]>
Commit | Line | Data |
---|---|---|
6f1ffb06 MA |
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 | /* | |
5dc8b736 | 23 | * Copyright (c) 2012, 2014 by Delphix. All rights reserved. |
95fd54a1 | 24 | * Copyright (c) 2013 Steven Hartland. All rights reserved. |
6f1ffb06 MA |
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 | * | |
b8fce77b | 47 | * - Programmatic Error Handling. libzfs_core communicates errors with |
6f1ffb06 MA |
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 | ||
e2454897 | 87 | static int g_fd = -1; |
6f1ffb06 MA |
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); | |
e2454897 GM |
112 | |
113 | if (g_refcount > 0) | |
114 | g_refcount--; | |
115 | ||
116 | if (g_refcount == 0 && g_fd != -1) { | |
6f1ffb06 | 117 | (void) close(g_fd); |
e2454897 GM |
118 | g_fd = -1; |
119 | } | |
6f1ffb06 MA |
120 | (void) pthread_mutex_unlock(&g_lock); |
121 | } | |
122 | ||
123 | static int | |
124 | lzc_ioctl(zfs_ioc_t ioc, const char *name, | |
125 | nvlist_t *source, nvlist_t **resultp) | |
126 | { | |
13fe0198 | 127 | zfs_cmd_t zc = {"\0"}; |
6f1ffb06 MA |
128 | int error = 0; |
129 | char *packed; | |
130 | size_t size; | |
131 | ||
132 | ASSERT3S(g_refcount, >, 0); | |
e2454897 | 133 | VERIFY3S(g_fd, !=, -1); |
6f1ffb06 MA |
134 | |
135 | (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); | |
136 | ||
137 | packed = fnvlist_pack(source, &size); | |
138 | zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; | |
139 | zc.zc_nvlist_src_size = size; | |
140 | ||
141 | if (resultp != NULL) { | |
13fe0198 | 142 | *resultp = NULL; |
6f1ffb06 MA |
143 | zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024); |
144 | zc.zc_nvlist_dst = (uint64_t)(uintptr_t) | |
145 | malloc(zc.zc_nvlist_dst_size); | |
146 | if (zc.zc_nvlist_dst == (uint64_t)0) { | |
147 | error = ENOMEM; | |
148 | goto out; | |
149 | } | |
150 | } | |
151 | ||
152 | while (ioctl(g_fd, ioc, &zc) != 0) { | |
153 | if (errno == ENOMEM && resultp != NULL) { | |
154 | free((void *)(uintptr_t)zc.zc_nvlist_dst); | |
155 | zc.zc_nvlist_dst_size *= 2; | |
156 | zc.zc_nvlist_dst = (uint64_t)(uintptr_t) | |
157 | malloc(zc.zc_nvlist_dst_size); | |
158 | if (zc.zc_nvlist_dst == (uint64_t)0) { | |
159 | error = ENOMEM; | |
160 | goto out; | |
161 | } | |
162 | } else { | |
163 | error = errno; | |
164 | break; | |
165 | } | |
166 | } | |
167 | if (zc.zc_nvlist_dst_filled) { | |
168 | *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, | |
169 | zc.zc_nvlist_dst_size); | |
6f1ffb06 MA |
170 | } |
171 | ||
172 | out: | |
173 | fnvlist_pack_free(packed, size); | |
174 | free((void *)(uintptr_t)zc.zc_nvlist_dst); | |
175 | return (error); | |
176 | } | |
177 | ||
178 | int | |
e67a7ffb | 179 | lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props) |
6f1ffb06 MA |
180 | { |
181 | int error; | |
182 | nvlist_t *args = fnvlist_alloc(); | |
e67a7ffb | 183 | fnvlist_add_int32(args, "type", (dmu_objset_type_t)type); |
6f1ffb06 MA |
184 | if (props != NULL) |
185 | fnvlist_add_nvlist(args, "props", props); | |
186 | error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL); | |
187 | nvlist_free(args); | |
188 | return (error); | |
189 | } | |
190 | ||
191 | int | |
192 | lzc_clone(const char *fsname, const char *origin, | |
193 | nvlist_t *props) | |
194 | { | |
195 | int error; | |
196 | nvlist_t *args = fnvlist_alloc(); | |
197 | fnvlist_add_string(args, "origin", origin); | |
198 | if (props != NULL) | |
199 | fnvlist_add_nvlist(args, "props", props); | |
200 | error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL); | |
201 | nvlist_free(args); | |
202 | return (error); | |
203 | } | |
204 | ||
205 | /* | |
206 | * Creates snapshots. | |
207 | * | |
208 | * The keys in the snaps nvlist are the snapshots to be created. | |
209 | * They must all be in the same pool. | |
210 | * | |
211 | * The props nvlist is properties to set. Currently only user properties | |
212 | * are supported. { user:prop_name -> string value } | |
213 | * | |
214 | * The returned results nvlist will have an entry for each snapshot that failed. | |
215 | * The value will be the (int32) error code. | |
