libzfs_core.c revision 325534
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, 2016 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#define _IN_LIBZFS_CORE_ 76 77#include <libzfs_core.h> 78#include <ctype.h> 79#include <unistd.h> 80#include <stdlib.h> 81#include <string.h> 82#include <errno.h> 83#include <fcntl.h> 84#include <pthread.h> 85#include <sys/nvpair.h> 86#include <sys/param.h> 87#include <sys/types.h> 88#include <sys/stat.h> 89#include <sys/zfs_ioctl.h> 90#include "libzfs_core_compat.h" 91#include "libzfs_compat.h" 92 93#ifdef __FreeBSD__ 94extern int zfs_ioctl_version; 95#endif 96 97static int g_fd; 98static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER; 99static int g_refcount; 100 101int 102libzfs_core_init(void) 103{ 104 (void) pthread_mutex_lock(&g_lock); 105 if (g_refcount == 0) { 106 g_fd = open("/dev/zfs", O_RDWR); 107 if (g_fd < 0) { 108 (void) pthread_mutex_unlock(&g_lock); 109 return (errno); 110 } 111 } 112 g_refcount++; 113 (void) pthread_mutex_unlock(&g_lock); 114 115 return (0); 116} 117 118void 119libzfs_core_fini(void) 120{ 121 (void) pthread_mutex_lock(&g_lock); 122 ASSERT3S(g_refcount, >, 0); 123 g_refcount--; 124 if (g_refcount == 0) 125 (void) close(g_fd); 126 (void) pthread_mutex_unlock(&g_lock); 127} 128 129static int 130lzc_ioctl(zfs_ioc_t ioc, const char *name, 131 nvlist_t *source, nvlist_t **resultp) 132{ 133 zfs_cmd_t zc = { 0 }; 134 int error = 0; 135 char *packed; 136#ifdef __FreeBSD__ 137 nvlist_t *oldsource; 138#endif 139 size_t size; 140 141 ASSERT3S(g_refcount, >, 0); 142 143 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 144 145#ifdef __FreeBSD__ 146 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) 147 zfs_ioctl_version = get_zfs_ioctl_version(); 148 149 if (zfs_ioctl_version < ZFS_IOCVER_LZC) { 150 oldsource = source; 151 error = lzc_compat_pre(&zc, &ioc, &source); 152 if (error) 153 return (error); 154 } 155#endif 156 157 packed = fnvlist_pack(source, &size); 158 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 159 zc.zc_nvlist_src_size = size; 160 161 if (resultp != NULL) { 162 *resultp = NULL; 163 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024); 164 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 165 malloc(zc.zc_nvlist_dst_size); 166#ifdef illumos 167 if (zc.zc_nvlist_dst == NULL) { 168#else 169 if (zc.zc_nvlist_dst == 0) { 170#endif 171 error = ENOMEM; 172 goto out; 173 } 174 } 175 176 while (ioctl(g_fd, ioc, &zc) != 0) { 177 /* 178 * If ioctl exited with ENOMEM, we retry the ioctl after 179 * increasing the size of the destination nvlist. 180 * 181 * Channel programs that exit with ENOMEM probably ran over the 182 * lua memory sandbox; they should not be retried. 183 */ 184 if (errno == ENOMEM && resultp != NULL && 185 ioc != ZFS_IOC_CHANNEL_PROGRAM) { 186 free((void *)(uintptr_t)zc.zc_nvlist_dst); 187 zc.zc_nvlist_dst_size *= 2; 188 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 189 malloc(zc.zc_nvlist_dst_size); 190#ifdef illumos 191 if (zc.zc_nvlist_dst == NULL) { 192#else 193 if (zc.zc_nvlist_dst == 0) { 194#endif 195 error = ENOMEM; 196 goto out; 197 } 198 } else { 199 error = errno; 200 break; 201 } 202 } 203 204#ifdef __FreeBSD__ 205 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 206 lzc_compat_post(&zc, ioc); 207#endif 208 if (zc.zc_nvlist_dst_filled) { 209 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 210 zc.zc_nvlist_dst_size); 211 } 212#ifdef __FreeBSD__ 213 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 214 lzc_compat_outnvl(&zc, ioc, resultp); 215#endif 216out: 217#ifdef __FreeBSD__ 218 if (zfs_ioctl_version < ZFS_IOCVER_LZC) { 219 if (source != oldsource) 220 nvlist_free(source); 221 source = oldsource; 222 } 223#endif 224 fnvlist_pack_free(packed, size); 225 free((void *)(uintptr_t)zc.zc_nvlist_dst); 226 return (error); 227} 228 229int 230lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props) 231{ 232 int error; 233 nvlist_t *args = fnvlist_alloc(); 234 fnvlist_add_int32(args, "type", (dmu_objset_type_t)type); 235 if (props != NULL) 236 fnvlist_add_nvlist(args, "props", props); 237 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL); 238 nvlist_free(args); 239 return (error); 240} 241 242int 243lzc_clone(const char *fsname, const char *origin, 244 nvlist_t *props) 245{ 246 int error; 247 nvlist_t *args = fnvlist_alloc(); 248 fnvlist_add_string(args, "origin", origin); 249 if (props != NULL) 250 fnvlist_add_nvlist(args, "props", props); 251 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL); 252 nvlist_free(args); 253 return (error); 254} 255 256int 257lzc_promote(const char *fsname, char *snapnamebuf, int snapnamelen) 258{ 259 /* 260 * The promote ioctl is still legacy, so we need to construct our 261 * own zfs_cmd_t rather than using lzc_ioctl(). 262 */ 263 zfs_cmd_t zc = { 0 }; 264 265 ASSERT3S(g_refcount, >, 0); 266 VERIFY3S(g_fd, !=, -1); 267 268 (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name)); 269 if (ioctl(g_fd, ZFS_IOC_PROMOTE, &zc) != 0) { 270 int error = errno; 271 if (error == EEXIST && snapnamebuf != NULL) 272 (void) strlcpy(snapnamebuf, zc.zc_string, snapnamelen); 273 return (error); 274 } 275 return (0); 276} 277 278/* 279 * Creates snapshots. 280 * 281 * The keys in the snaps nvlist are the snapshots to be created. 282 * They must all be in the same pool. 283 * 284 * The props nvlist is properties to set. Currently only user properties 285 * are supported. { user:prop_name -> string value } 286 * 287 * The returned results nvlist will have an entry for each snapshot that failed. 288 * The value will be the (int32) error code. 289 * 290 * The return value will be 0 if all snapshots were created, otherwise it will 291 * be the errno of a (unspecified) snapshot that failed. 292 */ 293int 294lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) 295{ 296 nvpair_t *elem; 297 nvlist_t *args; 298 int error; 299 char pool[ZFS_MAX_DATASET_NAME_LEN]; 300 301 *errlist = NULL; 302 303 /* determine the pool name */ 304 elem = nvlist_next_nvpair(snaps, NULL); 305 if (elem == NULL) 306 return (0); 307 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 308 pool[strcspn(pool, "/@")] = '\0'; 309 310 args = fnvlist_alloc(); 311 fnvlist_add_nvlist(args, "snaps", snaps); 312 if (props != NULL) 313 fnvlist_add_nvlist(args, "props", props); 314 315 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist); 316 nvlist_free(args); 317 318 return (error); 319} 320 321/* 322 * Destroys snapshots. 323 * 324 * The keys in the snaps nvlist are the snapshots to be destroyed. 325 * They must all be in the same pool. 326 * 327 * Snapshots that do not exist will be silently ignored. 328 * 329 * If 'defer' is not set, and a snapshot has user holds or clones, the 330 * destroy operation will fail and none of the snapshots will be 331 * destroyed. 332 * 333 * If 'defer' is set, and a snapshot has user holds or clones, it will be 334 * marked for deferred destruction, and will be destroyed when the last hold 335 * or clone is removed/destroyed. 336 * 337 * The return value will be 0 if all snapshots were destroyed (or marked for 338 * later destruction if 'defer' is set) or didn't exist to begin with. 339 * 340 * Otherwise the return value will be the errno of a (unspecified) snapshot 341 * that failed, no snapshots will be destroyed, and the errlist will have an 342 * entry for each snapshot that failed. The value in the errlist will be 343 * the (int32) error code. 344 */ 345int 346lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist) 347{ 348 nvpair_t *elem; 349 nvlist_t *args; 350 int error; 351 char pool[ZFS_MAX_DATASET_NAME_LEN]; 352 353 /* determine the pool name */ 354 elem = nvlist_next_nvpair(snaps, NULL); 355 if (elem == NULL) 356 return (0); 357 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 358 pool[strcspn(pool, "/@")] = '\0'; 359 360 args = fnvlist_alloc(); 361 fnvlist_add_nvlist(args, "snaps", snaps); 