libzfs_core.c revision 331393
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#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 = -1; 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); 123 124 if (g_refcount > 0) 125 g_refcount--; 126 127 if (g_refcount == 0 && g_fd != -1) { 128 (void) close(g_fd); 129 g_fd = -1; 130 } 131 (void) pthread_mutex_unlock(&g_lock); 132} 133 134static int 135lzc_ioctl(zfs_ioc_t ioc, const char *name, 136 nvlist_t *source, nvlist_t **resultp) 137{ 138 zfs_cmd_t zc = { 0 }; 139 int error = 0; 140 char *packed; 141#ifdef __FreeBSD__ 142 nvlist_t *oldsource; 143#endif 144 size_t size; 145 146 ASSERT3S(g_refcount, >, 0); 147 VERIFY3S(g_fd, !=, -1); 148 149 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name)); 150 151#ifdef __FreeBSD__ 152 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF) 153 zfs_ioctl_version = get_zfs_ioctl_version(); 154 155 if (zfs_ioctl_version < ZFS_IOCVER_LZC) { 156 oldsource = source; 157 error = lzc_compat_pre(&zc, &ioc, &source); 158 if (error) 159 return (error); 160 } 161#endif 162 163 packed = fnvlist_pack(source, &size); 164 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 165 zc.zc_nvlist_src_size = size; 166 167 if (resultp != NULL) { 168 *resultp = NULL; 169 if (ioc == ZFS_IOC_CHANNEL_PROGRAM) { 170 zc.zc_nvlist_dst_size = fnvlist_lookup_uint64(source, 171 ZCP_ARG_MEMLIMIT); 172 } else { 173 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024); 174 } 175 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 176 malloc(zc.zc_nvlist_dst_size); 177#ifdef illumos 178 if (zc.zc_nvlist_dst == NULL) { 179#else 180 if (zc.zc_nvlist_dst == 0) { 181#endif 182 error = ENOMEM; 183 goto out; 184 } 185 } 186 187 while (ioctl(g_fd, ioc, &zc) != 0) { 188 /* 189 * If ioctl exited with ENOMEM, we retry the ioctl after 190 * increasing the size of the destination nvlist. 191 * 192 * Channel programs that exit with ENOMEM ran over the 193 * lua memory sandbox; they should not be retried. 194 */ 195 if (errno == ENOMEM && resultp != NULL && 196 ioc != ZFS_IOC_CHANNEL_PROGRAM) { 197 free((void *)(uintptr_t)zc.zc_nvlist_dst); 198 zc.zc_nvlist_dst_size *= 2; 199 zc.zc_nvlist_dst = (uint64_t)(uintptr_t) 200 malloc(zc.zc_nvlist_dst_size); 201#ifdef illumos 202 if (zc.zc_nvlist_dst == NULL) { 203#else 204 if (zc.zc_nvlist_dst == 0) { 205#endif 206 error = ENOMEM; 207 goto out; 208 } 209 } else { 210 error = errno; 211 break; 212 } 213 } 214 215#ifdef __FreeBSD__ 216 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 217 lzc_compat_post(&zc, ioc); 218#endif 219 if (zc.zc_nvlist_dst_filled) { 220 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst, 221 zc.zc_nvlist_dst_size); 222 } 223#ifdef __FreeBSD__ 224 if (zfs_ioctl_version < ZFS_IOCVER_LZC) 225 lzc_compat_outnvl(&zc, ioc, resultp); 226#endif 227out: 228#ifdef __FreeBSD__ 229 if (zfs_ioctl_version < ZFS_IOCVER_LZC) { 230 if (source != oldsource) 231 nvlist_free(source); 232 source = oldsource; 233 } 234#endif 235 fnvlist_pack_free(packed, size); 236 free((void *)(uintptr_t)zc.zc_nvlist_dst); 237 return (error); 238} 239 240int 241lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props) 242{ 243 int error; 244 nvlist_t *args = fnvlist_alloc(); 245 fnvlist_add_int32(args, "type", (dmu_objset_type_t)type); 246 if (props != NULL) 247 fnvlist_add_nvlist(args, "props", props); 248 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL); 249 nvlist_free(args); 250 return (error); 251} 252 253int 254lzc_clone(const char *fsname, const char *origin, 255 nvlist_t *props) 256{ 257 int error; 258 nvlist_t *args = fnvlist_alloc(); 259 fnvlist_add_string(args, "origin", origin); 260 if (props != NULL) 261 fnvlist_add_nvlist(args, "props", props); 262 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL); 263 nvlist_free(args); 264 return (error); 265} 266 267int 268lzc_promote(const char *fsname, char *snapnamebuf, int snapnamelen) 269{ 270 /* 271 * The promote ioctl is still legacy, so we need to construct our 272 * own zfs_cmd_t rather than using lzc_ioctl(). 