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, 2017 by Delphix. All rights reserved. 24 * Copyright (c) 2013 Steven Hartland. All rights reserved. 25 * Copyright (c) 2014 Integros [integros.com] 26 * Copyright 2017 RackTop Systems. 27 */ 28 29 /* 30 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs. 31 * It has the following characteristics: 32 * 33 * - Thread Safe. libzfs_core is accessible concurrently from multiple 34 * threads. This is accomplished primarily by avoiding global data 35 * (e.g. caching). Since it's thread-safe, there is no reason for a 36 * process to have multiple libzfs "instances". Therefore, we store 37 * our few pieces of data (e.g. the file descriptor) in global 38 * variables. The fd is reference-counted so that the libzfs_core 39 * library can be "initialized" multiple times (e.g. by different 40 * consumers within the same process). 41 * 42 * - Committed Interface. The libzfs_core interface will be committed, 43 * therefore consumers can compile against it and be confident that 44 * their code will continue to work on future releases of this code. 45 * Currently, the interface is Evolving (not Committed), but we intend 46 * to commit to it once it is more complete and we determine that it 47 * meets the needs of all consumers. 48 * 49 * - Programatic Error Handling. libzfs_core communicates errors with 50 * defined error numbers, and doesn't print anything to stdout/stderr. 51 * 52 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments 53 * to/from the kernel ioctls. There is generally a 1:1 correspondence 54 * between libzfs_core functions and ioctls to /dev/zfs. 55 * 56 * - Clear Atomicity. Because libzfs_core functions are generally 1:1 57 * with kernel ioctls, and kernel ioctls are general atomic, each 58 * libzfs_core function is atomic. For example, creating multiple 59 * snapshots with a single call to lzc_snapshot() is atomic -- it 60 * can't fail with only some of the requested snapshots created, even 61 * in the event of power loss or system crash. 62 * 63 * - Continued libzfs Support. Some higher-level operations (e.g. 64 * support for "zfs send -R") are too complicated to fit the scope of 65 * libzfs_core. This functionality will continue to live in libzfs. 66 * Where appropriate, libzfs will use the underlying atomic operations 67 * of libzfs_core. For example, libzfs may implement "zfs send -R | 68 * zfs receive" by using individual "send one snapshot", rename, 69 * destroy, and "receive one snapshot" operations in libzfs_core. 70 * /sbin/zfs and /zbin/zpool will link with both libzfs and 71 * libzfs_core. Other consumers should aim to use only libzfs_core, 72 * since that will be the supported, stable interface going forwards. 73 */ 74 75 #include <libzfs_core.h> 76 #include <ctype.h> 77 #include <unistd.h> 78 #include <stdlib.h> 79 #include <string.h> 80 #include <errno.h> 81 #include <fcntl.h> 82 #include <pthread.h> 83 #include <sys/nvpair.h> 84 #include <sys/param.h> 85 #include <sys/types.h> 86 #include <sys/stat.h> 87 #include <sys/zfs_ioctl.h> 88 89 static int g_fd = -1; 90 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER; 91 static int g_refcount; 92 93 int 94 libzfs_core_init(void) 95 { 96 (void) pthread_mutex_lock(&g_lock); 97 if (g_refcount == 0) { 98 g_fd = open("/dev/zfs", O_RDWR); 99 if (g_fd < 0) { 100 (void) pthread_mutex_unlock(&g_lock); 101 return (errno); 102 } 103 } 104 g_refcount++; 105 (void) pthread_mutex_unlock(&g_lock); 106 return (0); 107 } 108 109 void 110 libzfs_core_fini(void) 111 { 112 (void) pthread_mutex_lock(&g_lock); 113 ASSERT3S(g_refcount, >, 0); 114 115 if (g_refcount > 0) 116 g_refcount--; 117 118 if (g_refcount == 0 && g_fd != -1) { 119 (void) close(g_fd); 120 g_fd = -1; 121 } 122 (void) pthread_mutex_unlock(&g_lock); 123 } 124 125 static int 126 lzc_ioctl(zfs_ioc_t ioc, const char *name, 127 nvlist_t *source, nvlist_t **resultp) 128 { 129 zfs_cmd_t zc = { 0 }; 130 int error = 0; 131 char *packed; 132 size_t size; 133 134 ASSERT3S(g_refcount, >, 0); 135 VERIFY3S(g_fd, !