proc-identity fix saved-set-ID update in setreuid and setregid

[doanbaotrung]

sys: guard saved ID update in setreuid/setregid with euid/egid != -1

Modify DEFINE_SETRE macro to ensure that the saved-set-ID is only
updated by the effective ID change check when the new effective ID
is actually being set (i.e. e != -1).

This prevents setreuid(-1, -1) and setregid(-1, -1) from spuriously
clobbering the saved-set-ID when the current effective ID differs
from the real ID, matching Linux behavior.

Fix #140

---

Summary by cubic

Fix saved-set-ID handling in setreuid(2) and setregid(2) to match Linux rules. The saved ID now updates from the new effective ID after applying real/effective changes, including when only the real ID is set.

• Bug Fixes
  • Updated DEFINE_SETRE to compute saved from the final eff after independently applying r and e.
  • Fixes the missed setreuid(r, -1) path; same change applies to setregid. Fixes setreuid / setregid Saved ID is Not Updated on Real ID Change when e == -1 #140.

^{Written for commit a6d01d9. Summary will update on new commits.}

[jserv]

Saved-set-ID condition drops the e != -1 guard.

This clause fires even when the effective ID is not being changed. When e == -1, eff still holds the current, unchanged effective ID; if it differs from the real ID, saved gets spuriously rewritten. So setreuid(-1, -1) on a process whose euid != ruid (e.g. ruid=1000, euid=0, suid=500 reached via a prior setresuid) clobbers the saved-set-ID even though a (-1, -1) call must be a no-op — contradicting this PR's own description table (setreuid(-1,-1) -> no change). This is privilege-relevant because uid_is_permitted/gid_is_permitted later key off emu_suid/emu_sgid.

Linux (kernel/sys.c __sys_setreuid, and the setreuid(2) man page) guards the second clause with euid != -1:

if (r != (uint32_t) -1 || (e != (uint32_t) -1 && eff != old_real))
    saved = eff;

The core intent of the PR — updating the saved-set-ID for setreuid(r, -1) — is correct; this is the one remaining defect, and it applies to setregid too via the shared macro.

[doanbaotrung]

Fixed.

Modify DEFINE_SETRE macro to ensure that the saved-set-ID is only
updated by the effective ID change check when the new effective ID
is actually being set (i.e. e != -1).

This prevents setreuid(-1, -1) and setregid(-1, -1) from spuriously
clobbering the saved-set-ID when the current effective ID differs
from the real ID, matching Linux behavior.

[jserv]

Thank @doanbaotrung for contributing!