1407075 - Add an alternative signal stack in our codebase that does not depend on breakpad

Status: RESOLVED WONTFIX

(Toolkit :: Crash Reporting, enhancement)

Description

[Christian Holler (:decoder)]

Currently, breakpad is calling sigaltstack() to make sure we have an alternative signal stack for processing signals. With the changes in bug 1403668, all of our SIGSEGV handlers will use the alternative signal stack to ensure that they are working even in the case of a stack overflow.

However, in build types that both have the crash reporter disabled and AddressSanitizer (which also sets up an alternative signal stack) not enabled, this will not work and fallback to the old default stack behavior.

As proposed by Ted and Nathan, we should probably factor out the sigaltstack code from breakpad and include it directly in our codebase unconditionally (or in non-asan builds in case that it interferes with ASan).

Comment 1

[Selena Deckelmann :selenamarie :selena]

Ted -- what kind of priority should we put on this? And who should do this work?

Comment 2

[(not currently active) Ted Mielczarek]

This is low-priority cleanup. decoder's patch in bug 1403668 fixed the pressing issue where we weren't handling stack overflows properly. The proposed change here would fix build configurations that we don't ship, and could consolidate some duplicated code.

Comment 3

[Firefox Bug Husbandry Bot]

https://wiki.mozilla.org/Bug_Triage/Projects/Bug_Handling/Bug_Husbandry#Update_affecteds

Comment 4

[Gabriele Svelto [:gsvelto] (PTO)]

I've implemented something like this in bug 1678152. It covers all thread except for the main one (because Breakpad handles that) but it should be easy to both factor this code out and put it in a separate library that can be included even when the crash reporter is disabled. Given this bug is 5 years old is this something we should prioritize?

Comment 5

[Christian Holler (:decoder)]

From a fuzzing perspective, we are no longer using any builds that do not have a sanitizer component of some sort enabled (at minimum, they have the UBSan runtime), so from a fuzzing perspective we can wontfix this.

Comment 6

[Gabriele Svelto [:gsvelto] (PTO)]

Thanks!