362 if (defer) 363 fnvlist_add_boolean(args, "defer"); 364 365 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist); 366 nvlist_free(args); 367 368 return (error); 369} 370 371int 372lzc_snaprange_space(const char *firstsnap, const char *lastsnap, 373 uint64_t *usedp) 374{ 375 nvlist_t *args; 376 nvlist_t *result; 377 int err; 378 char fs[ZFS_MAX_DATASET_NAME_LEN]; 379 char *atp; 380 381 /* determine the fs name */ 382 (void) strlcpy(fs, firstsnap, sizeof (fs)); 383 atp = strchr(fs, '@'); 384 if (atp == NULL) 385 return (EINVAL); 386 *atp = '\0'; 387 388 args = fnvlist_alloc(); 389 fnvlist_add_string(args, "firstsnap", firstsnap); 390 391 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result); 392 nvlist_free(args); 393 if (err == 0) 394 *usedp = fnvlist_lookup_uint64(result, "used"); 395 fnvlist_free(result); 396 397 return (err); 398} 399 400boolean_t 401lzc_exists(const char *dataset) 402{ 403 /* 404 * The objset_stats ioctl is still legacy, so we need to construct our 405 * own zfs_cmd_t rather than using lzc_ioctl(). 406 */ 407 zfs_cmd_t zc = { 0 }; 408 409 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); 410 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0); 411} 412 413/* 414 * Create "user holds" on snapshots. If there is a hold on a snapshot, 415 * the snapshot can not be destroyed. (However, it can be marked for deletion 416 * by lzc_destroy_snaps(defer=B_TRUE).) 417 * 418 * The keys in the nvlist are snapshot names. 419 * The snapshots must all be in the same pool. 420 * The value is the name of the hold (string type). 421 * 422 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL). 423 * In this case, when the cleanup_fd is closed (including on process 424 * termination), the holds will be released. If the system is shut down 425 * uncleanly, the holds will be released when the pool is next opened 426 * or imported. 427 * 428 * Holds for snapshots which don't exist will be skipped and have an entry 429 * added to errlist, but will not cause an overall failure. 430 * 431 * The return value will be 0 if all holds, for snapshots that existed, 432 * were succesfully created. 433 * 434 * Otherwise the return value will be the errno of a (unspecified) hold that 435 * failed and no holds will be created. 436 * 437 * In all cases the errlist will have an entry for each hold that failed 438 * (name = snapshot), with its value being the error code (int32). 439 */ 440int 441lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist) 442{ 443 char pool[ZFS_MAX_DATASET_NAME_LEN]; 444 nvlist_t *args; 445 nvpair_t *elem; 446 int error; 447 448 /* determine the pool name */ 449 elem = nvlist_next_nvpair(holds, NULL); 450 if (elem == NULL) 451 return (0); 452 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 453 pool[strcspn(pool, "/@")] = '\0'; 454 455 args = fnvlist_alloc(); 456 fnvlist_add_nvlist(args, "holds", holds); 457 if (cleanup_fd != -1) 458 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd); 459 460 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist); 461 nvlist_free(args); 462 return (error); 463} 464 465/* 466 * Release "user holds" on snapshots. If the snapshot has been marked for 467 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have 468 * any clones, and all the user holds are removed, then the snapshot will be 469 * destroyed. 470 * 471 * The keys in the nvlist are snapshot names. 472 * The snapshots must all be in the same pool. 473 * The value is a nvlist whose keys are the holds to remove. 474 * 475 * Holds which failed to release because they didn't exist will have an entry 476 * added to errlist, but will not cause an overall failure. 477 * 478 * The return value will be 0 if the nvl holds was empty or all holds that 479 * existed, were successfully removed. 480 * 481 * Otherwise the return value will be the errno of a (unspecified) hold that 482 * failed to release and no holds will be released. 