=, -1); 136 137 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 138 139 packed = fnvlist_pack(source, &size); 140 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 141 zc.zc_nvlist_src_size = size; 142 143 if (resultp != NULL) { 144 *resultp = NULL; 145 if (ioc == ZFS_IOC_CHANNEL_PROGRAM) { 146 zc.zc_nvlist_dst_size = fnvlist_lookup_uint64(source, 147 ZCP_ARG_MEMLIMIT); 148 } else { 149 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024); 150 } 151 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 152 malloc(zc.zc_nvlist_dst_size); 153 if (zc.zc_nvlist_dst == NULL) { 154 error = ENOMEM; 155 goto out; 156 } 157 } 158 159 while (ioctl(g_fd, ioc, &zc) != 0) { 160 /* 161 * If ioctl exited with ENOMEM, we retry the ioctl after 162 * increasing the size of the destination nvlist. 163 * 164 * Channel programs that exit with ENOMEM ran over the 165 * lua memory sandbox; they should not be retried. 166 */ 167 if (errno == ENOMEM && resultp != NULL && 168 ioc != ZFS_IOC_CHANNEL_PROGRAM) { 169 free((void *)(uintptr_t)zc.zc_nvlist_dst); 170 zc.zc_nvlist_dst_size *= 2; 171 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 172 malloc(zc.zc_nvlist_dst_size); 173 if (zc.zc_nvlist_dst == NULL) { 174 error = ENOMEM; 175 goto out; 176 } 177 } else { 178 error = errno; 179 break; 180 } 181 } 182 if (zc.zc_nvlist_dst_filled) { 183 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 184 zc.zc_nvlist_dst_size); 185 } 186 187 out: 188 fnvlist_pack_free(packed, size); 189 free((void *)(uintptr_t)zc.zc_nvlist_dst); 190 return (error); 191 } 192 193 int 194 lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props) 195 { 196 int error; 197 nvlist_t *args = fnvlist_alloc(); 198 fnvlist_add_int32(args, "type", (dmu_objset_type_t)type); 199 if (props != NULL) 200 fnvlist_add_nvlist(args, "props", props); 201 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL); 202 nvlist_free(args); 203 return (error); 204 } 205 206 int 207 lzc_clone(const char *fsname, const char *origin, 208 nvlist_t *props) 209 { 210 int error; 211 nvlist_t *args = fnvlist_alloc(); 212 fnvlist_add_string(args, "origin", origin); 213 if (props != NULL) 214 fnvlist_add_nvlist(args, "props", props); 215 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL); 216 nvlist_free(args); 217 return (error); 218 } 219 220 int 221 lzc_promote(const char *fsname, char *snapnamebuf, int snapnamelen) 222 { 223 /* 224 * The promote ioctl is still legacy, so we need to construct our 225 * own zfs_cmd_t rather than using lzc_ioctl(). 226 */ 227 zfs_cmd_t zc = { 0 }; 228 229 ASSERT3S(g_refcount, >, 0); 230 VERIFY3S(g_fd, !=, -1); 231 232 (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name)); 233 if (ioctl(g_fd, ZFS_IOC_PROMOTE, &zc) != 0) { 234 int error = errno; 235 if (error == EEXIST && snapnamebuf != NULL) 236 (void) strlcpy(snapnamebuf, zc.zc_string, snapnamelen); 237 return (error); 238 } 239 return (0); 240 } 241 242 int 243 lzc_remap(const char *fsname) 244 { 245 int error; 246 nvlist_t *args = fnvlist_alloc(); 247 error = lzc_ioctl(ZFS_IOC_REMAP, fsname, args, NULL); 248 nvlist_free(args); 249 return (error); 250 } 251 252 int 253 lzc_rename(const char *source, const char *target) 254 { 255 zfs_cmd_t zc = { 0 }; 256 int error; 257 258 ASSERT3S(g_refcount, >, 0); 259 VERIFY3S(g_fd, !=, -1); 260 261 (void) strlcpy(zc.zc_name, source, sizeof (zc.zc_name)); 262 (void) strlcpy(zc.zc_value, target, sizeof (zc.zc_value)); 263 error = ioctl(g_fd, ZFS_IOC_RENAME, &zc); 264 if (error != 0) 265 error = errno; 266 return (error); 267 } 268 269 int 270 lzc_destroy(const char *fsname) 271 { 272 int error; 273 274 nvlist_t *args = fnvlist_alloc(); 275 error = lzc_ioctl(ZFS_IOC_DESTROY, fsname, args, NULL); 276 nvlist_free(args); 277 return (error); 278 } 279 280 /* 281 * Creates snapshots. 282 * 283 * The keys in the snaps nvlist are the snapshots to be created. 284 * They must all be in the same pool. 285 * 286 * The props nvlist is properties to set. Currently only user properties 287 * are supported. { user:prop_name -> string value } 288 * 289 * The returned results nvlist will have an entry for each snapshot that failed. 290 * The value will be the (int32) error code. 291 * 292 * The return value will be 0 if all snapshots were created, otherwise it will 293 * be the errno of a (unspecified) snapshot that failed. 