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