483 * 484 * In all cases the errlist will have an entry for each hold that failed to 485 * to release. 486 */ 487int 488lzc_release(nvlist_t *holds, nvlist_t **errlist) 489{ 490 char pool[ZFS_MAX_DATASET_NAME_LEN]; 491 nvpair_t *elem; 492 493 /* determine the pool name */ 494 elem = nvlist_next_nvpair(holds, NULL); 495 if (elem == NULL) 496 return (0); 497 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 498 pool[strcspn(pool, "/@")] = '\0'; 499 500 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist)); 501} 502 503/* 504 * Retrieve list of user holds on the specified snapshot. 505 * 506 * On success, *holdsp will be set to a nvlist which the caller must free. 507 * The keys are the names of the holds, and the value is the creation time 508 * of the hold (uint64) in seconds since the epoch. 509 */ 510int 511lzc_get_holds(const char *snapname, nvlist_t **holdsp) 512{ 513 int error; 514 nvlist_t *innvl = fnvlist_alloc(); 515 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp); 516 fnvlist_free(innvl); 517 return (error); 518} 519 520/* 521 * Generate a zfs send stream for the specified snapshot and write it to 522 * the specified file descriptor. 523 * 524 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap") 525 * 526 * If "from" is NULL, a full (non-incremental) stream will be sent. 527 * If "from" is non-NULL, it must be the full name of a snapshot or 528 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or 529 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or 530 * bookmark must represent an earlier point in the history of "snapname"). 531 * It can be an earlier snapshot in the same filesystem or zvol as "snapname", 532 * or it can be the origin of "snapname"'s filesystem, or an earlier 533 * snapshot in the origin, etc. 534 * 535 * "fd" is the file descriptor to write the send stream to. 536 * 537 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted 538 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT 539 * records with drr_blksz > 128K. 540 * 541 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted 542 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA, 543 * which the receiving system must support (as indicated by support 544 * for the "embedded_data" feature). 545 */ 546int 547lzc_send(const char *snapname, const char *from, int fd, 548 enum lzc_send_flags flags) 549{ 550 return (lzc_send_resume(snapname, from, fd, flags, 0, 0)); 551} 552 553int 554lzc_send_resume(const char *snapname, const char *from, int fd, 555 enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff) 556{ 557 nvlist_t *args; 558 int err; 559 560 args = fnvlist_alloc(); 561 fnvlist_add_int32(args, "fd", fd); 562 if (from != NULL) 563 fnvlist_add_string(args, "fromsnap", from); 564 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 565 fnvlist_add_boolean(args, "largeblockok"); 566 if (flags & LZC_SEND_FLAG_EMBED_DATA) 567 fnvlist_add_boolean(args, "embedok"); 568 if (flags & LZC_SEND_FLAG_COMPRESS) 569 fnvlist_add_boolean(args, "compressok"); 570 if (resumeobj != 0 || resumeoff != 0) { 571 fnvlist_add_uint64(args, "resume_object", resumeobj); 572 fnvlist_add_uint64(args, "resume_offset", resumeoff); 573 } 574 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL); 575 nvlist_free(args); 576 return (err); 577} 578 579/* 580 * "from" can be NULL, a snapshot, or a bookmark. 581 * 582 * If from is NULL, a full (non-incremental) stream will be estimated. This 583 * is calculated very efficiently. 584 * 585 * If from is a snapshot, lzc_send_space uses the deadlists attached to 586 * each snapshot to efficiently estimate the stream size. 587 * 588 * If from is a bookmark, the indirect blocks in the destination snapshot 589 * are traversed, looking for blocks with a birth time since the creation TXG of 590 * the snapshot this bookmark was created from. This will result in 591 * significantly more I/O and be less efficient than a send space estimation on 592 * an equivalent snapshot. 