]>
Commit | Line | Data |
---|---|---|
9be53773 SH |
1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * (C) Copyright IBM Corp. 2007, 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Daniel Lezcano <daniel.lezcano at free.fr> | |
8 | * | |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
250b1eec | 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
9be53773 SH |
22 | */ |
23 | ||
24 | /* | |
25 | * this is all just a first shot for experiment. If we go this route, much | |
26 | * shoudl change. bdev should be a directory with per-bdev file. Things which | |
27 | * I'm doing by calling out to userspace should sometimes be done through | |
28 | * libraries like liblvm2 | |
29 | */ | |
30 | #define _GNU_SOURCE | |
31 | #include <stdio.h> | |
32 | #include <unistd.h> | |
33 | #include <errno.h> | |
34 | #include <sched.h> | |
35 | #include <sys/mount.h> | |
36 | #include <sys/wait.h> | |
37 | #include <libgen.h> | |
eddaaafd SH |
38 | #include <linux/loop.h> |
39 | #include <dirent.h> | |
9be53773 SH |
40 | #include "lxc.h" |
41 | #include "config.h" | |
42 | #include "conf.h" | |
43 | #include "bdev.h" | |
44 | #include "log.h" | |
45 | #include "error.h" | |
46 | #include "utils.h" | |
47 | #include "namespace.h" | |
48 | #include "parse.h" | |
6a44839f | 49 | #include "utils.h" |
9be53773 | 50 | |
bff13ba2 SG |
51 | #ifndef BLKGETSIZE64 |
52 | #define BLKGETSIZE64 _IOR(0x12,114,size_t) | |
53 | #endif | |
54 | ||
55 | #ifndef LO_FLAGS_AUTOCLEAR | |
56 | #define LO_FLAGS_AUTOCLEAR 4 | |
57 | #endif | |
58 | ||
9be53773 SH |
59 | lxc_log_define(bdev, lxc); |
60 | ||
9be53773 SH |
61 | static int do_rsync(const char *src, const char *dest) |
62 | { | |
63 | // call out to rsync | |
64 | pid_t pid; | |
65 | char *s; | |
66 | size_t l; | |
67 | ||
68 | pid = fork(); | |
69 | if (pid < 0) | |
70 | return -1; | |
71 | if (pid > 0) | |
72 | return wait_for_pid(pid); | |
73 | l = strlen(src) + 2; | |
74 | s = malloc(l); | |
75 | if (!s) | |
76 | exit(1); | |
77 | strcpy(s, src); | |
78 | s[l-2] = '/'; | |
79 | s[l-1] = '\0'; | |
80 | ||
ca52dcb5 SH |
81 | execlp("rsync", "rsync", "-a", s, dest, (char *)NULL); |
82 | exit(1); | |
9be53773 SH |
83 | } |
84 | ||
eddaaafd SH |
85 | /* |
86 | * return block size of dev->src | |
87 | */ | |
88 | static int blk_getsize(struct bdev *bdev, unsigned long *size) | |
9be53773 SH |
89 | { |
90 | int fd, ret; | |
eddaaafd SH |
91 | char *path = bdev->src; |
92 | ||
93 | if (strcmp(bdev->type, "loop") == 0) | |
94 | path = bdev->src + 5; | |
9be53773 SH |
95 | |
96 | fd = open(path, O_RDONLY); | |
42fb4b15 | 97 | if (fd < 0) |
9be53773 SH |
98 | return -1; |
99 | ret = ioctl(fd, BLKGETSIZE64, size); | |
100 | close(fd); | |
101 | return ret; | |
102 | } | |
103 | ||
104 | /* | |
105 | * These are copied from conf.c. However as conf.c will be moved to using | |
106 | * the callback system, they can be pulled from there eventually, so we | |
107 | * don't need to pollute utils.c with these low level functions | |
108 | */ | |
109 | static int find_fstype_cb(char* buffer, void *data) | |
110 | { | |
111 | struct cbarg { | |
112 | const char *rootfs; | |
113 | const char *target; | |
114 | int mntopt; | |
115 | } *cbarg = data; | |
116 | ||
117 | char *fstype; | |
118 | ||
119 | /* we don't try 'nodev' entries */ | |
120 | if (strstr(buffer, "nodev")) | |
121 | return 0; | |
122 | ||
123 | fstype = buffer; | |
124 | fstype += lxc_char_left_gc(fstype, strlen(fstype)); | |
125 | fstype[lxc_char_right_gc(fstype, strlen(fstype))] = '\0'; | |
126 | ||
127 | DEBUG("trying to mount '%s'->'%s' with fstype '%s'", | |
128 | cbarg->rootfs, cbarg->target, fstype); | |
129 | ||
130 | if (mount(cbarg->rootfs, cbarg->target, fstype, cbarg->mntopt, NULL)) { | |
131 | DEBUG("mount failed with error: %s", strerror(errno)); | |
132 | return 0; | |
133 | } | |
134 | ||
135 | INFO("mounted '%s' on '%s', with fstype '%s'", | |
136 | cbarg->rootfs, cbarg->target, fstype); | |
137 | ||
138 | return 1; | |
139 | } | |
140 | ||
141 | static int mount_unknow_fs(const char *rootfs, const char *target, int mntopt) | |
142 | { | |
143 | int i; | |
144 | ||
145 | struct cbarg { | |
146 | const char *rootfs; | |
147 | const char *target; | |
148 | int mntopt; | |
149 | } cbarg = { | |
150 | .rootfs = rootfs, | |
151 | .target = target, | |
152 | .mntopt = mntopt, | |
153 | }; | |
154 | ||
155 | /* | |
156 | * find the filesystem type with brute force: | |
157 | * first we check with /etc/filesystems, in case the modules | |
158 | * are auto-loaded and fall back to the supported kernel fs | |
159 | */ | |
160 | char *fsfile[] = { | |
161 | "/etc/filesystems", | |
162 | "/proc/filesystems", | |
163 | }; | |
164 | ||
165 | for (i = 0; i < sizeof(fsfile)/sizeof(fsfile[0]); i++) { | |
166 | ||
167 | int ret; | |
168 | ||
169 | if (access(fsfile[i], F_OK)) | |
170 | continue; | |
171 | ||
172 | ret = lxc_file_for_each_line(fsfile[i], find_fstype_cb, &cbarg); | |
173 | if (ret < 0) { | |
174 | ERROR("failed to parse '%s'", fsfile[i]); | |
175 | return -1; | |
176 | } | |
177 | ||
178 | if (ret) | |
179 | return 0; | |
180 | } | |
181 | ||
182 | ERROR("failed to determine fs type for '%s'", rootfs); | |
183 | return -1; | |
184 | } | |
185 | ||
186 | static int do_mkfs(const char *path, const char *fstype) | |
187 | { | |
188 | pid_t pid; | |
189 | ||
190 | if ((pid = fork()) < 0) { | |
191 | ERROR("error forking"); | |
192 | return -1; | |
193 | } | |
194 | if (pid > 0) | |
195 | return wait_for_pid(pid); | |
196 | ||
eddaaafd SH |
197 | // If the file is not a block device, we don't want mkfs to ask |
198 | // us about whether to proceed. | |
199 | close(0); | |
200 | close(1); | |
201 | close(2); | |
202 | open("/dev/zero", O_RDONLY); | |
203 | open("/dev/null", O_RDWR); | |
204 | open("/dev/null", O_RDWR); | |
ca52dcb5 SH |
205 | execlp("mkfs", "mkfs", "-t", fstype, path, NULL); |
206 | exit(1); | |
9be53773 SH |
207 | } |
208 | ||
209 | static char *linkderef(char *path, char *dest) | |
210 | { | |
211 | struct stat sbuf; | |
212 | ssize_t ret; | |
213 | ||
214 | ret = stat(path, &sbuf); | |
215 | if (ret < 0) | |
216 | return NULL; | |
217 | if (!S_ISLNK(sbuf.st_mode)) | |
218 | return path; | |
219 | ret = readlink(path, dest, MAXPATHLEN); | |
220 | if (ret < 0) { | |
221 | SYSERROR("error reading link %s", path); | |
222 | return NULL; | |
223 | } else if (ret >= MAXPATHLEN) { | |
224 | ERROR("link in %s too long", path); | |
225 | return NULL; | |
226 | } | |
227 | dest[ret] = '\0'; | |
228 | return dest; | |
229 | } | |
230 | ||
231 | /* | |
232 | * Given a bdev (presumably blockdev-based), detect the fstype | |
233 | * by trying mounting (in a private mntns) it. | |
234 | * @bdev: bdev to investigate | |
235 | * @type: preallocated char* in which to write the fstype | |
236 | * @len: length of passed in char* | |
237 | * Returns length of fstype, of -1 on error | |
238 | */ | |
239 | static int detect_fs(struct bdev *bdev, char *type, int len) | |
240 | { | |
241 | int p[2], ret; | |
242 | size_t linelen; | |
243 | pid_t pid; | |
244 | FILE *f; | |
245 | char *sp1, *sp2, *sp3, *line = NULL; | |
5d9598d7 | 246 | char *srcdev; |
9be53773 SH |
247 | |
248 | if (!bdev || !bdev->src || !bdev->dest) | |
249 | return -1; | |
250 | ||
5d9598d7 | 251 | srcdev = bdev->src; |
eddaaafd SH |
252 | if (strcmp(bdev->type, "loop") == 0) |
253 | srcdev = bdev->src + 5; | |
254 | ||
9be53773 SH |
255 | if (pipe(p) < 0) |
256 | return -1; | |
257 | if ((pid = fork()) < 0) | |
258 | return -1; | |
259 | if (pid > 0) { | |
260 | int status; | |
261 | close(p[1]); | |
262 | memset(type, 0, len); | |
263 | ret = read(p[0], type, len-1); | |
264 | close(p[0]); | |
265 | if (ret < 0) { | |
266 | SYSERROR("error reading from pipe"); | |
267 | wait(&status); | |
268 | return -1; | |
269 | } else if (ret == 0) { | |
270 | ERROR("child exited early - fstype not found"); | |
271 | wait(&status); | |
272 | return -1; | |
273 | } | |
274 | wait(&status); | |
275 | type[len-1] = '\0'; | |
eddaaafd | 276 | INFO("detected fstype %s for %s", type, srcdev); |
9be53773 SH |
277 | return ret; |
278 | } | |
279 | ||
280 | if (unshare(CLONE_NEWNS) < 0) | |
281 | exit(1); | |
282 | ||
eddaaafd | 283 | ret = mount_unknow_fs(srcdev, bdev->dest, 0); |
9be53773 | 284 | if (ret < 0) { |
