Disconnected inodes after test xfs/261

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

That said, if unmount is not leaving the log clean, then we've still got an issue we need to get to the bottom of.

Cheers,

Dave.

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

Quoted 42 lines · Dave Chinner

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

Quoted 4 lines

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

No, the problem is still there after mounting and unmounting the
filesystem.

Given what Michael wrote: I'm running xfs_repair version 3.2.1, filesystem is V4.

Quoted 47 lines · Jan Kara

Quoted 42 lines · Dave Chinner

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

Quoted 4 lines

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

Quoted 5 lines

No, the problem is still there after mounting and unmounting the
filesystem.

Given what Michael wrote: I'm running xfs_repair version 3.2.1, filesystem is V4.

Oh, and what might be related: Test xfs/071 passes but xfs_repair
complains like:
*** xfs_repair -n output ***
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
inode 131 - extent offset too large - start 14, count 1, offset 2251799813685247 correcting nextents for inode 131
bad data fork in inode 131
would have cleared inode 131

Quoted 42 lines · Dave Chinner

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

Quoted 4 lines

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

No, the problem is still there after mounting and unmounting the
filesystem.

Given what Michael wrote: I'm running xfs_repair version 3.2.1, filesystem is V4.

When I look via xfs_db at the inode I can see nlink is 1 which looks strange. So maybe the problem is somewhere else than I thought:
xfs_db> inode 132
xfs_db> p
core.magic = 0x494e
core.mode = 0100000
core.version = 2
core.format = 2 (extents)
core.nlinkv2 = 1
core.onlink = 0
core.projid_lo = 0
core.projid_hi = 0
core.uid = 0
core.gid = 0
core.flushiter = 1
core.atime.sec = Thu Dec 18 11:08:55 2014
core.atime.nsec = 510013169
core.mtime.sec = Thu Dec 18 11:08:55 2014
core.mtime.nsec = 510013169
core.ctime.sec = Thu Dec 18 11:08:55 2014
core.ctime.nsec = 510013169
core.size = 0
core.nblocks = 1
core.extsize = 0
core.nextents = 1
core.naextents = 0
core.forkoff = 0
core.aformat = 2 (extents)
core.dmevmask = 0
core.dmstate = 0
core.newrtbm = 0
core.prealloc = 0
core.realtime = 0
core.immutable = 0
core.append = 0
core.sync = 0
core.noatime = 0
core.nodump = 0
core.rtinherit = 0
core.projinherit = 0
core.nosymlinks = 0
core.extsz = 0
core.extszinherit = 0
core.nodefrag = 0
core.filestream = 0
core.gen = 0
next_unlinked = null
u.bmx[0] = [startoff,startblock,blockcount,extentflag] 0:[0,13,1,0]

I have taken xfs_metadump just after test xfs/261 completed and xfs_repair reported error. It is attached.

Quoted 102 lines · Jan Kara

Quoted 42 lines · Dave Chinner

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

Quoted 4 lines

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

No, the problem is still there after mounting and unmounting the
filesystem.

Given what Michael wrote: I'm running xfs_repair version 3.2.1, filesystem is V4.

When I look via xfs_db at the inode I can see nlink is 1 which looks strange. So maybe the problem is somewhere else than I thought:
xfs_db> inode 132
xfs_db> p
core.magic = 0x494e
core.mode = 0100000
core.version = 2
core.format = 2 (extents)
core.nlinkv2 = 1
core.onlink = 0
core.projid_lo = 0
core.projid_hi = 0
core.uid = 0
core.gid = 0
core.flushiter = 1
core.atime.sec = Thu Dec 18 11:08:55 2014
core.atime.nsec = 510013169
core.mtime.sec = Thu Dec 18 11:08:55 2014
core.mtime.nsec = 510013169
core.ctime.sec = Thu Dec 18 11:08:55 2014
core.ctime.nsec = 510013169
core.size = 0
core.nblocks = 1
core.extsize = 0
core.nextents = 1
core.naextents = 0
core.forkoff = 0
core.aformat = 2 (extents)
core.dmevmask = 0
core.dmstate = 0
core.newrtbm = 0
core.prealloc = 0
core.realtime = 0
core.immutable = 0
core.append = 0
core.sync = 0
core.noatime = 0
core.nodump = 0
core.rtinherit = 0
core.projinherit = 0
core.nosymlinks = 0
core.extsz = 0
core.extszinherit = 0
core.nodefrag = 0
core.filestream = 0
core.gen = 0
next_unlinked = null
u.bmx[0] = [startoff,startblock,blockcount,extentflag] 0:[0,13,1,0]

Quoted 2 lines

I have taken xfs_metadump just after test xfs/261 completed and xfs_repair reported error. It is attached.