593 */ 594int 595lzc_send_space(const char *snapname, const char *from, 596 enum lzc_send_flags flags, uint64_t *spacep) 597{ 598 nvlist_t *args; 599 nvlist_t *result; 600 int err; 601 602 args = fnvlist_alloc(); 603 if (from != NULL) 604 fnvlist_add_string(args, "from", from); 605 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 606 fnvlist_add_boolean(args, "largeblockok"); 607 if (flags & LZC_SEND_FLAG_EMBED_DATA) 608 fnvlist_add_boolean(args, "embedok"); 609 if (flags & LZC_SEND_FLAG_COMPRESS) 610 fnvlist_add_boolean(args, "compressok"); 611 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result); 612 nvlist_free(args); 613 if (err == 0) 614 *spacep = fnvlist_lookup_uint64(result, "space"); 615 nvlist_free(result); 616 return (err); 617} 618 619static int 620recv_read(int fd, void *buf, int ilen) 621{ 622 char *cp = buf; 623 int rv; 624 int len = ilen; 625 626 do { 627 rv = read(fd, cp, len); 628 cp += rv; 629 len -= rv; 630 } while (rv > 0); 631 632 if (rv < 0 || len != 0) 633 return (EIO); 634 635 return (0); 636} 637 638static int 639recv_impl(const char *snapname, nvlist_t *props, const char *origin, 640 boolean_t force, boolean_t resumable, int fd, 641 const dmu_replay_record_t *begin_record) 642{ 643 /* 644 * The receive ioctl is still legacy, so we need to construct our own 645 * zfs_cmd_t rather than using zfsc_ioctl(). 646 */ 647 zfs_cmd_t zc = { 0 }; 648 char *atp; 649 char *packed = NULL; 650 size_t size; 651 int error; 652 653 ASSERT3S(g_refcount, >, 0); 654 655 /* zc_name is name of containing filesystem */ 656 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name)); 657 atp = strchr(zc.zc_name, '@'); 658 if (atp == NULL) 659 return (EINVAL); 660 *atp = '\0'; 661 662 /* if the fs does not exist, try its parent. */ 663 if (!lzc_exists(zc.zc_name)) { 664 char *slashp = strrchr(zc.zc_name, '/'); 665 if (slashp == NULL) 666 return (ENOENT); 667 *slashp = '\0'; 668 669 } 670 671 /* zc_value is full name of the snapshot to create */ 672 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); 673 674 if (props != NULL) { 675 /* zc_nvlist_src is props to set */ 676 packed = fnvlist_pack(props, &size); 677 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 678 zc.zc_nvlist_src_size = size; 679 } 680 681 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */ 682 if (origin != NULL) 683 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string)); 684 685 /* zc_begin_record is non-byteswapped BEGIN record */ 686 if (begin_record == NULL) { 687 error = recv_read(fd, &zc.zc_begin_record, 688 sizeof (zc.zc_begin_record)); 689 if (error != 0) 690 goto out; 691 } else { 692 zc.zc_begin_record = *begin_record; 693 } 694 695 /* zc_cookie is fd to read from */ 696 zc.zc_cookie = fd; 697 698 /* zc guid is force flag */ 699 zc.zc_guid = force; 700 701 zc.zc_resumable = resumable; 702 703 /* zc_cleanup_fd is unused */ 704 zc.zc_cleanup_fd = -1; 705 706 error = ioctl(g_fd, ZFS_IOC_RECV, &zc); 707 if (error != 0) 708 error = errno; 709 710out: 711 if (packed != NULL) 712 fnvlist_pack_free(packed, size); 713 free((void*)(uintptr_t)zc.zc_nvlist_dst); 714 return (error); 715} 716 717/* 718 * The simplest receive case: receive from the specified fd, creating the 719 * specified snapshot. Apply the specified properties as "received" properties 720 * (which can be overridden by locally-set properties). If the stream is a 721 * clone, its origin snapshot must be specified by 'origin'. The 'force' 722 * flag will cause the target filesystem to be rolled back or destroyed if 723 * necessary to receive. 724 * 725 * Return 0 on success or an errno on failure. 726 * 727 * Note: this interface does not work on dedup'd streams 728 * (those with DMU_BACKUP_FEATURE_DEDUP). 