Verify the release bundle hash before extracting (both images)

simplerisk-minimal/Dockerfile

@@ -3,22 +3,20 @@ ARG php_version=8.5
 
 FROM alpine/curl:8.12.1 AS downloader
 
-# PREGA_BUNDLE_FALLBACK is a CI-ONLY switch, default false. Pre-GA the prod
-# bundle for a new version does not exist yet (it lands in public/bundles/ only
-# at GA). CI sets this true so a pre-GA build can fall back to the testing
-# bundle. A released image is ALWAYS built from the prod bundle and NEVER
-# silently falls back to testing bytes (even on a transient prod failure).
+# PREGA_BUNDLE_FALLBACK is a CI-ONLY switch, default false. Pre-GA the prod bundle
+# for a new version does not exist yet (it lands in public/bundles/ only at GA). CI
+# sets this true so a pre-GA build can fall back to the testing bundle WITHOUT hash
+# verification (the release has no published hash yet). A released image is ALWAYS
+# built from the VERIFIED prod bundle and NEVER from unverified testing bytes.
 ARG PREGA_BUNDLE_FALLBACK=false
 
 SHELL [ "/bin/ash", "-eo", "pipefail", "-c" ]
 
-RUN mkdir -p /var/www && \
-    if curl -fsSL https://simplerisk-downloads.s3.amazonaws.com/public/bundles/simplerisk-20260519-001.tgz -o /tmp/bundle.tgz; then true; \
-    elif [ "$PREGA_BUNDLE_FALLBACK" = "true" ]; then \
-      echo "prod bundle absent — pre-GA CI fallback to bundles-test"; \
-      curl -fsSL https://bundles-test.simplerisk.com/simplerisk-20260519-001.tgz -o /tmp/bundle.tgz; \
-    else echo "prod bundle simplerisk-20260519-001.tgz not found and PREGA_BUNDLE_FALLBACK=false" && exit 1; fi && \
-    tar xzf /tmp/bundle.tgz -C /var/www && rm -f /tmp/bundle.tgz
+# Download the prod bundle, verify its published sha256 (md5 fallback) from the
+# updates feed, then extract -- fail-closed unless PREGA_BUNDLE_FALLBACK allows the
+# pre-GA path. See common/download_and_verify_bundle.sh.
+COPY common/download_and_verify_bundle.sh /download_and_verify_bundle.sh
+RUN PREGA_BUNDLE_FALLBACK="$PREGA_BUNDLE_FALLBACK" sh /download_and_verify_bundle.sh 20260519-001
 
 FROM php:${php_version}-apache
 

simplerisk-minimal/common/download_and_verify_bundle.sh

@@ -0,0 +1,61 @@
+#!/bin/sh
+# Download the released SimpleRisk bundle for the given version, verify it against
+# the sha256 (md5 fallback) published in the production updates feed, then extract
+# it into /var/www. Run from the alpine/curl downloader stage of the generated
+# Dockerfile.
+#
+# Fail-closed by default: a RELEASED image (PREGA_BUNDLE_FALLBACK unset/false) is
+# built ONLY from the prod bundle and ONLY if it matches its published hash -- a
+# missing prod bundle, a missing feed hash, or a mismatch aborts the build, so a
+# swapped S3 object can never be baked into a published image. The bundle (S3
+# public/bundles) and its hash (served updates feed) are stored independently.
+#
+# PRE-GA CI ONLY: before GA the prod bundle/hash for a new version do not exist
+# yet (they land at GA). When PREGA_BUNDLE_FALLBACK=true AND the prod bundle is
+# absent, fall back to the testing bundle WITHOUT verification (there is no
+# published hash to check yet) and warn loudly. This path is never taken for a
+# released image.
+set -eu
+
+VERSION="${1:?usage: download_and_verify_bundle.sh <YYYYMMDD-NNN>}"
+case "$VERSION" in
+  [0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]-[0-9][0-9][0-9]) : ;;
+  *) echo "ERROR: bad version format: $VERSION" >&2; exit 1 ;;
+esac
+
+FEED="https://updates.simplerisk.com/releases.xml"
+PROD_URL="https://simplerisk-downloads.s3.amazonaws.com/public/bundles/simplerisk-${VERSION}.tgz"
+TEST_URL="https://bundles-test.simplerisk.com/simplerisk-${VERSION}.tgz"
+TGZ="/tmp/simplerisk-${VERSION}.tgz"
+
+if curl -fsSL -o "$TGZ" "$PROD_URL"; then
+  echo "Downloaded prod bundle for ${VERSION}; resolving published hash from ${FEED} ..."
+  ENTRY="$(curl -fsSL "$FEED" | sed -n "/<release version=\"${VERSION}\">/,/<\/release>/p")"
+  EXPECTED="$(printf '%s\n' "$ENTRY" | grep -oE '<bundle_sha256>[0-9a-f]{64}</bundle_sha256>' | grep -oE '[0-9a-f]{64}' | head -1 || true)"
+  ALGO=sha256
+  if [ -z "$EXPECTED" ]; then
+    EXPECTED="$(printf '%s\n' "$ENTRY" | grep -oE '<bundle_md5>[0-9a-f]{32}</bundle_md5>' | grep -oE '[0-9a-f]{32}' | head -1 || true)"
+    ALGO=md5
+  fi
+  if [ -z "$EXPECTED" ]; then
+    echo "ERROR: no bundle_sha256 or bundle_md5 for ${VERSION} in ${FEED} -- refusing to extract an unverifiable prod bundle" >&2
+    exit 1
+  fi
+  ACTUAL="$(${ALGO}sum "$TGZ" | cut -d' ' -f1)"
+  if [ "$ACTUAL" != "$EXPECTED" ]; then
+    echo "ERROR: bundle ${ALGO} mismatch for ${VERSION} -- expected ${EXPECTED}, got ${ACTUAL}" >&2
+    exit 1
+  fi
+  echo "Bundle ${ALGO} verified (${ACTUAL})."
+elif [ "${PREGA_BUNDLE_FALLBACK:-false}" = "true" ]; then
+  echo "WARNING: prod bundle simplerisk-${VERSION}.tgz absent -- PRE-GA CI fallback to bundles-test (UNVERIFIED: the release has no published hash yet)." >&2
+  curl -fsSL -o "$TGZ" "$TEST_URL"
+else
+  echo "ERROR: prod bundle simplerisk-${VERSION}.tgz not found and PREGA_BUNDLE_FALLBACK != true -- refusing to build a release from unverified bytes" >&2
+  exit 1
+fi
+
+mkdir -p /var/www
+tar xzf "$TGZ" -C /var/www
+rm -f "$TGZ"
+echo "Extracted bundle to /var/www."

simplerisk-minimal/generate_dockerfile.sh

@@ -34,22 +34,20 @@ if [ "$source_mode" == "download" ]; then
 	cat << EOF >> "${SCRIPT_LOCATION}/Dockerfile"
 FROM alpine/curl:8.12.1 AS downloader
 
-# PREGA_BUNDLE_FALLBACK is a CI-ONLY switch, default false. Pre-GA the prod
-# bundle for a new version does not exist yet (it lands in public/bundles/ only
-# at GA). CI sets this true so a pre-GA build can fall back to the testing
-# bundle. A released image is ALWAYS built from the prod bundle and NEVER
-# silently falls back to testing bytes (even on a transient prod failure).
+# PREGA_BUNDLE_FALLBACK is a CI-ONLY switch, default false. Pre-GA the prod bundle
+# for a new version does not exist yet (it lands in public/bundles/ only at GA). CI
+# sets this true so a pre-GA build can fall back to the testing bundle WITHOUT hash
+# verification (the release has no published hash yet). A released image is ALWAYS
+# built from the VERIFIED prod bundle and NEVER from unverified testing bytes.
 ARG PREGA_BUNDLE_FALLBACK=false
 
 SHELL [ "/bin/ash", "-eo", "pipefail", "-c" ]
 
