]>
Commit | Line | Data |
---|---|---|
432cea00 LP |
1 | /*** |
2 | This file is part of systemd. | |
3 | ||
4 | Copyright 2015 Lennart Poettering | |
5 | ||
6 | systemd is free software; you can redistribute it and/or modify it | |
7 | under the terms of the GNU Lesser General Public License as published by | |
8 | the Free Software Foundation; either version 2.1 of the License, or | |
9 | (at your option) any later version. | |
10 | ||
11 | systemd is distributed in the hope that it will be useful, but | |
12 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | Lesser General Public License for more details. | |
15 | ||
16 | You should have received a copy of the GNU Lesser General Public License | |
17 | along with systemd; If not, see <http://www.gnu.org/licenses/>. | |
18 | ***/ | |
19 | ||
a8fbdf54 TA |
20 | #include <errno.h> |
21 | #include <fcntl.h> | |
22 | #include <linux/loop.h> | |
23 | #include <signal.h> | |
24 | #include <stdbool.h> | |
25 | #include <stdio.h> | |
26 | #include <stdlib.h> | |
e306723e | 27 | #include <sys/file.h> |
a8fbdf54 | 28 | #include <sys/ioctl.h> |
07630cea | 29 | #include <sys/mount.h> |
432cea00 | 30 | #include <sys/prctl.h> |
a8fbdf54 TA |
31 | #include <sys/stat.h> |
32 | #include <sys/statfs.h> | |
432cea00 | 33 | #include <sys/statvfs.h> |
a8fbdf54 TA |
34 | #include <unistd.h> |
35 | ||
36 | #include "sd-bus-protocol.h" | |
37 | #include "sd-bus.h" | |
432cea00 | 38 | |
b5efdb8a | 39 | #include "alloc-util.h" |
07630cea | 40 | #include "btrfs-util.h" |
3ffd4af2 | 41 | #include "fd-util.h" |
0d39fa9c | 42 | #include "fileio.h" |
f4f15635 | 43 | #include "fs-util.h" |
cd2eb9e9 | 44 | #include "lockfile-util.h" |
a8fbdf54 | 45 | #include "log.h" |
3ffd4af2 | 46 | #include "machine-pool.h" |
a8fbdf54 TA |
47 | #include "macro.h" |
48 | #include "missing.h" | |
432cea00 | 49 | #include "mkdir.h" |
f4f15635 | 50 | #include "mount-util.h" |
6bedfcbb | 51 | #include "parse-util.h" |
432cea00 | 52 | #include "path-util.h" |
07630cea | 53 | #include "process-util.h" |
24882e06 | 54 | #include "signal-util.h" |
8fcde012 | 55 | #include "stat-util.h" |
07630cea | 56 | #include "string-util.h" |
432cea00 LP |
57 | |
58 | #define VAR_LIB_MACHINES_SIZE_START (1024UL*1024UL*500UL) | |
59 | #define VAR_LIB_MACHINES_FREE_MIN (1024UL*1024UL*750UL) | |
60 | ||
61 | static int check_btrfs(void) { | |
62 | struct statfs sfs; | |
63 | ||
64 | if (statfs("/var/lib/machines", &sfs) < 0) { | |
65 | if (errno != ENOENT) | |
66 | return -errno; | |
67 | ||
68 | if (statfs("/var/lib", &sfs) < 0) | |
69 | return -errno; | |
70 | } | |
71 | ||
72 | return F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC); | |
73 | } | |
74 | ||
4cee5eed | 75 | static int setup_machine_raw(uint64_t size, sd_bus_error *error) { |
432cea00 LP |
76 | _cleanup_free_ char *tmp = NULL; |
77 | _cleanup_close_ int fd = -1; | |
78 | struct statvfs ss; | |
79 | pid_t pid = 0; | |
80 | siginfo_t si; | |
81 | int r; | |
82 | ||
83 | /* We want to be able to make use of btrfs-specific file | |
84 | * system features, in particular subvolumes, reflinks and | |
85 | * quota. Hence, if we detect that /var/lib/machines.raw is | |
