cryptsetup-var: encrypt a grown /var in place, report conversion size/progress

meta-avocado-nxp/stone/stone-imx95-frdm.json

@@ -132,8 +132,6 @@
         {
           "name": "var",
           "image": "var",
-          "size": 512,
-          "size_unit": "mebibytes",
           "expand": "true"
         }
       ]

meta-avocado/recipes-core/cryptsetup-var/files/cryptsetup-var.sh

@@ -217,13 +217,45 @@ maybe_resize() {
 # Interrupted mid-way (power cut), LUKS2 records the reencryption in its
 # metadata; open_var resumes it before opening. A filesystem whose size cannot
 # be read gets the whole-partition reencryption instead - slower, still correct.
+#
+# A /var that has already been GROWN to fill the partition (a deployed device
+# whose OTA turns encryption on) has no free tail for the header. btrfs can
+# shrink online, so give the header its 32 MiB back first: mount, resize by
+# the deficit, unmount, then reencrypt as usual. Anything else about the
+# migration is the same; it just starts from a filesystem that no longer ends
+# at the partition's last sector.
+btrfs_total_bytes() {
+    btrfs inspect-internal dump-super "$VAR_DEV" 2>/dev/null | awk '/^total_bytes/{print $2}'
+}
+
+shrink_btrfs_for_header() {
+    shrink_mib="$1"
+    mnt=/run/cryptsetup-var-shrink
+    echo "cryptsetup-var: btrfs on $VAR_DEV fills the partition - shrinking it by ${shrink_mib} MiB to make room for the LUKS2 header"
+    mkdir -p "$mnt"
+    if ! mount -t btrfs "$VAR_DEV" "$mnt"; then
+        echo "cryptsetup-var: cannot mount $VAR_DEV to shrink it - cannot encrypt in place" >&2
+        exit 1
+    fi
+    if ! btrfs filesystem resize "-${shrink_mib}M" "$mnt"; then
+        umount "$mnt"
+        echo "cryptsetup-var: btrfs refused to shrink by ${shrink_mib} MiB (tail in use or too full) - cannot encrypt in place" >&2
+        exit 1
+    fi
+    umount "$mnt"
+}
+
 encrypt_in_place() {
     fstype="$1"
     part_bytes=$(blockdev --getsize64 "$VAR_DEV")
     fs_bytes=""
     if [ "$fstype" = "btrfs" ] && command -v btrfs >/dev/null 2>&1; then
-        fs_bytes=$(btrfs inspect-internal dump-super "$VAR_DEV" 2>/dev/null \
-                    | awk '/^total_bytes/{print $2}')
+        fs_bytes=$(btrfs_total_bytes)
+        if [ -n "$fs_bytes" ] && [ "$fs_bytes" -gt 0 ] 2>/dev/null \
+           && [ $(( part_bytes - fs_bytes )) -lt 33554432 ]; then
+            shrink_btrfs_for_header $(( (33554432 - (part_bytes - fs_bytes) + 1048575) / 1048576 ))
+            fs_bytes=$(btrfs_total_bytes)
+        fi
     fi
     size_opt=""
     if [ -n "$fs_bytes" ] && [ "$fs_bytes" -gt 0 ] 2>/dev/null; then
@@ -237,15 +269,26 @@ encrypt_in_place() {
         echo "cryptsetup-var: $fstype on $VAR_DEV leaves no 32 MiB for a LUKS header - cannot encrypt in place" >&2
         exit 1
     fi
+    # Tell whoever is watching the console what is about to happen: a seeded
+    # /var can be tens of GiB, so this step can run for many minutes with
+    # nothing else on screen. It is a one-time migration and resumable after a
+    # power cut (resume_reencrypt). Periodic progress needs cryptsetup >= 2.4.
+    if [ -n "$size_opt" ]; then work_mib=$want_mib; else work_mib=$(( part_bytes / 1048576 )); fi
+    progress_opt=""
+    if cryptsetup --help 2>&1 | grep -q -- --progress-frequency; then
+        progress_opt="--progress-frequency 30"
+    fi
     echo "cryptsetup-var: first boot - encrypting existing $fstype on $VAR_DEV in place${size_opt:+ ($size_opt)}"
-    # shellcheck disable=SC2086 # size_opt is two words on purpose
+    echo "cryptsetup-var: ${work_mib} MiB to convert - one-time migration, resumes after a power cut${progress_opt:+, progress every 30 s}"
+    # shellcheck disable=SC2086 # size_opt / progress_opt are two words on purpose
     cryptsetup reencrypt --encrypt \
         --type luks2 \
         --cipher aes-xts-plain64 \
         --key-size 512 \
         --hash sha256 \
         --reduce-device-size 32M \
         $size_opt \
+        $progress_opt \
         --key-file "$KEY_FILE" \
         --batch-mode \
         "$VAR_DEV"