273 */ 274 zfs_cmd_t zc = { 0 }; 275 276 ASSERT3S(g_refcount, >, 0); 277 VERIFY3S(g_fd, !=, -1); 278 279 (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name)); 280 if (ioctl(g_fd, ZFS_IOC_PROMOTE, &zc) != 0) { 281 int error = errno; 282 if (error == EEXIST && snapnamebuf != NULL) 283 (void) strlcpy(snapnamebuf, zc.zc_string, snapnamelen); 284 return (error); 285 } 286 return (0); 287} 288 289/* 290 * Creates snapshots. 291 * 292 * The keys in the snaps nvlist are the snapshots to be created. 293 * They must all be in the same pool. 294 * 295 * The props nvlist is properties to set. Currently only user properties 296 * are supported. { user:prop_name -> string value } 297 * 298 * The returned results nvlist will have an entry for each snapshot that failed. 299 * The value will be the (int32) error code. 300 * 301 * The return value will be 0 if all snapshots were created, otherwise it will 302 * be the errno of a (unspecified) snapshot that failed. 303 */ 304int 305lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist) 306{ 307 nvpair_t *elem; 308 nvlist_t *args; 309 int error; 310 char pool[ZFS_MAX_DATASET_NAME_LEN]; 311 312 *errlist = NULL; 313 314 /* determine the pool name */ 315 elem = nvlist_next_nvpair(snaps, NULL); 316 if (elem == NULL) 317 return (0); 318 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 319 pool[strcspn(pool, "/@")] = '\0'; 320 321 args = fnvlist_alloc(); 322 fnvlist_add_nvlist(args, "snaps", snaps); 323 if (props != NULL) 324 fnvlist_add_nvlist(args, "props", props); 325 326 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist); 327 nvlist_free(args); 328 329 return (error); 330} 331 332/* 333 * Destroys snapshots. 334 * 335 * The keys in the snaps nvlist are the snapshots to be destroyed. 336 * They must all be in the same pool. 337 * 338 * Snapshots that do not exist will be silently ignored. 339 * 340 * If 'defer' is not set, and a snapshot has user holds or clones, the 341 * destroy operation will fail and none of the snapshots will be 342 * destroyed. 343 * 344 * If 'defer' is set, and a snapshot has user holds or clones, it will be 345 * marked for deferred destruction, and will be destroyed when the last hold 346 * or clone is removed/destroyed. 347 * 348 * The return value will be 0 if all snapshots were destroyed (or marked for 349 * later destruction if 'defer' is set) or didn't exist to begin with. 350 * 351 * Otherwise the return value will be the errno of a (unspecified) snapshot 352 * that failed, no snapshots will be destroyed, and the errlist will have an 353 * entry for each snapshot that failed. The value in the errlist will be 354 * the (int32) error code. 355 */ 356int 357lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist) 358{ 359 nvpair_t *elem; 360 nvlist_t *args; 361 int error; 362 char pool[ZFS_MAX_DATASET_NAME_LEN]; 363 364 /* determine the pool name */ 365 elem = nvlist_next_nvpair(snaps, NULL); 366 if (elem == NULL) 367 return (0); 368 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 369 pool[strcspn(pool, "/@")] = '\0'; 370 371 args = fnvlist_alloc(); 372 fnvlist_add_nvlist(args, "snaps", snaps); 373 if (defer) 374 fnvlist_add_boolean(args, "defer"); 375 376 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist); 377 nvlist_free(args); 378 379 return (error); 380} 381 382int 383lzc_snaprange_space(const char *firstsnap, const char *lastsnap, 384 uint64_t *usedp) 385{ 386 nvlist_t *args; 387 nvlist_t *result; 388 int err; 389 char fs[ZFS_MAX_DATASET_NAME_LEN]; 390 char *atp; 391 392 /* determine the fs name */ 393 (void) strlcpy(fs, firstsnap, sizeof (fs)); 394 atp = strchr(fs, '@'); 395 if (atp == NULL) 396 return (EINVAL); 397 *atp = '\0'; 398 399 args = fnvlist_alloc(); 400 fnvlist_add_string(args, "firstsnap", firstsnap); 401 402 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result); 403 nvlist_free(args); 404 if (err == 0) 405 *usedp = fnvlist_lookup_uint64(result, "used"); 406 fnvlist_free(result); 407 408 return (err); 409} 410 411boolean_t 412lzc_exists(const char *dataset) 413{ 414 /* 415 * The objset_stats ioctl is still legacy, so we need to construct our 416 * own zfs_cmd_t rather than using lzc_ioctl(). 