blob: d6c5465b372b4c3dd818d3daff4d134012489360 [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>
Jens Axboeff002b32020-02-07 16:05:21 -070077#include <linux/fs_struct.h>
Pavel Begunkov7d67af22020-02-24 11:32:45 +030078#include <linux/splice.h>
Jens Axboeb41e9852020-02-17 09:52:41 -070079#include <linux/task_work.h>
Jens Axboebcf5a062020-05-22 09:24:42 -060080#include <linux/pagemap.h>
Jens Axboe0f212202020-09-13 13:09:39 -060081#include <linux/io_uring.h>
Dennis Zhou91d8f512020-09-16 13:41:05 -070082#include <linux/blk-cgroup.h>
Jens Axboe4ea33a92020-10-15 13:46:44 -060083#include <linux/audit.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070084
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020085#define CREATE_TRACE_POINTS
86#include <trace/events/io_uring.h>
87
Jens Axboe2b188cc2019-01-07 10:46:33 -070088#include <uapi/linux/io_uring.h>
89
90#include "internal.h"
Jens Axboe561fb042019-10-24 07:25:42 -060091#include "io-wq.h"
Jens Axboe2b188cc2019-01-07 10:46:33 -070092
Daniel Xu5277dea2019-09-14 14:23:45 -070093#define IORING_MAX_ENTRIES 32768
Jens Axboe33a107f2019-10-04 12:10:03 -060094#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
Jens Axboe65e19f52019-10-26 07:20:21 -060095
96/*
97 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
98 */
99#define IORING_FILE_TABLE_SHIFT 9
100#define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT)
101#define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1)
102#define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE)
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200103#define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \
104 IORING_REGISTER_LAST + IORING_OP_LAST)
Jens Axboe2b188cc2019-01-07 10:46:33 -0700105
Pavel Begunkovb16fed662021-02-18 18:29:40 +0000106#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
107 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
108 IOSQE_BUFFER_SELECT)
109
Jens Axboe2b188cc2019-01-07 10:46:33 -0700110struct io_uring {
111 u32 head ____cacheline_aligned_in_smp;
112 u32 tail ____cacheline_aligned_in_smp;
113};
114
Stefan Bühler1e84b972019-04-24 23:54:16 +0200115/*
Hristo Venev75b28af2019-08-26 17:23:46 +0000116 * This data is shared with the application through the mmap at offsets
117 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200118 *
119 * The offsets to the member fields are published through struct
120 * io_sqring_offsets when calling io_uring_setup.
121 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000122struct io_rings {
Stefan Bühler1e84b972019-04-24 23:54:16 +0200123 /*
124 * Head and tail offsets into the ring; the offsets need to be
125 * masked to get valid indices.
126 *
Hristo Venev75b28af2019-08-26 17:23:46 +0000127 * The kernel controls head of the sq ring and the tail of the cq ring,
128 * and the application controls tail of the sq ring and the head of the
129 * cq ring.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200130 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000131 struct io_uring sq, cq;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200132 /*
Hristo Venev75b28af2019-08-26 17:23:46 +0000133 * Bitmasks to apply to head and tail offsets (constant, equals
Stefan Bühler1e84b972019-04-24 23:54:16 +0200134 * ring_entries - 1)
135 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000136 u32 sq_ring_mask, cq_ring_mask;
137 /* Ring sizes (constant, power of 2) */
138 u32 sq_ring_entries, cq_ring_entries;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200139 /*
140 * Number of invalid entries dropped by the kernel due to
141 * invalid index stored in array
142 *
143 * Written by the kernel, shouldn't be modified by the
144 * application (i.e. get number of "new events" by comparing to
145 * cached value).
146 *
147 * After a new SQ head value was read by the application this
148 * counter includes all submissions that were dropped reaching
149 * the new SQ head (and possibly more).
150 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000151 u32 sq_dropped;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200152 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200153 * Runtime SQ flags
Stefan Bühler1e84b972019-04-24 23:54:16 +0200154 *
155 * Written by the kernel, shouldn't be modified by the
156 * application.
157 *
158 * The application needs a full memory barrier before checking
159 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
160 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000161 u32 sq_flags;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200162 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200163 * Runtime CQ flags
164 *
165 * Written by the application, shouldn't be modified by the
166 * kernel.
167 */
168 u32 cq_flags;
169 /*
Stefan Bühler1e84b972019-04-24 23:54:16 +0200170 * Number of completion events lost because the queue was full;
171 * this should be avoided by the application by making sure
LimingWu0b4295b2019-12-05 20:18:18 +0800172 * there are not more requests pending than there is space in
Stefan Bühler1e84b972019-04-24 23:54:16 +0200173 * the completion queue.
174 *
175 * Written by the kernel, shouldn't be modified by the
176 * application (i.e. get number of "new events" by comparing to
177 * cached value).
178 *
179 * As completion events come in out of order this counter is not
180 * ordered with any other data.
181 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000182 u32 cq_overflow;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200183 /*
184 * Ring buffer of completion events.
185 *
186 * The kernel writes completion events fresh every time they are
187 * produced, so the application is allowed to modify pending
188 * entries.
189 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000190 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700191};
192
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000193enum io_uring_cmd_flags {
194 IO_URING_F_NONBLOCK = 1,
Pavel Begunkov889fca72021-02-10 00:03:09 +0000195 IO_URING_F_COMPLETE_DEFER = 2,
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000196};
197
Jens Axboeedafcce2019-01-09 09:16:05 -0700198struct io_mapped_ubuf {
199 u64 ubuf;
200 size_t len;
201 struct bio_vec *bvec;
202 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600203 unsigned long acct_pages;
Jens Axboeedafcce2019-01-09 09:16:05 -0700204};
205
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000206struct io_ring_ctx;
207
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000208struct io_rsrc_put {
209 struct list_head list;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000210 union {
211 void *rsrc;
212 struct file *file;
213 };
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000214};
215
216struct fixed_rsrc_table {
Jens Axboe65e19f52019-10-26 07:20:21 -0600217 struct file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700218};
219
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000220struct fixed_rsrc_ref_node {
Xiaoguang Wang05589552020-03-31 14:05:18 +0800221 struct percpu_ref refs;
222 struct list_head node;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000223 struct list_head rsrc_list;
224 struct fixed_rsrc_data *rsrc_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000225 void (*rsrc_put)(struct io_ring_ctx *ctx,
226 struct io_rsrc_put *prsrc);
Jens Axboe4a38aed22020-05-14 17:21:15 -0600227 struct llist_node llist;
Pavel Begunkove2978222020-11-18 14:56:26 +0000228 bool done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800229};
230
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000231struct fixed_rsrc_data {
232 struct fixed_rsrc_table *table;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700233 struct io_ring_ctx *ctx;
234
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000235 struct fixed_rsrc_ref_node *node;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700236 struct percpu_ref refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700237 struct completion done;
Hao Xu8bad28d2021-02-19 17:19:36 +0800238 bool quiesce;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700239};
240
Jens Axboe5a2e7452020-02-23 16:23:11 -0700241struct io_buffer {
242 struct list_head list;
243 __u64 addr;
244 __s32 len;
245 __u16 bid;
246};
247
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200248struct io_restriction {
249 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
250 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
251 u8 sqe_flags_allowed;
252 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200253 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200254};
255
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700256enum {
257 IO_SQ_THREAD_SHOULD_STOP = 0,
258 IO_SQ_THREAD_SHOULD_PARK,
259};
260
Jens Axboe534ca6d2020-09-02 13:52:19 -0600261struct io_sq_data {
262 refcount_t refs;
Jens Axboe69fb2132020-09-14 11:16:23 -0600263 struct mutex lock;
264
265 /* ctx's that are using this sqd */
266 struct list_head ctx_list;
267 struct list_head ctx_new_list;
268 struct mutex ctx_lock;
269
Jens Axboe534ca6d2020-09-02 13:52:19 -0600270 struct task_struct *thread;
271 struct wait_queue_head wait;
Xiaoguang Wang08369242020-11-03 14:15:59 +0800272
273 unsigned sq_thread_idle;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700274 int sq_cpu;
275 pid_t task_pid;
276
277 unsigned long state;
278 struct completion startup;
279 struct completion completion;
280 struct completion exited;
Jens Axboe534ca6d2020-09-02 13:52:19 -0600281};
282
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000283#define IO_IOPOLL_BATCH 8
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000284#define IO_COMPL_BATCH 32
Pavel Begunkov6ff119a2021-02-10 00:03:18 +0000285#define IO_REQ_CACHE_SIZE 32
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000286#define IO_REQ_ALLOC_BATCH 8
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000287
288struct io_comp_state {
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000289 struct io_kiocb *reqs[IO_COMPL_BATCH];
Jens Axboe1b4c3512021-02-10 00:03:19 +0000290 unsigned int nr;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700291 unsigned int locked_free_nr;
292 /* inline/task_work completion list, under ->uring_lock */
Jens Axboe1b4c3512021-02-10 00:03:19 +0000293 struct list_head free_list;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700294 /* IRQ completion list, under ->completion_lock */
295 struct list_head locked_free_list;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000296};
297
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000298struct io_submit_link {
299 struct io_kiocb *head;
300 struct io_kiocb *last;
301};
302
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000303struct io_submit_state {
304 struct blk_plug plug;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000305 struct io_submit_link link;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000306
307 /*
308 * io_kiocb alloc cache
309 */
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000310 void *reqs[IO_REQ_CACHE_SIZE];
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000311 unsigned int free_reqs;
312
313 bool plug_started;
314
315 /*
316 * Batch completion logic
317 */
318 struct io_comp_state comp;
319
320 /*
321 * File reference cache
322 */
323 struct file *file;
324 unsigned int fd;
325 unsigned int file_refs;
326 unsigned int ios_left;
327};
328
Jens Axboe2b188cc2019-01-07 10:46:33 -0700329struct io_ring_ctx {
330 struct {
331 struct percpu_ref refs;
332 } ____cacheline_aligned_in_smp;
333
334 struct {
335 unsigned int flags;
Randy Dunlape1d85332020-02-05 20:57:10 -0800336 unsigned int compat: 1;
Randy Dunlape1d85332020-02-05 20:57:10 -0800337 unsigned int cq_overflow_flushed: 1;
338 unsigned int drain_next: 1;
339 unsigned int eventfd_async: 1;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200340 unsigned int restricted: 1;
Jens Axboe5f3f26f2021-02-25 10:17:46 -0700341 unsigned int sqo_exec: 1;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700342
Hristo Venev75b28af2019-08-26 17:23:46 +0000343 /*
344 * Ring buffer of indices into array of io_uring_sqe, which is
345 * mmapped by the application using the IORING_OFF_SQES offset.
346 *
347 * This indirection could e.g. be used to assign fixed
348 * io_uring_sqe entries to operations and only submit them to
349 * the queue when needed.
350 *
351 * The kernel modifies neither the indices array nor the entries
352 * array.
353 */
354 u32 *sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700355 unsigned cached_sq_head;
356 unsigned sq_entries;
357 unsigned sq_mask;
Jens Axboe6c271ce2019-01-10 11:22:30 -0700358 unsigned sq_thread_idle;
Jens Axboe498ccd92019-10-25 10:04:25 -0600359 unsigned cached_sq_dropped;
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +0100360 unsigned cached_cq_overflow;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700361 unsigned long sq_check_overflow;
Jens Axboede0617e2019-04-06 21:51:27 -0600362
Jens Axboee9418942021-02-19 12:33:30 -0700363 /* hashed buffered write serialization */
364 struct io_wq_hash *hash_map;
365
Jens Axboede0617e2019-04-06 21:51:27 -0600366 struct list_head defer_list;
Jens Axboe5262f562019-09-17 12:26:57 -0600367 struct list_head timeout_list;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -0700368 struct list_head cq_overflow_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700369
Jens Axboead3eb2c2019-12-18 17:12:20 -0700370 struct io_uring_sqe *sq_sqes;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700371 } ____cacheline_aligned_in_smp;
372
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700373 struct {
374 struct mutex uring_lock;
375 wait_queue_head_t wait;
376 } ____cacheline_aligned_in_smp;
377
378 struct io_submit_state submit_state;
379
Hristo Venev75b28af2019-08-26 17:23:46 +0000380 struct io_rings *rings;
381
Jens Axboe2aede0e2020-09-14 10:45:53 -0600382 /* Only used for accounting purposes */
383 struct mm_struct *mm_account;
384
Jens Axboe534ca6d2020-09-02 13:52:19 -0600385 struct io_sq_data *sq_data; /* if using sq thread polling */
386
Jens Axboe90554202020-09-03 12:12:41 -0600387 struct wait_queue_head sqo_sq_wait;
Jens Axboe69fb2132020-09-14 11:16:23 -0600388 struct list_head sqd_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700389
Jens Axboe6b063142019-01-10 22:13:58 -0700390 /*
391 * If used, fixed file set. Writers must ensure that ->refs is dead,
392 * readers must ensure that ->refs is alive as long as the file* is
393 * used. Only updated through io_uring_register(2).
394 */
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000395 struct fixed_rsrc_data *file_data;
Jens Axboe6b063142019-01-10 22:13:58 -0700396 unsigned nr_user_files;
397
Jens Axboeedafcce2019-01-09 09:16:05 -0700398 /* if used, fixed mapped user buffers */
399 unsigned nr_user_bufs;
400 struct io_mapped_ubuf *user_bufs;
401
Jens Axboe2b188cc2019-01-07 10:46:33 -0700402 struct user_struct *user;
403
Jens Axboe0f158b42020-05-14 17:18:39 -0600404 struct completion ref_comp;
405 struct completion sq_thread_comp;
Jens Axboe206aefd2019-11-07 18:27:42 -0700406
407#if defined(CONFIG_UNIX)
408 struct socket *ring_sock;
409#endif
410
Jens Axboe5a2e7452020-02-23 16:23:11 -0700411 struct idr io_buffer_idr;
412
Jens Axboe071698e2020-01-28 10:04:42 -0700413 struct idr personality_idr;
414
Jens Axboe206aefd2019-11-07 18:27:42 -0700415 struct {
416 unsigned cached_cq_tail;
417 unsigned cq_entries;
418 unsigned cq_mask;
419 atomic_t cq_timeouts;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -0500420 unsigned cq_last_tm_flush;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700421 unsigned long cq_check_overflow;
Jens Axboe206aefd2019-11-07 18:27:42 -0700422 struct wait_queue_head cq_wait;
423 struct fasync_struct *cq_fasync;
424 struct eventfd_ctx *cq_ev_fd;
425 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700426
427 struct {
Jens Axboe2b188cc2019-01-07 10:46:33 -0700428 spinlock_t completion_lock;
Jens Axboee94f1412019-12-19 12:06:02 -0700429
Jens Axboedef596e2019-01-09 08:59:42 -0700430 /*
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300431 * ->iopoll_list is protected by the ctx->uring_lock for
Jens Axboedef596e2019-01-09 08:59:42 -0700432 * io_uring instances that don't use IORING_SETUP_SQPOLL.
433 * For SQPOLL, only the single threaded io_sq_thread() will
434 * manipulate the list, hence no extra locking is needed there.
435 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300436 struct list_head iopoll_list;
Jens Axboe78076bb2019-12-04 19:56:40 -0700437 struct hlist_head *cancel_hash;
438 unsigned cancel_hash_bits;
Jens Axboee94f1412019-12-19 12:06:02 -0700439 bool poll_multi_file;
Jens Axboefcb323c2019-10-24 12:39:47 -0600440
441 spinlock_t inflight_lock;
442 struct list_head inflight_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700443 } ____cacheline_aligned_in_smp;
Jens Axboe85faa7b2020-04-09 18:14:00 -0600444
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000445 struct delayed_work rsrc_put_work;
446 struct llist_head rsrc_put_llist;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +0000447 struct list_head rsrc_ref_list;
448 spinlock_t rsrc_ref_lock;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600449
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200450 struct io_restriction restrictions;
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700451
Jens Axboe7c25c0d2021-02-16 07:17:00 -0700452 /* exit task_work */
453 struct callback_head *exit_task_work;
454
Jens Axboee9418942021-02-19 12:33:30 -0700455 struct wait_queue_head hash_wait;
456
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700457 /* Keep this last, we don't need it for the fast path */
458 struct work_struct exit_work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700459};
460
Jens Axboe09bb8392019-03-13 12:39:28 -0600461/*
462 * First field must be the file pointer in all the
463 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
464 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700465struct io_poll_iocb {
466 struct file *file;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000467 struct wait_queue_head *head;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700468 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600469 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700470 bool canceled;
Jens Axboe392edb42019-12-09 17:52:20 -0700471 struct wait_queue_entry wait;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700472};
473
Pavel Begunkov018043b2020-10-27 23:17:18 +0000474struct io_poll_remove {
475 struct file *file;
476 u64 addr;
477};
478
Jens Axboeb5dba592019-12-11 14:02:38 -0700479struct io_close {
480 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700481 int fd;
482};
483
Jens Axboead8a48a2019-11-15 08:49:11 -0700484struct io_timeout_data {
485 struct io_kiocb *req;
486 struct hrtimer timer;
487 struct timespec64 ts;
488 enum hrtimer_mode mode;
489};
490
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700491struct io_accept {
492 struct file *file;
493 struct sockaddr __user *addr;
494 int __user *addr_len;
495 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600496 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700497};
498
499struct io_sync {
500 struct file *file;
501 loff_t len;
502 loff_t off;
503 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700504 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700505};
506
Jens Axboefbf23842019-12-17 18:45:56 -0700507struct io_cancel {
508 struct file *file;
509 u64 addr;
510};
511
Jens Axboeb29472e2019-12-17 18:50:29 -0700512struct io_timeout {
513 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300514 u32 off;
515 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300516 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000517 /* head of the link, used by linked timeouts only */
518 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700519};
520
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100521struct io_timeout_rem {
522 struct file *file;
523 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000524
525 /* timeout update */
526 struct timespec64 ts;
527 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100528};
529
Jens Axboe9adbd452019-12-20 08:45:55 -0700530struct io_rw {
531 /* NOTE: kiocb has the file as the first member, so don't do it here */
532 struct kiocb kiocb;
533 u64 addr;
534 u64 len;
535};
536
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700537struct io_connect {
538 struct file *file;
539 struct sockaddr __user *addr;
540 int addr_len;
541};
542
Jens Axboee47293f2019-12-20 08:58:21 -0700543struct io_sr_msg {
544 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700545 union {
Pavel Begunkov270a5942020-07-12 20:41:04 +0300546 struct user_msghdr __user *umsg;
Jens Axboefddafac2020-01-04 20:19:44 -0700547 void __user *buf;
548 };
Jens Axboee47293f2019-12-20 08:58:21 -0700549 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700550 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700551 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700552 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700553};
554
Jens Axboe15b71ab2019-12-11 11:20:36 -0700555struct io_open {
556 struct file *file;
557 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700558 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700559 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600560 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700561};
562
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000563struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700564 struct file *file;
565 u64 arg;
566 u32 nr_args;
567 u32 offset;
568};
569
Jens Axboe4840e412019-12-25 22:03:45 -0700570struct io_fadvise {
571 struct file *file;
572 u64 offset;
573 u32 len;
574 u32 advice;
575};
576
Jens Axboec1ca7572019-12-25 22:18:28 -0700577struct io_madvise {
578 struct file *file;
579 u64 addr;
580 u32 len;
581 u32 advice;
582};
583
Jens Axboe3e4827b2020-01-08 15:18:09 -0700584struct io_epoll {
585 struct file *file;
586 int epfd;
587 int op;
588 int fd;
589 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700590};
591
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300592struct io_splice {
593 struct file *file_out;
594 struct file *file_in;
595 loff_t off_out;
596 loff_t off_in;
597 u64 len;
598 unsigned int flags;
599};
600
Jens Axboeddf0322d2020-02-23 16:41:33 -0700601struct io_provide_buf {
602 struct file *file;
603 __u64 addr;
604 __s32 len;
605 __u32 bgid;
606 __u16 nbufs;
607 __u16 bid;
608};
609
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700610struct io_statx {
611 struct file *file;
612 int dfd;
613 unsigned int mask;
614 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700615 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700616 struct statx __user *buffer;
617};
618
Jens Axboe36f4fa62020-09-05 11:14:22 -0600619struct io_shutdown {
620 struct file *file;
621 int how;
622};
623
Jens Axboe80a261f2020-09-28 14:23:58 -0600624struct io_rename {
625 struct file *file;
626 int old_dfd;
627 int new_dfd;
628 struct filename *oldpath;
629 struct filename *newpath;
630 int flags;
631};
632
Jens Axboe14a11432020-09-28 14:27:37 -0600633struct io_unlink {
634 struct file *file;
635 int dfd;
636 int flags;
637 struct filename *filename;
638};
639
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300640struct io_completion {
641 struct file *file;
642 struct list_head list;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +0300643 int cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300644};
645
Jens Axboef499a022019-12-02 16:28:46 -0700646struct io_async_connect {
647 struct sockaddr_storage address;
648};
649
Jens Axboe03b12302019-12-02 18:50:25 -0700650struct io_async_msghdr {
651 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000652 /* points to an allocated iov, if NULL we use fast_iov instead */
653 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700654 struct sockaddr __user *uaddr;
655 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700656 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700657};
658
Jens Axboef67676d2019-12-02 11:03:47 -0700659struct io_async_rw {
660 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600661 const struct iovec *free_iovec;
662 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600663 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600664 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700665};
666
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300667enum {
668 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
669 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
670 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
671 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
672 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700673 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300674
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300675 REQ_F_FAIL_LINK_BIT,
676 REQ_F_INFLIGHT_BIT,
677 REQ_F_CUR_POS_BIT,
678 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300679 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300680 REQ_F_ISREG_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300681 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700682 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700683 REQ_F_BUFFER_SELECTED_BIT,
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600684 REQ_F_NO_FILE_TABLE_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100685 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000686 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700687
688 /* not a real bit, just to check we're not overflowing the space */
689 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300690};
691
692enum {
693 /* ctx owns file */
694 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
695 /* drain existing IO first */
696 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
697 /* linked sqes */
698 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
699 /* doesn't sever on completion < 0 */
700 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
701 /* IOSQE_ASYNC */
702 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700703 /* IOSQE_BUFFER_SELECT */
704 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300705
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300706 /* fail rest of links */
707 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
708 /* on inflight list */
709 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
710 /* read/write uses file position */
711 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
712 /* must not punt to workers */
713 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100714 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300715 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300716 /* regular file */
717 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300718 /* needs cleanup */
719 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700720 /* already went through poll handler */
721 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700722 /* buffer already selected */
723 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600724 /* doesn't need file table for this request */
725 REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100726 /* linked timeout is active, i.e. prepared by link's head */
727 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000728 /* completion is deferred through io_comp_state */
729 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700730};
731
732struct async_poll {
733 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600734 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300735};
736
Jens Axboe7cbf1722021-02-10 00:03:20 +0000737struct io_task_work {
738 struct io_wq_work_node node;
739 task_work_func_t func;
740};
741
Jens Axboe09bb8392019-03-13 12:39:28 -0600742/*
743 * NOTE! Each of the iocb union members has the file pointer
744 * as the first entry in their struct definition. So you can
745 * access the file pointer through any of the sub-structs,
746 * or directly as just 'ki_filp' in this struct.
747 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700748struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700749 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600750 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700751 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700752 struct io_poll_iocb poll;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000753 struct io_poll_remove poll_remove;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700754 struct io_accept accept;
755 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700756 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700757 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100758 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700759 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700760 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700761 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700762 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000763 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700764 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700765 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700766 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300767 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700768 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700769 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600770 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600771 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600772 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300773 /* use only after cleaning per-op data, see io_clean_op() */
774 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700775 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700776
Jens Axboee8c2bc12020-08-15 18:44:09 -0700777 /* opcode allocated if it needs to store data for async defer */
778 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700779 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800780 /* polled IO has completed */
781 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700782
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700783 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300784 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700785
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300786 struct io_ring_ctx *ctx;
787 unsigned int flags;
788 refcount_t refs;
789 struct task_struct *task;
790 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700791
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000792 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000793 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700794
Pavel Begunkovd21ffe72020-07-13 23:37:10 +0300795 /*
796 * 1. used with ctx->iopoll_list with reads/writes
797 * 2. to track reqs with ->files (see io_op_def::file_table)
798 */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300799 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000800 union {
801 struct io_task_work io_task_work;
802 struct callback_head task_work;
803 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300804 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
805 struct hlist_node hash_node;
806 struct async_poll *apoll;
807 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700808};
809
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300810struct io_defer_entry {
811 struct list_head list;
812 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300813 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300814};
815
Jens Axboed3656342019-12-18 09:50:26 -0700816struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700817 /* needs req->file assigned */
818 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700819 /* hash wq insertion if file is a regular file */
820 unsigned hash_reg_file : 1;
821 /* unbound wq insertion if file is a non-regular file */
822 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700823 /* opcode is not supported by this kernel */
824 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700825 /* set if opcode supports polled "wait" */
826 unsigned pollin : 1;
827 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700828 /* op supports buffer selection */
829 unsigned buffer_select : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700830 /* must always have async data allocated */
831 unsigned needs_async_data : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600832 /* should block plug */
833 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700834 /* size of async data needed, if any */
835 unsigned short async_size;
Jens Axboed3656342019-12-18 09:50:26 -0700836};
837
Jens Axboe09186822020-10-13 15:01:40 -0600838static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300839 [IORING_OP_NOP] = {},
840 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700841 .needs_file = 1,
842 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700843 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700844 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700845 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600846 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700847 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700848 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300849 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700850 .needs_file = 1,
851 .hash_reg_file = 1,
852 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700853 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700854 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600855 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700856 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700857 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300858 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700859 .needs_file = 1,
860 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300861 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700862 .needs_file = 1,
863 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700864 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600865 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700866 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700867 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300868 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700869 .needs_file = 1,
870 .hash_reg_file = 1,
871 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700872 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600873 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700874 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700875 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300876 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700877 .needs_file = 1,
878 .unbound_nonreg_file = 1,
879 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300880 [IORING_OP_POLL_REMOVE] = {},
881 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700882 .needs_file = 1,
883 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300884 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700885 .needs_file = 1,
886 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700887 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700888 .needs_async_data = 1,
889 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700890 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300891 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700892 .needs_file = 1,
893 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700894 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700895 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700896 .needs_async_data = 1,
897 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700898 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300899 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700900 .needs_async_data = 1,
901 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700902 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000903 [IORING_OP_TIMEOUT_REMOVE] = {
904 /* used by timeout updates' prep() */
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000905 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300906 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700907 .needs_file = 1,
908 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700909 .pollin = 1,
Jens Axboed3656342019-12-18 09:50:26 -0700910 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300911 [IORING_OP_ASYNC_CANCEL] = {},
912 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700913 .needs_async_data = 1,
914 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700915 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300916 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700917 .needs_file = 1,
918 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700919 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700920 .needs_async_data = 1,
921 .async_size = sizeof(struct io_async_connect),
Jens Axboed3656342019-12-18 09:50:26 -0700922 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300923 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700924 .needs_file = 1,
925 },
Jens Axboe44526be2021-02-15 13:32:18 -0700926 [IORING_OP_OPENAT] = {},
927 [IORING_OP_CLOSE] = {},
928 [IORING_OP_FILES_UPDATE] = {},
929 [IORING_OP_STATX] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300930 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700931 .needs_file = 1,
932 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700933 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700934 .buffer_select = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600935 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700936 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700937 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300938 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700939 .needs_file = 1,
940 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700941 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600942 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700943 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700944 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300945 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700946 .needs_file = 1,
947 },
Jens Axboe44526be2021-02-15 13:32:18 -0700948 [IORING_OP_MADVISE] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300949 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700950 .needs_file = 1,
951 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700952 .pollout = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700953 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300954 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700955 .needs_file = 1,
956 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700957 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700958 .buffer_select = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700959 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300960 [IORING_OP_OPENAT2] = {
Jens Axboecebdb982020-01-08 17:59:24 -0700961 },
Jens Axboe3e4827b2020-01-08 15:18:09 -0700962 [IORING_OP_EPOLL_CTL] = {
963 .unbound_nonreg_file = 1,
Jens Axboe3e4827b2020-01-08 15:18:09 -0700964 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300965 [IORING_OP_SPLICE] = {
966 .needs_file = 1,
967 .hash_reg_file = 1,
968 .unbound_nonreg_file = 1,
Jens Axboeddf0322d2020-02-23 16:41:33 -0700969 },
970 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -0700971 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +0300972 [IORING_OP_TEE] = {
973 .needs_file = 1,
974 .hash_reg_file = 1,
975 .unbound_nonreg_file = 1,
976 },
Jens Axboe36f4fa62020-09-05 11:14:22 -0600977 [IORING_OP_SHUTDOWN] = {
978 .needs_file = 1,
979 },
Jens Axboe44526be2021-02-15 13:32:18 -0700980 [IORING_OP_RENAMEAT] = {},
981 [IORING_OP_UNLINKAT] = {},
Jens Axboed3656342019-12-18 09:50:26 -0700982};
983
Pavel Begunkov9936c7c2021-02-04 13:51:56 +0000984static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
985 struct task_struct *task,
986 struct files_struct *files);
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700987static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000988static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node);
Pavel Begunkovbc9744c2021-01-15 17:37:49 +0000989static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Pavel Begunkov1ffc5422020-12-30 21:34:15 +0000990 struct io_ring_ctx *ctx);
Pavel Begunkovf2303b12021-02-20 18:03:49 +0000991static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +0000992
Pavel Begunkov23faba32021-02-11 18:28:22 +0000993static bool io_rw_reissue(struct io_kiocb *req);
Jens Axboe78e19bb2019-11-06 15:21:34 -0700994static void io_cqring_fill_event(struct io_kiocb *req, long res);
Jackie Liuec9c02a2019-11-08 23:50:36 +0800995static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +0100996static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec40f6372020-06-25 15:39:59 -0600997static void io_double_put_req(struct io_kiocb *req);
Jens Axboec7dae4b2021-02-09 19:53:37 -0700998static void io_dismantle_req(struct io_kiocb *req);
999static void io_put_task(struct task_struct *task, int nr);
1000static void io_queue_next(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001001static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001002static void __io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001003static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -07001004static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001005 struct io_uring_rsrc_update *ip,
Jens Axboe05f3fb32019-12-09 11:22:50 -07001006 unsigned nr_args);
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001007static void __io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001008static struct file *io_file_get(struct io_submit_state *state,
1009 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001010static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001011static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001012
Pavel Begunkov847595d2021-02-04 13:52:06 +00001013static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
1014 struct iov_iter *iter, bool needs_lock);
Jens Axboeff6165b2020-08-13 09:47:43 -06001015static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
1016 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06001017 struct iov_iter *iter, bool force);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001018static void io_req_task_queue(struct io_kiocb *req);
Jens Axboe65453d12021-02-10 00:03:21 +00001019static void io_submit_flush_completions(struct io_comp_state *cs,
1020 struct io_ring_ctx *ctx);
Jens Axboe9a56a232019-01-09 09:06:50 -07001021
Jens Axboe2b188cc2019-01-07 10:46:33 -07001022static struct kmem_cache *req_cachep;
1023
Jens Axboe09186822020-10-13 15:01:40 -06001024static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001025
1026struct sock *io_uring_get_socket(struct file *file)
1027{
1028#if defined(CONFIG_UNIX)
1029 if (file->f_op == &io_uring_fops) {
1030 struct io_ring_ctx *ctx = file->private_data;
1031
1032 return ctx->ring_sock->sk;
1033 }
1034#endif
1035 return NULL;
1036}
1037EXPORT_SYMBOL(io_uring_get_socket);
1038
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001039#define io_for_each_link(pos, head) \
1040 for (pos = (head); pos; pos = pos->link)
1041
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001042static inline void io_clean_op(struct io_kiocb *req)
1043{
Pavel Begunkov9d5c8192021-01-24 15:08:14 +00001044 if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED))
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001045 __io_clean_op(req);
1046}
1047
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001048static inline void io_set_resource_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001049{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001050 struct io_ring_ctx *ctx = req->ctx;
1051
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001052 if (!req->fixed_rsrc_refs) {
1053 req->fixed_rsrc_refs = &ctx->file_data->node->refs;
1054 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001055 }
1056}
1057
Pavel Begunkov08d23632020-11-06 13:00:22 +00001058static bool io_match_task(struct io_kiocb *head,
1059 struct task_struct *task,
1060 struct files_struct *files)
1061{
1062 struct io_kiocb *req;
1063
Jens Axboe84965ff2021-01-23 15:51:11 -07001064 if (task && head->task != task) {
1065 /* in terms of cancelation, always match if req task is dead */
1066 if (head->task->flags & PF_EXITING)
1067 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001068 return false;
Jens Axboe84965ff2021-01-23 15:51:11 -07001069 }
Pavel Begunkov08d23632020-11-06 13:00:22 +00001070 if (!files)
1071 return true;
1072
1073 io_for_each_link(req, head) {
Jens Axboe02a13672021-01-23 15:49:31 -07001074 if (req->file && req->file->f_op == &io_uring_fops)
1075 return true;
Jens Axboe4379bf82021-02-15 13:40:22 -07001076 if (req->task->files == files)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001077 return true;
1078 }
1079 return false;
1080}
1081
Jens Axboec40f6372020-06-25 15:39:59 -06001082static inline void req_set_fail_links(struct io_kiocb *req)
1083{
1084 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1085 req->flags |= REQ_F_FAIL_LINK;
1086}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001087
Jens Axboe2b188cc2019-01-07 10:46:33 -07001088static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1089{
1090 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1091
Jens Axboe0f158b42020-05-14 17:18:39 -06001092 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001093}
1094
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001095static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1096{
1097 return !req->timeout.off;
1098}
1099
Jens Axboe2b188cc2019-01-07 10:46:33 -07001100static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1101{
1102 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001103 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001104
1105 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1106 if (!ctx)
1107 return NULL;
1108
Jens Axboe78076bb2019-12-04 19:56:40 -07001109 /*
1110 * Use 5 bits less than the max cq entries, that should give us around
1111 * 32 entries per hash list if totally full and uniformly spread.
1112 */
1113 hash_bits = ilog2(p->cq_entries);
1114 hash_bits -= 5;
1115 if (hash_bits <= 0)
1116 hash_bits = 1;
1117 ctx->cancel_hash_bits = hash_bits;
1118 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1119 GFP_KERNEL);
1120 if (!ctx->cancel_hash)
1121 goto err;
1122 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1123
Roman Gushchin21482892019-05-07 10:01:48 -07001124 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001125 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1126 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001127
1128 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001129 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001130 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001131 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001132 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001133 init_completion(&ctx->ref_comp);
1134 init_completion(&ctx->sq_thread_comp);
Jens Axboe5a2e7452020-02-23 16:23:11 -07001135 idr_init(&ctx->io_buffer_idr);
Jens Axboe071698e2020-01-28 10:04:42 -07001136 idr_init(&ctx->personality_idr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001137 mutex_init(&ctx->uring_lock);
1138 init_waitqueue_head(&ctx->wait);
1139 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001140 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001141 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001142 INIT_LIST_HEAD(&ctx->timeout_list);
Jens Axboefcb323c2019-10-24 12:39:47 -06001143 spin_lock_init(&ctx->inflight_lock);
1144 INIT_LIST_HEAD(&ctx->inflight_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001145 spin_lock_init(&ctx->rsrc_ref_lock);
1146 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001147 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1148 init_llist_head(&ctx->rsrc_put_llist);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001149 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001150 INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001151 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001152err:
Jens Axboe78076bb2019-12-04 19:56:40 -07001153 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001154 kfree(ctx);
1155 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001156}
1157
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001158static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001159{
Jens Axboe2bc99302020-07-09 09:43:27 -06001160 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1161 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001162
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001163 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001164 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001165 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001166
Bob Liu9d858b22019-11-13 18:06:25 +08001167 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001168}
1169
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001170static void io_req_clean_work(struct io_kiocb *req)
Jens Axboecccf0ee2020-01-27 16:34:48 -07001171{
Pavel Begunkov34e08fe2021-02-01 18:59:53 +00001172 if (req->flags & REQ_F_INFLIGHT) {
1173 struct io_ring_ctx *ctx = req->ctx;
1174 struct io_uring_task *tctx = req->task->io_uring;
1175 unsigned long flags;
Jens Axboe51a4cc12020-08-10 10:55:56 -06001176
Pavel Begunkov34e08fe2021-02-01 18:59:53 +00001177 spin_lock_irqsave(&ctx->inflight_lock, flags);
1178 list_del(&req->inflight_entry);
1179 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1180 req->flags &= ~REQ_F_INFLIGHT;
1181 if (atomic_read(&tctx->in_idle))
1182 wake_up(&tctx->wait);
1183 }
Jens Axboe1e6fa522020-10-15 08:46:24 -06001184}
1185
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001186static void io_req_track_inflight(struct io_kiocb *req)
1187{
1188 struct io_ring_ctx *ctx = req->ctx;
1189
1190 if (!(req->flags & REQ_F_INFLIGHT)) {
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001191 req->flags |= REQ_F_INFLIGHT;
1192
1193 spin_lock_irq(&ctx->inflight_lock);
1194 list_add(&req->inflight_entry, &ctx->inflight_list);
1195 spin_unlock_irq(&ctx->inflight_lock);
1196 }
1197}
1198
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001199static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001200{
Jens Axboed3656342019-12-18 09:50:26 -07001201 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001202 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001203
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001204 if (req->flags & REQ_F_FORCE_ASYNC)
1205 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1206
Jens Axboed3656342019-12-18 09:50:26 -07001207 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001208 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001209 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboed3656342019-12-18 09:50:26 -07001210 } else {
1211 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001212 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001213 }
Jens Axboe561fb042019-10-24 07:25:42 -06001214}
1215
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001216static void io_prep_async_link(struct io_kiocb *req)
1217{
1218 struct io_kiocb *cur;
1219
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001220 io_for_each_link(cur, req)
1221 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001222}
1223
Jens Axboe7271ef32020-08-10 09:55:22 -06001224static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001225{
Jackie Liua197f662019-11-08 08:09:12 -07001226 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001227 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001228 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe561fb042019-10-24 07:25:42 -06001229
Jens Axboe3bfe6102021-02-16 14:15:30 -07001230 BUG_ON(!tctx);
1231 BUG_ON(!tctx->io_wq);
Jens Axboe561fb042019-10-24 07:25:42 -06001232
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001233 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1234 &req->work, req->flags);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001235 io_wq_enqueue(tctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001236 return link;
Jens Axboe18d9be12019-09-10 09:13:05 -06001237}
1238
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001239static void io_queue_async_work(struct io_kiocb *req)
1240{
Jens Axboe7271ef32020-08-10 09:55:22 -06001241 struct io_kiocb *link;
1242
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001243 /* init ->work of the whole link before punting */
1244 io_prep_async_link(req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001245 link = __io_queue_async_work(req);
1246
1247 if (link)
1248 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001249}
1250
Jens Axboe5262f562019-09-17 12:26:57 -06001251static void io_kill_timeout(struct io_kiocb *req)
1252{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001253 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001254 int ret;
1255
Jens Axboee8c2bc12020-08-15 18:44:09 -07001256 ret = hrtimer_try_to_cancel(&io->timer);
Jens Axboe5262f562019-09-17 12:26:57 -06001257 if (ret != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001258 atomic_set(&req->ctx->cq_timeouts,
1259 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001260 list_del_init(&req->timeout.list);
Jens Axboe78e19bb2019-11-06 15:21:34 -07001261 io_cqring_fill_event(req, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001262 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001263 }
1264}
1265
Jens Axboe76e1b642020-09-26 15:05:03 -06001266/*
1267 * Returns true if we found and killed one or more timeouts
1268 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00001269static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
1270 struct files_struct *files)
Jens Axboe5262f562019-09-17 12:26:57 -06001271{
1272 struct io_kiocb *req, *tmp;
Jens Axboe76e1b642020-09-26 15:05:03 -06001273 int canceled = 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001274
1275 spin_lock_irq(&ctx->completion_lock);
Jens Axboef3606e32020-09-22 08:18:24 -06001276 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00001277 if (io_match_task(req, tsk, files)) {
Jens Axboef3606e32020-09-22 08:18:24 -06001278 io_kill_timeout(req);
Jens Axboe76e1b642020-09-26 15:05:03 -06001279 canceled++;
1280 }
Jens Axboef3606e32020-09-22 08:18:24 -06001281 }
Jens Axboe5262f562019-09-17 12:26:57 -06001282 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe76e1b642020-09-26 15:05:03 -06001283 return canceled != 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001284}
1285
Pavel Begunkov04518942020-05-26 20:34:05 +03001286static void __io_queue_deferred(struct io_ring_ctx *ctx)
1287{
1288 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001289 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1290 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001291
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001292 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001293 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001294 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001295 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001296 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001297 } while (!list_empty(&ctx->defer_list));
1298}
1299
Pavel Begunkov360428f2020-05-30 14:54:17 +03001300static void io_flush_timeouts(struct io_ring_ctx *ctx)
1301{
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001302 u32 seq;
1303
1304 if (list_empty(&ctx->timeout_list))
1305 return;
1306
1307 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
1308
1309 do {
1310 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001311 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001312 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001313
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001314 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001315 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001316
1317 /*
1318 * Since seq can easily wrap around over time, subtract
1319 * the last seq at which timeouts were flushed before comparing.
