472 lines · c
1#ifndef IOU_CORE_H2#define IOU_CORE_H3 4#include <linux/errno.h>5#include <linux/lockdep.h>6#include <linux/resume_user_mode.h>7#include <linux/kasan.h>8#include <linux/poll.h>9#include <linux/io_uring_types.h>10#include <uapi/linux/eventpoll.h>11#include "io-wq.h"12#include "slist.h"13#include "filetable.h"14 15#ifndef CREATE_TRACE_POINTS16#include <trace/events/io_uring.h>17#endif18 19enum {20 IOU_OK = 0,21 IOU_ISSUE_SKIP_COMPLETE = -EIOCBQUEUED,22 23 /*24 * Requeue the task_work to restart operations on this request. The25 * actual value isn't important, should just be not an otherwise26 * valid error code, yet less than -MAX_ERRNO and valid internally.27 */28 IOU_REQUEUE = -3072,29 30 /*31 * Intended only when both IO_URING_F_MULTISHOT is passed32 * to indicate to the poll runner that multishot should be33 * removed and the result is set on req->cqe.res.34 */35 IOU_STOP_MULTISHOT = -ECANCELED,36};37 38struct io_wait_queue {39 struct wait_queue_entry wq;40 struct io_ring_ctx *ctx;41 unsigned cq_tail;42 unsigned cq_min_tail;43 unsigned nr_timeouts;44 int hit_timeout;45 ktime_t min_timeout;46 ktime_t timeout;47 struct hrtimer t;48 49#ifdef CONFIG_NET_RX_BUSY_POLL50 ktime_t napi_busy_poll_dt;51 bool napi_prefer_busy_poll;52#endif53};54 55static inline bool io_should_wake(struct io_wait_queue *iowq)56{57 struct io_ring_ctx *ctx = iowq->ctx;58 int dist = READ_ONCE(ctx->rings->cq.tail) - (int) iowq->cq_tail;59 60 /*61 * Wake up if we have enough events, or if a timeout occurred since we62 * started waiting. For timeouts, we always want to return to userspace,63 * regardless of event count.64 */65 return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;66}67 68bool io_cqe_cache_refill(struct io_ring_ctx *ctx, bool overflow);69int io_run_task_work_sig(struct io_ring_ctx *ctx) HWJS_SUSPENDS;70void io_req_defer_failed(struct io_kiocb *req, s32 res) HWJS_SUSPENDS;71bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags) HWJS_SUSPENDS;72void io_add_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags);73bool io_req_post_cqe(struct io_kiocb *req, s32 res, u32 cflags) HWJS_SUSPENDS;74void __io_commit_cqring_flush(struct io_ring_ctx *ctx) HWJS_SUSPENDS;75 76struct file *io_file_get_normal(struct io_kiocb *req, int fd);77struct file *io_file_get_fixed(struct io_kiocb *req, int fd,78 unsigned issue_flags) HWJS_SUSPENDS;79 80void __io_req_task_work_add(struct io_kiocb *req, unsigned flags) HWJS_SUSPENDS;81void io_req_task_work_add_remote(struct io_kiocb *req, struct io_ring_ctx *ctx,82 unsigned flags) HWJS_SUSPENDS;83bool io_alloc_async_data(struct io_kiocb *req);84void io_req_task_queue(struct io_kiocb *req) HWJS_SUSPENDS;85void io_req_task_complete(struct io_kiocb *req, struct io_tw_state *ts);86void io_req_task_queue_fail(struct io_kiocb *req, int ret) HWJS_SUSPENDS;87void io_req_task_submit(struct io_kiocb *req, struct io_tw_state *ts) HWJS_SUSPENDS;88struct llist_node *io_handle_tw_list(struct llist_node *node, unsigned int *count, unsigned int max_entries) HWJS_SUSPENDS;89struct llist_node *tctx_task_work_run(struct io_uring_task *tctx, unsigned int max_entries, unsigned int *count) HWJS_SUSPENDS;90void tctx_task_work(struct callback_head *cb) HWJS_SUSPENDS;91__cold void io_uring_cancel_generic(bool cancel_all, struct io_sq_data *sqd) HWJS_SUSPENDS;92int io_uring_alloc_task_context(struct task_struct *task,93 struct io_ring_ctx *ctx) HWJS_SUSPENDS;94 95int io_ring_add_registered_file(struct io_uring_task *tctx, struct file *file,96 int start, int end);97void io_req_queue_iowq(struct io_kiocb *req) HWJS_SUSPENDS;98 99int io_poll_issue(struct io_kiocb *req, struct io_tw_state *ts) HWJS_SUSPENDS;100int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr) HWJS_SUSPENDS;101int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin) HWJS_SUSPENDS;102void __io_submit_flush_completions(struct io_ring_ctx *ctx) HWJS_SUSPENDS;103 104struct io_wq_work *io_wq_free_work(struct io_wq_work *work) HWJS_SUSPENDS;105void io_wq_submit_work(struct io_wq_work *work) HWJS_SUSPENDS;106 107void io_free_req(struct io_kiocb *req) HWJS_SUSPENDS;108void io_queue_next(struct io_kiocb *req) HWJS_SUSPENDS;109void io_task_refs_refill(struct io_uring_task *tctx);110bool __io_alloc_req_refill(struct io_ring_ctx *ctx);111 112bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task,113 bool cancel_all);114 115void io_activate_pollwq(struct io_ring_ctx *ctx) HWJS_SUSPENDS;116 117static inline void io_lockdep_assert_cq_locked(struct io_ring_ctx *ctx)118{119#if defined(CONFIG_PROVE_LOCKING)120 lockdep_assert(in_task());121 122 if (ctx->flags & IORING_SETUP_IOPOLL) {123 lockdep_assert_held(&ctx->uring_lock);124 } else if (!ctx->task_complete) {125 lockdep_assert_held(&ctx->completion_lock);126 } else if (ctx->submitter_task) {127 /*128 * ->submitter_task may be NULL and we can still post a CQE,129 * if the ring has been setup with IORING_SETUP_R_DISABLED.130 * Not from an SQE, as those cannot be submitted, but via131 * updating tagged resources.132 */133 if (ctx->submitter_task->flags & PF_EXITING)134 lockdep_assert(current_work());135 else136 lockdep_assert(current == ctx->submitter_task);137 }138#endif139}140 141static inline void io_req_task_work_add(struct io_kiocb *req)142{143 __io_req_task_work_add(req, 0);144}145 146static inline void io_submit_flush_completions(struct io_ring_ctx *ctx)147{148 if (!wq_list_empty(&ctx->submit_state.compl_reqs) ||149 ctx->submit_state.cq_flush)150 __io_submit_flush_completions(ctx);151}152 153#define io_for_each_link(pos, head) \154 for (pos = (head); pos; pos = pos->link)155 156static inline bool io_get_cqe_overflow(struct io_ring_ctx *ctx,157 struct io_uring_cqe **ret,158 bool overflow)159{160 io_lockdep_assert_cq_locked(ctx);161 162 if (unlikely(ctx->cqe_cached >= ctx->cqe_sentinel)) {163 if (unlikely(!io_cqe_cache_refill(ctx, overflow)))164 return false;165 }166 *ret = ctx->cqe_cached;167 ctx->cached_cq_tail++;168 ctx->cqe_cached++;169 if (ctx->flags & IORING_SETUP_CQE32)170 ctx->cqe_cached++;171 return true;172}173 174static inline bool io_get_cqe(struct io_ring_ctx *ctx, struct io_uring_cqe **ret)175{176 return io_get_cqe_overflow(ctx, ret, false);177}178 179static __always_inline bool io_fill_cqe_req(struct io_ring_ctx *ctx,180 struct io_kiocb *req)181{182 struct io_uring_cqe *cqe;183 184 /*185 * If we can't get a cq entry, userspace overflowed the186 * submission (by quite a lot). Increment the overflow count in187 * the ring.188 */189 if (unlikely(!io_get_cqe(ctx, &cqe)))190 return false;191 192 if (trace_io_uring_complete_enabled())193 trace_io_uring_complete(req->ctx, req, req->cqe.user_data,194 req->cqe.res, req->cqe.flags,195 req->big_cqe.extra1, req->big_cqe.extra2);196 197 memcpy(cqe, &req->cqe, sizeof(*cqe));198 if (ctx->flags & IORING_SETUP_CQE32) {199 memcpy(cqe->big_cqe, &req->big_cqe, sizeof(*cqe));200 memset(&req->big_cqe, 0, sizeof(req->big_cqe));201 }202 return true;203}204 205static inline void req_set_fail(struct io_kiocb *req)206{207 req->flags |= REQ_F_FAIL;208 if (req->flags & REQ_F_CQE_SKIP) {209 req->flags &= ~REQ_F_CQE_SKIP;210 req->flags |= REQ_F_SKIP_LINK_CQES;211 }212}213 214static inline void io_req_set_res(struct io_kiocb *req, s32 res, u32 cflags)215{216 req->cqe.res = res;217 req->cqe.flags = cflags;218}219 220static inline bool req_has_async_data(struct io_kiocb *req)221{222 return req->flags & REQ_F_ASYNC_DATA;223}224 225static inline void io_put_file(struct io_kiocb *req)226{227 if (!