On 05/18/2016 10:19 AM, Soumya Koduri wrote: > Hi Dan, > > Reviving this old thread. > > On 02/02/2016 07:44 PM, Daniel Gryniewicz wrote: >> On 02/02/2016 05:34 AM, Soumya Koduri wrote: >>> Thank you all for your inputs. I shall check if there is any easy way of >>> re-producing it and then test with Daniel's patch added. >> >> That patch applies on top of mdcache, so it will need modification to >> work on top of stock ganesha. I can do that, if you need it, once you >> have a reproducer. >> >>> >>> On 02/01/2016 10:10 PM, Frank Filz wrote: >>>> There might be a refcount leak. I think we do need to clean this stuff >>>> up, >>>> and maybe change how we handle killed cache inode entries. The >>>> process is >>>> now inline, it is supposed to clean up on the final unref. >>> >>> Okay. But in case if there are any special entries (like Matt had >>> mentioned below) which do not get cleaned up in normal code path, >>> shouldn't we forcefully clean them up as part of unexport? >>> >> >> An entry can be on multiple exports, so unconditionally removing it on >> unexport will break things. >> >> More likely, we should be removing it from the export anyway? That way >> it will be cleaned up when the last ref is dropped. >> >> That code looks like it's assuming another thread is actively cleaning >> up the entry, but I think that may not be the case. >> > > We seem to have got definite steps to reproduce this issue (at-least > with FSAL_GLUSTER) with v2.3-stable branch [1]. Could you please provide > info on what thread should be cleaning up such entries. > > Steps to reproduce > > 1.) create a volume and start a volume > 2.) export the volume > 3.) run the pynfs test on it(only subset is enough, for e.g., "link" test) > 4.) try to unexport the volume using dbus. >
Is there a way to test on 2.4 after the next -dev? I just closed the last refcount issue I've been able to find on there, and it'd be nice to know if that one is fixed too. I assumed these issues were introduced by MDCACHE, but it's possible one or more of them were already extant in cache_inode, and I just found them. In the mean time, I'll give a shot at 2.3.x and see if I can reproduce locally. Daniel ------------------------------------------------------------------------------ Mobile security can be enabling, not merely restricting. Employees who bring their own devices (BYOD) to work are irked by the imposition of MDM restrictions. Mobile Device Manager Plus allows you to control only the apps on BYO-devices by containerizing them, leaving personal data untouched! https://ad.doubleclick.net/ddm/clk/304595813;131938128;j _______________________________________________ Nfs-ganesha-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/nfs-ganesha-devel
