DF-3839 / verdict.json
{ "finding_id": "DF-3839", "status": "inconclusive", "reproduced": 0, "impact": "none", "confidence": "certain", "verdict": "Source-level certain, runtime inconclusive (guest virtio-only, ENOENT in run.log; clean compile in build.log). offset=0xfffffff0/size=0x20 wraps the u32 bound check to 0x10 so it passes for any BO, then offset64 = reloc->gpu_offset + offset programs a vertex buffer ~4GB past the BO (u64, no wrap on the GPU side) -> cross-process GPU memory disclosure via vertex fetch; when the wrapped check fires with offset > bo_size, the forced size at :2419 underflows to a huge value. Evergreen SET_RESOURCE instance of the DF-3766 (r100 INDX_BUFFER +1 wrap) family.", "exploit_chain": "", "evidence": "README.md (what to look at), VERDICT.md (line-by-line trace), trigger.c/mkbadfw.c, build.log (clean guest compile, gcc 8.3), run.log (honest ENOENT / generator RUN_OK), env.txt (guest has no DRI, no iwi NIC), fix.diff", "kernel_refs": [ "sys/dev/drm/radeon/evergreen_cs.c:2416", "sys/dev/drm/radeon/evergreen_cs.c:2419", "sys/dev/drm/radeon/evergreen_cs.c:2422", "sys/dev/drm/radeon/evergreen_cs.c:2405" ], "poc_changes": "PoC authored fresh for this finding (cross-file family seed): raw DRM ioctl ABI (cmd low byte 0x66 dispatches RADEON_CS on DF drm_ioctl.c:809), chunk structs and packet encodings verified against sys/dev/drm/include/uapi/drm/radeon_drm.h; iwi mkbadfw.c emits the v3 combined-image header per if_iwireg.h:161-166.", "attempts": 1, "guest_uname": "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", "runtime_sec": 30, "guest_dirty": 0, "build_cmd": "cc -O2 -o trigger trigger.c", "run_cmd": "./trigger /dev/dri/card0", "code_hash": "594c02d7c340cc996c7a3cedc8500537619945b9c609b05ac4c88d508c4cf2f5", "notes": "Hardware-gated: audit QEMU guest is virtio-only (no /dev/dri, hw.drm absent, no iwi PCI NIC) - runtime verification impossible in this environment; classification per orchestrator enum (no not_testable/skipped). Guest was NOT dirtied; no kernel rebuild performed (nothing reproduced to validate a fix against).", "recommended_fix": "Widen the (size+offset) comparison to u64 and reject offset >= bo_size before the forced-size write.", "fix_status": "not_applicable", "fix_baseline_reproduced": 0, "fix_patched_reproduced": 0, "fix_verdict": "Fix validation impossible on this guest: the vulnerable device (AMD radeon CS / Intel 2200BG) does not exist, so neither baseline nor patched kernel can exercise the path. fix.diff authored against the read-only sys/ tree, never applied.", "fix_evidence": "fix.diff in this pack; structurally mirrors the bounds checks the same driver family already enforces elsewhere (radeon_cs_packet_parse fit check; iwn ring-index masking)." } |