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, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 */
27
28 /*
29 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs.
30 * It has the following characteristics:
31 *
32 * - Thread Safe. libzfs_core is accessible concurrently from multiple
33 * threads. This is accomplished primarily by avoiding global data
34 * (e.g. caching). Since it's thread-safe, there is no reason for a
35 * process to have multiple libzfs "instances". Therefore, we store
36 * our few pieces of data (e.g. the file descriptor) in global
37 * variables. The fd is reference-counted so that the libzfs_core
38 * library can be "initialized" multiple times (e.g. by different
39 * consumers within the same process).
40 *
41 * - Committed Interface. The libzfs_core interface will be committed,
42 * therefore consumers can compile against it and be confident that
43 * their code will continue to work on future releases of this code.
44 * Currently, the interface is Evolving (not Committed), but we intend
45 * to commit to it once it is more complete and we determine that it
46 * meets the needs of all consumers.
47 *
48 * - Programatic Error Handling. libzfs_core communicates errors with
49 * defined error numbers, and doesn't print anything to stdout/stderr.
50 *
51 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments
52 * to/from the kernel ioctls. There is generally a 1:1 correspondence
53 * between libzfs_core functions and ioctls to /dev/zfs.
54 *
55 * - Clear Atomicity. Because libzfs_core functions are generally 1:1
56 * with kernel ioctls, and kernel ioctls are general atomic, each
57 * libzfs_core function is atomic. For example, creating multiple
58 * snapshots with a single call to lzc_snapshot() is atomic -- it
59 * can't fail with only some of the requested snapshots created, even
60 * in the event of power loss or system crash.
61 *
62 * - Continued libzfs Support. Some higher-level operations (e.g.
63 * support for "zfs send -R") are too complicated to fit the scope of
64 * libzfs_core. This functionality will continue to live in libzfs.
65 * Where appropriate, libzfs will use the underlying atomic operations
66 * of libzfs_core. For example, libzfs may implement "zfs send -R |
67 * zfs receive" by using individual "send one snapshot", rename,
68 * destroy, and "receive one snapshot" operations in libzfs_core.
69 * /sbin/zfs and /zbin/zpool will link with both libzfs and
70 * libzfs_core. Other consumers should aim to use only libzfs_core,
71 * since that will be the supported, stable interface going forwards.
72 */
73
74 #define _IN_LIBZFS_CORE_
75
76 #include <libzfs_core.h>
77 #include <ctype.h>
78 #include <unistd.h>
79 #include <stdlib.h>
80 #include <string.h>
81 #include <errno.h>
82 #include <fcntl.h>
83 #include <pthread.h>
84 #include <sys/nvpair.h>
85 #include <sys/param.h>
86 #include <sys/types.h>
87 #include <sys/stat.h>
88 #include <sys/zfs_ioctl.h>
89 #include "libzfs_core_compat.h"
90 #include "libzfs_compat.h"
91
92 #ifdef __FreeBSD__
93 extern int zfs_ioctl_version;
94 #endif
95
96 static int g_fd;
97 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER;
98 static int g_refcount;
99
100 int
libzfs_core_init(void)101 libzfs_core_init(void)
102 {
103 (void) pthread_mutex_lock(&g_lock);
104 if (g_refcount == 0) {
105 g_fd = open("/dev/zfs", O_RDWR);
106 if (g_fd < 0) {
107 (void) pthread_mutex_unlock(&g_lock);
108 return (errno);
109 }
110 }
111 g_refcount++;
112 (void) pthread_mutex_unlock(&g_lock);
113
114 return (0);
115 }
116
117 void
libzfs_core_fini(void)118 libzfs_core_fini(void)
119 {
120 (void) pthread_mutex_lock(&g_lock);
121 ASSERT3S(g_refcount, >, 0);
122 g_refcount--;
123 if (g_refcount == 0)
124 (void) close(g_fd);
125 (void) pthread_mutex_unlock(&g_lock);
126 }
127
128 static int
lzc_ioctl(zfs_ioc_t ioc,const char * name,nvlist_t * source,nvlist_t ** resultp)129 lzc_ioctl(zfs_ioc_t ioc, const char *name,
130 nvlist_t *source, nvlist_t **resultp)
131 {
132 zfs_cmd_t zc = { 0 };
133 int error = 0;
134 char *packed;
135 #ifdef __FreeBSD__
