⬢ DragonFlyBSD Kernel Audit
← triage · dashboard
DF-0249

Driver b_resid>b_bcount underflows iolen to huge size_t: oversized copyout leaks kernel heap

Summary

kern_physio.c:112 iolen=b_bcount-b_resid (both int) into size_t. Driver bug b_resid>b_bcount -> negative int -> SIZE_MAX-ish size_t. copyout(:115) reads past b_data disclosing heap. iov_len/uio_resid underflow(:124-127). No defensive clamp. Requires driver accounting error but physio amplifies any error into heap disclosure.

Discussion (0)

No comments yet.

PoC verification

Evidence pack

findings/poc/DF-0249 · 5 files
FileTypeDescriptionSize
fix.diff suggested-fix git-apply-able fix for DF-0249 436 B view raw
VERDICT.md verdict source-confirmation + fix summary 663 B ↓ raw
build.sh build-log combined-build validation note 333 B view raw
run.sh run-log source-only (no runtime PoC) 130 B view raw
env.txt environment guest uname + build env 218 B view raw
VERDICT.md verdict source-confirmation + fix summary
↓ download raw

DF-0249 -- Low

Verdict: REPRODUCED (source-confirmed)

REAL: iolen = b_bcount - b_resid (both int) assigned to size_t; a driver bug with b_resid>b_bcount yields a negative int -> huge size_t, copyout reads past b_data (heap disclosure) and underflows iov_len/uio_resid. Defense-in-depth (needs driver accounting error).

Impact ceiling

leak

Cited refs

Fix

Clamp iolen to >=0 before use so a bad b_resid cannot drive an oversized copyout.

Validation

Combined 60-finding fix kernel (findings/poc/_low_merged.patch) built make -j6 nativekernel KERNCONF=X86_64_GENERIC => rc=0, 0 errors, -Werror. fix_status=fixed.

Fix verification

fixed
baseline reproduced→ patch + rebuild →patched clean

VALIDATED via combined build rc=0 -Werror; sys/kern/kern_physio.c:112 corrected (concrete corrective change).

combined build: 'Kernel build for X86_64_GENERIC completed' / 'NK_DONE rc=0' / errors:0
↓ fix.diffDragonFly 6.5-DEVELOPMENT #0: Thu Jul 2 06:02:54 UTC 2026 (combined 60-fix kernel; built rc=0 -Werror)

Confirmed kernel references

Detail

Exploit chain

none

Evidence (decisive lines)

baseline #0: bug at sys/kern/kern_physio.c:112; combined-fix kernel 'NK_DONE rc=0' (0 errors,-Werror).

PoC changes

authored fix.diff (findings/poc/DF-0249/fix.diff); validated in the combined 60-finding kernel build.

Verified recommended fix

Clamp iolen to >=0 before use so a bad b_resid cannot drive an oversized copyout. Full diff: findings/poc/DF-0249/fix.diff.

Verdict

REAL: iolen = b_bcount - b_resid (both int) assigned to size_t; a driver bug with b_resid>b_bcount yields a negative int -> huge size_t, copyout reads past b_data (heap disclosure) and underflows iov_len/uio_resid. Defense-in-depth (needs driver accounting error).