meta-avocado/recipes-core/cryptsetup-var/tests/test-cryptsetup-var-inplace.sh

@@ -27,6 +27,7 @@ cat > "$work/bin/cryptsetup" <<'S'
 #!/bin/sh
 echo "cryptsetup $*" >> "$LOG"
 case "$1" in
+  --help)     echo "  --progress-frequency=secs" ;;
   isLuks)     [ -e "$STATE/luks" ] ;;
   luksDump)   cat "$STATE/dump" 2>/dev/null; exit 0 ;;
   luksFormat) touch "$STATE/luks" ;;
@@ -46,8 +47,13 @@ S
 cat > "$work/bin/btrfs" <<'S'
 #!/bin/sh
 echo "btrfs $*" >> "$LOG"
-case "$1" in inspect-internal) printf 'total_bytes\t\t%s\n' "$(cat "$STATE/fsbytes")" ;; esac
+case "$1" in
+  inspect-internal) printf 'total_bytes\t\t%s\n' "$(cat "$STATE/fsbytes")" ;;
+  filesystem) [ "$2" = resize ] && { n="${3#-}"; n="${n%M}"; echo $(( $(cat "$STATE/fsbytes") - n * 1048576 )) > "$STATE/fsbytes"; } ;;
+esac
 S
+printf '#!/bin/sh\necho "mount $*" >> "$LOG"\n' > "$work/bin/mount"
+printf '#!/bin/sh\necho "umount $*" >> "$LOG"\n' > "$work/bin/umount"
 printf '#!/bin/sh\necho "0 4194304 crypt aes-xts-plain64 :64:logon:x 0 %s 32768 1 allow_discards"\n' "8:0" > "$work/bin/dmsetup"
 printf '#!/bin/sh\nexit 0\n' > "$work/bin/modprobe"
 printf '#!/bin/sh\nexit 0\n' > "$work/bin/systemd-cryptenroll"
@@ -66,8 +72,8 @@ run() { # run <state-dir>
 # --- Case 1: flashed 128 MiB btrfs -> in-place reencrypt confined to fs+32M ---
 s="$work/c1"; mkdir -p "$s"; echo btrfs > "$s/fstype"; echo 134217728 > "$s/fsbytes"
 run "$s" || bad "case 1 script exit"
-if grep -q "^cryptsetup reencrypt --encrypt --type luks2 --cipher aes-xts-plain64 --key-size 512 --hash sha256 --reduce-device-size 32M --device-size 160M --key-file .* --batch-mode $blk\$" "$s/log"; then
-  ok "plaintext btrfs is re-encrypted in place, confined to fs + 32 MiB (160M)"
+if grep -q "^cryptsetup reencrypt --encrypt --type luks2 --cipher aes-xts-plain64 --key-size 512 --hash sha256 --reduce-device-size 32M --device-size 160M --progress-frequency 30 --key-file .* --batch-mode $blk\$" "$s/log"; then
+  ok "plaintext btrfs is re-encrypted in place, confined to fs + 32 MiB (160M), with progress"
 else bad "unexpected reencrypt command: $(grep reencrypt "$s/log" || echo none)"; fi
 grep -q "^cryptsetup luksFormat" "$s/log" && bad "luksFormat ran over a flashed filesystem" || ok "luksFormat never touches a flashed filesystem"
 grep -q "^cryptsetup luksOpen --key-file" "$s/log" && ok "container opened with the recovery key afterwards" || bad "container not opened"
@@ -85,4 +91,22 @@ if grep -q "^cryptsetup reencrypt --resume-only" "$s/log" && [ "$(grep -n 'reenc
   ok "interrupted reencryption is resumed before the container is opened"
 else bad "resume not attempted first: $(grep -n 'reencrypt\|luksOpen' "$s/log")"; fi
 
+# --- Case 4: btrfs grown to fill the 2 GiB partition (deployed device, OTA turns
+# encryption on) -> shrink it by the 32 MiB header deficit first, then reencrypt ---
+s="$work/c4"; mkdir -p "$s"; echo btrfs > "$s/fstype"; echo 2147483648 > "$s/fsbytes"
+run "$s" || bad "case 4 script exit"
+if grep -q "^btrfs filesystem resize -32M /run/cryptsetup-var-shrink$" "$s/log"; then
+  ok "grown btrfs is shrunk by exactly the 32 MiB header deficit"
+else bad "no/unexpected shrink: $(grep 'btrfs filesystem' "$s/log" || echo none)"; fi
+seq_ok=1
+m=$(grep -n '^mount -t btrfs' "$s/log" | head -1 | cut -d: -f1); r=$(grep -n '^btrfs filesystem resize' "$s/log" | head -1 | cut -d: -f1)
+u=$(grep -n '^umount' "$s/log" | head -1 | cut -d: -f1); e=$(grep -n '^cryptsetup reencrypt --encrypt' "$s/log" | head -1 | cut -d: -f1)
+{ [ -n "$m" ] && [ -n "$r" ] && [ -n "$u" ] && [ -n "$e" ] && [ "$m" -lt "$r" ] && [ "$r" -lt "$u" ] && [ "$u" -lt "$e" ]; } || seq_ok=0
+[ "$seq_ok" = 1 ] && ok "shrink is mount -> resize -> umount, all before the reencrypt" || bad "wrong shrink ordering: $(grep -n 'mount\|resize\|reencrypt --encrypt' "$s/log")"
+grep -q "^cryptsetup reencrypt --encrypt .*--device-size 2048M " "$s/log" && ok "reencrypt then covers the shrunk fs + header (2048M = whole partition)" || bad "unexpected device-size after shrink: $(grep 'reencrypt --encrypt' "$s/log")"
+grep -q "MiB to convert" "$s/out" && ok "operator is told how much is being converted" || bad "no size/progress line on the console"
+
+# --- Case 5: flashed small btrfs (case 1 shape) must NOT be shrunk ---
+grep -q "^btrfs filesystem resize" "$work/c1/log" && bad "small flashed btrfs was shrunk needlessly" || ok "a filesystem with free tail is left alone"
+
 echo; echo "passed: $pass  failed: $fail"; [ "$fail" -eq 0 ]