From 38cf95e0faf265b6ac8b1a3f1d869fce53d91b8c Mon Sep 17 00:00:00 2001
From: Gerd Hoffmann <kraxel@redhat.com>
Date: Thu, 27 Oct 2011 11:11:16 +0200
Subject: [PATCH] migration: flush migration data to disk.
RH-Author: Gerd Hoffmann <kraxel@redhat.com>
Message-id: <1319713876-25239-1-git-send-email-kraxel@redhat.com>
Patchwork-id: 34691
O-Subject: [RHEL-6.2 kvm PATCH] migration: flush migration data to disk.
Bugzilla: 721114
RH-Acked-by: Juan Quintela <quintela@redhat.com>
RH-Acked-by: Avi Kivity <avi@redhat.com>
RH-Acked-by: Orit Wasserman <owasserm@redhat.com>
This patch increases robustness when migrating to a file with
two little changes:
(1) Before closing the migration file handle checks if it happens to be
a regular file and if so it issues a fsync. This way the data is
flushed to disk before qemu sends the migration completed event.
(2) It adds error checking. In case either fsync or close syscall
fails pass up the error (and fail migration).
[ v2: return -errno instead of -1 ]
Cc: Juan Quintela <quintela@redhat.com>
Cc: Jiri Denemark <jdenemar@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
upstream: http://patchwork.ozlabs.org/patch/122074/
bugzilla: 721114 - qemu fails to restore guests that were previously
suspended on host shutdown
Extra note: It is not clear yet why this patch actually helps. It is
certainly a good thing to do from a rubustness perspective. But unless
disk errors or power failures are involved it should make no difference
in theory, the data should land savely on disk no matter what. In
practice it does make a difference for not yet known reasons.
scratch build #1 (used by Jiri for the tests, includes instrumentation):
http://brewweb.devel.redhat.com/brew/taskinfo?taskID=3743586
http://git.engineering.redhat.com/?p=users/ghoffman/rhel6/qemu-kvm.git;a=shortlog;h=bz721114-loadvm-failure
scratch build #2 (just this patch):
http://brewweb.devel.redhat.com/brew/taskinfo?taskID=3748758
http://git.engineering.redhat.com/?p=users/ghoffman/rhel6/qemu-kvm.git;a=shortlog;h=bz721114-loadvm-failure-backport
Conflicts:
migration-fd.c
---
migration-fd.c | 23 ++++++++++++++++++++++-
1 files changed, 22 insertions(+), 1 deletions(-)
Signed-off-by: Michal Novotny <minovotn@redhat.com>
---
migration-fd.c | 23 ++++++++++++++++++++++-
1 files changed, 22 insertions(+), 1 deletions(-)
diff --git a/migration-fd.c b/migration-fd.c
index 9feb8cd..0f0bdc4 100644
--- a/migration-fd.c
+++ b/migration-fd.c
@@ -43,10 +43,31 @@ static int fd_write(FdMigrationState *s, const void * buf, size_t size)
static int fd_close(FdMigrationState *s)
{
+ struct stat st;
+ int ret;
+
DPRINTF("fd_close\n");
if (s->fd != -1) {
- close(s->fd);
+ ret = fstat(s->fd, &st);
+ if (ret == 0 && S_ISREG(st.st_mode)) {
+ /*
+ * If the file handle is a regular file make sure the
+ * data is flushed to disk before signaling success.
+ */
+ ret = fsync(s->fd);
+ if (ret != 0) {
+ ret = -errno;
+ perror("migration-fd: fsync");
+ return ret;
+ }
+ }
+ ret = close(s->fd);
s->fd = -1;
+ if (ret != 0) {
+ ret = -errno;
+ perror("migration-fd: close");
+ return ret;
+ }
}
return 0;
}
--
1.7.4.4