-RUN mkdir -p /var/www && \\
-    if curl -fsSL https://simplerisk-downloads.s3.amazonaws.com/public/bundles/simplerisk-$release.tgz -o /tmp/bundle.tgz; then true; \\
-    elif [ "\$PREGA_BUNDLE_FALLBACK" = "true" ]; then \\
-      echo "prod bundle absent — pre-GA CI fallback to bundles-test"; \\
-      curl -fsSL https://bundles-test.simplerisk.com/simplerisk-$release.tgz -o /tmp/bundle.tgz; \\
-    else echo "prod bundle simplerisk-$release.tgz not found and PREGA_BUNDLE_FALLBACK=false" && exit 1; fi && \\
-    tar xzf /tmp/bundle.tgz -C /var/www && rm -f /tmp/bundle.tgz
+# Download the prod bundle, verify its published sha256 (md5 fallback) from the
+# updates feed, then extract -- fail-closed unless PREGA_BUNDLE_FALLBACK allows the
+# pre-GA path. See common/download_and_verify_bundle.sh.
+COPY common/download_and_verify_bundle.sh /download_and_verify_bundle.sh
+RUN PREGA_BUNDLE_FALLBACK="\$PREGA_BUNDLE_FALLBACK" sh /download_and_verify_bundle.sh $release
 
 EOF
 fi

simplerisk/Dockerfile

@@ -4,11 +4,15 @@ ARG ubuntu_version_code=noble
 FROM alpine/curl:8.12.1 AS downloader
 
 ARG DB_LANG=en
+# CI-ONLY pre-GA switch (default false) — see common/download_and_verify_bundle.sh.
+ARG PREGA_BUNDLE_FALLBACK=false
 
 SHELL [ "/bin/ash", "-eo", "pipefail", "-c" ]
 
-RUN mkdir -p /var/www && \
-    curl -sL https://simplerisk-downloads.s3.amazonaws.com/public/bundles/simplerisk-20260519-001.tgz | tar xz -C /var/www && \
+# Download the prod bundle, verify its published sha256 (md5 fallback) from the
+# updates feed, then extract (fail-closed) -- then fetch the release SQL schema.
+COPY common/download_and_verify_bundle.sh /download_and_verify_bundle.sh
+RUN PREGA_BUNDLE_FALLBACK="$PREGA_BUNDLE_FALLBACK" sh /download_and_verify_bundle.sh 20260519-001 && \
     curl -sL "https://github.com/simplerisk/database/raw/master/simplerisk-$DB_LANG-20260519-001.sql" > /simplerisk.sql
 
