CONFIG_UNIX)
Eric Biggers355e8d22019-06-12 14:58:43 -07008475 if (ctx->ring_sock) {
8476 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008477 sock_release(ctx->ring_sock);
Eric Biggers355e8d22019-06-12 14:58:43 -07008478 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008479#endif
8480
Hristo Venev75b28af2019-08-26 17:23:46 +00008481 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008482 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008483
8484 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008485 free_uid(ctx->user);
Jens Axboe4010fec2021-02-27 15:04:18 -07008486 io_req_caches_free(ctx);
Jens Axboee9418942021-02-19 12:33:30 -07008487 if (ctx->hash_map)
8488 io_wq_put_hash(ctx->hash_map);
Jens Axboe78076bb2019-12-04 19:56:40 -07008489 kfree(ctx->cancel_hash);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008490 kfree(ctx);
8491}
8492
8493static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8494{
8495 struct io_ring_ctx *ctx = file->private_data;
8496 __poll_t mask = 0;
8497
8498 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008499 /*
8500 * synchronizes with barrier from wq_has_sleeper call in
8501 * io_commit_cqring
8502 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008503 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008504 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008505 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008506
8507 /*
8508 * Don't flush cqring overflow list here, just do a simple check.
8509 * Otherwise there could possible be ABBA deadlock:
8510 * CPU0 CPU1
8511 * ---- ----
8512 * lock(&ctx->uring_lock);
8513 * lock(&ep->mtx);
8514 * lock(&ctx->uring_lock);
8515 * lock(&ep->mtx);
8516 *
8517 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8518 * pushs them to do the flush.
8519 */
8520 if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008521 mask |= EPOLLIN | EPOLLRDNORM;
8522
8523 return mask;
8524}
8525
8526static int io_uring_fasync(int fd, struct file *file, int on)
8527{
8528 struct io_ring_ctx *ctx = file->private_data;
8529
8530 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8531}
8532
Yejune Deng0bead8c2020-12-24 11:02:20 +08008533static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008534{
Jens Axboe4379bf82021-02-15 13:40:22 -07008535 const struct cred *creds;
Jens Axboe071698e2020-01-28 10:04:42 -07008536
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008537 creds = xa_erase(&ctx->personalities, id);
Jens Axboe4379bf82021-02-15 13:40:22 -07008538 if (creds) {
8539 put_cred(creds);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008540 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008541 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008542
8543 return -EINVAL;
8544}
8545
Pavel Begunkov9b465712021-03-15 14:23:07 +00008546static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008547{
Pavel Begunkov9b465712021-03-15 14:23:07 +00008548 return io_run_task_work_head(&ctx->exit_task_work);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008549}
8550
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008551struct io_tctx_exit {
8552 struct callback_head task_work;
8553 struct completion completion;
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008554 struct io_ring_ctx *ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008555};
8556
8557static void io_tctx_exit_cb(struct callback_head *cb)
8558{
8559 struct io_uring_task *tctx = current->io_uring;
8560 struct io_tctx_exit *work;
8561
8562 work = container_of(cb, struct io_tctx_exit, task_work);
8563 /*
8564 * When @in_idle, we're in cancellation and it's racy to remove the
8565 * node. It'll be removed by the end of cancellation, just ignore it.
8566 */
8567 if (!atomic_read(&tctx->in_idle))
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008568 io_uring_del_task_file((unsigned long)work->ctx);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008569 complete(&work->completion);
8570}
8571
Jens Axboe85faa7b2020-04-09 18:14:00 -06008572static void io_ring_exit_work(struct work_struct *work)
8573{
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008574 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008575 unsigned long timeout = jiffies + HZ * 60 * 5;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008576 struct io_tctx_exit exit;
8577 struct io_tctx_node *node;
8578 int ret;
Jens Axboe85faa7b2020-04-09 18:14:00 -06008579
Pavel Begunkova185f1d2021-03-23 10:52:38 +00008580 /* prevent SQPOLL from submitting new requests */
8581 if (ctx->sq_data) {
8582 io_sq_thread_park(ctx->sq_data);
8583 list_del_init(&ctx->sqd_list);
8584 io_sqd_update_thread_idle(ctx->sq_data);
8585 io_sq_thread_unpark(ctx->sq_data);
8586 }
8587
Jens Axboe56952e92020-06-17 15:00:04 -06008588 /*
8589 * If we're doing polled IO and end up having requests being
8590 * submitted async (out-of-line), then completions can come in while
8591 * we're waiting for refs to drop. We need to reap these manually,
8592 * as nobody else will be looking for them.
8593 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008594 do {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008595 io_uring_try_cancel_requests(ctx, NULL, NULL);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008596
8597 WARN_ON_ONCE(time_after(jiffies, timeout));
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008598 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008599
8600 mutex_lock(&ctx->uring_lock);
8601 while (!list_empty(&ctx->tctx_list)) {
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008602 WARN_ON_ONCE(time_after(jiffies, timeout));
8603
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008604 node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
8605 ctx_node);
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008606 exit.ctx = ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008607 init_completion(&exit.completion);
8608 init_task_work(&exit.task_work, io_tctx_exit_cb);
8609 ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
8610 if (WARN_ON_ONCE(ret))
8611 continue;
8612 wake_up_process(node->task);
8613
8614 mutex_unlock(&ctx->uring_lock);
8615 wait_for_completion(&exit.completion);
8616 cond_resched();
8617 mutex_lock(&ctx->uring_lock);
8618 }
8619 mutex_unlock(&ctx->uring_lock);
8620
Jens Axboe85faa7b2020-04-09 18:14:00 -06008621 io_ring_ctx_free(ctx);
8622}
8623
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008624/* Returns true if we found and killed one or more timeouts */
8625static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
8626 struct files_struct *files)
8627{
8628 struct io_kiocb *req, *tmp;
8629 int canceled = 0;
8630
8631 spin_lock_irq(&ctx->completion_lock);
8632 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
8633 if (io_match_task(req, tsk, files)) {
8634 io_kill_timeout(req, -ECANCELED);
8635 canceled++;
8636 }
8637 }
Pavel Begunkov51520422021-03-29 11:39:29 +01008638 if (canceled != 0)
8639 io_commit_cqring(ctx);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008640 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008641 if (canceled != 0)
8642 io_cqring_ev_posted(ctx);
8643 return canceled != 0;
8644}
8645
Jens Axboe2b188cc2019-01-07 10:46:33 -07008646static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8647{
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008648 unsigned long index;
8649 struct creds *creds;
8650
Jens Axboe2b188cc2019-01-07 10:46:33 -07008651 mutex_lock(&ctx->uring_lock);
8652 percpu_ref_kill(&ctx->refs);
Pavel Begunkovcda286f2020-12-17 00:24:35 +00008653 /* if force is set, the ring is going away. always drop after that */
8654 ctx->cq_overflow_flushed = 1;
Pavel Begunkov634578f2020-12-06 22:22:44 +00008655 if (ctx->rings)
Pavel Begunkov6c503152021-01-04 20:36:36 +00008656 __io_cqring_overflow_flush(ctx, true, NULL, NULL);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008657 xa_for_each(&ctx->personalities, index, creds)
8658 io_unregister_personality(ctx, index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008659 mutex_unlock(&ctx->uring_lock);
8660
Pavel Begunkov6b819282020-11-06 13:00:25 +00008661 io_kill_timeouts(ctx, NULL, NULL);
8662 io_poll_remove_all(ctx, NULL, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008663
Jens Axboe15dff282019-11-13 09:09:23 -07008664 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008665 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008666
Jens Axboe85faa7b2020-04-09 18:14:00 -06008667 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008668 /*
8669 * Use system_unbound_wq to avoid spawning tons of event kworkers
8670 * if we're exiting a ton of rings at the same time. It just adds
8671 * noise and overhead, there's no discernable change in runtime
8672 * over using system_wq.