1320 * Assuming not more than 2^31-1 events have happened since,
1321 * these subtractions won't have wrapped, so we can check if
1322 * target is in [last_seq, current_seq] by comparing the two.
1323 */
1324 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1325 events_got = seq - ctx->cq_last_tm_flush;
1326 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001327 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001328
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001329 list_del_init(&req->timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001330 io_kill_timeout(req);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001331 } while (!list_empty(&ctx->timeout_list));
1332
1333 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001334}
1335
Jens Axboede0617e2019-04-06 21:51:27 -06001336static void io_commit_cqring(struct io_ring_ctx *ctx)
1337{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001338 io_flush_timeouts(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001339
1340 /* order cqe stores with ring update */
1341 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001342
Pavel Begunkov04518942020-05-26 20:34:05 +03001343 if (unlikely(!list_empty(&ctx->defer_list)))
1344 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001345}
1346
Jens Axboe90554202020-09-03 12:12:41 -06001347static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1348{
1349 struct io_rings *r = ctx->rings;
1350
1351 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1352}
1353
Pavel Begunkov888aae22021-01-19 13:32:39 +00001354static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1355{
1356 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1357}
1358
Jens Axboe2b188cc2019-01-07 10:46:33 -07001359static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
1360{
Hristo Venev75b28af2019-08-26 17:23:46 +00001361 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001362 unsigned tail;
1363
Stefan Bühler115e12e2019-04-24 23:54:18 +02001364 /*
1365 * writes to the cq entry need to come after reading head; the
1366 * control dependency is enough as we're using WRITE_ONCE to
1367 * fill the cq entry
1368 */
Pavel Begunkov888aae22021-01-19 13:32:39 +00001369 if (__io_cqring_events(ctx) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001370 return NULL;
1371
Pavel Begunkov888aae22021-01-19 13:32:39 +00001372 tail = ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001373 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001374}
1375
Jens Axboef2842ab2020-01-08 11:04:00 -07001376static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1377{
Jens Axboef0b493e2020-02-01 21:30:11 -07001378 if (!ctx->cq_ev_fd)
1379 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001380 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1381 return false;
Jens Axboef2842ab2020-01-08 11:04:00 -07001382 if (!ctx->eventfd_async)
1383 return true;
Jens Axboeb41e9852020-02-17 09:52:41 -07001384 return io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001385}
1386
Jens Axboeb41e9852020-02-17 09:52:41 -07001387static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001388{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001389 /* see waitqueue_active() comment */
1390 smp_mb();
1391
Jens Axboe8c838782019-03-12 15:48:16 -06001392 if (waitqueue_active(&ctx->wait))
1393 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001394 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1395 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001396 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001397 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001398 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001399 wake_up_interruptible(&ctx->cq_wait);
1400 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1401 }
Jens Axboe8c838782019-03-12 15:48:16 -06001402}
1403
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001404static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1405{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001406 /* see waitqueue_active() comment */
1407 smp_mb();
1408
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001409 if (ctx->flags & IORING_SETUP_SQPOLL) {
1410 if (waitqueue_active(&ctx->wait))
1411 wake_up(&ctx->wait);
1412 }
1413 if (io_should_trigger_evfd(ctx))
1414 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001415 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001416 wake_up_interruptible(&ctx->cq_wait);
1417 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1418 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001419}
1420
Jens Axboec4a2ed72019-11-21 21:01:26 -07001421/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c503152021-01-04 20:36:36 +00001422static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1423 struct task_struct *tsk,
1424 struct files_struct *files)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001425{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001426 struct io_rings *rings = ctx->rings;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001427 struct io_kiocb *req, *tmp;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001428 struct io_uring_cqe *cqe;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001429 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001430 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001431 LIST_HEAD(list);
1432
Pavel Begunkove23de152020-12-17 00:24:37 +00001433 if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
1434 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001435
Jens Axboeb18032b2021-01-24 16:58:56 -07001436 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001437 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboee6c8aa92020-09-28 13:10:13 -06001438 list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00001439 if (!io_match_task(req, tsk, files))
Jens Axboee6c8aa92020-09-28 13:10:13 -06001440 continue;
1441
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001442 cqe = io_get_cqring(ctx);
1443 if (!cqe && !force)
1444 break;
1445
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001446 list_move(&req->compl.list, &list);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001447 if (cqe) {
1448 WRITE_ONCE(cqe->user_data, req->user_data);
1449 WRITE_ONCE(cqe->res, req->result);
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001450 WRITE_ONCE(cqe->flags, req->compl.cflags);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001451 } else {
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001452 ctx->cached_cq_overflow++;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001453 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001454 ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001455 }
Jens Axboeb18032b2021-01-24 16:58:56 -07001456 posted = true;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001457 }
1458
Pavel Begunkov09e88402020-12-17 00:24:38 +00001459 all_flushed = list_empty(&ctx->cq_overflow_list);
1460 if (all_flushed) {
1461 clear_bit(0, &ctx->sq_check_overflow);
1462 clear_bit(0, &ctx->cq_check_overflow);
1463 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1464 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001465
Jens Axboeb18032b2021-01-24 16:58:56 -07001466 if (posted)
1467 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001468 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001469 if (posted)
1470 io_cqring_ev_posted(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001471
1472 while (!list_empty(&list)) {
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001473 req = list_first_entry(&list, struct io_kiocb, compl.list);
1474 list_del(&req->compl.list);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001475 io_put_req(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001476 }
Jens Axboec4a2ed72019-11-21 21:01:26 -07001477
Pavel Begunkov09e88402020-12-17 00:24:38 +00001478 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001479}
1480
Pavel Begunkov6c503152021-01-04 20:36:36 +00001481static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1482 struct task_struct *tsk,
1483 struct files_struct *files)
1484{
1485 if (test_bit(0, &ctx->cq_check_overflow)) {
1486 /* iopoll syncs against uring_lock, not completion_lock */
1487 if (ctx->flags & IORING_SETUP_IOPOLL)
1488 mutex_lock(&ctx->uring_lock);
1489 __io_cqring_overflow_flush(ctx, force, tsk, files);
1490 if (ctx->flags & IORING_SETUP_IOPOLL)
1491 mutex_unlock(&ctx->uring_lock);
1492 }
1493}
1494
Jens Axboebcda7ba2020-02-23 16:42:51 -07001495static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001496{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001497 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001498 struct io_uring_cqe *cqe;
1499
Jens Axboe78e19bb2019-11-06 15:21:34 -07001500 trace_io_uring_complete(ctx, req->user_data, res);
Jens Axboe51c3ff62019-11-03 06:52:50 -07001501
Jens Axboe2b188cc2019-01-07 10:46:33 -07001502 /*
1503 * If we can't get a cq entry, userspace overflowed the
1504 * submission (by quite a lot). Increment the overflow count in
1505 * the ring.
1506 */
1507 cqe = io_get_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001508 if (likely(cqe)) {
Jens Axboe78e19bb2019-11-06 15:21:34 -07001509 WRITE_ONCE(cqe->user_data, req->user_data);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001510 WRITE_ONCE(cqe->res, res);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001511 WRITE_ONCE(cqe->flags, cflags);
Jens Axboefdaf0832020-10-30 09:37:30 -06001512 } else if (ctx->cq_overflow_flushed ||
1513 atomic_read(&req->task->io_uring->in_idle)) {
Jens Axboe0f212202020-09-13 13:09:39 -06001514 /*
1515 * If we're in ring overflow flush mode, or in task cancel mode,
1516 * then we cannot store the request for later flushing, we need
1517 * to drop it on the floor.
1518 */
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001519 ctx->cached_cq_overflow++;
1520 WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001521 } else {
Jens Axboead3eb2c2019-12-18 17:12:20 -07001522 if (list_empty(&ctx->cq_overflow_list)) {
1523 set_bit(0, &ctx->sq_check_overflow);
1524 set_bit(0, &ctx->cq_check_overflow);
Xiaoguang Wang6d5f9042020-07-09 09:15:29 +08001525 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
Jens Axboead3eb2c2019-12-18 17:12:20 -07001526 }
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001527 io_clean_op(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001528 req->result = res;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001529 req->compl.cflags = cflags;
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001530 refcount_inc(&req->refs);
1531 list_add_tail(&req->compl.list, &ctx->cq_overflow_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001532 }
1533}
1534
Jens Axboebcda7ba2020-02-23 16:42:51 -07001535static void io_cqring_fill_event(struct io_kiocb *req, long res)
1536{
1537 __io_cqring_fill_event(req, res, 0);
1538}
1539
Jens Axboec7dae4b2021-02-09 19:53:37 -07001540static inline void io_req_complete_post(struct io_kiocb *req, long res,
1541 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001542{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001543 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001544 unsigned long flags;
1545
1546 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001547 __io_cqring_fill_event(req, res, cflags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001548 io_commit_cqring(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001549 /*
1550 * If we're the last reference to this request, add to our locked
1551 * free_list cache.
1552 */
1553 if (refcount_dec_and_test(&req->refs)) {
1554 struct io_comp_state *cs = &ctx->submit_state.comp;
1555
1556 io_dismantle_req(req);
1557 io_put_task(req->task, 1);
1558 list_add(&req->compl.list, &cs->locked_free_list);
1559 cs->locked_free_nr++;
1560 } else
1561 req = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001562 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1563
Jens Axboe8c838782019-03-12 15:48:16 -06001564 io_cqring_ev_posted(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001565 if (req) {
1566 io_queue_next(req);
1567 percpu_ref_put(&ctx->refs);
1568 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001569}
1570
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001571static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001572 unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001573{
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001574 io_clean_op(req);
1575 req->result = res;
1576 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001577 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001578}
Jens Axboe2b188cc2019-01-07 10:46:33 -07001579
Pavel Begunkov889fca72021-02-10 00:03:09 +00001580static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1581 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001582{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001583 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1584 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001585 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001586 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001587}
Jens Axboebcda7ba2020-02-23 16:42:51 -07001588
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001589static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001590{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001591 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001592}
1593
Jens Axboec7dae4b2021-02-09 19:53:37 -07001594static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001595{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001596 struct io_submit_state *state = &ctx->submit_state;
1597 struct io_comp_state *cs = &state->comp;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001598 struct io_kiocb *req = NULL;
1599
Jens Axboec7dae4b2021-02-09 19:53:37 -07001600 /*
1601 * If we have more than a batch's worth of requests in our IRQ side
1602 * locked cache, grab the lock and move them over to our submission
1603 * side cache.
1604 */
1605 if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) {
1606 spin_lock_irq(&ctx->completion_lock);
1607 list_splice_init(&cs->locked_free_list, &cs->free_list);
1608 cs->locked_free_nr = 0;
1609 spin_unlock_irq(&ctx->completion_lock);
1610 }
1611
1612 while (!list_empty(&cs->free_list)) {
1613 req = list_first_entry(&cs->free_list, struct io_kiocb,
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001614 compl.list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001615 list_del(&req->compl.list);
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001616 state->reqs[state->free_reqs++] = req;
1617 if (state->free_reqs == ARRAY_SIZE(state->reqs))
1618 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001619 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001620
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001621 return req != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001622}
1623
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001624static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001625{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001626 struct io_submit_state *state = &ctx->submit_state;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001627
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001628 BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs));
Jens Axboe2b188cc2019-01-07 10:46:33 -07001629
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001630 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001631 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07001632 int ret;
1633
Jens Axboec7dae4b2021-02-09 19:53:37 -07001634 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001635 goto got_req;
1636
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001637 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
1638 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06001639
1640 /*
1641 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1642 * retry single alloc to be on the safe side.
1643 */
1644 if (unlikely(ret <= 0)) {
1645 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1646 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00001647 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06001648 ret = 1;
1649 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001650 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001651 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001652got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03001653 state->free_reqs--;
1654 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001655}
1656
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001657static inline void io_put_file(struct io_kiocb *req, struct file *file,
1658 bool fixed)
1659{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001660 if (!fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001661 fput(file);
1662}
1663
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001664static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001665{
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001666 io_clean_op(req);
Pavel Begunkov929a3af2020-02-19 00:19:09 +03001667
Jens Axboee8c2bc12020-08-15 18:44:09 -07001668 if (req->async_data)
1669 kfree(req->async_data);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001670 if (req->file)
1671 io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001672 if (req->fixed_rsrc_refs)
1673 percpu_ref_put(req->fixed_rsrc_refs);
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001674 io_req_clean_work(req);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001675}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001676
Pavel Begunkov7c660732021-01-25 11:42:21 +00001677static inline void io_put_task(struct task_struct *task, int nr)
1678{
1679 struct io_uring_task *tctx = task->io_uring;
1680
1681 percpu_counter_sub(&tctx->inflight, nr);
1682 if (unlikely(atomic_read(&tctx->in_idle)))
1683 wake_up(&tctx->wait);
1684 put_task_struct_many(task, nr);
1685}
1686
Pavel Begunkov216578e2020-10-13 09:44:00 +01001687static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001688{
Jens Axboe51a4cc12020-08-10 10:55:56 -06001689 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001690
Pavel Begunkov216578e2020-10-13 09:44:00 +01001691 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00001692 io_put_task(req->task, 1);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001693
Pavel Begunkov3893f392021-02-10 00:03:15 +00001694 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001695 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001696}
1697
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001698static inline void io_remove_next_linked(struct io_kiocb *req)
1699{
1700 struct io_kiocb *nxt = req->link;
1701
1702 req->link = nxt->link;
1703 nxt->link = NULL;
1704}
1705
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001706static void io_kill_linked_timeout(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001707{
Jackie Liua197f662019-11-08 08:09:12 -07001708 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001709 struct io_kiocb *link;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001710 bool cancelled = false;
1711 unsigned long flags;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001712
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001713 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001714 link = req->link;
1715
Pavel Begunkov900fad42020-10-19 16:39:16 +01001716 /*
1717 * Can happen if a linked timeout fired and link had been like
1718 * req -> link t-out -> link t-out [-> ...]
1719 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001720 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1721 struct io_timeout_data *io = link->async_data;
1722 int ret;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001723
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001724 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00001725 link->timeout.head = NULL;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001726 ret = hrtimer_try_to_cancel(&io->timer);
1727 if (ret != -1) {
1728 io_cqring_fill_event(link, -ECANCELED);
1729 io_commit_cqring(ctx);
1730 cancelled = true;
1731 }
1732 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001733 req->flags &= ~REQ_F_LINK_TIMEOUT;
Pavel Begunkov216578e2020-10-13 09:44:00 +01001734 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeab0b6452020-06-30 08:43:15 -06001735
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001736 if (cancelled) {
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001737 io_cqring_ev_posted(ctx);
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001738 io_put_req(link);
1739 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001740}
1741
Jens Axboe4d7dd462019-11-20 13:03:52 -07001742
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001743static void io_fail_links(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001744{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001745 struct io_kiocb *link, *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07001746 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001747 unsigned long flags;
Jens Axboe9e645e112019-05-10 16:07:28 -06001748
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001749 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001750 link = req->link;
1751 req->link = NULL;
Jens Axboe9e645e112019-05-10 16:07:28 -06001752
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001753 while (link) {
1754 nxt = link->link;
1755 link->link = NULL;
1756
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001757 trace_io_uring_fail_link(req, link);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001758 io_cqring_fill_event(link, -ECANCELED);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001759
Jens Axboe1575f212021-02-27 15:20:49 -07001760 io_put_req_deferred(link, 2);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001761 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001762 }
Jens Axboe2665abf2019-11-05 12:40:47 -07001763 io_commit_cqring(ctx);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001764 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001765
Jens Axboe2665abf2019-11-05 12:40:47 -07001766 io_cqring_ev_posted(ctx);
Jens Axboe9e645e112019-05-10 16:07:28 -06001767}
1768
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001769static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001770{
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001771 if (req->flags & REQ_F_LINK_TIMEOUT)
1772 io_kill_linked_timeout(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07001773
Jens Axboe9e645e112019-05-10 16:07:28 -06001774 /*
1775 * If LINK is set, we have dependent requests in this chain. If we
1776 * didn't fail this request, queue the first one up, moving any other
1777 * dependencies to the next request. In case of failure, fail the rest
1778 * of the chain.
1779 */
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001780 if (likely(!(req->flags & REQ_F_FAIL_LINK))) {
1781 struct io_kiocb *nxt = req->link;
1782
1783 req->link = NULL;
1784 return nxt;
1785 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03001786 io_fail_links(req);
1787 return NULL;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001788}
Jens Axboe2665abf2019-11-05 12:40:47 -07001789
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001790static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001791{
Pavel Begunkovcdbff982021-02-12 18:41:16 +00001792 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK))))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001793 return NULL;
1794 return __io_req_find_next(req);
1795}
1796
Pavel Begunkov2c323952021-02-28 22:04:53 +00001797static void ctx_flush_and_put(struct io_ring_ctx *ctx)
1798{
1799 if (!ctx)
1800 return;
1801 if (ctx->submit_state.comp.nr) {
1802 mutex_lock(&ctx->uring_lock);
1803 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
1804 mutex_unlock(&ctx->uring_lock);
1805 }
1806 percpu_ref_put(&ctx->refs);
1807}
1808
Jens Axboe7cbf1722021-02-10 00:03:20 +00001809static bool __tctx_task_work(struct io_uring_task *tctx)
1810{
Jens Axboe65453d12021-02-10 00:03:21 +00001811 struct io_ring_ctx *ctx = NULL;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001812 struct io_wq_work_list list;
1813 struct io_wq_work_node *node;
1814
1815 if (wq_list_empty(&tctx->task_list))
1816 return false;
1817
Jens Axboe0b81e802021-02-16 10:33:53 -07001818 spin_lock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001819 list = tctx->task_list;
1820 INIT_WQ_LIST(&tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001821 spin_unlock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001822
1823 node = list.first;
1824 while (node) {
1825 struct io_wq_work_node *next = node->next;
1826 struct io_kiocb *req;
1827
1828 req = container_of(node, struct io_kiocb, io_task_work.node);
Pavel Begunkov2c323952021-02-28 22:04:53 +00001829 if (req->ctx != ctx) {
1830 ctx_flush_and_put(ctx);
1831 ctx = req->ctx;
1832 percpu_ref_get(&ctx->refs);
1833 }
1834
Jens Axboe7cbf1722021-02-10 00:03:20 +00001835 req->task_work.func(&req->task_work);
1836 node = next;
Jens Axboe65453d12021-02-10 00:03:21 +00001837 }
1838
Pavel Begunkov2c323952021-02-28 22:04:53 +00001839 ctx_flush_and_put(ctx);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001840 return list.first != NULL;
1841}
1842
1843static void tctx_task_work(struct callback_head *cb)
1844{
1845 struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work);
1846
Jens Axboe1d5f3602021-02-26 14:54:16 -07001847 clear_bit(0, &tctx->task_state);
1848
Jens Axboe7cbf1722021-02-10 00:03:20 +00001849 while (__tctx_task_work(tctx))
1850 cond_resched();
Jens Axboe7cbf1722021-02-10 00:03:20 +00001851}
1852
1853static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req,
1854 enum task_work_notify_mode notify)
1855{
1856 struct io_uring_task *tctx = tsk->io_uring;
1857 struct io_wq_work_node *node, *prev;
Jens Axboe0b81e802021-02-16 10:33:53 -07001858 unsigned long flags;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001859 int ret;
1860
1861 WARN_ON_ONCE(!tctx);
1862
Jens Axboe0b81e802021-02-16 10:33:53 -07001863 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001864 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001865 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001866
1867 /* task_work already pending, we're done */
1868 if (test_bit(0, &tctx->task_state) ||
1869 test_and_set_bit(0, &tctx->task_state))
1870 return 0;
1871
1872 if (!task_work_add(tsk, &tctx->task_work, notify))
1873 return 0;
1874
1875 /*
1876 * Slow path - we failed, find and delete work. if the work is not
1877 * in the list, it got run and we're fine.
1878 */
1879 ret = 0;
Jens Axboe0b81e802021-02-16 10:33:53 -07001880 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001881 wq_list_for_each(node, prev, &tctx->task_list) {
1882 if (&req->io_task_work.node == node) {
1883 wq_list_del(&tctx->task_list, node, prev);
1884 ret = 1;
1885 break;
1886 }
1887 }
Jens Axboe0b81e802021-02-16 10:33:53 -07001888 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001889 clear_bit(0, &tctx->task_state);
1890 return ret;
1891}
1892
Jens Axboe355fb9e2020-10-22 20:19:35 -06001893static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboec2c4c832020-07-01 15:37:11 -06001894{
1895 struct task_struct *tsk = req->task;
1896 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe91989c72020-10-16 09:02:26 -06001897 enum task_work_notify_mode notify;
1898 int ret;
Jens Axboec2c4c832020-07-01 15:37:11 -06001899
Jens Axboe6200b0a2020-09-13 14:38:30 -06001900 if (tsk->flags & PF_EXITING)
1901 return -ESRCH;
1902
Jens Axboec2c4c832020-07-01 15:37:11 -06001903 /*
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001904 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
1905 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
1906 * processing task_work. There's no reliable way to tell if TWA_RESUME
1907 * will do the job.
Jens Axboec2c4c832020-07-01 15:37:11 -06001908 */
Jens Axboe91989c72020-10-16 09:02:26 -06001909 notify = TWA_NONE;
Jens Axboe355fb9e2020-10-22 20:19:35 -06001910 if (!(ctx->flags & IORING_SETUP_SQPOLL))
Jens Axboec2c4c832020-07-01 15:37:11 -06001911 notify = TWA_SIGNAL;
1912
Jens Axboe7cbf1722021-02-10 00:03:20 +00001913 ret = io_task_work_add(tsk, req, notify);
Jens Axboec2c4c832020-07-01 15:37:11 -06001914 if (!ret)
1915 wake_up_process(tsk);
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001916
Jens Axboec2c4c832020-07-01 15:37:11 -06001917 return ret;
1918}
1919
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001920static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00001921 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001922{
Jens Axboe7c25c0d2021-02-16 07:17:00 -07001923 struct io_ring_ctx *ctx = req->ctx;
1924 struct callback_head *head;
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001925
1926 init_task_work(&req->task_work, cb);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07001927 do {
1928 head = READ_ONCE(ctx->exit_task_work);
1929 req->task_work.next = head;
1930 } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001931}
1932
Jens Axboec40f6372020-06-25 15:39:59 -06001933static void __io_req_task_cancel(struct io_kiocb *req, int error)
1934{
1935 struct io_ring_ctx *ctx = req->ctx;
1936
1937 spin_lock_irq(&ctx->completion_lock);
1938 io_cqring_fill_event(req, error);
1939 io_commit_cqring(ctx);
1940 spin_unlock_irq(&ctx->completion_lock);
1941
1942 io_cqring_ev_posted(ctx);
1943 req_set_fail_links(req);
1944 io_double_put_req(req);
1945}
1946
1947static void io_req_task_cancel(struct callback_head *cb)
1948{
1949 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06001950 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06001951
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00001952 mutex_lock(&ctx->uring_lock);
Pavel Begunkova3df76982021-02-18 22:32:52 +00001953 __io_req_task_cancel(req, req->result);
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00001954 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06001955 percpu_ref_put(&ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06001956}
1957
1958static void __io_req_task_submit(struct io_kiocb *req)
1959{
1960 struct io_ring_ctx *ctx = req->ctx;
1961
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00001962 /* ctx stays valid until unlock, even if we drop all ours ctx->refs */
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001963 mutex_lock(&ctx->uring_lock);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00001964 if (!(current->flags & PF_EXITING) && !current->in_execve)
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001965 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001966 else
Jens Axboec40f6372020-06-25 15:39:59 -06001967 __io_req_task_cancel(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001968 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06001969}
1970
Jens Axboec40f6372020-06-25 15:39:59 -06001971static void io_req_task_submit(struct callback_head *cb)
1972{
1973 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
1974
1975 __io_req_task_submit(req);
1976}
1977
1978static void io_req_task_queue(struct io_kiocb *req)
1979{
Jens Axboec40f6372020-06-25 15:39:59 -06001980 int ret;
1981
Jens Axboe7cbf1722021-02-10 00:03:20 +00001982 req->task_work.func = io_req_task_submit;
Jens Axboe355fb9e2020-10-22 20:19:35 -06001983 ret = io_req_task_work_add(req);
Jens Axboec40f6372020-06-25 15:39:59 -06001984 if (unlikely(ret)) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00001985 req->result = -ECANCELED;
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00001986 percpu_ref_get(&req->ctx->refs);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001987 io_req_task_work_add_fallback(req, io_req_task_cancel);
Jens Axboec40f6372020-06-25 15:39:59 -06001988 }
Jens Axboec40f6372020-06-25 15:39:59 -06001989}
1990
Pavel Begunkova3df76982021-02-18 22:32:52 +00001991static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
1992{
1993 percpu_ref_get(&req->ctx->refs);
1994 req->result = ret;
1995 req->task_work.func = io_req_task_cancel;
1996
1997 if (unlikely(io_req_task_work_add(req)))
1998 io_req_task_work_add_fallback(req, io_req_task_cancel);
1999}
2000
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002001static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002002{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002003 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002004
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002005 if (nxt)
2006 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002007}
2008
Jens Axboe9e645e112019-05-10 16:07:28 -06002009static void io_free_req(struct io_kiocb *req)
2010{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002011 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002012 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002013}
2014
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002015struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002016 struct task_struct *task;
2017 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002018 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002019};
2020
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002021static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002022{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002023 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002024 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002025 rb->task = NULL;
2026}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002027
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002028static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2029 struct req_batch *rb)
2030{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002031 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002032 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002033 if (rb->ctx_refs)
2034 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002035}
2036
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002037static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2038 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002039{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002040 io_queue_next(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002041
Jens Axboee3bc8e92020-09-24 08:45:57 -06002042 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002043 if (rb->task)
2044 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002045 rb->task = req->task;
2046 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002047 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002048 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002049 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002050
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002051 io_dismantle_req(req);
Pavel Begunkovbd759042021-02-12 03:23:50 +00002052 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002053 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002054 else
2055 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002056}
2057
Pavel Begunkov905c1722021-02-10 00:03:14 +00002058static void io_submit_flush_completions(struct io_comp_state *cs,
2059 struct io_ring_ctx *ctx)
2060{
2061 int i, nr = cs->nr;
2062 struct io_kiocb *req;
2063 struct req_batch rb;
2064
2065 io_init_req_batch(&rb);
2066 spin_lock_irq(&ctx->completion_lock);
2067 for (i = 0; i < nr; i++) {
2068 req = cs->reqs[i];
2069 __io_cqring_fill_event(req, req->result, req->compl.cflags);
2070 }
2071 io_commit_cqring(ctx);
2072 spin_unlock_irq(&ctx->completion_lock);
2073
2074 io_cqring_ev_posted(ctx);
2075 for (i = 0; i < nr; i++) {
2076 req = cs->reqs[i];
2077
2078 /* submission and completion refs */
2079 if (refcount_sub_and_test(2, &req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002080 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002081 }
2082
2083 io_req_free_batch_finish(ctx, &rb);
2084 cs->nr = 0;
Jens Axboee65ef562019-03-12 10:16:44 -06002085}
2086
Jens Axboeba816ad2019-09-28 11:36:45 -06002087/*
2088 * Drop reference to request, return next in chain (if there is one) if this
2089 * was the last reference to this request.
2090 */
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002091static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002092{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002093 struct io_kiocb *nxt = NULL;
2094
Jens Axboe2a44f462020-02-25 13:25:41 -07002095 if (refcount_dec_and_test(&req->refs)) {
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002096 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002097 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002098 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002099 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002100}
2101
Jens Axboe2b188cc2019-01-07 10:46:33 -07002102static void io_put_req(struct io_kiocb *req)
2103{
Jens Axboedef596e2019-01-09 08:59:42 -07002104 if (refcount_dec_and_test(&req->refs))
2105 io_free_req(req);
2106}
2107
Pavel Begunkov216578e2020-10-13 09:44:00 +01002108static void io_put_req_deferred_cb(struct callback_head *cb)
2109{
2110 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2111
2112 io_free_req(req);
2113}
2114
2115static void io_free_req_deferred(struct io_kiocb *req)
2116{
2117 int ret;
2118
Jens Axboe7cbf1722021-02-10 00:03:20 +00002119 req->task_work.func = io_put_req_deferred_cb;
Jens Axboe355fb9e2020-10-22 20:19:35 -06002120 ret = io_req_task_work_add(req);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002121 if (unlikely(ret))
2122 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002123}
2124
2125static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2126{
2127 if (refcount_sub_and_test(refs, &req->refs))
2128 io_free_req_deferred(req);
2129}
2130
Jens Axboe978db572019-11-14 22:39:04 -07002131static void io_double_put_req(struct io_kiocb *req)
2132{
2133 /* drop both submit and complete references */
2134 if (refcount_sub_and_test(2, &req->refs))
2135 io_free_req(req);
2136}
2137
Pavel Begunkov6c503152021-01-04 20:36:36 +00002138static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002139{
2140 /* See comment at the top of this file */
2141 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002142 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002143}
2144
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002145static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2146{
2147 struct io_rings *rings = ctx->rings;
2148
2149 /* make sure SQ entry isn't read before tail */
2150 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2151}
2152
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002153static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002154{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002155 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002156
Jens Axboebcda7ba2020-02-23 16:42:51 -07002157 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2158 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002159 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002160 kfree(kbuf);
2161 return cflags;
2162}
2163
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002164static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2165{
2166 struct io_buffer *kbuf;
2167
2168 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2169 return io_put_kbuf(req, kbuf);
2170}
2171
Jens Axboe4c6e2772020-07-01 11:29:10 -06002172static inline bool io_run_task_work(void)
2173{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002174 /*
2175 * Not safe to run on exiting task, and the task_work handling will
2176 * not add work to such a task.
2177 */
2178 if (unlikely(current->flags & PF_EXITING))
2179 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002180 if (current->task_works) {
2181 __set_current_state(TASK_RUNNING);
2182 task_work_run();
2183 return true;
2184 }
2185
2186 return false;
2187}
2188
Jens Axboedef596e2019-01-09 08:59:42 -07002189/*
2190 * Find and free completed poll iocbs
2191 */
2192static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2193 struct list_head *done)
2194{
Jens Axboe8237e042019-12-28 10:48:22 -07002195 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002196 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002197
2198 /* order with ->result store in io_complete_rw_iopoll() */
2199 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002200
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002201 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002202 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002203 int cflags = 0;
2204
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002205 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002206 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002207
Pavel Begunkovf1613402021-02-11 18:28:21 +00002208 if (READ_ONCE(req->result) == -EAGAIN) {
2209 req->iopoll_completed = 0;
Pavel Begunkov23faba32021-02-11 18:28:22 +00002210 if (io_rw_reissue(req))
Pavel Begunkovf1613402021-02-11 18:28:21 +00002211 continue;
2212 }
2213
Jens Axboebcda7ba2020-02-23 16:42:51 -07002214 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002215 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002216
2217 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002218 (*nr_events)++;
2219
Pavel Begunkovc3524382020-06-28 12:52:32 +03002220 if (refcount_dec_and_test(&req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002221 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002222 }
Jens Axboedef596e2019-01-09 08:59:42 -07002223
Jens Axboe09bb8392019-03-13 12:39:28 -06002224 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002225 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002226 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002227}
2228
Jens Axboedef596e2019-01-09 08:59:42 -07002229static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2230 long min)
2231{
2232 struct io_kiocb *req, *tmp;
2233 LIST_HEAD(done);
2234 bool spin;
2235 int ret;
2236
2237 /*
2238 * Only spin for completions if we don't have multiple devices hanging
2239 * off our complete list, and we're under the requested amount.
2240 */
2241 spin = !ctx->poll_multi_file && *nr_events < min;
2242
2243 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002244 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002245 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002246
2247 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002248 * Move completed and retryable entries to our local lists.
2249 * If we find a request that requires polling, break out
2250 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002251 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002252 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002253 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002254 continue;
2255 }
2256 if (!list_empty(&done))
2257 break;
2258
2259 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2260 if (ret < 0)
2261 break;
2262
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002263 /* iopoll may have completed current req */
2264 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002265 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002266
Jens Axboedef596e2019-01-09 08:59:42 -07002267 if (ret && spin)
2268 spin = false;
2269 ret = 0;
2270 }
2271
2272 if (!list_empty(&done))
2273 io_iopoll_complete(ctx, nr_events, &done);
2274
2275 return ret;
2276}
2277
2278/*
Brian Gianforcarod195a662019-12-13 03:09:50 -08002279 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
Jens Axboedef596e2019-01-09 08:59:42 -07002280 * non-spinning poll check - we'll still enter the driver poll loop, but only
2281 * as a non-spinning completion check.
2282 */
2283static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
2284 long min)
2285{
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002286 while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
Jens Axboedef596e2019-01-09 08:59:42 -07002287 int ret;
2288
2289 ret = io_do_iopoll(ctx, nr_events, min);
2290 if (ret < 0)
2291 return ret;
Pavel Begunkoveba0a4d2020-07-06 17:59:30 +03002292 if (*nr_events >= min)
Jens Axboedef596e2019-01-09 08:59:42 -07002293 return 0;
2294 }
2295
2296 return 1;
2297}
2298
2299/*
2300 * We can't just wait for polled events to come to us, we have to actively
2301 * find and complete them.
2302 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002303static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002304{
2305 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2306 return;
2307
2308 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002309 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002310 unsigned int nr_events = 0;
2311
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002312 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002313
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002314 /* let it sleep and repeat later if can't complete a request */
2315 if (nr_events == 0)
2316 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002317 /*
2318 * Ensure we allow local-to-the-cpu processing to take place,
2319 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002320 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002321 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002322 if (need_resched()) {
2323 mutex_unlock(&ctx->uring_lock);
2324 cond_resched();
2325 mutex_lock(&ctx->uring_lock);
2326 }
Jens Axboedef596e2019-01-09 08:59:42 -07002327 }
2328 mutex_unlock(&ctx->uring_lock);
2329}
2330
Pavel Begunkov7668b922020-07-07 16:36:21 +03002331static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002332{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002333 unsigned int nr_events = 0;
Jens Axboe2b2ed972019-10-25 10:06:15 -06002334 int iters = 0, ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002335
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002336 /*
2337 * We disallow the app entering submit/complete with polling, but we
2338 * still need to lock the ring to prevent racing with polled issue
2339 * that got punted to a workqueue.
2340 */
2341 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002342 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002343 /*
Jens Axboea3a0e432019-08-20 11:03:11 -06002344 * Don't enter poll loop if we already have events pending.
2345 * If we do, we can potentially be spinning for commands that
2346 * already triggered a CQE (eg in error).
2347 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00002348 if (test_bit(0, &ctx->cq_check_overflow))
2349 __io_cqring_overflow_flush(ctx, false, NULL, NULL);
2350 if (io_cqring_events(ctx))
Jens Axboea3a0e432019-08-20 11:03:11 -06002351 break;
2352
2353 /*
Jens Axboe500f9fb2019-08-19 12:15:59 -06002354 * If a submit got punted to a workqueue, we can have the
2355 * application entering polling for a command before it gets
2356 * issued. That app will hold the uring_lock for the duration
2357 * of the poll right here, so we need to take a breather every
2358 * now and then to ensure that the issue has a chance to add
2359 * the poll to the issued list. Otherwise we can spin here
2360 * forever, while the workqueue is stuck trying to acquire the
2361 * very same mutex.
