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> |
ab1c84d8 PB |
6 | #include <linux/io_uring_types.h> |
7 | #include "io-wq.h" | |
a6b21fbb | 8 | #include "slist.h" |
ab1c84d8 | 9 | #include "filetable.h" |
de23077e | 10 | |
f3b44f92 JA |
11 | #ifndef CREATE_TRACE_POINTS |
12 | #include <trace/events/io_uring.h> | |
13 | #endif | |
14 | ||
97b388d7 JA |
15 | enum { |
16 | IOU_OK = 0, | |
17 | IOU_ISSUE_SKIP_COMPLETE = -EIOCBQUEUED, | |
114eccdf DY |
18 | |
19 | /* | |
20 | * Intended only when both REQ_F_POLLED and REQ_F_APOLL_MULTISHOT | |
21 | * are set to indicate to the poll runner that multishot should be | |
22 | * removed and the result is set on req->cqe.res. | |
23 | */ | |
24 | IOU_STOP_MULTISHOT = -ECANCELED, | |
97b388d7 JA |
25 | }; |
26 | ||
aa1df3a3 | 27 | struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx, bool overflow); |
68494a65 | 28 | bool io_req_cqe_overflow(struct io_kiocb *req); |
c0e0d6ba | 29 | int io_run_task_work_sig(struct io_ring_ctx *ctx); |
8ac5d85a | 30 | int __io_run_local_work(struct io_ring_ctx *ctx, bool locked); |
c0e0d6ba | 31 | int io_run_local_work(struct io_ring_ctx *ctx); |
9046c641 PB |
32 | void io_req_complete_failed(struct io_kiocb *req, s32 res); |
33 | void __io_req_complete(struct io_kiocb *req, unsigned issue_flags); | |
34 | void io_req_complete_post(struct io_kiocb *req); | |
35 | void __io_req_complete_post(struct io_kiocb *req); | |
52120f0f DY |
36 | bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags, |
37 | bool allow_overflow); | |
eb42cebb PB |
38 | bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags, |
39 | bool allow_overflow); | |
9046c641 PB |
40 | void __io_commit_cqring_flush(struct io_ring_ctx *ctx); |
41 | ||
42 | struct page **io_pin_pages(unsigned long ubuf, unsigned long len, int *npages); | |
43 | ||
44 | struct file *io_file_get_normal(struct io_kiocb *req, int fd); | |
45 | struct file *io_file_get_fixed(struct io_kiocb *req, int fd, | |
46 | unsigned issue_flags); | |
47 | ||
f6b543fd JA |
48 | static inline bool io_req_ffs_set(struct io_kiocb *req) |
49 | { | |
50 | return req->flags & REQ_F_FIXED_FILE; | |
51 | } | |
52 | ||
9046c641 PB |
53 | bool io_is_uring_fops(struct file *file); |
54 | bool io_alloc_async_data(struct io_kiocb *req); | |
55 | void io_req_task_work_add(struct io_kiocb *req); | |
9046c641 PB |
56 | void io_req_tw_post_queue(struct io_kiocb *req, s32 res, u32 cflags); |
57 | void io_req_task_queue(struct io_kiocb *req); | |
58 | void io_queue_iowq(struct io_kiocb *req, bool *dont_use); | |
59 | void io_req_task_complete(struct io_kiocb *req, bool *locked); | |
60 | void io_req_task_queue_fail(struct io_kiocb *req, int ret); | |
61 | void io_req_task_submit(struct io_kiocb *req, bool *locked); | |
62 | void tctx_task_work(struct callback_head *cb); | |
63 | __cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); | |
64 | int io_uring_alloc_task_context(struct task_struct *task, | |
65 | struct io_ring_ctx *ctx); | |
66 | ||
67 | int io_poll_issue(struct io_kiocb *req, bool *locked); | |
68 | int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr); | |
69 | int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin); | |
70 | void io_free_batch_list(struct io_ring_ctx *ctx, struct io_wq_work_node *node); | |