8673 */
8674 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008675}
8676
8677static int io_uring_release(struct inode *inode, struct file *file)
8678{
8679 struct io_ring_ctx *ctx = file->private_data;
8680
8681 file->private_data = NULL;
8682 io_ring_ctx_wait_and_kill(ctx);
8683 return 0;
8684}
8685
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008686struct io_task_cancel {
8687 struct task_struct *task;
8688 struct files_struct *files;
8689};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008690
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008691static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008692{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008693 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008694 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008695 bool ret;
8696
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008697 if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008698 unsigned long flags;
8699 struct io_ring_ctx *ctx = req->ctx;
8700
8701 /* protect against races with linked timeouts */
8702 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008703 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008704 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8705 } else {
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008706 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008707 }
8708 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008709}
8710
Pavel Begunkove1915f72021-03-11 23:29:35 +00008711static bool io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008712 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008713 struct files_struct *files)
8714{
Pavel Begunkove1915f72021-03-11 23:29:35 +00008715 struct io_defer_entry *de;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008716 LIST_HEAD(list);
8717
8718 spin_lock_irq(&ctx->completion_lock);
8719 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00008720 if (io_match_task(de->req, task, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008721 list_cut_position(&list, &ctx->defer_list, &de->list);
8722 break;
8723 }
8724 }
8725 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkove1915f72021-03-11 23:29:35 +00008726 if (list_empty(&list))
8727 return false;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008728
8729 while (!list_empty(&list)) {
8730 de = list_first_entry(&list, struct io_defer_entry, list);
8731 list_del_init(&de->list);
Pavel Begunkovf41db2732021-02-28 22:35:12 +00008732 io_req_complete_failed(de->req, -ECANCELED);
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008733 kfree(de);
8734 }
Pavel Begunkove1915f72021-03-11 23:29:35 +00008735 return true;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008736}
8737
Pavel Begunkov1b007642021-03-06 11:02:17 +00008738static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data)
8739{
8740 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8741
8742 return req->ctx == data;
8743}
8744
8745static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
8746{
8747 struct io_tctx_node *node;
8748 enum io_wq_cancel cret;
8749 bool ret = false;
8750
8751 mutex_lock(&ctx->uring_lock);
8752 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
8753 struct io_uring_task *tctx = node->task->io_uring;
8754
8755 /*
8756 * io_wq will stay alive while we hold uring_lock, because it's
8757 * killed after ctx nodes, which requires to take the lock.
8758 */
8759 if (!tctx || !tctx->io_wq)
8760 continue;
8761 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true);
8762 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8763 }
8764 mutex_unlock(&ctx->uring_lock);
8765
8766 return ret;
8767}
8768
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008769static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
8770 struct task_struct *task,
8771 struct files_struct *files)
8772{
8773 struct io_task_cancel cancel = { .task = task, .files = files, };
Pavel Begunkov1b007642021-03-06 11:02:17 +00008774 struct io_uring_task *tctx = task ? task->io_uring : NULL;
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008775
8776 while (1) {
8777 enum io_wq_cancel cret;
8778 bool ret = false;
8779
Pavel Begunkov1b007642021-03-06 11:02:17 +00008780 if (!task) {
8781 ret |= io_uring_try_cancel_iowq(ctx);
8782 } else if (tctx && tctx->io_wq) {
8783 /*
8784 * Cancels requests of all rings, not only @ctx, but
8785 * it's fine as the task is in exit/exec.
