block: bdrv_reopen_multiple(): move bdrv_flush to separate pre-prepare
During reopen we may add backing bs from other aio context, which may
lead to changing original context of top bs.
We are going to move graph modification to prepare stage. So, it will
be possible that bdrv_flush() in bdrv_reopen_prepare called on bs in
non-original aio context, which we didn't aquire which leads to crash.
To avoid this problem move bdrv_flush() to be a separate reopen stage
before bdrv_reopen_prepare().
This doesn't seem correct to acquire only one aio context and not all
contexts participating in reopen. But it's not obvious how to do it
correctly, keeping in mind:
1. rules of bdrv_set_aio_context_ignore() that requires new_context
lock not being held
2. possible deadlocks because of holding all (or several?) AioContext
locks
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-30-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
diff --git a/block.c b/block.c
index bdfe59d..357ec1b 100644
--- a/block.c
+++ b/block.c
@@ -4275,6 +4275,14 @@
assert(bs_queue != NULL);
QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
+ ret = bdrv_flush(bs_entry->state.bs);
+ if (ret < 0) {
+ error_setg_errno(errp, -ret, "Error flushing drive");
+ goto cleanup;
+ }
+ }
+
+ QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
assert(bs_entry->state.bs->quiesce_counter > 0);
if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
goto cleanup;
@@ -4669,12 +4677,6 @@
bdrv_reopen_perm(queue, reopen_state->bs,
&reopen_state->perm, &reopen_state->shared_perm);
- ret = bdrv_flush(reopen_state->bs);
- if (ret) {
- error_setg_errno(errp, -ret, "Error flushing drive");
- goto error;
- }
-
if (drv->bdrv_reopen_prepare) {
/*
* If a driver-specific option is missing, it means that we
