{
  "finding_id": "DF-4242",
  "status": "reproduced",
  "reproduced": 1,
  "impact": "panic",
  "confidence": "certain",
  "verdict": "REPRODUCED: with an in-range, EOC-set status packet and five device-fed register bytes (offsets 2..5 spelling DPTH plus HA_SERROR in HA_RSTATUS), the production dpt_intr (0xffffffff80439670 in the shipped kernel, panic site dpt_intr+0x2c2) executes panic(\"DPT Rebooted\") -- serial console shows the production dpt9: HBA rebooted ... This is incomplete, fix me banner immediately followed by panic: DPT Rebooted with the backtrace through rig_ioctl. The panic fires at dpt_scsi.c:1566 BEFORE ccb is dereferenced (line 1571), so no in-flight I/O is needed: a rebooting (or hostile) HBA alone takes the whole kernel down. FIX VALIDATED on a rebuilt kernel with the identical trigger aimed at an ACTIVE in-flight CCB (so it passes the DF-4241 guard and reaches the reboot branch): production dpt_intr prints the same banner, drains the interrupt, continues -- NO panic; the ioctl returned to userland (RC=0) and the guest stayed up, with in-flight commands left to dpttimeout/CAM recovery.",
  "exploit_chain": "device (rebooting or hostile) presents HA_SERROR + DPTH signature on a completion -> production dpt_intr panic(\"DPT Rebooted\") -> kernel down. Single-shot, zero-privilege-from-the-OS-side, purely device-fed DoS.",
  "evidence": "run.log (serial console: dpt9: HBA rebooted. + This is incomplete, fix me + panic: DPT Rebooted, backtrace dpt_intr+0x2c2 <- rig_ioctl <- dev_dioctl); panic.txt (panic block); fix_validation.log (same trigger on patched kernel: banner + Encountered bogus status packet drain, [dptrig] PANIC: dpt_intr RETURNED, guest up)",
  "kernel_refs": [
    "sys/dev/raid/dpt/dpt_scsi.c:1551",
    "sys/dev/raid/dpt/dpt_scsi.c:1558",
    "sys/dev/raid/dpt/dpt_scsi.c:1566",
    "sys/dev/raid/dpt/dpt_scsi.c:159"
  ],
  "poc_changes": "No seed PoC existed. Added the panic ioctl to the production-handler dispatch rig: writes the device-fed status packet + poisons the virtual register file (DPTH / HA_SERROR), then calls production dpt_intr. On the fix-validation run the trigger was aimed at an ACTIVE ccb (queued through production dpt_action) so the fixed reboot branch itself is exercised rather than being pre-empted by the DF-4241 inactive-CCB guard.",
  "attempts": 2,
  "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": 900,
  "guest_dirty": 0,
  "build_cmd": "cd /root/rig && make && cc -O2 -Wall -o dfdpt_ctl dfdpt_ctl.c && kldload ./dfdpt_rig.ko && ./dfdpt_ctl prime",
  "run_cmd": "./dfdpt_ctl panic 0   (panics the kernel; capture serial console)",
  "code_hash": "93d90103a5d6faf59e478ac7d74fbb3fcd04cb6fb799c747dc9e515eedf1d4ee",
  "notes": "Impact panic = deterministic full-kernel DoS from device-fed input; sibling of DF-4042/DF-4098 (hostile-HBA panic family). The ctls literal message dpt_intr RETURNED (unexpected!) is the SUCCESS marker on the patched kernel (message predates the fix). Guest reset clean afterwards.",
  "recommended_fix": "Replace panic() with interrupt drain + continue; let CAM timeouts recover in-flight commands (fix.diff).",
  "fix_status": "fixed",
  "fix_baseline_reproduced": 1,
  "fix_patched_reproduced": 0,
  "fix_verdict": "Validated on the rebuilt kernel: identical trigger on an ACTIVE ccb prints the reboot banner but drains and continues -- no panic, ioctl returns, guest stays up.",
  "fix_evidence": "fix_validation.log"
}