On 11/17/2015 10:22 AM, Stefan Hajnoczi wrote:
On Mon, Nov 16, 2015 at 06:24:36PM +0300, Denis V. Lunev wrote:
+int bdrv_all_find_snapshot(const char *name, bool skip_read_only,
+ BlockDriverState **first_bad_bs)
+{
+ QEMUSnapshotInfo sn;
+ int err = 0;
+ BlockDriverState *bs = NULL;
+
+ while (err == 0 && (bs = bdrv_next(bs))) {
+ AioContext *ctx = bdrv_get_aio_context(bs);
+
+ if (skip_read_only &&
+ (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs))) {
These must be called with AioContext acquired.
AFAIK no. this is called without that in bdrv jobs code.
bdrv_is_read_only is a simple flag, which is filled on open
bdrv_is_inserted is defined for CDROM only and does
not rely on AIO context stuff.
+ continue;
+ }
+
+ aio_context_acquire(ctx);
+ if (bdrv_can_snapshot(bs)) {
The !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) checks above are
redundant since bdrv_can_snapshot() checks too:
int bdrv_can_snapshot(BlockDriverState *bs)
{
BlockDriver *drv = bs->drv;
if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
return 0;
}
It also means that the "skip_read_only" name is inaccurate. Read-only
drives are always skipped, regardless of skip_read_only's value.
The skip_read_only argument can be dropped and the earlier
!bdrv_is_inserted(bs) || bdrv_is_read_only(bs) check can be dropped too.
I have though on this several times and once again I am wrong :(
This looks like a punishment for my sins. OK :(
@@ -2168,21 +2157,7 @@ void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
available_snapshots = g_new0(int, nb_sns);
total = 0;
for (i = 0; i < nb_sns; i++) {
- sn = &sn_tab[i];
- available = 1;
- bs1 = NULL;
-
- while ((bs1 = bdrv_next(bs1))) {
- if (bdrv_can_snapshot(bs1) && bs1 != bs) {
- ret = bdrv_snapshot_find(bs1, sn_info, sn->id_str);
- if (ret < 0) {
- available = 0;
- break;
- }
- }
- }
-
- if (available) {
+ if (bdrv_all_find_snapshot(sn_tab[i].id_str, false, &bs1) == 0) {
bdrv_all_find_snapshot() doesn't do the bs1 != bs exclusion so the new
code behaves differently from the old code. That seems like a bug.
no. The result will be the same. This is a minor optimisation:
- we get the first block device we can make snapshot on
- we get snapshot list on that device
- after that we start iterations over the list and removing not
available snapshots on other devices
This means that if bs == bs1 the check will be "always true"
