Use an caching apt proxy instead of copying /var/cache/apt/archives/*.deb

- only download Release files once and not by apt as well as with curl
   and thus avoid a mirror push happening between both downloads
 - no heuristic needed to place the file in their correct mirror
   location
 - no manual checksum checking
 - only throttle download speed when actually downloading and not when
   retrieving files from the cache
 - no translation of filenames between how the epoch colon is stored in
   files in /var/cache/apt/archives versus how it is stored in files on
   the mirrors
 - no special handling of stable update and security mirrors
 - implemented in Python instead of shell and thus an order of magnitude
   faster

caching_proxy.py

@@ -0,0 +1,108 @@
+#!/usr/bin/env python3
+
+import sys
+import os
+import time
+import http.client
+import http.server
+from io import StringIO
+import pathlib
+import urllib.parse
+
+oldcachedir = None
+newcachedir = None
+readonly = False
+
+
+class ProxyRequestHandler(http.server.BaseHTTPRequestHandler):
+    def do_GET(self):
+        assert int(self.headers.get("Content-Length", 0)) == 0
+        assert self.headers["Host"]
+        pathprefix = "http://" + self.headers["Host"] + "/"
+        assert self.path.startswith(pathprefix)
+        sanitizedpath = urllib.parse.unquote(self.path.removeprefix(pathprefix))
+        oldpath = oldcachedir / sanitizedpath
+        newpath = newcachedir / sanitizedpath
+
+        if not readonly:
+            newpath.parent.mkdir(parents=True, exist_ok=True)
+
+        # just send back to client
+        if newpath.exists():
+            print(f"proxy cached: {self.path}", file=sys.stderr)
+            self.wfile.write(b"HTTP/1.1 200 OK\r\n")
+            self.send_header("Content-Length", newpath.stat().st_size)
+            self.end_headers()
+            with newpath.open(mode="rb") as new:
+                while True:
+                    buf = new.read(64 * 1024)  # same as shutil uses
+                    if not buf:
+                        break
+                    self.wfile.write(buf)
+            self.wfile.flush()
+            return
+
+        if readonly:
+            newpath = pathlib.Path("/dev/null")
+
+        # copy from oldpath to newpath and send back to client
+        if oldpath.exists():
+            print(f"proxy cached: {self.path}", file=sys.stderr)
+            self.wfile.write(b"HTTP/1.1 200 OK\r\n")
+            self.send_header("Content-Length", oldpath.stat().st_size)
+            self.end_headers()
+            with oldpath.open(mode="rb") as old, newpath.open(mode="wb") as new:
+                while True:
+                    buf = old.read(64 * 1024)  # same as shutil uses
+                    if not buf:
+                        break
+                    self.wfile.write(buf)
+                    new.write(buf)
+            self.wfile.flush()
+            return
+
+        # download fresh copy
+        try:
+            print(f"\rproxy download: {self.path}", file=sys.stderr)
+            conn = http.client.HTTPConnection(self.headers["Host"], timeout=5)
+            conn.request("GET", self.path, None, dict(self.headers))
+            res = conn.getresponse()
+            assert (res.status, res.reason) == (200, "OK"), (res.status, res.reason)
+            self.wfile.write(b"HTTP/1.1 200 OK\r\n")
+            for k, v in res.getheaders():
+                # do not allow a persistent connection
+                if k == "connection":
+                    continue
+                self.send_header(k, v)
+            self.end_headers()
+            with newpath.open(mode="wb") as f:
+                while True:
+                    buf = res.read(64 * 1024)  # same as shutil uses
+                    if not buf:
+                        break
+                    self.wfile.write(buf)
+                    f.write(buf)
+                    time.sleep(64 / 1024)  # 1024 kB/s
+            self.wfile.flush()
+        except Exception as e:
+            self.send_error(502)
+
+
+def main():
+    global oldcachedir, newcachedir, readonly
+    if sys.argv[1] == "--readonly":
+        readonly = True
+        oldcachedir = pathlib.Path(sys.argv[2])
+        newcachedir = pathlib.Path(sys.argv[3])
+    else:
+        oldcachedir = pathlib.Path(sys.argv[1])
+        newcachedir = pathlib.Path(sys.argv[2])
+    print(f"starting caching proxy for {newcachedir}", file=sys.stderr)
+    httpd = http.server.ThreadingHTTPServer(
+        server_address=("", 8080), RequestHandlerClass=ProxyRequestHandler
+    )
+    httpd.serve_forever()
+
+
+if __name__ == "__main__":
+    main()

