Print this page
8264 want support for promoting datasets in libzfs_core
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/lib/libzfs_core/common/libzfs_core.c
+++ new/usr/src/lib/libzfs_core/common/libzfs_core.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
|
↓ open down ↓ |
15 lines elided |
↑ open up ↑ |
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 25 * Copyright (c) 2014 Integros [integros.com]
26 + * Copyright 2017 RackTop Systems.
26 27 */
27 28
28 29 /*
29 30 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs.
30 31 * It has the following characteristics:
31 32 *
32 33 * - Thread Safe. libzfs_core is accessible concurrently from multiple
33 34 * threads. This is accomplished primarily by avoiding global data
34 35 * (e.g. caching). Since it's thread-safe, there is no reason for a
35 36 * process to have multiple libzfs "instances". Therefore, we store
36 37 * our few pieces of data (e.g. the file descriptor) in global
37 38 * variables. The fd is reference-counted so that the libzfs_core
38 39 * library can be "initialized" multiple times (e.g. by different
39 40 * consumers within the same process).
40 41 *
41 42 * - Committed Interface. The libzfs_core interface will be committed,
42 43 * therefore consumers can compile against it and be confident that
43 44 * their code will continue to work on future releases of this code.
44 45 * Currently, the interface is Evolving (not Committed), but we intend
45 46 * to commit to it once it is more complete and we determine that it
46 47 * meets the needs of all consumers.
47 48 *
48 49 * - Programatic Error Handling. libzfs_core communicates errors with
49 50 * defined error numbers, and doesn't print anything to stdout/stderr.
50 51 *
51 52 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments
52 53 * to/from the kernel ioctls. There is generally a 1:1 correspondence
53 54 * between libzfs_core functions and ioctls to /dev/zfs.
54 55 *
55 56 * - Clear Atomicity. Because libzfs_core functions are generally 1:1
56 57 * with kernel ioctls, and kernel ioctls are general atomic, each
57 58 * libzfs_core function is atomic. For example, creating multiple
58 59 * snapshots with a single call to lzc_snapshot() is atomic -- it
59 60 * can't fail with only some of the requested snapshots created, even
60 61 * in the event of power loss or system crash.
61 62 *
62 63 * - Continued libzfs Support. Some higher-level operations (e.g.
63 64 * support for "zfs send -R") are too complicated to fit the scope of
64 65 * libzfs_core. This functionality will continue to live in libzfs.
65 66 * Where appropriate, libzfs will use the underlying atomic operations
66 67 * of libzfs_core. For example, libzfs may implement "zfs send -R |
67 68 * zfs receive" by using individual "send one snapshot", rename,
68 69 * destroy, and "receive one snapshot" operations in libzfs_core.
69 70 * /sbin/zfs and /zbin/zpool will link with both libzfs and
70 71 * libzfs_core. Other consumers should aim to use only libzfs_core,
71 72 * since that will be the supported, stable interface going forwards.
