cairn/actions/cairn-zig-fuzz-afl/action.yml

452 lines
17 KiB
YAML

name: 'Cairn Zig Fuzz (AFL++)'
description: 'Build and run Zig AFL++ fuzz targets, reporting crashes and corpus to Cairn. Each target is built twice: `zig build fuzz -Dfuzz-target=<name>` (AFL-instrumented) and `zig build fuzz-replay -Dfuzz-target=<name>` (plain, for crash replay).'
inputs:
cairn_server:
description: 'Cairn server URL'
required: true
repo:
description: 'Repository name'
required: true
owner:
description: 'Repository owner'
required: true
commit:
description: 'Commit SHA'
required: false
default: '${{ github.sha }}'
targets:
description: |
Line-delimited fuzz target names. Each line is passed as -Dfuzz-target=<name>.
Example:
lexer
parser
varint_decode
required: true
fuzz_binary:
description: 'Binary name in the build output bin/ directory (auto-detected if only one)'
required: false
default: ''
seed_dir:
description: |
Directory containing per-target seed inputs. Each subdirectory should match
a target name from the targets input. Files are merged with the downloaded corpus.
Example layout:
seeds/lexer/valid-token.txt
seeds/parser/minimal-program.txt
required: false
default: ''
duration:
description: 'Fuzz duration per target in seconds'
required: false
default: '600'
afl_args:
description: 'Extra afl-fuzz arguments'
required: false
default: ''
target:
description: 'Target platform metadata'
required: false
default: ''
cairn_version:
description: 'Cairn CLI version'
required: false
default: 'latest'
runs:
using: 'composite'
steps:
- name: Setup AFL++
uses: https://git.ts.mattnite.net/mattnite/cairn/actions/setup-afl@main
- name: Check prerequisites
shell: sh
run: |
set -eu
command -v zig >/dev/null 2>&1 || { echo "ERROR: zig not found in PATH. Install Zig before using this action (e.g. https://codeberg.org/mlugg/setup-zig@v2)."; exit 1; }
command -v afl-cc >/dev/null 2>&1 || { echo "ERROR: afl-cc not found in PATH after setup-afl step."; exit 1; }
echo "zig $(zig version), afl-cc found"
- name: Setup Cairn CLI
uses: https://git.ts.mattnite.net/mattnite/cairn/actions/setup-cairn@main
with:
version: ${{ inputs.cairn_version }}
server_url: ${{ inputs.cairn_server }}
- name: Build and fuzz with AFL++
shell: bash
env:
CAIRN_SERVER: ${{ inputs.cairn_server }}
REPO: ${{ inputs.repo }}
OWNER: ${{ inputs.owner }}
COMMIT: ${{ inputs.commit }}
TARGETS: ${{ inputs.targets }}
FUZZ_BINARY: ${{ inputs.fuzz_binary }}
SEED_DIR: ${{ inputs.seed_dir }}
DURATION: ${{ inputs.duration }}
EXTRA_AFL_ARGS: ${{ inputs.afl_args }}
TARGET_PLATFORM: ${{ inputs.target }}
run: |
set -eu
echo "Cairn CLI version: $(cairn version)"
NCPU=$(nproc 2>/dev/null || echo 1)
echo "Available cores: ${NCPU}"
RESULTS_DIR=$(mktemp -d)
trap 'rm -rf "${RESULTS_DIR}"' EXIT
# ── Collect target names ──
TARGET_NAMES=()
while IFS= read -r line; do
line=$(echo "${line}" | sed 's/#.*//' | xargs)
[ -z "${line}" ] && continue
TARGET_NAMES+=("${line}")
done <<EOF
${TARGETS}
EOF
TARGET_COUNT=${#TARGET_NAMES[@]}
echo "Targets: ${TARGET_COUNT}, parallel slots: ${NCPU}"
# ── Phase 1: Build all targets sequentially ──
echo ""
echo "=========================================="
echo "Building ${TARGET_COUNT} target(s)"
echo "=========================================="