216 | * | |
217 | * The return value will be 0 if all snapshots were created, otherwise it will | |
13fe0198 | 218 | * be the errno of a (unspecified) snapshot that failed. |
6f1ffb06 MA |
219 | */ |
220 | int | |
221 | lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) | |
222 | { | |
223 | nvpair_t *elem; | |
224 | nvlist_t *args; | |
225 | int error; | |
eca7b760 | 226 | char pool[ZFS_MAX_DATASET_NAME_LEN]; |
6f1ffb06 MA |
227 | |
228 | *errlist = NULL; | |
229 | ||
230 | /* determine the pool name */ | |
231 | elem = nvlist_next_nvpair(snaps, NULL); | |
232 | if (elem == NULL) | |
233 | return (0); | |
234 | (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); | |
235 | pool[strcspn(pool, "/@")] = '\0'; | |
236 | ||
237 | args = fnvlist_alloc(); | |
238 | fnvlist_add_nvlist(args, "snaps", snaps); | |
239 | if (props != NULL) | |
240 | fnvlist_add_nvlist(args, "props", props); | |
241 | ||
242 | error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist); | |
243 | nvlist_free(args); | |
244 | ||
245 | return (error); | |
246 | } | |
247 | ||
248 | /* | |
249 | * Destroys snapshots. | |
250 | * | |
251 | * The keys in the snaps nvlist are the snapshots to be destroyed. | |
252 | * They must all be in the same pool. | |
253 | * | |
254 | * Snapshots that do not exist will be silently ignored. | |
255 | * | |
256 | * If 'defer' is not set, and a snapshot has user holds or clones, the | |
257 | * destroy operation will fail and none of the snapshots will be | |
258 | * destroyed. | |
259 | * | |
260 | * If 'defer' is set, and a snapshot has user holds or clones, it will be | |
261 | * marked for deferred destruction, and will be destroyed when the last hold | |
262 | * or clone is removed/destroyed. | |
263 | * | |
264 | * The return value will be 0 if all snapshots were destroyed (or marked for | |
1a077756 | 265 | * later destruction if 'defer' is set) or didn't exist to begin with. |
6f1ffb06 | 266 | * |
13fe0198 | 267 | * Otherwise the return value will be the errno of a (unspecified) snapshot |
6f1ffb06 MA |
268 | * that failed, no snapshots will be destroyed, and the errlist will have an |
269 | * entry for each snapshot that failed. The value in the errlist will be | |
270 | * the (int32) error code. | |
271 | */ | |
272 | int | |
273 | lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist) | |
274 | { | |
275 | nvpair_t *elem; | |
276 | nvlist_t *args; | |
277 | int error; | |
eca7b760 | 278 | char pool[ZFS_MAX_DATASET_NAME_LEN]; |
6f1ffb06 MA |
279 | |
280 | /* determine the pool name */ | |
281 | elem = nvlist_next_nvpair(snaps, NULL); | |
282 | if (elem == NULL) | |
283 | return (0); | |
284 | (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); | |
285 | pool[strcspn(pool, "/@")] = '\0'; | |
286 | ||
287 | args = fnvlist_alloc(); | |
288 | fnvlist_add_nvlist(args, "snaps", snaps); | |
289 | if (defer) | |
290 | fnvlist_add_boolean(args, "defer"); | |
291 | ||
292 | error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist); | |
293 | nvlist_free(args); | |
294 | ||
295 | return (error); | |
6f1ffb06 MA |
296 | } |
297 | ||
298 | int | |
299 | lzc_snaprange_space(const char *firstsnap, const char *lastsnap, | |
300 | uint64_t *usedp) | |
301 | { | |
302 | nvlist_t *args; | |
303 | nvlist_t *result; | |
304 | int err; | |
eca7b760 | 305 | char fs[ZFS_MAX_DATASET_NAME_LEN]; |
6f1ffb06 MA |
306 | char *atp; |
307 | ||
308 | /* determine the fs name */ | |
309 | (void) strlcpy(fs, firstsnap, sizeof (fs)); | |
310 | atp = strchr(fs, '@'); | |
311 | if (atp == NULL) | |
312 | return (EINVAL); | |
313 | *atp = '\0'; | |
314 | ||
315 | args = fnvlist_alloc(); | |
316 | fnvlist_add_string(args, "firstsnap", firstsnap); | |
317 | ||
318 | err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result); | |
319 | nvlist_free(args); | |
320 | if (err == 0) | |
321 | *usedp = fnvlist_lookup_uint64(result, "used"); | |
322 | fnvlist_free(result); | |
323 | ||
324 | return (err); | |
325 | } | |
326 | ||
327 | boolean_t | |
328 | lzc_exists(const char *dataset) | |
329 | { | |
330 | /* | |
331 | * The objset_stats ioctl is still legacy, so we need to construct our | |
