]>
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 | ||
22 | #include <sys/prctl.h> | |
23 | #include <sys/vfs.h> | |
24 | #include <sys/statvfs.h> | |
25 | #include <sys/mount.h> | |
26 | ||
27 | #include "util.h" | |
28 | #include "mkdir.h" | |
29 | #include "btrfs-util.h" | |
30 | #include "path-util.h" | |
31 | #include "machine-pool.h" | |
32 | ||
33 | #define VAR_LIB_MACHINES_SIZE_START (1024UL*1024UL*500UL) | |
34 | #define VAR_LIB_MACHINES_FREE_MIN (1024UL*1024UL*750UL) | |
35 | ||
36 | static int check_btrfs(void) { | |
37 | struct statfs sfs; | |
38 | ||
39 | if (statfs("/var/lib/machines", &sfs) < 0) { | |
40 | if (errno != ENOENT) | |
41 | return -errno; | |
42 | ||
43 | if (statfs("/var/lib", &sfs) < 0) | |
44 | return -errno; | |
45 | } | |
46 | ||
47 | return F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC); | |
48 | } | |
49 | ||
4cee5eed | 50 | static int setup_machine_raw(uint64_t size, sd_bus_error *error) { |
432cea00 LP |
51 | _cleanup_free_ char *tmp = NULL; |
52 | _cleanup_close_ int fd = -1; | |
53 | struct statvfs ss; | |
54 | pid_t pid = 0; | |
55 | siginfo_t si; | |
56 | int r; | |
57 | ||
58 | /* We want to be able to make use of btrfs-specific file | |
59 | * system features, in particular subvolumes, reflinks and | |
60 | * quota. Hence, if we detect that /var/lib/machines.raw is | |
61 | * not located on btrfs, let's create a loopback file, place a | |
62 | * btrfs file system into it, and mount it to | |
63 | * /var/lib/machines. */ | |
64 | ||
65 | fd = open("/var/lib/machines.raw", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
66 | if (fd >= 0) { | |
67 | r = fd; | |
68 | fd = -1; | |
69 | return r; | |
70 | } | |
71 | ||
72 | if (errno != ENOENT) | |
73 | return sd_bus_error_set_errnof(error, errno, "Failed to open /var/lib/machines.raw: %m"); | |
74 | ||
75 | r = tempfn_xxxxxx("/var/lib/machines.raw", &tmp); | |
76 | if (r < 0) | |
77 | return r; | |
78 | ||
79 | (void) mkdir_p_label("/var/lib", 0755); | |
80 | fd = open(tmp, O_RDWR|O_CREAT|O_EXCL|O_NOCTTY|O_CLOEXEC, 0600); | |
81 | if (fd < 0) | |
82 | return sd_bus_error_set_errnof(error, errno, "Failed to create /var/lib/machines.raw: %m"); | |
83 | ||
84 | if (fstatvfs(fd, &ss) < 0) { | |
85 | r = sd_bus_error_set_errnof(error, errno, "Failed to determine free space on /var/lib/machines.raw: %m"); | |
86 | goto fail; | |
87 | } | |
88 | ||
89 | if (ss.f_bsize * ss.f_bavail < VAR_LIB_MACHINES_FREE_MIN) { | |
90 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "Not enough free disk space to set up /var/lib/machines."); | |
91 | goto fail; | |
92 | } | |
93 | ||
4cee5eed | 94 | if (ftruncate(fd, size) < 0) { |
432cea00 LP |
95 | r = sd_bus_error_set_errnof(error, errno, "Failed to enlarge /var/lib/machines.raw: %m"); |
96 | goto fail; | |
97 | } | |
98 | ||
99 | pid = fork(); | |
100 | if (pid < 0) { | |
101 | r = sd_bus_error_set_errnof(error, errno, "Failed to fork mkfs.btrfs: %m"); | |
102 | goto fail; | |
103 | } | |
104 | ||
105 | if (pid == 0) { | |
106 | ||
107 | /* Child */ | |
108 | ||
109 | reset_all_signal_handlers(); | |
110 | reset_signal_mask(); | |
111 | assert_se(prctl(PR_SET_PDEATHSIG, SIGTERM) == 0); | |
112 | ||
