DLPX-98872 linux-pkg: generate per-package CycloneDX SBOM via Syft deb scan and publish S3 sidecar

.github/scripts/verify-sbom-scan-flag.sh

@@ -0,0 +1,34 @@
+#!/bin/bash -ex
+
+set -o pipefail
+
+cd "$(git rev-parse --show-toplevel)"
+
+#
+# Every package must explicitly set SBOM_DEEP_SCAN to "true" or "false" in
+# its config.sh -- see docs on generate_sbom() in lib/common.sh. There is
+# no default: a package that hasn't been classified yet must fail CI
+# rather than silently ship without a CycloneDX sidecar or without an
+# explicit decision that it doesn't need one.
+#
+unclassified=$(./query-packages.sh list -o name,sbom-deep-scan all |
+	awk -F'\t' '$2 == "none" { print $1 }')
+
+if [[ -n "$unclassified" ]]; then
+	echo "The following packages have not set SBOM_DEEP_SCAN (\"true\" or" \
+		"\"false\") in their config.sh:"
+	echo "$unclassified"
+	echo
+	echo "Set it to \"true\" for 1st-party packages, so that the" \
+		"third-party components they package internally (jars, npm" \
+		"modules, Rust crates, ...) are included in the product's" \
+		"aggregate SBOM."
+	echo "Set it to \"false\" for 3rd-party forks of Debian packages, and" \
+		"for packages that are not included in a shipping product:" \
+		"those are already covered as a flat pkg:deb component by" \
+		"appliance-build's image-level scan, so a deep scan here would" \
+		"add nothing."
+	exit 1
+fi
+
+echo "All packages have classified SBOM_DEEP_SCAN"

.github/workflows/main.yml

@@ -29,3 +29,8 @@ jobs:
     steps:
       - uses: actions/checkout@v1
       - run: ./.github/scripts/verify-query-packages.sh
+  verify-sbom-scan-flag:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v1
+      - run: ./.github/scripts/verify-sbom-scan-flag.sh

buildpkg.sh

@@ -137,6 +137,9 @@ stage build
 logmust cd "$WORKDIR"
 stage store_build_info
 
+logmust cd "$WORKDIR"
+stage generate_sbom
+
 logmust cd "$WORKDIR"
 stage post_build_checks
 

lib/common.sh

@@ -619,6 +619,66 @@ function install_shfmt() {
 	echo "shfmt version $(shfmt -version) is installed."
 }
 
+#
+# Install the tooling generate_sbom() needs. That is a default hook, shared
+# unmodified by every package that sets SBOM_DEEP_SCAN, so the tools it runs
+# belong with the rest of the generic build tooling installed here rather than
+# being declared as a dependency by each of those packages individually.
+#
+# Unlike everything else installed from setup.sh these are Delphix-built
+# packages with no apt source, so they are fetched from the same S3 location
+# fetch_dependencies() pulls a package's dependencies from, then installed by
+# path. apt rather than dpkg, because delphix-cyclonedx-cli has real
+# dependencies (libicu and the usual shared libraries) that dpkg will not
+# resolve.
+#
+# Best-effort by design: this runs before *every* package build, including the
+# builds of syft and cyclonedx-cli themselves, and on a branch where neither
+# has been published yet there is nothing to fetch. Failing hard would break
+# every build on such a branch rather than just SBOM generation, so a missing
+# artifact warns and moves on; generate_sbom() checks for the tools itself and
+# fails loudly, for the only builds that actually need them.
+#
+function install_sbom_tools() {
+	local pkg s3url tmpdir
+	local debs=()
+
+	tmpdir="$(mktemp -d)" || die "Failed to create a temporary directory"
+
+	for pkg in syft cyclonedx-cli; do
+		#
+		# Run in a command substitution so that a failure to resolve
+		# the URL (get_package_dependency_s3_url dies when a package
+		# has no published artifacts) leaves $s3url empty here instead
+		# of aborting setup.
+		#
+		s3url="$(
+			get_package_dependency_s3_url "$pkg" >/dev/null 2>&1
+			echo "$_RET"
+		)"
+		if [[ -z "$s3url" ]]; then
+			echo "WARNING: no published artifacts found for '$pkg';" \
+				"skipping the SBOM tooling install. Builds of" \
+				"packages that set SBOM_DEEP_SCAN will fail until" \
+				"'$pkg' has been built for this branch."
+			logmust rm -rf "$tmpdir"
+			return 0
+		fi
+
+		[[ "$s3url" != */ ]] && s3url="$s3url/"
+		logmust mkdir -p "$tmpdir/$pkg"
+		logmust aws s3 cp --only-show-errors --recursive \
+			"$s3url" "$tmpdir/$pkg/"
+	done
+
+	debs=("$tmpdir"/*/*.deb)
+	[[ -e "${debs[0]}" ]] ||
+		die "No .deb found in the fetched syft/cyclonedx-cli artifacts"
+
+	logmust install_pkgs "${debs[@]}"
+	logmust rm -rf "$tmpdir"
+}
+
 #
 # Install kernel headers packages for all target kernels.
 # The kernel packages are fetched from S3.
@@ -1461,6 +1521,115 @@ function store_build_info() {
 	fi
 }
 
