1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2015 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23 #include <linux/loop.h>
29 #include <sys/ioctl.h>
30 #include <sys/mount.h>
31 #include <sys/prctl.h>
33 #include <sys/statfs.h>
34 #include <sys/statvfs.h>
37 #include "sd-bus-protocol.h"
40 #include "alloc-util.h"
41 #include "btrfs-util.h"
45 #include "lockfile-util.h"
47 #include "machine-pool.h"
51 #include "mount-util.h"
52 #include "parse-util.h"
53 #include "path-util.h"
54 #include "process-util.h"
55 #include "signal-util.h"
56 #include "stat-util.h"
57 #include "string-util.h"
59 #define VAR_LIB_MACHINES_SIZE_START (1024UL*1024UL*500UL)
60 #define VAR_LIB_MACHINES_FREE_MIN (1024UL*1024UL*750UL)
62 static int check_btrfs(void) {
65 if (statfs("/var/lib/machines", &sfs
) < 0) {
69 if (statfs("/var/lib", &sfs
) < 0)
73 return F_TYPE_EQUAL(sfs
.f_type
, BTRFS_SUPER_MAGIC
);
76 static int setup_machine_raw(uint64_t size
, sd_bus_error
*error
) {
77 _cleanup_free_
char *tmp
= NULL
;
78 _cleanup_close_
int fd
= -1;
84 /* We want to be able to make use of btrfs-specific file
85 * system features, in particular subvolumes, reflinks and
86 * quota. Hence, if we detect that /var/lib/machines.raw is
87 * not located on btrfs, let's create a loopback file, place a
88 * btrfs file system into it, and mount it to
89 * /var/lib/machines. */
91 fd
= open("/var/lib/machines.raw", O_RDWR
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
99 return sd_bus_error_set_errnof(error
, errno
, "Failed to open /var/lib/machines.raw: %m");
101 r
= tempfn_xxxxxx("/var/lib/machines.raw", NULL
, &tmp
);
105 (void) mkdir_p_label("/var/lib", 0755);
106 fd
= open(tmp
, O_RDWR
|O_CREAT
|O_EXCL
|O_NOCTTY
|O_CLOEXEC
, 0600);
108 return sd_bus_error_set_errnof(error
, errno
, "Failed to create /var/lib/machines.raw: %m");
110 if (fstatvfs(fd
, &ss
) < 0) {
111 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to determine free space on /var/lib/machines.raw: %m");
115 if (ss
.f_bsize
* ss
.f_bavail
< VAR_LIB_MACHINES_FREE_MIN
) {
116 r
= sd_bus_error_setf(error
, SD_BUS_ERROR_FAILED
, "Not enough free disk space to set up /var/lib/machines.");
120 if (ftruncate(fd
, size
) < 0) {
121 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to enlarge /var/lib/machines.raw: %m");
127 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to fork mkfs.btrfs: %m");
135 (void) reset_all_signal_handlers();
136 (void) reset_signal_mask();
137 assert_se(prctl(PR_SET_PDEATHSIG
, SIGTERM
) == 0);
141 execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp
, NULL
);
148 r
= wait_for_terminate(pid
, &si
);
150 sd_bus_error_set_errnof(error
, r
, "Failed to wait for mkfs.btrfs: %m");
156 if (si
.si_code
!= CLD_EXITED
) {
157 r
= sd_bus_error_setf(error
, SD_BUS_ERROR_FAILED
, "mkfs.btrfs died abnormally.");
160 if (si
.si_status
== 99) {
161 r
= sd_bus_error_set_errnof(error
, ENOENT
, "Cannot set up /var/lib/machines, mkfs.btrfs is missing");
164 if (si
.si_status
!= 0) {
165 r
= sd_bus_error_setf(error
, SD_BUS_ERROR_FAILED
, "mkfs.btrfs failed with error code %i", si
.si_status
);
169 r
= rename_noreplace(AT_FDCWD
, tmp
, AT_FDCWD
, "/var/lib/machines.raw");
171 sd_bus_error_set_errnof(error
, r
, "Failed to move /var/lib/machines.raw into place: %m");
184 kill_and_sigcont(pid
