[PATCH] ceph: fix incorrectly decoding the mdsmap bug

[PATCH] ceph: fix incorrectly decoding the mdsmap bug

[xiubli]

From: Xiubo Li <xiubli@redhat.com>

When decreasing the 'max_mds' in the cephfs cluster, when the extra
MDS or MDSes are not removed yet, the mdsmap may only decreased the
'max_mds' but still having the that or those MDSes 'in' or in the
export targets list.

Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/mdsmap.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
index 61d67cbcb367..30387733765d 100644
--- a/fs/ceph/mdsmap.c
+++ b/fs/ceph/mdsmap.c
@@ -263,10 +263,6 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
 				goto nomem;
 			for (j = 0; j < num_export_targets; j++) {
 				target = ceph_decode_32(&pexport_targets);
-				if (target >= m->possible_max_rank) {
-					err = -EIO;
-					goto corrupt;
-				}
 				info->export_targets[j] = target;
 			}
 		} else {
-- 
2.31.1

Re: [PATCH] ceph: fix incorrectly decoding the mdsmap bug

[Jeff Layton]

On Fri, 2021-11-05 at 17:34 +0800, xiubli@redhat.com wrote:
> From: Xiubo Li <xiubli@redhat.com>
> 
> When decreasing the 'max_mds' in the cephfs cluster, when the extra
> MDS or MDSes are not removed yet, the mdsmap may only decreased the
> 'max_mds' but still having the that or those MDSes 'in' or in the
> export targets list.
> 
> Signed-off-by: Xiubo Li <xiubli@redhat.com>
> ---
>  fs/ceph/mdsmap.c | 4 ----
>  1 file changed, 4 deletions(-)
> 
> diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
> index 61d67cbcb367..30387733765d 100644
> --- a/fs/ceph/mdsmap.c
> +++ b/fs/ceph/mdsmap.c
> @@ -263,10 +263,6 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
>  				goto nomem;
>  			for (j = 0; j < num_export_targets; j++) {
>  				target = ceph_decode_32(&pexport_targets);
> -				if (target >= m->possible_max_rank) {
> -					err = -EIO;
> -					goto corrupt;
> -				}
>  				info->export_targets[j] = target;
>  			}
>  		} else {

Thanks Xiubo, looks good. Given the severity when mdsmap decoding fails,
I think we should probably mark this for stable too. Let me know if you
have any objections.
-- 
Jeff Layton <jlayton@kernel.org>

Re: [PATCH] ceph: fix incorrectly decoding the mdsmap bug

[Xiubo Li]

On 11/5/21 6:29 PM, Jeff Layton wrote:
> On Fri, 2021-11-05 at 17:34 +0800, xiubli@redhat.com wrote:
>> From: Xiubo Li <xiubli@redhat.com>
>>
>> When decreasing the 'max_mds' in the cephfs cluster, when the extra
>> MDS or MDSes are not removed yet, the mdsmap may only decreased the
>> 'max_mds' but still having the that or those MDSes 'in' or in the
>> export targets list.
>>
>> Signed-off-by: Xiubo Li <xiubli@redhat.com>
>> ---
>>   fs/ceph/mdsmap.c | 4 ----
>>   1 file changed, 4 deletions(-)
>>
>> diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
>> index 61d67cbcb367..30387733765d 100644
>> --- a/fs/ceph/mdsmap.c
>> +++ b/fs/ceph/mdsmap.c
>> @@ -263,10 +263,6 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
>>   				goto nomem;
>>   			for (j = 0; j < num_export_targets; j++) {
>>   				target = ceph_decode_32(&pexport_targets);
>> -				if (target >= m->possible_max_rank) {
>> -					err = -EIO;
>> -					goto corrupt;
>> -				}
>>   				info->export_targets[j] = target;
>>   			}
>>   		} else {
> Thanks Xiubo, looks good. Given the severity when mdsmap decoding fails,
> I think we should probably mark this for stable too. Let me know if you
> have any objections.
Yeah, it should be.