Merge tag 'io_uring-5.11-2021-01-10' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "A bit larger than I had hoped at this point, but it's all changes that
  will be directed towards stable anyway. In detail:

   - Fix a merge window regression on error return (Matthew)

   - Remove useless variable declaration/assignment (Ye Bin)

   - IOPOLL fixes (Pavel)

   - Exit and cancelation fixes (Pavel)

   - fasync lockdep complaint fix (Pavel)

   - Ensure SQPOLL is synchronized with creator life time (Pavel)"

* tag 'io_uring-5.11-2021-01-10' of git://git.kernel.dk/linux-block:
  io_uring: stop SQPOLL submit on creator's death
  io_uring: add warn_once for io_uring_flush()
  io_uring: inline io_uring_attempt_task_drop()
  io_uring: io_rw_reissue lockdep annotations
  io_uring: synchronise ev_posted() with waitqueues
  io_uring: dont kill fasync under completion_lock
  io_uring: trigger eventfd for IOPOLL
  io_uring: Fix return value from alloc_fixed_file_ref_node
  io_uring: Delete useless variable ‘id’ in io_prep_async_work
  io_uring: cancel more aggressively in exit_work
  io_uring: drop file refs after task cancel
  io_uring: patch up IOPOLL overflow_flush sync
  io_uring: synchronise IOPOLL on task_submit fail

fs/io_uring.c

@@ -262,6 +262,7 @@ struct io_ring_ctx {
 		unsigned int		drain_next: 1;
 		unsigned int		eventfd_async: 1;
 		unsigned int		restricted: 1;
+		unsigned int		sqo_dead: 1;
 
 		/*
 		 * Ring buffer of indices into array of io_uring_sqe, which is
@@ -992,6 +993,9 @@ enum io_mem_account {
 	ACCT_PINNED,
 };
 
+static void __io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
+					    struct task_struct *task);
+
 static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node);
 static struct fixed_file_ref_node *alloc_fixed_file_ref_node(
 			struct io_ring_ctx *ctx);
@@ -1342,11 +1346,6 @@ static void __io_commit_cqring(struct io_ring_ctx *ctx)
 
 	/* order cqe stores with ring update */
 	smp_store_release(&rings->cq.tail, ctx->cached_cq_tail);
-
-	if (wq_has_sleeper(&ctx->cq_wait)) {
-		wake_up_interruptible(&ctx->cq_wait);
-		kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
-	}
 }
 
 static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req)
@@ -1520,10 +1519,8 @@ static void io_prep_async_work(struct io_kiocb *req)
 {
 	const struct io_op_def *def = &io_op_defs[req->opcode];
 	struct io_ring_ctx *ctx = req->ctx;
-	struct io_identity *id;
 
 	io_req_init_async(req);
-	id = req->work.identity;
 
 	if (req->flags & REQ_F_FORCE_ASYNC)
 		req->work.flags |= IO_WQ_WORK_CONCURRENT;
@@ -1704,16 +1701,40 @@ static inline unsigned __io_cqring_events(struct io_ring_ctx *ctx)
 
 static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
 {
+	/* see waitqueue_active() comment */
+	smp_mb();
+
 	if (waitqueue_active(&ctx->wait))
 		wake_up(&ctx->wait);
 	if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
 		wake_up(&ctx->sq_data->wait);
 	if (io_should_trigger_evfd(ctx))
 		eventfd_signal(ctx->cq_ev_fd, 1);
+	if (waitqueue_active(&ctx->cq_wait)) {
+		wake_up_interruptible(&ctx->cq_wait);
+		kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
+	}
+}
+
+static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
+{
+	/* see waitqueue_active() comment */
+	smp_mb();
+
+	if (ctx->flags & IORING_SETUP_SQPOLL) {
+		if (waitqueue_active(&ctx->wait))
+			wake_up(&ctx->wait);
+	}
+	if (io_should_trigger_evfd(ctx))
+		eventfd_signal(ctx->cq_ev_fd, 1);
+	if (waitqueue_active(&ctx->cq_wait)) {
+		wake_up_interruptible(&ctx->cq_wait);
+		kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
+	}
 }
 
