1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2015 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/mount.h>
23 #include <sys/prctl.h>
24 #include <sys/statvfs.h>
27 #include "btrfs-util.h"
29 #include "lockfile-util.h"
30 #include "machine-pool.h"
32 #include "path-util.h"
33 #include "process-util.h"
34 #include "signal-util.h"
35 #include "string-util.h"
38 #define VAR_LIB_MACHINES_SIZE_START (1024UL*1024UL*500UL)
39 #define VAR_LIB_MACHINES_FREE_MIN (1024UL*1024UL*750UL)
41 static int check_btrfs(void) {
44 if (statfs("/var/lib/machines", &sfs
) < 0) {
48 if (statfs("/var/lib", &sfs
) < 0)
52 return F_TYPE_EQUAL(sfs
.f_type
, BTRFS_SUPER_MAGIC
);
55 static int setup_machine_raw(uint64_t size
, sd_bus_error
*error
) {
56 _cleanup_free_
char *tmp
= NULL
;
57 _cleanup_close_
int fd
= -1;
63 /* We want to be able to make use of btrfs-specific file
64 * system features, in particular subvolumes, reflinks and
65 * quota. Hence, if we detect that /var/lib/machines.raw is
66 * not located on btrfs, let's create a loopback file, place a
67 * btrfs file system into it, and mount it to
68 * /var/lib/machines. */
70 fd
= open("/var/lib/machines.raw", O_RDWR
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
78 return sd_bus_error_set_errnof(error
, errno
, "Failed to open /var/lib/machines.raw: %m");
80 r
= tempfn_xxxxxx("/var/lib/machines.raw", NULL
, &tmp
);
84 (void) mkdir_p_label("/var/lib", 0755);
85 fd
= open(tmp
, O_RDWR
|O_CREAT
|O_EXCL
|O_NOCTTY
|O_CLOEXEC
, 0600);
87 return sd_bus_error_set_errnof(error
, errno
, "Failed to create /var/lib/machines.raw: %m");
89 if (fstatvfs(fd
, &ss
) < 0) {
90 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to determine free space on /var/lib/machines.raw: %m");
94 if (ss
.f_bsize
* ss
.f_bavail
< VAR_LIB_MACHINES_FREE_MIN
) {
95 r
= sd_bus_error_setf(error
, SD_BUS_ERROR_FAILED
, "Not enough free disk space to set up /var/lib/machines.");
99 if (ftruncate(fd
, size
) < 0) {
100 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to enlarge /var/lib/machines.raw: %m");
106 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to fork mkfs.btrfs: %m");
114 (void) reset_all_signal_handlers();
115 (void) reset_signal_mask();
116 assert_se(prctl(PR_SET_PDEATHSIG
, SIGTERM
) == 0);
120 execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp
, NULL
);
127 r
= wait_for_terminate(pid
, &si
);
129 sd_bus_error_set_errnof(error
, r
, "Failed to wait for mkfs.btrfs: %m");
135 if (si
.si_code
!= CLD_EXITED
) {
136 r
= sd_bus_error_setf(error
, SD_BUS_ERROR_FAILED
, "mkfs.btrfs died abnormally.");
139 if (si
.si_status
== 99) {
140 r
= sd_bus_error_set_errnof(error
, ENOENT
, "Cannot set up /var/lib/machines, mkfs.btrfs is missing");
143 if (si
.si_status
!= 0) {
144 r
= sd_bus_error_setf(error
, SD_BUS_ERROR_FAILED
, "mkfs.btrfs failed with error code %i", si
.si_status
);
148 r
= rename_noreplace(AT_FDCWD
, tmp
, AT_FDCWD
, "/var/lib/machines.raw");
150 sd_bus_error_set_errnof(error
, r
, "Failed to move /var/lib/machines.raw into place: %m");
163 kill_and_sigcont(pid
, SIGKILL
);
168 int setup_machine_directory(uint64_t size
, sd_bus_error
*error
) {
169 _cleanup_release_lock_file_ LockFile lock_file