294 */ 295 int 296 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) 297 { 298 nvpair_t *elem; 299 nvlist_t *args; 300 int error; 301 char pool[ZFS_MAX_DATASET_NAME_LEN]; 302 303 *errlist = NULL; 304 305 /* determine the pool name */ 306 elem = nvlist_next_nvpair(snaps, NULL); 307 if (elem == NULL) 308 return (0); 309 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 310 pool[strcspn(pool, "/@")] = '\0'; 311 312 args = fnvlist_alloc(); 313 fnvlist_add_nvlist(args, "snaps", snaps); 314 if (props != NULL) 315 fnvlist_add_nvlist(args, "props", props); 316 317 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist); 318 nvlist_free(args); 319 320 return (error); 321 } 322 323 /* 324 * Destroys snapshots. 325 * 326 * The keys in the snaps nvlist are the snapshots to be destroyed. 327 * They must all be in the same pool. 328 * 329 * Snapshots that do not exist will be silently ignored. 330 * 331 * If 'defer' is not set, and a snapshot has user holds or clones, the 332 * destroy operation will fail and none of the snapshots will be 333 * destroyed. 334 * 335 * If 'defer' is set, and a snapshot has user holds or clones, it will be 336 * marked for deferred destruction, and will be destroyed when the last hold 337 * or clone is removed/destroyed. 338 * 339 * The return value will be 0 if all snapshots were destroyed (or marked for 340 * later destruction if 'defer' is set) or didn't exist to begin with. 341 * 342 * Otherwise the return value will be the errno of a (unspecified) snapshot 343 * that failed, no snapshots will be destroyed, and the errlist will have an 344 * entry for each snapshot that failed. The value in the errlist will be 345 * the (int32) error code. 346 */ 347 int 348 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist) 349 { 350 nvpair_t *elem; 351 nvlist_t *args; 352 int error; 353 char pool[ZFS_MAX_DATASET_NAME_LEN]; 354 355 /* determine the pool name */ 356 elem = nvlist_next_nvpair(snaps, NULL); 357 if (elem == NULL) 358 return (0); 359 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 360 pool[strcspn(pool, "/@")] = '\0'; 361 362 args = fnvlist_alloc(); 363 fnvlist_add_nvlist(args, "snaps", snaps); 364 if (defer) 365 fnvlist_add_boolean(args, "defer"); 366 367 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist); 368 nvlist_free(args); 369 370 return (error); 371 } 372 373 int 374 lzc_snaprange_space(const char *firstsnap, const char *lastsnap, 375 uint64_t *usedp) 376 { 377 nvlist_t *args; 378 nvlist_t *result; 379 int err; 380 char fs[ZFS_MAX_DATASET_NAME_LEN]; 381 char *atp; 382 383 /* determine the fs name */ 384 (void) strlcpy(fs, firstsnap, sizeof (fs)); 385 atp = strchr(fs, '@'); 386 if (atp == NULL) 387 return (EINVAL); 388 *atp = '\0'; 389 390 args = fnvlist_alloc(); 391 fnvlist_add_string(args, "firstsnap", firstsnap); 392 393 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result); 394 nvlist_free(args); 395 if (err == 0) 396 *usedp = fnvlist_lookup_uint64(result, "used"); 397 fnvlist_free(result); 398 399 return (err); 400 } 401 402 boolean_t 403 lzc_exists(const char *dataset) 404 { 405 /* 406 * The objset_stats ioctl is still legacy, so we need to construct our 407 * own zfs_cmd_t rather than using lzc_ioctl(). 408 */ 409 zfs_cmd_t zc = { 0 }; 410 411 ASSERT3S(g_refcount, >, 0); 412 VERIFY3S(g_fd, !=, -1); 413 414 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); 415 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0); 416 } 417 418 /* 419 * Create "user holds" on snapshots. If there is a hold on a snapshot, 420 * the snapshot can not be destroyed. (However, it can be marked for deletion 421 * by lzc_destroy_snaps(defer=B_TRUE).) 422 * 423 * The keys in the nvlist are snapshot names. 424 * The snapshots must all be in the same pool. 425 * The value is the name of the hold (string type). 426 * 427 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL). 428 * In this case, when the cleanup_fd is closed (including on process 429 * termination), the holds will be released. If the system is shut down 430 * uncleanly, the holds will be released when the pool is next opened 431 * or imported. 432 * 433 * Holds for snapshots which don't exist will be skipped and have an entry 434 * added to errlist, but will not cause an overall failure. 435 * 436 * The return value will be 0 if all holds, for snapshots that existed, 437 * were succesfully created. 438 * 439 * Otherwise the return value will be the errno of a (unspecified) hold that 440 * failed and no holds will be created. 441 * 442 * In all cases the errlist will have an entry for each hold that failed 443 * (name = snapshot), with its value being the error code (int32). 