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 0 if all snapshots were destroyed (or marked for
259 * later destruction if 'defer' is set) or didn't exist to begin with.
260 *
261 * Otherwise the return value will be the errno of a (unspecified) snapshot
262 * that failed, no snapshots will be destroyed, and the errlist will have an
263 * entry for each snapshot that failed. The value in the errlist will be
264 * the (int32) error code.
265 */
266 int
267 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
268 {
269 nvpair_t *elem;
270 nvlist_t *args;
271 int error;
272 char pool[MAXNAMELEN];
273
274 /* determine the pool name */
275 elem = nvlist_next_nvpair(snaps, NULL);
276 if (elem == NULL)
277 return (0);
278 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
279 pool[strcspn(pool, "/@")] = '\0';
280
281 args = fnvlist_alloc();
282 fnvlist_add_nvlist(args, "snaps", snaps);
283 if (defer)
284 fnvlist_add_boolean(args, "defer");
285
286 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
287 nvlist_free(args);
288
289 return (error);
290 }
291
292 int
293 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
294 uint64_t *usedp)
295 {
296 nvlist_t *args;
297 nvlist_t *result;
298 int err;
299 char fs[MAXNAMELEN];
300 char *atp;
301
302 /* determine the fs name */
303 (void) strlcpy(fs, firstsnap, sizeof (fs));
304 atp = strchr(fs, '@');
305 if (atp == NULL)
306 return (EINVAL);
307 *atp = '\0';
308
309 args = fnvlist_alloc();
310 fnvlist_add_string(args, "firstsnap", firstsnap);
311
312 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
313 nvlist_free(args);
314 if (err == 0)
315 *usedp = fnvlist_lookup_uint64(result, "used");
316 fnvlist_free(result);
317
318 return (err);
319 }
320
321 boolean_t
322 lzc_exists(const char *dataset)
323 {
324 /*
325 * The objset_stats ioctl is still legacy, so we need to construct our
326 * own zfs_cmd_t rather than using zfsc_ioctl().
327 */
328 zfs_cmd_t zc = { 0 };
329
330 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
331 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
332 }
333
334 /*
335 * Create "user holds" on snapshots. If there is a hold on a snapshot,
336 * the snapshot can not be destroyed. (However, it can be marked for deletion
337 * by lzc_destroy_snaps(defer=B_TRUE).)
338 *
339 * The keys in the nvlist are snapshot names.
340 * The snapshots must all be in the same pool.
341 * The value is the name of the hold (string type).
342 *
343 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
344 * In this case, when the cleanup_fd is closed (including on process
345 * termination), the holds will be released. If the system is shut down
346 * uncleanly, the holds will be released when the pool is next opened
347 * or imported.
348 *
349 * Holds for snapshots which don't exist will be skipped and have an entry
350 * added to errlist, but will not cause an overall failure.
351 *
352 * The return value will be 0 if all holds, for snapshots that existed,
353 * were succesfully created.
354 *
355 * Otherwise the return value will be the errno of a (unspecified) hold that
356 * failed and no holds will be created.
357 *
358 * In all cases the errlist will have an entry for each hold that failed
359 * (name = snapshot), with its value being the error code (int32).
360 */
361 int
362 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
363 {
364 char pool[MAXNAMELEN];
365 nvlist_t *args;
366 nvpair_t *elem;
367 int error;
368
369 /* determine the pool name */
370 elem = nvlist_next_nvpair(holds, NULL);
371 if (elem == NULL)
372 return (0);
373 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
374 pool[strcspn(pool, "/@")] = '\0';
375
376 args = fnvlist_alloc();
377 fnvlist_add_nvlist(args, "holds", holds);
378 if (cleanup_fd != -1)
379 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
380
381 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
382 nvlist_free(args);
383 return (error);
384 }
385
386 /*
387 * Release "user holds" on snapshots. If the snapshot has been marked for
388 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
389 * any clones, and all the user holds are removed, then the snapshot will be
390 * destroyed.
391 *
392 * The keys in the nvlist are snapshot names.
393 * The snapshots must all be in the same pool.
394 * The value is a nvlist whose keys are the holds to remove.
395 *
396 * Holds which failed to release because they didn't exist will have an entry
397 * added to errlist, but will not cause an overall failure.
398 *
399 * The return value will be 0 if the nvl holds was empty or all holds that
400 * existed, were successfully removed.
401 *
402 * Otherwise the return value will be the errno of a (unspecified) hold that
403 * failed to release and no holds will be released.
404 *
405 * In all cases the errlist will have an entry for each hold that failed to
406 * to release.
407 */
408 int
409 lzc_release(nvlist_t *holds, nvlist_t **errlist)
410 {
411 char pool[MAXNAMELEN];
412 nvpair_t *elem;
413
414 /* determine the pool name */
415 elem = nvlist_next_nvpair(holds, NULL);
416 if (elem == NULL)
417 return (0);
418 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
419 pool[strcspn(pool, "/@")] = '\0';
420
421 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
422 }
423
424 /*
425 * Retrieve list of user holds on the specified snapshot.
