Commit | Line | Data |
---|---|---|
de23077e JA |
1 | #ifndef IOU_CORE_H |
2 | #define IOU_CORE_H | |
3 | ||
4 | #include <linux/errno.h> | |
cd40cae2 | 5 | #include <linux/lockdep.h> |
b5d3ae20 | 6 | #include <linux/resume_user_mode.h> |
c1755c25 | 7 | #include <linux/kasan.h> |
ab1c84d8 | 8 | #include <linux/io_uring_types.h> |
44648532 | 9 | #include <uapi/linux/eventpoll.h> |
ab1c84d8 | 10 | #include "io-wq.h" |
a6b21fbb | 11 | #include "slist.h" |
ab1c84d8 | 12 | #include "filetable.h" |
de23077e | 13 | |
f3b44f92 JA |
14 | #ifndef CREATE_TRACE_POINTS |
15 | #include <trace/events/io_uring.h> | |
16 | #endif | |
17 | ||
8501fe70 PB |
18 | enum { |
19 | /* don't use deferred task_work */ | |
20 | IOU_F_TWQ_FORCE_NORMAL = 1, | |
8751d154 PB |
21 | |
22 | /* | |
23 | * A hint to not wake right away but delay until there are enough of | |
24 | * tw's queued to match the number of CQEs the task is waiting for. | |
25 | * | |
26 | * Must not be used wirh requests generating more than one CQE. | |
27 | * It's also ignored unless IORING_SETUP_DEFER_TASKRUN is set. | |
28 | */ | |
29 | IOU_F_TWQ_LAZY_WAKE = 2, | |
8501fe70 PB |
30 | }; |
31 | ||
97b388d7 JA |
32 | enum { |
33 | IOU_OK = 0, | |
34 | IOU_ISSUE_SKIP_COMPLETE = -EIOCBQUEUED, | |
114eccdf DY |
35 | |
36 | /* | |
91482864 PB |
37 | * Intended only when both IO_URING_F_MULTISHOT is passed |
38 | * to indicate to the poll runner that multishot should be | |
114eccdf DY |
39 | * removed and the result is set on req->cqe.res. |
40 | */ | |
41 | IOU_STOP_MULTISHOT = -ECANCELED, | |
97b388d7 JA |
42 | }; |
43 | ||
aa1df3a3 | 44 | struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx, bool overflow); |
68494a65 | 45 | bool io_req_cqe_overflow(struct io_kiocb *req); |
c0e0d6ba | 46 | int io_run_task_work_sig(struct io_ring_ctx *ctx); |
973fc83f | 47 | void io_req_defer_failed(struct io_kiocb *req, s32 res); |
1bec951c | 48 | void io_req_complete_post(struct io_kiocb *req, unsigned issue_flags); |
b529c96a | 49 | bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags); |
9b8c5475 DY |
50 | bool io_aux_cqe(struct io_ring_ctx *ctx, bool defer, u64 user_data, s32 res, u32 cflags, |
51 | bool allow_overflow); | |
9046c641 PB |
52 | void __io_commit_cqring_flush(struct io_ring_ctx *ctx); |
53 | ||
54 | struct page **io_pin_pages(unsigned long ubuf, unsigned long len, int *npages); | |
55 | ||
56 | struct file *io_file_get_normal(struct io_kiocb *req, int fd); | |
57 | struct file *io_file_get_fixed(struct io_kiocb *req, int fd, | |
58 | unsigned issue_flags); | |
59 | ||
f6b543fd JA |
60 | static inline bool io_req_ffs_set(struct io_kiocb *req) |
61 | { | |
62 | return req->flags & REQ_F_FIXED_FILE; | |
63 | } | |
64 | ||
8501fe70 | 65 | void __io_req_task_work_add(struct io_kiocb *req, unsigned flags); |
9046c641 PB |
66 | bool io_is_uring_fops(struct file *file); |
67 | bool io_alloc_async_data(struct io_kiocb *req); | |
9046c641 | 68 | void io_req_task_queue(struct io_kiocb *req); |
a282967c PB |
69 | void io_queue_iowq(struct io_kiocb *req, struct io_tw_state *ts_dont_use); |
70 | void io_req_task_complete(struct io_kiocb *req, struct io_tw_state *ts); | |
9046c641 | 71 | void io_req_task_queue_fail(struct io_kiocb *req, int ret); |