417 */ 418 zfs_cmd_t zc = { 0 }; 419 420 ASSERT3S(g_refcount, >, 0); 421 VERIFY3S(g_fd, !=, -1); 422 423 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name)); 424 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0); 425} 426 427/* 428 * Create "user holds" on snapshots. If there is a hold on a snapshot, 429 * the snapshot can not be destroyed. (However, it can be marked for deletion 430 * by lzc_destroy_snaps(defer=B_TRUE).) 431 * 432 * The keys in the nvlist are snapshot names. 433 * The snapshots must all be in the same pool. 434 * The value is the name of the hold (string type). 435 * 436 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL). 437 * In this case, when the cleanup_fd is closed (including on process 438 * termination), the holds will be released. If the system is shut down 439 * uncleanly, the holds will be released when the pool is next opened 440 * or imported. 441 * 442 * Holds for snapshots which don't exist will be skipped and have an entry 443 * added to errlist, but will not cause an overall failure. 444 * 445 * The return value will be 0 if all holds, for snapshots that existed, 446 * were succesfully created. 447 * 448 * Otherwise the return value will be the errno of a (unspecified) hold that 449 * failed and no holds will be created. 450 * 451 * In all cases the errlist will have an entry for each hold that failed 452 * (name = snapshot), with its value being the error code (int32). 453 */ 454int 455lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist) 456{ 457 char pool[ZFS_MAX_DATASET_NAME_LEN]; 458 nvlist_t *args; 459 nvpair_t *elem; 460 int error; 461 462 /* determine the pool name */ 463 elem = nvlist_next_nvpair(holds, NULL); 464 if (elem == NULL) 465 return (0); 466 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 467 pool[strcspn(pool, "/@")] = '\0'; 468 469 args = fnvlist_alloc(); 470 fnvlist_add_nvlist(args, "holds", holds); 471 if (cleanup_fd != -1) 472 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd); 473 474 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist); 475 nvlist_free(args); 476 return (error); 477} 478 479/* 480 * Release "user holds" on snapshots. If the snapshot has been marked for 481 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have 482 * any clones, and all the user holds are removed, then the snapshot will be 483 * destroyed. 484 * 485 * The keys in the nvlist are snapshot names. 486 * The snapshots must all be in the same pool. 487 * The value is a nvlist whose keys are the holds to remove. 488 * 489 * Holds which failed to release because they didn't exist will have an entry 490 * added to errlist, but will not cause an overall failure. 491 * 492 * The return value will be 0 if the nvl holds was empty or all holds that 493 * existed, were successfully removed. 494 * 495 * Otherwise the return value will be the errno of a (unspecified) hold that 496 * failed to release and no holds will be released. 497 * 498 * In all cases the errlist will have an entry for each hold that failed to 499 * to release. 500 */ 501int 502lzc_release(nvlist_t *holds, nvlist_t **errlist) 503{ 504 char pool[ZFS_MAX_DATASET_NAME_LEN]; 505 nvpair_t *elem; 506 507 /* determine the pool name */ 508 elem = nvlist_next_nvpair(holds, NULL); 509 if (elem == NULL) 510 return (0); 511 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 512 pool[strcspn(pool, "/@")] = '\0'; 513 514 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist)); 515} 516 517/* 518 * Retrieve list of user holds on the specified snapshot. 519 * 520 * On success, *holdsp will be set to a nvlist which the caller must free. 521 * The keys are the names of the holds, and the value is the creation time 522 * of the hold (uint64) in seconds since the epoch. 