 /* Returns true if there are no backlogged entries after the flush */
-static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
+static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
 				       struct task_struct *tsk,
 				       struct files_struct *files)
 {
@@ -1768,6 +1789,20 @@ static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
 	return all_flushed;
 }
 
+static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
+				     struct task_struct *tsk,
+				     struct files_struct *files)
+{
+	if (test_bit(0, &ctx->cq_check_overflow)) {
+		/* iopoll syncs against uring_lock, not completion_lock */
+		if (ctx->flags & IORING_SETUP_IOPOLL)
+			mutex_lock(&ctx->uring_lock);
+		__io_cqring_overflow_flush(ctx, force, tsk, files);
+		if (ctx->flags & IORING_SETUP_IOPOLL)
+			mutex_unlock(&ctx->uring_lock);
+	}
+}
+
 static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
@@ -2127,14 +2162,14 @@ static void __io_req_task_submit(struct io_kiocb *req)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 
-	if (!__io_sq_thread_acquire_mm(ctx) &&
-	    !__io_sq_thread_acquire_files(ctx)) {
 	mutex_lock(&ctx->uring_lock);
+	if (!ctx->sqo_dead &&
+	    !__io_sq_thread_acquire_mm(ctx) &&
+	    !__io_sq_thread_acquire_files(ctx))
 		__io_queue_sqe(req, NULL);
-		mutex_unlock(&ctx->uring_lock);
-	} else {
+	else
 		__io_req_task_cancel(req, -EFAULT);
-	}
+	mutex_unlock(&ctx->uring_lock);
 }
 
 static void io_req_task_submit(struct callback_head *cb)
@@ -2313,20 +2348,8 @@ static void io_double_put_req(struct io_kiocb *req)
 		io_free_req(req);
 }
 
-static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush)
+static unsigned io_cqring_events(struct io_ring_ctx *ctx)
 {
-	if (test_bit(0, &ctx->cq_check_overflow)) {
-		/*
-		 * noflush == true is from the waitqueue handler, just ensure
-		 * we wake up the task, and the next invocation will flush the
-		 * entries. We cannot safely to it from here.
-		 */
-		if (noflush)
-			return -1U;
-
-		io_cqring_overflow_flush(ctx, false, NULL, NULL);
-	}
-
 	/* See comment at the top of this file */
 	smp_rmb();
 	return __io_cqring_events(ctx);
@@ -2424,8 +2447,7 @@ static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
 	}
 
 	io_commit_cqring(ctx);
-	if (ctx->flags & IORING_SETUP_SQPOLL)
-		io_cqring_ev_posted(ctx);
+	io_cqring_ev_posted_iopoll(ctx);
 	io_req_free_batch_finish(ctx, &rb);
 
 	if (!list_empty(&again))
@@ -2551,7 +2573,9 @@ static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
 		 * If we do, we can potentially be spinning for commands that
 		 * already triggered a CQE (eg in error).
 		 */
-		if (io_cqring_events(ctx, false))
+		if (test_bit(0, &ctx->cq_check_overflow))
+			__io_cqring_overflow_flush(ctx, false, NULL, NULL);
+		if (io_cqring_events(ctx))
 			break;
 
 		/*
@@ -2668,6 +2692,8 @@ static bool io_rw_reissue(struct io_kiocb *req, long res)
 	if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker())
 		return false;
 
+	lockdep_assert_held(&req->ctx->uring_lock);
+
 	ret = io_sq_thread_acquire_mm_files(req->ctx, req);
 
 	if (io_resubmit_prep(req, ret)) {
@@ -6826,7 +6852,7 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
 
 	/* if we have a backlog and couldn't flush it all, return BUSY */
 	if (test_bit(0, &ctx->sq_check_overflow)) {
-		if (!io_cqring_overflow_flush(ctx, false, NULL, NULL))
+		if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
 			return -EBUSY;
 	}
 
@@ -6928,7 +6954,8 @@ static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
 		if (!list_empty(&ctx->iopoll_list))
 			io_do_iopoll(ctx, &nr_events, 0);
 
