]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
432cea00 LP |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
5 | Copyright 2015 Lennart Poettering | |
432cea00 LP |
6 | ***/ |
7 | ||
a8fbdf54 TA |
8 | #include <errno.h> |
9 | #include <fcntl.h> | |
10 | #include <linux/loop.h> | |
11 | #include <signal.h> | |
12 | #include <stdbool.h> | |
13 | #include <stdio.h> | |
14 | #include <stdlib.h> | |
e306723e | 15 | #include <sys/file.h> |
a8fbdf54 | 16 | #include <sys/ioctl.h> |
07630cea | 17 | #include <sys/mount.h> |
432cea00 | 18 | #include <sys/prctl.h> |
a8fbdf54 TA |
19 | #include <sys/stat.h> |
20 | #include <sys/statfs.h> | |
432cea00 | 21 | #include <sys/statvfs.h> |
a8fbdf54 TA |
22 | #include <unistd.h> |
23 | ||
24 | #include "sd-bus-protocol.h" | |
25 | #include "sd-bus.h" | |
432cea00 | 26 | |
b5efdb8a | 27 | #include "alloc-util.h" |
07630cea | 28 | #include "btrfs-util.h" |
3ffd4af2 | 29 | #include "fd-util.h" |
0d39fa9c | 30 | #include "fileio.h" |
f4f15635 | 31 | #include "fs-util.h" |
de2e28d8 | 32 | #include "label.h" |
cd2eb9e9 | 33 | #include "lockfile-util.h" |
a8fbdf54 | 34 | #include "log.h" |
3ffd4af2 | 35 | #include "machine-pool.h" |
a8fbdf54 TA |
36 | #include "macro.h" |
37 | #include "missing.h" | |
432cea00 | 38 | #include "mkdir.h" |
f4f15635 | 39 | #include "mount-util.h" |
6bedfcbb | 40 | #include "parse-util.h" |
432cea00 | 41 | #include "path-util.h" |
07630cea | 42 | #include "process-util.h" |
24882e06 | 43 | #include "signal-util.h" |
8fcde012 | 44 | #include "stat-util.h" |
07630cea | 45 | #include "string-util.h" |
432cea00 LP |
46 | |
47 | #define VAR_LIB_MACHINES_SIZE_START (1024UL*1024UL*500UL) | |
48 | #define VAR_LIB_MACHINES_FREE_MIN (1024UL*1024UL*750UL) | |
49 | ||
50 | static int check_btrfs(void) { | |
51 | struct statfs sfs; | |
52 | ||
53 | if (statfs("/var/lib/machines", &sfs) < 0) { | |
54 | if (errno != ENOENT) | |
55 | return -errno; | |
56 | ||
57 | if (statfs("/var/lib", &sfs) < 0) | |
58 | return -errno; | |
59 | } | |
60 | ||
61 | return F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC); | |
62 | } | |
63 | ||
4cee5eed | 64 | static int setup_machine_raw(uint64_t size, sd_bus_error *error) { |
432cea00 LP |
65 | _cleanup_free_ char *tmp = NULL; |
66 | _cleanup_close_ int fd = -1; | |
67 | struct statvfs ss; | |
68 | pid_t pid = 0; | |
432cea00 LP |
69 | int r; |
70 | ||
71 | /* We want to be able to make use of btrfs-specific file | |
72 | * system features, in particular subvolumes, reflinks and | |
73 | * quota. Hence, if we detect that /var/lib/machines.raw is | |
74 | * not located on btrfs, let's create a loopback file, place a | |
75 | * btrfs file system into it, and mount it to | |
76 | * /var/lib/machines. */ | |
77 | ||
78 | fd = open("/var/lib/machines.raw", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
c10d6bdb LP |
79 | if (fd >= 0) |
80 | return TAKE_FD(fd); | |
432cea00 LP |
81 | |
82 | if (errno != ENOENT) | |
83 | return sd_bus_error_set_errnof(error, errno, "Failed to open /var/lib/machines.raw: %m"); | |
