Lines Matching +full:poll +full:- +full:retry +full:- +full:count
1 // SPDX-License-Identifier: GPL-2.0
14 * through a control-dependency in io_get_cqe (smp_store_release to
25 * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application
39 * Copyright (C) 2018-2019 Jens Axboe
40 * Copyright (c) 2018-2019 Christoph Hellwig
89 #include "io-wq.h"
100 #define IO_RSRC_TAG_TABLE_SHIFT (PAGE_SHIFT - 3)
102 #define IO_RSRC_TAG_TABLE_MASK (IO_RSRC_TAG_TABLE_MAX - 1)
138 * ring_entries - 1)
324 /* inline/task_work completion list, under ->uring_lock */
331 /* const or read-mostly hot data */
386 /* IRQ completion list, under ->completion_lock */
415 * ->iopoll_list is protected by the ctx->uring_lock for
787 /* already went through poll handler */
810 struct io_poll_iocb poll; member
832 * access the file pointer through any of the sub-structs,
839 struct io_poll_iocb poll; member
863 /* use only after cleaning per-op data, see io_clean_op() */
885 /* used with ctx->iopoll_list with reads/writes */
888 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
914 /* needs req->file assigned */
918 /* unbound wq insertion if file is a non-regular file */
1122 if (file->f_op == &io_uring_fops) { in io_uring_get_socket()
1123 struct io_ring_ctx *ctx = file->private_data; in io_uring_get_socket()
1125 return ctx->ring_sock->sk; in io_uring_get_socket()
1135 mutex_lock(&ctx->uring_lock); in io_tw_lock()
1141 for (pos = (head); pos; pos = pos->link)
1148 ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u)
1152 WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); in req_ref_inc_not_zero()
1153 return atomic_inc_not_zero(&req->refs); in req_ref_inc_not_zero()
1158 if (likely(!(req->flags & REQ_F_REFCOUNT))) in req_ref_put_and_test()
1162 return atomic_dec_and_test(&req->refs); in req_ref_put_and_test()
1167 WARN_ON_ONCE(!(req->flags & REQ_F_REFCOUNT)); in req_ref_get()
1169 atomic_inc(&req->refs); in req_ref_get()
1174 if (!(req->flags & REQ_F_REFCOUNT)) { in __io_req_set_refcount()
1175 req->flags |= REQ_F_REFCOUNT; in __io_req_set_refcount()
1176 atomic_set(&req->refs, nr); in __io_req_set_refcount()
1187 struct io_ring_ctx *ctx = req->ctx; in io_req_set_rsrc_node()
1189 if (!req->fixed_rsrc_refs) { in io_req_set_rsrc_node()
1190 req->fixed_rsrc_refs = &ctx->rsrc_node->refs; in io_req_set_rsrc_node()
1191 percpu_ref_get(req->fixed_rsrc_refs); in io_req_set_rsrc_node()
1199 /* already at zero, wait for ->release() */ in io_refs_resurrect()
1209 __must_hold(&req->ctx->timeout_lock) in io_match_task()
1213 if (task && head->task != task) in io_match_task()
1219 if (req->flags & REQ_F_INFLIGHT) in io_match_task()
1230 if (req->flags & REQ_F_INFLIGHT) in io_match_linked()
1245 if (task && head->task != task) in io_match_task_safe()
1250 if (head->flags & REQ_F_LINK_TIMEOUT) { in io_match_task_safe()
1251 struct io_ring_ctx *ctx = head->ctx; in io_match_task_safe()
1254 spin_lock_irq(&ctx->timeout_lock); in io_match_task_safe()
1256 spin_unlock_irq(&ctx->timeout_lock); in io_match_task_safe()
1265 req->flags |= REQ_F_FAIL; in req_set_fail()
1271 req->result = res; in req_fail_link_node()
1278 complete(&ctx->ref_comp); in io_ring_ctx_ref_free()
1283 return !req->timeout.off; in io_is_timeout_noseq()
1290 struct llist_node *node = llist_del_all(&ctx->fallback_llist); in io_fallback_req_func()
1294 percpu_ref_get(&ctx->refs); in io_fallback_req_func()
1296 req->io_task_work.func(req, &locked); in io_fallback_req_func()
1299 if (ctx->submit_state.compl_nr) in io_fallback_req_func()
1301 mutex_unlock(&ctx->uring_lock); in io_fallback_req_func()
1303 percpu_ref_put(&ctx->refs); in io_fallback_req_func()
1320 hash_bits = ilog2(p->cq_entries); in io_ring_ctx_alloc()
1321 hash_bits -= 5; in io_ring_ctx_alloc()
1324 ctx->cancel_hash_bits = hash_bits; in io_ring_ctx_alloc()
1325 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head), in io_ring_ctx_alloc()
1327 if (!ctx->cancel_hash) in io_ring_ctx_alloc()
1329 __hash_init(ctx->cancel_hash, 1U << hash_bits); in io_ring_ctx_alloc()
1331 ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL); in io_ring_ctx_alloc()
1332 if (!ctx->dummy_ubuf) in io_ring_ctx_alloc()
1335 ctx->dummy_ubuf->ubuf = -1UL; in io_ring_ctx_alloc()
1337 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free, in io_ring_ctx_alloc()
1341 ctx->flags = p->flags; in io_ring_ctx_alloc()
1342 init_waitqueue_head(&ctx->sqo_sq_wait); in io_ring_ctx_alloc()
1343 INIT_LIST_HEAD(&ctx->sqd_list); in io_ring_ctx_alloc()
1344 init_waitqueue_head(&ctx->poll_wait); in io_ring_ctx_alloc()
1345 INIT_LIST_HEAD(&ctx->cq_overflow_list); in io_ring_ctx_alloc()
1346 init_completion(&ctx->ref_comp); in io_ring_ctx_alloc()
1347 xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1); in io_ring_ctx_alloc()
1348 xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1); in io_ring_ctx_alloc()
1349 mutex_init(&ctx->uring_lock); in io_ring_ctx_alloc()
1350 init_waitqueue_head(&ctx->cq_wait); in io_ring_ctx_alloc()
1351 spin_lock_init(&ctx->completion_lock); in io_ring_ctx_alloc()
1352 spin_lock_init(&ctx->timeout_lock); in io_ring_ctx_alloc()
1353 INIT_LIST_HEAD(&ctx->iopoll_list); in io_ring_ctx_alloc()
1354 INIT_LIST_HEAD(&ctx->defer_list); in io_ring_ctx_alloc()
1355 INIT_LIST_HEAD(&ctx->timeout_list); in io_ring_ctx_alloc()
1356 INIT_LIST_HEAD(&ctx->ltimeout_list); in io_ring_ctx_alloc()
1357 spin_lock_init(&ctx->rsrc_ref_lock); in io_ring_ctx_alloc()
1358 INIT_LIST_HEAD(&ctx->rsrc_ref_list); in io_ring_ctx_alloc()
1359 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work); in io_ring_ctx_alloc()
1360 init_llist_head(&ctx->rsrc_put_llist); in io_ring_ctx_alloc()
1361 INIT_LIST_HEAD(&ctx->tctx_list); in io_ring_ctx_alloc()
1362 INIT_LIST_HEAD(&ctx->submit_state.free_list); in io_ring_ctx_alloc()
1363 INIT_LIST_HEAD(&ctx->locked_free_list); in io_ring_ctx_alloc()
1364 INIT_DELAYED_WORK(&ctx->fallback_work, io_fallback_req_func); in io_ring_ctx_alloc()
1367 kfree(ctx->dummy_ubuf); in io_ring_ctx_alloc()
1368 kfree(ctx->cancel_hash); in io_ring_ctx_alloc()
1375 struct io_rings *r = ctx->rings; in io_account_cq_overflow()
1377 WRITE_ONCE(r->cq_overflow, READ_ONCE(r->cq_overflow) + 1); in io_account_cq_overflow()
1378 ctx->cq_extra--; in io_account_cq_overflow()
1383 if (unlikely(req->flags & REQ_F_IO_DRAIN)) { in req_need_defer()
1384 struct io_ring_ctx *ctx = req->ctx; in req_need_defer()
1386 return seq + READ_ONCE(ctx->cq_extra) != ctx->cached_cq_tail; in req_need_defer()
1403 return IS_ENABLED(CONFIG_64BIT) && (req->flags & REQ_F_FIXED_FILE); in io_req_ffs_set()
1408 if (!(req->flags & REQ_F_INFLIGHT)) { in io_req_track_inflight()
1409 req->flags |= REQ_F_INFLIGHT; in io_req_track_inflight()
1410 atomic_inc(&req->task->io_uring->inflight_tracked); in io_req_track_inflight()
1416 if (WARN_ON_ONCE(!req->link)) in __io_prep_linked_timeout()
1419 req->flags &= ~REQ_F_ARM_LTIMEOUT; in __io_prep_linked_timeout()
1420 req->flags |= REQ_F_LINK_TIMEOUT; in __io_prep_linked_timeout()
1424 __io_req_set_refcount(req->link, 2); in __io_prep_linked_timeout()
1425 return req->link; in __io_prep_linked_timeout()
1430 if (likely(!(req->flags & REQ_F_ARM_LTIMEOUT))) in io_prep_linked_timeout()
1437 const struct io_op_def *def = &io_op_defs[req->opcode]; in io_prep_async_work()
1438 struct io_ring_ctx *ctx = req->ctx; in io_prep_async_work()
1440 if (!(req->flags & REQ_F_CREDS)) { in io_prep_async_work()
1441 req->flags |= REQ_F_CREDS; in io_prep_async_work()
1442 req->creds = get_current_cred(); in io_prep_async_work()
1445 req->work.list.next = NULL; in io_prep_async_work()
1446 req->work.flags = 0; in io_prep_async_work()
1447 if (req->flags & REQ_F_FORCE_ASYNC) in io_prep_async_work()
1448 req->work.flags |= IO_WQ_WORK_CONCURRENT; in io_prep_async_work()
1450 if (req->flags & REQ_F_ISREG) { in io_prep_async_work()
1451 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL)) in io_prep_async_work()
1452 io_wq_hash_work(&req->work, file_inode(req->file)); in io_prep_async_work()
1453 } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) { in io_prep_async_work()
1454 if (def->unbound_nonreg_file) in io_prep_async_work()
1455 req->work.flags |= IO_WQ_WORK_UNBOUND; in io_prep_async_work()
1463 if (req->flags & REQ_F_LINK_TIMEOUT) { in io_prep_async_link()
1464 struct io_ring_ctx *ctx = req->ctx; in io_prep_async_link()
1466 spin_lock_irq(&ctx->timeout_lock); in io_prep_async_link()
1469 spin_unlock_irq(&ctx->timeout_lock); in io_prep_async_link()
1478 struct io_ring_ctx *ctx = req->ctx; in io_queue_async_work()
1480 struct io_uring_task *tctx = req->task->io_uring; in io_queue_async_work()
1486 BUG_ON(!tctx->io_wq); in io_queue_async_work()
1488 /* init ->work of the whole link before punting */ in io_queue_async_work()
1494 * canceled. That will make io-wq go through the usual work cancel in io_queue_async_work()
1498 if (WARN_ON_ONCE(!same_thread_group(req->task, current))) in io_queue_async_work()
1499 req->work.flags |= IO_WQ_WORK_CANCEL; in io_queue_async_work()
1501 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req, in io_queue_async_work()
1502 &req->work, req->flags); in io_queue_async_work()
1503 io_wq_enqueue(tctx->io_wq, &req->work); in io_queue_async_work()
1509 __must_hold(&req->ctx->completion_lock) in io_kill_timeout()
1510 __must_hold(&req->ctx->timeout_lock) in io_kill_timeout()
1512 struct io_timeout_data *io = req->async_data; in io_kill_timeout()
1514 if (hrtimer_try_to_cancel(&io->timer) != -1) { in io_kill_timeout()
1517 atomic_set(&req->ctx->cq_timeouts, in io_kill_timeout()
1518 atomic_read(&req->ctx->cq_timeouts) + 1); in io_kill_timeout()
1519 list_del_init(&req->timeout.list); in io_kill_timeout()
1527 while (!list_empty(&ctx->defer_list)) { in io_queue_deferred()
1528 struct io_defer_entry *de = list_first_entry(&ctx->defer_list, in io_queue_deferred()
1531 if (req_need_defer(de->req, de->seq)) in io_queue_deferred()
1533 list_del_init(&de->list); in io_queue_deferred()
1534 io_req_task_queue(de->req); in io_queue_deferred()
1540 __must_hold(&ctx->completion_lock) in io_flush_timeouts()
1542 u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); in io_flush_timeouts()
1545 spin_lock_irq(&ctx->timeout_lock); in io_flush_timeouts()
1546 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { in io_flush_timeouts()
1555 * Assuming not more than 2^31-1 events have happened since, in io_flush_timeouts()
1559 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush; in io_flush_timeouts()
1560 events_got = seq - ctx->cq_last_tm_flush; in io_flush_timeouts()
1566 ctx->cq_last_tm_flush = seq; in io_flush_timeouts()
1567 spin_unlock_irq(&ctx->timeout_lock); in io_flush_timeouts()
1572 if (ctx->off_timeout_used) in __io_commit_cqring_flush()
1574 if (ctx->drain_active) in __io_commit_cqring_flush()
1580 if (unlikely(ctx->off_timeout_used || ctx->drain_active)) in io_commit_cqring()
1583 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail); in io_commit_cqring()
1588 struct io_rings *r = ctx->rings; in io_sqring_full()
1590 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == ctx->sq_entries; in io_sqring_full()
1595 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head); in __io_cqring_events()
1600 struct io_rings *rings = ctx->rings; in io_get_cqe()
1601 unsigned tail, mask = ctx->cq_entries - 1; in io_get_cqe()
1608 if (__io_cqring_events(ctx) == ctx->cq_entries) in io_get_cqe()
1611 tail = ctx->cached_cq_tail++; in io_get_cqe()
1612 return &rings->cqes[tail & mask]; in io_get_cqe()
1617 if (likely(!ctx->cq_ev_fd)) in io_should_trigger_evfd()
1619 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED) in io_should_trigger_evfd()
1621 return !ctx->eventfd_async || io_wq_current_is_worker(); in io_should_trigger_evfd()
1626 * Some applications rely on the eventfd notification count only changing
1629 * hence the eventfd count) and number of CQEs posted to the CQ ring.
1638 if (wq_has_sleeper(&ctx->cq_wait)) in io_cqring_ev_posted()
1639 __wake_up(&ctx->cq_wait, TASK_NORMAL, 0, in io_cqring_ev_posted()
1641 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait)) in io_cqring_ev_posted()
1642 wake_up(&ctx->sq_data->wait); in io_cqring_ev_posted()
1644 eventfd_signal_mask(ctx->cq_ev_fd, 1, EPOLL_URING_WAKE); in io_cqring_ev_posted()
1645 if (waitqueue_active(&ctx->poll_wait)) in io_cqring_ev_posted()
1646 __wake_up(&ctx->poll_wait, TASK_INTERRUPTIBLE, 0, in io_cqring_ev_posted()
1655 if (ctx->flags & IORING_SETUP_SQPOLL) { in io_cqring_ev_posted_iopoll()
1656 if (waitqueue_active(&ctx->cq_wait)) in io_cqring_ev_posted_iopoll()
1657 __wake_up(&ctx->cq_wait, TASK_NORMAL, 0, in io_cqring_ev_posted_iopoll()
1661 eventfd_signal_mask(ctx->cq_ev_fd, 1, EPOLL_URING_WAKE); in io_cqring_ev_posted_iopoll()
1662 if (waitqueue_active(&ctx->poll_wait)) in io_cqring_ev_posted_iopoll()
1663 __wake_up(&ctx->poll_wait, TASK_INTERRUPTIBLE, 0, in io_cqring_ev_posted_iopoll()
1672 if (!force && __io_cqring_events(ctx) == ctx->cq_entries) in __io_cqring_overflow_flush()
1676 spin_lock(&ctx->completion_lock); in __io_cqring_overflow_flush()
1677 while (!list_empty(&ctx->cq_overflow_list)) { in __io_cqring_overflow_flush()
1683 ocqe = list_first_entry(&ctx->cq_overflow_list, in __io_cqring_overflow_flush()
1686 memcpy(cqe, &ocqe->cqe, sizeof(*cqe)); in __io_cqring_overflow_flush()
1691 list_del(&ocqe->list); in __io_cqring_overflow_flush()
1695 all_flushed = list_empty(&ctx->cq_overflow_list); in __io_cqring_overflow_flush()
1697 clear_bit(0, &ctx->check_cq_overflow); in __io_cqring_overflow_flush()
1698 WRITE_ONCE(ctx->rings->sq_flags, in __io_cqring_overflow_flush()
1699 ctx->rings->sq_flags & ~IORING_SQ_CQ_OVERFLOW); in __io_cqring_overflow_flush()
1704 spin_unlock(&ctx->completion_lock); in __io_cqring_overflow_flush()
1714 if (test_bit(0, &ctx->check_cq_overflow)) { in io_cqring_overflow_flush()
1716 if (ctx->flags & IORING_SETUP_IOPOLL) in io_cqring_overflow_flush()
1717 mutex_lock(&ctx->uring_lock); in io_cqring_overflow_flush()
1719 if (ctx->flags & IORING_SETUP_IOPOLL) in io_cqring_overflow_flush()
1720 mutex_unlock(&ctx->uring_lock); in io_cqring_overflow_flush()
1729 struct io_uring_task *tctx = task->io_uring; in io_put_task()
1732 tctx->cached_refs += nr; in io_put_task()
1734 percpu_counter_sub(&tctx->inflight, nr); in io_put_task()
1735 if (unlikely(atomic_read(&tctx->in_idle))) in io_put_task()
1736 wake_up(&tctx->wait); in io_put_task()
1743 unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR; in io_task_refs_refill()
1745 percpu_counter_add(&tctx->inflight, refill); in io_task_refs_refill()
1746 refcount_add(refill, ¤t->usage); in io_task_refs_refill()
1747 tctx->cached_refs += refill; in io_task_refs_refill()
1752 struct io_uring_task *tctx = current->io_uring; in io_get_task_refs()
1754 tctx->cached_refs -= nr; in io_get_task_refs()
1755 if (unlikely(tctx->cached_refs < 0)) in io_get_task_refs()
1761 struct io_uring_task *tctx = task->io_uring; in io_uring_drop_tctx_refs()
1762 unsigned int refs = tctx->cached_refs; in io_uring_drop_tctx_refs()
1765 tctx->cached_refs = 0; in io_uring_drop_tctx_refs()
1766 percpu_counter_sub(&tctx->inflight, refs); in io_uring_drop_tctx_refs()
1786 if (list_empty(&ctx->cq_overflow_list)) { in io_cqring_event_overflow()
1787 set_bit(0, &ctx->check_cq_overflow); in io_cqring_event_overflow()
1788 WRITE_ONCE(ctx->rings->sq_flags, in io_cqring_event_overflow()
1789 ctx->rings->sq_flags | IORING_SQ_CQ_OVERFLOW); in io_cqring_event_overflow()
1792 ocqe->cqe.user_data = user_data; in io_cqring_event_overflow()
1793 ocqe->cqe.res = res; in io_cqring_event_overflow()
1794 ocqe->cqe.flags = cflags; in io_cqring_event_overflow()
1795 list_add_tail(&ocqe->list, &ctx->cq_overflow_list); in io_cqring_event_overflow()
1808 * submission (by quite a lot). Increment the overflow count in in __io_fill_cqe()
1813 WRITE_ONCE(cqe->user_data, user_data); in __io_fill_cqe()
1814 WRITE_ONCE(cqe->res, res); in __io_fill_cqe()
1815 WRITE_ONCE(cqe->flags, cflags); in __io_fill_cqe()
1823 __io_fill_cqe(req->ctx, req->user_data, res, cflags); in io_fill_cqe_req()
1829 ctx->cq_extra++; in io_fill_cqe_aux()
1836 struct io_ring_ctx *ctx = req->ctx; in io_req_complete_post()
1838 spin_lock(&ctx->completion_lock); in io_req_complete_post()
1839 __io_fill_cqe(ctx, req->user_data, res, cflags); in io_req_complete_post()
1845 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { in io_req_complete_post()
1846 if (req->flags & IO_DISARM_MASK) in io_req_complete_post()
1848 if (req->link) { in io_req_complete_post()
1849 io_req_task_queue(req->link); in io_req_complete_post()
1850 req->link = NULL; in io_req_complete_post()
1854 io_put_task(req->task, 1); in io_req_complete_post()
1855 list_add(&req->inflight_entry, &ctx->locked_free_list); in io_req_complete_post()
1856 ctx->locked_free_nr++; in io_req_complete_post()
1858 if (!percpu_ref_tryget(&ctx->refs)) in io_req_complete_post()
1862 spin_unlock(&ctx->completion_lock); in io_req_complete_post()
1866 percpu_ref_put(&ctx->refs); in io_req_complete_post()
1872 return req->flags & IO_REQ_CLEAN_FLAGS; in io_req_needs_clean()
1880 req->result = res; in io_req_complete_state()
1881 req->compl.cflags = cflags; in io_req_complete_state()
1882 req->flags |= REQ_F_COMPLETE_INLINE; in io_req_complete_state()
1911 req->flags &= ~REQ_F_HARDLINK; in io_req_complete_fail_submit()
1912 req->flags |= REQ_F_LINK; in io_req_complete_fail_submit()
1913 io_req_complete_failed(req, req->result); in io_req_complete_fail_submit()
1922 req->ctx = ctx; in io_preinit_req()
1923 req->link = NULL; in io_preinit_req()
1924 req->async_data = NULL; in io_preinit_req()
1926 req->result = 0; in io_preinit_req()
1932 spin_lock(&ctx->completion_lock); in io_flush_cached_locked_reqs()
1933 list_splice_init(&ctx->locked_free_list, &state->free_list); in io_flush_cached_locked_reqs()
1934 ctx->locked_free_nr = 0; in io_flush_cached_locked_reqs()
1935 spin_unlock(&ctx->completion_lock); in io_flush_cached_locked_reqs()
1941 struct io_submit_state *state = &ctx->submit_state; in io_flush_cached_reqs()
1949 if (READ_ONCE(ctx->locked_free_nr) > IO_COMPL_BATCH) in io_flush_cached_reqs()
1952 nr = state->free_reqs; in io_flush_cached_reqs()
1953 while (!list_empty(&state->free_list)) { in io_flush_cached_reqs()
1954 struct io_kiocb *req = list_first_entry(&state->free_list, in io_flush_cached_reqs()
1957 list_del(&req->inflight_entry); in io_flush_cached_reqs()
1958 state->reqs[nr++] = req; in io_flush_cached_reqs()
1959 if (nr == ARRAY_SIZE(state->reqs)) in io_flush_cached_reqs()
1963 state->free_reqs = nr; in io_flush_cached_reqs()
1970 * Because of that, io_alloc_req() should be called only under ->uring_lock
1974 __must_hold(&ctx->uring_lock) in io_alloc_req()
1976 struct io_submit_state *state = &ctx->submit_state; in io_alloc_req()
1980 BUILD_BUG_ON(ARRAY_SIZE(state->reqs) < IO_REQ_ALLOC_BATCH); in io_alloc_req()
1982 if (likely(state->free_reqs || io_flush_cached_reqs(ctx))) in io_alloc_req()
1986 state->reqs); in io_alloc_req()
1989 * Bulk alloc is all-or-nothing. If we fail to get a batch, in io_alloc_req()
1990 * retry single alloc to be on the safe side. in io_alloc_req()
1993 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp); in io_alloc_req()
1994 if (!state->reqs[0]) in io_alloc_req()
2000 io_preinit_req(state->reqs[i], ctx); in io_alloc_req()
2001 state->free_reqs = ret; in io_alloc_req()
2003 state->free_reqs--; in io_alloc_req()
2004 return state->reqs[state->free_reqs]; in io_alloc_req()
2015 unsigned int flags = req->flags; in io_dismantle_req()
2020 io_put_file(req->file); in io_dismantle_req()
2021 if (req->fixed_rsrc_refs) in io_dismantle_req()
2022 percpu_ref_put(req->fixed_rsrc_refs); in io_dismantle_req()
2023 if (req->async_data) { in io_dismantle_req()
2024 kfree(req->async_data); in io_dismantle_req()
2025 req->async_data = NULL; in io_dismantle_req()
2031 struct io_ring_ctx *ctx = req->ctx; in __io_free_req()
2034 io_put_task(req->task, 1); in __io_free_req()
2036 spin_lock(&ctx->completion_lock); in __io_free_req()
2037 list_add(&req->inflight_entry, &ctx->locked_free_list); in __io_free_req()
2038 ctx->locked_free_nr++; in __io_free_req()
2039 spin_unlock(&ctx->completion_lock); in __io_free_req()
2041 percpu_ref_put(&ctx->refs); in __io_free_req()
2046 struct io_kiocb *nxt = req->link; in io_remove_next_linked()
2048 req->link = nxt->link; in io_remove_next_linked()
2049 nxt->link = NULL; in io_remove_next_linked()
2053 __must_hold(&req->ctx->completion_lock) in io_kill_linked_timeout()
2054 __must_hold(&req->ctx->timeout_lock) in io_kill_linked_timeout()
2056 struct io_kiocb *link = req->link; in io_kill_linked_timeout()
2058 if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { in io_kill_linked_timeout()
2059 struct io_timeout_data *io = link->async_data; in io_kill_linked_timeout()
2062 link->timeout.head = NULL; in io_kill_linked_timeout()
2063 if (hrtimer_try_to_cancel(&io->timer) != -1) { in io_kill_linked_timeout()
2064 list_del(&link->timeout.list); in io_kill_linked_timeout()
2065 io_fill_cqe_req(link, -ECANCELED, 0); in io_kill_linked_timeout()
2074 __must_hold(&req->ctx->completion_lock) in io_fail_links()
2076 struct io_kiocb *nxt, *link = req->link; in io_fail_links()
2078 req->link = NULL; in io_fail_links()
2080 long res = -ECANCELED; in io_fail_links()
2082 if (link->flags & REQ_F_FAIL) in io_fail_links()
2083 res = link->result; in io_fail_links()
2085 nxt = link->link; in io_fail_links()