8786 */
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008787 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008788 &cancel, true);
8789 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8790 }
8791
8792 /* SQPOLL thread does its own polling */
Jens Axboed052d1d2021-03-11 10:49:20 -07008793 if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) ||
8794 (ctx->sq_data && ctx->sq_data->thread == current)) {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008795 while (!list_empty_careful(&ctx->iopoll_list)) {
8796 io_iopoll_try_reap_events(ctx);
8797 ret = true;
8798 }
8799 }
8800
Pavel Begunkove1915f72021-03-11 23:29:35 +00008801 ret |= io_cancel_defer_files(ctx, task, files);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008802 ret |= io_poll_remove_all(ctx, task, files);
8803 ret |= io_kill_timeouts(ctx, task, files);
8804 ret |= io_run_task_work();
Pavel Begunkovba50a032021-02-26 15:47:56 +00008805 ret |= io_run_ctx_fallback(ctx);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008806 io_cqring_overflow_flush(ctx, true, task, files);
8807 if (!ret)
8808 break;
8809 cond_resched();
8810 }
8811}
8812
Pavel Begunkovca70f002021-01-26 15:28:27 +00008813static int io_uring_count_inflight(struct io_ring_ctx *ctx,
8814 struct task_struct *task,
8815 struct files_struct *files)
8816{
8817 struct io_kiocb *req;
8818 int cnt = 0;
8819
8820 spin_lock_irq(&ctx->inflight_lock);
8821 list_for_each_entry(req, &ctx->inflight_list, inflight_entry)
8822 cnt += io_match_task(req, task, files);
8823 spin_unlock_irq(&ctx->inflight_lock);
8824 return cnt;
8825}
8826
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008827static void io_uring_cancel_files(struct io_ring_ctx *ctx,
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008828 struct task_struct *task,
Jens Axboefcb323c2019-10-24 12:39:47 -06008829 struct files_struct *files)
8830{
Jens Axboefcb323c2019-10-24 12:39:47 -06008831 while (!list_empty_careful(&ctx->inflight_list)) {
Xiaoguang Wangd8f1b972020-04-26 15:54:43 +08008832 DEFINE_WAIT(wait);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008833 int inflight;
Jens Axboefcb323c2019-10-24 12:39:47 -06008834
Pavel Begunkovca70f002021-01-26 15:28:27 +00008835 inflight = io_uring_count_inflight(ctx, task, files);
8836 if (!inflight)
Jens Axboefcb323c2019-10-24 12:39:47 -06008837 break;
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008838
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008839 io_uring_try_cancel_requests(ctx, task, files);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008840
8841 prepare_to_wait(&task->io_uring->wait, &wait,
8842 TASK_UNINTERRUPTIBLE);
8843 if (inflight == io_uring_count_inflight(ctx, task, files))
8844 schedule();
Pavel Begunkovc98de082020-11-15 12:56:32 +00008845 finish_wait(&task->io_uring->wait, &wait);
Jens Axboe0f212202020-09-13 13:09:39 -06008846 }
Jens Axboe0f212202020-09-13 13:09:39 -06008847}
8848
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008849static int __io_uring_add_task_file(struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008850{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008851 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008852 struct io_tctx_node *node;
Pavel Begunkova528b042020-12-21 18:34:04 +00008853 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008854
8855 if (unlikely(!tctx)) {
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008856 ret = io_uring_alloc_task_context(current, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06008857 if (unlikely(ret))
8858 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008859 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008860 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008861 if (!xa_load(&tctx->xa, (unsigned long)ctx)) {
8862 node = kmalloc(sizeof(*node), GFP_KERNEL);
8863 if (!node)
8864 return -ENOMEM;
8865 node->ctx = ctx;
8866 node->task = current;
Jens Axboe0f212202020-09-13 13:09:39 -06008867
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008868 ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx,
8869 node, GFP_KERNEL));
8870 if (ret) {
8871 kfree(node);
8872 return ret;
Jens Axboe0f212202020-09-13 13:09:39 -06008873 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008874
8875 mutex_lock(&ctx->uring_lock);
8876 list_add(&node->ctx_node, &ctx->tctx_list);
8877 mutex_unlock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06008878 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008879 tctx->last = ctx;
Jens Axboe0f212202020-09-13 13:09:39 -06008880 return 0;
8881}
8882
8883/*
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008884 * Note that this task has used io_uring. We use it for cancelation purposes.
8885 */
8886static inline int io_uring_add_task_file(struct io_ring_ctx *ctx)
8887{
8888 struct io_uring_task *tctx = current->io_uring;
8889
8890 if (likely(tctx && tctx->last == ctx))
8891 return 0;
8892 return __io_uring_add_task_file(ctx);
8893}
8894
8895/*
Jens Axboe0f212202020-09-13 13:09:39 -06008896 * Remove this io_uring_file -> task mapping.
8897 */
Pavel Begunkov29412672021-03-06 11:02:11 +00008898static void io_uring_del_task_file(unsigned long index)
Jens Axboe0f212202020-09-13 13:09:39 -06008899{
8900 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008901 struct io_tctx_node *node;
Pavel Begunkov29412672021-03-06 11:02:11 +00008902
Pavel Begunkoveebd2e32021-03-06 11:02:14 +00008903 if (!tctx)
8904 return;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008905 node = xa_erase(&tctx->xa, index);
8906 if (!node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008907 return;
Jens Axboe0f212202020-09-13 13:09:39 -06008908
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008909 WARN_ON_ONCE(current != node->task);
8910 WARN_ON_ONCE(list_empty(&node->ctx_node));
8911
8912 mutex_lock(&node->ctx->uring_lock);
8913 list_del(&node->ctx_node);
8914 mutex_unlock(&node->ctx->uring_lock);
8915
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008916 if (tctx->last == node->ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008917 tctx->last = NULL;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008918 kfree(node);
Jens Axboe0f212202020-09-13 13:09:39 -06008919}
8920
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008921static void io_uring_clean_tctx(struct io_uring_task *tctx)
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008922{
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008923 struct io_tctx_node *node;
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008924 unsigned long index;
8925
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008926 xa_for_each(&tctx->xa, index, node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008927 io_uring_del_task_file(index);
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008928 if (tctx->io_wq) {
8929 io_wq_put_and_exit(tctx->io_wq);
8930 tctx->io_wq = NULL;
8931 }
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008932}
8933
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008934static s64 tctx_inflight(struct io_uring_task *tctx)
8935{
8936 return percpu_counter_sum(&tctx->inflight);
8937}
8938
8939static void io_sqpoll_cancel_cb(struct callback_head *cb)
8940{
8941 struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work);
8942 struct io_ring_ctx *ctx = work->ctx;
8943 struct io_sq_data *sqd = ctx->sq_data;
8944
8945 if (sqd->thread)
8946 io_uring_cancel_sqpoll(ctx);
8947 complete(&work->completion);
8948}
8949
8950static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx)
8951{
8952 struct io_sq_data *sqd = ctx->sq_data;
8953 struct io_tctx_exit work = { .ctx = ctx, };
8954 struct task_struct *task;
8955
8956 io_sq_thread_park(sqd);
8957 list_del_init(&ctx->sqd_list);
8958 io_sqd_update_thread_idle(sqd);
8959 task = sqd->thread;
8960 if (task) {
8961 init_completion(&work.completion);
8962 init_task_work(&work.task_work, io_sqpoll_cancel_cb);
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00008963 io_task_work_add_head(&sqd->park_task_work, &work.task_work);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008964 wake_up_process(task);
8965 }
8966 io_sq_thread_unpark(sqd);
8967
8968 if (task)
8969 wait_for_completion(&work.completion);
8970}
8971
Jens Axboe0f212202020-09-13 13:09:39 -06008972void __io_uring_files_cancel(struct files_struct *files)
8973{
8974 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008975 struct io_tctx_node *node;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008976 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06008977
8978 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008979 atomic_inc(&tctx->in_idle);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008980 xa_for_each(&tctx->xa, index, node) {
8981 struct io_ring_ctx *ctx = node->ctx;
8982
8983 if (ctx->sq_data) {
8984 io_sqpoll_cancel_sync(ctx);
8985 continue;
8986 }
8987 io_uring_cancel_files(ctx, current, files);
8988 if (!files)
8989 io_uring_try_cancel_requests(ctx, current, NULL);
8990 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008991 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008992
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008993 if (files)
8994 io_uring_clean_tctx(tctx);
Jens Axboefdaf0832020-10-30 09:37:30 -06008995}
8996
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008997/* should only be called by SQPOLL task */
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008998static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx)
8999{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07009000 struct io_sq_data *sqd = ctx->sq_data;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00009001 struct io_uring_task *tctx = current->io_uring;
Jens Axboefdaf0832020-10-30 09:37:30 -06009002 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009003 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06009004
Pavel Begunkov521d6a72021-03-11 23:29:38 +00009005 WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current);
9006
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009007 atomic_inc(&tctx->in_idle);
9008 do {
9009 /* read completions before cancelations */
9010 inflight = tctx_inflight(tctx);
9011 if (!inflight)
9012 break;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00009013 io_uring_try_cancel_requests(ctx, current, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06009014
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009015 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
9016 /*
9017 * If we've seen completions, retry without waiting. This
9018 * avoids a race where a completion comes in before we did
9019 * prepare_to_wait().
9020 */
9021 if (inflight == tctx_inflight(tctx))
9022 schedule();
9023 finish_wait(&tctx->wait, &wait);
9024 } while (1);
9025 atomic_dec(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06009026}
9027
Jens Axboe0f212202020-09-13 13:09:39 -06009028/*
9029 * Find any io_uring fd that this task has registered or done IO on, and cancel
9030 * requests.
9031 */
9032void __io_uring_task_cancel(void)
9033{
9034 struct io_uring_task *tctx = current->io_uring;
9035 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06009036 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06009037
9038 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06009039 atomic_inc(&tctx->in_idle);
Pavel Begunkov5a978dc2021-03-27 09:59:30 +00009040 __io_uring_files_cancel(NULL);
9041
Jens Axboed8a6df12020-10-15 16:24:45 -06009042 do {
Jens Axboe0f212202020-09-13 13:09:39 -06009043 /* read completions before cancelations */
Jens Axboefdaf0832020-10-30 09:37:30 -06009044 inflight = tctx_inflight(tctx);
Jens Axboed8a6df12020-10-15 16:24:45 -06009045 if (!inflight)
9046 break;
Jens Axboe0f212202020-09-13 13:09:39 -06009047 __io_uring_files_cancel(NULL);
9048
9049 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
9050
9051 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009052 * If we've seen completions, retry without waiting. This
9053 * avoids a race where a completion comes in before we did
9054 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06009055 */
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009056 if (inflight == tctx_inflight(tctx))
9057 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00009058 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06009059 } while (1);
Jens Axboe0f212202020-09-13 13:09:39 -06009060
Jens Axboefdaf0832020-10-30 09:37:30 -06009061 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009062
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00009063 io_uring_clean_tctx(tctx);
9064 /* all current's requests should be gone, we can kill tctx */
9065 __io_uring_free(current);
Pavel Begunkov44e728b2020-06-15 10:24:04 +03009066}
9067
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009068static void *io_uring_validate_mmap_request(struct file *file,
9069 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009070{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009071 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009072 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009073 struct page *page;
9074 void *ptr;
9075
9076 switch (offset) {
9077 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00009078 case IORING_OFF_CQ_RING:
9079 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009080 break;
9081 case IORING_OFF_SQES:
9082 ptr = ctx->sq_sqes;
9083 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009084 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009085 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009086 }
9087
9088 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07009089 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009090 return ERR_PTR(-EINVAL);
9091
9092 return ptr;
9093}
9094
9095#ifdef CONFIG_MMU
9096
9097static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9098{
9099 size_t sz = vma->vm_end - vma->vm_start;
9100 unsigned long pfn;
9101 void *ptr;
9102
9103 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
9104 if (IS_ERR(ptr))
9105 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009106
9107 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
9108 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
9109}
9110
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009111#else /* !CONFIG_MMU */
9112
9113static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9114{
9115 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
9116}
9117
9118static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
9119{
9120 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
9121}
9122
9123static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
9124 unsigned long addr, unsigned long len,
9125 unsigned long pgoff, unsigned long flags)
9126{
9127 void *ptr;
9128
9129 ptr = io_uring_validate_mmap_request(file, pgoff, len);
9130 if (IS_ERR(ptr))
9131 return PTR_ERR(ptr);
9132
9133 return (unsigned long) ptr;
9134}
9135
9136#endif /* !CONFIG_MMU */
9137
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009138static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06009139{
9140 DEFINE_WAIT(wait);
9141
9142 do {
9143 if (!io_sqring_full(ctx))
9144 break;
Jens Axboe90554202020-09-03 12:12:41 -06009145 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
9146
9147 if (!io_sqring_full(ctx))
9148 break;
Jens Axboe90554202020-09-03 12:12:41 -06009149 schedule();
9150 } while (!signal_pending(current));
9151
9152 finish_wait(&ctx->sqo_sq_wait, &wait);
Yang Li51993282021-03-09 14:30:41 +08009153 return 0;
Jens Axboe90554202020-09-03 12:12:41 -06009154}
9155
Hao Xuc73ebb62020-11-03 10:54:37 +08009156static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9157 struct __kernel_timespec __user **ts,
9158 const sigset_t __user **sig)
9159{
9160 struct io_uring_getevents_arg arg;
9161
9162 /*
9163 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9164 * is just a pointer to the sigset_t.
9165 */
9166 if (!(flags & IORING_ENTER_EXT_ARG)) {
9167 *sig = (const sigset_t __user *) argp;
9168 *ts = NULL;
9169 return 0;
9170 }
9171
9172 /*
9173 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9174 * timespec and sigset_t pointers if good.
