1de23077eSJens Axboe #ifndef IOU_CORE_H 2de23077eSJens Axboe #define IOU_CORE_H 3de23077eSJens Axboe 4de23077eSJens Axboe #include <linux/errno.h> 5cd40cae2SJens Axboe #include <linux/lockdep.h> 6b5d3ae20SJens Axboe #include <linux/resume_user_mode.h> 7c1755c25SBreno Leitao #include <linux/kasan.h> 8ab1c84d8SPavel Begunkov #include <linux/io_uring_types.h> 944648532SJens Axboe #include <uapi/linux/eventpoll.h> 10ab1c84d8SPavel Begunkov #include "io-wq.h" 11a6b21fbbSPavel Begunkov #include "slist.h" 12ab1c84d8SPavel Begunkov #include "filetable.h" 13de23077eSJens Axboe 14f3b44f92SJens Axboe #ifndef CREATE_TRACE_POINTS 15f3b44f92SJens Axboe #include <trace/events/io_uring.h> 16f3b44f92SJens Axboe #endif 17f3b44f92SJens Axboe 1897b388d7SJens Axboe enum { 198751d154SPavel Begunkov /* 208751d154SPavel Begunkov * A hint to not wake right away but delay until there are enough of 218751d154SPavel Begunkov * tw's queued to match the number of CQEs the task is waiting for. 228751d154SPavel Begunkov * 238751d154SPavel Begunkov * Must not be used wirh requests generating more than one CQE. 248751d154SPavel Begunkov * It's also ignored unless IORING_SETUP_DEFER_TASKRUN is set. 258751d154SPavel Begunkov */ 26*91c7884aSPavel Begunkov IOU_F_TWQ_LAZY_WAKE = 1, 278501fe70SPavel Begunkov }; 288501fe70SPavel Begunkov 298501fe70SPavel Begunkov enum { 3097b388d7SJens Axboe IOU_OK = 0, 3197b388d7SJens Axboe IOU_ISSUE_SKIP_COMPLETE = -EIOCBQUEUED, 32114eccdfSDylan Yudaken 33114eccdfSDylan Yudaken /* 3491482864SPavel Begunkov * Intended only when both IO_URING_F_MULTISHOT is passed 3591482864SPavel Begunkov * to indicate to the poll runner that multishot should be 36114eccdfSDylan Yudaken * removed and the result is set on req->cqe.res. 37114eccdfSDylan Yudaken */ 38114eccdfSDylan Yudaken IOU_STOP_MULTISHOT = -ECANCELED, 3997b388d7SJens Axboe }; 4097b388d7SJens Axboe 41aa1df3a3SPavel Begunkov struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx, bool overflow); 4268494a65SPavel Begunkov bool io_req_cqe_overflow(struct io_kiocb *req); 43c0e0d6baSDylan Yudaken int io_run_task_work_sig(struct io_ring_ctx *ctx); 44973fc83fSDylan Yudaken void io_req_defer_failed(struct io_kiocb *req, s32 res); 451bec951cSPavel Begunkov void io_req_complete_post(struct io_kiocb *req, unsigned issue_flags); 46b529c96aSDylan Yudaken bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags); 47d86eaed1SJens Axboe bool io_aux_cqe(const struct io_kiocb *req, bool defer, s32 res, u32 cflags, 48eb42cebbSPavel Begunkov bool allow_overflow); 499046c641SPavel Begunkov void __io_commit_cqring_flush(struct io_ring_ctx *ctx); 509046c641SPavel Begunkov 519046c641SPavel Begunkov struct page **io_pin_pages(unsigned long ubuf, unsigned long len, int *npages); 529046c641SPavel Begunkov 539046c641SPavel Begunkov struct file *io_file_get_normal(struct io_kiocb *req, int fd); 549046c641SPavel Begunkov struct file *io_file_get_fixed(struct io_kiocb *req, int fd, 559046c641SPavel Begunkov unsigned