-		if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)))
+		if (to_submit && !ctx->sqo_dead &&
+		    likely(!percpu_ref_is_dying(&ctx->refs)))
 			ret = io_submit_sqes(ctx, to_submit);
 		mutex_unlock(&ctx->uring_lock);
 	}
@@ -7089,7 +7116,7 @@ struct io_wait_queue {
 	unsigned nr_timeouts;
 };
 
-static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush)
+static inline bool io_should_wake(struct io_wait_queue *iowq)
 {
 	struct io_ring_ctx *ctx = iowq->ctx;
 
@@ -7098,7 +7125,7 @@ static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush)
 	 * started waiting. For timeouts, we always want to return to userspace,
 	 * regardless of event count.
 	 */
-	return io_cqring_events(ctx, noflush) >= iowq->to_wait ||
+	return io_cqring_events(ctx) >= iowq->to_wait ||
 			atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
 }
 
@@ -7108,11 +7135,13 @@ static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
 	struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
 							wq);
 
-	/* use noflush == true, as we can't safely rely on locking context */
-	if (!io_should_wake(iowq, true))
-		return -1;
-
+	/*
+	 * Cannot safely flush overflowed CQEs from here, ensure we wake up
+	 * the task, and the next invocation will do it.
+	 */
+	if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
 		return autoremove_wake_function(curr, mode, wake_flags, key);
+	return -1;
 }
 
 static int io_run_task_work_sig(void)
@@ -7149,7 +7178,8 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	int ret = 0;
 
 	do {
-		if (io_cqring_events(ctx, false) >= min_events)
+		io_cqring_overflow_flush(ctx, false, NULL, NULL);
+		if (io_cqring_events(ctx) >= min_events)
 			return 0;
 		if (!io_run_task_work())
 			break;
@@ -7177,6 +7207,7 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 	iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
 	trace_io_uring_cqring_wait(ctx, min_events);
 	do {
+		io_cqring_overflow_flush(ctx, false, NULL, NULL);
 		prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
 						TASK_INTERRUPTIBLE);
 		/* make sure we run task_work before checking for signals */
@@ -7185,8 +7216,10 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 			continue;
 		else if (ret < 0)
 			break;
-		if (io_should_wake(&iowq, false))
+		if (io_should_wake(&iowq))
 			break;
+		if (test_bit(0, &ctx->cq_check_overflow))
+			continue;
 		if (uts) {
 			timeout = schedule_timeout(timeout);
 			if (timeout == 0) {
@@ -7684,12 +7717,12 @@ static struct fixed_file_ref_node *alloc_fixed_file_ref_node(
 
 	ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
 	if (!ref_node)
-		return ERR_PTR(-ENOMEM);
+		return NULL;
 
 	if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero,
 			    0, GFP_KERNEL)) {
 		kfree(ref_node);
-		return ERR_PTR(-ENOMEM);
+		return NULL;
 	}
 	INIT_LIST_HEAD(&ref_node->node);
 	INIT_LIST_HEAD(&ref_node->file_list);
@@ -7783,9 +7816,9 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
 	}
 
 	ref_node = alloc_fixed_file_ref_node(ctx);
-	if (IS_ERR(ref_node)) {
+	if (!ref_node) {
 		io_sqe_files_unregister(ctx);
-		return PTR_ERR(ref_node);
+		return -ENOMEM;
 	}
 
 	io_sqe_files_set_node(file_data, ref_node);
@@ -7885,8 +7918,8 @@ static int __io_sqe_files_update(struct io_ring_ctx *ctx,
 		return -EINVAL;
 
 	ref_node = alloc_fixed_file_ref_node(ctx);
-	if (IS_ERR(ref_node))
-		return PTR_ERR(ref_node);
+	if (!ref_node)
+		return -ENOMEM;
 
 	done = 0;
 	fds = u64_to_user_ptr(up->fds);
@@ -8624,7 +8657,8 @@ static __poll_t io_uring_poll(struct file *file, poll_table *wait)
 	smp_rmb();
 	if (!io_sqring_full(ctx))
 		mask |= EPOLLOUT | EPOLLWRNORM;
-	if (io_cqring_events(ctx, false))
+	io_cqring_overflow_flush(ctx, false, NULL, NULL);
+	if (io_cqring_events(ctx))
 		mask |= EPOLLIN | EPOLLRDNORM;
 