332 | * own zfs_cmd_t rather than using zfsc_ioctl(). | |
333 | */ | |
13fe0198 | 334 | zfs_cmd_t zc = {"\0"}; |
6f1ffb06 | 335 | |
e2454897 GM |
336 | ASSERT3S(g_refcount, >, 0); |
337 | VERIFY3S(g_fd, !=, -1); | |
338 | ||
6f1ffb06 MA |
339 | (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); |
340 | return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0); | |
341 | } | |
342 | ||
13fe0198 MA |
343 | /* |
344 | * Create "user holds" on snapshots. If there is a hold on a snapshot, | |
345 | * the snapshot can not be destroyed. (However, it can be marked for deletion | |
346 | * by lzc_destroy_snaps(defer=B_TRUE).) | |
347 | * | |
348 | * The keys in the nvlist are snapshot names. | |
349 | * The snapshots must all be in the same pool. | |
350 | * The value is the name of the hold (string type). | |
351 | * | |
352 | * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL). | |
353 | * In this case, when the cleanup_fd is closed (including on process | |
354 | * termination), the holds will be released. If the system is shut down | |
355 | * uncleanly, the holds will be released when the pool is next opened | |
356 | * or imported. | |
357 | * | |
95fd54a1 | 358 | * Holds for snapshots which don't exist will be skipped and have an entry |
1a077756 | 359 | * added to errlist, but will not cause an overall failure. |
95fd54a1 | 360 | * |
1a077756 | 361 | * The return value will be 0 if all holds, for snapshots that existed, |
b8fce77b | 362 | * were successfully created. |
95fd54a1 SH |
363 | * |
364 | * Otherwise the return value will be the errno of a (unspecified) hold that | |
365 | * failed and no holds will be created. | |
366 | * | |
367 | * In all cases the errlist will have an entry for each hold that failed | |
368 | * (name = snapshot), with its value being the error code (int32). | |
13fe0198 MA |
369 | */ |
370 | int | |
371 | lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist) | |
372 | { | |
eca7b760 | 373 | char pool[ZFS_MAX_DATASET_NAME_LEN]; |
13fe0198 MA |
374 | nvlist_t *args; |
375 | nvpair_t *elem; | |
376 | int error; | |
377 | ||
378 | /* determine the pool name */ | |
379 | elem = nvlist_next_nvpair(holds, NULL); | |
380 | if (elem == NULL) | |
381 | return (0); | |
382 | (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); | |
383 | pool[strcspn(pool, "/@")] = '\0'; | |
384 | ||
385 | args = fnvlist_alloc(); | |
386 | fnvlist_add_nvlist(args, "holds", holds); | |
387 | if (cleanup_fd != -1) | |
388 | fnvlist_add_int32(args, "cleanup_fd", cleanup_fd); | |
389 | ||
390 | error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist); | |
391 | nvlist_free(args); | |
392 | return (error); | |
393 | } | |
394 | ||
395 | /* | |
396 | * Release "user holds" on snapshots. If the snapshot has been marked for | |
397 | * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have | |
398 | * any clones, and all the user holds are removed, then the snapshot will be | |
399 | * destroyed. | |
400 | * | |
401 | * The keys in the nvlist are snapshot names. | |
402 | * The snapshots must all be in the same pool. | |
d5884c34 | 403 | * The value is an nvlist whose keys are the holds to remove. |
13fe0198 | 404 | * |
95fd54a1 | 405 | * Holds which failed to release because they didn't exist will have an entry |
1a077756 | 406 | * added to errlist, but will not cause an overall failure. |
95fd54a1 SH |
407 | * |
408 | * The return value will be 0 if the nvl holds was empty or all holds that | |
1a077756 | 409 | * existed, were successfully removed. |
95fd54a1 SH |
410 | * |
411 | * Otherwise the return value will be the errno of a (unspecified) hold that | |
412 | * failed to release and no holds will be released. | |
413 | * | |
414 | * In all cases the errlist will have an entry for each hold that failed to | |
415 | * to release. | |
13fe0198 MA |
416 | */ |
417 | int | |
418 | lzc_release(nvlist_t *holds, nvlist_t **errlist) | |
419 | { | |
eca7b760 | 420 | char pool[ZFS_MAX_DATASET_NAME_LEN]; |
13fe0198 MA |
421 | nvpair_t *elem; |
422 | ||
423 | /* determine the pool name */ | |
424 | elem = nvlist_next_nvpair(holds, NULL); | |
