Skip to contentSkip to conversation

nlm svid changes from pid --> nlm_lockowner

DISCUSSION Samuel Li 2 participants 2 messages Jan 11, 2013

Subsystems (2)
lore
Share & tools
lore

Samuel Li

author
#0 lore raw

A simple question with NLM svid implementation with after 2.6.16 or later.

In the old days when linux client request and NLM lock it submits a svid with the content of the client pid. However this appears to have changed a few year ago and I get an incremental number starting from 0.

from the NLM server side we only report the content of svid. I am trying to find a way to identify the end user application request.

for example

[sli@fed16-vm nfs3]# ./gwaiting & ./gwaiting & ./gwaiting & (excl lock programs)
[2] 977 <--pid
[3] 978
[4] 979

tshark -i eth0 -R "nlm" -V | grep -i svid
OOPS: dissector table "sctp.ppi" doesn't exist
Protocol being registered is "Datagram Transport Layer Security" Running as user "root" and group "root". This could be dangerous. Capturing on eth0

          svid: 4
          svid: 3
          svid: 5
          svid: 3
          svid: 4
          svid: 4
          svid: 4
          svid: 5
          svid: 5

How do I determine from the client side who owns these NLM processes?

The file lock in question was
inode 4318756868 so I could determine the lock.

cat /proc/locks
1: POSIX ADVISORY WRITE 979 00:2a:4315692708 0 EOF 1: -> ACCESS ADVISORY WRITE 978 00:2a:4315692708 0 EOF

2: POSIX  ADVISORY  WRITE 1778 00:10:16424 0 EOF
3: POSIX  ADVISORY  WRITE 1677 00:10:18376 0 EOF

Now assume I did not have the output of /proc/locks and only nlm_lockowner number how would I debug this?

I found the change in

commit 7bab377fcb495ee2e5a1cd69d235f8d84c76e3af

Author: Trond Myklebust <[email protected]>
Date:   Mon Mar 20 13:44:06 2006 -0500

lockd: Don't expose the process pid to the NLM server

Instead we use the nlm_lockowner->pid.

Signature · 6 lines
Best regards,

Samuel Li
Escalation Engineer
EMC²| Isilon Storage Division
[email protected]
#1 lore raw
Quoted 37 lines · Samuel Li

A simple question with NLM svid implementation with after 2.6.16 or later.

In the old days when linux client request and NLM lock it submits a svid with the content of the client pid. However this appears to have changed a few year ago and I get an incremental number starting from 0.

from the NLM server side we only report the content of svid. I am trying to find a way to identify the end user application request.

for example

[sli@fed16-vm nfs3]# ./gwaiting & ./gwaiting & ./gwaiting & (excl lock programs)
[2] 977 <--pid
[3] 978
[4] 979

tshark -i eth0 -R "nlm" -V | grep -i svid
OOPS: dissector table "sctp.ppi" doesn't exist
Protocol being registered is "Datagram Transport Layer Security" Running as user "root" and group "root". This could be dangerous. Capturing on eth0

           svid: 4
           svid: 3
           svid: 5
           svid: 3
           svid: 4
           svid: 4
           svid: 4
           svid: 5
           svid: 5

How do I determine from the client side who owns these NLM processes?

The file lock in question was
inode 4318756868 so I could determine the lock.

Quoted 8 lines

cat /proc/locks
1: POSIX ADVISORY WRITE 979 00:2a:4315692708 0 EOF 1: -> ACCESS ADVISORY WRITE 978 00:2a:4315692708 0 EOF

2: POSIX  ADVISORY  WRITE 1778 00:10:16424 0 EOF
3: POSIX  ADVISORY  WRITE 1677 00:10:18376 0 EOF

Now assume I did not have the output of /proc/locks and only nlm_lockowner number how would I debug this?

The POSIX lock inheritance rules under the clone() system call (where POSIX locks must follow the file descriptor table and _not_ the pid inheritance rules) mean that the svid does not map naturally into a pid. So post-2.6.16 we no longer try to pretend that is the case.

The bottom line is that you can use the svid as a debugging tool for figuring out which locks belong to the same lock owner, but you cannot map it to a pid.

Signature · 6 lines
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
www.netapp.com