issue_flags); 569046c641SPavel Begunkov 578501fe70SPavel Begunkov void __io_req_task_work_add(struct io_kiocb *req, unsigned flags); 589046c641SPavel Begunkov bool io_is_uring_fops(struct file *file); 599046c641SPavel Begunkov bool io_alloc_async_data(struct io_kiocb *req); 609046c641SPavel Begunkov void io_req_task_queue(struct io_kiocb *req); 61a282967cSPavel Begunkov void io_queue_iowq(struct io_kiocb *req, struct io_tw_state *ts_dont_use); 62a282967cSPavel Begunkov void io_req_task_complete(struct io_kiocb *req, struct io_tw_state *ts); 639046c641SPavel Begunkov void io_req_task_queue_fail(struct io_kiocb *req, int ret); 64a282967cSPavel Begunkov void io_req_task_submit(struct io_kiocb *req, struct io_tw_state *ts); 659046c641SPavel Begunkov void tctx_task_work(struct callback_head *cb); 669046c641SPavel Begunkov __cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); 679046c641SPavel Begunkov int io_uring_alloc_task_context(struct task_struct *task, 689046c641SPavel Begunkov struct io_ring_ctx *ctx); 699046c641SPavel Begunkov 706e76ac59SJosh Triplett int io_ring_add_registered_file(struct io_uring_task *tctx, struct file *file, 716e76ac59SJosh Triplett int start, int end); 726e76ac59SJosh Triplett 73a282967cSPavel Begunkov int io_poll_issue(struct io_kiocb *req, struct io_tw_state *ts); 749046c641SPavel Begunkov int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr); 759046c641SPavel Begunkov int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin); 769046c641SPavel Begunkov void io_free_batch_list(struct io_ring_ctx *ctx, struct io_wq_work_node *node); 779046c641SPavel Begunkov int io_req_prep_async(struct io_kiocb *req); 789046c641SPavel Begunkov 799046c641SPavel Begunkov struct io_wq_work *io_wq_free_work(struct io_wq_work *work); 809046c641SPavel Begunkov void io_wq_submit_work(struct io_wq_work *work); 819046c641SPavel Begunkov 829046c641SPavel Begunkov void io_free_req(struct io_kiocb *req); 839046c641SPavel Begunkov void io_queue_next(struct io_kiocb *req); 8463809137SPavel Begunkov void io_task_refs_refill(struct io_uring_task *tctx); 85bd1a3783SPavel Begunkov bool __io_alloc_req_refill(struct io_ring_ctx *ctx); 869046c641SPavel Begunkov 879046c641SPavel Begunkov bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, 889046c641SPavel Begunkov bool cancel_all); 899046c641SPavel Begunkov 90f26cc959SPavel Begunkov #define io_lockdep_assert_cq_locked(ctx) \ 91f26cc959SPavel Begunkov do { \ 928ce4269eSPavel Begunkov lockdep_assert(in_task()); \ 938ce4269eSPavel Begunkov \ 94f26cc959SPavel Begunkov if (ctx->flags & IORING_SETUP_IOPOLL) { \ 95f26cc959SPavel Begunkov lockdep_assert_held(&ctx->uring_lock); \ 96f26cc959SPavel Begunkov } else if (!ctx->task_complete) { \ 97f26cc959SPavel Begunkov lockdep_assert_held(&ctx->completion_lock); \ 98f26cc959SPavel Begunkov } else if (ctx->submitter_task->flags & PF_EXITING) { \ 99f26cc959SPavel Begunkov lockdep_assert(current_work()); \ 100f26cc959SPavel Begunkov } else { \ 101f26cc959SPavel