72 73 */
73 74
74 75 #include <libzfs_core.h>
75 76 #include <ctype.h>
76 77 #include <unistd.h>
77 78 #include <stdlib.h>
78 79 #include <string.h>
79 80 #include <errno.h>
80 81 #include <fcntl.h>
81 82 #include <pthread.h>
82 83 #include <sys/nvpair.h>
83 84 #include <sys/param.h>
84 85 #include <sys/types.h>
85 86 #include <sys/stat.h>
86 87 #include <sys/zfs_ioctl.h>
87 88
88 89 static int g_fd = -1;
89 90 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER;
90 91 static int g_refcount;
91 92
92 93 int
93 94 libzfs_core_init(void)
94 95 {
95 96 (void) pthread_mutex_lock(&g_lock);
96 97 if (g_refcount == 0) {
97 98 g_fd = open("/dev/zfs", O_RDWR);
98 99 if (g_fd < 0) {
99 100 (void) pthread_mutex_unlock(&g_lock);
100 101 return (errno);
101 102 }
102 103 }
103 104 g_refcount++;
104 105 (void) pthread_mutex_unlock(&g_lock);
105 106 return (0);
106 107 }
107 108
108 109 void
109 110 libzfs_core_fini(void)
110 111 {
111 112 (void) pthread_mutex_lock(&g_lock);
112 113 ASSERT3S(g_refcount, >, 0);
113 114
114 115 if (g_refcount > 0)
115 116 g_refcount--;
116 117
117 118 if (g_refcount == 0 && g_fd != -1) {
118 119 (void) close(g_fd);
119 120 g_fd = -1;
120 121 }
121 122 (void) pthread_mutex_unlock(&g_lock);
122 123 }
123 124
124 125 static int
125 126 lzc_ioctl(zfs_ioc_t ioc, const char *name,
126 127 nvlist_t *source, nvlist_t **resultp)
127 128 {
128 129 zfs_cmd_t zc = { 0 };
129 130 int error = 0;
130 131 char *packed;
131 132 size_t size;
132 133
133 134 ASSERT3S(g_refcount, >, 0);
134 135 VERIFY3S(g_fd, !=, -1);
135 136
136 137 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
137 138
138 139 packed = fnvlist_pack(source, &size);
139 140 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
140 141 zc.zc_nvlist_src_size = size;
141 142
142 143 if (resultp != NULL) {
143 144 *resultp = NULL;
144 145 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
145 146 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
146 147 malloc(zc.zc_nvlist_dst_size);
147 148 if (zc.zc_nvlist_dst == NULL) {
148 149 error = ENOMEM;
149 150 goto out;
150 151 }
151 152 }
152 153
153 154 while (ioctl(g_fd, ioc, &zc) != 0) {
154 155 if (errno == ENOMEM && resultp != NULL) {
155 156 free((void *)(uintptr_t)zc.zc_nvlist_dst);
156 157 zc.zc_nvlist_dst_size *= 2;
157 158 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
158 159 malloc(zc.zc_nvlist_dst_size);
159 160 if (zc.zc_nvlist_dst == NULL) {
160 161 error = ENOMEM;
161 162 goto out;
162 163 }
163 164 } else {
164 165 error = errno;
165 166 break;
166 167 }
167 168 }
168 169 if (zc.zc_nvlist_dst_filled) {
169 170 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
170 171 zc.zc_nvlist_dst_size);
171 172 }
172 173
173 174 out:
174 175 fnvlist_pack_free(packed, size);
175 176 free((void *)(uintptr_t)zc.zc_nvlist_dst);
176 177 return (error);
177 178 }
178 179
179 180 int
180 181 lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props)
181 182 {
182 183 int error;
183 184 nvlist_t *args = fnvlist_alloc();
184 185 fnvlist_add_int32(args, "type", (dmu_objset_type_t)type);
185 186 if (props != NULL)
186 187 fnvlist_add_nvlist(args, "props", props);
187 188 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
188 189 nvlist_free(args);
189 190 return (error);
190 191 }
191 192
192 193 int
193 194 lzc_clone(const char *fsname, const char *origin,
194 195 nvlist_t *props)
195 196 {
|
↓ open down ↓ |
160 lines elided |
↑ open up ↑ |
196 197 int error;
197 198 nvlist_t *args = fnvlist_alloc();
198 199 fnvlist_add_string(args, "origin", origin);
199 200 if (props != NULL)
200 201 fnvlist_add_nvlist(args, "props", props);
201 202 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
202 203 nvlist_free(args);
203 204 return (error);
204 205 }
205 206
207 +int
208 +lzc_promote(const char *fsname, char *snapnamebuf, int snapnamelen)
209 +{
210 + /*
211 + * The promote ioctl is still legacy, so we need to construct our
212 + * own zfs_cmd_t rather than using lzc_ioctl().