2086 link->link = NULL; in io_fail_links()
2096 __must_hold(&req->ctx->completion_lock) in io_disarm_next()
2100 if (req->flags & REQ_F_ARM_LTIMEOUT) { in io_disarm_next()
2101 struct io_kiocb *link = req->link; in io_disarm_next()
2103 req->flags &= ~REQ_F_ARM_LTIMEOUT; in io_disarm_next()
2104 if (link && link->opcode == IORING_OP_LINK_TIMEOUT) { in io_disarm_next()
2106 io_fill_cqe_req(link, -ECANCELED, 0); in io_disarm_next()
2110 } else if (req->flags & REQ_F_LINK_TIMEOUT) { in io_disarm_next()
2111 struct io_ring_ctx *ctx = req->ctx; in io_disarm_next()
2113 spin_lock_irq(&ctx->timeout_lock); in io_disarm_next()
2115 spin_unlock_irq(&ctx->timeout_lock); in io_disarm_next()
2117 if (unlikely((req->flags & REQ_F_FAIL) && in io_disarm_next()
2118 !(req->flags & REQ_F_HARDLINK))) { in io_disarm_next()
2119 posted |= (req->link != NULL); in io_disarm_next()
2135 if (req->flags & IO_DISARM_MASK) { in __io_req_find_next()
2136 struct io_ring_ctx *ctx = req->ctx; in __io_req_find_next()
2139 spin_lock(&ctx->completion_lock); in __io_req_find_next()
2142 io_commit_cqring(req->ctx); in __io_req_find_next()
2143 spin_unlock(&ctx->completion_lock); in __io_req_find_next()
2147 nxt = req->link; in __io_req_find_next()
2148 req->link = NULL; in __io_req_find_next()
2154 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK)))) in io_req_find_next()
2164 if (ctx->submit_state.compl_nr) in ctx_flush_and_put()
2166 mutex_unlock(&ctx->uring_lock); in ctx_flush_and_put()
2169 percpu_ref_put(&ctx->refs); in ctx_flush_and_put()
2182 if (!tctx->task_list.first && locked && ctx->submit_state.compl_nr) in tctx_task_work()
2185 spin_lock_irq(&tctx->task_lock); in tctx_task_work()
2186 node = tctx->task_list.first; in tctx_task_work()
2187 INIT_WQ_LIST(&tctx->task_list); in tctx_task_work()
2189 tctx->task_running = false; in tctx_task_work()
2190 spin_unlock_irq(&tctx->task_lock); in tctx_task_work()
2195 struct io_wq_work_node *next = node->next; in tctx_task_work()
2199 if (req->ctx != ctx) { in tctx_task_work()
2201 ctx = req->ctx; in tctx_task_work()
2203 locked = mutex_trylock(&ctx->uring_lock); in tctx_task_work()
2204 percpu_ref_get(&ctx->refs); in tctx_task_work()
2206 req->io_task_work.func(req, &locked); in tctx_task_work()
2215 /* relaxed read is enough as only the task itself sets ->in_idle */ in tctx_task_work()
2216 if (unlikely(atomic_read(&tctx->in_idle))) in tctx_task_work()
2222 struct task_struct *tsk = req->task; in io_req_task_work_add()
2223 struct io_uring_task *tctx = tsk->io_uring; in io_req_task_work_add()
2231 spin_lock_irqsave(&tctx->task_lock, flags); in io_req_task_work_add()
2232 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list); in io_req_task_work_add()
2233 running = tctx->task_running; in io_req_task_work_add()
2235 tctx->task_running = true; in io_req_task_work_add()
2236 spin_unlock_irqrestore(&tctx->task_lock, flags); in io_req_task_work_add()
2248 notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL; in io_req_task_work_add()
2249 if (!task_work_add(tsk, &tctx->task_work, notify)) { in io_req_task_work_add()
2254 spin_lock_irqsave(&tctx->task_lock, flags); in io_req_task_work_add()
2255 tctx->task_running = false; in io_req_task_work_add()
2256 node = tctx->task_list.first; in io_req_task_work_add()
2257 INIT_WQ_LIST(&tctx->task_list); in io_req_task_work_add()
2258 spin_unlock_irqrestore(&tctx->task_lock, flags); in io_req_task_work_add()
2262 node = node->next; in io_req_task_work_add()
2263 if (llist_add(&req->io_task_work.fallback_node, in io_req_task_work_add()
2264 &req->ctx->fallback_llist)) in io_req_task_work_add()
2265 schedule_delayed_work(&req->ctx->fallback_work, 1); in io_req_task_work_add()
2271 struct io_ring_ctx *ctx = req->ctx; in io_req_task_cancel()
2275 io_req_complete_failed(req, req->result); in io_req_task_cancel()
2280 struct io_ring_ctx *ctx = req->ctx; in io_req_task_submit()
2283 /* req->task == current here, checking PF_EXITING is safe */ in io_req_task_submit()
2284 if (likely(!(req->task->flags & PF_EXITING))) in io_req_task_submit()
2287 io_req_complete_failed(req, -EFAULT); in io_req_task_submit()
2292 req->result = ret; in io_req_task_queue_fail()
2293 req->io_task_work.func = io_req_task_cancel; in io_req_task_queue_fail()
2299 req->io_task_work.func = io_req_task_submit; in io_req_task_queue()
2305 req->io_task_work.func = io_queue_async_work; in io_req_task_queue_reissue()
2336 rb->task_refs = 0; in io_init_req_batch()
2337 rb->ctx_refs = 0; in io_init_req_batch()
2338 rb->task = NULL; in io_init_req_batch()
2344 if (rb->ctx_refs) in io_req_free_batch_finish()
2345 percpu_ref_put_many(&ctx->refs, rb->ctx_refs); in io_req_free_batch_finish()
2346 if (rb->task) in io_req_free_batch_finish()
2347 io_put_task(rb->task, rb->task_refs); in io_req_free_batch_finish()
2356 if (req->task != rb->task) { in io_req_free_batch()
2357 if (rb->task) in io_req_free_batch()
2358 io_put_task(rb->task, rb->task_refs); in io_req_free_batch()
2359 rb->task = req->task; in io_req_free_batch()
2360 rb->task_refs = 0; in io_req_free_batch()
2362 rb->task_refs++; in io_req_free_batch()
2363 rb->ctx_refs++; in io_req_free_batch()
2365 if (state->free_reqs != ARRAY_SIZE(state->reqs)) in io_req_free_batch()
2366 state->reqs[state->free_reqs++] = req; in io_req_free_batch()
2368 list_add(&req->inflight_entry, &state->free_list); in io_req_free_batch()
2372 __must_hold(&ctx->uring_lock) in io_submit_flush_completions()
2374 struct io_submit_state *state = &ctx->submit_state; in io_submit_flush_completions()
2375 int i, nr = state->compl_nr; in io_submit_flush_completions()
2378 spin_lock(&ctx->completion_lock); in io_submit_flush_completions()
2380 struct io_kiocb *req = state->compl_reqs[i]; in io_submit_flush_completions()
2382 __io_fill_cqe(ctx, req->user_data, req->result, in io_submit_flush_completions()
2383 req->compl.cflags); in io_submit_flush_completions()
2386 spin_unlock(&ctx->completion_lock); in io_submit_flush_completions()
2391 struct io_kiocb *req = state->compl_reqs[i]; in io_submit_flush_completions()
2394 io_req_free_batch(&rb, req, &ctx->submit_state); in io_submit_flush_completions()
2398 state->compl_nr = 0; in io_submit_flush_completions()
2425 req->io_task_work.func = io_free_req_work; in io_put_req_deferred()
2439 struct io_rings *rings = ctx->rings; in io_sqring_entries()
2442 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head; in io_sqring_entries()
2449 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT; in io_put_kbuf()
2451 req->flags &= ~REQ_F_BUFFER_SELECTED; in io_put_kbuf()
2460 if (likely(!(req->flags & REQ_F_BUFFER_SELECTED))) in io_put_rw_kbuf()
2462 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; in io_put_rw_kbuf()
2468 if (test_thread_flag(TIF_NOTIFY_SIGNAL) || current->task_works) { in io_run_task_work()
2478 * Find and free completed poll iocbs
2486 /* order with ->result store in io_complete_rw_iopoll() */ in io_iopoll_complete()
2492 list_del(&req->inflight_entry); in io_iopoll_complete()
2494 io_fill_cqe_req(req, req->result, io_put_rw_kbuf(req)); in io_iopoll_complete()
2498 io_req_free_batch(&rb, req, &ctx->submit_state); in io_iopoll_complete()
2517 spin = !ctx->poll_multi_queue && *nr_events < min; in io_do_iopoll()
2519 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) { in io_do_iopoll()
2520 struct kiocb *kiocb = &req->rw.kiocb; in io_do_iopoll()
2528 if (READ_ONCE(req->iopoll_completed)) { in io_do_iopoll()
2529 list_move_tail(&req->inflight_entry, &done); in io_do_iopoll()
2535 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin); in io_do_iopoll()
2542 if (READ_ONCE(req->iopoll_completed)) in io_do_iopoll()
2543 list_move_tail(&req->inflight_entry, &done); in io_do_iopoll()
2558 if (!(ctx->flags & IORING_SETUP_IOPOLL)) in io_iopoll_try_reap_events()
2561 mutex_lock(&ctx->uring_lock); in io_iopoll_try_reap_events()
2562 while (!list_empty(&ctx->iopoll_list)) { in io_iopoll_try_reap_events()
2571 * Ensure we allow local-to-the-cpu processing to take place, in io_iopoll_try_reap_events()
2576 mutex_unlock(&ctx->uring_lock); in io_iopoll_try_reap_events()
2578 mutex_lock(&ctx->uring_lock); in io_iopoll_try_reap_events()
2581 mutex_unlock(&ctx->uring_lock); in io_iopoll_try_reap_events()
2594 mutex_lock(&ctx->uring_lock); in io_iopoll_check()
2596 * Don't enter poll loop if we already have events pending. in io_iopoll_check()
2600 if (test_bit(0, &ctx->check_cq_overflow)) in io_iopoll_check()
2609 * of the poll right here, so we need to take a breather every in io_iopoll_check()
2611 * the poll to the issued list. Otherwise we can spin here in io_iopoll_check()
2615 if (list_empty(&ctx->iopoll_list)) { in io_iopoll_check()
2616 u32 tail = ctx->cached_cq_tail; in io_iopoll_check()
2618 mutex_unlock(&ctx->uring_lock); in io_iopoll_check()
2620 mutex_lock(&ctx->uring_lock); in io_iopoll_check()
2623 if (tail != ctx->cached_cq_tail || in io_iopoll_check()
2624 list_empty(&ctx->iopoll_list)) in io_iopoll_check()
2630 mutex_unlock(&ctx->uring_lock); in io_iopoll_check()
2640 if (req->flags & REQ_F_ISREG) { in kiocb_end_write()
2641 struct super_block *sb = file_inode(req->file)->i_sb; in kiocb_end_write()
2651 struct io_async_rw *rw = req->async_data; in io_resubmit_prep()
2655 iov_iter_restore(&rw->iter, &rw->iter_state); in io_resubmit_prep()
2661 umode_t mode = file_inode(req->file)->i_mode; in io_rw_should_reissue()
2662 struct io_ring_ctx *ctx = req->ctx; in io_rw_should_reissue()
2666 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() && in io_rw_should_reissue()
2667 !(ctx->flags & IORING_SETUP_IOPOLL))) in io_rw_should_reissue()
2670 * If ref is dying, we might be running poll reap from the exit work. in io_rw_should_reissue()
2672 * -EAGAIN. in io_rw_should_reissue()
2674 if (percpu_ref_is_dying(&ctx->refs)) in io_rw_should_reissue()
2677 * Play it safe and assume not safe to re-import and reissue if we're in io_rw_should_reissue()
2680 if (!same_thread_group(req->task, current) || !in_task()) in io_rw_should_reissue()
2697 if (req->rw.kiocb.ki_flags & IOCB_WRITE) { in __io_complete_rw_common()
2699 fsnotify_modify(req->file); in __io_complete_rw_common()
2701 fsnotify_access(req->file); in __io_complete_rw_common()
2703 if (res != req->result) { in __io_complete_rw_common()
2704 if ((res == -EAGAIN || res == -EOPNOTSUPP) && in __io_complete_rw_common()
2706 req->flags |= REQ_F_REISSUE; in __io_complete_rw_common()
2710 req->result = res; in __io_complete_rw_common()
2717 struct io_async_rw *io = req->async_data; in io_fixup_rw_res()
2720 if (io && io->bytes_done > 0) { in io_fixup_rw_res()
2722 res = io->bytes_done; in io_fixup_rw_res()
2724 res += io->bytes_done; in io_fixup_rw_res()
2732 int res = req->result; in io_req_task_complete()
2735 struct io_ring_ctx *ctx = req->ctx; in io_req_task_complete()
2736 struct io_submit_state *state = &ctx->submit_state; in io_req_task_complete()
2739 state->compl_reqs[state->compl_nr++] = req; in io_req_task_complete()
2740 if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) in io_req_task_complete()
2761 req->result = io_fixup_rw_res(req, res); in io_complete_rw()
2762 req->io_task_work.func = io_req_task_complete; in io_complete_rw()
2770 if (kiocb->ki_flags & IOCB_WRITE) in io_complete_rw_iopoll()
2772 if (unlikely(res != req->result)) { in io_complete_rw_iopoll()
2773 if (res == -EAGAIN && io_rw_should_reissue(req)) { in io_complete_rw_iopoll()
2774 req->flags |= REQ_F_REISSUE; in io_complete_rw_iopoll()
2779 WRITE_ONCE(req->result, res); in io_complete_rw_iopoll()
2780 /* order with io_iopoll_complete() checking ->result */ in io_complete_rw_iopoll()
2782 WRITE_ONCE(req->iopoll_completed, 1); in io_complete_rw_iopoll()
2786 * After the iocb has been issued, it's safe to be found on the poll list.
2793 struct io_ring_ctx *ctx = req->ctx; in io_iopoll_req_issued()
2798 mutex_lock(&ctx->uring_lock); in io_iopoll_req_issued()
2805 if (list_empty(&ctx->iopoll_list)) { in io_iopoll_req_issued()
2806 ctx->poll_multi_queue = false; in io_iopoll_req_issued()
2807 } else if (!ctx->poll_multi_queue) { in io_iopoll_req_issued()
2811 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb, in io_iopoll_req_issued()
2814 if (list_req->file != req->file) { in io_iopoll_req_issued()
2815 ctx->poll_multi_queue = true; in io_iopoll_req_issued()
2817 queue_num0 = blk_qc_t_to_queue_num(list_req->rw.kiocb.ki_cookie); in io_iopoll_req_issued()
2818 queue_num1 = blk_qc_t_to_queue_num(req->rw.kiocb.ki_cookie); in io_iopoll_req_issued()
2820 ctx->poll_multi_queue = true; in io_iopoll_req_issued()
2828 if (READ_ONCE(req->iopoll_completed)) in io_iopoll_req_issued()
2829 list_add(&req->inflight_entry, &ctx->iopoll_list); in io_iopoll_req_issued()
2831 list_add_tail(&req->inflight_entry, &ctx->iopoll_list); in io_iopoll_req_issued()
2840 if ((ctx->flags & IORING_SETUP_SQPOLL) && in io_iopoll_req_issued()
2841 wq_has_sleeper(&ctx->sq_data->wait)) in io_iopoll_req_issued()
2842 wake_up(&ctx->sq_data->wait); in io_iopoll_req_issued()
2844 mutex_unlock(&ctx->uring_lock); in io_iopoll_req_issued()
2860 umode_t mode = file_inode(file)->i_mode; in __io_file_supports_nowait()
2864 io_bdev_nowait(I_BDEV(file->f_mapping->host))) in __io_file_supports_nowait()
2872 io_bdev_nowait(file->f_inode->i_sb->s_bdev) && in __io_file_supports_nowait()
2873 file->f_op != &io_uring_fops) in __io_file_supports_nowait()
2878 /* any ->read/write should understand O_NONBLOCK */ in __io_file_supports_nowait()
2879 if (file->f_flags & O_NONBLOCK) in __io_file_supports_nowait()
2882 if (!(file->f_mode & FMODE_NOWAIT)) in __io_file_supports_nowait()
2886 return file->f_op->read_iter != NULL; in __io_file_supports_nowait()
2888 return file->f_op->write_iter != NULL; in __io_file_supports_nowait()
2893 if (rw == READ && (req->flags & REQ_F_NOWAIT_READ)) in io_file_supports_nowait()
2895 else if (rw == WRITE && (req->flags & REQ_F_NOWAIT_WRITE)) in io_file_supports_nowait()
2898 return __io_file_supports_nowait(req->file, rw); in io_file_supports_nowait()
2904 struct io_ring_ctx *ctx = req->ctx; in io_prep_rw()
2905 struct kiocb *kiocb = &req->rw.kiocb; in io_prep_rw()
2906 struct file *file = req->file; in io_prep_rw()
2910 if (!io_req_ffs_set(req) && S_ISREG(file_inode(file)->i_mode)) in io_prep_rw()
2911 req->flags |= REQ_F_ISREG; in io_prep_rw()
2913 kiocb->ki_pos = READ_ONCE(sqe->off); in io_prep_rw()
2914 if (kiocb->ki_pos == -1) { in io_prep_rw()
2915 if (!(file->f_mode & FMODE_STREAM)) { in io_prep_rw()
2916 req->flags |= REQ_F_CUR_POS; in io_prep_rw()
2917 kiocb->ki_pos = file->f_pos; in io_prep_rw()
2919 kiocb->ki_pos = 0; in io_prep_rw()
2922 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp)); in io_prep_rw()
2923 kiocb->ki_flags = iocb_flags(kiocb->ki_filp); in io_prep_rw()
2924 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags)); in io_prep_rw()
2929 * If the file is marked O_NONBLOCK, still allow retry for it if it in io_prep_rw()
2931 * reliably. If not, or it IOCB_NOWAIT is set, don't retry. in io_prep_rw()
2933 if ((kiocb->ki_flags & IOCB_NOWAIT) || in io_prep_rw()
2934 ((file->f_flags & O_NONBLOCK) && !io_file_supports_nowait(req, rw))) in io_prep_rw()
2935 req->flags |= REQ_F_NOWAIT; in io_prep_rw()
2937 ioprio = READ_ONCE(sqe->ioprio); in io_prep_rw()
2943 kiocb->ki_ioprio = ioprio; in io_prep_rw()
2945 kiocb->ki_ioprio = get_current_ioprio(); in io_prep_rw()
2947 if (ctx->flags & IORING_SETUP_IOPOLL) { in io_prep_rw()
2948 if (!(kiocb->ki_flags & IOCB_DIRECT) || in io_prep_rw()
2949 !kiocb->ki_filp->f_op->iopoll) in io_prep_rw()
2950 return -EOPNOTSUPP; in io_prep_rw()
2952 kiocb->ki_flags |= IOCB_HIPRI; in io_prep_rw()
2953 kiocb->ki_complete = io_complete_rw_iopoll; in io_prep_rw()
2954 req->iopoll_completed = 0; in io_prep_rw()
2956 if (kiocb->ki_flags & IOCB_HIPRI) in io_prep_rw()
2957 return -EINVAL; in io_prep_rw()
2958 kiocb->ki_complete = io_complete_rw; in io_prep_rw()
2961 /* used for fixed read/write too - just read unconditionally */ in io_prep_rw()
2962 req->buf_index = READ_ONCE(sqe->buf_index); in io_prep_rw()
2963 req->imu = NULL; in io_prep_rw()
2965 if (req->opcode == IORING_OP_READ_FIXED || in io_prep_rw()
2966 req->opcode == IORING_OP_WRITE_FIXED) { in io_prep_rw()
2967 struct io_ring_ctx *ctx = req->ctx; in io_prep_rw()
2970 if (unlikely(req->buf_index >= ctx->nr_user_bufs)) in io_prep_rw()
2971 return -EFAULT; in io_prep_rw()
2972 index = array_index_nospec(req->buf_index, ctx->nr_user_bufs); in io_prep_rw()
2973 req->imu = ctx->user_bufs[index]; in io_prep_rw()
2977 req->rw.addr = READ_ONCE(sqe->addr); in io_prep_rw()
2978 req->rw.len = READ_ONCE(sqe->len); in io_prep_rw()
2985 case -EIOCBQUEUED: in io_rw_done()
2987 case -ERESTARTSYS: in io_rw_done()
2988 case -ERESTARTNOINTR: in io_rw_done()
2989 case -ERESTARTNOHAND: in io_rw_done()
2990 case -ERESTART_RESTARTBLOCK: in io_rw_done()
2996 ret = -EINTR; in io_rw_done()
2999 kiocb->ki_complete(kiocb, ret, 0); in io_rw_done()
3008 if (req->flags & REQ_F_CUR_POS) in kiocb_done()
3009 req->file->f_pos = kiocb->ki_pos; in kiocb_done()
3010 if (ret >= 0 && (kiocb->ki_complete == io_complete_rw)) in kiocb_done()
3015 if (req->flags & REQ_F_REISSUE) { in kiocb_done()
3016 req->flags &= ~REQ_F_REISSUE; in kiocb_done()
3021 struct io_ring_ctx *ctx = req->ctx; in kiocb_done()
3026 mutex_lock(&ctx->uring_lock); in kiocb_done()
3028 mutex_unlock(&ctx->uring_lock); in kiocb_done()
3039 size_t len = req->rw.len; in __io_import_fixed()
3040 u64 buf_end, buf_addr = req->rw.addr; in __io_import_fixed()
3044 return -EFAULT; in __io_import_fixed()
3046 if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end)) in __io_import_fixed()
3047 return -EFAULT; in __io_import_fixed()
3053 offset = buf_addr - imu->ubuf; in __io_import_fixed()
3054 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len); in __io_import_fixed()
3059 * using the latter parts of a big fixed buffer - it iterates in __io_import_fixed()
3073 const struct bio_vec *bvec = imu->bvec; in __io_import_fixed()
3075 if (offset <= bvec->bv_len) { in __io_import_fixed()
3081 offset -= bvec->bv_len; in __io_import_fixed()
3084 iter->bvec = bvec + seg_skip; in __io_import_fixed()
3085 iter->nr_segs -= seg_skip; in __io_import_fixed()
3086 iter->count -= bvec->bv_len + offset; in __io_import_fixed()
3087 iter->iov_offset = offset & ~PAGE_MASK; in __io_import_fixed()
3096 if (WARN_ON_ONCE(!req->imu)) in io_import_fixed()
3097 return -EFAULT; in io_import_fixed()
3098 return __io_import_fixed(req, rw, iter, req->imu); in io_import_fixed()
3104 mutex_unlock(&ctx->uring_lock); in io_ring_submit_unlock()
3116 mutex_lock(&ctx->uring_lock); in io_ring_submit_lock()
3125 if (req->flags & REQ_F_BUFFER_SELECTED) in io_buffer_select()
3128 io_ring_submit_lock(req->ctx, needs_lock); in io_buffer_select()
3130 lockdep_assert_held(&req->ctx->uring_lock); in io_buffer_select()
3132 head = xa_load(&req->ctx->io_buffers, bgid); in io_buffer_select()
3134 if (!list_empty(&head->list)) { in io_buffer_select()
3135 kbuf = list_last_entry(&head->list, struct io_buffer, in io_buffer_select()
3137 list_del(&kbuf->list); in io_buffer_select()
3140 xa_erase(&req->ctx->io_buffers, bgid); in io_buffer_select()
3142 if (*len > kbuf->len) in io_buffer_select()
3143 *len = kbuf->len; in io_buffer_select()
3145 kbuf = ERR_PTR(-ENOBUFS); in io_buffer_select()
3148 io_ring_submit_unlock(req->ctx, needs_lock); in io_buffer_select()
3159 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; in io_rw_buffer_select()
3160 bgid = req->buf_index; in io_rw_buffer_select()
3164 req->rw.addr = (u64) (unsigned long) kbuf; in io_rw_buffer_select()
3165 req->flags |= REQ_F_BUFFER_SELECTED; in io_rw_buffer_select()
3166 return u64_to_user_ptr(kbuf->addr); in io_rw_buffer_select()
3178 uiov = u64_to_user_ptr(req->rw.addr); in io_compat_import()
3180 return -EFAULT; in io_compat_import()
3181 if (__get_user(clen, &uiov->iov_len)) in io_compat_import()
3182 return -EFAULT; in io_compat_import()
3184 return -EINVAL; in io_compat_import()
3199 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr); in __io_iov_buffer_select()
3204 return -EFAULT; in __io_iov_buffer_select()
3208 return -EINVAL; in __io_iov_buffer_select()
3220 if (req->flags & REQ_F_BUFFER_SELECTED) { in io_iov_buffer_select()
3223 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr; in io_iov_buffer_select()
3224 iov[0].iov_base = u64_to_user_ptr(kbuf->addr); in io_iov_buffer_select()
3225 iov[0].iov_len = kbuf->len; in io_iov_buffer_select()
3228 if (req->rw.len != 1) in io_iov_buffer_select()
3229 return -EINVAL; in io_iov_buffer_select()
3232 if (req->ctx->compat) in io_iov_buffer_select()
3242 void __user *buf = u64_to_user_ptr(req->rw.addr); in io_import_iovec()
3243 size_t sqe_len = req->rw.len; in io_import_iovec()
3244 u8 opcode = req->opcode; in io_import_iovec()
3253 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT)) in io_import_iovec()
3254 return -EINVAL; in io_import_iovec()
3257 if (req->flags & REQ_F_BUFFER_SELECT) { in io_import_iovec()
3261 req->rw.len = sqe_len; in io_import_iovec()
3269 if (req->flags & REQ_F_BUFFER_SELECT) { in io_import_iovec()
3272 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len); in io_import_iovec()
3278 req->ctx->compat); in io_import_iovec()
3283 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos; in io_kiocb_ppos()
3287 * For files that don't have ->read_iter() and ->write_iter(), handle them
3288 * by looping over ->read() or ->write() manually.
