1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2013 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
22 #include <sys/statfs.h>
28 #include "btrfs-util.h"
29 #include "path-util.h"
32 #include "machine-image.h"
34 static const char image_search_path[] =
36 "/var/lib/container\0"
37 "/usr/local/lib/machines\0"
38 "/usr/lib/machines\0";
40 Image *image_unref(Image *i) {
60 _cleanup_(image_unrefp) Image *i = NULL;
63 assert(t < _IMAGE_TYPE_MAX);
73 i->read_only = read_only;
76 i->size = i->size_exclusive = (uint64_t) -1;
77 i->limit = i->limit_exclusive = (uint64_t) -1;
79 i->name = strdup(pretty);
84 i->path = strjoin(path, "/", filename, NULL);
86 i->path = strdup(filename);
91 path_kill_slashes(i->path);
99 static int image_make(
103 const char *filename,
112 /* We explicitly *do* follow symlinks here, since we want to
113 * allow symlinking trees into /var/lib/machines/, and treat
116 if (fstatat(dfd, filename, &st, 0) < 0)
120 (path && path_startswith(path, "/usr")) ||
121 (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
123 if (S_ISDIR(st.st_mode)) {
124 _cleanup_close_ int fd = -1;
125 unsigned file_attr = 0;
133 fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
137 /* btrfs subvolumes have inode 256 */
138 if (st.st_ino == 256) {
141 if (fstatfs(fd, &sfs) < 0)
144 if (F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC)) {
145 BtrfsSubvolInfo info;
146 BtrfsQuotaInfo quota;
148 /* It's a btrfs subvolume */
150 r = btrfs_subvol_get_info_fd(fd, &info);
154 r = image_new(IMAGE_SUBVOLUME,
158 info.read_only || read_only,
165 r = btrfs_subvol_get_quota_fd(fd, "a);
167 (*ret)->size = quota.referred;
168 (*ret)->size_exclusive = quota.exclusive;
170 (*ret)->limit = quota.referred_max;
171 (*ret)->limit_exclusive = quota.exclusive_max;
178 /* If the IMMUTABLE bit is set, we consider the
179 * directory read-only. Since the ioctl is not
180 * supported everywhere we ignore failures. */
181 (void) read_attr_fd(fd, &file_attr);
183 /* It's just a normal directory. */
184 r = image_new(IMAGE_DIRECTORY,
188 read_only || (file_attr & FS_IMMUTABLE_FL),
197 } else if (S_ISREG(st.st_mode) && endswith(filename, ".raw")) {
200 /* It's a RAW disk image */
205 fd_getcrtime_at(dfd, filename, &crtime, 0);
208 pretty = strndupa(filename, strlen(filename) - 4);
210 r = image_new(IMAGE_RAW,
214 !(st.st_mode & 0222) || read_only,
216 timespec_load(&st.st_mtim),
221 (*ret)->size = (*ret)->size_exclusive = st.st_blocks * 512;
222 (*ret)->limit = (*ret)->limit_exclusive = st.st_size;
230 int image_find(const char *name, Image **ret) {
236 /* There are no images with invalid names */
237 if (!image_name_is_valid(name))
240 NULSTR_FOREACH(path, image_search_path) {
241 _cleanup_closedir_ DIR *d = NULL;
251 r = image_make(NULL, dirfd(d), path, name, ret);
252 if (r == 0 || r == -ENOENT) {
253 _cleanup_free_ char *raw = NULL;
255 raw = strappend(name, ".raw");
259 r = image_make(NULL, dirfd(d), path, raw, ret);
260 if (r == 0 || r == -ENOENT)
269 if (streq(name, ".host"))
270 return image_make(".host", AT_FDCWD, NULL, "/", ret);
275 int image_discover(Hashmap *h) {
281 NULSTR_FOREACH(path, image_search_path) {
282 _cleanup_closedir_ DIR *d = NULL;
293 FOREACH_DIRENT_ALL(de, d, return -errno) {
294 _cleanup_(image_unrefp) Image *image = NULL;
296 if (!image_name_is_valid(de->d_name))
299 if (hashmap_contains(h, de->d_name))
302 r = image_make(NULL, dirfd(d), path, de->d_name, &image);
303 if (r == 0 || r == -ENOENT)
308 r = hashmap_put(h, image->name, image);
316 if (!hashmap_contains(h, ".host")) {
317 _cleanup_(image_unrefp) Image *image = NULL;
319 r = image_make(".host", AT_FDCWD, NULL, "/", &image);
323 r = hashmap_put(h, image->name, image);
334 void image_hashmap_free(Hashmap *map) {
337 while ((i = hashmap_steal_first(map)))
343 int image_remove(Image *i) {
344 _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
349 if (path_equal(i->path, "/") ||
350 path_startswith(i->path, "/usr"))
353 /* Make sure we don't interfere with a running nspawn */
354 r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
360 case IMAGE_SUBVOLUME:
361 return btrfs_subvol_remove(i->path);
363 case IMAGE_DIRECTORY:
364 /* Allow deletion of read-only directories */
365 (void) chattr_path(i->path, false, FS_IMMUTABLE_FL);
370 return rm_rf_dangerous(i->path, false, true, false);
377 int image_rename(Image *i, const char *new_name) {
378 _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
379 _cleanup_free_ char *new_path = NULL, *nn = NULL;
380 unsigned file_attr = 0;
385 if (!image_name_is_valid(new_name))
388 if (path_equal(i->path, "/") ||
389 path_startswith(i->path, "/usr"))
392 /* Make sure we don't interfere with a running nspawn */
393 r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
397 /* Make sure nobody takes the new name, between the time we
398 * checked it is currently unused in all search paths, and the
399 * time we take possesion of it */
400 r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
404 r = image_find(new_name, NULL);
412 case IMAGE_DIRECTORY:
413 /* Turn of the immutable bit while we rename the image, so that we can rename it */
414 (void) read_attr_path(i->path, &file_attr);
416 if (file_attr & FS_IMMUTABLE_FL)
417 (void) chattr_path(i->path, false, FS_IMMUTABLE_FL);
421 case IMAGE_SUBVOLUME:
422 new_path = file_in_same_dir(i->path, new_name);
428 fn = strappenda(new_name, ".raw");
429 new_path = file_in_same_dir(i->path, fn);
440 nn = strdup(new_name);
444 if (renameat2(AT_FDCWD, i->path, AT_FDCWD, new_path, RENAME_NOREPLACE) < 0)
447 /* Restore the immutable bit, if it was set before */
448 if (file_attr & FS_IMMUTABLE_FL)
449 (void) chattr_path(new_path, true, FS_IMMUTABLE_FL);
462 int image_clone(Image *i, const char *new_name, bool read_only) {
463 _cleanup_release_lock_file_ LockFile name_lock = LOCK_FILE_INIT;
464 const char *new_path;
469 if (!image_name_is_valid(new_name))
472 /* Make sure nobody takes the new name, between the time we
473 * checked it is currently unused in all search paths, and the
474 * time we take possesion of it */
475 r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
479 r = image_find(new_name, NULL);
487 case IMAGE_SUBVOLUME:
488 case IMAGE_DIRECTORY:
489 new_path = strappenda("/var/lib/machines/", new_name);
491 r = btrfs_subvol_snapshot(i->path, new_path, read_only, true);
495 new_path = strappenda("/var/lib/machines/", new_name, ".raw");
497 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, false, FS_NOCOW_FL);
510 int image_read_only(Image *i, bool b) {
511 _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
515 if (path_equal(i->path, "/") ||
516 path_startswith(i->path, "/usr"))
519 /* Make sure we don't interfere with a running nspawn */
520 r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
526 case IMAGE_SUBVOLUME:
527 r = btrfs_subvol_set_read_only(i->path, b);
533 case IMAGE_DIRECTORY:
534 /* For simple directory trees we cannot use the access
535 mode of the top-level directory, since it has an
536 effect on the container itself. However, we can
537 use the "immutable" flag, to at least make the
538 top-level directory read-only. It's not as good as
539 a read-only subvolume, but at least something, and
540 we can read the value back.*/
542 r = chattr_path(i->path, b, FS_IMMUTABLE_FL);
551 if (stat(i->path, &st) < 0)
554 if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
557 /* If the images is now read-only, it's a good time to
558 * defrag it, given that no write patterns will
559 * fragment it again. */
561 (void) btrfs_defrag(i->path);
572 int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
573 _cleanup_free_ char *p = NULL;
574 LockFile t = LOCK_FILE_INIT;
582 /* Locks an image path. This actually creates two locks: one
583 * "local" one, next to the image path itself, which might be
584 * shared via NFS. And another "global" one, in /run, that
585 * uses the device/inode number. This has the benefit that we
586 * can even lock a tree that is a mount point, correctly. */
588 if (path_equal(path, "/"))
591 if (!path_is_absolute(path))
594 if (stat(path, &st) >= 0) {
595 if (asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino) < 0)
599 r = make_lock_file_for(path, operation, &t);
604 mkdir_p("/run/systemd/nspawn/locks", 0600);
606 r = make_lock_file(p, operation, global);
608 release_lock_file(&t);
617 int image_name_lock(const char *name, int operation, LockFile *ret) {
623 /* Locks an image name, regardless of the precise path used. */
625 if (!image_name_is_valid(name))
628 if (streq(name, ".host"))
631 mkdir_p("/run/systemd/nspawn/locks", 0600);
632 p = strappenda("/run/systemd/nspawn/locks/name-", name);
634 return make_lock_file(p, operation, ret);
637 bool image_name_is_valid(const char *s) {
638 if (!filename_is_valid(s))
641 if (string_has_cc(s, NULL))
644 if (!utf8_is_valid(s))
647 /* Temporary files for atomically creating new files */
648 if (startswith(s, ".#"))
654 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
655 [IMAGE_DIRECTORY] = "directory",
656 [IMAGE_SUBVOLUME] = "subvolume",
660 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);