coverage.py

@@ -39,7 +39,7 @@ all_formats = ["auto", "directory", "tar", "squashfs", "ext2", "null"]
 mirror = os.getenv("mirror", "http://127.0.0.1/debian")
 hostarch = subprocess.check_output(["dpkg", "--print-architecture"]).decode().strip()
 
-release_path = f"./shared/cache/debian/dists/{default_dist}/Release"
+release_path = f"./shared/cache/debian/dists/{default_dist}/InRelease"
 if not os.path.exists(release_path):
     print("path doesn't exist:", release_path, file=sys.stderr)
     print("run ./make_mirror.sh first", file=sys.stderr)

coverage.sh

@@ -22,8 +22,10 @@ if [ -e ./mmdebstrap ]; then
 	perlcritic --severity 4 --verbose 8 ./mmdebstrap
 fi
 
-[ -e ./tarfilter ] && black --check ./tarfilter
+for f in tarfilter coverage.py caching_proxy.py; do
-[ -e ./coverage.py ] && black --check ./coverage.py
+	[ -e "./$f" ] || continue
+	black --check "./$f"
+done
 
 shellcheck --exclude=SC2016 coverage.sh make_mirror.sh run_null.sh run_qemu.sh gpgvnoexpkeysig hooks/*/*.sh
 

make_mirror.sh

@@ -92,76 +92,11 @@ deletecache() {
 }
 
 cleanup_newcachedir() {
+	kill "$PROXYPID" || :
 	echo "running cleanup_newcachedir"
 	deletecache "$newcachedir"
 }
 
-get_oldaptnames() {
-	if [ ! -e "$1/$2" ]; then
-		return
-	fi
-	xz -dc "$1/$2" \
-		| grep-dctrl --no-field-names --show-field=Package,Version,Architecture,Filename '' \
-		| paste -sd "    \n" \
-		| while read -r name ver arch fname; do
-			if [ ! -e "$1/$fname" ]; then
-				continue
-			fi
-			# apt stores deb files with the colon encoded as %3a while
-			# mirrors do not contain the epoch at all #645895
-			case "$ver" in *:*) ver="${ver%%:*}%3a${ver#*:}";; esac
-			aptname="$rootdir/var/cache/apt/archives/${name}_${ver}_${arch}.deb"
-			# we have to cp and not mv because other
-			# distributions might still need this file
-			# we have to cp and not symlink because apt
-			# doesn't recognize symlinks
-			cp --link "$1/$fname" "$aptname"
-			echo "$aptname"
-		done
-}
-
-get_newaptnames() {
-	if [ ! -e "$1/$2" ]; then
-		return
-	fi
-	# skip empty files by trying to uncompress the first byte of the payload
-	if [ "$(xz -dc "$1/$2" | head -c1 | wc -c)" -eq 0 ]; then
-		return
-	fi
-	xz -dc "$1/$2" \
-		| grep-dctrl --no-field-names --show-field=Package,Version,Architecture,Filename,SHA256 '' \
-		| paste -sd "     \n" \
-		| while read -r name ver arch fname hash; do
-			# sanity check for the hash because sometimes the
-			# archive switches the hash algorithm
-			if [ "${#hash}" -ne 64 ]; then
-				echo "expected hash length of 64 but got ${#hash} for: $hash" >&2
-				exit 1
-			fi
-			dir="${fname%/*}"
-			# apt stores deb files with the colon encoded as %3a while
-			# mirrors do not contain the epoch at all #645895
-			case "$ver" in *:*) ver="${ver%%:*}%3a${ver#*:}";; esac
-			aptname="$rootdir/var/cache/apt/archives/${name}_${ver}_${arch}.deb"
-			if [ -e "$aptname" ]; then
-				# make sure that we found the right file by checking its hash
-				echo "$hash  $aptname" | sha256sum --check >&2
-				mkdir -p "$1/$dir"
-				# since we move hardlinks around, the same hardlink might've been
-				# moved already into the same place by another distribution.
-				# mv(1) refuses to copy A to B if both are hardlinks of each other.
-				if [ -e "$aptname" ] && [ -e "$1/$fname" ] && [ "$(stat -c "%d %i" "$aptname")" = "$(stat -c "%d %i" "$1/$fname")" ]; then
-					# both files are already the same so we just need to
-					# delete the source
-					rm "$aptname"
-				else
-					mv "$aptname" "$1/$fname"
-				fi
-				echo "$aptname"
-			fi
-		done
-}
-
 cleanupapt() {
 	echo "running cleanupapt" >&2
 	if [ ! -e "$rootdir" ]; then
@@ -175,12 +110,11 @@ cleanupapt() {
 		"$rootdir/var/lib/dpkg/status" \
 		"$rootdir/var/lib/dpkg/lock-frontend" \
 		"$rootdir/var/lib/dpkg/lock" \
+		"$rootdir/var/lib/apt/lists/lock" \
 		"$rootdir/etc/apt/apt.conf" \
 		"$rootdir/etc/apt/sources.list.d/"* \
 		"$rootdir/etc/apt/preferences.d/"* \
 		"$rootdir/etc/apt/sources.list" \
-		"$rootdir/oldaptnames" \
-		"$rootdir/newaptnames" \
 		"$rootdir/var/cache/apt/archives/lock"; do
 		if [ ! -e "$f" ]; then
 			echo "does not exist: $f" >&2
@@ -229,9 +163,7 @@ Apt::Get::Download-Only true;
 Acquire::Languages "none";
 Dir::Etc::Trusted "/etc/apt/trusted.gpg";
 Dir::Etc::TrustedParts "/etc/apt/trusted.gpg.d";
-Acquire::http::Dl-Limit "1000";
+Acquire::http::Proxy "http://127.0.0.1:8080/";
-Acquire::https::Dl-Limit "1000";
-Acquire::Retries "5";
 END
 