136 nvlist_t *oldsource;
137 #endif
138 size_t size;
139
140 ASSERT3S(g_refcount, >, 0);
141
142 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
143
144 #ifdef __FreeBSD__
145 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
146 zfs_ioctl_version = get_zfs_ioctl_version();
147
148 if (zfs_ioctl_version < ZFS_IOCVER_LZC) {
149 oldsource = source;
150 error = lzc_compat_pre(&zc, &ioc, &source);
151 if (error)
152 return (error);
153 }
154 #endif
155
156 packed = fnvlist_pack(source, &size);
157 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
158 zc.zc_nvlist_src_size = size;
159
160 if (resultp != NULL) {
161 *resultp = NULL;
162 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
163 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
164 malloc(zc.zc_nvlist_dst_size);
165 #ifdef illumos
166 if (zc.zc_nvlist_dst == NULL) {
167 #else
168 if (zc.zc_nvlist_dst == 0) {
169 #endif
170 error = ENOMEM;
171 goto out;
172 }
173 }
174
175 while (ioctl(g_fd, ioc, &zc) != 0) {
176 if (errno == ENOMEM && resultp != NULL) {
177 free((void *)(uintptr_t)zc.zc_nvlist_dst);
178 zc.zc_nvlist_dst_size *= 2;
179 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
180 malloc(zc.zc_nvlist_dst_size);
181 #ifdef illumos
182 if (zc.zc_nvlist_dst == NULL) {
183 #else
184 if (zc.zc_nvlist_dst == 0) {
185 #endif
186 error = ENOMEM;
187 goto out;
188 }
189 } else {
190 error = errno;
191 break;
192 }
193 }
194
195 #ifdef __FreeBSD__
196 if (zfs_ioctl_version < ZFS_IOCVER_LZC)
197 lzc_compat_post(&zc, ioc);
198 #endif
199 if (zc.zc_nvlist_dst_filled) {
200 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
201 zc.zc_nvlist_dst_size);
202 }
203 #ifdef __FreeBSD__
204 if (zfs_ioctl_version < ZFS_IOCVER_LZC)
205 lzc_compat_outnvl(&zc, ioc, resultp);
206 #endif
207 out:
208 #ifdef __FreeBSD__
209 if (zfs_ioctl_version < ZFS_IOCVER_LZC) {
210 if (source != oldsource)
211 nvlist_free(source);
212 source = oldsource;
213 }
214 #endif
215 fnvlist_pack_free(packed, size);
216 free((void *)(uintptr_t)zc.zc_nvlist_dst);
217 return (error);
218 }
219
220 int
221 lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props)
222 {
223 int error;
224 nvlist_t *args = fnvlist_alloc();
225 fnvlist_add_int32(args, "type", (dmu_objset_type_t)type);
226 if (props != NULL)
227 fnvlist_add_nvlist(args, "props", props);
228 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
229 nvlist_free(args);
230 return (error);
231 }
232
233 int
234 lzc_clone(const char *fsname, const char *origin,
235 nvlist_t *props)
236 {
237 int error;
238 nvlist_t *args = fnvlist_alloc();
239 fnvlist_add_string(args, "origin", origin);
240 if (props != NULL)
241 fnvlist_add_nvlist(args, "props", props);
242 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
243 nvlist_free(args);
244 return (error);
245 }
246
247 /*
248 * Creates snapshots.
249 *
250 * The keys in the snaps nvlist are the snapshots to be created.
251 * They must all be in the same pool.
252 *
253 * The props nvlist is properties to set. Currently only user properties
254 * are supported. { user:prop_name -> string value }
255 *
256 * The returned results nvlist will have an entry for each snapshot that failed.
257 * The value will be the (int32) error code.
258 *
259 * The return value will be 0 if all snapshots were created, otherwise it will
260 * be the errno of a (unspecified) snapshot that failed.
261 */
262 int
263 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
264 {
265 nvpair_t *elem;
266 nvlist_t *args;
267 int error;
268 char pool[ZFS_MAX_DATASET_NAME_LEN];
269
270 *errlist = NULL;
271
272 /* determine the pool name */
273 elem = nvlist_next_nvpair(snaps, NULL);
274 if (elem == NULL)
275 return (0);
276 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
277 pool[strcspn(pool, "/@")] = '\0';
278
279 args = fnvlist_alloc();
280 fnvlist_add_nvlist(args, "snaps", snaps);
281 if (props != NULL)
282 fnvlist_add_nvlist(args, "props", props);
283
284 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
285 nvlist_free(args);
286
287 return (error);
288 }
289
290 /*
291 * Destroys snapshots.
