4 #include <linux/errno.h>
5 #include <linux/lockdep.h>
6 #include <linux/io_uring_types.h>
11 #ifndef CREATE_TRACE_POINTS
12 #include <trace/events/io_uring.h>
17 IOU_ISSUE_SKIP_COMPLETE = -EIOCBQUEUED,
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.
24 IOU_STOP_MULTISHOT = -ECANCELED,
27 struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx);
28 bool io_req_cqe_overflow(struct io_kiocb *req);
29 int io_run_task_work_sig(void);
30 void io_req_complete_failed(struct io_kiocb *req, s32 res);
31 void __io_req_complete(struct io_kiocb *req, unsigned issue_flags);
32 void io_req_complete_post(struct io_kiocb *req);
33 void __io_req_complete_post(struct io_kiocb *req);
34 bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags,
36 bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags,
38 void __io_commit_cqring_flush(struct io_ring_ctx *ctx);
40 struct page **io_pin_pages(unsigned long ubuf, unsigned long len, int *npages);
42 struct file *io_file_get_normal(struct io_kiocb *req, int fd);
43 struct file *io_file_get_fixed(struct io_kiocb *req, int fd,
44 unsigned issue_flags);
46 static inline bool io_req_ffs_set(struct io_kiocb *req)
48 return req->flags & REQ_F_FIXED_FILE;
51 bool io_is_uring_fops(struct file *file);
52 bool io_alloc_async_data(struct io_kiocb *req);
53 void io_req_task_work_add(struct io_kiocb *req);
54 void io_req_tw_post_queue(struct io_kiocb *req, s32 res, u32 cflags);
55 void io_req_task_queue(struct io_kiocb *req);
56 void io_queue_iowq(struct io_kiocb *req, bool *dont_use);
57 void io_req_task_complete(struct io_kiocb *req, bool *locked);
58 void io_req_task_queue_fail(struct io_kiocb *req, int ret);
59 void io_req_task_submit(struct io_kiocb *req, bool *locked);
60 void tctx_task_work(struct callback_head *cb);
61 __cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd);
62 int io_uring_alloc_task_context(struct task_struct *task,
63 struct io_ring_ctx *ctx);
65 int io_poll_issue(struct io_kiocb *req, bool *locked);
66 int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr);
67 int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin);
68 void io_free_batch_list(struct io_ring_ctx *ctx, struct io_wq_work_node *node);
69 int io_req_prep_async(struct io_kiocb *req);
71 struct io_wq_work *io_wq_free_work(struct io_wq_work *work);
72 void io_wq_submit_work(struct io_wq_work *work);
74 void io_free_req(struct io_kiocb *req);
75 void io_queue_next(struct io_kiocb *req);
76 void __io_put_task(struct task_struct *task, int nr);
77 void io_task_refs_refill(struct io_uring_task *tctx);
78 bool __io_alloc_req_refill(struct io_ring_ctx *ctx);
80 bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task,
83 #define io_for_each_link(pos, head) \
84 for (pos = (head); pos; pos = pos->link)
86 static inline void io_cq_lock(struct io_ring_ctx *ctx)
87 __acquires(ctx->completion_lock)
89 spin_lock(&ctx->completion_lock);
92 void io_cq_unlock_post(struct io_ring_ctx *ctx);
94 static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx)
96 if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) {
97 struct io_uring_cqe *cqe = ctx->cqe_cached;
99 ctx->cached_cq_tail++;
101 if (ctx->flags & IORING_SETUP_CQE32)
106 return __io_get_cqe(ctx);
109 static inline bool __io_fill_cqe_req(struct io_ring_ctx *ctx,
110 struct io_kiocb *req)
112 struct io_uring_cqe *cqe;
115 * If we can't get a cq entry, userspace overflowed the
116 * submission (by quite a lot). Increment the overflow count in
119 cqe = io_get_cqe(ctx);
121 return io_req_cqe_overflow(req);
123 trace_io_uring_complete(req->ctx, req, req->cqe.user_data,
124 req->cqe.res, req->cqe.flags,
125 (req->flags & REQ_F_CQE32_INIT) ? req->extra1 : 0,
126 (req->flags & REQ_F_CQE32_INIT) ? req->extra2 : 0);
128 memcpy(cqe, &req->cqe, sizeof(*cqe));
130 if (ctx->flags & IORING_SETUP_CQE32) {
131 u64 extra1 = 0, extra2 = 0;
133 if (req->flags & REQ_F_CQE32_INIT) {
134 extra1 = req->extra1;
135 extra2 = req->extra2;
138 WRITE_ONCE(cqe->big_cqe[0], extra1);