84 | ||
14bcf25c | 85 | r = tempfn_xxxxxx("/var/lib/machines.raw", NULL, &tmp); |
432cea00 LP |
86 | if (r < 0) |
87 | return r; | |
88 | ||
89 | (void) mkdir_p_label("/var/lib", 0755); | |
90 | fd = open(tmp, O_RDWR|O_CREAT|O_EXCL|O_NOCTTY|O_CLOEXEC, 0600); | |
91 | if (fd < 0) | |
92 | return sd_bus_error_set_errnof(error, errno, "Failed to create /var/lib/machines.raw: %m"); | |
93 | ||
94 | if (fstatvfs(fd, &ss) < 0) { | |
95 | r = sd_bus_error_set_errnof(error, errno, "Failed to determine free space on /var/lib/machines.raw: %m"); | |
96 | goto fail; | |
97 | } | |
98 | ||
99 | if (ss.f_bsize * ss.f_bavail < VAR_LIB_MACHINES_FREE_MIN) { | |
100 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "Not enough free disk space to set up /var/lib/machines."); | |
101 | goto fail; | |
102 | } | |
103 | ||
4cee5eed | 104 | if (ftruncate(fd, size) < 0) { |
432cea00 LP |
105 | r = sd_bus_error_set_errnof(error, errno, "Failed to enlarge /var/lib/machines.raw: %m"); |
106 | goto fail; | |
107 | } | |
108 | ||
4c253ed1 LP |
109 | r = safe_fork("(mkfs)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &pid); |
110 | if (r < 0) { | |
111 | sd_bus_error_set_errnof(error, r, "Failed to fork mkfs.btrfs: %m"); | |
432cea00 LP |
112 | goto fail; |
113 | } | |
4c253ed1 | 114 | if (r == 0) { |
432cea00 LP |
115 | |
116 | /* Child */ | |
117 | ||
432cea00 LP |
118 | fd = safe_close(fd); |
119 | ||
120 | execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp, NULL); | |
121 | if (errno == ENOENT) | |
579afbea | 122 | _exit(99); |
432cea00 LP |
123 | |
124 | _exit(EXIT_FAILURE); | |
125 | } | |
126 | ||
2e87a1fd | 127 | r = wait_for_terminate_and_check("mkfs", pid, 0); |
432cea00 LP |
128 | pid = 0; |
129 | ||
2e87a1fd LP |
130 | if (r < 0) { |
131 | sd_bus_error_set_errnof(error, r, "Failed to wait for mkfs.btrfs: %m"); | |
432cea00 LP |
132 | goto fail; |
133 | } | |
2e87a1fd | 134 | if (r == 99) { |
432cea00 LP |
135 | r = sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing"); |
136 | goto fail; | |
137 | } | |
2e87a1fd LP |
138 | if (r != EXIT_SUCCESS) { |
139 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs failed with error code %i", r); | |
432cea00 LP |
140 | goto fail; |
141 | } | |
142 | ||
f85ef957 AC |
143 | r = rename_noreplace(AT_FDCWD, tmp, AT_FDCWD, "/var/lib/machines.raw"); |
144 | if (r < 0) { | |
145 | sd_bus_error_set_errnof(error, r, "Failed to move /var/lib/machines.raw into place: %m"); | |
432cea00 LP |
146 | goto fail; |
147 | } | |
148 | ||
c10d6bdb | 149 | return TAKE_FD(fd); |
432cea00 LP |
150 | |
151 | fail: | |
132764a2 | 152 | unlink_noerrno(tmp); |
432cea00 LP |
153 | |
154 | if (pid > 1) | |
155 | kill_and_sigcont(pid, SIGKILL); | |
156 | ||
157 | return r; | |
158 | } | |
159 | ||
4cee5eed | 160 | int setup_machine_directory(uint64_t size, sd_bus_error *error) { |
8e766630 | 161 | _cleanup_(release_lock_file) LockFile lock_file = LOCK_FILE_INIT; |
432cea00 LP |
162 | struct loop_info64 info = { |
163 | .lo_flags = LO_FLAGS_AUTOCLEAR, | |