523 */ 524int 525lzc_get_holds(const char *snapname, nvlist_t **holdsp) 526{ 527 int error; 528 nvlist_t *innvl = fnvlist_alloc(); 529 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp); 530 fnvlist_free(innvl); 531 return (error); 532} 533 534/* 535 * Generate a zfs send stream for the specified snapshot and write it to 536 * the specified file descriptor. 537 * 538 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap") 539 * 540 * If "from" is NULL, a full (non-incremental) stream will be sent. 541 * If "from" is non-NULL, it must be the full name of a snapshot or 542 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or 543 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or 544 * bookmark must represent an earlier point in the history of "snapname"). 545 * It can be an earlier snapshot in the same filesystem or zvol as "snapname", 546 * or it can be the origin of "snapname"'s filesystem, or an earlier 547 * snapshot in the origin, etc. 548 * 549 * "fd" is the file descriptor to write the send stream to. 550 * 551 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted 552 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT 553 * records with drr_blksz > 128K. 554 * 555 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted 556 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA, 557 * which the receiving system must support (as indicated by support 558 * for the "embedded_data" feature). 559 */ 560int 561lzc_send(const char *snapname, const char *from, int fd, 562 enum lzc_send_flags flags) 563{ 564 return (lzc_send_resume(snapname, from, fd, flags, 0, 0)); 565} 566 567int 568lzc_send_resume(const char *snapname, const char *from, int fd, 569 enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff) 570{ 571 nvlist_t *args; 572 int err; 573 574 args = fnvlist_alloc(); 575 fnvlist_add_int32(args, "fd", fd); 576 if (from != NULL) 577 fnvlist_add_string(args, "fromsnap", from); 578 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 579 fnvlist_add_boolean(args, "largeblockok"); 580 if (flags & LZC_SEND_FLAG_EMBED_DATA) 581 fnvlist_add_boolean(args, "embedok"); 582 if (flags & LZC_SEND_FLAG_COMPRESS) 583 fnvlist_add_boolean(args, "compressok"); 584 if (resumeobj != 0 || resumeoff != 0) { 585 fnvlist_add_uint64(args, "resume_object", resumeobj); 586 fnvlist_add_uint64(args, "resume_offset", resumeoff); 587 } 588 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL); 589 nvlist_free(args); 590 return (err); 591} 592 593/* 594 * "from" can be NULL, a snapshot, or a bookmark. 595 * 596 * If from is NULL, a full (non-incremental) stream will be estimated. This 597 * is calculated very efficiently. 598 * 599 * If from is a snapshot, lzc_send_space uses the deadlists attached to 600 * each snapshot to efficiently estimate the stream size. 601 * 602 * If from is a bookmark, the indirect blocks in the destination snapshot 603 * are traversed, looking for blocks with a birth time since the creation TXG of 604 * the snapshot this bookmark was created from. This will result in 605 * significantly more I/O and be less efficient than a send space estimation on 606 * an equivalent snapshot. 607 */ 608int 609lzc_send_space(const char *snapname, const char *from, 610 enum lzc_send_flags flags, uint64_t *spacep) 611{ 612 nvlist_t *args; 613 nvlist_t *result; 614 int err; 615 616 args = fnvlist_alloc(); 617 if (from != NULL) 618 fnvlist_add_string(args, "from", from); 619 if (flags & LZC_SEND_FLAG_LARGE_BLOCK) 620 fnvlist_add_boolean(args, "largeblockok"); 621 if (flags & LZC_SEND_FLAG_EMBED_DATA) 622 fnvlist_add_boolean(args, "embedok"); 623 if (flags & LZC_SEND_FLAG_COMPRESS) 624 fnvlist_add_boolean(args, "compressok"); 625 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result); 626 nvlist_free(args); 627 if (err == 0) 628 *spacep = fnvlist_lookup_uint64(result, "space"); 629 nvlist_free(result); 630 return (err); 631} 632 633static int 634recv_read(int fd, void *buf, int ilen) 635{ 636 char *cp = buf; 637 int rv; 638 int len = ilen; 639 640 do { 641 rv = read(fd, cp, len); 642 cp += rv; 643 len -= rv; 644 } while (rv > 0); 645 646 if (rv < 0 || len != 0) 647 return (EIO); 648 649 return (0); 650} 651 652static int 653recv_impl(const char *snapname, nvlist_t *props, const char *origin, 654 boolean_t force, boolean_t resumable, int fd, 655 const dmu_replay_record_t *begin_record) 656{ 657 /* 658 * The receive ioctl is still legacy, so we need to construct our own 659 * zfs_cmd_t rather than using zfsc_ioctl(). 