Looking for opinions: Best stack protection for Linux?

chort · 10-24-2004, 01:06 AM

I know of several options for kernel protection on Linux, such as PAX, openwall, and grsecurity that includes PAX, but does anyone have practical experience comparing the different patches available and what their effectiveness and performance impact is? Specifically I'm looking for something that would work with a 2.6 kernel on Fedora. I don't actually have a Fedora box of my own to test with, but I'm trying to select a solution to use for a production quality software product. I'm not terribly interested in grsec because it looks like it would have a pretty huge impact on the system and might cause us a lot of debugging to get everything running smoothly.

I'm really interested in hearing from people who have actually applied them to systems.

unSpawn · 11-09-2004, 02:21 PM

Any luck with choosing a security framework?

SciYro · 11-09-2004, 05:31 PM

my experience with grsec has been rather ok (first time i set it up i accidentally turned on one of the security options, that caused xdm to randomly not do its job (i for get, did it crash, or did it randomly just not do its stuff after someone logged off)

anyways, I'm on a athlon 2600 to i don't think i would see a performance hit, altho i do have it setup to log almost everything (even exec ... geez i got really really long logs), just be careful and read the help for every option because they can break some programs ....

tho i haven't tried other security options ...

chort · 11-10-2004, 02:43 PM

That's really why I wanted to stay away from grsec if possible. We really don't need all the options that grsec provides because our OS will not support users. We do have a real serious concern about breaking the functionality of our software. We just need buffer overflow protection (not reliant on the NX bit, i.e. most work on x86 CPUs in general, not just AMD64) and possibly systrace policies. The lowest impact on the OS is desired because we don't want to vastly complicate our build process.

Any input is still appreciated, but apparently there isn't a whole lot of real-world experience deploying these various patches on production (rather than personal) machines.