164 | }; | |
165 | _cleanup_close_ int fd = -1, control = -1, loop = -1; | |
166 | _cleanup_free_ char* loopdev = NULL; | |
5bcd08db | 167 | char tmpdir[] = "/tmp/machine-pool.XXXXXX", *mntdir = NULL; |
432cea00 | 168 | bool tmpdir_made = false, mntdir_made = false, mntdir_mounted = false; |
26166c88 | 169 | char buf[FORMAT_BYTES_MAX]; |
432cea00 LP |
170 | int r, nr = -1; |
171 | ||
4cee5eed LP |
172 | /* btrfs cannot handle file systems < 16M, hence use this as minimum */ |
173 | if (size == (uint64_t) -1) | |
174 | size = VAR_LIB_MACHINES_SIZE_START; | |
175 | else if (size < 16*1024*1024) | |
176 | size = 16*1024*1024; | |
177 | ||
403e5b32 LP |
178 | /* Make sure we only set the directory up once at a time */ |
179 | r = make_lock_file("/run/systemd/machines.lock", LOCK_EX, &lock_file); | |
180 | if (r < 0) | |
181 | return r; | |
182 | ||
432cea00 LP |
183 | r = check_btrfs(); |
184 | if (r < 0) | |
185 | return sd_bus_error_set_errnof(error, r, "Failed to determine whether /var/lib/machines is located on btrfs: %m"); | |
186 | if (r > 0) { | |
187 | (void) btrfs_subvol_make_label("/var/lib/machines"); | |
188 | ||
189 | r = btrfs_quota_enable("/var/lib/machines", true); | |
190 | if (r < 0) | |
5bcd08db | 191 | log_warning_errno(r, "Failed to enable quota for /var/lib/machines, ignoring: %m"); |
432cea00 | 192 | |
5bcd08db LP |
193 | r = btrfs_subvol_auto_qgroup("/var/lib/machines", 0, true); |
194 | if (r < 0) | |
195 | log_warning_errno(r, "Failed to set up default quota hierarchy for /var/lib/machines, ignoring: %m"); | |
196 | ||
197 | return 1; | |
198 | } | |
199 | ||
e1873695 | 200 | if (path_is_mount_point("/var/lib/machines", NULL, AT_SYMLINK_FOLLOW) > 0) { |
5bcd08db | 201 | log_debug("/var/lib/machines is already a mount point, not creating loopback file for it."); |
432cea00 LP |
202 | return 0; |
203 | } | |
204 | ||
5bcd08db LP |
205 | r = dir_is_populated("/var/lib/machines"); |
206 | if (r < 0 && r != -ENOENT) | |
207 | return r; | |
208 | if (r > 0) { | |
209 | log_debug("/var/log/machines is already populated, not creating loopback file for it."); | |
210 | return 0; | |
211 | } | |
212 | ||
213 | r = mkfs_exists("btrfs"); | |
c3b0e5ac EV |
214 | if (r == 0) |
215 | return sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing"); | |
5bcd08db LP |
216 | if (r < 0) |
217 | return r; | |
432cea00 | 218 | |
4cee5eed | 219 | fd = setup_machine_raw(size, error); |
432cea00 LP |
220 | if (fd < 0) |
221 | return fd; | |
222 | ||
223 | control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK); | |
224 | if (control < 0) | |
225 | return sd_bus_error_set_errnof(error, errno, "Failed to open /dev/loop-control: %m"); | |
226 | ||
227 | nr = ioctl(control, LOOP_CTL_GET_FREE); | |
228 | if (nr < 0) | |
229 | return sd_bus_error_set_errnof(error, errno, "Failed to allocate loop device: %m"); | |
230 | ||
231 | if (asprintf(&loopdev, "/dev/loop%i", nr) < 0) { | |
232 | r = -ENOMEM; | |
233 | goto fail; | |
234 | } | |
235 | ||
236 | loop = open(loopdev, O_CLOEXEC|O_RDWR|O_NOCTTY|O_NONBLOCK); | |
