feat(cpio): improve initramfs image performance and efficiency via cpio reflinks

Makefile

@@ -102,6 +102,16 @@ ifeq ($(HAVE_SHFMT),yes)
 	shfmt -w -s .
 endif
 
+src/dracut-cpio/target/release/dracut-cpio: src/dracut-cpio/src/main.rs
+	cargo --offline build --release --manifest-path src/dracut-cpio/Cargo.toml
+
+dracut-cpio: src/dracut-cpio/target/release/dracut-cpio
+	ln -fs $< $@
+
+ifeq ($(enable_dracut_cpio),yes)
+all: dracut-cpio
+endif
+
 doc: $(manpages) dracut.html
 
 ifneq ($(enable_documentation),no)
@@ -196,6 +206,9 @@ endif
 	if [ -f dracut-util ]; then \
 		install -m 0755 dracut-util $(DESTDIR)$(pkglibdir)/dracut-util; \
 	fi
+ifeq ($(enable_dracut_cpio),yes)
+	install -m 0755 dracut-cpio $(DESTDIR)$(pkglibdir)/dracut-cpio
+endif
 	mkdir -p $(DESTDIR)${prefix}/lib/kernel/install.d
 	install -m 0755 install.d/50-dracut.install $(DESTDIR)${prefix}/lib/kernel/install.d/50-dracut.install
 	install -m 0755 install.d/51-dracut-rescue.install $(DESTDIR)${prefix}/lib/kernel/install.d/51-dracut-rescue.install
@@ -222,6 +235,7 @@ clean:
 	$(RM) dracut-util util/util $(UTIL_OBJECTS)
 	$(RM) $(manpages) dracut.html
 	$(RM) dracut.pc
+	$(RM) dracut-cpio src/dracut-cpio/target/release/dracut-cpio*
 	$(MAKE) -C test clean
 
 dist: dracut-$(DRACUT_MAIN_VERSION).tar.xz

configure

@@ -6,6 +6,7 @@ echo \#buildapi-variable-no-builddir > /dev/null
 prefix=/usr
 
 enable_documentation=yes
+enable_dracut_cpio=no
 
 CC="${CC:-cc}"
 PKG_CONFIG="${PKG_CONFIG:-pkg-config}"
@@ -48,6 +49,7 @@ while (($# > 0)); do
         --infodir) read_arg infodir "$@" || shift ;;
         --systemdsystemunitdir) read_arg systemdsystemunitdir "$@" || shift ;;
         --bashcompletiondir) read_arg bashcompletiondir "$@" || shift ;;
+        --enable-dracut-cpio) enable_dracut_cpio=yes ;;
         *) echo "Ignoring unknown option '$1'" ;;
     esac
     shift
@@ -102,6 +104,15 @@ if test $found = no; then
     exit 1
 fi
 
+if test "$enable_dracut_cpio" = "yes"; then
+    cargo --version > /dev/null
+    ret=$?
+    if test $ret -ne 0; then
+        echo "dracut couldn't find cargo for dracut-cpio build"
+        exit 1
+    fi
+fi
+
 cat > Makefile.inc.$$ << EOF
 prefix ?= ${prefix}
 libdir ?= ${libdir:-${prefix}/lib}
@@ -110,6 +121,7 @@ sysconfdir ?= ${sysconfdir:-${prefix}/etc}
 sbindir ?= ${sbindir:-${prefix}/sbin}
 mandir ?= ${mandir:-${prefix}/share/man}
 enable_documentation ?= ${enable_documentation:-yes}
+enable_dracut_cpio ?= ${enable_dracut_cpio}
 bindir ?= ${bindir:-${prefix}/bin}
 KMOD_CFLAGS ?= $(${PKG_CONFIG} --cflags " libkmod >= 23 ")
 KMOD_LIBS ?= $(${PKG_CONFIG} --libs " libkmod >= 23 ")

