Linda Walsh wrote:
Carlos E. R. wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256
Hi,
Since yesterday (after an update), I have experimented twice the same problem, on return from hibernation. XFS fails completely, lots of kernel messages. System has to be restarted, recovery is impossible otherwise.
I, as well, would suggest asking about the problem on the xfs list.
They aren't as anal about tainted kernels, & Dave Chinner goes out of is way to look at some of these problems (along with other xfs devels, but I think he's current lead -- not sure/ no quoting me on that!).
But more to the point, he posted some patches for review yesterday morning involving writes at EOF and interlock problems, that might address or overlap your issue...
(People do direct I/O via mapping to the file past the end of file at the same time the file is being extended... things get tricky -- direct I/O doesn't like it's view being 'initialized to zero' after EOF -- even though it shouldn't have been able to map past EOF to begin with... hmmm...
Sounds like multiple programs trying to do write-activity on the same file at the same time.... which OF COURSE produces race conditions. Sounds like the root of such a problem lies in poor I/O management among multiple processes which should be communicating with each other in some way (semaphores?) to indicate when one of them wants to change the file. I don't know if any amount of filesystem code can prevent problems caused by appication programmers who don't even understand that they are creating race conditions in their own file I/O.
Code looks really hairy... one of the other devs just wrote up an FAQ on XFS's speculative preallocation.... next will be one on it's precognitive-disk writing strategy... ;-)
-- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org