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 */ 2691c7884aSPavel 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 /* 34*7cbd3aa5SJens Axboe * Requeue the task_work to restart operations on this request. The 35*7cbd3aa5SJens Axboe * actual value isn't important, should just be not an otherwise 36*7cbd3aa5SJens Axboe * valid error code, yet less than -MAX_ERRNO and valid internally. 37*7cbd3aa5SJens Axboe */ 38*7cbd3aa5SJens Axboe IOU_REQUEUE = -3072, 39*7cbd3aa5SJens Axboe 40*7cbd3aa5SJens Axboe /* 4191482864SPavel Begunkov * Intended only when both IO_URING_F_MULTISHOT is passed 4291482864SPavel Begunkov * to indicate to the poll runner that multishot should be 43114eccdfSDylan Yudaken * removed and the result is set on req->cqe.res. 44114eccdfSDylan Yudaken */ 45114eccdfSDylan Yudaken IOU_STOP_MULTISHOT = -ECANCELED, 4697b388d7SJens Axboe }; 4797b388d7SJens Axboe 4820d6b633SPavel Begunkov bool io_cqe_cache_refill(struct io_ring_ctx *ctx, bool overflow); 49056695bfSPavel Begunkov void io_req_cqe_overflow(struct io_kiocb *req); 50c0e0d6baSDylan Yudaken int io_run_task_work_sig(struct io_ring_ctx *ctx); 51973fc83fSDylan Yudaken void io_req_defer_failed(struct io_kiocb *req, s32 res); 521bec951cSPavel Begunkov void io_req_complete_post(struct io_kiocb *req, unsigned issue_flags); 53b529c96aSDylan Yudaken bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags); 54b6b2bb58SPavel Begunkov bool io_fill_cqe_req_aux(struct io_kiocb *req, bool defer, s32 res, u32 cflags); 559046c641SPavel Begunkov void __io_commit_cqring_flush(struct io_ring_ctx *ctx); 569046c641SPavel Begunkov 579046c641SPavel Begunkov struct page **io_pin_pages(unsigned long ubuf, unsigned long len, int *npages); 589046c641SPavel Begunkov 599046c641SPavel Begunkov struct file *io_file_get_normal(struct io_kiocb *req, int fd); 609046c641SPavel Begunkov struct file *io_file_get_fixed(struct io_kiocb *req, int fd, 619046c641SPavel Begunkov unsigned issue_flags); 629046c641SPavel Begunkov 638501fe70SPavel Begunkov void __io_req_task_work_add(struct io_kiocb *req, unsigned flags); 649046c641SPavel Begunkov bool io_is_uring_fops(struct file *file); 659046c641SPavel Begunkov bool io_alloc_async_data(struct io_kiocb *req); 669046c641SPavel Begunkov void io_req_task_queue(struct io_kiocb *req); 67a282967cSPavel Begunkov void io_queue_iowq(struct io_kiocb *req, struct io_tw_state *ts_dont_use); 68a282967cSPavel Begunkov void io_req_task_complete(struct io_kiocb *req, struct io_tw_state *ts); 699046c641SPavel Begunkov void io_req_task_queue_fail(struct io_kiocb *req, int ret); 70a282967cSPavel Begunkov void io_req_task_submit(struct io_kiocb *req, struct io_tw_state *ts); 719046c641SPavel Begunkov void tctx_task_work(struct callback_head *cb); 729046c641SPavel Begunkov __cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); 739046c641SPavel Begunkov int io_uring_alloc_task_context(struct task_struct *task, 749046c641SPavel Begunkov struct io_ring_ctx *ctx); 759046c641SPavel Begunkov 766e76ac59SJosh Triplett int io_ring_add_registered_file(struct io_uring_task *tctx, struct file *file, 776e76ac59SJosh Triplett int start, int end); 786e76ac59SJosh Triplett 79a282967cSPavel Begunkov int io_poll_issue(struct io_kiocb *req, struct io_tw_state *ts); 809046c641SPavel Begunkov