2362 */
2363 if (!(++iters & 7)) {
2364 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002365 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002366 mutex_lock(&ctx->uring_lock);
2367 }
2368
Pavel Begunkov7668b922020-07-07 16:36:21 +03002369 ret = io_iopoll_getevents(ctx, &nr_events, min);
Jens Axboedef596e2019-01-09 08:59:42 -07002370 if (ret <= 0)
2371 break;
2372 ret = 0;
Pavel Begunkov7668b922020-07-07 16:36:21 +03002373 } while (min && !nr_events && !need_resched());
Jens Axboedef596e2019-01-09 08:59:42 -07002374
Jens Axboe500f9fb2019-08-19 12:15:59 -06002375 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002376 return ret;
2377}
2378
Jens Axboe491381ce2019-10-17 09:20:46 -06002379static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002380{
Jens Axboe491381ce2019-10-17 09:20:46 -06002381 /*
2382 * Tell lockdep we inherited freeze protection from submission
2383 * thread.
2384 */
2385 if (req->flags & REQ_F_ISREG) {
2386 struct inode *inode = file_inode(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002387
Jens Axboe491381ce2019-10-17 09:20:46 -06002388 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002389 }
Jens Axboe491381ce2019-10-17 09:20:46 -06002390 file_end_write(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002391}
2392
Jens Axboeb63534c2020-06-04 11:28:00 -06002393#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002394static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002395{
2396 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Colin Ian King4a245472021-02-10 20:00:07 +00002397 int rw, ret;
Jens Axboeb63534c2020-06-04 11:28:00 -06002398 struct iov_iter iter;
Jens Axboeb63534c2020-06-04 11:28:00 -06002399
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002400 /* already prepared */
2401 if (req->async_data)
2402 return true;
Jens Axboeb63534c2020-06-04 11:28:00 -06002403
2404 switch (req->opcode) {
2405 case IORING_OP_READV:
2406 case IORING_OP_READ_FIXED:
2407 case IORING_OP_READ:
2408 rw = READ;
2409 break;
2410 case IORING_OP_WRITEV:
2411 case IORING_OP_WRITE_FIXED:
2412 case IORING_OP_WRITE:
2413 rw = WRITE;
2414 break;
2415 default:
2416 printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n",
2417 req->opcode);
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002418 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002419 }
2420
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002421 ret = io_import_iovec(rw, req, &iovec, &iter, false);
2422 if (ret < 0)
2423 return false;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00002424 return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false);
Jens Axboeb63534c2020-06-04 11:28:00 -06002425}
Jens Axboeb63534c2020-06-04 11:28:00 -06002426
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002427static bool io_rw_should_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002428{
Jens Axboe355afae2020-09-02 09:30:31 -06002429 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002430 struct io_ring_ctx *ctx = req->ctx;
Jens Axboeb63534c2020-06-04 11:28:00 -06002431
Jens Axboe355afae2020-09-02 09:30:31 -06002432 if (!S_ISBLK(mode) && !S_ISREG(mode))
2433 return false;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002434 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() &&
2435 !(ctx->flags & IORING_SETUP_IOPOLL)))
Jens Axboeb63534c2020-06-04 11:28:00 -06002436 return false;
Jens Axboe7c977a52021-02-23 19:17:35 -07002437 /*
2438 * If ref is dying, we might be running poll reap from the exit work.
2439 * Don't attempt to reissue from that path, just let it fail with
2440 * -EAGAIN.
2441 */
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002442 if (percpu_ref_is_dying(&ctx->refs))
2443 return false;
2444 return true;
2445}
2446#endif
2447
2448static bool io_rw_reissue(struct io_kiocb *req)
2449{
2450#ifdef CONFIG_BLOCK
2451 if (!io_rw_should_reissue(req))
Jens Axboe7c977a52021-02-23 19:17:35 -07002452 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002453
Pavel Begunkov55e6ac12021-01-08 20:57:22 +00002454 lockdep_assert_held(&req->ctx->uring_lock);
2455
Jens Axboe37d1e2e2021-02-17 21:03:43 -07002456 if (io_resubmit_prep(req)) {
Jens Axboefdee9462020-08-27 16:46:24 -06002457 refcount_inc(&req->refs);
2458 io_queue_async_work(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002459 return true;
Jens Axboefdee9462020-08-27 16:46:24 -06002460 }
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002461 req_set_fail_links(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002462#endif
2463 return false;
2464}
2465
Jens Axboea1d7c392020-06-22 11:09:46 -06002466static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002467 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002468{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002469 int cflags = 0;
2470
Pavel Begunkov23faba32021-02-11 18:28:22 +00002471 if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req))
2472 return;
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002473 if (res != req->result)
2474 req_set_fail_links(req);
Pavel Begunkov23faba32021-02-11 18:28:22 +00002475
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002476 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2477 kiocb_end_write(req);
2478 if (req->flags & REQ_F_BUFFER_SELECTED)
2479 cflags = io_put_rw_kbuf(req);
2480 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002481}
2482
2483static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2484{
Jens Axboe9adbd452019-12-20 08:45:55 -07002485 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002486
Pavel Begunkov889fca72021-02-10 00:03:09 +00002487 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002488}
2489
Jens Axboedef596e2019-01-09 08:59:42 -07002490static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2491{
Jens Axboe9adbd452019-12-20 08:45:55 -07002492 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002493
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002494#ifdef CONFIG_BLOCK
2495 /* Rewind iter, if we have one. iopoll path resubmits as usual */
2496 if (res == -EAGAIN && io_rw_should_reissue(req)) {
2497 struct io_async_rw *rw = req->async_data;
2498
2499 if (rw)
2500 iov_iter_revert(&rw->iter,
2501 req->result - iov_iter_count(&rw->iter));
2502 else if (!io_resubmit_prep(req))
2503 res = -EIO;
2504 }
2505#endif
2506
Jens Axboe491381ce2019-10-17 09:20:46 -06002507 if (kiocb->ki_flags & IOCB_WRITE)
2508 kiocb_end_write(req);
Jens Axboedef596e2019-01-09 08:59:42 -07002509
Xiaoguang Wang2d7d6792020-06-16 02:06:37 +08002510 if (res != -EAGAIN && res != req->result)
Jens Axboe4e88d6e2019-12-07 20:59:47 -07002511 req_set_fail_links(req);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002512
2513 WRITE_ONCE(req->result, res);
2514 /* order with io_poll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002515 smp_wmb();
2516 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002517}
2518
2519/*
2520 * After the iocb has been issued, it's safe to be found on the poll list.
2521 * Adding the kiocb to the list AFTER submission ensures that we don't
2522 * find it from a io_iopoll_getevents() thread before the issuer is done
2523 * accessing the kiocb cookie.
2524 */
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002525static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
Jens Axboedef596e2019-01-09 08:59:42 -07002526{
2527 struct io_ring_ctx *ctx = req->ctx;
2528
2529 /*
2530 * Track whether we have multiple files in our lists. This will impact
2531 * how we do polling eventually, not spinning if we're on potentially
2532 * different devices.
2533 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002534 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002535 ctx->poll_multi_file = false;
2536 } else if (!ctx->poll_multi_file) {
2537 struct io_kiocb *list_req;
2538
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002539 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002540 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002541 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002542 ctx->poll_multi_file = true;
2543 }
2544
2545 /*
2546 * For fast devices, IO may have already completed. If it has, add
2547 * it to the front so we find it first.
2548 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002549 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002550 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002551 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002552 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002553
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002554 /*
2555 * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread
2556 * task context or in io worker task context. If current task context is
2557 * sq thread, we don't need to check whether should wake up sq thread.
2558 */
2559 if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002560 wq_has_sleeper(&ctx->sq_data->wait))
2561 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002562}
2563
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002564static inline void io_state_file_put(struct io_submit_state *state)
2565{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002566 if (state->file_refs) {
2567 fput_many(state->file, state->file_refs);
2568 state->file_refs = 0;
2569 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002570}
2571
2572/*
2573 * Get as many references to a file as we have IOs left in this submission,
2574 * assuming most submissions are for one file, or at least that each file
2575 * has more than one submission.
2576 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002577static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002578{
2579 if (!state)
2580 return fget(fd);
2581
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002582 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002583 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002584 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002585 return state->file;
2586 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002587 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002588 }
2589 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002590 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002591 return NULL;
2592
2593 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002594 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002595 return state->file;
2596}
2597
Jens Axboe4503b762020-06-01 10:00:27 -06002598static bool io_bdev_nowait(struct block_device *bdev)
2599{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002600 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002601}
2602
Jens Axboe2b188cc2019-01-07 10:46:33 -07002603/*
2604 * If we tracked the file through the SCM inflight mechanism, we could support
2605 * any file. For now, just ensure that anything potentially problematic is done
2606 * inline.
2607 */
Jens Axboeaf197f52020-04-28 13:15:06 -06002608static bool io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002609{
2610 umode_t mode = file_inode(file)->i_mode;
2611
Jens Axboe4503b762020-06-01 10:00:27 -06002612 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002613 if (IS_ENABLED(CONFIG_BLOCK) &&
2614 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002615 return true;
2616 return false;
2617 }
2618 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002619 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002620 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002621 if (IS_ENABLED(CONFIG_BLOCK) &&
2622 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002623 file->f_op != &io_uring_fops)
2624 return true;
2625 return false;
2626 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002627
Jens Axboec5b85622020-06-09 19:23:05 -06002628 /* any ->read/write should understand O_NONBLOCK */
2629 if (file->f_flags & O_NONBLOCK)
2630 return true;
2631
Jens Axboeaf197f52020-04-28 13:15:06 -06002632 if (!(file->f_mode & FMODE_NOWAIT))
2633 return false;
2634
2635 if (rw == READ)
2636 return file->f_op->read_iter != NULL;
2637
2638 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002639}
2640
Pavel Begunkova88fc402020-09-30 22:57:53 +03002641static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002642{
Jens Axboedef596e2019-01-09 08:59:42 -07002643 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002644 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002645 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002646 unsigned ioprio;
2647 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002648
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002649 if (S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002650 req->flags |= REQ_F_ISREG;
2651
Jens Axboe2b188cc2019-01-07 10:46:33 -07002652 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002653 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002654 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002655 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002656 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002657 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002658 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2659 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2660 if (unlikely(ret))
2661 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002662
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002663 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
2664 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
2665 req->flags |= REQ_F_NOWAIT;
2666
Jens Axboe2b188cc2019-01-07 10:46:33 -07002667 ioprio = READ_ONCE(sqe->ioprio);
2668 if (ioprio) {
2669 ret = ioprio_check_cap(ioprio);
2670 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002671 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002672
2673 kiocb->ki_ioprio = ioprio;
2674 } else
2675 kiocb->ki_ioprio = get_current_ioprio();
2676
Jens Axboedef596e2019-01-09 08:59:42 -07002677 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002678 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2679 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002680 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002681
Jens Axboedef596e2019-01-09 08:59:42 -07002682 kiocb->ki_flags |= IOCB_HIPRI;
2683 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002684 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002685 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002686 if (kiocb->ki_flags & IOCB_HIPRI)
2687 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002688 kiocb->ki_complete = io_complete_rw;
2689 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002690
Jens Axboe3529d8c2019-12-19 18:24:38 -07002691 req->rw.addr = READ_ONCE(sqe->addr);
2692 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002693 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002694 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002695}
2696
2697static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2698{
2699 switch (ret) {
2700 case -EIOCBQUEUED:
2701 break;
2702 case -ERESTARTSYS:
2703 case -ERESTARTNOINTR:
2704 case -ERESTARTNOHAND:
2705 case -ERESTART_RESTARTBLOCK:
2706 /*
2707 * We can't just restart the syscall, since previously
2708 * submitted sqes may already be in progress. Just fail this
2709 * IO with EINTR.
2710 */
2711 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002712 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002713 default:
2714 kiocb->ki_complete(kiocb, ret, 0);
2715 }
2716}
2717
Jens Axboea1d7c392020-06-22 11:09:46 -06002718static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002719 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06002720{
Jens Axboeba042912019-12-25 16:33:42 -07002721 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002722 struct io_async_rw *io = req->async_data;
Jens Axboeba042912019-12-25 16:33:42 -07002723
Jens Axboe227c0c92020-08-13 11:51:40 -06002724 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002725 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002726 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002727 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002728 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002729 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002730 }
2731
Jens Axboeba042912019-12-25 16:33:42 -07002732 if (req->flags & REQ_F_CUR_POS)
2733 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002734 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00002735 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002736 else
2737 io_rw_done(kiocb, ret);
2738}
2739
Pavel Begunkov847595d2021-02-04 13:52:06 +00002740static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07002741{
Jens Axboe9adbd452019-12-20 08:45:55 -07002742 struct io_ring_ctx *ctx = req->ctx;
2743 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002744 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03002745 u16 index, buf_index = req->buf_index;
Jens Axboeedafcce2019-01-09 09:16:05 -07002746 size_t offset;
2747 u64 buf_addr;
2748
Jens Axboeedafcce2019-01-09 09:16:05 -07002749 if (unlikely(buf_index >= ctx->nr_user_bufs))
2750 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07002751 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2752 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07002753 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002754
2755 /* overflow */
2756 if (buf_addr + len < buf_addr)
2757 return -EFAULT;
2758 /* not inside the mapped region */
2759 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
2760 return -EFAULT;
2761
2762 /*
2763 * May not be a start of buffer, set size appropriately
2764 * and advance us to the beginning.
2765 */
2766 offset = buf_addr - imu->ubuf;
2767 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002768
2769 if (offset) {
2770 /*
2771 * Don't use iov_iter_advance() here, as it's really slow for
2772 * using the latter parts of a big fixed buffer - it iterates
2773 * over each segment manually. We can cheat a bit here, because
2774 * we know that:
2775 *
2776 * 1) it's a BVEC iter, we set it up
2777 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2778 * first and last bvec
2779 *
2780 * So just find our index, and adjust the iterator afterwards.
2781 * If the offset is within the first bvec (or the whole first
2782 * bvec, just use iov_iter_advance(). This makes it easier
2783 * since we can just skip the first segment, which may not
2784 * be PAGE_SIZE aligned.
2785 */
2786 const struct bio_vec *bvec = imu->bvec;
2787
2788 if (offset <= bvec->bv_len) {
2789 iov_iter_advance(iter, offset);
2790 } else {
2791 unsigned long seg_skip;
2792
2793 /* skip first vec */
2794 offset -= bvec->bv_len;
2795 seg_skip = 1 + (offset >> PAGE_SHIFT);
2796
2797 iter->bvec = bvec + seg_skip;
2798 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002799 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002800 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002801 }
2802 }
2803
Pavel Begunkov847595d2021-02-04 13:52:06 +00002804 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07002805}
2806
Jens Axboebcda7ba2020-02-23 16:42:51 -07002807static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2808{
2809 if (needs_lock)
2810 mutex_unlock(&ctx->uring_lock);
2811}
2812
2813static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2814{
2815 /*
2816 * "Normal" inline submissions always hold the uring_lock, since we
2817 * grab it from the system call. Same is true for the SQPOLL offload.
2818 * The only exception is when we've detached the request and issue it
2819 * from an async worker thread, grab the lock for that case.
2820 */
2821 if (needs_lock)
2822 mutex_lock(&ctx->uring_lock);
2823}
2824
2825static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2826 int bgid, struct io_buffer *kbuf,
2827 bool needs_lock)
2828{
2829 struct io_buffer *head;
2830
2831 if (req->flags & REQ_F_BUFFER_SELECTED)
2832 return kbuf;
2833
2834 io_ring_submit_lock(req->ctx, needs_lock);
2835
2836 lockdep_assert_held(&req->ctx->uring_lock);
2837
2838 head = idr_find(&req->ctx->io_buffer_idr, bgid);
2839 if (head) {
2840 if (!list_empty(&head->list)) {
2841 kbuf = list_last_entry(&head->list, struct io_buffer,
2842 list);
2843 list_del(&kbuf->list);
2844 } else {
2845 kbuf = head;
2846 idr_remove(&req->ctx->io_buffer_idr, bgid);
2847 }
2848 if (*len > kbuf->len)
2849 *len = kbuf->len;
2850 } else {
2851 kbuf = ERR_PTR(-ENOBUFS);
2852 }
2853
2854 io_ring_submit_unlock(req->ctx, needs_lock);
2855
2856 return kbuf;
2857}
2858
Jens Axboe4d954c22020-02-27 07:31:19 -07002859static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2860 bool needs_lock)
2861{
2862 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002863 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002864
2865 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002866 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002867 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2868 if (IS_ERR(kbuf))
2869 return kbuf;
2870 req->rw.addr = (u64) (unsigned long) kbuf;
2871 req->flags |= REQ_F_BUFFER_SELECTED;
2872 return u64_to_user_ptr(kbuf->addr);
2873}
2874
2875#ifdef CONFIG_COMPAT
2876static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2877 bool needs_lock)
2878{
2879 struct compat_iovec __user *uiov;
2880 compat_ssize_t clen;
2881 void __user *buf;
2882 ssize_t len;
2883
2884 uiov = u64_to_user_ptr(req->rw.addr);
2885 if (!access_ok(uiov, sizeof(*uiov)))
2886 return -EFAULT;
2887 if (__get_user(clen, &uiov->iov_len))
2888 return -EFAULT;
2889 if (clen < 0)
2890 return -EINVAL;
2891
2892 len = clen;
2893 buf = io_rw_buffer_select(req, &len, needs_lock);
2894 if (IS_ERR(buf))
2895 return PTR_ERR(buf);
2896 iov[0].iov_base = buf;
2897 iov[0].iov_len = (compat_size_t) len;
2898 return 0;
2899}
2900#endif
2901
2902static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2903 bool needs_lock)
2904{
2905 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
2906 void __user *buf;
2907 ssize_t len;
2908
2909 if (copy_from_user(iov, uiov, sizeof(*uiov)))
2910 return -EFAULT;
2911
2912 len = iov[0].iov_len;
2913 if (len < 0)
2914 return -EINVAL;
2915 buf = io_rw_buffer_select(req, &len, needs_lock);
2916 if (IS_ERR(buf))
2917 return PTR_ERR(buf);
2918 iov[0].iov_base = buf;
2919 iov[0].iov_len = len;
2920 return 0;
2921}
2922
2923static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2924 bool needs_lock)
2925{
Jens Axboedddb3e22020-06-04 11:27:01 -06002926 if (req->flags & REQ_F_BUFFER_SELECTED) {
2927 struct io_buffer *kbuf;
2928
2929 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2930 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
2931 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07002932 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06002933 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00002934 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07002935 return -EINVAL;
2936
2937#ifdef CONFIG_COMPAT
2938 if (req->ctx->compat)
2939 return io_compat_import(req, iov, needs_lock);
2940#endif
2941
2942 return __io_iov_buffer_select(req, iov, needs_lock);
2943}
2944
Pavel Begunkov847595d2021-02-04 13:52:06 +00002945static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
2946 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002947{
Jens Axboe9adbd452019-12-20 08:45:55 -07002948 void __user *buf = u64_to_user_ptr(req->rw.addr);
2949 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00002950 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07002951 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07002952
Pavel Begunkov7d009162019-11-25 23:14:40 +03002953 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07002954 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07002955 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07002956 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002957
Jens Axboebcda7ba2020-02-23 16:42:51 -07002958 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002959 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07002960 return -EINVAL;
2961
Jens Axboe3a6820f2019-12-22 15:19:35 -07002962 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002963 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07002964 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03002965 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07002966 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06002967 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002968 }
2969
Jens Axboe3a6820f2019-12-22 15:19:35 -07002970 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
2971 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00002972 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07002973 }
2974
Jens Axboe4d954c22020-02-27 07:31:19 -07002975 if (req->flags & REQ_F_BUFFER_SELECT) {
2976 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00002977 if (!ret)
2978 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07002979 *iovec = NULL;
2980 return ret;
2981 }
2982
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02002983 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
2984 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002985}
2986
Jens Axboe0fef9482020-08-26 10:36:20 -06002987static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
2988{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03002989 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06002990}
2991
Jens Axboe32960612019-09-23 11:05:34 -06002992/*
2993 * For files that don't have ->read_iter() and ->write_iter(), handle them
2994 * by looping over ->read() or ->write() manually.
2995 */
Jens Axboe4017eb92020-10-22 14:14:12 -06002996static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06002997{
Jens Axboe4017eb92020-10-22 14:14:12 -06002998 struct kiocb *kiocb = &req->rw.kiocb;
2999 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003000 ssize_t ret = 0;
3001
3002 /*
3003 * Don't support polled IO through this interface, and we can't
3004 * support non-blocking either. For the latter, this just causes
3005 * the kiocb to be handled from an async context.
3006 */
3007 if (kiocb->ki_flags & IOCB_HIPRI)
3008 return -EOPNOTSUPP;
3009 if (kiocb->ki_flags & IOCB_NOWAIT)
3010 return -EAGAIN;
3011
3012 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003013 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003014 ssize_t nr;
3015
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003016 if (!iov_iter_is_bvec(iter)) {
3017 iovec = iov_iter_iovec(iter);
3018 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003019 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3020 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003021 }
3022
Jens Axboe32960612019-09-23 11:05:34 -06003023 if (rw == READ) {
3024 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003025 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003026 } else {
3027 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003028 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003029 }
3030
3031 if (nr < 0) {
3032 if (!ret)
3033 ret = nr;
3034 break;
3035 }
3036 ret += nr;
3037 if (nr != iovec.iov_len)
3038 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003039 req->rw.len -= nr;
3040 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003041 iov_iter_advance(iter, nr);
3042 }
3043
3044 return ret;
3045}
3046
Jens Axboeff6165b2020-08-13 09:47:43 -06003047static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3048 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003049{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003050 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003051
Jens Axboeff6165b2020-08-13 09:47:43 -06003052 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003053 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003054 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003055 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003056 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003057 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003058 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003059 unsigned iov_off = 0;
3060
3061 rw->iter.iov = rw->fast_iov;
3062 if (iter->iov != fast_iov) {
3063 iov_off = iter->iov - fast_iov;
3064 rw->iter.iov += iov_off;
3065 }
3066 if (rw->fast_iov != fast_iov)
3067 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003068 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003069 } else {
3070 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003071 }
3072}
3073
Jens Axboee8c2bc12020-08-15 18:44:09 -07003074static inline int __io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003075{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003076 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3077 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3078 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003079}
3080
Jens Axboee8c2bc12020-08-15 18:44:09 -07003081static int io_alloc_async_data(struct io_kiocb *req)
Jens Axboef67676d2019-12-02 11:03:47 -07003082{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003083 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboed3656342019-12-18 09:50:26 -07003084 return 0;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003085
Jens Axboee8c2bc12020-08-15 18:44:09 -07003086 return __io_alloc_async_data(req);
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003087}
3088
Jens Axboeff6165b2020-08-13 09:47:43 -06003089static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3090 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003091 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003092{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003093 if (!force && !io_op_defs[req->opcode].needs_async_data)
Jens Axboe74566df2020-01-13 19:23:24 -07003094 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003095 if (!req->async_data) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003096 if (__io_alloc_async_data(req)) {
3097 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003098 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003099 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003100
Jens Axboeff6165b2020-08-13 09:47:43 -06003101 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003102 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003103 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003104}
3105
Pavel Begunkov73debe62020-09-30 22:57:54 +03003106static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003107{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003108 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003109 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003110 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003111
Pavel Begunkov2846c482020-11-07 13:16:27 +00003112 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003113 if (unlikely(ret < 0))
3114 return ret;
3115
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003116 iorw->bytes_done = 0;
3117 iorw->free_iovec = iov;
3118 if (iov)
3119 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003120 return 0;
3121}
3122
Pavel Begunkov73debe62020-09-30 22:57:54 +03003123static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003124{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003125 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3126 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003127 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003128}
3129
Jens Axboec1dd91d2020-08-03 16:43:59 -06003130/*
3131 * This is our waitqueue callback handler, registered through lock_page_async()
3132 * when we initially tried to do the IO with the iocb armed our waitqueue.
3133 * This gets called when the page is unlocked, and we generally expect that to
3134 * happen when the page IO is completed and the page is now uptodate. This will
3135 * queue a task_work based retry of the operation, attempting to copy the data
3136 * again. If the latter fails because the page was NOT uptodate, then we will
3137 * do a thread based blocking retry of the operation. That's the unexpected
3138 * slow path.
3139 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003140static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3141 int sync, void *arg)
3142{
3143 struct wait_page_queue *wpq;
3144 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003145 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003146
3147 wpq = container_of(wait, struct wait_page_queue, wait);
3148
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003149 if (!wake_page_match(wpq, key))
3150 return 0;
3151
Hao Xuc8d317a2020-09-29 20:00:45 +08003152 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003153 list_del_init(&wait->entry);
3154
Jens Axboebcf5a062020-05-22 09:24:42 -06003155 /* submit ref gets dropped, acquire a new one */
3156 refcount_inc(&req->refs);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003157 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003158 return 1;
3159}
3160
Jens Axboec1dd91d2020-08-03 16:43:59 -06003161/*
3162 * This controls whether a given IO request should be armed for async page
3163 * based retry. If we return false here, the request is handed to the async
3164 * worker threads for retry. If we're doing buffered reads on a regular file,
3165 * we prepare a private wait_page_queue entry and retry the operation. This
3166 * will either succeed because the page is now uptodate and unlocked, or it
3167 * will register a callback when the page is unlocked at IO completion. Through
3168 * that callback, io_uring uses task_work to setup a retry of the operation.
3169 * That retry will attempt the buffered read again. The retry will generally
3170 * succeed, or in rare cases where it fails, we then fall back to using the
3171 * async worker threads for a blocking retry.
3172 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003173static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003174{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003175 struct io_async_rw *rw = req->async_data;
3176 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003177 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003178
3179 /* never retry for NOWAIT, we just complete with -EAGAIN */
3180 if (req->flags & REQ_F_NOWAIT)
3181 return false;
3182
Jens Axboe227c0c92020-08-13 11:51:40 -06003183 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003184 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003185 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003186
Jens Axboebcf5a062020-05-22 09:24:42 -06003187 /*
3188 * just use poll if we can, and don't attempt if the fs doesn't
3189 * support callback based unlocks
3190 */
3191 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3192 return false;
3193
Jens Axboe3b2a4432020-08-16 10:58:43 -07003194 wait->wait.func = io_async_buf_func;
3195 wait->wait.private = req;
3196 wait->wait.flags = 0;
3197 INIT_LIST_HEAD(&wait->wait.entry);
3198 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003199 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003200 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003201 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003202}
3203
3204static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3205{
3206 if (req->file->f_op->read_iter)
3207 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003208 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003209 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003210 else
3211 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003212}
3213
Pavel Begunkov889fca72021-02-10 00:03:09 +00003214static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003215{
3216 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003217 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003218 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003219 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003220 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003221 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003222
Pavel Begunkov2846c482020-11-07 13:16:27 +00003223 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003224 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003225 iovec = NULL;
3226 } else {
3227 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3228 if (ret < 0)
3229 return ret;
3230 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003231 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003232 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003233
Jens Axboefd6c2e42019-12-18 12:19:41 -07003234 /* Ensure we clear previously set non-block flag */
3235 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003236 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003237 else
3238 kiocb->ki_flags |= IOCB_NOWAIT;
3239
Pavel Begunkov24c74672020-06-21 13:09:51 +03003240 /* If the file doesn't support async, just async punt */
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003241 if (force_nonblock && !io_file_supports_async(req->file, READ)) {
3242 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003243 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003244 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003245
Pavel Begunkov632546c2020-11-07 13:16:26 +00003246 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003247 if (unlikely(ret)) {
3248 kfree(iovec);
3249 return ret;
3250 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003251
Jens Axboe227c0c92020-08-13 11:51:40 -06003252 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003253
Pavel Begunkov57cd6572021-02-01 18:59:56 +00003254 if (ret == -EIOCBQUEUED) {
Jens Axboe3e6a0d32021-03-01 13:56:00 -07003255 if (req->async_data)
3256 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003257 goto out_free;
Jens Axboe227c0c92020-08-13 11:51:40 -06003258 } else if (ret == -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003259 /* IOPOLL retry should happen for io-wq threads */
3260 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003261 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003262 /* no retry on NONBLOCK nor RWF_NOWAIT */
3263 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003264 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003265 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003266 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003267 ret = 0;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003268 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003269 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003270 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003271 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003272 }
3273
Jens Axboe227c0c92020-08-13 11:51:40 -06003274 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003275 if (ret2)
3276 return ret2;
3277
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003278 iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003279 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003280 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003281 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003282
Pavel Begunkovb23df912021-02-04 13:52:04 +00003283 do {
3284 io_size -= ret;
3285 rw->bytes_done += ret;
3286 /* if we can retry, do so with the callbacks armed */
3287 if (!io_rw_should_retry(req)) {
3288 kiocb->ki_flags &= ~IOCB_WAITQ;
3289 return -EAGAIN;
3290 }
3291
3292 /*
3293 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3294 * we get -EIOCBQUEUED, then we'll get a notification when the
3295 * desired page gets unlocked. We can also get a partial read
3296 * here, and if we do, then just retry at the new offset.
3297 */
3298 ret = io_iter_do_read(req, iter);
3299 if (ret == -EIOCBQUEUED)
3300 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003301 /* we got some bytes, but not all. retry. */
Pavel Begunkovb23df912021-02-04 13:52:04 +00003302 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003303done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003304 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003305out_free:
3306 /* it's faster to check here then delegate to kfree */
3307 if (iovec)
3308 kfree(iovec);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003309 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003310}
3311
Pavel Begunkov73debe62020-09-30 22:57:54 +03003312static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003313{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003314 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3315 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003316 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003317}
3318
Pavel Begunkov889fca72021-02-10 00:03:09 +00003319static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003320{
3321 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003322 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003323 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003324 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003325 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003326 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003327
Pavel Begunkov2846c482020-11-07 13:16:27 +00003328 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003329 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003330 iovec = NULL;
3331 } else {
3332 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3333 if (ret < 0)
3334 return ret;
3335 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003336 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003337 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003338
Jens Axboefd6c2e42019-12-18 12:19:41 -07003339 /* Ensure we clear previously set non-block flag */
3340 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003341 kiocb->ki_flags &= ~IOCB_NOWAIT;
3342 else
3343 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003344
Pavel Begunkov24c74672020-06-21 13:09:51 +03003345 /* If the file doesn't support async, just async punt */
Jens Axboeaf197f52020-04-28 13:15:06 -06003346 if (force_nonblock && !io_file_supports_async(req->file, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003347 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003348
Jens Axboe10d59342019-12-09 20:16:22 -07003349 /* file path doesn't support NOWAIT for non-direct_IO */
3350 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3351 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003352 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003353
Pavel Begunkov632546c2020-11-07 13:16:26 +00003354 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003355 if (unlikely(ret))
3356 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003357
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003358 /*
3359 * Open-code file_start_write here to grab freeze protection,
3360 * which will be released by another thread in
3361 * io_complete_rw(). Fool lockdep by telling it the lock got
3362 * released so that it doesn't complain about the held lock when
3363 * we return to userspace.
3364 */
3365 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003366 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003367 __sb_writers_release(file_inode(req->file)->i_sb,
3368 SB_FREEZE_WRITE);
3369 }
3370 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003371
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003372 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003373 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003374 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003375 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003376 else
3377 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003378
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003379 /*
3380 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3381 * retry them without IOCB_NOWAIT.
3382 */
3383 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3384 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003385 /* no retry on NONBLOCK nor RWF_NOWAIT */
3386 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003387 goto done;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07003388 if (ret2 == -EIOCBQUEUED && req->async_data)
3389 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003390 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003391 /* IOPOLL retry should happen for io-wq threads */
3392 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3393 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003394done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003395 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003396 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003397copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003398 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003399 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003400 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003401 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003402 }
Jens Axboe31b51512019-01-18 22:56:34 -07003403out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003404 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003405 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003406 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003407 return ret;
3408}
3409
Jens Axboe80a261f2020-09-28 14:23:58 -06003410static int io_renameat_prep(struct io_kiocb *req,
3411 const struct io_uring_sqe *sqe)
3412{
3413 struct io_rename *ren = &req->rename;
3414 const char __user *oldf, *newf;
3415
3416 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3417 return -EBADF;
3418
3419 ren->old_dfd = READ_ONCE(sqe->fd);
3420 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3421 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3422 ren->new_dfd = READ_ONCE(sqe->len);
3423 ren->flags = READ_ONCE(sqe->rename_flags);
3424
3425 ren->oldpath = getname(oldf);
3426 if (IS_ERR(ren->oldpath))
3427 return PTR_ERR(ren->oldpath);
3428
3429 ren->newpath = getname(newf);
3430 if (IS_ERR(ren->newpath)) {
3431 putname(ren->oldpath);
3432 return PTR_ERR(ren->newpath);
3433 }
3434
3435 req->flags |= REQ_F_NEED_CLEANUP;
3436 return 0;
3437}
3438
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003439static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003440{
3441 struct io_rename *ren = &req->rename;
3442 int ret;
3443
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003444 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003445 return -EAGAIN;
3446
3447 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3448 ren->newpath, ren->flags);
3449
3450 req->flags &= ~REQ_F_NEED_CLEANUP;
3451 if (ret < 0)
3452 req_set_fail_links(req);
3453 io_req_complete(req, ret);
3454 return 0;
3455}
3456
Jens Axboe14a11432020-09-28 14:27:37 -06003457static int io_unlinkat_prep(struct io_kiocb *req,
3458 const struct io_uring_sqe *sqe)
3459{
3460 struct io_unlink *un = &req->unlink;
3461 const char __user *fname;
3462
3463 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3464 return -EBADF;
3465
3466 un->dfd = READ_ONCE(sqe->fd);
3467
3468 un->flags = READ_ONCE(sqe->unlink_flags);
3469 if (un->flags & ~AT_REMOVEDIR)
3470 return -EINVAL;
3471
3472 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3473 un->filename = getname(fname);
3474 if (IS_ERR(un->filename))
3475 return PTR_ERR(un->filename);
3476
3477 req->flags |= REQ_F_NEED_CLEANUP;
3478 return 0;
3479}
3480
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003481static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003482{
3483 struct io_unlink *un = &req->unlink;
3484 int ret;
3485
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003486 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003487 return -EAGAIN;
3488
3489 if (un->flags & AT_REMOVEDIR)
3490 ret = do_rmdir(un->dfd, un->filename);
3491 else
3492 ret = do_unlinkat(un->dfd, un->filename);
3493
3494 req->flags &= ~REQ_F_NEED_CLEANUP;
3495 if (ret < 0)
3496 req_set_fail_links(req);
3497 io_req_complete(req, ret);
3498 return 0;
3499}
3500
Jens Axboe36f4fa62020-09-05 11:14:22 -06003501static int io_shutdown_prep(struct io_kiocb *req,
3502 const struct io_uring_sqe *sqe)
3503{
3504#if defined(CONFIG_NET)
3505 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3506 return -EINVAL;
3507 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3508 sqe->buf_index)
3509 return -EINVAL;
3510
3511 req->shutdown.how = READ_ONCE(sqe->len);
3512 return 0;
3513#else
3514 return -EOPNOTSUPP;
3515#endif
3516}
3517
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003518static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003519{
3520#if defined(CONFIG_NET)
3521 struct socket *sock;
3522 int ret;
3523
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003524 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003525 return -EAGAIN;
3526
Linus Torvalds48aba792020-12-16 12:44:05 -08003527 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003528 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003529 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003530
3531 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003532 if (ret < 0)
3533 req_set_fail_links(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003534 io_req_complete(req, ret);
3535 return 0;
3536#else
3537 return -EOPNOTSUPP;
3538#endif
3539}
3540
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003541static int __io_splice_prep(struct io_kiocb *req,
3542 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003543{
3544 struct io_splice* sp = &req->splice;
3545 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003546
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003547 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3548 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003549
3550 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003551 sp->len = READ_ONCE(sqe->len);
3552 sp->flags = READ_ONCE(sqe->splice_flags);
3553
3554 if (unlikely(sp->flags & ~valid_flags))
3555 return -EINVAL;
3556
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003557 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3558 (sp->flags & SPLICE_F_FD_IN_FIXED));
3559 if (!sp->file_in)
3560 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003561 req->flags |= REQ_F_NEED_CLEANUP;
3562
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003563 if (!S_ISREG(file_inode(sp->file_in)->i_mode)) {
3564 /*
3565 * Splice operation will be punted aync, and here need to
3566 * modify io_wq_work.flags, so initialize io_wq_work firstly.
3567 */
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003568 req->work.flags |= IO_WQ_WORK_UNBOUND;
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003569 }
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003570
3571 return 0;
3572}
3573
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003574static int io_tee_prep(struct io_kiocb *req,
3575 const struct io_uring_sqe *sqe)
3576{
3577 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3578 return -EINVAL;
3579 return __io_splice_prep(req, sqe);
3580}
3581
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003582static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003583{
3584 struct io_splice *sp = &req->splice;
3585 struct file *in = sp->file_in;
3586 struct file *out = sp->file_out;
3587 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3588 long ret = 0;
3589
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003590 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003591 return -EAGAIN;
3592 if (sp->len)
3593 ret = do_tee(in, out, sp->len, flags);
3594
3595 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3596 req->flags &= ~REQ_F_NEED_CLEANUP;
3597
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003598 if (ret != sp->len)
3599 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003600 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003601 return 0;
3602}
3603
3604static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3605{
3606 struct io_splice* sp = &req->splice;
3607
3608 sp->off_in = READ_ONCE(sqe->splice_off_in);
3609 sp->off_out = READ_ONCE(sqe->off);
3610 return __io_splice_prep(req, sqe);
3611}
3612
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003613static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003614{
3615 struct io_splice *sp = &req->splice;
3616 struct file *in = sp->file_in;
3617 struct file *out = sp->file_out;
3618 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3619 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003620 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003621
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003622 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003623 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003624
3625 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3626 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003627
Jens Axboe948a7742020-05-17 14:21:38 -06003628 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003629 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003630
3631 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3632 req->flags &= ~REQ_F_NEED_CLEANUP;
3633
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003634 if (ret != sp->len)
3635 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003636 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003637 return 0;
3638}
3639
Jens Axboe2b188cc2019-01-07 10:46:33 -07003640/*
3641 * IORING_OP_NOP just posts a completion event, nothing else.
