contrib/dracut: 90: mount essential datasets under root

This partly mirrors what the i-t script does (though that mounts all
children, recursively) ‒ /etc, /usr, /lib*, and /bin are all essential,
if present, to successfully invoke the real init, which will then mount
everything else it might need in the right order

The following extreme-case set-up boots w/o issues now:
  /               zoot            zfs  rw,relatime,xattr,noacl
  ├─/etc          zoot/etc        zfs  rw,relatime,xattr,noacl
  ├─/usr          zoot/usr        zfs  rw,relatime,xattr,noacl
  │ └─/usr/local  zoot/usr/local  zfs  rw,relatime,xattr,noacl
  ├─/var          zoot/var        zfs  rw,relatime,xattr,noacl
  │ ├─/var/lib    zoot/var/lib    zfs  rw,relatime,xattr,noacl
  │ ├─/var/log    zoot/var/log    zfs  rw,relatime,xattr,posixacl
  │ ├─/var/cache  zoot/var/cache  zfs  rw,relatime,xattr,noacl
  │ └─/var/tmp    zoot/var/tmp    zfs  rw,relatime,xattr,noacl
  ├─/home         zoot/home       zfs  rw,relatime,xattr,noacl
  │ └─/home/nab   zoot/home/nab   zfs  rw,relatime,xattr,noacl
  ├─/boot         zoot/boot       zfs  rw,relatime,xattr,noacl
  ├─/root         zoot/home/root  zfs  rw,relatime,xattr,noacl
  ├─/opt          zoot/opt        zfs  rw,relatime,xattr,noacl
  └─/srv          zoot/srv        zfs  rw,relatime,xattr,noacl

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes #11898

contrib/dracut/90zfs/zfs-generator.sh.in

@@ -11,12 +11,13 @@ GENERATOR_DIR="$1"
 
 [ -f /lib/dracut-lib.sh ] && dracutlib=/lib/dracut-lib.sh
 [ -f /usr/lib/dracut/modules.d/99base/dracut-lib.sh ] && dracutlib=/usr/lib/dracut/modules.d/99base/dracut-lib.sh
-
 command -v getarg >/dev/null 2>&1 || {
     [ -n "$debug" ] && echo "zfs-generator: loading Dracut library from $dracutlib" >> /dev/kmsg
     . "$dracutlib"
 }
 
+. /lib/dracut-zfs-lib.sh
+
 [ -z "$root" ]       && root=$(getarg root=)
 [ -z "$rootfstype" ] && rootfstype=$(getarg rootfstype=)
 [ -z "$rootflags" ]  && rootflags=$(getarg rootflags=)
@@ -28,40 +29,84 @@ command -v getarg >/dev/null 2>&1 || {
 	[ "$rootfstype" != "zfs" ] &&
 	exit 0
 
-rootfstype=zfs
 case ",${rootflags}," in
 	*,zfsutil,*) ;;
 	,,)	rootflags=zfsutil ;;
 	*)	rootflags="zfsutil,${rootflags}" ;;
 esac
 
+if [ "${root}" != "zfs:AUTO" ]; then
+  root="${root##zfs:}"
+  root="${root##ZFS=}"
+fi
+
 [ -n "$debug" ] && echo "zfs-generator: writing extension for sysroot.mount to $GENERATOR_DIR/sysroot.mount.d/zfs-enhancement.conf" >> /dev/kmsg
 
-[ -d "$GENERATOR_DIR" ] || mkdir "$GENERATOR_DIR"
-[ -d "$GENERATOR_DIR"/sysroot.mount.d ] || mkdir "$GENERATOR_DIR"/sysroot.mount.d
 
+mkdir -p "$GENERATOR_DIR"/sysroot.mount.d "$GENERATOR_DIR"/initrd-root-fs.target.requires "$GENERATOR_DIR"/dracut-pre-mount.service.d
 {
     echo "[Unit]"
     echo "Before=initrd-root-fs.target"
     echo "After=zfs-import.target"
+    echo
     echo "[Mount]"
-    if [ "${root}" = "zfs:AUTO" ] ; then
+    if [ "${root}" = "zfs:AUTO" ]; then
       echo "PassEnvironment=BOOTFS"
       echo 'What=${BOOTFS}'
     else
-      root="${root##zfs:}"
-      root="${root##ZFS=}"
       echo "What=${root}"
     fi
-    echo "Type=${rootfstype}"
+    echo "Type=zfs"
     echo "Options=${rootflags}"
 } > "$GENERATOR_DIR"/sysroot.mount.d/zfs-enhancement.conf
-
+ln -fs ../sysroot.mount "$GENERATOR_DIR"/initrd-root-fs.target.requires/sysroot.mount
-[ -d "$GENERATOR_DIR"/initrd-root-fs.target.requires ] || mkdir -p "$GENERATOR_DIR"/initrd-root-fs.target.requires
-ln -s ../sysroot.mount "$GENERATOR_DIR"/initrd-root-fs.target.requires/sysroot.mount
 
