There seems to be a bug in the qnx nfs stack. Client side is a recent
ubuntu version. There's a dir exported on the qnx side, containing small
files (a few dozen), ~2.2k each (varies). Occasionally, reading a file
on the linux side results in about the first half or so from the correct
file, then the rest is from another file in that same dir.

The first 1536 (1024+512) bytes are from the correct file, the next 641
bytes are from the wrong file. The last 36 bytes are repeated from the
end of the wrong file. Size of correct and corrupt file is the same, 2213.

How I determined the problem was on the qnx side: I umounted then
remounted the nfs share on the linux side and the file was still
corrupt. The file was correct when read locally on the qnx side (using
cat). I then ran umount, restarted the nfs procs on the qnx side, then
mounted again and the file was then correct on the linux side, over nfs.

The qnx fs block size (in this case) appears to be 512. I'm guessing
that a block pointer somewhere is being corrupt, since this isn't
occurring at a 4k page boundary.

--
Tom Vier <tom@triadsys.com>
Senior Net/Sys Admin
Triad Systems Engineering, Inc.