kern/43413: accounting suspend/resume broken

[dholland%eecs.harvard.edu@localhost]

>Number:         43413
>Category:       kern
>Synopsis:       accounting suspend/resume broken
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Jun 04 05:30:00 +0000 2010
>Originator:     David A. Holland
>Release:        NetBSD 5.1_RC2
>Organization:
xenotelepathy lab
>Environment:
System: NetBSD amberdon.eecs.harvard.edu 5.1_RC2 NetBSD 5.1_RC2 (GENERIC) #0: 
Fri May 21 00:29:51 UTC 2010 
builds%b7.netbsd.org@localhost:/home/builds/ab/netbsd-5-1-RC2/amd64/201005210534Z-obj/home/builds/ab/netbsd-5-1-RC2/src/sys/arch/amd64/compile/GENERIC
 amd64
Architecture: x86_64
Machine: amd64
>Description:

Last night while running stress builds the root filesystem filled up
with pacct data.

According to the logs, at 04:09:12 the kernel printed "accounting
suspended"; at 04:09:42 it printed "accounting resumed". The first
disk full message appeared at 04:10:12, from sa (still running from
/etc/daily, I guess), followed by more such messages from the stress
builds. The kernel first printed "Accounting: write failed 28" at
04:10:36. (28 is ENOSPC.) Many more disk full messages, including
from accounting, follow, until the machine fell over some ten minutes
later. The last entry in the log file before rebooting is

        Jun  3 04:20:24 amberdon /netbsd: <

which is moderately suspicious in its own right. Since the log is on
a wapbl volume and not softupdates, this may merely be fs corruption;
however, I don't think so because it isn't on a block boundary.

>How-To-Repeat:

Fill / with accounting turned on. Watch it not suspend and resume
accounting properly.

Having accounting turned on while running stress builds was stupid, but
evidently I forgot to turn it off beforehand.

>Fix:
Dunno.