113 | fd = safe_close(fd); | |
114 | ||
115 | execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp, NULL); | |
116 | if (errno == ENOENT) | |
117 | return 99; | |
118 | ||
119 | _exit(EXIT_FAILURE); | |
120 | } | |
121 | ||
122 | r = wait_for_terminate(pid, &si); | |
123 | if (r < 0) { | |
124 | sd_bus_error_set_errnof(error, r, "Failed to wait for mkfs.btrfs: %m"); | |
125 | goto fail; | |
126 | } | |
127 | ||
128 | pid = 0; | |
129 | ||
130 | if (si.si_code != CLD_EXITED) { | |
131 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs died abnormally."); | |
132 | goto fail; | |
133 | } | |
134 | if (si.si_status == 99) { | |
135 | r = sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing"); | |
136 | goto fail; | |
137 | } | |
138 | if (si.si_status != 0) { | |
139 | r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs failed with error code %i", si.si_status); | |
140 | goto fail; | |
141 | } | |
142 | ||
143 | if (renameat2(AT_FDCWD, tmp, AT_FDCWD, "/var/lib/machines.raw", RENAME_NOREPLACE) < 0) { | |
144 | r = sd_bus_error_set_errnof(error, errno, "Failed to move /var/lib/machines.raw into place: %m"); | |
145 | goto fail; | |
146 | } | |
147 | ||
148 | r = fd; | |
149 | fd = -1; | |
150 | ||
151 | return r; | |
152 | ||
153 | fail: | |
132764a2 | 154 | unlink_noerrno(tmp); |
432cea00 LP |
155 | |
156 | if (pid > 1) | |
157 | kill_and_sigcont(pid, SIGKILL); | |
158 | ||
159 | return r; | |
160 | } | |
161 | ||
4cee5eed | 162 | int setup_machine_directory(uint64_t size, sd_bus_error *error) { |
403e5b32 | 163 | _cleanup_release_lock_file_ LockFile lock_file = LOCK_FILE_INIT; |
432cea00 LP |
164 | struct loop_info64 info = { |
165 | .lo_flags = LO_FLAGS_AUTOCLEAR, | |
166 | }; | |
167 | _cleanup_close_ int fd = -1, control = -1, loop = -1; | |
168 | _cleanup_free_ char* loopdev = NULL; | |
169 | char tmpdir[] = "/tmp/import-mount.XXXXXX", *mntdir = NULL; | |
170 | bool tmpdir_made = false, mntdir_made = false, mntdir_mounted = false; | |
26166c88 | 171 | char buf[FORMAT_BYTES_MAX]; |
432cea00 LP |
172 | int r, nr = -1; |
173 | ||
4cee5eed LP |
174 | /* btrfs cannot handle file systems < 16M, hence use this as minimum */ |
175 | if (size == (uint64_t) -1) | |
176 | size = VAR_LIB_MACHINES_SIZE_START; | |
177 | else if (size < 16*1024*1024) | |
178 | size = 16*1024*1024; | |
179 | ||
403e5b32 LP |
180 | /* Make sure we only set the directory up once at a time */ |
181 | r = make_lock_file("/run/systemd/machines.lock", LOCK_EX, &lock_file); | |
182 | if (r < 0) | |
183 | return r; | |
184 | ||
432cea00 LP |
185 | r = check_btrfs(); |
186 | if (r < 0) | |
187 | return sd_bus_error_set_errnof(error, r, "Failed to determine whether /var/lib/machines is located on btrfs: %m"); | |
188 | if (r > 0) { | |
189 | (void) btrfs_subvol_make_label("/var/lib/machines"); | |
190 | ||
191 | r = btrfs_quota_enable("/var/lib/machines", true); | |
192 | if (r < 0) | |
193 | log_warning_errno(r, "Failed to enable quota, ignoring: %m"); | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
198 | if (path_is_mount_point("/var/lib/machines", true) > 0 || | |
199 | dir_is_empty("/var/lib/machines") == 0) | |
