Excuse me if this has been dealt with but I can see nothng on the web.
SGI ICE Tempo uses ganglia as monitoring tool and that in turn uses pcp.
On earlier revs the use of root user allowed quota stats to be collected but in
newer revs the use of pcp user means quotactl fails.
For example. At older revs:pcp-3.5.5-1sgi706r3.sles11sp2
13:18:02 quotactl(Q_XGETQUOTA|0x2 /* ???QUOTA */, "/dev/sda6", 144, {version=1,
flags=XFS_PROJ_QUOTA, fieldmask=0, id=144, blk_hardlimit=2097152,
blk_softlimit=0, ino_hardlimit=0, ino_softlimit=0, bcount=0, icount=0, ...}) =
0
but at newer revs:pcp-3.6.10-1sgi708rp2.sles11sp2
06:47:13 quotactl(Q_XGETQUOTA|0x2 /* ???QUOTA */, "/dev/sda8", 42,
0x7fff0a0ed320) = -1 EPERM (Operation not permitted)
both from a pmval -S1 quota.project.space.hard type enquiry.
A quick test of changing pcp user to 0 uid and restart the pmcd works as
before.
I see the pmcd has lost the -U user ability and it performs a setuid (104) pcp
switch:
root 11353 8099 0 07:16 pts/5 00:00:00 strace -o /tmp/pmcd.trace -ff
-t ./pmcd start
pcp 11420 1 0 07:16 ? 00:00:00 /usr/lib64/pcp/bin/pmcd
root 15662 12170 0 07:18 pts/7 00:00:00 grep pmcd
r1lead:/tmp # grep setuid pmcd.trace.11420
07:16:06 setuid(104) = 0
r1lead:/tmp # grep pcp /etc/passwd
pcp:x:104:108:Performance Co-Pilot:/var/lib/pcp:/sbin/nologin
Did you have a mechanism in mind to overcome this and I've missed it? Have I
missed the documentation of the fix or the fix itself?
Thanks, MikeH mikeh@sgi.com