conf: support idmapping directories

src/lxc/conf.c

@@ -488,11 +488,18 @@ int run_script(const char *name, const char *section, const char *script, ...)
  */
 int lxc_rootfs_prepare(struct lxc_rootfs *rootfs, bool userns)
 {
-	__do_close int dfd_path = -EBADF, fd_pin = -EBADF;
+	__do_close int dfd_path = -EBADF, fd_pin = -EBADF, fd_userns = -EBADF;
 	int ret;
 	struct stat st;
 	struct statfs stfs;
 
+	if (!is_empty_string(rootfs->mnt_opts.userns_path)) {
+		fd_userns = open_at(-EBADF, rootfs->mnt_opts.userns_path,
+				    PROTECT_OPEN_WITH_TRAILING_SYMLINKS, 0, 0);
+		if (fd_userns < 0)
+			return syserror("Failed to open user namespace");
+	}
+
 	if (rootfs->path) {
 		if (rootfs->bdev_type &&
 		    (strequal(rootfs->bdev_type, "overlay") ||
@@ -504,13 +511,17 @@ int lxc_rootfs_prepare(struct lxc_rootfs *rootfs, bool userns)
 		dfd_path = open_at(-EBADF, "/", PROTECT_OPATH_FILE, PROTECT_LOOKUP_ABSOLUTE, 0);
 	}
 	if (dfd_path < 0)
-		return log_error_errno(-errno, errno, "Failed to open \"%s\"", rootfs->path);
+		return syserror("Failed to open \"%s\"", rootfs->path);
 
-	if (!rootfs->path)
-		return log_trace(0, "Not pinning because container does not have a rootfs");
+	if (!rootfs->path) {
+		TRACE("Not pinning because container does not have a rootfs");
+		goto out;
+	}
 
-	if (userns)
-		return log_trace(0, "Not pinning because container runs in user namespace");
+	if (userns) {
+		TRACE("Not pinning because container runs in user namespace");
+		goto out;
+	}
 
 	ret = fstat(dfd_path, &st);
 	if (ret < 0)
@@ -524,7 +535,7 @@ int lxc_rootfs_prepare(struct lxc_rootfs *rootfs, bool userns)
 			 PROTECT_LOOKUP_BENEATH,
 			 S_IWUSR | S_IRUSR);
 	if (fd_pin < 0)
-		return log_error_errno(-errno, errno, "Failed to pin rootfs");
+		return syserror("Failed to pin rootfs");
 
 	TRACE("Pinned rootfs %d(.lxc_keep)", fd_pin);
 
@@ -546,6 +557,7 @@ int lxc_rootfs_prepare(struct lxc_rootfs *rootfs, bool userns)
 
 out:
 	rootfs->fd_path_pin = move_fd(fd_pin);
+	rootfs->mnt_opts.userns_fd = move_fd(fd_userns);
 	return 0;
 }
 
@@ -2105,6 +2117,7 @@ const char *lxc_mount_options_info[LXC_MOUNT_MAX] = {
 /* Remove "optional", "create=dir", and "create=file" from mntopt */
 int parse_lxc_mntopts(struct lxc_mount_options *opts, char *mnt_opts)
 {
+	__do_close int fd_userns = -EBADF;
 
 	for (size_t i = LXC_MOUNT_CREATE_DIR; i < LXC_MOUNT_MAX; i++) {
 		const char *opt_name = lxc_mount_options_info[i];
@@ -2140,7 +2153,12 @@ int parse_lxc_mntopts(struct lxc_mount_options *opts, char *mnt_opts)
 			if (is_empty_string(opts->userns_path))
 				return syserror_set(-EINVAL, "Missing idmap path for \"idmap=<path>\" LXC specific mount option");
 
