errors during umount make SysRq fail
Hi,
having removed an USB disk while umount for it was still running (yes, that was stupid) I noticed that umount for this device hangs forever in D state. That would be ok (consequences for user error), however *all* other umounts I attempt also hang in D state and SysRq-U also hangs, resulting in a broken system on the next reboot.
I assume the locking against concurrent umount is there to protect against non-trivial namespace problems and makes sense for normal umounting, but IIRC SysRq-U is there to ensure consistent filesystems on the next startup. Would it make sense to allow SysRq-U to break these locks?
Similar problem exists with SysRq-S. If syncing of one device hangs, it will never proceed to the next one in the list. I agree that one is not trivial (stacked devices, loop et al), but can't we make a best effort to sync at least the physical devices in the machine? Please don't shoot me for talking about physical devices, I know there are some really grey areas trying to define that.
Regards,
Carl-Daniel
Signature · 1 line
http://www.hailfinger.org/
Hi!
Quoted 3 lines
having removed an USB disk while umount for it was still running (yes, that was stupid) I noticed that umount for this device hangs forever in D state. That would be ok (consequences for user error), however
Actually, I do not think that is okay. USB disk removed while you are unmounted it is quite simple case of disk error. umount should handle it.
Pavel
Signature · 2 lines
People were complaining that M$ turns users into beta-testers... ...jr ghea gurz vagb qrirybcref, naq gurl frrz gb yvxr vg gung jnl!
Quoted 9 lines · Pavel Machek
Hi!
Quoted 3 lines
having removed an USB disk while umount for it was still running (yes, that was stupid) I noticed that umount for this device hangs forever in D state. That would be ok (consequences for user error), however
Actually, I do not think that is okay. USB disk removed while you are unmounted it is quite simple case of disk error. umount should handle it.
Nice theory but 2.6.9 has refcount errors in the eh thread and some other problems that mean this doesn't happen. It ought to be ok in 10rc1 providing all the patches are merged now