213 + */
214 + zfs_cmd_t zc = { 0 };
215 +
216 + ASSERT3S(g_refcount, >, 0);
217 + VERIFY3S(g_fd, !=, -1);
218 +
219 + (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name));
220 + if (ioctl(g_fd, ZFS_IOC_PROMOTE, &zc) != 0) {
221 + if (errno == EEXIST && snapnamebuf != NULL)
222 + (void) strlcpy(snapnamebuf, zc.zc_string, snapnamelen);
223 + return (errno);
224 + }
225 + return (0);
226 +}
227 +
206 228 /*
207 229 * Creates snapshots.
208 230 *
209 231 * The keys in the snaps nvlist are the snapshots to be created.
210 232 * They must all be in the same pool.
211 233 *
212 234 * The props nvlist is properties to set. Currently only user properties
213 235 * are supported. { user:prop_name -> string value }
214 236 *
215 237 * The returned results nvlist will have an entry for each snapshot that failed.
216 238 * The value will be the (int32) error code.
217 239 *
218 240 * The return value will be 0 if all snapshots were created, otherwise it will
219 241 * be the errno of a (unspecified) snapshot that failed.
220 242 */
221 243 int
222 244 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
223 245 {
224 246 nvpair_t *elem;
225 247 nvlist_t *args;
226 248 int error;
227 249 char pool[ZFS_MAX_DATASET_NAME_LEN];
228 250
229 251 *errlist = NULL;
230 252
231 253 /* determine the pool name */
232 254 elem = nvlist_next_nvpair(snaps, NULL);
233 255 if (elem == NULL)
234 256 return (0);
235 257 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
236 258 pool[strcspn(pool, "/@")] = '\0';
237 259
238 260 args = fnvlist_alloc();
239 261 fnvlist_add_nvlist(args, "snaps", snaps);
240 262 if (props != NULL)
241 263 fnvlist_add_nvlist(args, "props", props);
242 264
243 265 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
244 266 nvlist_free(args);
245 267
246 268 return (error);
247 269 }
248 270
249 271 /*
250 272 * Destroys snapshots.
251 273 *
252 274 * The keys in the snaps nvlist are the snapshots to be destroyed.
253 275 * They must all be in the same pool.
254 276 *
255 277 * Snapshots that do not exist will be silently ignored.
256 278 *
257 279 * If 'defer' is not set, and a snapshot has user holds or clones, the
258 280 * destroy operation will fail and none of the snapshots will be
259 281 * destroyed.
260 282 *
261 283 * If 'defer' is set, and a snapshot has user holds or clones, it will be
262 284 * marked for deferred destruction, and will be destroyed when the last hold
263 285 * or clone is removed/destroyed.
264 286 *
265 287 * The return value will be 0 if all snapshots were destroyed (or marked for
266 288 * later destruction if 'defer' is set) or didn't exist to begin with.
267 289 *
268 290 * Otherwise the return value will be the errno of a (unspecified) snapshot
269 291 * that failed, no snapshots will be destroyed, and the errlist will have an
270 292 * entry for each snapshot that failed. The value in the errlist will be
271 293 * the (int32) error code.