 
-[ -d "$GENERATOR_DIR"/dracut-pre-mount.service.d ] || mkdir "$GENERATOR_DIR"/dracut-pre-mount.service.d
+if [ "${root}" = "zfs:AUTO" ]; then
+  {
+      echo "[Unit]"
+      echo "Before=initrd-root-fs.target"
+      echo "After=sysroot.mount"
+      echo "DefaultDependencies=no"
+      echo
+      echo "[Service]"
+      echo "Type=oneshot"
+      echo "PassEnvironment=BOOTFS"
+      echo "ExecStart=/bin/sh -c '" '                                        \
+        . /lib/dracut-zfs-lib.sh;                                            \
+        _zfs_nonroot_necessities_cb() {                                      \
+            zfs mount | grep -m1 -q "^$1 " && return 0;                      \
+            echo "Mounting $1 on /sysroot$2";                                \
+            mount -o zfsutil -t zfs "$1" "/sysroot$2";                       \
+        };                                                                   \
+        for_relevant_root_children "${BOOTFS}" _zfs_nonroot_necessities_cb;' \
+      "'"
+  } > "$GENERATOR_DIR"/zfs-nonroot-necessities.service
+  ln -fs ../zfs-nonroot-necessities.service "$GENERATOR_DIR"/initrd-root-fs.target.requires/zfs-nonroot-necessities.service
+else
+  # We can solve this statically at generation time, so do!
+  _zfs_generator_cb() {
+      dset="${1}"
+      mpnt="${2}"
+      unit="sysroot$(echo "$mpnt" | sed 's;/;-;g').mount"
+
+      {
+          echo "[Unit]"
+          echo "Before=initrd-root-fs.target"
+          echo "After=sysroot.mount"
+          echo
+          echo "[Mount]"
+          echo "Where=/sysroot${mpnt}"
+          echo "What=${dset}"
+          echo "Type=zfs"
+          echo "Options=zfsutil"
+      } > "$GENERATOR_DIR/${unit}"
+      ln -fs ../"${unit}" "$GENERATOR_DIR"/initrd-root-fs.target.requires/"${unit}"
+  }
+
+  for_relevant_root_children "${root}" _zfs_generator_cb
+fi
+
 
 {
     echo "[Unit]"

contrib/dracut/90zfs/zfs-lib.sh.in

@@ -23,6 +23,7 @@ command -v getargbool >/dev/null || {
 OLDIFS="${IFS}"
 NEWLINE="
 "
+TAB="	"
 
 ZPOOL_IMPORT_OPTS=""
 if getargbool 0 zfs_force -y zfs.force -y zfsforce ; then
@@ -71,20 +72,56 @@ import_pool() {
     return 0
 }
 
+_mount_dataset_cb() {
+    mount -o zfsutil -t zfs "${1}" "${NEWROOT}${2}"
+}
+
 # mount_dataset DATASET
 #   mounts the given zfs dataset.
 mount_dataset() {
     dataset="${1}"
     mountpoint="$(zfs get -H -o value mountpoint "${dataset}")"
+    ret=0
 
     # We need zfsutil for non-legacy mounts and not for legacy mounts.
     if [ "${mountpoint}" = "legacy" ] ; then
-        mount -t zfs "${dataset}" "${NEWROOT}"
+        mount -t zfs "${dataset}" "${NEWROOT}" || ret=$?
     else
-        mount -o zfsutil -t zfs "${dataset}" "${NEWROOT}"
+        mount -o zfsutil -t zfs "${dataset}" "${NEWROOT}" || ret=$?
+
+        if [ "$ret" = "0" ]; then
+            for_relevant_root_children "${dataset}" _mount_dataset_cb || ret=$?
+        fi
     fi
 
-    return $?
+    return ${ret}
+}
+
+# for_relevant_root_children DATASET EXEC
+#   Runs "EXEC dataset mountpoint" for all children of DATASET that are needed for system bringup
+#   Used by zfs-generator.sh and friends, too!
+for_relevant_root_children() {
+    dataset="${1}"
+    exec="${2}"
+
+    zfs list -t filesystem -Ho name,mountpoint,canmount -r "${dataset}" |
+        (
+            _ret=0
+            while IFS="${TAB}" read -r dataset mountpoint canmount; do
+                [ "$canmount" != "on" ] && continue
+
+                case "$mountpoint" in
+                    /etc|/bin|/lib|/lib??|/libx32|/usr)
+                        # If these aren't mounted we may not be able to get to the real init at all, or pollute the dataset holding the rootfs
+                        "${exec}" "${dataset}" "${mountpoint}" || _ret=$?
+                        ;;
+                    *)
+                        # Up to the real init to remount everything else it might need
+                        ;;
+                esac
+            done
+            exit ${_ret}
+        )
 }
 
 # export_all OPTS