[Gluster-users] [Gluster-devel] NFS File handles change during upgrade from glusterfs version 3.4.2 to 3.5.3

[Niels de Vos]

On Tue, Feb 24, 2015 at 01:26:08PM +0530, Vijay Bellur wrote:
> On 02/24/2015 02:57 AM, A Ghoshal wrote:
> >Hello,
> >
> >After upgrading from 3.4.2 to 3.5.3, I noticed that all my NFS clients
> >(mounted over tcp) on remote nodes turned stale. I investigated this and
> >got the following recurrent logs from the NFS server:
> >
> >[2015-02-23 20:40:25.834071] W [nfs3-helpers.c:3401:nfs3_log_common_res]
> >0-nfs-nfsv3: XID: 7ead95cb, GETATTR: NFS: 10001(Illegal NFS file
> >handle), POSIX: 14(Bad address)
> >[2015-02-23 20:40:25.834167] E [nfs3.c:301:__nfs3_get_volume_id]
> >(-->/usr/lib64/glusterfs/3.5.3/xlator/nfs/server.so(nfs3_getattr+0x4cb)
> >[0x7fc728b3631a]
> >(-->/usr/lib64/glusterfs/3.5.3/xlator/nfs/server.so(nfs3_getattr_reply+0x37)
> >[0x7fc728b35873]
> >(-->/usr/lib64/glusterfs/3.5.3/xlator/nfs/server.so(nfs3_request_xlator_deviceid+0xb0)
> >[0x7fc728b357ba]))) 0-nfs-nfsv3: invalid argument: xl
> >[2015-02-23 20:40:25.834801] E [nfs3.c:840:nfs3_getattr] 0-nfs-nfsv3:
> >*Bad Handle*
> >
> >Upon investigation, it seems to me that the trouble is with procedure
> >nfs3_fh_validate(). In 3.4.2, the validation is against the following
> >identifiers:
> >
> >#define GF_NFSFH_IDENT0     ':'
> >#define GF_NFSFH_IDENT1     'O'
> >#define GF_NFSFH_IDENT_SIZE   (sizeof(char) * 2)
> >#define GF_NFSFH_STATIC_SIZE  (GF_NFSFH_IDENT_SIZE + (2*sizeof (uuid_t)))
> >
> >While, on 3.5.3, this has expanded to
> >
> >#define GF_NFSFH_IDENT0     ':'
> >#define GF_NFSFH_IDENT1     'O'
> >#define GF_NFSFH_IDENT2     'G'
> >#define GF_NFSFH_IDENT3     'L'
> >#define GF_NFSFH_IDENT_SIZE   (sizeof(char) * 4)
> >#define GF_NFSFH_STATIC_SIZE  (GF_NFSFH_IDENT_SIZE + (2*sizeof (uuid_t)))
> >
> >Due to this, I have to unmount+mount all my nfs clients to get them back
> >into service. Could somebody help me understand why this change was
> >introduced (maybe a reference of the bug ID)? Also, any chance that
> >there's a way to get around this?
> >
> >
> 
> git blame indicates that this change has been introduced as part of [1]. Not
> sure if there is an easy way to get around this without re-mounting.

No, unmounting on the NFS-clients is needed with this change. It should
have been listed in the release notes for updating to 3.5, but that is
obviously missing :-/ I'll make sure to add something for it in the next
notes for 3.5.4.

Thanks,
Niels

> 
> -Vijay
> 
> [1] http://review.gluster.org/4918
> 
> 
> 
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: not available
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20150224/9f34c483/attachment.sig>