3292 struct kiocb *kiocb = &req->rw.kiocb; in loop_rw_iter()
3293 struct file *file = req->file; in loop_rw_iter()
3298 * support non-blocking either. For the latter, this just causes in loop_rw_iter()
3301 if (kiocb->ki_flags & IOCB_HIPRI) in loop_rw_iter()
3302 return -EOPNOTSUPP; in loop_rw_iter()
3303 if (kiocb->ki_flags & IOCB_NOWAIT) in loop_rw_iter()
3304 return -EAGAIN; in loop_rw_iter()
3313 iovec.iov_base = u64_to_user_ptr(req->rw.addr); in loop_rw_iter()
3314 iovec.iov_len = req->rw.len; in loop_rw_iter()
3318 nr = file->f_op->read(file, iovec.iov_base, in loop_rw_iter()
3321 nr = file->f_op->write(file, iovec.iov_base, in loop_rw_iter()
3334 req->rw.addr += nr; in loop_rw_iter()
3335 req->rw.len -= nr; in loop_rw_iter()
3336 if (!req->rw.len) in loop_rw_iter()
3349 struct io_async_rw *rw = req->async_data; in io_req_map_rw()
3351 memcpy(&rw->iter, iter, sizeof(*iter)); in io_req_map_rw()
3352 rw->free_iovec = iovec; in io_req_map_rw()
3353 rw->bytes_done = 0; in io_req_map_rw()
3360 rw->iter.iov = rw->fast_iov; in io_req_map_rw()
3361 if (iter->iov != fast_iov) { in io_req_map_rw()
3362 iov_off = iter->iov - fast_iov; in io_req_map_rw()
3363 rw->iter.iov += iov_off; in io_req_map_rw()
3365 if (rw->fast_iov != fast_iov) in io_req_map_rw()
3366 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off, in io_req_map_rw()
3367 sizeof(struct iovec) * iter->nr_segs); in io_req_map_rw()
3369 req->flags |= REQ_F_NEED_CLEANUP; in io_req_map_rw()
3375 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size); in io_alloc_async_data()
3376 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL); in io_alloc_async_data()
3377 return req->async_data == NULL; in io_alloc_async_data()
3384 if (!force && !io_op_defs[req->opcode].needs_async_setup) in io_setup_async_rw()
3386 if (!req->async_data) { in io_setup_async_rw()
3391 return -ENOMEM; in io_setup_async_rw()
3395 iorw = req->async_data; in io_setup_async_rw()
3397 iov_iter_save_state(&iorw->iter, &iorw->iter_state); in io_setup_async_rw()
3404 struct io_async_rw *iorw = req->async_data; in io_rw_prep_async()
3405 struct iovec *iov = iorw->fast_iov; in io_rw_prep_async()
3408 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false); in io_rw_prep_async()
3412 iorw->bytes_done = 0; in io_rw_prep_async()
3413 iorw->free_iovec = iov; in io_rw_prep_async()
3415 req->flags |= REQ_F_NEED_CLEANUP; in io_rw_prep_async()
3416 iov_iter_save_state(&iorw->iter, &iorw->iter_state); in io_rw_prep_async()
3422 if (unlikely(!(req->file->f_mode & FMODE_READ))) in io_read_prep()
3423 return -EBADF; in io_read_prep()
3432 * queue a task_work based retry of the operation, attempting to copy the data
3434 * do a thread based blocking retry of the operation. That's the unexpected
3441 struct io_kiocb *req = wait->private; in io_async_buf_func()
3449 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ; in io_async_buf_func()
3450 list_del_init(&wait->entry); in io_async_buf_func()
3457 * based retry. If we return false here, the request is handed to the async
3458 * worker threads for retry. If we're doing buffered reads on a regular file,
3459 * we prepare a private wait_page_queue entry and retry the operation. This
3462 * that callback, io_uring uses task_work to setup a retry of the operation.
3463 * That retry will attempt the buffered read again. The retry will generally
3465 * async worker threads for a blocking retry.
3469 struct io_async_rw *rw = req->async_data; in io_rw_should_retry()
3470 struct wait_page_queue *wait = &rw->wpq; in io_rw_should_retry()
3471 struct kiocb *kiocb = &req->rw.kiocb; in io_rw_should_retry()
3473 /* never retry for NOWAIT, we just complete with -EAGAIN */ in io_rw_should_retry()
3474 if (req->flags & REQ_F_NOWAIT) in io_rw_should_retry()
3478 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI)) in io_rw_should_retry()
3482 * just use poll if we can, and don't attempt if the fs doesn't in io_rw_should_retry()
3485 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC)) in io_rw_should_retry()
3488 wait->wait.func = io_async_buf_func; in io_rw_should_retry()
3489 wait->wait.private = req; in io_rw_should_retry()
3490 wait->wait.flags = 0; in io_rw_should_retry()
3491 INIT_LIST_HEAD(&wait->wait.entry); in io_rw_should_retry()
3492 kiocb->ki_flags |= IOCB_WAITQ; in io_rw_should_retry()
3493 kiocb->ki_flags &= ~IOCB_NOWAIT; in io_rw_should_retry()
3494 kiocb->ki_waitq = wait; in io_rw_should_retry()
3500 if (req->file->f_op->read_iter) in io_iter_do_read()
3501 return call_read_iter(req->file, &req->rw.kiocb, iter); in io_iter_do_read()
3502 else if (req->file->f_op->read) in io_iter_do_read()
3505 return -EINVAL; in io_iter_do_read()
3510 return req->flags & REQ_F_ISREG || in need_read_all()
3511 S_ISBLK(file_inode(req->file)->i_mode); in need_read_all()
3517 struct kiocb *kiocb = &req->rw.kiocb; in io_read()
3519 struct io_async_rw *rw = req->async_data; in io_read()
3525 iter = &rw->iter; in io_read()
3526 state = &rw->iter_state; in io_read()
3541 req->result = iov_iter_count(iter); in io_read()
3543 /* Ensure we clear previously set non-block flag */ in io_read()
3545 kiocb->ki_flags &= ~IOCB_NOWAIT; in io_read()
3547 kiocb->ki_flags |= IOCB_NOWAIT; in io_read()
3552 return ret ?: -EAGAIN; in io_read()
3555 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), req->result); in io_read()
3563 if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) { in io_read()
3564 req->flags &= ~REQ_F_REISSUE; in io_read()
3565 /* IOPOLL retry should happen for io-wq threads */ in io_read()
3566 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_read()
3568 /* no retry on NONBLOCK nor RWF_NOWAIT */ in io_read()
3569 if (req->flags & REQ_F_NOWAIT) in io_read()
3572 } else if (ret == -EIOCBQUEUED) { in io_read()
3574 } else if (ret <= 0 || ret == req->result || !force_nonblock || in io_read()
3575 (req->flags & REQ_F_NOWAIT) || !need_read_all(req)) { in io_read()
3576 /* read all, failed, already did sync or don't want to retry */ in io_read()
3592 rw = req->async_data; in io_read()
3597 if (iter != &rw->iter) { in io_read()
3598 iter = &rw->iter; in io_read()
3599 state = &rw->iter_state; in io_read()
3611 rw->bytes_done += ret; in io_read()
3614 /* if we can retry, do so with the callbacks armed */ in io_read()
3616 kiocb->ki_flags &= ~IOCB_WAITQ; in io_read()
3617 return -EAGAIN; in io_read()
3620 req->result = iov_iter_count(iter); in io_read()
3622 * Now retry read with the IOCB_WAITQ parts set in the iocb. If in io_read()
3623 * we get -EIOCBQUEUED, then we'll get a notification when the in io_read()
3625 * here, and if we do, then just retry at the new offset. in io_read()
3628 if (ret == -EIOCBQUEUED) in io_read()
3630 /* we got some bytes, but not all. retry. */ in io_read()
3631 kiocb->ki_flags &= ~IOCB_WAITQ; in io_read()
3645 if (unlikely(!(req->file->f_mode & FMODE_WRITE))) in io_write_prep()
3646 return -EBADF; in io_write_prep()
3653 struct kiocb *kiocb = &req->rw.kiocb; in io_write()
3655 struct io_async_rw *rw = req->async_data; in io_write()
3661 iter = &rw->iter; in io_write()
3662 state = &rw->iter_state; in io_write()
3672 req->result = iov_iter_count(iter); in io_write()
3674 /* Ensure we clear previously set non-block flag */ in io_write()
3676 kiocb->ki_flags &= ~IOCB_NOWAIT; in io_write()
3678 kiocb->ki_flags |= IOCB_NOWAIT; in io_write()
3684 /* file path doesn't support NOWAIT for non-direct_IO */ in io_write()
3685 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) && in io_write()
3686 (req->flags & REQ_F_ISREG)) in io_write()
3689 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), req->result); in io_write()
3694 * Open-code file_start_write here to grab freeze protection, in io_write()
3700 if (req->flags & REQ_F_ISREG) { in io_write()
3701 sb_start_write(file_inode(req->file)->i_sb); in io_write()
3702 __sb_writers_release(file_inode(req->file)->i_sb, in io_write()
3705 kiocb->ki_flags |= IOCB_WRITE; in io_write()
3707 if (req->file->f_op->write_iter) in io_write()
3708 ret2 = call_write_iter(req->file, kiocb, iter); in io_write()
3709 else if (req->file->f_op->write) in io_write()
3712 ret2 = -EINVAL; in io_write()
3714 if (req->flags & REQ_F_REISSUE) { in io_write()
3715 req->flags &= ~REQ_F_REISSUE; in io_write()
3716 ret2 = -EAGAIN; in io_write()
3720 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just in io_write()
3721 * retry them without IOCB_NOWAIT. in io_write()
3723 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT)) in io_write()
3724 ret2 = -EAGAIN; in io_write()
3725 /* no retry on NONBLOCK nor RWF_NOWAIT */ in io_write()
3726 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT)) in io_write()
3728 if (!force_nonblock || ret2 != -EAGAIN) { in io_write()
3729 /* IOPOLL retry should happen for io-wq threads */ in io_write()
3730 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN) in io_write()
3739 if (kiocb->ki_flags & IOCB_WRITE) in io_write()
3741 return -EAGAIN; in io_write()
3755 struct io_rename *ren = &req->rename; in io_renameat_prep()
3758 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_renameat_prep()
3759 return -EINVAL; in io_renameat_prep()
3760 if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) in io_renameat_prep()
3761 return -EINVAL; in io_renameat_prep()
3762 if (unlikely(req->flags & REQ_F_FIXED_FILE)) in io_renameat_prep()
3763 return -EBADF; in io_renameat_prep()
3765 ren->old_dfd = READ_ONCE(sqe->fd); in io_renameat_prep()
3766 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_renameat_prep()
3767 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2)); in io_renameat_prep()
3768 ren->new_dfd = READ_ONCE(sqe->len); in io_renameat_prep()
3769 ren->flags = READ_ONCE(sqe->rename_flags); in io_renameat_prep()
3771 ren->oldpath = getname(oldf); in io_renameat_prep()
3772 if (IS_ERR(ren->oldpath)) in io_renameat_prep()
3773 return PTR_ERR(ren->oldpath); in io_renameat_prep()
3775 ren->newpath = getname(newf); in io_renameat_prep()
3776 if (IS_ERR(ren->newpath)) { in io_renameat_prep()
3777 putname(ren->oldpath); in io_renameat_prep()
3778 return PTR_ERR(ren->newpath); in io_renameat_prep()
3781 req->flags |= REQ_F_NEED_CLEANUP; in io_renameat_prep()
3787 struct io_rename *ren = &req->rename; in io_renameat()
3791 return -EAGAIN; in io_renameat()
3793 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd, in io_renameat()
3794 ren->newpath, ren->flags); in io_renameat()
3796 req->flags &= ~REQ_F_NEED_CLEANUP; in io_renameat()
3806 struct io_unlink *un = &req->unlink; in io_unlinkat_prep()
3809 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_unlinkat_prep()
3810 return -EINVAL; in io_unlinkat_prep()
3811 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index || in io_unlinkat_prep()
3812 sqe->splice_fd_in) in io_unlinkat_prep()
3813 return -EINVAL; in io_unlinkat_prep()
3814 if (unlikely(req->flags & REQ_F_FIXED_FILE)) in io_unlinkat_prep()
3815 return -EBADF; in io_unlinkat_prep()
3817 un->dfd = READ_ONCE(sqe->fd); in io_unlinkat_prep()
3819 un->flags = READ_ONCE(sqe->unlink_flags); in io_unlinkat_prep()
3820 if (un->flags & ~AT_REMOVEDIR) in io_unlinkat_prep()
3821 return -EINVAL; in io_unlinkat_prep()
3823 fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_unlinkat_prep()
3824 un->filename = getname(fname); in io_unlinkat_prep()
3825 if (IS_ERR(un->filename)) in io_unlinkat_prep()
3826 return PTR_ERR(un->filename); in io_unlinkat_prep()
3828 req->flags |= REQ_F_NEED_CLEANUP; in io_unlinkat_prep()
3834 struct io_unlink *un = &req->unlink; in io_unlinkat()
3838 return -EAGAIN; in io_unlinkat()
3840 if (un->flags & AT_REMOVEDIR) in io_unlinkat()
3841 ret = do_rmdir(un->dfd, un->filename); in io_unlinkat()
3843 ret = do_unlinkat(un->dfd, un->filename); in io_unlinkat()
3845 req->flags &= ~REQ_F_NEED_CLEANUP; in io_unlinkat()
3856 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_shutdown_prep()
3857 return -EINVAL; in io_shutdown_prep()
3858 if (unlikely(sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags || in io_shutdown_prep()
3859 sqe->buf_index || sqe->splice_fd_in)) in io_shutdown_prep()
3860 return -EINVAL; in io_shutdown_prep()
3862 req->shutdown.how = READ_ONCE(sqe->len); in io_shutdown_prep()
3865 return -EOPNOTSUPP; in io_shutdown_prep()
3876 return -EAGAIN; in io_shutdown()
3878 sock = sock_from_file(req->file, &ret); in io_shutdown()
3882 ret = __sys_shutdown_sock(sock, req->shutdown.how); in io_shutdown()
3888 return -EOPNOTSUPP; in io_shutdown()
3895 struct io_splice *sp = &req->splice; in __io_splice_prep()
3898 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in __io_splice_prep()
3899 return -EINVAL; in __io_splice_prep()
3901 sp->len = READ_ONCE(sqe->len); in __io_splice_prep()
3902 sp->flags = READ_ONCE(sqe->splice_flags); in __io_splice_prep()
3903 if (unlikely(sp->flags & ~valid_flags)) in __io_splice_prep()
3904 return -EINVAL; in __io_splice_prep()
3905 sp->splice_fd_in = READ_ONCE(sqe->splice_fd_in); in __io_splice_prep()
3912 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off)) in io_tee_prep()
3913 return -EINVAL; in io_tee_prep()
3919 struct io_splice *sp = &req->splice; in io_tee()
3920 struct file *out = sp->file_out; in io_tee()
3921 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; in io_tee()
3926 return -EAGAIN; in io_tee()
3928 in = io_file_get(req->ctx, req, sp->splice_fd_in, in io_tee()
3929 (sp->flags & SPLICE_F_FD_IN_FIXED)); in io_tee()
3931 ret = -EBADF; in io_tee()
3935 if (sp->len) in io_tee()
3936 ret = do_tee(in, out, sp->len, flags); in io_tee()
3938 if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) in io_tee()
3941 if (ret != sp->len) in io_tee()
3949 struct io_splice *sp = &req->splice; in io_splice_prep()
3951 sp->off_in = READ_ONCE(sqe->splice_off_in); in io_splice_prep()
3952 sp->off_out = READ_ONCE(sqe->off); in io_splice_prep()
3958 struct io_splice *sp = &req->splice; in io_splice()
3959 struct file *out = sp->file_out; in io_splice()
3960 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED; in io_splice()
3966 return -EAGAIN; in io_splice()
3968 in = io_file_get(req->ctx, req, sp->splice_fd_in, in io_splice()
3969 (sp->flags & SPLICE_F_FD_IN_FIXED)); in io_splice()
3971 ret = -EBADF; in io_splice()
3975 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in; in io_splice()
3976 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out; in io_splice()
3978 if (sp->len) in io_splice()
3979 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags); in io_splice()
3981 if (!(sp->flags & SPLICE_F_FD_IN_FIXED)) in io_splice()
3984 if (ret != sp->len) in io_splice()
3995 struct io_ring_ctx *ctx = req->ctx; in io_nop()
3997 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) in io_nop()
3998 return -EINVAL; in io_nop()
4006 struct io_ring_ctx *ctx = req->ctx; in io_fsync_prep()
4008 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) in io_fsync_prep()
4009 return -EINVAL; in io_fsync_prep()
4010 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || in io_fsync_prep()
4011 sqe->splice_fd_in)) in io_fsync_prep()
4012 return -EINVAL; in io_fsync_prep()
4014 req->sync.flags = READ_ONCE(sqe->fsync_flags); in io_fsync_prep()
4015 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC)) in io_fsync_prep()
4016 return -EINVAL; in io_fsync_prep()
4018 req->sync.off = READ_ONCE(sqe->off); in io_fsync_prep()
4019 req->sync.len = READ_ONCE(sqe->len); in io_fsync_prep()
4025 loff_t end = req->sync.off + req->sync.len; in io_fsync()
4030 return -EAGAIN; in io_fsync()
4032 ret = vfs_fsync_range(req->file, req->sync.off, in io_fsync()
4034 req->sync.flags & IORING_FSYNC_DATASYNC); in io_fsync()
4044 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags || in io_fallocate_prep()
4045 sqe->splice_fd_in) in io_fallocate_prep()
4046 return -EINVAL; in io_fallocate_prep()
4047 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_fallocate_prep()
4048 return -EINVAL; in io_fallocate_prep()
4050 req->sync.off = READ_ONCE(sqe->off); in io_fallocate_prep()
4051 req->sync.len = READ_ONCE(sqe->addr); in io_fallocate_prep()
4052 req->sync.mode = READ_ONCE(sqe->len); in io_fallocate_prep()
4062 return -EAGAIN; in io_fallocate()
4063 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off, in io_fallocate()
4064 req->sync.len); in io_fallocate()
4068 fsnotify_modify(req->file); in io_fallocate()
4078 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in __io_openat_prep()
4079 return -EINVAL; in __io_openat_prep()
4080 if (unlikely(sqe->ioprio || sqe->buf_index)) in __io_openat_prep()
4081 return -EINVAL; in __io_openat_prep()
4082 if (unlikely(req->flags & REQ_F_FIXED_FILE)) in __io_openat_prep()
4083 return -EBADF; in __io_openat_prep()
4086 if (!(req->open.how.flags & O_PATH) && force_o_largefile()) in __io_openat_prep()
4087 req->open.how.flags |= O_LARGEFILE; in __io_openat_prep()
4089 req->open.dfd = READ_ONCE(sqe->fd); in __io_openat_prep()
4090 fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); in __io_openat_prep()
4091 req->open.filename = getname(fname); in __io_openat_prep()
4092 if (IS_ERR(req->open.filename)) { in __io_openat_prep()
4093 ret = PTR_ERR(req->open.filename); in __io_openat_prep()
4094 req->open.filename = NULL; in __io_openat_prep()
4098 req->open.file_slot = READ_ONCE(sqe->file_index); in __io_openat_prep()
4099 if (req->open.file_slot && (req->open.how.flags & O_CLOEXEC)) in __io_openat_prep()
4100 return -EINVAL; in __io_openat_prep()
4102 req->open.nofile = rlimit(RLIMIT_NOFILE); in __io_openat_prep()
4103 req->flags |= REQ_F_NEED_CLEANUP; in __io_openat_prep()
4109 u64 mode = READ_ONCE(sqe->len); in io_openat_prep()
4110 u64 flags = READ_ONCE(sqe->open_flags); in io_openat_prep()
4112 req->open.how = build_open_how(flags, mode); in io_openat_prep()
4122 how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); in io_openat2_prep()
4123 len = READ_ONCE(sqe->len); in io_openat2_prep()
4125 return -EINVAL; in io_openat2_prep()
4127 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how, in io_openat2_prep()
4140 bool fixed = !!req->open.file_slot; in io_openat2()
4143 ret = build_open_flags(&req->open.how, &op); in io_openat2()
4147 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED; in io_openat2()
4151 * it'll always -EAGAIN in io_openat2()
4153 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE)) in io_openat2()
4154 return -EAGAIN; in io_openat2()
4160 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile); in io_openat2()
4165 file = do_filp_open(req->open.dfd, req->open.filename, &op); in io_openat2()
4170 * path. So just put it, and we'll get a new one when we retry. in io_openat2()
4176 /* only retry if RESOLVE_CACHED wasn't already set by application */ in io_openat2()
4177 if (ret == -EAGAIN && in io_openat2()
4179 return -EAGAIN; in io_openat2()
4184 file->f_flags &= ~O_NONBLOCK; in io_openat2()
4191 req->open.file_slot - 1); in io_openat2()
4193 putname(req->open.filename); in io_openat2()
4194 req->flags &= ~REQ_F_NEED_CLEANUP; in io_openat2()
4209 struct io_provide_buf *p = &req->pbuf; in io_remove_buffers_prep()
4212 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off || in io_remove_buffers_prep()
4213 sqe->splice_fd_in) in io_remove_buffers_prep()
4214 return -EINVAL; in io_remove_buffers_prep()
4216 tmp = READ_ONCE(sqe->fd); in io_remove_buffers_prep()
4218 return -EINVAL; in io_remove_buffers_prep()
4221 p->nbufs = tmp; in io_remove_buffers_prep()
4222 p->bgid = READ_ONCE(sqe->buf_group); in io_remove_buffers_prep()
4236 while (!list_empty(&buf->list)) { in __io_remove_buffers()
4239 nxt = list_first_entry(&buf->list, struct io_buffer, list); in __io_remove_buffers()
4240 list_del(&nxt->list); in __io_remove_buffers()
4248 xa_erase(&ctx->io_buffers, bgid); in __io_remove_buffers()
4255 struct io_provide_buf *p = &req->pbuf; in io_remove_buffers()
4256 struct io_ring_ctx *ctx = req->ctx; in io_remove_buffers()
4263 lockdep_assert_held(&ctx->uring_lock); in io_remove_buffers()
4265 ret = -ENOENT; in io_remove_buffers()
4266 head = xa_load(&ctx->io_buffers, p->bgid); in io_remove_buffers()
4268 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs); in io_remove_buffers()
4282 struct io_provide_buf *p = &req->pbuf; in io_provide_buffers_prep()
4285 if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) in io_provide_buffers_prep()
4286 return -EINVAL; in io_provide_buffers_prep()
4288 tmp = READ_ONCE(sqe->fd); in io_provide_buffers_prep()
4290 return -E2BIG; in io_provide_buffers_prep()
4291 p->nbufs = tmp; in io_provide_buffers_prep()
4292 p->addr = READ_ONCE(sqe->addr); in io_provide_buffers_prep()
4293 p->len = READ_ONCE(sqe->len); in io_provide_buffers_prep()
4295 if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs, in io_provide_buffers_prep()
4297 return -EOVERFLOW; in io_provide_buffers_prep()
4298 if (check_add_overflow((unsigned long)p->addr, size, &tmp_check)) in io_provide_buffers_prep()
4299 return -EOVERFLOW; in io_provide_buffers_prep()
4301 size = (unsigned long)p->len * p->nbufs; in io_provide_buffers_prep()
