Merge tag 'io_uring-5.7-2020-05-22' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "A small collection of small fixes that should go into this release:

   - Two fixes for async request preparation (Pavel)

   - Busy clear fix for SQPOLL (Xiaoguang)

   - Don't use kiocb->private for O_DIRECT buf index, some file systems
     use it (Bijan)

   - Kill dead check in io_splice()

   - Ensure sqo_wait is initialized early

   - Cancel task_work if we fail adding to original process

   - Only add (IO)pollable requests to iopoll list, fixing a regression
     in this merge window"

* tag 'io_uring-5.7-2020-05-22' of git://git.kernel.dk/linux-block:
  io_uring: reset -EBUSY error when io sq thread is waken up
  io_uring: don't add non-IO requests to iopoll pending list
  io_uring: don't use kiocb.private to store buf_index
  io_uring: cancel work if task_work_add() fails
  io_uring: remove dead check in io_splice()
  io_uring: fix FORCE_ASYNC req preparation
  io_uring: don't prepare DRAIN reqs twice
  io_uring: initialize ctx->sqo_wait earlier

Author: torvalds

fs/io_uring.c

@@ -619,6 +619,8 @@ struct io_kiocb {
 	bool				needs_fixed_file;
 	u8				opcode;
 
+	u16				buf_index;
+
 	struct io_ring_ctx	*ctx;
 	struct list_head	list;
 	unsigned int		flags;
@@ -924,6 +926,7 @@ static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 		goto err;
 
 	ctx->flags = p->flags;
+	init_waitqueue_head(&ctx->sqo_wait);
 	init_waitqueue_head(&ctx->cq_wait);
 	INIT_LIST_HEAD(&ctx->cq_overflow_list);
 	init_completion(&ctx->completions[0]);
@@ -2100,9 +2103,7 @@ static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 
 	req->rw.addr = READ_ONCE(sqe->addr);
 	req->rw.len = READ_ONCE(sqe->len);
-	/* we own ->private, reuse it for the buffer index  / buffer ID */
-	req->rw.kiocb.private = (void *) (unsigned long)
-					READ_ONCE(sqe->buf_index);
+	req->buf_index = READ_ONCE(sqe->buf_index);
 	return 0;
 }
 
@@ -2145,15 +2146,15 @@ static ssize_t io_import_fixed(struct io_kiocb *req, int rw,
 	struct io_ring_ctx *ctx = req->ctx;
 	size_t len = req->rw.len;
 	struct io_mapped_ubuf *imu;
-	unsigned index, buf_index;
+	u16 index, buf_index;
 	size_t offset;
 	u64 buf_addr;
 
 	/* attempt to use fixed buffers without having provided iovecs */
 	if (unlikely(!ctx->user_bufs))
 		return -EFAULT;
 
-	buf_index = (unsigned long) req->rw.kiocb.private;
+	buf_index = req->buf_index;
 	if (unlikely(buf_index >= ctx->nr_user_bufs))
 		return -EFAULT;
 
@@ -2269,10 +2270,10 @@ static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
 					bool needs_lock)
 {
 	struct io_buffer *kbuf;
-	int bgid;
+	u16 bgid;
 
 	kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
-	bgid = (int) (unsigned long) req->rw.kiocb.private;
+	bgid = req->buf_index;
 	kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
 	if (IS_ERR(kbuf))
 		return kbuf;
@@ -2363,7 +2364,7 @@ static ssize_t io_import_iovec(int rw, struct io_kiocb *req,
 	}
 
 	/* buffer index only valid with fixed read/write, or buffer select  */
-	if (req->rw.kiocb.private && !(req->flags & REQ_F_BUFFER_SELECT))
+	if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
 		return -EINVAL;
 
 	if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
@@ -2771,11 +2772,8 @@ static int io_splice(struct io_kiocb *req, bool force_nonblock)
 	poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
 	poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
 
-	if (sp->len) {
+	if (sp->len)
 		ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
-		if (force_nonblock && ret == -EAGAIN)
-			return -EAGAIN;
-	}
 
 	io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
 	req->flags &= ~REQ_F_NEED_CLEANUP;
@@ -4137,12 +4135,14 @@ static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
 	req->result = mask;
 	init_task_work(&req->task_work, func);
 	/*
-	 * If this fails, then the task is exiting. Punt to one of the io-wq
-	 * threads to ensure the work gets run, we can't always rely on exit
-	 * cancelation taking care of this.
+	 * If this fails, then the task is exiting. When a task exits, the
+	 * work gets canceled, so just cancel this request as well instead
+	 * of executing it. We can't safely execute it anyway, as we may not
+	 * have the needed state needed for it anyway.
 	 */
 	ret = task_work_add(tsk, &req->task_work, true);
 	if (unlikely(ret)) {
+		WRITE_ONCE(poll->canceled, true);
 		tsk = io_wq_get_task(req->ctx->io_wq);
 		task_work_add(tsk, &req->task_work, true);
 	}
@@ -5013,12 +5013,13 @@ static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list))
 		return 0;
 
-	if (!req->io && io_alloc_async_ctx(req))
-		return -EAGAIN;
-
-	ret = io_req_defer_prep(req, sqe);
-	if (ret < 0)
-		return ret;
+	if (!req->io) {
+		if (io_alloc_async_ctx(req))
+			return -EAGAIN;
+		ret = io_req_defer_prep(req, sqe);
+		if (ret < 0)
+			return ret;
+	}
 
 	spin_lock_irq(&ctx->completion_lock);
 	if (!req_need_defer(req) && list_empty(&ctx->defer_list)) {
@@ -5305,7 +5306,8 @@ static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 	if (ret)
 		return ret;
 
-	if (ctx->flags & IORING_SETUP_IOPOLL) {
+	/* If the op doesn't have a file, we're not polling for it */
+	if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
 		const bool in_async = io_wq_current_is_worker();
 
 		if (req->result == -EAGAIN)
@@ -5606,9 +5608,15 @@ static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 			io_double_put_req(req);
 		}
 	} else if (req->flags & REQ_F_FORCE_ASYNC) {
-		ret = io_req_defer_prep(req, sqe);
-		if (unlikely(ret < 0))
-			goto fail_req;
+		if (!req->io) {
+			ret = -EAGAIN;
+			if (io_alloc_async_ctx(req))
+				goto fail_req;
+			ret = io_req_defer_prep(req, sqe);
+			if (unlikely(ret < 0))
+				goto fail_req;
+		}
+
 		/*
 		 * Never try inline submit of IOSQE_ASYNC is set, go straight
 		 * to async execution.
@@ -6024,6 +6032,7 @@ static int io_sq_thread(void *data)
 				finish_wait(&ctx->sqo_wait, &wait);
 
 				ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
+				ret = 0;
 				continue;
 			}
 			finish_wait(&ctx->sqo_wait, &wait);
@@ -6837,7 +6846,6 @@ static int io_sq_offload_start(struct io_ring_ctx *ctx,
 {
 	int ret;
 
-	init_waitqueue_head(&ctx->sqo_wait);
 	mmgrab(current->mm);
 	ctx->sqo_mm = current->mm;