[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[PATCH] block/io_uring: resubmit when result is -EAGAIN

From: Fabian Ebner
Subject: [PATCH] block/io_uring: resubmit when result is -EAGAIN
Date: Wed, 28 Jul 2021 12:35:18 +0200

Quoting from [0]:

 Some setups, like SCSI, can throw spurious -EAGAIN off the softirq
 completion path. Normally we expect this to happen inline as part
 of submission, but apparently SCSI has a weird corner case where it
 can happen as part of normal completions.

Host kernels without patch [0] can panic when this happens [1], and
resubmitting makes the panic more likely. On the other hand, for
kernels with patch [0], resubmitting ensures that a block job is not
aborted just because of such spurious errors.


  #9 [ffffb732000c8b70] asm_exc_page_fault at ffffffffa4800ade
 #10 [ffffb732000c8bf8] io_prep_async_work at ffffffffa3d89c16
 #11 [ffffb732000c8c50] io_rw_reissue at ffffffffa3d8b2e1
 #12 [ffffb732000c8c78] io_complete_rw at ffffffffa3d8baa8
 #13 [ffffb732000c8c98] blkdev_bio_end_io at ffffffffa3d62a80
 #14 [ffffb732000c8cc8] bio_endio at ffffffffa3f4e800
 #15 [ffffb732000c8ce8] dec_pending at ffffffffa432f854
 #16 [ffffb732000c8d30] clone_endio at ffffffffa433170c
 #17 [ffffb732000c8d70] bio_endio at ffffffffa3f4e800
 #18 [ffffb732000c8d90] blk_update_request at ffffffffa3f53a37
 #19 [ffffb732000c8dd0] scsi_end_request at ffffffffa4233a5c
 #20 [ffffb732000c8e08] scsi_io_completion at ffffffffa423432c
 #21 [ffffb732000c8e58] scsi_finish_command at ffffffffa422c527
 #22 [ffffb732000c8e88] scsi_softirq_done at ffffffffa42341e4

Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>

I'm new to this code and io_uring, so I don't know what all the
implications are, but retrying upon EAGAIN does not sound like
a bad thing to my inexperienced ears.

Some more context, leading up to this patch:

We had some users reporting issues after we switched to using io_uring
by default. Namely, kernel panics [2] for some, and failing block jobs
[3] (with a custom backup mechanism we implemented on top of QEMU's
block layer) for others.

I had luck and managed to reprouce the issue, and it was a failed
block job about half of the time and a kernel panic the other half.
When using a host kernel with [0], it's a failed block job all the
time, and this patch attempts to fix that, by resubmitting instead
of bubbling up the error.


 block/io_uring.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/block/io_uring.c b/block/io_uring.c
index 00a3ee9fb8..77d162cb24 100644
--- a/block/io_uring.c
+++ b/block/io_uring.c
@@ -165,7 +165,7 @@ static void luring_process_completions(LuringState *s)
         total_bytes = ret + luringcb->total_read;
         if (ret < 0) {
-            if (ret == -EINTR) {
+            if (ret == -EINTR || ret == -EAGAIN) {
                 luring_resubmit(s, luringcb);

reply via email to

[Prev in Thread] Current Thread [Next in Thread]