3642 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003643static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003644{
3645 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003646
Jens Axboedef596e2019-01-09 08:59:42 -07003647 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3648 return -EINVAL;
3649
Pavel Begunkov889fca72021-02-10 00:03:09 +00003650 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003651 return 0;
3652}
3653
Pavel Begunkov1155c762021-02-18 18:29:38 +00003654static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003655{
Jens Axboe6b063142019-01-10 22:13:58 -07003656 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003657
Jens Axboe09bb8392019-03-13 12:39:28 -06003658 if (!req->file)
3659 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003660
Jens Axboe6b063142019-01-10 22:13:58 -07003661 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003662 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003663 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003664 return -EINVAL;
3665
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003666 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3667 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3668 return -EINVAL;
3669
3670 req->sync.off = READ_ONCE(sqe->off);
3671 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003672 return 0;
3673}
3674
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003675static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07003676{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003677 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003678 int ret;
3679
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003680 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003681 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003682 return -EAGAIN;
3683
Jens Axboe9adbd452019-12-20 08:45:55 -07003684 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003685 end > 0 ? end : LLONG_MAX,
3686 req->sync.flags & IORING_FSYNC_DATASYNC);
3687 if (ret < 0)
3688 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003689 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003690 return 0;
3691}
3692
Jens Axboed63d1b52019-12-10 10:38:56 -07003693static int io_fallocate_prep(struct io_kiocb *req,
3694 const struct io_uring_sqe *sqe)
3695{
3696 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3697 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003698 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3699 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003700
3701 req->sync.off = READ_ONCE(sqe->off);
3702 req->sync.len = READ_ONCE(sqe->addr);
3703 req->sync.mode = READ_ONCE(sqe->len);
3704 return 0;
3705}
3706
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003707static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07003708{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003709 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003710
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003711 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003712 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003713 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003714 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3715 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003716 if (ret < 0)
3717 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003718 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003719 return 0;
3720}
3721
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003722static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003723{
Jens Axboef8748882020-01-08 17:47:02 -07003724 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003725 int ret;
3726
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003727 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003728 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003729 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003730 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003731
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003732 /* open.how should be already initialised */
3733 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003734 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003735
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003736 req->open.dfd = READ_ONCE(sqe->fd);
3737 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003738 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003739 if (IS_ERR(req->open.filename)) {
3740 ret = PTR_ERR(req->open.filename);
3741 req->open.filename = NULL;
3742 return ret;
3743 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003744 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003745 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003746 return 0;
3747}
3748
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003749static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3750{
3751 u64 flags, mode;
3752
Jens Axboe14587a462020-09-05 11:36:08 -06003753 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003754 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003755 mode = READ_ONCE(sqe->len);
3756 flags = READ_ONCE(sqe->open_flags);
3757 req->open.how = build_open_how(flags, mode);
3758 return __io_openat_prep(req, sqe);
3759}
3760
Jens Axboecebdb982020-01-08 17:59:24 -07003761static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3762{
3763 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003764 size_t len;
3765 int ret;
3766
Jens Axboe14587a462020-09-05 11:36:08 -06003767 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003768 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003769 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3770 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003771 if (len < OPEN_HOW_SIZE_VER0)
3772 return -EINVAL;
3773
3774 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3775 len);
3776 if (ret)
3777 return ret;
3778
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003779 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003780}
3781
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003782static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003783{
3784 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003785 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003786 bool nonblock_set;
3787 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003788 int ret;
3789
Jens Axboecebdb982020-01-08 17:59:24 -07003790 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003791 if (ret)
3792 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003793 nonblock_set = op.open_flag & O_NONBLOCK;
3794 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003795 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003796 /*
3797 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
3798 * it'll always -EAGAIN
3799 */
3800 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
3801 return -EAGAIN;
3802 op.lookup_flags |= LOOKUP_CACHED;
3803 op.open_flag |= O_NONBLOCK;
3804 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07003805
Jens Axboe4022e7a2020-03-19 19:23:18 -06003806 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003807 if (ret < 0)
3808 goto err;
3809
3810 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Jens Axboe3a81fd02020-12-10 12:25:36 -07003811 /* only retry if RESOLVE_CACHED wasn't already set by application */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003812 if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) &&
3813 file == ERR_PTR(-EAGAIN)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003814 /*
3815 * We could hang on to this 'fd', but seems like marginal
3816 * gain for something that is now known to be a slower path.
3817 * So just put it, and we'll get a new one when we retry.
3818 */
3819 put_unused_fd(ret);
3820 return -EAGAIN;
3821 }
3822
Jens Axboe15b71ab2019-12-11 11:20:36 -07003823 if (IS_ERR(file)) {
3824 put_unused_fd(ret);
3825 ret = PTR_ERR(file);
3826 } else {
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003827 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
Jens Axboe3a81fd02020-12-10 12:25:36 -07003828 file->f_flags &= ~O_NONBLOCK;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003829 fsnotify_open(file);
3830 fd_install(ret, file);
3831 }
3832err:
3833 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003834 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003835 if (ret < 0)
3836 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003837 io_req_complete(req, ret);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003838 return 0;
3839}
3840
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003841static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07003842{
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003843 return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK);
Jens Axboecebdb982020-01-08 17:59:24 -07003844}
3845
Jens Axboe067524e2020-03-02 16:32:28 -07003846static int io_remove_buffers_prep(struct io_kiocb *req,
3847 const struct io_uring_sqe *sqe)
3848{
3849 struct io_provide_buf *p = &req->pbuf;
3850 u64 tmp;
3851
3852 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3853 return -EINVAL;
3854
3855 tmp = READ_ONCE(sqe->fd);
3856 if (!tmp || tmp > USHRT_MAX)
3857 return -EINVAL;
3858
3859 memset(p, 0, sizeof(*p));
3860 p->nbufs = tmp;
3861 p->bgid = READ_ONCE(sqe->buf_group);
3862 return 0;
3863}
3864
3865static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3866 int bgid, unsigned nbufs)
3867{
3868 unsigned i = 0;
3869
3870 /* shouldn't happen */
3871 if (!nbufs)
3872 return 0;
3873
3874 /* the head kbuf is the list itself */
3875 while (!list_empty(&buf->list)) {
3876 struct io_buffer *nxt;
3877
3878 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3879 list_del(&nxt->list);
3880 kfree(nxt);
3881 if (++i == nbufs)
3882 return i;
3883 }
3884 i++;
3885 kfree(buf);
3886 idr_remove(&ctx->io_buffer_idr, bgid);
3887
3888 return i;
3889}
3890
Pavel Begunkov889fca72021-02-10 00:03:09 +00003891static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07003892{
3893 struct io_provide_buf *p = &req->pbuf;
3894 struct io_ring_ctx *ctx = req->ctx;
3895 struct io_buffer *head;
3896 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003897 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07003898
3899 io_ring_submit_lock(ctx, !force_nonblock);
3900
3901 lockdep_assert_held(&ctx->uring_lock);
3902
3903 ret = -ENOENT;
3904 head = idr_find(&ctx->io_buffer_idr, p->bgid);
3905 if (head)
3906 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07003907 if (ret < 0)
3908 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003909
3910 /* need to hold the lock to complete IOPOLL requests */
3911 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00003912 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003913 io_ring_submit_unlock(ctx, !force_nonblock);
3914 } else {
3915 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00003916 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003917 }
Jens Axboe067524e2020-03-02 16:32:28 -07003918 return 0;
3919}
3920
Jens Axboeddf0322d2020-02-23 16:41:33 -07003921static int io_provide_buffers_prep(struct io_kiocb *req,
3922 const struct io_uring_sqe *sqe)
3923{
3924 struct io_provide_buf *p = &req->pbuf;
3925 u64 tmp;
3926
3927 if (sqe->ioprio || sqe->rw_flags)
3928 return -EINVAL;
3929
3930 tmp = READ_ONCE(sqe->fd);
3931 if (!tmp || tmp > USHRT_MAX)
3932 return -E2BIG;
3933 p->nbufs = tmp;
3934 p->addr = READ_ONCE(sqe->addr);
3935 p->len = READ_ONCE(sqe->len);
3936
Bijan Mottahedehefe68c12020-06-04 18:01:52 -07003937 if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs)))
Jens Axboeddf0322d2020-02-23 16:41:33 -07003938 return -EFAULT;
3939
3940 p->bgid = READ_ONCE(sqe->buf_group);
3941 tmp = READ_ONCE(sqe->off);
3942 if (tmp > USHRT_MAX)
3943 return -E2BIG;
3944 p->bid = tmp;
3945 return 0;
3946}
3947
3948static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
3949{
3950 struct io_buffer *buf;
3951 u64 addr = pbuf->addr;
3952 int i, bid = pbuf->bid;
3953
3954 for (i = 0; i < pbuf->nbufs; i++) {
3955 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
3956 if (!buf)
3957 break;
3958
3959 buf->addr = addr;
3960 buf->len = pbuf->len;
3961 buf->bid = bid;
3962 addr += pbuf->len;
3963 bid++;
3964 if (!*head) {
3965 INIT_LIST_HEAD(&buf->list);
3966 *head = buf;
3967 } else {
3968 list_add_tail(&buf->list, &(*head)->list);
3969 }
3970 }
3971
3972 return i ? i : -ENOMEM;
3973}
3974
Pavel Begunkov889fca72021-02-10 00:03:09 +00003975static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07003976{
3977 struct io_provide_buf *p = &req->pbuf;
3978 struct io_ring_ctx *ctx = req->ctx;
3979 struct io_buffer *head, *list;
3980 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003981 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003982
3983 io_ring_submit_lock(ctx, !force_nonblock);
3984
3985 lockdep_assert_held(&ctx->uring_lock);
3986
3987 list = head = idr_find(&ctx->io_buffer_idr, p->bgid);
3988
3989 ret = io_add_buffers(p, &head);
3990 if (ret < 0)
3991 goto out;
3992
3993 if (!list) {
3994 ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1,
3995 GFP_KERNEL);
3996 if (ret < 0) {
Jens Axboe067524e2020-03-02 16:32:28 -07003997 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07003998 goto out;
3999 }
4000 }
4001out:
Jens Axboeddf0322d2020-02-23 16:41:33 -07004002 if (ret < 0)
4003 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004004
4005 /* need to hold the lock to complete IOPOLL requests */
4006 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00004007 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004008 io_ring_submit_unlock(ctx, !force_nonblock);
4009 } else {
4010 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004011 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004012 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07004013 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004014}
4015
Jens Axboe3e4827b2020-01-08 15:18:09 -07004016static int io_epoll_ctl_prep(struct io_kiocb *req,
4017 const struct io_uring_sqe *sqe)
4018{
4019#if defined(CONFIG_EPOLL)
4020 if (sqe->ioprio || sqe->buf_index)
4021 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004022 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004023 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004024
4025 req->epoll.epfd = READ_ONCE(sqe->fd);
4026 req->epoll.op = READ_ONCE(sqe->len);
4027 req->epoll.fd = READ_ONCE(sqe->off);
4028
4029 if (ep_op_has_event(req->epoll.op)) {
4030 struct epoll_event __user *ev;
4031
4032 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4033 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4034 return -EFAULT;
4035 }
4036
4037 return 0;
4038#else
4039 return -EOPNOTSUPP;
4040#endif
4041}
4042
Pavel Begunkov889fca72021-02-10 00:03:09 +00004043static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004044{
4045#if defined(CONFIG_EPOLL)
4046 struct io_epoll *ie = &req->epoll;
4047 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004048 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004049
4050 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4051 if (force_nonblock && ret == -EAGAIN)
4052 return -EAGAIN;
4053
4054 if (ret < 0)
4055 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004056 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004057 return 0;
4058#else
4059 return -EOPNOTSUPP;
4060#endif
4061}
4062
Jens Axboec1ca7572019-12-25 22:18:28 -07004063static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4064{
4065#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4066 if (sqe->ioprio || sqe->buf_index || sqe->off)
4067 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004068 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4069 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004070
4071 req->madvise.addr = READ_ONCE(sqe->addr);
4072 req->madvise.len = READ_ONCE(sqe->len);
4073 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4074 return 0;
4075#else
4076 return -EOPNOTSUPP;
4077#endif
4078}
4079
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004080static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004081{
4082#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4083 struct io_madvise *ma = &req->madvise;
4084 int ret;
4085
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004086 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004087 return -EAGAIN;
4088
Minchan Kim0726b012020-10-17 16:14:50 -07004089 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004090 if (ret < 0)
4091 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004092 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004093 return 0;
4094#else
4095 return -EOPNOTSUPP;
4096#endif
4097}
4098
Jens Axboe4840e412019-12-25 22:03:45 -07004099static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4100{
4101 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4102 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004103 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4104 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004105
4106 req->fadvise.offset = READ_ONCE(sqe->off);
4107 req->fadvise.len = READ_ONCE(sqe->len);
4108 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4109 return 0;
4110}
4111
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004112static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004113{
4114 struct io_fadvise *fa = &req->fadvise;
4115 int ret;
4116
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004117 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004118 switch (fa->advice) {
4119 case POSIX_FADV_NORMAL:
4120 case POSIX_FADV_RANDOM:
4121 case POSIX_FADV_SEQUENTIAL:
4122 break;
4123 default:
4124 return -EAGAIN;
4125 }
4126 }
Jens Axboe4840e412019-12-25 22:03:45 -07004127
4128 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4129 if (ret < 0)
4130 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004131 io_req_complete(req, ret);
Jens Axboe4840e412019-12-25 22:03:45 -07004132 return 0;
4133}
4134
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004135static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4136{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004137 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004138 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004139 if (sqe->ioprio || sqe->buf_index)
4140 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004141 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004142 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004143
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004144 req->statx.dfd = READ_ONCE(sqe->fd);
4145 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004146 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004147 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4148 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004149
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004150 return 0;
4151}
4152
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004153static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004154{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004155 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004156 int ret;
4157
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004158 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004159 /* only need file table for an actual valid fd */
4160 if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD)
4161 req->flags |= REQ_F_NO_FILE_TABLE;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004162 return -EAGAIN;
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004163 }
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004164
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004165 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4166 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004167
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004168 if (ret < 0)
4169 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004170 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004171 return 0;
4172}
4173
Jens Axboeb5dba592019-12-11 14:02:38 -07004174static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4175{
Jens Axboe14587a462020-09-05 11:36:08 -06004176 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004177 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004178 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4179 sqe->rw_flags || sqe->buf_index)
4180 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004181 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004182 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004183
4184 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004185 return 0;
4186}
4187
Pavel Begunkov889fca72021-02-10 00:03:09 +00004188static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004189{
Jens Axboe9eac1902021-01-19 15:50:37 -07004190 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004191 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004192 struct fdtable *fdt;
4193 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -07004194 int ret;
4195
Jens Axboe9eac1902021-01-19 15:50:37 -07004196 file = NULL;
4197 ret = -EBADF;
4198 spin_lock(&files->file_lock);
4199 fdt = files_fdtable(files);
4200 if (close->fd >= fdt->max_fds) {
4201 spin_unlock(&files->file_lock);
4202 goto err;
4203 }
4204 file = fdt->fd[close->fd];
4205 if (!file) {
4206 spin_unlock(&files->file_lock);
4207 goto err;
4208 }
4209
4210 if (file->f_op == &io_uring_fops) {
4211 spin_unlock(&files->file_lock);
4212 file = NULL;
4213 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004214 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004215
4216 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004217 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004218 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004219 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004220 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004221
Jens Axboe9eac1902021-01-19 15:50:37 -07004222 ret = __close_fd_get_file(close->fd, &file);
4223 spin_unlock(&files->file_lock);
4224 if (ret < 0) {
4225 if (ret == -ENOENT)
4226 ret = -EBADF;
4227 goto err;
4228 }
4229
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004230 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004231 ret = filp_close(file, current->files);
4232err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004233 if (ret < 0)
4234 req_set_fail_links(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004235 if (file)
4236 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004237 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004238 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004239}
4240
Pavel Begunkov1155c762021-02-18 18:29:38 +00004241static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004242{
4243 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004244
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004245 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4246 return -EINVAL;
4247 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4248 return -EINVAL;
4249
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004250 req->sync.off = READ_ONCE(sqe->off);
4251 req->sync.len = READ_ONCE(sqe->len);
4252 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004253 return 0;
4254}
4255
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004256static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004257{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004258 int ret;
4259
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004260 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004261 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004262 return -EAGAIN;
4263
Jens Axboe9adbd452019-12-20 08:45:55 -07004264 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004265 req->sync.flags);
4266 if (ret < 0)
4267 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004268 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004269 return 0;
4270}
4271
YueHaibing469956e2020-03-04 15:53:52 +08004272#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004273static int io_setup_async_msg(struct io_kiocb *req,
4274 struct io_async_msghdr *kmsg)
4275{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004276 struct io_async_msghdr *async_msg = req->async_data;
4277
4278 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004279 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004280 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004281 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004282 return -ENOMEM;
4283 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004284 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004285 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004286 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004287 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004288 /* if were using fast_iov, set it to the new one */
4289 if (!async_msg->free_iov)
4290 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4291
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004292 return -EAGAIN;
4293}
4294
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004295static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4296 struct io_async_msghdr *iomsg)
4297{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004298 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004299 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004300 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004301 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004302}
4303
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004304static int io_sendmsg_prep_async(struct io_kiocb *req)
4305{
4306 int ret;
4307
4308 if (!io_op_defs[req->opcode].needs_async_data)
4309 return 0;
4310 ret = io_sendmsg_copy_hdr(req, req->async_data);
4311 if (!ret)
4312 req->flags |= REQ_F_NEED_CLEANUP;
4313 return ret;
4314}
4315
Jens Axboe3529d8c2019-12-19 18:24:38 -07004316static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004317{
Jens Axboee47293f2019-12-20 08:58:21 -07004318 struct io_sr_msg *sr = &req->sr_msg;
Jens Axboe03b12302019-12-02 18:50:25 -07004319
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004320 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4321 return -EINVAL;
4322
Jens Axboee47293f2019-12-20 08:58:21 -07004323 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004324 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004325 sr->len = READ_ONCE(sqe->len);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004326
Jens Axboed8768362020-02-27 14:17:49 -07004327#ifdef CONFIG_COMPAT
4328 if (req->ctx->compat)
4329 sr->msg_flags |= MSG_CMSG_COMPAT;
4330#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004331 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004332}
4333
Pavel Begunkov889fca72021-02-10 00:03:09 +00004334static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004335{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004336 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004337 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004338 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004339 int ret;
4340
Florent Revestdba4a922020-12-04 12:36:04 +01004341 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004342 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004343 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004344
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004345 kmsg = req->async_data;
4346 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004347 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004348 if (ret)
4349 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004350 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004351 }
4352
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004353 flags = req->sr_msg.msg_flags;
4354 if (flags & MSG_DONTWAIT)
4355 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004356 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004357 flags |= MSG_DONTWAIT;
4358
4359 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004360 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004361 return io_setup_async_msg(req, kmsg);
4362 if (ret == -ERESTARTSYS)
4363 ret = -EINTR;
4364
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004365 /* fast path, check for non-NULL to avoid function call */
4366 if (kmsg->free_iov)
4367 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004368 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboefddafac2020-01-04 20:19:44 -07004369 if (ret < 0)
4370 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004371 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004372 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004373}
4374
Pavel Begunkov889fca72021-02-10 00:03:09 +00004375static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004376{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004377 struct io_sr_msg *sr = &req->sr_msg;
4378 struct msghdr msg;
4379 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004380 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004381 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004382 int ret;
4383
Florent Revestdba4a922020-12-04 12:36:04 +01004384 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004385 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004386 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004387
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004388 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4389 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004390 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004391
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004392 msg.msg_name = NULL;
4393 msg.msg_control = NULL;
4394 msg.msg_controllen = 0;
4395 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004396
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004397 flags = req->sr_msg.msg_flags;
4398 if (flags & MSG_DONTWAIT)
4399 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004400 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004401 flags |= MSG_DONTWAIT;
Jens Axboe03b12302019-12-02 18:50:25 -07004402
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004403 msg.msg_flags = flags;
4404 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004405 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004406 return -EAGAIN;
4407 if (ret == -ERESTARTSYS)
4408 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004409
Jens Axboe03b12302019-12-02 18:50:25 -07004410 if (ret < 0)
4411 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004412 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004413 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004414}
4415
Pavel Begunkov1400e692020-07-12 20:41:05 +03004416static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4417 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004418{
4419 struct io_sr_msg *sr = &req->sr_msg;
4420 struct iovec __user *uiov;
4421 size_t iov_len;
4422 int ret;
4423
Pavel Begunkov1400e692020-07-12 20:41:05 +03004424 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4425 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004426 if (ret)
4427 return ret;
4428
4429 if (req->flags & REQ_F_BUFFER_SELECT) {
4430 if (iov_len > 1)
4431 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004432 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004433 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004434 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004435 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004436 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004437 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004438 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004439 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004440 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004441 if (ret > 0)
4442 ret = 0;
4443 }
4444
4445 return ret;
4446}
4447
4448#ifdef CONFIG_COMPAT
4449static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004450 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004451{
4452 struct compat_msghdr __user *msg_compat;
4453 struct io_sr_msg *sr = &req->sr_msg;
4454 struct compat_iovec __user *uiov;
4455 compat_uptr_t ptr;
4456 compat_size_t len;
4457 int ret;
4458
Pavel Begunkov270a5942020-07-12 20:41:04 +03004459 msg_compat = (struct compat_msghdr __user *) sr->umsg;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004460 ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr,
Jens Axboe52de1fe2020-02-27 10:15:42 -07004461 &ptr, &len);
4462 if (ret)
4463 return ret;
4464
4465 uiov = compat_ptr(ptr);
4466 if (req->flags & REQ_F_BUFFER_SELECT) {
4467 compat_ssize_t clen;
4468
4469 if (len > 1)
4470 return -EINVAL;
4471 if (!access_ok(uiov, sizeof(*uiov)))
4472 return -EFAULT;
4473 if (__get_user(clen, &uiov->iov_len))
4474 return -EFAULT;
4475 if (clen < 0)
4476 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004477 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004478 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004479 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004480 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004481 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004482 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004483 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004484 if (ret < 0)
4485 return ret;
4486 }
4487
4488 return 0;
4489}
Jens Axboe03b12302019-12-02 18:50:25 -07004490#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004491
Pavel Begunkov1400e692020-07-12 20:41:05 +03004492static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4493 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004494{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004495 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004496
4497#ifdef CONFIG_COMPAT
4498 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004499 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004500#endif
4501
Pavel Begunkov1400e692020-07-12 20:41:05 +03004502 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004503}
4504
Jens Axboebcda7ba2020-02-23 16:42:51 -07004505static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004506 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004507{
4508 struct io_sr_msg *sr = &req->sr_msg;
4509 struct io_buffer *kbuf;
4510
Jens Axboebcda7ba2020-02-23 16:42:51 -07004511 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4512 if (IS_ERR(kbuf))
4513 return kbuf;
4514
4515 sr->kbuf = kbuf;
4516 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004517 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004518}
4519
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004520static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4521{
4522 return io_put_kbuf(req, req->sr_msg.kbuf);
4523}
4524
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004525static int io_recvmsg_prep_async(struct io_kiocb *req)
Jens Axboe03b12302019-12-02 18:50:25 -07004526{
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004527 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004528
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004529 if (!io_op_defs[req->opcode].needs_async_data)
4530 return 0;
4531 ret = io_recvmsg_copy_hdr(req, req->async_data);
4532 if (!ret)
4533 req->flags |= REQ_F_NEED_CLEANUP;
4534 return ret;
4535}
4536
4537static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4538{
4539 struct io_sr_msg *sr = &req->sr_msg;
4540
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004541 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4542 return -EINVAL;
4543
Jens Axboe3529d8c2019-12-19 18:24:38 -07004544 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004545 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004546 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004547 sr->bgid = READ_ONCE(sqe->buf_group);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004548
Jens Axboed8768362020-02-27 14:17:49 -07004549#ifdef CONFIG_COMPAT
4550 if (req->ctx->compat)
4551 sr->msg_flags |= MSG_CMSG_COMPAT;
4552#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004553 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004554}
4555
Pavel Begunkov889fca72021-02-10 00:03:09 +00004556static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004557{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004558 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004559 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004560 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004561 unsigned flags;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004562 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004563 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004564
Florent Revestdba4a922020-12-04 12:36:04 +01004565 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004566 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004567 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004568
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004569 kmsg = req->async_data;
4570 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004571 ret = io_recvmsg_copy_hdr(req, &iomsg);
4572 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004573 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004574 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004575 }
4576
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004577 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004578 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004579 if (IS_ERR(kbuf))
4580 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004581 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004582 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4583 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004584 1, req->sr_msg.len);
4585 }
4586
4587 flags = req->sr_msg.msg_flags;
4588 if (flags & MSG_DONTWAIT)
4589 req->flags |= REQ_F_NOWAIT;
4590 else if (force_nonblock)
4591 flags |= MSG_DONTWAIT;
4592
4593 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4594 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004595 if (force_nonblock && ret == -EAGAIN)
4596 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004597 if (ret == -ERESTARTSYS)
4598 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004599
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004600 if (req->flags & REQ_F_BUFFER_SELECTED)
4601 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004602 /* fast path, check for non-NULL to avoid function call */
4603 if (kmsg->free_iov)
4604 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004605 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004606 if (ret < 0)
4607 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004608 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004609 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004610}
4611
Pavel Begunkov889fca72021-02-10 00:03:09 +00004612static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004613{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004614 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004615 struct io_sr_msg *sr = &req->sr_msg;
4616 struct msghdr msg;
4617 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004618 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004619 struct iovec iov;
4620 unsigned flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004621 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004622 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004623
Florent Revestdba4a922020-12-04 12:36:04 +01004624 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004625 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004626 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004627
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004628 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004629 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004630 if (IS_ERR(kbuf))
4631 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004632 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004633 }
4634
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004635 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004636 if (unlikely(ret))
4637 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004638
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004639 msg.msg_name = NULL;
4640 msg.msg_control = NULL;
4641 msg.msg_controllen = 0;
4642 msg.msg_namelen = 0;
4643 msg.msg_iocb = NULL;
4644 msg.msg_flags = 0;
4645
4646 flags = req->sr_msg.msg_flags;
4647 if (flags & MSG_DONTWAIT)
4648 req->flags |= REQ_F_NOWAIT;
4649 else if (force_nonblock)
4650 flags |= MSG_DONTWAIT;
4651
4652 ret = sock_recvmsg(sock, &msg, flags);
4653 if (force_nonblock && ret == -EAGAIN)
4654 return -EAGAIN;
4655 if (ret == -ERESTARTSYS)
4656 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004657out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004658 if (req->flags & REQ_F_BUFFER_SELECTED)
4659 cflags = io_put_recv_kbuf(req);
Jens Axboefddafac2020-01-04 20:19:44 -07004660 if (ret < 0)
4661 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004662 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07004663 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004664}
4665
Jens Axboe3529d8c2019-12-19 18:24:38 -07004666static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004667{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004668 struct io_accept *accept = &req->accept;
4669
Jens Axboe14587a462020-09-05 11:36:08 -06004670 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06004671 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004672 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004673 return -EINVAL;
4674
Jens Axboed55e5f52019-12-11 16:12:15 -07004675 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4676 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004677 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004678 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004679 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004680}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004681
Pavel Begunkov889fca72021-02-10 00:03:09 +00004682static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004683{
4684 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004685 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004686 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004687 int ret;
4688
Jiufei Xuee697dee2020-06-10 13:41:59 +08004689 if (req->file->f_flags & O_NONBLOCK)
4690 req->flags |= REQ_F_NOWAIT;
4691
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004692 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004693 accept->addr_len, accept->flags,
4694 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004695 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004696 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004697 if (ret < 0) {
4698 if (ret == -ERESTARTSYS)
4699 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004700 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004701 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00004702 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004703 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004704}
4705
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004706static int io_connect_prep_async(struct io_kiocb *req)
4707{
4708 struct io_async_connect *io = req->async_data;
4709 struct io_connect *conn = &req->connect;
4710
4711 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
4712}
4713
Jens Axboe3529d8c2019-12-19 18:24:38 -07004714static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004715{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004716 struct io_connect *conn = &req->connect;
Jens Axboef499a022019-12-02 16:28:46 -07004717
Jens Axboe14587a462020-09-05 11:36:08 -06004718 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004719 return -EINVAL;
4720 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4721 return -EINVAL;
4722
Jens Axboe3529d8c2019-12-19 18:24:38 -07004723 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4724 conn->addr_len = READ_ONCE(sqe->addr2);
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004725 return 0;
Jens Axboef499a022019-12-02 16:28:46 -07004726}
4727
Pavel Begunkov889fca72021-02-10 00:03:09 +00004728static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004729{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004730 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004731 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004732 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004733 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004734
Jens Axboee8c2bc12020-08-15 18:44:09 -07004735 if (req->async_data) {
4736 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004737 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004738 ret = move_addr_to_kernel(req->connect.addr,
4739 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004740 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004741 if (ret)
4742 goto out;
4743 io = &__io;
4744 }
4745
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004746 file_flags = force_nonblock ? O_NONBLOCK : 0;
4747
Jens Axboee8c2bc12020-08-15 18:44:09 -07004748 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004749 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004750 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004751 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004752 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004753 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004754 ret = -ENOMEM;
4755 goto out;
4756 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004757 io = req->async_data;
4758 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004759 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004760 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004761 if (ret == -ERESTARTSYS)
4762 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004763out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004764 if (ret < 0)
4765 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004766 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004767 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004768}
YueHaibing469956e2020-03-04 15:53:52 +08004769#else /* !CONFIG_NET */
Jens Axboe99a10082021-02-19 09:35:19 -07004770#define IO_NETOP_FN(op) \
4771static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \
4772{ \
4773 return -EOPNOTSUPP; \
Jens Axboef8e85cf2019-11-23 14:24:24 -07004774}
4775
Jens Axboe99a10082021-02-19 09:35:19 -07004776#define IO_NETOP_PREP(op) \
4777IO_NETOP_FN(op) \
4778static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \
4779{ \
4780 return -EOPNOTSUPP; \
4781} \
4782
4783#define IO_NETOP_PREP_ASYNC(op) \
4784IO_NETOP_PREP(op) \
4785static int io_##op##_prep_async(struct io_kiocb *req) \
4786{ \
4787 return -EOPNOTSUPP; \
YueHaibing469956e2020-03-04 15:53:52 +08004788}
4789
Jens Axboe99a10082021-02-19 09:35:19 -07004790IO_NETOP_PREP_ASYNC(sendmsg);
4791IO_NETOP_PREP_ASYNC(recvmsg);
4792IO_NETOP_PREP_ASYNC(connect);
4793IO_NETOP_PREP(accept);
4794IO_NETOP_FN(send);
4795IO_NETOP_FN(recv);
YueHaibing469956e2020-03-04 15:53:52 +08004796#endif /* CONFIG_NET */
Jens Axboe17f2fe32019-10-17 14:42:58 -06004797
Jens Axboed7718a92020-02-14 22:23:12 -07004798struct io_poll_table {
4799 struct poll_table_struct pt;
4800 struct io_kiocb *req;
4801 int error;
4802};
4803
Jens Axboed7718a92020-02-14 22:23:12 -07004804static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4805 __poll_t mask, task_work_func_t func)
4806{
Jens Axboeaa96bf82020-04-03 11:26:26 -06004807 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004808
4809 /* for instances that support it check for an event match first: */
4810 if (mask && !(mask & poll->events))
4811 return 0;
4812
4813 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4814
4815 list_del_init(&poll->wait.entry);
4816
Jens Axboed7718a92020-02-14 22:23:12 -07004817 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00004818 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06004819 percpu_ref_get(&req->ctx->refs);
4820
Jens Axboed7718a92020-02-14 22:23:12 -07004821 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004822 * If this fails, then the task is exiting. When a task exits, the
4823 * work gets canceled, so just cancel this request as well instead
4824 * of executing it. We can't safely execute it anyway, as we may not
4825 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004826 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06004827 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004828 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06004829 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00004830 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004831 }
Jens Axboed7718a92020-02-14 22:23:12 -07004832 return 1;
4833}
4834
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004835static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4836 __acquires(&req->ctx->completion_lock)
4837{
4838 struct io_ring_ctx *ctx = req->ctx;
4839
4840 if (!req->result && !READ_ONCE(poll->canceled)) {
4841 struct poll_table_struct pt = { ._key = poll->events };
4842
4843 req->result = vfs_poll(req->file, &pt) & poll->events;
4844 }
4845
4846 spin_lock_irq(&ctx->completion_lock);
4847 if (!req->result && !READ_ONCE(poll->canceled)) {
4848 add_wait_queue(poll->head, &poll->wait);
4849 return true;
4850 }
4851
4852 return false;
4853}
4854
Jens Axboed4e7cd32020-08-15 11:44:50 -07004855static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004856{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004857 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004858 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004859 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004860 return req->apoll->double_poll;
4861}
4862
4863static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4864{
4865 if (req->opcode == IORING_OP_POLL_ADD)
4866 return &req->poll;
4867 return &req->apoll->poll;
4868}
4869
4870static void io_poll_remove_double(struct io_kiocb *req)
4871{
4872 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004873
4874 lockdep_assert_held(&req->ctx->completion_lock);
4875
4876 if (poll && poll->head) {
4877 struct wait_queue_head *head = poll->head;
4878
4879 spin_lock(&head->lock);
4880 list_del_init(&poll->wait.entry);
4881 if (poll->wait.private)
4882 refcount_dec(&req->refs);
4883 poll->head = NULL;
4884 spin_unlock(&head->lock);
4885 }
4886}
4887
4888static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
4889{
4890 struct io_ring_ctx *ctx = req->ctx;
4891
Jens Axboed4e7cd32020-08-15 11:44:50 -07004892 io_poll_remove_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004893 req->poll.done = true;
4894 io_cqring_fill_event(req, error ? error : mangle_poll(mask));
4895 io_commit_cqring(ctx);
4896}
4897
Jens Axboe18bceab2020-05-15 11:56:54 -06004898static void io_poll_task_func(struct callback_head *cb)
4899{
4900 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004901 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004902 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004903
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004904 if (io_poll_rewait(req, &req->poll)) {
4905 spin_unlock_irq(&ctx->completion_lock);
4906 } else {
4907 hash_del(&req->hash_node);
4908 io_poll_complete(req, req->result, 0);
4909 spin_unlock_irq(&ctx->completion_lock);
4910
4911 nxt = io_put_req_find_next(req);
4912 io_cqring_ev_posted(ctx);
4913 if (nxt)
4914 __io_req_task_submit(nxt);
4915 }
4916
Jens Axboe6d816e02020-08-11 08:04:14 -06004917 percpu_ref_put(&ctx->refs);
Jens Axboe18bceab2020-05-15 11:56:54 -06004918}
4919
4920static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4921 int sync, void *key)
4922{
4923 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004924 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004925 __poll_t mask = key_to_poll(key);
4926
4927 /* for instances that support it check for an event match first: */
4928 if (mask && !(mask & poll->events))
4929 return 0;
4930
Jens Axboe8706e042020-09-28 08:38:54 -06004931 list_del_init(&wait->entry);
4932
Jens Axboe807abcb2020-07-17 17:09:27 -06004933 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004934 bool done;
4935
Jens Axboe807abcb2020-07-17 17:09:27 -06004936 spin_lock(&poll->head->lock);
4937 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06004938 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06004939 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07004940 /* make sure double remove sees this as being gone */
4941 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06004942 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06004943 if (!done) {
4944 /* use wait func handler, so it matches the rq type */
4945 poll->wait.func(&poll->wait, mode, sync, key);
4946 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004947 }
4948 refcount_dec(&req->refs);
4949 return 1;
4950}
4951
4952static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
4953 wait_queue_func_t wake_func)
4954{
4955 poll->head = NULL;
4956 poll->done = false;
4957 poll->canceled = false;
4958 poll->events = events;
4959 INIT_LIST_HEAD(&poll->wait.entry);
4960 init_waitqueue_func_entry(&poll->wait, wake_func);
4961}
4962
4963static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06004964 struct wait_queue_head *head,
4965 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06004966{
4967 struct io_kiocb *req = pt->req;
4968
4969 /*
4970 * If poll->head is already set, it's because the file being polled
4971 * uses multiple waitqueues for poll handling (eg one for read, one
4972 * for write). Setup a separate io_poll_iocb if this happens.