a282967c | 72 | void io_req_task_submit(struct io_kiocb *req, struct io_tw_state *ts); |
9046c641 PB |
73 | void tctx_task_work(struct callback_head *cb); |
74 | __cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); | |
75 | int io_uring_alloc_task_context(struct task_struct *task, | |
76 | struct io_ring_ctx *ctx); | |
77 | ||
a282967c | 78 | int io_poll_issue(struct io_kiocb *req, struct io_tw_state *ts); |
9046c641 PB |
79 | int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr); |
80 | int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin); | |
81 | void io_free_batch_list(struct io_ring_ctx *ctx, struct io_wq_work_node *node); | |
82 | int io_req_prep_async(struct io_kiocb *req); | |
83 | ||
84 | struct io_wq_work *io_wq_free_work(struct io_wq_work *work); | |
85 | void io_wq_submit_work(struct io_wq_work *work); | |
86 | ||
87 | void io_free_req(struct io_kiocb *req); | |
88 | void io_queue_next(struct io_kiocb *req); | |
63809137 | 89 | void io_task_refs_refill(struct io_uring_task *tctx); |
bd1a3783 | 90 | bool __io_alloc_req_refill(struct io_ring_ctx *ctx); |
9046c641 PB |
91 | |
92 | bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, | |
93 | bool cancel_all); | |
94 | ||
f26cc959 PB |
95 | #define io_lockdep_assert_cq_locked(ctx) \ |
96 | do { \ | |
8ce4269e PB |
97 | lockdep_assert(in_task()); \ |
98 | \ | |
f26cc959 PB |
99 | if (ctx->flags & IORING_SETUP_IOPOLL) { \ |
100 | lockdep_assert_held(&ctx->uring_lock); \ | |
101 | } else if (!ctx->task_complete) { \ | |
102 | lockdep_assert_held(&ctx->completion_lock); \ | |
103 | } else if (ctx->submitter_task->flags & PF_EXITING) { \ | |
104 | lockdep_assert(current_work()); \ | |
105 | } else { \ | |
106 | lockdep_assert(current == ctx->submitter_task); \ | |
107 | } \ | |
108 | } while (0) | |
109 | ||
e52d2e58 PB |
110 | static inline void io_req_task_work_add(struct io_kiocb *req) |
111 | { | |
8501fe70 | 112 | __io_req_task_work_add(req, 0); |
e52d2e58 PB |
113 | } |
114 | ||
9046c641 PB |
115 | #define io_for_each_link(pos, head) \ |
116 | for (pos = (head); pos; pos = pos->link) | |
f3b44f92 | 117 | |
25399321 PB |
118 | void io_cq_unlock_post(struct io_ring_ctx *ctx); |
119 | ||
aa1df3a3 PB |
120 | static inline struct io_uring_cqe *io_get_cqe_overflow(struct io_ring_ctx *ctx, |
121 | bool overflow) | |
f3b44f92 | 122 | { |
f26cc959 PB |
123 | io_lockdep_assert_cq_locked(ctx); |
124 | ||
f3b44f92 JA |
125 | if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) { |
126 | struct io_uring_cqe *cqe = ctx->cqe_cached; | |
127 | ||
f3b44f92 JA |
128 | ctx->cached_cq_tail++; |
129 | ctx->cqe_cached++; | |
b3659a65 PB |
130 | if (ctx->flags & IORING_SETUP_CQE32) |
131 | ctx->cqe_cached++; | |
f3b44f92 JA |
132 | return cqe; |
133 | } | |
134 | ||
aa1df3a3 PB |
135 | return __io_get_cqe(ctx, overflow); |
136 | } | |
137 | ||
138 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) | |
139 | { | |
140 | return io_get_cqe_overflow(ctx, false); | |
f3b44f92 JA |
141 | } |
142 | ||
143 | static inline bool __io_fill_cqe_req(struct io_ring_ctx *ctx, | |
144 | struct io_kiocb *req) | |
145 | { | |
146 | struct io_uring_cqe *cqe; | |
147 | ||
e8c328c3 PB |
148 | /* |
149 | * If we can't get a cq entry, userspace overflowed the | |
150 | * submission (by quite a lot). Increment the overflow count in | |
