Blame SOURCES/kvm-block-Fix-invalidate_cache-error-path-for-parent-act.patch

383d26
From b7d8518d4d2e5d53c2f8eb6cc78d99a69dc4feff Mon Sep 17 00:00:00 2001
383d26
From: Kevin Wolf <kwolf@redhat.com>
383d26
Date: Mon, 4 Feb 2019 16:06:04 +0100
383d26
Subject: [PATCH 7/8] block: Fix invalidate_cache error path for parent
383d26
 activation
383d26
383d26
RH-Author: Kevin Wolf <kwolf@redhat.com>
383d26
Message-id: <20190204160604.2723-2-kwolf@redhat.com>
383d26
Patchwork-id: 84202
383d26
O-Subject: [RHEL-7.7 qemu-kvm-rhev PATCH 1/1] block: Fix invalidate_cache error path for parent activation
383d26
Bugzilla: 1531888
383d26
RH-Acked-by: Markus Armbruster <armbru@redhat.com>
383d26
RH-Acked-by: John Snow <jsnow@redhat.com>
383d26
RH-Acked-by: Max Reitz <mreitz@redhat.com>
383d26
383d26
bdrv_co_invalidate_cache() clears the BDRV_O_INACTIVE flag before
383d26
actually activating a node so that the correct permissions etc. are
383d26
taken. In case of errors, the flag must be restored so that the next
383d26
call to bdrv_co_invalidate_cache() retries activation.
383d26
383d26
Restoring the flag was missing in the error path for a failed
383d26
parent->role->activate() call. The consequence is that this attempt to
383d26
activate all images correctly fails because we still set errp, however
383d26
on the next attempt BDRV_O_INACTIVE is already clear, so we return
383d26
success without actually retrying the failed action.
383d26
383d26
An example where this is observable in practice is migration to a QEMU
383d26
instance that has a raw format block node attached to a guest device
383d26
with share-rw=off (the default) while another process holds
383d26
BLK_PERM_WRITE for the same image. In this case, all activation steps
383d26
before parent->role->activate() succeed because raw can tolerate other
383d26
writers to the image. Only the parent callback (in particular
383d26
blk_root_activate()) tries to implement the share-rw=on property and
383d26
requests exclusive write permissions. This fails when the migration
383d26
completes and correctly displays an error. However, a manual 'cont' will
383d26
incorrectly resume the VM without calling blk_root_activate() again.
383d26
383d26
This case is described in more detail in the following bug report:
383d26
https://bugzilla.redhat.com/show_bug.cgi?id=1531888
383d26
383d26
Fix this by correctly restoring the BDRV_O_INACTIVE flag in the error
383d26
path.
383d26
383d26
Cc: qemu-stable@nongnu.org
383d26
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
383d26
Tested-by: Markus Armbruster <armbru@redhat.com>
383d26
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
383d26
(cherry picked from commit 78fc3b3a26c145eebcdee992988644974b243a74)
383d26
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
383d26
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
383d26
---
383d26
 block.c | 1 +
383d26
 1 file changed, 1 insertion(+)
383d26
383d26
diff --git a/block.c b/block.c
383d26
index da9b1a6..ce85c65 100644
383d26
--- a/block.c
383d26
+++ b/block.c
383d26
@@ -4410,6 +4410,7 @@ static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
383d26
         if (parent->role->activate) {
383d26
             parent->role->activate(parent, &local_err);
383d26
             if (local_err) {
383d26
+                bs->open_flags |= BDRV_O_INACTIVE;
383d26
                 error_propagate(errp, local_err);
383d26
                 return;
383d26
             }
383d26
-- 
383d26
1.8.3.1
383d26