272 294 */
273 295 int
274 296 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
275 297 {
276 298 nvpair_t *elem;
277 299 nvlist_t *args;
278 300 int error;
279 301 char pool[ZFS_MAX_DATASET_NAME_LEN];
280 302
281 303 /* determine the pool name */
282 304 elem = nvlist_next_nvpair(snaps, NULL);
283 305 if (elem == NULL)
284 306 return (0);
285 307 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
286 308 pool[strcspn(pool, "/@")] = '\0';
287 309
288 310 args = fnvlist_alloc();
289 311 fnvlist_add_nvlist(args, "snaps", snaps);
290 312 if (defer)
291 313 fnvlist_add_boolean(args, "defer");
292 314
293 315 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
294 316 nvlist_free(args);
295 317
296 318 return (error);
297 319 }
298 320
299 321 int
300 322 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
301 323 uint64_t *usedp)
302 324 {
303 325 nvlist_t *args;
304 326 nvlist_t *result;
305 327 int err;
306 328 char fs[ZFS_MAX_DATASET_NAME_LEN];
307 329 char *atp;
308 330
309 331 /* determine the fs name */
310 332 (void) strlcpy(fs, firstsnap, sizeof (fs));
311 333 atp = strchr(fs, '@');
312 334 if (atp == NULL)
313 335 return (EINVAL);
314 336 *atp = '\0';
315 337
316 338 args = fnvlist_alloc();
317 339 fnvlist_add_string(args, "firstsnap", firstsnap);
318 340
319 341 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
320 342 nvlist_free(args);
321 343 if (err == 0)
322 344 *usedp = fnvlist_lookup_uint64(result, "used");
|
↓ open down ↓ |
107 lines elided |
↑ open up ↑ |
323 345 fnvlist_free(result);
324 346
325 347 return (err);
326 348 }
327 349
328 350 boolean_t
329 351 lzc_exists(const char *dataset)
330 352 {
331 353 /*
332 354 * The objset_stats ioctl is still legacy, so we need to construct our
333 - * own zfs_cmd_t rather than using zfsc_ioctl().
355 + * own zfs_cmd_t rather than using lzc_ioctl().
334 356 */
335 357 zfs_cmd_t zc = { 0 };
336 358
337 359 ASSERT3S(g_refcount, >, 0);
338 360 VERIFY3S(g_fd, !=, -1);
339 361
340 362 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
341 363 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
342 364 }
343 365
344 366 /*
345 367 * Create "user holds" on snapshots. If there is a hold on a snapshot,
346 368 * the snapshot can not be destroyed. (However, it can be marked for deletion
347 369 * by lzc_destroy_snaps(defer=B_TRUE).)
348 370 *
349 371 * The keys in the nvlist are snapshot names.
350 372 * The snapshots must all be in the same pool.
351 373 * The value is the name of the hold (string type).
352 374 *
353 375 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
354 376 * In this case, when the cleanup_fd is closed (including on process
355 377 * termination), the holds will be released. If the system is shut down
356 378 * uncleanly, the holds will be released when the pool is next opened
357 379 * or imported.
358 380 *
359 381 * Holds for snapshots which don't exist will be skipped and have an entry
360 382 * added to errlist, but will not cause an overall failure.
361 383 *
362 384 * The return value will be 0 if all holds, for snapshots that existed,
363 385 * were succesfully created.
364 386 *
365 387 * Otherwise the return value will be the errno of a (unspecified) hold that
366 388 * failed and no holds will be created.
367 389 *
368 390 * In all cases the errlist will have an entry for each hold that failed
369 391 * (name = snapshot), with its value being the error code (int32).
370 392 */
371 393 int
372 394 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
373 395 {
374 396 char pool[ZFS_MAX_DATASET_NAME_LEN];
375 397 nvlist_t *args;
376 398 nvpair_t *elem;
377 399 int error;
378 400
379 401 /* determine the pool name */
380 402 elem = nvlist_next_nvpair(holds, NULL);
381 403 if (elem == NULL)
382 404 return (0);
383 405 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
384 406 pool[strcspn(pool, "/@")] = '\0';
385 407
386 408 args = fnvlist_alloc();
387 409 fnvlist_add_nvlist(args, "holds", holds);
388 410 if (cleanup_fd != -1)
389 411 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
390 412
391 413 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
392 414 nvlist_free(args);
393 415 return (error);
394 416 }
395 417
396 418 /*
397 419 * Release "user holds" on snapshots. If the snapshot has been marked for
398 420 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
399 421 * any clones, and all the user holds are removed, then the snapshot will be
400 422 * destroyed.
401 423 *
402 424 * The keys in the nvlist are snapshot names.
403 425 * The snapshots must all be in the same pool.