4973 */
4974 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01004975 struct io_poll_iocb *poll_one = poll;
4976
Jens Axboe18bceab2020-05-15 11:56:54 -06004977 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06004978 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004979 pt->error = -EINVAL;
4980 return;
4981 }
Jens Axboe1c3b3e62021-02-28 16:07:30 -07004982 /* double add on the same waitqueue head, ignore */
4983 if (poll->head == head)
4984 return;
Jens Axboe18bceab2020-05-15 11:56:54 -06004985 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
4986 if (!poll) {
4987 pt->error = -ENOMEM;
4988 return;
4989 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01004990 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboe18bceab2020-05-15 11:56:54 -06004991 refcount_inc(&req->refs);
4992 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06004993 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06004994 }
4995
4996 pt->error = 0;
4997 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08004998
4999 if (poll->events & EPOLLEXCLUSIVE)
5000 add_wait_queue_exclusive(head, &poll->wait);
5001 else
5002 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005003}
5004
5005static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5006 struct poll_table_struct *p)
5007{
5008 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005009 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005010
Jens Axboe807abcb2020-07-17 17:09:27 -06005011 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005012}
5013
Jens Axboed7718a92020-02-14 22:23:12 -07005014static void io_async_task_func(struct callback_head *cb)
5015{
5016 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5017 struct async_poll *apoll = req->apoll;
5018 struct io_ring_ctx *ctx = req->ctx;
5019
5020 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5021
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005022 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005023 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe6d816e02020-08-11 08:04:14 -06005024 percpu_ref_put(&ctx->refs);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005025 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005026 }
5027
Jens Axboe31067252020-05-17 17:43:31 -06005028 /* If req is still hashed, it cannot have been canceled. Don't check. */
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005029 if (hash_hashed(&req->hash_node))
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005030 hash_del(&req->hash_node);
Jens Axboe2bae0472020-04-13 11:16:34 -06005031
Jens Axboed4e7cd32020-08-15 11:44:50 -07005032 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005033 spin_unlock_irq(&ctx->completion_lock);
5034
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005035 if (!READ_ONCE(apoll->poll.canceled))
5036 __io_req_task_submit(req);
5037 else
5038 __io_req_task_cancel(req, -ECANCELED);
Dan Carpenteraa340842020-07-08 21:47:11 +03005039
Jens Axboe6d816e02020-08-11 08:04:14 -06005040 percpu_ref_put(&ctx->refs);
Jens Axboe807abcb2020-07-17 17:09:27 -06005041 kfree(apoll->double_poll);
Jens Axboe31067252020-05-17 17:43:31 -06005042 kfree(apoll);
Jens Axboed7718a92020-02-14 22:23:12 -07005043}
5044
5045static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5046 void *key)
5047{
5048 struct io_kiocb *req = wait->private;
5049 struct io_poll_iocb *poll = &req->apoll->poll;
5050
5051 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5052 key_to_poll(key));
5053
5054 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5055}
5056
5057static void io_poll_req_insert(struct io_kiocb *req)
5058{
5059 struct io_ring_ctx *ctx = req->ctx;
5060 struct hlist_head *list;
5061
5062 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5063 hlist_add_head(&req->hash_node, list);
5064}
5065
5066static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5067 struct io_poll_iocb *poll,
5068 struct io_poll_table *ipt, __poll_t mask,
5069 wait_queue_func_t wake_func)
5070 __acquires(&ctx->completion_lock)
5071{
5072 struct io_ring_ctx *ctx = req->ctx;
5073 bool cancel = false;
5074
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005075 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005076 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005077 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005078 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005079
5080 ipt->pt._key = mask;
5081 ipt->req = req;
5082 ipt->error = -EINVAL;
5083
Jens Axboed7718a92020-02-14 22:23:12 -07005084 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5085
5086 spin_lock_irq(&ctx->completion_lock);
5087 if (likely(poll->head)) {
5088 spin_lock(&poll->head->lock);
5089 if (unlikely(list_empty(&poll->wait.entry))) {
5090 if (ipt->error)
5091 cancel = true;
5092 ipt->error = 0;
5093 mask = 0;
5094 }
5095 if (mask || ipt->error)
5096 list_del_init(&poll->wait.entry);
5097 else if (cancel)
5098 WRITE_ONCE(poll->canceled, true);
5099 else if (!poll->done) /* actually waiting for an event */
5100 io_poll_req_insert(req);
5101 spin_unlock(&poll->head->lock);
5102 }
5103
5104 return mask;
5105}
5106
5107static bool io_arm_poll_handler(struct io_kiocb *req)
5108{
5109 const struct io_op_def *def = &io_op_defs[req->opcode];
5110 struct io_ring_ctx *ctx = req->ctx;
5111 struct async_poll *apoll;
5112 struct io_poll_table ipt;
5113 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005114 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005115
5116 if (!req->file || !file_can_poll(req->file))
5117 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005118 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005119 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005120 if (def->pollin)
5121 rw = READ;
5122 else if (def->pollout)
5123 rw = WRITE;
5124 else
5125 return false;
5126 /* if we can't nonblock try, then no point in arming a poll handler */
5127 if (!io_file_supports_async(req->file, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005128 return false;
5129
5130 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5131 if (unlikely(!apoll))
5132 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005133 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005134
5135 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005136 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005137
Nathan Chancellor8755d972020-03-02 16:01:19 -07005138 mask = 0;
Jens Axboed7718a92020-02-14 22:23:12 -07005139 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005140 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005141 if (def->pollout)
5142 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005143
5144 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5145 if ((req->opcode == IORING_OP_RECVMSG) &&
5146 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5147 mask &= ~POLLIN;
5148
Jens Axboed7718a92020-02-14 22:23:12 -07005149 mask |= POLLERR | POLLPRI;
5150
5151 ipt.pt._qproc = io_async_queue_proc;
5152
5153 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5154 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005155 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005156 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005157 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe807abcb2020-07-17 17:09:27 -06005158 kfree(apoll->double_poll);
Jens Axboed7718a92020-02-14 22:23:12 -07005159 kfree(apoll);
5160 return false;
5161 }
5162 spin_unlock_irq(&ctx->completion_lock);
5163 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5164 apoll->poll.events);
5165 return true;
5166}
5167
5168static bool __io_poll_remove_one(struct io_kiocb *req,
5169 struct io_poll_iocb *poll)
5170{
Jens Axboeb41e9852020-02-17 09:52:41 -07005171 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005172
5173 spin_lock(&poll->head->lock);
5174 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005175 if (!list_empty(&poll->wait.entry)) {
5176 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005177 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005178 }
5179 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005180 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005181 return do_complete;
5182}
5183
5184static bool io_poll_remove_one(struct io_kiocb *req)
5185{
5186 bool do_complete;
5187
Jens Axboed4e7cd32020-08-15 11:44:50 -07005188 io_poll_remove_double(req);
5189
Jens Axboed7718a92020-02-14 22:23:12 -07005190 if (req->opcode == IORING_OP_POLL_ADD) {
5191 do_complete = __io_poll_remove_one(req, &req->poll);
5192 } else {
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005193 struct async_poll *apoll = req->apoll;
5194
Jens Axboed7718a92020-02-14 22:23:12 -07005195 /* non-poll requests have submit ref still */
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005196 do_complete = __io_poll_remove_one(req, &apoll->poll);
5197 if (do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005198 io_put_req(req);
Jens Axboe807abcb2020-07-17 17:09:27 -06005199 kfree(apoll->double_poll);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005200 kfree(apoll);
5201 }
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005202 }
5203
Jens Axboeb41e9852020-02-17 09:52:41 -07005204 if (do_complete) {
5205 io_cqring_fill_event(req, -ECANCELED);
5206 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005207 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005208 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005209 }
5210
5211 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005212}
5213
Jens Axboe76e1b642020-09-26 15:05:03 -06005214/*
5215 * Returns true if we found and killed one or more poll requests
5216 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005217static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
5218 struct files_struct *files)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005219{
Jens Axboe78076bb2019-12-04 19:56:40 -07005220 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005221 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005222 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005223
5224 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005225 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5226 struct hlist_head *list;
5227
5228 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005229 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00005230 if (io_match_task(req, tsk, files))
Jens Axboef3606e32020-09-22 08:18:24 -06005231 posted += io_poll_remove_one(req);
5232 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005233 }
5234 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005235
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005236 if (posted)
5237 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005238
5239 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005240}
5241
Jens Axboe47f46762019-11-09 17:43:02 -07005242static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
5243{
Jens Axboe78076bb2019-12-04 19:56:40 -07005244 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005245 struct io_kiocb *req;
5246
Jens Axboe78076bb2019-12-04 19:56:40 -07005247 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5248 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005249 if (sqe_addr != req->user_data)
5250 continue;
5251 if (io_poll_remove_one(req))
Jens Axboeeac406c2019-11-14 12:09:58 -07005252 return 0;
Jens Axboeb41e9852020-02-17 09:52:41 -07005253 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005254 }
5255
5256 return -ENOENT;
5257}
5258
Jens Axboe3529d8c2019-12-19 18:24:38 -07005259static int io_poll_remove_prep(struct io_kiocb *req,
5260 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005261{
Jens Axboe221c5eb2019-01-17 09:41:58 -07005262 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5263 return -EINVAL;
5264 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
5265 sqe->poll_events)
5266 return -EINVAL;
5267
Pavel Begunkov018043b2020-10-27 23:17:18 +00005268 req->poll_remove.addr = READ_ONCE(sqe->addr);
Jens Axboe0969e782019-12-17 18:40:57 -07005269 return 0;
5270}
5271
5272/*
5273 * Find a running poll command that matches one specified in sqe->addr,
5274 * and remove it if found.
5275 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005276static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005277{
5278 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe0969e782019-12-17 18:40:57 -07005279 int ret;
5280
Jens Axboe221c5eb2019-01-17 09:41:58 -07005281 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov018043b2020-10-27 23:17:18 +00005282 ret = io_poll_cancel(ctx, req->poll_remove.addr);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005283 spin_unlock_irq(&ctx->completion_lock);
5284
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005285 if (ret < 0)
5286 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06005287 io_req_complete(req, ret);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005288 return 0;
5289}
5290
Jens Axboe221c5eb2019-01-17 09:41:58 -07005291static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5292 void *key)
5293{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005294 struct io_kiocb *req = wait->private;
5295 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005296
Jens Axboed7718a92020-02-14 22:23:12 -07005297 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005298}
5299
Jens Axboe221c5eb2019-01-17 09:41:58 -07005300static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5301 struct poll_table_struct *p)
5302{
5303 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5304
Jens Axboee8c2bc12020-08-15 18:44:09 -07005305 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c2019-11-14 12:09:58 -07005306}
5307
Jens Axboe3529d8c2019-12-19 18:24:38 -07005308static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005309{
5310 struct io_poll_iocb *poll = &req->poll;
Jiufei Xue5769a352020-06-17 17:53:55 +08005311 u32 events;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005312
5313 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5314 return -EINVAL;
5315 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
5316 return -EINVAL;
5317
Jiufei Xue5769a352020-06-17 17:53:55 +08005318 events = READ_ONCE(sqe->poll32_events);
5319#ifdef __BIG_ENDIAN
5320 events = swahw32(events);
5321#endif
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005322 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP |
5323 (events & EPOLLEXCLUSIVE);
Jens Axboe0969e782019-12-17 18:40:57 -07005324 return 0;
5325}
5326
Pavel Begunkov61e98202021-02-10 00:03:08 +00005327static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005328{
5329 struct io_poll_iocb *poll = &req->poll;
5330 struct io_ring_ctx *ctx = req->ctx;
5331 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005332 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005333
Jens Axboed7718a92020-02-14 22:23:12 -07005334 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005335
Jens Axboed7718a92020-02-14 22:23:12 -07005336 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5337 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005338
Jens Axboe8c838782019-03-12 15:48:16 -06005339 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005340 ipt.error = 0;
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005341 io_poll_complete(req, mask, 0);
Jens Axboe8c838782019-03-12 15:48:16 -06005342 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005343 spin_unlock_irq(&ctx->completion_lock);
5344
Jens Axboe8c838782019-03-12 15:48:16 -06005345 if (mask) {
5346 io_cqring_ev_posted(ctx);
Pavel Begunkov014db002020-03-03 21:33:12 +03005347 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005348 }
Jens Axboe8c838782019-03-12 15:48:16 -06005349 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005350}
5351
Jens Axboe5262f562019-09-17 12:26:57 -06005352static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5353{
Jens Axboead8a48a2019-11-15 08:49:11 -07005354 struct io_timeout_data *data = container_of(timer,
5355 struct io_timeout_data, timer);
5356 struct io_kiocb *req = data->req;
5357 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005358 unsigned long flags;
5359
Jens Axboe5262f562019-09-17 12:26:57 -06005360 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005361 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005362 atomic_set(&req->ctx->cq_timeouts,
5363 atomic_read(&req->ctx->cq_timeouts) + 1);
5364
Jens Axboe78e19bb2019-11-06 15:21:34 -07005365 io_cqring_fill_event(req, -ETIME);
Jens Axboe5262f562019-09-17 12:26:57 -06005366 io_commit_cqring(ctx);
5367 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5368
5369 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005370 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005371 io_put_req(req);
5372 return HRTIMER_NORESTART;
5373}
5374
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005375static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5376 __u64 user_data)
Jens Axboe47f46762019-11-09 17:43:02 -07005377{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005378 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005379 struct io_kiocb *req;
5380 int ret = -ENOENT;
5381
5382 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
5383 if (user_data == req->user_data) {
5384 ret = 0;
5385 break;
5386 }
5387 }
5388
5389 if (ret == -ENOENT)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005390 return ERR_PTR(ret);
Jens Axboef254ac02020-08-12 17:33:30 -06005391
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005392 io = req->async_data;
5393 ret = hrtimer_try_to_cancel(&io->timer);
5394 if (ret == -1)
5395 return ERR_PTR(-EALREADY);
5396 list_del_init(&req->timeout.list);
5397 return req;
5398}
5399
5400static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
5401{
5402 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5403
5404 if (IS_ERR(req))
5405 return PTR_ERR(req);
5406
5407 req_set_fail_links(req);
5408 io_cqring_fill_event(req, -ECANCELED);
5409 io_put_req_deferred(req, 1);
5410 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005411}
5412
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005413static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5414 struct timespec64 *ts, enum hrtimer_mode mode)
5415{
5416 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5417 struct io_timeout_data *data;
5418
5419 if (IS_ERR(req))
5420 return PTR_ERR(req);
5421
5422 req->timeout.off = 0; /* noseq */
5423 data = req->async_data;
5424 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5425 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5426 data->timer.function = io_timeout_fn;
5427 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5428 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005429}
5430
Jens Axboe3529d8c2019-12-19 18:24:38 -07005431static int io_timeout_remove_prep(struct io_kiocb *req,
5432 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005433{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005434 struct io_timeout_rem *tr = &req->timeout_rem;
5435
Jens Axboeb29472e2019-12-17 18:50:29 -07005436 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5437 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005438 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5439 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005440 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005441 return -EINVAL;
5442
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005443 tr->addr = READ_ONCE(sqe->addr);
5444 tr->flags = READ_ONCE(sqe->timeout_flags);
5445 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5446 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5447 return -EINVAL;
5448 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5449 return -EFAULT;
5450 } else if (tr->flags) {
5451 /* timeout removal doesn't support flags */
5452 return -EINVAL;
5453 }
5454
Jens Axboeb29472e2019-12-17 18:50:29 -07005455 return 0;
5456}
5457
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005458static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5459{
5460 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5461 : HRTIMER_MODE_REL;
5462}
5463
Jens Axboe11365042019-10-16 09:08:32 -06005464/*
5465 * Remove or update an existing timeout command
5466 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005467static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005468{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005469 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005470 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005471 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005472
Jens Axboe11365042019-10-16 09:08:32 -06005473 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005474 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005475 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005476 else
5477 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5478 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005479
Jens Axboe47f46762019-11-09 17:43:02 -07005480 io_cqring_fill_event(req, ret);
Jens Axboe11365042019-10-16 09:08:32 -06005481 io_commit_cqring(ctx);
5482 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005483 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005484 if (ret < 0)
5485 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005486 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005487 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005488}
5489
Jens Axboe3529d8c2019-12-19 18:24:38 -07005490static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005491 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005492{
Jens Axboead8a48a2019-11-15 08:49:11 -07005493 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005494 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005495 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005496
Jens Axboead8a48a2019-11-15 08:49:11 -07005497 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005498 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005499 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005500 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005501 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005502 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005503 flags = READ_ONCE(sqe->timeout_flags);
5504 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005505 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005506
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005507 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005508
Jens Axboee8c2bc12020-08-15 18:44:09 -07005509 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005510 return -ENOMEM;
5511
Jens Axboee8c2bc12020-08-15 18:44:09 -07005512 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005513 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005514
5515 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005516 return -EFAULT;
5517
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005518 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005519 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
5520 return 0;
5521}
5522
Pavel Begunkov61e98202021-02-10 00:03:08 +00005523static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005524{
Jens Axboead8a48a2019-11-15 08:49:11 -07005525 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005526 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005527 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005528 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005529
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005530 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005531
Jens Axboe5262f562019-09-17 12:26:57 -06005532 /*
5533 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005534 * timeout event to be satisfied. If it isn't set, then this is
5535 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005536 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005537 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005538 entry = ctx->timeout_list.prev;
5539 goto add;
5540 }
Jens Axboe5262f562019-09-17 12:26:57 -06005541
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005542 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5543 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005544
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005545 /* Update the last seq here in case io_flush_timeouts() hasn't.
5546 * This is safe because ->completion_lock is held, and submissions
5547 * and completions are never mixed in the same ->completion_lock section.
5548 */
5549 ctx->cq_last_tm_flush = tail;
5550
Jens Axboe5262f562019-09-17 12:26:57 -06005551 /*
5552 * Insertion sort, ensuring the first entry in the list is always
5553 * the one we need first.
5554 */
Jens Axboe5262f562019-09-17 12:26:57 -06005555 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005556 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5557 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005558
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005559 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005560 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005561 /* nxt.seq is behind @tail, otherwise would've been completed */
5562 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005563 break;
5564 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005565add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005566 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005567 data->timer.function = io_timeout_fn;
5568 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005569 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005570 return 0;
5571}
5572
Jens Axboe62755e32019-10-28 21:49:21 -06005573static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005574{
Jens Axboe62755e32019-10-28 21:49:21 -06005575 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Jens Axboede0617e2019-04-06 21:51:27 -06005576
Jens Axboe62755e32019-10-28 21:49:21 -06005577 return req->user_data == (unsigned long) data;
5578}
5579
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005580static int io_async_cancel_one(struct io_uring_task *tctx, void *sqe_addr)
Jens Axboe62755e32019-10-28 21:49:21 -06005581{
Jens Axboe62755e32019-10-28 21:49:21 -06005582 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005583 int ret = 0;
5584
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005585 if (!tctx->io_wq)
5586 return -ENOENT;
5587
5588 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, sqe_addr, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005589 switch (cancel_ret) {
5590 case IO_WQ_CANCEL_OK:
5591 ret = 0;
5592 break;
5593 case IO_WQ_CANCEL_RUNNING:
5594 ret = -EALREADY;
5595 break;
5596 case IO_WQ_CANCEL_NOTFOUND:
5597 ret = -ENOENT;
5598 break;
5599 }
5600
Jens Axboee977d6d2019-11-05 12:39:45 -07005601 return ret;
5602}
5603
Jens Axboe47f46762019-11-09 17:43:02 -07005604static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5605 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005606 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005607{
5608 unsigned long flags;
5609 int ret;
5610
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005611 ret = io_async_cancel_one(req->task->io_uring,
5612 (void *) (unsigned long) sqe_addr);
Jens Axboe47f46762019-11-09 17:43:02 -07005613 if (ret != -ENOENT) {
5614 spin_lock_irqsave(&ctx->completion_lock, flags);
5615 goto done;
5616 }
5617
5618 spin_lock_irqsave(&ctx->completion_lock, flags);
5619 ret = io_timeout_cancel(ctx, sqe_addr);
5620 if (ret != -ENOENT)
5621 goto done;
5622 ret = io_poll_cancel(ctx, sqe_addr);
5623done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005624 if (!ret)
5625 ret = success_ret;
Jens Axboe47f46762019-11-09 17:43:02 -07005626 io_cqring_fill_event(req, ret);
5627 io_commit_cqring(ctx);
5628 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5629 io_cqring_ev_posted(ctx);
5630
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005631 if (ret < 0)
5632 req_set_fail_links(req);
Pavel Begunkov014db002020-03-03 21:33:12 +03005633 io_put_req(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005634}
5635
Jens Axboe3529d8c2019-12-19 18:24:38 -07005636static int io_async_cancel_prep(struct io_kiocb *req,
5637 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005638{
Jens Axboefbf23842019-12-17 18:45:56 -07005639 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005640 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005641 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5642 return -EINVAL;
5643 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005644 return -EINVAL;
5645
Jens Axboefbf23842019-12-17 18:45:56 -07005646 req->cancel.addr = READ_ONCE(sqe->addr);
5647 return 0;
5648}
5649
Pavel Begunkov61e98202021-02-10 00:03:08 +00005650static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07005651{
5652 struct io_ring_ctx *ctx = req->ctx;
Jens Axboefbf23842019-12-17 18:45:56 -07005653
Pavel Begunkov014db002020-03-03 21:33:12 +03005654 io_async_find_and_cancel(ctx, req, req->cancel.addr, 0);
Jens Axboe62755e32019-10-28 21:49:21 -06005655 return 0;
5656}
5657
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005658static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07005659 const struct io_uring_sqe *sqe)
5660{
Jens Axboe6ca56f82020-09-18 16:51:19 -06005661 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
5662 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005663 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5664 return -EINVAL;
5665 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005666 return -EINVAL;
5667
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005668 req->rsrc_update.offset = READ_ONCE(sqe->off);
5669 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
5670 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005671 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005672 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005673 return 0;
5674}
5675
Pavel Begunkov889fca72021-02-10 00:03:09 +00005676static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005677{
5678 struct io_ring_ctx *ctx = req->ctx;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005679 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005680 int ret;
5681
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005682 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005683 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005684
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005685 up.offset = req->rsrc_update.offset;
5686 up.data = req->rsrc_update.arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005687
5688 mutex_lock(&ctx->uring_lock);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005689 ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005690 mutex_unlock(&ctx->uring_lock);
5691
5692 if (ret < 0)
5693 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005694 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005695 return 0;
5696}
5697
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005698static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005699{
Jens Axboed625c6e2019-12-17 19:53:05 -07005700 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005701 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005702 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005703 case IORING_OP_READV:
5704 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005705 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005706 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005707 case IORING_OP_WRITEV:
5708 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005709 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005710 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005711 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005712 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005713 case IORING_OP_POLL_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005714 return io_poll_remove_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005715 case IORING_OP_FSYNC:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005716 return io_fsync_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005717 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005718 return io_sfr_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005719 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005720 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005721 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005722 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005723 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005724 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005725 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005726 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005727 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005728 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005729 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005730 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005731 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005732 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005733 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005734 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005735 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005736 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005737 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005738 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005739 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005740 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005741 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005742 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005743 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005744 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005745 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005746 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005747 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005748 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005749 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005750 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005751 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005752 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005753 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005754 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005755 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005756 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005757 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005758 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005759 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005760 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005761 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005762 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06005763 case IORING_OP_SHUTDOWN:
5764 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06005765 case IORING_OP_RENAMEAT:
5766 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06005767 case IORING_OP_UNLINKAT:
5768 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005769 }
5770
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005771 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5772 req->opcode);
5773 return-EINVAL;
5774}
5775
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005776static int io_req_prep_async(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005777{
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005778 switch (req->opcode) {
5779 case IORING_OP_READV:
5780 case IORING_OP_READ_FIXED:
5781 case IORING_OP_READ:
5782 return io_rw_prep_async(req, READ);
5783 case IORING_OP_WRITEV:
5784 case IORING_OP_WRITE_FIXED:
5785 case IORING_OP_WRITE:
5786 return io_rw_prep_async(req, WRITE);
5787 case IORING_OP_SENDMSG:
5788 case IORING_OP_SEND:
5789 return io_sendmsg_prep_async(req);
5790 case IORING_OP_RECVMSG:
5791 case IORING_OP_RECV:
5792 return io_recvmsg_prep_async(req);
5793 case IORING_OP_CONNECT:
5794 return io_connect_prep_async(req);
5795 }
5796 return 0;
5797}
5798
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005799static int io_req_defer_prep(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005800{
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005801 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005802 return 0;
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005803 /* some opcodes init it during the inital prep */
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005804 if (req->async_data)
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005805 return 0;
5806 if (__io_alloc_async_data(req))
Jens Axboeb76da702019-11-20 13:05:32 -07005807 return -EAGAIN;
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005808 return io_req_prep_async(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005809}
5810
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005811static u32 io_get_sequence(struct io_kiocb *req)
5812{
5813 struct io_kiocb *pos;
5814 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005815 u32 total_submitted, nr_reqs = 0;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005816
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005817 io_for_each_link(pos, req)
5818 nr_reqs++;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005819
5820 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
5821 return total_submitted - nr_reqs;
5822}
5823
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005824static int io_req_defer(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005825{
5826 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005827 struct io_defer_entry *de;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005828 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005829 u32 seq;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005830
5831 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005832 if (likely(list_empty_careful(&ctx->defer_list) &&
5833 !(req->flags & REQ_F_IO_DRAIN)))
5834 return 0;
5835
5836 seq = io_get_sequence(req);
5837 /* Still a chance to pass the sequence check */
5838 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboe2b188cc2019-01-07 10:46:33 -07005839 return 0;
5840
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005841 ret = io_req_defer_prep(req);
5842 if (ret)
5843 return ret;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03005844 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005845 de = kmalloc(sizeof(*de), GFP_KERNEL);
5846 if (!de)
5847 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07005848
5849 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005850 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07005851 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005852 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03005853 io_queue_async_work(req);
5854 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07005855 }
5856
5857 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005858 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005859 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005860 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07005861 spin_unlock_irq(&ctx->completion_lock);
5862 return -EIOCBQUEUED;
5863}
Jens Axboeedafcce2019-01-09 09:16:05 -07005864
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03005865static void __io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005866{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005867 if (req->flags & REQ_F_BUFFER_SELECTED) {
5868 switch (req->opcode) {
5869 case IORING_OP_READV:
5870 case IORING_OP_READ_FIXED:
5871 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07005872 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005873 break;
5874 case IORING_OP_RECVMSG:
5875 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07005876 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005877 break;
5878 }
5879 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005880 }
5881
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005882 if (req->flags & REQ_F_NEED_CLEANUP) {
5883 switch (req->opcode) {
5884 case IORING_OP_READV:
5885 case IORING_OP_READ_FIXED:
5886 case IORING_OP_READ:
5887 case IORING_OP_WRITEV:
5888 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005889 case IORING_OP_WRITE: {
5890 struct io_async_rw *io = req->async_data;
5891 if (io->free_iovec)
5892 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005893 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005894 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005895 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005896 case IORING_OP_SENDMSG: {
5897 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00005898
5899 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005900 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005901 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005902 case IORING_OP_SPLICE:
5903 case IORING_OP_TEE:
5904 io_put_file(req, req->splice.file_in,
5905 (req->splice.flags & SPLICE_F_FD_IN_FIXED));
5906 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06005907 case IORING_OP_OPENAT:
5908 case IORING_OP_OPENAT2:
5909 if (req->open.filename)
5910 putname(req->open.filename);
5911 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06005912 case IORING_OP_RENAMEAT:
5913 putname(req->rename.oldpath);
5914 putname(req->rename.newpath);
5915 break;
Jens Axboe14a11432020-09-28 14:27:37 -06005916 case IORING_OP_UNLINKAT:
5917 putname(req->unlink.filename);
5918 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005919 }
5920 req->flags &= ~REQ_F_NEED_CLEANUP;
5921 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005922}
5923
Pavel Begunkov889fca72021-02-10 00:03:09 +00005924static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07005925{
Jens Axboeedafcce2019-01-09 09:16:05 -07005926 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5730b272021-02-27 15:57:30 -07005927 const struct cred *creds = NULL;
Jens Axboed625c6e2019-12-17 19:53:05 -07005928 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07005929
Jens Axboe5730b272021-02-27 15:57:30 -07005930 if (req->work.personality) {
5931 const struct cred *new_creds;
5932
5933 if (!(issue_flags & IO_URING_F_NONBLOCK))
5934 mutex_lock(&ctx->uring_lock);
5935 new_creds = idr_find(&ctx->personality_idr, req->work.personality);
5936 if (!(issue_flags & IO_URING_F_NONBLOCK))
5937 mutex_unlock(&ctx->uring_lock);
5938 if (!new_creds)
5939 return -EINVAL;
5940 creds = override_creds(new_creds);
5941 }
5942
Jens Axboed625c6e2019-12-17 19:53:05 -07005943 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07005944 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005945 ret = io_nop(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07005946 break;
5947 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07005948 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005949 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005950 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005951 break;
5952 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07005953 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005954 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005955 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005956 break;
5957 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005958 ret = io_fsync(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005959 break;
5960 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005961 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005962 break;
5963 case IORING_OP_POLL_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005964 ret = io_poll_remove(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005965 break;
5966 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005967 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005968 break;
5969 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005970 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005971 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005972 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005973 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005974 break;
5975 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005976 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005977 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005978 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005979 ret = io_recv(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005980 break;
5981 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005982 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005983 break;
5984 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005985 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005986 break;
5987 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005988 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005989 break;
5990 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005991 ret = io_connect(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005992 break;
5993 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005994 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005995 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07005996 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005997 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07005998 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07005999 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006000 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006001 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07006002 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006003 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07006004 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006005 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006006 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006007 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006008 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006009 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006010 break;
Jens Axboe4840e412019-12-25 22:03:45 -07006011 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006012 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006013 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006014 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006015 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006016 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006017 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006018 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006019 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006020 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006021 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006022 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006023 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006024 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006025 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006026 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006027 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006028 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006029 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006030 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006031 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006032 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006033 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006034 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006035 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006036 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006037 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006038 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006039 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006040 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006041 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006042 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006043 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006044 default:
6045 ret = -EINVAL;
6046 break;
Jens Axboe31b51512019-01-18 22:56:34 -07006047 }
6048
Jens Axboe5730b272021-02-27 15:57:30 -07006049 if (creds)
6050 revert_creds(creds);
6051
Jens Axboe2b188cc2019-01-07 10:46:33 -07006052 if (ret)
6053 return ret;
6054
Jens Axboeb5325762020-05-19 21:20:27 -06006055 /* If the op doesn't have a file, we're not polling for it */
6056 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006057 const bool in_async = io_wq_current_is_worker();
6058
Jens Axboe11ba8202020-01-15 21:51:17 -07006059 /* workqueue context doesn't hold uring_lock, grab it now */
6060 if (in_async)
6061 mutex_lock(&ctx->uring_lock);
6062
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08006063 io_iopoll_req_issued(req, in_async);
Jens Axboe11ba8202020-01-15 21:51:17 -07006064
6065 if (in_async)
6066 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006067 }
6068
6069 return 0;
6070}
6071
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006072static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006073{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006074 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006075 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006076 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006077
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006078 timeout = io_prep_linked_timeout(req);
6079 if (timeout)
6080 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006081
Jens Axboe4014d942021-01-19 15:53:54 -07006082 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006083 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006084
Jens Axboe561fb042019-10-24 07:25:42 -06006085 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006086 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006087 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006088 /*
6089 * We can get EAGAIN for polled IO even though we're
6090 * forcing a sync submission from here, since we can't
6091 * wait for request slots on the block side.
6092 */
6093 if (ret != -EAGAIN)
6094 break;
6095 cond_resched();
6096 } while (1);
6097 }
Jens Axboe31b51512019-01-18 22:56:34 -07006098
Pavel Begunkova3df76982021-02-18 22:32:52 +00006099 /* avoid locking problems by failing it from a clean context */
Jens Axboe561fb042019-10-24 07:25:42 -06006100 if (ret) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00006101 /* io-wq is going to take one down */
6102 refcount_inc(&req->refs);
6103 io_req_task_queue_fail(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006104 }
Jens Axboe31b51512019-01-18 22:56:34 -07006105}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006106
Jens Axboe65e19f52019-10-26 07:20:21 -06006107static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6108 int index)
Jens Axboe09bb8392019-03-13 12:39:28 -06006109{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006110 struct fixed_rsrc_table *table;
Jens Axboe65e19f52019-10-26 07:20:21 -06006111
Jens Axboe05f3fb32019-12-09 11:22:50 -07006112 table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT];
Xiaoming Ni84695082020-05-11 19:25:43 +08006113 return table->files[index & IORING_FILE_TABLE_MASK];
Jens Axboe65e19f52019-10-26 07:20:21 -06006114}
6115
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006116static struct file *io_file_get(struct io_submit_state *state,
6117 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006118{
6119 struct io_ring_ctx *ctx = req->ctx;
6120 struct file *file;
6121
6122 if (fixed) {
Pavel Begunkov479f5172020-10-10 18:34:07 +01006123 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006124 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006125 fd = array_index_nospec(fd, ctx->nr_user_files);
6126 file = io_file_from_index(ctx, fd);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00006127 io_set_resource_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006128 } else {
6129 trace_io_uring_file_get(ctx, fd);
6130 file = __io_file_get(state, fd);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006131 }
6132
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00006133 if (file && unlikely(file->f_op == &io_uring_fops))
6134 io_req_track_inflight(req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006135 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006136}
6137
Jens Axboe2665abf2019-11-05 12:40:47 -07006138static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6139{
Jens Axboead8a48a2019-11-15 08:49:11 -07006140 struct io_timeout_data *data = container_of(timer,
6141 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006142 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006143 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006144 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006145
6146 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006147 prev = req->timeout.head;
6148 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006149
6150 /*
6151 * We don't expect the list to be empty, that will only happen if we
6152 * race with the completion of the linked work.
6153 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006154 if (prev && refcount_inc_not_zero(&prev->refs))
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006155 io_remove_next_linked(prev);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006156 else
6157 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006158 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6159
6160 if (prev) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006161 req_set_fail_links(prev);
Pavel Begunkov014db002020-03-03 21:33:12 +03006162 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006163 io_put_req_deferred(prev, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006164 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006165 io_req_complete_post(req, -ETIME, 0);
6166 io_put_req_deferred(req, 1);
Jens Axboe2665abf2019-11-05 12:40:47 -07006167 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006168 return HRTIMER_NORESTART;
6169}
6170
Jens Axboe7271ef32020-08-10 09:55:22 -06006171static void __io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006172{
Jens Axboe76a46e02019-11-10 23:34:16 -07006173 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006174 * If the back reference is NULL, then our linked request finished
6175 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006176 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006177 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006178 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006179
Jens Axboead8a48a2019-11-15 08:49:11 -07006180 data->timer.function = io_link_timeout_fn;
6181 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6182 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006183 }
Jens Axboe7271ef32020-08-10 09:55:22 -06006184}
6185
6186static void io_queue_linked_timeout(struct io_kiocb *req)
6187{
6188 struct io_ring_ctx *ctx = req->ctx;
6189
6190 spin_lock_irq(&ctx->completion_lock);
6191 __io_queue_linked_timeout(req);
Jens Axboe76a46e02019-11-10 23:34:16 -07006192 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006193
Jens Axboe2665abf2019-11-05 12:40:47 -07006194 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006195 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006196}
6197
Jens Axboead8a48a2019-11-15 08:49:11 -07006198static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006199{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006200 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006201
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006202 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6203 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006204 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006205
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006206 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006207 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006208 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006209 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006210}
6211
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006212static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006213{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006214 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006215 int ret;
6216
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006217 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006218
6219 /*
6220 * We async punt it if the file wasn't marked NOWAIT, or if the file
6221 * doesn't support non-blocking read/write attempts
6222 */
Pavel Begunkov24c74672020-06-21 13:09:51 +03006223 if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006224 if (!io_arm_poll_handler(req)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006225 /*
6226 * Queued up for async execution, worker will release
6227 * submit reference when the iocb is actually submitted.
6228 */
6229 io_queue_async_work(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006230 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006231 } else if (likely(!ret)) {
6232 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006233 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006234 struct io_ring_ctx *ctx = req->ctx;
6235 struct io_comp_state *cs = &ctx->submit_state.comp;
Jens Axboee65ef562019-03-12 10:16:44 -06006236
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006237 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006238 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006239 io_submit_flush_completions(cs, ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006240 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006241 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006242 }
6243 } else {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006244 req_set_fail_links(req);
Jens Axboee65ef562019-03-12 10:16:44 -06006245 io_put_req(req);
Pavel Begunkov652532a2020-07-03 22:15:07 +03006246 io_req_complete(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006247 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006248 if (linked_timeout)
6249 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006250}
6251
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006252static void io_queue_sqe(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006253{
6254 int ret;
6255
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006256 ret = io_req_defer(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006257 if (ret) {
6258 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006259fail_req:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006260 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06006261 io_put_req(req);
6262 io_req_complete(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006263 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006264 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006265 ret = io_req_defer_prep(req);
6266 if (unlikely(ret))
6267 goto fail_req;
Jens Axboece35a472019-12-17 08:04:44 -07006268 io_queue_async_work(req);
6269 } else {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006270 __io_queue_sqe(req);
Jens Axboece35a472019-12-17 08:04:44 -07006271 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006272}
6273
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006274/*
6275 * Check SQE restrictions (opcode and flags).
6276 *
6277 * Returns 'true' if SQE is allowed, 'false' otherwise.
6278 */
6279static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6280 struct io_kiocb *req,
6281 unsigned int sqe_flags)
6282{
6283 if (!ctx->restricted)
6284 return true;
6285
6286 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6287 return false;
6288
6289 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6290 ctx->restrictions.sqe_flags_required)
6291 return false;
6292
6293 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6294 ctx->restrictions.sqe_flags_required))
6295 return false;
6296
6297 return true;
6298}
6299
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006300static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006301 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006302{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006303 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006304 unsigned int sqe_flags;
Jens Axboe5730b272021-02-27 15:57:30 -07006305 int ret = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006306
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006307 req->opcode = READ_ONCE(sqe->opcode);
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006308 /* same numerical values with corresponding REQ_F_*, safe to copy */
6309 req->flags = sqe_flags = READ_ONCE(sqe->flags);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006310 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006311 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006312 req->file = NULL;
6313 req->ctx = ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006314 req->link = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006315 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006316 /* one is dropped after submission, the other at completion */
6317 refcount_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006318 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006319 req->result = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006320
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006321 /* enforce forwards compatibility on users */
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006322 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
6323 req->flags = 0;
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006324 return -EINVAL;
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006325 }
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006326
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006327 if (unlikely(req->opcode >= IORING_OP_LAST))
6328 return -EINVAL;
6329
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006330 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6331 return -EACCES;
6332
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006333 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6334 !io_op_defs[req->opcode].buffer_select)
6335 return -EOPNOTSUPP;
6336
Jens Axboe5730b272021-02-27 15:57:30 -07006337 req->work.list.next = NULL;
6338 req->work.flags = 0;
6339 req->work.personality = READ_ONCE(sqe->personality);
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006340 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006341
Jens Axboe27926b62020-10-28 09:33:23 -06006342 /*
6343 * Plug now if we have more than 1 IO left after this, and the target
6344 * is potentially a read/write to block based storage.
6345 */
6346 if (!state->plug_started && state->ios_left > 1 &&
6347 io_op_defs[req->opcode].plug) {
6348 blk_start_plug(&state->plug);
6349 state->plug_started = true;
6350 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006351
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006352 if (io_op_defs[req->opcode].needs_file) {
6353 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006354
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006355 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006356 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006357 ret = -EBADF;
6358 }
6359
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006360 state->ios_left--;
6361 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006362}
6363
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006364static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006365 const struct io_uring_sqe *sqe)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006366{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006367 struct io_submit_link *link = &ctx->submit_state.link;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006368 int ret;
6369
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006370 ret = io_init_req(ctx, req, sqe);
6371 if (unlikely(ret)) {
6372fail_req:
6373 io_put_req(req);
6374 io_req_complete(req, ret);
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006375 if (link->head) {
6376 /* fail even hard links since we don't submit */
Pavel Begunkovcf109602021-02-18 18:29:43 +00006377 link->head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006378 io_put_req(link->head);
6379 io_req_complete(link->head, -ECANCELED);
6380 link->head = NULL;
6381 }
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006382 return ret;
6383 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006384 ret = io_req_prep(req, sqe);
6385 if (unlikely(ret))
6386 goto fail_req;
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006387
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006388 /* don't need @sqe from now on */
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006389 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
6390 true, ctx->flags & IORING_SETUP_SQPOLL);
6391
Jens Axboe6c271ce2019-01-10 11:22:30 -07006392 /*
6393 * If we already have a head request, queue this one for async
6394 * submittal once the head completes. If we don't have a head but
6395 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6396 * submitted sync once the chain is complete. If none of those
6397 * conditions are true (normal request), then just queue it.