 	return mask;
@@ -8663,7 +8697,7 @@ static void io_ring_exit_work(struct work_struct *work)
 	 * as nobody else will be looking for them.
 	 */
 	do {
-		io_iopoll_try_reap_events(ctx);
+		__io_uring_cancel_task_requests(ctx, NULL);
 	} while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
 	io_ring_ctx_free(ctx);
 }
@@ -8679,10 +8713,14 @@ static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
 {
 	mutex_lock(&ctx->uring_lock);
 	percpu_ref_kill(&ctx->refs);
+
+	if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead))
+		ctx->sqo_dead = 1;
+
 	/* if force is set, the ring is going away. always drop after that */
 	ctx->cq_overflow_flushed = 1;
 	if (ctx->rings)
-		io_cqring_overflow_flush(ctx, true, NULL, NULL);
+		__io_cqring_overflow_flush(ctx, true, NULL, NULL);
 	mutex_unlock(&ctx->uring_lock);
 
 	io_kill_timeouts(ctx, NULL, NULL);
@@ -8818,9 +8856,11 @@ static void __io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
 		enum io_wq_cancel cret;
 		bool ret = false;
 
-		cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, &cancel, true);
-		if (cret != IO_WQ_CANCEL_NOTFOUND)
-			ret = true;
+		if (ctx->io_wq) {
+			cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb,
+					       &cancel, true);
+			ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
+		}
 
 		/* SQPOLL thread does its own polling */
 		if (!(ctx->flags & IORING_SETUP_SQPOLL)) {
@@ -8839,6 +8879,18 @@ static void __io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
 	}
 }
 
+static void io_disable_sqo_submit(struct io_ring_ctx *ctx)
+{
+	WARN_ON_ONCE(ctx->sqo_task != current);
+
+	mutex_lock(&ctx->uring_lock);
+	ctx->sqo_dead = 1;
+	mutex_unlock(&ctx->uring_lock);
+
+	/* make sure callers enter the ring to get error */
+	io_ring_set_wakeup_flag(ctx);
+}
+
 /*
  * We need to iteratively cancel requests, in case a request has dependent
  * hard links. These persist even for failure of cancelations, hence keep
@@ -8850,15 +8902,15 @@ static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
 	struct task_struct *task = current;
 
 	if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
+		/* for SQPOLL only sqo_task has task notes */
+		io_disable_sqo_submit(ctx);
 		task = ctx->sq_data->thread;
 		atomic_inc(&task->io_uring->in_idle);
 		io_sq_thread_park(ctx->sq_data);
 	}
 
 	io_cancel_defer_files(ctx, task, files);
-	io_ring_submit_lock(ctx, (ctx->flags & IORING_SETUP_IOPOLL));
 	io_cqring_overflow_flush(ctx, true, task, files);
-	io_ring_submit_unlock(ctx, (ctx->flags & IORING_SETUP_IOPOLL));
 
 	if (!files)
 		__io_uring_cancel_task_requests(ctx, task);
@@ -8931,20 +8983,12 @@ static void io_uring_del_task_file(struct file *file)
 		fput(file);
 }
 
-/*
- * Drop task note for this file if we're the only ones that hold it after
- * pending fput()
- */
-static void io_uring_attempt_task_drop(struct file *file)
+static void io_uring_remove_task_files(struct io_uring_task *tctx)
 {
-	if (!current->io_uring)
-		return;
-	/*
-	 * fput() is pending, will be 2 if the only other ref is our potential
-	 * task file note. If the task is exiting, drop regardless of count.
-	 */
-	if (fatal_signal_pending(current) || (current->flags & PF_EXITING) ||
-	    atomic_long_read(&file->f_count) == 2)
+	struct file *file;
+	unsigned long index;
+
+	xa_for_each(&tctx->xa, index, file)
 		io_uring_del_task_file(file);
 }
 
@@ -8956,16 +9000,12 @@ void __io_uring_files_cancel(struct files_struct *files)
 
 	/* make sure overflow events are dropped */
 	atomic_inc(&tctx->in_idle);
+	xa_for_each(&tctx->xa, index, file)
+		io_uring_cancel_task_requests(file->private_data, files);
+	atomic_dec(&tctx->in_idle);
 