444 */ 445 int 446 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist) 447 { 448 char pool[ZFS_MAX_DATASET_NAME_LEN]; 449 nvlist_t *args; 450 nvpair_t *elem; 451 int error; 452 453 /* determine the pool name */ 454 elem = nvlist_next_nvpair(holds, NULL); 455 if (elem == NULL) 456 return (0); 457 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 458 pool[strcspn(pool, "/@")] = '\0'; 459 460 args = fnvlist_alloc(); 461 fnvlist_add_nvlist(args, "holds", holds); 462 if (cleanup_fd != -1) 463 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd); 464 465 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist); 466 nvlist_free(args); 467 return (error); 468 } 469 470 /* 471 * Release "user holds" on snapshots. If the snapshot has been marked for 472 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have 473 * any clones, and all the user holds are removed, then the snapshot will be 474 * destroyed. 475 * 476 * The keys in the nvlist are snapshot names. 477 * The snapshots must all be in the same pool. 478 * The value is a nvlist whose keys are the holds to remove. 479 * 480 * Holds which failed to release because they didn't exist will have an entry 481 * added to errlist, but will not cause an overall failure. 482 * 483 * The return value will be 0 if the nvl holds was empty or all holds that 484 * existed, were successfully removed. 485 * 486 * Otherwise the return value will be the errno of a (unspecified) hold that 487 * failed to release and no holds will be released. 488 * 489 * In all cases the errlist will have an entry for each hold that failed to 490 * to release. 491 */ 492 int 493 lzc_release(nvlist_t *holds, nvlist_t **errlist) 494 { 495 char pool[ZFS_MAX_DATASET_NAME_LEN]; 496 nvpair_t *elem; 497 498 /* determine the pool name */ 499 elem = nvlist_next_nvpair(holds, NULL); 500 if (elem == NULL) 501 return (0); 502 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 503 pool[strcspn(pool, "/@")] = '\0'; 504 505 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist)); 506 } 507 508 /* 509 * Retrieve list of user holds on the specified snapshot. 510 * 511 * On success, *holdsp will be set to a nvlist which the caller must free. 512 * The keys are the names of the holds, and the value is the creation time 513 * of the hold (uint64) in seconds since the epoch. 514 */ 515 int 516 lzc_get_holds(const char *snapname, nvlist_t **holdsp) 517 { 518 int error; 519 nvlist_t *innvl = fnvlist_alloc(); 520 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp); 521 fnvlist_free(innvl); 522 return (error); 523 } 524 525 /* 526 * Generate a zfs send stream for the specified snapshot and write it to 527 * the specified file descriptor. 528 * 529 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap") 530 * 531 * If "from" is NULL, a full (non-incremental) stream will be sent. 532 * If "from" is non-NULL, it must be the full name of a snapshot or 533 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or 534 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or 535 * bookmark must represent an earlier point in the history of "snapname"). 536 * It can be an earlier snapshot in the same filesystem or zvol as "snapname", 537 * or it can be the origin of "snapname"'s filesystem, or an earlier 538 * snapshot in the origin, etc. 539 * 540 * "fd" is the file descriptor to write the send stream to. 541 * 542 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted 543 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT 544 * records with drr_blksz > 128K. 545 * 546 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted 547 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA, 548 * which the receiving system must support (as indicated by support 549 * for the "embedded_data" feature). 550 */ 551 int 552 lzc_send(const char *snapname, const char *from, int fd, 553 enum lzc_send_flags flags) 554 { 555 return (lzc_send_resume(snapname, from, fd, flags, 0, 0)); 556 } 557 558 int 559 lzc_send_resume(const char *snapname, const char *from, int fd, 560 enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff) 561 { 562 nvlist_t *args; 563 int err; 564 565 args = fnvlist_alloc(); 566 fnvlist_add_int32(args, "fd", fd); 567 if (from != NULL) 568 fnvlist_add_string(args, "fromsnap", from); 569 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 570 fnvlist_add_boolean(args, "largeblockok"); 571 if (flags & LZC_SEND_FLAG_EMBED_DATA) 572 fnvlist_add_boolean(args, "embedok"); 573 if (flags & LZC_SEND_FLAG_COMPRESS) 574 fnvlist_add_boolean(args, "compressok"); 575 if (resumeobj != 0 || resumeoff != 0) { 576 fnvlist_add_uint64(args, "resume_object", resumeobj); 577 fnvlist_add_uint64(args, "resume_offset", resumeoff); 578 } 579 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL); 580 nvlist_free(args); 581 return (err); 582 } 583 584 /* 585 * "from" can be NULL, a snapshot, or a bookmark. 