int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr); 819046c641SPavel Begunkov int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin); 82ec26c225SPavel Begunkov void __io_submit_flush_completions(struct io_ring_ctx *ctx); 839046c641SPavel Begunkov int io_req_prep_async(struct io_kiocb *req); 849046c641SPavel Begunkov 859046c641SPavel Begunkov struct io_wq_work *io_wq_free_work(struct io_wq_work *work); 869046c641SPavel Begunkov void io_wq_submit_work(struct io_wq_work *work); 879046c641SPavel Begunkov 889046c641SPavel Begunkov void io_free_req(struct io_kiocb *req); 899046c641SPavel Begunkov void io_queue_next(struct io_kiocb *req); 9063809137SPavel Begunkov void io_task_refs_refill(struct io_uring_task *tctx); 91bd1a3783SPavel Begunkov bool __io_alloc_req_refill(struct io_ring_ctx *ctx); 929046c641SPavel Begunkov 939046c641SPavel Begunkov bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, 949046c641SPavel Begunkov bool cancel_all); 959046c641SPavel Begunkov 964ca5f54aSJens Axboe void *io_mem_alloc(size_t size); 974ca5f54aSJens Axboe void io_mem_free(void *ptr); 984ca5f54aSJens Axboe 991658633cSJens Axboe #if defined(CONFIG_PROVE_LOCKING) 1001658633cSJens Axboe static inline void io_lockdep_assert_cq_locked(struct io_ring_ctx *ctx) 1011658633cSJens Axboe { 1021658633cSJens Axboe lockdep_assert(in_task()); 1031658633cSJens Axboe 1041658633cSJens Axboe if (ctx->flags & IORING_SETUP_IOPOLL) { 1051658633cSJens Axboe lockdep_assert_held(&ctx->uring_lock); 1061658633cSJens Axboe } else if (!ctx->task_complete) { 1071658633cSJens Axboe lockdep_assert_held(&ctx->completion_lock); 1081658633cSJens Axboe } else if (ctx->submitter_task) { 1091658633cSJens Axboe /* 1101658633cSJens Axboe * ->submitter_task may be NULL and we can still post a CQE, 1111658633cSJens Axboe * if the ring has been setup with IORING_SETUP_R_DISABLED. 1121658633cSJens Axboe * Not from an SQE, as those cannot be submitted, but via 1131658633cSJens Axboe * updating tagged resources. 1141658633cSJens Axboe */ 1151658633cSJens Axboe if (ctx->submitter_task->flags & PF_EXITING) 1161658633cSJens Axboe lockdep_assert(current_work()); 1171658633cSJens Axboe else 1181658633cSJens Axboe lockdep_assert(current == ctx->submitter_task); 1191658633cSJens Axboe } 1201658633cSJens Axboe } 1211658633cSJens Axboe #else 1221658633cSJens Axboe static inline void io_lockdep_assert_cq_locked(struct io_ring_ctx *ctx) 1231658633cSJens Axboe { 1241658633cSJens Axboe } 1251658633cSJens Axboe #endif 126f26cc959SPavel Begunkov 127e52d2e58SPavel Begunkov static inline void io_req_task_work_add(struct io_kiocb *req) 128e52d2e58SPavel Begunkov { 1298501fe70SPavel Begunkov __io_req_task_work_add(req, 0); 130e52d2e58SPavel Begunkov } 131e52d2e58SPavel Begunkov 1329046c641SPavel Begunkov #define io_for_each_link(pos, head) \ 1339046c641SPavel Begunkov for (pos = (head); pos; pos = pos->link) 134f3b44f92SJens Axboe 13559fbc409SPavel Begunkov static inline bool io_get_cqe_overflow(struct io_ring_ctx *ctx, 13659fbc409SPavel Begunkov struct io_uring_cqe **ret, 137aa1df3a3SPavel Begunkov bool overflow) 138f3b44f92SJens Axboe { 139f26cc959SPavel Begunkov io_lockdep_assert_cq_locked(ctx); 140f26cc959SPavel Begunkov 14120d6b633SPavel Begunkov if (unlikely(ctx->cqe_cached >= ctx->cqe_sentinel)) { 14220d6b633SPavel Begunkov if (unlikely(!io_cqe_cache_refill(ctx, overflow))) 14359fbc409SPavel Begunkov return false; 14420d6b633SPavel Begunkov } 14559fbc409SPavel Begunkov *ret = ctx->cqe_cached; 146f3b44f92SJens Axboe ctx->cached_cq_tail++; 147f3b44f92SJens Axboe ctx->cqe_cached++; 148b3659a65SPavel Begunkov if (ctx->flags & IORING_SETUP_CQE32) 149b3659a65SPavel Begunkov ctx->cqe_cached++; 15059fbc409SPavel Begunkov return true; 151f3b44f92SJens Axboe } 152f3b44f92SJens Axboe 15359fbc409SPavel Begunkov static inline bool io_get_cqe(struct io_ring_ctx *ctx, struct io_uring_cqe **ret) 154aa1df3a3SPavel Begunkov { 15559fbc409SPavel Begunkov return io_get_cqe_overflow(ctx, ret, false); 156f3b44f92SJens Axboe } 157f3b44f92SJens Axboe 158093a650bSPavel Begunkov static __always_inline bool io_fill_cqe_req(struct io_ring_ctx *ctx, 159093a650bSPavel Begunkov struct io_kiocb *req) 160f3b44f92SJens Axboe { 161f3b44f92SJens Axboe struct io_uring_cqe *cqe; 162f3b44f92SJens Axboe 163f3b44f92SJens Axboe /* 164f3b44f92SJens Axboe * If we can't get a cq entry, userspace overflowed the 165f3b44f92SJens Axboe * submission (by quite a lot). Increment the overflow count in 166f3b44f92SJens Axboe * the ring. 167f3b44f92SJens Axboe */ 16859fbc409SPavel Begunkov if (unlikely(!io_get_cqe(ctx, &cqe))) 169f66f7342SPavel Begunkov return false; 170e0486f3fSDylan Yudaken 171a0727c73SPavel Begunkov if (trace_io_uring_complete_enabled()) 172e0486f3fSDylan Yudaken trace_io_uring_complete(req->ctx, req, req->cqe.user_data, 173e0486f3fSDylan Yudaken req->cqe.res, req->cqe.flags, 174b24c5d75SPavel Begunkov req->big_cqe.extra1, req->big_cqe.extra2); 175e0486f3fSDylan Yudaken 176f3b44f92SJens Axboe memcpy(cqe, &req->cqe, sizeof(*cqe)); 177e8c328c3SPavel Begunkov if (ctx->flags & IORING_SETUP_CQE32) { 178b24c5d75SPavel Begunkov memcpy(cqe->big_cqe, &req->big_cqe, sizeof(*cqe)); 179b24c5d75SPavel Begunkov memset(&req->big_cqe, 0, sizeof(req->big_cqe)); 180e8c328c3SPavel Begunkov } 181f3b44f92SJens Axboe return true; 182f3b44f92SJens Axboe } 183f3b44f92SJens Axboe 184531113bbSJens Axboe static inline void req_set_fail(struct io_kiocb *req) 185531113bbSJens Axboe { 186531113bbSJens Axboe req->flags |= REQ_F_FAIL; 187531113bbSJens Axboe if (req->flags & REQ_F_CQE_SKIP) { 188531113bbSJens Axboe req->flags &= ~REQ_F_CQE_SKIP; 189531113bbSJens Axboe req->flags |= REQ_F_SKIP_LINK_CQES; 190531113bbSJens Axboe } 191531113bbSJens Axboe } 192531113bbSJens Axboe 193de23077eSJens Axboe static inline void io_req_set_res(struct io_kiocb *req, s32 res, u32 cflags) 194de23077eSJens Axboe { 195de23077eSJens Axboe req->cqe.res = res; 196de23077eSJens Axboe req->cqe.flags = cflags; 197de23077eSJens Axboe } 198de23077eSJens Axboe 19999f15d8dSJens Axboe static inline bool req_has_async_data(struct io_kiocb *req) 20099f15d8dSJens Axboe { 20199f15d8dSJens Axboe return req->flags & REQ_F_ASYNC_DATA; 20299f15d8dSJens Axboe } 20399f15d8dSJens Axboe 20417bc2837SJens Axboe static inline void io_put_file(struct io_kiocb *req) 205531113bbSJens Axboe { 20617bc2837SJens Axboe if (!(req->flags & REQ_F_FIXED_FILE) && req->file) 20717bc2837SJens Axboe fput(req->file); 208531113bbSJens Axboe } 209531113bbSJens Axboe 210cd40cae2SJens Axboe static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx, 211cd40cae2SJens Axboe unsigned issue_flags) 212cd40cae2SJens Axboe { 213cd40cae2SJens Axboe lockdep_assert_held(&ctx->uring_lock); 214cd40cae2SJens Axboe if (issue_flags & IO_URING_F_UNLOCKED) 215cd40cae2SJens Axboe mutex_unlock(&ctx->uring_lock); 216cd40cae2SJens Axboe } 217cd40cae2SJens Axboe 218cd40cae2SJens Axboe static inline void io_ring_submit_lock(struct io_ring_ctx *ctx, 219cd40cae2SJens Axboe unsigned issue_flags) 220cd40cae2SJens Axboe { 221cd40cae2SJens Axboe /* 222cd40cae2SJens Axboe * "Normal" inline submissions always hold the uring_lock, since we 223cd40cae2SJens Axboe * grab it from the system call. Same is true for the SQPOLL offload. 224cd40cae2SJens Axboe * The only exception is when we've detached the request and issue it 225cd40cae2SJens Axboe * from an async worker thread, grab the lock for that case. 226cd40cae2SJens Axboe */ 227cd40cae2SJens Axboe if (issue_flags & IO_URING_F_UNLOCKED) 228cd40cae2SJens Axboe mutex_lock(&ctx->uring_lock); 229cd40cae2SJens Axboe lockdep_assert_held(&ctx->uring_lock); 230cd40cae2SJens Axboe } 231cd40cae2SJens Axboe 232f9ead18cSJens Axboe static inline void io_commit_cqring(struct io_ring_ctx *ctx) 233f9ead18cSJens Axboe { 234f9ead18cSJens Axboe /* order cqe stores with ring update */ 235f9ead18cSJens Axboe smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); 236f9ead18cSJens Axboe } 237f9ead18cSJens Axboe 2387b235dd8SPavel Begunkov static inline void io_poll_wq_wake(struct io_ring_ctx *ctx) 2397b235dd8SPavel Begunkov { 240bca39f39SPavel Begunkov if (wq_has_sleeper(&ctx->poll_wq)) 2417b235dd8SPavel Begunkov __wake_up(&ctx->poll_wq, TASK_NORMAL, 0, 2427b235dd8SPavel Begunkov poll_to_key(EPOLL_URING_WAKE | EPOLLIN)); 2437b235dd8SPavel Begunkov } 2447b235dd8SPavel Begunkov 2456e7248adSPavel Begunkov static inline void io_cqring_wake(struct io_ring_ctx *ctx) 246f3b44f92SJens Axboe { 247f3b44f92SJens Axboe /* 24844648532SJens Axboe * Trigger waitqueue handler on all waiters on our waitqueue. This 24944648532SJens Axboe * won't necessarily wake up all the tasks, io_should_wake() will make 25044648532SJens Axboe * that decision. 25144648532SJens Axboe * 25244648532SJens Axboe * Pass in EPOLLIN|EPOLL_URING_WAKE as the poll wakeup key. The latter 25344648532SJens Axboe * set in the mask so that if we recurse back into our own poll 25444648532SJens Axboe * waitqueue handlers, we know we have a dependency between eventfd or 25544648532SJens Axboe * epoll and should terminate multishot poll at that point. 256f3b44f92SJens Axboe */ 2576e7248adSPavel Begunkov if (wq_has_sleeper(&ctx->cq_wait)) 25844648532SJens Axboe __wake_up(&ctx->cq_wait, TASK_NORMAL, 0, 25944648532SJens Axboe poll_to_key(EPOLL_URING_WAKE | EPOLLIN)); 260f3b44f92SJens Axboe } 261f3b44f92SJens Axboe 26217437f31SJens Axboe static inline bool io_sqring_full(struct io_ring_ctx *ctx) 26317437f31SJens Axboe { 26417437f31SJens Axboe struct io_rings *r = ctx->rings; 26517437f31SJens Axboe 26617437f31SJens Axboe return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; 26717437f31SJens Axboe } 26817437f31SJens Axboe 26917437f31SJens Axboe static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) 27017437f31SJens Axboe { 27117437f31SJens Axboe struct io_rings *rings = ctx->rings; 272e3ef728fSJens Axboe unsigned int entries; 27317437f31SJens Axboe 27417437f31SJens Axboe /* make sure SQ entry isn't read