OK, so I understand better what's going on. The detached inodes are
actually inodes from quota files being created by quotacheck on mount. Test xfs/261 first mounts with uquota - that adds user quota ino and quota feature just fine. But then it mounts with gquota - now we go through different path since xfs_sb_version_hasquota() is already true. We find that group quota isn't initialized though and thus we still need to allocate inode for group quota file. However the change to sb_gquotaino isn't written out because the following test in xfs_sb_quota_to_disk() fails:

	if ((*fields & XFS_SB_GQUOTINO) &&
                                (from->sb_qflags & XFS_GQUOTA_ACCT))
                to->sb_gquotino = cpu_to_be64(from->sb_gquotino);

because from->sb_qflags doesn't have XFS_GQUOTA_ACCT set (only mp->m_qflags has XFS_GQUOTA_ACCT set).

Quoted 115 lines · Jan Kara

Quoted 102 lines · Jan Kara

Quoted 42 lines · Dave Chinner

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

Quoted 4 lines

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

No, the problem is still there after mounting and unmounting the
filesystem.

Given what Michael wrote: I'm running xfs_repair version 3.2.1, filesystem is V4.

When I look via xfs_db at the inode I can see nlink is 1 which looks strange. So maybe the problem is somewhere else than I thought:
xfs_db> inode 132
xfs_db> p
core.magic = 0x494e
core.mode = 0100000
core.version = 2
core.format = 2 (extents)
core.nlinkv2 = 1
core.onlink = 0
core.projid_lo = 0
core.projid_hi = 0
core.uid = 0
core.gid = 0
core.flushiter = 1
core.atime.sec = Thu Dec 18 11:08:55 2014
core.atime.nsec = 510013169
core.mtime.sec = Thu Dec 18 11:08:55 2014
core.mtime.nsec = 510013169
core.ctime.sec = Thu Dec 18 11:08:55 2014
core.ctime.nsec = 510013169
core.size = 0
core.nblocks = 1
core.extsize = 0
core.nextents = 1
core.naextents = 0
core.forkoff = 0
core.aformat = 2 (extents)
core.dmevmask = 0
core.dmstate = 0
core.newrtbm = 0
core.prealloc = 0
core.realtime = 0
core.immutable = 0
core.append = 0
core.sync = 0
core.noatime = 0
core.nodump = 0
core.rtinherit = 0
core.projinherit = 0
core.nosymlinks = 0
core.extsz = 0
core.extszinherit = 0
core.nodefrag = 0
core.filestream = 0
core.gen = 0
next_unlinked = null
u.bmx[0] = [startoff,startblock,blockcount,extentflag] 0:[0,13,1,0]

Quoted 2 lines

I have taken xfs_metadump just after test xfs/261 completed and xfs_repair reported error. It is attached.

OK, so I understand better what's going on. The detached inodes are
actually inodes from quota files being created by quotacheck on mount. Test xfs/261 first mounts with uquota - that adds user quota ino and quota feature just fine. But then it mounts with gquota - now we go through different path since xfs_sb_version_hasquota() is already true. We find that group quota isn't initialized though and thus we still need to allocate inode for group quota file. However the change to sb_gquotaino isn't written out because the following test in xfs_sb_quota_to_disk() fails:

Quoted 6 lines

	if ((*fields & XFS_SB_GQUOTINO) &&
                                (from->sb_qflags & XFS_GQUOTA_ACCT))
                to->sb_gquotino = cpu_to_be64(from->sb_gquotino);

because from->sb_qflags doesn't have XFS_GQUOTA_ACCT set (only mp->m_qflags has XFS_GQUOTA_ACCT set).

I think you've got that wrong, but at the same time got it right. ;)

That is, we get there through xfs_qm_mount_quotas(), which does this in the the write_changes section:

        spin_lock(&mp->m_sb_lock);
        sbf = mp->m_sb.sb_qflags;
        mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
        spin_unlock(&mp->m_sb_lock);

        if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
                if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {

Hence (from->sb_qflags & XFS_GQUOTA_ACCT) evaluates are true because we've copied mp->m_qflags into from->sb_qflags before logging the change. that's the bit you got wrong.

However, the code there only passes XFS_SB_QFLAGS to xfs_log_sb(), and that means (*fileds & XFS_SB_GQUOTINO) evaluates as false, and we fail to write the new group quota inode. SO you are right in that this is the check that is failing. :)

And, again, the reason I wasn't able to reproduce it is that I tested on a kernel running that "get rid of field based superblock logging", which meant the above check was simply:

	if (from->sb_qflags & XFS_GQUOTA_ACCT)
		to->sb_gquotino = cpu_to_be64(from->sb_gquotino);

and, from above, we know that this is set appropriately. Hence the group quota inode was updated and the test did not fail.