292 *
293 * The keys in the snaps nvlist are the snapshots to be destroyed.
294 * They must all be in the same pool.
295 *
296 * Snapshots that do not exist will be silently ignored.
297 *
298 * If 'defer' is not set, and a snapshot has user holds or clones, the
299 * destroy operation will fail and none of the snapshots will be
300 * destroyed.
301 *
302 * If 'defer' is set, and a snapshot has user holds or clones, it will be
303 * marked for deferred destruction, and will be destroyed when the last hold
304 * or clone is removed/destroyed.
305 *
306 * The return value will be 0 if all snapshots were destroyed (or marked for
307 * later destruction if 'defer' is set) or didn't exist to begin with.
308 *
309 * Otherwise the return value will be the errno of a (unspecified) snapshot
310 * that failed, no snapshots will be destroyed, and the errlist will have an
311 * entry for each snapshot that failed. The value in the errlist will be
312 * the (int32) error code.
313 */
314 int
315 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
316 {
317 nvpair_t *elem;
318 nvlist_t *args;
319 int error;
320 char pool[ZFS_MAX_DATASET_NAME_LEN];
321
322 /* determine the pool name */
323 elem = nvlist_next_nvpair(snaps, NULL);
324 if (elem == NULL)
325 return (0);
326 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
327 pool[strcspn(pool, "/@")] = '\0';
328
329 args = fnvlist_alloc();
330 fnvlist_add_nvlist(args, "snaps", snaps);
331 if (defer)
332 fnvlist_add_boolean(args, "defer");
333
334 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
335 nvlist_free(args);
336
337 return (error);
338 }
339
340 int
341 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
342 uint64_t *usedp)
343 {
344 nvlist_t *args;
345 nvlist_t *result;
346 int err;
347 char fs[ZFS_MAX_DATASET_NAME_LEN];
348 char *atp;
349
350 /* determine the fs name */
351 (void) strlcpy(fs, firstsnap, sizeof (fs));
352 atp = strchr(fs, '@');
353 if (atp == NULL)
354 return (EINVAL);
355 *atp = '\0';
356
357 args = fnvlist_alloc();
358 fnvlist_add_string(args, "firstsnap", firstsnap);
359
360 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
361 nvlist_free(args);
362 if (err == 0)
363 *usedp = fnvlist_lookup_uint64(result, "used");
364 fnvlist_free(result);
365
366 return (err);
367 }
368
369 boolean_t
370 lzc_exists(const char *dataset)
371 {
372 /*
373 * The objset_stats ioctl is still legacy, so we need to construct our
374 * own zfs_cmd_t rather than using zfsc_ioctl().
375 */
376 zfs_cmd_t zc = { 0 };
377
378 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
379 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
380 }
381
382 /*
383 * Create "user holds" on snapshots. If there is a hold on a snapshot,
384 * the snapshot can not be destroyed. (However, it can be marked for deletion
385 * by lzc_destroy_snaps(defer=B_TRUE).)
386 *
387 * The keys in the nvlist are snapshot names.
388 * The snapshots must all be in the same pool.
389 * The value is the name of the hold (string type).
390 *
391 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
392 * In this case, when the cleanup_fd is closed (including on process
393 * termination), the holds will be released. If the system is shut down
394 * uncleanly, the holds will be released when the pool is next opened
395 * or imported.
396 *
397 * Holds for snapshots which don't exist will be skipped and have an entry
398 * added to errlist, but will not cause an overall failure.
399 *
400 * The return value will be 0 if all holds, for snapshots that existed,
401 * were succesfully created.
402 *
403 * Otherwise the return value will be the errno of a (unspecified) hold that
404 * failed and no holds will be created.
405 *
406 * In all cases the errlist will have an entry for each hold that failed
407 * (name = snapshot), with its value being the error code (int32).
408 */
409 int
410 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
411 {
412 char pool[ZFS_MAX_DATASET_NAME_LEN];
413 nvlist_t *args;
414 nvpair_t *elem;
415 int error;
416
417 /* determine the pool name */
418 elem = nvlist_next_nvpair(holds, NULL);
419 if (elem == NULL)
420 return (0);
421 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
422 pool[strcspn(pool, "/@")] = '\0';
423
424 args = fnvlist_alloc();
425 fnvlist_add_nvlist(args, "holds", holds);
426 if (cleanup_fd != -1)
427 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
428
429 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
430 nvlist_free(args);
431 return (error);
432 }
433
434 /*
435 * Release "user holds" on snapshots. If the snapshot has been marked for
436 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
437 * any clones, and all the user holds are removed, then the snapshot will be
438 * destroyed.
