[PATCH v2] io_uring: re-issue plug based block requests that failed

[Jens Axboe <axboe@kernel.dk>]

Mark the plug with nowait == true, which will cause requests to avoid
blocking on request allocation. If they do, we catch them and add them
to the plug list. Once we finish the plug, re-issue requests that got
caught.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

---

Since v1:
- Properly re-prep on resubmit
- Sanity check for io_wq_current_is_worker()
- Sanity check we're just doing read/write for re-issue
- Ensure iov state is sane for resubmit

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 625578715d37..942984bda2f8 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1947,12 +1947,31 @@ static void io_complete_rw_common(struct kiocb *kiocb, long res)
 	__io_cqring_add_event(req, res, cflags);
 }
 
+static bool io_rw_reissue(struct io_kiocb *req, long res)
+{
+#ifdef CONFIG_BLOCK
+	struct blk_plug *plug;
+
+	if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker())
+		return false;
+
+	plug = current->plug;
+	if (plug && plug->nowait) {
+		list_add_tail(&req->list, &plug->nowait_list);
+		return true;
+	}
+#endif
+	return false;
+}
+
 static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
 {
 	struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
 
-	io_complete_rw_common(kiocb, res);
-	io_put_req(req);
+	if (!io_rw_reissue(req, res)) {
+		io_complete_rw_common(kiocb, res);
+		io_put_req(req);
+	}
 }
 
 static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
@@ -2629,6 +2648,7 @@ static int io_read(struct io_kiocb *req, bool force_nonblock)
 	iov_count = iov_iter_count(&iter);
 	ret = rw_verify_area(READ, req->file, &kiocb->ki_pos, iov_count);
 	if (!ret) {
+		unsigned long nr_segs = iter.nr_segs;
 		ssize_t ret2 = 0;
 
 		if (req->file->f_op->read_iter)
@@ -2640,6 +2660,8 @@ static int io_read(struct io_kiocb *req, bool force_nonblock)
 		if (!force_nonblock || (ret2 != -EAGAIN && ret2 != -EIO)) {
 			kiocb_done(kiocb, ret2);
 		} else {
+			iter.count = iov_count;
+			iter.nr_segs = nr_segs;
 copy_iov:
 			ret = io_setup_async_rw(req, io_size, iovec,
 						inline_vecs, &iter);
@@ -2726,6 +2748,7 @@ static int io_write(struct io_kiocb *req, bool force_nonblock)
 	iov_count = iov_iter_count(&iter);
 	ret = rw_verify_area(WRITE, req->file, &kiocb->ki_pos, iov_count);
 	if (!ret) {
+		unsigned long nr_segs = iter.nr_segs;
 		ssize_t ret2;
 
 		/*
@@ -2763,6 +2786,8 @@ static int io_write(struct io_kiocb *req, bool force_nonblock)
 		if (!force_nonblock || ret2 != -EAGAIN) {
 			kiocb_done(kiocb, ret2);
 		} else {
+			iter.count = iov_count;
+			iter.nr_segs = nr_segs;
 copy_iov:
 			ret = io_setup_async_rw(req, io_size, iovec,
 						inline_vecs, &iter);
@@ -5789,12 +5814,70 @@ static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 	return 0;
 }
 
+#ifdef CONFIG_BLOCK
+static bool io_resubmit_prep(struct io_kiocb *req)
+{
+	struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
+	struct iov_iter iter;
+	ssize_t ret;
+	int rw;
+
+	switch (req->opcode) {
+	case IORING_OP_READV:
+	case IORING_OP_READ_FIXED:
+	case IORING_OP_READ:
+		rw = READ;
+		break;
+	case IORING_OP_WRITEV:
+	case IORING_OP_WRITE_FIXED:
+	case IORING_OP_WRITE:
+		rw = WRITE;
+		break;
+	default:
+		printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n",
+				req->opcode);
+		goto end_req;
+	}
+
+	ret = io_import_iovec(rw, req, &iovec, &iter, false);
+	if (ret < 0)
+		goto end_req;
+	ret = io_setup_async_rw(req, ret, iovec, inline_vecs, &iter);
+	if (!ret)
+		return true;
+	kfree(iovec);
+end_req:
+	io_cqring_add_event(req, ret);
+	req_set_fail_links(req);
+	io_put_req(req);
+	return false;
+}
+
+static void io_resubmit_rw(struct list_head *list)
+{
+	struct io_kiocb *req;
+
+	while (!list_empty(list)) {
+		req = list_first_entry(list, struct io_kiocb, list);
+		list_del(&req->list);
+		if (io_resubmit_prep(req)) {
+			refcount_inc(&req->refs);
+			io_queue_async_work(req);
+		}
+	}
+}
+#endif
+
 /*
  * Batched submission is done, ensure local IO is flushed out.
  */
 static void io_submit_state_end(struct io_submit_state *state)
 {
 	blk_finish_plug(&state->plug);
+#ifdef CONFIG_BLOCK
+	if (unlikely(!list_empty(&state->plug.nowait_list)))
+		io_resubmit_rw(&state->plug.nowait_list);
+#endif
 	io_state_file_put(state);
 	if (state->free_reqs)
 		kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs);
@@ -5807,6 +5890,10 @@ static void io_submit_state_start(struct io_submit_state *state,
 				  unsigned int max_ios)
 {
 	blk_start_plug(&state->plug);
+#ifdef CONFIG_BLOCK
+	INIT_LIST_HEAD(&state->plug.nowait_list);
+	state->plug.nowait = true;
+#endif
 	state->free_reqs = 0;
 	state->file = NULL;
 	state->ios_left = max_ios;

-- 
Jens Axboe