6398 */
6399 if (link->head) {
6400 struct io_kiocb *head = link->head;
6401
6402 /*
6403 * Taking sequential execution of a link, draining both sides
6404 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6405 * requests in the link. So, it drains the head and the
6406 * next after the link request. The last one is done via
6407 * drain_next flag to persist the effect across calls.
6408 */
6409 if (req->flags & REQ_F_IO_DRAIN) {
6410 head->flags |= REQ_F_IO_DRAIN;
6411 ctx->drain_next = 1;
6412 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006413 ret = io_req_defer_prep(req);
Pavel Begunkovcf109602021-02-18 18:29:43 +00006414 if (unlikely(ret))
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006415 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006416 trace_io_uring_link(ctx, req, head);
6417 link->last->link = req;
6418 link->last = req;
6419
6420 /* last request of a link, enqueue the link */
6421 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006422 io_queue_sqe(head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006423 link->head = NULL;
6424 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006425 } else {
6426 if (unlikely(ctx->drain_next)) {
6427 req->flags |= REQ_F_IO_DRAIN;
6428 ctx->drain_next = 0;
6429 }
6430 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Jackie Liu4fe2c962019-09-09 20:50:40 +08006431 link->head = req;
6432 link->last = req;
6433 } else {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006434 io_queue_sqe(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006435 }
6436 }
6437
6438 return 0;
6439}
6440
6441/*
6442 * Batched submission is done, ensure local IO is flushed out.
6443 */
6444static void io_submit_state_end(struct io_submit_state *state,
6445 struct io_ring_ctx *ctx)
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006446{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006447 if (state->link.head)
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006448 io_queue_sqe(state->link.head);
Jens Axboe3529d8c2019-12-19 18:24:38 -07006449 if (state->comp.nr)
Jens Axboe9e645e112019-05-10 16:07:28 -06006450 io_submit_flush_completions(&state->comp, ctx);
Jackie Liua197f662019-11-08 08:09:12 -07006451 if (state->plug_started)
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006452 blk_finish_plug(&state->plug);
Jens Axboe75c6a032020-01-28 10:15:23 -07006453 io_state_file_put(state);
Jens Axboe9e645e112019-05-10 16:07:28 -06006454}
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006455
Jens Axboe9e645e112019-05-10 16:07:28 -06006456/*
6457 * Start submission side cache.
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006458 */
Jens Axboe9e645e112019-05-10 16:07:28 -06006459static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkov196be952019-11-07 01:41:06 +03006460 unsigned int max_ios)
Jens Axboe9e645e112019-05-10 16:07:28 -06006461{
6462 state->plug_started = false;
Jens Axboebcda7ba2020-02-23 16:42:51 -07006463 state->ios_left = max_ios;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006464 /* set only head, no need to init link_last in advance */
6465 state->link.head = NULL;
Jens Axboe75c6a032020-01-28 10:15:23 -07006466}
6467
Jens Axboe193155c2020-02-22 23:22:19 -07006468static void io_commit_sqring(struct io_ring_ctx *ctx)
6469{
Jens Axboe75c6a032020-01-28 10:15:23 -07006470 struct io_rings *rings = ctx->rings;
6471
6472 /*
Jens Axboe193155c2020-02-22 23:22:19 -07006473 * Ensure any loads from the SQEs are done at this point,
Jens Axboe75c6a032020-01-28 10:15:23 -07006474 * since once we write the new head, the application could
6475 * write new data to them.
Pavel Begunkov6b47ee62020-01-18 20:22:41 +03006476 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006477 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboebcda7ba2020-02-23 16:42:51 -07006478}
6479
Jens Axboe9e645e112019-05-10 16:07:28 -06006480/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006481 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe9e645e112019-05-10 16:07:28 -06006482 * that is mapped by userspace. This means that care needs to be taken to
6483 * ensure that reads are stable, as we cannot rely on userspace always
Jens Axboe78e19bb2019-11-06 15:21:34 -07006484 * being a good citizen. If members of the sqe are validated and then later
6485 * used, it's important that those reads are done through READ_ONCE() to
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006486 * prevent a re-load down the line.
Jens Axboe9e645e112019-05-10 16:07:28 -06006487 */
6488static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe9e645e112019-05-10 16:07:28 -06006489{
6490 u32 *sq_array = ctx->sq_array;
6491 unsigned head;
6492
6493 /*
6494 * The cached sq head (or cq tail) serves two purposes:
6495 *
6496 * 1) allows us to batch the cost of updating the user visible
Pavel Begunkov9d763772019-12-17 02:22:07 +03006497 * head updates.
Jens Axboe9e645e112019-05-10 16:07:28 -06006498 * 2) allows the kernel side to track the head on its own, even
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006499 * though the application is the one updating it.
6500 */
6501 head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]);
6502 if (likely(head < ctx->sq_entries))
6503 return &ctx->sq_sqes[head];
6504
6505 /* drop invalid entries */
Pavel Begunkov711be032020-01-17 03:57:59 +03006506 ctx->cached_sq_dropped++;
6507 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
6508 return NULL;
6509}
Jens Axboeb7bb4f72019-12-15 22:13:43 -07006510
Jens Axboe0f212202020-09-13 13:09:39 -06006511static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006512{
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006513 int submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006514
Jens Axboec4a2ed72019-11-21 21:01:26 -07006515 /* if we have a backlog and couldn't flush it all, return BUSY */
Jens Axboead3eb2c2019-12-18 17:12:20 -07006516 if (test_bit(0, &ctx->sq_check_overflow)) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006517 if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
Jens Axboead3eb2c2019-12-18 17:12:20 -07006518 return -EBUSY;
6519 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006520
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006521 /* make sure SQ entry isn't read before tail */
6522 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006523
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006524 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6525 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006526
Jens Axboed8a6df12020-10-15 16:24:45 -06006527 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006528 refcount_add(nr, &current->usage);
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006529 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006530
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006531 while (submitted < nr) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006532 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006533 struct io_kiocb *req;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006534
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006535 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006536 if (unlikely(!req)) {
6537 if (!submitted)
6538 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006539 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006540 }
Pavel Begunkov4fccfcb2021-02-12 11:55:17 +00006541 sqe = io_get_sqe(ctx);
6542 if (unlikely(!sqe)) {
6543 kmem_cache_free(req_cachep, req);
6544 break;
6545 }
Jens Axboed3656342019-12-18 09:50:26 -07006546 /* will complete beyond this point, count as submitted */
6547 submitted++;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006548 if (io_submit_sqe(ctx, req, sqe))
Jens Axboed3656342019-12-18 09:50:26 -07006549 break;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006550 }
6551
Pavel Begunkov9466f432020-01-25 22:34:01 +03006552 if (unlikely(submitted != nr)) {
6553 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006554 struct io_uring_task *tctx = current->io_uring;
6555 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006556
Jens Axboed8a6df12020-10-15 16:24:45 -06006557 percpu_ref_put_many(&ctx->refs, unused);
6558 percpu_counter_sub(&tctx->inflight, unused);
6559 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006560 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006561
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006562 io_submit_state_end(&ctx->submit_state, ctx);
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006563 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6564 io_commit_sqring(ctx);
6565
Jens Axboe6c271ce2019-01-10 11:22:30 -07006566 return submitted;
6567}
6568
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006569static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6570{
6571 /* Tell userspace we may need a wakeup call */
6572 spin_lock_irq(&ctx->completion_lock);
6573 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6574 spin_unlock_irq(&ctx->completion_lock);
6575}
6576
6577static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6578{
6579 spin_lock_irq(&ctx->completion_lock);
6580 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6581 spin_unlock_irq(&ctx->completion_lock);
6582}
6583
Xiaoguang Wang08369242020-11-03 14:15:59 +08006584static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006585{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006586 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006587 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006588
Jens Axboec8d1ba52020-09-14 11:07:26 -06006589 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006590 /* if we're handling multiple rings, cap submit size for fairness */
6591 if (cap_entries && to_submit > 8)
6592 to_submit = 8;
6593
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006594 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6595 unsigned nr_events = 0;
6596
Xiaoguang Wang08369242020-11-03 14:15:59 +08006597 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006598 if (!list_empty(&ctx->iopoll_list))
6599 io_do_iopoll(ctx, &nr_events, 0);
6600
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00006601 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)))
Xiaoguang Wang08369242020-11-03 14:15:59 +08006602 ret = io_submit_sqes(ctx, to_submit);
6603 mutex_unlock(&ctx->uring_lock);
6604 }
Jens Axboe90554202020-09-03 12:12:41 -06006605
6606 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
6607 wake_up(&ctx->sqo_sq_wait);
6608
Xiaoguang Wang08369242020-11-03 14:15:59 +08006609 return ret;
6610}
6611
6612static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
6613{
6614 struct io_ring_ctx *ctx;
6615 unsigned sq_thread_idle = 0;
6616
6617 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6618 if (sq_thread_idle < ctx->sq_thread_idle)
6619 sq_thread_idle = ctx->sq_thread_idle;
6620 }
6621
6622 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006623}
6624
Jens Axboe69fb2132020-09-14 11:16:23 -06006625static void io_sqd_init_new(struct io_sq_data *sqd)
6626{
6627 struct io_ring_ctx *ctx;
6628
6629 while (!list_empty(&sqd->ctx_new_list)) {
6630 ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006631 list_move_tail(&ctx->sqd_list, &sqd->ctx_list);
6632 complete(&ctx->sq_thread_comp);
6633 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006634
6635 io_sqd_update_thread_idle(sqd);
Jens Axboe69fb2132020-09-14 11:16:23 -06006636}
6637
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006638static bool io_sq_thread_should_stop(struct io_sq_data *sqd)
6639{
6640 return test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
6641}
6642
6643static bool io_sq_thread_should_park(struct io_sq_data *sqd)
6644{
6645 return test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
6646}
6647
6648static void io_sq_thread_parkme(struct io_sq_data *sqd)
6649{
6650 for (;;) {
6651 /*
6652 * TASK_PARKED is a special state; we must serialize against
6653 * possible pending wakeups to avoid store-store collisions on
6654 * task->state.
6655 *
6656 * Such a collision might possibly result in the task state
6657 * changin from TASK_PARKED and us failing the
6658 * wait_task_inactive() in kthread_park().
6659 */
6660 set_special_state(TASK_PARKED);
6661 if (!test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state))
6662 break;
6663
6664 /*
6665 * Thread is going to call schedule(), do not preempt it,
6666 * or the caller of kthread_park() may spend more time in
6667 * wait_task_inactive().
6668 */
6669 preempt_disable();
6670 complete(&sqd->completion);
6671 schedule_preempt_disabled();
6672 preempt_enable();
6673 }
6674 __set_current_state(TASK_RUNNING);
6675}
6676
Jens Axboe6c271ce2019-01-10 11:22:30 -07006677static int io_sq_thread(void *data)
6678{
Jens Axboe69fb2132020-09-14 11:16:23 -06006679 struct io_sq_data *sqd = data;
6680 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d92052020-11-12 14:55:59 +08006681 unsigned long timeout = 0;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006682 char buf[TASK_COMM_LEN];
Xiaoguang Wang08369242020-11-03 14:15:59 +08006683 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006684
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006685 sprintf(buf, "iou-sqp-%d", sqd->task_pid);
6686 set_task_comm(current, buf);
6687 sqd->thread = current;
6688 current->pf_io_worker = NULL;
Jens Axboe28cea78a2020-09-14 10:51:17 -06006689
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006690 if (sqd->sq_cpu != -1)
6691 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
6692 else
6693 set_cpus_allowed_ptr(current, cpu_online_mask);
6694 current->flags |= PF_NO_SETAFFINITY;
6695
6696 complete(&sqd->completion);
6697
6698 wait_for_completion(&sqd->startup);
6699
6700 while (!io_sq_thread_should_stop(sqd)) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006701 int ret;
6702 bool cap_entries, sqt_spin, needs_sched;
Jens Axboec1edbf52019-11-10 16:56:04 -07006703
6704 /*
Jens Axboe69fb2132020-09-14 11:16:23 -06006705 * Any changes to the sqd lists are synchronized through the
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006706 * thread parking. This synchronizes the thread vs users,
Jens Axboe69fb2132020-09-14 11:16:23 -06006707 * the users are synchronized on the sqd->ctx_lock.
Jens Axboec1edbf52019-11-10 16:56:04 -07006708 */
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006709 if (io_sq_thread_should_park(sqd)) {
6710 io_sq_thread_parkme(sqd);
6711 continue;
Xiaoguang Wang65b2b212020-11-19 17:44:46 +08006712 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006713 if (unlikely(!list_empty(&sqd->ctx_new_list))) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006714 io_sqd_init_new(sqd);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006715 timeout = jiffies + sqd->sq_thread_idle;
6716 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006717 if (fatal_signal_pending(current))
6718 break;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006719 sqt_spin = false;
Jens Axboee95eee22020-09-08 09:11:32 -06006720 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006721 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006722 ret = __io_sq_thread(ctx, cap_entries);
6723 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
6724 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006725 }
6726
Xiaoguang Wang08369242020-11-03 14:15:59 +08006727 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006728 io_run_task_work();
6729 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08006730 if (sqt_spin)
6731 timeout = jiffies + sqd->sq_thread_idle;
6732 continue;
6733 }
6734
Xiaoguang Wang08369242020-11-03 14:15:59 +08006735 needs_sched = true;
6736 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
6737 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6738 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6739 !list_empty_careful(&ctx->iopoll_list)) {
6740 needs_sched = false;
6741 break;
6742 }
6743 if (io_sqring_entries(ctx)) {
6744 needs_sched = false;
6745 break;
6746 }
6747 }
6748
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006749 if (needs_sched && !io_sq_thread_should_park(sqd)) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006750 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6751 io_ring_set_wakeup_flag(ctx);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006752
Jens Axboe69fb2132020-09-14 11:16:23 -06006753 schedule();
Jens Axboe69fb2132020-09-14 11:16:23 -06006754 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6755 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006756 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006757
6758 finish_wait(&sqd->wait, &wait);
6759 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006760 }
6761
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006762 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6763 io_uring_cancel_sqpoll(ctx);
6764
Jens Axboe4c6e2772020-07-01 11:29:10 -06006765 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07006766
Jens Axboe86293972021-02-26 13:46:49 -07006767 if (io_sq_thread_should_park(sqd))
6768 io_sq_thread_parkme(sqd);
6769
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006770 /*
6771 * Clear thread under lock so that concurrent parks work correctly
6772 */
Jens Axboe86293972021-02-26 13:46:49 -07006773 complete(&sqd->completion);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006774 mutex_lock(&sqd->lock);
6775 sqd->thread = NULL;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07006776 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6777 ctx->sqo_exec = 1;
6778 io_ring_set_wakeup_flag(ctx);
6779 }
Jens Axboe06058632019-04-13 09:26:03 -06006780
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006781 complete(&sqd->exited);
Jens Axboee54945a2021-02-26 11:27:15 -07006782 mutex_unlock(&sqd->lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006783 do_exit(0);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006784}
6785
Jens Axboebda52162019-09-24 13:47:15 -06006786struct io_wait_queue {
6787 struct wait_queue_entry wq;
6788 struct io_ring_ctx *ctx;
6789 unsigned to_wait;
6790 unsigned nr_timeouts;
6791};
6792
Pavel Begunkov6c503152021-01-04 20:36:36 +00006793static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06006794{
6795 struct io_ring_ctx *ctx = iowq->ctx;
6796
6797 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006798 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006799 * started waiting. For timeouts, we always want to return to userspace,
6800 * regardless of event count.
6801 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00006802 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006803 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6804}
6805
6806static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6807 int wake_flags, void *key)
6808{
6809 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6810 wq);
6811
Pavel Begunkov6c503152021-01-04 20:36:36 +00006812 /*
6813 * Cannot safely flush overflowed CQEs from here, ensure we wake up
6814 * the task, and the next invocation will do it.
6815 */
6816 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
6817 return autoremove_wake_function(curr, mode, wake_flags, key);
6818 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06006819}
6820
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006821static int io_run_task_work_sig(void)
6822{
6823 if (io_run_task_work())
6824 return 1;
6825 if (!signal_pending(current))
6826 return 0;
Jens Axboe792ee0f62020-10-22 20:17:18 -06006827 if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL))
6828 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006829 return -EINTR;
6830}
6831
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006832/* when returns >0, the caller should retry */
6833static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
6834 struct io_wait_queue *iowq,
6835 signed long *timeout)
6836{
6837 int ret;
6838
6839 /* make sure we run task_work before checking for signals */
6840 ret = io_run_task_work_sig();
6841 if (ret || io_should_wake(iowq))
6842 return ret;
6843 /* let the caller flush overflows, retry */
6844 if (test_bit(0, &ctx->cq_check_overflow))
6845 return 1;
6846
6847 *timeout = schedule_timeout(*timeout);
6848 return !*timeout ? -ETIME : 1;
6849}
6850
Jens Axboe2b188cc2019-01-07 10:46:33 -07006851/*
6852 * Wait until events become available, if we don't already have some. The
6853 * application must reap them itself, as they reside on the shared cq ring.
6854 */
6855static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08006856 const sigset_t __user *sig, size_t sigsz,
6857 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006858{
Jens Axboebda52162019-09-24 13:47:15 -06006859 struct io_wait_queue iowq = {
6860 .wq = {
6861 .private = current,
6862 .func = io_wake_function,
6863 .entry = LIST_HEAD_INIT(iowq.wq.entry),
6864 },
6865 .ctx = ctx,
6866 .to_wait = min_events,
6867 };
Hristo Venev75b28af2019-08-26 17:23:46 +00006868 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006869 signed long timeout = MAX_SCHEDULE_TIMEOUT;
6870 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006871
Jens Axboeb41e9852020-02-17 09:52:41 -07006872 do {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006873 io_cqring_overflow_flush(ctx, false, NULL, NULL);
6874 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07006875 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06006876 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07006877 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07006878 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006879
6880 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006881#ifdef CONFIG_COMPAT
6882 if (in_compat_syscall())
6883 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07006884 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006885 else
6886#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07006887 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006888
Jens Axboe2b188cc2019-01-07 10:46:33 -07006889 if (ret)
6890 return ret;
6891 }
6892
Hao Xuc73ebb62020-11-03 10:54:37 +08006893 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006894 struct timespec64 ts;
6895
Hao Xuc73ebb62020-11-03 10:54:37 +08006896 if (get_timespec64(&ts, uts))
6897 return -EFAULT;
6898 timeout = timespec64_to_jiffies(&ts);
6899 }
6900
Jens Axboebda52162019-09-24 13:47:15 -06006901 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02006902 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06006903 do {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006904 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Jens Axboebda52162019-09-24 13:47:15 -06006905 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
6906 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006907 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
6908 finish_wait(&ctx->wait, &iowq.wq);
6909 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06006910
Jens Axboeb7db41c2020-07-04 08:55:50 -06006911 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006912
Hristo Venev75b28af2019-08-26 17:23:46 +00006913 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006914}
6915
Jens Axboe6b063142019-01-10 22:13:58 -07006916static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
6917{
6918#if defined(CONFIG_UNIX)
6919 if (ctx->ring_sock) {
6920 struct sock *sock = ctx->ring_sock->sk;
6921 struct sk_buff *skb;
6922
6923 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
6924 kfree_skb(skb);
6925 }
6926#else
6927 int i;
6928
Jens Axboe65e19f52019-10-26 07:20:21 -06006929 for (i = 0; i < ctx->nr_user_files; i++) {
6930 struct file *file;
6931
6932 file = io_file_from_index(ctx, i);
6933 if (file)
6934 fput(file);
6935 }
Jens Axboe6b063142019-01-10 22:13:58 -07006936#endif
6937}
6938
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00006939static void io_rsrc_data_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006940{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006941 struct fixed_rsrc_data *data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006942
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006943 data = container_of(ref, struct fixed_rsrc_data, refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006944 complete(&data->done);
6945}
6946
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006947static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
Pavel Begunkov1642b442020-12-30 21:34:14 +00006948{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006949 spin_lock_bh(&ctx->rsrc_ref_lock);
Pavel Begunkov1642b442020-12-30 21:34:14 +00006950}
6951
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006952static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
Jens Axboe6b063142019-01-10 22:13:58 -07006953{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006954 spin_unlock_bh(&ctx->rsrc_ref_lock);
6955}
6956
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00006957static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx,
6958 struct fixed_rsrc_data *rsrc_data,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006959 struct fixed_rsrc_ref_node *ref_node)
Jens Axboe6b063142019-01-10 22:13:58 -07006960{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006961 io_rsrc_ref_lock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006962 rsrc_data->node = ref_node;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00006963 list_add_tail(&ref_node->node, &ctx->rsrc_ref_list);
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006964 io_rsrc_ref_unlock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006965 percpu_ref_get(&rsrc_data->refs);
Jens Axboe6b063142019-01-10 22:13:58 -07006966}
6967
Hao Xu8bad28d2021-02-19 17:19:36 +08006968static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data)
Jens Axboe6b063142019-01-10 22:13:58 -07006969{
Hao Xu8bad28d2021-02-19 17:19:36 +08006970 struct fixed_rsrc_ref_node *ref_node = NULL;
Jens Axboe65e19f52019-10-26 07:20:21 -06006971
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006972 io_rsrc_ref_lock(ctx);
Pavel Begunkov1e5d7702020-11-18 14:56:25 +00006973 ref_node = data->node;
Pavel Begunkove6cb0072021-02-20 18:03:47 +00006974 data->node = NULL;
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006975 io_rsrc_ref_unlock(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08006976 if (ref_node)
6977 percpu_ref_kill(&ref_node->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08006978}
Xiaoguang Wang05589552020-03-31 14:05:18 +08006979
Hao Xu8bad28d2021-02-19 17:19:36 +08006980static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data,
6981 struct io_ring_ctx *ctx,
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006982 void (*rsrc_put)(struct io_ring_ctx *ctx,
6983 struct io_rsrc_put *prsrc))
Hao Xu8bad28d2021-02-19 17:19:36 +08006984{
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006985 struct fixed_rsrc_ref_node *backup_node;
Hao Xu8bad28d2021-02-19 17:19:36 +08006986 int ret;
Xiaoguang Wang05589552020-03-31 14:05:18 +08006987
Hao Xu8bad28d2021-02-19 17:19:36 +08006988 if (data->quiesce)
6989 return -ENXIO;
6990
6991 data->quiesce = true;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00006992 do {
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006993 ret = -ENOMEM;
6994 backup_node = alloc_fixed_rsrc_ref_node(ctx);
6995 if (!backup_node)
6996 break;
6997 backup_node->rsrc_data = data;
6998 backup_node->rsrc_put = rsrc_put;
6999
Hao Xu8bad28d2021-02-19 17:19:36 +08007000 io_sqe_rsrc_kill_node(ctx, data);
7001 percpu_ref_kill(&data->refs);
7002 flush_delayed_work(&ctx->rsrc_put_work);
7003
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007004 ret = wait_for_completion_interruptible(&data->done);
7005 if (!ret)
7006 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007007
Jens Axboecb5e1b82021-02-25 07:37:35 -07007008 percpu_ref_resurrect(&data->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08007009 io_sqe_rsrc_set_node(ctx, data, backup_node);
7010 backup_node = NULL;
Jens Axboecb5e1b82021-02-25 07:37:35 -07007011 reinit_completion(&data->done);
Hao Xu8bad28d2021-02-19 17:19:36 +08007012 mutex_unlock(&ctx->uring_lock);
7013 ret = io_run_task_work_sig();
7014 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007015 } while (ret >= 0);
Hao Xu8bad28d2021-02-19 17:19:36 +08007016 data->quiesce = false;
7017
7018 if (backup_node)
7019 destroy_fixed_rsrc_ref_node(backup_node);
7020 return ret;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007021}
7022
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007023static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx)
7024{
7025 struct fixed_rsrc_data *data;
7026
7027 data = kzalloc(sizeof(*data), GFP_KERNEL);
7028 if (!data)
7029 return NULL;
7030
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007031 if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero,
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007032 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
7033 kfree(data);
7034 return NULL;
7035 }
7036 data->ctx = ctx;
7037 init_completion(&data->done);
7038 return data;
7039}
7040
7041static void free_fixed_rsrc_data(struct fixed_rsrc_data *data)
7042{
7043 percpu_ref_exit(&data->refs);
7044 kfree(data->table);
7045 kfree(data);
7046}
7047
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007048static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7049{
7050 struct fixed_rsrc_data *data = ctx->file_data;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007051 unsigned nr_tables, i;
7052 int ret;
7053
Hao Xu8bad28d2021-02-19 17:19:36 +08007054 /*
7055 * percpu_ref_is_dying() is to stop parallel files unregister
7056 * Since we possibly drop uring lock later in this function to
7057 * run task work.
7058 */
7059 if (!data || percpu_ref_is_dying(&data->refs))
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007060 return -ENXIO;
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007061 ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put);
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007062 if (ret)
7063 return ret;
7064
Jens Axboe6b063142019-01-10 22:13:58 -07007065 __io_sqe_files_unregister(ctx);
Jens Axboe65e19f52019-10-26 07:20:21 -06007066 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
7067 for (i = 0; i < nr_tables; i++)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007068 kfree(data->table[i].files);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007069 free_fixed_rsrc_data(data);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007070 ctx->file_data = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07007071 ctx->nr_user_files = 0;
7072 return 0;
7073}
7074
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007075static void io_sq_thread_unpark(struct io_sq_data *sqd)
7076 __releases(&sqd->lock)
7077{
7078 if (!sqd->thread)
7079 return;
7080 if (sqd->thread == current)
7081 return;
7082 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
7083 wake_up_state(sqd->thread, TASK_PARKED);
7084 mutex_unlock(&sqd->lock);
7085}
7086
7087static bool io_sq_thread_park(struct io_sq_data *sqd)
7088 __acquires(&sqd->lock)
7089{
7090 if (sqd->thread == current)
7091 return true;
7092 mutex_lock(&sqd->lock);
7093 if (!sqd->thread) {
7094 mutex_unlock(&sqd->lock);
7095 return false;
7096 }
7097 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
7098 wake_up_process(sqd->thread);
7099 wait_for_completion(&sqd->completion);
7100 return true;
7101}
7102
7103static void io_sq_thread_stop(struct io_sq_data *sqd)
7104{
Jens Axboee54945a2021-02-26 11:27:15 -07007105 if (test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state))
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007106 return;
Jens Axboee54945a2021-02-26 11:27:15 -07007107 mutex_lock(&sqd->lock);
7108 if (sqd->thread) {
7109 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
7110 WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state));
7111 wake_up_process(sqd->thread);
7112 mutex_unlock(&sqd->lock);
7113 wait_for_completion(&sqd->exited);
7114 WARN_ON_ONCE(sqd->thread);
7115 } else {
7116 mutex_unlock(&sqd->lock);
7117 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007118}
7119
Jens Axboe534ca6d2020-09-02 13:52:19 -06007120static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007121{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007122 if (refcount_dec_and_test(&sqd->refs)) {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007123 io_sq_thread_stop(sqd);
7124 kfree(sqd);
7125 }
7126}
7127
7128static void io_sq_thread_finish(struct io_ring_ctx *ctx)
7129{
7130 struct io_sq_data *sqd = ctx->sq_data;
7131
7132 if (sqd) {
Jens Axboeeb858902021-02-25 10:13:29 -07007133 complete(&sqd->startup);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007134 if (sqd->thread) {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007135 wait_for_completion(&ctx->sq_thread_comp);
7136 io_sq_thread_park(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007137 }
7138
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007139 mutex_lock(&sqd->ctx_lock);
7140 list_del(&ctx->sqd_list);
7141 io_sqd_update_thread_idle(sqd);
7142 mutex_unlock(&sqd->ctx_lock);
7143
7144 if (sqd->thread)
7145 io_sq_thread_unpark(sqd);
7146
7147 io_put_sq_data(sqd);
7148 ctx->sq_data = NULL;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007149 }
7150}
7151
Jens Axboeaa061652020-09-02 14:50:27 -06007152static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7153{
7154 struct io_ring_ctx *ctx_attach;
7155 struct io_sq_data *sqd;
7156 struct fd f;
7157
7158 f = fdget(p->wq_fd);
7159 if (!f.file)
7160 return ERR_PTR(-ENXIO);
7161 if (f.file->f_op != &io_uring_fops) {
7162 fdput(f);
7163 return ERR_PTR(-EINVAL);
7164 }
7165
7166 ctx_attach = f.file->private_data;
7167 sqd = ctx_attach->sq_data;
7168 if (!sqd) {
7169 fdput(f);
7170 return ERR_PTR(-EINVAL);
7171 }
7172
7173 refcount_inc(&sqd->refs);
7174 fdput(f);
7175 return sqd;
7176}
7177
Jens Axboe534ca6d2020-09-02 13:52:19 -06007178static struct io_sq_data *io_get_sq_data(struct io_uring_params *p)
7179{
7180 struct io_sq_data *sqd;
7181
Jens Axboeaa061652020-09-02 14:50:27 -06007182 if (p->flags & IORING_SETUP_ATTACH_WQ)
7183 return io_attach_sq_data(p);
7184
Jens Axboe534ca6d2020-09-02 13:52:19 -06007185 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7186 if (!sqd)
7187 return ERR_PTR(-ENOMEM);
7188
7189 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007190 INIT_LIST_HEAD(&sqd->ctx_list);
7191 INIT_LIST_HEAD(&sqd->ctx_new_list);
7192 mutex_init(&sqd->ctx_lock);
7193 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007194 init_waitqueue_head(&sqd->wait);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007195 init_completion(&sqd->startup);
7196 init_completion(&sqd->completion);
7197 init_completion(&sqd->exited);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007198 return sqd;
7199}
7200
Jens Axboe6b063142019-01-10 22:13:58 -07007201#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007202/*
7203 * Ensure the UNIX gc is aware of our file set, so we are certain that
7204 * the io_uring can be safely unregistered on process exit, even if we have
7205 * loops in the file referencing.
7206 */
7207static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7208{
7209 struct sock *sk = ctx->ring_sock->sk;
7210 struct scm_fp_list *fpl;
7211 struct sk_buff *skb;
Jens Axboe08a45172019-10-03 08:11:03 -06007212 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007213
Jens Axboe6b063142019-01-10 22:13:58 -07007214 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7215 if (!fpl)
7216 return -ENOMEM;
7217
7218 skb = alloc_skb(0, GFP_KERNEL);
7219 if (!skb) {
7220 kfree(fpl);
7221 return -ENOMEM;
7222 }
7223
7224 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007225
Jens Axboe08a45172019-10-03 08:11:03 -06007226 nr_files = 0;
Jens Axboe62e398b2021-02-21 16:19:37 -07007227 fpl->user = get_uid(current_user());
Jens Axboe6b063142019-01-10 22:13:58 -07007228 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007229 struct file *file = io_file_from_index(ctx, i + offset);
7230
7231 if (!file)
Jens Axboe08a45172019-10-03 08:11:03 -06007232 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007233 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a45172019-10-03 08:11:03 -06007234 unix_inflight(fpl->user, fpl->fp[nr_files]);
7235 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007236 }
7237
Jens Axboe08a45172019-10-03 08:11:03 -06007238 if (nr_files) {
7239 fpl->max = SCM_MAX_FD;
7240 fpl->count = nr_files;
7241 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007242 skb->destructor = unix_destruct_scm;
Jens Axboe08a45172019-10-03 08:11:03 -06007243 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7244 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007245
Jens Axboe08a45172019-10-03 08:11:03 -06007246 for (i = 0; i < nr_files; i++)
7247 fput(fpl->fp[i]);
7248 } else {
7249 kfree_skb(skb);
7250 kfree(fpl);
7251 }
Jens Axboe6b063142019-01-10 22:13:58 -07007252
7253 return 0;
7254}
7255
7256/*
7257 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7258 * causes regular reference counting to break down. We rely on the UNIX
7259 * garbage collection to take care of this problem for us.
7260 */
7261static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7262{
7263 unsigned left, total;
7264 int ret = 0;
7265
7266 total = 0;
7267 left = ctx->nr_user_files;
7268 while (left) {
7269 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007270
7271 ret = __io_sqe_files_scm(ctx, this_files, total);
7272 if (ret)
7273 break;
7274 left -= this_files;
7275 total += this_files;
7276 }
7277
7278 if (!ret)
7279 return 0;
7280
7281 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007282 struct file *file = io_file_from_index(ctx, total);
7283
7284 if (file)
7285 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007286 total++;
7287 }
7288
7289 return ret;
7290}
7291#else
7292static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7293{
7294 return 0;
7295}
7296#endif
7297
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007298static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data,
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007299 unsigned nr_tables, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007300{
7301 int i;
7302
7303 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007304 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007305 unsigned this_files;
7306
7307 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
7308 table->files = kcalloc(this_files, sizeof(struct file *),
7309 GFP_KERNEL);
7310 if (!table->files)
7311 break;
7312 nr_files -= this_files;
7313 }
7314
7315 if (i == nr_tables)
7316 return 0;
7317
7318 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007319 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007320 kfree(table->files);
7321 }
7322 return 1;
7323}
7324
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007325static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007326{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007327 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007328#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007329 struct sock *sock = ctx->ring_sock->sk;
7330 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7331 struct sk_buff *skb;
7332 int i;
7333
7334 __skb_queue_head_init(&list);
7335
7336 /*
7337 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7338 * remove this entry and rearrange the file array.
7339 */
7340 skb = skb_dequeue(head);
7341 while (skb) {
7342 struct scm_fp_list *fp;
7343
7344 fp = UNIXCB(skb).fp;
7345 for (i = 0; i < fp->count; i++) {
7346 int left;
7347
7348 if (fp->fp[i] != file)
7349 continue;
7350
7351 unix_notinflight(fp->user, fp->fp[i]);
7352 left = fp->count - 1 - i;
7353 if (left) {
7354 memmove(&fp->fp[i], &fp->fp[i + 1],
7355 left * sizeof(struct file *));
7356 }
7357 fp->count--;
7358 if (!fp->count) {
7359 kfree_skb(skb);
7360 skb = NULL;
7361 } else {
7362 __skb_queue_tail(&list, skb);
7363 }
7364 fput(file);
7365 file = NULL;
7366 break;
7367 }
7368
7369 if (!file)
7370 break;
7371
7372 __skb_queue_tail(&list, skb);
7373
7374 skb = skb_dequeue(head);
7375 }
7376
7377 if (skb_peek(&list)) {
7378 spin_lock_irq(&head->lock);
7379 while ((skb = __skb_dequeue(&list)) != NULL)
7380 __skb_queue_tail(head, skb);
7381 spin_unlock_irq(&head->lock);
7382 }
7383#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007384 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007385#endif
7386}
7387
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007388static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007389{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007390 struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data;
7391 struct io_ring_ctx *ctx = rsrc_data->ctx;
7392 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007393
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007394 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7395 list_del(&prsrc->list);
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007396 ref_node->rsrc_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007397 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007398 }
7399
Xiaoguang Wang05589552020-03-31 14:05:18 +08007400 percpu_ref_exit(&ref_node->refs);
7401 kfree(ref_node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007402 percpu_ref_put(&rsrc_data->refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007403}
7404
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007405static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007406{
7407 struct io_ring_ctx *ctx;
7408 struct llist_node *node;
7409
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007410 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7411 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007412
7413 while (node) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007414 struct fixed_rsrc_ref_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007415 struct llist_node *next = node->next;
7416
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007417 ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist);
7418 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007419 node = next;
7420 }
7421}
7422
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007423static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data,
7424 unsigned i)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007425{
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007426 struct fixed_rsrc_table *table;
7427
7428 table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT];
7429 return &table->files[i & IORING_FILE_TABLE_MASK];
7430}
7431
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007432static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007433{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007434 struct fixed_rsrc_ref_node *ref_node;
7435 struct fixed_rsrc_data *data;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007436 struct io_ring_ctx *ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007437 bool first_add = false;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007438 int delay = HZ;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007439
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007440 ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs);
7441 data = ref_node->rsrc_data;
Pavel Begunkove2978222020-11-18 14:56:26 +00007442 ctx = data->ctx;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007443
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007444 io_rsrc_ref_lock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007445 ref_node->done = true;
7446
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007447 while (!list_empty(&ctx->rsrc_ref_list)) {
7448 ref_node = list_first_entry(&ctx->rsrc_ref_list,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007449 struct fixed_rsrc_ref_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007450 /* recycle ref nodes in order */
7451 if (!ref_node->done)
7452 break;
7453 list_del(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007454 first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007455 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007456 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007457
7458 if (percpu_ref_is_dying(&data->refs))
Jens Axboe4a38aed22020-05-14 17:21:15 -06007459 delay = 0;
7460
Jens Axboe4a38aed22020-05-14 17:21:15 -06007461 if (!delay)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007462 mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007463 else if (first_add)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007464 queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007465}
7466
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007467static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Xiaoguang Wang05589552020-03-31 14:05:18 +08007468 struct io_ring_ctx *ctx)
7469{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007470 struct fixed_rsrc_ref_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007471
7472 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7473 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007474 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007475
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007476 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007477 0, GFP_KERNEL)) {
7478 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007479 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007480 }
7481 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007482 INIT_LIST_HEAD(&ref_node->rsrc_list);
Pavel Begunkove2978222020-11-18 14:56:26 +00007483 ref_node->done = false;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007484 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007485}
7486
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007487static void init_fixed_file_ref_node(struct io_ring_ctx *ctx,
7488 struct fixed_rsrc_ref_node *ref_node)
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007489{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007490 ref_node->rsrc_data = ctx->file_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007491 ref_node->rsrc_put = io_ring_file_put;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007492}
7493
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007494static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007495{
7496 percpu_ref_exit(&ref_node->refs);
7497 kfree(ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007498}
7499
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007500
Jens Axboe05f3fb32019-12-09 11:22:50 -07007501static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7502 unsigned nr_args)
7503{
7504 __s32 __user *fds = (__s32 __user *) arg;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007505 unsigned nr_tables, i;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007506 struct file *file;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007507 int fd, ret = -ENOMEM;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007508 struct fixed_rsrc_ref_node *ref_node;
7509 struct fixed_rsrc_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007510
7511 if (ctx->file_data)
7512 return -EBUSY;
7513 if (!nr_args)
7514 return -EINVAL;
7515 if (nr_args > IORING_MAX_FIXED_FILES)
7516 return -EMFILE;
7517
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007518 file_data = alloc_fixed_rsrc_data(ctx);
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007519 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007520 return -ENOMEM;
Dan Carpenter13770a72021-02-01 15:23:42 +03007521 ctx->file_data = file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007522
7523 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
Colin Ian King035fbaf2020-10-12 15:03:41 +01007524 file_data->table = kcalloc(nr_tables, sizeof(*file_data->table),
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007525 GFP_KERNEL);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007526 if (!file_data->table)
7527 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007528
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007529 if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args))
Jens Axboe05f3fb32019-12-09 11:22:50 -07007530 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007531
Jens Axboe05f3fb32019-12-09 11:22:50 -07007532 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007533 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7534 ret = -EFAULT;
7535 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007536 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007537 /* allow sparse sets */
7538 if (fd == -1)
7539 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007540
Jens Axboe05f3fb32019-12-09 11:22:50 -07007541 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007542 ret = -EBADF;
7543 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007544 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007545
7546 /*
7547 * Don't allow io_uring instances to be registered. If UNIX
7548 * isn't enabled, then this causes a reference cycle and this
7549 * instance can never get freed. If UNIX is enabled we'll
7550 * handle it just fine, but there's still no point in allowing
7551 * a ring fd as it doesn't support regular read/write anyway.