4302 if (!access_ok(u64_to_user_ptr(p->addr), size)) in io_provide_buffers_prep()
4303 return -EFAULT; in io_provide_buffers_prep()
4305 p->bgid = READ_ONCE(sqe->buf_group); in io_provide_buffers_prep()
4306 tmp = READ_ONCE(sqe->off); in io_provide_buffers_prep()
4308 return -E2BIG; in io_provide_buffers_prep()
4309 p->bid = tmp; in io_provide_buffers_prep()
4316 u64 addr = pbuf->addr; in io_add_buffers()
4317 int i, bid = pbuf->bid; in io_add_buffers()
4319 for (i = 0; i < pbuf->nbufs; i++) { in io_add_buffers()
4324 buf->addr = addr; in io_add_buffers()
4325 buf->len = min_t(__u32, pbuf->len, MAX_RW_COUNT); in io_add_buffers()
4326 buf->bid = bid; in io_add_buffers()
4327 addr += pbuf->len; in io_add_buffers()
4330 INIT_LIST_HEAD(&buf->list); in io_add_buffers()
4333 list_add_tail(&buf->list, &(*head)->list); in io_add_buffers()
4338 return i ? i : -ENOMEM; in io_add_buffers()
4343 struct io_provide_buf *p = &req->pbuf; in io_provide_buffers()
4344 struct io_ring_ctx *ctx = req->ctx; in io_provide_buffers()
4351 lockdep_assert_held(&ctx->uring_lock); in io_provide_buffers()
4353 list = head = xa_load(&ctx->io_buffers, p->bgid); in io_provide_buffers()
4357 ret = xa_insert(&ctx->io_buffers, p->bgid, head, in io_provide_buffers()
4360 __io_remove_buffers(ctx, head, p->bgid, -1U); in io_provide_buffers()
4374 if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) in io_epoll_ctl_prep()
4375 return -EINVAL; in io_epoll_ctl_prep()
4376 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_epoll_ctl_prep()
4377 return -EINVAL; in io_epoll_ctl_prep()
4379 req->epoll.epfd = READ_ONCE(sqe->fd); in io_epoll_ctl_prep()
4380 req->epoll.op = READ_ONCE(sqe->len); in io_epoll_ctl_prep()
4381 req->epoll.fd = READ_ONCE(sqe->off); in io_epoll_ctl_prep()
4383 if (ep_op_has_event(req->epoll.op)) { in io_epoll_ctl_prep()
4386 ev = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_epoll_ctl_prep()
4387 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev))) in io_epoll_ctl_prep()
4388 return -EFAULT; in io_epoll_ctl_prep()
4393 return -EOPNOTSUPP; in io_epoll_ctl_prep()
4400 struct io_epoll *ie = &req->epoll; in io_epoll_ctl()
4404 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock); in io_epoll_ctl()
4405 if (force_nonblock && ret == -EAGAIN) in io_epoll_ctl()
4406 return -EAGAIN; in io_epoll_ctl()
4413 return -EOPNOTSUPP; in io_epoll_ctl()
4420 if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->splice_fd_in) in io_madvise_prep()
4421 return -EINVAL; in io_madvise_prep()
4422 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_madvise_prep()
4423 return -EINVAL; in io_madvise_prep()
4425 req->madvise.addr = READ_ONCE(sqe->addr); in io_madvise_prep()
4426 req->madvise.len = READ_ONCE(sqe->len); in io_madvise_prep()
4427 req->madvise.advice = READ_ONCE(sqe->fadvise_advice); in io_madvise_prep()
4430 return -EOPNOTSUPP; in io_madvise_prep()
4437 struct io_madvise *ma = &req->madvise; in io_madvise()
4441 return -EAGAIN; in io_madvise()
4443 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice); in io_madvise()
4449 return -EOPNOTSUPP; in io_madvise()
4455 if (sqe->ioprio || sqe->buf_index || sqe->addr || sqe->splice_fd_in) in io_fadvise_prep()
4456 return -EINVAL; in io_fadvise_prep()
4457 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_fadvise_prep()
4458 return -EINVAL; in io_fadvise_prep()
4460 req->fadvise.offset = READ_ONCE(sqe->off); in io_fadvise_prep()
4461 req->fadvise.len = READ_ONCE(sqe->len); in io_fadvise_prep()
4462 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice); in io_fadvise_prep()
4468 struct io_fadvise *fa = &req->fadvise; in io_fadvise()
4472 switch (fa->advice) { in io_fadvise()
4478 return -EAGAIN; in io_fadvise()
4482 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice); in io_fadvise()
4491 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_statx_prep()
4492 return -EINVAL; in io_statx_prep()
4493 if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) in io_statx_prep()
4494 return -EINVAL; in io_statx_prep()
4495 if (req->flags & REQ_F_FIXED_FILE) in io_statx_prep()
4496 return -EBADF; in io_statx_prep()
4498 req->statx.dfd = READ_ONCE(sqe->fd); in io_statx_prep()
4499 req->statx.mask = READ_ONCE(sqe->len); in io_statx_prep()
4500 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_statx_prep()
4501 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2)); in io_statx_prep()
4502 req->statx.flags = READ_ONCE(sqe->statx_flags); in io_statx_prep()
4509 struct io_statx *ctx = &req->statx; in io_statx()
4513 return -EAGAIN; in io_statx()
4515 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask, in io_statx()
4516 ctx->buffer); in io_statx()
4526 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_close_prep()
4527 return -EINVAL; in io_close_prep()
4528 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len || in io_close_prep()
4529 sqe->rw_flags || sqe->buf_index) in io_close_prep()
4530 return -EINVAL; in io_close_prep()
4531 if (req->flags & REQ_F_FIXED_FILE) in io_close_prep()
4532 return -EBADF; in io_close_prep()
4534 req->close.fd = READ_ONCE(sqe->fd); in io_close_prep()
4535 req->close.file_slot = READ_ONCE(sqe->file_index); in io_close_prep()
4536 if (req->close.file_slot && req->close.fd) in io_close_prep()
4537 return -EINVAL; in io_close_prep()
4544 struct files_struct *files = current->files; in io_close()
4545 struct io_close *close = &req->close; in io_close()
4548 int ret = -EBADF; in io_close()
4550 if (req->close.file_slot) { in io_close()
4555 spin_lock(&files->file_lock); in io_close()
4557 if (close->fd >= fdt->max_fds) { in io_close()
4558 spin_unlock(&files->file_lock); in io_close()
4561 file = fdt->fd[close->fd]; in io_close()
4562 if (!file || file->f_op == &io_uring_fops) { in io_close()
4563 spin_unlock(&files->file_lock); in io_close()
4569 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { in io_close()
4570 spin_unlock(&files->file_lock); in io_close()
4571 return -EAGAIN; in io_close()
4574 ret = __close_fd_get_file(close->fd, &file); in io_close()
4575 spin_unlock(&files->file_lock); in io_close()
4577 if (ret == -ENOENT) in io_close()
4578 ret = -EBADF; in io_close()
4582 /* No ->flush() or already async, safely close from here */ in io_close()
4583 ret = filp_close(file, current->files); in io_close()
4595 struct io_ring_ctx *ctx = req->ctx; in io_sfr_prep()
4597 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL)) in io_sfr_prep()
4598 return -EINVAL; in io_sfr_prep()
4599 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index || in io_sfr_prep()
4600 sqe->splice_fd_in)) in io_sfr_prep()
4601 return -EINVAL; in io_sfr_prep()
4603 req->sync.off = READ_ONCE(sqe->off); in io_sfr_prep()
4604 req->sync.len = READ_ONCE(sqe->len); in io_sfr_prep()
4605 req->sync.flags = READ_ONCE(sqe->sync_range_flags); in io_sfr_prep()
4615 return -EAGAIN; in io_sync_file_range()
4617 ret = sync_file_range(req->file, req->sync.off, req->sync.len, in io_sync_file_range()
4618 req->sync.flags); in io_sync_file_range()
4629 struct io_async_msghdr *async_msg = req->async_data; in io_setup_async_msg()
4632 return -EAGAIN; in io_setup_async_msg()
4634 kfree(kmsg->free_iov); in io_setup_async_msg()
4635 return -ENOMEM; in io_setup_async_msg()
4637 async_msg = req->async_data; in io_setup_async_msg()
4638 req->flags |= REQ_F_NEED_CLEANUP; in io_setup_async_msg()
4640 if (async_msg->msg.msg_name) in io_setup_async_msg()
4641 async_msg->msg.msg_name = &async_msg->addr; in io_setup_async_msg()
4643 if (!async_msg->free_iov) in io_setup_async_msg()
4644 async_msg->msg.msg_iter.iov = async_msg->fast_iov; in io_setup_async_msg()
4646 return -EAGAIN; in io_setup_async_msg()
4652 iomsg->msg.msg_name = &iomsg->addr; in io_sendmsg_copy_hdr()
4653 iomsg->free_iov = iomsg->fast_iov; in io_sendmsg_copy_hdr()
4654 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg, in io_sendmsg_copy_hdr()
4655 req->sr_msg.msg_flags, &iomsg->free_iov); in io_sendmsg_copy_hdr()
4662 ret = io_sendmsg_copy_hdr(req, req->async_data); in io_sendmsg_prep_async()
4664 req->flags |= REQ_F_NEED_CLEANUP; in io_sendmsg_prep_async()
4670 struct io_sr_msg *sr = &req->sr_msg; in io_sendmsg_prep()
4672 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_sendmsg_prep()
4673 return -EINVAL; in io_sendmsg_prep()
4674 if (unlikely(sqe->addr2 || sqe->file_index)) in io_sendmsg_prep()
4675 return -EINVAL; in io_sendmsg_prep()
4676 if (unlikely(sqe->addr2 || sqe->file_index || sqe->ioprio)) in io_sendmsg_prep()
4677 return -EINVAL; in io_sendmsg_prep()
4679 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_sendmsg_prep()
4680 sr->len = READ_ONCE(sqe->len); in io_sendmsg_prep()
4681 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; in io_sendmsg_prep()
4682 if (sr->msg_flags & MSG_DONTWAIT) in io_sendmsg_prep()
4683 req->flags |= REQ_F_NOWAIT; in io_sendmsg_prep()
4686 if (req->ctx->compat) in io_sendmsg_prep()
4687 sr->msg_flags |= MSG_CMSG_COMPAT; in io_sendmsg_prep()
4700 sock = sock_from_file(req->file, &ret); in io_sendmsg()
4704 kmsg = req->async_data; in io_sendmsg()
4712 flags = req->sr_msg.msg_flags; in io_sendmsg()
4716 min_ret = iov_iter_count(&kmsg->msg.msg_iter); in io_sendmsg()
4718 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags); in io_sendmsg()
4719 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) in io_sendmsg()
4721 if (ret == -ERESTARTSYS) in io_sendmsg()
4722 ret = -EINTR; in io_sendmsg()
4724 /* fast path, check for non-NULL to avoid function call */ in io_sendmsg()
4725 if (kmsg->free_iov) in io_sendmsg()
4726 kfree(kmsg->free_iov); in io_sendmsg()
4727 req->flags &= ~REQ_F_NEED_CLEANUP; in io_sendmsg()
4736 struct io_sr_msg *sr = &req->sr_msg; in io_send()
4744 sock = sock_from_file(req->file, &ret); in io_send()
4748 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter); in io_send()
4757 flags = req->sr_msg.msg_flags; in io_send()
4765 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN) in io_send()
4766 return -EAGAIN; in io_send()
4767 if (ret == -ERESTARTSYS) in io_send()
4768 ret = -EINTR; in io_send()
4779 struct io_sr_msg *sr = &req->sr_msg; in __io_recvmsg_copy_hdr()
4784 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg, in __io_recvmsg_copy_hdr()
4785 &iomsg->uaddr, &uiov, &iov_len); in __io_recvmsg_copy_hdr()
4789 if (req->flags & REQ_F_BUFFER_SELECT) { in __io_recvmsg_copy_hdr()
4791 return -EINVAL; in __io_recvmsg_copy_hdr()
4792 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov))) in __io_recvmsg_copy_hdr()
4793 return -EFAULT; in __io_recvmsg_copy_hdr()
4794 sr->len = iomsg->fast_iov[0].iov_len; in __io_recvmsg_copy_hdr()
4795 iomsg->free_iov = NULL; in __io_recvmsg_copy_hdr()
4797 iomsg->free_iov = iomsg->fast_iov; in __io_recvmsg_copy_hdr()
4799 &iomsg->free_iov, &iomsg->msg.msg_iter, in __io_recvmsg_copy_hdr()
4812 struct io_sr_msg *sr = &req->sr_msg; in __io_compat_recvmsg_copy_hdr()
4818 ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr, in __io_compat_recvmsg_copy_hdr()
4824 if (req->flags & REQ_F_BUFFER_SELECT) { in __io_compat_recvmsg_copy_hdr()
4828 return -EINVAL; in __io_compat_recvmsg_copy_hdr()
4830 return -EFAULT; in __io_compat_recvmsg_copy_hdr()
4831 if (__get_user(clen, &uiov->iov_len)) in __io_compat_recvmsg_copy_hdr()
4832 return -EFAULT; in __io_compat_recvmsg_copy_hdr()
4834 return -EINVAL; in __io_compat_recvmsg_copy_hdr()
4835 sr->len = clen; in __io_compat_recvmsg_copy_hdr()
4836 iomsg->free_iov = NULL; in __io_compat_recvmsg_copy_hdr()
4838 iomsg->free_iov = iomsg->fast_iov; in __io_compat_recvmsg_copy_hdr()
4840 UIO_FASTIOV, &iomsg->free_iov, in __io_compat_recvmsg_copy_hdr()
4841 &iomsg->msg.msg_iter, true); in __io_compat_recvmsg_copy_hdr()
4853 iomsg->msg.msg_name = &iomsg->addr; in io_recvmsg_copy_hdr()
4856 if (req->ctx->compat) in io_recvmsg_copy_hdr()
4866 struct io_sr_msg *sr = &req->sr_msg; in io_recv_buffer_select()
4869 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock); in io_recv_buffer_select()
4873 sr->kbuf = kbuf; in io_recv_buffer_select()
4874 req->flags |= REQ_F_BUFFER_SELECTED; in io_recv_buffer_select()
4880 return io_put_kbuf(req, req->sr_msg.kbuf); in io_put_recv_kbuf()
4887 ret = io_recvmsg_copy_hdr(req, req->async_data); in io_recvmsg_prep_async()
4889 req->flags |= REQ_F_NEED_CLEANUP; in io_recvmsg_prep_async()
4895 struct io_sr_msg *sr = &req->sr_msg; in io_recvmsg_prep()
4897 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_recvmsg_prep()
4898 return -EINVAL; in io_recvmsg_prep()
4899 if (unlikely(sqe->addr2 || sqe->file_index)) in io_recvmsg_prep()
4900 return -EINVAL; in io_recvmsg_prep()
4901 if (unlikely(sqe->addr2 || sqe->file_index || sqe->ioprio)) in io_recvmsg_prep()
4902 return -EINVAL; in io_recvmsg_prep()
4904 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_recvmsg_prep()
4905 sr->len = READ_ONCE(sqe->len); in io_recvmsg_prep()
4906 sr->bgid = READ_ONCE(sqe->buf_group); in io_recvmsg_prep()
4907 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL; in io_recvmsg_prep()
4908 if (sr->msg_flags & MSG_DONTWAIT) in io_recvmsg_prep()
4909 req->flags |= REQ_F_NOWAIT; in io_recvmsg_prep()
4912 if (req->ctx->compat) in io_recvmsg_prep()
4913 sr->msg_flags |= MSG_CMSG_COMPAT; in io_recvmsg_prep()
4928 sock = sock_from_file(req->file, &ret); in io_recvmsg()
4932 kmsg = req->async_data; in io_recvmsg()
4940 if (req->flags & REQ_F_BUFFER_SELECT) { in io_recvmsg()
4944 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr); in io_recvmsg()
4945 kmsg->fast_iov[0].iov_len = req->sr_msg.len; in io_recvmsg()
4946 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, in io_recvmsg()
4947 1, req->sr_msg.len); in io_recvmsg()
4950 flags = req->sr_msg.msg_flags; in io_recvmsg()
4954 min_ret = iov_iter_count(&kmsg->msg.msg_iter); in io_recvmsg()
4956 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg, in io_recvmsg()
4957 kmsg->uaddr, flags); in io_recvmsg()
4958 if (force_nonblock && ret == -EAGAIN) in io_recvmsg()
4960 if (ret == -ERESTARTSYS) in io_recvmsg()
4961 ret = -EINTR; in io_recvmsg()
4963 if (req->flags & REQ_F_BUFFER_SELECTED) in io_recvmsg()
4965 /* fast path, check for non-NULL to avoid function call */ in io_recvmsg()
4966 if (kmsg->free_iov) in io_recvmsg()
4967 kfree(kmsg->free_iov); in io_recvmsg()
4968 req->flags &= ~REQ_F_NEED_CLEANUP; in io_recvmsg()
4969 if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC)))) in io_recvmsg()
4978 struct io_sr_msg *sr = &req->sr_msg; in io_recv()
4980 void __user *buf = sr->buf; in io_recv()
4988 sock = sock_from_file(req->file, &ret); in io_recv()
4992 if (req->flags & REQ_F_BUFFER_SELECT) { in io_recv()
4996 buf = u64_to_user_ptr(kbuf->addr); in io_recv()
4999 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter); in io_recv()
5010 flags = req->sr_msg.msg_flags; in io_recv()
5017 if (force_nonblock && ret == -EAGAIN) in io_recv()
5018 return -EAGAIN; in io_recv()
5019 if (ret == -ERESTARTSYS) in io_recv()
5020 ret = -EINTR; in io_recv()
5022 if (req->flags & REQ_F_BUFFER_SELECTED) in io_recv()
5032 struct io_accept *accept = &req->accept; in io_accept_prep()
5034 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_accept_prep()
5035 return -EINVAL; in io_accept_prep()
5036 if (sqe->ioprio || sqe->len || sqe->buf_index) in io_accept_prep()
5037 return -EINVAL; in io_accept_prep()
5039 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_accept_prep()
5040 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2)); in io_accept_prep()
5041 accept->flags = READ_ONCE(sqe->accept_flags); in io_accept_prep()
5042 accept->nofile = rlimit(RLIMIT_NOFILE); in io_accept_prep()
5044 accept->file_slot = READ_ONCE(sqe->file_index); in io_accept_prep()
5045 if (accept->file_slot && (accept->flags & SOCK_CLOEXEC)) in io_accept_prep()
5046 return -EINVAL; in io_accept_prep()
5047 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) in io_accept_prep()
5048 return -EINVAL; in io_accept_prep()
5049 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK)) in io_accept_prep()
5050 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK; in io_accept_prep()
5056 struct io_accept *accept = &req->accept; in io_accept()
5059 bool fixed = !!accept->file_slot; in io_accept()
5063 if (req->file->f_flags & O_NONBLOCK) in io_accept()
5064 req->flags |= REQ_F_NOWAIT; in io_accept()
5067 fd = __get_unused_fd_flags(accept->flags, accept->nofile); in io_accept()
5071 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len, in io_accept()
5072 accept->flags); in io_accept()
5078 if (ret == -EAGAIN && force_nonblock) in io_accept()
5079 return -EAGAIN; in io_accept()
5080 if (ret == -ERESTARTSYS) in io_accept()
5081 ret = -EINTR; in io_accept()
5088 accept->file_slot - 1); in io_accept()
5096 struct io_async_connect *io = req->async_data; in io_connect_prep_async()
5097 struct io_connect *conn = &req->connect; in io_connect_prep_async()
5099 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address); in io_connect_prep_async()
5104 struct io_connect *conn = &req->connect; in io_connect_prep()
5106 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_connect_prep()
5107 return -EINVAL; in io_connect_prep()
5108 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags || in io_connect_prep()
5109 sqe->splice_fd_in) in io_connect_prep()
5110 return -EINVAL; in io_connect_prep()
5112 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr)); in io_connect_prep()
5113 conn->addr_len = READ_ONCE(sqe->addr2); in io_connect_prep()
5124 if (req->async_data) { in io_connect()
5125 io = req->async_data; in io_connect()
5127 ret = move_addr_to_kernel(req->connect.addr, in io_connect()
5128 req->connect.addr_len, in io_connect()
5137 ret = __sys_connect_file(req->file, &io->address, in io_connect()
5138 req->connect.addr_len, file_flags); in io_connect()
5139 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) { in io_connect()
5140 if (req->async_data) in io_connect()
5141 return -EAGAIN; in io_connect()
5143 ret = -ENOMEM; in io_connect()
5146 memcpy(req->async_data, &__io, sizeof(__io)); in io_connect()
5147 return -EAGAIN; in io_connect()
5149 if (ret == -ERESTARTSYS) in io_connect()
5150 ret = -EINTR; in io_connect()
5161 return -EOPNOTSUPP; \
5168 return -EOPNOTSUPP; \
5175 return -EOPNOTSUPP; \
5198 * We usually have 1-2 refs taken, 128 is more than enough and we want to
5209 * grabbing the ownership. Instead of incrementing set a retry flag in io_poll_get_ownership_slowpath()
5212 v = atomic_fetch_or(IO_POLL_RETRY_FLAG, &req->poll_refs); in io_poll_get_ownership_slowpath()
5215 return !(atomic_fetch_inc(&req->poll_refs) & IO_POLL_REF_MASK); in io_poll_get_ownership_slowpath()
5219 * If refs part of ->poll_refs (see IO_POLL_REF_MASK) is 0, it's free. We can
5222 * arming poll and wakeups.
5226 if (unlikely(atomic_read(&req->poll_refs) >= IO_POLL_REF_BIAS)) in io_poll_get_ownership()
5228 return !(atomic_fetch_inc(&req->poll_refs) & IO_POLL_REF_MASK); in io_poll_get_ownership()
5233 atomic_or(IO_POLL_CANCEL_FLAG, &req->poll_refs); in io_poll_mark_cancelled()
5238 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */ in io_poll_get_double()
5239 if (req->opcode == IORING_OP_POLL_ADD) in io_poll_get_double()
5240 return req->async_data; in io_poll_get_double()
5241 return req->apoll->double_poll; in io_poll_get_double()
5246 if (req->opcode == IORING_OP_POLL_ADD) in io_poll_get_single()
5247 return &req->poll; in io_poll_get_single()
5248 return &req->apoll->poll; in io_poll_get_single()
5253 struct io_ring_ctx *ctx = req->ctx; in io_poll_req_insert()
5256 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)]; in io_poll_req_insert()
5257 hlist_add_head(&req->hash_node, list); in io_poll_req_insert()
5260 static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events, in io_init_poll_iocb() argument
5263 poll->head = NULL; in io_init_poll_iocb()
5266 poll->events = events | IO_POLL_UNMASK; in io_init_poll_iocb()
5267 INIT_LIST_HEAD(&poll->wait.entry); in io_init_poll_iocb()
5268 init_waitqueue_func_entry(&poll->wait, wake_func); in io_init_poll_iocb()
5271 static inline void io_poll_remove_entry(struct io_poll_iocb *poll) in io_poll_remove_entry() argument
5273 struct wait_queue_head *head = smp_load_acquire(&poll->head); in io_poll_remove_entry()
5276 spin_lock_irq(&head->lock); in io_poll_remove_entry()
5277 list_del_init(&poll->wait.entry); in io_poll_remove_entry()
5278 poll->head = NULL; in io_poll_remove_entry()
5279 spin_unlock_irq(&head->lock); in io_poll_remove_entry()
5285 struct io_poll_iocb *poll = io_poll_get_single(req); in io_poll_remove_entries() local
5294 * all users of wake_up_pollfree() will RCU-delay the actual free. If in io_poll_remove_entries()
5296 * non-NULL, we can then lock it without the memory being freed out from in io_poll_remove_entries()
5304 io_poll_remove_entry(poll); in io_poll_remove_entries()
5311 * All poll tw should go through this. Checks for poll events, manages
5316 * the request, then the mask is stored in req->result.