71 | int io_req_prep_async(struct io_kiocb *req); | |
72 | ||
73 | struct io_wq_work *io_wq_free_work(struct io_wq_work *work); | |
74 | void io_wq_submit_work(struct io_wq_work *work); | |
75 | ||
76 | void io_free_req(struct io_kiocb *req); | |
77 | void io_queue_next(struct io_kiocb *req); | |
e70cb608 | 78 | void __io_put_task(struct task_struct *task, int nr); |
63809137 | 79 | void io_task_refs_refill(struct io_uring_task *tctx); |
bd1a3783 | 80 | bool __io_alloc_req_refill(struct io_ring_ctx *ctx); |
9046c641 PB |
81 | |
82 | bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, | |
83 | bool cancel_all); | |
84 | ||
85 | #define io_for_each_link(pos, head) \ | |
86 | for (pos = (head); pos; pos = pos->link) | |
f3b44f92 | 87 | |
25399321 PB |
88 | static inline void io_cq_lock(struct io_ring_ctx *ctx) |
89 | __acquires(ctx->completion_lock) | |
90 | { | |
91 | spin_lock(&ctx->completion_lock); | |
92 | } | |
93 | ||
94 | void io_cq_unlock_post(struct io_ring_ctx *ctx); | |
95 | ||
aa1df3a3 PB |
96 | static inline struct io_uring_cqe *io_get_cqe_overflow(struct io_ring_ctx *ctx, |
97 | bool overflow) | |
f3b44f92 JA |
98 | { |
99 | if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) { | |
100 | struct io_uring_cqe *cqe = ctx->cqe_cached; | |
101 | ||
f3b44f92 JA |
102 | ctx->cached_cq_tail++; |
103 | ctx->cqe_cached++; | |
b3659a65 PB |
104 | if (ctx->flags & IORING_SETUP_CQE32) |
105 | ctx->cqe_cached++; | |
f3b44f92 JA |
106 | return cqe; |
107 | } | |
108 | ||
aa1df3a3 PB |
109 | return __io_get_cqe(ctx, overflow); |
110 | } | |
111 | ||
112 | static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) | |
113 | { | |
114 | return io_get_cqe_overflow(ctx, false); | |
f3b44f92 JA |
115 | } |
116 | ||
117 | static inline bool __io_fill_cqe_req(struct io_ring_ctx *ctx, | |
118 | struct io_kiocb *req) | |
119 | { | |
120 | struct io_uring_cqe *cqe; | |
121 | ||
e8c328c3 PB |
122 | /* |
123 | * If we can't get a cq entry, userspace overflowed the | |
124 | * submission (by quite a lot). Increment the overflow count in | |
125 | * the ring. | |
126 | */ | |
127 | cqe = io_get_cqe(ctx); | |
128 | if (unlikely(!cqe)) | |
129 | return io_req_cqe_overflow(req); | |
e0486f3f DY |
130 | |
131 | trace_io_uring_complete(req->ctx, req, req->cqe.user_data, | |
132 | req->cqe.res, req->cqe.flags, | |
133 | (req->flags & REQ_F_CQE32_INIT) ? req->extra1 : 0, | |
134 | (req->flags & REQ_F_CQE32_INIT) ? req->extra2 : 0); | |
135 | ||
e8c328c3 PB |
136 | memcpy(cqe, &req->cqe, sizeof(*cqe)); |
137 | ||
138 | if (ctx->flags & IORING_SETUP_CQE32) { | |
f3b44f92 JA |
139 | u64 extra1 = 0, extra2 = 0; |
140 | ||
141 | if (req->flags & REQ_F_CQE32_INIT) { | |
142 | extra1 = req->extra1; | |
143 | extra2 = req->extra2; | |
144 | } | |
145 | ||
e8c328c3 PB |
146 | WRITE_ONCE(cqe->big_cqe[0], extra1); |
147 | WRITE_ONCE(cqe->big_cqe[1], extra2); | |
f3b44f92 | 148 | } |
e8c328c3 | 149 | return true; |
f3b44f92 JA |
150 | } |
151 | ||
531113bb JA |
152 | static inline void req_set_fail(struct io_kiocb *req) |
153 | { | |
154 | req->flags |= REQ_F_FAIL; | |
155 | if (req->flags & REQ_F_CQE_SKIP) { | |
156 | req->flags &= ~REQ_F_CQE_SKIP; | |
157 | req->flags |= REQ_F_SKIP_LINK_CQES; | |
158 | } | |
159 | } | |
