RE: [PATCH v2] migration: Don't allow migration if vm is in POSTMIGRATE state

[Tuguoyi]

On December 07, 2020 6:06 PM Vladimir Sementsov-Ogievskiy wrote:
> 07.12.2020 10:44, Tuguoyi wrote:
> > The following steps will cause qemu assertion failure:
> > - pause vm by executing 'virsh suspend'
> > - create external snapshot of memory and disk using 'virsh
> snapshot-create-as'
> > - doing the above operation again will cause qemu crash
> >
> > The backtrace looks like:
> > #0  0x00007fbf958c5c37 in raise () from /lib/x86_64-linux-gnu/libc.so.6
> > #1  0x00007fbf958c9028 in abort () from /lib/x86_64-linux-gnu/libc.so.6
> > #2  0x00007fbf958bebf6 in ?? () from /lib/x86_64-linux-gnu/libc.so.6
> > #3  0x00007fbf958beca2 in __assert_fail () from
> /lib/x86_64-linux-gnu/libc.so.6
> > #4  0x000055ca8decd39d in bdrv_inactivate_recurse (bs=0x55ca90c80400)
> at /build/qemu-5.0/block.c:5724
> > #5  0x000055ca8dece967 in bdrv_inactivate_all () at
> /build//qemu-5.0/block.c:5792
> > #6  0x000055ca8de5539d in
> qemu_savevm_state_complete_precopy_non_iterable (inactivate_disks=true,
> in_postcopy=false, f=0x55ca907044b0)
> >      at /build/qemu-5.0/migration/savevm.c:1401
> > #7  qemu_savevm_state_complete_precopy (f=0x55ca907044b0,
> iterable_only=iterable_only@entry=false,
> inactivate_disks=inactivate_disks@entry=true)
> >      at /build/qemu-5.0/migration/savevm.c:1453
> > #8  0x000055ca8de4f581 in migration_completion (s=0x55ca8f64d9f0) at
> /build/qemu-5.0/migration/migration.c:2941
> > #9  migration_iteration_run (s=0x55ca8f64d9f0) at
> /build/qemu-5.0/migration/migration.c:3295
> > #10 migration_thread (opaque=opaque@entry=0x55ca8f64d9f0) at
> /build/qemu-5.0/migration/migration.c:3459
> > #11 0x000055ca8dfc6716 in qemu_thread_start (args=<optimized out>) at
> /build/qemu-5.0/util/qemu-thread-posix.c:519
> > #12 0x00007fbf95c5f184 in start_thread () from
> /lib/x86_64-linux-gnu/libpthread.so.0
> > #13 0x00007fbf9598cbed in clone () from /lib/x86_64-linux-gnu/libc.so.6
> >
> > When the first migration completes, bs->open_flags will set
> BDRV_O_INACTIVE
> > flag by bdrv_inactivate_all(), and during the second migration the
> > bdrv_inactivate_recurse assert that the bs->open_flags is already
> > BDRV_O_INACTIVE enabled which cause crash.
> >
> > As Vladimir suggested, this patch just make migration job error-out with a
> > message in  migrate_fd_connect() if the vm is in
> RUN_STATE_POSTMIGRATE state.
> >
> > Signed-off-by: Tuguoyi <tu.guoyi@h3c.com>
> > ---
> >   migration/migration.c | 7 +++++++
> >   1 file changed, 7 insertions(+)
> >
> > diff --git a/migration/migration.c b/migration/migration.c
> > index 87a9b59..4091678 100644
> > --- a/migration/migration.c
> > +++ b/migration/migration.c
> > @@ -3622,6 +3622,13 @@ void migrate_fd_connect(MigrationState *s,
> Error *error_in)
> >           return;
> >       }
> >
> > +    if (runstate_check(RUN_STATE_POSTMIGRATE)) {
> > +        error_report("Can't migrate the vm that is in POSTMIGRATE
> state");
> > +        migrate_set_state(&s->state, s->state,
> MIGRATION_STATUS_FAILED);
> > +        migrate_fd_cleanup(s);
> > +        return;
> > +    }
> > +
> >       if (resume) {
> >           /* This is a resumed migration */
> >           rate_limit = s->parameters.max_postcopy_bandwidth /
> >
> 
> 
> I think, correct place for the check migrate_prepare, as it is called for any 
> kind
> of migration, not only fd_. And in it we have already check for wrong state:
> 
>        if (runstate_check(RUN_STATE_INMIGRATE)) {
>            error_setg(errp, "Guest is waiting for an incoming migration");
>            return false;
>        }
> 
> and no additional state change and cleanup is needed.

Thank you for your advise, I'll send another patch. 

> --
> Best regards,
> Vladimir