 # Using Ubuntu image

simplerisk/common/download_and_verify_bundle.sh

@@ -0,0 +1,61 @@
+#!/bin/sh
+# Download the released SimpleRisk bundle for the given version, verify it against
+# the sha256 (md5 fallback) published in the production updates feed, then extract
+# it into /var/www. Run from the alpine/curl downloader stage of the generated
+# Dockerfile.
+#
+# Fail-closed by default: a RELEASED image (PREGA_BUNDLE_FALLBACK unset/false) is
+# built ONLY from the prod bundle and ONLY if it matches its published hash -- a
+# missing prod bundle, a missing feed hash, or a mismatch aborts the build, so a
+# swapped S3 object can never be baked into a published image. The bundle (S3
+# public/bundles) and its hash (served updates feed) are stored independently.
+#
+# PRE-GA CI ONLY: before GA the prod bundle/hash for a new version do not exist
+# yet (they land at GA). When PREGA_BUNDLE_FALLBACK=true AND the prod bundle is
+# absent, fall back to the testing bundle WITHOUT verification (there is no
+# published hash to check yet) and warn loudly. This path is never taken for a
+# released image.
+set -eu
+
+VERSION="${1:?usage: download_and_verify_bundle.sh <YYYYMMDD-NNN>}"
+case "$VERSION" in
+  [0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]-[0-9][0-9][0-9]) : ;;
+  *) echo "ERROR: bad version format: $VERSION" >&2; exit 1 ;;
+esac
+
+FEED="https://updates.simplerisk.com/releases.xml"
+PROD_URL="https://simplerisk-downloads.s3.amazonaws.com/public/bundles/simplerisk-${VERSION}.tgz"
+TEST_URL="https://bundles-test.simplerisk.com/simplerisk-${VERSION}.tgz"
+TGZ="/tmp/simplerisk-${VERSION}.tgz"
+
+if curl -fsSL -o "$TGZ" "$PROD_URL"; then
+  echo "Downloaded prod bundle for ${VERSION}; resolving published hash from ${FEED} ..."
+  ENTRY="$(curl -fsSL "$FEED" | sed -n "/<release version=\"${VERSION}\">/,/<\/release>/p")"
+  EXPECTED="$(printf '%s\n' "$ENTRY" | grep -oE '<bundle_sha256>[0-9a-f]{64}</bundle_sha256>' | grep -oE '[0-9a-f]{64}' | head -1 || true)"
+  ALGO=sha256
+  if [ -z "$EXPECTED" ]; then
+    EXPECTED="$(printf '%s\n' "$ENTRY" | grep -oE '<bundle_md5>[0-9a-f]{32}</bundle_md5>' | grep -oE '[0-9a-f]{32}' | head -1 || true)"
+    ALGO=md5
+  fi
+  if [ -z "$EXPECTED" ]; then
+    echo "ERROR: no bundle_sha256 or bundle_md5 for ${VERSION} in ${FEED} -- refusing to extract an unverifiable prod bundle" >&2
+    exit 1
+  fi
+  ACTUAL="$(${ALGO}sum "$TGZ" | cut -d' ' -f1)"
+  if [ "$ACTUAL" != "$EXPECTED" ]; then
+    echo "ERROR: bundle ${ALGO} mismatch for ${VERSION} -- expected ${EXPECTED}, got ${ACTUAL}" >&2
+    exit 1
+  fi
+  echo "Bundle ${ALGO} verified (${ACTUAL})."
+elif [ "${PREGA_BUNDLE_FALLBACK:-false}" = "true" ]; then
+  echo "WARNING: prod bundle simplerisk-${VERSION}.tgz absent -- PRE-GA CI fallback to bundles-test (UNVERIFIED: the release has no published hash yet)." >&2
+  curl -fsSL -o "$TGZ" "$TEST_URL"
+else
+  echo "ERROR: prod bundle simplerisk-${VERSION}.tgz not found and PREGA_BUNDLE_FALLBACK != true -- refusing to build a release from unverified bytes" >&2
+  exit 1
+fi
+
+mkdir -p /var/www
+tar xzf "$TGZ" -C /var/www
+rm -f "$TGZ"
+echo "Extracted bundle to /var/www."

simplerisk/generate_dockerfile.sh

@@ -22,11 +22,15 @@ if [ "$release" != "testing" ]; then
 FROM alpine/curl:8.12.1 AS downloader
 
 ARG DB_LANG=en
+# CI-ONLY pre-GA switch (default false) — see common/download_and_verify_bundle.sh.
+ARG PREGA_BUNDLE_FALLBACK=false
 
 SHELL [ "/bin/ash", "-eo", "pipefail", "-c" ]
 
-RUN mkdir -p /var/www && \\
-    curl -sL https://simplerisk-downloads.s3.amazonaws.com/public/bundles/simplerisk-$release.tgz | tar xz -C /var/www && \\
+# Download the prod bundle, verify its published sha256 (md5 fallback) from the
+# updates feed, then extract (fail-closed) -- then fetch the release SQL schema.
+COPY common/download_and_verify_bundle.sh /download_and_verify_bundle.sh
+RUN PREGA_BUNDLE_FALLBACK="\$PREGA_BUNDLE_FALLBACK" sh /download_and_verify_bundle.sh $release && \\
     curl -sL "https://github.com/simplerisk/database/raw/master/simplerisk-\$DB_LANG-$release.sql" > /simplerisk.sql
 
 EOF