5320 struct io_ring_ctx *ctx = req->ctx; in io_poll_check_events()
5321 struct io_poll_iocb *poll = io_poll_get_single(req); in io_poll_check_events() local
5324 /* req->task == current here, checking PF_EXITING is safe */ in io_poll_check_events()
5325 if (unlikely(req->task->flags & PF_EXITING)) in io_poll_check_events()
5329 v = atomic_read(&req->poll_refs); in io_poll_check_events()
5335 return -ECANCELED; in io_poll_check_events()
5342 req->result = 0; in io_poll_check_events()
5344 req->result = 0; in io_poll_check_events()
5350 atomic_andnot(IO_POLL_RETRY_FLAG, &req->poll_refs); in io_poll_check_events()
5354 if (!req->result) { in io_poll_check_events()
5355 struct poll_table_struct pt = { ._key = poll->events }; in io_poll_check_events()
5357 req->result = vfs_poll(req->file, &pt) & poll->events; in io_poll_check_events()
5361 if (req->result && !(poll->events & EPOLLONESHOT)) { in io_poll_check_events()
5362 __poll_t mask = mangle_poll(req->result & poll->events); in io_poll_check_events()
5365 spin_lock(&ctx->completion_lock); in io_poll_check_events()
5366 filled = io_fill_cqe_aux(ctx, req->user_data, mask, in io_poll_check_events()
5369 spin_unlock(&ctx->completion_lock); in io_poll_check_events()
5371 return -ECANCELED; in io_poll_check_events()
5373 } else if (req->result) { in io_poll_check_events()
5378 req->result = 0; in io_poll_check_events()
5381 * Release all references, retry if someone tried to restart in io_poll_check_events()
5384 } while (atomic_sub_return(v & IO_POLL_REF_MASK, &req->poll_refs) & in io_poll_check_events()
5392 struct io_ring_ctx *ctx = req->ctx; in io_poll_task_func()
5400 req->result = mangle_poll(req->result & req->poll.events); in io_poll_task_func()
5402 req->result = ret; in io_poll_task_func()
5407 spin_lock(&ctx->completion_lock); in io_poll_task_func()
5408 hash_del(&req->hash_node); in io_poll_task_func()
5409 spin_unlock(&ctx->completion_lock); in io_poll_task_func()
5410 io_req_complete_post(req, req->result, 0); in io_poll_task_func()
5415 struct io_ring_ctx *ctx = req->ctx; in io_apoll_task_func()
5423 spin_lock(&ctx->completion_lock); in io_apoll_task_func()
5424 hash_del(&req->hash_node); in io_apoll_task_func()
5425 spin_unlock(&ctx->completion_lock); in io_apoll_task_func()
5435 req->result = mask; in __io_poll_execute()
5436 if (req->opcode == IORING_OP_POLL_ADD) in __io_poll_execute()
5437 req->io_task_work.func = io_poll_task_func; in __io_poll_execute()
5439 req->io_task_work.func = io_apoll_task_func; in __io_poll_execute()
5441 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask); in __io_poll_execute()
5461 struct io_kiocb *req = wait->private; in io_poll_wake()
5462 struct io_poll_iocb *poll = container_of(wait, struct io_poll_iocb, in io_poll_wake() local
5478 list_del_init(&poll->wait.entry); in io_poll_wake()
5482 * as req->head is NULL'ed out, the request can be in io_poll_wake()
5486 smp_store_release(&poll->head, NULL); in io_poll_wake()
5491 if (mask && !(mask & poll->events)) in io_poll_wake()
5496 * If we trigger a multishot poll off our own wakeup path, in io_poll_wake()
5501 poll->events |= EPOLLONESHOT; in io_poll_wake()
5508 static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt, in __io_queue_proc() argument
5512 struct io_kiocb *req = pt->req; in __io_queue_proc()
5515 * The file being polled uses multiple waitqueues for poll handling in __io_queue_proc()
5519 if (unlikely(pt->nr_entries)) { in __io_queue_proc()
5520 struct io_poll_iocb *first = poll; in __io_queue_proc()
5523 if (first->head == head) in __io_queue_proc()
5527 if ((*poll_ptr)->head == head) in __io_queue_proc()
5529 pt->error = -EINVAL; in __io_queue_proc()
5533 poll = kmalloc(sizeof(*poll), GFP_ATOMIC); in __io_queue_proc()
5534 if (!poll) { in __io_queue_proc()
5535 pt->error = -ENOMEM; in __io_queue_proc()
5538 io_init_poll_iocb(poll, first->events, first->wait.func); in __io_queue_proc()
5539 *poll_ptr = poll; in __io_queue_proc()
5542 pt->nr_entries++; in __io_queue_proc()
5543 poll->head = head; in __io_queue_proc()
5544 poll->wait.private = req; in __io_queue_proc()
5546 if (poll->events & EPOLLEXCLUSIVE) in __io_queue_proc()
5547 add_wait_queue_exclusive(head, &poll->wait); in __io_queue_proc()
5549 add_wait_queue(head, &poll->wait); in __io_queue_proc()
5557 __io_queue_proc(&pt->req->poll, pt, head, in io_poll_queue_proc()
5558 (struct io_poll_iocb **) &pt->req->async_data); in io_poll_queue_proc()
5562 struct io_poll_iocb *poll, in __io_arm_poll_handler() argument
5565 struct io_ring_ctx *ctx = req->ctx; in __io_arm_poll_handler()
5567 INIT_HLIST_NODE(&req->hash_node); in __io_arm_poll_handler()
5568 io_init_poll_iocb(poll, mask, io_poll_wake); in __io_arm_poll_handler()
5569 poll->file = req->file; in __io_arm_poll_handler()
5570 poll->wait.private = req; in __io_arm_poll_handler()
5572 ipt->pt._key = mask; in __io_arm_poll_handler()
5573 ipt->req = req; in __io_arm_poll_handler()
5574 ipt->error = 0; in __io_arm_poll_handler()
5575 ipt->nr_entries = 0; in __io_arm_poll_handler()
5579 * with poll arming. see io_poll_get_ownership(). in __io_arm_poll_handler()
5581 atomic_set(&req->poll_refs, 1); in __io_arm_poll_handler()
5582 mask = vfs_poll(req->file, &ipt->pt) & poll->events; in __io_arm_poll_handler()
5584 if (mask && (poll->events & EPOLLONESHOT)) { in __io_arm_poll_handler()
5589 if (!mask && unlikely(ipt->error || !ipt->nr_entries)) { in __io_arm_poll_handler()
5591 if (!ipt->error) in __io_arm_poll_handler()
5592 ipt->error = -EINVAL; in __io_arm_poll_handler()
5596 spin_lock(&ctx->completion_lock); in __io_arm_poll_handler()
5598 spin_unlock(&ctx->completion_lock); in __io_arm_poll_handler()
5602 if (unlikely(ipt->error || !ipt->nr_entries)) { in __io_arm_poll_handler()
5603 poll->events |= EPOLLONESHOT; in __io_arm_poll_handler()
5604 ipt->error = 0; in __io_arm_poll_handler()
5612 * poll was waken up, queue up a tw, it'll deal with it. in __io_arm_poll_handler()
5614 if (atomic_cmpxchg(&req->poll_refs, 1, 0) != 1) in __io_arm_poll_handler()
5623 struct async_poll *apoll = pt->req->apoll; in io_async_queue_proc()
5625 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll); in io_async_queue_proc()
5636 const struct io_op_def *def = &io_op_defs[req->opcode]; in io_arm_poll_handler()
5637 struct io_ring_ctx *ctx = req->ctx; in io_arm_poll_handler()
5643 if (!req->file || !file_can_poll(req->file)) in io_arm_poll_handler()
5645 if (req->flags & REQ_F_POLLED) in io_arm_poll_handler()
5647 if (!def->pollin && !def->pollout) in io_arm_poll_handler()
5650 if (def->pollin) { in io_arm_poll_handler()
5654 if ((req->opcode == IORING_OP_RECVMSG) && in io_arm_poll_handler()
5655 (req->sr_msg.msg_flags & MSG_ERRQUEUE)) in io_arm_poll_handler()
5664 apoll->double_poll = NULL; in io_arm_poll_handler()
5665 req->apoll = apoll; in io_arm_poll_handler()
5666 req->flags |= REQ_F_POLLED; in io_arm_poll_handler()
5669 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask); in io_arm_poll_handler()
5673 trace_io_uring_poll_arm(ctx, req, req->opcode, req->user_data, in io_arm_poll_handler()
5674 mask, apoll->poll.events); in io_arm_poll_handler()
5679 * Returns true if we found and killed one or more poll requests
5689 spin_lock(&ctx->completion_lock); in io_poll_remove_all()
5690 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { in io_poll_remove_all()
5693 list = &ctx->cancel_hash[i]; in io_poll_remove_all()
5696 hlist_del_init(&req->hash_node); in io_poll_remove_all()
5702 spin_unlock(&ctx->completion_lock); in io_poll_remove_all()
5708 __must_hold(&ctx->completion_lock) in io_poll_find()
5713 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)]; in io_poll_find()
5715 if (sqe_addr != req->user_data) in io_poll_find()
5717 if (poll_only && req->opcode != IORING_OP_POLL_ADD) in io_poll_find()
5725 __must_hold(&ctx->completion_lock) in io_poll_disarm()
5730 hash_del(&req->hash_node); in io_poll_disarm()
5736 __must_hold(&ctx->completion_lock) in io_poll_cancel()
5741 return -ENOENT; in io_poll_cancel()
5751 events = READ_ONCE(sqe->poll32_events); in io_poll_parse_events()
5763 struct io_poll_update *upd = &req->poll_update; in io_poll_update_prep()
5766 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_poll_update_prep()
5767 return -EINVAL; in io_poll_update_prep()
5768 if (sqe->ioprio || sqe->buf_index || sqe->splice_fd_in) in io_poll_update_prep()
5769 return -EINVAL; in io_poll_update_prep()
5770 flags = READ_ONCE(sqe->len); in io_poll_update_prep()
5773 return -EINVAL; in io_poll_update_prep()
5776 return -EINVAL; in io_poll_update_prep()
5778 upd->old_user_data = READ_ONCE(sqe->addr); in io_poll_update_prep()
5779 upd->update_events = flags & IORING_POLL_UPDATE_EVENTS; in io_poll_update_prep()
5780 upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA; in io_poll_update_prep()
5782 upd->new_user_data = READ_ONCE(sqe->off); in io_poll_update_prep()
5783 if (!upd->update_user_data && upd->new_user_data) in io_poll_update_prep()
5784 return -EINVAL; in io_poll_update_prep()
5785 if (upd->update_events) in io_poll_update_prep()
5786 upd->events = io_poll_parse_events(sqe, flags); in io_poll_update_prep()
5787 else if (sqe->poll32_events) in io_poll_update_prep()
5788 return -EINVAL; in io_poll_update_prep()
5795 struct io_poll_iocb *poll = &req->poll; in io_poll_add_prep() local
5798 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_poll_add_prep()
5799 return -EINVAL; in io_poll_add_prep()
5800 if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr) in io_poll_add_prep()
5801 return -EINVAL; in io_poll_add_prep()
5802 flags = READ_ONCE(sqe->len); in io_poll_add_prep()
5804 return -EINVAL; in io_poll_add_prep()
5807 poll->events = io_poll_parse_events(sqe, flags); in io_poll_add_prep()
5813 struct io_poll_iocb *poll = &req->poll; in io_poll_add() local
5819 ret = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events); in io_poll_add()
5830 struct io_ring_ctx *ctx = req->ctx; in io_poll_update()
5834 spin_lock(&ctx->completion_lock); in io_poll_update()
5835 preq = io_poll_find(ctx, req->poll_update.old_user_data, true); in io_poll_update()
5837 spin_unlock(&ctx->completion_lock); in io_poll_update()
5838 ret = preq ? -EALREADY : -ENOENT; in io_poll_update()
5841 spin_unlock(&ctx->completion_lock); in io_poll_update()
5843 if (req->poll_update.update_events || req->poll_update.update_user_data) { in io_poll_update()
5845 if (req->poll_update.update_events) { in io_poll_update()
5846 preq->poll.events &= ~0xffff; in io_poll_update()
5847 preq->poll.events |= req->poll_update.events & 0xffff; in io_poll_update()
5848 preq->poll.events |= IO_POLL_UNMASK; in io_poll_update()
5850 if (req->poll_update.update_user_data) in io_poll_update()
5851 preq->user_data = req->poll_update.new_user_data; in io_poll_update()
5854 /* successfully updated, don't complete poll request */ in io_poll_update()
5859 io_req_complete(preq, -ECANCELED); in io_poll_update()
5871 io_req_complete_post(req, -ETIME, 0); in io_req_task_timeout()
5878 struct io_kiocb *req = data->req; in io_timeout_fn()
5879 struct io_ring_ctx *ctx = req->ctx; in io_timeout_fn()
5882 spin_lock_irqsave(&ctx->timeout_lock, flags); in io_timeout_fn()
5883 list_del_init(&req->timeout.list); in io_timeout_fn()
5884 atomic_set(&req->ctx->cq_timeouts, in io_timeout_fn()
5885 atomic_read(&req->ctx->cq_timeouts) + 1); in io_timeout_fn()
5886 spin_unlock_irqrestore(&ctx->timeout_lock, flags); in io_timeout_fn()
5888 req->io_task_work.func = io_req_task_timeout; in io_timeout_fn()
5895 __must_hold(&ctx->timeout_lock) in io_timeout_extract()
5901 list_for_each_entry(req, &ctx->timeout_list, timeout.list) { in io_timeout_extract()
5902 found = user_data == req->user_data; in io_timeout_extract()
5907 return ERR_PTR(-ENOENT); in io_timeout_extract()
5909 io = req->async_data; in io_timeout_extract()
5910 if (hrtimer_try_to_cancel(&io->timer) == -1) in io_timeout_extract()
5911 return ERR_PTR(-EALREADY); in io_timeout_extract()
5912 list_del_init(&req->timeout.list); in io_timeout_extract()
5917 __must_hold(&ctx->completion_lock) in io_timeout_cancel()
5918 __must_hold(&ctx->timeout_lock) in io_timeout_cancel()
5926 io_fill_cqe_req(req, -ECANCELED, 0); in io_timeout_cancel()
5933 switch (data->flags & IORING_TIMEOUT_CLOCK_MASK) { in io_timeout_get_clock()
5949 __must_hold(&ctx->timeout_lock) in io_linked_timeout_update()
5955 list_for_each_entry(req, &ctx->ltimeout_list, timeout.list) { in io_linked_timeout_update()
5956 found = user_data == req->user_data; in io_linked_timeout_update()
5961 return -ENOENT; in io_linked_timeout_update()
5963 io = req->async_data; in io_linked_timeout_update()
5964 if (hrtimer_try_to_cancel(&io->timer) == -1) in io_linked_timeout_update()
5965 return -EALREADY; in io_linked_timeout_update()
5966 hrtimer_init(&io->timer, io_timeout_get_clock(io), mode); in io_linked_timeout_update()
5967 io->timer.function = io_link_timeout_fn; in io_linked_timeout_update()
5968 hrtimer_start(&io->timer, timespec64_to_ktime(*ts), mode); in io_linked_timeout_update()
5974 __must_hold(&ctx->timeout_lock) in io_timeout_update()
5982 req->timeout.off = 0; /* noseq */ in io_timeout_update()
5983 data = req->async_data; in io_timeout_update()
5984 list_add_tail(&req->timeout.list, &ctx->timeout_list); in io_timeout_update()
5985 hrtimer_init(&data->timer, io_timeout_get_clock(data), mode); in io_timeout_update()
5986 data->timer.function = io_timeout_fn; in io_timeout_update()
5987 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode); in io_timeout_update()
5994 struct io_timeout_rem *tr = &req->timeout_rem; in io_timeout_remove_prep()
5996 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_timeout_remove_prep()
5997 return -EINVAL; in io_timeout_remove_prep()
5998 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) in io_timeout_remove_prep()
5999 return -EINVAL; in io_timeout_remove_prep()
6000 if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->splice_fd_in) in io_timeout_remove_prep()
6001 return -EINVAL; in io_timeout_remove_prep()
6003 tr->ltimeout = false; in io_timeout_remove_prep()
6004 tr->addr = READ_ONCE(sqe->addr); in io_timeout_remove_prep()
6005 tr->flags = READ_ONCE(sqe->timeout_flags); in io_timeout_remove_prep()
6006 if (tr->flags & IORING_TIMEOUT_UPDATE_MASK) { in io_timeout_remove_prep()
6007 if (hweight32(tr->flags & IORING_TIMEOUT_CLOCK_MASK) > 1) in io_timeout_remove_prep()
6008 return -EINVAL; in io_timeout_remove_prep()
6009 if (tr->flags & IORING_LINK_TIMEOUT_UPDATE) in io_timeout_remove_prep()
6010 tr->ltimeout = true; in io_timeout_remove_prep()
6011 if (tr->flags & ~(IORING_TIMEOUT_UPDATE_MASK|IORING_TIMEOUT_ABS)) in io_timeout_remove_prep()
6012 return -EINVAL; in io_timeout_remove_prep()
6013 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2))) in io_timeout_remove_prep()
6014 return -EFAULT; in io_timeout_remove_prep()
6015 } else if (tr->flags) { in io_timeout_remove_prep()
6017 return -EINVAL; in io_timeout_remove_prep()
6034 struct io_timeout_rem *tr = &req->timeout_rem; in io_timeout_remove()
6035 struct io_ring_ctx *ctx = req->ctx; in io_timeout_remove()
6038 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE)) { in io_timeout_remove()
6039 spin_lock(&ctx->completion_lock); in io_timeout_remove()
6040 spin_lock_irq(&ctx->timeout_lock); in io_timeout_remove()
6041 ret = io_timeout_cancel(ctx, tr->addr); in io_timeout_remove()
6042 spin_unlock_irq(&ctx->timeout_lock); in io_timeout_remove()
6043 spin_unlock(&ctx->completion_lock); in io_timeout_remove()
6045 enum hrtimer_mode mode = io_translate_timeout_mode(tr->flags); in io_timeout_remove()
6047 spin_lock_irq(&ctx->timeout_lock); in io_timeout_remove()
6048 if (tr->ltimeout) in io_timeout_remove()
6049 ret = io_linked_timeout_update(ctx, tr->addr, &tr->ts, mode); in io_timeout_remove()
6051 ret = io_timeout_update(ctx, tr->addr, &tr->ts, mode); in io_timeout_remove()
6052 spin_unlock_irq(&ctx->timeout_lock); in io_timeout_remove()
6066 u32 off = READ_ONCE(sqe->off); in io_timeout_prep()
6068 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_timeout_prep()
6069 return -EINVAL; in io_timeout_prep()
6070 if (sqe->ioprio || sqe->buf_index || sqe->len != 1 || in io_timeout_prep()
6071 sqe->splice_fd_in) in io_timeout_prep()
6072 return -EINVAL; in io_timeout_prep()
6074 return -EINVAL; in io_timeout_prep()
6075 flags = READ_ONCE(sqe->timeout_flags); in io_timeout_prep()
6077 return -EINVAL; in io_timeout_prep()
6080 return -EINVAL; in io_timeout_prep()
6082 INIT_LIST_HEAD(&req->timeout.list); in io_timeout_prep()
6083 req->timeout.off = off; in io_timeout_prep()
6084 if (unlikely(off && !req->ctx->off_timeout_used)) in io_timeout_prep()
6085 req->ctx->off_timeout_used = true; in io_timeout_prep()
6087 if (!req->async_data && io_alloc_async_data(req)) in io_timeout_prep()
6088 return -ENOMEM; in io_timeout_prep()
6090 data = req->async_data; in io_timeout_prep()
6091 data->req = req; in io_timeout_prep()
6092 data->flags = flags; in io_timeout_prep()
6094 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr))) in io_timeout_prep()
6095 return -EFAULT; in io_timeout_prep()
6097 INIT_LIST_HEAD(&req->timeout.list); in io_timeout_prep()
6098 data->mode = io_translate_timeout_mode(flags); in io_timeout_prep()
6099 hrtimer_init(&data->timer, io_timeout_get_clock(data), data->mode); in io_timeout_prep()
6102 struct io_submit_link *link = &req->ctx->submit_state.link; in io_timeout_prep()
6104 if (!link->head) in io_timeout_prep()
6105 return -EINVAL; in io_timeout_prep()
6106 if (link->last->opcode == IORING_OP_LINK_TIMEOUT) in io_timeout_prep()
6107 return -EINVAL; in io_timeout_prep()
6108 req->timeout.head = link->last; in io_timeout_prep()
6109 link->last->flags |= REQ_F_ARM_LTIMEOUT; in io_timeout_prep()
6116 struct io_ring_ctx *ctx = req->ctx; in io_timeout()
6117 struct io_timeout_data *data = req->async_data; in io_timeout()
6119 u32 tail, off = req->timeout.off; in io_timeout()
6121 spin_lock_irq(&ctx->timeout_lock); in io_timeout()
6124 * sqe->off holds how many events that need to occur for this in io_timeout()
6129 entry = ctx->timeout_list.prev; in io_timeout()
6133 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts); in io_timeout()
6134 req->timeout.target_seq = tail + off; in io_timeout()
6137 * This is safe because ->completion_lock is held, and submissions in io_timeout()
6138 * and completions are never mixed in the same ->completion_lock section. in io_timeout()
6140 ctx->cq_last_tm_flush = tail; in io_timeout()
6146 list_for_each_prev(entry, &ctx->timeout_list) { in io_timeout()
6153 if (off >= nxt->timeout.target_seq - tail) in io_timeout()
6157 list_add(&req->timeout.list, entry); in io_timeout()
6158 data->timer.function = io_timeout_fn; in io_timeout()
6159 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode); in io_timeout()
6160 spin_unlock_irq(&ctx->timeout_lock); in io_timeout()
6174 return req->ctx == cd->ctx && req->user_data == cd->user_data; in io_cancel_cb()
6184 if (!tctx || !tctx->io_wq) in io_async_cancel_one()
6185 return -ENOENT; in io_async_cancel_one()
6187 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false); in io_async_cancel_one()
6193 ret = -EALREADY; in io_async_cancel_one()
6196 ret = -ENOENT; in io_async_cancel_one()
6205 struct io_ring_ctx *ctx = req->ctx; in io_try_cancel_userdata()
6208 WARN_ON_ONCE(!io_wq_current_is_worker() && req->task != current); in io_try_cancel_userdata()
6210 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx); in io_try_cancel_userdata()
6211 if (ret != -ENOENT) in io_try_cancel_userdata()
6214 spin_lock(&ctx->completion_lock); in io_try_cancel_userdata()
6215 spin_lock_irq(&ctx->timeout_lock); in io_try_cancel_userdata()
6217 spin_unlock_irq(&ctx->timeout_lock); in io_try_cancel_userdata()
6218 if (ret != -ENOENT) in io_try_cancel_userdata()
6222 spin_unlock(&ctx->completion_lock); in io_try_cancel_userdata()
6229 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_async_cancel_prep()
6230 return -EINVAL; in io_async_cancel_prep()
6231 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) in io_async_cancel_prep()
6232 return -EINVAL; in io_async_cancel_prep()
6233 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags || in io_async_cancel_prep()
6234 sqe->splice_fd_in) in io_async_cancel_prep()
6235 return -EINVAL; in io_async_cancel_prep()
6237 req->cancel.addr = READ_ONCE(sqe->addr); in io_async_cancel_prep()
6243 struct io_ring_ctx *ctx = req->ctx; in io_async_cancel()
6244 u64 sqe_addr = req->cancel.addr; in io_async_cancel()
6249 if (ret != -ENOENT) in io_async_cancel()
6252 /* slow path, try all io-wq's */ in io_async_cancel()
6254 ret = -ENOENT; in io_async_cancel()
6255 list_for_each_entry(node, &ctx->tctx_list, ctx_node) { in io_async_cancel()
6256 struct io_uring_task *tctx = node->task->io_uring; in io_async_cancel()
6258 ret = io_async_cancel_one(tctx, req->cancel.addr, ctx); in io_async_cancel()
6259 if (ret != -ENOENT) in io_async_cancel()
6273 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT))) in io_rsrc_update_prep()
6274 return -EINVAL; in io_rsrc_update_prep()
6275 if (sqe->ioprio || sqe->rw_flags || sqe->splice_fd_in) in io_rsrc_update_prep()
6276 return -EINVAL; in io_rsrc_update_prep()
6278 req->rsrc_update.offset = READ_ONCE(sqe->off); in io_rsrc_update_prep()
6279 req->rsrc_update.nr_args = READ_ONCE(sqe->len); in io_rsrc_update_prep()
6280 if (!req->rsrc_update.nr_args) in io_rsrc_update_prep()
6281 return -EINVAL; in io_rsrc_update_prep()
6282 req->rsrc_update.arg = READ_ONCE(sqe->addr); in io_rsrc_update_prep()
6288 struct io_ring_ctx *ctx = req->ctx; in io_files_update()
6292 up.offset = req->rsrc_update.offset; in io_files_update()
6293 up.data = req->rsrc_update.arg; in io_files_update()
6301 &up, req->rsrc_update.nr_args); in io_files_update()
6312 switch (req->opcode) { in io_req_prep()
6384 req->opcode); in io_req_prep()
6385 return -EINVAL; in io_req_prep()
6390 if (!io_op_defs[req->opcode].needs_async_setup) in io_req_prep_async()
6392 if (WARN_ON_ONCE(req->async_data)) in io_req_prep_async()
6393 return -EFAULT; in io_req_prep_async()
6395 return -EAGAIN; in io_req_prep_async()
6397 switch (req->opcode) { in io_req_prep_async()
6410 req->opcode); in io_req_prep_async()
6411 return -EFAULT; in io_req_prep_async()
6416 u32 seq = req->ctx->cached_sq_head; in io_get_sequence()
6420 seq--; in io_get_sequence()
6427 struct io_ring_ctx *ctx = req->ctx; in io_drain_req()
6432 if (req->flags & REQ_F_FAIL) { in io_drain_req()
6443 if (ctx->drain_next) { in io_drain_req()
6444 req->flags |= REQ_F_IO_DRAIN; in io_drain_req()
6445 ctx->drain_next = false; in io_drain_req()
6448 io_for_each_link(pos, req->link) { in io_drain_req()
6449 if (pos->flags & REQ_F_IO_DRAIN) { in io_drain_req()
6450 ctx->drain_next = true; in io_drain_req()
6451 req->flags |= REQ_F_IO_DRAIN; in io_drain_req()
6457 spin_lock(&ctx->completion_lock); in io_drain_req()
6458 if (likely(list_empty_careful(&ctx->defer_list) && in io_drain_req()
6459 !(req->flags & REQ_F_IO_DRAIN))) { in io_drain_req()
6460 spin_unlock(&ctx->completion_lock); in io_drain_req()
6461 ctx->drain_active = false; in io_drain_req()
6464 spin_unlock(&ctx->completion_lock); in io_drain_req()
6468 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list)) in io_drain_req()
6477 ret = -ENOMEM; in io_drain_req()
6483 spin_lock(&ctx->completion_lock); in io_drain_req()
6484 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { in io_drain_req()
6485 spin_unlock(&ctx->completion_lock); in io_drain_req()
6491 trace_io_uring_defer(ctx, req, req->user_data); in io_drain_req()
6492 de->req = req; in io_drain_req()
6493 de->seq = seq; in io_drain_req()
6494 list_add_tail(&de->list, &ctx->defer_list); in io_drain_req()
6495 spin_unlock(&ctx->completion_lock); in io_drain_req()
6501 if (req->flags & REQ_F_BUFFER_SELECTED) { in io_clean_op()
6502 switch (req->opcode) { in io_clean_op()
6506 kfree((void *)(unsigned long)req->rw.addr); in io_clean_op()
6510 kfree(req->sr_msg.kbuf); in io_clean_op()
6515 if (req->flags & REQ_F_NEED_CLEANUP) { in io_clean_op()
6516 switch (req->opcode) { in io_clean_op()
6523 struct io_async_rw *io = req->async_data; in io_clean_op()
6525 kfree(io->free_iovec); in io_clean_op()
6530 struct io_async_msghdr *io = req->async_data; in io_clean_op()
6532 kfree(io->free_iov); in io_clean_op()
6537 if (req->open.filename) in io_clean_op()
6538 putname(req->open.filename); in io_clean_op()
6541 putname(req->rename.oldpath); in io_clean_op()
6542 putname(req->rename.newpath); in io_clean_op()
6545 putname(req->unlink.filename); in io_clean_op()
6549 if ((req->flags & REQ_F_POLLED) && req->apoll) { in io_clean_op()
6550 kfree(req->apoll->double_poll); in io_clean_op()
6551 kfree(req->apoll); in io_clean_op()
6552 req->apoll = NULL; in io_clean_op()
6554 if (req->flags & REQ_F_INFLIGHT) { in io_clean_op()
6555 struct io_uring_task *tctx = req->task->io_uring; in io_clean_op()
6557 atomic_dec(&tctx->inflight_tracked); in io_clean_op()
6559 if (req->flags & REQ_F_CREDS) in io_clean_op()
6560 put_cred(req->creds); in io_clean_op()
6562 req->flags &= ~IO_REQ_CLEAN_FLAGS; in io_clean_op()
6567 struct io_ring_ctx *ctx = req->ctx; in io_issue_sqe()
6571 if ((req->flags & REQ_F_CREDS) && req->creds != current_cred()) in io_issue_sqe()
6572 creds = override_creds(req->creds); in io_issue_sqe()
6574 switch (req->opcode) { in io_issue_sqe()
6676 ret = -EINVAL; in io_issue_sqe()
6685 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) in io_issue_sqe()
6696 return req ? &req->work : NULL; in io_wq_free_work()
6705 /* one will be dropped by ->io_free_work() after returning to io-wq */ in io_wq_submit_work()
6706 if (!(req->flags & REQ_F_REFCOUNT)) in io_wq_submit_work()
6715 /* either cancelled or io-wq is dying, so don't touch tctx->iowq */ in io_wq_submit_work()
6716 if (work->flags & IO_WQ_WORK_CANCEL) in io_wq_submit_work()
6717 ret = -ECANCELED; in io_wq_submit_work()
6727 if (ret != -EAGAIN || !(req->ctx->flags & IORING_SETUP_IOPOLL)) in io_wq_submit_work()
6741 return &table->files[i]; in io_fixed_file_slot()
6747 struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index); in io_file_from_index()
6749 return (struct file *) (slot->file_ptr & FFS_MASK); in io_file_from_index()
6760 if (S_ISREG(file_inode(file)->i_mode)) in io_fixed_file_set()
6762 file_slot->file_ptr = file_ptr; in io_fixed_file_set()
6771 if (unlikely((unsigned int)fd >= ctx->nr_user_files)) in io_file_get_fixed()
6773 fd = array_index_nospec(fd, ctx->nr_user_files); in io_file_get_fixed()
6774 file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr; in io_file_get_fixed()
6778 req->flags |= (file_ptr << REQ_F_NOWAIT_READ_BIT); in io_file_get_fixed()
6791 if (file && unlikely(file->f_op == &io_uring_fops)) in io_file_get_normal()
6807 struct io_kiocb *prev = req->timeout.prev; in io_req_task_link_timeout()
6808 int ret = -ENOENT; in io_req_task_link_timeout()
6811 if (!(req->task->flags & PF_EXITING)) in io_req_task_link_timeout()
6812 ret = io_try_cancel_userdata(req, prev->user_data); in io_req_task_link_timeout()
6813 io_req_complete_post(req, ret ?: -ETIME, 0); in io_req_task_link_timeout()
6816 io_req_complete_post(req, -ETIME, 0); in io_req_task_link_timeout()
6824 struct io_kiocb *prev, *req = data->req; in io_link_timeout_fn()
6825 struct io_ring_ctx *ctx = req->ctx; in io_link_timeout_fn()
6828 spin_lock_irqsave(&ctx->timeout_lock, flags); in io_link_timeout_fn()
6829 prev = req->timeout.head; in io_link_timeout_fn()
6830 req->timeout.head = NULL; in io_link_timeout_fn()
6841 list_del(&req->timeout.list); in io_link_timeout_fn()
6842 req->timeout.prev = prev; in io_link_timeout_fn()
6843 spin_unlock_irqrestore(&ctx->timeout_lock, flags); in io_link_timeout_fn()
6845 req->io_task_work.func = io_req_task_link_timeout; in io_link_timeout_fn()
6852 struct io_ring_ctx *ctx = req->ctx; in io_queue_linked_timeout()
6854 spin_lock_irq(&ctx->timeout_lock); in io_queue_linked_timeout()
6859 if (req->timeout.head) { in io_queue_linked_timeout()
6860 struct io_timeout_data *data = req->async_data; in io_queue_linked_timeout()
6862 data->timer.function = io_link_timeout_fn; in io_queue_linked_timeout()
6863 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), in io_queue_linked_timeout()
6864 data->mode); in io_queue_linked_timeout()
6865 list_add_tail(&req->timeout.list, &ctx->ltimeout_list); in io_queue_linked_timeout()
6867 spin_unlock_irq(&ctx->timeout_lock); in io_queue_linked_timeout()
6873 __must_hold(&req->ctx->uring_lock) in __io_queue_sqe()
6883 * doesn't support non-blocking read/write attempts in __io_queue_sqe()
6886 if (req->flags & REQ_F_COMPLETE_INLINE) { in __io_queue_sqe()
6887 struct io_ring_ctx *ctx = req->ctx; in __io_queue_sqe()
6888 struct io_submit_state *state = &ctx->submit_state; in __io_queue_sqe()
6890 state->compl_reqs[state->compl_nr++] = req; in __io_queue_sqe()
6891 if (state->compl_nr == ARRAY_SIZE(state->compl_reqs)) in __io_queue_sqe()
6899 } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) { in __io_queue_sqe()
6924 __must_hold(&req->ctx->uring_lock) in io_queue_sqe()
6926 if (unlikely(req->ctx->drain_active) && io_drain_req(req)) in io_queue_sqe()
6929 if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL)))) { in io_queue_sqe()
6931 } else if (req->flags & REQ_F_FAIL) { in io_queue_sqe()
6952 if (likely(!ctx->restricted)) in io_check_restriction()
6955 if (!test_bit(req->opcode, ctx->restrictions.sqe_op)) in io_check_restriction()
6958 if ((sqe_flags & ctx->restrictions.sqe_flags_required) != in io_check_restriction()
6959 ctx->restrictions.sqe_flags_required) in io_check_restriction()
6962 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed | in io_check_restriction()
6963 ctx->restrictions.sqe_flags_required)) in io_check_restriction()
6971 __must_hold(&ctx->uring_lock) in io_init_req()
6977 /* req is partially pre-initialised, see io_preinit_req() */ in io_init_req()
6978 req->opcode = READ_ONCE(sqe->opcode); in io_init_req()
6980 req->flags = sqe_flags = READ_ONCE(sqe->flags); in io_init_req()
6981 req->user_data = READ_ONCE(sqe->user_data); in io_init_req()
6982 req->file = NULL; in io_init_req()
6983 req->fixed_rsrc_refs = NULL; in io_init_req()
6984 req->task = current; in io_init_req()
6988 return -EINVAL; in io_init_req()
6989 if (unlikely(req->opcode >= IORING_OP_LAST)) in io_init_req()
6990 return -EINVAL; in io_init_req()
6992 return -EACCES; in io_init_req()
6995 !io_op_defs[req->opcode].buffer_select) in io_init_req()
6996 return -EOPNOTSUPP; in io_init_req()
6998 ctx->drain_active = true; in io_init_req()
7000 personality = READ_ONCE(sqe->personality); in io_init_req()
7002 req->creds = xa_load(&ctx->personalities, personality); in io_init_req()
7003 if (!req->creds) in io_init_req()
7004 return -EINVAL; in io_init_req()
7005 get_cred(req->creds); in io_init_req()
7006 req->flags |= REQ_F_CREDS; in io_init_req()
7008 state = &ctx->submit_state; in io_init_req()
7014 if (!state->plug_started && state->ios_left > 1 && in io_init_req()
7015 io_op_defs[req->opcode].plug) { in io_init_req()
7016 blk_start_plug(&state->plug); in io_init_req()
7017 state->plug_started = true; in io_init_req()
7020 if (io_op_defs[req->opcode].needs_file) { in io_init_req()
7021 req->file = io_file_get(ctx, req, READ_ONCE(sqe->fd), in io_init_req()
7023 if (unlikely(!req->file)) in io_init_req()
7024 ret = -EBADF; in io_init_req()
7027 state->ios_left--; in io_init_req()
7033 __must_hold(&ctx->uring_lock) in io_submit_sqe()
7035 struct io_submit_link *link = &ctx->submit_state.link; in io_submit_sqe()
7042 if (link->head) { in io_submit_sqe()
7047 * so let's leverage req->result to distinguish if a head in io_submit_sqe()
7052 if (!(link->head->flags & REQ_F_FAIL)) in io_submit_sqe()
7053 req_fail_link_node(link->head, -ECANCELED); in io_submit_sqe()
7054 } else if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { in io_submit_sqe()
7070 trace_io_uring_submit_sqe(ctx, req, req->opcode, req->user_data, in io_submit_sqe()
7071 req->flags, true, in io_submit_sqe()
7072 ctx->flags & IORING_SETUP_SQPOLL); in io_submit_sqe()
7081 if (link->head) { in io_submit_sqe()
7082 struct io_kiocb *head = link->head; in io_submit_sqe()
7084 if (!(req->flags & REQ_F_FAIL)) { in io_submit_sqe()
7088 if (!(head->flags & REQ_F_FAIL)) in io_submit_sqe()
7089 req_fail_link_node(head, -ECANCELED); in io_submit_sqe()
7093 link->last->link = req; in io_submit_sqe()
7094 link->last = req; in io_submit_sqe()
7097 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) { in io_submit_sqe()
7098 link->head = NULL; in io_submit_sqe()
7102 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) { in io_submit_sqe()
7103 link->head = req; in io_submit_sqe()
7104 link->last = req; in io_submit_sqe()
7119 if (state->link.head) in io_submit_state_end()
7120 io_queue_sqe(state->link.head); in io_submit_state_end()
7121 if (state->compl_nr) in io_submit_state_end()
7123 if (state->plug_started) in io_submit_state_end()
7124 blk_finish_plug(&state->plug); in io_submit_state_end()
7133 state->plug_started = false; in io_submit_state_start()
7134 state->ios_left = max_ios; in io_submit_state_start()
7136 state->link.head = NULL; in io_submit_state_start()
7141 struct io_rings *rings = ctx->rings; in io_commit_sqring()
7148 smp_store_release(&rings->sq.head, ctx->cached_sq_head); in io_commit_sqring()
7157 * prevent a re-load down the line.