eddaaafd | 285 | ERROR("failed mounting %s onto %s to detect fstype", srcdev, bdev->dest); |
9be53773 SH |
286 | exit(1); |
287 | } | |
288 | // if symlink, get the real dev name | |
289 | char devpath[MAXPATHLEN]; | |
eddaaafd | 290 | char *l = linkderef(srcdev, devpath); |
9be53773 SH |
291 | if (!l) |
292 | exit(1); | |
293 | f = fopen("/proc/self/mounts", "r"); | |
294 | if (!f) | |
295 | exit(1); | |
296 | while (getline(&line, &linelen, f) != -1) { | |
297 | sp1 = index(line, ' '); | |
298 | if (!sp1) | |
299 | exit(1); | |
300 | *sp1 = '\0'; | |
301 | if (strcmp(line, l)) | |
302 | continue; | |
303 | sp2 = index(sp1+1, ' '); | |
304 | if (!sp2) | |
305 | exit(1); | |
306 | *sp2 = '\0'; | |
307 | sp3 = index(sp2+1, ' '); | |
308 | if (!sp3) | |
309 | exit(1); | |
310 | *sp3 = '\0'; | |
311 | sp2++; | |
312 | if (write(p[1], sp2, strlen(sp2)) != strlen(sp2)) | |
313 | exit(1); | |
314 | exit(0); | |
315 | } | |
316 | exit(1); | |
317 | } | |
318 | ||
319 | struct bdev_type { | |
320 | char *name; | |
321 | struct bdev_ops *ops; | |
322 | }; | |
323 | ||
324 | static int is_dir(const char *path) | |
325 | { | |
326 | struct stat statbuf; | |
327 | int ret = stat(path, &statbuf); | |
328 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) | |
329 | return 1; | |
330 | return 0; | |
331 | } | |
332 | ||
333 | static int dir_detect(const char *path) | |
334 | { | |
335 | if (strncmp(path, "dir:", 4) == 0) | |
336 | return 1; // take their word for it | |
337 | if (is_dir(path)) | |
338 | return 1; | |
339 | return 0; | |
340 | } | |
341 | ||
342 | // | |
343 | // XXXXXXX plain directory bind mount ops | |
344 | // | |
60bf62d4 | 345 | static int dir_mount(struct bdev *bdev) |
9be53773 SH |
346 | { |
347 | if (strcmp(bdev->type, "dir")) | |
348 | return -22; | |
349 | if (!bdev->src || !bdev->dest) | |
350 | return -22; | |
351 | return mount(bdev->src, bdev->dest, "bind", MS_BIND | MS_REC, NULL); | |
352 | } | |
353 | ||
60bf62d4 | 354 | static int dir_umount(struct bdev *bdev) |
9be53773 SH |
355 | { |
356 | if (strcmp(bdev->type, "dir")) | |
357 | return -22; | |
358 | if (!bdev->src || !bdev->dest) | |
359 | return -22; | |
360 | return umount(bdev->dest); | |
361 | } | |
362 | ||
363 | /* the bulk of this needs to become a common helper */ | |
364 | static char *dir_new_path(char *src, const char *oldname, const char *name, | |
365 | const char *oldpath, const char *lxcpath) | |
366 | { | |
367 | char *ret, *p, *p2; | |
368 | int l1, l2, nlen; | |
369 | ||
370 | nlen = strlen(src) + 1; | |
371 | l1 = strlen(oldpath); | |
372 | p = src; | |
373 | /* if src starts with oldpath, look for oldname only after | |
374 | * that path */ | |
375 | if (strncmp(src, oldpath, l1) == 0) { | |
376 | p += l1; | |
377 | nlen += (strlen(lxcpath) - l1); | |
378 | } | |
379 | l2 = strlen(oldname); | |
380 | while ((p = strstr(p, oldname)) != NULL) { | |
381 | p += l2; | |
382 | nlen += strlen(name) - l2; | |
383 | } | |
384 | ||
385 | ret = malloc(nlen); | |
386 | if (!ret) | |
387 | return NULL; | |
388 | ||
389 | p = ret; | |
390 | if (strncmp(src, oldpath, l1) == 0) { | |
391 | p += sprintf(p, "%s", lxcpath); | |
392 | src += l1; | |
393 | } | |
394 | ||
395 | while ((p2 = strstr(src, oldname)) != NULL) { | |
396 | strncpy(p, src, p2-src); // copy text up to oldname | |
397 | p += p2-src; // move target pointer (p) | |
398 | p += sprintf(p, "%s", name); // print new name in place of oldname | |
399 | src = p2 + l2; // move src to end of oldname | |
400 | } | |
401 | sprintf(p, "%s", src); // copy the rest of src | |
402 | return ret; | |
403 | } | |
404 | ||
405 | /* | |
406 | * for a simple directory bind mount, we substitute the old container | |
407 | * name and paths for the new | |
408 | */ | |
409 | static int dir_clonepaths(struct bdev *orig, struct bdev *new, const char *oldname, | |
410 | const char *cname, const char *oldpath, const char *lxcpath, int snap, | |
411 | unsigned long newsize) | |
412 | { | |
ca52dcb5 SH |
413 | int len, ret; |
414 | ||
9be53773 SH |
415 | if (snap) { |
416 | ERROR("directories cannot be snapshotted. Try overlayfs."); | |
417 | return -1; | |
418 | } | |
419 | ||
9be53773 SH |
420 | if (!orig->dest || !orig->src) |
421 | return -1; | |
9be53773 | 422 | |
ca52dcb5 SH |
423 | len = strlen(lxcpath) + strlen(cname) + strlen("rootfs") + 3; |
424 | new->src = malloc(len); | |
9be53773 SH |
425 | if (!new->src) |
426 | return -1; | |
ca52dcb5 SH |
427 | ret = snprintf(new->src, len, "%s/%s/rootfs", lxcpath, cname); |
428 | if (ret < 0 || ret >= len) | |
429 | return -1; | |
430 | if ((new->dest = strdup(new->src)) == NULL) | |
431 | return -1; | |
9be53773 SH |
432 | |
433 | return 0; | |
434 | } | |
435 | ||
60bf62d4 SH |
436 | static int dir_destroy(struct bdev *orig) |
437 | { | |
438 | if (!lxc_rmdir_onedev(orig->src)) | |
439 | return -1; | |
440 | return 0; | |
441 | } | |
442 | ||
1897e3bc SH |
443 | static int dir_create(struct bdev *bdev, const char *dest, const char *n, |
444 | struct bdev_specs *specs) | |
445 | { | |
446 | bdev->src = strdup(dest); | |
447 | bdev->dest = strdup(dest); | |
448 | if (!bdev->src || !bdev->dest) { | |
449 | ERROR("Out of memory"); | |
450 | return -1; | |
451 | } | |
452 | ||
453 | if (mkdir_p(bdev->src, 0755) < 0) { | |
454 | ERROR("Error creating %s\n", bdev->src); | |
455 | return -1; | |
456 | } | |
457 | if (mkdir_p(bdev->dest, 0755) < 0) { | |
458 | ERROR("Error creating %s\n", bdev->dest); | |
459 | return -1; | |
460 | } | |
461 | ||
462 | return 0; | |
463 | } | |
464 | ||
9be53773 SH |
465 | struct bdev_ops dir_ops = { |
466 | .detect = &dir_detect, | |
467 | .mount = &dir_mount, | |
468 | .umount = &dir_umount, | |
469 | .clone_paths = &dir_clonepaths, | |
60bf62d4 | 470 | .destroy = &dir_destroy, |
1897e3bc | 471 | .create = &dir_create, |
9be53773 SH |
472 | }; |
473 | ||
3baa76fe SH |
474 | |
475 | // | |
476 | // XXXXXXX zfs ops | |
477 | // There are two ways we could do this. We could always specify the | |
478 | // 'zfs device' (i.e. tank/lxc lxc/container) as rootfs. But instead | |
479 | // (at least right now) we have lxc-create specify $lxcpath/$lxcname/rootfs | |
480 | // as the mountpoint, so that it is always mounted. | |
481 | // | |
482 | // That means 'mount' is really never needed and could be noop, but for the | |
483 | // sake of flexibility let's always bind-mount. | |
484 | // | |
485 | ||
60bf62d4 | 486 | static int zfs_list_entry(const char *path, char *output, size_t inlen) |
3baa76fe SH |
487 | { |
488 | FILE *f; | |
489 | int found=0; | |
490 | ||
0fc0d057 | 491 | if ((f = popen("zfs list 2> /dev/null", "r")) == NULL) { |
3baa76fe SH |
492 | SYSERROR("popen failed"); |
493 | return 0; | |
494 | } | |
60bf62d4 | 495 | while (fgets(output, inlen, f)) { |
3baa76fe SH |
496 | if (strstr(output, path)) { |
497 | found = 1; | |
498 | break; | |
499 | } | |
500 | } | |
501 | (void) pclose(f); | |
502 | ||
503 | return found; | |
504 | } | |
505 | ||
506 | static int zfs_detect(const char *path) | |
507 | { | |
508 | char *output = malloc(LXC_LOG_BUFFER_SIZE); | |
509 | int found; | |
510 | ||
511 | if (!output) { | |
512 | ERROR("out of memory"); | |
513 | return 0; | |
514 | } | |
60bf62d4 | 515 | found = zfs_list_entry(path, output, LXC_LOG_BUFFER_SIZE); |
3baa76fe SH |
516 | free(output); |
517 | return found; | |
518 | } | |
519 | ||
60bf62d4 | 520 | static int zfs_mount(struct bdev *bdev) |
3baa76fe SH |
521 | { |
522 | if (strcmp(bdev->type, "zfs")) | |
523 | return -22; | |
524 | if (!bdev->src || !bdev->dest) | |
525 | return -22; | |
526 | return mount(bdev->src, bdev->dest, "bind", MS_BIND | MS_REC, NULL); | |
527 | } | |
528 | ||
60bf62d4 | 529 | static int zfs_umount(struct bdev *bdev) |
3baa76fe SH |
530 | { |
531 | if (strcmp(bdev->type, "zfs")) | |
532 | return -22; | |
533 | if (!bdev->src || !bdev->dest) | |
534 | return -22; | |
535 | return umount(bdev->dest); | |
536 | } | |
537 | ||
538 | static int zfs_clone(const char *opath, const char *npath, const char *oname, | |
539 | const char *nname, const char *lxcpath, int snapshot) | |
540 | { | |
541 | // use the 'zfs list | grep opath' entry to get the zfsroot | |
542 | char output[MAXPATHLEN], option[MAXPATHLEN], *p; | |
31a95fec | 543 | const char *zfsroot = output; |
3baa76fe SH |
544 | int ret; |
545 | pid_t pid; | |
546 | ||
60bf62d4 | 547 | if (zfs_list_entry(opath, output, MAXPATHLEN)) { |