151 | * the ring. | |
152 | */ | |
153 | cqe = io_get_cqe(ctx); | |
154 | if (unlikely(!cqe)) | |
f66f7342 | 155 | return false; |
e0486f3f DY |
156 | |
157 | trace_io_uring_complete(req->ctx, req, req->cqe.user_data, | |
158 | req->cqe.res, req->cqe.flags, | |
159 | (req->flags & REQ_F_CQE32_INIT) ? req->extra1 : 0, | |
160 | (req->flags & REQ_F_CQE32_INIT) ? req->extra2 : 0); | |
161 | ||
e8c328c3 PB |
162 | memcpy(cqe, &req->cqe, sizeof(*cqe)); |
163 | ||
164 | if (ctx->flags & IORING_SETUP_CQE32) { | |
f3b44f92 JA |
165 | u64 extra1 = 0, extra2 = 0; |
166 | ||
167 | if (req->flags & REQ_F_CQE32_INIT) { | |
168 | extra1 = req->extra1; | |
169 | extra2 = req->extra2; | |
170 | } | |
171 | ||
e8c328c3 PB |
172 | WRITE_ONCE(cqe->big_cqe[0], extra1); |
173 | WRITE_ONCE(cqe->big_cqe[1], extra2); | |
f3b44f92 | 174 | } |
e8c328c3 | 175 | return true; |
f3b44f92 JA |
176 | } |
177 | ||
f66f7342 PB |
178 | static inline bool io_fill_cqe_req(struct io_ring_ctx *ctx, |
179 | struct io_kiocb *req) | |
180 | { | |
181 | if (likely(__io_fill_cqe_req(ctx, req))) | |
182 | return true; | |
183 | return io_req_cqe_overflow(req); | |
184 | } | |
185 | ||
531113bb JA |
186 | static inline void req_set_fail(struct io_kiocb *req) |
187 | { | |
188 | req->flags |= REQ_F_FAIL; | |
189 | if (req->flags & REQ_F_CQE_SKIP) { | |
190 | req->flags &= ~REQ_F_CQE_SKIP; | |
191 | req->flags |= REQ_F_SKIP_LINK_CQES; | |
192 | } | |
193 | } | |
194 | ||
de23077e JA |
195 | static inline void io_req_set_res(struct io_kiocb *req, s32 res, u32 cflags) |
196 | { | |
197 | req->cqe.res = res; | |
198 | req->cqe.flags = cflags; | |
199 | } | |
200 | ||
99f15d8d JA |
201 | static inline bool req_has_async_data(struct io_kiocb *req) |
202 | { | |
203 | return req->flags & REQ_F_ASYNC_DATA; | |
204 | } | |
205 | ||
531113bb JA |
206 | static inline void io_put_file(struct file *file) |
207 | { | |
208 | if (file) | |
209 | fput(file); | |
210 | } | |
211 | ||
cd40cae2 JA |
212 | static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx, |
213 | unsigned issue_flags) | |
214 | { | |
215 | lockdep_assert_held(&ctx->uring_lock); | |
216 | if (issue_flags & IO_URING_F_UNLOCKED) | |
217 | mutex_unlock(&ctx->uring_lock); | |
218 | } | |
219 | ||
220 | static inline void io_ring_submit_lock(struct io_ring_ctx *ctx, | |
221 | unsigned issue_flags) | |
222 | { | |
223 | /* | |
224 | * "Normal" inline submissions always hold the uring_lock, since we | |
225 | * grab it from the system call. Same is true for the SQPOLL offload. | |
226 | * The only exception is when we've detached the request and issue it | |
227 | * from an async worker thread, grab the lock for that case. | |
228 | */ | |
229 | if (issue_flags & IO_URING_F_UNLOCKED) | |
230 | mutex_lock(&ctx->uring_lock); | |
231 | lockdep_assert_held(&ctx->uring_lock); | |
232 | } | |
233 | ||
f9ead18c JA |
234 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) |
235 | { | |
236 | /* order cqe stores with ring update */ | |
237 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
238 | } | |
239 | ||
7b235dd8 PB |
240 | static inline void io_poll_wq_wake(struct io_ring_ctx *ctx) |
241 | { | |
bca39f39 | 242 | if (wq_has_sleeper(&ctx->poll_wq)) |
7b235dd8 PB |