425 | if (elem == NULL) | |
426 | return (0); | |
427 | (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); | |
428 | pool[strcspn(pool, "/@")] = '\0'; | |
429 | ||
430 | return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist)); | |
431 | } | |
432 | ||
433 | /* | |
434 | * Retrieve list of user holds on the specified snapshot. | |
435 | * | |
d5884c34 | 436 | * On success, *holdsp will be set to an nvlist which the caller must free. |
13fe0198 MA |
437 | * The keys are the names of the holds, and the value is the creation time |
438 | * of the hold (uint64) in seconds since the epoch. | |
439 | */ | |
440 | int | |
441 | lzc_get_holds(const char *snapname, nvlist_t **holdsp) | |
442 | { | |
443 | int error; | |
444 | nvlist_t *innvl = fnvlist_alloc(); | |
445 | error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp); | |
446 | fnvlist_free(innvl); | |
447 | return (error); | |
448 | } | |
449 | ||
6f1ffb06 | 450 | /* |
9b67f605 MA |
451 | * Generate a zfs send stream for the specified snapshot and write it to |
452 | * the specified file descriptor. | |
da536844 MA |
453 | * |
454 | * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap") | |
455 | * | |
456 | * If "from" is NULL, a full (non-incremental) stream will be sent. | |
457 | * If "from" is non-NULL, it must be the full name of a snapshot or | |
458 | * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or | |
459 | * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or | |
460 | * bookmark must represent an earlier point in the history of "snapname"). | |
461 | * It can be an earlier snapshot in the same filesystem or zvol as "snapname", | |
462 | * or it can be the origin of "snapname"'s filesystem, or an earlier | |
463 | * snapshot in the origin, etc. | |
464 | * | |
465 | * "fd" is the file descriptor to write the send stream to. | |
9b67f605 | 466 | * |
f1512ee6 MA |
467 | * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted |
468 | * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT | |
469 | * records with drr_blksz > 128K. | |
470 | * | |
9b67f605 MA |
471 | * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted |
472 | * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA, | |
473 | * which the receiving system must support (as indicated by support | |
474 | * for the "embedded_data" feature). | |
6f1ffb06 MA |
475 | */ |
476 | int | |
9b67f605 MA |
477 | lzc_send(const char *snapname, const char *from, int fd, |
478 | enum lzc_send_flags flags) | |
47dfff3b MA |
479 | { |
480 | return (lzc_send_resume(snapname, from, fd, flags, 0, 0)); | |
481 | } | |
482 | ||
483 | int | |
484 | lzc_send_resume(const char *snapname, const char *from, int fd, | |
485 | enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff) | |
6f1ffb06 MA |
486 | { |
487 | nvlist_t *args; | |
488 | int err; | |
489 | ||
490 | args = fnvlist_alloc(); | |
491 | fnvlist_add_int32(args, "fd", fd); | |
da536844 MA |
492 | if (from != NULL) |
493 | fnvlist_add_string(args, "fromsnap", from); | |
f1512ee6 MA |
494 | if (flags & LZC_SEND_FLAG_LARGE_BLOCK) |
495 | fnvlist_add_boolean(args, "largeblockok"); | |
2aa34383 DK |
496 | if (flags & LZC_SEND_FLAG_COMPRESS) |
497 | fnvlist_add_boolean(args, "compressok"); | |
9b67f605 MA |
498 | if (flags & LZC_SEND_FLAG_EMBED_DATA) |
499 | fnvlist_add_boolean(args, "embedok"); | |
47dfff3b MA |
500 | if (resumeobj != 0 || resumeoff != 0) { |
501 | fnvlist_add_uint64(args, "resume_object", resumeobj); | |
502 | fnvlist_add_uint64(args, "resume_offset", resumeoff); | |
503 | } | |
6f1ffb06 MA |
504 | err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL); |
505 | nvlist_free(args); | |
506 | return (err); | |
507 | } | |
508 | ||
509 | /* | |
5dc8b736 MG |
510 | * "from" can be NULL, a snapshot, or a bookmark. |
511 | * | |
512 | * If from is NULL, a full (non-incremental) stream will be estimated. This | |
513 | * is calculated very efficiently. | |
514 | * | |
515 | * If from is a snapshot, lzc_send_space uses the deadlists attached to | |
516 | * each snapshot to efficiently estimate the stream size. | |
517 | * | |
