2009-04-22 21:26:50

by Andrew Morton

[permalink] [raw]
Subject: Fw: 2.6.28.9: EXT3/NFS inodes corruption


Is it nfsd, or is it htree?


Begin forwarded message:

Date: Mon, 20 Apr 2009 18:20:18 +0200
From: Sylvain Rochet <[email protected]>
To: [email protected]
Subject: 2.6.28.9: EXT3/NFS inodes corruption


Hi,


We(TuxFamily) are having some inodes corruptions on a NFS server.


So, let's start with the facts.


==== NFS Server

Linux bazooka 2.6.28.9 #1 SMP Mon Mar 30 12:58:22 CEST 2009 x86_64 GNU/Linux

[email protected]:/usr/src# grep EXT3 /boot/config-2.6.28.9
CONFIG_EXT3_FS=y
CONFIG_EXT3_FS_XATTR=y
CONFIG_EXT3_FS_POSIX_ACL=y
CONFIG_EXT3_FS_SECURITY=y

[email protected]:/usr/src# grep NFS /boot/config-2.6.28.9
CONFIG_NFS_FS=y
CONFIG_NFS_V3=y
CONFIG_NFS_V3_ACL=y
# CONFIG_NFS_V4 is not set
CONFIG_NFSD=y
CONFIG_NFSD_V2_ACL=y
CONFIG_NFSD_V3=y
CONFIG_NFSD_V3_ACL=y
# CONFIG_NFSD_V4 is not set
CONFIG_NFS_ACL_SUPPORT=y
CONFIG_NFS_COMMON=y

==> We upgraded from 2.6.26.5 to 2.6.28.9, problem's still here


/dev/md10 on /data type ext3 (rw,noatime,nodiratime,grpquota,commit=5,data=ordered)

==> We used data=writeback, we fallback to data=ordered,
problem's still here


# /etc/exports
/data *(rw,no_root_squash,async,no_subtree_check)
... and lots of exports of subdirs of /data, exported the same way


Process about NFS, on the NFS server.

[email protected]:~# ps aux | grep -E '(nfsd]|lockd]|statd|mountd|idmapd|rquotad|portmap)$'
daemon 1226 0.0 0.0 4824 452 ? Ss Apr11 0:06 /sbin/portmap
root 1703 0.0 0.0 0 0 ? S< 01:29 0:09 [lockd]
root 1704 0.3 0.0 0 0 ? D< 01:29 3:29 [nfsd]
root 1705 0.3 0.0 0 0 ? S< 01:29 3:34 [nfsd]
root 1706 0.3 0.0 0 0 ? S< 01:29 3:32 [nfsd]
root 1707 0.3 0.0 0 0 ? S< 01:29 3:30 [nfsd]
root 1708 0.3 0.0 0 0 ? D< 01:29 3:43 [nfsd]
root 1709 0.3 0.0 0 0 ? D< 01:29 3:43 [nfsd]
root 1710 0.3 0.0 0 0 ? D< 01:29 3:39 [nfsd]
root 1711 0.3 0.0 0 0 ? D< 01:29 3:42 [nfsd]
root 1715 0.0 0.0 5980 576 ? Ss 01:29 0:00 /usr/sbin/rpc.mountd
statd 1770 0.0 0.0 8072 648 ? Ss Apr11 0:00 /sbin/rpc.statd
root 1776 0.0 0.0 23180 536 ? Ss Apr11 0:00 /usr/sbin/rpc.idmapd
root 1785 0.0 0.0 6148 552 ? Ss Apr11 0:00 /usr/sbin/rpc.rquotad

==> We used to run tenths of nfsd daemons, we fallback to 8,
the default, problem's still here
==> There are some 'D' processes because of a running data-check


Block device health:

Apr 3 00:28:20 bazooka kernel: md: data-check of RAID array md10
Apr 3 05:11:59 bazooka kernel: md: md10: data-check done.

Apr 5 01:06:01 bazooka kernel: md: data-check of RAID array md10
Apr 5 05:49:42 bazooka kernel: md: md10: data-check done.

