cephfs: expose snapshot creation time as new ceph.snap.btime vxattr #27077
Conversation
There was a problem hiding this comment.
Implementation LGTM.
Reviewed-by: Greg Farnum gfarnum@redhat.com
|
The updated test compares the ceph.snap.btime between an older and newer snapshot. |
|
New version adds Fixes: https://tracker.ceph.com/issues/38838 tags and reworks the infomap.size() == 1 logic. |
|
LGTM |
|
Thanks for the feedback so far - the corresponding kernel changes were proposed (and subsequently acked) via https://www.spinics.net/lists/ceph-devel/msg44894.html . @batrick any further comment on the proposed $secs.$nsecs format? |
|
I've posted the Samba changes to https://lists.samba.org/archive/samba-technical/2019-March/133089.html - would be good to get final feedback on the time format soon if possible, as that appears to be the only thing holding up all three patch-sets. |
* refs/pull/27077/head: test: add libcephfs snap.btime xattr test client: add ceph.snap.btime vxattr mds: carry snapshot creation time with InodeStat Reviewed-by: Patrick Donnelly <pdonnell@redhat.com> Reviewed-by: Greg Farnum <gfarnum@redhat.com>
* refs/pull/27077/head: test: add libcephfs snap.btime xattr test client: add ceph.snap.btime vxattr mds: carry snapshot creation time with InodeStat Reviewed-by: Patrick Donnelly <pdonnell@redhat.com> Reviewed-by: Greg Farnum <gfarnum@redhat.com>
|
Investigating a possible QA failure caused by this PR: http://pulpito.ceph.com/pdonnell-2019-04-02_00:10:11-fs-master-distro-basic-smithi/ |
|
I think I've finally worked out what's going on here: Client::_listxattr() returns a length based on the static _vxattrs_name_size() value. Client::_vxattrs_calcu_name_size() only takes into account whether the vxattr is hidden, not whether the vxattr.exists_cb() returns true/false. |
|
It seems that "ceph.snap.btime" is the only vxattr to set hidden=false and fill the exists_cb(), so I don't think this client bug affects existing releases.
I'll proceed with option (1), but any other feedback would still be appreciated. |
|
Changes since previous version:
|
This allows for a future vxattr exposing the snapshot creation time to users. The InodeStat encoding version is bumped to account for the new snap_btime member. Fixes: https://tracker.ceph.com/issues/38838 Signed-off-by: David Disseldorp <ddiss@suse.de>
The ceph.snap.btime vxattr carries the snapshot creation time for files and directories residing within a snapshot. Fixes: https://tracker.ceph.com/issues/38838 Signed-off-by: David Disseldorp <ddiss@suse.de>
Test that the new "ceph.snap.btime" xattr is present following snapshot, and that the corresponding $secs.$nsecs value changes between old and new snapshots. Fixes: https://tracker.ceph.com/issues/38838 Signed-off-by: David Disseldorp <ddiss@suse.de>
Client::_listxattr() incorrectly returns a length based on the static _vxattrs_name_size() value. _vxattrs_calcu_name_size() only takes into account whether vxattrs are hidden, ignoring vxattr.exists_cb(). When filling the xattr buffer, Client::_listxattr() checks vxattr.hidden and vxattr.exists_cb(). When the latter returns false (as is the case for ceph.snap.btime on non-snapshots), we return a length which is larger than the amount that we wrote to the buffer. Fix this behaviour by always calculating the vxattrs length at runtime, taking both vxattr.hidden and vxattr.exists_cb() into account. Signed-off-by: David Disseldorp <ddiss@suse.de>
_listxattr() now calculates the length of vxattrs dynamically, so these helpers, which incorrectly ignore vxattr.exists_cb(), can be removed. Signed-off-by: David Disseldorp <ddiss@suse.de>
Additionally check that: - a zero buflen returns the needed length - a non-zero buflen less than needed fails - the list-length is non-zero following setxattr Signed-off-by: David Disseldorp <ddiss@suse.de>
Extend LibCephFS.SnapXattrs to also test ceph_listxattr() against snapshot and non-snapshot files. Signed-off-by: David Disseldorp <ddiss@suse.de>
|
Changes since previous version:
|
FWIW, it looks like the kernel ceph_listxattr() function has similar behaviour, so I'll respin my kernel patchset with a corresponding fix once this gets merged. |
* refs/pull/27077/head: test: check listattr for snapshot btime entry test: extend LibCephFS.Xattrs test client: remove unused vxattr length helpers client: fix _listxattr() vxattr buffer length calculation test: add libcephfs snap.btime xattr test client: add ceph.snap.btime vxattr mds: carry snapshot creation time with InodeStat Reviewed-by: Patrick Donnelly <pdonnell@redhat.com> Reviewed-by: Greg Farnum <gfarnum@redhat.com>
Samba requires the snapshot creation time for the purpose of tracking and identifying Previous Versions for files and directories.
Fixes: https://tracker.ceph.com/issues/38838