On Thu, Aug 23, 2012 at 02:28:07PM +0200, Benoît Canet wrote:
> Usage:
> (qemu) migrate file:/path/to/vm_statefile
>
> Signed-off-by: Benoit Canet <ben...@irqsave.net>
> ---
> migration-fd.c | 4 ++--
> migration.c | 20 +++++++++++++++++++-
> migration.h | 2 +-
> 3 files changed, 22 insertions(+), 4 deletions(-)
>
> diff --git a/migration-fd.c b/migration-fd.c
> index 50138ed..d39e44a 100644
> --- a/migration-fd.c
> +++ b/migration-fd.c
> @@ -73,9 +73,9 @@ static int fd_close(MigrationState *s)
> return 0;
> }
>
> -int fd_start_outgoing_migration(MigrationState *s, const char *fdname)
> +int fd_start_outgoing_migration(MigrationState *s, const char *fdname, int
> fd)
> {
> - s->fd = monitor_get_fd(cur_mon, fdname);
> + s->fd = fd ? fd : monitor_get_fd(cur_mon, fdname);
> if (s->fd == -1) {
> DPRINTF("fd_migration: invalid file descriptor identifier\n");
> goto err_after_get_fd;
> diff --git a/migration.c b/migration.c
> index 1edeec5..679847d 100644
> --- a/migration.c
> +++ b/migration.c
> @@ -239,9 +239,14 @@ void
> qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
> static int migrate_fd_cleanup(MigrationState *s)
> {
> int ret = 0;
> + struct stat st;
>
> qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
>
> + if (!fstat(s->fd, &st) && S_ISREG(st.st_mode)) {
> + fsync(s->fd);
> + }
> +
> if (s->file) {
> DPRINTF("closing file\n");
> ret = qemu_fclose(s->file);
> @@ -475,6 +480,17 @@ void migrate_del_blocker(Error *reason)
> migration_blockers = g_slist_remove(migration_blockers, reason);
> }
>
> +static int file_start_outgoing_migration(MigrationState *s,
> + const char *filename)
> +{
> + int fd;
> + fd = open(filename, O_CREAT|O_TRUNC|O_WRONLY, S_IRUSR|S_IWUSR);
> + if (fd < 0) {
> + return -errno;
> + }
> + return fd_start_outgoing_migration(s, NULL, fd);
'fd_start_outgoing_migration' requires that the FD you give it
supports non-blocking I/O. File descriptors opened from plain
files or block devices do not honour that requirement. So this
proposed code will cause the entire QEMU process to block while
migration is taking place. This is why no on has ever implemented
the 'file:' protocol in QEMU before.
To deal with this issue you'd either have to use the POSIX
async I/O APIs (or QEMU's internal equivalent), or spawn a
separate 'dd' helper process and give QEMU a pipe FD instead.
The latter is what libvirt does to implement migrate to file.
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://autobuild.org -o- http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|