7161 unsigned head, mask = ctx->sq_entries - 1; in io_get_sqe()
7162 unsigned sq_idx = ctx->cached_sq_head++ & mask; in io_get_sqe()
7172 head = READ_ONCE(ctx->sq_array[sq_idx]); in io_get_sqe()
7173 if (likely(head < ctx->sq_entries)) in io_get_sqe()
7174 return &ctx->sq_sqes[head]; in io_get_sqe()
7177 ctx->cq_extra--; in io_get_sqe()
7178 WRITE_ONCE(ctx->rings->sq_dropped, in io_get_sqe()
7179 READ_ONCE(ctx->rings->sq_dropped) + 1); in io_get_sqe()
7184 __must_hold(&ctx->uring_lock) in io_submit_sqes()
7189 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx)); in io_submit_sqes()
7190 if (!percpu_ref_tryget_many(&ctx->refs, nr)) in io_submit_sqes()
7191 return -EAGAIN; in io_submit_sqes()
7194 io_submit_state_start(&ctx->submit_state, nr); in io_submit_sqes()
7202 submitted = -EAGAIN; in io_submit_sqes()
7207 list_add(&req->inflight_entry, &ctx->submit_state.free_list); in io_submit_sqes()
7210 /* will complete beyond this point, count as submitted */ in io_submit_sqes()
7217 int ref_used = (submitted == -EAGAIN) ? 0 : submitted; in io_submit_sqes()
7218 int unused = nr - ref_used; in io_submit_sqes()
7220 current->io_uring->cached_refs += unused; in io_submit_sqes()
7221 percpu_ref_put_many(&ctx->refs, unused); in io_submit_sqes()
7224 io_submit_state_end(&ctx->submit_state, ctx); in io_submit_sqes()
7233 return READ_ONCE(sqd->state); in io_sqd_events_pending()
7239 spin_lock(&ctx->completion_lock); in io_ring_set_wakeup_flag()
7240 WRITE_ONCE(ctx->rings->sq_flags, in io_ring_set_wakeup_flag()
7241 ctx->rings->sq_flags | IORING_SQ_NEED_WAKEUP); in io_ring_set_wakeup_flag()
7242 spin_unlock(&ctx->completion_lock); in io_ring_set_wakeup_flag()
7247 spin_lock(&ctx->completion_lock); in io_ring_clear_wakeup_flag()
7248 WRITE_ONCE(ctx->rings->sq_flags, in io_ring_clear_wakeup_flag()
7249 ctx->rings->sq_flags & ~IORING_SQ_NEED_WAKEUP); in io_ring_clear_wakeup_flag()
7250 spin_unlock(&ctx->completion_lock); in io_ring_clear_wakeup_flag()
7263 if (!list_empty(&ctx->iopoll_list) || to_submit) { in __io_sq_thread()
7267 if (ctx->sq_creds != current_cred()) in __io_sq_thread()
7268 creds = override_creds(ctx->sq_creds); in __io_sq_thread()
7270 mutex_lock(&ctx->uring_lock); in __io_sq_thread()
7271 if (!list_empty(&ctx->iopoll_list)) in __io_sq_thread()
7278 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) && in __io_sq_thread()
7279 !(ctx->flags & IORING_SETUP_R_DISABLED)) in __io_sq_thread()
7281 mutex_unlock(&ctx->uring_lock); in __io_sq_thread()
7283 if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait)) in __io_sq_thread()
7284 wake_up(&ctx->sqo_sq_wait); in __io_sq_thread()
7297 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) in io_sqd_update_thread_idle()
7298 sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle); in io_sqd_update_thread_idle()
7299 sqd->sq_thread_idle = sq_thread_idle; in io_sqd_update_thread_idle()
7307 if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) || in io_sqd_handle_event()
7309 mutex_unlock(&sqd->lock); in io_sqd_handle_event()
7313 mutex_lock(&sqd->lock); in io_sqd_handle_event()
7315 return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); in io_sqd_handle_event()
7326 snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid); in io_sq_thread()
7329 if (sqd->sq_cpu != -1) in io_sq_thread()
7330 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu)); in io_sq_thread()
7333 current->flags |= PF_NO_SETAFFINITY; in io_sq_thread()
7335 mutex_lock(&sqd->lock); in io_sq_thread()
7342 timeout = jiffies + sqd->sq_thread_idle; in io_sq_thread()
7345 cap_entries = !list_is_singular(&sqd->ctx_list); in io_sq_thread()
7346 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { in io_sq_thread()
7349 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list))) in io_sq_thread()
7358 timeout = jiffies + sqd->sq_thread_idle; in io_sq_thread()
7362 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE); in io_sq_thread()
7363 if (!io_sqd_events_pending(sqd) && !current->task_works) { in io_sq_thread()
7366 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) { in io_sq_thread()
7369 if ((ctx->flags & IORING_SETUP_IOPOLL) && in io_sq_thread()
7370 !list_empty_careful(&ctx->iopoll_list)) { in io_sq_thread()
7381 mutex_unlock(&sqd->lock); in io_sq_thread()
7383 mutex_lock(&sqd->lock); in io_sq_thread()
7385 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) in io_sq_thread()
7389 finish_wait(&sqd->wait, &wait); in io_sq_thread()
7390 timeout = jiffies + sqd->sq_thread_idle; in io_sq_thread()
7394 sqd->thread = NULL; in io_sq_thread()
7395 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) in io_sq_thread()
7398 mutex_unlock(&sqd->lock); in io_sq_thread()
7400 complete(&sqd->exited); in io_sq_thread()
7413 struct io_ring_ctx *ctx = iowq->ctx; in io_should_wake()
7414 int dist = ctx->cached_cq_tail - (int) iowq->cq_tail; in io_should_wake()
7419 * regardless of event count. in io_should_wake()
7421 return dist >= 0 || atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts; in io_should_wake()
7434 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->check_cq_overflow)) in io_wake_function()
7436 return -1; in io_wake_function()
7446 return -ERESTARTSYS; in io_run_task_work_sig()
7447 return -EINTR; in io_run_task_work_sig()
7450 /* when returns >0, the caller should retry */
7461 /* let the caller flush overflows, retry */ in io_cqring_wait_schedule()
7462 if (test_bit(0, &ctx->check_cq_overflow)) in io_cqring_wait_schedule()
7466 return -ETIME; in io_cqring_wait_schedule()
7479 struct io_rings *rings = ctx->rings; in io_cqring_wait()
7495 return -EFAULT; in io_cqring_wait()
7516 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts); in io_cqring_wait()
7517 iowq.cq_tail = READ_ONCE(ctx->rings->cq.head) + min_events; in io_cqring_wait()
7523 ret = -EBUSY; in io_cqring_wait()
7526 prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq, in io_cqring_wait()
7529 finish_wait(&ctx->cq_wait, &iowq.wq); in io_cqring_wait()
7533 restore_saved_sigmask_unless(ret == -EINTR); in io_cqring_wait()
7535 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0; in io_cqring_wait()
7565 size -= this_size; in io_alloc_page_table()
7572 percpu_ref_exit(&ref_node->refs); in io_rsrc_node_destroy()
7579 struct io_ring_ctx *ctx = node->rsrc_data->ctx; in io_rsrc_node_ref_zero()
7584 spin_lock_irqsave(&ctx->rsrc_ref_lock, flags); in io_rsrc_node_ref_zero()
7585 node->done = true; in io_rsrc_node_ref_zero()
7587 /* if we are mid-quiesce then do not delay */ in io_rsrc_node_ref_zero()
7588 if (node->rsrc_data->quiesce) in io_rsrc_node_ref_zero()
7591 while (!list_empty(&ctx->rsrc_ref_list)) { in io_rsrc_node_ref_zero()
7592 node = list_first_entry(&ctx->rsrc_ref_list, in io_rsrc_node_ref_zero()
7595 if (!node->done) in io_rsrc_node_ref_zero()
7597 list_del(&node->node); in io_rsrc_node_ref_zero()
7598 first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist); in io_rsrc_node_ref_zero()
7600 spin_unlock_irqrestore(&ctx->rsrc_ref_lock, flags); in io_rsrc_node_ref_zero()
7603 mod_delayed_work(system_wq, &ctx->rsrc_put_work, delay); in io_rsrc_node_ref_zero()
7614 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero, in io_rsrc_node_alloc()
7619 INIT_LIST_HEAD(&ref_node->node); in io_rsrc_node_alloc()
7620 INIT_LIST_HEAD(&ref_node->rsrc_list); in io_rsrc_node_alloc()
7621 ref_node->done = false; in io_rsrc_node_alloc()
7628 WARN_ON_ONCE(!ctx->rsrc_backup_node); in io_rsrc_node_switch()
7629 WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node); in io_rsrc_node_switch()
7632 struct io_rsrc_node *rsrc_node = ctx->rsrc_node; in io_rsrc_node_switch()
7634 rsrc_node->rsrc_data = data_to_kill; in io_rsrc_node_switch()
7635 spin_lock_irq(&ctx->rsrc_ref_lock); in io_rsrc_node_switch()
7636 list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list); in io_rsrc_node_switch()
7637 spin_unlock_irq(&ctx->rsrc_ref_lock); in io_rsrc_node_switch()
7639 atomic_inc(&data_to_kill->refs); in io_rsrc_node_switch()
7640 percpu_ref_kill(&rsrc_node->refs); in io_rsrc_node_switch()
7641 ctx->rsrc_node = NULL; in io_rsrc_node_switch()
7644 if (!ctx->rsrc_node) { in io_rsrc_node_switch()
7645 ctx->rsrc_node = ctx->rsrc_backup_node; in io_rsrc_node_switch()
7646 ctx->rsrc_backup_node = NULL; in io_rsrc_node_switch()
7652 if (ctx->rsrc_backup_node) in io_rsrc_node_switch_start()
7654 ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx); in io_rsrc_node_switch_start()
7655 return ctx->rsrc_backup_node ? 0 : -ENOMEM; in io_rsrc_node_switch_start()
7662 /* As we may drop ->uring_lock, other task may have started quiesce */ in io_rsrc_ref_quiesce()
7663 if (data->quiesce) in io_rsrc_ref_quiesce()
7664 return -ENXIO; in io_rsrc_ref_quiesce()
7666 data->quiesce = true; in io_rsrc_ref_quiesce()
7674 if (atomic_dec_and_test(&data->refs)) in io_rsrc_ref_quiesce()
7676 mutex_unlock(&ctx->uring_lock); in io_rsrc_ref_quiesce()
7677 flush_delayed_work(&ctx->rsrc_put_work); in io_rsrc_ref_quiesce()
7678 ret = wait_for_completion_interruptible(&data->done); in io_rsrc_ref_quiesce()
7680 mutex_lock(&ctx->uring_lock); in io_rsrc_ref_quiesce()
7681 if (atomic_read(&data->refs) > 0) { in io_rsrc_ref_quiesce()
7686 mutex_unlock(&ctx->uring_lock); in io_rsrc_ref_quiesce()
7692 atomic_inc(&data->refs); in io_rsrc_ref_quiesce()
7693 /* wait for all works potentially completing data->done */ in io_rsrc_ref_quiesce()
7694 flush_delayed_work(&ctx->rsrc_put_work); in io_rsrc_ref_quiesce()
7695 reinit_completion(&data->done); in io_rsrc_ref_quiesce()
7698 mutex_lock(&ctx->uring_lock); in io_rsrc_ref_quiesce()
7700 data->quiesce = false; in io_rsrc_ref_quiesce()
7710 return &data->tags[table_idx][off]; in io_get_tag_slot()
7715 size_t size = data->nr * sizeof(data->tags[0][0]); in io_rsrc_data_free()
7717 if (data->tags) in io_rsrc_data_free()
7718 io_free_page_table((void **)data->tags, size); in io_rsrc_data_free()
7727 int ret = -ENOMEM; in io_rsrc_data_alloc()
7732 return -ENOMEM; in io_rsrc_data_alloc()
7733 data->tags = (u64 **)io_alloc_page_table(nr * sizeof(data->tags[0][0])); in io_rsrc_data_alloc()
7734 if (!data->tags) { in io_rsrc_data_alloc()
7736 return -ENOMEM; in io_rsrc_data_alloc()
7739 data->nr = nr; in io_rsrc_data_alloc()
7740 data->ctx = ctx; in io_rsrc_data_alloc()
7741 data->do_put = do_put; in io_rsrc_data_alloc()
7743 ret = -EFAULT; in io_rsrc_data_alloc()
7753 atomic_set(&data->refs, 1); in io_rsrc_data_alloc()
7754 init_completion(&data->done); in io_rsrc_data_alloc()
7764 table->files = kvcalloc(nr_files, sizeof(table->files[0]), in io_alloc_file_tables()
7766 return !!table->files; in io_alloc_file_tables()
7771 kvfree(table->files); in io_free_file_tables()
7772 table->files = NULL; in io_free_file_tables()
7778 if (ctx->ring_sock) { in __io_sqe_files_unregister()
7779 struct sock *sock = ctx->ring_sock->sk; in __io_sqe_files_unregister()
7782 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL) in __io_sqe_files_unregister()
7788 for (i = 0; i < ctx->nr_user_files; i++) { in __io_sqe_files_unregister()
7796 io_free_file_tables(&ctx->file_table); in __io_sqe_files_unregister()
7797 io_rsrc_data_free(ctx->file_data); in __io_sqe_files_unregister()
7798 ctx->file_data = NULL; in __io_sqe_files_unregister()
7799 ctx->nr_user_files = 0; in __io_sqe_files_unregister()
7804 unsigned nr = ctx->nr_user_files; in io_sqe_files_unregister()
7807 if (!ctx->file_data) in io_sqe_files_unregister()
7808 return -ENXIO; in io_sqe_files_unregister()
7811 * Quiesce may unlock ->uring_lock, and while it's not held in io_sqe_files_unregister()
7814 ctx->nr_user_files = 0; in io_sqe_files_unregister()
7815 ret = io_rsrc_ref_quiesce(ctx->file_data, ctx); in io_sqe_files_unregister()
7816 ctx->nr_user_files = nr; in io_sqe_files_unregister()
7823 __releases(&sqd->lock) in io_sq_thread_unpark()
7825 WARN_ON_ONCE(sqd->thread == current); in io_sq_thread_unpark()
7831 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); in io_sq_thread_unpark()
7832 if (atomic_dec_return(&sqd->park_pending)) in io_sq_thread_unpark()
7833 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); in io_sq_thread_unpark()
7834 mutex_unlock(&sqd->lock); in io_sq_thread_unpark()
7838 __acquires(&sqd->lock) in io_sq_thread_park()
7840 WARN_ON_ONCE(sqd->thread == current); in io_sq_thread_park()
7842 atomic_inc(&sqd->park_pending); in io_sq_thread_park()
7843 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state); in io_sq_thread_park()
7844 mutex_lock(&sqd->lock); in io_sq_thread_park()
7845 if (sqd->thread) in io_sq_thread_park()
7846 wake_up_process(sqd->thread); in io_sq_thread_park()
7851 WARN_ON_ONCE(sqd->thread == current); in io_sq_thread_stop()
7852 WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)); in io_sq_thread_stop()
7854 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state); in io_sq_thread_stop()
7855 mutex_lock(&sqd->lock); in io_sq_thread_stop()
7856 if (sqd->thread) in io_sq_thread_stop()
7857 wake_up_process(sqd->thread); in io_sq_thread_stop()
7858 mutex_unlock(&sqd->lock); in io_sq_thread_stop()
7859 wait_for_completion(&sqd->exited); in io_sq_thread_stop()
7864 if (refcount_dec_and_test(&sqd->refs)) { in io_put_sq_data()
7865 WARN_ON_ONCE(atomic_read(&sqd->park_pending)); in io_put_sq_data()
7874 struct io_sq_data *sqd = ctx->sq_data; in io_sq_thread_finish()
7878 list_del_init(&ctx->sqd_list); in io_sq_thread_finish()
7883 ctx->sq_data = NULL; in io_sq_thread_finish()
7893 f = fdget(p->wq_fd); in io_attach_sq_data()
7895 return ERR_PTR(-ENXIO); in io_attach_sq_data()
7896 if (f.file->f_op != &io_uring_fops) { in io_attach_sq_data()
7898 return ERR_PTR(-EINVAL); in io_attach_sq_data()
7901 ctx_attach = f.file->private_data; in io_attach_sq_data()
7902 sqd = ctx_attach->sq_data; in io_attach_sq_data()
7905 return ERR_PTR(-EINVAL); in io_attach_sq_data()
7907 if (sqd->task_tgid != current->tgid) { in io_attach_sq_data()
7909 return ERR_PTR(-EPERM); in io_attach_sq_data()
7912 refcount_inc(&sqd->refs); in io_attach_sq_data()
7923 if (p->flags & IORING_SETUP_ATTACH_WQ) { in io_get_sq_data()
7930 if (PTR_ERR(sqd) != -EPERM) in io_get_sq_data()
7936 return ERR_PTR(-ENOMEM); in io_get_sq_data()
7938 atomic_set(&sqd->park_pending, 0); in io_get_sq_data()
7939 refcount_set(&sqd->refs, 1); in io_get_sq_data()
7940 INIT_LIST_HEAD(&sqd->ctx_list); in io_get_sq_data()
7941 mutex_init(&sqd->lock); in io_get_sq_data()
7942 init_waitqueue_head(&sqd->wait); in io_get_sq_data()
7943 init_completion(&sqd->exited); in io_get_sq_data()
7955 struct sock *sk = ctx->ring_sock->sk; in __io_sqe_files_scm()
7962 return -ENOMEM; in __io_sqe_files_scm()
7967 return -ENOMEM; in __io_sqe_files_scm()
7970 skb->sk = sk; in __io_sqe_files_scm()
7971 skb->scm_io_uring = 1; in __io_sqe_files_scm()
7974 fpl->user = get_uid(current_user()); in __io_sqe_files_scm()
7980 fpl->fp[nr_files] = get_file(file); in __io_sqe_files_scm()
7981 unix_inflight(fpl->user, fpl->fp[nr_files]); in __io_sqe_files_scm()
7986 fpl->max = SCM_MAX_FD; in __io_sqe_files_scm()
7987 fpl->count = nr_files; in __io_sqe_files_scm()
7989 skb->destructor = unix_destruct_scm; in __io_sqe_files_scm()
7990 refcount_add(skb->truesize, &sk->sk_wmem_alloc); in __io_sqe_files_scm()
7991 skb_queue_head(&sk->sk_receive_queue, skb); in __io_sqe_files_scm()
8001 free_uid(fpl->user); in __io_sqe_files_scm()
8019 left = ctx->nr_user_files; in io_sqe_files_scm()
8026 left -= this_files; in io_sqe_files_scm()
8033 while (total < ctx->nr_user_files) { in io_sqe_files_scm()
8052 struct file *file = prsrc->file; in io_rsrc_file_put()
8054 struct sock *sock = ctx->ring_sock->sk; in io_rsrc_file_put()
8055 struct sk_buff_head list, *head = &sock->sk_receive_queue; in io_rsrc_file_put()
8070 for (i = 0; i < fp->count; i++) { in io_rsrc_file_put()
8073 if (fp->fp[i] != file) in io_rsrc_file_put()
8076 unix_notinflight(fp->user, fp->fp[i]); in io_rsrc_file_put()
8077 left = fp->count - 1 - i; in io_rsrc_file_put()
8079 memmove(&fp->fp[i], &fp->fp[i + 1], in io_rsrc_file_put()
8082 fp->count--; in io_rsrc_file_put()
8083 if (!fp->count) { in io_rsrc_file_put()
8103 spin_lock_irq(&head->lock); in io_rsrc_file_put()
8106 spin_unlock_irq(&head->lock); in io_rsrc_file_put()
8115 struct io_rsrc_data *rsrc_data = ref_node->rsrc_data; in __io_rsrc_put_work()
8116 struct io_ring_ctx *ctx = rsrc_data->ctx; in __io_rsrc_put_work()
8119 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) { in __io_rsrc_put_work()
8120 list_del(&prsrc->list); in __io_rsrc_put_work()
8122 if (prsrc->tag) { in __io_rsrc_put_work()
8123 bool lock_ring = ctx->flags & IORING_SETUP_IOPOLL; in __io_rsrc_put_work()
8126 spin_lock(&ctx->completion_lock); in __io_rsrc_put_work()
8127 io_fill_cqe_aux(ctx, prsrc->tag, 0, 0); in __io_rsrc_put_work()
8129 spin_unlock(&ctx->completion_lock); in __io_rsrc_put_work()
8134 rsrc_data->do_put(ctx, prsrc); in __io_rsrc_put_work()
8139 if (atomic_dec_and_test(&rsrc_data->refs)) in __io_rsrc_put_work()
8140 complete(&rsrc_data->done); in __io_rsrc_put_work()
8149 node = llist_del_all(&ctx->rsrc_put_llist); in io_rsrc_put_work()
8153 struct llist_node *next = node->next; in io_rsrc_put_work()
8169 if (ctx->file_data) in io_sqe_files_register()
8170 return -EBUSY; in io_sqe_files_register()
8172 return -EINVAL; in io_sqe_files_register()
8174 return -EMFILE; in io_sqe_files_register()
8176 return -EMFILE; in io_sqe_files_register()
8181 &ctx->file_data); in io_sqe_files_register()
8185 ret = -ENOMEM; in io_sqe_files_register()
8186 if (!io_alloc_file_tables(&ctx->file_table, nr_args)) in io_sqe_files_register()
8189 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) { in io_sqe_files_register()
8191 ret = -EFAULT; in io_sqe_files_register()
8195 if (fd == -1) { in io_sqe_files_register()
8196 ret = -EINVAL; in io_sqe_files_register()
8197 if (unlikely(*io_get_tag_slot(ctx->file_data, i))) in io_sqe_files_register()
8203 ret = -EBADF; in io_sqe_files_register()
8214 if (file->f_op == &io_uring_fops) { in io_sqe_files_register()
8218 io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file); in io_sqe_files_register()
8230 for (i = 0; i < ctx->nr_user_files; i++) { in io_sqe_files_register()
8235 io_free_file_tables(&ctx->file_table); in io_sqe_files_register()
8236 ctx->nr_user_files = 0; in io_sqe_files_register()
8238 io_rsrc_data_free(ctx->file_data); in io_sqe_files_register()
8239 ctx->file_data = NULL; in io_sqe_files_register()
8247 struct sock *sock = ctx->ring_sock->sk; in io_sqe_file_register()
8248 struct sk_buff_head *head = &sock->sk_receive_queue; in io_sqe_file_register()
8256 spin_lock_irq(&head->lock); in io_sqe_file_register()
8261 if (fpl->count < SCM_MAX_FD) { in io_sqe_file_register()
8263 spin_unlock_irq(&head->lock); in io_sqe_file_register()
8264 fpl->fp[fpl->count] = get_file(file); in io_sqe_file_register()
8265 unix_inflight(fpl->user, fpl->fp[fpl->count]); in io_sqe_file_register()
8266 fpl->count++; in io_sqe_file_register()
8267 spin_lock_irq(&head->lock); in io_sqe_file_register()
8273 spin_unlock_irq(&head->lock); in io_sqe_file_register()
8294 return -ENOMEM; in io_queue_rsrc_removal()
8296 prsrc->tag = *tag_slot; in io_queue_rsrc_removal()
8298 prsrc->rsrc = rsrc; in io_queue_rsrc_removal()
8299 list_add(&prsrc->list, &node->rsrc_list); in io_queue_rsrc_removal()
8306 struct io_ring_ctx *ctx = req->ctx; in io_install_fixed_file()
8310 int ret = -EBADF; in io_install_fixed_file()
8313 if (file->f_op == &io_uring_fops) in io_install_fixed_file()
8315 ret = -ENXIO; in io_install_fixed_file()
8316 if (!ctx->file_data) in io_install_fixed_file()
8318 ret = -EINVAL; in io_install_fixed_file()
8319 if (slot_index >= ctx->nr_user_files) in io_install_fixed_file()
8322 slot_index = array_index_nospec(slot_index, ctx->nr_user_files); in io_install_fixed_file()
8323 file_slot = io_fixed_file_slot(&ctx->file_table, slot_index); in io_install_fixed_file()
8325 if (file_slot->file_ptr) { in io_install_fixed_file()
8332 old_file = (struct file *)(file_slot->file_ptr & FFS_MASK); in io_install_fixed_file()
8333 ret = io_queue_rsrc_removal(ctx->file_data, slot_index, in io_install_fixed_file()
8334 ctx->rsrc_node, old_file); in io_install_fixed_file()
8337 file_slot->file_ptr = 0; in io_install_fixed_file()
8341 *io_get_tag_slot(ctx->file_data, slot_index) = 0; in io_install_fixed_file()
8345 file_slot->file_ptr = 0; in io_install_fixed_file()
8352 io_rsrc_node_switch(ctx, ctx->file_data); in io_install_fixed_file()
8361 unsigned int offset = req->close.file_slot - 1; in io_close_fixed()
8362 struct io_ring_ctx *ctx = req->ctx; in io_close_fixed()
8368 ret = -ENXIO; in io_close_fixed()
8369 if (unlikely(!ctx->file_data)) in io_close_fixed()
8371 ret = -EINVAL; in io_close_fixed()
8372 if (offset >= ctx->nr_user_files) in io_close_fixed()
8378 offset = array_index_nospec(offset, ctx->nr_user_files); in io_close_fixed()
8379 file_slot = io_fixed_file_slot(&ctx->file_table, offset); in io_close_fixed()
8380 ret = -EBADF; in io_close_fixed()
8381 if (!file_slot->file_ptr) in io_close_fixed()
