suspend-to-disk / hibernate corrupts filesystems

[karmaking]

Since recently, I get serious file system corruptions while disk-hibernating/resuming.

Every 3 or 4 times I hibernate, I get a kernel panic because of unmountable root partition and then have to manually fsck (which reports broken superblocks). In these cases, also the home and boot partitions (the latter not being mounted during normal work) will be corrupted.

hibernate.log, messages, dmesg don't show anything obviously related. All SMART values are healthy.

Where to start looking?

amd64, tuxonice-sources-2.6.28-r8, hibernate-script-1.99-r1, hibernate to swap. No initrd, no modules. All ext3.

Thx & cheers
Daniel

[Hu]

What kernel command line do you use to resume? Is there any chance you are mounting the filesystems before the system tries to resume?

[karmaking]

Hmm that's nothing special:

[Hu]

How do you know it is happening during hibernation rather than when trying to resume? TuxOnIce does not unmount filesystems during hibernation, so any filesystems mounted read-write during hibernation will be inconsistent while the machine is halted. When the system resumes, the saved kernel must be made active without mounting the inconsistent filesystems. Your command line looks fine. What does the kernel output during a failed resume? Specifically, what does it say about the swap device?

[karmaking]

Couldn't reproduce it now (it never works when someone's watching :)

I do understand how the mechanism is working. I guess it is that TuxOnIce for some reason refuses to use the saved cache on resume and that the systems wants to boot normally, which it can't because the fs is inconsistent so badly, that it can't fix it on the fly (invalid superblock).

That's why my thought was that it's not sync'ing the fs to disk properly on hibernate or something (?)

Will post kernel output as soon as it appears to happen again.