Hi,
With both Centos 5.1 (2.6-18.21) & 5.0 NFS clients I am seeing a lock-up during read() calls. from crash I can see that the program is hung at sync_page()

I have 2 sets of program. Program -1 monitors program-2.

Program -1 periodically every 10 sec does NFS statfs. If a response doesn't comeback within 5 sec after 2 retries will kill program-2.

Program -2 does follows

fd = open(foo.bar, O_RDWR,0)
read(fd, buf, 256, 0); <= This locks-up
while (1){
fd-x = open(foo.bar.tmp, O_CREAT|O_RDWR, S_IWXRU)
write(fd-x, buf, 256, 0);
fdatasync(fd-x)
close(fd-x)
rename(foo.bar.tmp, foo.bar)
sleep 5
}

Now when the Program -2 is executing, pull the ethernet cable on the NFS server. Seeing that the Program-1 will not get a response for statfs, it will do a kill -9 of program-2. then do a umount -l and then after a min bring the NFS server up, remount the backend vol & restart the program-2. At this time it will lock-up at sync_page during the one and only read() call.

This is easily reproducible within a couple of tries within an hour or two. With nfs_debug turned on we couldn't reproduce this. My NFS client is 2 x Quad core 2 GHz
X86_64 with 16GB of memory. I have other CPU hoggers on the box which pegs the cpu at 750%. In addition I have couple of rsync sessions copying from one directory to another on the same NFS server. (rsync does lots of rename) My NFS server is a 800MHz laptop this is to simulate slow NFS server conditions.

Do anyone have a suggestion of how to workaround this issue or some surgical patches. Since after ths sync() lock-up subsequent lsof or any filesystem op to that directory
locks up the system. only way to recover is to reboot the box

Thanks in Advance
Nalan