+#
+# Generate a CycloneDX SBOM sidecar for each of this package's built
+# .deb(s) by running Syft against it. Only packages that bundle
+# third-party composition (jars, npm, wheels, Rust crates, ...) opt in
+# via SBOM_DEEP_SCAN="true" in their config.sh -- everything else is
+# left as a flat pkg:deb component by appliance-build's base chroot
+# scan, so a deb scan here would add nothing. Each <deb-filename>.cdx.json
+# is dropped in $WORKDIR/artifacts/ alongside the .deb it describes --
+# a strict 1:1 mapping, no merging across a package's .deb(s) -- where
+# it's picked up by the same S3 sync as every other build artifact, no
+# separate upload path needed.
+#
+function generate_sbom() {
+	if [[ "$SBOM_DEEP_SCAN" != "true" ]]; then
+		return 0
+	fi
+
+	local debs=("$WORKDIR/artifacts/"*.deb)
+	if [[ ! -e "${debs[0]}" ]]; then
+		die "SBOM_DEEP_SCAN is set but no .deb was found in" \
+			"'$WORKDIR/artifacts'"
+	fi
+
+	#
+	# syft/cyclonedx-cli are part of the generic build tooling installed
+	# by setup.sh (install_sbom_tools()) before any package is built,
+	# rather than something each SBOM_DEEP_SCAN package declares for
+	# itself -- this is a default hook, so what it needs is its own
+	# concern, not its callers'. That install is best-effort, so check
+	# here rather than letting "command not found" surface from the
+	# middle of a scan.
+	#
+	local tool
+	for tool in syft cyclonedx-cli; do
+		command -v "$tool" >/dev/null ||
+			die "'$tool' is not installed, so no SBOM can be" \
+				"generated for '$PACKAGE'. It is provisioned by" \
+				"install_sbom_tools() in setup.sh; check that" \
+				"run's output for why it was skipped."
+	done
+
+	#
+	# One sidecar per .deb, not per package: a package that emits more
+	# than one .deb (e.g. "zfs" splits into zfs-dkms, zfsutils-linux,
+	# etc.) gets one <deb-filename>.deb.cdx.json per .deb, each a
+	# standalone document scoped to that .deb alone. No merging across
+	# .debs -- keeps a strict 1:1 mapping between a .deb and its BOM,
+	# with the .deb's own filename as the common prefix.
+	#
+	local deb
+	for deb in "${debs[@]}"; do
+		local sbom_file deb_version extract_dir
+		sbom_file="$WORKDIR/artifacts/$(basename "$deb").cdx.json"
+		#
+		# Read the version back out of the .deb itself, rather than
+		# relying on $PACKAGE_VERSION: by this point in the build,
+		# $PACKAGE_VERSION may no longer hold the final,
+		# revision-suffixed version set_changelog() wrote into the
+		# package (e.g. it's empty for packages that don't set it
+		# explicitly themselves, unlike syft/cyclonedx-cli's own
+		# config.sh). dpkg-deb reads the actual, authoritative
+		# version of the artifact being scanned.
+		#
+		deb_version="$(dpkg-deb -f "$deb" Version)"
+
+		#
+		# Scan the .deb's extracted payload rather than the .deb file.
+		# "syft scan <file>.deb" only identifies the archive: its
+		# deb-archive-cataloger reads the control metadata and emits a
+		# single pkg:deb component, never descending into data.tar.*,
+		# so none of the bundled jars/wheels/modules this sidecar
+		# exists to capture are found. That produced valid but empty
+		# documents -- one component for a 1.2GB application -- which
+		# is no more than appliance-build's image-level dpkg scan
+		# already gives for free. Extracting first is what the design
+		# spec prescribed for exactly this case.
+		#
+		extract_dir="$(logmust mktemp -d)"
+		logmust dpkg-deb -x "$deb" "$extract_dir"
+
+		#
+		# Full catalogers here, deliberately unlike appliance-build's
+		# 95-generate-sbom.binary hook, which restricts to dpkg. That
+		# restriction exists because a jar sitting somewhere on a whole
+		# rootfs cannot be attributed to the package that placed it;
+		# within a single package's own extracted payload everything
+		# found belongs to that package by construction, which is the
+		# entire point of scanning here rather than at image level.
+		#
+		# SYFT_FILE_METADATA_SELECTION=none suppresses Syft's default
+		# per-file "file" component (with SHA-1/SHA-256 hashes and the
+		# scanned path baked in) -- with a whole extracted payload to
+		# walk that would otherwise emit an entry per file.
+		#
+		SYFT_FILE_METADATA_SELECTION=none logmust syft scan "dir:$extract_dir" \
+			--source-name "$PACKAGE" \
+			--source-version "$deb_version" \
+			-o "cyclonedx-json@1.6=$sbom_file"
+
+		logmust rm -rf "$extract_dir"
+
+		logmust cyclonedx-cli validate \
+			--input-file "$sbom_file" \
+			--input-format json \
+			--input-version v1_6 \
+			--fail-on-errors
+	done
+}
+
 function set_secret_build_args() {
 	_SECRET_BUILD_ARGS=()
 

packages/bcc/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/bcc.git"
+SBOM_DEEP_SCAN="false"
 