237 | if (loop < 0) { | |
238 | r = sd_bus_error_set_errnof(error, errno, "Failed to open loopback device: %m"); | |
239 | goto fail; | |
240 | } | |
241 | ||
242 | if (ioctl(loop, LOOP_SET_FD, fd) < 0) { | |
243 | r = sd_bus_error_set_errnof(error, errno, "Failed to bind loopback device: %m"); | |
244 | goto fail; | |
245 | } | |
246 | ||
247 | if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0) { | |
248 | r = sd_bus_error_set_errnof(error, errno, "Failed to enable auto-clear for loopback device: %m"); | |
249 | goto fail; | |
250 | } | |
251 | ||
252 | /* We need to make sure the new /var/lib/machines directory | |
253 | * has an access mode of 0700 at the time it is first made | |
254 | * available. mkfs will create it with 0755 however. Hence, | |
255 | * let's mount the directory into an inaccessible directory | |
256 | * below /tmp first, fix the access mode, and move it to the | |
257 | * public place then. */ | |
258 | ||
259 | if (!mkdtemp(tmpdir)) { | |
260 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount parent directory: %m"); | |
261 | goto fail; | |
262 | } | |
263 | tmpdir_made = true; | |
264 | ||
265 | mntdir = strjoina(tmpdir, "/mnt"); | |
266 | if (mkdir(mntdir, 0700) < 0) { | |
267 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount directory: %m"); | |
268 | goto fail; | |
269 | } | |
270 | mntdir_made = true; | |
271 | ||
272 | if (mount(loopdev, mntdir, "btrfs", 0, NULL) < 0) { | |
273 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount loopback device: %m"); | |
274 | goto fail; | |
275 | } | |
276 | mntdir_mounted = true; | |
277 | ||
278 | r = btrfs_quota_enable(mntdir, true); | |
279 | if (r < 0) | |
280 | log_warning_errno(r, "Failed to enable quota, ignoring: %m"); | |
281 | ||
5bcd08db LP |
282 | r = btrfs_subvol_auto_qgroup(mntdir, 0, true); |
283 | if (r < 0) | |
284 | log_warning_errno(r, "Failed to set up default quota hierarchy, ignoring: %m"); | |
285 | ||
432cea00 LP |
286 | if (chmod(mntdir, 0700) < 0) { |
287 | r = sd_bus_error_set_errnof(error, errno, "Failed to fix owner: %m"); | |
288 | goto fail; | |
289 | } | |
290 | ||
291 | (void) mkdir_p_label("/var/lib/machines", 0700); | |
292 | ||
293 | if (mount(mntdir, "/var/lib/machines", NULL, MS_BIND, NULL) < 0) { | |
294 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount directory into right place: %m"); | |
295 | goto fail; | |
296 | } | |
297 | ||
26166c88 LP |
298 | (void) syncfs(fd); |
299 | ||
300 | 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)); | |
301 | ||
432cea00 LP |
302 | (void) umount2(mntdir, MNT_DETACH); |
303 | (void) rmdir(mntdir); | |
304 | (void) rmdir(tmpdir); | |
305 | ||
5bcd08db | 306 | return 1; |
432cea00 LP |
307 | |
308 | fail: | |
309 | if (mntdir_mounted) | |
310 | (void) umount2(mntdir, MNT_DETACH); | |
311 | ||
312 | if (mntdir_made) | |
313 | (void) rmdir(mntdir); | |
314 | if (tmpdir_made) | |
315 | (void) rmdir(tmpdir); | |
316 | ||
317 | if (loop >= 0) { | |
318 | (void) ioctl(loop, LOOP_CLR_FD); | |