FYI, I noticed a bunch of these sb logging field bugs a while back, which is why I wrote the patch to simply get rid of it. You've found two of those bugs in the past few days....

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

That said, if unmount is not leaving the log clean, then we've still got an issue we need to get to the bottom of.

Cheers,

Dave.

Quoted 47 lines · Dave Chinner

Quoted 32 lines · Jan Kara

Hello,

in my test KVM with today's Linus' kernel I'm getting xfs_repair
complaint about disconnected inodes after the test xfs/261 finishes (with success). xfs_repair output is like:
xfs_repair -n /dev/vdb2
Phase 1 - find and verify superblock...
Phase 2 - using internal log
- scan filesystem freespace and inode maps...
- found root inode chunk
Phase 3 - for each AG...
- scan (but don't clear) agi unlinked lists...
- process known inodes and perform inode discovery...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
- process newly discovered inodes...
Phase 4 - check for duplicate blocks...
- setting up duplicate extent list...
- check for inodes claiming duplicate blocks...
- agno = 0
- agno = 1
- agno = 2
- agno = 3
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
- traversing filesystem ...
- traversal finished ...
- moving disconnected inodes to lost+found ...
disconnected inode 132, would move to lost+found disconnected inode 133, would move to lost+found

Quoted 8 lines

Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.
---
Given how trivial test xfs/261 is, it seems like created private mtab files that also get unlinked don't get added to AGI unlinked list before umount. I didn't have a detailed look whether that's possible or not and probably won't get to it before Christmas. So I'm sending this just in case someone more knowledgeable has ideas earlier...

I don't see that here. If you mount/unmount the filesystem, does the warning go away? i.e. xfs_repair -n ignores the contents of the log, so if the unlinked list transactions are in the log then log recovery will make everything good again.

Quoted 6 lines

That said, if unmount is not leaving the log clean, then we've still got an issue we need to get to the bottom of.

Cheers,

Dave.

OK, I played with this a bit, though in all but one case, my XFS stack is new. A bisect will be done to find when this started. kernel 3.10.63 here at work is OK, where new xfsprogs is installed.

In trying out new kernels, I hit the "kernel too old" wall on boot. Therefore, a new system was brewed from slackware-current, and xfstests was installed on top of that. This result is from the "huge-smp" vendor kernel, as it came from the original binary packages:

root@kyhorse:/var/lib/xfstests# uname -a
Linux kyhorse 3.14.24-smp #2 SMP Fri Nov 14 18:34:21 CST 2014 i686 Intel(R) Pentium(R) 4 CPU 1.80GHz GenuineIntel GNU/Linux

root@kyhorse:/var/lib/xfstests# ./check xfs/261
FSTYP         -- xfs (non-debug)
PLATFORM      -- Linux/i686 kyhorse 3.14.24-smp
MKFS_OPTIONS  -- -f -bsize=4096 /dev/sda7
MOUNT_OPTIONS -- /dev/sda7 /mnt/scratch

xfs/261 2s ... 3s
_check_xfs_filesystem: filesystem on /dev/sda7 is inconsistent (c) (see /var/lib/xfstests/results//xfs/261.full) _check_xfs_filesystem: filesystem on /dev/sda7 is inconsistent (r) (see /var/lib/xfstests/results//xfs/261.full)
Ran: xfs/261
Passed all 1 tests

root@kyhorse:/var/lib/xfstests# xfs_repair $SCRQ ATCH_DEV
Phase 1 - find and verify superblock...
Phase 2 - using internal log
        - zero log...
        - scan filesystem freespace and inode maps...
        - found root inode chunk
Phase 3 - for each AG...
        - scan and clear agi unlinked lists...
        - process known inodes and perform inode discovery...
        - agno = 0
        - agno = 1
        - agno = 2
        - agno = 3
        - process newly discovered inodes...
Phase 4 - check for duplicate blocks...
        - setting up duplicate extent list...
        - check for inodes claiming duplicate blocks...
        - agno = 0
        - agno = 1
        - agno = 2
        - agno = 3
Phase 5 - rebuild AG headers and trees...
        - reset superblock...
Phase 6 - check inode connectivity...
        - resetting contents of realtime bitmap and summary inodes
        - traversing filesystem ...
        - traversal finished ...
        - moving disconnected inodes to lost+found ...
disconnected inode 132, moving to lost+found
disconnected inode 133, moving to lost+found
Phase 7 - verify and correct link counts...
Note - quota info will be regenerated on next quota mount.
done

A mount/umount sequence did not change matters. xfsprogs appears to be version 3.1.11, glancing from the tarball names in the a/ directory here:

ftp://ftp.slackware.com/pub/slackware/slackware-current/slackware/

The new setup was just made bootable before I came in to work, and the bisect will start tonight and will hopefully be posted by Monday night.