, SIGKILL
);
189 int setup_machine_directory(uint64_t size
, sd_bus_error
*error
) {
190 _cleanup_release_lock_file_ LockFile lock_file
= LOCK_FILE_INIT
;
191 struct loop_info64 info
= {
192 .lo_flags
= LO_FLAGS_AUTOCLEAR
,
194 _cleanup_close_
int fd
= -1, control
= -1, loop
= -1;
195 _cleanup_free_
char* loopdev
= NULL
;
196 char tmpdir
[] = "/tmp/machine-pool.XXXXXX", *mntdir
= NULL
;
197 bool tmpdir_made
= false, mntdir_made
= false, mntdir_mounted
= false;
198 char buf
[FORMAT_BYTES_MAX
];
201 /* btrfs cannot handle file systems < 16M, hence use this as minimum */
202 if (size
== (uint64_t) -1)
203 size
= VAR_LIB_MACHINES_SIZE_START
;
204 else if (size
< 16*1024*1024)
207 /* Make sure we only set the directory up once at a time */
208 r
= make_lock_file("/run/systemd/machines.lock", LOCK_EX
, &lock_file
);
214 return sd_bus_error_set_errnof(error
, r
, "Failed to determine whether /var/lib/machines is located on btrfs: %m");
216 (void) btrfs_subvol_make_label("/var/lib/machines");
218 r
= btrfs_quota_enable("/var/lib/machines", true);
220 log_warning_errno(r
, "Failed to enable quota for /var/lib/machines, ignoring: %m");
222 r
= btrfs_subvol_auto_qgroup("/var/lib/machines", 0, true);
224 log_warning_errno(r
, "Failed to set up default quota hierarchy for /var/lib/machines, ignoring: %m");
229 if (path_is_mount_point("/var/lib/machines", NULL
, AT_SYMLINK_FOLLOW
) > 0) {
230 log_debug("/var/lib/machines is already a mount point, not creating loopback file for it.");
234 r
= dir_is_populated("/var/lib/machines");
235 if (r
< 0 && r
!= -ENOENT
)
238 log_debug("/var/log/machines is already populated, not creating loopback file for it.");
242 r
= mkfs_exists("btrfs");
244 return sd_bus_error_set_errnof(error
, ENOENT
, "Cannot set up /var/lib/machines, mkfs.btrfs is missing");
248 fd
= setup_machine_raw(size
, error
);
252 control
= open("/dev/loop-control", O_RDWR
|O_CLOEXEC
|O_NOCTTY
|O_NONBLOCK
);
254 return sd_bus_error_set_errnof(error
, errno
, "Failed to open /dev/loop-control: %m");
256 nr
= ioctl(control
, LOOP_CTL_GET_FREE
);
258 return sd_bus_error_set_errnof(error
, errno
, "Failed to allocate loop device: %m");
260 if (asprintf(&loopdev
, "/dev/loop%i", nr
) < 0) {
265 loop
= open(loopdev
, O_CLOEXEC
|O_RDWR
|O_NOCTTY
|O_NONBLOCK
);
267 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to open loopback device: %m");
271 if (ioctl(loop
, LOOP_SET_FD
, fd
) < 0) {
272 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to bind loopback device: %m");
276 if (ioctl(loop
, LOOP_SET_STATUS64
, &info
) < 0) {
277 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to enable auto-clear for loopback device: %m");
281 /* We need to make sure the new /var/lib/machines directory
282 * has an access mode of 0700 at the time it is first made
283 * available. mkfs will create it with 0755 however. Hence,
284 * let's mount the directory into an inaccessible directory
285 * below /tmp first, fix the access mode, and move it to the
286 * public place then. */
288 if (!mkdtemp(tmpdir
)) {
289 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to create temporary mount parent directory: %m");
294 mntdir
= strjoina(tmpdir
, "/mnt");
295 if (mkdir(mntdir
, 0700) < 0) {
296 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to create temporary mount directory: %m");
