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> 6ab1c84d8SPavel Begunkov #include <linux/io_uring_types.h> 7ab1c84d8SPavel Begunkov #include "io-wq.h" 8ab1c84d8SPavel Begunkov #include "filetable.h" 9de23077eSJens Axboe 10f3b44f92SJens Axboe #ifndef CREATE_TRACE_POINTS 11f3b44f92SJens Axboe #include <trace/events/io_uring.h> 12f3b44f92SJens Axboe #endif 13f3b44f92SJens Axboe 1497b388d7SJens Axboe enum { 1597b388d7SJens Axboe IOU_OK = 0, 1697b388d7SJens Axboe IOU_ISSUE_SKIP_COMPLETE = -EIOCBQUEUED, 1797b388d7SJens Axboe }; 1897b388d7SJens Axboe 19faf88ddeSPavel Begunkov struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx); 2068494a65SPavel Begunkov bool io_req_cqe_overflow(struct io_kiocb *req); 219046c641SPavel Begunkov int io_run_task_work_sig(void); 229046c641SPavel Begunkov void io_req_complete_failed(struct io_kiocb *req, s32 res); 239046c641SPavel Begunkov void __io_req_complete(struct io_kiocb *req, unsigned issue_flags); 249046c641SPavel Begunkov void io_req_complete_post(struct io_kiocb *req); 259046c641SPavel Begunkov void __io_req_complete_post(struct io_kiocb *req); 269046c641SPavel Begunkov bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags); 279046c641SPavel Begunkov void __io_commit_cqring_flush(struct io_ring_ctx *ctx); 289046c641SPavel Begunkov 299046c641SPavel Begunkov struct page **io_pin_pages(unsigned long ubuf, unsigned long len, int *npages); 309046c641SPavel Begunkov 319046c641SPavel Begunkov struct file *io_file_get_normal(struct io_kiocb *req, int fd); 329046c641SPavel Begunkov struct file *io_file_get_fixed(struct io_kiocb *req, int fd, 339046c641SPavel Begunkov unsigned issue_flags); 349046c641SPavel Begunkov 359046c641SPavel Begunkov bool io_is_uring_fops(struct file *file); 369046c641SPavel Begunkov bool io_alloc_async_data(struct io_kiocb *req); 379046c641SPavel Begunkov void io_req_task_work_add(struct io_kiocb *req); 389046c641SPavel Begunkov void io_req_task_prio_work_add(struct io_kiocb *req); 399046c641SPavel Begunkov void io_req_tw_post_queue(struct io_kiocb *req, s32 res, u32 cflags); 409046c641SPavel Begunkov void io_req_task_queue(struct io_kiocb *req); 419046c641SPavel Begunkov void io_queue_iowq(struct io_kiocb *req, bool *dont_use); 429046c641SPavel Begunkov void io_req_task_complete(struct io_kiocb *req, bool *locked); 439046c641SPavel Begunkov void io_req_task_queue_fail(struct io_kiocb *req, int ret); 449046c641SPavel Begunkov void io_req_task_submit(struct io_kiocb *req, bool *locked); 459046c641SPavel Begunkov void tctx_task_work(struct callback_head *cb); 469046c641SPavel Begunkov __cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd); 479046c641SPavel Begunkov int io_uring_alloc_task_context(struct task_struct *task, 489046c641SPavel Begunkov struct io_ring_ctx *ctx); 499046c641SPavel Begunkov 509046c641SPavel Begunkov int io_poll_issue(struct io_kiocb *req, bool *locked); 519046c641SPavel Begunkov int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr); 529046c641SPavel Begunkov int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin); 539046c641SPavel Begunkov void io_free_batch_list(struct io_ring_ctx *ctx, struct io_wq_work_node *node); 549046c641SPavel Begunkov int io_req_prep_async(struct io_kiocb *req); 559046c641SPavel Begunkov 569046c641SPavel Begunkov struct io_wq_work *io_wq_free_work(struct io_wq_work *work); 579046c641SPavel Begunkov void io_wq_submit_work(struct io_wq_work *work); 589046c641SPavel Begunkov 599046c641SPavel Begunkov void io_free_req(struct io_kiocb *req); 609046c641SPavel Begunkov void io_queue_next(struct io_kiocb *req); 619046c641SPavel Begunkov 629046c641SPavel Begunkov bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task, 639046c641SPavel Begunkov bool cancel_all); 649046c641SPavel Begunkov 659046c641SPavel Begunkov #define io_for_each_link(pos, head) \ 669046c641SPavel Begunkov for (pos = (head); pos; pos = pos->link) 67f3b44f92SJens Axboe 6825399321SPavel Begunkov static inline void io_cq_lock(struct io_ring_ctx *ctx) 6925399321SPavel Begunkov __acquires(ctx->completion_lock) 7025399321SPavel Begunkov { 7125399321SPavel Begunkov spin_lock(&ctx->completion_lock); 7225399321SPavel Begunkov } 7325399321SPavel Begunkov 7425399321SPavel Begunkov void io_cq_unlock_post(struct io_ring_ctx *ctx); 7525399321SPavel Begunkov 76f3b44f92SJens Axboe static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx) 77f3b44f92SJens Axboe { 78f3b44f92SJens Axboe if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) { 79f3b44f92SJens Axboe struct io_uring_cqe *cqe = ctx->cqe_cached; 80f3b44f92SJens Axboe 81f3b44f92SJens Axboe ctx->cached_cq_tail++; 82f3b44f92SJens Axboe ctx->cqe_cached++; 83b3659a65SPavel Begunkov if (ctx->flags & IORING_SETUP_CQE32) 84b3659a65SPavel Begunkov ctx->cqe_cached++; 85f3b44f92SJens Axboe return cqe; 86f3b44f92SJens Axboe } 87f3b44f92SJens Axboe 88f3b44f92SJens Axboe return __io_get_cqe(ctx); 89f3b44f92SJens Axboe } 90f3b44f92SJens Axboe 91f3b44f92SJens Axboe static inline bool __io_fill_cqe_req(struct io_ring_ctx *ctx, 92f3b44f92SJens Axboe struct io_kiocb *req) 93f3b44f92SJens Axboe { 94f3b44f92SJens Axboe struct io_uring_cqe *cqe; 95f3b44f92SJens Axboe 96f3b44f92SJens Axboe trace_io_uring_complete(req->ctx, req, req->cqe.user_data, 97ae5735c6SPavel Begunkov req->cqe.res, req->cqe.flags, 98ae5735c6SPavel Begunkov (req->flags & REQ_F_CQE32_INIT) ? req->extra1 : 0, 99ae5735c6SPavel Begunkov (req->flags & REQ_F_CQE32_INIT) ? req->extra2 : 0); 100f3b44f92SJens Axboe /* 101f3b44f92SJens Axboe * If we can't get a cq entry, userspace overflowed the 102f3b44f92SJens Axboe * submission (by quite a lot). Increment the overflow count in 103f3b44f92SJens Axboe * the ring. 104f3b44f92SJens Axboe */ 105f3b44f92SJens Axboe cqe = io_get_cqe(ctx); 106e8c328c3SPavel Begunkov if (unlikely(!cqe)) 107e8c328c3SPavel Begunkov return io_req_cqe_overflow(req); 108f3b44f92SJens Axboe memcpy(cqe, &req->cqe, sizeof(*cqe)); 109e8c328c3SPavel Begunkov 110e8c328c3SPavel Begunkov if (ctx->flags & IORING_SETUP_CQE32) { 111f3b44f92SJens Axboe u64 extra1 = 0, extra2 = 0; 112f3b44f92SJens Axboe 113f3b44f92SJens Axboe if (req->flags & REQ_F_CQE32_INIT) { 114f3b44f92SJens Axboe extra1 = req->extra1; 115f3b44f92SJens Axboe extra2 = req->extra2; 116f3b44f92SJens Axboe } 117f3b44f92SJens Axboe 118f3b44f92SJens Axboe WRITE_ONCE(cqe->big_cqe[0], extra1); 119f3b44f92SJens Axboe WRITE_ONCE(cqe->big_cqe[1], extra2); 120e8c328c3SPavel Begunkov } 121f3b44f92SJens Axboe return true; 122f3b44f92SJens Axboe } 123f3b44f92SJens Axboe 124531113bbSJens Axboe static inline void req_set_fail(struct io_kiocb *req) 125531113bbSJens Axboe { 126531113bbSJens Axboe req->flags |= REQ_F_FAIL; 127531113bbSJens Axboe if (req->flags & REQ_F_CQE_SKIP) { 128531113bbSJens Axboe req->flags &= ~REQ_F_CQE_SKIP; 129531113bbSJens Axboe req->flags |= REQ_F_SKIP_LINK_CQES; 130531113bbSJens Axboe } 131531113bbSJens Axboe } 132531113bbSJens Axboe 133de23077eSJens Axboe static inline void io_req_set_res(struct io_kiocb *req, s32 res, u32 cflags) 134de23077eSJens Axboe { 135de23077eSJens Axboe req->cqe.res = res; 136de23077eSJens Axboe req->cqe.flags = cflags; 137de23077eSJens Axboe } 138de23077eSJens Axboe 13999f15d8dSJens Axboe static inline bool req_has_async_data(struct io_kiocb *req) 14099f15d8dSJens Axboe { 14199f15d8dSJens Axboe return req->flags & REQ_F_ASYNC_DATA; 14299f15d8dSJens Axboe } 14399f15d8dSJens Axboe 144531113bbSJens Axboe static inline void io_put_file(struct file *file) 145531113bbSJens Axboe { 146531113bbSJens Axboe if (file) 147531113bbSJens Axboe fput(file); 148531113bbSJens Axboe } 149531113bbSJens Axboe 150cd40cae2SJens Axboe static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx, 151cd40cae2SJens Axboe unsigned issue_flags) 152cd40cae2SJens Axboe { 153cd40cae2SJens Axboe lockdep_assert_held(&ctx->uring_lock); 154cd40cae2SJens Axboe if (issue_flags & IO_URING_F_UNLOCKED) 155cd40cae2SJens Axboe mutex_unlock(&ctx->uring_lock); 156cd40cae2SJens Axboe } 157cd40cae2SJens Axboe 158cd40cae2SJens Axboe static inline void io_ring_submit_lock(struct io_ring_ctx *ctx, 159cd40cae2SJens Axboe unsigned issue_flags) 160cd40cae2SJens Axboe { 161cd40cae2SJens Axboe /* 162cd40cae2SJens Axboe * "Normal" inline submissions always hold the uring_lock, since we 163cd40cae2SJens Axboe * grab it from the system call. Same is true for the SQPOLL offload. 