Begunkov lockdep_assert(current == ctx->submitter_task); \ 102f26cc959SPavel Begunkov } \ 103f26cc959SPavel Begunkov } while (0) 104f26cc959SPavel Begunkov 105e52d2e58SPavel Begunkov static inline void io_req_task_work_add(struct io_kiocb *req) 106e52d2e58SPavel Begunkov { 1078501fe70SPavel Begunkov __io_req_task_work_add(req, 0); 108e52d2e58SPavel Begunkov } 109e52d2e58SPavel Begunkov 1109046c641SPavel Begunkov #define io_for_each_link(pos, head) \ 1119046c641SPavel Begunkov for (pos = (head); pos; pos = pos->link) 112f3b44f92SJens Axboe 11325399321SPavel Begunkov void io_cq_unlock_post(struct io_ring_ctx *ctx); 11425399321SPavel Begunkov 115aa1df3a3SPavel Begunkov static inline struct io_uring_cqe *io_get_cqe_overflow(struct io_ring_ctx *ctx, 116aa1df3a3SPavel Begunkov bool overflow) 117f3b44f92SJens Axboe { 118f26cc959SPavel Begunkov io_lockdep_assert_cq_locked(ctx); 119f26cc959SPavel Begunkov 120f3b44f92SJens Axboe if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) { 121f3b44f92SJens Axboe struct io_uring_cqe *cqe = ctx->cqe_cached; 122f3b44f92SJens Axboe 123f3b44f92SJens Axboe ctx->cached_cq_tail++; 124f3b44f92SJens Axboe ctx->cqe_cached++; 125b3659a65SPavel Begunkov if (ctx->flags & IORING_SETUP_CQE32) 126b3659a65SPavel Begunkov ctx->cqe_cached++; 127f3b44f92SJens Axboe return cqe; 128f3b44f92SJens Axboe } 129f3b44f92SJens Axboe 130aa1df3a3SPavel Begunkov return __io_get_cqe(ctx, overflow); 131aa1df3a3SPavel Begunkov } 132aa1df3a3SPavel Begunkov 133aa1df3a3SPavel Begunkov static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) 134aa1df3a3SPavel Begunkov { 135aa1df3a3SPavel Begunkov return io_get_cqe_overflow(ctx, false); 136f3b44f92SJens Axboe } 137f3b44f92SJens Axboe 138f3b44f92SJens Axboe static inline bool __io_fill_cqe_req(struct io_ring_ctx *ctx, 139f3b44f92SJens Axboe struct io_kiocb *req) 140f3b44f92SJens Axboe { 141f3b44f92SJens Axboe struct io_uring_cqe *cqe; 142f3b44f92SJens Axboe 143f3b44f92SJens Axboe /* 144f3b44f92SJens Axboe * If we can't get a cq entry, userspace overflowed the 145f3b44f92SJens Axboe * submission (by quite a lot). Increment the overflow count in 146f3b44f92SJens Axboe * the ring. 147f3b44f92SJens Axboe */ 148f3b44f92SJens Axboe cqe = io_get_cqe(ctx); 149e8c328c3SPavel Begunkov if (unlikely(!cqe)) 150f66f7342SPavel Begunkov return false; 151e0486f3fSDylan Yudaken 152e0486f3fSDylan Yudaken trace_io_uring_complete(req->ctx, req, req->cqe.user_data, 153e0486f3fSDylan Yudaken req->cqe.res, req->cqe.flags, 154e0486f3fSDylan Yudaken (req->flags & REQ_F_CQE32_INIT) ? req->extra1 : 0, 155e0486f3fSDylan Yudaken (req->flags & REQ_F_CQE32_INIT) ? req->extra2 : 0); 156e0486f3fSDylan Yudaken 157f3b44f92SJens Axboe memcpy(cqe, &req->cqe, sizeof(*cqe)); 158e8c328c3SPavel Begunkov 159e8c328c3SPavel Begunkov if (ctx->flags & IORING_SETUP_CQE32) { 160f3b44f92SJens Axboe u64 extra1 = 0, extra2 = 0; 161f3b44f92SJens Axboe 162f3b44f92SJens Axboe if (req->flags & REQ_F_CQE32_INIT) { 163f3b44f92SJens Axboe extra1 = req->extra1; 164f3b44f92SJens Axboe extra2 = req->extra2; 165f3b44f92SJens Axboe } 166f3b44f92SJens Axboe 167f3b44f92SJens Axboe WRITE_ONCE(cqe->big_cqe[0], extra1); 168f3b44f92SJens Axboe WRITE_ONCE(cqe->big_cqe[1], extra2); 169e8c328c3SPavel Begunkov } 170f3b44f92SJens Axboe return true; 171f3b44f92SJens Axboe } 172f3b44f92SJens Axboe 173f66f7342SPavel Begunkov static inline bool io_fill_cqe_req(struct io_ring_ctx *ctx, 174f66f7342SPavel Begunkov struct io_kiocb *req) 175f66f7342SPavel Begunkov { 176f66f7342SPavel Begunkov if (likely(__io_fill_cqe_req(ctx, req))) 177f66f7342SPavel Begunkov return true; 178f66f7342SPavel Begunkov return io_req_cqe_overflow(req); 179f66f7342SPavel Begunkov } 180f66f7342SPavel Begunkov 181531113bbSJens Axboe static inline void req_set_fail(struct io_kiocb *req) 182531113bbSJens Axboe { 183531113bbSJens Axboe req->flags |= REQ_F_FAIL; 184531113bbSJens Axboe if (req->flags & REQ_F_CQE_SKIP) { 185531113bbSJens Axboe req->flags &= ~REQ_F_CQE_SKIP; 186531113bbSJens Axboe req->flags |= REQ_F_SKIP_LINK_CQES; 187531113bbSJens Axboe } 188531113bbSJens Axboe } 189531113bbSJens Axboe 190de23077eSJens Axboe static inline void io_req_set_res(struct io_kiocb *req, s32 res, u32 cflags) 191de23077eSJens Axboe { 192de23077eSJens Axboe req->cqe.res = res; 193de23077eSJens Axboe req->cqe.flags = cflags; 194de23077eSJens Axboe } 195de23077eSJens Axboe 19699f15d8dSJens Axboe static inline bool req_has_async_data(struct io_kiocb *req) 19799f15d8dSJens Axboe { 19899f15d8dSJens Axboe return req->flags & REQ_F_ASYNC_DATA; 19999f15d8dSJens Axboe } 20099f15d8dSJens Axboe 201531113bbSJens Axboe static inline void io_put_file(struct file *file) 202531113bbSJens Axboe { 203531113bbSJens Axboe if (file) 204531113bbSJens Axboe fput(file); 205531113bbSJens Axboe } 206531113bbSJens Axboe 207cd40cae2SJens Axboe static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx, 208cd40cae2SJens Axboe unsigned issue_flags) 209cd40cae2SJens Axboe { 210cd40cae2SJens Axboe lockdep_assert_held(&ctx->uring_lock); 211cd40cae2SJens Axboe if (issue_flags & IO_URING_F_UNLOCKED) 212cd40cae2SJens Axboe mutex_unlock(&ctx->uring_lock); 213cd40cae2SJens Axboe } 214cd40cae2SJens Axboe 215cd40cae2SJens Axboe static inline void io_ring_submit_lock(struct io_ring_ctx *ctx, 216cd40cae2SJens Axboe unsigned issue_flags) 217cd40cae2SJens Axboe { 218cd40cae2SJens Axboe /* 219cd40cae2SJens Axboe * "Normal" inline submissions always hold the uring_lock, since we 220cd40cae2SJens Axboe * grab it from the system call. Same is true for the SQPOLL offload. 