426 *
427 * On success, *holdsp will be set to a nvlist which the caller must free.
428 * The keys are the names of the holds, and the value is the creation time
429 * of the hold (uint64) in seconds since the epoch.
430 */
431 int
432 lzc_get_holds(const char *snapname, nvlist_t **holdsp)
433 {
434 int error;
435 nvlist_t *innvl = fnvlist_alloc();
436 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp);
437 fnvlist_free(innvl);
438 return (error);
439 }
440
441 /*
442 * If fromsnap is NULL, a full (non-incremental) stream will be sent.
443 */
444 int
445 lzc_send(const char *snapname, const char *fromsnap, int fd)
446 {
447 nvlist_t *args;
448 int err;
449
450 args = fnvlist_alloc();
451 fnvlist_add_int32(args, "fd", fd);
452 if (fromsnap != NULL)
453 fnvlist_add_string(args, "fromsnap", fromsnap);
454 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
455 nvlist_free(args);
456 return (err);
457 }
458
459 /*
460 * If fromsnap is NULL, a full (non-incremental) stream will be estimated.
461 */
462 int
463 lzc_send_space(const char *snapname, const char *fromsnap, uint64_t *spacep)
464 {
465 nvlist_t *args;
466 nvlist_t *result;
467 int err;
468
469 args = fnvlist_alloc();
470 if (fromsnap != NULL)
471 fnvlist_add_string(args, "fromsnap", fromsnap);
472 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
473 nvlist_free(args);
474 if (err == 0)
475 *spacep = fnvlist_lookup_uint64(result, "space");
476 nvlist_free(result);
477 return (err);
478 }
479
480 static int
481 recv_read(int fd, void *buf, int ilen)
482 {
483 char *cp = buf;
484 int rv;
485 int len = ilen;
486
487 do {
488 rv = read(fd, cp, len);
489 cp += rv;
490 len -= rv;
491 } while (rv > 0);
492
493 if (rv < 0 || len != 0)
494 return (EIO);
495
496 return (0);
497 }
498
499 /*
500 * The simplest receive case: receive from the specified fd, creating the
501 * specified snapshot. Apply the specified properties a "received" properties
502 * (which can be overridden by locally-set properties). If the stream is a
503 * clone, its origin snapshot must be specified by 'origin'. The 'force'
504 * flag will cause the target filesystem to be rolled back or destroyed if
505 * necessary to receive.
506 *
507 * Return 0 on success or an errno on failure.
508 *
509 * Note: this interface does not work on dedup'd streams
510 * (those with DMU_BACKUP_FEATURE_DEDUP).
511 */
512 int
513 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
514 boolean_t force, int fd)
515 {
516 /*
517 * The receive ioctl is still legacy, so we need to construct our own
518 * zfs_cmd_t rather than using zfsc_ioctl().
519 */
520 zfs_cmd_t zc = { 0 };
521 char *atp;
522 char *packed = NULL;
523 size_t size;
524 dmu_replay_record_t drr;
525 int error;
526
527 ASSERT3S(g_refcount, >, 0);
528
529 /* zc_name is name of containing filesystem */
530 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
531 atp = strchr(zc.zc_name, '@');
532 if (atp == NULL)
533 return (EINVAL);
534 *atp = '\0';
535
536 /* if the fs does not exist, try its parent. */
537 if (!lzc_exists(zc.zc_name)) {
538 char *slashp = strrchr(zc.zc_name, '/');
539 if (slashp == NULL)
540 return (ENOENT);
541 *slashp = '\0';
542
543 }
544
545 /* zc_value is full name of the snapshot to create */
546 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
547
548 if (props != NULL) {
549 /* zc_nvlist_src is props to set */
550 packed = fnvlist_pack(props, &size);
551 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
552 zc.zc_nvlist_src_size = size;
553 }
554
555 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
556 if (origin != NULL)
557 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
558
559 /* zc_begin_record is non-byteswapped BEGIN record */
560 error = recv_read(fd, &drr, sizeof (drr));
561 if (error != 0)
562 goto out;
563 zc.zc_begin_record = drr.drr_u.drr_begin;
564
565 /* zc_cookie is fd to read from */
566 zc.zc_cookie = fd;
567
568 /* zc guid is force flag */
569 zc.zc_guid = force;
570
571 /* zc_cleanup_fd is unused */
572 zc.zc_cleanup_fd = -1;
573
574 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
575 if (error != 0)
576 error = errno;
577
578 out:
579 if (packed != NULL)
580 fnvlist_pack_free(packed, size);
581 free((void*)(uintptr_t)zc.zc_nvlist_dst);
582 return (error);
583 }