Bug #21361 | Patch: create coredumps on linux when --user or --group is used | ||
---|---|---|---|
Submitted: | 31 Jul 2006 11:55 | Modified: | 5 Sep 2006 13:39 |
Reporter: | Michal Marek | Email Updates: | |
Status: | Duplicate | Impact on me: | |
Category: | MySQL Server | Severity: | S4 (Feature request) |
Version: | 5.0.22 | OS: | Linux (Linux) |
Assigned to: | CPU Architecture: | Any |
[31 Jul 2006 11:55]
Michal Marek
[31 Jul 2006 11:56]
Michal Marek
patch to allow coredumps
Attachment: mysql-5.0.22-coredump.patch (text/x-patch), 1.19 KiB.
[31 Jul 2006 12:30]
Hartmut Holzgraefe
The call is indeed linux specific (introduced with kernel 2.4), and given it's recent security implications i'd say it should not be used. Starting the server using the right user id right away is not that much of an issue when core dumps are needed and other settings like ulimit need to be tweaked, too ...
[31 Jul 2006 12:35]
Valeriy Kravchuk
So, it is a feature request and, according to Hartmut, it is unlikely that it will be implemented.
[1 Aug 2006 13:37]
Michal Marek
> The call is indeed linux specific (introduced with kernel 2.4), > and given it's recent security implications i'd say it should > not be used. Which security implications? mysqld runs under a dedicated user ID, so in fact only root will be able to signal it and read the coredump. And the patch calls prctl() only if the --core-file option is given, ie. when the user explicitly asks for a coredump.
[5 Sep 2006 13:37]
Domas Mituzas
prctl() security issues are false assumptions. mysqld does not have setuid bit on binary, it is setuid to non-privileged user, and kernel bugs are kernel bugs - mysqld is not required to do anything in order for them to be exploitable.
[5 Sep 2006 13:39]
Domas Mituzas
Duplicate of Bug#21723