86 | * not located on btrfs, let's create a loopback file, place a | |
87 | * btrfs file system into it, and mount it to | |
88 | * /var/lib/machines. */ | |
89 | ||
90 | fd = open("/var/lib/machines.raw", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
91 | if (fd >= 0) { | |
92 | r = fd; | |
93 | fd = -1; | |
94 | return r; | |
95 | } | |
96 | ||
97 | if (errno != ENOENT) | |
98 | return sd_bus_error_set_errnof(error, errno, "Failed to open /var/lib/machines.raw: %m"); | |
99 | ||
14bcf25c | 100 | r = tempfn_xxxxxx("/var/lib/machines.raw", NULL, &tmp); |
432cea00 LP |
101 | if (r < 0) |
102 | return r; | |
103 | ||
104 | (void) mkdir_p_label("/var/lib", 0755); | |
105 | fd = open(tmp, O_RDWR|O_CREAT|O_EXCL|O_NOCTTY|O_CLOEXEC, 0600); | |
106 | if (fd < 0) | |
107 | return sd_bus_error_set_errnof(error, errno, "Failed to create /var/lib/machines.raw: %m"); | |
108 | ||
109 | if (fstatvfs(fd, &ss) < 0) { | |
110 | r = sd_bus_error_set_errnof(error, errno, "Failed to determine free space on /var/lib/machines.raw: %m"); | |
111 | goto fail; | |
112 | } | |
113 | ||
114 | if (ss.f_bsize * ss.f_bavail < VAR_LIB_MACHINES_FREE_MIN) { | |
115 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "Not enough free disk space to set up /var/lib/machines."); | |
116 | goto fail; | |
117 | } | |
118 | ||
4cee5eed | 119 | if (ftruncate(fd, size) < 0) { |
432cea00 LP |
120 | r = sd_bus_error_set_errnof(error, errno, "Failed to enlarge /var/lib/machines.raw: %m"); |
121 | goto fail; | |
122 | } | |
123 | ||
124 | pid = fork(); | |
125 | if (pid < 0) { | |
126 | r = sd_bus_error_set_errnof(error, errno, "Failed to fork mkfs.btrfs: %m"); | |
127 | goto fail; | |
128 | } | |
129 | ||
130 | if (pid == 0) { | |
131 | ||
132 | /* Child */ | |
133 | ||
ce30c8dc LP |
134 | (void) reset_all_signal_handlers(); |
135 | (void) reset_signal_mask(); | |
432cea00 LP |
136 | assert_se(prctl(PR_SET_PDEATHSIG, SIGTERM) == 0); |
137 | ||
138 | fd = safe_close(fd); | |
139 | ||
140 | execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp, NULL); | |
141 | if (errno == ENOENT) | |
142 | return 99; | |
143 | ||
144 | _exit(EXIT_FAILURE); | |
145 | } | |
146 | ||
147 | r = wait_for_terminate(pid, &si); | |
148 | if (r < 0) { | |
149 | sd_bus_error_set_errnof(error, r, "Failed to wait for mkfs.btrfs: %m"); | |
150 | goto fail; | |
151 | } | |
152 | ||
153 | pid = 0; | |
154 | ||
155 | if (si.si_code != CLD_EXITED) { | |
156 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs died abnormally."); | |
157 | goto fail; | |
158 | } | |
159 | if (si.si_status == 99) { | |
160 | r = sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing"); | |
161 | goto fail; | |
162 | } | |
163 | if (si.si_status != 0) { | |
164 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs failed with error code %i", si.si_status); | |
165 | goto fail; | |
166 | } | |
167 | ||
f85ef957 AC |
168 | r = rename_noreplace(AT_FDCWD, tmp, AT_FDCWD, "/var/lib/machines.raw"); |
169 | if (r < 0) { | |
170 | sd_bus_error_set_errnof(error, r, "Failed to move /var/lib/machines.raw into place: %m"); | |
432cea00 LP |
171 | goto fail; |
172 | } | |
173 | ||
174 | r = fd; | |
175 | fd = -1; | |
176 | ||
177 | return r; | |