Apr 20 16:27:33 bazooka kernel: md: data-check of RAID array md10

md10 : active raid6 sda[0] sdl[11] sdk[10] sdj[9] sdi[8] sdh[7] sdg[6] sdf[5] sde[4] sdd[3] sdc[2] sdb[1]
1433738880 blocks level 6, 64k chunk, algorithm 2 [12/12] [UUUUUUUUUUUU]
[======>..............] check = 30.1% (43176832/143373888) finish=208.1min speed=8020K/sec

==> Everything seems fine


# df -m
/dev/md10 1378166 87170 1290997 7% /data

# df -i
/dev/md10 179224576 3454822 175769754 2% /data



==== NFS Clients

6x Linux cognac 2.6.28.9-grsec #1 SMP Sun Apr 12 13:06:49 CEST 2009 i686 GNU/Linux
5x Linux martini 2.6.28.9-grsec #1 SMP Tue Apr 14 00:01:30 UTC 2009 i686 GNU/Linux
2x Linux armagnac 2.6.28.9 #1 SMP Tue Apr 14 08:59:12 CEST 2009 i686 GNU/Linux

[email protected]:~$ grep NFS /boot/config-2.6.28.9
CONFIG_NFS_FS=y
CONFIG_NFS_V3=y
CONFIG_NFS_V3_ACL=y
# CONFIG_NFS_V4 is not set
CONFIG_NFSD=y
CONFIG_NFSD_V2_ACL=y
CONFIG_NFSD_V3=y
CONFIG_NFSD_V3_ACL=y
# CONFIG_NFSD_V4 is not set
CONFIG_NFS_ACL_SUPPORT=y
CONFIG_NFS_COMMON=y

==> We upgraded from 2.6.23.16 and 2.6.24.2 (yeah, vmsplice upgrade
;-) to 2.6.28.9, problem's still here


x.x.x.x:/data/... on /data/... type nfs (rw,noexec,nosuid,nodev,async,hard,nfsvers=3,udp,intr,rsize=32768,wsize=32768,timeo=20,addr=x.x.x.x)

==> All NFS exports are mounted this way, sometimes with the 'sync'
option, like web sessions.
==> Those are often mounted from outside of chroots into chroots,
useless detail I think


Process about NFS, on the NFS clients.

[email protected]:~# ps aux | grep -E '(nfsd]|lockd]|statd|mountd|idmapd|rquotad|portmap)$'
daemon 349 0.0 0.0 1904 536 ? Ss Apr12 0:00 /sbin/portmap
statd 360 0.0 0.1 3452 1152 ? Ss Apr12 0:00 /sbin/rpc.statd
root 1190 0.0 0.0 0 0 ? S< Apr12 0:00 [lockd]



==== So, now, going into the problem

The kernel log is not really nice with us, here on the NFS Server:

Mar 22 06:47:14 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Unrecognised inode hash code 52
Mar 22 06:47:14 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Corrupt dir inode 40420228, running e2fsck is recommended.
Mar 22 06:47:16 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Unrecognised inode hash code 52
Mar 22 06:47:16 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Corrupt dir inode 40420228, running e2fsck is recommended.
Mar 22 06:47:19 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Unrecognised inode hash code 52
Mar 22 06:47:19 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Corrupt dir inode 40420228, running e2fsck is recommended.
Mar 22 06:47:19 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Unrecognised inode hash code 52
Mar 22 06:47:19 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Corrupt dir inode 40420228, running e2fsck is recommended.
Mar 22 06:47:19 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Unrecognised inode hash code 52
And so on...

And more recently...
Apr 2 22:19:01 bazooka kernel: EXT3-fs warning (device md10): ext3_unlink: Deleting nonexistent file (40780223), 0
Apr 2 22:19:02 bazooka kernel: EXT3-fs warning (device md10): ext3_unlink: Deleting nonexistent file (40491685), 0
Apr 11 07:23:02 bazooka kernel: EXT3-fs warning (device md10): ext3_unlink: Deleting nonexistent file (174301379), 0
Apr 20 08:13:32 bazooka kernel: EXT3-fs warning (device md10): ext3_unlink: Deleting nonexistent file (54942021), 0


Not much stuff in the kernel log of NFS clients, history is quite lost,
but we got some of them:

....................: NFS: Buggy server - nlink == 0!