# Helper to find the single executable in a bin/ directory.
find_bin() {
local BIN_DIR="$1"
local LABEL="$2"
if [ -n "${FUZZ_BINARY}" ]; then
echo "${BIN_DIR}/${FUZZ_BINARY}"
return
fi
local BIN
BIN=$(find "${BIN_DIR}" -maxdepth 1 -type f -executable)
local COUNT
COUNT=$(echo "${BIN}" | wc -l)
if [ "${COUNT}" -eq 0 ] || [ -z "${BIN}" ]; then
echo "ERROR: No executable found in ${BIN_DIR} (${LABEL})" >&2
return 1
elif [ "${COUNT}" -gt 1 ]; then
echo "ERROR: Multiple executables in ${BIN_DIR} (${LABEL}), specify fuzz_binary input" >&2
return 1
fi
echo "${BIN}"
}
declare -A TARGET_BINS
declare -A REPLAY_BINS
for i in "${!TARGET_NAMES[@]}"; do
FUZZ_TARGET="${TARGET_NAMES[$i]}"
NUM=$((i + 1))
WORK="work/${FUZZ_TARGET}"
mkdir -p "${WORK}"
echo "[${NUM}/${TARGET_COUNT}] Building ${FUZZ_TARGET}..."
# Each target gets its own output dir to avoid binary name collisions.
# 1) AFL-instrumented binary for fuzzing.
AFL_OUT="${WORK}/afl-out"
rm -rf "${AFL_OUT}"
zig build fuzz -Dfuzz-target="${FUZZ_TARGET}" --prefix "${AFL_OUT}"
FUZZ_BIN=$(find_bin "${AFL_OUT}/bin" "afl")
if [ ! -x "${FUZZ_BIN}" ]; then
echo "ERROR: Fuzz binary not found or not executable: ${FUZZ_BIN}"
exit 1
fi
TARGET_BINS["${FUZZ_TARGET}"]="${FUZZ_BIN}"
echo " AFL binary: ${FUZZ_BIN}"
# 2) Plain binary for crash replay (no AFL instrumentation).
REPLAY_OUT="${WORK}/replay-out"
rm -rf "${REPLAY_OUT}"
zig build fuzz-replay -Dfuzz-target="${FUZZ_TARGET}" --prefix "${REPLAY_OUT}"
REPLAY_BIN=$(find_bin "${REPLAY_OUT}/bin" "replay")
if [ ! -x "${REPLAY_BIN}" ]; then
echo "ERROR: Replay binary not found or not executable: ${REPLAY_BIN}"
exit 1
fi
REPLAY_BINS["${FUZZ_TARGET}"]="${REPLAY_BIN}"
echo " Replay binary: ${REPLAY_BIN}"
done
# ── Phase 2: Prepare seeds and Cairn targets (sequential, network I/O) ──
echo ""
echo "=========================================="
echo "Preparing corpus for ${TARGET_COUNT} target(s)"
echo "=========================================="
declare -A CAIRN_TARGET_IDS
declare -A RUN_IDS
for i in "${!TARGET_NAMES[@]}"; do
FUZZ_TARGET="${TARGET_NAMES[$i]}"
NUM=$((i + 1))
FUZZ_BIN="${TARGET_BINS[${FUZZ_TARGET}]}"
SEEDS="work/${FUZZ_TARGET}/seeds"
mkdir -p "${SEEDS}"
echo "[${NUM}/${TARGET_COUNT}] Preparing ${FUZZ_TARGET}..."
# Ensure Cairn target.
CAIRN_TARGET_ID=$(cairn target ensure \
-server "${CAIRN_SERVER}" \
-repo "${REPO}" \
-owner "${OWNER}" \
-name "${FUZZ_TARGET}" \
-type fuzz)
CAIRN_TARGET_IDS["${FUZZ_TARGET}"]="${CAIRN_TARGET_ID}"