178 | ||
179 | fail: | |
132764a2 | 180 | unlink_noerrno(tmp); |
432cea00 LP |
181 | |
182 | if (pid > 1) | |
183 | kill_and_sigcont(pid, SIGKILL); | |
184 | ||
185 | return r; | |
186 | } | |
187 | ||
4cee5eed | 188 | int setup_machine_directory(uint64_t size, sd_bus_error *error) { |
403e5b32 | 189 | _cleanup_release_lock_file_ LockFile lock_file = LOCK_FILE_INIT; |
432cea00 LP |
190 | struct loop_info64 info = { |
191 | .lo_flags = LO_FLAGS_AUTOCLEAR, | |
192 | }; | |
193 | _cleanup_close_ int fd = -1, control = -1, loop = -1; | |
194 | _cleanup_free_ char* loopdev = NULL; | |
5bcd08db | 195 | char tmpdir[] = "/tmp/machine-pool.XXXXXX", *mntdir = NULL; |
432cea00 | 196 | bool tmpdir_made = false, mntdir_made = false, mntdir_mounted = false; |
26166c88 | 197 | char buf[FORMAT_BYTES_MAX]; |
432cea00 LP |
198 | int r, nr = -1; |
199 | ||
4cee5eed LP |
200 | /* btrfs cannot handle file systems < 16M, hence use this as minimum */ |
201 | if (size == (uint64_t) -1) | |
202 | size = VAR_LIB_MACHINES_SIZE_START; | |
203 | else if (size < 16*1024*1024) | |
204 | size = 16*1024*1024; | |
205 | ||
403e5b32 LP |
206 | /* Make sure we only set the directory up once at a time */ |
207 | r = make_lock_file("/run/systemd/machines.lock", LOCK_EX, &lock_file); | |
208 | if (r < 0) | |
209 | return r; | |
210 | ||
432cea00 LP |
211 | r = check_btrfs(); |
212 | if (r < 0) | |
213 | return sd_bus_error_set_errnof(error, r, "Failed to determine whether /var/lib/machines is located on btrfs: %m"); | |
214 | if (r > 0) { | |
215 | (void) btrfs_subvol_make_label("/var/lib/machines"); | |
216 | ||
217 | r = btrfs_quota_enable("/var/lib/machines", true); | |
218 | if (r < 0) | |
5bcd08db | 219 | log_warning_errno(r, "Failed to enable quota for /var/lib/machines, ignoring: %m"); |
432cea00 | 220 | |
5bcd08db LP |
221 | r = btrfs_subvol_auto_qgroup("/var/lib/machines", 0, true); |
222 | if (r < 0) | |
223 | log_warning_errno(r, "Failed to set up default quota hierarchy for /var/lib/machines, ignoring: %m"); | |
224 | ||
225 | return 1; | |
226 | } | |
227 | ||
228 | if (path_is_mount_point("/var/lib/machines", AT_SYMLINK_FOLLOW) > 0) { | |
229 | log_debug("/var/lib/machines is already a mount point, not creating loopback file for it."); | |
432cea00 LP |
230 | return 0; |
231 | } | |
232 | ||
5bcd08db LP |
233 | r = dir_is_populated("/var/lib/machines"); |
234 | if (r < 0 && r != -ENOENT) | |
235 | return r; | |
236 | if (r > 0) { | |
237 | log_debug("/var/log/machines is already populated, not creating loopback file for it."); | |
238 | return 0; | |
239 | } | |
240 | ||
241 | r = mkfs_exists("btrfs"); | |
c3b0e5ac EV |
242 | if (r == 0) |
243 | return sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing"); | |
5bcd08db LP |
244 | if (r < 0) |
245 | return r; | |
432cea00 | 246 | |
4cee5eed | 247 | fd = setup_machine_raw(size, error); |
432cea00 LP |
248 | if (fd < 0) |
249 | return fd; | |
250 | ||
251 | control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK); | |
252 | if (control < 0) | |
253 | return sd_bus_error_set_errnof(error, errno, "Failed to open /dev/loop-control: %m"); | |
254 | ||
255 | nr = ioctl(control, LOOP_CTL_GET_FREE); | |
256 | if (nr < 0) | |
257 | return sd_bus_error_set_errnof(error, errno, "Failed to allocate loop device: %m"); | |
258 | ||
259 | if (asprintf(&loopdev, "/dev/loop%i", nr) < 0) { | |
260 | r = -ENOMEM; | |
261 | goto fail; | |
262 | } | |
263 | ||
264 | loop = open(loopdev, O_CLOEXEC|O_RDWR|O_NOCTTY|O_NONBLOCK); | |
265 | if (loop < 0) { | |
266 | r = sd_bus_error_set_errnof(error, errno, "Failed to open loopback device: %m"); | |
267 | goto fail; | |
268 | } | |
269 | ||
270 | if (ioctl(loop, LOOP_SET_FD, fd) < 0) { | |
271 | r = sd_bus_error_set_errnof(error, errno, "Failed to bind loopback device: %m"); | |
272 | goto fail; | |
273 | } | |
274 | ||
275 | if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0) { | |
276 | r = sd_bus_error_set_errnof(error, errno, "Failed to enable auto-clear for loopback device: %m"); | |
277 | goto fail; | |
278 | } | |
279 | ||
280 | /* We need to make sure the new /var/lib/machines directory | |
281 | * has an access mode of 0700 at the time it is first made | |
282 | * available. mkfs will create it with 0755 however. Hence, | |
283 | * let's mount the directory into an inaccessible directory | |
284 | * below /tmp first, fix the access mode, and move it to the | |
285 | * public place then. */ | |
286 | ||
287 | if (!mkdtemp(tmpdir)) { | |
288 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount parent directory: %m"); | |
289 | goto fail; | |
290 | } | |
291 | tmpdir_made = true; | |
292 | ||
293 | mntdir = strjoina(tmpdir, "/mnt"); | |
294 | if (mkdir(mntdir, 0700) < 0) { | |
295 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount directory: %m"); | |
296 | goto fail; | |
297 | } | |
298 | mntdir_made = true; | |
299 | ||
300 | if (mount(loopdev, mntdir, "btrfs", 0, NULL) < 0) { | |
301 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount loopback device: %m"); | |
302 | goto fail; | |
303 | } | |
304 | mntdir_mounted = true; | |
305 | ||
306 | r = btrfs_quota_enable(mntdir, true); | |
307 | if (r < 0) | |
308 | log_warning_errno(r, "Failed to enable quota, ignoring: %m"); | |
309 | ||
5bcd08db LP |
310 | r = btrfs_subvol_auto_qgroup(mntdir, 0, true); |
311 | if (r < 0) | |
312 | log_warning_errno(r, "Failed to set up default quota hierarchy, ignoring: %m"); | |
313 | ||
432cea00 LP |
314 | if (chmod(mntdir, 0700) < 0) { |
315 | r = sd_bus_error_set_errnof(error, errno, "Failed to fix owner: %m"); | |
316 | goto fail; | |
317 | } | |
318 | ||
319 | (void) mkdir_p_label("/var/lib/machines", 0700); | |
320 | ||
321 | if (mount(mntdir, "/var/lib/machines", NULL, MS_BIND, NULL) < 0) { | |
322 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount directory into right place: %m"); | |
323 | goto fail; | |
324 | } | |
325 | ||
26166c88 LP |
326 | (void) syncfs(fd); |
327 | ||
328 | 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)); | |
329 | ||
432cea00 LP |
330 | (void) umount2(mntdir, MNT_DETACH); |
331 | (void) rmdir(mntdir); | |
332 | (void) rmdir(tmpdir); | |
333 | ||
5bcd08db | 334 | return 1; |
432cea00 LP |
335 | |
336 | fail: | |
337 | if (mntdir_mounted) | |
338 | (void) umount2(mntdir, MNT_DETACH); | |
339 | ||
340 | if (mntdir_made) | |
341 | (void) rmdir(mntdir); | |
342 | if (tmpdir_made) | |
343 | (void) rmdir(tmpdir); | |
344 | ||
345 | if (loop >= 0) { | |
346 | (void) ioctl(loop, LOOP_CLR_FD); | |
347 | loop = safe_close(loop); | |
348 | } | |
349 | ||
350 | if (control >= 0 && nr >= 0) | |
351 | (void) ioctl(control, LOOP_CTL_REMOVE, nr); | |
352 | ||
353 | return r; | |
354 | } | |
26166c88 LP |
355 | |
356 | static int sync_path(const char *p) { | |
357 | _cleanup_close_ int fd = -1; | |
358 | ||
359 | fd = open(p, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
360 | if (fd < 0) | |
361 | return -errno; | |
362 | ||
363 | if (syncfs(fd) < 0) | |
364 | return -errno; | |
365 | ||
366 | return 0; | |
367 | } | |
368 | ||
369 | int grow_machine_directory(void) { | |
370 | char buf[FORMAT_BYTES_MAX]; | |
371 | struct statvfs a, b; | |
372 | uint64_t old_size, new_size, max_add; | |
373 | int r; | |
374 | ||
375 | /* Ensure the disk space data is accurate */ | |
376 | sync_path("/var/lib/machines"); | |
377 | sync_path("/var/lib/machines.raw"); | |
378 | ||
379 | if (statvfs("/var/lib/machines.raw", &a) < 0) | |
380 | return -errno; | |
381 | ||
382 | if (statvfs("/var/lib/machines", &b) < 0) | |
383 | return -errno; | |
384 | ||
385 | /* Don't grow if not enough disk space is available on the host */ | |
386 | if (((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) <= VAR_LIB_MACHINES_FREE_MIN) | |
387 | return 0; | |
388 | ||
389 | /* Don't grow if at least 1/3th of the fs is still free */ | |
390 | if (b.f_bavail > b.f_blocks / 3) | |
391 | return 0; | |
392 | ||
a8eaaee7 | 393 | /* Calculate how much we are willing to add at most */ |
26166c88 LP |
394 | max_add = ((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) - VAR_LIB_MACHINES_FREE_MIN; |
395 | ||
396 | /* Calculate the old size */ | |
397 | old_size = (uint64_t) b.f_blocks * (uint64_t) b.f_bsize; | |
398 | ||
399 | /* Calculate the new size as three times the size of what is used right now */ | |
400 | new_size = ((uint64_t) b.f_blocks - (uint64_t) b.f_bavail) * (uint64_t) b.f_bsize * 3; | |
401 | ||
402 | /* Always, grow at least to the start size */ | |
403 | if (new_size < VAR_LIB_MACHINES_SIZE_START) | |
404 | new_size = VAR_LIB_MACHINES_SIZE_START; | |
405 | ||
406 | /* If the new size is smaller than the old size, don't grow */ | |
407 | if (new_size < old_size) | |
408 | return 0; | |
409 | ||
410 | /* Ensure we never add more than the maximum */ | |
411 | if (new_size > old_size + max_add) | |
412 | new_size = old_size + max_add; | |
413 | ||
414 | r = btrfs_resize_loopback("/var/lib/machines", new_size, true); | |
415 | if (r <= 0) | |
416 | return r; | |
417 | ||
5bcd08db LP |
418 | /* Also bump the quota, of both the subvolume leaf qgroup, as |
419 | * well as of any subtree quota group by the same id but a | |
420 | * higher level, if it exists. */ | |
421 | (void) btrfs_qgroup_set_limit("/var/lib/machines", 0, new_size); | |
422 | (void) btrfs_subvol_set_subtree_quota_limit("/var/lib/machines", 0, new_size); | |
26166c88 LP |
423 | |
424 | log_info("Grew /var/lib/machines btrfs loopback file system to %s.", format_bytes(buf, sizeof(buf), new_size)); | |
425 | return 1; | |
426 | } |