221cd40cae2SJens Axboe * The only exception is when we've detached the request and issue it 222cd40cae2SJens Axboe * from an async worker thread, grab the lock for that case. 223cd40cae2SJens Axboe */ 224cd40cae2SJens Axboe if (issue_flags & IO_URING_F_UNLOCKED) 225cd40cae2SJens Axboe mutex_lock(&ctx->uring_lock); 226cd40cae2SJens Axboe lockdep_assert_held(&ctx->uring_lock); 227cd40cae2SJens Axboe } 228cd40cae2SJens Axboe 229f9ead18cSJens Axboe static inline void io_commit_cqring(struct io_ring_ctx *ctx) 230f9ead18cSJens Axboe { 231f9ead18cSJens Axboe /* order cqe stores with ring update */ 232f9ead18cSJens Axboe smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); 233f9ead18cSJens Axboe } 234f9ead18cSJens Axboe 2357b235dd8SPavel Begunkov static inline void io_poll_wq_wake(struct io_ring_ctx *ctx) 2367b235dd8SPavel Begunkov { 237bca39f39SPavel Begunkov if (wq_has_sleeper(&ctx->poll_wq)) 2387b235dd8SPavel Begunkov __wake_up(&ctx->poll_wq, TASK_NORMAL, 0, 2397b235dd8SPavel Begunkov poll_to_key(EPOLL_URING_WAKE | EPOLLIN)); 2407b235dd8SPavel Begunkov } 2417b235dd8SPavel Begunkov 2426e7248adSPavel Begunkov static inline void io_cqring_wake(struct io_ring_ctx *ctx) 243f3b44f92SJens Axboe { 244f3b44f92SJens Axboe /* 24544648532SJens Axboe * Trigger waitqueue handler on all waiters on our waitqueue. This 24644648532SJens Axboe * won't necessarily wake up all the tasks, io_should_wake() will make 24744648532SJens Axboe * that decision. 24844648532SJens Axboe * 24944648532SJens Axboe * Pass in EPOLLIN|EPOLL_URING_WAKE as the poll wakeup key. The latter 25044648532SJens Axboe * set in the mask so that if we recurse back into our own poll 25144648532SJens Axboe * waitqueue handlers, we know we have a dependency between eventfd or 25244648532SJens Axboe * epoll and should terminate multishot poll at that point. 253f3b44f92SJens Axboe */ 2546e7248adSPavel Begunkov if (wq_has_sleeper(&ctx->cq_wait)) 25544648532SJens Axboe __wake_up(&ctx->cq_wait, TASK_NORMAL, 0, 25644648532SJens Axboe poll_to_key(EPOLL_URING_WAKE | EPOLLIN)); 257f3b44f92SJens Axboe } 258f3b44f92SJens Axboe 25917437f31SJens Axboe static inline bool io_sqring_full(struct io_ring_ctx *ctx) 26017437f31SJens Axboe { 26117437f31SJens Axboe struct io_rings *r = ctx->rings; 26217437f31SJens Axboe 26317437f31SJens Axboe return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; 26417437f31SJens Axboe } 26517437f31SJens Axboe 26617437f31SJens Axboe static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) 26717437f31SJens Axboe { 26817437f31SJens Axboe struct io_rings *rings = ctx->rings; 269e3ef728fSJens Axboe unsigned int entries; 27017437f31SJens Axboe 27117437f31SJens Axboe /* make sure SQ entry isn't read before tail */ 272e3ef728fSJens Axboe entries = smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; 273e3ef728fSJens Axboe return min(entries, ctx->sq_entries); 27417437f31SJens Axboe } 27517437f31SJens Axboe 276c0e0d6baSDylan Yudaken static inline int io_run_task_work(void) 27717437f31SJens Axboe { 2787cfe7a09SJens Axboe /* 2797cfe7a09SJens Axboe * Always check-and-clear the task_work notification signal. With how 2807cfe7a09SJens Axboe * signaling works for task_work, we can find it set with nothing to 2817cfe7a09SJens Axboe * run. We need to clear it for that case, like get_signal() does. 2827cfe7a09SJens Axboe */ 28346a525e1SJens Axboe if (test_thread_flag(TIF_NOTIFY_SIGNAL)) 28417437f31SJens Axboe clear_notify_signal(); 285b5d3ae20SJens Axboe /* 286b5d3ae20SJens Axboe * PF_IO_WORKER never returns to userspace, so check here if we have 287b5d3ae20SJens Axboe * notify work that needs processing. 