= LOCK_FILE_INIT
;
170 struct loop_info64 info
= {
171 .lo_flags
= LO_FLAGS_AUTOCLEAR
,
173 _cleanup_close_
int fd
= -1, control
= -1, loop
= -1;
174 _cleanup_free_
char* loopdev
= NULL
;
175 char tmpdir
[] = "/tmp/machine-pool.XXXXXX", *mntdir
= NULL
;
176 bool tmpdir_made
= false, mntdir_made
= false, mntdir_mounted
= false;
177 char buf
[FORMAT_BYTES_MAX
];
180 /* btrfs cannot handle file systems < 16M, hence use this as minimum */
181 if (size
== (uint64_t) -1)
182 size
= VAR_LIB_MACHINES_SIZE_START
;
183 else if (size
< 16*1024*1024)
186 /* Make sure we only set the directory up once at a time */
187 r
= make_lock_file("/run/systemd/machines.lock", LOCK_EX
, &lock_file
);
193 return sd_bus_error_set_errnof(error
, r
, "Failed to determine whether /var/lib/machines is located on btrfs: %m");
195 (void) btrfs_subvol_make_label("/var/lib/machines");
197 r
= btrfs_quota_enable("/var/lib/machines", true);
199 log_warning_errno(r
, "Failed to enable quota for /var/lib/machines, ignoring: %m");
201 r
= btrfs_subvol_auto_qgroup("/var/lib/machines", 0, true);
203 log_warning_errno(r
, "Failed to set up default quota hierarchy for /var/lib/machines, ignoring: %m");
208 if (path_is_mount_point("/var/lib/machines", AT_SYMLINK_FOLLOW
) > 0) {
209 log_debug("/var/lib/machines is already a mount point, not creating loopback file for it.");
213 r
= dir_is_populated("/var/lib/machines");
214 if (r
< 0 && r
!= -ENOENT
)
217 log_debug("/var/log/machines is already populated, not creating loopback file for it.");
221 r
= mkfs_exists("btrfs");
223 log_debug("mkfs.btrfs is missing, cannot create loopback file for /var/lib/machines.");
229 fd
= setup_machine_raw(size
, error
);
233 control
= open("/dev/loop-control", O_RDWR
|O_CLOEXEC
|O_NOCTTY
|O_NONBLOCK
);
235 return sd_bus_error_set_errnof(error
, errno
, "Failed to open /dev/loop-control: %m");
237 nr
= ioctl(control
, LOOP_CTL_GET_FREE
);
239 return sd_bus_error_set_errnof(error
, errno
, "Failed to allocate loop device: %m");
241 if (asprintf(&loopdev
, "/dev/loop%i", nr
) < 0) {
246 loop
= open(loopdev
, O_CLOEXEC
|O_RDWR
|O_NOCTTY
|O_NONBLOCK
);
248 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to open loopback device: %m");
252 if (ioctl(loop
, LOOP_SET_FD
, fd
) < 0) {
253 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to bind loopback device: %m");
257 if (ioctl(loop
, LOOP_SET_STATUS64
, &info
) < 0) {
258 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to enable auto-clear for loopback device: %m");
262 /* We need to make sure the new /var/lib/machines directory
263 * has an access mode of 0700 at the time it is first made
264 * available. mkfs will create it with 0755 however. Hence,
265 * let's mount the directory into an inaccessible directory
266 * below /tmp first, fix the access mode, and move it to the
267 * public place then. */
269 if (!mkdtemp(tmpdir
)) {
270 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to create temporary mount parent directory: %m");
275 mntdir
= strjoina(tmpdir
, "/mnt");
276 if (mkdir(mntdir
, 0700) < 0) {
277 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to create temporary mount directory: %m");
282 if (mount(loopdev
