Re: Strange Syslog messages - this is NOT an isolated case!

Re: Strange Syslog messages - this is NOT an isolated case!

[Fionn Behrens]

Hi Sean, hi all,

I had a bad system crash last night and had to rebuild-tree afterwards. After
completion I noticed that I always got errors trying to list the /lost+found
directory on the rebuilt partition:

root@rtfm[lost+found] ls -la
ls: 33277_18: Permission denied
ls: 33277_21: Permission denied
total 723505
drwx------    2 root     root          232 Sep 30 09:36 .
drwxrwxr-x   14 root     staff         312 Sep 29 21:19 ..
-rw-r-----    1 fionn    staff        1680 Sep 28  2000 10_5545
-rw-r--r--    1 root     root          738 Apr 23 15:17 33277_16
-rwxr-xr-x    1 fionn    staff     5528244 Mar 16  2001 41953_37254
-rwxr-xr-x    1 fionn    staff    733839360 Dec 13  2001 4625_4657
-rw-------    1 fionn    staff       12288 Aug 28  2001 48215_50042
-rwxr-xr-x    1 fionn    staff     1478380 Dec 17  2001 mplayer

Obviously there are links to some non-existent directory items.
This is the accompanying part of syslog:

Sep 30 11:01:14 rtfm kernel: Using rupasov hash to sort names
Sep 30 11:01:14 rtfm kernel: reiserfs: using 3.5.x disk format
Sep 30 11:01:14 rtfm kernel: ReiserFS version 3.6.25
Sep 30 11:01:19 rtfm kernel: is_leaf: wrong item type for item *3.5*[33277 21
0x1 IND], item_len 4, item_location 1846, free_space(entry_count) 0
Sep 30 11:01:19 rtfm kernel: vs-5150: search_by_key: invalid format found in
block 12064. Fsck?
Sep 30 11:01:19 rtfm kernel: vs-13070: reiserfs_read_inode2: i/o failure
occurred trying to find stat data of [33277 18 0x0 SD]
Sep 30 11:01:19 rtfm kernel: is_leaf: wrong item type for item *3.5*[33277 21
0x1 IND], item_len 4, item_location 1846, free_space(entry_count) 0
Sep 30 11:01:19 rtfm kernel: vs-5150: search_by_key: invalid format found in
block 12064. Fsck?
Sep 30 11:01:19 rtfm kernel: vs-13070: reiserfs_read_inode2: i/o failure
occurred trying to find stat data of [33277 21 0x0 SD]

This looks strangely familiar, doesnt it?

Having gotten curious, I did a ls -lR on the drive and uncovered about a dozen
similar problems outside /lost+found. I have not yet changed anything on the
drive and hope you can give me some advice as I consider this a SERIOUS
problem (wheter my data is in danger or not, problems should be found by an
fsck). As a satiric sidenote, there are messages appearing in the syslog,
suggesting an fsck. But fsck wont find a problem.

So, bottom line:
* The drive is pretty new. badblocks does NOT find anything. In fact it ran   
  flawlessly until I rebuilt the tree.
* reiserfsck --fix-fixable claims there are NO errors.
* I have 2.4.19, too
* I am using reiserfsprogs 3.6.3 from the Debian
  reiserfsprogs 3.6.3-1 package.

If one of the core developers gets interested in the problem we might be able
to find an arrangement for hin to look after the issue on_site. I will gladly
supply any debug output on request.

With kind regards,
		Fionn

P.S.: Sean, do you happen to have tried either 2.4.19-ac4 lately or are you
      using the kernel-preemption-rml patch? Are you using LVM? 
      These are the only unusual things I had running lately and although
      I dont suspect it has to do something with this I'd like to
      back-check for similarities...

P.P.S: As a side note: Is there _any_ way to convert old 3.5/rupasov volumes
       to newer formats without copying half a terabyte over and over again?
-- 
I believe we have been called by history to lead the world.
                                                       G.W. Bush, 2002-03-01

Re: Re: Strange Syslog messages - this is NOT an isolated case!

[Oleg Drokin]

Hello!

On Mon, Sep 30, 2002 at 12:57:53PM +0200, Fionn Behrens wrote:

> * reiserfsck --fix-fixable claims there are NO errors.
> * I have 2.4.19, too
> * I am using reiserfsprogs 3.6.3 from the Debian
>   reiserfsprogs 3.6.3-1 package.

