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