1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 */
26
27 /*
28 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs.
29 * It has the following characteristics:
30 *
31 * - Thread Safe. libzfs_core is accessible concurrently from multiple
32 * threads. This is accomplished primarily by avoiding global data
33 * (e.g. caching). Since it's thread-safe, there is no reason for a
34 * process to have multiple libzfs "instances". Therefore, we store
35 * our few pieces of data (e.g. the file descriptor) in global
36 * variables. The fd is reference-counted so that the libzfs_core
37 * library can be "initialized" multiple times (e.g. by different
38 * consumers within the same process).
39 *
40 * - Committed Interface. The libzfs_core interface will be committed,
41 * therefore consumers can compile against it and be confident that
42 * their code will continue to work on future releases of this code.
43 * Currently, the interface is Evolving (not Committed), but we intend
44 * to commit to it once it is more complete and we determine that it
45 * meets the needs of all consumers.
46 *
47 * - Programatic Error Handling. libzfs_core communicates errors with
48 * defined error numbers, and doesn't print anything to stdout/stderr.
49 *
50 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments
51 * to/from the kernel ioctls. There is generally a 1:1 correspondence
52 * between libzfs_core functions and ioctls to /dev/zfs.
53 *
54 * - Clear Atomicity. Because libzfs_core functions are generally 1:1
55 * with kernel ioctls, and kernel ioctls are general atomic, each
56 * libzfs_core function is atomic. For example, creating multiple
57 * snapshots with a single call to lzc_snapshot() is atomic -- it
58 * can't fail with only some of the requested snapshots created, even
59 * in the event of power loss or system crash.
60 *
61 * - Continued libzfs Support. Some higher-level operations (e.g.
62 * support for "zfs send -R") are too complicated to fit the scope of
63 * libzfs_core. This functionality will continue to live in libzfs.
64 * Where appropriate, libzfs will use the underlying atomic operations
65 * of libzfs_core. For example, libzfs may implement "zfs send -R |
66 * zfs receive" by using individual "send one snapshot", rename,
67 * destroy, and "receive one snapshot" operations in libzfs_core.
68 * /sbin/zfs and /zbin/zpool will link with both libzfs and
69 * libzfs_core. Other consumers should aim to use only libzfs_core,
70 * since that will be the supported, stable interface going forwards.
71 */
72
73 #include <libzfs_core.h>
74 #include <ctype.h>
75 #include <unistd.h>
76 #include <stdlib.h>
77 #include <string.h>
78 #include <errno.h>
79 #include <fcntl.h>
80 #include <pthread.h>
81 #include <sys/nvpair.h>
82 #include <sys/param.h>
83 #include <sys/types.h>
84 #include <sys/stat.h>
85 #include <sys/zfs_ioctl.h>
86
87 static int g_fd;
88 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER;
89 static int g_refcount;
90
91 int
92 libzfs_core_init(void)
93 {
94 (void) pthread_mutex_lock(&g_lock);
95 if (g_refcount == 0) {
96 g_fd = open("/dev/zfs", O_RDWR);
97 if (g_fd < 0) {
98 (void) pthread_mutex_unlock(&g_lock);
99 return (errno);
100 }
101 }
102 g_refcount++;
103 (void) pthread_mutex_unlock(&g_lock);
104 return (0);
105 }
106
107 void
108 libzfs_core_fini(void)
109 {
110 (void) pthread_mutex_lock(&g_lock);
111 ASSERT3S(g_refcount, >, 0);
112 g_refcount--;
113 if (g_refcount == 0)
114 (void) close(g_fd);
115 (void) pthread_mutex_unlock(&g_lock);
116 }
117
118 static int
119 lzc_ioctl(zfs_ioc_t ioc, const char *name,
120 nvlist_t *source, nvlist_t **resultp)
121 {
122 zfs_cmd_t zc = { 0 };
123 int error = 0;
124 char *packed;
125 size_t size;
126
127 ASSERT3S(g_refcount, >, 0);
128
129 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
130
131 packed = fnvlist_pack(source, &size);
132 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
133 zc.zc_nvlist_src_size = size;
134
135 if (resultp != NULL) {