9175 */
9176 if (*argsz != sizeof(arg))
9177 return -EINVAL;
9178 if (copy_from_user(&arg, argp, sizeof(arg)))
9179 return -EFAULT;
9180 *sig = u64_to_user_ptr(arg.sigmask);
9181 *argsz = arg.sigmask_sz;
9182 *ts = u64_to_user_ptr(arg.ts);
9183 return 0;
9184}
9185
Jens Axboe2b188cc2019-01-07 10:46:33 -07009186SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009187 u32, min_complete, u32, flags, const void __user *, argp,
9188 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009189{
9190 struct io_ring_ctx *ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009191 int submitted = 0;
9192 struct fd f;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009193 long ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009194
Jens Axboe4c6e2772020-07-01 11:29:10 -06009195 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009196
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009197 if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
9198 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009199 return -EINVAL;
9200
9201 f = fdget(fd);
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009202 if (unlikely(!f.file))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009203 return -EBADF;
9204
9205 ret = -EOPNOTSUPP;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009206 if (unlikely(f.file->f_op != &io_uring_fops))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009207 goto out_fput;
9208
9209 ret = -ENXIO;
9210 ctx = f.file->private_data;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009211 if (unlikely(!percpu_ref_tryget(&ctx->refs)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009212 goto out_fput;
9213
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009214 ret = -EBADFD;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009215 if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED))
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009216 goto out;
9217
Jens Axboe6c271ce2019-01-10 11:22:30 -07009218 /*
9219 * For SQ polling, the thread will do all submissions and completions.
9220 * Just return the requested submit count, and wake the thread if
9221 * we were asked to.
9222 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009223 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009224 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00009225 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009226
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009227 ret = -EOWNERDEAD;
Stefan Metzmacher04147482021-03-07 11:54:29 +01009228 if (unlikely(ctx->sq_data->thread == NULL)) {
9229 goto out;
9230 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009231 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009232 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009233 if (flags & IORING_ENTER_SQ_WAIT) {
9234 ret = io_sqpoll_wait_sq(ctx);
9235 if (ret)
9236 goto out;
9237 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009238 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009239 } else if (to_submit) {
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009240 ret = io_uring_add_task_file(ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06009241 if (unlikely(ret))
9242 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009243 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009244 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009245 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009246
9247 if (submitted != to_submit)
9248 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009249 }
9250 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009251 const sigset_t __user *sig;
9252 struct __kernel_timespec __user *ts;
9253
9254 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9255 if (unlikely(ret))
9256 goto out;
9257
Jens Axboe2b188cc2019-01-07 10:46:33 -07009258 min_complete = min(min_complete, ctx->cq_entries);
9259
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009260 /*
9261 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9262 * space applications don't need to do io completion events
9263 * polling again, they can rely on io_sq_thread to do polling
9264 * work, which can reduce cpu usage and uring_lock contention.
9265 */
9266 if (ctx->flags & IORING_SETUP_IOPOLL &&
9267 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009268 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009269 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009270 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009271 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009272 }
9273
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009274out:
Pavel Begunkov6805b322019-10-08 02:18:42 +03009275 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009276out_fput:
9277 fdput(f);
9278 return submitted ? submitted : ret;
9279}
9280
Tobias Klauserbebdb652020-02-26 18:38:32 +01009281#ifdef CONFIG_PROC_FS
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009282static int io_uring_show_cred(struct seq_file *m, unsigned int id,
9283 const struct cred *cred)
Jens Axboe87ce9552020-01-30 08:25:34 -07009284{
Jens Axboe87ce9552020-01-30 08:25:34 -07009285 struct user_namespace *uns = seq_user_ns(m);
9286 struct group_info *gi;
9287 kernel_cap_t cap;
9288 unsigned __capi;
9289 int g;
9290
9291 seq_printf(m, "%5d\n", id);
9292 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9293 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9294 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9295 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9296 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9297 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9298 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9299 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9300 seq_puts(m, "\n\tGroups:\t");
9301 gi = cred->group_info;
9302 for (g = 0; g < gi->ngroups; g++) {
9303 seq_put_decimal_ull(m, g ? " " : "",
9304 from_kgid_munged(uns, gi->gid[g]));
9305 }
9306 seq_puts(m, "\n\tCapEff:\t");
9307 cap = cred->cap_effective;
9308 CAP_FOR_EACH_U32(__capi)
9309 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9310 seq_putc(m, '\n');
9311 return 0;
9312}
9313
9314static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9315{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009316 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009317 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009318 int i;
9319
Jens Axboefad8e0d2020-09-28 08:57:48 -06009320 /*
9321 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9322 * since fdinfo case grabs it in the opposite direction of normal use
9323 * cases. If we fail to get the lock, we just don't iterate any
9324 * structures that could be going away outside the io_uring mutex.
9325 */
9326 has_lock = mutex_trylock(&ctx->uring_lock);
9327
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009328 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
Joseph Qidbbe9c62020-09-29 09:01:22 -06009329 sq = ctx->sq_data;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009330 if (!sq->thread)
9331 sq = NULL;
9332 }
Joseph Qidbbe9c62020-09-29 09:01:22 -06009333
9334 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9335 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009336 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009337 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Jens Axboe7b29f922021-03-12 08:30:14 -07009338 struct file *f = io_file_from_index(ctx, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009339
Jens Axboe87ce9552020-01-30 08:25:34 -07009340 if (f)
9341 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9342 else
9343 seq_printf(m, "%5u: <none>\n", i);
9344 }
9345 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009346 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009347 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
9348
9349 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf,
9350 (unsigned int) buf->len);
9351 }
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009352 if (has_lock && !xa_empty(&ctx->personalities)) {
9353 unsigned long index;
9354 const struct cred *cred;
9355
Jens Axboe87ce9552020-01-30 08:25:34 -07009356 seq_printf(m, "Personalities:\n");
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009357 xa_for_each(&ctx->personalities, index, cred)
9358 io_uring_show_cred(m, index, cred);
Jens Axboe87ce9552020-01-30 08:25:34 -07009359 }
Jens Axboed7718a92020-02-14 22:23:12 -07009360 seq_printf(m, "PollList:\n");
9361 spin_lock_irq(&ctx->completion_lock);
9362 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9363 struct hlist_head *list = &ctx->cancel_hash[i];
9364 struct io_kiocb *req;
9365
9366 hlist_for_each_entry(req, list, hash_node)
9367 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9368 req->task->task_works != NULL);
9369 }
9370 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009371 if (has_lock)
9372 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009373}
9374
9375static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9376{
9377 struct io_ring_ctx *ctx = f->private_data;
9378
9379 if (percpu_ref_tryget(&ctx->refs)) {
9380 __io_uring_show_fdinfo(ctx, m);
9381 percpu_ref_put(&ctx->refs);
9382 }
9383}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009384#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009385
Jens Axboe2b188cc2019-01-07 10:46:33 -07009386static const struct file_operations io_uring_fops = {
9387 .release = io_uring_release,
9388 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009389#ifndef CONFIG_MMU
9390 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9391 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9392#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009393 .poll = io_uring_poll,