31a95fec SH |
548 | // zfsroot is output up to ' ' |
549 | if ((p = index(output, ' ')) == NULL) | |
550 | return -1; | |
551 | *p = '\0'; | |
c32981c3 | 552 | if ((p = strrchr(output, '/')) == NULL) |
31a95fec SH |
553 | return -1; |
554 | *p = '\0'; | |
555 | } else | |
556 | zfsroot = default_zfs_root(); | |
3baa76fe SH |
557 | |
558 | ret = snprintf(option, MAXPATHLEN, "-omountpoint=%s/%s/rootfs", | |
559 | lxcpath, nname); | |
560 | if (ret < 0 || ret >= MAXPATHLEN) | |
561 | return -1; | |
562 | ||
3baa76fe SH |
563 | // zfs create -omountpoint=$lxcpath/$lxcname $zfsroot/$nname |
564 | if (!snapshot) { | |
565 | if ((pid = fork()) < 0) | |
566 | return -1; | |
567 | if (!pid) { | |
568 | char dev[MAXPATHLEN]; | |
31a95fec | 569 | ret = snprintf(dev, MAXPATHLEN, "%s/%s", zfsroot, nname); |
3baa76fe SH |
570 | if (ret < 0 || ret >= MAXPATHLEN) |
571 | exit(1); | |
ca52dcb5 SH |
572 | execlp("zfs", "zfs", "create", option, dev, NULL); |
573 | exit(1); | |
3baa76fe SH |
574 | } |
575 | return wait_for_pid(pid); | |
576 | } else { | |
577 | // if snapshot, do | |
578 | // 'zfs snapshot zfsroot/oname@nname | |
579 | // zfs clone zfsroot/oname@nname zfsroot/nname | |
580 | char path1[MAXPATHLEN], path2[MAXPATHLEN]; | |
581 | ||
31a95fec | 582 | ret = snprintf(path1, MAXPATHLEN, "%s/%s@%s", zfsroot, |
3baa76fe SH |
583 | oname, nname); |
584 | if (ret < 0 || ret >= MAXPATHLEN) | |
585 | return -1; | |
31a95fec | 586 | (void) snprintf(path2, MAXPATHLEN, "%s/%s", zfsroot, nname); |
3baa76fe SH |
587 | |
588 | // if the snapshot exists, delete it | |
589 | if ((pid = fork()) < 0) | |
590 | return -1; | |
591 | if (!pid) { | |
ca52dcb5 SH |
592 | execlp("zfs", "zfs", "destroy", path1, NULL); |
593 | exit(1); | |
3baa76fe SH |
594 | } |
595 | // it probably doesn't exist so destroy probably will fail. | |
596 | (void) wait_for_pid(pid); | |
597 | ||
598 | // run first (snapshot) command | |
599 | if ((pid = fork()) < 0) | |
600 | return -1; | |
601 | if (!pid) { | |
ca52dcb5 SH |
602 | execlp("zfs", "zfs", "snapshot", path1, NULL); |
603 | exit(1); | |
3baa76fe SH |
604 | } |
605 | if (wait_for_pid(pid) < 0) | |
606 | return -1; | |
607 | ||
608 | // run second (clone) command | |
609 | if ((pid = fork()) < 0) | |
610 | return -1; | |
611 | if (!pid) { | |
ca52dcb5 SH |
612 | execlp("zfs", "zfs", "clone", option, path1, path2, NULL); |
613 | exit(1); | |
3baa76fe SH |
614 | } |
615 | return wait_for_pid(pid); | |
616 | } | |
617 | } | |
618 | ||
619 | static int zfs_clonepaths(struct bdev *orig, struct bdev *new, const char *oldname, | |
620 | const char *cname, const char *oldpath, const char *lxcpath, int snap, | |
621 | unsigned long newsize) | |
622 | { | |
ca52dcb5 SH |
623 | int len, ret; |
624 | ||
3baa76fe SH |
625 | if (!orig->src || !orig->dest) |
626 | return -1; | |
627 | ||
ca52dcb5 SH |
628 | if (snap && strcmp(orig->type, "zfs")) { |
629 | ERROR("zfs snapshot from %s backing store is not supported", | |
3baa76fe SH |
630 | orig->type); |
631 | return -1; | |
632 | } | |
633 | ||
ca52dcb5 SH |
634 | len = strlen(lxcpath) + strlen(cname) + strlen("rootfs") + 3; |
635 | new->src = malloc(len); | |
3baa76fe SH |
636 | if (!new->src) |
637 | return -1; | |
ca52dcb5 SH |
638 | ret = snprintf(new->src, len, "%s/%s/rootfs", lxcpath, cname); |
639 | if (ret < 0 || ret >= len) | |
640 | return -1; | |
641 | if ((new->dest = strdup(new->src)) == NULL) | |
642 | return -1; | |
3baa76fe SH |
643 | |
644 | return zfs_clone(orig->src, new->src, oldname, cname, lxcpath, snap); | |
645 | } | |
646 | ||
60bf62d4 SH |
647 | /* |
648 | * TODO: detect whether this was a clone, and if so then also delete the | |
649 | * snapshot it was based on, so that we don't hold the original | |
650 | * container busy. | |
651 | */ | |
652 | static int zfs_destroy(struct bdev *orig) | |
653 | { | |
654 | pid_t pid; | |
655 | char output[MAXPATHLEN], *p; | |
656 | ||
657 | if ((pid = fork()) < 0) | |
658 | return -1; | |
659 | if (pid) | |
660 | return wait_for_pid(pid); | |
661 | ||
662 | if (!zfs_list_entry(orig->src, output, MAXPATHLEN)) { | |
663 | ERROR("Error: zfs entry for %s not found", orig->src); | |
664 | return -1; | |
665 | } | |
666 | ||
667 | // zfs mount is output up to ' ' | |
668 | if ((p = index(output, ' ')) == NULL) | |
669 | return -1; | |
670 | *p = '\0'; | |
671 | ||
672 | execlp("zfs", "zfs", "destroy", output, NULL); | |
673 | exit(1); | |
674 | } | |
675 | ||
1897e3bc SH |
676 | static int zfs_create(struct bdev *bdev, const char *dest, const char *n, |
677 | struct bdev_specs *specs) | |
678 | { | |
679 | const char *zfsroot; | |
680 | char option[MAXPATHLEN]; | |
681 | int ret; | |
682 | pid_t pid; | |
683 | ||
684 | if (!specs || !specs->u.zfs.zfsroot) | |
685 | zfsroot = default_zfs_root(); | |
686 | else | |
687 | zfsroot = specs->u.zfs.zfsroot; | |
688 | ||
689 | if (!(bdev->dest = strdup(dest))) { | |
690 | ERROR("No mount target specified or out of memory"); | |
691 | return -1; | |
692 | } | |
693 | if (!(bdev->src = strdup(bdev->dest))) { | |
694 | ERROR("out of memory"); | |
695 | return -1; | |
696 | } | |
697 | ||
698 | ret = snprintf(option, MAXPATHLEN, "-omountpoint=%s", bdev->dest); | |
699 | if (ret < 0 || ret >= MAXPATHLEN) | |
700 | return -1; | |
701 | if ((pid = fork()) < 0) | |
702 | return -1; | |
703 | if (pid) | |
704 | return wait_for_pid(pid); | |
705 | ||
706 | char dev[MAXPATHLEN]; | |
707 | ret = snprintf(dev, MAXPATHLEN, "%s/%s", zfsroot, n); | |
708 | if (ret < 0 || ret >= MAXPATHLEN) | |
709 | exit(1); | |
710 | execlp("zfs", "zfs", "create", option, dev, NULL); | |
711 | exit(1); | |
712 | } | |
713 | ||
3baa76fe SH |
714 | struct bdev_ops zfs_ops = { |
715 | .detect = &zfs_detect, | |
716 | .mount = &zfs_mount, | |
717 | .umount = &zfs_umount, | |
718 | .clone_paths = &zfs_clonepaths, | |
60bf62d4 | 719 | .destroy = &zfs_destroy, |
1897e3bc | 720 | .create = &zfs_create, |
3baa76fe SH |
721 | }; |
722 | ||
9be53773 SH |
723 | // |
724 | // LVM ops | |
725 | // | |
726 | ||
727 | /* | |
728 | * Look at /sys/dev/block/maj:min/dm/uuid. If it contains the hardcoded LVM | |
729 | * prefix "LVM-", then this is an lvm2 LV | |
730 | */ | |
731 | static int lvm_detect(const char *path) | |
732 | { | |
733 | char devp[MAXPATHLEN], buf[4]; | |
734 | FILE *fout; | |
735 | int ret; | |
736 | struct stat statbuf; | |
737 | ||
738 | if (strncmp(path, "lvm:", 4) == 0) | |
739 | return 1; // take their word for it | |
740 | ||
741 | ret = stat(path, &statbuf); | |
742 | if (ret != 0) | |
743 | return 0; | |
744 | if (!S_ISBLK(statbuf.st_mode)) | |
745 | return 0; | |
746 | ||
747 | ret = snprintf(devp, MAXPATHLEN, "/sys/dev/block/%d:%d/dm/uuid", | |
748 | major(statbuf.st_rdev), minor(statbuf.st_rdev)); | |
749 | if (ret < 0 || ret >= MAXPATHLEN) { | |
750 | ERROR("lvm uuid pathname too long"); | |
751 | return 0; | |
752 | } | |
753 | fout = fopen(devp, "r"); | |
754 | if (!fout) | |
755 | return 0; | |
756 | ret = fread(buf, 1, 4, fout); | |
757 | fclose(fout); | |
758 | if (ret != 4 || strncmp(buf, "LVM-", 4) != 0) | |
759 | return 0; | |
760 | return 1; | |
761 | } | |
762 | ||
763 | static int lvm_mount(struct bdev *bdev) | |
764 | { | |
765 | if (strcmp(bdev->type, "lvm")) | |
766 | return -22; | |
767 | if (!bdev->src || !bdev->dest) | |
768 | return -22; | |
769 | /* if we might pass in data sometime, then we'll have to enrich | |
770 | * mount_unknow_fs */ | |
771 | return mount_unknow_fs(bdev->src, bdev->dest, 0); | |
772 | } | |
773 | ||
774 | static int lvm_umount(struct bdev *bdev) | |
775 | { | |
776 | if (strcmp(bdev->type, "lvm")) | |
777 | return -22; | |
778 | if (!bdev->src || !bdev->dest) | |
779 | return -22; | |
780 | return umount(bdev->dest); | |
781 | } | |
782 | ||
783 | /* | |
784 | * path must be '/dev/$vg/$lv', $vg must be an existing VG, and $lv must | |
785 | * not yet exist. This function will attempt to create /dev/$vg/$lv of | |
786 | * size $size. | |
787 | */ | |
1897e3bc | 788 | static int do_lvm_create(const char *path, unsigned long size) |
9be53773 SH |
789 | { |
790 | int ret, pid; | |
791 | char sz[24], *pathdup, *vg, *lv; | |
792 | ||
793 | if ((pid = fork()) < 0) { | |
794 | SYSERROR("failed fork"); | |
795 | return -1; | |
796 | } | |
797 | if (pid > 0) | |
798 | return wait_for_pid(pid); | |
799 | ||
