[Fwd: PATCH for Corrupted IO on all block devices]
PATCH 1/1 [Fwd: PATCH for Corrupted IO on all block devices]
Quoted 31 lines · David J. Picard
This is happening on an e2fs file system, I haven't tried it with others, but the code is pretty clear in
elevator.cabout putting the reads ahead of the write if they are pushed into the queue close enough to each other.Quoted 9 lines · Linus Torvalds
Quoted 4 lines · David J. Picard
Basically, what is happening is the read requests are being pushed to the front of the IO queue - before the preceding write for the same sector.
This is a bug in the USER, not in the code.
Quoted 8 lines
The locking is NOT supposed to be done at the elevator level (or, indeed at ANY _io_ level), but must be done by upper layers.
If upper layers do not do this locking, then THAT is the bug.
What filesystem do you see the bug with?
Linus
--
David J. Picard
[email protected]If you can keep your head when all about you are losing theirs,
then you clearly don't understand the situation.
Signature · 5 lines
David J. Picard [email protected] If you can keep your head when all about you are losing theirs, then you clearly don't understand the situation.