404 426 * The value is a nvlist whose keys are the holds to remove.
405 427 *
406 428 * Holds which failed to release because they didn't exist will have an entry
407 429 * added to errlist, but will not cause an overall failure.
408 430 *
409 431 * The return value will be 0 if the nvl holds was empty or all holds that
410 432 * existed, were successfully removed.
411 433 *
412 434 * Otherwise the return value will be the errno of a (unspecified) hold that
413 435 * failed to release and no holds will be released.
414 436 *
415 437 * In all cases the errlist will have an entry for each hold that failed to
416 438 * to release.
417 439 */
418 440 int
419 441 lzc_release(nvlist_t *holds, nvlist_t **errlist)
420 442 {
421 443 char pool[ZFS_MAX_DATASET_NAME_LEN];
422 444 nvpair_t *elem;
423 445
424 446 /* determine the pool name */
425 447 elem = nvlist_next_nvpair(holds, NULL);
426 448 if (elem == NULL)
427 449 return (0);
428 450 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
429 451 pool[strcspn(pool, "/@")] = '\0';
430 452
431 453 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
432 454 }
433 455
434 456 /*
435 457 * Retrieve list of user holds on the specified snapshot.
436 458 *
437 459 * On success, *holdsp will be set to a nvlist which the caller must free.
438 460 * The keys are the names of the holds, and the value is the creation time
439 461 * of the hold (uint64) in seconds since the epoch.
440 462 */
441 463 int
442 464 lzc_get_holds(const char *snapname, nvlist_t **holdsp)
443 465 {
444 466 int error;
445 467 nvlist_t *innvl = fnvlist_alloc();
446 468 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp);
447 469 fnvlist_free(innvl);
448 470 return (error);
449 471 }
450 472
451 473 /*
452 474 * Generate a zfs send stream for the specified snapshot and write it to
453 475 * the specified file descriptor.
454 476 *
455 477 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap")
456 478 *
457 479 * If "from" is NULL, a full (non-incremental) stream will be sent.
458 480 * If "from" is non-NULL, it must be the full name of a snapshot or
459 481 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or
460 482 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or
461 483 * bookmark must represent an earlier point in the history of "snapname").
462 484 * It can be an earlier snapshot in the same filesystem or zvol as "snapname",
463 485 * or it can be the origin of "snapname"'s filesystem, or an earlier
464 486 * snapshot in the origin, etc.
465 487 *
466 488 * "fd" is the file descriptor to write the send stream to.
467 489 *
468 490 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted
469 491 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT
470 492 * records with drr_blksz > 128K.
471 493 *
472 494 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted
473 495 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA,
474 496 * which the receiving system must support (as indicated by support
475 497 * for the "embedded_data" feature).
476 498 */
477 499 int
478 500 lzc_send(const char *snapname, const char *from, int fd,
479 501 enum lzc_send_flags flags)
480 502 {
481 503 return (lzc_send_resume(snapname, from, fd, flags, 0, 0));
482 504 }
483 505
484 506 int
485 507 lzc_send_resume(const char *snapname, const char *from, int fd,
486 508 enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff)
487 509 {
488 510 nvlist_t *args;
489 511 int err;
490 512
491 513 args = fnvlist_alloc();
492 514 fnvlist_add_int32(args, "fd", fd);
493 515 if (from != NULL)
494 516 fnvlist_add_string(args, "fromsnap", from);
495 517 if (flags & LZC_SEND_FLAG_LARGE_BLOCK)
496 518 fnvlist_add_boolean(args, "largeblockok");
497 519 if (flags & LZC_SEND_FLAG_EMBED_DATA)
498 520 fnvlist_add_boolean(args, "embedok");
499 521 if (flags & LZC_SEND_FLAG_COMPRESS)
500 522 fnvlist_add_boolean(args, "compressok");
501 523 if (resumeobj != 0 || resumeoff != 0) {
502 524 fnvlist_add_uint64(args, "resume_object", resumeobj);
503 525 fnvlist_add_uint64(args, "resume_offset", resumeoff);
504 526 }
505 527 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
506 528 nvlist_free(args);
507 529 return (err);
508 530 }
509 531
510 532 /*
511 533 * "from" can be NULL, a snapshot, or a bookmark.