800 | // lvcreate default size is in M, not bytes. | |
801 | ret = snprintf(sz, 24, "%lu", size/1000000); | |
802 | if (ret < 0 || ret >= 24) | |
803 | exit(1); | |
804 | ||
805 | pathdup = strdup(path); | |
806 | if (!pathdup) | |
807 | exit(1); | |
c32981c3 | 808 | lv = strrchr(pathdup, '/'); |
9be53773 SH |
809 | if (!lv) { |
810 | free(pathdup); | |
811 | exit(1); | |
812 | } | |
813 | *lv = '\0'; | |
814 | lv++; | |
c32981c3 | 815 | vg = strrchr(pathdup, '/'); |
9be53773 SH |
816 | if (!vg) |
817 | exit(1); | |
818 | vg++; | |
ca52dcb5 | 819 | execlp("lvcreate", "lvcreate", "-L", sz, vg, "-n", lv, (char *)NULL); |
9be53773 | 820 | free(pathdup); |
ca52dcb5 | 821 | exit(1); |
9be53773 SH |
822 | } |
823 | ||
824 | static int lvm_snapshot(const char *orig, const char *path, unsigned long size) | |
825 | { | |
826 | int ret, pid; | |
827 | char sz[24], *pathdup, *lv; | |
828 | ||
829 | if ((pid = fork()) < 0) { | |
830 | SYSERROR("failed fork"); | |
831 | return -1; | |
832 | } | |
833 | if (pid > 0) | |
834 | return wait_for_pid(pid); | |
835 | // lvcreate default size is in M, not bytes. | |
836 | ret = snprintf(sz, 24, "%lu", size/1000000); | |
837 | if (ret < 0 || ret >= 24) | |
838 | exit(1); | |
839 | ||
840 | pathdup = strdup(path); | |
841 | if (!pathdup) | |
842 | exit(1); | |
c32981c3 | 843 | lv = strrchr(pathdup, '/'); |
9be53773 SH |
844 | if (!lv) { |
845 | free(pathdup); | |
846 | exit(1); | |
847 | } | |
848 | *lv = '\0'; | |
849 | lv++; | |
850 | ||
851 | ret = execlp("lvcreate", "lvcreate", "-s", "-L", sz, "-n", lv, orig, (char *)NULL); | |
852 | free(pathdup); | |
ca52dcb5 SH |
853 | exit(1); |
854 | } | |
855 | ||
856 | // this will return 1 for physical disks, qemu-nbd, loop, etc | |
857 | // right now only lvm is a block device | |
858 | static int is_blktype(struct bdev *b) | |
859 | { | |
860 | if (strcmp(b->type, "lvm") == 0) | |
861 | return 1; | |
862 | return 0; | |
9be53773 SH |
863 | } |
864 | ||
865 | static int lvm_clonepaths(struct bdev *orig, struct bdev *new, const char *oldname, | |
866 | const char *cname, const char *oldpath, const char *lxcpath, int snap, | |
867 | unsigned long newsize) | |
868 | { | |
869 | char fstype[100]; | |
870 | unsigned long size = newsize; | |
ca52dcb5 | 871 | int len, ret; |
9be53773 SH |
872 | |
873 | if (!orig->src || !orig->dest) | |
874 | return -1; | |
875 | ||
876 | if (strcmp(orig->type, "lvm")) { | |
31a95fec SH |
877 | const char *vg; |
878 | ||
ca52dcb5 SH |
879 | if (snap) { |
880 | ERROR("LVM snapshot from %s backing store is not supported", | |
881 | orig->type); | |
882 | return -1; | |
883 | } | |
31a95fec SH |
884 | vg = default_lvm_vg(); |
885 | len = strlen("/dev/") + strlen(vg) + strlen(cname) + 2; | |
ca52dcb5 SH |
886 | if ((new->src = malloc(len)) == NULL) |
887 | return -1; | |
31a95fec | 888 | ret = snprintf(new->src, len, "/dev/%s/%s", vg, cname); |
ca52dcb5 SH |
889 | if (ret < 0 || ret >= len) |
890 | return -1; | |
891 | } else { | |
892 | new->src = dir_new_path(orig->src, oldname, cname, oldpath, lxcpath); | |
893 | if (!new->src) | |
894 | return -1; | |
9be53773 SH |
895 | } |
896 | ||
897 | if (orig->data) { | |
898 | new->data = strdup(orig->data); | |
899 | if (!new->data) | |
900 | return -1; | |
901 | } | |
ca52dcb5 SH |
902 | |
903 | len = strlen(lxcpath) + strlen(cname) + strlen("rootfs") + 3; | |
904 | new->dest = malloc(len); | |
9be53773 SH |
905 | if (!new->dest) |
906 | return -1; | |
ca52dcb5 SH |
907 | ret = snprintf(new->dest, len, "%s/%s/rootfs", lxcpath, cname); |
908 | if (ret < 0 || ret >= len) | |
9be53773 | 909 | return -1; |
ca52dcb5 | 910 | if (mkdir_p(new->dest, 0755) < 0) |
9be53773 SH |
911 | return -1; |
912 | ||
ca52dcb5 | 913 | if (is_blktype(orig)) { |
eddaaafd | 914 | if (!newsize && blk_getsize(orig, &size) < 0) { |
ca52dcb5 SH |
915 | ERROR("Error getting size of %s", orig->src); |
916 | return -1; | |
917 | } | |
918 | if (detect_fs(orig, fstype, 100) < 0) { | |
919 | INFO("could not find fstype for %s, using ext3", orig->src); | |
920 | return -1; | |
921 | } | |
922 | } else { | |
923 | sprintf(fstype, "ext3"); | |
924 | if (!newsize) | |
925 | size = 1000000000; // default to 1G | |
9be53773 | 926 | } |
ca52dcb5 | 927 | |
9be53773 SH |
928 | if (snap) { |
929 | if (lvm_snapshot(orig->src, new->src, size) < 0) { | |
930 | ERROR("could not create %s snapshot of %s", new->src, orig->src); | |
931 | return -1; | |
932 | } | |
933 | } else { | |
1897e3bc | 934 | if (do_lvm_create(new->src, size) < 0) { |
9be53773 SH |
935 | ERROR("Error creating new lvm blockdev"); |
936 | return -1; | |
937 | } | |
9be53773 SH |
938 | if (do_mkfs(new->src, fstype) < 0) { |
939 | ERROR("Error creating filesystem type %s on %s", fstype, | |
940 | new->src); | |
941 | return -1; | |
942 | } | |
943 | } | |
944 | ||
945 | return 0; | |
946 | } | |
947 | ||
60bf62d4 SH |
948 | static int lvm_destroy(struct bdev *orig) |
949 | { | |
950 | pid_t pid; | |
951 | ||
952 | if ((pid = fork()) < 0) | |
953 | return -1; | |
954 | if (!pid) { | |
955 | execlp("lvremove", "lvremove", "-f", orig->src, NULL); | |
956 | exit(1); | |
957 | } | |
958 | return wait_for_pid(pid); | |
959 | } | |
960 | ||
eddaaafd SH |
961 | #define DEFAULT_FS_SIZE 1024000000 |
962 | #define DEFAULT_FSTYPE "ext3" | |
1897e3bc SH |
963 | static int lvm_create(struct bdev *bdev, const char *dest, const char *n, |
964 | struct bdev_specs *specs) | |
965 | { | |
966 | const char *vg, *fstype, *lv = n; | |
967 | unsigned long sz; | |
968 | int ret, len; | |
969 | ||
970 | if (!specs) | |
971 | return -1; | |
972 | ||
973 | vg = specs->u.lvm.vg; | |
974 | if (!vg) | |
975 | vg = default_lvm_vg(); | |
976 | ||
977 | /* /dev/$vg/$lv */ | |
978 | if (specs->u.lvm.lv) | |
979 | lv = specs->u.lvm.lv; | |
980 | len = strlen(vg) + strlen(lv) + 7; | |
981 | bdev->src = malloc(len); | |
982 | if (!bdev->src) | |
983 | return -1; | |
984 | ||
985 | ret = snprintf(bdev->src, len, "/dev/%s/%s", vg, lv); | |
986 | if (ret < 0 || ret >= len) | |
987 | return -1; | |
988 | ||
989 | // lvm.fssize is in bytes. | |
990 | sz = specs->u.lvm.fssize; | |
991 | if (!sz) | |
eddaaafd | 992 | sz = DEFAULT_FS_SIZE; |
1897e3bc SH |
993 | |
994 | INFO("Error creating new lvm blockdev %s size %lu", bdev->src, sz); | |
995 | if (do_lvm_create(bdev->src, sz) < 0) { | |
996 | ERROR("Error creating new lvm blockdev %s size %lu", bdev->src, sz); | |
997 | return -1; | |
998 | } | |
999 | ||
1000 | fstype = specs->u.lvm.fstype; | |
1001 | if (!fstype) | |
eddaaafd | 1002 | fstype = DEFAULT_FSTYPE; |
1897e3bc SH |
1003 | if (do_mkfs(bdev->src, fstype) < 0) { |
1004 | ERROR("Error creating filesystem type %s on %s", fstype, | |
1005 | bdev->src); | |
1006 | return -1; | |
1007 | } | |
1008 | if (!(bdev->dest = strdup(dest))) | |
1009 | return -1; | |
1010 | ||
1011 | if (mkdir_p(bdev->dest, 0755) < 0) { | |
1012 | ERROR("Error creating %s\n", bdev->dest); | |
1013 | return -1; | |
1014 | } | |
1015 | ||
1016 | return 0; | |
1017 | } | |
1018 | ||
9be53773 SH |
1019 | struct bdev_ops lvm_ops = { |
1020 | .detect = &lvm_detect, | |
1021 | .mount = &lvm_mount, | |
1022 | .umount = &lvm_umount, | |
1023 | .clone_paths = &lvm_clonepaths, | |
60bf62d4 | 1024 | .destroy = &lvm_destroy, |
1897e3bc | 1025 | .create = &lvm_create, |
9be53773 SH |
1026 | }; |
1027 | ||
1028 | // | |
1029 | // btrfs ops | |
1030 | // | |
1031 | ||
1032 | struct btrfs_ioctl_space_info { | |
1033 | unsigned long long flags; | |
1034 | unsigned long long total_bytes; | |
1035 | unsigned long long used_bytes; | |
1036 | }; | |
1037 | ||
1038 | struct btrfs_ioctl_space_args { | |
1039 | unsigned long long space_slots; | |
1040 | unsigned long long total_spaces; | |
1041 | struct btrfs_ioctl_space_info spaces[0]; | |
1042 | }; | |
1043 | ||
1044 | #define BTRFS_IOCTL_MAGIC 0x94 | |
1045 | #define BTRFS_IOC_SUBVOL_GETFLAGS _IOR(BTRFS_IOCTL_MAGIC, 25, unsigned long long) | |
1046 | #define BTRFS_IOC_SPACE_INFO _IOWR(BTRFS_IOCTL_MAGIC, 20, \ | |
1047 | struct btrfs_ioctl_space_args) | |
1048 | ||
1897e3bc | 1049 | static bool is_btrfs_fs(const char *path) |
9be53773 | 1050 | { |
9be53773 SH |
1051 | int fd, ret; |
1052 | struct btrfs_ioctl_space_args sargs; | |
1053 | ||
1054 | // make sure this is a btrfs filesystem | |
1055 | fd = open(path, O_RDONLY); | |
1056 | if (fd < 0) | |