518 | * If from is a bookmark, the indirect blocks in the destination snapshot | |
519 | * are traversed, looking for blocks with a birth time since the creation TXG of | |
520 | * the snapshot this bookmark was created from. This will result in | |
521 | * significantly more I/O and be less efficient than a send space estimation on | |
522 | * an equivalent snapshot. | |
6f1ffb06 MA |
523 | */ |
524 | int | |
2aa34383 DK |
525 | lzc_send_space(const char *snapname, const char *from, |
526 | enum lzc_send_flags flags, uint64_t *spacep) | |
6f1ffb06 MA |
527 | { |
528 | nvlist_t *args; | |
529 | nvlist_t *result; | |
530 | int err; | |
531 | ||
532 | args = fnvlist_alloc(); | |
5dc8b736 MG |
533 | if (from != NULL) |
534 | fnvlist_add_string(args, "from", from); | |
2aa34383 DK |
535 | if (flags & LZC_SEND_FLAG_LARGE_BLOCK) |
536 | fnvlist_add_boolean(args, "largeblockok"); | |
537 | if (flags & LZC_SEND_FLAG_EMBED_DATA) | |
538 | fnvlist_add_boolean(args, "embedok"); | |
539 | if (flags & LZC_SEND_FLAG_COMPRESS) | |
540 | fnvlist_add_boolean(args, "compressok"); | |
6f1ffb06 MA |
541 | err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result); |
542 | nvlist_free(args); | |
543 | if (err == 0) | |
544 | *spacep = fnvlist_lookup_uint64(result, "space"); | |
545 | nvlist_free(result); | |
546 | return (err); | |
547 | } | |
548 | ||
549 | static int | |
550 | recv_read(int fd, void *buf, int ilen) | |
551 | { | |
552 | char *cp = buf; | |
553 | int rv; | |
554 | int len = ilen; | |
555 | ||
556 | do { | |
557 | rv = read(fd, cp, len); | |
558 | cp += rv; | |
559 | len -= rv; | |
560 | } while (rv > 0); | |
561 | ||
562 | if (rv < 0 || len != 0) | |
563 | return (EIO); | |
564 | ||
565 | return (0); | |
566 | } | |
567 | ||
43e52edd BB |
568 | /* |
569 | * Linux adds ZFS_IOC_RECV_NEW for resumable streams and preserves the legacy | |
570 | * ZFS_IOC_RECV user/kernel interface. The new interface supports all stream | |
571 | * options but is currently only used for resumable streams. This way updated | |
572 | * user space utilities will interoperate with older kernel modules. | |
573 | * | |
574 | * Non-Linux OpenZFS platforms have opted to modify the legacy interface. | |
575 | */ | |
47dfff3b | 576 | static int |
fd41e935 | 577 | recv_impl(const char *snapname, nvlist_t *props, const char *origin, |
43e52edd BB |
578 | boolean_t force, boolean_t resumable, int input_fd, |
579 | const dmu_replay_record_t *begin_record, int cleanup_fd, | |
580 | uint64_t *read_bytes, uint64_t *errflags, uint64_t *action_handle, | |
581 | nvlist_t **errors) | |
6f1ffb06 | 582 | { |
43e52edd BB |
583 | dmu_replay_record_t drr; |
584 | char fsname[MAXPATHLEN]; | |
6f1ffb06 | 585 | char *atp; |
6f1ffb06 MA |
586 | int error; |
587 | ||
e2454897 GM |
588 | ASSERT3S(g_refcount, >, 0); |
589 | VERIFY3S(g_fd, !=, -1); | |
590 | ||
43e52edd BB |
591 | /* Set 'fsname' to the name of containing filesystem */ |
592 | (void) strlcpy(fsname, snapname, sizeof (fsname)); | |
593 | atp = strchr(fsname, '@'); | |
6f1ffb06 MA |
594 | if (atp == NULL) |
595 | return (EINVAL); | |
596 | *atp = '\0'; | |
597 | ||
43e52edd BB |
598 | /* If the fs does not exist, try its parent. */ |
599 | if (!lzc_exists(fsname)) { | |
600 | char *slashp = strrchr(fsname, '/'); | |
6f1ffb06 MA |
601 | if (slashp == NULL) |
602 | return (ENOENT); | |
603 | *slashp = '\0'; | |
43e52edd | 604 | } |
6f1ffb06 | 605 | |
43e52edd BB |
606 | /* |
607 | * The begin_record is normally a non-byteswapped BEGIN record. | |
608 | * For resumable streams it may be set to any non-byteswapped | |
609 | * dmu_replay_record_t. | |
610 | */ | |
611 | if (begin_record == NULL) { | |
612 | error = recv_read(input_fd, &drr, sizeof (drr)); | |
613 | if (error != 0) | |
614 | return (error); | |
615 | } else { | |
616 | drr = *begin_record; | |
6f1ffb06 MA |
617 | } |
618 | ||
43e52edd BB |
619 | if (resumable) { |
620 | nvlist_t *outnvl = NULL; | |
621 | nvlist_t *innvl = fnvlist_alloc(); | |
6f1ffb06 | 622 | |
43e52edd | 623 | fnvlist_add_string(innvl, "snapname", snapname); |
6f1ffb06 | 624 | |
43e52edd BB |
625 | if (props != NULL) |
626 | fnvlist_add_nvlist(innvl, "props", props); | |
