Hi,

I have a small C++ program which basically copies data from one
directory to a destination directory.. using i/ofstream to do the job.
I am mounting a NFS share (Fedora Core 5) with the following client
options:

-o proto=tcp,soft,intr,timeo=100

What I am noticing is that if i yank a network cable to simulate an
error, the client does not seem to get an I/O error from the kernel
UNLESS i specify the 'noac' option in the mount command. The noac
option causes the file transfer to slow down greatly, I assume because
it forces async writes. Though I am confused as to why this is
necessary for the kernel to report an I/O error to the process after
major timeout.

Has anyone run into this who can maybe shed some light? I thought any
soft mount was supposed to report an I/O error to the application if
RPC/NFS timed out.