⬢ DragonFlyBSD Kernel Audit
← triage · dashboard
DF-3371

ip_setmoptions IP_ADD_MEMBERSHIP: route reference dropped before rt_ifp is read — use-after-release race and permanent rtentry leak when the lookup held the last reference

Summary

In the no-interface/no-index ADD_MEMBERSHIP path, ip_setmoptions does 'rt = rtlookup(...); --rt->rt_refcnt; ifp = rt->rt_ifp;'. rtlookup returns the rtentry with a reference held, and the correct release idiom is RTFREE(). The bare '--' (a) leaks the rtentry permanently when our reference was the last (e.g. a fresh protocol-cloned route, which RTL_DOCLONE creates refcounted), and (b) dereferences rt->rt_ifp AFTER giving up the reference, so a concurrent RTFREE from another CPU can free the rtentry inside the window and in_addmulti() then runs on a garbage ifp. Unpriv user racing route-table churn during IP_ADD_MEMBERSHIP: UAF read -> garbage interface pointer (tiny window, Low/likely); with cloning routes, repeated ADD_MEMBERSHIP leaks cloned rtentries (kmem exhaustion). Analysis-only (no pack). Fix: copy rt_ifp first, then RTFREE(rt). [VERIFIED pass2: about 370k join attempts raced against route churn no hit - two-instruction window line-confirmed; capture-ifp-before-rtfree fix compile-validated]

Discussion (0)

No comments yet.

PoC verification

Evidence pack

findings/poc/DF-3371 · 5 files
FileTypeDescriptionSize
README.md — 973 B ↓ raw
VERDICT.md — 2.1 KB ↓ raw
manifest.json — 666 B view raw
race_mreq.c — 3.5 KB view raw
verdict.json — 1.6 KB view raw
README.md
↓ download raw

DF-3371 -- verification pack

Method: Code-line analysis + attempted live reproduction Status: NOT REPRODUCED | impact: none | confidence: likely

What this is

IP_ADD_MEMBERSHIP with imr_address=INADDR_ANY does --rt->rt_refcnt then reads rt->rt_ifp (ip_output.c:1655-1659): a manual decrement + use-after-drop - if a concurrent RTM_DELETE freed the entry in between, both the decrement and the read touch freed memory. Stress attempted: ~370k joins raced against route(8) add/delete churn of 239.1/16 (and an in-process RTM socket writer; its RTM_DELETE formatting hit ESRCH, so the external route(8) churn carried the delete side). Window is 2 instructions - not hit; mechanism line-confirmed.

Build

cc -O -o <prog> <prog>.c   (on the guest)

Run

25 (root)

(success criterion: see VERDICT.md)

Evidence

VERDICT.md run logs: joins=159670/173573/216187 across three runs, no crash

Kernel references: sys/netinet/ip_output.c:1655-1659

VERDICT.md
↓ download raw

DF-3371 -- VERDICT

NOT REPRODUCED (impact=none, confidence=likely) Guest: DragonFly dfbsd 6.5-DEVELOPMENT DragonFly 6.5-DEVELOPMENT #0: Thu Jul 2 06:02:54 UTC 2026 root@dfbsd:/usr/obj/usr/src/sys/X86_64_GENERIC x86_64

Summary

IP_ADD_MEMBERSHIP with imr_address=INADDR_ANY does --rt->rt_refcnt then reads rt->rt_ifp (ip_output.c:1655-1659): a manual decrement + use-after-drop - if a concurrent RTM_DELETE freed the entry in between, both the decrement and the read touch freed memory. Stress attempted: ~370k joins raced against route(8) add/delete churn of 239.1/16 (and an in-process RTM socket writer; its RTM_DELETE formatting hit ESRCH, so the external route(8) churn carried the delete side). Window is 2 instructions - not hit; mechanism line-confirmed.

How it was verified

  • Method: Code-line analysis + attempted live reproduction
  • Kernel references: sys/netinet/ip_output.c:1655-1659
  • Evidence: VERDICT.md run logs: joins=159670/173573/216187 across three runs, no crash
  • Reachability: local-unpriv where noted in the summary; root-config for tun/lagg/carp/ipfw3/gre items; remote-facing for the packet parsers (wlan harnesses model the RX/TX path verbatim because the guest has no 802.11 hardware; precedent DF-0393/DF-0594/DF-0621/DF-0350).
  • Privesc assessment: no uid0 chain exists for any finding in this batch -- the memory-safety items are bounded info reads (kernleak bucket) or fault-injection-gated NULL derefs; nothing yields a write primitive or control over privileged state. The one panic-class live item (DF-3414) is root-only (SIOCIFDESTROY).

Fix

fix_status: not_testable -- Fix: rtfree(rt) AFTER capturing ifp (two-line reorder); needs fault injection to A/B.

Fix status (batch validation)

Applied with the 36 sibling fix.diffs of this batch in findings/poc/_batch41/all_fixes.patch; single in-guest make -j6 nativekernel KERNCONF=X86_64_GENERIC completed RC=0 with zero errors under -Werror (full log: findings/poc/_batch41/fix_build8.log). Behavior A/B intentionally skipped for this Info-tier item.

(No fix.diff for this item: documented recommendation only - see Fix above.)

Fix verification

not_testable

Fix: rtfree(rt) AFTER capturing ifp (two-line reorder); needs fault injection to A/B.

per-fix-DF-3371

Confirmed kernel references

Detail

Evidence (decisive lines)

VERDICT.md run logs: joins=159670/173573/216187 across three runs, no crash

Verdict

IP_ADD_MEMBERSHIP with imr_address=INADDR_ANY does --rt->rt_refcnt then reads rt->rt_ifp (ip_output.c:1655-1659): a manual decrement + use-after-drop - if a concurrent RTM_DELETE freed the entry in between, both the decrement and the read touch freed memory. Stress attempted: ~370k joins raced against route(8) add/delete churn of 239.1/16 (and an in-process RTM socket writer; its RTM_DELETE formatting hit ESRCH, so the external route(8) churn carried the delete side). Window is 2 instructions - not hit; mechanism line-confirmed.