8384 file = (struct file *)(file_slot->file_ptr & FFS_MASK); in io_close_fixed()
8385 ret = io_queue_rsrc_removal(ctx->file_data, offset, ctx->rsrc_node, file); in io_close_fixed()
8389 file_slot->file_ptr = 0; in io_close_fixed()
8390 io_rsrc_node_switch(ctx, ctx->file_data); in io_close_fixed()
8401 u64 __user *tags = u64_to_user_ptr(up->tags); in __io_sqe_files_update()
8402 __s32 __user *fds = u64_to_user_ptr(up->data); in __io_sqe_files_update()
8403 struct io_rsrc_data *data = ctx->file_data; in __io_sqe_files_update()
8410 if (!ctx->file_data) in __io_sqe_files_update()
8411 return -ENXIO; in __io_sqe_files_update()
8412 if (up->offset + nr_args > ctx->nr_user_files) in __io_sqe_files_update()
8413 return -EINVAL; in __io_sqe_files_update()
8420 err = -EFAULT; in __io_sqe_files_update()
8423 if ((fd == IORING_REGISTER_FILES_SKIP || fd == -1) && tag) { in __io_sqe_files_update()
8424 err = -EINVAL; in __io_sqe_files_update()
8430 i = array_index_nospec(up->offset + done, ctx->nr_user_files); in __io_sqe_files_update()
8431 file_slot = io_fixed_file_slot(&ctx->file_table, i); in __io_sqe_files_update()
8433 if (file_slot->file_ptr) { in __io_sqe_files_update()
8434 file = (struct file *)(file_slot->file_ptr & FFS_MASK); in __io_sqe_files_update()
8435 err = io_queue_rsrc_removal(data, i, ctx->rsrc_node, file); in __io_sqe_files_update()
8438 file_slot->file_ptr = 0; in __io_sqe_files_update()
8441 if (fd != -1) { in __io_sqe_files_update()
8444 err = -EBADF; in __io_sqe_files_update()
8455 if (file->f_op == &io_uring_fops) { in __io_sqe_files_update()
8457 err = -EBADF; in __io_sqe_files_update()
8464 file_slot->file_ptr = 0; in __io_sqe_files_update()
8483 mutex_lock(&ctx->uring_lock); in io_init_wq_offload()
8484 hash = ctx->hash_map; in io_init_wq_offload()
8488 mutex_unlock(&ctx->uring_lock); in io_init_wq_offload()
8489 return ERR_PTR(-ENOMEM); in io_init_wq_offload()
8491 refcount_set(&hash->refs, 1); in io_init_wq_offload()
8492 init_waitqueue_head(&hash->wait); in io_init_wq_offload()
8493 ctx->hash_map = hash; in io_init_wq_offload()
8495 mutex_unlock(&ctx->uring_lock); in io_init_wq_offload()
8503 concurrency = min(ctx->sq_entries, 4 * num_online_cpus()); in io_init_wq_offload()
8516 return -ENOMEM; in io_uring_alloc_task_context()
8518 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL); in io_uring_alloc_task_context()
8524 tctx->io_wq = io_init_wq_offload(ctx, task); in io_uring_alloc_task_context()
8525 if (IS_ERR(tctx->io_wq)) { in io_uring_alloc_task_context()
8526 ret = PTR_ERR(tctx->io_wq); in io_uring_alloc_task_context()
8527 percpu_counter_destroy(&tctx->inflight); in io_uring_alloc_task_context()
8532 xa_init(&tctx->xa); in io_uring_alloc_task_context()
8533 init_waitqueue_head(&tctx->wait); in io_uring_alloc_task_context()
8534 atomic_set(&tctx->in_idle, 0); in io_uring_alloc_task_context()
8535 atomic_set(&tctx->inflight_tracked, 0); in io_uring_alloc_task_context()
8536 task->io_uring = tctx; in io_uring_alloc_task_context()
8537 spin_lock_init(&tctx->task_lock); in io_uring_alloc_task_context()
8538 INIT_WQ_LIST(&tctx->task_list); in io_uring_alloc_task_context()
8539 init_task_work(&tctx->task_work, tctx_task_work); in io_uring_alloc_task_context()
8545 struct io_uring_task *tctx = tsk->io_uring; in __io_uring_free()
8547 WARN_ON_ONCE(!xa_empty(&tctx->xa)); in __io_uring_free()
8548 WARN_ON_ONCE(tctx->io_wq); in __io_uring_free()
8549 WARN_ON_ONCE(tctx->cached_refs); in __io_uring_free()
8551 percpu_counter_destroy(&tctx->inflight); in __io_uring_free()
8553 tsk->io_uring = NULL; in __io_uring_free()
8562 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == in io_sq_offload_create()
8566 f = fdget(p->wq_fd); in io_sq_offload_create()
8568 return -ENXIO; in io_sq_offload_create()
8569 if (f.file->f_op != &io_uring_fops) { in io_sq_offload_create()
8571 return -EINVAL; in io_sq_offload_create()
8575 if (ctx->flags & IORING_SETUP_SQPOLL) { in io_sq_offload_create()
8586 ctx->sq_creds = get_current_cred(); in io_sq_offload_create()
8587 ctx->sq_data = sqd; in io_sq_offload_create()
8588 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle); in io_sq_offload_create()
8589 if (!ctx->sq_thread_idle) in io_sq_offload_create()
8590 ctx->sq_thread_idle = HZ; in io_sq_offload_create()
8593 list_add(&ctx->sqd_list, &sqd->ctx_list); in io_sq_offload_create()
8596 ret = (attached && !sqd->thread) ? -ENXIO : 0; in io_sq_offload_create()
8604 if (p->flags & IORING_SETUP_SQ_AFF) { in io_sq_offload_create()
8605 int cpu = p->sq_thread_cpu; in io_sq_offload_create()
8607 ret = -EINVAL; in io_sq_offload_create()
8610 sqd->sq_cpu = cpu; in io_sq_offload_create()
8612 sqd->sq_cpu = -1; in io_sq_offload_create()
8615 sqd->task_pid = current->pid; in io_sq_offload_create()
8616 sqd->task_tgid = current->tgid; in io_sq_offload_create()
8623 sqd->thread = tsk; in io_sq_offload_create()
8628 } else if (p->flags & IORING_SETUP_SQ_AFF) { in io_sq_offload_create()
8630 ret = -EINVAL; in io_sq_offload_create()
8636 complete(&ctx->sq_data->exited); in io_sq_offload_create()
8645 atomic_long_sub(nr_pages, &user->locked_vm); in __io_unaccount_mem()
8657 cur_pages = atomic_long_read(&user->locked_vm); in __io_account_mem()
8660 return -ENOMEM; in __io_account_mem()
8661 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, in __io_account_mem()
8669 if (ctx->user) in io_unaccount_mem()
8670 __io_unaccount_mem(ctx->user, nr_pages); in io_unaccount_mem()
8672 if (ctx->mm_account) in io_unaccount_mem()
8673 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm); in io_unaccount_mem()
8680 if (ctx->user) { in io_account_mem()
8681 ret = __io_account_mem(ctx->user, nr_pages); in io_account_mem()
8686 if (ctx->mm_account) in io_account_mem()
8687 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm); in io_account_mem()
8745 if (imu != ctx->dummy_ubuf) { in io_buffer_unmap()
8746 for (i = 0; i < imu->nr_bvecs; i++) in io_buffer_unmap()
8747 unpin_user_page(imu->bvec[i].bv_page); in io_buffer_unmap()
8748 if (imu->acct_pages) in io_buffer_unmap()
8749 io_unaccount_mem(ctx, imu->acct_pages); in io_buffer_unmap()
8757 io_buffer_unmap(ctx, &prsrc->buf); in io_rsrc_buf_put()
8758 prsrc->buf = NULL; in io_rsrc_buf_put()
8765 for (i = 0; i < ctx->nr_user_bufs; i++) in __io_sqe_buffers_unregister()
8766 io_buffer_unmap(ctx, &ctx->user_bufs[i]); in __io_sqe_buffers_unregister()
8767 kfree(ctx->user_bufs); in __io_sqe_buffers_unregister()
8768 io_rsrc_data_free(ctx->buf_data); in __io_sqe_buffers_unregister()
8769 ctx->user_bufs = NULL; in __io_sqe_buffers_unregister()
8770 ctx->buf_data = NULL; in __io_sqe_buffers_unregister()
8771 ctx->nr_user_bufs = 0; in __io_sqe_buffers_unregister()
8776 unsigned nr = ctx->nr_user_bufs; in io_sqe_buffers_unregister()
8779 if (!ctx->buf_data) in io_sqe_buffers_unregister()
8780 return -ENXIO; in io_sqe_buffers_unregister()
8783 * Quiesce may unlock ->uring_lock, and while it's not held in io_sqe_buffers_unregister()
8786 ctx->nr_user_bufs = 0; in io_sqe_buffers_unregister()
8787 ret = io_rsrc_ref_quiesce(ctx->buf_data, ctx); in io_sqe_buffers_unregister()
8788 ctx->nr_user_bufs = nr; in io_sqe_buffers_unregister()
8800 if (ctx->compat) { in io_copy_iov()
8806 return -EFAULT; in io_copy_iov()
8808 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base); in io_copy_iov()
8809 dst->iov_len = ciov.iov_len; in io_copy_iov()
8815 return -EFAULT; in io_copy_iov()
8842 for (i = 0; i < ctx->nr_user_bufs; i++) { in headpage_already_acct()
8843 struct io_mapped_ubuf *imu = ctx->user_bufs[i]; in headpage_already_acct()
8845 for (j = 0; j < imu->nr_bvecs; j++) { in headpage_already_acct()
8846 if (!PageCompound(imu->bvec[j].bv_page)) in headpage_already_acct()
8848 if (compound_head(imu->bvec[j].bv_page) == hpage) in headpage_already_acct()
8862 imu->acct_pages = 0; in io_buffer_account_pin()
8865 imu->acct_pages++; in io_buffer_account_pin()
8875 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT; in io_buffer_account_pin()
8879 if (!imu->acct_pages) in io_buffer_account_pin()
8882 ret = io_account_mem(ctx, imu->acct_pages); in io_buffer_account_pin()
8884 imu->acct_pages = 0; in io_buffer_account_pin()
8899 if (!iov->iov_base) { in io_sqe_buffer_register()
8900 *pimu = ctx->dummy_ubuf; in io_sqe_buffer_register()
8904 ubuf = (unsigned long) iov->iov_base; in io_sqe_buffer_register()
8905 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; in io_sqe_buffer_register()
8907 nr_pages = end - start; in io_sqe_buffer_register()
8910 ret = -ENOMEM; in io_sqe_buffer_register()
8926 mmap_read_lock(current->mm); in io_sqe_buffer_register()
8936 if (vma->vm_file && in io_sqe_buffer_register()
8937 !is_file_hugepages(vma->vm_file)) { in io_sqe_buffer_register()
8938 ret = -EOPNOTSUPP; in io_sqe_buffer_register()
8943 ret = pret < 0 ? pret : -EFAULT; in io_sqe_buffer_register()
8945 mmap_read_unlock(current->mm); in io_sqe_buffer_register()
8963 size = iov->iov_len; in io_sqe_buffer_register()
8967 vec_len = min_t(size_t, size, PAGE_SIZE - off); in io_sqe_buffer_register()
8968 imu->bvec[i].bv_page = pages[i]; in io_sqe_buffer_register()
8969 imu->bvec[i].bv_len = vec_len; in io_sqe_buffer_register()
8970 imu->bvec[i].bv_offset = off; in io_sqe_buffer_register()
8972 size -= vec_len; in io_sqe_buffer_register()
8975 imu->ubuf = ubuf; in io_sqe_buffer_register()
8976 imu->ubuf_end = ubuf + iov->iov_len; in io_sqe_buffer_register()
8977 imu->nr_bvecs = nr_pages; in io_sqe_buffer_register()
8990 ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL); in io_buffers_map_alloc()
8991 return ctx->user_bufs ? 0 : -ENOMEM; in io_buffers_map_alloc()
8996 unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1); in io_buffer_validate()
9000 * constraints here, we'll -EINVAL later when IO is in io_buffer_validate()
9003 if (!iov->iov_base) in io_buffer_validate()
9004 return iov->iov_len ? -EFAULT : 0; in io_buffer_validate()
9005 if (!iov->iov_len) in io_buffer_validate()
9006 return -EFAULT; in io_buffer_validate()
9009 if (iov->iov_len > SZ_1G) in io_buffer_validate()
9010 return -EFAULT; in io_buffer_validate()
9012 if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp)) in io_buffer_validate()
9013 return -EOVERFLOW; in io_buffer_validate()
9026 if (ctx->user_bufs) in io_sqe_buffers_register()
9027 return -EBUSY; in io_sqe_buffers_register()
9029 return -EINVAL; in io_sqe_buffers_register()
9042 for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) { in io_sqe_buffers_register()
9050 ret = -EINVAL; in io_sqe_buffers_register()
9054 ret = io_sqe_buffer_register(ctx, &iov, &ctx->user_bufs[i], in io_sqe_buffers_register()
9060 WARN_ON_ONCE(ctx->buf_data); in io_sqe_buffers_register()
9062 ctx->buf_data = data; in io_sqe_buffers_register()
9074 u64 __user *tags = u64_to_user_ptr(up->tags); in __io_sqe_buffers_update()
9075 struct iovec iov, __user *iovs = u64_to_user_ptr(up->data); in __io_sqe_buffers_update()
9081 if (!ctx->buf_data) in __io_sqe_buffers_update()
9082 return -ENXIO; in __io_sqe_buffers_update()
9083 if (up->offset + nr_args > ctx->nr_user_bufs) in __io_sqe_buffers_update()
9084 return -EINVAL; in __io_sqe_buffers_update()
9088 int offset = up->offset + done; in __io_sqe_buffers_update()
9095 err = -EFAULT; in __io_sqe_buffers_update()
9102 err = -EINVAL; in __io_sqe_buffers_update()
9109 i = array_index_nospec(offset, ctx->nr_user_bufs); in __io_sqe_buffers_update()
9110 if (ctx->user_bufs[i] != ctx->dummy_ubuf) { in __io_sqe_buffers_update()
9111 err = io_queue_rsrc_removal(ctx->buf_data, i, in __io_sqe_buffers_update()
9112 ctx->rsrc_node, ctx->user_bufs[i]); in __io_sqe_buffers_update()
9117 ctx->user_bufs[i] = NULL; in __io_sqe_buffers_update()
9121 ctx->user_bufs[i] = imu; in __io_sqe_buffers_update()
9122 *io_get_tag_slot(ctx->buf_data, offset) = tag; in __io_sqe_buffers_update()
9126 io_rsrc_node_switch(ctx, ctx->buf_data); in __io_sqe_buffers_update()
9135 if (ctx->cq_ev_fd) in io_eventfd_register()
9136 return -EBUSY; in io_eventfd_register()
9139 return -EFAULT; in io_eventfd_register()
9141 ctx->cq_ev_fd = eventfd_ctx_fdget(fd); in io_eventfd_register()
9142 if (IS_ERR(ctx->cq_ev_fd)) { in io_eventfd_register()
9143 int ret = PTR_ERR(ctx->cq_ev_fd); in io_eventfd_register()
9145 ctx->cq_ev_fd = NULL; in io_eventfd_register()
9154 if (ctx->cq_ev_fd) { in io_eventfd_unregister()
9155 eventfd_ctx_put(ctx->cq_ev_fd); in io_eventfd_unregister()
9156 ctx->cq_ev_fd = NULL; in io_eventfd_unregister()
9160 return -ENXIO; in io_eventfd_unregister()
9168 xa_for_each(&ctx->io_buffers, index, buf) in io_destroy_buffers()
9169 __io_remove_buffers(ctx, buf, index, -1U); in io_destroy_buffers()
9177 list_del(&req->inflight_entry); in io_req_cache_free()
9184 struct io_submit_state *state = &ctx->submit_state; in io_req_caches_free()
9186 mutex_lock(&ctx->uring_lock); in io_req_caches_free()
9188 if (state->free_reqs) { in io_req_caches_free()
9189 kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs); in io_req_caches_free()
9190 state->free_reqs = 0; in io_req_caches_free()
9194 io_req_cache_free(&state->free_list); in io_req_caches_free()
9195 mutex_unlock(&ctx->uring_lock); in io_req_caches_free()
9200 if (data && !atomic_dec_and_test(&data->refs)) in io_wait_rsrc_data()
9201 wait_for_completion(&data->done); in io_wait_rsrc_data()
9209 io_wait_rsrc_data(ctx->buf_data); in io_ring_ctx_free()
9210 io_wait_rsrc_data(ctx->file_data); in io_ring_ctx_free()
9212 mutex_lock(&ctx->uring_lock); in io_ring_ctx_free()
9213 if (ctx->buf_data) in io_ring_ctx_free()
9215 if (ctx->file_data) in io_ring_ctx_free()
9217 if (ctx->rings) in io_ring_ctx_free()
9219 mutex_unlock(&ctx->uring_lock); in io_ring_ctx_free()
9222 if (ctx->sq_creds) in io_ring_ctx_free()
9223 put_cred(ctx->sq_creds); in io_ring_ctx_free()
9226 if (ctx->rsrc_node) in io_ring_ctx_free()
9227 io_rsrc_node_destroy(ctx->rsrc_node); in io_ring_ctx_free()
9228 if (ctx->rsrc_backup_node) in io_ring_ctx_free()
9229 io_rsrc_node_destroy(ctx->rsrc_backup_node); in io_ring_ctx_free()
9230 flush_delayed_work(&ctx->rsrc_put_work); in io_ring_ctx_free()
9232 WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list)); in io_ring_ctx_free()
9233 WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist)); in io_ring_ctx_free()
9236 if (ctx->ring_sock) { in io_ring_ctx_free()
9237 ctx->ring_sock->file = NULL; /* so that iput() is called */ in io_ring_ctx_free()
9238 sock_release(ctx->ring_sock); in io_ring_ctx_free()
9241 WARN_ON_ONCE(!list_empty(&ctx->ltimeout_list)); in io_ring_ctx_free()
9243 if (ctx->mm_account) { in io_ring_ctx_free()
9244 mmdrop(ctx->mm_account); in io_ring_ctx_free()
9245 ctx->mm_account = NULL; in io_ring_ctx_free()
9248 io_mem_free(ctx->rings); in io_ring_ctx_free()
9249 io_mem_free(ctx->sq_sqes); in io_ring_ctx_free()
9251 percpu_ref_exit(&ctx->refs); in io_ring_ctx_free()
9252 free_uid(ctx->user); in io_ring_ctx_free()
9254 if (ctx->hash_map) in io_ring_ctx_free()
9255 io_wq_put_hash(ctx->hash_map); in io_ring_ctx_free()
9256 kfree(ctx->cancel_hash); in io_ring_ctx_free()
9257 kfree(ctx->dummy_ubuf); in io_ring_ctx_free()
9263 struct io_ring_ctx *ctx = file->private_data; in io_uring_poll()
9266 poll_wait(file, &ctx->poll_wait, wait); in io_uring_poll()
9279 * ---- ---- in io_uring_poll()
9280 * lock(&ctx->uring_lock); in io_uring_poll()
9281 * lock(&ep->mtx); in io_uring_poll()
9282 * lock(&ctx->uring_lock); in io_uring_poll()
9283 * lock(&ep->mtx); in io_uring_poll()
9288 if (io_cqring_events(ctx) || test_bit(0, &ctx->check_cq_overflow)) in io_uring_poll()
9298 creds = xa_erase(&ctx->personalities, id); in io_unregister_personality()
9304 return -EINVAL; in io_unregister_personality()
9315 struct io_uring_task *tctx = current->io_uring; in io_tctx_exit_cb()
9325 if (tctx && !atomic_read(&tctx->in_idle)) in io_tctx_exit_cb()
9326 io_uring_del_tctx_node((unsigned long)work->ctx); in io_tctx_exit_cb()
9327 complete(&work->completion); in io_tctx_exit_cb()
9334 return req->ctx == data; in io_cancel_ctx_cb()
9348 * submitted async (out-of-line), then completions can come in while in io_ring_exit_work()
9354 if (ctx->sq_data) { in io_ring_exit_work()
9355 struct io_sq_data *sqd = ctx->sq_data; in io_ring_exit_work()
9359 tsk = sqd->thread; in io_ring_exit_work()
9360 if (tsk && tsk->io_uring && tsk->io_uring->io_wq) in io_ring_exit_work()
9361 io_wq_cancel_cb(tsk->io_uring->io_wq, in io_ring_exit_work()
9370 } while (!wait_for_completion_timeout(&ctx->ref_comp, interval)); in io_ring_exit_work()
9381 mutex_lock(&ctx->uring_lock); in io_ring_exit_work()
9382 while (!list_empty(&ctx->tctx_list)) { in io_ring_exit_work()
9385 node = list_first_entry(&ctx->tctx_list, struct io_tctx_node, in io_ring_exit_work()
9388 list_rotate_left(&ctx->tctx_list); in io_ring_exit_work()
9389 ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL); in io_ring_exit_work()
9392 wake_up_process(node->task); in io_ring_exit_work()
9394 mutex_unlock(&ctx->uring_lock); in io_ring_exit_work()
9396 mutex_lock(&ctx->uring_lock); in io_ring_exit_work()
9398 mutex_unlock(&ctx->uring_lock); in io_ring_exit_work()
9399 spin_lock(&ctx->completion_lock); in io_ring_exit_work()
9400 spin_unlock(&ctx->completion_lock); in io_ring_exit_work()
9412 spin_lock(&ctx->completion_lock); in io_kill_timeouts()
9413 spin_lock_irq(&ctx->timeout_lock); in io_kill_timeouts()
9414 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) { in io_kill_timeouts()
9416 io_kill_timeout(req, -ECANCELED); in io_kill_timeouts()
9420 spin_unlock_irq(&ctx->timeout_lock); in io_kill_timeouts()
9423 spin_unlock(&ctx->completion_lock); in io_kill_timeouts()
9434 mutex_lock(&ctx->uring_lock); in io_ring_ctx_wait_and_kill()
9435 percpu_ref_kill(&ctx->refs); in io_ring_ctx_wait_and_kill()
9436 if (ctx->rings) in io_ring_ctx_wait_and_kill()
9438 xa_for_each(&ctx->personalities, index, creds) in io_ring_ctx_wait_and_kill()
9440 mutex_unlock(&ctx->uring_lock); in io_ring_ctx_wait_and_kill()
9448 INIT_WORK(&ctx->exit_work, io_ring_exit_work); in io_ring_ctx_wait_and_kill()
9455 queue_work(system_unbound_wq, &ctx->exit_work); in io_ring_ctx_wait_and_kill()
9460 struct io_ring_ctx *ctx = file->private_data; in io_uring_release()
9462 file->private_data = NULL; in io_uring_release()
9477 return io_match_task_safe(req, cancel->task, cancel->all); in io_cancel_task_cb()
9486 spin_lock(&ctx->completion_lock); in io_cancel_defer_files()
9487 list_for_each_entry_reverse(de, &ctx->defer_list, list) { in io_cancel_defer_files()
9488 if (io_match_task_safe(de->req, task, cancel_all)) { in io_cancel_defer_files()
9489 list_cut_position(&list, &ctx->defer_list, &de->list); in io_cancel_defer_files()
9493 spin_unlock(&ctx->completion_lock); in io_cancel_defer_files()
9499 list_del_init(&de->list); in io_cancel_defer_files()
9500 io_req_complete_failed(de->req, -ECANCELED); in io_cancel_defer_files()
9512 mutex_lock(&ctx->uring_lock); in io_uring_try_cancel_iowq()
9513 list_for_each_entry(node, &ctx->tctx_list, ctx_node) { in io_uring_try_cancel_iowq()
9514 struct io_uring_task *tctx = node->task->io_uring; in io_uring_try_cancel_iowq()
9520 if (!tctx || !tctx->io_wq) in io_uring_try_cancel_iowq()
9522 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true); in io_uring_try_cancel_iowq()
9525 mutex_unlock(&ctx->uring_lock); in io_uring_try_cancel_iowq()
9535 struct io_uring_task *tctx = task ? task->io_uring : NULL; in io_uring_try_cancel_requests()
9543 } else if (tctx && tctx->io_wq) { in io_uring_try_cancel_requests()
9548 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb, in io_uring_try_cancel_requests()
9554 if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) || in io_uring_try_cancel_requests()
9555 (ctx->sq_data && ctx->sq_data->thread == current)) { in io_uring_try_cancel_requests()
9556 while (!list_empty_careful(&ctx->iopoll_list)) { in io_uring_try_cancel_requests()
9575 struct io_uring_task *tctx = current->io_uring; in __io_uring_add_tctx_node()
9584 tctx = current->io_uring; in __io_uring_add_tctx_node()
9585 if (ctx->iowq_limits_set) { in __io_uring_add_tctx_node()
9586 unsigned int limits[2] = { ctx->iowq_limits[0], in __io_uring_add_tctx_node()
9587 ctx->iowq_limits[1], }; in __io_uring_add_tctx_node()
9589 ret = io_wq_max_workers(tctx->io_wq, limits); in __io_uring_add_tctx_node()
9594 if (!xa_load(&tctx->xa, (unsigned long)ctx)) { in __io_uring_add_tctx_node()
9597 return -ENOMEM; in __io_uring_add_tctx_node()
9598 node->ctx = ctx; in __io_uring_add_tctx_node()
9599 node->task = current; in __io_uring_add_tctx_node()
9601 ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx, in __io_uring_add_tctx_node()
9608 mutex_lock(&ctx->uring_lock); in __io_uring_add_tctx_node()
9609 list_add(&node->ctx_node, &ctx->tctx_list); in __io_uring_add_tctx_node()
9610 mutex_unlock(&ctx->uring_lock); in __io_uring_add_tctx_node()
9612 tctx->last = ctx; in __io_uring_add_tctx_node()
9621 struct io_uring_task *tctx = current->io_uring; in io_uring_add_tctx_node()
9623 if (likely(tctx && tctx->last == ctx)) in io_uring_add_tctx_node()
9629 * Remove this io_uring_file -> task mapping.
