On Thu, Oct 21, 2010 at 05:26:52PM +0200, Jan Schermer wrote: > Update: It's not in SSH/crypto but a network problem > > I tried netcat over network and the file also got corrupted (5x OK, 1x > corrupt, 1GB file - former swap file, so half random data and half zeroes). > There are ~10 bytes corrupted in the middle of the file, very close > together (on one page in vbindiff) - so probably one packet/fragment/frame. > I also have tcpdump record of the whole connection - but nothing > apparently fishy there. > > Right now I'm testing without netfilter enabled and so far so good (also > had to reboot so it might have "fixed" itself). > Memtest done without a problem via memtester on 75% of memory, proper > memtest86 will be done tonight. > > Any suggestions where to go next? I'm thinking of making a 1GB plaintext > file so that the corruption will be readable and searchable in a data > stream and I can inspect the corrupted packets - but what to look for?
I've seen previous reports of bugs in the TSO[*] implementation in the similar atl1e driver and/or hardware, so perhaps the atl1c driver or hardware is also broken. Unfortunately there is currently no way to disable TSO in the driver, but we can change that. Please build and install a kernel package with the attached patch, following the instructions at <http://kernel-handbook.alioth.debian.org/ch-common-tasks.html#s-common-official> and then try disabling TSO using the ethtool command ('ethtool -K eth0 tso off'). Ben. [*] http://en.wikipedia.org/wiki/TCP_segmentation_offloading -- Ben Hutchings We get into the habit of living before acquiring the habit of thinking. - Albert Camus
signature.asc
Description: Digital signature