# Start a run.
RUN_ID=$(cairn run start \
-server "${CAIRN_SERVER}" \
-target-id "${CAIRN_TARGET_ID}" \
-commit "${COMMIT}")
RUN_IDS["${FUZZ_TARGET}"]="${RUN_ID}"
echo " Target ID: ${CAIRN_TARGET_ID}, Run ID: ${RUN_ID}"
# Copy per-target seeds from repo.
if [ -n "${SEED_DIR}" ] && [ -d "${SEED_DIR}/${FUZZ_TARGET}" ]; then
REPO_SEED_COUNT=$(find "${SEED_DIR}/${FUZZ_TARGET}" -maxdepth 1 -type f | wc -l)
if [ "${REPO_SEED_COUNT}" -gt 0 ]; then
cp "${SEED_DIR}/${FUZZ_TARGET}"/* "${SEEDS}/" 2>/dev/null || true
echo " Copied ${REPO_SEED_COUNT} repo seeds"
fi
fi
# Download existing corpus.
DL_START=$(date +%s)
cairn corpus download \
-server "${CAIRN_SERVER}" \
-target-id "${CAIRN_TARGET_ID}" \
-dir "${SEEDS}" || true
DL_ELAPSED=$(( $(date +%s) - DL_START ))
DL_COUNT=$(find "${SEEDS}" -maxdepth 1 -type f | wc -l)
echo " Corpus: ${DL_COUNT} entries (${DL_ELAPSED}s)"
if [ "$(find "${SEEDS}" -maxdepth 1 -type f | wc -l)" -eq 0 ]; then
printf 'A' > "${SEEDS}/seed-0"
fi
# Minimize corpus.
SEED_COUNT=$(find "${SEEDS}" -maxdepth 1 -type f | wc -l)
if [ "${SEED_COUNT}" -gt 1 ]; then
echo " Minimizing corpus (${SEED_COUNT} inputs)..."
CMIN_START=$(date +%s)
MINIMIZED="work/${FUZZ_TARGET}/minimized"
rm -rf "${MINIMIZED}"
mkdir -p "${MINIMIZED}"
if afl-cmin -i "${SEEDS}" -o "${MINIMIZED}" -- "${FUZZ_BIN}" >/dev/null 2>&1; then
CMIN_ELAPSED=$(( $(date +%s) - CMIN_START ))
MINIMIZED_COUNT=$(find "${MINIMIZED}" -maxdepth 1 -type f | wc -l)
echo " Minimized: ${SEED_COUNT} -> ${MINIMIZED_COUNT} inputs (${CMIN_ELAPSED}s)"
rm -rf "${SEEDS}"
mv "${MINIMIZED}" "${SEEDS}"
else
echo " afl-cmin failed, using unminimized corpus"
rm -rf "${MINIMIZED}"
fi
fi
done
# ── Phase 3: Fuzz all targets in parallel ──
echo ""
echo "=========================================="
echo "Fuzzing ${TARGET_COUNT} target(s) in parallel (${NCPU} cores, ${DURATION}s each)"
echo "=========================================="
# Function that runs in a background subshell per target.
fuzz_target() {
local FUZZ_TARGET="$1"
local FUZZ_BIN="$2"
local SEEDS="$3"
local FINDINGS="$4"
local RESULT_FILE="$5"
local SEED_COUNT="$6"
mkdir -p "${FINDINGS}"
echo "[${FUZZ_TARGET}] Starting AFL++ (${SEED_COUNT} seeds, ${DURATION}s)"
local FUZZ_START=$(date +%s)
local AFL_EXIT=0
{
AFL_NO_UI=1 \
AFL_SKIP_CPUFREQ=1 \
AFL_I_DONT_CARE_ABOUT_MISSING_CRASHES=1 \
AFL_NO_CRASH_README=1 \
afl-fuzz \
-V "${DURATION}" \
-i "${SEEDS}" \
-o "${FINDINGS}" \
${EXTRA_AFL_ARGS} \
-- "${FUZZ_BIN}"
} >/dev/null 2>&1 || AFL_EXIT=$?
local FUZZ_ELAPSED=$(( $(date +%s) - FUZZ_START ))
# Summary from fuzzer_stats.
local STATS_FILE="${FINDINGS}/default/fuzzer_stats"
if [ -f "${STATS_FILE}" ]; then
local EXECS=$(grep -oP 'execs_done\s*:\s*\K\d+' "${STATS_FILE}" || echo "?")
