***/
#include <sys/statfs.h>
+#include <linux/fs.h>
#include <fcntl.h>
-#include "strv.h"
#include "utf8.h"
#include "btrfs-util.h"
#include "path-util.h"
#include "copy.h"
+#include "mkdir.h"
+#include "rm-rf.h"
#include "machine-image.h"
static const char image_search_path[] =
i->read_only = read_only;
i->crtime = crtime;
i->mtime = mtime;
+ i->usage = i->usage_exclusive = (uint64_t) -1;
+ i->limit = i->limit_exclusive = (uint64_t) -1;
i->name = strdup(pretty);
if (!i->name)
assert(filename);
/* We explicitly *do* follow symlinks here, since we want to
- * allow symlinking trees into /var/lib/container/, and treat
+ * allow symlinking trees into /var/lib/machines/, and treat
* them normally. */
if (fstatat(dfd, filename, &st, 0) < 0)
(faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS);
if (S_ISDIR(st.st_mode)) {
+ _cleanup_close_ int fd = -1;
+ unsigned file_attr = 0;
if (!ret)
return 1;
if (!pretty)
pretty = filename;
+ fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
+ if (fd < 0)
+ return -errno;
+
/* btrfs subvolumes have inode 256 */
if (st.st_ino == 256) {
- _cleanup_close_ int fd = -1;
struct statfs sfs;
- fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY);
- if (fd < 0)
- return -errno;
-
if (fstatfs(fd, &sfs) < 0)
return -errno;
if (F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC)) {
BtrfsSubvolInfo info;
+ BtrfsQuotaInfo quota;
/* It's a btrfs subvolume */
if (r < 0)
return r;
+ r = btrfs_subvol_get_quota_fd(fd, "a);
+ if (r >= 0) {
+ (*ret)->usage = quota.referenced;
+ (*ret)->usage_exclusive = quota.exclusive;
+
+ (*ret)->limit = quota.referenced_max;
+ (*ret)->limit_exclusive = quota.exclusive_max;
+ }
+
return 1;
}
}
- /* It's just a normal directory. */
+ /* If the IMMUTABLE bit is set, we consider the
+ * directory read-only. Since the ioctl is not
+ * supported everywhere we ignore failures. */
+ (void) read_attr_fd(fd, &file_attr);
+ /* It's just a normal directory. */
r = image_new(IMAGE_DIRECTORY,
pretty,
path,
filename,
- read_only,
+ read_only || (file_attr & FS_IMMUTABLE_FL),
0,
0,
ret);
return 1;
- } else if (S_ISREG(st.st_mode) && endswith(filename, ".gpt")) {
+ } else if (S_ISREG(st.st_mode) && endswith(filename, ".raw")) {
usec_t crtime = 0;
- /* It's a GPT block device */
+ /* It's a RAW disk image */
if (!ret)
return 1;
if (!pretty)
pretty = strndupa(filename, strlen(filename) - 4);
- r = image_new(IMAGE_GPT,
+ r = image_new(IMAGE_RAW,
pretty,
path,
filename,
if (r < 0)
return r;
+ (*ret)->usage = (*ret)->usage_exclusive = st.st_blocks * 512;
+ (*ret)->limit = (*ret)->limit_exclusive = st.st_size;
+
return 1;
}
r = image_make(NULL, dirfd(d), path, name, ret);
if (r == 0 || r == -ENOENT) {
- _cleanup_free_ char *gpt = NULL;
+ _cleanup_free_ char *raw = NULL;
- gpt = strappend(name, ".gpt");
- if (!gpt)
+ raw = strappend(name, ".raw");
+ if (!raw)
return -ENOMEM;
- r = image_make(NULL, dirfd(d), path, gpt, ret);
+ r = image_make(NULL, dirfd(d), path, raw, ret);
if (r == 0 || r == -ENOENT)
continue;
}
}
int image_remove(Image *i) {
+ _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
+ int r;
+
assert(i);
if (path_equal(i->path, "/") ||
path_startswith(i->path, "/usr"))
return -EROFS;
+ /* Make sure we don't interfere with a running nspawn */
+ r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
+ if (r < 0)
+ return r;
+
switch (i->type) {
case IMAGE_SUBVOLUME:
return btrfs_subvol_remove(i->path);