, mntdir
, "btrfs", 0, NULL
) < 0) {
283 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to mount loopback device: %m");
286 mntdir_mounted
= true;
288 r
= btrfs_quota_enable(mntdir
, true);
290 log_warning_errno(r
, "Failed to enable quota, ignoring: %m");
292 r
= btrfs_subvol_auto_qgroup(mntdir
, 0, true);
294 log_warning_errno(r
, "Failed to set up default quota hierarchy, ignoring: %m");
296 if (chmod(mntdir
, 0700) < 0) {
297 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to fix owner: %m");
301 (void) mkdir_p_label("/var/lib/machines", 0700);
303 if (mount(mntdir
, "/var/lib/machines", NULL
, MS_BIND
, NULL
) < 0) {
304 r
= sd_bus_error_set_errnof(error
, errno
, "Failed to mount directory into right place: %m");
310 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
));
312 (void) umount2(mntdir
, MNT_DETACH
);
313 (void) rmdir(mntdir
);
314 (void) rmdir(tmpdir
);
320 (void) umount2(mntdir
, MNT_DETACH
);
323 (void) rmdir(mntdir
);
325 (void) rmdir(tmpdir
);
328 (void) ioctl(loop
, LOOP_CLR_FD
);
329 loop
= safe_close(loop
);
332 if (control
>= 0 && nr
>= 0)
333 (void) ioctl(control
, LOOP_CTL_REMOVE
, nr
);
338 static int sync_path(const char *p
) {
339 _cleanup_close_
int fd
= -1;
341 fd
= open(p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
351 int grow_machine_directory(void) {
352 char buf
[FORMAT_BYTES_MAX
];
354 uint64_t old_size
, new_size
, max_add
;
357 /* Ensure the disk space data is accurate */
358 sync_path("/var/lib/machines");
359 sync_path("/var/lib/machines.raw");
361 if (statvfs("/var/lib/machines.raw", &a
) < 0)
364 if (statvfs("/var/lib/machines", &b
) < 0)
367 /* Don't grow if not enough disk space is available on the host */
368 if (((uint64_t) a
.f_bavail
* (uint64_t) a
.f_bsize
) <= VAR_LIB_MACHINES_FREE_MIN
)
371 /* Don't grow if at least 1/3th of the fs is still free */
372 if (b
.f_bavail
> b
.f_blocks
/ 3)
375 /* Calculate how much we are willing to add at maximum */
376 max_add
= ((uint64_t) a
.f_bavail
* (uint64_t) a
.f_bsize
) - VAR_LIB_MACHINES_FREE_MIN
;
378 /* Calculate the old size */
379 old_size
= (uint64_t) b
.f_blocks
* (uint64_t) b
.f_bsize
;
381 /* Calculate the new size as three times the size of what is used right now */
382 new_size
= ((uint64_t) b
.f_blocks
- (uint64_t) b
.f_bavail
) * (uint64_t) b
.f_bsize
* 3;
384 /* Always, grow at least to the start size */
385 if (new_size
< VAR_LIB_MACHINES_SIZE_START
)
386 new_size
= VAR_LIB_MACHINES_SIZE_START
;
388 /* If the new size is smaller than the old size, don't grow */
389 if (new_size
< old_size
)
392 /* Ensure we never add more than the maximum */
393 if (new_size
> old_size
+ max_add
)
394 new_size
= old_size
+ max_add
;
396 r
= btrfs_resize_loopback("/var/lib/machines", new_size
, true);
400 /* Also bump the quota, of both the subvolume leaf qgroup, as
401 * well as of any subtree quota group by the same id but a
402 * higher level, if it exists. */
403 (void) btrfs_qgroup_set_limit("/var/lib/machines", 0, new_size
);
404 (void) btrfs_subvol_set_subtree_quota_limit("/var/lib/machines", 0, new_size
);
406 log_info("Grew /var/lib/machines btrfs loopback file system to %s.", format_bytes(buf
, sizeof(buf
), new_size
));