160 | ||
de23077e JA |
161 | static inline void io_req_set_res(struct io_kiocb *req, s32 res, u32 cflags) |
162 | { | |
163 | req->cqe.res = res; | |
164 | req->cqe.flags = cflags; | |
165 | } | |
166 | ||
99f15d8d JA |
167 | static inline bool req_has_async_data(struct io_kiocb *req) |
168 | { | |
169 | return req->flags & REQ_F_ASYNC_DATA; | |
170 | } | |
171 | ||
531113bb JA |
172 | static inline void io_put_file(struct file *file) |
173 | { | |
174 | if (file) | |
175 | fput(file); | |
176 | } | |
177 | ||
cd40cae2 JA |
178 | static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx, |
179 | unsigned issue_flags) | |
180 | { | |
181 | lockdep_assert_held(&ctx->uring_lock); | |
182 | if (issue_flags & IO_URING_F_UNLOCKED) | |
183 | mutex_unlock(&ctx->uring_lock); | |
184 | } | |
185 | ||
186 | static inline void io_ring_submit_lock(struct io_ring_ctx *ctx, | |
187 | unsigned issue_flags) | |
188 | { | |
189 | /* | |
190 | * "Normal" inline submissions always hold the uring_lock, since we | |
191 | * grab it from the system call. Same is true for the SQPOLL offload. | |
192 | * The only exception is when we've detached the request and issue it | |
193 | * from an async worker thread, grab the lock for that case. | |
194 | */ | |
195 | if (issue_flags & IO_URING_F_UNLOCKED) | |
196 | mutex_lock(&ctx->uring_lock); | |
197 | lockdep_assert_held(&ctx->uring_lock); | |
198 | } | |
199 | ||
f9ead18c JA |
200 | static inline void io_commit_cqring(struct io_ring_ctx *ctx) |
201 | { | |
202 | /* order cqe stores with ring update */ | |
203 | smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); | |
204 | } | |
205 | ||
f3b44f92 JA |
206 | static inline void io_cqring_wake(struct io_ring_ctx *ctx) |
207 | { | |
208 | /* | |
209 | * wake_up_all() may seem excessive, but io_wake_function() and | |
210 | * io_should_wake() handle the termination of the loop and only | |
211 | * wake as many waiters as we need to. | |
212 | */ | |
213 | if (wq_has_sleeper(&ctx->cq_wait)) | |
214 | wake_up_all(&ctx->cq_wait); | |
215 | } | |
216 | ||
17437f31 JA |
217 | static inline bool io_sqring_full(struct io_ring_ctx *ctx) |
218 | { | |
219 | struct io_rings *r = ctx->rings; | |
220 | ||
221 | return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; | |
222 | } | |
223 | ||
224 | static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) | |
225 | { | |
226 | struct io_rings *rings = ctx->rings; | |
227 | ||
228 | /* make sure SQ entry isn't read before tail */ | |
229 | return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; | |
230 | } | |
231 | ||
c0e0d6ba | 232 | static inline int io_run_task_work(void) |
17437f31 | 233 | { |
625d38b3 | 234 | if (test_thread_flag(TIF_NOTIFY_SIGNAL)) { |
17437f31 JA |
235 | __set_current_state(TASK_RUNNING); |
236 | clear_notify_signal(); | |
237 | if (task_work_pending(current)) | |
238 | task_work_run(); | |
c0e0d6ba | 239 | return 1; |
17437f31 JA |
240 | } |
241 | ||
c0e0d6ba DY |
242 | return 0; |
243 | } | |
244 | ||
dac6a0ea JA |
245 | static inline bool io_task_work_pending(struct io_ring_ctx *ctx) |
246 | { | |
247 | return test_thread_flag(TIF_NOTIFY_SIGNAL) || | |
248 | !wq_list_empty(&ctx->work_llist); | |
249 | } | |
250 | ||
c0e0d6ba DY |
251 | static inline int io_run_task_work_ctx(struct io_ring_ctx *ctx) |
