It'll be funny if it was not so sad [LWN.net]

It'll be funny if it was not so sad

Posted Jun 18, 2009 12:16 UTC (Thu) by khim (subscriber, #9252)
In reply to: Maybe btrfs has no fsck, by nye
Parent article: What ever happened to chunkfs?

The 'problem' of mistaking a superblock in some image you have somewhere for the start of a new fs is *exactly what you asked it to do*, so those who complained so loudly about it really have nobody to blame but themselves.

Yup. They did one mistake, but that mistake was grave: they assumed they can trust reiserfs. I've seen few cases where tiny nimber of badblocks killed reiserfs completely: nothing except this "cobble it together if you can" option worked and "cobbled together" filesystem was a mess (because there were some virtual images on that filesystem). Note: this exactly type of corruption SSD shows in real world. If reiserfs's "gentle" fsck does not work and if "last resort" approach is unusable then the only solution is to switch to other filesystem...

(Log in to post comments)

The problem is: disk DO fail and reiserfs is TOTALLY unready

Posted Jun 19, 2009 10:58 UTC (Fri) by khim (subscriber, #9252) [Link]

Meh, the only FS I've ever lost data to - aside from understandably unrecoverable data on damaged disk blocks

But the bad blocks DO exist in real life - you can not just ignore them! Reiserfs design NEVER ever considered this facet of life: if you have ONE bad block in wrong place - you are screwed 100%. When disk size is 512 bytes and HDD size of 2TiB loss of a 0.0000000003% of your data means 100% of your stuff is lost. This is not even funny.

XFS is also not a good idea (I was biten by it too) - but here we have bad implementation, not bad design. Implementation can be fixed, design mistake is unfixable.

The problem is: disk DO fail and reiserfs is TOTALLY unready

Posted Jun 22, 2009 16:22 UTC (Mon) by nye (guest, #51576) [Link]

>But the bad blocks DO exist in real life - you can not just ignore them!

I certainly don't disagree; I was referring to the data actually on the damaged part of the disk, which I wouldn't reasonably expect to be able to recover without great expense.

It'll be funny if it was not so sad

Posted Jun 21, 2009 6:47 UTC (Sun) by cventers (subscriber, #31465) [Link]

I was a reiser3 user for quite a while and often sung its praises. A few
years into that stretch, my PC started having stability problems which I
tracked to bad RAM.

I caught the bad RAM pretty quickly, and considered myself lucky that I
hadn't obviously lost any big chunks of data... I had seen the ReiserFS
journal check making some noise in dmesg but everything seemed to work.

However, replacing the RAM didn't solve the stability problem. The nature
of the problem changed... it became a random system freeze. At the time, I
didn't realize that I had a new problem - hidden filesystem corruption.

After a couple of big scares with "md" after the system had randomly
frozen, I made a full backup of the filesystem. I continued using the
computer, but the stability problem seemed to be getting worse. I
installed a brand new monster power supply and over the course of the next
month or two I burned a lot of money replacing the rest of the system,
thoroughly confused that I hadn't nailed the problem. (Mockingly, it often
seemed that replacing a part would make the problem go away for a day or
two, leading me to believe I'd fixed it until it slapped me in the face in
the middle of my work yet again.)

My full filesystem backup became handy after I was unable to bring the
filesystem online one time. reiserfsck made lots of noise about problems
with my data and was unable to repair it. I was frustrated to have lost a
month's worth of data, but thrilled that I had a backup at all.

Sadly, I lost the filesystem a few more times and burned even more time
and money on the computer before I realized that with all the hardware
having been replaced, I needed to consider what I had considered to be the
unlikely cause: the software. I became suspicious of reiserfs. This time,
rather than restoring again from my old reiserfs image, I made an ext3
partition, mounted the reiserfs image read-only and migrated.

My system never froze again.

I don't know enough about the reiserfs design to know how plausible my
hypothesis is, but it seems that the bad RAM I dealt with a long time ago
had led to a reiserfs filesystem which was "doomed". I assume the bad RAM
provided the initial corruption, some sort of corruption that made the
reiserfs kernel code fall on its face. Sometimes, the system accessed the
"wrong" bit of corrupted data and the kernel would panic or hang somewhere
inside reiserfs, spreading the corruption in the process.

There's a shocking bit of irony in this particular failure mode. Because
the backup I always restored from was a reiserfs image taken with dd, the
only way I was ever going to escape the crashes and repeated loss of my
data was to abandon reiserfs.