200 | return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "/var/lib/machines is not a btrfs file system. Operation is not supported on legacy file systems."); | |
201 | ||
4cee5eed | 202 | fd = setup_machine_raw(size, error); |
432cea00 LP |
203 | if (fd < 0) |
204 | return fd; | |
205 | ||
206 | control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK); | |
207 | if (control < 0) | |
208 | return sd_bus_error_set_errnof(error, errno, "Failed to open /dev/loop-control: %m"); | |
209 | ||
210 | nr = ioctl(control, LOOP_CTL_GET_FREE); | |
211 | if (nr < 0) | |
212 | return sd_bus_error_set_errnof(error, errno, "Failed to allocate loop device: %m"); | |
213 | ||
214 | if (asprintf(&loopdev, "/dev/loop%i", nr) < 0) { | |
215 | r = -ENOMEM; | |
216 | goto fail; | |
217 | } | |
218 | ||
219 | loop = open(loopdev, O_CLOEXEC|O_RDWR|O_NOCTTY|O_NONBLOCK); | |
220 | if (loop < 0) { | |
221 | r = sd_bus_error_set_errnof(error, errno, "Failed to open loopback device: %m"); | |
222 | goto fail; | |
223 | } | |
224 | ||
225 | if (ioctl(loop, LOOP_SET_FD, fd) < 0) { | |
226 | r = sd_bus_error_set_errnof(error, errno, "Failed to bind loopback device: %m"); | |
227 | goto fail; | |
228 | } | |
229 | ||
230 | if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0) { | |
231 | r = sd_bus_error_set_errnof(error, errno, "Failed to enable auto-clear for loopback device: %m"); | |
232 | goto fail; | |
233 | } | |
234 | ||
235 | /* We need to make sure the new /var/lib/machines directory | |
236 | * has an access mode of 0700 at the time it is first made | |
237 | * available. mkfs will create it with 0755 however. Hence, | |
238 | * let's mount the directory into an inaccessible directory | |
239 | * below /tmp first, fix the access mode, and move it to the | |
240 | * public place then. */ | |
241 | ||
242 | if (!mkdtemp(tmpdir)) { | |
243 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount parent directory: %m"); | |
244 | goto fail; | |
245 | } | |
246 | tmpdir_made = true; | |
247 | ||
248 | mntdir = strjoina(tmpdir, "/mnt"); | |
249 | if (mkdir(mntdir, 0700) < 0) { | |
250 | r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount directory: %m"); | |
251 | goto fail; | |
252 | } | |
253 | mntdir_made = true; | |
254 | ||
255 | if (mount(loopdev, mntdir, "btrfs", 0, NULL) < 0) { | |
256 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount loopback device: %m"); | |
257 | goto fail; | |
258 | } | |
259 | mntdir_mounted = true; | |
260 | ||
261 | r = btrfs_quota_enable(mntdir, true); | |
262 | if (r < 0) | |
263 | log_warning_errno(r, "Failed to enable quota, ignoring: %m"); | |
264 | ||
265 | if (chmod(mntdir, 0700) < 0) { | |
266 | r = sd_bus_error_set_errnof(error, errno, "Failed to fix owner: %m"); | |
267 | goto fail; | |
268 | } | |
269 | ||
270 | (void) mkdir_p_label("/var/lib/machines", 0700); | |
271 | ||
272 | if (mount(mntdir, "/var/lib/machines", NULL, MS_BIND, NULL) < 0) { | |
273 | r = sd_bus_error_set_errnof(error, errno, "Failed to mount directory into right place: %m"); | |
274 | goto fail; | |
275 | } | |
276 | ||
26166c88 LP |
277 | (void) syncfs(fd); |
278 | ||
279 | 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)); | |
280 | ||
432cea00 LP |