6f1ffb06 | 627 | |
43e52edd BB |
628 | if (origin != NULL && strlen(origin)) |
629 | fnvlist_add_string(innvl, "origin", origin); | |
630 | ||
631 | fnvlist_add_byte_array(innvl, "begin_record", | |
02730c33 | 632 | (uchar_t *)&drr, sizeof (drr)); |
43e52edd BB |
633 | |
634 | fnvlist_add_int32(innvl, "input_fd", input_fd); | |
635 | ||
636 | if (force) | |
637 | fnvlist_add_boolean(innvl, "force"); | |
638 | ||
639 | if (resumable) | |
640 | fnvlist_add_boolean(innvl, "resumable"); | |
641 | ||
642 | if (cleanup_fd >= 0) | |
643 | fnvlist_add_int32(innvl, "cleanup_fd", cleanup_fd); | |
644 | ||
645 | if (action_handle != NULL) | |
646 | fnvlist_add_uint64(innvl, "action_handle", | |
647 | *action_handle); | |
648 | ||
649 | error = lzc_ioctl(ZFS_IOC_RECV_NEW, fsname, innvl, &outnvl); | |
650 | ||
651 | if (error == 0 && read_bytes != NULL) | |
652 | error = nvlist_lookup_uint64(outnvl, "read_bytes", | |
653 | read_bytes); | |
654 | ||
655 | if (error == 0 && errflags != NULL) | |
656 | error = nvlist_lookup_uint64(outnvl, "error_flags", | |
657 | errflags); | |
658 | ||
659 | if (error == 0 && action_handle != NULL) | |
660 | error = nvlist_lookup_uint64(outnvl, "action_handle", | |
661 | action_handle); | |
662 | ||
663 | if (error == 0 && errors != NULL) { | |
664 | nvlist_t *nvl; | |
665 | error = nvlist_lookup_nvlist(outnvl, "errors", &nvl); | |
666 | if (error == 0) | |
667 | *errors = fnvlist_dup(nvl); | |
668 | } | |
669 | ||
670 | fnvlist_free(innvl); | |
671 | fnvlist_free(outnvl); | |
fd41e935 | 672 | } else { |
43e52edd BB |
673 | zfs_cmd_t zc = {"\0"}; |
674 | char *packed = NULL; | |
675 | size_t size; | |
6f1ffb06 | 676 | |
43e52edd | 677 | ASSERT3S(g_refcount, >, 0); |
6f1ffb06 | 678 | |
43e52edd BB |
679 | (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_value)); |
680 | (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); | |
6f1ffb06 | 681 | |
43e52edd BB |
682 | if (props != NULL) { |
683 | packed = fnvlist_pack(props, &size); | |
684 | zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; | |
685 | zc.zc_nvlist_src_size = size; | |
686 | } | |
47dfff3b | 687 | |
43e52edd BB |
688 | if (origin != NULL) |
689 | (void) strlcpy(zc.zc_string, origin, | |
690 | sizeof (zc.zc_string)); | |
6f1ffb06 | 691 | |
43e52edd BB |
692 | ASSERT3S(drr.drr_type, ==, DRR_BEGIN); |
693 | zc.zc_begin_record = drr.drr_u.drr_begin; | |
694 | zc.zc_guid = force; | |
695 | zc.zc_cookie = input_fd; | |
696 | zc.zc_cleanup_fd = -1; | |
697 | zc.zc_action_handle = 0; | |
698 | ||
699 | if (cleanup_fd >= 0) | |
700 | zc.zc_cleanup_fd = cleanup_fd; | |
701 | ||
702 | if (action_handle != NULL) | |
703 | zc.zc_action_handle = *action_handle; | |
704 | ||
705 | zc.zc_nvlist_dst_size = 128 * 1024; | |
706 | zc.zc_nvlist_dst = (uint64_t)(uintptr_t) | |
707 | malloc(zc.zc_nvlist_dst_size); | |
708 | ||
709 | error = ioctl(g_fd, ZFS_IOC_RECV, &zc); | |
710 | if (error != 0) { | |
711 | error = errno; | |
712 | } else { | |
713 | if (read_bytes != NULL) | |
714 | *read_bytes = zc.zc_cookie; | |
715 | ||
716 | if (errflags != NULL) | |
717 | *errflags = zc.zc_obj; | |
718 | ||
719 | if (action_handle != NULL) | |
720 | *action_handle = zc.zc_action_handle; | |
721 | ||
722 | if (errors != NULL) | |
723 | VERIFY0(nvlist_unpack( | |
724 | (void *)(uintptr_t)zc.zc_nvlist_dst, | |
725 | zc.zc_nvlist_dst_size, errors, KM_SLEEP)); | |
726 | } | |
727 | ||
728 | if (packed != NULL) | |
729 | fnvlist_pack_free(packed, size); | |
730 | free((void *)(uintptr_t)zc.zc_nvlist_dst); | |
731 | } | |
6f1ffb06 | 732 | |
6f1ffb06 MA |
733 | return (error); |
734 | } | |
46ba1e59 | 735 | |
47dfff3b MA |
736 | /* |
737 | * The simplest receive case: receive from the specified fd, creating the | |
738 | * specified snapshot. Apply the specified properties as "received" properties | |
739 | * (which can be overridden by locally-set properties). If the stream is a | |
740 | * clone, its origin snapshot must be specified by 'origin'. The 'force' | |
741 | * flag will cause the target filesystem to be rolled back or destroyed if | |
742 | * necessary to receive. | |
743 | * | |
