Fw: cp hang while copying from cifs

[Steven French]

Alexey,
Your error indicates that a file read request to windows server is unable
to be sent.  RC -11 means "EAGAIN indicating that the tcp/ip stack or
network card can't send the request or the server won't acknowledge the
data we sent from the client).  Normally if the socket or adapter card is
dead we would get a different error back.

If this is reasonably current kernel/cifs code could we track this via
bugzilla.samba.org by opening a bug #.

I need a few pieces of information that would help though:
1) the oops information in the kernel error log (dmesg)
2) what is the version of cifs.ko (you can see this at the top of the
output of  "cat /proc/fs/cifs/DebugData" except for old versions of cifs
where you have to do modinfo of cifs.ko)
3) Do you have any indication that your system is low on memory?
4) Could the network route to your windows server be unusual (e.g. gigabit
through slower ethernet that might cause frames to get tossed regularly) or
is there data indicating that your network adapter or the servers is flaky?

Your describe three symptoms:
a) An oops (I need the data to see if it is even in cifs)

b) The error message about "cannot stat" may be related to a readdir
problem that has long since been fixed - the last readdir fix IIRC was in
cifs version 1.37 but there was something in the cifs version 1.35
changelog that reminded me of what you describe.

c) [TCP] "Send Error in Read":  I have had one other case reported where
EAGAIN keeps coming back forever from the tcp socket (perhaps flaky
hardware or buggy eth driver).
I am interested to see if setting a lower sk_sndtimeo (socket send timeout)
will help - and to verify that we don't have an out of memory case here
where the tcp stack doesn't have sufficient memory to alloc buffers to send
the read request.


Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench at-sign us dot ibm dot com