 	: > "$rootdir/var/lib/dpkg/status"
@@ -256,26 +188,6 @@ END
 
 	APT_CONFIG="$rootdir/etc/apt/apt.conf" apt-get update
 
-	# before downloading packages and before replacing the old Packages
-	# file, copy all old *.deb packages from the mirror to
-	# /var/cache/apt/archives so that apt will not re-download *.deb
-	# packages that we already have
-	{
-		get_oldaptnames "$oldmirrordir" "dists/$dist/main/binary-$nativearch/Packages.xz"
-		case "$dist" in oldstable|stable)
-			get_oldaptnames "$oldmirrordir" "dists/$dist-updates/main/binary-$nativearch/Packages.xz"
-			;;
-		esac
-		case "$dist" in
-			oldstable)
-				get_oldaptnames "$oldcachedir/debian-security" "dists/$dist/updates/main/binary-$nativearch/Packages.xz"
-				;;
-			stable)
-				get_oldaptnames "$oldcachedir/debian-security" "dists/$dist-security/main/binary-$nativearch/Packages.xz"
-				;;
-		esac
-	} | sort -u > "$rootdir/oldaptnames"
-
 	pkgs=$(APT_CONFIG="$rootdir/etc/apt/apt.conf" apt-get indextargets \
 		--format '$(FILENAME)' 'Created-By: Packages' "Architecture: $nativearch" \
 		| xargs --delimiter='\n' /usr/lib/apt/apt-helper cat-file \
@@ -296,73 +208,8 @@ END
 	# shellcheck disable=SC2086
 	APT_CONFIG="$rootdir/etc/apt/apt.conf" apt-get --yes install $pkgs
 
