Subaudit/update_pltbind.c contains unreachable code

[SamHerts]

Priority: Low

Spindle has two interception paths: the standard LD_AUDIT path (used on x86_64 and aarch64) and the SubAudit path (OPT_SUBAUDIT) which has only remaining use case of ppc64le.

Bug: spindleint_none is an unreachable enum state.

Spindle/src/client/subaudit/update_pltbind.c

Line 213 in 0380a7c

has_spindleint = ld_preload && strstr(ld_preload, "libspindleint.so") ? spindleint_present : spindleint_unset;

Sets has_spindleint to either spindleint_present or spindleint_unset, but

Spindle/src/client/subaudit/update_pltbind.c

Lines 231 to 232 in 0380a7c

	if (has_spindleint == spindleint_none)
	redirect_interceptions(update_list[i]);

checks for spindleint_none.

Because has_spindleint is never set to spindleint_none, two things follow:

1. getenv("LD_PRELOAD") + strstr is called on every la_activity event for the lifetime of the process, rather than being cached after the first check.
2. The redirect_interceptions() call at line 231 is unreachable dead code — spindleint_none is defined in the enum and has a consumer, but is never assigned.

Fix:

has_spindleint = ld_preload && strstr(ld_preload, "libspindleint.so") ? spindleint_present : spindleint_none;

Potential follow-up:

• Confirm whether redirect_interceptions() / spindleint_none represents intentional design or can be removed.

[mplegendre]

Are you using subaudit on any system? I'm wondering if it'd be worth just deleting the whole subaudit branch.

Most recently, on ppc64, subaudit worked around a GLIBC bug where glibc's PLT binding had a race condition in the AUDITing case and randomly crashed when doing a binding. Subaudit is an alternate (and hacky) way to use LD_AUDIT that avoids the GLIBC bug.

That bug was in RHEL7, and fixed by RHEL8. If we ever go back to a ppc64 system I'd expect to not need it.

[SamHerts]

I've got no use for ppc64 or the subaudit branch, so removing unused code would definitely help simplify.