7552 */
7553 if (file->f_op == &io_uring_fops) {
7554 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007555 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007556 }
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007557 *io_fixed_file_slot(file_data, i) = file;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007558 }
7559
Jens Axboe05f3fb32019-12-09 11:22:50 -07007560 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007561 if (ret) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07007562 io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007563 return ret;
7564 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007565
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007566 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007567 if (!ref_node) {
Xiaoguang Wang05589552020-03-31 14:05:18 +08007568 io_sqe_files_unregister(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007569 return -ENOMEM;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007570 }
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007571 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007572
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007573 io_sqe_rsrc_set_node(ctx, file_data, ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007574 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007575out_fput:
7576 for (i = 0; i < ctx->nr_user_files; i++) {
7577 file = io_file_from_index(ctx, i);
7578 if (file)
7579 fput(file);
7580 }
7581 for (i = 0; i < nr_tables; i++)
7582 kfree(file_data->table[i].files);
7583 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007584out_free:
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007585 free_fixed_rsrc_data(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007586 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007587 return ret;
7588}
7589
Jens Axboec3a31e62019-10-03 13:59:56 -06007590static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7591 int index)
7592{
7593#if defined(CONFIG_UNIX)
7594 struct sock *sock = ctx->ring_sock->sk;
7595 struct sk_buff_head *head = &sock->sk_receive_queue;
7596 struct sk_buff *skb;
7597
7598 /*
7599 * See if we can merge this file into an existing skb SCM_RIGHTS
7600 * file set. If there's no room, fall back to allocating a new skb
7601 * and filling it in.
7602 */
7603 spin_lock_irq(&head->lock);
7604 skb = skb_peek(head);
7605 if (skb) {
7606 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7607
7608 if (fpl->count < SCM_MAX_FD) {
7609 __skb_unlink(skb, head);
7610 spin_unlock_irq(&head->lock);
7611 fpl->fp[fpl->count] = get_file(file);
7612 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7613 fpl->count++;
7614 spin_lock_irq(&head->lock);
7615 __skb_queue_head(head, skb);
7616 } else {
7617 skb = NULL;
7618 }
7619 }
7620 spin_unlock_irq(&head->lock);
7621
7622 if (skb) {
7623 fput(file);
7624 return 0;
7625 }
7626
7627 return __io_sqe_files_scm(ctx, 1, index);
7628#else
7629 return 0;
7630#endif
7631}
7632
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007633static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007634{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007635 struct io_rsrc_put *prsrc;
7636 struct fixed_rsrc_ref_node *ref_node = data->node;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007637
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007638 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7639 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007640 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007641
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007642 prsrc->rsrc = rsrc;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007643 list_add(&prsrc->list, &ref_node->rsrc_list);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007644
Hillf Dantona5318d32020-03-23 17:47:15 +08007645 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007646}
7647
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007648static inline int io_queue_file_removal(struct fixed_rsrc_data *data,
7649 struct file *file)
7650{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007651 return io_queue_rsrc_removal(data, (void *)file);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007652}
7653
Jens Axboe05f3fb32019-12-09 11:22:50 -07007654static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007655 struct io_uring_rsrc_update *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07007656 unsigned nr_args)
7657{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007658 struct fixed_rsrc_data *data = ctx->file_data;
7659 struct fixed_rsrc_ref_node *ref_node;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007660 struct file *file, **file_slot;
Jens Axboec3a31e62019-10-03 13:59:56 -06007661 __s32 __user *fds;
7662 int fd, i, err;
7663 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007664 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007665
Jens Axboe05f3fb32019-12-09 11:22:50 -07007666 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06007667 return -EOVERFLOW;
7668 if (done > ctx->nr_user_files)
7669 return -EINVAL;
7670
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007671 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007672 if (!ref_node)
7673 return -ENOMEM;
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007674 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007675
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007676 fds = u64_to_user_ptr(up->data);
Pavel Begunkov67973b92021-01-26 13:51:09 +00007677 for (done = 0; done < nr_args; done++) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007678 err = 0;
7679 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
7680 err = -EFAULT;
7681 break;
7682 }
noah4e0377a2021-01-26 15:23:28 -05007683 if (fd == IORING_REGISTER_FILES_SKIP)
7684 continue;
7685
Pavel Begunkov67973b92021-01-26 13:51:09 +00007686 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007687 file_slot = io_fixed_file_slot(ctx->file_data, i);
7688
7689 if (*file_slot) {
7690 err = io_queue_file_removal(data, *file_slot);
Hillf Dantona5318d32020-03-23 17:47:15 +08007691 if (err)
7692 break;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007693 *file_slot = NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007694 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007695 }
7696 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007697 file = fget(fd);
7698 if (!file) {
7699 err = -EBADF;
7700 break;
7701 }
7702 /*
7703 * Don't allow io_uring instances to be registered. If
7704 * UNIX isn't enabled, then this causes a reference
7705 * cycle and this instance can never get freed. If UNIX
7706 * is enabled we'll handle it just fine, but there's
7707 * still no point in allowing a ring fd as it doesn't
7708 * support regular read/write anyway.
7709 */
7710 if (file->f_op == &io_uring_fops) {
7711 fput(file);
7712 err = -EBADF;
7713 break;
7714 }
Jens Axboee68a3ff2021-02-11 07:45:08 -07007715 *file_slot = file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007716 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007717 if (err) {
Jens Axboee68a3ff2021-02-11 07:45:08 -07007718 *file_slot = NULL;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007719 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007720 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007721 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007722 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007723 }
7724
Xiaoguang Wang05589552020-03-31 14:05:18 +08007725 if (needs_switch) {
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007726 percpu_ref_kill(&data->node->refs);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007727 io_sqe_rsrc_set_node(ctx, data, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007728 } else
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007729 destroy_fixed_rsrc_ref_node(ref_node);
Jens Axboec3a31e62019-10-03 13:59:56 -06007730
7731 return done ? done : err;
7732}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007733
Jens Axboe05f3fb32019-12-09 11:22:50 -07007734static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
7735 unsigned nr_args)
7736{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007737 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007738
7739 if (!ctx->file_data)
7740 return -ENXIO;
7741 if (!nr_args)
7742 return -EINVAL;
7743 if (copy_from_user(&up, arg, sizeof(up)))
7744 return -EFAULT;
7745 if (up.resv)
7746 return -EINVAL;
7747
7748 return __io_sqe_files_update(ctx, &up, nr_args);
7749}
Jens Axboec3a31e62019-10-03 13:59:56 -06007750
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007751static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007752{
7753 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7754
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007755 req = io_put_req_find_next(req);
7756 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07007757}
7758
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007759static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx)
Pavel Begunkov24369c22020-01-28 03:15:48 +03007760{
Jens Axboee9418942021-02-19 12:33:30 -07007761 struct io_wq_hash *hash;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007762 struct io_wq_data data;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007763 unsigned int concurrency;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007764
Jens Axboee9418942021-02-19 12:33:30 -07007765 hash = ctx->hash_map;
7766 if (!hash) {
7767 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
7768 if (!hash)
7769 return ERR_PTR(-ENOMEM);
7770 refcount_set(&hash->refs, 1);
7771 init_waitqueue_head(&hash->wait);
7772 ctx->hash_map = hash;
7773 }
7774
7775 data.hash = hash;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007776 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007777 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007778
Jens Axboed25e3a32021-02-16 11:41:41 -07007779 /* Do QD, or 4 * CPUS, whatever is smallest */
7780 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
Pavel Begunkov24369c22020-01-28 03:15:48 +03007781
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007782 return io_wq_create(concurrency, &data);
Pavel Begunkov24369c22020-01-28 03:15:48 +03007783}
7784
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007785static int io_uring_alloc_task_context(struct task_struct *task,
7786 struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06007787{
7788 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007789 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007790
7791 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
7792 if (unlikely(!tctx))
7793 return -ENOMEM;
7794
Jens Axboed8a6df12020-10-15 16:24:45 -06007795 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7796 if (unlikely(ret)) {
7797 kfree(tctx);
7798 return ret;
7799 }
7800
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007801 tctx->io_wq = io_init_wq_offload(ctx);
7802 if (IS_ERR(tctx->io_wq)) {
7803 ret = PTR_ERR(tctx->io_wq);
7804 percpu_counter_destroy(&tctx->inflight);
7805 kfree(tctx);
7806 return ret;
7807 }
7808
Jens Axboe0f212202020-09-13 13:09:39 -06007809 xa_init(&tctx->xa);
7810 init_waitqueue_head(&tctx->wait);
7811 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06007812 atomic_set(&tctx->in_idle, 0);
7813 tctx->sqpoll = false;
Jens Axboe0f212202020-09-13 13:09:39 -06007814 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00007815 spin_lock_init(&tctx->task_lock);
7816 INIT_WQ_LIST(&tctx->task_list);
7817 tctx->task_state = 0;
7818 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06007819 return 0;
7820}
7821
7822void __io_uring_free(struct task_struct *tsk)
7823{
7824 struct io_uring_task *tctx = tsk->io_uring;
7825
7826 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00007827 WARN_ON_ONCE(tctx->io_wq);
7828
Jens Axboed8a6df12020-10-15 16:24:45 -06007829 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06007830 kfree(tctx);
7831 tsk->io_uring = NULL;
7832}
7833
Jens Axboe5f3f26f2021-02-25 10:17:46 -07007834static int io_sq_thread_fork(struct io_sq_data *sqd, struct io_ring_ctx *ctx)
7835{
7836 int ret;
7837
7838 clear_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
7839 reinit_completion(&sqd->completion);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00007840 ctx->sqo_exec = 0;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07007841 sqd->task_pid = current->pid;
7842 current->flags |= PF_IO_WORKER;
7843 ret = io_wq_fork_thread(io_sq_thread, sqd);
7844 current->flags &= ~PF_IO_WORKER;
7845 if (ret < 0) {
7846 sqd->thread = NULL;
7847 return ret;
7848 }
7849 wait_for_completion(&sqd->completion);
7850 return io_uring_alloc_task_context(sqd->thread, ctx);
7851}
7852
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007853static int io_sq_offload_create(struct io_ring_ctx *ctx,
7854 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007855{
7856 int ret;
7857
Jens Axboed25e3a32021-02-16 11:41:41 -07007858 /* Retain compatibility with failing for an invalid attach attempt */
7859 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
7860 IORING_SETUP_ATTACH_WQ) {
7861 struct fd f;
7862
7863 f = fdget(p->wq_fd);
7864 if (!f.file)
7865 return -ENXIO;
7866 if (f.file->f_op != &io_uring_fops) {
7867 fdput(f);
7868 return -EINVAL;
7869 }
7870 fdput(f);
7871 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07007872 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe534ca6d2020-09-02 13:52:19 -06007873 struct io_sq_data *sqd;
7874
Jens Axboe3ec482d2019-04-08 10:51:01 -06007875 ret = -EPERM;
Jens Axboece59fc62020-09-02 13:28:09 -06007876 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE))
Jens Axboe3ec482d2019-04-08 10:51:01 -06007877 goto err;
7878
Jens Axboe534ca6d2020-09-02 13:52:19 -06007879 sqd = io_get_sq_data(p);
7880 if (IS_ERR(sqd)) {
7881 ret = PTR_ERR(sqd);
7882 goto err;
7883 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007884
Jens Axboe534ca6d2020-09-02 13:52:19 -06007885 ctx->sq_data = sqd;
Jens Axboe69fb2132020-09-14 11:16:23 -06007886 io_sq_thread_park(sqd);
7887 mutex_lock(&sqd->ctx_lock);
7888 list_add(&ctx->sqd_list, &sqd->ctx_new_list);
7889 mutex_unlock(&sqd->ctx_lock);
7890 io_sq_thread_unpark(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007891
Jens Axboe917257d2019-04-13 09:28:55 -06007892 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
7893 if (!ctx->sq_thread_idle)
7894 ctx->sq_thread_idle = HZ;
7895
Jens Axboeaa061652020-09-02 14:50:27 -06007896 if (sqd->thread)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007897 return 0;
Jens Axboeaa061652020-09-02 14:50:27 -06007898
Jens Axboe6c271ce2019-01-10 11:22:30 -07007899 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06007900 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007901
Jens Axboe917257d2019-04-13 09:28:55 -06007902 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06007903 if (cpu >= nr_cpu_ids)
7904 goto err;
Shenghui Wang7889f442019-05-07 16:03:19 +08007905 if (!cpu_online(cpu))
Jens Axboe917257d2019-04-13 09:28:55 -06007906 goto err;
7907
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007908 sqd->sq_cpu = cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007909 } else {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007910 sqd->sq_cpu = -1;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007911 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007912
7913 sqd->task_pid = current->pid;
7914 current->flags |= PF_IO_WORKER;
7915 ret = io_wq_fork_thread(io_sq_thread, sqd);
7916 current->flags &= ~PF_IO_WORKER;
7917 if (ret < 0) {
Jens Axboe534ca6d2020-09-02 13:52:19 -06007918 sqd->thread = NULL;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007919 goto err;
7920 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007921 wait_for_completion(&sqd->completion);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007922 ret = io_uring_alloc_task_context(sqd->thread, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06007923 if (ret)
7924 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007925 } else if (p->flags & IORING_SETUP_SQ_AFF) {
7926 /* Can't have SQ_AFF without SQPOLL */
7927 ret = -EINVAL;
7928 goto err;
7929 }
7930
Jens Axboe2b188cc2019-01-07 10:46:33 -07007931 return 0;
7932err:
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007933 io_sq_thread_finish(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007934 return ret;
7935}
7936
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007937static void io_sq_offload_start(struct io_ring_ctx *ctx)
7938{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007939 struct io_sq_data *sqd = ctx->sq_data;
7940
Jens Axboe3ebba792021-02-28 15:32:18 -07007941 ctx->flags &= ~IORING_SETUP_R_DISABLED;
Jens Axboeeb858902021-02-25 10:13:29 -07007942 if (ctx->flags & IORING_SETUP_SQPOLL)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007943 complete(&sqd->startup);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007944}
7945
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007946static inline void __io_unaccount_mem(struct user_struct *user,
7947 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007948{
7949 atomic_long_sub(nr_pages, &user->locked_vm);
7950}
7951
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007952static inline int __io_account_mem(struct user_struct *user,
7953 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007954{
7955 unsigned long page_limit, cur_pages, new_pages;
7956
7957 /* Don't allow more pages than we can safely lock */
7958 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
7959
7960 do {
7961 cur_pages = atomic_long_read(&user->locked_vm);
7962 new_pages = cur_pages + nr_pages;
7963 if (new_pages > page_limit)
7964 return -ENOMEM;
7965 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
7966 new_pages) != cur_pages);
7967
7968 return 0;
7969}
7970
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007971static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007972{
Jens Axboe62e398b2021-02-21 16:19:37 -07007973 if (ctx->user)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007974 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007975
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007976 if (ctx->mm_account)
7977 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007978}
7979
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007980static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007981{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007982 int ret;
7983
Jens Axboe62e398b2021-02-21 16:19:37 -07007984 if (ctx->user) {
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007985 ret = __io_account_mem(ctx->user, nr_pages);
7986 if (ret)
7987 return ret;
7988 }
7989
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007990 if (ctx->mm_account)
7991 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007992
7993 return 0;
7994}
7995
Jens Axboe2b188cc2019-01-07 10:46:33 -07007996static void io_mem_free(void *ptr)
7997{
Mark Rutland52e04ef2019-04-30 17:30:21 +01007998 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007999
Mark Rutland52e04ef2019-04-30 17:30:21 +01008000 if (!ptr)
8001 return;
8002
8003 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008004 if (put_page_testzero(page))
8005 free_compound_page(page);
8006}
8007
8008static void *io_mem_alloc(size_t size)
8009{
8010 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008011 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008012
8013 return (void *) __get_free_pages(gfp_flags, get_order(size));
8014}
8015
Hristo Venev75b28af2019-08-26 17:23:46 +00008016static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
8017 size_t *sq_offset)
8018{
8019 struct io_rings *rings;
8020 size_t off, sq_array_size;
8021
8022 off = struct_size(rings, cqes, cq_entries);
8023 if (off == SIZE_MAX)
8024 return SIZE_MAX;
8025
8026#ifdef CONFIG_SMP
8027 off = ALIGN(off, SMP_CACHE_BYTES);
8028 if (off == 0)
8029 return SIZE_MAX;
8030#endif
8031
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02008032 if (sq_offset)
8033 *sq_offset = off;
8034
Hristo Venev75b28af2019-08-26 17:23:46 +00008035 sq_array_size = array_size(sizeof(u32), sq_entries);
8036 if (sq_array_size == SIZE_MAX)
8037 return SIZE_MAX;
8038
8039 if (check_add_overflow(off, sq_array_size, &off))
8040 return SIZE_MAX;
8041
Hristo Venev75b28af2019-08-26 17:23:46 +00008042 return off;
8043}
8044
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008045static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07008046{
8047 int i, j;
8048
8049 if (!ctx->user_bufs)
8050 return -ENXIO;
8051
8052 for (i = 0; i < ctx->nr_user_bufs; i++) {
8053 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8054
8055 for (j = 0; j < imu->nr_bvecs; j++)
John Hubbardf1f6a7d2020-01-30 22:13:35 -08008056 unpin_user_page(imu->bvec[j].bv_page);
Jens Axboeedafcce2019-01-09 09:16:05 -07008057
Jens Axboede293932020-09-17 16:19:16 -06008058 if (imu->acct_pages)
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008059 io_unaccount_mem(ctx, imu->acct_pages);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008060 kvfree(imu->bvec);
Jens Axboeedafcce2019-01-09 09:16:05 -07008061 imu->nr_bvecs = 0;
8062 }
8063
8064 kfree(ctx->user_bufs);
8065 ctx->user_bufs = NULL;
8066 ctx->nr_user_bufs = 0;
8067 return 0;
8068}
8069
8070static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8071 void __user *arg, unsigned index)
8072{
8073 struct iovec __user *src;
8074
8075#ifdef CONFIG_COMPAT
8076 if (ctx->compat) {
8077 struct compat_iovec __user *ciovs;
8078 struct compat_iovec ciov;
8079
8080 ciovs = (struct compat_iovec __user *) arg;
8081 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8082 return -EFAULT;
8083
Jens Axboed55e5f52019-12-11 16:12:15 -07008084 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008085 dst->iov_len = ciov.iov_len;
8086 return 0;
8087 }
8088#endif
8089 src = (struct iovec __user *) arg;
8090 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8091 return -EFAULT;
8092 return 0;
8093}
8094
Jens Axboede293932020-09-17 16:19:16 -06008095/*
8096 * Not super efficient, but this is just a registration time. And we do cache
8097 * the last compound head, so generally we'll only do a full search if we don't
8098 * match that one.
8099 *
8100 * We check if the given compound head page has already been accounted, to
8101 * avoid double accounting it. This allows us to account the full size of the
8102 * page, not just the constituent pages of a huge page.
8103 */
8104static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8105 int nr_pages, struct page *hpage)
8106{
8107 int i, j;
8108
8109 /* check current page array */
8110 for (i = 0; i < nr_pages; i++) {
8111 if (!PageCompound(pages[i]))
8112 continue;
8113 if (compound_head(pages[i]) == hpage)
8114 return true;
8115 }
8116
8117 /* check previously registered pages */
8118 for (i = 0; i < ctx->nr_user_bufs; i++) {
8119 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8120
8121 for (j = 0; j < imu->nr_bvecs; j++) {
8122 if (!PageCompound(imu->bvec[j].bv_page))
8123 continue;
8124 if (compound_head(imu->bvec[j].bv_page) == hpage)
8125 return true;
8126 }
8127 }
8128
8129 return false;
8130}
8131
8132static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8133 int nr_pages, struct io_mapped_ubuf *imu,
8134 struct page **last_hpage)
8135{
8136 int i, ret;
8137
8138 for (i = 0; i < nr_pages; i++) {
8139 if (!PageCompound(pages[i])) {
8140 imu->acct_pages++;
8141 } else {
8142 struct page *hpage;
8143
8144 hpage = compound_head(pages[i]);
8145 if (hpage == *last_hpage)
8146 continue;
8147 *last_hpage = hpage;
8148 if (headpage_already_acct(ctx, pages, i, hpage))
8149 continue;
8150 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8151 }
8152 }
8153
8154 if (!imu->acct_pages)
8155 return 0;
8156
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008157 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008158 if (ret)
8159 imu->acct_pages = 0;
8160 return ret;
8161}
8162
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008163static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
8164 struct io_mapped_ubuf *imu,
8165 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008166{
8167 struct vm_area_struct **vmas = NULL;
8168 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008169 unsigned long off, start, end, ubuf;
8170 size_t size;
8171 int ret, pret, nr_pages, i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008172
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008173 ubuf = (unsigned long) iov->iov_base;
8174 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8175 start = ubuf >> PAGE_SHIFT;
8176 nr_pages = end - start;
8177
8178 ret = -ENOMEM;
8179
8180 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8181 if (!pages)
8182 goto done;
8183
8184 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8185 GFP_KERNEL);
8186 if (!vmas)
8187 goto done;
8188
8189 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
8190 GFP_KERNEL);
8191 if (!imu->bvec)
8192 goto done;
8193
8194 ret = 0;
8195 mmap_read_lock(current->mm);
8196 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8197 pages, vmas);
8198 if (pret == nr_pages) {
8199 /* don't support file backed memory */
8200 for (i = 0; i < nr_pages; i++) {
8201 struct vm_area_struct *vma = vmas[i];
8202
8203 if (vma->vm_file &&
8204 !is_file_hugepages(vma->vm_file)) {
8205 ret = -EOPNOTSUPP;
8206 break;
8207 }
8208 }
8209 } else {
8210 ret = pret < 0 ? pret : -EFAULT;
8211 }
8212 mmap_read_unlock(current->mm);
8213 if (ret) {
8214 /*
8215 * if we did partial map, or found file backed vmas,
8216 * release any pages we did get
8217 */
8218 if (pret > 0)
8219 unpin_user_pages(pages, pret);
8220 kvfree(imu->bvec);
8221 goto done;
8222 }
8223
8224 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8225 if (ret) {
8226 unpin_user_pages(pages, pret);
8227 kvfree(imu->bvec);
8228 goto done;
8229 }
8230
8231 off = ubuf & ~PAGE_MASK;
8232 size = iov->iov_len;
8233 for (i = 0; i < nr_pages; i++) {
8234 size_t vec_len;
8235
8236 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8237 imu->bvec[i].bv_page = pages[i];
8238 imu->bvec[i].bv_len = vec_len;
8239 imu->bvec[i].bv_offset = off;
8240 off = 0;
8241 size -= vec_len;
8242 }
8243 /* store original address for later verification */
8244 imu->ubuf = ubuf;
8245 imu->len = iov->iov_len;
8246 imu->nr_bvecs = nr_pages;
8247 ret = 0;
8248done:
8249 kvfree(pages);
8250 kvfree(vmas);
8251 return ret;
8252}
8253
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008254static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008255{
Jens Axboeedafcce2019-01-09 09:16:05 -07008256 if (ctx->user_bufs)
8257 return -EBUSY;
8258 if (!nr_args || nr_args > UIO_MAXIOV)
8259 return -EINVAL;
8260
8261 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
8262 GFP_KERNEL);
8263 if (!ctx->user_bufs)
8264 return -ENOMEM;
8265
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008266 return 0;
8267}
8268
8269static int io_buffer_validate(struct iovec *iov)
8270{
8271 /*
8272 * Don't impose further limits on the size and buffer
8273 * constraints here, we'll -EINVAL later when IO is
8274 * submitted if they are wrong.
8275 */
8276 if (!iov->iov_base || !iov->iov_len)
8277 return -EFAULT;
8278
8279 /* arbitrary limit, but we need something */
8280 if (iov->iov_len > SZ_1G)
8281 return -EFAULT;
8282
8283 return 0;
8284}
8285
8286static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
8287 unsigned int nr_args)
8288{
8289 int i, ret;
8290 struct iovec iov;
8291 struct page *last_hpage = NULL;
8292
8293 ret = io_buffers_map_alloc(ctx, nr_args);
8294 if (ret)
8295 return ret;
8296
Jens Axboeedafcce2019-01-09 09:16:05 -07008297 for (i = 0; i < nr_args; i++) {
8298 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
Jens Axboeedafcce2019-01-09 09:16:05 -07008299
8300 ret = io_copy_iov(ctx, &iov, arg, i);
8301 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008302 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008303
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008304 ret = io_buffer_validate(&iov);
8305 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008306 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008307
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008308 ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage);
8309 if (ret)
8310 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008311
8312 ctx->nr_user_bufs++;
8313 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008314
8315 if (ret)
8316 io_sqe_buffers_unregister(ctx);
8317
Jens Axboeedafcce2019-01-09 09:16:05 -07008318 return ret;
8319}
8320
Jens Axboe9b402842019-04-11 11:45:41 -06008321static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8322{
8323 __s32 __user *fds = arg;
8324 int fd;
8325
8326 if (ctx->cq_ev_fd)
8327 return -EBUSY;
8328
8329 if (copy_from_user(&fd, fds, sizeof(*fds)))
8330 return -EFAULT;
8331
8332 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8333 if (IS_ERR(ctx->cq_ev_fd)) {
8334 int ret = PTR_ERR(ctx->cq_ev_fd);
8335 ctx->cq_ev_fd = NULL;
8336 return ret;
8337 }
8338
8339 return 0;
8340}
8341
8342static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8343{
8344 if (ctx->cq_ev_fd) {
8345 eventfd_ctx_put(ctx->cq_ev_fd);
8346 ctx->cq_ev_fd = NULL;
8347 return 0;
8348 }
8349
8350 return -ENXIO;
8351}
8352
Jens Axboe5a2e7452020-02-23 16:23:11 -07008353static int __io_destroy_buffers(int id, void *p, void *data)
8354{
8355 struct io_ring_ctx *ctx = data;
8356 struct io_buffer *buf = p;
8357
Jens Axboe067524e2020-03-02 16:32:28 -07008358 __io_remove_buffers(ctx, buf, id, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008359 return 0;
8360}
8361
8362static void io_destroy_buffers(struct io_ring_ctx *ctx)
8363{
8364 idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx);
8365 idr_destroy(&ctx->io_buffer_idr);
8366}
8367
Jens Axboe68e68ee2021-02-13 09:00:02 -07008368static void io_req_cache_free(struct list_head *list, struct task_struct *tsk)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008369{
Jens Axboe68e68ee2021-02-13 09:00:02 -07008370 struct io_kiocb *req, *nxt;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008371
Jens Axboe68e68ee2021-02-13 09:00:02 -07008372 list_for_each_entry_safe(req, nxt, list, compl.list) {
8373 if (tsk && req->task != tsk)
8374 continue;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008375 list_del(&req->compl.list);
8376 kmem_cache_free(req_cachep, req);
8377 }
8378}
8379
Jens Axboe4010fec2021-02-27 15:04:18 -07008380static void io_req_caches_free(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008381{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008382 struct io_submit_state *submit_state = &ctx->submit_state;
Pavel Begunkove5547d22021-02-23 22:17:20 +00008383 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008384
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008385 mutex_lock(&ctx->uring_lock);
8386
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008387 if (submit_state->free_reqs) {
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008388 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8389 submit_state->reqs);
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008390 submit_state->free_reqs = 0;
8391 }
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008392
8393 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkove5547d22021-02-23 22:17:20 +00008394 list_splice_init(&cs->locked_free_list, &cs->free_list);
8395 cs->locked_free_nr = 0;
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008396 spin_unlock_irq(&ctx->completion_lock);
8397
Pavel Begunkove5547d22021-02-23 22:17:20 +00008398 io_req_cache_free(&cs->free_list, NULL);
8399
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008400 mutex_unlock(&ctx->uring_lock);
8401}
8402
Jens Axboe2b188cc2019-01-07 10:46:33 -07008403static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8404{
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00008405 /*
8406 * Some may use context even when all refs and requests have been put,
8407 * and they are free to do so while still holding uring_lock, see
8408 * __io_req_task_submit(). Wait for them to finish.
8409 */
8410 mutex_lock(&ctx->uring_lock);
8411 mutex_unlock(&ctx->uring_lock);
8412
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008413 io_sq_thread_finish(ctx);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008414 io_sqe_buffers_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008415
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008416 if (ctx->mm_account) {
Jens Axboe2aede0e2020-09-14 10:45:53 -06008417 mmdrop(ctx->mm_account);
8418 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008419 }
Jens Axboedef596e2019-01-09 08:59:42 -07008420
Hao Xu8bad28d2021-02-19 17:19:36 +08008421 mutex_lock(&ctx->uring_lock);
Jens Axboe6b063142019-01-10 22:13:58 -07008422 io_sqe_files_unregister(ctx);
Hao Xu8bad28d2021-02-19 17:19:36 +08008423 mutex_unlock(&ctx->uring_lock);
Jens Axboe9b402842019-04-11 11:45:41 -06008424 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008425 io_destroy_buffers(ctx);
Jens Axboe41726c92020-02-23 13:11:42 -07008426 idr_destroy(&ctx->personality_idr);
Jens Axboedef596e2019-01-09 08:59:42 -07008427
Jens Axboe2b188cc2019-01-07 10:46:33 -07008428#if defined(CONFIG_UNIX)
Eric Biggers355e8d22019-06-12 14:58:43 -07008429 if (ctx->ring_sock) {
8430 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008431 sock_release(ctx->ring_sock);
Eric Biggers355e8d22019-06-12 14:58:43 -07008432 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008433#endif
8434
Hristo Venev75b28af2019-08-26 17:23:46 +00008435 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008436 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008437
8438 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008439 free_uid(ctx->user);
Jens Axboe4010fec2021-02-27 15:04:18 -07008440 io_req_caches_free(ctx);
Jens Axboee9418942021-02-19 12:33:30 -07008441 if (ctx->hash_map)
8442 io_wq_put_hash(ctx->hash_map);
Jens Axboe78076bb2019-12-04 19:56:40 -07008443 kfree(ctx->cancel_hash);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008444 kfree(ctx);
8445}
8446
8447static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8448{
8449 struct io_ring_ctx *ctx = file->private_data;
8450 __poll_t mask = 0;
8451
8452 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008453 /*
8454 * synchronizes with barrier from wq_has_sleeper call in
8455 * io_commit_cqring
8456 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008457 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008458 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008459 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008460
8461 /*
8462 * Don't flush cqring overflow list here, just do a simple check.
8463 * Otherwise there could possible be ABBA deadlock:
8464 * CPU0 CPU1
8465 * ---- ----
8466 * lock(&ctx->uring_lock);
8467 * lock(&ep->mtx);
8468 * lock(&ctx->uring_lock);
8469 * lock(&ep->mtx);
8470 *
8471 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8472 * pushs them to do the flush.
8473 */
8474 if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008475 mask |= EPOLLIN | EPOLLRDNORM;
8476
8477 return mask;
8478}
8479
8480static int io_uring_fasync(int fd, struct file *file, int on)
8481{
8482 struct io_ring_ctx *ctx = file->private_data;
8483
8484 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8485}
8486
Yejune Deng0bead8c2020-12-24 11:02:20 +08008487static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008488{
Jens Axboe4379bf82021-02-15 13:40:22 -07008489 const struct cred *creds;
Jens Axboe071698e2020-01-28 10:04:42 -07008490
Jens Axboe4379bf82021-02-15 13:40:22 -07008491 creds = idr_remove(&ctx->personality_idr, id);
8492 if (creds) {
8493 put_cred(creds);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008494 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008495 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008496
8497 return -EINVAL;
8498}
8499
8500static int io_remove_personalities(int id, void *p, void *data)
8501{
8502 struct io_ring_ctx *ctx = data;
8503
8504 io_unregister_personality(ctx, id);
Jens Axboe071698e2020-01-28 10:04:42 -07008505 return 0;
8506}
8507
Pavel Begunkovba50a032021-02-26 15:47:56 +00008508static bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008509{
Pavel Begunkov28c47212021-02-28 22:04:54 +00008510 struct callback_head *work, *next;
Pavel Begunkovba50a032021-02-26 15:47:56 +00008511 bool executed = false;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008512
8513 do {
Pavel Begunkov28c47212021-02-28 22:04:54 +00008514 work = xchg(&ctx->exit_task_work, NULL);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008515 if (!work)
8516 break;
8517
8518 do {
8519 next = work->next;
8520 work->func(work);
8521 work = next;
8522 cond_resched();
8523 } while (work);
Pavel Begunkovba50a032021-02-26 15:47:56 +00008524 executed = true;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008525 } while (1);
Pavel Begunkovba50a032021-02-26 15:47:56 +00008526
8527 return executed;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008528}
8529
Jens Axboe85faa7b2020-04-09 18:14:00 -06008530static void io_ring_exit_work(struct work_struct *work)
8531{
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008532 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx,
8533 exit_work);
Jens Axboe85faa7b2020-04-09 18:14:00 -06008534
Jens Axboe56952e92020-06-17 15:00:04 -06008535 /*
8536 * If we're doing polled IO and end up having requests being
8537 * submitted async (out-of-line), then completions can come in while
8538 * we're waiting for refs to drop. We need to reap these manually,
8539 * as nobody else will be looking for them.
8540 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008541 do {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008542 io_uring_try_cancel_requests(ctx, NULL, NULL);
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008543 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Jens Axboe85faa7b2020-04-09 18:14:00 -06008544 io_ring_ctx_free(ctx);
8545}
8546
Jens Axboe2b188cc2019-01-07 10:46:33 -07008547static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8548{
8549 mutex_lock(&ctx->uring_lock);
8550 percpu_ref_kill(&ctx->refs);
Pavel Begunkovcda286f2020-12-17 00:24:35 +00008551 /* if force is set, the ring is going away. always drop after that */
8552 ctx->cq_overflow_flushed = 1;
Pavel Begunkov634578f2020-12-06 22:22:44 +00008553 if (ctx->rings)
Pavel Begunkov6c503152021-01-04 20:36:36 +00008554 __io_cqring_overflow_flush(ctx, true, NULL, NULL);
Pavel Begunkov5c766a92021-01-19 13:32:36 +00008555 idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008556 mutex_unlock(&ctx->uring_lock);
8557
Pavel Begunkov6b819282020-11-06 13:00:25 +00008558 io_kill_timeouts(ctx, NULL, NULL);
8559 io_poll_remove_all(ctx, NULL, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008560
Jens Axboe15dff282019-11-13 09:09:23 -07008561 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008562 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008563
Jens Axboe85faa7b2020-04-09 18:14:00 -06008564 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008565 /*
8566 * Use system_unbound_wq to avoid spawning tons of event kworkers
8567 * if we're exiting a ton of rings at the same time. It just adds
8568 * noise and overhead, there's no discernable change in runtime
8569 * over using system_wq.
8570 */
8571 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008572}
8573
8574static int io_uring_release(struct inode *inode, struct file *file)
8575{
8576 struct io_ring_ctx *ctx = file->private_data;
8577
8578 file->private_data = NULL;
8579 io_ring_ctx_wait_and_kill(ctx);
8580 return 0;
8581}
8582
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008583struct io_task_cancel {
8584 struct task_struct *task;
8585 struct files_struct *files;
8586};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008587
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008588static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008589{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008590 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008591 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008592 bool ret;
8593
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008594 if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008595 unsigned long flags;
8596 struct io_ring_ctx *ctx = req->ctx;
8597
8598 /* protect against races with linked timeouts */
8599 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008600 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008601 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8602 } else {
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008603 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008604 }
8605 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008606}
8607
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008608static void io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008609 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008610 struct files_struct *files)
8611{
8612 struct io_defer_entry *de = NULL;
8613 LIST_HEAD(list);
8614
8615 spin_lock_irq(&ctx->completion_lock);
8616 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00008617 if (io_match_task(de->req, task, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008618 list_cut_position(&list, &ctx->defer_list, &de->list);
8619 break;
8620 }
8621 }
8622 spin_unlock_irq(&ctx->completion_lock);
8623
8624 while (!list_empty(&list)) {
8625 de = list_first_entry(&list, struct io_defer_entry, list);
8626 list_del_init(&de->list);
8627 req_set_fail_links(de->req);
8628 io_put_req(de->req);
8629 io_req_complete(de->req, -ECANCELED);
8630 kfree(de);
8631 }
8632}
8633
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008634static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
8635 struct task_struct *task,
8636 struct files_struct *files)
8637{
8638 struct io_task_cancel cancel = { .task = task, .files = files, };
Pavel Begunkov64c72122021-03-01 18:20:45 +00008639 struct task_struct *tctx_task = task ?: current;
8640 struct io_uring_task *tctx = tctx_task->io_uring;
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008641
8642 while (1) {
8643 enum io_wq_cancel cret;
8644 bool ret = false;
8645
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008646 if (tctx && tctx->io_wq) {
8647 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008648 &cancel, true);
8649 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8650 }
8651
8652 /* SQPOLL thread does its own polling */
8653 if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) {
8654 while (!list_empty_careful(&ctx->iopoll_list)) {
8655 io_iopoll_try_reap_events(ctx);
8656 ret = true;
8657 }
8658 }
8659
8660 ret |= io_poll_remove_all(ctx, task, files);
8661 ret |= io_kill_timeouts(ctx, task, files);
8662 ret |= io_run_task_work();
Pavel Begunkovba50a032021-02-26 15:47:56 +00008663 ret |= io_run_ctx_fallback(ctx);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008664 io_cqring_overflow_flush(ctx, true, task, files);
8665 if (!ret)
8666 break;
8667 cond_resched();
8668 }
8669}
8670
Pavel Begunkovca70f002021-01-26 15:28:27 +00008671static int io_uring_count_inflight(struct io_ring_ctx *ctx,
8672 struct task_struct *task,
8673 struct files_struct *files)
8674{
8675 struct io_kiocb *req;
8676 int cnt = 0;
8677
8678 spin_lock_irq(&ctx->inflight_lock);
8679 list_for_each_entry(req, &ctx->inflight_list, inflight_entry)
8680 cnt += io_match_task(req, task, files);
8681 spin_unlock_irq(&ctx->inflight_lock);
8682 return cnt;
8683}
8684
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008685static void io_uring_cancel_files(struct io_ring_ctx *ctx,
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008686 struct task_struct *task,
Jens Axboefcb323c2019-10-24 12:39:47 -06008687 struct files_struct *files)
8688{
Jens Axboefcb323c2019-10-24 12:39:47 -06008689 while (!list_empty_careful(&ctx->inflight_list)) {
Xiaoguang Wangd8f1b972020-04-26 15:54:43 +08008690 DEFINE_WAIT(wait);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008691 int inflight;
Jens Axboefcb323c2019-10-24 12:39:47 -06008692
Pavel Begunkovca70f002021-01-26 15:28:27 +00008693 inflight = io_uring_count_inflight(ctx, task, files);
8694 if (!inflight)
Jens Axboefcb323c2019-10-24 12:39:47 -06008695 break;
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008696
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008697 io_uring_try_cancel_requests(ctx, task, files);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008698
Pavel Begunkov34343782021-02-10 11:45:42 +00008699 if (ctx->sq_data)
8700 io_sq_thread_unpark(ctx->sq_data);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008701 prepare_to_wait(&task->io_uring->wait, &wait,
8702 TASK_UNINTERRUPTIBLE);
8703 if (inflight == io_uring_count_inflight(ctx, task, files))
8704 schedule();
Pavel Begunkovc98de082020-11-15 12:56:32 +00008705 finish_wait(&task->io_uring->wait, &wait);
Pavel Begunkov34343782021-02-10 11:45:42 +00008706 if (ctx->sq_data)
8707 io_sq_thread_park(ctx->sq_data);
Jens Axboe0f212202020-09-13 13:09:39 -06008708 }
Jens Axboe0f212202020-09-13 13:09:39 -06008709}
8710
8711/*
8712 * We need to iteratively cancel requests, in case a request has dependent
8713 * hard links. These persist even for failure of cancelations, hence keep
8714 * looping until none are found.
