Commit d93231a6bc8a ("ceph: prevent a client from exceeding the MDS
maximum xattr size") was merged before the corresponding MDS-side changes
have been merged. With the introduction of 'bal_rank_mask' in the mdsmap,
the decoding of maps with v>=17 is now incorrect. Fix this by skipping
the 'bal_rank_mask' string decoding.
Fixes: d93231a6bc8a ("ceph: prevent a client from exceeding the MDS maximum xattr size")
Signed-off-by: Luís Henriques <[email protected]>
---
Hi!
This inconsistency was introduced by ceph PR #43284; I think that, before
picking this patch, we need to get PR #46357 merged to avoid new
problems.
Cheers,
--
Luís
fs/ceph/mdsmap.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
index 3fbabc98e1f7..fe4f1a6c3465 100644
--- a/fs/ceph/mdsmap.c
+++ b/fs/ceph/mdsmap.c
@@ -379,6 +379,8 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
ceph_decode_skip_8(p, end, bad_ext);
/* required_client_features */
ceph_decode_skip_set(p, end, 64, bad_ext);
+ /* bal_rank_mask */
+ ceph_decode_skip_string(p, end, bad_ext);
ceph_decode_64_safe(p, end, m->m_max_xattr_size, bad_ext);
} else {
/* This forces the usage of the (sync) SETXATTR Op */
On 27/10/2022 23:28, Luís Henriques wrote:
> Commit d93231a6bc8a ("ceph: prevent a client from exceeding the MDS
> maximum xattr size") was merged before the corresponding MDS-side changes
> have been merged. With the introduction of 'bal_rank_mask' in the mdsmap,
> the decoding of maps with v>=17 is now incorrect. Fix this by skipping
> the 'bal_rank_mask' string decoding.
>
> Fixes: d93231a6bc8a ("ceph: prevent a client from exceeding the MDS maximum xattr size")
> Signed-off-by: Luís Henriques <[email protected]>
> ---
> Hi!
>
> This inconsistency was introduced by ceph PR #43284; I think that, before
> picking this patch, we need to get PR #46357 merged to avoid new
> problems.
>
> Cheers,
> --
> Luís
>
> fs/ceph/mdsmap.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
> index 3fbabc98e1f7..fe4f1a6c3465 100644
> --- a/fs/ceph/mdsmap.c
> +++ b/fs/ceph/mdsmap.c
> @@ -379,6 +379,8 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
> ceph_decode_skip_8(p, end, bad_ext);
> /* required_client_features */
> ceph_decode_skip_set(p, end, 64, bad_ext);
> + /* bal_rank_mask */
> + ceph_decode_skip_string(p, end, bad_ext);
> ceph_decode_64_safe(p, end, m->m_max_xattr_size, bad_ext);
> } else {
> /* This forces the usage of the (sync) SETXATTR Op */
>
Luis,
Because the ceph PR #43284 will break kclient here and your xattr size
patch got merged long time ago, we should fix it in ceph. More detail
please see my comments in:
https://github.com/ceph/ceph/pull/46357#issuecomment-1294290492
Thanks!
- Xiubo
On Fri, Oct 28, 2022 at 09:28:37AM +0800, Xiubo Li wrote:
>
> On 27/10/2022 23:28, Lu?s Henriques wrote:
> > Commit d93231a6bc8a ("ceph: prevent a client from exceeding the MDS
> > maximum xattr size") was merged before the corresponding MDS-side changes
> > have been merged. With the introduction of 'bal_rank_mask' in the mdsmap,
> > the decoding of maps with v>=17 is now incorrect. Fix this by skipping
> > the 'bal_rank_mask' string decoding.
> >
> > Fixes: d93231a6bc8a ("ceph: prevent a client from exceeding the MDS maximum xattr size")
> > Signed-off-by: Lu?s Henriques <[email protected]>
> > ---
> > Hi!
> >
> > This inconsistency was introduced by ceph PR #43284; I think that, before
> > picking this patch, we need to get PR #46357 merged to avoid new
> > problems.
> >
> > Cheers,
> > --
> > Lu?s
> >
> > fs/ceph/mdsmap.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
> > index 3fbabc98e1f7..fe4f1a6c3465 100644
> > --- a/fs/ceph/mdsmap.c
> > +++ b/fs/ceph/mdsmap.c
> > @@ -379,6 +379,8 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
> > ceph_decode_skip_8(p, end, bad_ext);
> > /* required_client_features */
> > ceph_decode_skip_set(p, end, 64, bad_ext);
> > + /* bal_rank_mask */
> > + ceph_decode_skip_string(p, end, bad_ext);
> > ceph_decode_64_safe(p, end, m->m_max_xattr_size, bad_ext);
> > } else {
> > /* This forces the usage of the (sync) SETXATTR Op */
> >
> Luis,
>
> Because the ceph PR #43284 will break kclient here and your xattr size patch
> got merged long time ago, we should fix it in ceph. More detail please see
> my comments in:
>
> https://github.com/ceph/ceph/pull/46357#issuecomment-1294290492
OK, agreed. I'll update this PR to try to fix it on the MDS side
instead. And let's try to have it merged as soon as possible to prevent
further issues.
Cheers,
--
Lu?s
On 28/10/2022 17:05, Lu?s Henriques wrote:
> On Fri, Oct 28, 2022 at 09:28:37AM +0800, Xiubo Li wrote:
>> On 27/10/2022 23:28, Lu?s Henriques wrote:
>>> Commit d93231a6bc8a ("ceph: prevent a client from exceeding the MDS
>>> maximum xattr size") was merged before the corresponding MDS-side changes
>>> have been merged. With the introduction of 'bal_rank_mask' in the mdsmap,
>>> the decoding of maps with v>=17 is now incorrect. Fix this by skipping
>>> the 'bal_rank_mask' string decoding.
>>>
>>> Fixes: d93231a6bc8a ("ceph: prevent a client from exceeding the MDS maximum xattr size")
>>> Signed-off-by: Lu?s Henriques <[email protected]>
>>> ---
>>> Hi!
>>>
>>> This inconsistency was introduced by ceph PR #43284; I think that, before
>>> picking this patch, we need to get PR #46357 merged to avoid new
>>> problems.
>>>
>>> Cheers,
>>> --
>>> Lu?s
>>>
>>> fs/ceph/mdsmap.c | 2 ++
>>> 1 file changed, 2 insertions(+)
>>>
>>> diff --git a/fs/ceph/mdsmap.c b/fs/ceph/mdsmap.c
>>> index 3fbabc98e1f7..fe4f1a6c3465 100644
>>> --- a/fs/ceph/mdsmap.c
>>> +++ b/fs/ceph/mdsmap.c
>>> @@ -379,6 +379,8 @@ struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2)
>>> ceph_decode_skip_8(p, end, bad_ext);
>>> /* required_client_features */
>>> ceph_decode_skip_set(p, end, 64, bad_ext);
>>> + /* bal_rank_mask */
>>> + ceph_decode_skip_string(p, end, bad_ext);
>>> ceph_decode_64_safe(p, end, m->m_max_xattr_size, bad_ext);
>>> } else {
>>> /* This forces the usage of the (sync) SETXATTR Op */
>>>
>> Luis,
>>
>> Because the ceph PR #43284 will break kclient here and your xattr size patch
>> got merged long time ago, we should fix it in ceph. More detail please see
>> my comments in:
>>
>> https://github.com/ceph/ceph/pull/46357#issuecomment-1294290492
> OK, agreed. I'll update this PR to try to fix it on the MDS side
> instead. And let's try to have it merged as soon as possible to prevent
> further issues.
Sounds good!
- Xiubo
> Cheers,
> --
> Lu?s
>