1897e3bc | 1057 | return false; |
9be53773 SH |
1058 | sargs.space_slots = 0; |
1059 | sargs.total_spaces = 0; | |
1060 | ret = ioctl(fd, BTRFS_IOC_SPACE_INFO, &sargs); | |
1061 | close(fd); | |
1062 | if (ret < 0) | |
1897e3bc SH |
1063 | return false; |
1064 | ||
1065 | return true; | |
1066 | } | |
1067 | ||
1068 | static int btrfs_detect(const char *path) | |
1069 | { | |
1070 | struct stat st; | |
1071 | int ret; | |
1072 | ||
1073 | if (!is_btrfs_fs(path)) | |
9be53773 SH |
1074 | return 0; |
1075 | ||
1076 | // and make sure it's a subvolume. | |
1077 | ret = stat(path, &st); | |
1078 | if (ret < 0) | |
1079 | return 0; | |
1080 | ||
1081 | if (st.st_ino == 256 && S_ISDIR(st.st_mode)) | |
1082 | return 1; | |
1083 | ||
1084 | return 0; | |
1085 | } | |
1086 | ||
60bf62d4 | 1087 | static int btrfs_mount(struct bdev *bdev) |
9be53773 SH |
1088 | { |
1089 | if (strcmp(bdev->type, "btrfs")) | |
1090 | return -22; | |
1091 | if (!bdev->src || !bdev->dest) | |
1092 | return -22; | |
1093 | return mount(bdev->src, bdev->dest, "bind", MS_BIND | MS_REC, NULL); | |
1094 | } | |
1095 | ||
60bf62d4 | 1096 | static int btrfs_umount(struct bdev *bdev) |
9be53773 SH |
1097 | { |
1098 | if (strcmp(bdev->type, "btrfs")) | |
1099 | return -22; | |
1100 | if (!bdev->src || !bdev->dest) | |
1101 | return -22; | |
1102 | return umount(bdev->dest); | |
1103 | } | |
1104 | ||
1105 | #define BTRFS_SUBVOL_NAME_MAX 4039 | |
1106 | #define BTRFS_PATH_NAME_MAX 4087 | |
1107 | ||
1108 | struct btrfs_ioctl_vol_args { | |
1109 | signed long long fd; | |
1110 | char name[BTRFS_PATH_NAME_MAX + 1]; | |
1111 | }; | |
1112 | ||
1113 | #define BTRFS_IOCTL_MAGIC 0x94 | |
1114 | #define BTRFS_IOC_SUBVOL_CREATE_V2 _IOW(BTRFS_IOCTL_MAGIC, 24, \ | |
1115 | struct btrfs_ioctl_vol_args_v2) | |
1116 | #define BTRFS_IOC_SNAP_CREATE_V2 _IOW(BTRFS_IOCTL_MAGIC, 23, \ | |
1117 | struct btrfs_ioctl_vol_args_v2) | |
1118 | #define BTRFS_IOC_SUBVOL_CREATE _IOW(BTRFS_IOCTL_MAGIC, 14, \ | |
1119 | struct btrfs_ioctl_vol_args) | |
60bf62d4 SH |
1120 | #define BTRFS_IOC_SNAP_DESTROY _IOW(BTRFS_IOCTL_MAGIC, 15, \ |
1121 | struct btrfs_ioctl_vol_args) | |
9be53773 SH |
1122 | |
1123 | #define BTRFS_QGROUP_INHERIT_SET_LIMITS (1ULL << 0) | |
1124 | ||
1125 | struct btrfs_ioctl_vol_args_v2 { | |
1126 | signed long long fd; | |
1127 | unsigned long long transid; | |
1128 | unsigned long long flags; | |
1129 | union { | |
1130 | struct { | |
1131 | unsigned long long size; | |
1132 | //struct btrfs_qgroup_inherit *qgroup_inherit; | |
1133 | void *qgroup_inherit; | |
1134 | }; | |
1135 | unsigned long long unused[4]; | |
1136 | }; | |
1137 | char name[BTRFS_SUBVOL_NAME_MAX + 1]; | |
1138 | }; | |
1139 | ||
1140 | static int btrfs_subvolume_create(const char *path) | |
1141 | { | |
1142 | int ret, fd = -1; | |
1143 | struct btrfs_ioctl_vol_args args; | |
1144 | char *p, *newfull = strdup(path); | |
1145 | ||
1146 | if (!newfull) { | |
1147 | ERROR("Error: out of memory"); | |
1148 | return -1; | |
1149 | } | |
1150 | ||
c32981c3 | 1151 | p = strrchr(newfull, '/'); |
9be53773 SH |
1152 | if (!p) { |
1153 | ERROR("bad path: %s", path); | |
1154 | return -1; | |
1155 | } | |
1156 | *p = '\0'; | |
1157 | ||
1158 | if ((fd = open(newfull, O_RDONLY)) < 0) { | |
1159 | ERROR("Error opening %s", newfull); | |
1160 | free(newfull); | |
1161 | return -1; | |
1162 | } | |
1163 | ||
1164 | memset(&args, 0, sizeof(args)); | |
1165 | strncpy(args.name, p+1, BTRFS_SUBVOL_NAME_MAX); | |
1166 | args.name[BTRFS_SUBVOL_NAME_MAX-1] = 0; | |
1167 | ret = ioctl(fd, BTRFS_IOC_SUBVOL_CREATE, &args); | |
1168 | INFO("btrfs: snapshot create ioctl returned %d", ret); | |
9be53773 SH |
1169 | |
1170 | free(newfull); | |
1171 | close(fd); | |
1172 | return ret; | |
1173 | } | |
1174 | ||
1175 | static int btrfs_snapshot(const char *orig, const char *new) | |
1176 | { | |
1177 | int fd = -1, fddst = -1, ret = -1; | |
1178 | struct btrfs_ioctl_vol_args_v2 args; | |
1179 | char *newdir, *newname, *newfull = NULL; | |
1180 | ||
1181 | newfull = strdup(new); | |
1182 | if (!newfull) { | |
1183 | ERROR("Error: out of memory"); | |
1184 | goto out; | |
1185 | } | |
1186 | // make sure the directory doesn't already exist | |
1187 | if (rmdir(newfull) < 0 && errno != -ENOENT) { | |
1188 | SYSERROR("Error removing empty new rootfs"); | |
1189 | goto out; | |
1190 | } | |
1191 | newname = basename(newfull); | |
1192 | newdir = dirname(newfull); | |
1193 | fd = open(orig, O_RDONLY); | |
1194 | if (fd < 0) { | |
1195 | SYSERROR("Error opening original rootfs %s", orig); | |
1196 | goto out; | |
1197 | } | |
1198 | fddst = open(newdir, O_RDONLY); | |
1199 | if (fddst < 0) { | |
1200 | SYSERROR("Error opening new container dir %s", newdir); | |
1201 | goto out; | |
1202 | } | |
1203 | ||
1204 | memset(&args, 0, sizeof(args)); | |
1205 | args.fd = fd; | |
1206 | strncpy(args.name, newname, BTRFS_SUBVOL_NAME_MAX); | |
1207 | args.name[BTRFS_SUBVOL_NAME_MAX-1] = 0; | |
1208 | ret = ioctl(fddst, BTRFS_IOC_SNAP_CREATE_V2, &args); | |
1209 | INFO("btrfs: snapshot create ioctl returned %d", ret); | |
1210 | ||
1211 | out: | |
1212 | if (fddst != -1) | |
1213 | close(fddst); | |
1214 | if (fd != -1) | |
1215 | close(fd); | |
1216 | if (newfull) | |
1217 | free(newfull); | |
1218 | return ret; | |
1219 | } | |
1220 | ||
1221 | static int btrfs_clonepaths(struct bdev *orig, struct bdev *new, const char *oldname, | |
1222 | const char *cname, const char *oldpath, const char *lxcpath, int snap, | |
1223 | unsigned long newsize) | |
1224 | { | |
1225 | if (!orig->dest || !orig->src) | |
1226 | return -1; | |
1227 | ||
1228 | if (strcmp(orig->type, "btrfs")) { | |
ca52dcb5 SH |
1229 | int len, ret; |
1230 | if (snap) { | |
1231 | ERROR("btrfs snapshot from %s backing store is not supported", | |
1232 | orig->type); | |
1233 | return -1; | |
1234 | } | |
1235 | len = strlen(lxcpath) + strlen(cname) + strlen("rootfs") + 3; | |
1236 | new->src = malloc(len); | |
1237 | if (!new->src) | |
1238 | return -1; | |
1239 | ret = snprintf(new->src, len, "%s/%s/rootfs", lxcpath, cname); | |
1240 | if (ret < 0 || ret >= len) | |
1241 | return -1; | |
1242 | } else { | |
1243 | // in case rootfs is in custom path, reuse it | |
1244 | if ((new->src = dir_new_path(orig->src, oldname, cname, oldpath, lxcpath)) == NULL) | |
1245 | return -1; | |
9be53773 | 1246 | |
ca52dcb5 | 1247 | } |
9be53773 | 1248 | |
ca52dcb5 | 1249 | if ((new->dest = strdup(new->src)) == NULL) |
9be53773 SH |
1250 | return -1; |
1251 | ||
1252 | if (orig->data && (new->data = strdup(orig->data)) == NULL) | |
1253 | return -1; | |
1254 | ||
1255 | if (snap) | |
1256 | return btrfs_snapshot(orig->dest, new->dest); | |
1257 | ||
1258 | if (rmdir(new->dest) < 0 && errno != -ENOENT) { | |
1259 | SYSERROR("removing %s\n", new->dest); | |
1260 | return -1; | |
1261 | } | |
1262 | ||
1263 | return btrfs_subvolume_create(new->dest); | |
1264 | } | |
1265 | ||
60bf62d4 SH |
1266 | static int btrfs_destroy(struct bdev *orig) |
1267 | { | |
1268 | int ret, fd = -1; | |
1269 | struct btrfs_ioctl_vol_args args; | |
1270 | char *path = orig->src; | |
1271 | char *p, *newfull = strdup(path); | |
1272 | ||
1273 | if (!newfull) { | |
1274 | ERROR("Error: out of memory"); | |
1275 | return -1; | |
1276 | } | |
1277 | ||
c32981c3 | 1278 | p = strrchr(newfull, '/'); |
60bf62d4 SH |
1279 | if (!p) { |
1280 | ERROR("bad path: %s", path); | |
1281 | return -1; | |
1282 | } | |
1283 | *p = '\0'; | |
1284 | ||
1285 | if ((fd = open(newfull, O_RDONLY)) < 0) { | |
1286 | ERROR("Error opening %s", newfull); | |
1287 | free(newfull); | |
1288 | return -1; | |
1289 | } | |
1290 | ||
1291 | memset(&args, 0, sizeof(args)); | |
1292 | strncpy(args.name, p+1, BTRFS_SUBVOL_NAME_MAX); | |
1293 | args.name[BTRFS_SUBVOL_NAME_MAX-1] = 0; | |
1294 | ret = ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &args); | |
1295 | INFO("btrfs: snapshot create ioctl returned %d", ret); | |
1296 | ||
1297 | free(newfull); | |
1298 | close(fd); | |
1299 | return ret; | |
1300 | } | |
1301 | ||
1897e3bc SH |
1302 | static int btrfs_create(struct bdev *bdev, const char *dest, const char *n, |
1303 | struct bdev_specs *specs) | |
1304 | { | |
1305 | bdev->src = strdup(dest); | |
1306 | bdev->dest = strdup(dest); | |
1307 | if (!bdev->src || !bdev->dest) | |
1308 | return -1; | |
1309 | return btrfs_subvolume_create(bdev->dest); | |