660 */ 661 zfs_cmd_t zc = { 0 }; 662 char *atp; 663 char *packed = NULL; 664 size_t size; 665 int error; 666 667 ASSERT3S(g_refcount, >, 0); 668 VERIFY3S(g_fd, !=, -1); 669 670 /* zc_name is name of containing filesystem */ 671 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name)); 672 atp = strchr(zc.zc_name, '@'); 673 if (atp == NULL) 674 return (EINVAL); 675 *atp = '\0'; 676 677 /* if the fs does not exist, try its parent. */ 678 if (!lzc_exists(zc.zc_name)) { 679 char *slashp = strrchr(zc.zc_name, '/'); 680 if (slashp == NULL) 681 return (ENOENT); 682 *slashp = '\0'; 683 684 } 685 686 /* zc_value is full name of the snapshot to create */ 687 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value)); 688 689 if (props != NULL) { 690 /* zc_nvlist_src is props to set */ 691 packed = fnvlist_pack(props, &size); 692 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed; 693 zc.zc_nvlist_src_size = size; 694 } 695 696 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */ 697 if (origin != NULL) 698 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string)); 699 700 /* zc_begin_record is non-byteswapped BEGIN record */ 701 if (begin_record == NULL) { 702 error = recv_read(fd, &zc.zc_begin_record, 703 sizeof (zc.zc_begin_record)); 704 if (error != 0) 705 goto out; 706 } else { 707 zc.zc_begin_record = *begin_record; 708 } 709 710 /* zc_cookie is fd to read from */ 711 zc.zc_cookie = fd; 712 713 /* zc guid is force flag */ 714 zc.zc_guid = force; 715 716 zc.zc_resumable = resumable; 717 718 /* zc_cleanup_fd is unused */ 719 zc.zc_cleanup_fd = -1; 720 721 error = ioctl(g_fd, ZFS_IOC_RECV, &zc); 722 if (error != 0) 723 error = errno; 724 725out: 726 if (packed != NULL) 727 fnvlist_pack_free(packed, size); 728 free((void*)(uintptr_t)zc.zc_nvlist_dst); 729 return (error); 730} 731 732/* 733 * The simplest receive case: receive from the specified fd, creating the 734 * specified snapshot. Apply the specified properties as "received" properties 735 * (which can be overridden by locally-set properties). If the stream is a 736 * clone, its origin snapshot must be specified by 'origin'. The 'force' 737 * flag will cause the target filesystem to be rolled back or destroyed if 738 * necessary to receive. 739 * 740 * Return 0 on success or an errno on failure. 741 * 742 * Note: this interface does not work on dedup'd streams 743 * (those with DMU_BACKUP_FEATURE_DEDUP). 744 */ 745int 746lzc_receive(const char *snapname, nvlist_t *props, const char *origin, 747 boolean_t force, int fd) 748{ 749 return (recv_impl(snapname, props, origin, force, B_FALSE, fd, NULL)); 750} 751 752/* 753 * Like lzc_receive, but if the receive fails due to premature stream 754 * termination, the intermediate state will be preserved on disk. In this 755 * case, ECKSUM will be returned. The receive may subsequently be resumed 756 * with a resuming send stream generated by lzc_send_resume(). 757 */ 758int 759lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin, 760 boolean_t force, int fd) 761{ 762 return (recv_impl(snapname, props, origin, force, B_TRUE, fd, NULL)); 763} 764 765/* 766 * Like lzc_receive, but allows the caller to read the begin record and then to 767 * pass it in. That could be useful if the caller wants to derive, for example, 768 * the snapname or the origin parameters based on the information contained in 769 * the begin record. 770 * The begin record must be in its original form as read from the stream, 771 * in other words, it should not be byteswapped. 772 * 773 * The 'resumable' parameter allows to obtain the same behavior as with 774 * lzc_receive_resumable. 