288b5d3ae20SJens Axboe */ 289b5d3ae20SJens Axboe if (current->flags & PF_IO_WORKER && 2902f2bb1ffSJens Axboe test_thread_flag(TIF_NOTIFY_RESUME)) { 2912f2bb1ffSJens Axboe __set_current_state(TASK_RUNNING); 292b5d3ae20SJens Axboe resume_user_mode_work(NULL); 2932f2bb1ffSJens Axboe } 2947cfe7a09SJens Axboe if (task_work_pending(current)) { 29546a525e1SJens Axboe __set_current_state(TASK_RUNNING); 29617437f31SJens Axboe task_work_run(); 297c0e0d6baSDylan Yudaken return 1; 29817437f31SJens Axboe } 29917437f31SJens Axboe 300c0e0d6baSDylan Yudaken return 0; 301c0e0d6baSDylan Yudaken } 302c0e0d6baSDylan Yudaken 303dac6a0eaSJens Axboe static inline bool io_task_work_pending(struct io_ring_ctx *ctx) 304dac6a0eaSJens Axboe { 3056434ec01SJens Axboe return task_work_pending(current) || !wq_list_empty(&ctx->work_llist); 306dac6a0eaSJens Axboe } 307dac6a0eaSJens Axboe 308a282967cSPavel Begunkov static inline void io_tw_lock(struct io_ring_ctx *ctx, struct io_tw_state *ts) 309aa1e90f6SPavel Begunkov { 310a282967cSPavel Begunkov if (!ts->locked) { 311aa1e90f6SPavel Begunkov mutex_lock(&ctx->uring_lock); 312a282967cSPavel Begunkov ts->locked = true; 313aa1e90f6SPavel Begunkov } 314aa1e90f6SPavel Begunkov } 315aa1e90f6SPavel Begunkov 3169da070b1SPavel Begunkov /* 3179da070b1SPavel Begunkov * Don't complete immediately but use deferred completion infrastructure. 3189da070b1SPavel Begunkov * Protected by ->uring_lock and can only be used either with 3199da070b1SPavel Begunkov * IO_URING_F_COMPLETE_DEFER or inside a tw handler holding the mutex. 3209da070b1SPavel Begunkov */ 3219da070b1SPavel Begunkov static inline void io_req_complete_defer(struct io_kiocb *req) 3229da070b1SPavel Begunkov __must_hold(&req->ctx->uring_lock) 323aa1e90f6SPavel Begunkov { 324aa1e90f6SPavel Begunkov struct io_submit_state *state = &req->ctx->submit_state; 325aa1e90f6SPavel Begunkov 3269da070b1SPavel Begunkov lockdep_assert_held(&req->ctx->uring_lock); 3279da070b1SPavel Begunkov 328aa1e90f6SPavel Begunkov wq_list_add_tail(&req->comp_list, &state->compl_reqs); 329aa1e90f6SPavel Begunkov } 330aa1e90f6SPavel Begunkov 33146929b08SPavel Begunkov static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx) 33246929b08SPavel Begunkov { 333bca39f39SPavel Begunkov if (unlikely(ctx->off_timeout_used || ctx->drain_active || 334bca39f39SPavel Begunkov ctx->has_evfd || ctx->poll_activated)) 33546929b08SPavel Begunkov __io_commit_cqring_flush(ctx); 33646929b08SPavel Begunkov } 33746929b08SPavel Begunkov 33863809137SPavel Begunkov static inline void io_get_task_refs(int nr) 33963809137SPavel Begunkov { 34063809137SPavel Begunkov struct io_uring_task *tctx = current->io_uring; 34163809137SPavel Begunkov 34263809137SPavel Begunkov tctx->cached_refs -= nr; 34363809137SPavel Begunkov if (unlikely(tctx->cached_refs < 0)) 34463809137SPavel Begunkov io_task_refs_refill(tctx); 34563809137SPavel Begunkov } 34663809137SPavel Begunkov 347bd1a3783SPavel Begunkov static inline bool io_req_cache_empty(struct io_ring_ctx *ctx) 348bd1a3783SPavel Begunkov { 349bd1a3783SPavel Begunkov return !ctx->submit_state.free_list.next; 350bd1a3783SPavel Begunkov } 351bd1a3783SPavel Begunkov 352c1755c25SBreno Leitao extern struct kmem_cache *req_cachep; 353c1755c25SBreno Leitao 354c8576f3eSPavel Begunkov static inline struct io_kiocb *io_extract_req(struct io_ring_ctx *ctx) 355bd1a3783SPavel Begunkov { 356c1755c25SBreno Leitao struct io_kiocb *req; 357bd1a3783SPavel Begunkov 358c1755c25SBreno Leitao req = container_of(ctx->submit_state.free_list.next, struct io_kiocb, comp_list); 359c1755c25SBreno Leitao kasan_unpoison_object_data(req_cachep, req); 360c1755c25SBreno Leitao wq_stack_extract(&ctx->submit_state.free_list); 361c1755c25SBreno Leitao return req; 362bd1a3783SPavel Begunkov } 363bd1a3783SPavel Begunkov 364c8576f3eSPavel Begunkov static inline bool io_alloc_req(struct io_ring_ctx *ctx, struct io_kiocb **req) 365c8576f3eSPavel Begunkov { 366c8576f3eSPavel Begunkov if (unlikely(io_req_cache_empty(ctx))) { 367c8576f3eSPavel Begunkov if (!__io_alloc_req_refill(ctx)) 368c8576f3eSPavel Begunkov return false; 369c8576f3eSPavel Begunkov } 370c8576f3eSPavel Begunkov *req = io_extract_req(ctx); 371c8576f3eSPavel Begunkov return true; 372c8576f3eSPavel Begunkov } 373c8576f3eSPavel Begunkov 374140102aeSPavel Begunkov static inline bool io_allowed_defer_tw_run(struct io_ring_ctx *ctx) 375140102aeSPavel Begunkov { 376140102aeSPavel Begunkov return likely(ctx->submitter_task == current); 377140102aeSPavel Begunkov } 378140102aeSPavel Begunkov 37976de6749SPavel Begunkov static inline bool io_allowed_run_tw(struct io_ring_ctx *ctx) 38076de6749SPavel Begunkov { 3816567506bSPavel Begunkov return likely(!(ctx->flags & IORING_SETUP_DEFER_TASKRUN) || 3826567506bSPavel Begunkov ctx->submitter_task == current); 38376de6749SPavel Begunkov } 38476de6749SPavel Begunkov 385833b5dffSPavel Begunkov static inline void io_req_queue_tw_complete(struct io_kiocb *req, s32 res) 386833b5dffSPavel Begunkov { 387833b5dffSPavel Begunkov io_req_set_res(req, res, 0); 388833b5dffSPavel Begunkov req->io_task_work.func = io_req_task_complete; 389833b5dffSPavel Begunkov io_req_task_work_add(req); 390833b5dffSPavel Begunkov } 391833b5dffSPavel Begunkov 39296c7d4f8SBreno Leitao /* 39396c7d4f8SBreno Leitao * IORING_SETUP_SQE128 contexts allocate twice the normal SQE size for each 39496c7d4f8SBreno Leitao * slot. 39596c7d4f8SBreno Leitao */ 39696c7d4f8SBreno Leitao static inline size_t uring_sqe_size(struct io_ring_ctx *ctx) 39796c7d4f8SBreno Leitao { 39896c7d4f8SBreno Leitao if (ctx->flags & IORING_SETUP_SQE128) 39996c7d4f8SBreno Leitao return 2 * sizeof(struct io_uring_sqe); 40096c7d4f8SBreno Leitao return sizeof(struct io_uring_sqe); 40196c7d4f8SBreno Leitao } 402de23077eSJens Axboe #endif 403