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

Hi Tao, all,

    I guess it's a memory(or ext3/kenrel) issue. Beause in one
machine, after report this issue, the partition is made to readonly,
we use debugfs to "ls dir", and it's fine. It can list all files
without error. If the disk has issue, when we using ls command, it
will give error also. (The dir name is also using debugfs to get by
issue inode ID.)

    Is there the possibility: one thread(A) is read_dir(directly read
from buffer head), and another thread(B) is creating item, and fill
this buffer header at the same time. During create item, first modify
the last item's rec_len(let it point to next item which initially is
zero), then fill this added new item. Suppose the seq is below :
   1) B: modify last item's rec_len
   2) A: Read last item, rec_len is modified already by B, and it
identify next item is existing.
   3) A: Read new item, all feilds are zero still.
   3) B: fill new item with correct value.

   This may cause problem. Sorry I am not still checking the code
properly. Raise this suppose is just hope ext3 experts can help to
think whether such concurring scenario has problem or not ?
   Any idea or clue is welcome.

Regards & Thanks a lot.
Wuqx

On Tue, Dec 4, 2012 at 11:29 PM, Tao Ma [off-list ref] wrote:

Hi zefan,
On 12/04/2012 09:54 PM, Li Zefan wrote:

quoted

quoted

quoted

We have many x86 boards, and we've been using 2.6.16.60 for a long
time. Before time we occasionally found ext3 was switched to read-only
while services were running, and we took it for granted it must be
some hardware problems.
    But recently this issue happens frequently, both in old boards and
new boards. We've analyzed logs, and in one board we did find
exceptional reboot (but ext3 error happened 9 days after), and in
another board we found mptbase recovery routine, but in all other
boards there's no suspicious output at all.
    The only change with the system is some application updates, and
apps now put more IO burden on disks.
    The error always happened in ext3_readdir, like this:

EXT3-fs error (device sda7): ext3_readdir: bad entry in directory#6685458: rec_len is smaller than minimal - offset=3860, inode=0, rec_len=0, name_len=0
Aborting journal on device sda7.
EXT3-fs error (device sda7) in start_transaction: Readonly filesystem
Aborting journal on device sda7.
ext3_abort called.
EXT3-fs error (device sda7): ext3_journal_start_sb: Detected aborted journal
Remounting filesystem read-only
__journal_remove_journal_head: freeing b_committed_data

We highly doubt it's hardware failures with this frequency in mind, so
we're wondering regarding to this issue if there's some ext3 bug-fix
having merged into mainline but not in our old kernel?

Absolutely there are.  There have been 87 changes just to namei.c since 2.6.16.
You could look through git logs to see if anything looks applicable.

You might try:

ef2b02d3e617cb0400eedf2668f86215e1b0e6af ext34: ensure do_split leaves enough free space in both blocks

I've been asked to investigate this issue. Thanks for the reply!

I found this fix while searching for similar bug reports, but I don't think it
worths trying as we don't use dir_index feature.

I've collected some logs in different machines, and the error was always
triggered in ext3_readdir:

EXT3-fs error (device sda7): ext3_readdir: bad entry in directory #6685458: rec_len is smaller than minimal - offset=3860, inode=0, rec_len=0, name_len=0
EXT3-fs error (device sda7): ext3_readdir: bad entry in directory #9650541: rec_len is smaller than minimal - offset=3960, inode=0, rec_len=0, name_len=0
EXT3-fs error (device sda7): ext3_readdir: bad entry in directory #11124783: rec_len is smaller than minimal - offset=4072, inode=0, rec_len=0, name_len=0
EXT3-fs error (device sda7): ext3_readdir: bad entry in directory #52740880: rec_len is smaller than minimal - offset=4024, inode=0, rec_len=0, name_len=0
EXT3-fs error (device sda7): ext3_readdir: bad entry in directory #52740880: rec_len is smaller than minimal - offset=4084, inode=0, rec_len=0, name_len=0

The last two errors happened on the same machine, and the same inode! One
happened in 11/22 (I was told they had run fsck later on), and one in 12/01.

So now this directory has been fscked to be right? You can try by just
ls this directory and check whether there are any errors in dmesg.

Having said that, as this error happens 2 times for the same inode,
maybe there is a kernel bug. At least as Ted said in another mail, the
end of this buffer head seems to be cleared. So I guess next time when
you see this error, please do:
1. use debugfs to find the disk layout for this dir
2. read the blocks from the block device directly
3. check whether the end of a block(from offset to the end) is zeroed.
4. If yes, I guess there should be a kernel bug and we can go on to
investigate the code.

Thanks
Tao

quoted

The offset is always a bit smaller than blocksize, and all the fields are 0.
I dumped one of the dirs, and only ~1.6K was used (fsck reported no error).

In some machines fsck reported no error at all, and in others filesystems
were corrupted though fixable.

I didn't see any other error messages before this error at all.

Does this remind you of some old ext3 bug?

I'll send you fsck output, dir contents and other logs if u'r interested.

quoted

but to be honest, sticking with such an old kernel means you are largely on your own, or may need contract help if you can't resolve it.

There're numerous machines running old kernels, and many of them are hard to
change. :(

Yesterday they upgrade apps on ~30 machines, and soon after that 5 machines
had filesystem corrupted. However they won't stop upgrading other machines!

On the other hand, we can hardly reproduce this bug in the lab.

So this is critical and urgent. Any help is appreciated.

Regards
Li Zefan

--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html