512 534 *
513 535 * If from is NULL, a full (non-incremental) stream will be estimated. This
514 536 * is calculated very efficiently.
515 537 *
516 538 * If from is a snapshot, lzc_send_space uses the deadlists attached to
517 539 * each snapshot to efficiently estimate the stream size.
518 540 *
519 541 * If from is a bookmark, the indirect blocks in the destination snapshot
520 542 * are traversed, looking for blocks with a birth time since the creation TXG of
521 543 * the snapshot this bookmark was created from. This will result in
522 544 * significantly more I/O and be less efficient than a send space estimation on
523 545 * an equivalent snapshot.
524 546 */
525 547 int
526 548 lzc_send_space(const char *snapname, const char *from,
527 549 enum lzc_send_flags flags, uint64_t *spacep)
528 550 {
529 551 nvlist_t *args;
530 552 nvlist_t *result;
531 553 int err;
532 554
533 555 args = fnvlist_alloc();
534 556 if (from != NULL)
535 557 fnvlist_add_string(args, "from", from);
536 558 if (flags & LZC_SEND_FLAG_LARGE_BLOCK)
537 559 fnvlist_add_boolean(args, "largeblockok");
538 560 if (flags & LZC_SEND_FLAG_EMBED_DATA)
539 561 fnvlist_add_boolean(args, "embedok");
540 562 if (flags & LZC_SEND_FLAG_COMPRESS)
541 563 fnvlist_add_boolean(args, "compressok");
542 564 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
543 565 nvlist_free(args);
544 566 if (err == 0)
545 567 *spacep = fnvlist_lookup_uint64(result, "space");
546 568 nvlist_free(result);
547 569 return (err);
548 570 }
549 571
550 572 static int
551 573 recv_read(int fd, void *buf, int ilen)
552 574 {
553 575 char *cp = buf;
554 576 int rv;
555 577 int len = ilen;
556 578
557 579 do {
558 580 rv = read(fd, cp, len);
559 581 cp += rv;
560 582 len -= rv;
561 583 } while (rv > 0);
562 584
563 585 if (rv < 0 || len != 0)
564 586 return (EIO);
565 587
566 588 return (0);
567 589 }
568 590
569 591 static int
570 592 recv_impl(const char *snapname, nvlist_t *props, const char *origin,
571 593 boolean_t force, boolean_t resumable, int fd,
572 594 const dmu_replay_record_t *begin_record)
573 595 {
574 596 /*
575 597 * The receive ioctl is still legacy, so we need to construct our own
576 598 * zfs_cmd_t rather than using zfsc_ioctl().
