mysqld doesn't dump core after setuid

[Raghavendra D Prabhu]

This is related to http://bugs.mysql.com/21723 but over there it
is classified as a feature request. (which doesn't make sense
since setuid/setguid is usually how people run mysqld).

Basically, unless you do "echo 2 > /proc/sys/fs/suid_dumpable" and "
echo “1″ > /proc/sys/kernel/core_uses_pid" (optional), you will
not get core (both require root).

Noticed this during:
/tmp/bld/sql/mysqld.cc:1713:7: warning: will never be executed
[-Wunreachable-code]
  if (test_flags & TEST_CORE_ON_SIGNAL)
      ^~~~~~~~~~

(The code is unreachable because of "unireg_abort(1);" above).

It seems fixed in mariadb codebase:

 allow_coredumps(); is called in the tail of
set_user/set_effective_user:

static inline void allow_coredumps()
{
#ifdef PR_SET_DUMPABLE
  if (test_flags & TEST_CORE_ON_SIGNAL)
  {
    /* inform kernel that process is dumpable */
    (void) prctl(PR_SET_DUMPABLE, 1);
  }
#endif
}

(the core_uses_pid is required since without it mysqld won't
overwrite a bare corefile if already exists -
http://www.linuxinsight.com/proc_sys_fs_suid_dumpable.html)

What's more is that core-file can even be made dynamic (but
outside scope of current bug) since all it is does is do a prctl.

[Raghavendra D Prabhu (raghavendra-prabhu)]

This has not been a problem on jenkins for instance since over there we don't use setuid (jenkins user runs mysqld as same user unlike in centos etc. where init scripts are used).

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1496