136 *resultp = NULL;
137 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
138 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
139 malloc(zc.zc_nvlist_dst_size);
140 if (zc.zc_nvlist_dst == NULL) {
141 error = ENOMEM;
142 goto out;
143 }
144 }
145
146 while (ioctl(g_fd, ioc, &zc) != 0) {
147 if (errno == ENOMEM && resultp != NULL) {
148 free((void *)(uintptr_t)zc.zc_nvlist_dst);
149 zc.zc_nvlist_dst_size *= 2;
150 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
151 malloc(zc.zc_nvlist_dst_size);
152 if (zc.zc_nvlist_dst == NULL) {
153 error = ENOMEM;
154 goto out;
155 }
156 } else {
157 error = errno;
158 break;
159 }
160 }
161 if (zc.zc_nvlist_dst_filled) {
162 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
163 zc.zc_nvlist_dst_size);
164 }
165
166 out:
167 fnvlist_pack_free(packed, size);
168 free((void *)(uintptr_t)zc.zc_nvlist_dst);
169 return (error);
170 }
171
172 int
173 lzc_create(const char *fsname, dmu_objset_type_t type, nvlist_t *props)
174 {
175 int error;
176 nvlist_t *args = fnvlist_alloc();
177 fnvlist_add_int32(args, "type", type);
178 if (props != NULL)
179 fnvlist_add_nvlist(args, "props", props);
180 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
181 nvlist_free(args);
182 return (error);
183 }
184
185 int
186 lzc_clone(const char *fsname, const char *origin,
187 nvlist_t *props)
188 {
189 int error;
190 nvlist_t *args = fnvlist_alloc();
191 fnvlist_add_string(args, "origin", origin);
192 if (props != NULL)
193 fnvlist_add_nvlist(args, "props", props);
194 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
195 nvlist_free(args);
196 return (error);
197 }
198
199 /*
200 * Creates snapshots.
201 *
202 * The keys in the snaps nvlist are the snapshots to be created.
203 * They must all be in the same pool.
204 *
205 * The props nvlist is properties to set. Currently only user properties
206 * are supported. { user:prop_name -> string value }
207 *
208 * The returned results nvlist will have an entry for each snapshot that failed.
209 * The value will be the (int32) error code.
210 *
211 * The return value will be 0 if all snapshots were created, otherwise it will
212 * be the errno of a (unspecified) snapshot that failed.
213 */
214 int
215 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
216 {
217 nvpair_t *elem;
218 nvlist_t *args;
219 int error;
220 char pool[MAXNAMELEN];
221
222 *errlist = NULL;
223
224 /* determine the pool name */
225 elem = nvlist_next_nvpair(snaps, NULL);
226 if (elem == NULL)
227 return (0);
228 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
229 pool[strcspn(pool, "/@")] = '\0';
230
231 args = fnvlist_alloc();
232 fnvlist_add_nvlist(args, "snaps", snaps);
233 if (props != NULL)
234 fnvlist_add_nvlist(args, "props", props);
235
236 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
237 nvlist_free(args);
238
239 return (error);
240 }
241
242 /*
243 * Destroys snapshots.
244 *
245 * The keys in the snaps nvlist are the snapshots to be destroyed.
246 * They must all be in the same pool.
247 *
248 * Snapshots that do not exist will be silently ignored.
249 *
250 * If 'defer' is not set, and a snapshot has user holds or clones, the
251 * destroy operation will fail and none of the snapshots will be
252 * destroyed.
253 *
254 * If 'defer' is set, and a snapshot has user holds or clones, it will be
255 * marked for deferred destruction, and will be destroyed when the last hold
256 * or clone is removed/destroyed.
257 *
258 * The return value will be ENOENT if none of the snapshots existed.
259 *
260 * The return value will be 0 if all snapshots were destroyed (or marked for
261 * later destruction if 'defer' is set) or didn't exist to begin with and
262 * at least one snapshot was destroyed.
263 *
264 * Otherwise the return value will be the errno of a (unspecified) snapshot
265 * that failed, no snapshots will be destroyed, and the errlist will have an
266 * entry for each snapshot that failed. The value in the errlist will be
267 * the (int32) error code.