-	xa_for_each(&tctx->xa, index, file) {
-		struct io_ring_ctx *ctx = file->private_data;
-
-		io_uring_cancel_task_requests(ctx, files);
 	if (files)
-			io_uring_del_task_file(file);
-	}
-
-	atomic_dec(&tctx->in_idle);
+		io_uring_remove_task_files(tctx);
 }
 
 static s64 tctx_inflight(struct io_uring_task *tctx)
@@ -9028,11 +9068,39 @@ void __io_uring_task_cancel(void)
 	} while (1);
 
 	atomic_dec(&tctx->in_idle);
+
+	io_uring_remove_task_files(tctx);
 }
 
 static int io_uring_flush(struct file *file, void *data)
 {
-	io_uring_attempt_task_drop(file);
+	struct io_uring_task *tctx = current->io_uring;
+	struct io_ring_ctx *ctx = file->private_data;
+
+	if (!tctx)
+		return 0;
+
+	/* we should have cancelled and erased it before PF_EXITING */
+	WARN_ON_ONCE((current->flags & PF_EXITING) &&
+		     xa_load(&tctx->xa, (unsigned long)file));
+
+	/*
+	 * fput() is pending, will be 2 if the only other ref is our potential
+	 * task file note. If the task is exiting, drop regardless of count.
+	 */
+	if (atomic_long_read(&file->f_count) != 2)
+		return 0;
+
+	if (ctx->flags & IORING_SETUP_SQPOLL) {
+		/* there is only one file note, which is owned by sqo_task */
+		WARN_ON_ONCE((ctx->sqo_task == current) ==
+			     !xa_load(&tctx->xa, (unsigned long)file));
+
+		io_disable_sqo_submit(ctx);
+	}
+
+	if (!(ctx->flags & IORING_SETUP_SQPOLL) || ctx->sqo_task == current)
+		io_uring_del_task_file(file);
 	return 0;
 }
 
@@ -9106,8 +9174,9 @@ static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
 
 #endif /* !CONFIG_MMU */
 
-static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
+static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
 {
+	int ret = 0;
 	DEFINE_WAIT(wait);
 
 	do {
@@ -9116,6 +9185,11 @@ static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
 
 		prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
 
+		if (unlikely(ctx->sqo_dead)) {
+			ret = -EOWNERDEAD;
+			goto out;
+		}
+
 		if (!io_sqring_full(ctx))
 			break;
 
@@ -9123,6 +9197,8 @@ static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
 	} while (!signal_pending(current));
 
 	finish_wait(&ctx->sqo_sq_wait, &wait);
+out:
+	return ret;
 }
 
 static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
@@ -9194,17 +9270,18 @@ SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
 	 */
 	ret = 0;
 	if (ctx->flags & IORING_SETUP_SQPOLL) {
-		if (!list_empty_careful(&ctx->cq_overflow_list)) {
-			bool needs_lock = ctx->flags & IORING_SETUP_IOPOLL;
-
-			io_ring_submit_lock(ctx, needs_lock);
 		io_cqring_overflow_flush(ctx, false, NULL, NULL);
-			io_ring_submit_unlock(ctx, needs_lock);
-		}
+
+		ret = -EOWNERDEAD;
+		if (unlikely(ctx->sqo_dead))
+			goto out;
 		if (flags & IORING_ENTER_SQ_WAKEUP)
 			wake_up(&ctx->sq_data->wait);
-		if (flags & IORING_ENTER_SQ_WAIT)
-			io_sqpoll_wait_sq(ctx);
+		if (flags & IORING_ENTER_SQ_WAIT) {
+			ret = io_sqpoll_wait_sq(ctx);
+			if (ret)
+				goto out;
+		}
 		submitted = to_submit;
 	} else if (to_submit) {
 		ret = io_uring_add_task_file(ctx, f.file);
@@ -9631,6 +9708,7 @@ static int io_uring_create(unsigned entries, struct io_uring_params *p,
 	trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
 	return ret;
 err:
+	io_disable_sqo_submit(ctx);
 	io_ring_ctx_wait_and_kill(ctx);
 	return ret;
 }