775 */ 776int 777lzc_receive_with_header(const char *snapname, nvlist_t *props, 778 const char *origin, boolean_t force, boolean_t resumable, int fd, 779 const dmu_replay_record_t *begin_record) 780{ 781 if (begin_record == NULL) 782 return (EINVAL); 783 return (recv_impl(snapname, props, origin, force, resumable, fd, 784 begin_record)); 785} 786 787/* 788 * Roll back this filesystem or volume to its most recent snapshot. 789 * If snapnamebuf is not NULL, it will be filled in with the name 790 * of the most recent snapshot. 791 * Note that the latest snapshot may change if a new one is concurrently 792 * created or the current one is destroyed. lzc_rollback_to can be used 793 * to roll back to a specific latest snapshot. 794 * 795 * Return 0 on success or an errno on failure. 796 */ 797int 798lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen) 799{ 800 nvlist_t *args; 801 nvlist_t *result; 802 int err; 803 804 args = fnvlist_alloc(); 805 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 806 nvlist_free(args); 807 if (err == 0 && snapnamebuf != NULL) { 808 const char *snapname = fnvlist_lookup_string(result, "target"); 809 (void) strlcpy(snapnamebuf, snapname, snapnamelen); 810 } 811 nvlist_free(result); 812 813 return (err); 814} 815 816/* 817 * Roll back this filesystem or volume to the specified snapshot, 818 * if possible. 819 * 820 * Return 0 on success or an errno on failure. 821 */ 822int 823lzc_rollback_to(const char *fsname, const char *snapname) 824{ 825 nvlist_t *args; 826 nvlist_t *result; 827 int err; 828 829 args = fnvlist_alloc(); 830 fnvlist_add_string(args, "target", snapname); 831 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result); 832 nvlist_free(args); 833 nvlist_free(result); 834 return (err); 835} 836 837/* 838 * Creates bookmarks. 839 * 840 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to 841 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and 842 * snapshots must be in the same pool. 843 * 844 * The returned results nvlist will have an entry for each bookmark that failed. 845 * The value will be the (int32) error code. 846 * 847 * The return value will be 0 if all bookmarks were created, otherwise it will 848 * be the errno of a (undetermined) bookmarks that failed. 849 */ 850int 851lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist) 852{ 853 nvpair_t *elem; 854 int error; 855 char pool[ZFS_MAX_DATASET_NAME_LEN]; 856 857 /* determine the pool name */ 858 elem = nvlist_next_nvpair(bookmarks, NULL); 859 if (elem == NULL) 860 return (0); 861 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 862 pool[strcspn(pool, "/#")] = '\0'; 863 864 error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist); 865 866 return (error); 867} 868 869/* 870 * Retrieve bookmarks. 871 * 872 * Retrieve the list of bookmarks for the given file system. The props 873 * parameter is an nvlist of property names (with no values) that will be 874 * returned for each bookmark. 875 * 876 * The following are valid properties on bookmarks, all of which are numbers 877 * (represented as uint64 in the nvlist) 878 * 879 * "guid" - globally unique identifier of the snapshot it refers to 880 * "createtxg" - txg when the snapshot it refers to was created 881 * "creation" - timestamp when the snapshot it refers to was created 882 * 883 * The format of the returned nvlist as follows: 884 * <short name of bookmark> -> { 885 * <name of property> -> { 886 * "value" -> uint64 887 * } 888 * } 889 */ 890int 891lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks) 892{ 893 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks)); 894} 895 896/* 897 * Destroys bookmarks. 898 * 899 * The keys in the bmarks nvlist are the bookmarks to be destroyed. 900 * They must all be in the same pool. Bookmarks are specified as 901 * <fs>#<bmark>. 902 * 903 * Bookmarks that do not exist will be silently ignored. 904 * 905 * The return value will be 0 if all bookmarks that existed were destroyed. 