268 */
269 int
270 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
271 {
272 nvpair_t *elem;
273 nvlist_t *args;
274 int error;
275 char pool[MAXNAMELEN];
276
277 /* determine the pool name */
278 elem = nvlist_next_nvpair(snaps, NULL);
279 if (elem == NULL)
280 return (0);
281 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
282 pool[strcspn(pool, "/@")] = '\0';
283
284 args = fnvlist_alloc();
285 fnvlist_add_nvlist(args, "snaps", snaps);
286 if (defer)
287 fnvlist_add_boolean(args, "defer");
288
289 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
290 nvlist_free(args);
291
292 return (error);
293 }
294
295 int
296 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
297 uint64_t *usedp)
298 {
299 nvlist_t *args;
300 nvlist_t *result;
301 int err;
302 char fs[MAXNAMELEN];
303 char *atp;
304
305 /* determine the fs name */
306 (void) strlcpy(fs, firstsnap, sizeof (fs));
307 atp = strchr(fs, '@');
308 if (atp == NULL)
309 return (EINVAL);
310 *atp = '\0';
311
312 args = fnvlist_alloc();
313 fnvlist_add_string(args, "firstsnap", firstsnap);
314
315 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
316 nvlist_free(args);
317 if (err == 0)
318 *usedp = fnvlist_lookup_uint64(result, "used");
319 fnvlist_free(result);
320
321 return (err);
322 }
323
324 boolean_t
325 lzc_exists(const char *dataset)
326 {
327 /*
328 * The objset_stats ioctl is still legacy, so we need to construct our
329 * own zfs_cmd_t rather than using zfsc_ioctl().
330 */
331 zfs_cmd_t zc = { 0 };
332
333 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
334 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
335 }
336
337 /*
338 * Create "user holds" on snapshots. If there is a hold on a snapshot,
339 * the snapshot can not be destroyed. (However, it can be marked for deletion
340 * by lzc_destroy_snaps(defer=B_TRUE).)
341 *
342 * The keys in the nvlist are snapshot names.
343 * The snapshots must all be in the same pool.
344 * The value is the name of the hold (string type).
345 *
346 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
347 * In this case, when the cleanup_fd is closed (including on process
348 * termination), the holds will be released. If the system is shut down
349 * uncleanly, the holds will be released when the pool is next opened
350 * or imported.
351 *
352 * Holds for snapshots which don't exist will be skipped and have an entry
353 * added to errlist, but will not cause an overall failure, except in the
354 * case that all holds where skipped.
355 *
356 * The return value will be ENOENT if none of the snapshots for the requested
357 * holds existed.
358 *
359 * The return value will be 0 if the nvl holds was empty or all holds, for
360 * snapshots that existed, were succesfully created and at least one hold
361 * was created.
362 *
363 * Otherwise the return value will be the errno of a (unspecified) hold that
364 * failed and no holds will be created.
365 *
366 * In all cases the errlist will have an entry for each hold that failed
367 * (name = snapshot), with its value being the error code (int32).
368 */
369 int
370 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
371 {
372 char pool[MAXNAMELEN];
373 nvlist_t *args;
374 nvpair_t *elem;
375 int error;
376
377 /* determine the pool name */
378 elem = nvlist_next_nvpair(holds, NULL);
379 if (elem == NULL)
380 return (0);
381 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
382 pool[strcspn(pool, "/@")] = '\0';
383
384 args = fnvlist_alloc();
385 fnvlist_add_nvlist(args, "holds", holds);
386 if (cleanup_fd != -1)
387 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
388
389 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
390 nvlist_free(args);
391 return (error);
392 }
393
394 /*
395 * Release "user holds" on snapshots. If the snapshot has been marked for
396 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
397 * any clones, and all the user holds are removed, then the snapshot will be
398 * destroyed.
399 *
400 * The keys in the nvlist are snapshot names.
401 * The snapshots must all be in the same pool.
402 * The value is a nvlist whose keys are the holds to remove.
403 *
404 * Holds which failed to release because they didn't exist will have an entry
405 * added to errlist, but will not cause an overall failure, except in the
406 * case that all releases where skipped.
407 *
408 * The return value will be ENOENT if none of the specified holds existed.
409 *
410 * The return value will be 0 if the nvl holds was empty or all holds, that
411 * existed, were succesfully removed and at least one hold was removed.
412 *
413 * Otherwise the return value will be the errno of a (unspecified) hold that
414 * failed to release and no holds will be released.
415 *
416 * In all cases the errlist will have an entry for each hold that failed to
417 * to release.
418 */
419 int
420 lzc_release(nvlist_t *holds, nvlist_t **errlist)
421 {
422 char pool[MAXNAMELEN];
423 nvpair_t *elem;
424
425 /* determine the pool name */
426 elem = nvlist_next_nvpair(holds, NULL);
427 if (elem == NULL)
428 return (0);
429 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
430 pool[strcspn(pool, "/@")] = '\0';
431
432 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
433 }
434
435 /*
436 * Retrieve list of user holds on the specified snapshot.
437 *
438 * On success, *holdsp will be set to a nvlist which the caller must free.