586 * 587 * If from is NULL, a full (non-incremental) stream will be estimated. This 588 * is calculated very efficiently. 589 * 590 * If from is a snapshot, lzc_send_space uses the deadlists attached to 591 * each snapshot to efficiently estimate the stream size. 592 * 593 * If from is a bookmark, the indirect blocks in the destination snapshot 594 * are traversed, looking for blocks with a birth time since the creation TXG of 595 * the snapshot this bookmark was created from. This will result in 596 * significantly more I/O and be less efficient than a send space estimation on 597 * an equivalent snapshot. 598 */ 599 int 600 lzc_send_space(const char *snapname, const char *from, 601 enum lzc_send_flags flags, uint64_t *spacep) 602 { 603 nvlist_t *args; 604 nvlist_t *result; 605 int err; 606 607 args = fnvlist_alloc(); 608 if (from != NULL) 609 fnvlist_add_string(args, "from", from); 610 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 611 fnvlist_add_boolean(args, "largeblockok"); 612 if (flags & LZC_SEND_FLAG_EMBED_DATA) 613 fnvlist_add_boolean(args, "embedok"); 614 if (flags & LZC_SEND_FLAG_COMPRESS) 615 fnvlist_add_boolean(args, "compressok"); 616 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result); 617 nvlist_free(args); 618 if (err == 0) 619 *spacep = fnvlist_lookup_uint64(result, "space"); 620 nvlist_free(result); 621 return (err); 622 } 623 624 static int 625 recv_read(int fd, void *buf, int ilen) 626 { 627 char *cp = buf; 628 int rv; 629 int len = ilen; 630 631 do { 632 rv = read(fd, cp, len); 633 cp += rv; 634 len -= rv; 635 } while (rv > 0); 636 637 if (rv < 0 || len != 0) 638 return (EIO); 639 640 return (0); 641 } 642 643 static int 644 recv_impl(const char *snapname, nvlist_t *props, const char *origin, 645 boolean_t force, boolean_t resumable, int fd, 646 const dmu_replay_record_t *begin_record) 647 { 648 /* 649 * The receive ioctl is still legacy, so we need to construct our own 650 * zfs_cmd_t rather than using zfsc_ioctl(). 651 */ 652 zfs_cmd_t zc = { 0 }; 653 char *atp; 654 char *packed = NULL; 655 size_t size; 656 int error; 657 658 ASSERT3S(g_refcount, >, 0); 659 VERIFY3S(g_fd, !=, -1); 660 661 /* zc_name is name of containing filesystem */ 662 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name)); 663 atp = strchr(zc.zc_name, '@'); 664 if (atp == NULL) 665 return (EINVAL); 666 *atp = '\0'; 667 668 /* if the fs does not exist, try its parent. */ 669 if (!lzc_exists(zc.zc_name)) { 670 char *slashp = strrchr(zc.zc_name, '/'); 671 if (slashp == NULL) 672 return (ENOENT); 673 *slashp = '\0'; 674 675 } 676 677 /* zc_value is full name of the snapshot to create */ 678 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); 679 680 if (props != NULL) { 681 /* zc_nvlist_src is props to set */ 682 packed = fnvlist_pack(props, &size); 683 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 684 zc.zc_nvlist_src_size = size; 685 } 686 687 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */ 688 if (origin != NULL) 689 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string)); 690 691 /* zc_begin_record is non-byteswapped BEGIN record */ 692 if (begin_record == NULL) { 693 error = recv_read(fd, &zc.zc_begin_record, 694 sizeof (zc.zc_begin_record)); 695 if (error != 0) 696 goto out; 697 } else { 698 zc.zc_begin_record = *begin_record; 699 } 700 701 /* zc_cookie is fd to read from */ 702 zc.zc_cookie = fd; 703 704 /* zc guid is force flag */ 705 zc.zc_guid = force; 706 707 zc.zc_resumable = resumable; 708 709 /* zc_cleanup_fd is unused */ 710 zc.zc_cleanup_fd = -1; 711 712 error = ioctl(g_fd, ZFS_IOC_RECV, &zc); 713 if (error != 0) 714 error = errno; 715 716 out: 717 if (packed != NULL) 718 fnvlist_pack_free(packed, size); 719 free((void*)(uintptr_t)zc.zc_nvlist_dst); 720 return (error); 721 } 722 723 /* 724 * The simplest receive case: receive from the specified fd, creating the 725 * specified snapshot. Apply the specified properties as "received" properties 726 * (which can be overridden by locally-set properties). If the stream is a 727 * clone, its origin snapshot must be specified by 'origin'. The 'force' 728 * flag will cause the target filesystem to be rolled back or destroyed if 729 * necessary to receive. 