Same setup as bug #4763. Two WinXP clients write massive amount of data into defferent directories on the same samba share. After many ours of transfer one of them stops with "specified network name is no longer reachable" messages.
This is samba 3.0.25b with the fix from bug #4763 for long notify responces.
Ethereal captures and log.smbd for the failing client are on ftp://188.8.131.52/samba1
.20 is samba server, .13 is the client. Packet #16134 and on.
Looks like the client is going "deaf" after the WriteX reply at packet 16134. You can see the Linux TCP retransmit going off 200ms later, and subsequently. The client must be missing this reply, as you see it setting the "delete on close" bit and then closing the file to delete what it thinks is the broken write.
It then sits there doing echo requests (probably from the SMB redirector level in the client). Looks like a Windows TCP stack problem to me...
I agree with the analysis, it looks like nothing smbd could be doing differently. I'll make it INVALID.