439 * The keys are the names of the holds, and the value is the creation time
440 * of the hold (uint64) in seconds since the epoch.
441 */
442 int
443 lzc_get_holds(const char *snapname, nvlist_t **holdsp)
444 {
445 int error;
446 nvlist_t *innvl = fnvlist_alloc();
447 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp);
448 fnvlist_free(innvl);
449 return (error);
450 }
451
452 /*
453 * If fromsnap is NULL, a full (non-incremental) stream will be sent.
454 */
455 int
456 lzc_send(const char *snapname, const char *fromsnap, int fd)
457 {
458 nvlist_t *args;
459 int err;
460
461 args = fnvlist_alloc();
462 fnvlist_add_int32(args, "fd", fd);
463 if (fromsnap != NULL)
464 fnvlist_add_string(args, "fromsnap", fromsnap);
465 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
466 nvlist_free(args);
467 return (err);
468 }
469
470 /*
471 * If fromsnap is NULL, a full (non-incremental) stream will be estimated.
472 */
473 int
474 lzc_send_space(const char *snapname, const char *fromsnap, uint64_t *spacep)
475 {
476 nvlist_t *args;
477 nvlist_t *result;
478 int err;
479
480 args = fnvlist_alloc();
481 if (fromsnap != NULL)
482 fnvlist_add_string(args, "fromsnap", fromsnap);
483 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
484 nvlist_free(args);
485 if (err == 0)
486 *spacep = fnvlist_lookup_uint64(result, "space");
487 nvlist_free(result);
488 return (err);
489 }
490
491 static int
492 recv_read(int fd, void *buf, int ilen)
493 {
494 char *cp = buf;
495 int rv;
496 int len = ilen;
497
498 do {
499 rv = read(fd, cp, len);
500 cp += rv;
501 len -= rv;
502 } while (rv > 0);
503
504 if (rv < 0 || len != 0)
505 return (EIO);
506
507 return (0);
508 }
509
510 /*
511 * The simplest receive case: receive from the specified fd, creating the
512 * specified snapshot. Apply the specified properties a "received" properties
513 * (which can be overridden by locally-set properties). If the stream is a
514 * clone, its origin snapshot must be specified by 'origin'. The 'force'
515 * flag will cause the target filesystem to be rolled back or destroyed if
516 * necessary to receive.
517 *
518 * Return 0 on success or an errno on failure.
519 *
520 * Note: this interface does not work on dedup'd streams
521 * (those with DMU_BACKUP_FEATURE_DEDUP).
522 */
523 int
524 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
525 boolean_t force, int fd)
526 {
527 /*
528 * The receive ioctl is still legacy, so we need to construct our own
529 * zfs_cmd_t rather than using zfsc_ioctl().
530 */
531 zfs_cmd_t zc = { 0 };
532 char *atp;
533 char *packed = NULL;
534 size_t size;
535 dmu_replay_record_t drr;
536 int error;
537
538 ASSERT3S(g_refcount, >, 0);
539
540 /* zc_name is name of containing filesystem */
541 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
542 atp = strchr(zc.zc_name, '@');
543 if (atp == NULL)
544 return (EINVAL);
545 *atp = '\0';
546
547 /* if the fs does not exist, try its parent. */
548 if (!lzc_exists(zc.zc_name)) {
549 char *slashp = strrchr(zc.zc_name, '/');
550 if (slashp == NULL)
551 return (ENOENT);
552 *slashp = '\0';
553
554 }
555
556 /* zc_value is full name of the snapshot to create */
557 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
558
559 if (props != NULL) {
560 /* zc_nvlist_src is props to set */
561 packed = fnvlist_pack(props, &size);
562 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
563 zc.zc_nvlist_src_size = size;
564 }
565
566 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
567 if (origin != NULL)
568 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
569
570 /* zc_begin_record is non-byteswapped BEGIN record */
571 error = recv_read(fd, &drr, sizeof (drr));
572 if (error != 0)
573 goto out;
574 zc.zc_begin_record = drr.drr_u.drr_begin;
575
576 /* zc_cookie is fd to read from */
577 zc.zc_cookie = fd;
578
579 /* zc guid is force flag */
580 zc.zc_guid = force;
581
582 /* zc_cleanup_fd is unused */
583 zc.zc_cleanup_fd = -1;
584
585 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
586 if (error != 0)
587 error = errno;
588
589 out:
590 if (packed != NULL)
591 fnvlist_pack_free(packed, size);
592 free((void*)(uintptr_t)zc.zc_nvlist_dst);
593 return (error);
594 }