-			TRACE("Parse LXC specific mount option \"idmap=%s\"", opts->userns_path);
+			close_prot_errno_disarm(fd_userns);
+			fd_userns = open(opts->userns_path, O_RDONLY | O_NOCTTY | O_CLOEXEC);
+			if (fd_userns < 0)
+				return syserror("Failed to open user namespace");
+
+			TRACE("Parse LXC specific mount option %d->\"idmap=%s\"", fd_userns, opts->userns_path);
 			break;
 		default:
 			return syserror_set(-EINVAL, "Unknown LXC specific mount option");
@@ -2726,6 +2744,7 @@ struct lxc_conf *lxc_conf_init(void)
 	new->rootfs.dfd_dev = -EBADF;
 	new->rootfs.dfd_host = -EBADF;
 	new->rootfs.fd_path_pin = -EBADF;
+	new->rootfs.mnt_opts.userns_fd = -EBADF;
 	new->logfd = -1;
 	lxc_list_init(&new->cgroup);
 	lxc_list_init(&new->cgroup2);

src/lxc/conf.h

@@ -198,6 +198,7 @@ struct lxc_mount_options {
 	int optional : 1;
 	int relative : 1;
 	char userns_path[PATH_MAX];
+	int userns_fd;
 };
 
 /* Defines a structure to store the rootfs location, the
@@ -575,12 +576,18 @@ static inline const char *get_rootfs_mnt(const struct lxc_rootfs *rootfs)
 	return !is_empty_string(rootfs->path) ? rootfs->mount : s;
 }
 
+static inline bool idmapped_rootfs_mnt(const struct lxc_rootfs *rootfs)
+{
+	return rootfs->mnt_opts.userns_fd >= 0;
+}
+
 static inline void put_lxc_rootfs(struct lxc_rootfs *rootfs, bool unpin)
 {
 	if (rootfs) {
 		close_prot_errno_disarm(rootfs->dfd_host);
 		close_prot_errno_disarm(rootfs->dfd_mnt);
 		close_prot_errno_disarm(rootfs->dfd_dev);
+		close_prot_errno_disarm(rootfs->mnt_opts.userns_fd);
 		if (unpin)
 			close_prot_errno_disarm(rootfs->fd_path_pin);
 	}

src/lxc/mount_utils.c

@@ -236,12 +236,15 @@ int fs_attach(int fd_fs,
 	return 0;
 }
 
-int fd_bind_mount(int dfd_from, const char *path_from,
-		  __u64 o_flags_from, __u64 resolve_flags_from,
-		  int dfd_to, const char *path_to,
-		  __u64 o_flags_to, __u64 resolve_flags_to,
-		  unsigned int attr_flags, bool recursive)
+static int __fd_bind_mount(int dfd_from, const char *path_from,
+			   __u64 o_flags_from, __u64 resolve_flags_from,
+			   int dfd_to, const char *path_to, __u64 o_flags_to,
+			   __u64 resolve_flags_to, unsigned int attr_flags,
+			   int userns_fd, bool recursive)
 {
+	struct lxc_mount_attr attr = {
+		.attr_set = attr_flags,
+	};
 	__do_close int __fd_from = -EBADF, __fd_to = -EBADF;
 	__do_close int fd_tree_from = -EBADF;
 	unsigned int open_tree_flags = AT_EMPTY_PATH | OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC;
@@ -266,7 +269,20 @@ int fd_bind_mount(int dfd_from, const char *path_from,
 
 	fd_tree_from = open_tree(fd_from, "", open_tree_flags);
 	if (fd_tree_from < 0)
-		return log_error_errno(-errno, errno, "Failed to create detached mount");
+		return syserror("Failed to create detached mount");
+
+	if (userns_fd >= 0) {
+		attr.attr_set	|= MOUNT_ATTR_IDMAP;
+		attr.userns_fd	= userns_fd;
+	}
+
+	if (attr.attr_set) {
+		ret = mount_setattr(fd_tree_from, "",
+				    AT_EMPTY_PATH | AT_RECURSIVE,
+				    &attr, sizeof(attr));
+		if (ret < 0)
+			return syserror("Failed to change mount attributes");
+	}
 
 	if (!is_empty_string(path_to)) {
 		struct lxc_open_how how = {
@@ -284,12 +300,34 @@ int fd_bind_mount(int dfd_from, const char *path_from,
 
 	ret = move_mount(fd_tree_from, "", fd_to, "", MOVE_MOUNT_F_EMPTY_PATH | MOVE_MOUNT_T_EMPTY_PATH);
 	if (ret)
-		return log_error_errno(-errno, errno, "Failed to attach detached mount %d to filesystem at %d", fd_tree_from, fd_to);
+		return syserror("Failed to attach detached mount %d to filesystem at %d", fd_tree_from, fd_to);
 
