[linux-cifs-client] Re: Bug in CIFS-client code ??

Steven French sfrench at us.ibm.com
Mon Nov 29 06:41:48 GMT 2004


> (here Kernel 2.4 -- to be precise:
> If the SMB-Server and the CIFS-client run on the same
> physical machine the client hangs in timeout procedures.

Can you tell the blocked request (there are potential deadlocks as far as 
I know - only in write cases in loopback (sever and client on same host) 
in which Samba server is low on memory and forces page out to cifs client 
on same host which is blocked on the same server).

cat /proc/fs/cifs/DebugData may give more useful information about the 
blocked requests

> Was this tested? (Client/Server = same machine)
Yes - heavily on 2.6 and somewhat on 2.4

> Is the behavior a bug ?
Probably - but not necessarily easy to fix without changing the samba 
memory allocation to preallocated pools or better - changing cifs 
writepage code to better handle async writebehind
Fortunately from my testing seems to only occur with very large writes in 
loopback (server and client on sane host) when host is low on memory.

Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench at-sign us dot ibm dot com
-------------- next part --------------
HTML attachment scrubbed and removed


More information about the linux-cifs-client mailing list