== Going deeper into the problem

Something like that is quite common:

[email protected]:/data/...# ls -la
total xxx
drwxrwx--- 2 xx xx 4096 2009-04-20 03:48 .
drwxr-xr-x 7 root root 4096 2007-01-21 13:15 ..
-rw-r--r-- 1 root root 0 2009-04-20 03:48 access.log
-rw-r--r-- 1 root root 70784145 2009-04-20 00:11 access.log.0
-rw-r--r-- 1 root root 6347007 2009-04-10 00:07 access.log.10.gz
-rw-r--r-- 1 root root 6866097 2009-04-09 00:08 access.log.11.gz
-rw-r--r-- 1 root root 6410119 2009-04-08 00:07 access.log.12.gz
-rw-r--r-- 1 root root 6488274 2009-04-07 00:08 access.log.13.gz
?--------- ? ? ? ? ? access.log.14.gz
?--------- ? ? ? ? ? access.log.15.gz
?--------- ? ? ? ? ? access.log.16.gz
?--------- ? ? ? ? ? access.log.17.gz
-rw-r--r-- 1 root root 6950626 2009-04-02 00:07 access.log.18.gz
?--------- ? ? ? ? ? access.log.19.gz
-rw-r--r-- 1 root root 6635884 2009-04-19 00:11 access.log.1.gz
?--------- ? ? ? ? ? access.log.20.gz
?--------- ? ? ? ? ? access.log.21.gz
?--------- ? ? ? ? ? access.log.22.gz
?--------- ? ? ? ? ? access.log.23.gz
?--------- ? ? ? ? ? access.log.24.gz
?--------- ? ? ? ? ? access.log.25.gz
?--------- ? ? ? ? ? access.log.26.gz
-rw-r--r-- 1 root root 6616546 2009-03-24 00:07 access.log.27.gz
?--------- ? ? ? ? ? access.log.28.gz
?--------- ? ? ? ? ? access.log.29.gz
-rw-r--r-- 1 root root 6671875 2009-04-18 00:12 access.log.2.gz
?--------- ? ? ? ? ? access.log.30.gz
-rw-r--r-- 1 root root 6347518 2009-04-17 00:10 access.log.3.gz
-rw-r--r-- 1 root root 6569714 2009-04-16 00:12 access.log.4.gz
-rw-r--r-- 1 root root 7170750 2009-04-15 00:11 access.log.5.gz
-rw-r--r-- 1 root root 6676518 2009-04-14 00:12 access.log.6.gz
-rw-r--r-- 1 root root 6167458 2009-04-13 00:11 access.log.7.gz
-rw-r--r-- 1 root root 5856576 2009-04-12 00:10 access.log.8.gz
-rw-r--r-- 1 root root 6644142 2009-04-11 00:07 access.log.9.gz


[email protected]:/data/...# cat * # output filtered, only errors
cat: access.log.14.gz: Stale NFS file handle
cat: access.log.15.gz: Stale NFS file handle
cat: access.log.16.gz: Stale NFS file handle
cat: access.log.17.gz: Stale NFS file handle
cat: access.log.19.gz: Stale NFS file handle
cat: access.log.20.gz: Stale NFS file handle
cat: access.log.21.gz: Stale NFS file handle
cat: access.log.22.gz: Stale NFS file handle
cat: access.log.23.gz: Stale NFS file handle
cat: access.log.24.gz: Stale NFS file handle
cat: access.log.25.gz: Stale NFS file handle
cat: access.log.26.gz: Stale NFS file handle
cat: access.log.28.gz: Stale NFS file handle
cat: access.log.29.gz: Stale NFS file handle
cat: access.log.30.gz: Stale NFS file handle


"Stale NFS file handle"... on the NFS Server... hummm...


== Other facts

fsck.ext3 fixed the filesystem but didn't fix the problem.

mkfs.ext3 didn't fix the problem either.

It only concerns files which have been recently modified, logs, awstats
hashfiles, websites caches, sessions, locks, and such.

It mainly happens to files which are created on the NFS server itself,
but it's not a hard rule.

Keeping inodes into servers' cache seems to prevent the problem to happen.
( yeah, # while true ; do ionice -c3 find /data -size +0 > /dev/null ; done )


Hummm, it seems to concern files which are quite near to each others,
let's check that:

Let's build up an inode "database"

# find /data -printf '%i %p\n' > /root/inodesnumbers


Let's check how inodes numbers are distributed:

# cat /root/inodesnumbers | perl -e 'use Data::Dumper; my @pof; while(<>){my ( $inode ) = ( $_ =~ /^(\d+)/ ); my $hop = int($inode/1000000); $pof[$hop]++; }; for (0 .. $#pof) { print $_." = ".($pof[$_]/10000)."%\n" }'
[... lot of quite unused inodes groups]
53 = 3.0371%
54 = 26.679% <= mailboxes
55 = 2.7026%
[... lot of quite unused inodes groups]
58 = 1.3262%
59 = 27.3211% <= mailing lists archives
60 = 5.5159%
[... lot of quite unused inodes groups]
171 = 0.0631%
172 = 0.1063%
173 = 27.2895% <=
174 = 44.0623% <=
175 = 45.6783% <= websites files
176 = 45.8247% <=
177 = 36.9376% <=
178 = 6.3294%
179 = 0.0442%

Hummm, all the files are using the same inodes "groups".
(groups of a million of inodes)

We use to fix broken folders by moving them to a quarantine folder and
by restoring disappeared files from the backup.