(req->flags & REQ_F_FIXED_FILE) && req->file)228 fput(req->file);229}230 231static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx,232 unsigned issue_flags)233{234 lockdep_assert_held(&ctx->uring_lock);235 if (unlikely(issue_flags & IO_URING_F_UNLOCKED))236 mutex_unlock(&ctx->uring_lock);237}238 239static inline void io_ring_submit_lock(struct io_ring_ctx *ctx,240 unsigned issue_flags)241{242 /*243 * "Normal" inline submissions always hold the uring_lock, since we244 * grab it from the system call. Same is true for the SQPOLL offload.245 * The only exception is when we've detached the request and issue it246 * from an async worker thread, grab the lock for that case.247 */248 if (unlikely(issue_flags & IO_URING_F_UNLOCKED))249 mutex_lock(&ctx->uring_lock);250 lockdep_assert_held(&ctx->uring_lock);251}252 253static inline void io_commit_cqring(struct io_ring_ctx *ctx)254{255 /* order cqe stores with ring update */256 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);257}258 259static inline void io_poll_wq_wake(struct io_ring_ctx *ctx)260{261 if (wq_has_sleeper(&ctx->poll_wq))262 __wake_up(&ctx->poll_wq, TASK_NORMAL, 0,263 poll_to_key(EPOLL_URING_WAKE | EPOLLIN));264}265 266static inline void io_cqring_wake(struct io_ring_ctx *ctx)267{268 /*269 * Trigger waitqueue handler on all waiters on our waitqueue. This270 * won't necessarily wake up all the tasks, io_should_wake() will make271 * that decision.272 *273 * Pass in EPOLLIN|EPOLL_URING_WAKE as the poll wakeup key. The latter274 * set in the mask so that if we recurse back into our own poll275 * waitqueue handlers, we know we have a dependency between eventfd or276 * epoll and should terminate multishot poll at that point.277 */278 if (wq_has_sleeper(&ctx->cq_wait))279 __wake_up(&ctx->cq_wait, TASK_NORMAL, 0,280 poll_to_key(EPOLL_URING_WAKE | EPOLLIN));281}282 283static inline bool io_sqring_full(struct io_ring_ctx *ctx)284{285 struct io_rings *r = ctx->rings;286 287 /*288 * SQPOLL must use the actual sqring head, as using the cached_sq_head289 * is race prone if the SQPOLL thread has grabbed entries but not yet290 * committed them to the ring. For !SQPOLL, this doesn't matter, but291 * since this helper is just used for SQPOLL sqring waits (or POLLOUT),292 * just read the actual sqring head unconditionally.293 */294 return READ_ONCE(r->sq.tail) - READ_ONCE(r->sq.head) == ctx->sq_entries;295}296 297static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)298{299 struct io_rings *rings = ctx->rings;300 unsigned int entries;301 302 /* make sure SQ entry isn't read before tail */303 entries = smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;304 return min(entries, ctx->sq_entries);305}306 307static inline int io_run_task_work(void)308{309 bool ret = false;310 311 /*312 * Always check-and-clear the task_work notification signal. With how313 * signaling works for task_work, we can find it set with nothing to314 * run. We need to clear it for that case, like get_signal() does.315 */316 if (test_thread_flag(TIF_NOTIFY_SIGNAL))317 clear_notify_signal();318 /*319 * PF_IO_WORKER never returns to userspace, so check here if we have320 * notify work that needs processing.321 */322 if (current->flags & PF_IO_WORKER) {323 if (test_thread_flag(TIF_NOTIFY_RESUME)) {324 __set_current_state(TASK_RUNNING);325 resume_user_mode_work(NULL);326 }327 if (current->io_uring) {328 unsigned int count = 0;329 330 __set_current_state(TASK_RUNNING);331 