dracut.sh

@@ -226,6 +226,7 @@ Creates initial ramdisk images for preloading modules
                          otherwise you will not be able to boot.
   --no-compress         Do not compress the generated initramfs.  This will
                          override any other compression options.
+  --enhanced-cpio       Attempt to reflink cpio file data using dracut-cpio.
   --list-modules        List all available dracut modules.
   -M, --show-modules    Print included module's name to standard output during
                          build.
@@ -412,6 +413,7 @@ rearrange_params() {
             --long zstd \
             --long no-compress \
             --long gzip \
+            --long enhanced-cpio \
             --long list-modules \
             --long show-modules \
             --long keep \
@@ -770,6 +772,7 @@ while :; do
         --zstd) compress_l="zstd" ;;
         --no-compress) _no_compress_l="cat" ;;
         --gzip) compress_l="gzip" ;;
+        --enhanced-cpio) enhanced_cpio_l="yes" ;;
         --list-modules) do_list="yes" ;;
         -M | --show-modules)
             show_modules_l="yes"
@@ -982,6 +985,7 @@ stdloglvl=$((stdloglvl + verbosity_mod_l))
 [[ $tmpdir ]] || tmpdir="$dracutsysrootdir"/var/tmp
 [[ $INITRD_COMPRESS ]] && compress=$INITRD_COMPRESS
 [[ $compress_l ]] && compress=$compress_l
+[[ $enhanced_cpio_l ]] && enhanced_cpio=$enhanced_cpio_l
 [[ $show_modules_l ]] && show_modules=$show_modules_l
 [[ $nofscks_l ]] && nofscks="yes"
 [[ $ro_mnt_l ]] && ro_mnt="yes"
@@ -1188,6 +1192,19 @@ else
     exit 1
 fi
 
+if [[ $enhanced_cpio == "yes" ]]; then
+    enhanced_cpio="$dracutbasedir/dracut-cpio"
+    if [[ -x $enhanced_cpio ]]; then
+        # align based on statfs optimal transfer size
+        cpio_align=$(stat --file-system -c "%s" -- "$initdir")
+    else
+        dinfo "--enhanced-cpio ignored due to lack of dracut-cpio"
+        unset enhanced_cpio
+    fi
+else
+    unset enhanced_cpio
+fi
+
 # shellcheck disable=SC2154
 if [[ $no_kernel != yes ]] && ! [[ -d $srcmods ]]; then
     printf "%s\n" "dracut: Cannot find module directory $srcmods" >&2
@@ -2252,6 +2269,8 @@ if dracut_module_included "squash"; then
 fi
 
 if [[ $do_strip == yes ]] && ! [[ $DRACUT_FIPS_MODE ]]; then
+    # stripping files negates (dedup) benefits of using reflink
+    [[ -n $enhanced_cpio ]] && ddebug "strip is enabled alongside cpio reflink"
     dinfo "*** Stripping files ***"
     find "$initdir" -type f \
         -executable -not -path '*/lib/modules/*.ko' -print0 \
@@ -2322,15 +2341,29 @@ if [[ $create_early_cpio == yes ]]; then
     fi
 
     # The microcode blob is _before_ the initramfs blob, not after
-    if ! (
-        umask 077
-        cd "$early_cpio_dir/d"
-        find . -print0 | sort -z \
-            | cpio ${CPIO_REPRODUCIBLE:+--reproducible} --null \
-                ${cpio_owner:+-R "$cpio_owner"} -H newc -o --quiet > "${DRACUT_TMPDIR}/initramfs.img"
-    ); then
-        dfatal "dracut: creation of $outfile failed"
-        exit 1
+    if [[ -n $enhanced_cpio ]]; then
+        if ! (
+            umask 077
+            cd "$early_cpio_dir/d"
+            find . -print0 | sort -z \
+                | $enhanced_cpio --null ${cpio_owner:+--owner "$cpio_owner"} \
+                    --mtime 0 --data-align "$cpio_align" --truncate-existing \
+                    "${DRACUT_TMPDIR}/initramfs.img"
+        ); then
+            dfatal "dracut-cpio: creation of $outfile failed"
+            exit 1
+        fi
+    else
+        if ! (
+            umask 077
+            cd "$early_cpio_dir/d"
+            find . -print0 | sort -z \
+                | cpio ${CPIO_REPRODUCIBLE:+--reproducible} --null \
+                    ${cpio_owner:+-R "$cpio_owner"} -H newc -o --quiet > "${DRACUT_TMPDIR}/initramfs.img"
+        ); then
+            dfatal "dracut: creation of $outfile failed"
+            exit 1
+        fi
     fi
 fi
 