1310 | } | |
1311 | ||
9be53773 SH |
1312 | struct bdev_ops btrfs_ops = { |
1313 | .detect = &btrfs_detect, | |
1314 | .mount = &btrfs_mount, | |
1315 | .umount = &btrfs_umount, | |
1316 | .clone_paths = &btrfs_clonepaths, | |
60bf62d4 | 1317 | .destroy = &btrfs_destroy, |
1897e3bc | 1318 | .create = &btrfs_create, |
9be53773 SH |
1319 | }; |
1320 | ||
eddaaafd SH |
1321 | // |
1322 | // loopback dev ops | |
1323 | // | |
1324 | static int loop_detect(const char *path) | |
1325 | { | |
1326 | if (strncmp(path, "loop:", 5) == 0) | |
1327 | return 1; | |
1328 | return 0; | |
1329 | } | |
1330 | ||
1331 | static int find_free_loopdev(int *retfd, char *namep) | |
1332 | { | |
1333 | struct dirent dirent, *direntp; | |
1334 | struct loop_info64 lo; | |
1335 | DIR *dir; | |
1336 | int fd = -1; | |
1337 | ||
1338 | if (!(dir = opendir("/dev"))) { | |
1339 | SYSERROR("Error opening /dev"); | |
1340 | return -1; | |
1341 | } | |
1342 | while (!readdir_r(dir, &dirent, &direntp)) { | |
1343 | ||
1344 | if (!direntp) | |
1345 | break; | |
1346 | if (strncmp(direntp->d_name, "loop", 4) != 0) | |
1347 | continue; | |
1348 | if ((fd = openat(dirfd(dir), direntp->d_name, O_RDWR)) < 0) | |
1349 | continue; | |
1350 | if (ioctl(fd, LOOP_GET_STATUS64, &lo) == 0 || errno != ENXIO) { | |
1351 | close(fd); | |
1352 | fd = -1; | |
1353 | continue; | |
1354 | } | |
1355 | // We can use this fd | |
1356 | snprintf(namep, 100, "/dev/%s", direntp->d_name); | |
1357 | break; | |
1358 | } | |
ca697342 | 1359 | closedir(dir); |
eddaaafd SH |
1360 | if (fd == -1) { |
1361 | ERROR("No loop device found"); | |
1362 | return -1; | |
1363 | } | |
eddaaafd SH |
1364 | |
1365 | *retfd = fd; | |
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | static int loop_mount(struct bdev *bdev) | |
1370 | { | |
1371 | int lfd, ffd = -1, ret = -1; | |
1372 | struct loop_info64 lo; | |
1373 | char loname[100]; | |
1374 | ||
1375 | if (strcmp(bdev->type, "loop")) | |
1376 | return -22; | |
1377 | if (!bdev->src || !bdev->dest) | |
1378 | return -22; | |
1379 | if (find_free_loopdev(&lfd, loname) < 0) | |
1380 | return -22; | |
1381 | ||
1382 | if ((ffd = open(bdev->src + 5, O_RDWR)) < 0) { | |
1383 | SYSERROR("Error opening backing file %s\n", bdev->src); | |
1384 | goto out; | |
1385 | } | |
1386 | ||
1387 | if (ioctl(lfd, LOOP_SET_FD, ffd) < 0) { | |
1388 | SYSERROR("Error attaching backing file to loop dev"); | |
1389 | goto out; | |
1390 | } | |
1391 | memset(&lo, 0, sizeof(lo)); | |
1392 | lo.lo_flags = LO_FLAGS_AUTOCLEAR; | |
1393 | if (ioctl(lfd, LOOP_SET_STATUS64, &lo) < 0) { | |
1394 | SYSERROR("Error setting autoclear on loop dev\n"); | |
1395 | goto out; | |
1396 | } | |
1397 | ||
1398 | ret = mount_unknow_fs(loname, bdev->dest, 0); | |
1399 | if (ret < 0) | |
1400 | ERROR("Error mounting %s\n", bdev->src); | |
1401 | else | |
1402 | bdev->lofd = lfd; | |
1403 | ||
1404 | out: | |
1405 | if (ffd > -1) | |
1406 | close(ffd); | |
1407 | if (ret < 0) { | |
1408 | close(lfd); | |
1409 | bdev->lofd = -1; | |
1410 | } | |
1411 | return ret; | |
1412 | } | |
1413 | ||
1414 | static int loop_umount(struct bdev *bdev) | |
1415 | { | |
1416 | int ret; | |
1417 | ||
1418 | if (strcmp(bdev->type, "loop")) | |
1419 | return -22; | |
1420 | if (!bdev->src || !bdev->dest) | |
1421 | return -22; | |
1422 | ret = umount(bdev->dest); | |
1423 | if (bdev->lofd >= 0) { | |
1424 | close(bdev->lofd); | |
1425 | bdev->lofd = -1; | |
1426 | } | |
1427 | return ret; | |
1428 | } | |
1429 | ||
1430 | static int do_loop_create(const char *path, unsigned long size, const char *fstype) | |
1431 | { | |
1432 | int fd; | |
1433 | // create the new loopback file. | |
1434 | fd = creat(path, S_IRUSR|S_IWUSR); | |
1435 | if (fd < 0) | |
1436 | return -1; | |
1437 | if (lseek(fd, size, SEEK_SET) < 0) { | |
1438 | SYSERROR("Error seeking to set new loop file size"); | |
1439 | close(fd); | |
1440 | return -1; | |
1441 | } | |
1442 | if (write(fd, "1", 1) != 1) { | |
1443 | SYSERROR("Error creating new loop file"); | |
1444 | close(fd); | |
1445 | return -1; | |
1446 | } | |
1447 | if (close(fd) < 0) { | |
1448 | SYSERROR("Error closing new loop file"); | |
1449 | return -1; | |
1450 | } | |
1451 | ||
1452 | // create an fs in the loopback file | |
1453 | if (do_mkfs(path, fstype) < 0) { | |
1454 | ERROR("Error creating filesystem type %s on %s", fstype, | |
1455 | path); | |
1456 | return -1; | |
1457 | } | |
1458 | ||
1459 | return 0; | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * No idea what the original blockdev will be called, but the copy will be | |
1464 | * called $lxcpath/$lxcname/rootdev | |
1465 | */ | |
1466 | static int loop_clonepaths(struct bdev *orig, struct bdev *new, const char *oldname, | |
1467 | const char *cname, const char *oldpath, const char *lxcpath, int snap, | |
1468 | unsigned long newsize) | |
1469 | { | |
1470 | char fstype[100]; | |
1471 | unsigned long size = newsize; | |
1472 | int len, ret; | |
1473 | char *srcdev; | |
1474 | ||
1475 | if (snap) { | |
1476 | ERROR("loop devices cannot be snapshotted."); | |
1477 | return -1; | |
1478 | } | |
1479 | ||
1480 | if (!orig->dest || !orig->src) | |
1481 | return -1; | |
1482 | ||
1483 | len = strlen(lxcpath) + strlen(cname) + strlen("rootdev") + 3; | |
1484 | srcdev = alloca(len); | |
1485 | ret = snprintf(srcdev, len, "%s/%s/rootdev", lxcpath, cname); | |
1486 | if (ret < 0 || ret >= len) | |
1487 | return -1; | |
1488 | ||
1489 | new->src = malloc(len + 5); | |
1490 | if (!new->src) | |
1491 | return -1; | |
1492 | ret = snprintf(new->src, len + 5, "loop:%s", srcdev); | |
1493 | if (ret < 0 || ret >= len + 5) | |
1494 | return -1; | |
1495 | ||
1496 | new->dest = malloc(len); | |
1497 | if (!new->dest) | |
1498 | return -1; | |
1499 | ret = snprintf(new->dest, len, "%s/%s/rootfs", lxcpath, cname); | |
1500 | if (ret < 0 || ret >= len) | |
1501 | return -1; | |
1502 | ||
1503 | // it's tempting to say: if orig->src == loopback and !newsize, then | |
1504 | // copy the loopback file. However, we'd have to make sure to | |
1505 | // correctly keep holes! So punt for now. | |
1506 | ||
1507 | if (is_blktype(orig)) { | |
1508 | if (!newsize && blk_getsize(orig, &size) < 0) { | |
1509 | ERROR("Error getting size of %s", orig->src); | |
1510 | return -1; | |
1511 | } | |
1512 | if (detect_fs(orig, fstype, 100) < 0) { | |
1513 | INFO("could not find fstype for %s, using %s", orig->src, | |
1514 | DEFAULT_FSTYPE); | |
1515 | return -1; | |
1516 | } | |
1517 | } else { | |
1518 | sprintf(fstype, "%s", DEFAULT_FSTYPE); | |
1519 | if (!newsize) | |
1520 | size = DEFAULT_FS_SIZE; // default to 1G | |
1521 | } | |
1522 | return do_loop_create(srcdev, size, fstype); | |
1523 | } | |
1524 | ||
1525 | static int loop_create(struct bdev *bdev, const char *dest, const char *n, | |
1526 | struct bdev_specs *specs) | |
1527 | { | |
1528 | const char *fstype; | |
1529 | unsigned long sz; | |
1530 | int ret, len; | |
1531 | char *srcdev; | |
1532 | ||
1533 | if (!specs) | |
1534 | return -1; | |
1535 | ||
1536 | // dest is passed in as $lxcpath / $lxcname / rootfs | |
1537 | // srcdev will be: $lxcpath / $lxcname / rootdev | |
1538 | // src will be 'loop:$srcdev' | |
1539 | len = strlen(dest) + 2; | |
1540 | srcdev = alloca(len); | |
1541 | ||
1542 | ret = snprintf(srcdev, len, "%s", dest); | |
1543 | if (ret < 0 || ret >= len) | |
1544 | return -1; | |
1545 | sprintf(srcdev + len - 4, "dev"); | |
1546 | ||
1547 | bdev->src = malloc(len + 5); | |
1548 | if (!bdev->src) | |
1549 | return -1; | |
1550 | ret = snprintf(bdev->src, len + 5, "loop:%s", srcdev); | |
1551 | if (ret < 0 || ret >= len + 5) | |
1552 | return -1; | |
1553 | ||
1554 | sz = specs->u.loop.fssize; | |
1555 | if (!sz) | |
1556 | sz = DEFAULT_FS_SIZE; | |
1557 | ||
1558 | fstype = specs->u.loop.fstype; | |
1559 | if (!fstype) | |
1560 | fstype = DEFAULT_FSTYPE; | |
1561 | ||
1562 | if (!(bdev->dest = strdup(dest))) | |
1563 | return -1; | |
1564 | ||
1565 | if (mkdir_p(bdev->dest, 0755) < 0) { | |
1566 | ERROR("Error creating %s\n", bdev->dest); | |
1567 | return -1; | |
1568 | } | |
1569 | ||
1570 | return do_loop_create(srcdev, sz, fstype); | |
1571 | } | |
1572 | ||
1573 | static int loop_destroy(struct bdev *orig) | |
1574 | { | |
1575 | return unlink(orig->src + 5); | |
1576 | } | |
1577 | ||
1578 | struct bdev_ops loop_ops = { | |
1579 | .detect = &loop_detect, | |
1580 | .mount = &loop_mount, | |
1581 | .umount = &loop_umount, | |
1582 | .clone_paths = &loop_clonepaths, | |
1583 | .destroy = &loop_destroy, | |
1584 | .create = &loop_create, | |
1585 | }; | |
1586 | ||
9be53773 SH |
1587 | // |
1588 | // overlayfs ops | |
1589 | // | |
1590 | ||
1591 | static int overlayfs_detect(const char *path) | |
1592 | { | |
1593 | if (strncmp(path, "overlayfs:", 10) == 0) | |
1594 | return 1; // take their word for it | |
1595 | return 0; | |
1596 | } | |
1597 | ||
1598 | // | |
1599 | // XXXXXXX plain directory bind mount ops | |
1600 | // | |
60bf62d4 | 1601 | static int overlayfs_mount(struct bdev *bdev) |
9be53773 SH |
1602 | { |
1603 | char *options, *dup, *lower, *upper; | |
1604 | int len; | |
1605 | int ret; | |
1606 | ||
1607 | if (strcmp(bdev->type, "overlayfs")) | |
1608 | return -22; | |
1609 | if (!bdev->src || !bdev->dest) | |
1610 | return -22; | |
1611 | ||
1612 | // separately mount it first | |
1613 | // mount -t overlayfs -oupperdir=${upper},lowerdir=${lower} lower dest | |
d74325c4 SG |
1614 | dup = alloca(strlen(bdev->src)+1); |
1615 | strcpy(dup, bdev->src); | |
9be53773 SH |
1616 | if (!(lower = index(dup, ':'))) |
1617 | return -22; | |
1618 | if (!(upper = index(++lower, ':'))) | |
1619 | return -22; | |
1620 | *upper = '\0'; | |
1621 | upper++; | |
1622 | ||
1623 | // TODO We should check whether bdev->src is a blockdev, and if so | |
1624 | // but for now, only support overlays of a basic directory | |
1625 | ||
1626 | len = strlen(lower) + strlen(upper) + strlen("upperdir=,lowerdir=") + 1; | |
1627 | options = alloca(len); | |
1628 | ret = snprintf(options, len, "upperdir=%s,lowerdir=%s", upper, lower); | |
1629 | if (ret < 0 || ret >= len) | |
1630 | return -1; | |
1631 | ret = mount(lower, bdev->dest, "overlayfs", MS_MGC_VAL, options); | |
1632 | if (ret < 0) | |
1633 | SYSERROR("overlayfs: error mounting %s onto %s options %s", | |
1634 | lower, bdev->dest, options); | |
1635 | else | |
1636 | INFO("overlayfs: mounted %s onto %s options %s", | |
1637 | lower, bdev->dest, options); | |
1638 | return ret; | |
1639 | } | |
1640 | ||
60bf62d4 | 1641 | static int overlayfs_umount(struct bdev *bdev) |
9be53773 SH |
1642 | { |
1643 | if (strcmp(bdev->type, "overlayfs")) | |
1644 | return -22; | |
1645 | if (!bdev->src || !bdev->dest) | |
1646 | return -22; | |
1647 | return umount(bdev->dest); | |
1648 | } | |
1649 | ||
1650 | static int overlayfs_clonepaths(struct bdev *orig, struct bdev *new, const char *oldname, | |
1651 | const char *cname, const char *oldpath, const char *lxcpath, int snap, | |
1652 | unsigned long newsize) | |
1653 | { | |
1654 | if (!snap) { | |
1655 | ERROR("overlayfs is only for snapshot clones"); | |
1656 | return -22; | |
1657 | } | |
1658 | ||
1659 | if (!orig->src || !orig->dest) | |
1660 | return -1; | |
1661 | ||
1662 | new->dest = dir_new_path(orig->dest, oldname, cname, oldpath, lxcpath); | |
1663 | if (!new->dest) | |
1664 | return -1; | |
1665 | if (mkdir_p(new->dest, 0755) < 0) | |
1666 | return -1; | |
1667 | ||
1668 | if (strcmp(orig->type, "dir") == 0) { | |
1669 | char *delta; | |
1670 | int ret, len; | |
5ca6c34b | 1671 | |
9be53773 SH |
1672 | // if we have /var/lib/lxc/c2/rootfs, then delta will be |
1673 | // /var/lib/lxc/c2/delta0 | |
1674 | delta = strdup(new->dest); | |
1675 | if (!delta) { | |
1676 | return -1; | |
1677 | } | |
1678 | if (strlen(delta) < 6) { | |
1679 | free(delta); | |
1680 | return -22; | |
1681 | } | |
1682 | strcpy(&delta[strlen(delta)-6], "delta0"); | |
1683 | if ((ret = mkdir(delta, 0755)) < 0) { | |
1684 | SYSERROR("error: mkdir %s", delta); | |
1685 | free(delta); | |
1686 | return -1; | |
1687 | } | |
1688 | ||
1689 | // the src will be 'overlayfs:lowerdir:upperdir' | |
1690 | len = strlen(delta) + strlen(orig->src) + 12; | |
1691 | new->src = malloc(len); | |
1692 | if (!new->src) { | |
1693 | free(delta); | |
1694 | return -ENOMEM; | |
1695 | } | |
1696 | ret = snprintf(new->src, len, "overlayfs:%s:%s", orig->src, delta); | |
1697 | free(delta); | |
1698 | if (ret < 0 || ret >= len) | |
1699 | return -ENOMEM; | |
9be53773 SH |
1700 | } else if (strcmp(orig->type, "overlayfs") == 0) { |
1701 | // What exactly do we want to do here? | |
1702 | // I think we want to use the original lowerdir, with a | |
1703 | // private delta which is originally rsynced from the | |
1704 | // original delta | |
1705 | char *osrc, *odelta, *nsrc, *ndelta; | |
1706 | int len, ret; | |
1707 | if (!(osrc = strdup(orig->src))) | |
1708 | return -22; | |
1709 | nsrc = index(osrc, ':') + 1; | |
1710 | if (nsrc != osrc + 10 || (odelta = index(nsrc, ':')) == NULL) { | |
1711 | free(osrc); | |
1712 | return -22; | |
1713 | } | |
1714 | *odelta = '\0'; | |
1715 | odelta++; | |
1716 | ndelta = dir_new_path(odelta, oldname, cname, oldpath, lxcpath); | |
1717 | if (!ndelta) { | |
1718 | free(osrc); | |
1719 | return -ENOMEM; | |
1720 | } | |
1721 | if (do_rsync(odelta, ndelta) < 0) { | |
91c908ee DE |
1722 | free(osrc); |
1723 | free(ndelta); | |
9be53773 SH |
1724 | ERROR("copying overlayfs delta"); |
1725 | return -1; | |
1726 | } | |
1727 | len = strlen(nsrc) + strlen(ndelta) + 12; | |
1728 | new->src = malloc(len); | |
1729 | if (!new->src) { | |
1730 | free(osrc); | |
1731 | free(ndelta); | |
1732 | return -ENOMEM; | |
1733 | } | |
1734 | ret = snprintf(new->src, len, "overlayfs:%s:%s", nsrc, ndelta); | |
1735 | free(osrc); | |
1736 | free(ndelta); | |
1737 | if (ret < 0 || ret >= len) | |
1738 | return -ENOMEM; | |
375c2258 SH |
1739 | } else { |
1740 | ERROR("overlayfs clone of %s container is not yet supported", | |
1741 | orig->type); | |
1742 | // Note, supporting this will require overlayfs_mount supporting | |
1743 | // mounting of the underlay. No big deal, just needs to be done. | |
1744 | return -1; | |
9be53773 SH |
1745 | } |
1746 | ||
1747 | return 0; | |
1748 | } | |
1749 | ||
60bf62d4 SH |
1750 | int overlayfs_destroy(struct bdev *orig) |
1751 | { | |
1752 | char *upper; | |
1753 | ||
1754 | if (strncmp(orig->src, "overlayfs:", 10) != 0) | |
1755 | return -22; | |
1756 | upper = index(orig->src + 10, ':'); | |
1757 | if (!upper) | |
1758 | return -22; | |
1759 | upper++; | |
1760 | return lxc_rmdir_onedev(upper); | |
1761 | } | |
1762 | ||
1897e3bc SH |
1763 | /* |
1764 | * to say 'lxc-create -t ubuntu -n o1 -B overlayfs' means you want | |
1765 | * $lxcpath/$lxcname/rootfs to have the created container, while all | |
1766 | * changes after starting the container are written to | |
1767 | * $lxcpath/$lxcname/delta0 | |
1768 | */ | |
1769 | static int overlayfs_create(struct bdev *bdev, const char *dest, const char *n, | |
1770 | struct bdev_specs *specs) | |
1771 | { | |
1772 | char *delta; | |
1773 | int ret, len = strlen(dest), newlen; | |
1774 | ||
1775 | if (len < 8 || strcmp(dest+len-7, "/rootfs") != 0) | |
1776 | return -1; | |
1777 | ||
1778 | if (!(bdev->dest = strdup(dest))) { | |
1779 | ERROR("Out of memory"); | |
1780 | return -1; | |
1781 | } | |
1782 | ||
d74325c4 SG |
1783 | delta = alloca(strlen(dest)+1); |
1784 | strcpy(delta, dest); | |
1897e3bc SH |
1785 | strcpy(delta+len-6, "delta0"); |
1786 | ||
1787 | if (mkdir_p(delta, 0755) < 0) { | |
1788 | ERROR("Error creating %s\n", delta); | |
1789 | return -1; | |
1790 | } | |
1791 | ||
1792 | /* overlayfs:lower:upper */ | |
1793 | newlen = (2 * len) + strlen("overlayfs:") + 2; | |
1794 | bdev->src = malloc(newlen); | |
1795 | if (!bdev->src) { | |
1796 | ERROR("Out of memory"); | |
1797 | return -1; | |
1798 | } | |
1799 | ret = snprintf(bdev->src, newlen, "overlayfs:%s:%s", dest, delta); | |
1800 | if (ret < 0 || ret >= newlen) | |
1801 | return -1; | |
1802 | ||
1803 | if (mkdir_p(bdev->dest, 0755) < 0) { | |
1804 | ERROR("Error creating %s\n", bdev->dest); | |
1805 | return -1; | |
1806 | } | |
1807 | ||
1808 | return 0; | |
1809 | } | |
1810 | ||
9be53773 SH |
1811 | struct bdev_ops overlayfs_ops = { |
1812 | .detect = &overlayfs_detect, | |
1813 | .mount = &overlayfs_mount, | |
1814 | .umount = &overlayfs_umount, | |
1815 | .clone_paths = &overlayfs_clonepaths, | |
60bf62d4 | 1816 | .destroy = &overlayfs_destroy, |
1897e3bc | 1817 | .create = &overlayfs_create, |
9be53773 SH |
1818 | }; |
1819 | ||
1820 | struct bdev_type bdevs[] = { | |
3baa76fe | 1821 | {.name = "zfs", .ops = &zfs_ops,}, |
9be53773 SH |
1822 | {.name = "lvm", .ops = &lvm_ops,}, |
1823 | {.name = "btrfs", .ops = &btrfs_ops,}, | |
1824 | {.name = "dir", .ops = &dir_ops,}, | |
1825 | {.name = "overlayfs", .ops = &overlayfs_ops,}, | |
eddaaafd | 1826 | {.name = "loop", .ops = &loop_ops,}, |
9be53773 SH |
1827 | }; |
1828 | ||
1829 | static const size_t numbdevs = sizeof(bdevs) / sizeof(struct bdev_type); | |
1830 | ||
1831 | void bdev_put(struct bdev *bdev) | |
1832 | { | |
1833 | if (bdev->data) | |
1834 | free(bdev->data); | |
1835 | if (bdev->src) | |
1836 | free(bdev->src); | |
1837 | if (bdev->dest) | |
1838 | free(bdev->dest); | |
1839 | free(bdev); | |
1840 | } | |
1841 | ||
1842 | struct bdev *bdev_get(const char *type) | |
1843 | { | |
1844 | int i; | |
1845 | struct bdev *bdev; | |
1846 | ||
1847 | for (i=0; i<numbdevs; i++) { | |
1848 | if (strcmp(bdevs[i].name, type) == 0) | |
1849 | break; | |
1850 | } | |
1851 | if (i == numbdevs) | |
1852 | return NULL; | |
1853 | bdev = malloc(sizeof(struct bdev)); | |
1854 | if (!bdev) | |
1855 | return NULL; | |
1856 | memset(bdev, 0, sizeof(struct bdev)); | |
1857 | bdev->ops = bdevs[i].ops; | |
1858 | bdev->type = bdevs[i].name; | |
1859 | return bdev; | |
1860 | } | |
1861 | ||
1862 | struct bdev *bdev_init(const char *src, const char *dst, const char *data) | |
1863 | { | |
1864 | int i; | |
1865 | struct bdev *bdev; | |
1866 | ||
1867 | for (i=0; i<numbdevs; i++) { | |
1868 | int r; | |
1869 | r = bdevs[i].ops->detect(src); | |
1870 | if (r) | |
1871 | break; | |
1872 | } | |
eddaaafd | 1873 | |
9be53773 SH |
1874 | if (i == numbdevs) |
1875 | return NULL; | |
1876 | bdev = malloc(sizeof(struct bdev)); | |
1877 | if (!bdev) | |
1878 | return NULL; | |
1879 | memset(bdev, 0, sizeof(struct bdev)); | |
1880 | bdev->ops = bdevs[i].ops; | |
1881 | bdev->type = bdevs[i].name; | |
1882 | if (data) | |
1883 | bdev->data = strdup(data); | |
1884 | if (src) | |
1885 | bdev->src = strdup(src); | |
1886 | if (dst) | |
1887 | bdev->dest = strdup(dst); | |
1888 | ||
1889 | return bdev; | |
1890 | } | |
1891 | ||
1892 | /* | |
1893 | * If we're not snaphotting, then bdev_copy becomes a simple case of mount | |
1894 | * the original, mount the new, and rsync the contents. | |
1895 | */ | |
1896 | struct bdev *bdev_copy(const char *src, const char *oldname, const char *cname, | |
1897 | const char *oldpath, const char *lxcpath, const char *bdevtype, | |
dfb31b25 SH |
1898 | int snap, const char *bdevdata, unsigned long newsize, |
1899 | int *needs_rdep) | |
9be53773 SH |
1900 | { |
1901 | struct bdev *orig, *new; | |
1902 | pid_t pid; | |
1903 | ||
1904 | /* if the container name doesn't show up in the rootfs path, then | |
1905 | * we don't know how to come up with a new name | |
1906 | */ | |
1907 | if (strstr(src, oldname) == NULL) { | |
1908 | ERROR("original rootfs path %s doesn't include container name %s", | |
1909 | src, oldname); | |
1910 | return NULL; | |
1911 | } | |
1912 | ||
1913 | orig = bdev_init(src, NULL, NULL); | |
1914 | if (!orig) { | |
1915 | ERROR("failed to detect blockdev type for %s\n", src); | |
1916 | return NULL; | |
1917 | } | |
1918 | ||
1919 | if (!orig->dest) { | |
1920 | int ret; | |
1921 | orig->dest = malloc(MAXPATHLEN); | |
1922 | if (!orig->dest) { | |
1923 | ERROR("out of memory"); | |
1924 | bdev_put(orig); | |
1925 | return NULL; | |
1926 | } | |
1927 | ret = snprintf(orig->dest, MAXPATHLEN, "%s/%s/rootfs", oldpath, oldname); | |
1928 | if (ret < 0 || ret >= MAXPATHLEN) { | |
1929 | ERROR("rootfs path too long"); | |
1930 | bdev_put(orig); | |
1931 | return NULL; | |
1932 | } | |
1933 | } | |
1934 | ||
e3fdf5cc SH |
1935 | /* |
1936 | * If newtype is NULL and snapshot is set, then use overlayfs | |
1937 | */ | |
1938 | if (!bdevtype && snap && strcmp(orig->type , "dir") == 0) | |
1939 | bdevtype = "overlayfs"; | |
1940 | ||
dfb31b25 SH |
1941 | *needs_rdep = 0; |
1942 | if (strcmp(orig->type, "dir") == 0 && | |
1943 | strcmp(bdevtype, "overlayfs") == 0) | |
1944 | *needs_rdep = 1; | |
1945 | ||
9be53773 SH |
1946 | new = bdev_get(bdevtype ? bdevtype : orig->type); |
1947 | if (!new) { | |
1948 | ERROR("no such block device type: %s", bdevtype ? bdevtype : orig->type); | |
1949 | bdev_put(orig); | |
1950 | return NULL; | |
1951 | } | |
1952 | ||
1953 | if (new->ops->clone_paths(orig, new, oldname, cname, oldpath, lxcpath, snap, newsize) < 0) { | |
1954 | ERROR("failed getting pathnames for cloned storage: %s\n", src); | |
1955 | bdev_put(orig); | |
1956 | bdev_put(new); | |
1957 | return NULL; | |
1958 | } | |
1959 | ||
1960 | pid = fork(); | |
1961 | if (pid < 0) { | |
1962 | SYSERROR("fork"); | |
1963 | bdev_put(orig); | |
1964 | bdev_put(new); | |
1965 | return NULL; | |
1966 | } | |
1967 | ||
1968 | if (pid > 0) { | |
1969 | int ret = wait_for_pid(pid); | |
1970 | bdev_put(orig); | |
1971 | if (ret < 0) { | |
1972 | bdev_put(new); | |
1973 | return NULL; | |
1974 | } | |
1975 | return new; | |
1976 | } | |
1977 | ||
1978 | if (unshare(CLONE_NEWNS) < 0) { | |
1979 | SYSERROR("unshare CLONE_NEWNS"); | |
1980 | exit(1); | |
1981 | } | |
1982 | if (snap) | |
1983 | exit(0); | |
1984 | ||
1985 | // If not a snapshot, copy the fs. | |
1986 | if (orig->ops->mount(orig) < 0) { | |
1987 | ERROR("failed mounting %s onto %s\n", src, orig->dest); | |
1988 | exit(1); | |
1989 | } | |
1990 | if (new->ops->mount(new) < 0) { | |
1991 | ERROR("failed mounting %s onto %s\n", new->src, new->dest); | |
1992 | exit(1); | |
1993 | } | |
1994 | if (do_rsync(orig->dest, new->dest) < 0) { | |
1995 | ERROR("rsyncing %s to %s\n", orig->src, new->src); | |
1996 | exit(1); | |
1997 | } | |
1998 | // don't bother umounting, ns exit will do that | |
1999 | ||
2000 | exit(0); | |
2001 | } | |
1897e3bc | 2002 | |
d44e88c2 SH |
2003 | static struct bdev * do_bdev_create(const char *dest, const char *type, |
2004 | const char *cname, struct bdev_specs *specs) | |
2005 | { | |
2006 | struct bdev *bdev = bdev_get(type); | |
2007 | if (!bdev) { | |
2008 | return NULL; | |
2009 | } | |
2010 | ||
2011 | if (bdev->ops->create(bdev, dest, cname, specs) < 0) { | |
2012 | bdev_put(bdev); | |
2013 | return NULL; | |
2014 | } | |
2015 | ||
2016 | return bdev; | |
2017 | } | |
2018 | ||
1897e3bc SH |
2019 | /* |
2020 | * bdev_create: | |
2021 | * Create a backing store for a container. | |
2022 | * If successfull, return a struct bdev *, with the bdev mounted and ready | |
2023 | * for use. Before completing, the caller will need to call the | |
2024 | * umount operation and bdev_put(). | |
2025 | * @dest: the mountpoint (i.e. /var/lib/lxc/$name/rootfs) | |
2026 | * @type: the bdevtype (dir, btrfs, zfs, etc) | |
2027 | * @cname: the container name | |
2028 | * @specs: details about the backing store to create, like fstype | |
2029 | */ | |
2030 | struct bdev *bdev_create(const char *dest, const char *type, | |
2031 | const char *cname, struct bdev_specs *specs) | |
2032 | { | |
2033 | struct bdev *bdev; | |
d44e88c2 | 2034 | char *best_options[] = {"btrfs", "zfs", "lvm", "dir", NULL}; |
1897e3bc | 2035 | |
d3060bd0 | 2036 | if (!type) |
d44e88c2 SH |
2037 | return do_bdev_create(dest, "dir", cname, specs); |
2038 | ||
2039 | if (strcmp(type, "best") == 0) { | |
2040 | int i; | |
2041 | // try for the best backing store type, according to our | |
2042 | // opinionated preferences | |
2043 | for (i=0; best_options[i]; i++) { | |
2044 | if ((bdev = do_bdev_create(dest, best_options[i], cname, specs))) | |
2045 | return bdev; | |
2046 | } | |
2047 | return NULL; // 'dir' should never fail, so this shouldn't happen | |
1897e3bc SH |
2048 | } |
2049 | ||
d44e88c2 SH |
2050 | // -B lvm,dir |
2051 | if (index(type, ',') != NULL) { | |
2052 | char *dup = alloca(strlen(type)+1), *saveptr, *token; | |
2053 | strcpy(dup, type); | |
2054 | for (token = strtok_r(dup, ",", &saveptr); token; | |
2055 | token = strtok_r(NULL, ",", &saveptr)) { | |
2056 | if ((bdev = do_bdev_create(dest, token, cname, specs))) | |
2057 | return bdev; | |
2058 | } | |
1897e3bc SH |
2059 | } |
2060 | ||
d44e88c2 | 2061 | return do_bdev_create(dest, type, cname, specs); |
1897e3bc SH |
2062 | } |
2063 | ||
2064 | char *overlayfs_getlower(char *p) | |
2065 | { | |
2066 | char *p1 = index(p, ':'); | |
2067 | if (p1) | |
2068 | *p1 = '\0'; | |
2069 | return p; | |
2070 | } |