Bug 176444

Summary: chkfontpath should check for running xfs before killing it
Product: [Fedora] Fedora Reporter: Ralf Ertzinger <redhat-bugzilla>
Component: chkfontpathAssignee: Mike A. Harris <mharris>
Status: CLOSED RAWHIDE QA Contact: Ben Levenson <benl>
Severity: low Docs Contact:
Priority: medium    
Version: rawhide   
Target Milestone: ---   
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: 1.10.1-1 Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2006-02-16 14:03:12 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:
Bug Depends On:    
Bug Blocks: 150221    

Description Ralf Ertzinger 2005-12-22 21:17:47 UTC
Description of problem:
I yesterday updated a whole bunch of packages and thus had run yum in single
user mode (to reduce the number of running programs). Every time chkfontpath was
run a message popped up explaining the usage of "kill". I think this is due to
the following line from chkfontpath (extracted with strings):

kill -USR1 `/sbin/pidof xfs` 2>&1 >/dev/null

This of course fails if no xfs is actually running. Maybe using "killall -q -s
USR1 xfs" would have the same effect?

Version-Release number of selected component (if applicable):
chkfontpath-1.10.0-4.1

How reproducible:
Always

Steps to Reproduce:
1. Update any package calling chkfontpath in it's %scripts without xfs running
2.
3.
  
Actual results:
Message about invalid "kill" usage

Expected results:
no error messages

Additional info:

Comment 1 Mike A. Harris 2006-02-01 02:23:26 UTC
Hrm...   Perhaps checking for the pid file, parsing it, and then testing
the pid would work better.  That would be a bit racey though, and there's
no guarantee that the pid pointed to by the pid file is actually still
xfs (ie: xfs crashes, some other app runs and gets the same pid)

Adding to FC5Target for closer review for FC5.

Thanks for the report.

Comment 2 Mike A. Harris 2006-02-16 13:52:46 UTC
"kill -q" does not exist, at least not on FC1 or FC4 on my systems, neither
in the manpage nor usage() info for kill


Here is the current code:

void restartXfs(void)
{
  struct stat st;
  FILE *pidfile;
  pid_t pid;
  char buf[MAXPIDBUF];

  /* Make sure /proc is mounted, and /sbin/pidof exists */
  if ((stat("/proc/version", &st) == 0) && (stat("/sbin/pidof", &st) == 0)) {
    system("kill -USR1 `/sbin/pidof xfs` 2>&1 >/dev/null");
  /* If not, then test if xfs subsystem is locked, and there is a pid file */
  } else if ((stat(XFS_SUBSYSLOCK, &st) == 0) && (stat(XFS_PIDFILE, &st) == 0)) {
    if (NULL != (pidfile = fopen(XFS_PIDFILE, "r"))) {
      fgets(buf, sizeof(buf), pidfile);
      fclose(pidfile);
      if(NULL != buf)
        pid = (pid_t) atol(buf);
      if(pid > 1)
        kill(pid, SIGUSR1);
    }
  }
}


The current method favours invoking the kill command if /proc is mounted,
so that it's getting the PID of a running xfs server, but falls back to
using the xfs pid file if /proc is not mounted, so it seems it already can
do what I suggested in comment #1, but only if /proc isn't there.


It appears the problem is due to the ordering of the commandline passed
to kill:

    system("kill -USR1 `/sbin/pidof xfs` 2>&1 >/dev/null");

should instead be:

    system("kill -USR1 `/sbin/pidof xfs` >/dev/null 2>&1");


Example:

pts/7 mharris@laser:/tmp/xfs-1.0.1/os$ ls io.c sdfasdf
/bin/ls: sdfasdf: No such file or directory
io.c
pts/7 mharris@laser:/tmp/xfs-1.0.1/os$ ls io.c sdfasdf 2>&1 >/dev/null
/bin/ls: sdfasdf: No such file or directory
pts/7 mharris@laser:/tmp/xfs-1.0.1/os$ ls io.c sdfasdf >/dev/null 2>&1
pts/7 mharris@laser:/tmp/xfs-1.0.1/os$


I'm committing this change to chkfontpath CVS, and it'll be in the next build.


Comment 3 Mike A. Harris 2006-02-16 14:03:12 UTC
Fixed in 1.10.1-1 and built in rawhide.

Comment 4 Ralf Ertzinger 2006-02-16 14:09:15 UTC
The "-q" is valid for killall only. But if a simple reordering of output
redirections has the same effect that's fine, too.