439 *
440 * The keys in the nvlist are snapshot names.
441 * The snapshots must all be in the same pool.
442 * The value is a nvlist whose keys are the holds to remove.
443 *
444 * Holds which failed to release because they didn't exist will have an entry
445 * added to errlist, but will not cause an overall failure.
446 *
447 * The return value will be 0 if the nvl holds was empty or all holds that
448 * existed, were successfully removed.
449 *
450 * Otherwise the return value will be the errno of a (unspecified) hold that
451 * failed to release and no holds will be released.
452 *
453 * In all cases the errlist will have an entry for each hold that failed to
454 * to release.
455 */
456 int
457 lzc_release(nvlist_t *holds, nvlist_t **errlist)
458 {
459 char pool[ZFS_MAX_DATASET_NAME_LEN];
460 nvpair_t *elem;
461
462 /* determine the pool name */
463 elem = nvlist_next_nvpair(holds, NULL);
464 if (elem == NULL)
465 return (0);
466 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
467 pool[strcspn(pool, "/@")] = '\0';
468
469 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
470 }
471
472 /*
473 * Retrieve list of user holds on the specified snapshot.
474 *
475 * On success, *holdsp will be set to a nvlist which the caller must free.
476 * The keys are the names of the holds, and the value is the creation time
477 * of the hold (uint64) in seconds since the epoch.
478 */
479 int
480 lzc_get_holds(const char *snapname, nvlist_t **holdsp)
481 {
482 int error;
483 nvlist_t *innvl = fnvlist_alloc();
484 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp);
485 fnvlist_free(innvl);
486 return (error);
487 }
488
489 /*
490 * Generate a zfs send stream for the specified snapshot and write it to
491 * the specified file descriptor.
492 *
493 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap")
494 *
495 * If "from" is NULL, a full (non-incremental) stream will be sent.
496 * If "from" is non-NULL, it must be the full name of a snapshot or
497 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or
498 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or
499 * bookmark must represent an earlier point in the history of "snapname").
500 * It can be an earlier snapshot in the same filesystem or zvol as "snapname",
501 * or it can be the origin of "snapname"'s filesystem, or an earlier
502 * snapshot in the origin, etc.
503 *
504 * "fd" is the file descriptor to write the send stream to.
505 *
506 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted
507 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT
508 * records with drr_blksz > 128K.
509 *
510 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted
511 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA,
512 * which the receiving system must support (as indicated by support
513 * for the "embedded_data" feature).
514 */
515 int
516 lzc_send(const char *snapname, const char *from, int fd,
517 enum lzc_send_flags flags)
518 {
519 return (lzc_send_resume(snapname, from, fd, flags, 0, 0));
520 }
521
522 int
523 lzc_send_resume(const char *snapname, const char *from, int fd,
524 enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff)
525 {
526 nvlist_t *args;
527 int err;
528
529 args = fnvlist_alloc();
530 fnvlist_add_int32(args, "fd", fd);
531 if (from != NULL)
532 fnvlist_add_string(args, "fromsnap", from);
533 if (flags & LZC_SEND_FLAG_LARGE_BLOCK)
534 fnvlist_add_boolean(args, "largeblockok");
535 if (flags & LZC_SEND_FLAG_EMBED_DATA)
536 fnvlist_add_boolean(args, "embedok");
537 if (resumeobj != 0 || resumeoff != 0) {
538 fnvlist_add_uint64(args, "resume_object", resumeobj);
539 fnvlist_add_uint64(args, "resume_offset", resumeoff);
540 }
541 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
542 nvlist_free(args);
543 return (err);
544 }
545
546 /*
547 * "from" can be NULL, a snapshot, or a bookmark.
548 *
549 * If from is NULL, a full (non-incremental) stream will be estimated. This
550 * is calculated very efficiently.
551 *
552 * If from is a snapshot, lzc_send_space uses the deadlists attached to
553 * each snapshot to efficiently estimate the stream size.
