[LU-8069] Allow remount to include "flock" and "localflock", as well as "lazystatfs" and "nolazystatfs" Created: 26/Apr/16  Updated: 29/Jan/22

Status: Open
Project: Lustre
Component/s: None
Affects Version/s: Lustre 2.8.0
Fix Version/s: None

Type: Improvement Priority: Minor
Reporter: Faye Gibbins Assignee: WC Triage
Resolution: Unresolved Votes: 0
Labels: easy
Environment:

RHEL 7.2


Issue Links:
Related
is related to LU-7759 umount hanging in modern distros when... Resolved
is related to LU-10885 enable flock by default Resolved
Business Value: 1
Severity: 3
Rank (Obsolete): 9223372036854775807

 Description   

We needed to remount a lustre file system on all the clients with the additional "flock" option.
But this is not supported by lustre and we had to take each host out of our cluster to achieve it.
Andreas Dilger suggested I raise this as a bug here to get this fixed in ll_remount_fs().
Unfortunately I'm not able to attach a patch too.



 Comments   
Comment by Andreas Dilger [ 26/Apr/16 ]

Notes for implementer:

  • turning on "flock" and "localflock" is relatively straight forward
  • turning off the flock support with "noflock" would likely be difficult since this might cause flock references to be leaked and/or break applications that have already gotten locks if all of the locks were cancelled
  • since the default client behaviour is to start with flock disabled it is OK to only allow enabling either -o remount,flock or -o remount,localflock and disallow -o remount,noflock (client would need to unmount to disable flock)
  • there isn't really any state to be saved for "lazystatfs", so it should be possible to enable/disable this feature at will
  • the default client behaviour at mount is changing from "nolazystatfs" to "lazystatfs" (http://review.whamcloud.com/19195) it should be possible to both enable and disable this feature via remount
Generated at Sat Feb 10 02:14:22 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.