164cd40cae2SJens Axboe * The only exception is when we've detached the request and issue it 165cd40cae2SJens Axboe * from an async worker thread, grab the lock for that case. 166cd40cae2SJens Axboe */ 167cd40cae2SJens Axboe if (issue_flags & IO_URING_F_UNLOCKED) 168cd40cae2SJens Axboe mutex_lock(&ctx->uring_lock); 169cd40cae2SJens Axboe lockdep_assert_held(&ctx->uring_lock); 170cd40cae2SJens Axboe } 171cd40cae2SJens Axboe 172f9ead18cSJens Axboe static inline void io_commit_cqring(struct io_ring_ctx *ctx) 173f9ead18cSJens Axboe { 174f9ead18cSJens Axboe /* order cqe stores with ring update */ 175f9ead18cSJens Axboe smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); 176f9ead18cSJens Axboe } 177f9ead18cSJens Axboe 178f3b44f92SJens Axboe static inline void io_cqring_wake(struct io_ring_ctx *ctx) 179f3b44f92SJens Axboe { 180f3b44f92SJens Axboe /* 181f3b44f92SJens Axboe * wake_up_all() may seem excessive, but io_wake_function() and 182f3b44f92SJens Axboe * io_should_wake() handle the termination of the loop and only 183f3b44f92SJens Axboe * wake as many waiters as we need to. 184f3b44f92SJens Axboe */ 185f3b44f92SJens Axboe if (wq_has_sleeper(&ctx->cq_wait)) 186f3b44f92SJens Axboe wake_up_all(&ctx->cq_wait); 187f3b44f92SJens Axboe } 188f3b44f92SJens Axboe 18917437f31SJens Axboe static inline bool io_sqring_full(struct io_ring_ctx *ctx) 19017437f31SJens Axboe { 19117437f31SJens Axboe struct io_rings *r = ctx->rings; 19217437f31SJens Axboe 19317437f31SJens Axboe return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; 19417437f31SJens Axboe } 19517437f31SJens Axboe 19617437f31SJens Axboe static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx) 19717437f31SJens Axboe { 19817437f31SJens Axboe struct io_rings *rings = ctx->rings; 19917437f31SJens Axboe 20017437f31SJens Axboe /* make sure SQ entry isn't read before tail */ 20117437f31SJens Axboe return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; 20217437f31SJens Axboe } 20317437f31SJens Axboe 20417437f31SJens Axboe static inline bool io_run_task_work(void) 20517437f31SJens Axboe { 20617437f31SJens Axboe if (test_thread_flag(TIF_NOTIFY_SIGNAL) || task_work_pending(current)) { 20717437f31SJens Axboe __set_current_state(TASK_RUNNING); 20817437f31SJens Axboe clear_notify_signal(); 20917437f31SJens Axboe if (task_work_pending(current)) 21017437f31SJens Axboe task_work_run(); 21117437f31SJens Axboe return true; 21217437f31SJens Axboe } 21317437f31SJens Axboe 21417437f31SJens Axboe return false; 21517437f31SJens Axboe } 21617437f31SJens Axboe 217aa1e90f6SPavel Begunkov static inline void io_tw_lock(struct io_ring_ctx *ctx, bool *locked) 218aa1e90f6SPavel Begunkov { 219aa1e90f6SPavel Begunkov if (!*locked) { 220aa1e90f6SPavel Begunkov mutex_lock(&ctx->uring_lock); 221aa1e90f6SPavel Begunkov *locked = true; 222aa1e90f6SPavel Begunkov } 223aa1e90f6SPavel Begunkov } 224aa1e90f6SPavel Begunkov 225aa1e90f6SPavel Begunkov static inline void io_req_add_compl_list(struct io_kiocb *req) 226aa1e90f6SPavel Begunkov { 227aa1e90f6SPavel Begunkov struct io_submit_state *state = &req->ctx->submit_state; 228aa1e90f6SPavel Begunkov 229aa1e90f6SPavel Begunkov wq_list_add_tail(&req->comp_list, &state->compl_reqs); 230aa1e90f6SPavel Begunkov } 231aa1e90f6SPavel Begunkov 232*46929b08SPavel Begunkov static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx) 233*46929b08SPavel Begunkov { 234*46929b08SPavel Begunkov if (unlikely(ctx->off_timeout_used || ctx->drain_active || ctx->has_evfd)) 235*46929b08SPavel Begunkov __io_commit_cqring_flush(ctx); 236*46929b08SPavel Begunkov } 237*46929b08SPavel Begunkov 238de23077eSJens Axboe #endif 239