It finally got me today.. I’d been very careful with ext4 after all the news about file system corruption (not that I’m blaming the file system) and haven’t been using it on critical systems.
In fact, I’m still not, but I have been testing it. So as I understand it, if the system crashes hard you might not get your data written to disk. Fair enough, we have magic keys to help combat that, and they work well – except when a crash takes out your keyboard input too.
So on the test Apple Xserve machine I’ve installed Fedora onto, I thought I’d install ext4.
Since installing it, I needed to backup some data using an LTO-4 SAS connected tape drive. Unfortunately the SAS card I have is PCI-Express and the IBM servers I use at work are PCI-X only. So I put the card into the Xserve and set up the tape drive (well set up is an exaggeration, it worked out of the box).
Everything has been going just swimmingly, however today I wanted to run some tests on various block sizes and that’s when the trouble started.
It wasn’t the SAS drive that caused it, I was in the process of running the tests I had previously run successfully on it, on the fibre channel connected IBM tape library. When it got to testing 2048, the Fedora machine hung (and hung hard). I was also erasing a tape in the SAS unit at the same time (which kept on going).
Long story short (although it’s a bit late now) magic keys would not save me and I was forced to power off. Upon reboot, my /etc/fstab file was corrupt – some sort of binary, text concoction which I noticed contained (among other things) “/dev/st0,” one of the drives I was using. How the hell that got in there I’ve no idea!
Since fixing the fstab, numerous other GNOME binaries are broken too.. reinstall time tomorrow, this time with ext3! 🙂
6 thoughts on “Bitten by the ext4 “bug””
Wow – that’s pretty bad! Not seen that yet, which kernel are you using ?
Can’t remember now.. the latest at the time, whatever that was.. 2.6.29.something.
I think there was a fair bit of work done in 2.6.30 to address those sorts of issues.
Indeed.. I’m hanging out for an updated kernel at some stage, but I think that Fedora, like Ubuntu and others, doesn’t upgrade major kernel releases. Would be nice to see some of those fixes backported though, if it were possible. Still, for now it’s ext3 for me 🙂
I lost an external to ext3. Couldn’t get any of the superblocks to recover. Kept telling me it was unable to write a superblock.
Problem is they kept making changes to ext3. They should lock it down and not touch it much like Reiserfs3.
http://forum.soft32.com/linux/PULL-ext3-improvements-31-rc8-ftopict492618.html
improvements?
Think they should just call it ext3.1
In all fairness to the developers, they did fork ext3 in order to create ext4, rather than extending it further for that very reason. The change to the default journalling mode in the kernel is in response to issues discovered once ext4 went mainstream. It’s a good change, but in the end it’s the distros which have to decide what to set 🙂
-c