744 | * Return 0 on success or an errno on failure. | |
745 | * | |
746 | * Note: this interface does not work on dedup'd streams | |
747 | * (those with DMU_BACKUP_FEATURE_DEDUP). | |
748 | */ | |
749 | int | |
750 | lzc_receive(const char *snapname, nvlist_t *props, const char *origin, | |
751 | boolean_t force, int fd) | |
752 | { | |
43e52edd BB |
753 | return (recv_impl(snapname, props, origin, force, B_FALSE, fd, |
754 | NULL, -1, NULL, NULL, NULL, NULL)); | |
47dfff3b MA |
755 | } |
756 | ||
757 | /* | |
758 | * Like lzc_receive, but if the receive fails due to premature stream | |
759 | * termination, the intermediate state will be preserved on disk. In this | |
760 | * case, ECKSUM will be returned. The receive may subsequently be resumed | |
761 | * with a resuming send stream generated by lzc_send_resume(). | |
762 | */ | |
763 | int | |
764 | lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin, | |
765 | boolean_t force, int fd) | |
766 | { | |
43e52edd BB |
767 | return (recv_impl(snapname, props, origin, force, B_TRUE, fd, |
768 | NULL, -1, NULL, NULL, NULL, NULL)); | |
fd41e935 BB |
769 | } |
770 | ||
771 | /* | |
772 | * Like lzc_receive, but allows the caller to read the begin record and then to | |
773 | * pass it in. That could be useful if the caller wants to derive, for example, | |
774 | * the snapname or the origin parameters based on the information contained in | |
775 | * the begin record. | |
776 | * The begin record must be in its original form as read from the stream, | |
777 | * in other words, it should not be byteswapped. | |
778 | * | |
779 | * The 'resumable' parameter allows to obtain the same behavior as with | |
780 | * lzc_receive_resumable. | |
781 | */ | |
782 | int | |
783 | lzc_receive_with_header(const char *snapname, nvlist_t *props, | |
784 | const char *origin, boolean_t force, boolean_t resumable, int fd, | |
785 | const dmu_replay_record_t *begin_record) | |
786 | { | |
787 | if (begin_record == NULL) | |
788 | return (EINVAL); | |
789 | return (recv_impl(snapname, props, origin, force, resumable, fd, | |
43e52edd BB |
790 | begin_record, -1, NULL, NULL, NULL, NULL)); |
791 | } | |
792 | ||
793 | /* | |
794 | * Like lzc_receive, but allows the caller to pass all supported arguments | |
795 | * and retrieve all values returned. The only additional input parameter | |
796 | * is 'cleanup_fd' which is used to set a cleanup-on-exit file descriptor. | |
797 | * | |
798 | * The following parameters all provide return values. Several may be set | |
799 | * in the failure case and will contain additional information. | |
800 | * | |
801 | * The 'read_bytes' value will be set to the total number of bytes read. | |
802 | * | |
803 | * The 'errflags' value will contain zprop_errflags_t flags which are | |
804 | * used to describe any failures. | |
805 | * | |
806 | * The 'action_handle' is used to pass the handle for this guid/ds mapping. | |
807 | * It should be set to zero on first call and will contain an updated handle | |
808 | * on success, it should be passed in subsequent calls. | |
809 | * | |
810 | * The 'errors' nvlist contains an entry for each unapplied received | |
811 | * property. Callers are responsible for freeing this nvlist. | |
812 | */ | |
813 | int lzc_receive_one(const char *snapname, nvlist_t *props, | |
814 | const char *origin, boolean_t force, boolean_t resumable, int input_fd, | |
815 | const dmu_replay_record_t *begin_record, int cleanup_fd, | |
816 | uint64_t *read_bytes, uint64_t *errflags, uint64_t *action_handle, | |
817 | nvlist_t **errors) | |
818 | { | |
819 | return (recv_impl(snapname, props, origin, force, resumable, | |
820 | input_fd, begin_record, cleanup_fd, read_bytes, errflags, | |
821 | action_handle, errors)); | |
47dfff3b MA |
822 | } |
823 | ||
46ba1e59 MA |
824 | /* |
825 | * Roll back this filesystem or volume to its most recent snapshot. | |
826 | * If snapnamebuf is not NULL, it will be filled in with the name | |
827 | * of the most recent snapshot. | |
828 | * | |
829 | * Return 0 on success or an errno on failure. | |
830 | */ | |
831 | int | |
832 | lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen) | |