case IMAGE_DIRECTORY:
- case IMAGE_GPT:
- return rm_rf_dangerous(i->path, false, true, false);
+ /* Allow deletion of read-only directories */
+ (void) chattr_path(i->path, false, FS_IMMUTABLE_FL);
+ return rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
+
+ case IMAGE_RAW:
+ if (unlink(i->path) < 0)
+ return -errno;
+
+ return 0;
default:
- return -ENOTSUP;
+ return -EOPNOTSUPP;
}
}
int image_rename(Image *i, const char *new_name) {
+ _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
_cleanup_free_ char *new_path = NULL, *nn = NULL;
+ unsigned file_attr = 0;
int r;
assert(i);
path_startswith(i->path, "/usr"))
return -EROFS;
+ /* Make sure we don't interfere with a running nspawn */
+ r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
+ if (r < 0)
+ return r;
+
+ /* Make sure nobody takes the new name, between the time we
+ * checked it is currently unused in all search paths, and the
+ * time we take possesion of it */
+ r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
+ if (r < 0)
+ return r;
+
r = image_find(new_name, NULL);
if (r < 0)
return r;
switch (i->type) {
- case IMAGE_SUBVOLUME:
case IMAGE_DIRECTORY:
+ /* Turn of the immutable bit while we rename the image, so that we can rename it */
+ (void) read_attr_path(i->path, &file_attr);
+
+ if (file_attr & FS_IMMUTABLE_FL)
+ (void) chattr_path(i->path, false, FS_IMMUTABLE_FL);
+
+ /* fall through */
+
+ case IMAGE_SUBVOLUME:
new_path = file_in_same_dir(i->path, new_name);
break;
- case IMAGE_GPT: {
+ case IMAGE_RAW: {
const char *fn;
- fn = strappenda(new_name, ".gpt");
+ fn = strjoina(new_name, ".raw");
new_path = file_in_same_dir(i->path, fn);
break;
}
default:
- return -ENOTSUP;
+ return -EOPNOTSUPP;
}
if (!new_path)
if (!nn)
return -ENOMEM;
- if (renameat2(AT_FDCWD, i->path, AT_FDCWD, new_path, RENAME_NOREPLACE) < 0)
- return -errno;
+ r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
+ if (r < 0)
+ return r;
+
+ /* Restore the immutable bit, if it was set before */
+ if (file_attr & FS_IMMUTABLE_FL)
+ (void) chattr_path(new_path, true, FS_IMMUTABLE_FL);
free(i->path);
i->path = new_path;
}
int image_clone(Image *i, const char *new_name, bool read_only) {
+ _cleanup_release_lock_file_ LockFile name_lock = LOCK_FILE_INIT;
const char *new_path;
int r;
if (!image_name_is_valid(new_name))
return -EINVAL;
+ /* Make sure nobody takes the new name, between the time we
+ * checked it is currently unused in all search paths, and the
+ * time we take possesion of it */
+ r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
+ if (r < 0)
+ return r;
+
r = image_find(new_name, NULL);
if (r < 0)
return r;
case IMAGE_SUBVOLUME:
case IMAGE_DIRECTORY:
- new_path = strappenda("/var/lib/container/", new_name);
+ new_path = strjoina("/var/lib/machines/", new_name);
r = btrfs_subvol_snapshot(i->path, new_path, read_only, true);
break;
- case IMAGE_GPT:
- new_path = strappenda("/var/lib/container/", new_name, ".gpt");
+ case IMAGE_RAW:
+ new_path = strjoina("/var/lib/machines/", new_name, ".raw");
- r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, false);
+ r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, false, FS_NOCOW_FL);
break;
default:
- return -ENOTSUP;
+ return -EOPNOTSUPP;
}
if (r < 0)
}
int image_read_only(Image *i, bool b) {
+ _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
int r;
assert(i);
path_startswith(i->path, "/usr"))
return -EROFS;