-	# to be able to also test gpg verification, we need to create a mirror
-	mkdir -p "$newmirrordir/dists/$dist/main/binary-$nativearch/"
-	curl --location "$mirror/dists/$dist/Release" > "$newmirrordir/dists/$dist/Release"
-	curl --location "$mirror/dists/$dist/Release.gpg" > "$newmirrordir/dists/$dist/Release.gpg"
-	curl --location "$mirror/dists/$dist/main/binary-$nativearch/Packages.xz" > "$newmirrordir/dists/$dist/main/binary-$nativearch/Packages.xz"
-	codename=$(awk '/^Codename: / { print $2; }' < "$newmirrordir/dists/$dist/Release")
-	[ -L "$newmirrordir/dists/$codename" ] || ln -s "$dist" "$newmirrordir/dists/$codename"
-	case "$dist" in oldstable|stable)
-		mkdir -p "$newmirrordir/dists/$dist-updates/main/binary-$nativearch/"
-		curl --location "$mirror/dists/$dist-updates/Release" > "$newmirrordir/dists/$dist-updates/Release"
-		curl --location "$mirror/dists/$dist-updates/Release.gpg" > "$newmirrordir/dists/$dist-updates/Release.gpg"
-		curl --location "$mirror/dists/$dist-updates/main/binary-$nativearch/Packages.xz" > "$newmirrordir/dists/$dist-updates/main/binary-$nativearch/Packages.xz"
-		[ -L "$newmirrordir/dists/$codename-updates" ] || ln -s "$dist-updates" "$newmirrordir/dists/$codename-updates"
-		;;
-	esac
-	case "$dist" in
-		oldstable)
-			mkdir -p "$newcachedir/debian-security/dists/$dist/updates/main/binary-$nativearch/"
-			curl --location "$security_mirror/dists/$dist/updates/Release" > "$newcachedir/debian-security/dists/$dist/updates/Release"
-			curl --location "$security_mirror/dists/$dist/updates/Release.gpg" > "$newcachedir/debian-security/dists/$dist/updates/Release.gpg"
-			curl --location "$security_mirror/dists/$dist/updates/main/binary-$nativearch/Packages.xz" > "$newcachedir/debian-security/dists/$dist/updates/main/binary-$nativearch/Packages.xz"
-			;;
-		stable)
-			mkdir -p "$newcachedir/debian-security/dists/$dist-security/main/binary-$nativearch/"
-			curl --location "$security_mirror/dists/$dist-security/Release" > "$newcachedir/debian-security/dists/$dist-security/Release"
-			curl --location "$security_mirror/dists/$dist-security/Release.gpg" > "$newcachedir/debian-security/dists/$dist-security/Release.gpg"
-			curl --location "$security_mirror/dists/$dist-security/main/binary-$nativearch/Packages.xz" > "$newcachedir/debian-security/dists/$dist-security/main/binary-$nativearch/Packages.xz"
-			[ -L "$newcachedir/debian-security/dists/$codename-security" ] || ln -s "$dist-security" "$newcachedir/debian-security/dists/$codename-security"
-			;;
-	esac
-
-	# the deb files downloaded by apt must be moved to their right locations in the
-	# pool directory
-	#
-	# Instead of parsing the Packages file, we could also attempt to move the deb
-	# files ourselves to the appropriate pool directories. But that approach
-	# requires re-creating the heuristic by which the directory is chosen, requires
-	# stripping the epoch from the filename and will break once mirrors change.
-	# This way, it doesn't matter where the mirror ends up storing the package.
-	{
-		get_newaptnames "$newmirrordir" "dists/$dist/main/binary-$nativearch/Packages.xz";
-		case "$dist" in oldstable|stable)
-			get_newaptnames "$newmirrordir" "dists/$dist-updates/main/binary-$nativearch/Packages.xz"
-			;;
-		esac
-		case "$dist" in
-			oldstable)
-				get_newaptnames "$newcachedir/debian-security" "dists/$dist/updates/main/binary-$nativearch/Packages.xz"
-				;;
-			stable)
-				get_newaptnames "$newcachedir/debian-security" "dists/$dist-security/main/binary-$nativearch/Packages.xz"
-				;;
-		esac
-	} | sort -u > "$rootdir/newaptnames"
-
 	rm "$rootdir/var/cache/apt/archives/lock"
 	rmdir "$rootdir/var/cache/apt/archives/partial"
-	# remove all packages that were in the old Packages file but not in the
-	# new one anymore
-	comm -23 "$rootdir/oldaptnames" "$rootdir/newaptnames" | xargs --delimiter="\n" --no-run-if-empty rm
-	# now the apt cache should be empty
-	if [ -n "$(ls -1qA "$rootdir/var/cache/apt/archives/")" ]; then
-		echo "$rootdir/var/cache/apt/archives not empty:"
-		ls -la "$rootdir/var/cache/apt/archives/"
-		exit 1
-	fi
-
 	APT_CONFIG="$rootdir/etc/apt/apt.conf" apt-get --option Dir::Etc::SourceList=/dev/null update
 	APT_CONFIG="$rootdir/etc/apt/apt.conf" apt-get clean
 