So, let's check corrupted inodes number from the quarantine folder:

[email protected]:/data/path/to/rep/of/quarantine/folders# find . -mindepth 1 -maxdepth 1 -printf '%i\n' | sort -n
174293418
174506030
174506056
174506073
174506081
174506733
174507694
174507708
174507888
174507985
174508077
174508083
176473056
176473062
176473064

Humm... those are quite near to each other 17450... 17647... and are of
course in the most used inodes "groups"...


Open question: are NFS clients can steal inodes numbers from each others ?


I am not sure whether my bug report is good, feel free to ask questions ;)

Best regards,
Sylvain


Attachments:
signature.asc (190.00 B)

2009-04-22 22:45:04

by Theodore Ts'o

[permalink] [raw]
Subject: Re: Fw: 2.6.28.9: EXT3/NFS inodes corruption

On Wed, Apr 22, 2009 at 02:24:24PM -0700, Andrew Morton wrote:
>
> Is it nfsd, or is it htree?

Well, I see evidence in the bug report of corrupted directory data
structures, so I don't think it's an NFS problem. I would want to
rule out hardware flakiness, though. This could easily be caused by a
hardware problem.

> The kernel log is not really nice with us, here on the NFS Server:
>
> Mar 22 06:47:14 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Unrecognised inode hash code 52
> Mar 22 06:47:14 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Corrupt dir inode 40420228, running e2fsck is recommended.

Evidence of a corrupted directory entry. We would need to look at the
directory to see whether the directory just ad a few bits flipped, or
is pure garbage. The ext3 htree code should do a better job printing
out diagnostics, and flagging the filesystem as corrupt here.

> Apr 2 22:19:02 bazooka kernel: EXT3-fs warning (device md10): ext3_unlink: Deleting nonexistent file (40491685), 0

More evidence of a corrupted directory.

> == Going deeper into the problem
>
> Something like that is quite common:
>
> [email protected]:/data/...# ls -la
> total xxx
> drwxrwx--- 2 xx xx 4096 2009-04-20 03:48 .
> drwxr-xr-x 7 root root 4096 2007-01-21 13:15 ..
> -rw-r--r-- 1 root root 0 2009-04-20 03:48 access.log
> -rw-r--r-- 1 root root 70784145 2009-04-20 00:11 access.log.0
> -rw-r--r-- 1 root root 6347007 2009-04-10 00:07 access.log.10.gz
> -rw-r--r-- 1 root root 6866097 2009-04-09 00:08 access.log.11.gz
> -rw-r--r-- 1 root root 6410119 2009-04-08 00:07 access.log.12.gz
> -rw-r--r-- 1 root root 6488274 2009-04-07 00:08 access.log.13.gz
> ?--------- ? ? ? ? ? access.log.14.gz
> ?--------- ? ? ? ? ? access.log.15.gz
> ?--------- ? ? ? ? ? access.log.16.gz

This is on the client side; what happens when you look at the same
directory from the server side?

>
> fsck.ext3 fixed the filesystem but didn't fix the problem.
>

What do you mean by that? That subsequently, you started seeing
filesystem corruptions again? Can you send me the output of
fsck.ext3? The sorts of filesystem corruption problems which are
fixed by e2fsck are important in figuring out what is going on.

What you if you run fsck.ext3 (aka e2fsck) twice. Once after fixing
fixing all of the problems, and then a second time afterwards. Do the
problems stay fixed?

Suppose you try mounting the filesystem read-only; are things stable
while it is mounted read-only.