577 599 */
578 600 zfs_cmd_t zc = { 0 };
579 601 char *atp;
580 602 char *packed = NULL;
581 603 size_t size;
582 604 int error;
583 605
584 606 ASSERT3S(g_refcount, >, 0);
585 607 VERIFY3S(g_fd, !=, -1);
586 608
587 609 /* zc_name is name of containing filesystem */
588 610 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
589 611 atp = strchr(zc.zc_name, '@');
590 612 if (atp == NULL)
591 613 return (EINVAL);
592 614 *atp = '\0';
593 615
594 616 /* if the fs does not exist, try its parent. */
595 617 if (!lzc_exists(zc.zc_name)) {
596 618 char *slashp = strrchr(zc.zc_name, '/');
597 619 if (slashp == NULL)
598 620 return (ENOENT);
599 621 *slashp = '\0';
600 622
601 623 }
602 624
603 625 /* zc_value is full name of the snapshot to create */
604 626 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
605 627
606 628 if (props != NULL) {
607 629 /* zc_nvlist_src is props to set */
608 630 packed = fnvlist_pack(props, &size);
609 631 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
610 632 zc.zc_nvlist_src_size = size;
611 633 }
612 634
613 635 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
614 636 if (origin != NULL)
615 637 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
616 638
617 639 /* zc_begin_record is non-byteswapped BEGIN record */
618 640 if (begin_record == NULL) {
619 641 error = recv_read(fd, &zc.zc_begin_record,
620 642 sizeof (zc.zc_begin_record));
621 643 if (error != 0)
622 644 goto out;
623 645 } else {
624 646 zc.zc_begin_record = *begin_record;
625 647 }
626 648
627 649 /* zc_cookie is fd to read from */
628 650 zc.zc_cookie = fd;
629 651
630 652 /* zc guid is force flag */
631 653 zc.zc_guid = force;
632 654
633 655 zc.zc_resumable = resumable;
634 656
635 657 /* zc_cleanup_fd is unused */
636 658 zc.zc_cleanup_fd = -1;
637 659
638 660 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
639 661 if (error != 0)
640 662 error = errno;
641 663
642 664 out:
643 665 if (packed != NULL)
644 666 fnvlist_pack_free(packed, size);
645 667 free((void*)(uintptr_t)zc.zc_nvlist_dst);
646 668 return (error);
647 669 }
648 670
649 671 /*
650 672 * The simplest receive case: receive from the specified fd, creating the
651 673 * specified snapshot. Apply the specified properties as "received" properties
652 674 * (which can be overridden by locally-set properties). If the stream is a
653 675 * clone, its origin snapshot must be specified by 'origin'. The 'force'
654 676 * flag will cause the target filesystem to be rolled back or destroyed if
655 677 * necessary to receive.
656 678 *
657 679 * Return 0 on success or an errno on failure.
658 680 *
659 681 * Note: this interface does not work on dedup'd streams
660 682 * (those with DMU_BACKUP_FEATURE_DEDUP).
661 683 */
662 684 int
663 685 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
664 686 boolean_t force, int fd)
665 687 {
666 688 return (recv_impl(snapname, props, origin, force, B_FALSE, fd, NULL));
667 689 }
668 690
669 691 /*
670 692 * Like lzc_receive, but if the receive fails due to premature stream
671 693 * termination, the intermediate state will be preserved on disk. In this
672 694 * case, ECKSUM will be returned. The receive may subsequently be resumed
673 695 * with a resuming send stream generated by lzc_send_resume().
674 696 */
675 697 int
676 698 lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin,
677 699 boolean_t force, int fd)
678 700 {
679 701 return (recv_impl(snapname, props, origin, force, B_TRUE, fd, NULL));
680 702 }
681 703
682 704 /*
683 705 * Like lzc_receive, but allows the caller to read the begin record and then to
684 706 * pass it in. That could be useful if the caller wants to derive, for example,
685 707 * the snapname or the origin parameters based on the information contained in
686 708 * the begin record.
687 709 * The begin record must be in its original form as read from the stream,
688 710 * in other words, it should not be byteswapped.
689 711 *
690 712 * The 'resumable' parameter allows to obtain the same behavior as with
691 713 * lzc_receive_resumable.
692 714 */
693 715 int
694 716 lzc_receive_with_header(const char *snapname, nvlist_t *props,
695 717 const char *origin, boolean_t force, boolean_t resumable, int fd,
696 718 const dmu_replay_record_t *begin_record)
697 719 {
698 720 if (begin_record == NULL)
699 721 return (EINVAL);
700 722 return (recv_impl(snapname, props, origin, force, resumable, fd,
701 723 begin_record));
702 724 }
703 725
704 726 /*
705 727 * Roll back this filesystem or volume to its most recent snapshot.
706 728 * If snapnamebuf is not NULL, it will be filled in with the name
707 729 * of the most recent snapshot.