9394 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009395#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009396 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009397#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009398};
9399
9400static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9401 struct io_uring_params *p)
9402{
Hristo Venev75b28af2019-08-26 17:23:46 +00009403 struct io_rings *rings;
9404 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009405
Jens Axboebd740482020-08-05 12:58:23 -06009406 /* make sure these are sane, as we already accounted them */
9407 ctx->sq_entries = p->sq_entries;
9408 ctx->cq_entries = p->cq_entries;
9409
Hristo Venev75b28af2019-08-26 17:23:46 +00009410 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9411 if (size == SIZE_MAX)
9412 return -EOVERFLOW;
9413
9414 rings = io_mem_alloc(size);
9415 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009416 return -ENOMEM;
9417
Hristo Venev75b28af2019-08-26 17:23:46 +00009418 ctx->rings = rings;
9419 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9420 rings->sq_ring_mask = p->sq_entries - 1;
9421 rings->cq_ring_mask = p->cq_entries - 1;
9422 rings->sq_ring_entries = p->sq_entries;
9423 rings->cq_ring_entries = p->cq_entries;
9424 ctx->sq_mask = rings->sq_ring_mask;
9425 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009426
9427 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009428 if (size == SIZE_MAX) {
9429 io_mem_free(ctx->rings);
9430 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009431 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009432 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009433
9434 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009435 if (!ctx->sq_sqes) {
9436 io_mem_free(ctx->rings);
9437 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009438 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009439 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009440
Jens Axboe2b188cc2019-01-07 10:46:33 -07009441 return 0;
9442}
9443
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009444static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9445{
9446 int ret, fd;
9447
9448 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9449 if (fd < 0)
9450 return fd;
9451
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009452 ret = io_uring_add_task_file(ctx);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009453 if (ret) {
9454 put_unused_fd(fd);
9455 return ret;
9456 }
9457 fd_install(fd, file);
9458 return fd;
9459}
9460
Jens Axboe2b188cc2019-01-07 10:46:33 -07009461/*
9462 * Allocate an anonymous fd, this is what constitutes the application
9463 * visible backing of an io_uring instance. The application mmaps this
9464 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9465 * we have to tie this fd to a socket for file garbage collection purposes.
9466 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009467static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009468{
9469 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009470#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009471 int ret;
9472
Jens Axboe2b188cc2019-01-07 10:46:33 -07009473 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9474 &ctx->ring_sock);
9475 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009476 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009477#endif
9478
Jens Axboe2b188cc2019-01-07 10:46:33 -07009479 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9480 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009481#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009482 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009483 sock_release(ctx->ring_sock);
9484 ctx->ring_sock = NULL;
9485 } else {
9486 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009487 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009488#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009489 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009490}
9491
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009492static int io_uring_create(unsigned entries, struct io_uring_params *p,
9493 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009494{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009495 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009496 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009497 int ret;
9498
Jens Axboe8110c1a2019-12-28 15:39:54 -07009499 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009500 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009501 if (entries > IORING_MAX_ENTRIES) {
9502 if (!(p->flags & IORING_SETUP_CLAMP))
9503 return -EINVAL;
9504 entries = IORING_MAX_ENTRIES;
9505 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009506
9507 /*
9508 * Use twice as many entries for the CQ ring. It's possible for the
9509 * application to drive a higher depth than the size of the SQ ring,
9510 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009511 * some flexibility in overcommitting a bit. If the application has
9512 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9513 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009514 */
9515 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009516 if (p->flags & IORING_SETUP_CQSIZE) {
9517 /*
9518 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9519 * to a power-of-two, if it isn't already. We do NOT impose
9520 * any cq vs sq ring sizing.
9521 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009522 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009523 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009524 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9525 if (!(p->flags & IORING_SETUP_CLAMP))
9526 return -EINVAL;
9527 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9528 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009529 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9530 if (p->cq_entries < p->sq_entries)
9531 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009532 } else {
9533 p->cq_entries = 2 * p->sq_entries;
9534 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009535
Jens Axboe2b188cc2019-01-07 10:46:33 -07009536 ctx = io_ring_ctx_alloc(p);
Jens Axboe62e398b2021-02-21 16:19:37 -07009537 if (!ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009538 return -ENOMEM;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009539 ctx->compat = in_compat_syscall();
Jens Axboe62e398b2021-02-21 16:19:37 -07009540 if (!capable(CAP_IPC_LOCK))
9541 ctx->user = get_uid(current_user());
Jens Axboe2aede0e2020-09-14 10:45:53 -06009542
9543 /*
9544 * This is just grabbed for accounting purposes. When a process exits,
9545 * the mm is exited and dropped before the files, hence we need to hang
9546 * on to this mm purely for the purposes of being able to unaccount
9547 * memory (locked/pinned vm). It's not used for anything else.
9548 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009549 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009550 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009551
Jens Axboe2b188cc2019-01-07 10:46:33 -07009552 ret = io_allocate_scq_urings(ctx, p);
9553 if (ret)
9554 goto err;
9555
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009556 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009557 if (ret)
9558 goto err;
9559
Jens Axboe2b188cc2019-01-07 10:46:33 -07009560 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009561 p->sq_off.head = offsetof(struct io_rings, sq.head);
9562 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9563 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9564 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9565 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9566 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9567 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009568
9569 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009570 p->cq_off.head = offsetof(struct io_rings, cq.head);
9571 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9572 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9573 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9574 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9575 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009576 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009577
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009578 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9579 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009580 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009581 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
Jens Axboe1c0aa1f2021-02-20 11:55:28 -07009582 IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009583
9584 if (copy_to_user(params, p, sizeof(*p))) {
9585 ret = -EFAULT;
9586 goto err;
9587 }
Jens Axboed1719f72020-07-30 13:43:53 -06009588
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009589 file = io_uring_get_file(ctx);
9590 if (IS_ERR(file)) {
9591 ret = PTR_ERR(file);
9592 goto err;
9593 }
9594
Jens Axboed1719f72020-07-30 13:43:53 -06009595 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009596 * Install ring fd as the very last thing, so we don't risk someone
9597 * having closed it before we finish setup
9598 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009599 ret = io_uring_install_fd(ctx, file);
9600 if (ret < 0) {
9601 /* fput will clean it up */
9602 fput(file);
9603 return ret;
9604 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009605
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009606 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009607 return ret;
9608err:
9609 io_ring_ctx_wait_and_kill(ctx);
9610 return ret;
9611}
9612
9613/*
9614 * Sets up an aio uring context, and returns the fd. Applications asks for a
9615 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9616 * params structure passed in.
9617 */
9618static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9619{
9620 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009621 int i;
9622
9623 if (copy_from_user(&p, params, sizeof(p)))
9624 return -EFAULT;
9625 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9626 if (p.resv[i])
9627 return -EINVAL;
9628 }
9629
Jens Axboe6c271ce2019-01-10 11:22:30 -07009630 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009631 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009632 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9633 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009634 return -EINVAL;
9635
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009636 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009637}
9638
9639SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9640 struct io_uring_params __user *, params)
9641{
9642 return io_uring_setup(entries, params);
9643}
9644
Jens Axboe66f4af92020-01-16 15:36:52 -07009645static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9646{
9647 struct io_uring_probe *p;
9648 size_t size;
9649 int i, ret;
9650
9651 size = struct_size(p, ops, nr_args);
9652 if (size == SIZE_MAX)
9653 return -EOVERFLOW;
9654 p = kzalloc(size, GFP_KERNEL);
9655 if (!p)
9656 return -ENOMEM;
9657
9658 ret = -EFAULT;
9659 if (copy_from_user(p, arg, size))
9660 goto out;
9661 ret = -EINVAL;
9662 if (memchr_inv(p, 0, size))
9663 goto out;
9664
9665 p->last_op = IORING_OP_LAST - 1;
9666 if (nr_args > IORING_OP_LAST)
9667 nr_args = IORING_OP_LAST;
9668
9669 for (i = 0; i < nr_args; i++) {
9670 p->ops[i].op = i;
9671 if (!io_op_defs[i].not_supported)
9672 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9673 }
9674 p->ops_len = i;
9675
9676 ret = 0;
9677 if (copy_to_user(arg, p, size))
9678 ret = -EFAULT;
9679out:
9680 kfree(p);
9681 return ret;
9682}
9683
Jens Axboe071698e2020-01-28 10:04:42 -07009684static int io_register_personality(struct io_ring_ctx *ctx)
9685{
Jens Axboe4379bf82021-02-15 13:40:22 -07009686 const struct cred *creds;
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009687 u32 id;
Jens Axboe1e6fa522020-10-15 08:46:24 -06009688 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009689
Jens Axboe4379bf82021-02-15 13:40:22 -07009690 creds = get_current_cred();
Jens Axboe1e6fa522020-10-15 08:46:24 -06009691
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009692 ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds,
9693 XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL);
9694 if (!ret)
9695 return id;
9696 put_cred(creds);
Jens Axboe1e6fa522020-10-15 08:46:24 -06009697 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009698}
9699
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009700static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9701 unsigned int nr_args)
9702{
9703 struct io_uring_restriction *res;
9704 size_t size;
9705 int i, ret;
9706
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009707 /* Restrictions allowed only if rings started disabled */
9708 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9709 return -EBADFD;
9710
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009711 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009712 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009713 return -EBUSY;
9714
9715 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9716 return -EINVAL;
9717
9718 size = array_size(nr_args, sizeof(*res));
9719 if (size == SIZE_MAX)
9720 return -EOVERFLOW;
9721
9722 res = memdup_user(arg, size);
9723 if (IS_ERR(res))
9724 return PTR_ERR(res);
9725
9726 ret = 0;
9727
9728 for (i = 0; i < nr_args; i++) {
9729 switch (res[i].opcode) {
9730 case IORING_RESTRICTION_REGISTER_OP:
9731 if (res[i].register_op >= IORING_REGISTER_LAST) {
9732 ret = -EINVAL;
9733 goto out;
9734 }
9735
9736 __set_bit(res[i].register_op,
9737 ctx->restrictions.register_op);
9738 break;
9739 case IORING_RESTRICTION_SQE_OP:
9740 if (res[i].sqe_op >= IORING_OP_LAST) {
9741 ret = -EINVAL;
9742 goto out;
9743 }
9744
9745 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9746 break;
9747 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9748 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9749 break;
9750 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9751 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9752 break;
9753 default:
9754 ret = -EINVAL;
9755 goto out;
9756 }
9757 }
9758
9759out:
9760 /* Reset all restrictions if an error happened */
9761 if (ret != 0)
9762 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9763 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009764 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009765
9766 kfree(res);
9767 return ret;
9768}
9769
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009770static int io_register_enable_rings(struct io_ring_ctx *ctx)
9771{
9772 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9773 return -EBADFD;
9774
9775 if (ctx->restrictions.registered)
9776 ctx->restricted = 1;
9777
Pavel Begunkov0298ef92021-03-08 13:20:57 +00009778 ctx->flags &= ~IORING_SETUP_R_DISABLED;
9779 if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait))
9780 wake_up(&ctx->sq_data->wait);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009781 return 0;
9782}
9783
Jens Axboe071698e2020-01-28 10:04:42 -07009784static bool io_register_op_must_quiesce(int op)
9785{
9786 switch (op) {
9787 case IORING_UNREGISTER_FILES:
9788 case IORING_REGISTER_FILES_UPDATE:
9789 case IORING_REGISTER_PROBE:
9790 case IORING_REGISTER_PERSONALITY:
9791 case IORING_UNREGISTER_PERSONALITY:
9792 return false;
9793 default:
9794 return true;
9795 }
9796}
9797
Jens Axboeedafcce2019-01-09 09:16:05 -07009798static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
9799 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -06009800 __releases(ctx->uring_lock)
9801 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -07009802{
9803 int ret;
9804
Jens Axboe35fa71a2019-04-22 10:23:23 -06009805 /*
9806 * We're inside the ring mutex, if the ref is already dying, then
9807 * someone else killed the ctx or is already going through
9808 * io_uring_register().