729 */ 730int 731lzc_receive(const char *snapname, nvlist_t *props, const char *origin, 732 boolean_t force, int fd) 733{ 734 return (recv_impl(snapname, props, origin, force, B_FALSE, fd, NULL)); 735} 736 737/* 738 * Like lzc_receive, but if the receive fails due to premature stream 739 * termination, the intermediate state will be preserved on disk. In this 740 * case, ECKSUM will be returned. The receive may subsequently be resumed 741 * with a resuming send stream generated by lzc_send_resume(). 742 */ 743int 744lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin, 745 boolean_t force, int fd) 746{ 747 return (recv_impl(snapname, props, origin, force, B_TRUE, fd, NULL)); 748} 749 750/* 751 * Like lzc_receive, but allows the caller to read the begin record and then to 752 * pass it in. That could be useful if the caller wants to derive, for example, 753 * the snapname or the origin parameters based on the information contained in 754 * the begin record. 755 * The begin record must be in its original form as read from the stream, 756 * in other words, it should not be byteswapped. 757 * 758 * The 'resumable' parameter allows to obtain the same behavior as with 759 * lzc_receive_resumable. 760 */ 761int 762lzc_receive_with_header(const char *snapname, nvlist_t *props, 763 const char *origin, boolean_t force, boolean_t resumable, int fd, 764 const dmu_replay_record_t *begin_record) 765{ 766 if (begin_record == NULL) 767 return (EINVAL); 768 return (recv_impl(snapname, props, origin, force, resumable, fd, 769 begin_record)); 770} 771 772/* 773 * Roll back this filesystem or volume to its most recent snapshot. 774 * If snapnamebuf is not NULL, it will be filled in with the name 775 * of the most recent snapshot. 776 * Note that the latest snapshot may change if a new one is concurrently 777 * created or the current one is destroyed. lzc_rollback_to can be used 778 * to roll back to a specific latest snapshot. 779 * 780 * Return 0 on success or an errno on failure. 781 */ 782int 783lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen) 784{ 785 nvlist_t *args; 786 nvlist_t *result; 787 int err; 788 789 args = fnvlist_alloc(); 790 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 791 nvlist_free(args); 792 if (err == 0 && snapnamebuf != NULL) { 793 const char *snapname = fnvlist_lookup_string(result, "target"); 794 (void) strlcpy(snapnamebuf, snapname, snapnamelen); 795 } 796 nvlist_free(result); 797 798 return (err); 799} 800 801/* 802 * Roll back this filesystem or volume to the specified snapshot, 803 * if possible. 804 * 805 * Return 0 on success or an errno on failure. 806 */ 807int 808lzc_rollback_to(const char *fsname, const char *snapname) 809{ 810 nvlist_t *args; 811 nvlist_t *result; 812 int err; 813 814 args = fnvlist_alloc(); 815 fnvlist_add_string(args, "target", snapname); 816 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 817 nvlist_free(args); 818 nvlist_free(result); 819 return (err); 820} 821 822/* 823 * Creates bookmarks. 824 * 825 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to 826 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and 827 * snapshots must be in the same pool. 828 * 829 * The returned results nvlist will have an entry for each bookmark that failed. 830 * The value will be the (int32) error code. 831 * 832 * The return value will be 0 if all bookmarks were created, otherwise it will 833 * be the errno of a (undetermined) bookmarks that failed. 834 */ 835int 836lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist) 837{ 838 nvpair_t *elem; 839 int error; 840 char pool[ZFS_MAX_DATASET_NAME_LEN]; 841 842 /* determine the pool name */ 843 elem = nvlist_next_nvpair(bookmarks, NULL); 844 if (elem == NULL) 845 return (0); 846 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 847 pool[strcspn(pool, "/#")] = '\0'; 848 849 error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist); 850 851 return (error); 852} 853 854/* 855 * Retrieve bookmarks. 856 * 857 * Retrieve the list of bookmarks for the given file system. The props 858 * parameter is an nvlist of property names (with no values) that will be 859 * returned for each bookmark. 