 	TRACE("Attach detached mount %d to filesystem at %d", fd_tree_from, fd_to);
 	return 0;
 }
 
+int fd_mount_idmapped(int dfd_from, const char *path_from,
+		      __u64 o_flags_from, __u64 resolve_flags_from,
+		      int dfd_to, const char *path_to,
+		      __u64 o_flags_to, __u64 resolve_flags_to,
+		      unsigned int attr_flags, int userns_fd, bool recursive)
+{
+	return __fd_bind_mount(dfd_from, path_from, o_flags_from, resolve_flags_from,
+			       dfd_to, path_to, o_flags_to, resolve_flags_to,
+			       attr_flags, userns_fd, recursive);
+}
+
+int fd_bind_mount(int dfd_from, const char *path_from,
+		  __u64 o_flags_from, __u64 resolve_flags_from,
+		  int dfd_to, const char *path_to,
+		  __u64 o_flags_to, __u64 resolve_flags_to,
+		  unsigned int attr_flags, bool recursive)
+{
+	return __fd_bind_mount(dfd_from, path_from, o_flags_from, resolve_flags_from,
+			       dfd_to, path_to, o_flags_to, resolve_flags_to,
+			       attr_flags, -EBADF, recursive);
+}
+
 int calc_remount_flags_new(int dfd_from, const char *path_from,
 			   __u64 o_flags_from, __u64 resolve_flags_from,
 			   bool remount, unsigned long cur_flags,

src/lxc/mount_utils.h

@@ -189,6 +189,13 @@ __hidden extern int fd_bind_mount(int dfd_from, const char *path_from,
 				  __u64 o_flags_to, __u64 resolve_flags_to,
 				  unsigned int attr_flags, bool recursive);
 
+__hidden extern int fd_mount_idmapped(int dfd_from, const char *path_from,
+				      __u64 o_flags_from, __u64 resolve_flags_from,
+				      int dfd_to, const char *path_to,
+				      __u64 o_flags_to, __u64 resolve_flags_to,
+				      unsigned int attr_flags, int userns_fd,
+				      bool recursive);
+
 __hidden extern int calc_remount_flags_new(int dfd_from, const char *path_from,
 					   __u64 o_flags_from,
 					   __u64 resolve_flags_from,

src/lxc/start.c

@@ -1645,16 +1645,6 @@ static int lxc_spawn(struct lxc_handler *handler)
 		goto out_delete_net;
 	}
 
-	/* If the rootfs is not a blockdev, prevent the container from marking
-	 * it readonly.
-	 * If the container is unprivileged then skip rootfs pinning.
-	 */
-	ret = lxc_rootfs_prepare(&conf->rootfs, wants_to_map_ids);
-	if (ret) {
-		ERROR("Failed to handle rootfs pinning for container \"%s\"", handler->name);
-		goto out_delete_net;
-	}
-
 	/* Create a process in a new set of namespaces. */
 	if (share_ns) {
 		pid_t attacher_pid;
@@ -2040,9 +2030,20 @@ int __lxc_start(struct lxc_handler *handler, struct lxc_operations *ops,
 		goto out_abort;
 	}
 
+	/* If the rootfs is not a blockdev, prevent the container from marking
+	 * it readonly.
+	 * If the container is unprivileged then skip rootfs pinning.
+	 */
+	ret = lxc_rootfs_prepare(&conf->rootfs, !lxc_list_empty(&conf->id_map));
+	if (ret) {
+		ERROR("Failed to handle rootfs pinning for container \"%s\"", handler->name);
+		ret = -1;
+		goto out_abort;
+	}
+
 	if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
 		/* If the backing store is a device, mount it here and now. */
-		if (rootfs_is_blockdev(conf)) {
+		if (idmapped_rootfs_mnt(&conf->rootfs) || rootfs_is_blockdev(conf)) {
 			ret = unshare(CLONE_NEWNS);
 			if (ret < 0) {
 				ERROR("Failed to unshare CLONE_NEWNS");

src/lxc/storage/dir.c

@@ -148,23 +148,46 @@ int dir_mount(struct lxc_storage *bdev)
 
 	src = lxc_storage_get_path(bdev->src, bdev->type);
 