243 | __wake_up(&ctx->poll_wq, TASK_NORMAL, 0, |
244 | poll_to_key(EPOLL_URING_WAKE | EPOLLIN)); | |
245 | } | |
246 | ||
6e7248ad | 247 | static inline void io_cqring_wake(struct io_ring_ctx *ctx) |
f3b44f92 JA |
248 | { |
249 | /* | |
44648532 JA |
250 | * Trigger waitqueue handler on all waiters on our waitqueue. This |
251 | * won't necessarily wake up all the tasks, io_should_wake() will make | |
252 | * that decision. | |
253 | * | |
254 | * Pass in EPOLLIN|EPOLL_URING_WAKE as the poll wakeup key. The latter | |
255 | * set in the mask so that if we recurse back into our own poll | |
256 | * waitqueue handlers, we know we have a dependency between eventfd or | |
257 | * epoll and should terminate multishot poll at that point. | |
f3b44f92 | 258 | */ |
6e7248ad | 259 | if (wq_has_sleeper(&ctx->cq_wait)) |
44648532 JA |
260 | __wake_up(&ctx->cq_wait, TASK_NORMAL, 0, |
261 | poll_to_key(EPOLL_URING_WAKE | EPOLLIN)); | |
f3b44f92 JA |
262 | } |
263 | ||
17437f31 JA |
264 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
265 | { | |
266 | struct io_rings *r = ctx->rings; | |
267 | ||
268 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; | |
269 | } | |
270 | ||
271 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) | |
272 | { | |
273 | struct io_rings *rings = ctx->rings; | |
e3ef728f | 274 | unsigned int entries; |
17437f31 JA |
275 | |
276 | /* make sure SQ entry isn't read before tail */ | |
e3ef728f JA |
277 | entries = smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; |
278 | return min(entries, ctx->sq_entries); | |
17437f31 JA |
279 | } |
280 | ||
c0e0d6ba | 281 | static inline int io_run_task_work(void) |
17437f31 | 282 | { |
7cfe7a09 JA |
283 | /* |
284 | * Always check-and-clear the task_work notification signal. With how | |
285 | * signaling works for task_work, we can find it set with nothing to | |
286 | * run. We need to clear it for that case, like get_signal() does. | |
287 | */ | |
288 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) | |
289 | clear_notify_signal(); | |
b5d3ae20 JA |
290 | /* |
291 | * PF_IO_WORKER never returns to userspace, so check here if we have | |
292 | * notify work that needs processing. | |
293 | */ | |
294 | if (current->flags & PF_IO_WORKER && | |
2f2bb1ff JA |
295 | test_thread_flag(TIF_NOTIFY_RESUME)) { |
296 | __set_current_state(TASK_RUNNING); | |
b5d3ae20 | 297 | resume_user_mode_work(NULL); |
2f2bb1ff | 298 | } |
46a525e1 | 299 | if (task_work_pending(current)) { |
17437f31 | 300 | __set_current_state(TASK_RUNNING); |
46a525e1 | 301 | task_work_run(); |
c0e0d6ba | 302 | return 1; |
17437f31 JA |
303 | } |
304 | ||
c0e0d6ba DY |
305 | return 0; |
306 | } | |
307 | ||
dac6a0ea JA |
308 | static inline bool io_task_work_pending(struct io_ring_ctx *ctx) |
309 | { | |
6434ec01 | 310 | return task_work_pending(current) || !wq_list_empty(&ctx->work_llist); |
dac6a0ea JA |
311 | } |
312 | ||
a282967c | 313 | static inline void io_tw_lock(struct io_ring_ctx *ctx, struct io_tw_state *ts) |
aa1e90f6 | 314 | { |
a282967c | 315 | if (!ts->locked) { |
aa1e90f6 | 316 | mutex_lock(&ctx->uring_lock); |
a282967c | 317 | ts->locked = true; |
aa1e90f6 PB |
318 | } |
319 | } | |
320 | ||
9da070b1 PB |
321 | /* |
322 | * Don't complete immediately but use deferred completion infrastructure. | |