Hm. Can you try reiserfsprogs 3.6.4-pre2 from our ftp server please?
( ftp://ftp.namesys.com/pub/reiserfsprogs/pre )

> P.P.S: As a side note: Is there _any_ way to convert old 3.5/rupasov volumes
>        to newer formats without copying half a terabyte over and over again?

There is no way to change hash, you can convert v3.5 fs to v3.6 one
by using -o conv swith to the mount command (only once).
Note that while FS will become v3.6 one and all new files will be created
in new format, all the old files will remain in old format until you remove and
recreate them.
And of course there is always a way to have whatever feature/tool you need
for reiserfs:
http://namesys.com/support.html

Bye,
    Oleg

Re: Re: Strange Syslog messages - this is NOT an isolated case!

[Vitaly Fertman]

> Having gotten curious, I did a ls -lR on the drive and uncovered about a
> dozen similar problems outside /lost+found. I have not yet changed anything
> on the drive and hope you can give me some advice as I consider this a
> SERIOUS problem (wheter my data is in danger or not, problems should be
> found by an fsck). As a satiric sidenote, there are messages appearing in
> the syslog, suggesting an fsck. But fsck wont find a problem.
>
> So, bottom line:
> * The drive is pretty new. badblocks does NOT find anything. In fact it ran
>   flawlessly until I rebuilt the tree.
> * reiserfsck --fix-fixable claims there are NO errors.
> * I have 2.4.19, too
> * I am using reiserfsprogs 3.6.3 from the Debian
>   reiserfsprogs 3.6.3-1 package.

It looks like a known bug which was fixed in last pre releases. Would you try
reiserfsprogs-3.6.4-pre2 and tell us about the result?

-- 

Thanks,
Vitaly Fertman

Re: Re: Strange Syslog messages - this is NOT an isolated case!

[Fionn Behrens]

On Mon, 30 Sep 2002 15:09:40 +0400
Oleg Drokin <green@namesys.com> wrote:

> Hello!
> 
> On Mon, Sep 30, 2002 at 12:57:53PM +0200, Fionn Behrens wrote:
> 
> > * reiserfsck --fix-fixable claims there are NO errors.
> > * I have 2.4.19, too
> > * I am using reiserfsprogs 3.6.3 from the Debian
> >   reiserfsprogs 3.6.3-1 package.
> 
> Hm. Can you try reiserfsprogs 3.6.4-pre2 from our ftp server please?
> ( ftp://ftp.namesys.com/pub/reiserfsprogs/pre )

Indeed this version of reiserfsck fixed the problem. Thanks for the link.

> > P.P.S: As a side note: Is there _any_ way to convert old 3.5/rupasov
> > volumes
> >        to newer formats without copying half a terabyte over and over
> >        again?
> 
> There is no way to change hash, you can convert v3.5 fs to v3.6 one
> by using -o conv swith to the mount command (only once).
> Note that while FS will become v3.6 one and all new files will be created
> in new format, all the old files will remain in old format until you remove
> and recreate them.

Thanks for the hint as well.

Is there any place to check for an explanation what exactly is different with
newfiles versus old files (save from the source that is)?

I am very pleased to report that it seems none of our files have taken damage
and I am impressed by the speed of response on this list. Thank you all.

Ah - and regarding the hash change: I guess it wont be done with $25, would it
;-) Maybe I am going to find that in a future version som day. Its not that I
would be in a hurry and need it badly. I was just curious.

With kind regards,
		Fionn
-- 
Because of the ad skips.... It's theft. [...] Any time you skip a
[TV] commercial or watch the button you're actually stealing the programming. 
[...] I guess there's a certain amount of tolerance for going to
the bathroom.              *** Jamie Kellner, AOL Time Warner  (serious!)

Re: Re: Strange Syslog messages - this is NOT an isolated case!

[Oleg Drokin]

Hello!

On Mon, Sep 30, 2002 at 03:56:29PM +0200, Fionn Behrens wrote:

> Is there any place to check for an explanation what exactly is different with
> newfiles versus old files (save from the source that is)?

Sure.
Here's some stats: http://www.namesys.com/faq.html#reiserfsspecs

Also on v3.6 FS's you can have volume labels/uuids, extended inode attributes,
better NFS interoperability and may be something else I forgot.
I guess I really need to add that into our FAQ, too.

> Ah - and regarding the hash change: I guess it wont be done with $25, would it

It's unlikely for such a feature to take only one hour of development.

Bye,
    Oleg