Re: [PATCH for-9.0 1/2] migration: Set migration error in migration_completion()

["Cédric Le Goater" <clg@redhat.com>]

On 3/28/24 16:50, Avihai Horon wrote:
> 
> On 28/03/2024 17:21, Cédric Le Goater wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> Hello Avihai,
>>
>> On 3/28/24 15:02, Avihai Horon wrote:
>>> After commit 9425ef3f990a ("migration: Use migrate_has_error() in
>>> close_return_path_on_source()"), close_return_path_on_source() assumes
>>> that migration error is set if an error occurs during migration.
>>>
>>> This may not be true if migration errors in migration_completion(). For
>>> example, if qemu_savevm_state_complete_precopy() errors, migration error
>>> will not be set
>>
>> Out of curiosity, could you describe a bit more the context ? Did
>> vfio_save_complete_precopy() fail ? why ?
> 
> Yep, vfio_save_complete_precopy() failed (but it failed while I was experimenting with an unofficial debug FW).
> 
>>
>> We should propagate errors of .save_live_complete_precopy() handlers as
>> it was done .save_setup handlers(). For 9.1.
> 
> Agreed.
> 
>>
>>> This in turn, will cause a migration hang bug, similar to the bug that
>>> was fixed by commit 22b04245f0d5 ("migration: Join the return path
>>> thread before releasing to_dst_file"), as shutdown() will not be issued
>>> for the return-path channel.
>>
>> yes, but this test :
>>
>>     if (ret < 0) {
>>         goto fail;
>>     }
>>
>> will skip the close_return_path_on_source() call. Won't it ? So I don't
>> understand how it can be an issue. Am I missing something ?
> 
> It will skip the close_return_path_on_source() call in migration_completion(), but there is another close_return_path_on_source() call in migrate_fd_cleanup().

OK. Found it. This is a code path I hadn't explored yet.

Acked-by: Cédric Le Goater <clg@redhat.com>

Thanks,

C.


> 
>>
>>> Fix it by ensuring migration error is set in case of error in
>>> migration_completion().
>>
>> Why didn't you add a reference to commit 9425ef3f990a ?
> 
> I thought this commit didn't introduce this bug, but looking again in the mailing list [1], it kinda did:
> The hang bug was fully fixed by commit 22b04245f0d ("migration: Join the return path thread before releasing to_dst_file") and then 9425ef3f990a re-introduced the bug, but only for migration_completion() case.
> So, you are right, a fixes line with 9425ef3f990a should be added.
> 
> Thanks.
> 
> [1] https://lore.kernel.org/all/20240226203122.22894-1-farosas@suse.de/
> 
>>
>>
>>> Signed-off-by: Avihai Horon <avihaih@nvidia.com>
>>> ---
>>>   migration/migration.c | 10 ++++++++++
>>>   1 file changed, 10 insertions(+)
>>>
>>> diff --git a/migration/migration.c b/migration/migration.c
>>> index 9fe8fd2afd7..b73ae3a72c4 100644
>>> --- a/migration/migration.c
>>> +++ b/migration/migration.c
>>> @@ -2799,6 +2799,7 @@ static void migration_completion(MigrationState *s)
>>>   {
>>>       int ret = 0;
>>>       int current_active_state = s->state;
>>> +    Error *local_err = NULL;
>>>
>>>       if (s->state == MIGRATION_STATUS_ACTIVE) {
>>>           ret = migration_completion_precopy(s, &current_active_state);
>>> @@ -2832,6 +2833,15 @@ static void migration_completion(MigrationState *s)
>>>       return;
>>>
>>>   fail:
>>> +    if (qemu_file_get_error_obj(s->to_dst_file, &local_err)) {
>>> +        migrate_set_error(s, local_err);
>>> +        error_free(local_err);
>>> +    } else if (ret) {
>>> +        error_setg_errno(&local_err, -ret, "Error in migration completion");
>>
>> The 'ret = -1' case could be improved with error_setg(). As a followup.
>>
>> Thanks,
>>
>> C.
>>
>>
>>
>>
>>> +        migrate_set_error(s, local_err);
>>> +        error_free(local_err);
>>> +    }
>>> +
>>>       migration_completion_failed(s, current_active_state);
>>>   }
>>>
>>
>