On Mon, 21 Jun 2010 10:50:48 +0200 Sebastien Clediere <Seb@xxxxxxxxxxxx> wrote: > > > On Wed, 28 Apr 2010 10:30:04 +0200 > > Seb <Seb@xxxxxxxxxxxx> wrote: > > > > > >> Jeff Layton wrote: > >> > >>> On Mon, 26 Apr 2010 21:22:16 +0200 > >>> Seb <Seb@xxxxxxxxxxxx> wrote: > >>> > >>> > >>> > >>>> Hi Jeff, > >>>> > >>>> Sorry to contact you direct. I tried to send this email to the cifs > >>>> client mailing list a number of times without luck. I registered 2 > >>>> different accounts, but my email is always rejected now! I think the > >>>> setting for spam of the list is a bit too aggressive. Anyway, that is > >>>> the message: > >>>> > >>>> Seb. > >>>> > >>>> > >>>> ======= > >>>> Going from kernel (x86) 2.6.27-8 (I am not sure which version of > >>>> mount.cifs) to 2.6.31-19 with mount.cifs version: 1.12-3.4.0, I > >>>> noticed an important decrease of performance while navigating CIFS > >>>> shares on a Windows 2003 server. Navigating means here either using > >>>> midnight commander, ls -l (with ls unaliased first...), rsync, etc.. > >>>> > >>>> I added noserverino,nolinux to the mount command, but it does not make > >>>> much difference. > >>>> > >>>> The kind of performance degradation I am talking about is in the order > >>>> of times ten or so. An rsync which takes less that 2 minutes on old > >>>> hardware, now takes over 15 minutes on newer hardware! > >>>> > >>>> Doing a bit of investigation with tcpdump, we can see a lot > >>>> QUERY_PATH_INFO requests happening which are not necessary as all the > >>>> information is already returned by FIND_FIRST2 requests. A trivial > >>>> example will illustrate it. Lets have a directory with 3 files inside. > >>>> > >>>> ls -l with my old machine, it gives: > >>>> Protocol Info > >>>> SMB Trans2 Request, QUERY_PATH_INFO, Query File All Info, Path: \Temp > >>>> SMB Trans2 Response, QUERY_PATH_INFO > >>>> SMB Trans2 Request, FIND_FIRST2, Pattern: \Temp\* > >>>> SMB Trans2 Response, FIND_FIRST2, Files: . .. file1 file2 file3 > >>>> > >>>> > >>>> With the new one, it gives: > >>>> Protocol Info > >>>> SMB Trans2 Request, QUERY_PATH_INFO, Query File All Info, Path: \Temp > >>>> SMB Trans2 Response, QUERY_PATH_INFO > >>>> SMB Trans2 Request, FIND_FIRST2, Pattern: \Temp\* > >>>> SMB Trans2 Response, FIND_FIRST2, Files: . .. file1 file2 file3 > >>>> SMB Trans2 Request, QUERY_PATH_INFO, Query File All Info, Path: > >>>> \Temp\file1 > >>>> SMB Trans2 Response, QUERY_PATH_INFO > >>>> SMB Trans2 Request, QUERY_PATH_INFO, Query File All Info, Path: > >>>> \Temp\file2 > >>>> SMB Trans2 Response, QUERY_PATH_INFO > >>>> SMB Trans2 Request, QUERY_PATH_INFO, Query File All Info, Path: > >>>> \Temp\file3 > >>>> SMB Trans2 Response, QUERY_PATH_INFO > >>>> > >>>> So for every file in the directory, as returned by the FIRST_FIND2 > >>>> response, it does a QUERY_PATH_INFO, which does not bring any new > >>>> information, all the attributes were already returned by FIRST_FIND2. > >>>> I attached to the email the tcpdump so you can really double check by > >>>> yourself that the QUERY_PATH info does not bring anything new. > >>>> > >>>> That is the cause of the slowness I notice. > >>>> Is that a known issue? > >>>> > >>>> Thanks, > >>>> > >>>> Seb. > >>>> > >>>> > >>>> > >>> Does your kernel have this commit? > >>> > >>> commit f12f98dba6ea1517cd7fbb912208893b9c014c15 > >>> Author: Jeff Layton <jlayton@xxxxxxxxxx> > >>> Date: Fri Feb 5 13:14:00 2010 -0500 > >>> > >>> cifs: fix length calculation for converted unicode readdir names > >>> > >>> ...that's most likely the cause of the problem. > >>> > >>> > >>> > >> Thanks a lot for answering my email Jeff. > >> I don't think the kernel I use has this commit as I am on Ubuntu Karmic > >> and it was based on "v2.6.31 final". > >> > >> Seb. > >> > >> > > On 04/28/2010 12:27 PM, Jeff Layton wrote > > Ok, then you'll probably want to apply that patch and test it out, or > > ask your distro to include it. That patch fixes a bug that can cause > > the symptoms you're seeing. > > > > > Hi Jeff, > > Sorry to bother you again. It took me a while to try out as I moved and > did not have a fast internet connexion. However, I have just upgraded my > system, and this kernel has this commit you mentioned. However, I still > see the exact same problem. > What can I do to help identify that bug? > (re-cc'ing linux-cifs mailing list) First, please cc the list rather than emailing me directly. I generally don't do personal one-on-one support unless you're paying me ;) You don't say what your upgraded kernel actually is, but you should probably start with testing a recent upstream kernel. I don't see this behavior with a quick test on a 2.6.34-ish kernel. If you see this problem with a 2.6.34-ish kernel, then I'd say to open a bug at bugzilla.samba.org and one of us will have a look when we have time. If you can't reproduce it on a more recent kernel, then you'll probably want to open a bug with your distro vendor, and see whether they can track down what fixed it then they can put the patches in the kernels they're shipping. -- Jeff Layton <jlayton@xxxxxxxxxx> -- To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