252 | { | |
253 | int ret = 0; | |
254 | int ret2; | |
255 | ||
256 | if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) | |
257 | ret = io_run_local_work(ctx); | |
258 | ||
259 | /* want to run this after in case more is added */ | |
260 | ret2 = io_run_task_work(); | |
261 | ||
262 | /* Try propagate error in favour of if tasks were run, | |
263 | * but still make sure to run them if requested | |
264 | */ | |
265 | if (ret >= 0) | |
266 | ret += ret2; | |
267 | ||
268 | return ret; | |
17437f31 JA |
269 | } |
270 | ||
aa1e90f6 PB |
271 | static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked) |
272 | { | |
273 | if (!*locked) { | |
274 | mutex_lock(&ctx->uring_lock); | |
275 | *locked = true; | |
276 | } | |
277 | } | |
278 | ||
9da070b1 PB |
279 | /* |
280 | * Don't complete immediately but use deferred completion infrastructure. | |
281 | * Protected by ->uring_lock and can only be used either with | |
282 | * IO_URING_F_COMPLETE_DEFER or inside a tw handler holding the mutex. | |
283 | */ | |
284 | static inline void io_req_complete_defer(struct io_kiocb *req) | |
285 | __must_hold(&req->ctx->uring_lock) | |
aa1e90f6 PB |
286 | { |
287 | struct io_submit_state *state = &req->ctx->submit_state; | |
288 | ||
9da070b1 PB |
289 | lockdep_assert_held(&req->ctx->uring_lock); |
290 | ||
aa1e90f6 PB |
291 | wq_list_add_tail(&req->comp_list, &state->compl_reqs); |
292 | } | |
293 | ||
46929b08 PB |
294 | static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx) |
295 | { | |
296 | if (unlikely(ctx->off_timeout_used || ctx->drain_active || ctx->has_evfd)) | |
297 | __io_commit_cqring_flush(ctx); | |
298 | } | |
299 | ||
e70cb608 PB |
300 | /* must to be called somewhat shortly after putting a request */ |
301 | static inline void io_put_task(struct task_struct *task, int nr) | |
302 | { | |
303 | if (likely(task == current)) | |
304 | task->io_uring->cached_refs += nr; | |
305 | else | |
306 | __io_put_task(task, nr); | |
307 | } | |
308 | ||
63809137 PB |
309 | static inline void io_get_task_refs(int nr) |
310 | { | |
311 | struct io_uring_task *tctx = current->io_uring; | |
312 | ||
313 | tctx->cached_refs -= nr; | |
314 | if (unlikely(tctx->cached_refs < 0)) | |
315 | io_task_refs_refill(tctx); | |
316 | } | |
317 | ||
bd1a3783 PB |
318 | static inline bool io_req_cache_empty(struct io_ring_ctx *ctx) |
319 | { | |
320 | return !ctx->submit_state.free_list.next; | |
321 | } | |
322 | ||
323 | static inline bool io_alloc_req_refill(struct io_ring_ctx *ctx) | |
324 | { | |
325 | if (unlikely(io_req_cache_empty(ctx))) | |
326 | return __io_alloc_req_refill(ctx); | |
327 | return true; | |
328 | } | |
329 | ||
330 | static inline struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx) | |
331 | { | |
332 | struct io_wq_work_node *node; | |
333 | ||
334 | node = wq_stack_extract(&ctx->submit_state.free_list); | |
335 | return container_of(node, struct io_kiocb, comp_list); | |
336 | } | |
337 | ||
76de6749 PB |
338 | static inline bool io_allowed_run_tw(struct io_ring_ctx *ctx) |
339 | { | |
6567506b PB |
340 | return likely(!(ctx->flags & IORING_SETUP_DEFER_TASKRUN) || |
341 | ctx->submitter_task == current); | |
76de6749 PB |
342 | } |
343 | ||
de23077e | 344 | #endif |