@@ -425,8 +272,8 @@ components=main
 : "${CMD:=./mmdebstrap}"
 : "${USE_HOST_APT_CONFIG:=no}"
 
-if [ -e "$oldmirrordir/dists/$DEFAULT_DIST/Release" ]; then
+if [ -e "$oldmirrordir/dists/$DEFAULT_DIST/InRelease" ]; then
-	http_code=$(curl --output /dev/null --silent --location --head --time-cond "$oldmirrordir/dists/$DEFAULT_DIST/Release" --write-out '%{http_code}' "$mirror/dists/$DEFAULT_DIST/Release")
+	http_code=$(curl --output /dev/null --silent --location --head --time-cond "$oldmirrordir/dists/$DEFAULT_DIST/InRelease" --write-out '%{http_code}' "$mirror/dists/$DEFAULT_DIST/InRelease")
 	case "$http_code" in
 		200) ;; # need update
 		304) echo up-to-date; exit 0;;
@@ -434,6 +281,19 @@ if [ -e "$oldmirrordir/dists/$DEFAULT_DIST/Release" ]; then
 	esac
 fi
 
+./caching_proxy.py "$oldcachedir" "$newcachedir" &
+PROXYPID=$!
+
+for i in $(seq 10); do
+	curl --proxy "http://127.0.0.1:8080/" --silent -o /dev/null "http://deb.debian.org/debian/dists/$DEFAULT_DIST/InRelease" && break
+	sleep 1
+done
+if [ ! -s "$newmirrordir/dists/$DEFAULT_DIST/InRelease" ]; then
+	echo "failed to start proxy" >&2
+	kill $PROXYPID
+	exit 1
+fi
+
 trap "cleanup_newcachedir" EXIT INT TERM
 
 mkdir -p "$newcachedir"
@@ -447,8 +307,12 @@ elif [ "$HOSTARCH" = arm64 ]; then
 	arches="$arches amd64 armhf"
 fi
 
-for nativearch in $arches; do
+# we need the split_inline_sig() function
-	for dist in oldstable stable testing unstable; do
+# shellcheck disable=SC1091
+. /usr/share/debootstrap/functions
+
+for dist in oldstable stable testing unstable; do
+	for nativearch in $arches; do
 		# non-host architectures are only downloaded for $DEFAULT_DIST
 		if [ "$nativearch" != "$HOSTARCH" ] && [ "$DEFAULT_DIST" != "$dist" ]; then
 			continue
@@ -483,8 +347,20 @@ END
 				;;
 		esac
 	done
+	codename=$(awk '/^Codename: / { print $2; }' < "$newmirrordir/dists/$dist/InRelease")
+	ln -s "$dist" "$newmirrordir/dists/$codename"
+
+	# split the InRelease file into Release and Release.gpg not because apt
+	# or debootstrap need it that way but because grep-dctrl does
+	split_inline_sig \
+		"$newmirrordir/dists/$dist/InRelease" \
+		"$newmirrordir/dists/$dist/Release" \
+		"$newmirrordir/dists/$dist/Release.gpg"
+	touch --reference="$newmirrordir/dists/$dist/InRelease" "$newmirrordir/dists/$dist/Release" "$newmirrordir/dists/$dist/Release.gpg"
 done
 
+kill $PROXYPID
+
 # Create some symlinks so that we can trick apt into accepting multiple apt
 # lines that point to the same repository but look different. This is to
 # avoid the warning:
@@ -530,6 +406,25 @@ if [ "$HAVE_QEMU" = "yes" ]; then
 			;;
 	esac
 