139 WRITE_ONCE(cqe->big_cqe[1], extra2);
144 static inline void req_set_fail(struct io_kiocb *req)
146 req->flags |= REQ_F_FAIL;
147 if (req->flags & REQ_F_CQE_SKIP) {
148 req->flags &= ~REQ_F_CQE_SKIP;
149 req->flags |= REQ_F_SKIP_LINK_CQES;
153 static inline void io_req_set_res(struct io_kiocb *req, s32 res, u32 cflags)
156 req->cqe.flags = cflags;
159 static inline bool req_has_async_data(struct io_kiocb *req)
161 return req->flags & REQ_F_ASYNC_DATA;
164 static inline void io_put_file(struct file *file)
170 static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx,
171 unsigned issue_flags)
173 lockdep_assert_held(&ctx->uring_lock);
174 if (issue_flags & IO_URING_F_UNLOCKED)
175 mutex_unlock(&ctx->uring_lock);
178 static inline void io_ring_submit_lock(struct io_ring_ctx *ctx,
179 unsigned issue_flags)
182 * "Normal" inline submissions always hold the uring_lock, since we
183 * grab it from the system call. Same is true for the SQPOLL offload.
184 * The only exception is when we've detached the request and issue it
185 * from an async worker thread, grab the lock for that case.
187 if (issue_flags & IO_URING_F_UNLOCKED)
188 mutex_lock(&ctx->uring_lock);
189 lockdep_assert_held(&ctx->uring_lock);
192 static inline void io_commit_cqring(struct io_ring_ctx *ctx)
194 /* order cqe stores with ring update */
195 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
198 static inline void io_cqring_wake(struct io_ring_ctx *ctx)
201 * wake_up_all() may seem excessive, but io_wake_function() and
202 * io_should_wake() handle the termination of the loop and only
203 * wake as many waiters as we need to.
205 if (wq_has_sleeper(&ctx->cq_wait))
206 wake_up_all(&ctx->cq_wait);
209 static inline bool io_sqring_full(struct io_ring_ctx *ctx)
211 struct io_rings *r = ctx->rings;
213 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries;
216 static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
218 struct io_rings *rings = ctx->rings;
220 /* make sure SQ entry isn't read before tail */
221 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
224 static inline bool io_run_task_work(void)
226 if (test_thread_flag(TIF_NOTIFY_SIGNAL)) {
227 __set_current_state(TASK_RUNNING);
228 clear_notify_signal();
229 if (task_work_pending(current))
237 static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked)
240 mutex_lock(&ctx->uring_lock);
246 * Don't complete immediately but use deferred completion infrastructure.
247 * Protected by ->uring_lock and can only be used either with
248 * IO_URING_F_COMPLETE_DEFER or inside a tw handler holding the mutex.
250 static inline void io_req_complete_defer(struct io_kiocb *req)
251 __must_hold(&req->ctx->uring_lock)
253 struct io_submit_state *state = &req->ctx->submit_state;
255 lockdep_assert_held(&req->ctx->uring_lock);
257 wq_list_add_tail(&req->comp_list, &state->compl_reqs);
260 static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx)
262 if (unlikely(ctx->off_timeout_used || ctx->drain_active || ctx->has_evfd))
263 __io_commit_cqring_flush(ctx);
266 /* must to be called somewhat shortly after putting a request */
267 static inline void io_put_task(struct task_struct *task, int nr)
269 if (likely(task == current))
270 task->io_uring->cached_refs += nr;
272 __io_put_task(task, nr);
275 static inline void io_get_task_refs(int nr)
277 struct io_uring_task *tctx = current->io_uring;
279 tctx->cached_refs -= nr;
280 if (unlikely(tctx->cached_refs < 0))
281 io_task_refs_refill(tctx);
284 static inline bool io_req_cache_empty(struct io_ring_ctx *ctx)
286 return !ctx->submit_state.free_list.next;
289 static inline bool io_alloc_req_refill(struct io_ring_ctx *ctx)
291 if (unlikely(io_req_cache_empty(ctx)))
292 return __io_alloc_req_refill(ctx);
296 static inline struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
298 struct io_wq_work_node *node;
300 node = wq_stack_extract(&ctx->submit_state.free_list);
301 return container_of(node, struct io_kiocb, comp_list);