-	ret = mount(src, bdev->dest, "bind", MS_BIND | MS_REC | mntflags | pflags, mntdata);
-	if (ret < 0)
-		return log_error_errno(-errno, errno, "Failed to mount \"%s\" on \"%s\"", src, bdev->dest);
-
-	if (ret == 0 && (mntflags & MS_RDONLY)) {
-		mflags = add_required_remount_flags(src, bdev->dest, MS_BIND | MS_REC | mntflags | pflags | MS_REMOUNT);
-		ret = mount(src, bdev->dest, "bind", mflags, mntdata);
+	if (can_use_bind_mounts()) {
+		__do_close int fd_source = -EBADF, fd_target = -EBADF;
+
+		fd_source = open_at(-EBADF, src, PROTECT_OPATH_DIRECTORY, PROTECT_LOOKUP_ABSOLUTE, 0);
+		if (fd_source < 0)
+			return syserror("Failed to open \"%s\"", src);
+
+		fd_target = open_at(-EBADF, bdev->dest, PROTECT_OPATH_DIRECTORY, PROTECT_LOOKUP_ABSOLUTE, 0);
+		if (fd_target < 0)
+			return syserror("Failed to open \"%s\"", bdev->dest);
+
+		ret = fd_mount_idmapped(fd_source, "", PROTECT_OPATH_DIRECTORY,
+					PROTECT_LOOKUP_BENEATH, fd_target, "",
+					PROTECT_OPATH_DIRECTORY,
+					PROTECT_LOOKUP_BENEATH, 0,
+					bdev->rootfs->mnt_opts.userns_fd, true);
+		if (ret < 0)
+			return syserror("Failed to mount \"%s\" onto \"%s\"", src, bdev->dest);
+	} else {
+		ret = mount(src, bdev->dest, "bind", MS_BIND | MS_REC | mntflags | pflags, mntdata);
 		if (ret < 0)
-			return log_error_errno(-errno, errno, "Failed to remount \"%s\" on \"%s\" read-only with options \"%s\", mount flags \"%lu\", and propagation flags \"%lu\"",
-					       src ? src : "(none)", bdev->dest ? bdev->dest : "(none)", mntdata, mflags, pflags);
-		else
-			DEBUG("Remounted \"%s\" on \"%s\" read-only with options \"%s\", mount flags \"%lu\", and propagation flags \"%lu\"",
-			      src ? src : "(none)", bdev->dest ? bdev->dest : "(none)", mntdata, mflags, pflags);
+			return log_error_errno(-errno, errno, "Failed to mount \"%s\" on \"%s\"", src, bdev->dest);
+
+		if (ret == 0 && (mntflags & MS_RDONLY)) {
+			mflags = add_required_remount_flags(src, bdev->dest, MS_BIND | MS_REC | mntflags | pflags | MS_REMOUNT);
+
+			ret = mount(src, bdev->dest, "bind", mflags, mntdata);
+			if (ret < 0)
+				return log_error_errno(-errno, errno, "Failed to remount \"%s\" on \"%s\" read-only with options \"%s\", mount flags \"%lu\", and propagation flags \"%lu\"",
+						       src ? src : "(none)", bdev->dest ? bdev->dest : "(none)", mntdata, mflags, pflags);
+			else
+				DEBUG("Remounted \"%s\" on \"%s\" read-only with options \"%s\", mount flags \"%lu\", and propagation flags \"%lu\"",
+				      src ? src : "(none)", bdev->dest ? bdev->dest : "(none)", mntdata, mflags, pflags);
+		}
+
+		TRACE("Mounted \"%s\" on \"%s\" with options \"%s\", mount flags \"%lu\", and propagation flags \"%lu\"",
+		      src ? src : "(none)", bdev->dest ? bdev->dest : "(none)", mntdata, mflags, pflags);
 	}
 
-	TRACE("Mounted \"%s\" on \"%s\" with options \"%s\", mount flags \"%lu\", and propagation flags \"%lu\"",
-	      src ? src : "(none)", bdev->dest ? bdev->dest : "(none)", mntdata, mflags, pflags);
+	TRACE("Mounted \"%s\" onto \"%s\"", src, bdev->dest);
 	return 0;
 }