281 | (void) umount2(mntdir, MNT_DETACH); |
282 | (void) rmdir(mntdir); | |
283 | (void) rmdir(tmpdir); | |
284 | ||
285 | return 0; | |
286 | ||
287 | fail: | |
288 | if (mntdir_mounted) | |
289 | (void) umount2(mntdir, MNT_DETACH); | |
290 | ||
291 | if (mntdir_made) | |
292 | (void) rmdir(mntdir); | |
293 | if (tmpdir_made) | |
294 | (void) rmdir(tmpdir); | |
295 | ||
296 | if (loop >= 0) { | |
297 | (void) ioctl(loop, LOOP_CLR_FD); | |
298 | loop = safe_close(loop); | |
299 | } | |
300 | ||
301 | if (control >= 0 && nr >= 0) | |
302 | (void) ioctl(control, LOOP_CTL_REMOVE, nr); | |
303 | ||
304 | return r; | |
305 | } | |
26166c88 LP |
306 | |
307 | static int sync_path(const char *p) { | |
308 | _cleanup_close_ int fd = -1; | |
309 | ||
310 | fd = open(p, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
311 | if (fd < 0) | |
312 | return -errno; | |
313 | ||
314 | if (syncfs(fd) < 0) | |
315 | return -errno; | |
316 | ||
317 | return 0; | |
318 | } | |
319 | ||
320 | int grow_machine_directory(void) { | |
321 | char buf[FORMAT_BYTES_MAX]; | |
322 | struct statvfs a, b; | |
323 | uint64_t old_size, new_size, max_add; | |
324 | int r; | |
325 | ||
326 | /* Ensure the disk space data is accurate */ | |
327 | sync_path("/var/lib/machines"); | |
328 | sync_path("/var/lib/machines.raw"); | |
329 | ||
330 | if (statvfs("/var/lib/machines.raw", &a) < 0) | |
331 | return -errno; | |
332 | ||
333 | if (statvfs("/var/lib/machines", &b) < 0) | |
334 | return -errno; | |
335 | ||
336 | /* Don't grow if not enough disk space is available on the host */ | |
337 | if (((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) <= VAR_LIB_MACHINES_FREE_MIN) | |
338 | return 0; | |
339 | ||
340 | /* Don't grow if at least 1/3th of the fs is still free */ | |
341 | if (b.f_bavail > b.f_blocks / 3) | |
342 | return 0; | |
343 | ||
344 | /* Calculate how much we are willing to add at maximum */ | |
345 | max_add = ((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) - VAR_LIB_MACHINES_FREE_MIN; | |
346 | ||
347 | /* Calculate the old size */ | |
348 | old_size = (uint64_t) b.f_blocks * (uint64_t) b.f_bsize; | |
349 | ||
350 | /* Calculate the new size as three times the size of what is used right now */ | |
351 | new_size = ((uint64_t) b.f_blocks - (uint64_t) b.f_bavail) * (uint64_t) b.f_bsize * 3; | |
352 | ||
353 | /* Always, grow at least to the start size */ | |
354 | if (new_size < VAR_LIB_MACHINES_SIZE_START) | |
355 | new_size = VAR_LIB_MACHINES_SIZE_START; | |
356 | ||
357 | /* If the new size is smaller than the old size, don't grow */ | |
358 | if (new_size < old_size) | |
359 | return 0; | |
360 | ||
361 | /* Ensure we never add more than the maximum */ | |
362 | if (new_size > old_size + max_add) | |
363 | new_size = old_size + max_add; | |
364 | ||
365 | r = btrfs_resize_loopback("/var/lib/machines", new_size, true); | |
366 | if (r <= 0) | |
367 | return r; | |
368 | ||
369 | r = btrfs_quota_limit("/var/lib/machines", new_size); | |
370 | if (r < 0) | |
371 | return r; | |
372 | ||
373 | log_info("Grew /var/lib/machines btrfs loopback file system to %s.", format_bytes(buf, sizeof(buf), new_size)); | |
374 | return 1; | |
375 | } |