323 | * Protected by ->uring_lock and can only be used either with | |
324 | * IO_URING_F_COMPLETE_DEFER or inside a tw handler holding the mutex. | |
325 | */ | |
326 | static inline void io_req_complete_defer(struct io_kiocb *req) | |
327 | __must_hold(&req->ctx->uring_lock) | |
aa1e90f6 PB |
328 | { |
329 | struct io_submit_state *state = &req->ctx->submit_state; | |
330 | ||
9da070b1 PB |
331 | lockdep_assert_held(&req->ctx->uring_lock); |
332 | ||
aa1e90f6 PB |
333 | wq_list_add_tail(&req->comp_list, &state->compl_reqs); |
334 | } | |
335 | ||
46929b08 PB |
336 | static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx) |
337 | { | |
bca39f39 PB |
338 | if (unlikely(ctx->off_timeout_used || ctx->drain_active || |
339 | ctx->has_evfd || ctx->poll_activated)) | |
46929b08 PB |
340 | __io_commit_cqring_flush(ctx); |
341 | } | |
342 | ||
63809137 PB |
343 | static inline void io_get_task_refs(int nr) |
344 | { | |
345 | struct io_uring_task *tctx = current->io_uring; | |
346 | ||
347 | tctx->cached_refs -= nr; | |
348 | if (unlikely(tctx->cached_refs < 0)) | |
349 | io_task_refs_refill(tctx); | |
350 | } | |
351 | ||
bd1a3783 PB |
352 | static inline bool io_req_cache_empty(struct io_ring_ctx *ctx) |
353 | { | |
354 | return !ctx->submit_state.free_list.next; | |
355 | } | |
356 | ||
c1755c25 BL |
357 | extern struct kmem_cache *req_cachep; |
358 | ||
c8576f3e | 359 | static inline struct io_kiocb *io_extract_req(struct io_ring_ctx *ctx) |
bd1a3783 | 360 | { |
c1755c25 | 361 | struct io_kiocb *req; |
bd1a3783 | 362 | |
c1755c25 BL |
363 | req = container_of(ctx->submit_state.free_list.next, struct io_kiocb, comp_list); |
364 | kasan_unpoison_object_data(req_cachep, req); | |
365 | wq_stack_extract(&ctx->submit_state.free_list); | |
366 | return req; | |
bd1a3783 PB |
367 | } |
368 | ||
c8576f3e PB |
369 | static inline bool io_alloc_req(struct io_ring_ctx *ctx, struct io_kiocb **req) |
370 | { | |
371 | if (unlikely(io_req_cache_empty(ctx))) { | |
372 | if (!__io_alloc_req_refill(ctx)) | |
373 | return false; | |
374 | } | |
375 | *req = io_extract_req(ctx); | |
376 | return true; | |
377 | } | |
378 | ||
140102ae PB |
379 | static inline bool io_allowed_defer_tw_run(struct io_ring_ctx *ctx) |
380 | { | |
381 | return likely(ctx->submitter_task == current); | |
382 | } | |
383 | ||
76de6749 PB |
384 | static inline bool io_allowed_run_tw(struct io_ring_ctx *ctx) |
385 | { | |
6567506b PB |
386 | return likely(!(ctx->flags & IORING_SETUP_DEFER_TASKRUN) || |
387 | ctx->submitter_task == current); | |
76de6749 PB |
388 | } |
389 | ||
833b5dff PB |
390 | static inline void io_req_queue_tw_complete(struct io_kiocb *req, s32 res) |
391 | { | |
392 | io_req_set_res(req, res, 0); | |
393 | req->io_task_work.func = io_req_task_complete; | |
394 | io_req_task_work_add(req); | |
395 | } | |
396 | ||
96c7d4f8 BL |
397 | /* |
398 | * IORING_SETUP_SQE128 contexts allocate twice the normal SQE size for each | |
399 | * slot. | |
400 | */ | |
401 | static inline size_t uring_sqe_size(struct io_ring_ctx *ctx) | |
402 | { | |
403 | if (ctx->flags & IORING_SETUP_SQE128) | |
404 | return 2 * sizeof(struct io_uring_sqe); | |
405 | return sizeof(struct io_uring_sqe); | |
406 | } | |
de23077e | 407 | #endif |