killproc routine removes PID file even if HUP is sent to process.

[David Ramsden]

Binary package hint: lsb-base

Whilst looking in to a bug with nagios2 (see bug #174466), I think there is a bug in the killproc() sub-routine within /lib/lsb/init-functions (from the lsb-base package). If an init script calls this sub-routine to issue a HUP siginal, the PID file is removed. For example:

killproc -p $THEPIDFILE $DAEMON 1

This causes an issue because the PID file is removed, even though a HUP is sent to the process and as a result, a future reload (e.g. /etc/init.d/nagios2 reload) causes two daemons to be spawned because the PID file has been removed so the daemon's init script believes the daemon has stopped.

Package: lsb-base
Priority: required
Section: misc
Installed-Size: 76
Maintainer: Ubuntu Core Developers <email address hidden>
Original-Maintainer: Chris Lawrence <email address hidden>
Architecture: all
Source: lsb
Version: 3.1-23.1ubuntu3
Replaces: lsb-core (<< 2.0-6), lsb (<< 2.0-6)
Depends: sed, ncurses-bin
Conflicts: lsb-core (<< 2.0-6), lsb (<< 2.0-6)
Filename: pool/main/l/lsb/lsb-base_3.1-23.1ubuntu3_all.deb
Size: 20176
MD5sum: cfcd198d912a3ac8846b58c883c9c14a
SHA1: 2dd7a35b92fb105ca4393a418dfcc869eab9b77e
SHA256: 3c3fc72e2eb258faed465ce840c6c7601ed681cecb800df0885d741f8fd87d3f
Description: Linux Standard Base 3.1 init script functionality
 The Linux Standard Base (http://www.linuxbase.org/) is a standard
 core system that third-party applications written for Linux can
 depend upon.
 .
 This package only includes the init-functions shell library, which
 may be used by other packages' initialization scripts for console
 logging and other purposes.
Bugs: mailto:<email address hidden>
Origin: Ubuntu

This was noted in gusty. Other releases have not been tested or checked.

[David Ramsden (david-hexstream)]

Attached is a patch that (for me) corrects the above problem in killproc() from /lib/lsb/init-functions. It checks to see what signal type was sent. If the signal type is 1 (HUP), it does not remove the PID file.

[David Ramsden (david-hexstream)]

I've had a look at the lsb-base package in Debian unstable and with lsb-base 3.1-24, this bug has been fixed.

http://packages.debian.org/changelogs/pool/main/l/lsb/lsb_3.1-24/changelog
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=416559

" * Don't remove the pidfile in killproc if a signal is specified; the
     specification does not specify this behavior, and it can cause
     problems with signals like HUP. (Closes: #416559)
   * And then, only remove the pidfile if the process is actually dead.
     (Closes: #410622)"

The fix in init-functions:

    if [ -z "$sig" ]; then
        pidofproc -p $pidfile "$1" >/dev/null || rm -f "$pidfile"
    fi

[Albert Damen (albrt)]

This was indeed fixed in lsb 3.1-24. Therefore I will mark this bug as fixed.
Unfortunately a new bug in lsb showed up: calling killproc with SIGHUP will now terminate the process. See bug 228460.