Skip to contentSkip to conversation

xfs_ncheck gives outdated and partial result

DISCUSSION Daniele P. 4 participants 4 messages Oct 20, 2006 – Oct 23, 2006

lore
Share & tools
lore

Daniele P.

author
#0 lore raw

Hi All,
I'm here to ask a couple of question regarding xfs_ncheck. I'm using xfs_db version 2.6.20

The first question is:
Is it right that I have to unmount/mount the filesystem before using it (1), or to get the partial right (3) answer? And why it's print only the first file name for inodes pointed by multiple files (2)?

(1)

# mkfs.xfs /dev/hda5
meta-data=/dev/hda5              isize=256    agcount=16, agsize=152617 blks
         =                       sectsz=512
data     =                       bsize=4096   blocks=2441872, imaxpct=25
         =                       sunit=0      swidth=0 blks, unwritten=1
naming   =version 2              bsize=4096
log      =internal log           bsize=4096   blocks=2560, version=1
         =                       sectsz=512   sunit=0 blks
realtime =none                   extsz=65536  blocks=0, rtextents=0
# mount /dev/hda5 /mnt/hda5/
# touch /mnt/hda5/a
# xfs_ncheck -i 131 /dev/hda5
ERROR: The filesystem has valuable metadata changes in a log which needs to
be replayed.  Mount the filesystem to replay the log, and unmount it before
re-running xfs_ncheck.  If you are unable to mount the filesystem, then use
the xfs_repair -L option to destroy the log and attempt a repair.
Note that destroying the log may cause corruption -- please attempt a mount
of the filesystem before doing this.
must run blockget -n first

(2)

# ln /mnt/hda5/a /mnt/hda5/b
# ls -i /mnt/hda5/
131 a  131 b
# umount /mnt/hda5
# mount /dev/hda5 /mnt/hda5
# xfs_ncheck -i 131 /dev/hda5
        131 a

(3)

# ln /mnt/hda5/a /mnt/hda5/c
# xfs_ncheck -i 131 /dev/hda5
        131 a
# rm /mnt/hda5/a
# xfs_ncheck -i 131 /dev/hda5
        131 a
# umount /mnt/hda5
# mount /dev/hda5 /mnt/hda5
# xfs_ncheck -i 131 /dev/hda5
        131 b

Thanks in advance,
Daniele P.

#1 lore raw

Hi Daniele,

Quoted 23 lines · Daniele P.

Hi All,
I'm here to ask a couple of question regarding xfs_ncheck. I'm using xfs_db version 2.6.20

The first question is:
Is it right that I have to unmount/mount the filesystem before using it (1), or to get the partial right (3) answer? And why it's print only the first file name for inodes pointed by multiple files (2)?

(1)

# mkfs.xfs /dev/hda5
meta-data=/dev/hda5              isize=256    agcount=16, agsize=152617 blks
         =                       sectsz=512
data     =                       bsize=4096   blocks=2441872, imaxpct=25
         =                       sunit=0      swidth=0 blks, unwritten=1
naming   =version 2              bsize=4096
log      =internal log           bsize=4096   blocks=2560, version=1
         =                       sectsz=512   sunit=0 blks
realtime =none                   extsz=65536  blocks=0, rtextents=0
# mount /dev/hda5 /mnt/hda5/
# touch /mnt/hda5/a
Quoted 8 lines
# xfs_ncheck -i 131 /dev/hda5
ERROR: The filesystem has valuable metadata changes in a log which needs to
be replayed.  Mount the filesystem to replay the log, and unmount it before
re-running xfs_ncheck.  If you are unable to mount the filesystem, then use
the xfs_repair -L option to destroy the log and attempt a repair.
Note that destroying the log may cause corruption -- please attempt a mount
of the filesystem before doing this.
must run blockget -n first

While it is possible to run xfs_ncheck on a mounted filesystem the results may be inaccurate and misleading. I believe xfs_ncheck accesses the filesystem directly without interpreting the log so if there are outstanding changes in the log then the filesystem may appear inconsistent. Unmounting the filesystem ensures that no changes will occur while xfs_ncheck is running and is the only way to be sure that it will provide an accurate report. If the filesystem is unmounted but there are still outstanding entries in the log then mount and unmount the filesystem again to replay the log.

Quoted 9 lines

(2)

# ln /mnt/hda5/a /mnt/hda5/b
# ls -i /mnt/hda5/
131 a  131 b
# umount /mnt/hda5
# mount /dev/hda5 /mnt/hda5
# xfs_ncheck -i 131 /dev/hda5
        131 a

This could just be a convenience feature to provide the first known path to the inode. Does anyone have a better answer?

Quoted 7 lines

(3) # ln /mnt/hda5/a /mnt/hda5/c

# xfs_ncheck -i 131 /dev/hda5
        131 a
# rm /mnt/hda5/a
# xfs_ncheck -i 131 /dev/hda5
        131 a

The log entry for the delete has not been replayed into the filesystem yet so xfs_ncheck still sees stale data.

Quoted 4 lines
# umount /mnt/hda5
# mount /dev/hda5 /mnt/hda5
# xfs_ncheck -i 131 /dev/hda5
        131 b

Remounting the filesystem forces the log to be replayed into the filesystem and now xfs_ncheck has a consistent view.

Quoted 5 lines

Thanks in advance,
Daniele P.

#2 lore raw
In reply to Daniele P.
Quoted 9 lines · Daniele P.

Hi All,
I'm here to ask a couple of question regarding xfs_ncheck. I'm using xfs_db version 2.6.20

The first question is:
Is it right that I have to unmount/mount the filesystem before using it (1), or to get the partial right (3) answer? And why it's print only the first file name for inodes pointed by multiple files (2)?

I think Lachlan had good answers to your questions... I'm curious, are you using xfs_ncheck on a mounted filesystem for any particular reason?
It seems that maybe if you are concerned about the out-of-date info on
a mounted filesystem, maybe you are attempting to use this from a script or something as part of a larger tool, and that may not be a good idea in general. xfs_ncheck, a wrapper around xfs_db, reads the block device directly, and nothing in the kernel guarantees consistency between the block device address space and the filesystem address space.

-Eric

#3 lore raw
In reply to Daniele P.
Quoted 4 lines · --On 20 October 2006 8:58:08 AM +0200 "Daniele P." <[email protected]> wrote:

Hi All,
I'm here to ask a couple of question regarding xfs_ncheck. I'm using xfs_db version 2.6.20

...

Quoted 3 lines

And why it's print only the first file name for inodes pointed by multiple files (2)?

I presume you are referring to hardlinks - with multiple directory entries pointing to the same inode.
Looking at db/check.c/addname_inode() it looks like it only stores the first name that it comes across during directory processing when scanning happens in the blockget command.

addname_inode(

       inodata_t       *id,
       char            *name,
       int             namelen)
{
       if (!nflag || id->name)  <---- if name exists, then return
               return;
       id->name = xmalloc(namelen + 1);
       memcpy(id->name, name, namelen);
       id->name[namelen] = '\0';
}

So we just get one valid path.

--Tim