@@ -2386,15 +2419,41 @@ if [[ $compress == $DRACUT_COMPRESS_ZSTD* ]] && ! check_kernel_config CONFIG_RD_
     compress="cat"
 fi
 
-if ! (
-    umask 077
-    cd "$initdir"
-    find . -print0 | sort -z \
-        | cpio ${CPIO_REPRODUCIBLE:+--reproducible} --null ${cpio_owner:+-R "$cpio_owner"} -H newc -o --quiet \
-        | $compress >> "${DRACUT_TMPDIR}/initramfs.img"
-); then
-    dfatal "dracut: creation of $outfile failed"
-    exit 1
+if [[ -n $enhanced_cpio ]]; then
+    if [[ $compress == "cat" ]]; then
+        # dracut-cpio appends by default, so any ucode remains
+        cpio_outfile="${DRACUT_TMPDIR}/initramfs.img"
+    else
+        ddebug "$compress compression enabled alongside cpio reflink"
+        # dracut-cpio doesn't output to stdout, so stage for compression
+        cpio_outfile="${DRACUT_TMPDIR}/initramfs.img.uncompressed"
+    fi
+
+    if ! (
+        umask 077
+        cd "$initdir"
+        find . -print0 | sort -z \
+            | $enhanced_cpio --null ${cpio_owner:+--owner "$cpio_owner"} \
+                --mtime 0 --data-align "$cpio_align" "$cpio_outfile" || exit 1
+        [[ $compress == "cat" ]] && exit 0
+        $compress < "$cpio_outfile" >> "${DRACUT_TMPDIR}/initramfs.img" \
+            && rm "$cpio_outfile"
+    ); then
+        dfatal "dracut-cpio: creation of $outfile failed"
+        exit 1
+    fi
+    unset cpio_outfile
+else
+    if ! (
+        umask 077
+        cd "$initdir"
+        find . -print0 | sort -z \
+            | cpio ${CPIO_REPRODUCIBLE:+--reproducible} --null ${cpio_owner:+-R "$cpio_owner"} -H newc -o --quiet \
+            | $compress >> "${DRACUT_TMPDIR}/initramfs.img"
+    ); then
+        dfatal "dracut: creation of $outfile failed"
+        exit 1
+    fi
 fi
 
 # shellcheck disable=SC2154

man/dracut.8.asc

@@ -530,6 +530,15 @@ will not be able to boot.
     Specifies the kernel image, which to include in the UEFI executable. The default is
     _/lib/modules/<KERNEL-VERSION>/vmlinuz_ or _/boot/vmlinuz-<KERNEL-VERSION>_
 
+**--enhanced-cpio**::
+    Attempt to use the dracut-cpio binary, which optimizes archive creation for
+    copy-on-write filesystems by using the copy_file_range(2) syscall via Rust's
+    io::copy(). When specified, initramfs archives are also padded to ensure
+    optimal data alignment for extent sharing. To retain reflink data
+    deduplication benefits, this should be used alongside the **--no-compress**
+    and **--no-strip** parameters, with initramfs source files, **--tmpdir**
+    staging area and destination all on the same copy-on-write capable filesystem.
+
 ENVIRONMENT
 -----------
 

shell-completion/bash/dracut

@@ -32,7 +32,8 @@ _dracut() {
             --local --hostonly --no-hostonly --fstab --help --bzip2 --lzma
             --xz --zstd --no-compress --gzip --list-modules --show-modules --keep
             --printsize --regenerate-all --noimageifnotneeded --early-microcode
-            --no-early-microcode --print-cmdline --reproducible --uefi'
+            --no-early-microcode --print-cmdline --reproducible --uefi
+            --enhanced-cpio'
         [ARG]='-a -m -o -d -I -k -c -L --kver --add --force-add --add-drivers
             --omit-drivers --modules --omit --drivers --filesystems --install
             --fwdir --libdirs --fscks --add-fstab --mount --device --nofscks

src/dracut-cpio/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+name = "dracut-cpio"
+description = "cpio archive generator for Dracut"
+authors = ["David Disseldorp"]
+license = "GPL-2.0"
+version = "0.1.0"
+edition = "2018"
+
+[dependencies]
+crosvm = { path = "third_party/crosvm" }
+# please avoid adding any more dependencies