Useless setrlimit(RLIMIT_CORE, ...) and typo in mainconfig.c

[Alan DeKok]

Enrik Berkhan wrote:
> when coredumps are enabled in the config file, the code in mainconfig.c
> does an setrlimit(RLIMIT_CORE, ...) using the core limits retrieved by
> an earlier getrlimit(RLIMIT_CORE, ...). This looks useless to me.

  It's there for paranoia.  Maybe the limits get reset after setuid().

> I
> think the original intention of the setrlimit() was to actually enable
> core dumping in case the server is started with core dump limit set to 0
> but coredumps enabled in the config file.

  No.  If the parent shell has the core limit set to zero, the server
should respect that.

> Further, a prctl() call to set the dumpable flag of the process is
> protected by HAVE_SYS_PRTCL_H which should be HAVE_SYS_PRCTL_H. Even

  Fixed.

> with the typo fixed, the code will never be run, because configure does
> not check for prctl.h yet.

  I can go fix that, too.

> I have attached a suggested patch (not including configure / prctl.h).

  I've committed some code clarifying the code paths, and doing
additional checks:

  - if debug_flag > 0, core dumps are always allowed (ulimit inherited
from parent shell)

  - if NOT debugging AND core dumps are disabled, AND we didn't do
setuid(), then disable core dumps.  (setuid disables core dumps by itself)

  - if we're in daemon mode, AND core dumps are enabled, AND we did
setuid(), then re-enable core dumps.

  - otherwise, core limit is inherited from parent process.

  Alan DeKok.