730 * 731 * Return 0 on success or an errno on failure. 732 * 733 * Note: this interface does not work on dedup'd streams 734 * (those with DMU_BACKUP_FEATURE_DEDUP). 735 */ 736 int 737 lzc_receive(const char *snapname, nvlist_t *props, const char *origin, 738 boolean_t force, int fd) 739 { 740 return (recv_impl(snapname, props, origin, force, B_FALSE, fd, NULL)); 741 } 742 743 /* 744 * Like lzc_receive, but if the receive fails due to premature stream 745 * termination, the intermediate state will be preserved on disk. In this 746 * case, ECKSUM will be returned. The receive may subsequently be resumed 747 * with a resuming send stream generated by lzc_send_resume(). 748 */ 749 int 750 lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin, 751 boolean_t force, int fd) 752 { 753 return (recv_impl(snapname, props, origin, force, B_TRUE, fd, NULL)); 754 } 755 756 /* 757 * Like lzc_receive, but allows the caller to read the begin record and then to 758 * pass it in. That could be useful if the caller wants to derive, for example, 759 * the snapname or the origin parameters based on the information contained in 760 * the begin record. 761 * The begin record must be in its original form as read from the stream, 762 * in other words, it should not be byteswapped. 763 * 764 * The 'resumable' parameter allows to obtain the same behavior as with 765 * lzc_receive_resumable. 766 */ 767 int 768 lzc_receive_with_header(const char *snapname, nvlist_t *props, 769 const char *origin, boolean_t force, boolean_t resumable, int fd, 770 const dmu_replay_record_t *begin_record) 771 { 772 if (begin_record == NULL) 773 return (EINVAL); 774 return (recv_impl(snapname, props, origin, force, resumable, fd, 775 begin_record)); 776 } 777 778 /* 779 * Roll back this filesystem or volume to its most recent snapshot. 780 * If snapnamebuf is not NULL, it will be filled in with the name 781 * of the most recent snapshot. 782 * Note that the latest snapshot may change if a new one is concurrently 783 * created or the current one is destroyed. lzc_rollback_to can be used 784 * to roll back to a specific latest snapshot. 785 * 786 * Return 0 on success or an errno on failure. 787 */ 788 int 789 lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen) 790 { 791 nvlist_t *args; 792 nvlist_t *result; 793 int err; 794 795 args = fnvlist_alloc(); 796 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 797 nvlist_free(args); 798 if (err == 0 && snapnamebuf != NULL) { 799 const char *snapname = fnvlist_lookup_string(result, "target"); 800 (void) strlcpy(snapnamebuf, snapname, snapnamelen); 801 } 802 nvlist_free(result); 803 804 return (err); 805 } 806 807 /* 808 * Roll back this filesystem or volume to the specified snapshot, 809 * if possible. 810 * 811 * Return 0 on success or an errno on failure. 812 */ 813 int 814 lzc_rollback_to(const char *fsname, const char *snapname) 815 { 816 nvlist_t *args; 817 nvlist_t *result; 818 int err; 819 820 args = fnvlist_alloc(); 821 fnvlist_add_string(args, "target", snapname); 822 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 823 nvlist_free(args); 824 nvlist_free(result); 825 return (err); 826 } 827 828 /* 829 * Creates bookmarks. 830 * 831 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to 832 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and 833 * snapshots must be in the same pool. 834 * 835 * The returned results nvlist will have an entry for each bookmark that failed. 836 * The value will be the (int32) error code. 837 * 838 * The return value will be 0 if all bookmarks were created, otherwise it will 839 * be the errno of a (undetermined) bookmarks that failed. 840 */ 841 int 842 lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist) 843 { 844 nvpair_t *elem; 845 int error; 846 char pool[ZFS_MAX_DATASET_NAME_LEN]; 847 848 /* determine the pool name */ 849 elem = nvlist_next_nvpair(bookmarks, NULL); 850 if (elem == NULL) 851 return (0); 852 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 853 pool[strcspn(pool, "/#")] = '\0'; 854 855 error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist); 856 857 return (error); 858 } 859 860 /* 861 * Retrieve bookmarks. 