Skip to contentSkip to conversation

io_uring: fix OP_ASYNC_CANCEL across tasks

PATCH In review Pavel Begunkov 2 participants 2 messages Mar 12, 2021 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

PATCH 1/1 io_uring: fix OP_ASYNC_CANCEL across tasks

IORING_OP_ASYNC_CANCEL tries io-wq cancellation only for current task. If it fails go over tctx_list and try it out for every single tctx.

---
 fs/io_uring.c | 41 ++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 40 insertions(+), 1 deletion(-)
fs/io_uring.c +40 -1
index 70286b393c0e..a4bce17af506 100644
@@ -5667,8 +5667,47 @@ static int io_async_cancel_prep(struct io_kiocb *req,
 static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
+	u64 sqe_addr = req->cancel.addr;
+	struct io_tctx_node *node;
+	int ret;
 
-	io_async_find_and_cancel(ctx, req, req->cancel.addr, 0);
+	/* tasks should wait for their io-wq threads, so safe w/o sync */
+	ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
+	spin_lock_irq(&ctx->completion_lock);
+	if (ret != -ENOENT)
+		goto done;
+	ret = io_timeout_cancel(ctx, sqe_addr);
+	if (ret != -ENOENT)
+		goto done;
+	ret = io_poll_cancel(ctx, sqe_addr);
+	if (ret != -ENOENT)
+		goto done;
+	spin_unlock_irq(&ctx->completion_lock);
+
+	/* slow path, try all io-wq's */
+	io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
+	ret = -ENOENT;
+	list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
+		struct io_uring_task *tctx = node->task->io_uring;
+
+		if (!tctx || !tctx->io_wq)
+			continue;
+		ret = io_async_cancel_one(tctx, req->cancel.addr, ctx);
+		if (ret != -ENOENT)
+			break;
+	}
+	io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
+
+	spin_lock_irq(&ctx->completion_lock);
+done:
+	io_cqring_fill_event(req, ret);
+	io_commit_cqring(ctx);
+	spin_unlock_irq(&ctx->completion_lock);
+	io_cqring_ev_posted(ctx);
+
+	if (ret < 0)
+		req_set_fail_links(req);
+	io_put_req(req);
 	return 0;
 }
 
Signature · 1 line
2.24.0
maintainer
#2 lore raw
Quoted 2 lines · Pavel Begunkov

IORING_OP_ASYNC_CANCEL tries io-wq cancellation only for current task. If it fails go over tctx_list and try it out for every single tctx.

Applied for 5.12, thanks.

Signature · 1 line
Jens Axboe