301 if (mount(loopdev
, mntdir
, "btrfs", 0, NULL
) < 0) {
302 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to mount loopback device: %m");
305 mntdir_mounted
= true;
307 r
= btrfs_quota_enable(mntdir
, true);
309 log_warning_errno(r
, "Failed to enable quota, ignoring: %m");
311 r
= btrfs_subvol_auto_qgroup(mntdir
, 0, true);
313 log_warning_errno(r
, "Failed to set up default quota hierarchy, ignoring: %m");
315 if (chmod(mntdir
, 0700) < 0) {
316 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to fix owner: %m");
320 (void) mkdir_p_label("/var/lib/machines", 0700);
322 if (mount(mntdir
, "/var/lib/machines", NULL
, MS_BIND
, NULL
) < 0) {
323 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to mount directory into right place: %m");
329 log_info("Set up /var/lib/machines as btrfs loopback file system of size %s mounted on /var/lib/machines.raw.", format_bytes(buf
, sizeof(buf
), size
));
331 (void) umount2(mntdir
, MNT_DETACH
);
332 (void) rmdir(mntdir
);
333 (void) rmdir(tmpdir
);
339 (void) umount2(mntdir
, MNT_DETACH
);
342 (void) rmdir(mntdir
);
344 (void) rmdir(tmpdir
);
347 (void) ioctl(loop
, LOOP_CLR_FD
);
348 loop
= safe_close(loop
);
351 if (control
>= 0 && nr
>= 0)
352 (void) ioctl(control
, LOOP_CTL_REMOVE
, nr
);
357 static int sync_path(const char *p
) {
358 _cleanup_close_
int fd
= -1;
360 fd
= open(p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
370 int grow_machine_directory(void) {
371 char buf
[FORMAT_BYTES_MAX
];
373 uint64_t old_size
, new_size
, max_add
;
376 /* Ensure the disk space data is accurate */
377 sync_path("/var/lib/machines");
378 sync_path("/var/lib/machines.raw");
380 if (statvfs("/var/lib/machines.raw", &a
) < 0)
383 if (statvfs("/var/lib/machines", &b
) < 0)
386 /* Don't grow if not enough disk space is available on the host */
387 if (((uint64_t) a
.f_bavail
* (uint64_t) a
.f_bsize
) <= VAR_LIB_MACHINES_FREE_MIN
)
390 /* Don't grow if at least 1/3th of the fs is still free */
391 if (b
.f_bavail
> b
.f_blocks
/ 3)
394 /* Calculate how much we are willing to add at most */
395 max_add
= ((uint64_t) a
.f_bavail
* (uint64_t) a
.f_bsize
) - VAR_LIB_MACHINES_FREE_MIN
;
397 /* Calculate the old size */
398 old_size
= (uint64_t) b
.f_blocks
* (uint64_t) b
.f_bsize
;
400 /* Calculate the new size as three times the size of what is used right now */
401 new_size
= ((uint64_t) b
.f_blocks
- (uint64_t) b
.f_bavail
) * (uint64_t) b
.f_bsize
* 3;
403 /* Always, grow at least to the start size */
404 if (new_size
< VAR_LIB_MACHINES_SIZE_START
)
405 new_size
= VAR_LIB_MACHINES_SIZE_START
;
407 /* If the new size is smaller than the old size, don't grow */
408 if (new_size
< old_size
)
411 /* Ensure we never add more than the maximum */
412 if (new_size
> old_size
+ max_add
)
413 new_size
= old_size
+ max_add
;
415 r
= btrfs_resize_loopback("/var/lib/machines", new_size
, true);
419 /* Also bump the quota, of both the subvolume leaf qgroup, as
420 * well as of any subtree quota group by the same id but a
421 * higher level, if it exists. */
422 (void) btrfs_qgroup_set_limit("/var/lib/machines", 0, new_size
);
423 (void) btrfs_subvol_set_subtree_quota_limit("/var/lib/machines", 0, new_size
);
425 log_info("Grew /var/lib/machines btrfs loopback file system to %s.", format_bytes(buf
, sizeof(buf
), new_size
));