862 * 863 * Retrieve the list of bookmarks for the given file system. The props 864 * parameter is an nvlist of property names (with no values) that will be 865 * returned for each bookmark. 866 * 867 * The following are valid properties on bookmarks, all of which are numbers 868 * (represented as uint64 in the nvlist) 869 * 870 * "guid" - globally unique identifier of the snapshot it refers to 871 * "createtxg" - txg when the snapshot it refers to was created 872 * "creation" - timestamp when the snapshot it refers to was created 873 * 874 * The format of the returned nvlist as follows: 875 * <short name of bookmark> -> { 876 * <name of property> -> { 877 * "value" -> uint64 878 * } 879 * } 880 */ 881 int 882 lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks) 883 { 884 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks)); 885 } 886 887 /* 888 * Destroys bookmarks. 889 * 890 * The keys in the bmarks nvlist are the bookmarks to be destroyed. 891 * They must all be in the same pool. Bookmarks are specified as 892 * <fs>#<bmark>. 893 * 894 * Bookmarks that do not exist will be silently ignored. 895 * 896 * The return value will be 0 if all bookmarks that existed were destroyed. 897 * 898 * Otherwise the return value will be the errno of a (undetermined) bookmark 899 * that failed, no bookmarks will be destroyed, and the errlist will have an 900 * entry for each bookmarks that failed. The value in the errlist will be 901 * the (int32) error code. 902 */ 903 int 904 lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist) 905 { 906 nvpair_t *elem; 907 int error; 908 char pool[ZFS_MAX_DATASET_NAME_LEN]; 909 910 /* determine the pool name */ 911 elem = nvlist_next_nvpair(bmarks, NULL); 912 if (elem == NULL) 913 return (0); 914 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 915 pool[strcspn(pool, "/#")] = '\0'; 916 917 error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist); 918 919 return (error); 920 } 921 922 static int 923 lzc_channel_program_impl(const char *pool, const char *program, boolean_t sync, 924 uint64_t instrlimit, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 925 { 926 int error; 927 nvlist_t *args; 928 929 args = fnvlist_alloc(); 930 fnvlist_add_string(args, ZCP_ARG_PROGRAM, program); 931 fnvlist_add_nvlist(args, ZCP_ARG_ARGLIST, argnvl); 932 fnvlist_add_boolean_value(args, ZCP_ARG_SYNC, sync); 933 fnvlist_add_uint64(args, ZCP_ARG_INSTRLIMIT, instrlimit); 934 fnvlist_add_uint64(args, ZCP_ARG_MEMLIMIT, memlimit); 935 error = lzc_ioctl(ZFS_IOC_CHANNEL_PROGRAM, pool, args, outnvl); 936 fnvlist_free(args); 937 938 return (error); 939 } 940 941 /* 942 * Executes a channel program. 943 * 944 * If this function returns 0 the channel program was successfully loaded and 945 * ran without failing. Note that individual commands the channel program ran 946 * may have failed and the channel program is responsible for reporting such 947 * errors through outnvl if they are important. 948 * 949 * This method may also return: 950 * 951 * EINVAL The program contains syntax errors, or an invalid memory or time 952 * limit was given. No part of the channel program was executed. 953 * If caused by syntax errors, 'outnvl' contains information about the 954 * errors. 955 * 956 * ECHRNG The program was executed, but encountered a runtime error, such as 957 * calling a function with incorrect arguments, invoking the error() 958 * function directly, failing an assert() command, etc. Some portion 959 * of the channel program may have executed and committed changes. 960 * Information about the failure can be found in 'outnvl'. 961 * 962 * ENOMEM The program fully executed, but the output buffer was not large 963 * enough to store the returned value. No output is returned through 964 * 'outnvl'. 965 * 966 * ENOSPC The program was terminated because it exceeded its memory usage 967 * limit. Some portion of the channel program may have executed and 968 * committed changes to disk. No output is returned through 'outnvl'. 969 * 970 * ETIME The program was terminated because it exceeded its Lua instruction 971 * limit. Some portion of the channel program may have executed and 972 * committed changes to disk. No output is returned through 'outnvl'. 