before tail */ 275e3ef728fSJens Axboe entries = smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; 276e3ef728fSJens Axboe return min(entries, ctx->sq_entries); 27717437f31SJens Axboe } 27817437f31SJens Axboe 279c0e0d6baSDylan Yudaken static inline int io_run_task_work(void) 28017437f31SJens Axboe { 2817cfe7a09SJens Axboe /* 2827cfe7a09SJens Axboe * Always check-and-clear the task_work notification signal. With how 2837cfe7a09SJens Axboe * signaling works for task_work, we can find it set with nothing to 2847cfe7a09SJens Axboe * run. We need to clear it for that case, like get_signal() does. 2857cfe7a09SJens Axboe */ 28646a525e1SJens Axboe if (test_thread_flag(TIF_NOTIFY_SIGNAL)) 28717437f31SJens Axboe clear_notify_signal(); 288b5d3ae20SJens Axboe /* 289b5d3ae20SJens Axboe * PF_IO_WORKER never returns to userspace, so check here if we have 290b5d3ae20SJens Axboe * notify work that needs processing. 291b5d3ae20SJens Axboe */ 292b5d3ae20SJens Axboe if (current->flags & PF_IO_WORKER && 2932f2bb1ffSJens Axboe test_thread_flag(TIF_NOTIFY_RESUME)) { 2942f2bb1ffSJens Axboe __set_current_state(TASK_RUNNING); 295b5d3ae20SJens Axboe resume_user_mode_work(NULL); 2962f2bb1ffSJens Axboe } 2977cfe7a09SJens Axboe if (task_work_pending(current)) { 29846a525e1SJens Axboe __set_current_state(TASK_RUNNING); 29917437f31SJens Axboe task_work_run(); 300c0e0d6baSDylan Yudaken return 1; 30117437f31SJens Axboe } 30217437f31SJens Axboe 303c0e0d6baSDylan Yudaken return 0; 304c0e0d6baSDylan Yudaken } 305c0e0d6baSDylan Yudaken 306dac6a0eaSJens Axboe static inline bool io_task_work_pending(struct io_ring_ctx *ctx) 307dac6a0eaSJens Axboe { 3086434ec01SJens Axboe return task_work_pending(current) || !wq_list_empty(&ctx->work_llist); 309dac6a0eaSJens Axboe } 310dac6a0eaSJens Axboe 311a282967cSPavel Begunkov static inline void io_tw_lock(struct io_ring_ctx *ctx, struct io_tw_state *ts) 312aa1e90f6SPavel Begunkov { 313a282967cSPavel Begunkov if (!ts->locked) { 314aa1e90f6SPavel Begunkov mutex_lock(&ctx->uring_lock); 315a282967cSPavel Begunkov ts->locked = true; 316aa1e90f6SPavel Begunkov } 317aa1e90f6SPavel Begunkov } 318aa1e90f6SPavel Begunkov 3199da070b1SPavel Begunkov /* 3209da070b1SPavel Begunkov * Don't complete immediately but use deferred completion infrastructure. 3219da070b1SPavel Begunkov * Protected by ->uring_lock and can only be used either with 3229da070b1SPavel Begunkov * IO_URING_F_COMPLETE_DEFER or inside a tw handler holding the mutex. 3239da070b1SPavel Begunkov */ 3249da070b1SPavel Begunkov static inline void io_req_complete_defer(struct io_kiocb *req) 3259da070b1SPavel Begunkov __must_hold(&req->ctx->uring_lock) 326aa1e90f6SPavel Begunkov { 327aa1e90f6SPavel Begunkov struct io_submit_state *state = &req->ctx->submit_state; 328aa1e90f6SPavel Begunkov 3299da070b1SPavel Begunkov lockdep_assert_held(&req->ctx->uring_lock); 3309da070b1SPavel Begunkov 331aa1e90f6SPavel Begunkov wq_list_add_tail(&req->comp_list, &state->compl_reqs); 332aa1e90f6SPavel Begunkov } 333aa1e90f6SPavel Begunkov 33446929b08SPavel Begunkov static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx) 33546929b08SPavel Begunkov { 336bca39f39SPavel Begunkov if (unlikely(ctx->off_timeout_used || ctx->drain_active || 337bca39f39SPavel Begunkov ctx->has_evfd || ctx->poll_activated)) 33846929b08SPavel