Re: [PATCH] ceph: avoid putting the realm twice when docoding snaps fails

[Xiubo Li <xiubli@redhat.com>]

On 07/11/2022 18:39, Luís Henriques wrote:
> On Mon, Nov 07, 2022 at 03:17:59PM +0800, xiubli@redhat.com wrote:
>> From: Xiubo Li <xiubli@redhat.com>
>>
>> When decoding the snaps fails it maybe leaving the 'first_realm'
>> and 'realm' pointing to the same snaprealm memory. And then it'll
>> put it twice and could cause random use-after-free, BUG_ON, etc
>> issues.
>>
>> Cc: stable@vger.kernel.org
>> URL: https://tracker.ceph.com/issues/57686
>> Signed-off-by: Xiubo Li <xiubli@redhat.com>
>> ---
>>   fs/ceph/snap.c | 6 ++++--
>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c
>> index 9bceed2ebda3..baf17df05107 100644
>> --- a/fs/ceph/snap.c
>> +++ b/fs/ceph/snap.c
>> @@ -849,10 +849,12 @@ int ceph_update_snap_trace(struct ceph_mds_client *mdsc,
>>   	if (realm_to_rebuild && p >= e)
>>   		rebuild_snap_realms(realm_to_rebuild, &dirty_realms);
>>   
>> -	if (!first_realm)
>> +	if (!first_realm) {
>>   		first_realm = realm;
>> -	else
>> +		realm = NULL;
>> +	} else {
>>   		ceph_put_snap_realm(mdsc, realm);
>> +	}
>>   
>>   	if (p < e)
>>   		goto more;
>> -- 
>> 2.31.1
>>
> This patch looks correct to me.  But I wonder if there's a deeper problem
> there (probably not on the kernel client).  Because the other question is:
> why are we failing to decode the snaps?  But I guess this fix is worth it
> anyway.

Yeah, good question.

At the same time the MDS also crashed [1][2] just before the kernel 
crash was triggered seconds later. And the metadata in cephfs was 
corrupted due to some reasons.

[1] https://tracker.ceph.com/issues/56140

[2] https://tracker.ceph.com/issues/54546

Thanks!

- Xiubo

> Reviewed-by: Luís Henriques <lhenriques@suse.de>
>
>
> Cheers,
> --
> Luís
>