973 */ 974 int 975 lzc_channel_program(const char *pool, const char *program, uint64_t instrlimit, 976 uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 977 { 978 return (lzc_channel_program_impl(pool, program, B_TRUE, instrlimit, 979 memlimit, argnvl, outnvl)); 980 } 981 982 /* 983 * Creates a checkpoint for the specified pool. 984 * 985 * If this function returns 0 the pool was successfully checkpointed. 986 * 987 * This method may also return: 988 * 989 * ZFS_ERR_CHECKPOINT_EXISTS 990 * The pool already has a checkpoint. A pools can only have one 991 * checkpoint at most, at any given time. 992 * 993 * ZFS_ERR_DISCARDING_CHECKPOINT 994 * ZFS is in the middle of discarding a checkpoint for this pool. 995 * The pool can be checkpointed again once the discard is done. 996 * 997 * ZFS_DEVRM_IN_PROGRESS 998 * A vdev is currently being removed. The pool cannot be 999 * checkpointed until the device removal is done. 1000 * 1001 * ZFS_VDEV_TOO_BIG 1002 * One or more top-level vdevs exceed the maximum vdev size 1003 * supported for this feature. 1004 */ 1005 int 1006 lzc_pool_checkpoint(const char *pool) 1007 { 1008 int error; 1009 1010 nvlist_t *result = NULL; 1011 nvlist_t *args = fnvlist_alloc(); 1012 1013 error = lzc_ioctl(ZFS_IOC_POOL_CHECKPOINT, pool, args, &result); 1014 1015 fnvlist_free(args); 1016 fnvlist_free(result); 1017 1018 return (error); 1019 } 1020 1021 /* 1022 * Discard the checkpoint from the specified pool. 1023 * 1024 * If this function returns 0 the checkpoint was successfully discarded. 1025 * 1026 * This method may also return: 1027 * 1028 * ZFS_ERR_NO_CHECKPOINT 1029 * The pool does not have a checkpoint. 1030 * 1031 * ZFS_ERR_DISCARDING_CHECKPOINT 1032 * ZFS is already in the middle of discarding the checkpoint. 1033 */ 1034 int 1035 lzc_pool_checkpoint_discard(const char *pool) 1036 { 1037 int error; 1038 1039 nvlist_t *result = NULL; 1040 nvlist_t *args = fnvlist_alloc(); 1041 1042 error = lzc_ioctl(ZFS_IOC_POOL_DISCARD_CHECKPOINT, pool, args, &result); 1043 1044 fnvlist_free(args); 1045 fnvlist_free(result); 1046 1047 return (error); 1048 } 1049 1050 /* 1051 * Executes a read-only channel program. 1052 * 1053 * A read-only channel program works programmatically the same way as a 1054 * normal channel program executed with lzc_channel_program(). The only 1055 * difference is it runs exclusively in open-context and therefore can 1056 * return faster. The downside to that, is that the program cannot change 1057 * on-disk state by calling functions from the zfs.sync submodule. 1058 * 1059 * The return values of this function (and their meaning) are exactly the 1060 * same as the ones described in lzc_channel_program(). 1061 */ 1062 int 1063 lzc_channel_program_nosync(const char *pool, const char *program, 1064 uint64_t timeout, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 1065 { 1066 return (lzc_channel_program_impl(pool, program, B_FALSE, timeout, 1067 memlimit, argnvl, outnvl)); 1068 } 1069 1070 /* 1071 * Changes initializing state. 1072 * 1073 * vdevs should be a list of (<key>, guid) where guid is a uint64 vdev GUID. 1074 * The key is ignored. 1075 * 1076 * If there are errors related to vdev arguments, per-vdev errors are returned 1077 * in an nvlist with the key "vdevs". Each error is a (guid, errno) pair where 1078 * guid is stringified with PRIu64, and errno is one of the following as 1079 * an int64_t: 1080 * - ENODEV if the device was not found 1081 * - EINVAL if the devices is not a leaf or is not concrete (e.g. missing) 1082 * - EROFS if the device is not writeable 1083 * - EBUSY start requested but the device is already being initialized 1084 * - ESRCH cancel/suspend requested but device is not being initialized 1085 * 1086 * If the errlist is empty, then return value will be: 1087 * - EINVAL if one or more arguments was invalid 1088 * - Other spa_open failures 1089 * - 0 if the operation succeeded 1090 */ 1091 int 1092 lzc_initialize(const char *poolname, pool_initialize_func_t cmd_type, 1093 nvlist_t *vdevs, nvlist_t **errlist) 1094 { 1095 int error; 1096 nvlist_t *args = fnvlist_alloc(); 1097 fnvlist_add_uint64(args, ZPOOL_INITIALIZE_COMMAND, (uint64_t)cmd_type); 1098 fnvlist_add_nvlist(args, ZPOOL_INITIALIZE_VDEVS, vdevs); 1099 1100 error = lzc_ioctl(ZFS_IOC_POOL_INITIALIZE, poolname, args, errlist); 1101 1102 fnvlist_free(args); 1103 1104 return (error); 1105 } 1106