+	# we use the caching proxy again when building the qemu image
+	#  - we can re-use the packages that were already downloaded earlier
+	#  - we make sure that the qemu image uses the same Release file even
+	#    if a mirror push happened between now and earlier
+	#  - we avoid polluting the mirror with the additional packages by
+	#    using --readonly
+	./caching_proxy.py --readonly "$oldcachedir" "$newcachedir" &
+	PROXYPID=$!
+
+	for i in $(seq 10); do
+		curl --proxy "http://127.0.0.1:8080/" --silent -o /dev/null "http://deb.debian.org/debian/dists/$DEFAULT_DIST/InRelease" && break
+		sleep 1
+	done
+	if [ ! -s "$newmirrordir/dists/$DEFAULT_DIST/InRelease" ]; then
+		echo "failed to start proxy" >&2
+		kill $PROXYPID
+		exit 1
+	fi
+
 	# We must not use any --dpkgopt here because any dpkg options still
 	# leak into the chroot with chrootless mode.
 	# We do not use our own package cache here because
@@ -571,11 +466,12 @@ if [ "$HAVE_QEMU" = "yes" ]; then
 		arches=$HOSTARCH
 	fi
 	$CMD --variant=apt --architectures="$arches" --include="$pkgs" \
-		--aptopt='Acquire::http::Dl-Limit "1000"' \
+		--setup-hook='echo "Acquire::http::Proxy \"http://127.0.0.1:8080/\";" > "$1/etc/apt/apt.conf.d/00proxy"' \
-		--aptopt='Acquire::https::Dl-Limit "1000"' \
+		--customize-hook='rm "$1/etc/apt/apt.conf.d/00proxy"' \
-		--aptopt='Acquire::Retries "5"' \
 		"$DEFAULT_DIST" - "$mirror" > "$tmpdir/debian-chroot.tar"
 
+	kill $PROXYPID
+
 	cat << END > "$tmpdir/mmdebstrap.service"
 [Unit]
 Description=mmdebstrap worker script

tests/as-debootstrap-unshare-wrapper

@@ -78,6 +78,7 @@ find /tmp/debian-debootstrap/run/ -mindepth 1 -maxdepth 1 ! -name lock -print0 |
 
 # debootstrap doesn't clean apt
 rm /tmp/debian-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_unstable_main_binary-{{ HOSTARCH }}_Packages \
+	/tmp/debian-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_unstable_InRelease \
 	/tmp/debian-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_unstable_Release \
 	/tmp/debian-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_unstable_Release.gpg
 

tests/check-against-debootstrap-dist

@@ -96,6 +96,7 @@ fi
 find /tmp/debian-{{ DIST }}-debootstrap/run/ -mindepth 1 -maxdepth 1 ! -name lock -print0 | xargs --no-run-if-empty -0 rm -r
 # debootstrap doesn't clean apt
 rm /tmp/debian-{{ DIST }}-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_{{ DIST }}_main_binary-{{ HOSTARCH }}_Packages \
+	/tmp/debian-{{ DIST }}-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_{{ DIST }}_InRelease \
 	/tmp/debian-{{ DIST }}-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_{{ DIST }}_Release \
 	/tmp/debian-{{ DIST }}-debootstrap/var/lib/apt/lists/127.0.0.1_debian_dists_{{ DIST }}_Release.gpg
 

tests/dist-using-codename

@@ -4,9 +4,9 @@
 
 set -eu
 export LC_ALL=C.UTF-8
-trap "rm -f Release; rm -rf /tmp/debian-chroot.tar /tmp/expected" EXIT INT TERM
+trap "rm -f InRelease; rm -rf /tmp/debian-chroot.tar /tmp/expected" EXIT INT TERM
-/usr/lib/apt/apt-helper download-file "{{ MIRROR }}/dists/{{ DIST }}/Release" Release
+/usr/lib/apt/apt-helper download-file "{{ MIRROR }}/dists/{{ DIST }}/InRelease" InRelease
-codename=$(awk '/^Codename: / { print $2; }' Release)
+codename=$(awk '/^Codename: / { print $2; }' InRelease)
 {{ CMD }} --mode={{ MODE }} --variant=apt "$codename" /tmp/debian-chroot.tar {{ MIRROR }}
 echo "deb {{ MIRROR }} $codename main" > /tmp/expected
 tar --to-stdout --extract --file /tmp/debian-chroot.tar ./etc/apt/sources.list \