9633 struct io_uring_task *tctx = current->io_uring; in io_uring_del_tctx_node()
9638 node = xa_erase(&tctx->xa, index); in io_uring_del_tctx_node()
9642 WARN_ON_ONCE(current != node->task); in io_uring_del_tctx_node()
9643 WARN_ON_ONCE(list_empty(&node->ctx_node)); in io_uring_del_tctx_node()
9645 mutex_lock(&node->ctx->uring_lock); in io_uring_del_tctx_node()
9646 list_del(&node->ctx_node); in io_uring_del_tctx_node()
9647 mutex_unlock(&node->ctx->uring_lock); in io_uring_del_tctx_node()
9649 if (tctx->last == node->ctx) in io_uring_del_tctx_node()
9650 tctx->last = NULL; in io_uring_del_tctx_node()
9656 struct io_wq *wq = tctx->io_wq; in io_uring_clean_tctx()
9660 xa_for_each(&tctx->xa, index, node) { in io_uring_clean_tctx()
9670 tctx->io_wq = NULL; in io_uring_clean_tctx()
9677 return atomic_read(&tctx->inflight_tracked); in tctx_inflight()
9678 return percpu_counter_sum(&tctx->inflight); in tctx_inflight()
9683 * requests. @sqd should be not-null IFF it's an SQPOLL thread cancellation.
9687 struct io_uring_task *tctx = current->io_uring; in io_uring_cancel_generic()
9692 WARN_ON_ONCE(sqd && sqd->thread != current); in io_uring_cancel_generic()
9694 if (!current->io_uring) in io_uring_cancel_generic()
9696 if (tctx->io_wq) in io_uring_cancel_generic()
9697 io_wq_exit_start(tctx->io_wq); in io_uring_cancel_generic()
9699 atomic_inc(&tctx->in_idle); in io_uring_cancel_generic()
9711 xa_for_each(&tctx->xa, index, node) { in io_uring_cancel_generic()
9713 if (node->ctx->sq_data) in io_uring_cancel_generic()
9715 io_uring_try_cancel_requests(node->ctx, current, in io_uring_cancel_generic()
9719 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) in io_uring_cancel_generic()
9724 prepare_to_wait(&tctx->wait, &wait, TASK_INTERRUPTIBLE); in io_uring_cancel_generic()
9729 * If we've seen completions, retry without waiting. This in io_uring_cancel_generic()
9735 finish_wait(&tctx->wait, &wait); in io_uring_cancel_generic()
9742 * ->in_idle set for normal exit. in io_uring_cancel_generic()
9744 atomic_dec(&tctx->in_idle); in io_uring_cancel_generic()
9758 struct io_ring_ctx *ctx = file->private_data; in io_uring_validate_mmap_request()
9766 ptr = ctx->rings; in io_uring_validate_mmap_request()
9769 ptr = ctx->sq_sqes; in io_uring_validate_mmap_request()
9772 return ERR_PTR(-EINVAL); in io_uring_validate_mmap_request()
9777 return ERR_PTR(-EINVAL); in io_uring_validate_mmap_request()
9786 size_t sz = vma->vm_end - vma->vm_start; in io_uring_mmap()
9790 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz); in io_uring_mmap()
9795 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot); in io_uring_mmap()
9802 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL; in io_uring_mmap()
9832 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE); in io_sqpoll_wait_sq()
9839 finish_wait(&ctx->sqo_sq_wait, &wait); in io_sqpoll_wait_sq()
9860 * EXT_ARG is set - ensure we agree on the size of it and copy in our in io_get_ext_arg()
9864 return -EINVAL; in io_get_ext_arg()
9866 return -EFAULT; in io_get_ext_arg()
9868 return -EINVAL; in io_get_ext_arg()
9888 return -EINVAL; in SYSCALL_DEFINE6()
9892 return -EBADF; in SYSCALL_DEFINE6()
9894 ret = -EOPNOTSUPP; in SYSCALL_DEFINE6()
9895 if (unlikely(f.file->f_op != &io_uring_fops)) in SYSCALL_DEFINE6()
9898 ret = -ENXIO; in SYSCALL_DEFINE6()
9899 ctx = f.file->private_data; in SYSCALL_DEFINE6()
9900 if (unlikely(!percpu_ref_tryget(&ctx->refs))) in SYSCALL_DEFINE6()
9903 ret = -EBADFD; in SYSCALL_DEFINE6()
9904 if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED)) in SYSCALL_DEFINE6()
9909 * Just return the requested submit count, and wake the thread if in SYSCALL_DEFINE6()
9913 if (ctx->flags & IORING_SETUP_SQPOLL) { in SYSCALL_DEFINE6()
9916 if (unlikely(ctx->sq_data->thread == NULL)) { in SYSCALL_DEFINE6()
9917 ret = -EOWNERDEAD; in SYSCALL_DEFINE6()
9921 wake_up(&ctx->sq_data->wait); in SYSCALL_DEFINE6()
9932 mutex_lock(&ctx->uring_lock); in SYSCALL_DEFINE6()
9934 mutex_unlock(&ctx->uring_lock); in SYSCALL_DEFINE6()
9947 min_complete = min(min_complete, ctx->cq_entries); in SYSCALL_DEFINE6()
9955 if (ctx->flags & IORING_SETUP_IOPOLL && in SYSCALL_DEFINE6()
9956 !(ctx->flags & IORING_SETUP_SQPOLL)) { in SYSCALL_DEFINE6()
9964 percpu_ref_put(&ctx->refs); in SYSCALL_DEFINE6()
9981 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid)); in io_uring_show_cred()
9982 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid)); in io_uring_show_cred()
9983 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid)); in io_uring_show_cred()
9984 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid)); in io_uring_show_cred()
9985 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid)); in io_uring_show_cred()
9986 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid)); in io_uring_show_cred()
9987 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid)); in io_uring_show_cred()
9988 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid)); in io_uring_show_cred()
9990 gi = cred->group_info; in io_uring_show_cred()
9991 for (g = 0; g < gi->ngroups; g++) { in io_uring_show_cred()
9993 from_kgid_munged(uns, gi->gid[g])); in io_uring_show_cred()
9996 cap = cred->cap_effective; in io_uring_show_cred()
9998 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8); in io_uring_show_cred()
10015 has_lock = mutex_trylock(&ctx->uring_lock); in __io_uring_show_fdinfo()
10017 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) { in __io_uring_show_fdinfo()
10018 sq = ctx->sq_data; in __io_uring_show_fdinfo()
10019 if (!sq->thread) in __io_uring_show_fdinfo()
10023 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1); in __io_uring_show_fdinfo()
10024 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1); in __io_uring_show_fdinfo()
10025 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files); in __io_uring_show_fdinfo()
10026 for (i = 0; has_lock && i < ctx->nr_user_files; i++) { in __io_uring_show_fdinfo()
10030 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname); in __io_uring_show_fdinfo()
10034 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs); in __io_uring_show_fdinfo()
10035 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) { in __io_uring_show_fdinfo()
10036 struct io_mapped_ubuf *buf = ctx->user_bufs[i]; in __io_uring_show_fdinfo()
10037 unsigned int len = buf->ubuf_end - buf->ubuf; in __io_uring_show_fdinfo()
10039 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len); in __io_uring_show_fdinfo()
10041 if (has_lock && !xa_empty(&ctx->personalities)) { in __io_uring_show_fdinfo()
10046 xa_for_each(&ctx->personalities, index, cred) in __io_uring_show_fdinfo()
10050 spin_lock(&ctx->completion_lock); in __io_uring_show_fdinfo()
10051 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) { in __io_uring_show_fdinfo()
10052 struct hlist_head *list = &ctx->cancel_hash[i]; in __io_uring_show_fdinfo()
10056 seq_printf(m, " op=%d, task_works=%d\n", req->opcode, in __io_uring_show_fdinfo()
10057 req->task->task_works != NULL); in __io_uring_show_fdinfo()
10059 spin_unlock(&ctx->completion_lock); in __io_uring_show_fdinfo()
10061 mutex_unlock(&ctx->uring_lock); in __io_uring_show_fdinfo()
10066 struct io_ring_ctx *ctx = f->private_data; in io_uring_show_fdinfo()
10068 if (percpu_ref_tryget(&ctx->refs)) { in io_uring_show_fdinfo()
10070 percpu_ref_put(&ctx->refs); in io_uring_show_fdinfo()
10082 .poll = io_uring_poll,
10095 ctx->sq_entries = p->sq_entries; in io_allocate_scq_urings()
10096 ctx->cq_entries = p->cq_entries; in io_allocate_scq_urings()
10098 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset); in io_allocate_scq_urings()
10100 return -EOVERFLOW; in io_allocate_scq_urings()
10104 return -ENOMEM; in io_allocate_scq_urings()
10106 ctx->rings = rings; in io_allocate_scq_urings()
10107 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset); in io_allocate_scq_urings()
10108 rings->sq_ring_mask = p->sq_entries - 1; in io_allocate_scq_urings()
10109 rings->cq_ring_mask = p->cq_entries - 1; in io_allocate_scq_urings()
10110 rings->sq_ring_entries = p->sq_entries; in io_allocate_scq_urings()
10111 rings->cq_ring_entries = p->cq_entries; in io_allocate_scq_urings()
10113 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries); in io_allocate_scq_urings()
10115 io_mem_free(ctx->rings); in io_allocate_scq_urings()
10116 ctx->rings = NULL; in io_allocate_scq_urings()
10117 return -EOVERFLOW; in io_allocate_scq_urings()
10120 ctx->sq_sqes = io_mem_alloc(size); in io_allocate_scq_urings()
10121 if (!ctx->sq_sqes) { in io_allocate_scq_urings()
10122 io_mem_free(ctx->rings); in io_allocate_scq_urings()
10123 ctx->rings = NULL; in io_allocate_scq_urings()
10124 return -ENOMEM; in io_allocate_scq_urings()
10160 &ctx->ring_sock); in io_uring_get_file()
10169 sock_release(ctx->ring_sock); in io_uring_get_file()
10170 ctx->ring_sock = NULL; in io_uring_get_file()
10172 ctx->ring_sock->file = file; in io_uring_get_file()
10186 return -EINVAL; in io_uring_create()
10188 if (!(p->flags & IORING_SETUP_CLAMP)) in io_uring_create()
10189 return -EINVAL; in io_uring_create()
10201 p->sq_entries = roundup_pow_of_two(entries); in io_uring_create()
10202 if (p->flags & IORING_SETUP_CQSIZE) { in io_uring_create()
10205 * to a power-of-two, if it isn't already. We do NOT impose in io_uring_create()
10208 if (!p->cq_entries) in io_uring_create()
10209 return -EINVAL; in io_uring_create()
10210 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) { in io_uring_create()
10211 if (!(p->flags & IORING_SETUP_CLAMP)) in io_uring_create()
10212 return -EINVAL; in io_uring_create()
10213 p->cq_entries = IORING_MAX_CQ_ENTRIES; in io_uring_create()
10215 p->cq_entries = roundup_pow_of_two(p->cq_entries); in io_uring_create()
10216 if (p->cq_entries < p->sq_entries) in io_uring_create()
10217 return -EINVAL; in io_uring_create()
10219 p->cq_entries = 2 * p->sq_entries; in io_uring_create()
10224 return -ENOMEM; in io_uring_create()
10225 ctx->compat = in_compat_syscall(); in io_uring_create()
10227 ctx->user = get_uid(current_user()); in io_uring_create()
10235 mmgrab(current->mm); in io_uring_create()
10236 ctx->mm_account = current->mm; in io_uring_create()
10251 memset(&p->sq_off, 0, sizeof(p->sq_off)); in io_uring_create()
10252 p->sq_off.head = offsetof(struct io_rings, sq.head); in io_uring_create()
10253 p->sq_off.tail = offsetof(struct io_rings, sq.tail); in io_uring_create()
10254 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask); in io_uring_create()
10255 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries); in io_uring_create()
10256 p->sq_off.flags = offsetof(struct io_rings, sq_flags); in io_uring_create()
10257 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped); in io_uring_create()
10258 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings; in io_uring_create()
10260 memset(&p->cq_off, 0, sizeof(p->cq_off)); in io_uring_create()
10261 p->cq_off.head = offsetof(struct io_rings, cq.head); in io_uring_create()
10262 p->cq_off.tail = offsetof(struct io_rings, cq.tail); in io_uring_create()
10263 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask); in io_uring_create()
10264 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries); in io_uring_create()
10265 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow); in io_uring_create()
10266 p->cq_off.cqes = offsetof(struct io_rings, cqes); in io_uring_create()
10267 p->cq_off.flags = offsetof(struct io_rings, cq_flags); in io_uring_create()
10269 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP | in io_uring_create()
10277 ret = -EFAULT; in io_uring_create()
10298 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags); in io_uring_create()
10316 return -EFAULT; in io_uring_setup()
10319 return -EINVAL; in io_uring_setup()
10326 return -EINVAL; in io_uring_setup()
10345 return -EOVERFLOW; in io_probe()
10348 return -ENOMEM; in io_probe()
10350 ret = -EFAULT; in io_probe()
10353 ret = -EINVAL; in io_probe()
10357 p->last_op = IORING_OP_LAST - 1; in io_probe()
10362 p->ops[i].op = i; in io_probe()
10364 p->ops[i].flags = IO_URING_OP_SUPPORTED; in io_probe()
10366 p->ops_len = i; in io_probe()
10370 ret = -EFAULT; in io_probe()
10384 ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds, in io_register_personality()
10385 XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL); in io_register_personality()
10401 if (!(ctx->flags & IORING_SETUP_R_DISABLED)) in io_register_restrictions()
10402 return -EBADFD; in io_register_restrictions()
10405 if (ctx->restrictions.registered) in io_register_restrictions()
10406 return -EBUSY; in io_register_restrictions()
10409 return -EINVAL; in io_register_restrictions()
10413 return -EOVERFLOW; in io_register_restrictions()
10425 ret = -EINVAL; in io_register_restrictions()
10430 ctx->restrictions.register_op); in io_register_restrictions()
10434 ret = -EINVAL; in io_register_restrictions()
10438 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op); in io_register_restrictions()
10441 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags; in io_register_restrictions()
10444 ctx->restrictions.sqe_flags_required = res[i].sqe_flags; in io_register_restrictions()
10447 ret = -EINVAL; in io_register_restrictions()
10455 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions)); in io_register_restrictions()
10457 ctx->restrictions.registered = true; in io_register_restrictions()
10465 if (!(ctx->flags & IORING_SETUP_R_DISABLED)) in io_register_enable_rings()
10466 return -EBADFD; in io_register_enable_rings()
10468 if (ctx->restrictions.registered) in io_register_enable_rings()
10469 ctx->restricted = 1; in io_register_enable_rings()
10471 ctx->flags &= ~IORING_SETUP_R_DISABLED; in io_register_enable_rings()
10472 if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait)) in io_register_enable_rings()
10473 wake_up(&ctx->sq_data->wait); in io_register_enable_rings()
10484 if (check_add_overflow(up->offset, nr_args, &tmp)) in __io_register_rsrc_update()
10485 return -EOVERFLOW; in __io_register_rsrc_update()
10496 return -EINVAL; in __io_register_rsrc_update()
10505 return -EINVAL; in io_register_files_update()
10508 return -EFAULT; in io_register_files_update()
10510 return -EINVAL; in io_register_files_update()
10520 return -EINVAL; in io_register_rsrc_update()
10522 return -EFAULT; in io_register_rsrc_update()
10524 return -EINVAL; in io_register_rsrc_update()
10535 return -EINVAL; in io_register_rsrc()
10539 return -EFAULT; in io_register_rsrc()
10541 return -EINVAL; in io_register_rsrc()
10551 return -EINVAL; in io_register_rsrc()
10557 struct io_uring_task *tctx = current->io_uring; in io_register_iowq_aff()
10561 if (!tctx || !tctx->io_wq) in io_register_iowq_aff()
10562 return -EINVAL; in io_register_iowq_aff()
10565 return -ENOMEM; in io_register_iowq_aff()
10585 return -EFAULT; in io_register_iowq_aff()
10588 ret = io_wq_cpu_affinity(tctx->io_wq, new_mask); in io_register_iowq_aff()
10595 struct io_uring_task *tctx = current->io_uring; in io_unregister_iowq_aff()
10597 if (!tctx || !tctx->io_wq) in io_unregister_iowq_aff()
10598 return -EINVAL; in io_unregister_iowq_aff()
10600 return io_wq_cpu_affinity(tctx->io_wq, NULL); in io_unregister_iowq_aff()
10605 __must_hold(&ctx->uring_lock) in io_register_iowq_max_workers()
10614 return -EFAULT; in io_register_iowq_max_workers()
10617 return -EINVAL; in io_register_iowq_max_workers()
10619 if (ctx->flags & IORING_SETUP_SQPOLL) { in io_register_iowq_max_workers()
10620 sqd = ctx->sq_data; in io_register_iowq_max_workers()
10623 * Observe the correct sqd->lock -> ctx->uring_lock in io_register_iowq_max_workers()
10627 refcount_inc(&sqd->refs); in io_register_iowq_max_workers()
10628 mutex_unlock(&ctx->uring_lock); in io_register_iowq_max_workers()
10629 mutex_lock(&sqd->lock); in io_register_iowq_max_workers()
10630 mutex_lock(&ctx->uring_lock); in io_register_iowq_max_workers()
10631 if (sqd->thread) in io_register_iowq_max_workers()
10632 tctx = sqd->thread->io_uring; in io_register_iowq_max_workers()
10635 tctx = current->io_uring; in io_register_iowq_max_workers()
10638 BUILD_BUG_ON(sizeof(new_count) != sizeof(ctx->iowq_limits)); in io_register_iowq_max_workers()
10642 ctx->iowq_limits[i] = new_count[i]; in io_register_iowq_max_workers()
10643 ctx->iowq_limits_set = true; in io_register_iowq_max_workers()
10645 ret = -EINVAL; in io_register_iowq_max_workers()
10646 if (tctx && tctx->io_wq) { in io_register_iowq_max_workers()
10647 ret = io_wq_max_workers(tctx->io_wq, new_count); in io_register_iowq_max_workers()
10655 mutex_unlock(&sqd->lock); in io_register_iowq_max_workers()
10660 return -EFAULT; in io_register_iowq_max_workers()
10667 list_for_each_entry(node, &ctx->tctx_list, ctx_node) { in io_register_iowq_max_workers()
10668 struct io_uring_task *tctx = node->task->io_uring; in io_register_iowq_max_workers()
10670 if (WARN_ON_ONCE(!tctx->io_wq)) in io_register_iowq_max_workers()
10674 new_count[i] = ctx->iowq_limits[i]; in io_register_iowq_max_workers()
10676 (void)io_wq_max_workers(tctx->io_wq, new_count); in io_register_iowq_max_workers()
10681 mutex_unlock(&sqd->lock); in io_register_iowq_max_workers()
10715 percpu_ref_kill(&ctx->refs); in io_ctx_quiesce()
10724 mutex_unlock(&ctx->uring_lock); in io_ctx_quiesce()
10726 ret = wait_for_completion_interruptible(&ctx->ref_comp); in io_ctx_quiesce()
10731 mutex_lock(&ctx->uring_lock); in io_ctx_quiesce()
10734 io_refs_resurrect(&ctx->refs, &ctx->ref_comp); in io_ctx_quiesce()
10740 __releases(ctx->uring_lock) in __io_uring_register()
10741 __acquires(ctx->uring_lock) in __io_uring_register()
10750 if (percpu_ref_is_dying(&ctx->refs)) in __io_uring_register()
10751 return -ENXIO; in __io_uring_register()
10753 if (ctx->restricted) { in __io_uring_register()
10755 return -EINVAL; in __io_uring_register()
10757 if (!test_bit(opcode, ctx->restrictions.register_op)) in __io_uring_register()
10758 return -EACCES; in __io_uring_register()
10772 ret = -EINVAL; in __io_uring_register()
10781 ret = -EINVAL; in __io_uring_register()
10791 ret = -EINVAL; in __io_uring_register()
10798 ctx->eventfd_async = 1; in __io_uring_register()
10800 ctx->eventfd_async = 0; in __io_uring_register()
10803 ret = -EINVAL; in __io_uring_register()
10809 ret = -EINVAL; in __io_uring_register()
10815 ret = -EINVAL; in __io_uring_register()
10821 ret = -EINVAL; in __io_uring_register()
10827 ret = -EINVAL; in __io_uring_register()
10850 ret = -EINVAL; in __io_uring_register()
10856 ret = -EINVAL; in __io_uring_register()
10862 ret = -EINVAL; in __io_uring_register()
10868 ret = -EINVAL; in __io_uring_register()
10874 percpu_ref_reinit(&ctx->refs); in __io_uring_register()
10875 reinit_completion(&ctx->ref_comp); in __io_uring_register()
10884 long ret = -EBADF; in SYSCALL_DEFINE4()
10889 return -EBADF; in SYSCALL_DEFINE4()
10891 ret = -EOPNOTSUPP; in SYSCALL_DEFINE4()
10892 if (f.file->f_op != &io_uring_fops) in SYSCALL_DEFINE4()
10895 ctx = f.file->private_data; in SYSCALL_DEFINE4()
10899 mutex_lock(&ctx->uring_lock); in SYSCALL_DEFINE4()
10901 mutex_unlock(&ctx->uring_lock); in SYSCALL_DEFINE4()
10902 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs, in SYSCALL_DEFINE4()
10903 ctx->cq_ev_fd != NULL, ret); in SYSCALL_DEFINE4()
10955 /* ->buf_index is u16 */ in io_uring_init()