local PATHS=$(grep -oP 'corpus_count\s*:\s*\K\d+' "${STATS_FILE}" || echo "?")
local CRASHES=$(grep -oP 'saved_crashes\s*:\s*\K\d+' "${STATS_FILE}" || echo "?")
local SPEED=$(grep -oP 'execs_per_sec\s*:\s*\K[\d.]+' "${STATS_FILE}" || echo "?")
local COVERAGE=$(grep -oP 'bitmap_cvg\s*:\s*\K[\d.]+%' "${STATS_FILE}" || echo "?")
echo "[${FUZZ_TARGET}] Finished in ${FUZZ_ELAPSED}s: ${EXECS} execs (${SPEED}/s), ${PATHS} paths, ${CRASHES} crashes, ${COVERAGE} coverage (exit ${AFL_EXIT})"
else
echo "[${FUZZ_TARGET}] AFL++ exited with code ${AFL_EXIT} after ${FUZZ_ELAPSED}s"
fi
# Count crashes for the result file.
local CRASH_COUNT=0
local CRASH_DIR="${FINDINGS}/default/crashes"
if [ -d "${CRASH_DIR}" ]; then
CRASH_COUNT=$(find "${CRASH_DIR}" -maxdepth 1 -type f -name 'id:*' | wc -l)
fi
echo "${CRASH_COUNT}" > "${RESULT_FILE}"
}
FUZZ_PHASE_START=$(date +%s)
PIDS=()
for i in "${!TARGET_NAMES[@]}"; do
FUZZ_TARGET="${TARGET_NAMES[$i]}"
FUZZ_BIN="${TARGET_BINS[${FUZZ_TARGET}]}"
SEEDS="work/${FUZZ_TARGET}/seeds"
FINDINGS="work/${FUZZ_TARGET}/findings"
RESULT_FILE="${RESULTS_DIR}/${FUZZ_TARGET}"
SEED_COUNT=$(find "${SEEDS}" -maxdepth 1 -type f | wc -l)
fuzz_target "${FUZZ_TARGET}" "${FUZZ_BIN}" "${SEEDS}" "${FINDINGS}" "${RESULT_FILE}" "${SEED_COUNT}" &
PIDS+=($!)
# Limit parallelism to available cores.
if [ "${#PIDS[@]}" -ge "${NCPU}" ]; then
wait "${PIDS[0]}"
PIDS=("${PIDS[@]:1}")
fi
done
# Wait for remaining jobs.
for pid in "${PIDS[@]}"; do
wait "${pid}"
done
FUZZ_PHASE_ELAPSED=$(( $(date +%s) - FUZZ_PHASE_START ))
echo "Fuzzing phase completed in ${FUZZ_PHASE_ELAPSED}s"
# ── Phase 4: Upload results sequentially ──
echo ""
echo "=========================================="
echo "Uploading results"
echo "=========================================="
TOTAL_CRASHES=0
for i in "${!TARGET_NAMES[@]}"; do
FUZZ_TARGET="${TARGET_NAMES[$i]}"
NUM=$((i + 1))
FUZZ_BIN="${TARGET_BINS[${FUZZ_TARGET}]}"
REPLAY_BIN="${REPLAY_BINS[${FUZZ_TARGET}]}"
RUN_ID="${RUN_IDS[${FUZZ_TARGET}]}"
CAIRN_TARGET_ID="${CAIRN_TARGET_IDS[${FUZZ_TARGET}]}"
FINDINGS="work/${FUZZ_TARGET}/findings"
echo "[${NUM}/${TARGET_COUNT}] ${FUZZ_TARGET}..."