 UPSTREAM_GIT_URL=https://github.com/iovisor/bcc.git
 UPSTREAM_GIT_BRANCH=master

packages/challenge-response/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/challenge-response.git"
+SBOM_DEEP_SCAN="false"
 
 function prepare() {
 	install_build_deps_from_control_file

packages/cloud-init/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/cloud-init.git"
+SBOM_DEEP_SCAN="false"
 
 UPSTREAM_SOURCE_PACKAGE=cloud-init
 

packages/connstat/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/connstat.git"
+SBOM_DEEP_SCAN="false"
 PACKAGE_DEPENDENCIES="@linux-kernel dwarves"
 
 function prepare() {

packages/containerized-masking/config.sh

@@ -40,6 +40,8 @@ DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/dms-core-gate.git"
 #
 PACKAGE_NEEDS_DOCKER="true"
 MEND_SCAN_APPLICABLE="true"
+SBOM_DEEP_SCAN="true"
+
 MEND_SCAN_IMAGES="'delphix-masking-proxy', 'delphix-masking-database', 'delphix-masking-app'"
 
 SKIP_COPYRIGHTS_CHECK=true

packages/crash-python/config.sh

@@ -17,6 +17,7 @@
 
 # shellcheck disable=SC2034
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/crash-python.git"
+SBOM_DEEP_SCAN="false"
 
 function prepare() {
 	logmust install_build_deps_from_control_file

packages/crypt-blowfish/config.sh

@@ -17,6 +17,8 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/crypt-blowfish.git"
+SBOM_DEEP_SCAN="false"
+
 SKIP_COPYRIGHTS_CHECK=true
 
 function build() {

packages/cyclonedx-cli/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/cyclonedx-cli.git"
+SBOM_DEEP_SCAN="false"
 
 function build() {
 	logmust mkdir -p "$WORKDIR/repo"

packages/delphix-go/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/delphix-go.git"
+SBOM_DEEP_SCAN="false"
 
 function build() {
 	logmust mkdir -p "$WORKDIR/repo"

packages/delphix-kernel/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/delphix-kernel.git"
+SBOM_DEEP_SCAN="false"
 PACKAGE_DEPENDENCIES="@linux-kernel"
 
 function prepare() {

packages/delphix-platform/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/delphix-platform.git"
+SBOM_DEEP_SCAN="false"
 
 function prepare() {
 	logmust cd "$WORKDIR/repo"

packages/delphix-rust/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/delphix-rust.git"
+SBOM_DEEP_SCAN="true"
 
 function build() {
 	logmust mkdir -p "$WORKDIR/repo"

packages/delphix-sso-app/config.sh

@@ -19,6 +19,7 @@
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/saml-app.git"
 MEND_SCAN_APPLICABLE="true"
 MEND_SCAN_USING_SUDO="true"
+SBOM_DEEP_SCAN="true"
 
 function prepare() {
 	logmust install_pkgs openjdk-17-jdk-headless:

packages/docker-python-image/config.sh

@@ -17,6 +17,8 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/docker-python-image.git"
+SBOM_DEEP_SCAN="false"
+
 #
 # debian/rules' override_dh_install runs 'docker pull' to fetch the python
 # image it repackages, so the build needs a docker daemon. The build container

packages/drgn/config.sh

@@ -17,6 +17,7 @@
 
 # shellcheck disable=SC2034
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/drgn.git"
+SBOM_DEEP_SCAN="false"
 PACKAGE_DEPENDENCIES="libkdumpfile"
 
 UPSTREAM_GIT_URL="https://github.com/osandov/drgn.git"

packages/dwarves/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/dwarves.git"
+SBOM_DEEP_SCAN="false"
 
 UPSTREAM_GIT_URL="https://github.com/acmel/dwarves.git"
 UPSTREAM_GIT_BRANCH="master"

packages/fluentd-gems/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/fluentd-gems.git"
+SBOM_DEEP_SCAN="false"
 
 function build() {
 	logmust mkdir -p "$WORKDIR/repo"

packages/gdb-python/config.sh

@@ -17,6 +17,7 @@
 
 # shellcheck disable=SC2034
 DEFAULT_PACKAGE_GIT_URL="https://github.com/delphix/gdb-python.git"
+SBOM_DEEP_SCAN="false"
 PACKAGE_DEPENDENCIES="libkdumpfile"
 
 function prepare() {

packages/grub2/config.sh

@@ -17,6 +17,7 @@
 # shellcheck disable=SC2034
 
 DEFAULT_PACKAGE_GIT_URL=none
+SBOM_DEEP_SCAN="false"
 SKIP_COPYRIGHTS_CHECK=true
 
 URI="s3://release-de-images/internal-artifacts/2025.3.0.1/1.0.53/input-artifacts/combined-packages/packages/grub2"