+ /* Make sure we don't interfere with a running nspawn */
+ r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
+ if (r < 0)
+ return r;
+
switch (i->type) {
case IMAGE_SUBVOLUME:
r = btrfs_subvol_set_read_only(i->path, b);
if (r < 0)
return r;
+
+ break;
+
+ case IMAGE_DIRECTORY:
+ /* For simple directory trees we cannot use the access
+ mode of the top-level directory, since it has an
+ effect on the container itself. However, we can
+ use the "immutable" flag, to at least make the
+ top-level directory read-only. It's not as good as
+ a read-only subvolume, but at least something, and
+ we can read the value back.*/
+
+ r = chattr_path(i->path, b, FS_IMMUTABLE_FL);
+ if (r < 0)
+ return r;
+
break;
- case IMAGE_GPT: {
+ case IMAGE_RAW: {
struct stat st;
if (stat(i->path, &st) < 0)
if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0)
return -errno;
+
+ /* If the images is now read-only, it's a good time to
+ * defrag it, given that no write patterns will
+ * fragment it again. */
+ if (b)
+ (void) btrfs_defrag(i->path);
break;
}
- case IMAGE_DIRECTORY:
default:
- return -ENOTSUP;
+ return -EOPNOTSUPP;
}
return 0;
}
+int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
+ _cleanup_free_ char *p = NULL;
+ LockFile t = LOCK_FILE_INIT;
+ struct stat st;
+ int r;
+
+ assert(path);
+ assert(global);
+ assert(local);
+
+ /* Locks an image path. This actually creates two locks: one
+ * "local" one, next to the image path itself, which might be
+ * shared via NFS. And another "global" one, in /run, that
+ * uses the device/inode number. This has the benefit that we
+ * can even lock a tree that is a mount point, correctly. */
+
+ if (path_equal(path, "/"))
+ return -EBUSY;
+
+ if (!path_is_absolute(path))
+ return -EINVAL;
+
+ if (stat(path, &st) >= 0) {
+ if (asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino) < 0)
+ return -ENOMEM;
+ }
+
+ r = make_lock_file_for(path, operation, &t);
+ if (r < 0)
+ return r;
+
+ if (p) {
+ mkdir_p("/run/systemd/nspawn/locks", 0600);
+
+ r = make_lock_file(p, operation, global);
+ if (r < 0) {
+ release_lock_file(&t);
+ return r;
+ }
+ }
+
+ *local = t;
+ return 0;
+}
+
+int image_set_limit(Image *i, uint64_t referenced_max) {
+ assert(i);
+
+ if (path_equal(i->path, "/") ||
+ path_startswith(i->path, "/usr"))
+ return -EROFS;
+
+ if (i->type != IMAGE_SUBVOLUME)
+ return -EOPNOTSUPP;
+
+ return btrfs_quota_limit(i->path, referenced_max);
+}
+
+int image_name_lock(const char *name, int operation, LockFile *ret) {
+ const char *p;
+
+ assert(name);
+ assert(ret);
+
+ /* Locks an image name, regardless of the precise path used. */
+
+ if (!image_name_is_valid(name))
+ return -EINVAL;
+
+ if (streq(name, ".host"))
+ return -EBUSY;
+
+ mkdir_p("/run/systemd/nspawn/locks", 0600);
+ p = strjoina("/run/systemd/nspawn/locks/name-", name);
+
+ return make_lock_file(p, operation, ret);
+}
+
+bool image_name_is_valid(const char *s) {
+ if (!filename_is_valid(s))
+ return false;
+
+ if (string_has_cc(s, NULL))
+ return false;
+
+ if (!utf8_is_valid(s))
+ return false;
+
+ /* Temporary files for atomically creating new files */
+ if (startswith(s, ".#"))
+ return false;
+
+ return true;
+}
+
static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
[IMAGE_DIRECTORY] = "directory",
[IMAGE_SUBVOLUME] = "subvolume",
- [IMAGE_GPT] = "gpt",
+ [IMAGE_RAW] = "raw",
};
DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);
~ianmdlvl / elogind.git / blobdiff