feat(dracut.sh): add "--cpio-reflink" option for requesting GNU cpio reflinks

The new GNU cpio "--reflink" parameter sees it use copy_file_range()
when copying between source and destination, allowing for copy-on-write
optimization if supported by the underlying filesystem (e.g. Btrfs or
XFS).

When calling cpio with --reflink the file list is piped through padcpio
to ensure optimal alignment for extent sharing. Microcode and initramfs
proper archives are chained together using the new cpio "--chain"
parameter, which ensures that subsequent archives are appended in a
reflink friendly fashion.

RFC: This shouldn't be merged until GNU cpio --reflink --chain patches
     have made it upstream.

Signed-off-by: David Disseldorp <ddiss@suse.de>

Author: ddiss

dracut-bash-completion.sh

@@ -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
+            --cpio-reflink'
         [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

dracut.8.asc

@@ -531,6 +531,15 @@ 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>_
 
+**--cpio-reflink**::
+    Attempt to use the GNU cpio --reflink option, which optimizes archive creation for
+    copy-on-write filesystems by using the copy_file_range(2) syscall. 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
 -----------
 

dracut.sh

@@ -228,6 +228,7 @@ 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.
+  --cpio-reflink        Request that cpio perform reflinks for file data.
   --list-modules        List all available dracut modules.
   -M, --show-modules    Print included module's name to standard output during
                          build.
@@ -414,6 +415,7 @@ rearrange_params() {
             --long zstd \
             --long no-compress \
             --long gzip \
+            --long cpio-reflink \
             --long list-modules \
             --long show-modules \
             --long keep \
@@ -772,6 +774,7 @@ while :; do
         --zstd) compress_l="zstd" ;;
         --no-compress) _no_compress_l="cat" ;;
         --gzip) compress_l="gzip" ;;
+        --cpio-reflink) cpio_reflink="yes" ;;
         --list-modules) do_list="yes" ;;
         -M | --show-modules)
             show_modules_l="yes"
@@ -1147,6 +1150,28 @@ trap 'exit 1;' SIGINT
 readonly initdir="${DRACUT_TMPDIR}/initramfs"
 mkdir -p "$initdir"
 
+if [[ $cpio_reflink == "yes" ]]; then
+    if [[ "$(cpio --help)" == *--reflink* ]]; then
+        # both XFS and Btrfs require data to be FS block-size aligned for proper
+        # extent sharing / reflinks. padcpio ensures this.
+        if [[ -d "$dracutbasedir/skipcpio" ]]; then
+            padcpio="$dracutbasedir/skipcpio/padcpio"
+        else
+            padcpio="$dracutbasedir/padcpio"
+        fi
+        if [[ -x "$padcpio" ]]; then
+            # align based on statfs optimal transfer size
+            padcpio_align=$(stat --file-system -c "%s" -- "$initdir")
+        else
+            dinfo "cpio-reflink ignored due to lack of padcpio"
+            unset cpio_reflink
+        fi
+    else
+        dinfo "cpio-reflink ignored due to lack of support in $(which cpio)"
+        unset cpio_reflink
+    fi
+fi
+
 # shellcheck disable=SC2154
 if [[ $early_microcode == yes ]] || { [[ $acpi_override == yes ]] && [[ -d $acpi_table_dir ]]; }; then
     readonly early_cpio_dir="${DRACUT_TMPDIR}/earlycpio"
@@ -2196,6 +2221,8 @@ if dracut_module_included "squash"; then
 fi
 
 if [[ $do_strip == yes ]] && ! [[ $DRACUT_FIPS_MODE ]]; then
+    # warn that stripping files negates (dedup) benefits of using reflink
+    [ -n "$cpio_reflink" ] && dinfo "inefficient: strip is enabled alongside cpio reflink"
     dinfo "*** Stripping files ***"
     find "$initdir" -type f \
         -executable -not -path '*/lib/modules/*.ko' -print0 \
@@ -2266,15 +2293,28 @@ 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 $outfile failed"
-        exit 1
+    if [[ -n "$cpio_reflink" ]]; then
+        if ! (
+            umask 077
+            cd "$early_cpio_dir/d"
+            find . -print0 | sort -z | "$padcpio" --min "$padcpio_align" --align "$padcpio_align" \
+                | cpio ${CPIO_REPRODUCIBLE:+--reproducible} --null \
+                    ${cpio_owner:+-R "$cpio_owner"} -H newc -o --quiet --reflink -O "${DRACUT_TMPDIR}/initramfs.img"
+        ); then
+            dfatal "dracut: creation of reflinked $outfile failed"
+            exit 1
+        fi
+    else
+        if ! (
+            umask 077
+            cd "$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
 
@@ -2325,15 +2365,33 @@ case $compress in
         ;;
 esac
 
-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 $outfile failed"
-    exit 1
+if [[ -n "$cpio_reflink" && "$compress" == "cat" ]]; then
+    # determine padding offset if we're appending to microcode
+    i=$(stat -c "%s" -- "${DRACUT_TMPDIR}/initramfs.img" 2>/dev/null)
+    if ! (
+        umask 077
+        cd "$initdir"
+        find . -print0 | sort -z \
+            | "$padcpio" --min "$padcpio_align" --align "$padcpio_align" ${i:+--offset "$i"} \
+            | cpio ${CPIO_REPRODUCIBLE:+--reproducible} --null \
+                   ${cpio_owner:+-R "$cpio_owner"} -H newc -o --quiet --reflink ${i:+--chain} \
+                   -O "${DRACUT_TMPDIR}/initramfs.img"
+    ); then
+        dfatal "dracut: creation of reflinked $outfile failed"
+        exit 1
+    fi
+else
+    [ -n "$cpio_reflink" ] && dinfo "cpio-reflink ignored due to compression"
+    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