linux-ext4 - Re: help about ext3 read-only issue on ext3(2.6.16.30)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20121220113213.GB3301@quack.suse.cz>
Date:	Thu, 20 Dec 2012 12:32:13 +0100
From:	Jan Kara <jack@...e.cz>
To:	Li Zefan <lizefan@...wei.com>
Cc:	Jan Kara <jack@...e.cz>, qixuan wu <wuqixuan@...il.com>,
	Tao Ma <tm@....ma>, Theodore Ts'o <tytso@....edu>,
	Eric Sandeen <sandeen@...hat.com>,
	Yafang Shao <laoar.shao@...il.com>,
	linux-fsdevel@...r.kernel.org, linux-ext4@...r.kernel.org,
	wuqixuan@...wei.com, xieshuangyi@...wei.com, tao.peng@....com
Subject: Re: help about ext3 read-only issue on ext3(2.6.16.30)

  Hello,

On Mon 17-12-12 18:51:27, Li Zefan wrote:
> >>> last_offset=-1, last_fpos=-1, f_pos=4024
> >>>
> >>> -1 means we hit the bug in the first iteration in the insde while in
> >>> ext3_readdir().
> >>>
> >>> I've checked how ext3_readdir() works and how f_pos, f_version and i_version
> >>> get initialized and modified. Now I'm lost. I really can't see how f_pos got
> >>> corrupted. :(
> >>   Hum, it looks really curious. So f_pos has been 4024 when we entered
> >> ext3_readdir()?
> > 
> > dunno. but what else can be
> > 
> >> Do you know what it was when we last left ext3_readdir()
> >> for that filp? You can store that value in some debug entry added to struct
> >> file... Also any chance we ever hit:
> >>                                 if (version != filp->f_version)
> >>                                         goto revalidate;
> >> I don't think it can ever happen since we hold i_mutex and
> >> generic_file_llseek() takes i_mutex as well. But better be sure.
> >>
> > 
> > Yesterday I've added more debug aids, which convers all the above information
> > mentioned. Actually the code tracks all the places that change f_pos, and
> > I think only lseek() and readdir() can change it.
> > 
> > Now I'm waiting for the bug to happen again, can be several days...
> > 
> 
> The bug was triggered again:
> 
> EXT3-fs error (device sda7): ext3_readdir: bad entry in directory #9372013: rec_len is smaller than minimal - offset=4028, inode=0, rec_len=0, name_len=0
> 
> And I've confirmed f_pos=4028 when we entered ext3_readdir(), while it
> should be 4096.
  OK, interesting.
 
> I wrote a simple ring buffer to track operations on log dirs, and from the
> ring buffer, we can see that there were no lseek, unlink, rename, etc.
> 
> This is correct:
> 
> dir=9372013, seq=1549, spot=readdir_1, f_pos=0, f_pos_delta=0
> dir=9372013, seq=1550, spot=readdir_3, f_pos=0, f_pos_delta=0
> dir=9372013, seq=1551, spot=readdir_5, f_pos=12, f_pos_delta=12
> dir=9372013, seq=1552, spot=readdir_5, f_pos=24, f_pos_delta=12
> ...
> dir=9372013, seq=1595, spot=readdir_5, f_pos=1488, f_pos_delta=28
> dir=9372013, seq=1596, spot=readdir_5, f_pos=1516, f_pos_delta=28
> dir=9372013, seq=1597, spot=readdir_5, f_pos=1556, f_pos_delta=40
> dir=9372013, seq=1598, spot=readdir_5, f_pos=1584, f_pos_delta=28
> ...
> dir=9372013, seq=1627, spot=readdir_5, f_pos=2392, f_pos_delta=28
> dir=9372013, seq=1628, spot=readdir_5, f_pos=4096, f_pos_delta=1704
> dir=9372013, seq=1629, spot=readdir_1, f_pos=4096, f_pos_delta=0
> 
> (readir_1 is the entry of readdir(), and readdir_3 is when we enter
> (f_version != i_version), and readdir_5 is we iterate the dir block)
  Can you post the readdir() code you now run including your logging?
Thanks!

> Then f_pos went wrong suddenly:
> 
> dir=9372013, seq=1676, spot=readdir_5, f_pos=1488, f_pos_delta=28
> dir=9372013, seq=1677, spot=readdir_5, f_pos=1516, f_pos_delta=28
> dir=9372013, seq=1678, spot=readdir_5, f_pos=1556, f_pos_delta=40
> dir=9372013, seq=1679, spot=readdir_5, f_pos=1516, f_pos_delta=28   <-- !!!!!!!!
> dir=9372013, seq=1680, spot=readdir_5, f_pos=1540, f_pos_delta=24
> ...
> dir=9372013, seq=1708, spot=readdir_5, f_pos=2324, f_pos_delta=28
> dir=9372013, seq=1709, spot=readdir_5, f_pos=4028, f_pos_delta=1704
> dir=9372013, seq=1710, spot=readdir_1, f_pos=4028, f_pos_delta=0
> 
> This is odd...
> 
> While f_pos was wrong, offset is always correct, and this is not some
> single-bit error in memory, so someone else changed f_pos? but we were
> holding i_mutex, and we see nothing else except readdir in the ring
> buffer...
  How do you know 'offset' was correct? I don't see it in the ring
buffer... Anyway, this all points to us taking the:
                               if (version != filp->f_version)
                                        goto revalidate;
branch when 'offset' is already advanced but f_pos isn't. Then we don't
enter:
	if (filp->f_version != inode->i_version) {
branch and thus f_pos and 'offset' stay out of sync and problems happen.

But how these strange issues with f_version happen is a mystery for me so
far.

								Honza
-- 
Jan Kara <jack@...e.cz>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html