860 * 861 * The following are valid properties on bookmarks, all of which are numbers 862 * (represented as uint64 in the nvlist) 863 * 864 * "guid" - globally unique identifier of the snapshot it refers to 865 * "createtxg" - txg when the snapshot it refers to was created 866 * "creation" - timestamp when the snapshot it refers to was created 867 * 868 * The format of the returned nvlist as follows: 869 * <short name of bookmark> -> { 870 * <name of property> -> { 871 * "value" -> uint64 872 * } 873 * } 874 */ 875int 876lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks) 877{ 878 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks)); 879} 880 881/* 882 * Destroys bookmarks. 883 * 884 * The keys in the bmarks nvlist are the bookmarks to be destroyed. 885 * They must all be in the same pool. Bookmarks are specified as 886 * <fs>#<bmark>. 887 * 888 * Bookmarks that do not exist will be silently ignored. 889 * 890 * The return value will be 0 if all bookmarks that existed were destroyed. 891 * 892 * Otherwise the return value will be the errno of a (undetermined) bookmark 893 * that failed, no bookmarks will be destroyed, and the errlist will have an 894 * entry for each bookmarks that failed. The value in the errlist will be 895 * the (int32) error code. 896 */ 897int 898lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist) 899{ 900 nvpair_t *elem; 901 int error; 902 char pool[ZFS_MAX_DATASET_NAME_LEN]; 903 904 /* determine the pool name */ 905 elem = nvlist_next_nvpair(bmarks, NULL); 906 if (elem == NULL) 907 return (0); 908 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 909 pool[strcspn(pool, "/#")] = '\0'; 910 911 error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist); 912 913 return (error); 914} 915 916/* 917 * Executes a channel program. 918 * 919 * If this function returns 0 the channel program was successfully loaded and 920 * ran without failing. Note that individual commands the channel program ran 921 * may have failed and the channel program is responsible for reporting such 922 * errors through outnvl if they are important. 923 * 924 * This method may also return: 925 * 926 * EINVAL The program contains syntax errors, or an invalid memory or time 927 * limit was given. No part of the channel program was executed. 928 * If caused by syntax errors, 'outnvl' contains information about the 929 * errors. 930 * 931 * EDOM The program was executed, but encountered a runtime error, such as 932 * calling a function with incorrect arguments, invoking the error() 933 * function directly, failing an assert() command, etc. Some portion 934 * of the channel program may have executed and committed changes. 935 * Information about the failure can be found in 'outnvl'. 936 * 937 * ENOMEM The program fully executed, but the output buffer was not large 938 * enough to store the returned value. No output is returned through 939 * 'outnvl'. 940 * 941 * ENOSPC The program was terminated because it exceeded its memory usage 942 * limit. Some portion of the channel program may have executed and 943 * committed changes to disk. No output is returned through 'outnvl'. 944 * 945 * ETIMEDOUT The program was terminated because it exceeded its Lua instruction 946 * limit. Some portion of the channel program may have executed and 947 * committed changes to disk. No output is returned through 'outnvl'. 948 */ 949int 950lzc_channel_program(const char *pool, const char *program, uint64_t instrlimit, 951 uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 952{ 953 int error; 954 nvlist_t *args; 955 956 args = fnvlist_alloc(); 957 fnvlist_add_string(args, ZCP_ARG_PROGRAM, program); 958 fnvlist_add_nvlist(args, ZCP_ARG_ARGLIST, argnvl); 959 fnvlist_add_uint64(args, ZCP_ARG_INSTRLIMIT, instrlimit); 960 fnvlist_add_uint64(args, ZCP_ARG_MEMLIMIT, memlimit); 961 error = lzc_ioctl(ZFS_IOC_CHANNEL_PROGRAM, pool, args, outnvl); 962 fnvlist_free(args); 963 964 return (error); 965} 966