906 * 907 * Otherwise the return value will be the errno of a (undetermined) bookmark 908 * that failed, no bookmarks will be destroyed, and the errlist will have an 909 * entry for each bookmarks that failed. The value in the errlist will be 910 * the (int32) error code. 911 */ 912int 913lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist) 914{ 915 nvpair_t *elem; 916 int error; 917 char pool[ZFS_MAX_DATASET_NAME_LEN]; 918 919 /* determine the pool name */ 920 elem = nvlist_next_nvpair(bmarks, NULL); 921 if (elem == NULL) 922 return (0); 923 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool)); 924 pool[strcspn(pool, "/#")] = '\0'; 925 926 error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist); 927 928 return (error); 929} 930 931static int 932lzc_channel_program_impl(const char *pool, const char *program, boolean_t sync, 933 uint64_t instrlimit, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 934{ 935 int error; 936 nvlist_t *args; 937 938 args = fnvlist_alloc(); 939 fnvlist_add_string(args, ZCP_ARG_PROGRAM, program); 940 fnvlist_add_nvlist(args, ZCP_ARG_ARGLIST, argnvl); 941 fnvlist_add_boolean_value(args, ZCP_ARG_SYNC, sync); 942 fnvlist_add_uint64(args, ZCP_ARG_INSTRLIMIT, instrlimit); 943 fnvlist_add_uint64(args, ZCP_ARG_MEMLIMIT, memlimit); 944 error = lzc_ioctl(ZFS_IOC_CHANNEL_PROGRAM, pool, args, outnvl); 945 fnvlist_free(args); 946 947 return (error); 948} 949 950/* 951 * Executes a channel program. 952 * 953 * If this function returns 0 the channel program was successfully loaded and 954 * ran without failing. Note that individual commands the channel program ran 955 * may have failed and the channel program is responsible for reporting such 956 * errors through outnvl if they are important. 957 * 958 * This method may also return: 959 * 960 * EINVAL The program contains syntax errors, or an invalid memory or time 961 * limit was given. No part of the channel program was executed. 962 * If caused by syntax errors, 'outnvl' contains information about the 963 * errors. 964 * 965 * EDOM The program was executed, but encountered a runtime error, such as 966 * calling a function with incorrect arguments, invoking the error() 967 * function directly, failing an assert() command, etc. Some portion 968 * of the channel program may have executed and committed changes. 969 * Information about the failure can be found in 'outnvl'. 970 * 971 * ENOMEM The program fully executed, but the output buffer was not large 972 * enough to store the returned value. No output is returned through 973 * 'outnvl'. 974 * 975 * ENOSPC The program was terminated because it exceeded its memory usage 976 * limit. Some portion of the channel program may have executed and 977 * committed changes to disk. No output is returned through 'outnvl'. 978 * 979 * ETIMEDOUT The program was terminated because it exceeded its Lua instruction 980 * limit. Some portion of the channel program may have executed and 981 * committed changes to disk. No output is returned through 'outnvl'. 982 */ 983int 984lzc_channel_program(const char *pool, const char *program, uint64_t instrlimit, 985 uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 986{ 987 return (lzc_channel_program_impl(pool, program, B_TRUE, instrlimit, 988 memlimit, argnvl, outnvl)); 989} 990 991/* 992 * Executes a read-only channel program. 993 * 994 * A read-only channel program works programmatically the same way as a 995 * normal channel program executed with lzc_channel_program(). The only 996 * difference is it runs exclusively in open-context and therefore can 997 * return faster. The downside to that, is that the program cannot change 998 * on-disk state by calling functions from the zfs.sync submodule. 999 * 1000 * The return values of this function (and their meaning) are exactly the 1001 * same as the ones described in lzc_channel_program(). 1002 */ 1003int 1004lzc_channel_program_nosync(const char *pool, const char *program, 1005 uint64_t timeout, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl) 1006{ 1007 return (lzc_channel_program_impl(pool, program, B_FALSE, timeout, 1008 memlimit, argnvl, outnvl)); 1009} 1010