Linux 2.0.x kernel vs select() - Symptom seen elsewhere?

[Richard Bollinger]

I see JRA checked in a few patches to lib/select.c with the intent to work around a bug in Linux
2.0.x .  I'd be interested in knowing what that known bug is - some references please.

We've seen a possibly related problem with 2.0.7 and 2.2.0-alphax on Linux 2.0.38 where we get the
following log messages when running an old DOS version of NetBench from a Win98 client:

[2001/02/23 09:02:04.482307, 0, pid=31051] smbd/oplock.c:oplock_break(768)
  oplock_break: receive_smb timed out after 30 seconds.
  oplock_break failed for file netbench/netbench.exe (dev = 900, inode = 181).
[2001/02/23 09:02:04.499823, 0, pid=31051] smbd/oplock.c:oplock_break(841)
  oplock_break: client failure in oplock break in file netbench/netbench.exe

After this message sequence, oplocks appear to be disabled for this connection, but if functions
normally otherwise.

The same client application accessing the same samba binaries running with a Linux 2.2.18 kernel run
fine, with full oplock functionality.  At Samba 2.0.5a the symptom doesn't occur with either kernel
version - it all works as advertised; though there were other oplock problems back then forcing us
to generally disable oplocks.

The recent changes to lib/select.c don't seem to affect this situation, but I thought I'd mention
our problem to see if it rings any bells.

Thanks, Rich Bollinger