> Let's check how inodes numbers are distributed:
>
> # cat /root/inodesnumbers | perl -e 'use Data::Dumper; my @pof; while(<>){my ( $inode ) = ( $_ =~ /^(\d+)/ ); my $hop = int($inode/1000000); $pof[$hop]++; }; for (0 .. $#pof) { print $_." = ".($pof[$_]/10000)."%\n" }'
> [... lot of quite unused inodes groups]
> 53 = 3.0371%
> 54 = 26.679% <= mailboxes
> 55 = 2.7026%
> [... lot of quite unused inodes groups]
> 58 = 1.3262%
> 59 = 27.3211% <= mailing lists archives
> 60 = 5.5159%
> [... lot of quite unused inodes groups]
> 171 = 0.0631%
> 172 = 0.1063%
> 173 = 27.2895% <=
> 174 = 44.0623% <=
> 175 = 45.6783% <= websites files
> 176 = 45.8247% <=
> 177 = 36.9376% <=
> 178 = 6.3294%
> 179 = 0.0442%

Yes, that's normal. BTW, you can get this sort of information much
more easily simply by using the "dumpe2fs" program.

> We use to fix broken folders by moving them to a quarantine folder and
> by restoring disappeared files from the backup.
>
> So, let's check corrupted inodes number from the quarantine folder:
>
> [email protected]:/data/path/to/rep/of/quarantine/folders# find . -mindepth 1 -maxdepth 1 -printf '%i\n' | sort -n
> 174293418
> 174506030
> 174506056
> 174506073
> 174506081
> 174506733
> 174507694
> 174507708
> 174507888
> 174507985
> 174508077
> 174508083
> 176473056
> 176473062
> 176473064
>
> Humm... those are quite near to each other 17450... 17647... and are of
> course in the most used inodes "groups"...

When you say "corrupted inodes", how are they corrupted? The errors
you showed on the server side looked like directory corruptions. Were
these inodes directories or data files?


This really smells like a hardware problem to me; my recommendation
would be to run memory tests and also hard drive tests. I'm going to
guess it's more likely the problem is with your hard drives as opposed
to memory --- that would be consistent with your observation that
trying to keep the inodes in memory seems to help.

- Ted

2009-04-23 00:11:39

by Theodore Y. Ts'o

[permalink] [raw]
Subject: Re: Fw: 2.6.28.9: EXT3/NFS inodes corruption

On Thu, Apr 23, 2009 at 01:48:23AM +0200, Sylvain Rochet wrote:
> >
> > This is on the client side; what happens when you look at the same
> > directory from the server side?
>
> This is on the server side ;)
>

On the server side, that means you also an inode table block look
corrupted. I'm pretty sure that if you used debugfs to examine those
blocks you would have seen that the inodes were completely garbaged.
Depending on the inode size, and assuming a 4k block size, there are
typically 128 or 64 inodes in a 4k block, so if you were to look at
the inodes by inode number, you normally find that adjacent inodes are
corrupted within a 4k block. Of course, this just tells us what had
gotten damaged; whether it was damanged by a kernel bug, a memory bug,
a hard drive or controller failure (and there are multiple types of
storage stack failures; complete garbage getting written into the
right place, and the right data getting written into the wrong place).

> Well, this is the inode numbers of directories with entries pointing on
> inexisting inodes, of course we cannot delete these directories anymore
> through a regular recursive deletion (well, without debugfs ;).
> Considering the amount of inodes, this is quite a very low corruption
> rate.

Well, sure, but any amount of corruption is extremely troubling....

> Yes, this is what we thought too, especially because we use ext3/nfs for
> a very long time without problem like that. I moved all the data to the
> backup array so we can now do read-write tests on the primary one
> without impacting much the production.
>
> So, let's check the raid6 array, well, this is going to take a few days.
>
> # badblocks -w -s /dev/md10
>
> If everything goes well I will check disk by disk.
>
> By the way, if such corruptions doesn't happen on the backup storage
> array we can conclude to a hardware problem around the primary one, but,
> we are not going to be able to conclude before a few weeks.

Good luck!!