708 730 *
709 731 * Return 0 on success or an errno on failure.
710 732 */
711 733 int
712 734 lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen)
713 735 {
714 736 nvlist_t *args;
715 737 nvlist_t *result;
716 738 int err;
717 739
718 740 args = fnvlist_alloc();
719 741 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result);
720 742 nvlist_free(args);
721 743 if (err == 0 && snapnamebuf != NULL) {
722 744 const char *snapname = fnvlist_lookup_string(result, "target");
723 745 (void) strlcpy(snapnamebuf, snapname, snapnamelen);
724 746 }
725 747 nvlist_free(result);
726 748
727 749 return (err);
728 750 }
729 751
730 752 /*
731 753 * Creates bookmarks.
732 754 *
733 755 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to
734 756 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and
735 757 * snapshots must be in the same pool.
736 758 *
737 759 * The returned results nvlist will have an entry for each bookmark that failed.
738 760 * The value will be the (int32) error code.
739 761 *
740 762 * The return value will be 0 if all bookmarks were created, otherwise it will
741 763 * be the errno of a (undetermined) bookmarks that failed.
742 764 */
743 765 int
744 766 lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist)
745 767 {
746 768 nvpair_t *elem;
747 769 int error;
748 770 char pool[ZFS_MAX_DATASET_NAME_LEN];
749 771
750 772 /* determine the pool name */
751 773 elem = nvlist_next_nvpair(bookmarks, NULL);
752 774 if (elem == NULL)
753 775 return (0);
754 776 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
755 777 pool[strcspn(pool, "/#")] = '\0';
756 778
757 779 error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist);
758 780
759 781 return (error);
760 782 }
761 783
762 784 /*
763 785 * Retrieve bookmarks.
764 786 *
765 787 * Retrieve the list of bookmarks for the given file system. The props
766 788 * parameter is an nvlist of property names (with no values) that will be
767 789 * returned for each bookmark.
768 790 *
769 791 * The following are valid properties on bookmarks, all of which are numbers
770 792 * (represented as uint64 in the nvlist)
771 793 *
772 794 * "guid" - globally unique identifier of the snapshot it refers to
773 795 * "createtxg" - txg when the snapshot it refers to was created
774 796 * "creation" - timestamp when the snapshot it refers to was created
775 797 *
776 798 * The format of the returned nvlist as follows:
777 799 * <short name of bookmark> -> {
778 800 * <name of property> -> {
779 801 * "value" -> uint64
780 802 * }
781 803 * }
782 804 */
783 805 int
784 806 lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks)
785 807 {
786 808 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks));
787 809 }
788 810
789 811 /*
790 812 * Destroys bookmarks.
791 813 *
792 814 * The keys in the bmarks nvlist are the bookmarks to be destroyed.
793 815 * They must all be in the same pool. Bookmarks are specified as
794 816 * <fs>#<bmark>.
795 817 *
796 818 * Bookmarks that do not exist will be silently ignored.
797 819 *
798 820 * The return value will be 0 if all bookmarks that existed were destroyed.
799 821 *
800 822 * Otherwise the return value will be the errno of a (undetermined) bookmark
801 823 * that failed, no bookmarks will be destroyed, and the errlist will have an
802 824 * entry for each bookmarks that failed. The value in the errlist will be
803 825 * the (int32) error code.
804 826 */
805 827 int
806 828 lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist)
807 829 {
808 830 nvpair_t *elem;
809 831 int error;
810 832 char pool[ZFS_MAX_DATASET_NAME_LEN];
811 833
812 834 /* determine the pool name */
813 835 elem = nvlist_next_nvpair(bmarks, NULL);
814 836 if (elem == NULL)
815 837 return (0);
816 838 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
817 839 pool[strcspn(pool, "/#")] = '\0';
818 840
819 841 error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist);
820 842
821 843 return (error);
822 844 }
|
↓ open down ↓ |
479 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX