]>
Commit | Line | Data |
---|---|---|
432cea00 LP |
1 | /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ |
2 | ||
3 | /*** | |
4 | This file is part of systemd. | |
5 | ||
6 | Copyright 2015 Lennart Poettering | |
7 | ||
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. | |
12 | ||
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. | |
17 | ||
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/>. | |
20 | ***/ | |
21 | ||
07630cea | 22 | #include <sys/mount.h> |
432cea00 | 23 | #include <sys/prctl.h> |
432cea00 | 24 | #include <sys/statvfs.h> |
07630cea | 25 | #include <sys/vfs.h> |
432cea00 | 26 | |
07630cea | 27 | #include "btrfs-util.h" |
3ffd4af2 | 28 | #include "fd-util.h" |
0d39fa9c | 29 | #include "fileio.h" |
cd2eb9e9 | 30 | #include "lockfile-util.h" |
3ffd4af2 | 31 | #include "machine-pool.h" |
432cea00 | 32 | #include "mkdir.h" |
6bedfcbb | 33 | #include "parse-util.h" |
432cea00 | 34 | #include "path-util.h" |
07630cea | 35 | #include "process-util.h" |
24882e06 | 36 | #include "signal-util.h" |
07630cea LP |
37 | #include "string-util.h" |
38 | #include "util.h" | |
4349cd7c | 39 | #include "mount-util.h" |
432cea00 LP |
40 | |
41 | #define VAR_LIB_MACHINES_SIZE_START (1024UL*1024UL*500UL) | |
42 | #define VAR_LIB_MACHINES_FREE_MIN (1024UL*1024UL*750UL) | |
43 | ||
44 | static int check_btrfs(void) { | |
45 | struct statfs sfs; | |
46 | ||
47 | if (statfs("/var/lib/machines", &sfs) < 0) { | |
48 | if (errno != ENOENT) | |
49 | return -errno; | |
50 | ||
51 | if (statfs("/var/lib", &sfs) < 0) | |
52 | return -errno; | |
53 | } | |
54 | ||
55 | return F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC); | |
56 | } | |
57 | ||
4cee5eed | 58 | static int setup_machine_raw(uint64_t size, sd_bus_error *error) { |
432cea00 LP |
59 | _cleanup_free_ char *tmp = NULL; |
60 | _cleanup_close_ int fd = -1; | |
61 | struct statvfs ss; | |
62 | pid_t pid = 0; | |
63 | siginfo_t si; | |
64 | int r; | |
65 | ||
66 | /* We want to be able to make use of btrfs-specific file | |
67 | * system features, in particular subvolumes, reflinks and | |
68 | * quota. Hence, if we detect that /var/lib/machines.raw is | |
69 | * not located on btrfs, let's create a loopback file, place a | |
70 | * btrfs file system into it, and mount it to | |
71 | * /var/lib/machines. */ | |
72 | ||
73 | fd = open("/var/lib/machines.raw", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
74 | if (fd >= 0) { | |
75 | r = fd; | |
76 | fd = -1; | |
77 | return r; | |
78 | } | |
79 | ||
80 | if (errno != ENOENT) | |
81 | return sd_bus_error_set_errnof(error, errno, "Failed to open /var/lib/machines.raw: %m"); | |
82 | ||
14bcf25c | 83 | r = tempfn_xxxxxx("/var/lib/machines.raw", NULL, &tmp); |
432cea00 LP |
84 | if (r < 0) |
85 | return r; | |
86 | ||
87 | (void) mkdir_p_label("/var/lib", 0755); | |
88 | fd = open(tmp, O_RDWR|O_CREAT|O_EXCL|O_NOCTTY|O_CLOEXEC, 0600); | |
89 | if (fd < 0) | |
90 | return sd_bus_error_set_errnof(error, errno, "Failed to create /var/lib/machines.raw: %m"); | |
91 | ||
92 | if (fstatvfs(fd, &ss) < 0) { | |
93 | r = sd_bus_error_set_errnof(error, errno, "Failed to determine free space on /var/lib/machines.raw: %m"); | |
94 | goto fail; | |
95 | } | |
96 | ||
97 | if (ss.f_bsize * ss.f_bavail < VAR_LIB_MACHINES_FREE_MIN) { | |
98 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "Not enough free disk space to set up /var/lib/machines."); | |
99 | goto fail; | |
100 | } | |
101 | ||
4cee5eed | 102 | if (ftruncate(fd, size) < 0) { |
432cea00 LP |
103 | r = sd_bus_error_set_errnof(error, errno, "Failed to enlarge /var/lib/machines.raw: %m"); |
104 | goto fail; | |
105 | } | |
106 | ||
107 | pid = fork(); | |
108 | if (pid < 0) { | |
109 | r = sd_bus_error_set_errnof(error, errno, "Failed to fork mkfs.btrfs: %m"); | |
110 | goto fail; | |
111 | } | |
112 | ||
113 | if (pid == 0) { | |
114 | ||
115 | /* Child */ | |
116 | ||
ce30c8dc LP |
117 | (void) reset_all_signal_handlers(); |
118 | (void) reset_signal_mask(); | |
432cea00 LP |
119 | assert_se(prctl(PR_SET_PDEATHSIG, SIGTERM) == 0); |
120 | ||
121 | fd = safe_close(fd); | |
122 | ||
123 | execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp, NULL); | |
124 | if (errno == ENOENT) | |
125 | return 99; | |
126 | ||
127 | _exit(EXIT_FAILURE); | |
128 | } | |
129 | ||
130 | r = wait_for_terminate(pid, &si); | |
131 | if (r < 0) { | |
132 | sd_bus_error_set_errnof(error, r, "Failed to wait for mkfs.btrfs: %m"); | |
133 | goto fail; | |
134 | } | |
135 | ||
136 | pid = 0; | |
137 | ||
138 | if (si.si_code != CLD_EXITED) { | |
139 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs died abnormally."); | |
140 | goto fail; | |
141 | } | |
142 | if (si.si_status == 99) { | |
143 | r = sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing"); | |
144 | goto fail; | |
145 | } | |
146 | if (si.si_status != 0) { | |
147 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs failed with error code %i", si.si_status); | |
148 | goto fail; | |
149 | } | |
150 | ||
f85ef957 AC |
151 | r = rename_noreplace(AT_FDCWD, tmp, AT_FDCWD, "/var/lib/machines.raw"); |
152 | if (r < 0) { | |
153 | sd_bus_error_set_errnof(error, r, "Failed to move /var/lib/machines.raw into place: %m"); | |
432cea00 LP |
154 | goto fail; |
155 | } | |
156 | ||
157 | r = fd; | |
158 | fd = -1; | |
159 | ||
160 | return r; | |
161 | ||
162 | fail: | |
132764a2 | 163 | unlink_noerrno(tmp); |
432cea00 LP |
164 | |
165 | if (pid > 1) | |
166 | kill_and_sigcont(pid, SIGKILL); | |
167 | ||
168 | return r; | |
169 | } | |
170 | ||
4cee5eed | 171 | int setup_machine_directory(uint64_t size, sd_bus_error *error) { |
403e5b32 | 172 | _cleanup_release_lock_file_ LockFile lock_file = LOCK_FILE_INIT; |
432cea00 LP |
173 | struct loop_info64 info = { |
174 | .lo_flags = LO_FLAGS_AUTOCLEAR, | |
175 | }; | |
176 | _cleanup_close_ int fd = -1, control = -1, loop = -1; | |
177 | _cleanup_free_ char* loopdev = NULL; | |
5bcd08db | 178 | char tmpdir[] = "/tmp/machine-pool.XXXXXX", *mntdir = NULL; |
432cea00 | 179 | bool tmpdir_made = false, mntdir_made = false, mntdir_mounted = false; |
26166c88 | 180 | char buf[FORMAT_BYTES_MAX]; |
432cea00 LP |
181 | int r, nr = -1; |
182 | ||
4cee5eed LP |
183 | /* btrfs cannot handle file systems < 16M, hence use this as minimum */ |
184 | if (size == (uint64_t) -1) | |
185 | size = VAR_LIB_MACHINES_SIZE_START; | |
186 | else if (size < 16*1024*1024) | |
187 | size = 16*1024*1024; | |
188 | ||
403e5b32 LP |
189 | /* Make sure we only set the directory up once at a time */ |
190 | r = make_lock_file("/run/systemd/machines.lock", LOCK_EX, &lock_file); | |
191 | if (r < 0) | |
192 | return r; | |
193 | ||
432cea00 LP |
194 | r = check_btrfs(); |
195 | if (r < 0) | |
196 | return sd_bus_error_set_errnof(error, r, "Failed to determine whether /var/lib/machines is located on btrfs: %m"); | |
197 | if (r > 0) { | |
198 | (void) btrfs_subvol_make_label("/var/lib/machines"); | |
199 | ||
200 | r = btrfs_quota_enable("/var/lib/machines", true); | |
201 | if (r < 0) | |
5bcd08db | 202 | log_warning_errno(r, "Failed to enable quota for /var/lib/machines, ignoring: %m"); |
432cea00 | 203 | |
5bcd08db LP |
204 | r = btrfs_subvol_auto_qgroup("/var/lib/machines", 0, true); |
205 | if (r < 0) | |
206 | log_warning_errno(r, "Failed to set up default quota hierarchy for /var/lib/machines, ignoring: %m"); | |
207 | ||
208 | return 1; | |
209 | } | |
210 | ||
211 | if (path_is_mount_point("/var/lib/machines", AT_SYMLINK_FOLLOW) > 0) { | |
212 | log_debug("/var/lib/machines is already a mount point, not creating loopback file for it."); | |
432cea00 LP |
213 | return 0; |
214 | } | |
215 | ||
5bcd08db LP |
216 | r = dir_is_populated("/var/lib/machines"); |
217 | if (r < 0 && r != -ENOENT) | |
218 | return r; | |
219 | if (r > 0) { | |
220 | log_debug("/var/log/machines is already populated, not creating loopback file for it."); | |
221 | return 0; | |
222 | } | |
223 | ||
224 | r = mkfs_exists("btrfs"); | |
225 | if (r == -ENOENT) { | |
226 | log_debug("mkfs.btrfs is missing, cannot create loopback file for /var/lib/machines."); | |
227 | return 0; | |
228 | } | |
229 | if (r < 0) | |
230 | return r; | |
432cea00 | 231 | |
4cee5eed | 232 | fd = setup_machine_raw(size, error); |
432cea00 LP |
233 | if (fd < 0) |
234 | return fd; | |
235 | ||
236 | control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK); | |
237 | if (control < 0) | |
238 | return sd_bus_error_set_errnof(error, errno, "Failed to open /dev/loop-control: %m"); | |
239 | ||
240 | nr = ioctl(control, LOOP_CTL_GET_FREE); | |
241 | if (nr < 0) | |
242 | return sd_bus_error_set_errnof(error, errno, "Failed to allocate loop device: %m"); | |
243 | ||
244 | if (asprintf(&loopdev, "/dev/loop%i", nr) < 0) { | |
245 | r = -ENOMEM; | |
246 | goto fail; | |
247 | } | |
248 | ||
249 | loop = open(loopdev, O_CLOEXEC|O_RDWR|O_NOCTTY|O_NONBLOCK); | |
250 | if (loop < 0) { | |
251 | r = sd_bus_error_set_errnof(error, errno, "Failed to open loopback device: %m"); | |
252 | goto fail; | |
253 | } | |
254 | ||
255 | if (ioctl(loop, LOOP_SET_FD, fd) < 0) { | |
256 | r = sd_bus_error_set_errnof(error, errno, "Failed to bind loopback device: %m"); | |
257 | goto fail; | |
258 | } | |
259 | ||
260 | if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0) { | |
261 | r = sd_bus_error_set_errnof(error, errno, "Failed to enable auto-clear for loopback device: %m"); | |
262 | goto fail; | |
263 | } | |
264 | ||
265 | /* We need to make sure the new /var/lib/machines directory | |
266 | * has an access mode of 0700 at the time it is first made | |
267 | * available. mkfs will create it with 0755 however. Hence, | |
268 | * let's mount the directory into an inaccessible directory | |
269 | * below /tmp first, fix the access mode, and move it to the | |
270 | * public place then. */ | |
271 | ||
272 | if (!mkdtemp(tmpdir)) { | |
273 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount parent directory: %m"); | |
274 | goto fail; | |
275 | } | |
276 | tmpdir_made = true; | |
277 | ||
278 | mntdir = strjoina(tmpdir, "/mnt"); | |
279 | if (mkdir(mntdir, 0700) < 0) { | |
280 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount directory: %m"); | |
281 | goto fail; | |
282 | } | |
283 | mntdir_made = true; | |
284 | ||
285 | if (mount(loopdev, mntdir, "btrfs", 0, NULL) < 0) { | |
286 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount loopback device: %m"); | |
287 | goto fail; | |
288 | } | |
289 | mntdir_mounted = true; | |
290 | ||
291 | r = btrfs_quota_enable(mntdir, true); | |
292 | if (r < 0) | |
293 | log_warning_errno(r, "Failed to enable quota, ignoring: %m"); | |
294 | ||
5bcd08db LP |
295 | r = btrfs_subvol_auto_qgroup(mntdir, 0, true); |
296 | if (r < 0) | |
297 | log_warning_errno(r, "Failed to set up default quota hierarchy, ignoring: %m"); | |
298 | ||
432cea00 LP |
299 | if (chmod(mntdir, 0700) < 0) { |
300 | r = sd_bus_error_set_errnof(error, errno, "Failed to fix owner: %m"); | |
301 | goto fail; | |
302 | } | |
303 | ||
304 | (void) mkdir_p_label("/var/lib/machines", 0700); | |
305 | ||
306 | if (mount(mntdir, "/var/lib/machines", NULL, MS_BIND, NULL) < 0) { | |
307 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount directory into right place: %m"); | |
308 | goto fail; | |
309 | } | |
310 | ||
26166c88 LP |
311 | (void) syncfs(fd); |
312 | ||
313 | 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)); | |
314 | ||
432cea00 LP |
315 | (void) umount2(mntdir, MNT_DETACH); |
316 | (void) rmdir(mntdir); | |
317 | (void) rmdir(tmpdir); | |
318 | ||
5bcd08db | 319 | return 1; |
432cea00 LP |
320 | |
321 | fail: | |
322 | if (mntdir_mounted) | |
323 | (void) umount2(mntdir, MNT_DETACH); | |
324 | ||
325 | if (mntdir_made) | |
326 | (void) rmdir(mntdir); | |
327 | if (tmpdir_made) | |
328 | (void) rmdir(tmpdir); | |
329 | ||
330 | if (loop >= 0) { | |
331 | (void) ioctl(loop, LOOP_CLR_FD); | |
332 | loop = safe_close(loop); | |
333 | } | |
334 | ||
335 | if (control >= 0 && nr >= 0) | |
336 | (void) ioctl(control, LOOP_CTL_REMOVE, nr); | |
337 | ||
338 | return r; | |
339 | } | |
26166c88 LP |
340 | |
341 | static int sync_path(const char *p) { | |
342 | _cleanup_close_ int fd = -1; | |
343 | ||
344 | fd = open(p, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
345 | if (fd < 0) | |
346 | return -errno; | |
347 | ||
348 | if (syncfs(fd) < 0) | |
349 | return -errno; | |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
354 | int grow_machine_directory(void) { | |
355 | char buf[FORMAT_BYTES_MAX]; | |
356 | struct statvfs a, b; | |
357 | uint64_t old_size, new_size, max_add; | |
358 | int r; | |
359 | ||
360 | /* Ensure the disk space data is accurate */ | |
361 | sync_path("/var/lib/machines"); | |
362 | sync_path("/var/lib/machines.raw"); | |
363 | ||
364 | if (statvfs("/var/lib/machines.raw", &a) < 0) | |
365 | return -errno; | |
366 | ||
367 | if (statvfs("/var/lib/machines", &b) < 0) | |
368 | return -errno; | |
369 | ||
370 | /* Don't grow if not enough disk space is available on the host */ | |
371 | if (((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) <= VAR_LIB_MACHINES_FREE_MIN) | |
372 | return 0; | |
373 | ||
374 | /* Don't grow if at least 1/3th of the fs is still free */ | |
375 | if (b.f_bavail > b.f_blocks / 3) | |
376 | return 0; | |
377 | ||
378 | /* Calculate how much we are willing to add at maximum */ | |
379 | max_add = ((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) - VAR_LIB_MACHINES_FREE_MIN; | |
380 | ||
381 | /* Calculate the old size */ | |
382 | old_size = (uint64_t) b.f_blocks * (uint64_t) b.f_bsize; | |
383 | ||
384 | /* Calculate the new size as three times the size of what is used right now */ | |
385 | new_size = ((uint64_t) b.f_blocks - (uint64_t) b.f_bavail) * (uint64_t) b.f_bsize * 3; | |
386 | ||
387 | /* Always, grow at least to the start size */ | |
388 | if (new_size < VAR_LIB_MACHINES_SIZE_START) | |
389 | new_size = VAR_LIB_MACHINES_SIZE_START; | |
390 | ||
391 | /* If the new size is smaller than the old size, don't grow */ | |
392 | if (new_size < old_size) | |
393 | return 0; | |
394 | ||
395 | /* Ensure we never add more than the maximum */ | |
396 | if (new_size > old_size + max_add) | |
397 | new_size = old_size + max_add; | |
398 | ||
399 | r = btrfs_resize_loopback("/var/lib/machines", new_size, true); | |
400 | if (r <= 0) | |
401 | return r; | |
402 | ||
5bcd08db LP |
403 | /* Also bump the quota, of both the subvolume leaf qgroup, as |
404 | * well as of any subtree quota group by the same id but a | |
405 | * higher level, if it exists. */ | |
406 | (void) btrfs_qgroup_set_limit("/var/lib/machines", 0, new_size); | |
407 | (void) btrfs_subvol_set_subtree_quota_limit("/var/lib/machines", 0, new_size); | |
26166c88 LP |
408 | |
409 | log_info("Grew /var/lib/machines btrfs loopback file system to %s.", format_bytes(buf, sizeof(buf), new_size)); | |
410 | return 1; | |
411 | } |