Distribution: Fedora Core 1, Fedora Core 2 Hardware Environment: P4 Prescott 3.4GHz, 4G PC3200, Intel865PERL Hardware Environment2: Athlon 64 3200+, 2G PC3200, MSI K8T-Neo Hardware Environment3: P4 Northwood 2.8G, 1G PC3200, MSI 865PE-Neo Kernel: 2.6.7 custom compile, preempt enabled (tried disabled, didn't matter), 32bit on P4, 64bit on A64. Problem Description: Under heavy load, smbfs grinds to a halt. Most obvious when using tail -f on a networked file (samba-shared from fileserver), after a couple of hours tail -f dies, gives input/output error. Even worse - under really heavy load, like trying to zip a couple of big files from network to a zipfile on the network, smbfs died completely, only fixable by hard reboot. Steps to reproduce: Unfortunately, not easy. I've tried to overload smbfs by running about a hundred tail -f simultaneously, didn't die. 5 minutes later my colleague tries to tail -f, almost instant failure. This is getting kinda ugly, three different linux-boxes using different fileservers (different version of linux/samba and win2k tested) showing the same behaviour, and not just a small bug - this makes it virtually impossible to work with. I'm out of ideas for workarounds, time to call in the big shots.
Does it happens in recent kernels? Have you tried to use cifs?
This bugreport was over two years ago, I avoided it back then by using NFS, which is far superior to samba when it comes to unix-like permissions. Since then, many things have changed, both to the kernel and to my hardware, which combined with the bad reproducability makes we doubt whether it is even remotely useful to try.
I'll set it as "not reproducible". If it happens again, it'll be much better to open a new bug