8715 */
8716static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
8717 struct files_struct *files)
8718{
8719 struct task_struct *task = current;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008720 bool did_park = false;
Jens Axboe0f212202020-09-13 13:09:39 -06008721
Jens Axboefdaf0832020-10-30 09:37:30 -06008722 if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00008723 /* never started, nothing to cancel */
8724 if (ctx->flags & IORING_SETUP_R_DISABLED) {
8725 io_sq_offload_start(ctx);
8726 return;
8727 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008728 did_park = io_sq_thread_park(ctx->sq_data);
8729 if (did_park) {
8730 task = ctx->sq_data->thread;
8731 atomic_inc(&task->io_uring->in_idle);
8732 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008733 }
Jens Axboe0f212202020-09-13 13:09:39 -06008734
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008735 io_cancel_defer_files(ctx, task, files);
Jens Axboe0f212202020-09-13 13:09:39 -06008736
Pavel Begunkov3a7efd12021-01-28 23:23:42 +00008737 io_uring_cancel_files(ctx, task, files);
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008738 if (!files)
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008739 io_uring_try_cancel_requests(ctx, task, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008740
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008741 if (did_park) {
Jens Axboefdaf0832020-10-30 09:37:30 -06008742 atomic_dec(&task->io_uring->in_idle);
Jens Axboefdaf0832020-10-30 09:37:30 -06008743 io_sq_thread_unpark(ctx->sq_data);
8744 }
Jens Axboe0f212202020-09-13 13:09:39 -06008745}
8746
8747/*
8748 * Note that this task has used io_uring. We use it for cancelation purposes.
8749 */
Jens Axboefdaf0832020-10-30 09:37:30 -06008750static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file)
Jens Axboe0f212202020-09-13 13:09:39 -06008751{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008752 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkova528b042020-12-21 18:34:04 +00008753 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008754
8755 if (unlikely(!tctx)) {
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008756 ret = io_uring_alloc_task_context(current, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06008757 if (unlikely(ret))
8758 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008759 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008760 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008761 if (tctx->last != file) {
8762 void *old = xa_load(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06008763
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008764 if (!old) {
Jens Axboe0f212202020-09-13 13:09:39 -06008765 get_file(file);
Pavel Begunkova528b042020-12-21 18:34:04 +00008766 ret = xa_err(xa_store(&tctx->xa, (unsigned long)file,
8767 file, GFP_KERNEL));
8768 if (ret) {
8769 fput(file);
8770 return ret;
8771 }
Jens Axboe0f212202020-09-13 13:09:39 -06008772 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008773 tctx->last = file;
Jens Axboe0f212202020-09-13 13:09:39 -06008774 }
8775
Jens Axboefdaf0832020-10-30 09:37:30 -06008776 /*
8777 * This is race safe in that the task itself is doing this, hence it
8778 * cannot be going through the exit/cancel paths at the same time.
8779 * This cannot be modified while exit/cancel is running.
8780 */
8781 if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL))
8782 tctx->sqpoll = true;
8783
Jens Axboe0f212202020-09-13 13:09:39 -06008784 return 0;
8785}
8786
8787/*
8788 * Remove this io_uring_file -> task mapping.
8789 */
8790static void io_uring_del_task_file(struct file *file)
8791{
8792 struct io_uring_task *tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008793
8794 if (tctx->last == file)
8795 tctx->last = NULL;
Matthew Wilcox (Oracle)5e2ed8c2020-10-09 13:49:53 +01008796 file = xa_erase(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06008797 if (file)
8798 fput(file);
8799}
8800
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008801static void io_uring_clean_tctx(struct io_uring_task *tctx)
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008802{
8803 struct file *file;
8804 unsigned long index;
8805
8806 xa_for_each(&tctx->xa, index, file)
8807 io_uring_del_task_file(file);
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008808 if (tctx->io_wq) {
8809 io_wq_put_and_exit(tctx->io_wq);
8810 tctx->io_wq = NULL;
8811 }
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008812}
8813
Jens Axboe0f212202020-09-13 13:09:39 -06008814void __io_uring_files_cancel(struct files_struct *files)
8815{
8816 struct io_uring_task *tctx = current->io_uring;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008817 struct file *file;
8818 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06008819
8820 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008821 atomic_inc(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008822 xa_for_each(&tctx->xa, index, file)
8823 io_uring_cancel_task_requests(file->private_data, files);
Jens Axboefdaf0832020-10-30 09:37:30 -06008824 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008825
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008826 if (files)
8827 io_uring_clean_tctx(tctx);
Jens Axboefdaf0832020-10-30 09:37:30 -06008828}
8829
8830static s64 tctx_inflight(struct io_uring_task *tctx)
8831{
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008832 return percpu_counter_sum(&tctx->inflight);
8833}
8834
8835static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx)
8836{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008837 struct io_sq_data *sqd = ctx->sq_data;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008838 struct io_uring_task *tctx;
Jens Axboefdaf0832020-10-30 09:37:30 -06008839 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008840 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06008841
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008842 if (!sqd)
8843 return;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008844 if (!io_sq_thread_park(sqd))
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008845 return;
8846 tctx = ctx->sq_data->thread->io_uring;
Jens Axboee54945a2021-02-26 11:27:15 -07008847 /* can happen on fork/alloc failure, just ignore that state */
8848 if (!tctx) {
8849 io_sq_thread_unpark(sqd);
8850 return;
8851 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008852
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008853 atomic_inc(&tctx->in_idle);
8854 do {
8855 /* read completions before cancelations */
8856 inflight = tctx_inflight(tctx);
8857 if (!inflight)
8858 break;
8859 io_uring_cancel_task_requests(ctx, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008860
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008861 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8862 /*
8863 * If we've seen completions, retry without waiting. This
8864 * avoids a race where a completion comes in before we did
8865 * prepare_to_wait().
8866 */
8867 if (inflight == tctx_inflight(tctx))
8868 schedule();
8869 finish_wait(&tctx->wait, &wait);
8870 } while (1);
8871 atomic_dec(&tctx->in_idle);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008872 io_sq_thread_unpark(sqd);
Jens Axboe0f212202020-09-13 13:09:39 -06008873}
8874
Jens Axboe0f212202020-09-13 13:09:39 -06008875/*
8876 * Find any io_uring fd that this task has registered or done IO on, and cancel
8877 * requests.
8878 */
8879void __io_uring_task_cancel(void)
8880{
8881 struct io_uring_task *tctx = current->io_uring;
8882 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008883 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06008884
8885 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008886 atomic_inc(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06008887
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008888 if (tctx->sqpoll) {
8889 struct file *file;
8890 unsigned long index;
8891
8892 xa_for_each(&tctx->xa, index, file)
8893 io_uring_cancel_sqpoll(file->private_data);
8894 }
Pavel Begunkov0b5cd6c2021-01-17 02:29:56 +00008895
Jens Axboed8a6df12020-10-15 16:24:45 -06008896 do {
Jens Axboe0f212202020-09-13 13:09:39 -06008897 /* read completions before cancelations */
Jens Axboefdaf0832020-10-30 09:37:30 -06008898 inflight = tctx_inflight(tctx);
Jens Axboed8a6df12020-10-15 16:24:45 -06008899 if (!inflight)
8900 break;
Jens Axboe0f212202020-09-13 13:09:39 -06008901 __io_uring_files_cancel(NULL);
8902
8903 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8904
8905 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008906 * If we've seen completions, retry without waiting. This
8907 * avoids a race where a completion comes in before we did
8908 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06008909 */
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008910 if (inflight == tctx_inflight(tctx))
8911 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00008912 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008913 } while (1);
Jens Axboe0f212202020-09-13 13:09:39 -06008914
Jens Axboefdaf0832020-10-30 09:37:30 -06008915 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008916
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008917 io_uring_clean_tctx(tctx);
8918 /* all current's requests should be gone, we can kill tctx */
8919 __io_uring_free(current);
Pavel Begunkov44e728b2020-06-15 10:24:04 +03008920}
8921
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008922static void *io_uring_validate_mmap_request(struct file *file,
8923 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008924{
Jens Axboe2b188cc2019-01-07 10:46:33 -07008925 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008926 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008927 struct page *page;
8928 void *ptr;
8929
8930 switch (offset) {
8931 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00008932 case IORING_OFF_CQ_RING:
8933 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008934 break;
8935 case IORING_OFF_SQES:
8936 ptr = ctx->sq_sqes;
8937 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008938 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008939 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008940 }
8941
8942 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07008943 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008944 return ERR_PTR(-EINVAL);
8945
8946 return ptr;
8947}
8948
8949#ifdef CONFIG_MMU
8950
8951static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8952{
8953 size_t sz = vma->vm_end - vma->vm_start;
8954 unsigned long pfn;
8955 void *ptr;
8956
8957 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
8958 if (IS_ERR(ptr))
8959 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008960
8961 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
8962 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
8963}
8964
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008965#else /* !CONFIG_MMU */
8966
8967static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8968{
8969 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
8970}
8971
8972static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
8973{
8974 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
8975}
8976
8977static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
8978 unsigned long addr, unsigned long len,
8979 unsigned long pgoff, unsigned long flags)
8980{
8981 void *ptr;
8982
8983 ptr = io_uring_validate_mmap_request(file, pgoff, len);
8984 if (IS_ERR(ptr))
8985 return PTR_ERR(ptr);
8986
8987 return (unsigned long) ptr;
8988}
8989
8990#endif /* !CONFIG_MMU */
8991
Pavel Begunkovd9d05212021-01-08 20:57:25 +00008992static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06008993{
Pavel Begunkovd9d05212021-01-08 20:57:25 +00008994 int ret = 0;
Jens Axboe90554202020-09-03 12:12:41 -06008995 DEFINE_WAIT(wait);
8996
8997 do {
8998 if (!io_sqring_full(ctx))
8999 break;
Jens Axboe90554202020-09-03 12:12:41 -06009000 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
9001
9002 if (!io_sqring_full(ctx))
9003 break;
Jens Axboe90554202020-09-03 12:12:41 -06009004 schedule();
9005 } while (!signal_pending(current));
9006
9007 finish_wait(&ctx->sqo_sq_wait, &wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009008 return ret;
Jens Axboe90554202020-09-03 12:12:41 -06009009}
9010
Hao Xuc73ebb62020-11-03 10:54:37 +08009011static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9012 struct __kernel_timespec __user **ts,
9013 const sigset_t __user **sig)
9014{
9015 struct io_uring_getevents_arg arg;
9016
9017 /*
9018 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9019 * is just a pointer to the sigset_t.
9020 */
9021 if (!(flags & IORING_ENTER_EXT_ARG)) {
9022 *sig = (const sigset_t __user *) argp;
9023 *ts = NULL;
9024 return 0;
9025 }
9026
9027 /*
9028 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9029 * timespec and sigset_t pointers if good.
9030 */
9031 if (*argsz != sizeof(arg))
9032 return -EINVAL;
9033 if (copy_from_user(&arg, argp, sizeof(arg)))
9034 return -EFAULT;
9035 *sig = u64_to_user_ptr(arg.sigmask);
9036 *argsz = arg.sigmask_sz;
9037 *ts = u64_to_user_ptr(arg.ts);
9038 return 0;
9039}
9040
Jens Axboe2b188cc2019-01-07 10:46:33 -07009041SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009042 u32, min_complete, u32, flags, const void __user *, argp,
9043 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009044{
9045 struct io_ring_ctx *ctx;
9046 long ret = -EBADF;
9047 int submitted = 0;
9048 struct fd f;
9049
Jens Axboe4c6e2772020-07-01 11:29:10 -06009050 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009051
Jens Axboe90554202020-09-03 12:12:41 -06009052 if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
Hao Xuc73ebb62020-11-03 10:54:37 +08009053 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009054 return -EINVAL;
9055
9056 f = fdget(fd);
9057 if (!f.file)
9058 return -EBADF;
9059
9060 ret = -EOPNOTSUPP;
9061 if (f.file->f_op != &io_uring_fops)
9062 goto out_fput;
9063
9064 ret = -ENXIO;
9065 ctx = f.file->private_data;
9066 if (!percpu_ref_tryget(&ctx->refs))
9067 goto out_fput;
9068
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009069 ret = -EBADFD;
9070 if (ctx->flags & IORING_SETUP_R_DISABLED)
9071 goto out;
9072
Jens Axboe6c271ce2019-01-10 11:22:30 -07009073 /*
9074 * For SQ polling, the thread will do all submissions and completions.
9075 * Just return the requested submit count, and wake the thread if
9076 * we were asked to.
9077 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009078 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009079 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00009080 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009081
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009082 if (unlikely(ctx->sqo_exec)) {
9083 ret = io_sq_thread_fork(ctx->sq_data, ctx);
9084 if (ret)
9085 goto out;
9086 ctx->sqo_exec = 0;
9087 }
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009088 ret = -EOWNERDEAD;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009089 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009090 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009091 if (flags & IORING_ENTER_SQ_WAIT) {
9092 ret = io_sqpoll_wait_sq(ctx);
9093 if (ret)
9094 goto out;
9095 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009096 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009097 } else if (to_submit) {
Jens Axboefdaf0832020-10-30 09:37:30 -06009098 ret = io_uring_add_task_file(ctx, f.file);
Jens Axboe0f212202020-09-13 13:09:39 -06009099 if (unlikely(ret))
9100 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009101 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009102 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009103 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009104
9105 if (submitted != to_submit)
9106 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009107 }
9108 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009109 const sigset_t __user *sig;
9110 struct __kernel_timespec __user *ts;
9111
9112 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9113 if (unlikely(ret))
9114 goto out;
9115
Jens Axboe2b188cc2019-01-07 10:46:33 -07009116 min_complete = min(min_complete, ctx->cq_entries);
9117
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009118 /*
9119 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9120 * space applications don't need to do io completion events
9121 * polling again, they can rely on io_sq_thread to do polling
9122 * work, which can reduce cpu usage and uring_lock contention.
9123 */
9124 if (ctx->flags & IORING_SETUP_IOPOLL &&
9125 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009126 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009127 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009128 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009129 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009130 }
9131
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009132out:
Pavel Begunkov6805b322019-10-08 02:18:42 +03009133 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009134out_fput:
9135 fdput(f);
9136 return submitted ? submitted : ret;
9137}
9138
Tobias Klauserbebdb652020-02-26 18:38:32 +01009139#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009140static int io_uring_show_cred(int id, void *p, void *data)
9141{
Jens Axboe4379bf82021-02-15 13:40:22 -07009142 const struct cred *cred = p;
Jens Axboe87ce9552020-01-30 08:25:34 -07009143 struct seq_file *m = data;
9144 struct user_namespace *uns = seq_user_ns(m);
9145 struct group_info *gi;
9146 kernel_cap_t cap;
9147 unsigned __capi;
9148 int g;
9149
9150 seq_printf(m, "%5d\n", id);
9151 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9152 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9153 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9154 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9155 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9156 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9157 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9158 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9159 seq_puts(m, "\n\tGroups:\t");
9160 gi = cred->group_info;
9161 for (g = 0; g < gi->ngroups; g++) {
9162 seq_put_decimal_ull(m, g ? " " : "",
9163 from_kgid_munged(uns, gi->gid[g]));
9164 }
9165 seq_puts(m, "\n\tCapEff:\t");
9166 cap = cred->cap_effective;
9167 CAP_FOR_EACH_U32(__capi)
9168 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9169 seq_putc(m, '\n');
9170 return 0;
9171}
9172
9173static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9174{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009175 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009176 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009177 int i;
9178
Jens Axboefad8e0d2020-09-28 08:57:48 -06009179 /*
9180 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9181 * since fdinfo case grabs it in the opposite direction of normal use
9182 * cases. If we fail to get the lock, we just don't iterate any
9183 * structures that could be going away outside the io_uring mutex.
9184 */
9185 has_lock = mutex_trylock(&ctx->uring_lock);
9186
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009187 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
Joseph Qidbbe9c62020-09-29 09:01:22 -06009188 sq = ctx->sq_data;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009189 if (!sq->thread)
9190 sq = NULL;
9191 }
Joseph Qidbbe9c62020-09-29 09:01:22 -06009192
9193 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9194 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009195 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009196 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Pavel Begunkovea64ec022021-02-04 13:52:07 +00009197 struct file *f = *io_fixed_file_slot(ctx->file_data, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009198
Jens Axboe87ce9552020-01-30 08:25:34 -07009199 if (f)
9200 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9201 else
9202 seq_printf(m, "%5u: <none>\n", i);
9203 }
9204 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009205 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009206 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
9207
9208 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf,
9209 (unsigned int) buf->len);
9210 }
Jens Axboefad8e0d2020-09-28 08:57:48 -06009211 if (has_lock && !idr_is_empty(&ctx->personality_idr)) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009212 seq_printf(m, "Personalities:\n");
9213 idr_for_each(&ctx->personality_idr, io_uring_show_cred, m);
9214 }
Jens Axboed7718a92020-02-14 22:23:12 -07009215 seq_printf(m, "PollList:\n");
9216 spin_lock_irq(&ctx->completion_lock);
9217 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9218 struct hlist_head *list = &ctx->cancel_hash[i];
9219 struct io_kiocb *req;
9220
9221 hlist_for_each_entry(req, list, hash_node)
9222 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9223 req->task->task_works != NULL);
9224 }
9225 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009226 if (has_lock)
9227 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009228}
9229
9230static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9231{
9232 struct io_ring_ctx *ctx = f->private_data;
9233
9234 if (percpu_ref_tryget(&ctx->refs)) {
9235 __io_uring_show_fdinfo(ctx, m);
9236 percpu_ref_put(&ctx->refs);
9237 }
9238}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009239#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009240
Jens Axboe2b188cc2019-01-07 10:46:33 -07009241static const struct file_operations io_uring_fops = {
9242 .release = io_uring_release,
9243 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009244#ifndef CONFIG_MMU
9245 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9246 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9247#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009248 .poll = io_uring_poll,
9249 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009250#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009251 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009252#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009253};
9254
9255static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9256 struct io_uring_params *p)
9257{
Hristo Venev75b28af2019-08-26 17:23:46 +00009258 struct io_rings *rings;
9259 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009260
Jens Axboebd740482020-08-05 12:58:23 -06009261 /* make sure these are sane, as we already accounted them */
9262 ctx->sq_entries = p->sq_entries;
9263 ctx->cq_entries = p->cq_entries;
9264
Hristo Venev75b28af2019-08-26 17:23:46 +00009265 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9266 if (size == SIZE_MAX)
9267 return -EOVERFLOW;
9268
9269 rings = io_mem_alloc(size);
9270 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009271 return -ENOMEM;
9272
Hristo Venev75b28af2019-08-26 17:23:46 +00009273 ctx->rings = rings;
9274 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9275 rings->sq_ring_mask = p->sq_entries - 1;
9276 rings->cq_ring_mask = p->cq_entries - 1;
9277 rings->sq_ring_entries = p->sq_entries;
9278 rings->cq_ring_entries = p->cq_entries;
9279 ctx->sq_mask = rings->sq_ring_mask;
9280 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009281
9282 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009283 if (size == SIZE_MAX) {
9284 io_mem_free(ctx->rings);
9285 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009286 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009287 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009288
9289 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009290 if (!ctx->sq_sqes) {
9291 io_mem_free(ctx->rings);
9292 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009293 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009294 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009295
Jens Axboe2b188cc2019-01-07 10:46:33 -07009296 return 0;
9297}
9298
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009299static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9300{
9301 int ret, fd;
9302
9303 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9304 if (fd < 0)
9305 return fd;
9306
9307 ret = io_uring_add_task_file(ctx, file);
9308 if (ret) {
9309 put_unused_fd(fd);
9310 return ret;
9311 }
9312 fd_install(fd, file);
9313 return fd;
9314}
9315
Jens Axboe2b188cc2019-01-07 10:46:33 -07009316/*
9317 * Allocate an anonymous fd, this is what constitutes the application
9318 * visible backing of an io_uring instance. The application mmaps this
9319 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9320 * we have to tie this fd to a socket for file garbage collection purposes.
9321 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009322static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009323{
9324 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009325#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009326 int ret;
9327
Jens Axboe2b188cc2019-01-07 10:46:33 -07009328 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9329 &ctx->ring_sock);
9330 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009331 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009332#endif
9333
Jens Axboe2b188cc2019-01-07 10:46:33 -07009334 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9335 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009336#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009337 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009338 sock_release(ctx->ring_sock);
9339 ctx->ring_sock = NULL;
9340 } else {
9341 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009342 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009343#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009344 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009345}
9346
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009347static int io_uring_create(unsigned entries, struct io_uring_params *p,
9348 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009349{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009350 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009351 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009352 int ret;
9353
Jens Axboe8110c1a2019-12-28 15:39:54 -07009354 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009355 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009356 if (entries > IORING_MAX_ENTRIES) {
9357 if (!(p->flags & IORING_SETUP_CLAMP))
9358 return -EINVAL;
9359 entries = IORING_MAX_ENTRIES;
9360 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009361
9362 /*
9363 * Use twice as many entries for the CQ ring. It's possible for the
9364 * application to drive a higher depth than the size of the SQ ring,
9365 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009366 * some flexibility in overcommitting a bit. If the application has
9367 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9368 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009369 */
9370 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009371 if (p->flags & IORING_SETUP_CQSIZE) {
9372 /*
9373 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9374 * to a power-of-two, if it isn't already. We do NOT impose
9375 * any cq vs sq ring sizing.
9376 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009377 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009378 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009379 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9380 if (!(p->flags & IORING_SETUP_CLAMP))
9381 return -EINVAL;
9382 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9383 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009384 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9385 if (p->cq_entries < p->sq_entries)
9386 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009387 } else {
9388 p->cq_entries = 2 * p->sq_entries;
9389 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009390
Jens Axboe2b188cc2019-01-07 10:46:33 -07009391 ctx = io_ring_ctx_alloc(p);
Jens Axboe62e398b2021-02-21 16:19:37 -07009392 if (!ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009393 return -ENOMEM;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009394 ctx->compat = in_compat_syscall();
Jens Axboe62e398b2021-02-21 16:19:37 -07009395 if (!capable(CAP_IPC_LOCK))
9396 ctx->user = get_uid(current_user());
Jens Axboe2aede0e2020-09-14 10:45:53 -06009397
9398 /*
9399 * This is just grabbed for accounting purposes. When a process exits,
9400 * the mm is exited and dropped before the files, hence we need to hang
9401 * on to this mm purely for the purposes of being able to unaccount
9402 * memory (locked/pinned vm). It's not used for anything else.
9403 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009404 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009405 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009406
Jens Axboe2b188cc2019-01-07 10:46:33 -07009407 ret = io_allocate_scq_urings(ctx, p);
9408 if (ret)
9409 goto err;
9410
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009411 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009412 if (ret)
9413 goto err;
9414
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009415 if (!(p->flags & IORING_SETUP_R_DISABLED))
9416 io_sq_offload_start(ctx);
9417
Jens Axboe2b188cc2019-01-07 10:46:33 -07009418 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009419 p->sq_off.head = offsetof(struct io_rings, sq.head);
9420 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9421 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9422 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9423 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9424 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9425 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009426
9427 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009428 p->cq_off.head = offsetof(struct io_rings, cq.head);
9429 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9430 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9431 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9432 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9433 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009434 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009435
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009436 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9437 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009438 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009439 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
Jens Axboe1c0aa1f2021-02-20 11:55:28 -07009440 IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009441
9442 if (copy_to_user(params, p, sizeof(*p))) {
9443 ret = -EFAULT;
9444 goto err;
9445 }
Jens Axboed1719f72020-07-30 13:43:53 -06009446
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009447 file = io_uring_get_file(ctx);
9448 if (IS_ERR(file)) {
9449 ret = PTR_ERR(file);
9450 goto err;
9451 }
9452
Jens Axboed1719f72020-07-30 13:43:53 -06009453 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009454 * Install ring fd as the very last thing, so we don't risk someone
9455 * having closed it before we finish setup
9456 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009457 ret = io_uring_install_fd(ctx, file);
9458 if (ret < 0) {
9459 /* fput will clean it up */
9460 fput(file);
9461 return ret;
9462 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009463
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009464 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009465 return ret;
9466err:
9467 io_ring_ctx_wait_and_kill(ctx);
9468 return ret;
9469}
9470
9471/*
9472 * Sets up an aio uring context, and returns the fd. Applications asks for a
9473 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9474 * params structure passed in.
9475 */
9476static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9477{
9478 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009479 int i;
9480
9481 if (copy_from_user(&p, params, sizeof(p)))
9482 return -EFAULT;
9483 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9484 if (p.resv[i])
9485 return -EINVAL;
9486 }
9487
Jens Axboe6c271ce2019-01-10 11:22:30 -07009488 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009489 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009490 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9491 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009492 return -EINVAL;
9493
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009494 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009495}
9496
9497SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9498 struct io_uring_params __user *, params)
9499{
9500 return io_uring_setup(entries, params);
9501}
9502
Jens Axboe66f4af92020-01-16 15:36:52 -07009503static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9504{
9505 struct io_uring_probe *p;
9506 size_t size;
9507 int i, ret;
9508
9509 size = struct_size(p, ops, nr_args);
9510 if (size == SIZE_MAX)
9511 return -EOVERFLOW;
9512 p = kzalloc(size, GFP_KERNEL);
9513 if (!p)
9514 return -ENOMEM;
9515
9516 ret = -EFAULT;
9517 if (copy_from_user(p, arg, size))
9518 goto out;
9519 ret = -EINVAL;
9520 if (memchr_inv(p, 0, size))
9521 goto out;
9522
9523 p->last_op = IORING_OP_LAST - 1;
9524 if (nr_args > IORING_OP_LAST)
9525 nr_args = IORING_OP_LAST;
9526
9527 for (i = 0; i < nr_args; i++) {
9528 p->ops[i].op = i;
9529 if (!io_op_defs[i].not_supported)
9530 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9531 }
9532 p->ops_len = i;
9533
9534 ret = 0;
9535 if (copy_to_user(arg, p, size))
9536 ret = -EFAULT;
9537out:
9538 kfree(p);
9539 return ret;
9540}
9541
Jens Axboe071698e2020-01-28 10:04:42 -07009542static int io_register_personality(struct io_ring_ctx *ctx)
9543{
Jens Axboe4379bf82021-02-15 13:40:22 -07009544 const struct cred *creds;
Jens Axboe1e6fa522020-10-15 08:46:24 -06009545 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009546
Jens Axboe4379bf82021-02-15 13:40:22 -07009547 creds = get_current_cred();
Jens Axboe1e6fa522020-10-15 08:46:24 -06009548
Jens Axboe4379bf82021-02-15 13:40:22 -07009549 ret = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1,
9550 USHRT_MAX, GFP_KERNEL);
9551 if (ret < 0)
9552 put_cred(creds);
Jens Axboe1e6fa522020-10-15 08:46:24 -06009553 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009554}
9555
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009556static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9557 unsigned int nr_args)
9558{
9559 struct io_uring_restriction *res;
9560 size_t size;
9561 int i, ret;
9562
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009563 /* Restrictions allowed only if rings started disabled */
9564 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9565 return -EBADFD;
9566
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009567 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009568 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009569 return -EBUSY;
9570
9571 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9572 return -EINVAL;
9573
9574 size = array_size(nr_args, sizeof(*res));
9575 if (size == SIZE_MAX)
9576 return -EOVERFLOW;
9577
9578 res = memdup_user(arg, size);
9579 if (IS_ERR(res))
9580 return PTR_ERR(res);
9581
9582 ret = 0;
9583
9584 for (i = 0; i < nr_args; i++) {
9585 switch (res[i].opcode) {
9586 case IORING_RESTRICTION_REGISTER_OP:
9587 if (res[i].register_op >= IORING_REGISTER_LAST) {
9588 ret = -EINVAL;
9589 goto out;
9590 }
9591
9592 __set_bit(res[i].register_op,
9593 ctx->restrictions.register_op);
9594 break;
9595 case IORING_RESTRICTION_SQE_OP:
9596 if (res[i].sqe_op >= IORING_OP_LAST) {
9597 ret = -EINVAL;
9598 goto out;
9599 }
9600
9601 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9602 break;
9603 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9604 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9605 break;
9606 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9607 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9608 break;
9609 default:
9610 ret = -EINVAL;
9611 goto out;
9612 }
9613 }
9614
9615out:
9616 /* Reset all restrictions if an error happened */
9617 if (ret != 0)
9618 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9619 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009620 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009621
9622 kfree(res);
9623 return ret;
9624}
9625
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009626static int io_register_enable_rings(struct io_ring_ctx *ctx)
9627{
9628 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9629 return -EBADFD;
9630
9631 if (ctx->restrictions.registered)
9632 ctx->restricted = 1;
9633
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009634 io_sq_offload_start(ctx);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009635 return 0;
9636}
9637
Jens Axboe071698e2020-01-28 10:04:42 -07009638static bool io_register_op_must_quiesce(int op)
9639{
9640 switch (op) {
9641 case IORING_UNREGISTER_FILES:
9642 case IORING_REGISTER_FILES_UPDATE:
9643 case IORING_REGISTER_PROBE:
9644 case IORING_REGISTER_PERSONALITY:
9645 case IORING_UNREGISTER_PERSONALITY:
9646 return false;
9647 default:
9648 return true;
9649 }
9650}
9651
Jens Axboeedafcce2019-01-09 09:16:05 -07009652static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
9653 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -06009654 __releases(ctx->uring_lock)
9655 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -07009656{
9657 int ret;
9658
Jens Axboe35fa71a2019-04-22 10:23:23 -06009659 /*
9660 * We're inside the ring mutex, if the ref is already dying, then
9661 * someone else killed the ctx or is already going through
9662 * io_uring_register().
9663 */
9664 if (percpu_ref_is_dying(&ctx->refs))
9665 return -ENXIO;
9666
Jens Axboe071698e2020-01-28 10:04:42 -07009667 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009668 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -06009669
Jens Axboe05f3fb32019-12-09 11:22:50 -07009670 /*
9671 * Drop uring mutex before waiting for references to exit. If
9672 * another thread is currently inside io_uring_enter() it might
9673 * need to grab the uring_lock to make progress. If we hold it
9674 * here across the drain wait, then we can deadlock. It's safe
9675 * to drop the mutex here, since no new references will come in
9676 * after we've killed the percpu ref.
9677 */
9678 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009679 do {
9680 ret = wait_for_completion_interruptible(&ctx->ref_comp);
9681 if (!ret)
9682 break;
Jens Axboeed6930c2020-10-08 19:09:46 -06009683 ret = io_run_task_work_sig();
9684 if (ret < 0)
9685 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009686 } while (1);
9687
Jens Axboe05f3fb32019-12-09 11:22:50 -07009688 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009689
Jens Axboec1503682020-01-08 08:26:07 -07009690 if (ret) {
9691 percpu_ref_resurrect(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009692 goto out_quiesce;
9693 }
9694 }
9695
9696 if (ctx->restricted) {
9697 if (opcode >= IORING_REGISTER_LAST) {
9698 ret = -EINVAL;
9699 goto out;
9700 }
9701
9702 if (!test_bit(opcode, ctx->restrictions.register_op)) {
9703 ret = -EACCES;
Jens Axboec1503682020-01-08 08:26:07 -07009704 goto out;
9705 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07009706 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009707
9708 switch (opcode) {
9709 case IORING_REGISTER_BUFFERS:
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009710 ret = io_sqe_buffers_register(ctx, arg, nr_args);
Jens Axboeedafcce2019-01-09 09:16:05 -07009711 break;
9712 case IORING_UNREGISTER_BUFFERS:
9713 ret = -EINVAL;
9714 if (arg || nr_args)
9715 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009716 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -07009717 break;
Jens Axboe6b063142019-01-10 22:13:58 -07009718 case IORING_REGISTER_FILES:
9719 ret = io_sqe_files_register(ctx, arg, nr_args);
9720 break;
9721 case IORING_UNREGISTER_FILES:
9722 ret = -EINVAL;
9723 if (arg || nr_args)
9724 break;
9725 ret = io_sqe_files_unregister(ctx);
9726 break;
Jens Axboec3a31e62019-10-03 13:59:56 -06009727 case IORING_REGISTER_FILES_UPDATE:
9728 ret = io_sqe_files_update(ctx, arg, nr_args);
9729 break;
Jens Axboe9b402842019-04-11 11:45:41 -06009730 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -07009731 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -06009732 ret = -EINVAL;
9733 if (nr_args != 1)
9734 break;
9735 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -07009736 if (ret)
9737 break;
9738 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
9739 ctx->eventfd_async = 1;
9740 else
9741 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -06009742 break;
9743 case IORING_UNREGISTER_EVENTFD:
9744 ret = -EINVAL;
9745 if (arg || nr_args)
9746 break;
9747 ret = io_eventfd_unregister(ctx);
9748 break;
Jens Axboe66f4af92020-01-16 15:36:52 -07009749 case IORING_REGISTER_PROBE:
9750 ret = -EINVAL;
9751 if (!arg || nr_args > 256)
9752 break;
9753 ret = io_probe(ctx, arg, nr_args);
9754 break;
Jens Axboe071698e2020-01-28 10:04:42 -07009755 case IORING_REGISTER_PERSONALITY:
9756 ret = -EINVAL;
9757 if (arg || nr_args)
9758 break;
9759 ret = io_register_personality(ctx);
9760 break;
9761 case IORING_UNREGISTER_PERSONALITY:
9762 ret = -EINVAL;
9763 if (arg)
9764 break;
9765 ret = io_unregister_personality(ctx, nr_args);
9766 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009767 case IORING_REGISTER_ENABLE_RINGS:
9768 ret = -EINVAL;
9769 if (arg || nr_args)
9770 break;
9771 ret = io_register_enable_rings(ctx);
9772 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009773 case IORING_REGISTER_RESTRICTIONS:
9774 ret = io_register_restrictions(ctx, arg, nr_args);
9775 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07009776 default:
9777 ret = -EINVAL;
9778 break;
9779 }
9780
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009781out:
Jens Axboe071698e2020-01-28 10:04:42 -07009782 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009783 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -07009784 percpu_ref_reinit(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009785out_quiesce:
Jens Axboe0f158b42020-05-14 17:18:39 -06009786 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009787 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009788 return ret;
9789}
9790
9791SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
9792 void __user *, arg, unsigned int, nr_args)
9793{
9794 struct io_ring_ctx *ctx;
9795 long ret = -EBADF;
9796 struct fd f;
9797
9798 f = fdget(fd);
9799 if (!f.file)
9800 return -EBADF;
9801
9802 ret = -EOPNOTSUPP;
9803 if (f.file->f_op != &io_uring_fops)
9804 goto out_fput;
9805
9806 ctx = f.file->private_data;
9807
Pavel Begunkovb6c23dd2021-02-20 15:17:18 +00009808 io_run_task_work();
9809
Jens Axboeedafcce2019-01-09 09:16:05 -07009810 mutex_lock(&ctx->uring_lock);
9811 ret = __io_uring_register(ctx, opcode, arg, nr_args);
9812 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009813 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
9814 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07009815out_fput:
9816 fdput(f);
9817 return ret;
9818}
9819
Jens Axboe2b188cc2019-01-07 10:46:33 -07009820static int __init io_uring_init(void)
9821{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009822#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
9823 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
9824 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
9825} while (0)
9826
9827#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
9828 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
9829 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
9830 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
9831 BUILD_BUG_SQE_ELEM(1, __u8, flags);
9832 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
9833 BUILD_BUG_SQE_ELEM(4, __s32, fd);
9834 BUILD_BUG_SQE_ELEM(8, __u64, off);
9835 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
9836 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009837 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009838 BUILD_BUG_SQE_ELEM(24, __u32, len);
9839 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
9840 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
9841 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
9842 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +08009843 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
9844 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009845 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
9846 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
9847 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
9848 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
9849 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
9850 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
9851 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
9852 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009853 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009854 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
9855 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
9856 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009857 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009858
Jens Axboed3656342019-12-18 09:50:26 -07009859 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -07009860 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe91f245d2021-02-09 13:48:50 -07009861 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
9862 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009863 return 0;
9864};
9865__initcall(io_uring_init);