- Ted

2009-04-22 23:48:23

by Sylvain Rochet

[permalink] [raw]
Subject: Re: Fw: 2.6.28.9: EXT3/NFS inodes corruption

Hi,


On Wed, Apr 22, 2009 at 06:44:55PM -0400, Theodore Tso wrote:
> On Wed, Apr 22, 2009 at 02:24:24PM -0700, Andrew Morton wrote:
> >
> > Is it nfsd, or is it htree?
>
> Well, I see evidence in the bug report of corrupted directory data
> structures, so I don't think it's an NFS problem. I would want to
> rule out hardware flakiness, though. This could easily be caused by a
> hardware problem.
>
> > The kernel log is not really nice with us, here on the NFS Server:
> >
> > Mar 22 06:47:14 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Unrecognised inode hash code 52
> > Mar 22 06:47:14 bazooka kernel: EXT3-fs warning (device md10): dx_probe: Corrupt dir inode 40420228, running e2fsck is recommended.
>
> Evidence of a corrupted directory entry. We would need to look at the
> directory to see whether the directory just ad a few bits flipped, or
> is pure garbage. The ext3 htree code should do a better job printing
> out diagnostics, and flagging the filesystem as corrupt here.
>
> > Apr 2 22:19:02 bazooka kernel: EXT3-fs warning (device md10): ext3_unlink: Deleting nonexistent file (40491685), 0
>
> More evidence of a corrupted directory.
>
> > == Going deeper into the problem
> >
> > Something like that is quite common:
> >
> > [email protected]:/data/...# ls -la
> > total xxx
> > drwxrwx--- 2 xx xx 4096 2009-04-20 03:48 .
> > drwxr-xr-x 7 root root 4096 2007-01-21 13:15 ..
> > -rw-r--r-- 1 root root 0 2009-04-20 03:48 access.log
> > -rw-r--r-- 1 root root 70784145 2009-04-20 00:11 access.log.0
> > -rw-r--r-- 1 root root 6347007 2009-04-10 00:07 access.log.10.gz
> > -rw-r--r-- 1 root root 6866097 2009-04-09 00:08 access.log.11.gz
> > -rw-r--r-- 1 root root 6410119 2009-04-08 00:07 access.log.12.gz
> > -rw-r--r-- 1 root root 6488274 2009-04-07 00:08 access.log.13.gz
> > ?--------- ? ? ? ? ? access.log.14.gz
> > ?--------- ? ? ? ? ? access.log.15.gz
> > ?--------- ? ? ? ? ? access.log.16.gz
>
> This is on the client side; what happens when you look at the same
> directory from the server side?

This is on the server side ;)


> > fsck.ext3 fixed the filesystem but didn't fix the problem.
>
> What do you mean by that? That subsequently, you started seeing
> filesystem corruptions again?

Yes, a few days later, sorry for being unclear.


> Can you send me the output of fsck.ext3? The sorts of filesystem
> corruption problems which are fixed by e2fsck are important in
> figuring out what is going on.

Unfortunately I can't, we fsck'ed it up quite in a hurry, but
/data/lost+found/ was filled up well with orphaned blocks which appeared
to be part of the disappeared files.

We first thought it was a problem caused by a not-so-recent power
outage, and that a simple fsck would fix that. But a further look up on
cron job mails told us we were wrong ;)


> What you if you run fsck.ext3 (aka e2fsck) twice. Once after fixing
> fixing all of the problems, and then a second time afterwards. Do the
> problems stay fixed?

We ran fsck two times in row, and the second check didn't find any
mistake. We thought, "so, it's fixed!"... erm. Actually it was one month
ago, corruption happens from time to time, several days to one week can
pass without worry.


> Suppose you try mounting the filesystem read-only; are things stable
> while it is mounted read-only.

Humm this is not easy to find out, we should wait at least one week to
conclude.


> > Let's check how inodes numbers are distributed:
> >
> > # cat /root/inodesnumbers | perl -e 'use Data::Dumper; my @pof; while(<>){my ( $inode ) = ( $_ =~ /^(\d+)/ ); my $hop = int($inode/1000000); $pof[$hop]++; }; for (0 .. $#pof) { print $_." = ".($pof[$_]/10000)."%\n" }'
> > [... lot of quite unused inodes groups]
> > 53 = 3.0371%
> > 54 = 26.679% <= mailboxes
> > 55 = 2.7026%
> > [... lot of quite unused inodes groups]
> > 58 = 1.3262%
> > 59 = 27.3211% <= mailing lists archives
> > 60 = 5.5159%
> > [... lot of quite unused inodes groups]
> > 171 = 0.0631%
> > 172 = 0.1063%
> > 173 = 27.2895% <=
> > 174 = 44.0623% <=
> > 175 = 45.6783% <= websites files
> > 176 = 45.8247% <=
> > 177 = 36.9376% <=
> > 178 = 6.3294%
> > 179 = 0.0442%
>
> Yes, that's normal. BTW, you can get this sort of information much
> more easily simply by using the "dumpe2fs" program.