9809 */
9810 if (percpu_ref_is_dying(&ctx->refs))
9811 return -ENXIO;
9812
Jens Axboe071698e2020-01-28 10:04:42 -07009813 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009814 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -06009815
Jens Axboe05f3fb32019-12-09 11:22:50 -07009816 /*
9817 * Drop uring mutex before waiting for references to exit. If
9818 * another thread is currently inside io_uring_enter() it might
9819 * need to grab the uring_lock to make progress. If we hold it
9820 * here across the drain wait, then we can deadlock. It's safe
9821 * to drop the mutex here, since no new references will come in
9822 * after we've killed the percpu ref.
9823 */
9824 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009825 do {
9826 ret = wait_for_completion_interruptible(&ctx->ref_comp);
9827 if (!ret)
9828 break;
Jens Axboeed6930c2020-10-08 19:09:46 -06009829 ret = io_run_task_work_sig();
9830 if (ret < 0)
9831 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009832 } while (1);
9833
Jens Axboe05f3fb32019-12-09 11:22:50 -07009834 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009835
Jens Axboec1503682020-01-08 08:26:07 -07009836 if (ret) {
9837 percpu_ref_resurrect(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009838 goto out_quiesce;
9839 }
9840 }
9841
9842 if (ctx->restricted) {
9843 if (opcode >= IORING_REGISTER_LAST) {
9844 ret = -EINVAL;
9845 goto out;
9846 }
9847
9848 if (!test_bit(opcode, ctx->restrictions.register_op)) {
9849 ret = -EACCES;
Jens Axboec1503682020-01-08 08:26:07 -07009850 goto out;
9851 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07009852 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009853
9854 switch (opcode) {
9855 case IORING_REGISTER_BUFFERS:
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009856 ret = io_sqe_buffers_register(ctx, arg, nr_args);
Jens Axboeedafcce2019-01-09 09:16:05 -07009857 break;
9858 case IORING_UNREGISTER_BUFFERS:
9859 ret = -EINVAL;
9860 if (arg || nr_args)
9861 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009862 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -07009863 break;
Jens Axboe6b063142019-01-10 22:13:58 -07009864 case IORING_REGISTER_FILES:
9865 ret = io_sqe_files_register(ctx, arg, nr_args);
9866 break;
9867 case IORING_UNREGISTER_FILES:
9868 ret = -EINVAL;
9869 if (arg || nr_args)
9870 break;
9871 ret = io_sqe_files_unregister(ctx);
9872 break;
Jens Axboec3a31e62019-10-03 13:59:56 -06009873 case IORING_REGISTER_FILES_UPDATE:
9874 ret = io_sqe_files_update(ctx, arg, nr_args);
9875 break;
Jens Axboe9b402842019-04-11 11:45:41 -06009876 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -07009877 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -06009878 ret = -EINVAL;
9879 if (nr_args != 1)
9880 break;
9881 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -07009882 if (ret)
9883 break;
9884 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
9885 ctx->eventfd_async = 1;
9886 else
9887 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -06009888 break;
9889 case IORING_UNREGISTER_EVENTFD:
9890 ret = -EINVAL;
9891 if (arg || nr_args)
9892 break;
9893 ret = io_eventfd_unregister(ctx);
9894 break;
Jens Axboe66f4af92020-01-16 15:36:52 -07009895 case IORING_REGISTER_PROBE:
9896 ret = -EINVAL;
9897 if (!arg || nr_args > 256)
9898 break;
9899 ret = io_probe(ctx, arg, nr_args);
9900 break;
Jens Axboe071698e2020-01-28 10:04:42 -07009901 case IORING_REGISTER_PERSONALITY:
9902 ret = -EINVAL;
9903 if (arg || nr_args)
9904 break;
9905 ret = io_register_personality(ctx);
9906 break;
9907 case IORING_UNREGISTER_PERSONALITY:
9908 ret = -EINVAL;
9909 if (arg)
9910 break;
9911 ret = io_unregister_personality(ctx, nr_args);
9912 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009913 case IORING_REGISTER_ENABLE_RINGS:
9914 ret = -EINVAL;
9915 if (arg || nr_args)
9916 break;
9917 ret = io_register_enable_rings(ctx);
9918 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009919 case IORING_REGISTER_RESTRICTIONS:
9920 ret = io_register_restrictions(ctx, arg, nr_args);
9921 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07009922 default:
9923 ret = -EINVAL;
9924 break;
9925 }
9926
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009927out:
Jens Axboe071698e2020-01-28 10:04:42 -07009928 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009929 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -07009930 percpu_ref_reinit(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009931out_quiesce:
Jens Axboe0f158b42020-05-14 17:18:39 -06009932 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009933 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009934 return ret;
9935}
9936
9937SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
9938 void __user *, arg, unsigned int, nr_args)
9939{
9940 struct io_ring_ctx *ctx;
9941 long ret = -EBADF;
9942 struct fd f;
9943
9944 f = fdget(fd);
9945 if (!f.file)
9946 return -EBADF;
9947
9948 ret = -EOPNOTSUPP;
9949 if (f.file->f_op != &io_uring_fops)
9950 goto out_fput;
9951
9952 ctx = f.file->private_data;
9953
Pavel Begunkovb6c23dd2021-02-20 15:17:18 +00009954 io_run_task_work();
9955
Jens Axboeedafcce2019-01-09 09:16:05 -07009956 mutex_lock(&ctx->uring_lock);
9957 ret = __io_uring_register(ctx, opcode, arg, nr_args);
9958 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009959 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
9960 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07009961out_fput:
9962 fdput(f);
9963 return ret;
9964}
9965
Jens Axboe2b188cc2019-01-07 10:46:33 -07009966static int __init io_uring_init(void)
9967{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009968#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
9969 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
9970 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
9971} while (0)
9972
9973#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
9974 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
9975 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
9976 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
9977 BUILD_BUG_SQE_ELEM(1, __u8, flags);
9978 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
9979 BUILD_BUG_SQE_ELEM(4, __s32, fd);
9980 BUILD_BUG_SQE_ELEM(8, __u64, off);
9981 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
9982 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009983 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009984 BUILD_BUG_SQE_ELEM(24, __u32, len);
9985 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
9986 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
9987 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
9988 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +08009989 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
9990 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009991 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
9992 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
9993 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
9994 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
9995 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
9996 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
9997 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
9998 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009999 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010000 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
10001 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
10002 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +030010003 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010004
Jens Axboed3656342019-12-18 09:50:26 -070010005 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -070010006 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe91f245d2021-02-09 13:48:50 -070010007 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
10008 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -070010009 return 0;
10010};
10011__initcall(io_uring_init);