554 *
555 * If from is a bookmark, the indirect blocks in the destination snapshot
556 * are traversed, looking for blocks with a birth time since the creation TXG of
557 * the snapshot this bookmark was created from. This will result in
558 * significantly more I/O and be less efficient than a send space estimation on
559 * an equivalent snapshot.
560 */
561 int
562 lzc_send_space(const char *snapname, const char *from, uint64_t *spacep)
563 {
564 nvlist_t *args;
565 nvlist_t *result;
566 int err;
567
568 args = fnvlist_alloc();
569 if (from != NULL)
570 fnvlist_add_string(args, "from", from);
571 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
572 nvlist_free(args);
573 if (err == 0)
574 *spacep = fnvlist_lookup_uint64(result, "space");
575 nvlist_free(result);
576 return (err);
577 }
578
579 static int
580 recv_read(int fd, void *buf, int ilen)
581 {
582 char *cp = buf;
583 int rv;
584 int len = ilen;
585
586 do {
587 rv = read(fd, cp, len);
588 cp += rv;
589 len -= rv;
590 } while (rv > 0);
591
592 if (rv < 0 || len != 0)
593 return (EIO);
594
595 return (0);
596 }
597
598 static int
599 recv_impl(const char *snapname, nvlist_t *props, const char *origin,
600 boolean_t force, boolean_t resumable, int fd,
601 const dmu_replay_record_t *begin_record)
602 {
603 /*
604 * The receive ioctl is still legacy, so we need to construct our own
605 * zfs_cmd_t rather than using zfsc_ioctl().
606 */
607 zfs_cmd_t zc = { 0 };
608 char *atp;
609 char *packed = NULL;
610 size_t size;
611 int error;
612
613 ASSERT3S(g_refcount, >, 0);
614
615 /* zc_name is name of containing filesystem */
616 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
617 atp = strchr(zc.zc_name, '@');
618 if (atp == NULL)
619 return (EINVAL);
620 *atp = '\0';
621
622 /* if the fs does not exist, try its parent. */
623 if (!lzc_exists(zc.zc_name)) {
624 char *slashp = strrchr(zc.zc_name, '/');
625 if (slashp == NULL)
626 return (ENOENT);
627 *slashp = '\0';
628
629 }
630
631 /* zc_value is full name of the snapshot to create */
632 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
633
634 if (props != NULL) {
635 /* zc_nvlist_src is props to set */
636 packed = fnvlist_pack(props, &size);
637 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
638 zc.zc_nvlist_src_size = size;
639 }
640
641 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
642 if (origin != NULL)
643 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
644
645 /* zc_begin_record is non-byteswapped BEGIN record */
646 if (begin_record == NULL) {
647 error = recv_read(fd, &zc.zc_begin_record,
648 sizeof (zc.zc_begin_record));
649 if (error != 0)
650 goto out;
651 } else {
652 zc.zc_begin_record = *begin_record;
653 }
654
655 /* zc_cookie is fd to read from */
656 zc.zc_cookie = fd;
657
658 /* zc guid is force flag */
659 zc.zc_guid = force;
660
661 zc.zc_resumable = resumable;
662
663 /* zc_cleanup_fd is unused */
664 zc.zc_cleanup_fd = -1;
665
666 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
667 if (error != 0)
668 error = errno;
669
670 out:
671 if (packed != NULL)
672 fnvlist_pack_free(packed, size);
673 free((void*)(uintptr_t)zc.zc_nvlist_dst);
674 return (error);
675 }
676
677 /*
678 * The simplest receive case: receive from the specified fd, creating the
679 * specified snapshot. Apply the specified properties as "received" properties
680 * (which can be overridden by locally-set properties). If the stream is a
681 * clone, its origin snapshot must be specified by 'origin'. The 'force'
682 * flag will cause the target filesystem to be rolled back or destroyed if
683 * necessary to receive.
684 *
685 * Return 0 on success or an errno on failure.
686 *
687 * Note: this interface does not work on dedup'd streams
688 * (those with DMU_BACKUP_FEATURE_DEDUP).
689 */
690 int
691 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
692 boolean_t force, int fd)
693 {
694 return (recv_impl(snapname, props, origin, force, B_FALSE, fd, NULL));
695 }
696
697 /*
698 * Like lzc_receive, but if the receive fails due to premature stream
699 * termination, the intermediate state will be preserved on disk. In this
700 * case, ECKSUM will be returned. The receive may subsequently be resumed
701 * with a resuming send stream generated by lzc_send_resume().