Begunkov __io_commit_cqring_flush(ctx); 33946929b08SPavel Begunkov } 34046929b08SPavel Begunkov 34163809137SPavel Begunkov static inline void io_get_task_refs(int nr) 34263809137SPavel Begunkov { 34363809137SPavel Begunkov struct io_uring_task *tctx = current->io_uring; 34463809137SPavel Begunkov 34563809137SPavel Begunkov tctx->cached_refs -= nr; 34663809137SPavel Begunkov if (unlikely(tctx->cached_refs < 0)) 34763809137SPavel Begunkov io_task_refs_refill(tctx); 34863809137SPavel Begunkov } 34963809137SPavel Begunkov 350bd1a3783SPavel Begunkov static inline bool io_req_cache_empty(struct io_ring_ctx *ctx) 351bd1a3783SPavel Begunkov { 352bd1a3783SPavel Begunkov return !ctx->submit_state.free_list.next; 353bd1a3783SPavel Begunkov } 354bd1a3783SPavel Begunkov 355c1755c25SBreno Leitao extern struct kmem_cache *req_cachep; 356c1755c25SBreno Leitao 357c8576f3eSPavel Begunkov static inline struct io_kiocb *io_extract_req(struct io_ring_ctx *ctx) 358bd1a3783SPavel Begunkov { 359c1755c25SBreno Leitao struct io_kiocb *req; 360bd1a3783SPavel Begunkov 361c1755c25SBreno Leitao req = container_of(ctx->submit_state.free_list.next, struct io_kiocb, comp_list); 362c1755c25SBreno Leitao wq_stack_extract(&ctx->submit_state.free_list); 363c1755c25SBreno Leitao return req; 364bd1a3783SPavel Begunkov } 365bd1a3783SPavel Begunkov 366c8576f3eSPavel Begunkov static inline bool io_alloc_req(struct io_ring_ctx *ctx, struct io_kiocb **req) 367c8576f3eSPavel Begunkov { 368c8576f3eSPavel Begunkov if (unlikely(io_req_cache_empty(ctx))) { 369c8576f3eSPavel Begunkov if (!__io_alloc_req_refill(ctx)) 370c8576f3eSPavel Begunkov return false; 371c8576f3eSPavel Begunkov } 372c8576f3eSPavel Begunkov *req = io_extract_req(ctx); 373c8576f3eSPavel Begunkov return true; 374c8576f3eSPavel Begunkov } 375c8576f3eSPavel Begunkov 376140102aeSPavel Begunkov static inline bool io_allowed_defer_tw_run(struct io_ring_ctx *ctx) 377140102aeSPavel Begunkov { 378140102aeSPavel Begunkov return likely(ctx->submitter_task == current); 379140102aeSPavel Begunkov } 380140102aeSPavel Begunkov 38176de6749SPavel Begunkov static inline bool io_allowed_run_tw(struct io_ring_ctx *ctx) 38276de6749SPavel Begunkov { 3836567506bSPavel Begunkov return likely(!(ctx->flags & IORING_SETUP_DEFER_TASKRUN) || 3846567506bSPavel Begunkov ctx->submitter_task == current); 38576de6749SPavel Begunkov } 38676de6749SPavel Begunkov 387833b5dffSPavel Begunkov static inline void io_req_queue_tw_complete(struct io_kiocb *req, s32 res) 388833b5dffSPavel Begunkov { 389833b5dffSPavel Begunkov io_req_set_res(req, res, 0); 390833b5dffSPavel Begunkov req->io_task_work.func = io_req_task_complete; 391833b5dffSPavel Begunkov io_req_task_work_add(req); 392833b5dffSPavel Begunkov } 393833b5dffSPavel Begunkov 39496c7d4f8SBreno Leitao /* 39596c7d4f8SBreno Leitao * IORING_SETUP_SQE128 contexts allocate twice the normal SQE size for each 39696c7d4f8SBreno Leitao * slot. 39796c7d4f8SBreno Leitao */ 39896c7d4f8SBreno Leitao static inline size_t uring_sqe_size(struct io_ring_ctx *ctx) 39996c7d4f8SBreno Leitao { 40096c7d4f8SBreno Leitao if (ctx->flags & IORING_SETUP_SQE128) 40196c7d4f8SBreno Leitao return 2 * sizeof(struct io_uring_sqe); 40296c7d4f8SBreno Leitao return sizeof(struct io_uring_sqe); 40396c7d4f8SBreno Leitao } 404de23077eSJens Axboe #endif 405