# Upload crashes.
CRASH_DIR="${FINDINGS}/default/crashes"
if [ -d "${CRASH_DIR}" ]; then
for crash_file in "${CRASH_DIR}"/id:*; do
[ -f "${crash_file}" ] || continue
CRASH_NAME=$(basename "${crash_file}")
SIG=""
case "${CRASH_NAME}" in
*,sig:*)
SIG=$(echo "${CRASH_NAME}" | sed 's/.*,sig:\([0-9]*\).*/\1/')
;;
esac
# Replay using the non-AFL binary to get a proper stack trace.
STACK_TRACE=""
CRASH_MSG="AFL++ crash (${FUZZ_TARGET}): ${CRASH_NAME}"
REPLAY_OUTPUT=$(timeout 10 "${REPLAY_BIN}" < "${crash_file}" 2>&1 || true)
if [ -n "${REPLAY_OUTPUT}" ]; then
STACK_TRACE="${REPLAY_OUTPUT}"
FIRST_LINE=$(echo "${REPLAY_OUTPUT}" | grep -m1 -iE 'panic|error|fault|abort|overflow|undefined|sanitizer|SUMMARY' || true)
if [ -n "${FIRST_LINE}" ]; then
CRASH_MSG="${FIRST_LINE}"
fi
echo " Replay (${CRASH_NAME}):"
echo "${REPLAY_OUTPUT}" | head -10 | sed 's/^/ /'
else
echo " Replay produced no output for ${CRASH_NAME}"
fi
echo " Uploading crash: ${CRASH_NAME}"
set -- -server "${CAIRN_SERVER}" -repo "${REPO}" -owner "${OWNER}" \
-commit "${COMMIT}" -run-id "${RUN_ID}" -type fuzz -file "${crash_file}" \
-kind crash \
-crash-message "${CRASH_MSG}"
if [ -n "${STACK_TRACE}" ]; then
set -- "$@" -stack-trace "${STACK_TRACE}"
fi
if [ -n "${TARGET_PLATFORM}" ]; then
set -- "$@" -target "${TARGET_PLATFORM}"
fi
if [ -n "${SIG}" ]; then
set -- "$@" -signal "${SIG}"
fi
cairn upload "$@"
TOTAL_CRASHES=$((TOTAL_CRASHES + 1))
done
fi
# Minimize and upload corpus.
QUEUE_DIR="${FINDINGS}/default/queue"
if [ -d "${QUEUE_DIR}" ]; then
QUEUE_COUNT=$(find "${QUEUE_DIR}" -maxdepth 1 -type f -name 'id:*' | wc -l)
if [ "${QUEUE_COUNT}" -gt 0 ]; then
UPLOAD_DIR="work/${FUZZ_TARGET}/corpus-upload"
rm -rf "${UPLOAD_DIR}"
mkdir -p "${UPLOAD_DIR}"
CMIN_START=$(date +%s)
if afl-cmin -i "${QUEUE_DIR}" -o "${UPLOAD_DIR}" -- "${FUZZ_BIN}" >/dev/null 2>&1; then
CMIN_ELAPSED=$(( $(date +%s) - CMIN_START ))
UPLOAD_COUNT=$(find "${UPLOAD_DIR}" -maxdepth 1 -type f | wc -l)
echo " Corpus minimized: ${QUEUE_COUNT} -> ${UPLOAD_COUNT} entries (${CMIN_ELAPSED}s)"
else
echo " afl-cmin failed, uploading full queue"
rm -rf "${UPLOAD_DIR}"
UPLOAD_DIR="${QUEUE_DIR}"
UPLOAD_COUNT="${QUEUE_COUNT}"
fi
echo " Uploading corpus (${UPLOAD_COUNT} entries)..."
cairn corpus upload \
-server "${CAIRN_SERVER}" \
-target-id "${CAIRN_TARGET_ID}" \
-run-id "${RUN_ID}" \
-dir "${UPLOAD_DIR}"
fi
fi
# Finish run.
cairn run finish -server "${CAIRN_SERVER}" -id "${RUN_ID}" || true
done
# ── Final report ──
echo ""
echo "=========================================="
echo "Fuzzed ${TARGET_COUNT} target(s), ${TOTAL_CRASHES} total crash(es)"
echo "=========================================="
if [ "${TOTAL_CRASHES}" -gt 0 ]; then
echo "WARNING: ${TOTAL_CRASHES} crash(es) found and uploaded to Cairn"
else
echo "OK: No crashes found"
fi