Merge pull request #34258 from yuwata/nspawn-volatile-u

nspawn: make --volatile work with -U

src/nspawn/nspawn-mount.c

@@ -1061,19 +1061,30 @@ bool has_custom_root_mount(const CustomMount *mounts, size_t n) {
         return false;
 }
 
-static int setup_volatile_state(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
-        _cleanup_free_ char *buf = NULL;
-        const char *p, *options;
+static int setup_volatile_state(const char *directory) {
         int r;
 
         assert(directory);
 
         /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
 
+        /* First, remount the root directory. */
         r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
         if (r < 0)
                 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
 
+        return 0;
+}
+
+static int setup_volatile_state_after_remount_idmap(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
+        _cleanup_free_ char *buf = NULL;
+        const char *p, *options;
+        int r;
+
+        assert(directory);
+
+        /* Then, after remount_idmap(), overmount /var/ with a tmpfs. */
+
         p = prefix_roota(directory, "/var");
         r = mkdir(p, 0755);
         if (r < 0 && errno != EEXIST)
@@ -1249,7 +1260,7 @@ int setup_volatile_mode(
                 return setup_volatile_yes(directory, uid_shift, selinux_apifs_context);
 
         case VOLATILE_STATE:
-                return setup_volatile_state(directory, uid_shift, selinux_apifs_context);
+                return setup_volatile_state(directory);
 
         case VOLATILE_OVERLAY:
                 return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context);
@@ -1259,6 +1270,22 @@ int setup_volatile_mode(
         }
 }
 
+int setup_volatile_mode_after_remount_idmap(
+                const char *directory,
+                VolatileMode mode,
+                uid_t uid_shift,
+                const char *selinux_apifs_context) {
+
+        switch (mode) {
+
+        case VOLATILE_STATE:
+                return setup_volatile_state_after_remount_idmap(directory, uid_shift, selinux_apifs_context);
+
+        default:
+                return 0;
+        }
+}
+
 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
         _cleanup_free_ char *root_new = NULL, *root_old = NULL;

src/nspawn/nspawn-mount.h

@@ -63,6 +63,11 @@ int mount_custom(const char *dest, CustomMount *mounts, size_t n, uid_t uid_shif
 bool has_custom_root_mount(const CustomMount *mounts, size_t n);
 
 int setup_volatile_mode(const char *directory, VolatileMode mode, uid_t uid_shift, const char *selinux_apifs_context);
+int setup_volatile_mode_after_remount_idmap(
+                const char *directory,
+                VolatileMode mode,
+                uid_t uid_shift,
+                const char *selinux_apifs_context);
 
 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s);
 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old);

src/nspawn/nspawn.c

@@ -4083,21 +4083,24 @@ static int outer_child(
         if (arg_userns_mode != USER_NAMESPACE_NO &&
             IN_SET(arg_userns_ownership, USER_NAMESPACE_OWNERSHIP_MAP, USER_NAMESPACE_OWNERSHIP_AUTO) &&
             arg_uid_shift != 0) {
-                _cleanup_free_ char *usr_subtree = NULL;
-                char *dirs[3];
-                size_t i = 0;
+                _cleanup_strv_free_ char **dirs = NULL;
 
-                dirs[i++] = (char*) directory;
-
-                if (dissected_image && dissected_image->partitions[PARTITION_USR].found) {
-                        usr_subtree = path_join(directory, "/usr");
-                        if (!usr_subtree)
+                if (arg_volatile_mode != VOLATILE_YES) {
+                        r = strv_extend(&dirs, directory);
+                        if (r < 0)
                                 return log_oom();
-
-                        dirs[i++] = usr_subtree;
                 }
 
-                dirs[i] = NULL;
+                if ((dissected_image && dissected_image->partitions[PARTITION_USR].found) ||
+                    arg_volatile_mode == VOLATILE_YES) {
+                        char *s = path_join(directory, "/usr");
+                        if (!s)
+                                return log_oom();
+
+                        r = strv_consume(&dirs, s);
+                        if (r < 0)
+                                return log_oom();
+                }
 
                 r = remount_idmap(dirs, arg_uid_shift, arg_uid_range, UID_INVALID, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
                 if (r == -EINVAL || ERRNO_IS_NEG_NOT_SUPPORTED(r)) {
@@ -4118,6 +4121,14 @@ static int outer_child(
                 }
         }
 
+        r = setup_volatile_mode_after_remount_idmap(
+                        directory,
+                        arg_volatile_mode,
+                        arg_uid_shift,
+                        arg_selinux_apifs_context);
+        if (r < 0)
+                return r;
+
         if (dissected_image) {
                 /* Now we know the uid shift, let's now mount everything else that might be in the image. */
                 r = dissected_image_mount_and_warn(

test/units/TEST-13-NSPAWN.nspawn.sh

@@ -144,6 +144,37 @@ testcase_sanity() {
     test ! -e "$root/var/also-nope"
     test ! -e "$root/usr/nope-too"
 
+    # --volatile= with -U
+    touch "$root/usr/has-usr"
+    # volatile(=yes): rootfs is tmpfs, /usr/ from the OS tree is mounted read only
+    systemd-nspawn --directory="$root"\
+                   --volatile \
+                   -U \
+                   bash -xec 'test -e /usr/has-usr; touch /usr/read-only && exit 1; touch /nope'
+    test ! -e "$root/nope"
+    test ! -e "$root/usr/read-only"
+    systemd-nspawn --directory="$root"\
+                   --volatile=yes \
+                   -U \
+                   bash -xec 'test -e /usr/has-usr; touch /usr/read-only && exit 1; touch /nope'
+    test ! -e "$root/nope"
+    test ! -e "$root/usr/read-only"
+    # volatile=state: rootfs is read-only, /var/ is tmpfs
+    systemd-nspawn --directory="$root" \
+                   --volatile=state \
+                   -U \
+                   bash -xec 'test -e /usr/has-usr; mountpoint /var; touch /read-only && exit 1; touch /var/nope'
+    test ! -e "$root/read-only"
+    test ! -e "$root/var/nope"
+    # volatile=overlay: tmpfs overlay is mounted over rootfs
+    systemd-nspawn --directory="$root" \
+                   --volatile=overlay \
+                   -U \
+                   bash -xec 'test -e /usr/has-usr; touch /nope; touch /var/also-nope; touch /usr/nope-too'
+    test ! -e "$root/nope"
+    test ! -e "$root/var/also-nope"
+    test ! -e "$root/usr/nope-too"
+
     # --machine=, --hostname=
     systemd-nspawn --directory="$root" \
                    --machine="foo-bar.baz" \