Messages in this thread | | | Date | Mon, 20 Mar 2000 18:48:51 -0800 | From | Simon Kirby <> | Subject | Re: [2.3.99pre2] [CORRUPTION] Doh! Corruption problems again... |
| |
On Mon, Mar 20, 2000 at 05:23:30PM -0800, Andre Hedrick wrote:
> PLease repeat on SCSI-SCSI, SCSI-IDE and IDE-SCSI. > There has been nothing in the driver that has change as it relates to IO. > I want to know if it is true for the subsystem or of it is changes in the > ./fs/* code......... > > ATA has 512-byte block sizes.
But hmm, cmp reports the files differ at char 15099, and nothing even close to that factors nicely:
[sroot@oof:/tmp]# cmp corrupted-pulsar.c original-pulsar.c corrupted-pulsar.c original-pulsar.c differ: char 15099, line 516
factoring: 15085: 5 7 431 15086: 2 19 397 15097: 31 487 15098: 2 7549 15099: 3 7 719 15100: 2 2 5 5 151
...Well, the good news is I've just reproduced it four times with a test thrashing script. It seems to be so easy to reproduce under memory pressure that a simple copy of the kernel usually shows between one and three files corrupted. I've reproduced it copying from a 1K blocksize EXT2 filesystem to a 2K blocksize EXT2 filesystem. It might be worth noting that I copied approximately 5 times as much data between two 4K blocksize EXT2 filesystems and saw no corruption in any of the copied data, but I saw about 6 files corrupted on the filesystems that weren't 4K. This could just be a coincidence, though, as the 4K filesystems were on a different drive (and on the same IDE bus, now that I think of it). Hmm...
In the case I reported the file corruption started at offset (zero-based) 15098 and ended at 16384, and in this case the file corruption started at offset 15456 and stopped at 16384.
In the second reproduction I see the end of this 9849 byte file overwritten with NULLs starting at offset 9216 (1024-aligned).
In the third I see it at 17408 (-> 1024) and ending at 20480 (-> 4096).
Maybe the strange unaligned blocks are the result of buffers being modified at the time same as they're being written?
Hmm...I'll go back and try the earlier kernels where I saw no corruption and see if I can reproduce it there.
Simon-
[ Stormix Technologies Inc. ][ NetNation Communcations Inc. ] [ sim@stormix.com ][ sim@netnation.com ] [ Opinions expressed are not necessarily those of my employers. ]
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |