Bug 127928

Summary:	creating a file in a 4755 directory hangs
Product:	[Retired] Red Hat Cluster Suite	Reporter:	Christine Caulfield <ccaulfie>
Component:	gfs	Assignee:	David Teigland <teigland>
Status:	CLOSED CURRENTRELEASE	QA Contact:	Cluster QE <mspqa-list>
Severity:	medium	Docs Contact:
Priority:	medium
Version:	4	CC:	fdeutsch, iggy
Target Milestone:	---
Target Release:	---
Hardware:	i686
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2004-08-30 22:06:53 UTC	Type:	---
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:

Description Christine Caulfield 2004-07-15 14:09:48 UTC

Description of problem:

If an unprivileged user tries to create a file in a directory with
4755 protection, the process hangs as does anything else that tries to
access that directory.

This only happens when running GFS with the DLM, not gulm. A user also
reports it happens using gfs_quota but I haven't checked this

# mkdir /mnt/gfs/test
# chown user /mnt/gfs/test
# chmod 4755 /mnt/gfs/test
# su - user
$ touch /mnt/gfs/test/file

Looking in /proc/cluster/dlm_locks shows no waiting or converting locks.

How reproducible:
Always

Comment 1 Christine Caulfield 2004-07-15 14:39:31 UTC

This is the content of dlm_debug just afterwards.

test2 rq 3 980088 "       8             3e8"
test2 send lu 980088 to 3
test2 lu rep 980088 fr 3 0
test2 cv 5 980088 "       8             3e8"
test2 un 980088 ref 1 flg 4 nodeid 0/-1 "       8             3
test2 rq 3 9800fd "       8             3e8"
test2 send lu 9800fd to 3
test2 lu rep 9800fd fr 3 0
test2 cv 5 9800fd "       8             3e8"
test2 un 9800fd ref 1 flg 4 nodeid 0/-1 "       8             3
test2 rq 3 aa017c "       8             3e8"
test2 send lu aa017c to 3
test2 lu rep aa017c fr 3 0
test2 cv 5 aa017c "       8             3e8"
test2 un aa017c ref 1 flg 4 nodeid 0/-1 "       8             3
test2 rq 3 a002d5 "       8             3e8"
test2 send lu a002d5 to 3
test2 lu rep a002d5 fr 3 0
test2 cv 5 a002d5 "       8             3e8"
test2 un a002d5 ref 1 flg 4 nodeid 0/-1 "       8             3
test2 rq 3 a902bf "       8             3e8"
test2 send lu a902bf to 3
test2 lu rep a902bf fr 3 0

touch, and the dlm processes seem to be looping here.

Comment 2 Christine Caulfield 2004-07-15 14:53:54 UTC

This seems to have been caused by the recent lock_dlm changes. If I
roll back to Monday it works OK.

Comment 3 David Teigland 2004-07-16 07:21:45 UTC

it works for me after the following fixes:

Three different problems resulting from recent change to quit doing 
a convert-to-NL on a gfs unlock:

- dlm_unlock wasn't passing VALBLK flag to write lvb on unlock

- wake/complete for synchronous (internal) unlocks rather than gfs
callback

- lm_hold_lvb() wasn't preserving lvb contents since a dlm_unlock
would free the lkb and rsb.  Now if there's an lvb to preserve,
convert to NL on a lm_unlock() and do dlm_unlock on lm_unhold_lvb().

Comment 4 Corey Marthaler 2004-08-30 22:06:53 UTC

fix verified.

Comment 5 Kiersten (Kerri) Anderson 2004-11-16 19:13:06 UTC

Updating version to the right level in the defects.  Sorry for the storm.