702 */
703 int
704 lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin,
705 boolean_t force, int fd)
706 {
707 return (recv_impl(snapname, props, origin, force, B_TRUE, fd, NULL));
708 }
709
710 /*
711 * Like lzc_receive, but allows the caller to read the begin record and then to
712 * pass it in. That could be useful if the caller wants to derive, for example,
713 * the snapname or the origin parameters based on the information contained in
714 * the begin record.
715 * The begin record must be in its original form as read from the stream,
716 * in other words, it should not be byteswapped.
717 *
718 * The 'resumable' parameter allows to obtain the same behavior as with
719 * lzc_receive_resumable.
720 */
721 int
722 lzc_receive_with_header(const char *snapname, nvlist_t *props,
723 const char *origin, boolean_t force, boolean_t resumable, int fd,
724 const dmu_replay_record_t *begin_record)
725 {
726 if (begin_record == NULL)
727 return (EINVAL);
728 return (recv_impl(snapname, props, origin, force, resumable, fd,
729 begin_record));
730 }
731
732 /*
733 * Roll back this filesystem or volume to its most recent snapshot.
734 * If snapnamebuf is not NULL, it will be filled in with the name
735 * of the most recent snapshot.
736 *
737 * Return 0 on success or an errno on failure.
738 */
739 int
740 lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen)
741 {
742 nvlist_t *args;
743 nvlist_t *result;
744 int err;
745
746 args = fnvlist_alloc();
747 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result);
748 nvlist_free(args);
749 if (err == 0 && snapnamebuf != NULL) {
750 const char *snapname = fnvlist_lookup_string(result, "target");
751 (void) strlcpy(snapnamebuf, snapname, snapnamelen);
752 }
753 return (err);
754 }
755
756 /*
757 * Creates bookmarks.
758 *
759 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to
760 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and
761 * snapshots must be in the same pool.
762 *
763 * The returned results nvlist will have an entry for each bookmark that failed.
764 * The value will be the (int32) error code.
765 *
766 * The return value will be 0 if all bookmarks were created, otherwise it will
767 * be the errno of a (undetermined) bookmarks that failed.
768 */
769 int
770 lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist)
771 {
772 nvpair_t *elem;
773 int error;
774 char pool[ZFS_MAX_DATASET_NAME_LEN];
775
776 /* determine the pool name */
777 elem = nvlist_next_nvpair(bookmarks, NULL);
778 if (elem == NULL)
779 return (0);
780 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
781 pool[strcspn(pool, "/#")] = '\0';
782
783 error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist);
784
785 return (error);
786 }
787
788 /*
789 * Retrieve bookmarks.
790 *
791 * Retrieve the list of bookmarks for the given file system. The props
792 * parameter is an nvlist of property names (with no values) that will be
793 * returned for each bookmark.
794 *
795 * The following are valid properties on bookmarks, all of which are numbers
796 * (represented as uint64 in the nvlist)
797 *
798 * "guid" - globally unique identifier of the snapshot it refers to
799 * "createtxg" - txg when the snapshot it refers to was created
800 * "creation" - timestamp when the snapshot it refers to was created
801 *
802 * The format of the returned nvlist as follows:
803 * <short name of bookmark> -> {
804 * <name of property> -> {
805 * "value" -> uint64
806 * }
807 * }
808 */
809 int
810 lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks)
811 {
812 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks));
813 }
814
815 /*
816 * Destroys bookmarks.
817 *
818 * The keys in the bmarks nvlist are the bookmarks to be destroyed.
819 * They must all be in the same pool. Bookmarks are specified as
820 * <fs>#<bmark>.
821 *
822 * Bookmarks that do not exist will be silently ignored.
823 *
824 * The return value will be 0 if all bookmarks that existed were destroyed.
825 *
826 * Otherwise the return value will be the errno of a (undetermined) bookmark
827 * that failed, no bookmarks will be destroyed, and the errlist will have an
828 * entry for each bookmarks that failed. The value in the errlist will be
829 * the (int32) error code.
830 */
831 int
832 lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist)
833 {
834 nvpair_t *elem;
835 int error;
836 char pool[ZFS_MAX_DATASET_NAME_LEN];
837
838 /* determine the pool name */
839 elem = nvlist_next_nvpair(bmarks, NULL);
840 if (elem == NULL)
841 return (0);
842 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
843 pool[strcspn(pool, "/#")] = '\0';
844
845 error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist);
846
847 return (error);
848 }
849