Here's the result of a few minutes of hacking up something similar to Apport based on the awesome (and much underused) Frysk debugging tool kit. It doesn't post any backtraces on any Internet servers and has no fancy UI -- but it automatically dumps a stacktrace of every crashing process on the system to syslog and stores all kinds of data in /tmp/core.*/ for later inspection.
#!/bin/bash set -e export PATH=/sbin:/bin:/usr/sbin:/usr/bin DIR="/tmp/core.$1.$2" umask 077 mkdir "$DIR" cat > "$DIR/core" exec &> "$DIR/dump.log" set +e echo "$1" > "$DIR/pid" echo "$2" > "$DIR/timestamp" echo "$3" > "$DIR/uid" echo "$4" > "$DIR/gid" echo "$5" > "$DIR/signal" echo "$6" > "$DIR/hostname" set -x fauxv "$DIR/core" > "$DIR/auxv" fexe "$DIR/core" > "$DIR/exe" fmaps "$DIR/core" > "$DIR/maps" PKGS=`/usr/bin/fdebuginfo "$DIR/core" | grep "\-\-\-" | cut -d ' ' -f 1 | sort | uniq | grep '^/'| xargs rpm -qf | sort | uniq` [ "x$PKGS" != x ] && debuginfo-install -y $PKGS fstack -rich "$DIR/core" > "$DIR/fstack" set +x ( echo "Application `cat "$DIR/exe"` (pid=$1,uid=$3,gid=$4) crashed with signal $5." echo "Stack trace follows:" cat "$DIR/fstack" echo "Auxiliary vector:" cat "$DIR/auxv" echo "Maps:" cat "$DIR/maps" echo "For details check $DIR" ) | logger -p local6.info -t "frysk-core-dump-$1"
Copy that into a file $SOMEWHERE/frysk-core-dump. Then do a chmod +x $SOMEWHERE/frysk-core-dump and a chown root:root $SOMEWHERE/frysk-core-dump. Now, tell the kernel that core dumps should be handed to this script:
# echo "|$SOMEWHERE/frysk-core-dump %p %t %u %g %s %h" > /proc/sys/kernel/core_pattern
Finally, increase RLIMIT_CORE to actually enable core dumps. ulimit -c unlimited is a good idea. This will enable them only for your shell and everything it spawns. In /etc/security/limits.conf you can enable them for all users. I haven't found out yet how to enable them globally in Fedora though, i.e. for every single process that is started after boot including system daemons.
You can test this with running sleep 4711 and then dumping core with C-\. The stacktrace should appear right-away in /var/log/messages.
This script will automatically try to install the debugging symbols for the crashing application via yum. In some cases it hence might take a while until the backtrace appears in syslog.
Don't forget to install Frysk before trying this script!
You can't believe how useful this script is. Something crashed and the backtrace is already waiting for you! It's a bugfixer's wet dream.
I am a bit surprised though that noone else came up with this before me. Or maybe I am just too dumb to use Google properly?