Yep, exactly.


> > We use to fix broken folders by moving them to a quarantine folder and
> > by restoring disappeared files from the backup.
> >
> > So, let's check corrupted inodes number from the quarantine folder:
> >
> > [email protected]:/data/path/to/rep/of/quarantine/folders# find . -mindepth 1 -maxdepth 1 -printf '%i\n' | sort -n
> > 174293418
> > 174506030
> > 174506056
> > 174506073
> > 174506081
> > 174506733
> > 174507694
> > 174507708
> > 174507888
> > 174507985
> > 174508077
> > 174508083
> > 176473056
> > 176473062
> > 176473064
> >
> > Humm... those are quite near to each other 17450... 17647... and are of
> > course in the most used inodes "groups"...
>
> When you say "corrupted inodes", how are they corrupted? The errors
> you showed on the server side looked like directory corruptions. Were
> these inodes directories or data files?

Well, this is the inode numbers of directories with entries pointing on
inexisting inodes, of course we cannot delete these directories anymore
through a regular recursive deletion (well, without debugfs ;).
Considering the amount of inodes, this is quite a very low corruption
rate.


> This really smells like a hardware problem to me; my recommendation
> would be to run memory tests and also hard drive tests. I'm going to
> guess it's more likely the problem is with your hard drives as opposed
> to memory --- that would be consistent with your observation that
> trying to keep the inodes in memory seems to help.

Yes, this is what we thought too, especially because we use ext3/nfs for
a very long time without problem like that. I moved all the data to the
backup array so we can now do read-write tests on the primary one
without impacting much the production.


So, let's check the raid6 array, well, this is going to take a few days.

# badblocks -w -s /dev/md10


If everything goes well I will check disk by disk.


By the way, if such corruptions doesn't happen on the backup storage
array we can conclude to a hardware problem around the primary one, but,
we are not going to be able to conclude before a few weeks.


Thanks Theodore, your help is appreciated ;)


Sylvain


Attachments:
(No filename) (6.46 kB)
signature.asc (189.00 B)
Digital signature
Download all attachments

2009-04-23 23:14:14

by Sylvain Rochet

[permalink] [raw]
Subject: Re: Fw: 2.6.28.9: EXT3/NFS inodes corruption

Hi,


On Wed, Apr 22, 2009 at 08:11:39PM -0400, Theodore Tso wrote:
>
> On the server side, that means you also an inode table block look
> corrupted. I'm pretty sure that if you used debugfs to examine those
> blocks you would have seen that the inodes were completely garbaged.

Yep, I destroyed all evidences by using badblocks in read-write mode,
but in case of real need of them we just have to put the production back
on the primary array and wait a few days.


> Depending on the inode size, and assuming a 4k block size, there are
> typically 128 or 64 inodes in a 4k block,

4k block size
128 bytes/inode

so 32 inodes per 4k block in our case ?

Since the new default is 256 bytes/inode and values of less than 128 are
not allowed, how is it possible to store 64 or 128 inodes in a 4k block ?
(Maybe I miss something :p)


> so if you were to look at the inodes by inode number, you normally
> find that adjacent inodes are corrupted within a 4k block. Of course,
> this just tells us what had gotten damaged; whether it was damanged by
> a kernel bug, a memory bug, a hard drive or controller failure (and
> there are multiple types of storage stack failures; complete garbage
> getting written into the right place, and the right data getting
> written into the wrong place).

Yes, this is not going to be easy to find out what is responsible, but
based on the probability of hardware that use to fail easily, let's
point out one of the harddrive :-)


> Well, sure, but any amount of corruption is extremely troubling....

Yep ;-)


> > By the way, if such corruptions doesn't happen on the backup storage
> > array we can conclude to a hardware problem around the primary one, but,
> > we are not going to be able to conclude before a few weeks.
>
> Good luck!!

Thanks, actually this isn't so bad, we enjoy having backup hardware
(The things we always consider as useless until we -really- need it --
"Who said like backups ? I heard it from the end of the room." ;-)

By the way, the badblocks check is going to take 12 days considering the
current rate. However I ran some data checks of the raid6 array in the
past, mainly when the filesystem was corrupted and every check
succeeded. Maybe the raid6 driver computed another parity strides by
reading corrupted data.


Sylvain


Attachments:
(No filename) (2.26 kB)
signature.asc (189.00 B)
Digital signature
Download all attachments