Controlling core files (Linux)

Core files get created when a program misbehaves due to a bug, or a violation of the cpu or memory protection mechanisms. The operating system kills the program and creates the core file.



If you don't want core files at all, set "ulimit -c 0" in your startup files. That's the default on many systems; in /etc/profile you may find

ulimit -S -c 0 > /dev/null 2>&1 
 

If you DO want core files, you need to reset that in your own .bash_profile:

ulimit -c 50000 
 

would allow core files but limit them to 50,000 bytes.



You have more control of core files in /proc/sys/kernel/



For example, you can do eliminate the tagged on pid by

echo "0" > /proc/sys/kernel/core_uses_pid 
 

Core files will then just be named "core". People do things like that so that a user can choose to put a non-writable file named "core" in directories where they don't want to generate core dumps. That could be a directory (mkdir core) or a file (touch core;chmod 000 core). I've seen it suggested that a symlink named core would redirect the dump to wherever it pointed, but I found that didn't work.



But perhaps more interesting is that you can do:

mkdir /tmp/corefiles 
chmod 777 /tmp/corefiles 
echo "/tmp/corefiles/core" > /proc/sys/kernel/core_pattern 
 

All corefiles then get tossed to /tmp/corefiles (don't change core_uses_pid if you do this).



Test this with a simple script:

# script that dumps core 
kill -s SIGSEGV $$ 
 

But wait, there's more (if your kernel is new enough). From "man proc":

/proc/sys/kernel/core_pattern 
    This file (new in Linux 2.5) provides finer control over the
    form of a core filename than the obsolete
    /proc/sys/kernel/core_uses_pid file described below. The name
    for a core file is controlled by defining a template in
    /proc/sys/kernel/core_pattern. The template can contain %
    specifiers which are substituted by the following values when
    a core file is created: 
 
     
  %%  A single % character 
  %p  PID of dumped process 
  %u  real UID of dumped process 
  %g  real GID of dumped process 
  %s  number of signal causing dump 
  %t  time of dump (secs since 0:00h, 1 Jan 1970) 
  %h  hostname (same as the 'nodename'  
      returned by uname(2)) 
  %e  executable filename 
     
 
    A single % at the end of the template is dropped from the core
    filename, as is the combination of a % followed by any character
    other than those listed above. All other characters in the
    template become a literal part of the core filename. The maximum
    size of the resulting core filename is 64 bytes. The default
    value in this file is "core". For backward compatibility, if
    /proc/sys/kernel/core_pattern does not include "%p" and
    /proc/sys/kernel/core_uses_pid is non-zero, then .PID will be
    appended to the core filename. 
 

If you are running a Linux kernel that doesn't support this, you'll get no core files at all, which is also what happens if the directory in core_pattern doesn't exist or isn't writable by the user dumping core. So that's yet another way to not dump core for certain users: set core_pattern to a directory that they can't write to, and give write permission to the users who you do want to create core files.





Use this to find core files and remove them:

find . | egrep "\/core\.[0-9]+$" | xargs rm -f

This works well as it finds only core files.