833 | { | |
834 | nvlist_t *args; | |
835 | nvlist_t *result; | |
836 | int err; | |
837 | ||
838 | args = fnvlist_alloc(); | |
839 | err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); | |
840 | nvlist_free(args); | |
841 | if (err == 0 && snapnamebuf != NULL) { | |
842 | const char *snapname = fnvlist_lookup_string(result, "target"); | |
843 | (void) strlcpy(snapnamebuf, snapname, snapnamelen); | |
844 | } | |
bb7ffdaf GM |
845 | nvlist_free(result); |
846 | ||
46ba1e59 MA |
847 | return (err); |
848 | } | |
da536844 MA |
849 | |
850 | /* | |
851 | * Creates bookmarks. | |
852 | * | |
853 | * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to | |
854 | * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and | |
855 | * snapshots must be in the same pool. | |
856 | * | |
857 | * The returned results nvlist will have an entry for each bookmark that failed. | |
858 | * The value will be the (int32) error code. | |
859 | * | |
860 | * The return value will be 0 if all bookmarks were created, otherwise it will | |
861 | * be the errno of a (undetermined) bookmarks that failed. | |
862 | */ | |
863 | int | |
864 | lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist) | |
865 | { | |
866 | nvpair_t *elem; | |
867 | int error; | |
eca7b760 | 868 | char pool[ZFS_MAX_DATASET_NAME_LEN]; |
da536844 MA |
869 | |
870 | /* determine the pool name */ | |
871 | elem = nvlist_next_nvpair(bookmarks, NULL); | |
872 | if (elem == NULL) | |
873 | return (0); | |
874 | (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); | |
875 | pool[strcspn(pool, "/#")] = '\0'; | |
876 | ||
877 | error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist); | |
878 | ||
879 | return (error); | |
880 | } | |
881 | ||
882 | /* | |
883 | * Retrieve bookmarks. | |
884 | * | |
885 | * Retrieve the list of bookmarks for the given file system. The props | |
886 | * parameter is an nvlist of property names (with no values) that will be | |
887 | * returned for each bookmark. | |
888 | * | |
889 | * The following are valid properties on bookmarks, all of which are numbers | |
890 | * (represented as uint64 in the nvlist) | |
891 | * | |
892 | * "guid" - globally unique identifier of the snapshot it refers to | |
893 | * "createtxg" - txg when the snapshot it refers to was created | |
894 | * "creation" - timestamp when the snapshot it refers to was created | |
895 | * | |
896 | * The format of the returned nvlist as follows: | |
897 | * <short name of bookmark> -> { | |
898 | * <name of property> -> { | |
899 | * "value" -> uint64 | |
900 | * } | |
901 | * } | |
902 | */ | |
903 | int | |
904 | lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks) | |
905 | { | |
906 | return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks)); | |
907 | } | |
908 | ||
909 | /* | |
910 | * Destroys bookmarks. | |
911 | * | |
912 | * The keys in the bmarks nvlist are the bookmarks to be destroyed. | |
913 | * They must all be in the same pool. Bookmarks are specified as | |
914 | * <fs>#<bmark>. | |
915 | * | |
916 | * Bookmarks that do not exist will be silently ignored. | |
917 | * | |
918 | * The return value will be 0 if all bookmarks that existed were destroyed. | |
919 | * | |
920 | * Otherwise the return value will be the errno of a (undetermined) bookmark | |
921 | * that failed, no bookmarks will be destroyed, and the errlist will have an | |
922 | * entry for each bookmarks that failed. The value in the errlist will be | |
923 | * the (int32) error code. | |
924 | */ | |
925 | int | |
926 | lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist) | |
927 | { | |
928 | nvpair_t *elem; | |
929 | int error; | |
eca7b760 | 930 | char pool[ZFS_MAX_DATASET_NAME_LEN]; |
da536844 MA |
931 | |
932 | /* determine the pool name */ | |
933 | elem = nvlist_next_nvpair(bmarks, NULL); | |
934 | if (elem == NULL) | |
935 | return (0); | |
936 | (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); | |
937 | pool[strcspn(pool, "/#")] = '\0'; | |
938 | ||
939 | error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist); | |
940 | ||
941 | return (error); | |
942 | } |