tctx_task_work_run(current->io_uring, UINT_MAX, &count);332 if (count)333 ret = true;334 }335 }336 if (task_work_pending(current)) {337 __set_current_state(TASK_RUNNING);338 task_work_run();339 ret = true;340 }341 342 return ret;343}344 345static inline bool io_task_work_pending(struct io_ring_ctx *ctx)346{347 return task_work_pending(current) || !llist_empty(&ctx->work_llist);348}349 350static inline void io_tw_lock(struct io_ring_ctx *ctx, struct io_tw_state *ts)351{352 lockdep_assert_held(&ctx->uring_lock);353}354 355/*356 * Don't complete immediately but use deferred completion infrastructure.357 * Protected by ->uring_lock and can only be used either with358 * IO_URING_F_COMPLETE_DEFER or inside a tw handler holding the mutex.359 */360static inline void io_req_complete_defer(struct io_kiocb *req)361 __must_hold(&req->ctx->uring_lock)362{363 struct io_submit_state *state = &req->ctx->submit_state;364 365 lockdep_assert_held(&req->ctx->uring_lock);366 367 wq_list_add_tail(&req->comp_list, &state->compl_reqs);368}369 370static inline void io_commit_cqring_flush(struct io_ring_ctx *ctx)371{372 if (unlikely(ctx->off_timeout_used || ctx->drain_active ||373 ctx->has_evfd || ctx->poll_activated))374 __io_commit_cqring_flush(ctx);375}376 377static inline void io_get_task_refs(int nr)378{379 struct io_uring_task *tctx = current->io_uring;380 381 tctx->cached_refs -= nr;382 if (unlikely(tctx->cached_refs < 0))383 io_task_refs_refill(tctx);384}385 386static inline bool io_req_cache_empty(struct io_ring_ctx *ctx)387{388 return !ctx->submit_state.free_list.next;389}390 391extern struct kmem_cache *req_cachep;392extern struct kmem_cache *io_buf_cachep;393 394static inline struct io_kiocb *io_extract_req(struct io_ring_ctx *ctx)395{396 struct io_kiocb *req;397 398 req = container_of(ctx->submit_state.free_list.next, struct io_kiocb, comp_list);399 wq_stack_extract(&ctx->submit_state.free_list);400 return req;401}402 403static inline bool io_alloc_req(struct io_ring_ctx *ctx, struct io_kiocb **req)404{405 if (unlikely(io_req_cache_empty(ctx))) {406 if (!__io_alloc_req_refill(ctx))407 return false;408 }409 *req = io_extract_req(ctx);410 return true;411}412 413static inline bool io_allowed_defer_tw_run(struct io_ring_ctx *ctx)414{415 return likely(ctx->submitter_task == current);416}417 418static inline bool io_allowed_run_tw(struct io_ring_ctx *ctx)419{420 return likely(!(ctx->flags & IORING_SETUP_DEFER_TASKRUN) ||421 ctx->submitter_task == current);422}423 424static inline void io_req_queue_tw_complete(struct io_kiocb *req, s32 res)425{426 io_req_set_res(req, res, 0);427 req->io_task_work.func = io_req_task_complete;428 io_req_task_work_add(req);429}430 431/*432 * IORING_SETUP_SQE128 contexts allocate twice the normal SQE size for each433 * slot.434 */435static inline size_t uring_sqe_size(struct io_ring_ctx *ctx)436{437 if (ctx->flags & IORING_SETUP_SQE128)438 return 2 * sizeof(struct io_uring_sqe);439 return sizeof(struct io_uring_sqe);440}441 442static inline bool io_file_can_poll(struct io_kiocb *req)443{444 if (req->flags & REQ_F_CAN_POLL)445 return true;446 if (req->file && file_can_poll(req->file)) {447 req->flags |= REQ_F_CAN_POLL;448 return true;449 }450 return false;451}452 453static inline ktime_t io_get_time(struct io_ring_ctx *ctx)454{455 if (ctx->clockid == CLOCK_MONOTONIC)456 return ktime_get();457 458 return ktime_get_with_offset(ctx->clock_offset);459}460 461enum {462 IO_CHECK_CQ_OVERFLOW_BIT,463 IO_CHECK_CQ_DROPPED_BIT,464};465 466static inline bool io_has_work(struct io_ring_ctx *ctx)467{468 return test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq) ||469 !llist_empty(&ctx->work_llist);470}471#endif472