|
 |
9ae3a8 |
From b72ee3ebf0f00962872c69ad3576e1c8c0a208d9 Mon Sep 17 00:00:00 2001
|
|
|
9ae3a8 |
From: Fam Zheng <famz@redhat.com>
|
|
|
9ae3a8 |
Date: Thu, 21 Sep 2017 14:23:45 -0300
|
|
|
9ae3a8 |
Subject: [PATCH] block: Limit multiwrite merge (downstream only)
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
RH-Author: Fam Zheng <famz@redhat.com>
|
|
|
9ae3a8 |
Message-id: <20170921142345.22892-1-famz@redhat.com>
|
|
|
9ae3a8 |
Patchwork-id: 76502
|
|
|
9ae3a8 |
O-Subject: [RHEL-7.4 qemu-kvm PATCH] block: Limit multiwrite merge (downstream only)
|
|
|
9ae3a8 |
Bugzilla: 1492559
|
|
|
9ae3a8 |
RH-Acked-by: Max Reitz <mreitz@redhat.com>
|
|
|
9ae3a8 |
RH-Acked-by: John Snow <jsnow@redhat.com>
|
|
|
9ae3a8 |
RH-Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
BZ: https://bugzilla.redhat.com/show_bug.cgi?id=1492559
|
|
|
9ae3a8 |
Brew: https://brewweb.engineering.redhat.com/brew/taskinfo?taskID=14090140
|
|
|
9ae3a8 |
Upstream: Fixed since 2.3, see below.
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
We don't limit the size of the final request of multiwrite_merge. The BZ
|
|
|
9ae3a8 |
has a relatively stable reproducer: Attach an iscsi:// lun as
|
|
|
9ae3a8 |
virtio-blk, then mkfs.ext4. When the guest kernel flushes the page
|
|
|
9ae3a8 |
cache, iscsi driver gets two huge requests that are rejected by the LIO
|
|
|
9ae3a8 |
iscsi target, then reports EIO.
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
Upstream QEMU and qemu-kvm-rhev don't have this problem because of two
|
|
|
9ae3a8 |
things:
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
1) multiread feature in virtio-blk added in 2.3 (95f7142abc8).
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
2) request fragmentation in block layer I/O in 2.7 (04ed95f484 and
|
|
|
9ae3a8 |
1a62d0accd).
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
For 1), we cannot do a faithful backport because it is a new feature,
|
|
|
9ae3a8 |
and is invasive. The change of this patch can be seen as doing a small
|
|
|
9ae3a8 |
part from it, though the implementation is totally different.
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
For 2), we would have a serious context conflict because upstream has a
|
|
|
9ae3a8 |
lot of intermediate changes. The patch could possibly be rewritten in
|
|
|
9ae3a8 |
downstream but is on the one hand much more risky than this patch as it
|
|
|
9ae3a8 |
touches the very center of the block I/O code, and on the other hand
|
|
|
9ae3a8 |
loses the meaning because the diff will completely deviate from
|
|
|
9ae3a8 |
upstream.
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
Signed-off-by: Fam Zheng <famz@redhat.com>
|
|
|
9ae3a8 |
Signed-off-by: Wainer dos Santos Moschetta <wainersm@redhat.com>
|
|
|
9ae3a8 |
---
|
|
|
9ae3a8 |
block.c | 5 +++++
|
|
|
9ae3a8 |
1 file changed, 5 insertions(+)
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
diff --git a/block.c b/block.c
|
|
|
9ae3a8 |
index bdcd741389..b0373d0c16 100644
|
|
|
9ae3a8 |
--- a/block.c
|
|
|
9ae3a8 |
+++ b/block.c
|
|
|
9ae3a8 |
@@ -4303,6 +4303,11 @@ static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
|
|
|
9ae3a8 |
merge = 0;
|
|
|
9ae3a8 |
}
|
|
|
9ae3a8 |
|
|
|
9ae3a8 |
+ if (reqs[outidx].qiov->size + reqs[i].qiov->size >=
|
|
|
9ae3a8 |
+ bs->bl.max_transfer_length * BDRV_SECTOR_SIZE) {
|
|
|
9ae3a8 |
+ merge = 0;
|
|
|
9ae3a8 |
+ }
|
|
|
9ae3a8 |
+
|
|
|
9ae3a8 |
if (merge) {
|
|
|
9ae3a8 |
size_t size;
|
|
|
9ae3a8 |
QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
|
|
|
9ae3a8 |
--
|
|
|
9ae3a8 |
2.13.5
|
|
|
9ae3a8 |
|