319 | loop = safe_close(loop); | |
320 | } | |
321 | ||
322 | if (control >= 0 && nr >= 0) | |
323 | (void) ioctl(control, LOOP_CTL_REMOVE, nr); | |
324 | ||
325 | return r; | |
326 | } | |
26166c88 LP |
327 | |
328 | static int sync_path(const char *p) { | |
329 | _cleanup_close_ int fd = -1; | |
330 | ||
331 | fd = open(p, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
332 | if (fd < 0) | |
333 | return -errno; | |
334 | ||
335 | if (syncfs(fd) < 0) | |
336 | return -errno; | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
341 | int grow_machine_directory(void) { | |
342 | char buf[FORMAT_BYTES_MAX]; | |
343 | struct statvfs a, b; | |
344 | uint64_t old_size, new_size, max_add; | |
345 | int r; | |
346 | ||
347 | /* Ensure the disk space data is accurate */ | |
348 | sync_path("/var/lib/machines"); | |
349 | sync_path("/var/lib/machines.raw"); | |
350 | ||
351 | if (statvfs("/var/lib/machines.raw", &a) < 0) | |
352 | return -errno; | |
353 | ||
354 | if (statvfs("/var/lib/machines", &b) < 0) | |
355 | return -errno; | |
356 | ||
357 | /* Don't grow if not enough disk space is available on the host */ | |
358 | if (((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) <= VAR_LIB_MACHINES_FREE_MIN) | |
359 | return 0; | |
360 | ||
361 | /* Don't grow if at least 1/3th of the fs is still free */ | |
362 | if (b.f_bavail > b.f_blocks / 3) | |
363 | return 0; | |
364 | ||
a8eaaee7 | 365 | /* Calculate how much we are willing to add at most */ |
26166c88 LP |
366 | max_add = ((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) - VAR_LIB_MACHINES_FREE_MIN; |
367 | ||
368 | /* Calculate the old size */ | |
369 | old_size = (uint64_t) b.f_blocks * (uint64_t) b.f_bsize; | |
370 | ||
371 | /* Calculate the new size as three times the size of what is used right now */ | |
372 | new_size = ((uint64_t) b.f_blocks - (uint64_t) b.f_bavail) * (uint64_t) b.f_bsize * 3; | |
373 | ||
374 | /* Always, grow at least to the start size */ | |
375 | if (new_size < VAR_LIB_MACHINES_SIZE_START) | |
376 | new_size = VAR_LIB_MACHINES_SIZE_START; | |
377 | ||
378 | /* If the new size is smaller than the old size, don't grow */ | |
379 | if (new_size < old_size) | |
380 | return 0; | |
381 | ||
382 | /* Ensure we never add more than the maximum */ | |
383 | if (new_size > old_size + max_add) | |
384 | new_size = old_size + max_add; | |
385 | ||
386 | r = btrfs_resize_loopback("/var/lib/machines", new_size, true); | |
387 | if (r <= 0) | |
388 | return r; | |
389 | ||
5bcd08db LP |
390 | /* Also bump the quota, of both the subvolume leaf qgroup, as |
391 | * well as of any subtree quota group by the same id but a | |
392 | * higher level, if it exists. */ | |
393 | (void) btrfs_qgroup_set_limit("/var/lib/machines", 0, new_size); | |
394 | (void) btrfs_subvol_set_subtree_quota_limit("/var/lib/machines", 0, new_size); | |
26166c88 LP |
395 | |
396 | log_info("Grew /var/lib/machines btrfs loopback file system to %s.", format_bytes(buf, sizeof(buf), new_size)); | |
397 | return 1; | |
398 | } |