Skip to contentSkip to conversation

Lock recursion in rpc_pipefs+auth_gss...

DISCUSSION Daniel Phillips 2 participants 3 messages Jan 19, 2006 – Jan 20, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi Trond,

Quoted 3 lines · You

I believe I've finally figured out what is causing the Oopses that Vince and Vincent were seeing. It all boils down to a mutex being held after the inode is released due to a deadlock situation.

That in itself should not cause an oops. I would think there is a reference-counting problem still lurking. I'm a little concerned that a patch like this one may just make the oops a lot rarer without solving it.

Regards,

Daniel

-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

#1 lore raw
Quoted 3 lines · Daniel Phillips

That in itself should not cause an oops. I would think there is a reference-counting problem still lurking. I'm a little concerned that a patch like this one may just make the oops a lot rarer without solving it.

The lock recursion is _real_: I've been able to trigger it on 2.6.16-rc1. The only question here is therefore whether or not it suffices to explain the Oops.

Cheers,
Trond

-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

#2 lore raw
Quoted 8 lines · Trond Myklebust
Quoted 3 lines · Daniel Phillips

That in itself should not cause an oops. I would think there is a reference-counting problem still lurking. I'm a little concerned that a patch like this one may just make the oops a lot rarer without solving it.

The lock recursion is _real_: I've been able to trigger it on 2.6.16-rc1. The only question here is therefore whether or not it suffices to explain the Oops.

I don't see how it could.

I do not doubt the veracity of the lock recursion, but holding the i_sem/mutex for a long time seems to be one of the things we need to do to trigger the oops, so this deadlock is our friend. Does it always trigger the oops? Do you have a recipe handy so we can try it here?

The symptoms we see suggest the pipe inode was freed while somebody was waiting to get the i_mutex. What prevents this?

Re the lock recursion itself, is there any good reason to serialize upcalls against downcalls in the first place?

Regards,

Daniel

-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs