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.
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.
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]
23 * Copyright (c) 2012, 2020 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 * Copyright (c) 2017 Datto Inc.
31 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs.
32 * It has the following characteristics:
34 * - Thread Safe. libzfs_core is accessible concurrently from multiple
35 * threads. This is accomplished primarily by avoiding global data
36 * (e.g. caching). Since it's thread-safe, there is no reason for a
37 * process to have multiple libzfs "instances". Therefore, we store
38 * our few pieces of data (e.g. the file descriptor) in global
39 * variables. The fd is reference-counted so that the libzfs_core
40 * library can be "initialized" multiple times (e.g. by different
41 * consumers within the same process).
43 * - Committed Interface. The libzfs_core interface will be committed,
44 * therefore consumers can compile against it and be confident that
45 * their code will continue to work on future releases of this code.
46 * Currently, the interface is Evolving (not Committed), but we intend
47 * to commit to it once it is more complete and we determine that it
48 * meets the needs of all consumers.
50 * - Programatic Error Handling. libzfs_core communicates errors with
51 * defined error numbers, and doesn't print anything to stdout/stderr.
53 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments
54 * to/from the kernel ioctls. There is generally a 1:1 correspondence
55 * between libzfs_core functions and ioctls to /dev/zfs.
57 * - Clear Atomicity. Because libzfs_core functions are generally 1:1
58 * with kernel ioctls, and kernel ioctls are general atomic, each
59 * libzfs_core function is atomic. For example, creating multiple
60 * snapshots with a single call to lzc_snapshot() is atomic -- it
61 * can't fail with only some of the requested snapshots created, even
62 * in the event of power loss or system crash.
64 * - Continued libzfs Support. Some higher-level operations (e.g.
65 * support for "zfs send -R") are too complicated to fit the scope of
66 * libzfs_core. This functionality will continue to live in libzfs.
67 * Where appropriate, libzfs will use the underlying atomic operations
68 * of libzfs_core. For example, libzfs may implement "zfs send -R |
69 * zfs receive" by using individual "send one snapshot", rename,
70 * destroy, and "receive one snapshot" operations in libzfs_core.
71 * /sbin/zfs and /zbin/zpool will link with both libzfs and
72 * libzfs_core. Other consumers should aim to use only libzfs_core,
73 * since that will be the supported, stable interface going forwards.
76 #define _IN_LIBZFS_CORE_
78 #include <libzfs_core.h>
89 #include <sys/nvpair.h>
90 #include <sys/param.h>
91 #include <sys/types.h>
93 #include <sys/zfs_ioctl.h>
94 #include "libzfs_core_compat.h"
95 #include "libzfs_compat.h"
98 extern int zfs_ioctl_version;
101 static int g_fd = -1;
102 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER;
103 static int g_refcount;
106 static zfs_ioc_t fail_ioc_cmd;
107 static zfs_errno_t fail_ioc_err;
110 libzfs_core_debug_ioc(void)
113 * To test running newer user space binaries with kernel's
114 * that don't yet support an ioctl or a new ioctl arg we
115 * provide an override to intentionally fail an ioctl.
118 * The override variable, ZFS_IOC_TEST, is of the form "cmd:err"
120 * For example, to fail a ZFS_IOC_POOL_CHECKPOINT with a
121 * ZFS_ERR_IOC_CMD_UNAVAIL, the string would be "0x5a4d:1029"
123 * $ sudo sh -c "ZFS_IOC_TEST=0x5a4d:1029 zpool checkpoint tank"
124 * cannot checkpoint 'tank': the loaded zfs module does not support
125 * this operation. A reboot may be required to enable this operation.
127 if (fail_ioc_cmd == 0) {
128 char *ioc_test = getenv("ZFS_IOC_TEST");
129 unsigned int ioc_num = 0, ioc_err = 0;
131 if (ioc_test != NULL &&
132 sscanf(ioc_test, "%i:%i", &ioc_num, &ioc_err) == 2 &&
133 ioc_num < ZFS_IOC_LAST) {
134 fail_ioc_cmd = ioc_num;
135 fail_ioc_err = ioc_err;
142 libzfs_core_init(void)
144 (void) pthread_mutex_lock(&g_lock);
145 if (g_refcount == 0) {
146 g_fd = open("/dev/zfs", O_RDWR);
148 (void) pthread_mutex_unlock(&g_lock);
155 libzfs_core_debug_ioc();
157 (void) pthread_mutex_unlock(&g_lock);
163 libzfs_core_fini(void)
165 (void) pthread_mutex_lock(&g_lock);
166 ASSERT3S(g_refcount, >, 0);
171 if (g_refcount == 0 && g_fd != -1) {
175 (void) pthread_mutex_unlock(&g_lock);
179 lzc_ioctl(zfs_ioc_t ioc, const char *name,
180 nvlist_t *source, nvlist_t **resultp)
182 zfs_cmd_t zc = { 0 };
190 ASSERT3S(g_refcount, >, 0);
191 VERIFY3S(g_fd, !=, -1);
194 if (ioc == fail_ioc_cmd)
195 return (fail_ioc_err);
199 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
202 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
203 zfs_ioctl_version = get_zfs_ioctl_version();
205 if (zfs_ioctl_version < ZFS_IOCVER_LZC) {
207 error = lzc_compat_pre(&zc, &ioc, &source);
213 if (source != NULL) {
214 packed = fnvlist_pack(source, &size);
215 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
216 zc.zc_nvlist_src_size = size;
219 if (resultp != NULL) {
221 if (ioc == ZFS_IOC_CHANNEL_PROGRAM) {
222 zc.zc_nvlist_dst_size = fnvlist_lookup_uint64(source,
225 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
227 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
228 malloc(zc.zc_nvlist_dst_size);
230 if (zc.zc_nvlist_dst == NULL) {
232 if (zc.zc_nvlist_dst == 0) {
239 while (ioctl(g_fd, ioc, &zc) != 0) {
241 * If ioctl exited with ENOMEM, we retry the ioctl after
242 * increasing the size of the destination nvlist.
244 * Channel programs that exit with ENOMEM ran over the
245 * lua memory sandbox; they should not be retried.
247 if (errno == ENOMEM && resultp != NULL &&
248 ioc != ZFS_IOC_CHANNEL_PROGRAM) {
249 free((void *)(uintptr_t)zc.zc_nvlist_dst);
250 zc.zc_nvlist_dst_size *= 2;
251 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
252 malloc(zc.zc_nvlist_dst_size);
254 if (zc.zc_nvlist_dst == NULL) {
256 if (zc.zc_nvlist_dst == 0) {
268 if (zfs_ioctl_version < ZFS_IOCVER_LZC)
269 lzc_compat_post(&zc, ioc);
271 if (zc.zc_nvlist_dst_filled) {
272 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
273 zc.zc_nvlist_dst_size);
276 if (zfs_ioctl_version < ZFS_IOCVER_LZC)
277 lzc_compat_outnvl(&zc, ioc, resultp);
281 if (zfs_ioctl_version < ZFS_IOCVER_LZC) {
282 if (source != oldsource)
287 fnvlist_pack_free(packed, size);
288 free((void *)(uintptr_t)zc.zc_nvlist_dst);
293 lzc_create(const char *fsname, enum lzc_dataset_type type, nvlist_t *props)
296 nvlist_t *args = fnvlist_alloc();
297 fnvlist_add_int32(args, "type", (dmu_objset_type_t)type);
299 fnvlist_add_nvlist(args, "props", props);
300 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
306 lzc_clone(const char *fsname, const char *origin,
310 nvlist_t *args = fnvlist_alloc();
311 fnvlist_add_string(args, "origin", origin);
313 fnvlist_add_nvlist(args, "props", props);
314 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
320 lzc_promote(const char *fsname, char *snapnamebuf, int snapnamelen)
323 * The promote ioctl is still legacy, so we need to construct our
324 * own zfs_cmd_t rather than using lzc_ioctl().
326 zfs_cmd_t zc = { 0 };
328 ASSERT3S(g_refcount, >, 0);
329 VERIFY3S(g_fd, !=, -1);
331 (void) strlcpy(zc.zc_name, fsname, sizeof (zc.zc_name));
332 if (ioctl(g_fd, ZFS_IOC_PROMOTE, &zc) != 0) {
334 if (error == EEXIST && snapnamebuf != NULL)
335 (void) strlcpy(snapnamebuf, zc.zc_string, snapnamelen);
342 lzc_remap(const char *fsname)
345 nvlist_t *args = fnvlist_alloc();
346 error = lzc_ioctl(ZFS_IOC_REMAP, fsname, args, NULL);
352 lzc_rename(const char *source, const char *target)
354 zfs_cmd_t zc = { 0 };
357 ASSERT3S(g_refcount, >, 0);
358 VERIFY3S(g_fd, !=, -1);
360 (void) strlcpy(zc.zc_name, source, sizeof (zc.zc_name));
361 (void) strlcpy(zc.zc_value, target, sizeof (zc.zc_value));
362 error = ioctl(g_fd, ZFS_IOC_RENAME, &zc);
369 lzc_destroy(const char *fsname)
373 nvlist_t *args = fnvlist_alloc();
374 error = lzc_ioctl(ZFS_IOC_DESTROY, fsname, args, NULL);
382 * The keys in the snaps nvlist are the snapshots to be created.
383 * They must all be in the same pool.
385 * The props nvlist is properties to set. Currently only user properties
386 * are supported. { user:prop_name -> string value }
388 * The returned results nvlist will have an entry for each snapshot that failed.
389 * The value will be the (int32) error code.
391 * The return value will be 0 if all snapshots were created, otherwise it will
392 * be the errno of a (unspecified) snapshot that failed.
395 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
400 char pool[ZFS_MAX_DATASET_NAME_LEN];
404 /* determine the pool name */
405 elem = nvlist_next_nvpair(snaps, NULL);
408 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
409 pool[strcspn(pool, "/@")] = '\0';
411 args = fnvlist_alloc();
412 fnvlist_add_nvlist(args, "snaps", snaps);
414 fnvlist_add_nvlist(args, "props", props);
416 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
423 * Destroys snapshots.
425 * The keys in the snaps nvlist are the snapshots to be destroyed.
426 * They must all be in the same pool.
428 * Snapshots that do not exist will be silently ignored.
430 * If 'defer' is not set, and a snapshot has user holds or clones, the
431 * destroy operation will fail and none of the snapshots will be
434 * If 'defer' is set, and a snapshot has user holds or clones, it will be
435 * marked for deferred destruction, and will be destroyed when the last hold
436 * or clone is removed/destroyed.
438 * The return value will be 0 if all snapshots were destroyed (or marked for
439 * later destruction if 'defer' is set) or didn't exist to begin with.
441 * Otherwise the return value will be the errno of a (unspecified) snapshot
442 * that failed, no snapshots will be destroyed, and the errlist will have an
443 * entry for each snapshot that failed. The value in the errlist will be
444 * the (int32) error code.
447 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
452 char pool[ZFS_MAX_DATASET_NAME_LEN];
454 /* determine the pool name */
455 elem = nvlist_next_nvpair(snaps, NULL);
458 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
459 pool[strcspn(pool, "/@")] = '\0';
461 args = fnvlist_alloc();
462 fnvlist_add_nvlist(args, "snaps", snaps);
464 fnvlist_add_boolean(args, "defer");
466 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
473 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
479 char fs[ZFS_MAX_DATASET_NAME_LEN];
482 /* determine the fs name */
483 (void) strlcpy(fs, firstsnap, sizeof (fs));
484 atp = strchr(fs, '@');
489 args = fnvlist_alloc();
490 fnvlist_add_string(args, "firstsnap", firstsnap);
492 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
495 *usedp = fnvlist_lookup_uint64(result, "used");
496 fnvlist_free(result);
502 lzc_exists(const char *dataset)
505 * The objset_stats ioctl is still legacy, so we need to construct our
506 * own zfs_cmd_t rather than using lzc_ioctl().
508 zfs_cmd_t zc = { 0 };
510 ASSERT3S(g_refcount, >, 0);
511 VERIFY3S(g_fd, !=, -1);
513 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
514 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
519 * It was added to preserve the function signature in case it is
520 * needed in the future.
524 lzc_sync(const char *pool_name, nvlist_t *innvl, nvlist_t **outnvl)
526 return (lzc_ioctl(ZFS_IOC_POOL_SYNC, pool_name, innvl, NULL));
530 * Create "user holds" on snapshots. If there is a hold on a snapshot,
531 * the snapshot can not be destroyed. (However, it can be marked for deletion
532 * by lzc_destroy_snaps(defer=B_TRUE).)
534 * The keys in the nvlist are snapshot names.
535 * The snapshots must all be in the same pool.
536 * The value is the name of the hold (string type).
538 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
539 * In this case, when the cleanup_fd is closed (including on process
540 * termination), the holds will be released. If the system is shut down
541 * uncleanly, the holds will be released when the pool is next opened
544 * Holds for snapshots which don't exist will be skipped and have an entry
545 * added to errlist, but will not cause an overall failure.
547 * The return value will be 0 if all holds, for snapshots that existed,
548 * were succesfully created.
550 * Otherwise the return value will be the errno of a (unspecified) hold that
551 * failed and no holds will be created.
553 * In all cases the errlist will have an entry for each hold that failed
554 * (name = snapshot), with its value being the error code (int32).
557 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
559 char pool[ZFS_MAX_DATASET_NAME_LEN];
564 /* determine the pool name */
565 elem = nvlist_next_nvpair(holds, NULL);
568 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
569 pool[strcspn(pool, "/@")] = '\0';
571 args = fnvlist_alloc();
572 fnvlist_add_nvlist(args, "holds", holds);
573 if (cleanup_fd != -1)
574 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
576 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
582 * Release "user holds" on snapshots. If the snapshot has been marked for
583 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
584 * any clones, and all the user holds are removed, then the snapshot will be
587 * The keys in the nvlist are snapshot names.
588 * The snapshots must all be in the same pool.
589 * The value is a nvlist whose keys are the holds to remove.
591 * Holds which failed to release because they didn't exist will have an entry
592 * added to errlist, but will not cause an overall failure.
594 * The return value will be 0 if the nvl holds was empty or all holds that
595 * existed, were successfully removed.
597 * Otherwise the return value will be the errno of a (unspecified) hold that
598 * failed to release and no holds will be released.
600 * In all cases the errlist will have an entry for each hold that failed to
604 lzc_release(nvlist_t *holds, nvlist_t **errlist)
606 char pool[ZFS_MAX_DATASET_NAME_LEN];
609 /* determine the pool name */
610 elem = nvlist_next_nvpair(holds, NULL);
613 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
614 pool[strcspn(pool, "/@")] = '\0';
616 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
620 * Retrieve list of user holds on the specified snapshot.
622 * On success, *holdsp will be set to a nvlist which the caller must free.
623 * The keys are the names of the holds, and the value is the creation time
624 * of the hold (uint64) in seconds since the epoch.
627 lzc_get_holds(const char *snapname, nvlist_t **holdsp)
629 return (lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, NULL, holdsp));
633 * Generate a zfs send stream for the specified snapshot and write it to
634 * the specified file descriptor.
636 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap")
638 * If "from" is NULL, a full (non-incremental) stream will be sent.
639 * If "from" is non-NULL, it must be the full name of a snapshot or
640 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or
641 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or
642 * bookmark must represent an earlier point in the history of "snapname").
643 * It can be an earlier snapshot in the same filesystem or zvol as "snapname",
644 * or it can be the origin of "snapname"'s filesystem, or an earlier
645 * snapshot in the origin, etc.
647 * "fd" is the file descriptor to write the send stream to.
649 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted
650 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT
651 * records with drr_blksz > 128K.
653 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted
654 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA,
655 * which the receiving system must support (as indicated by support
656 * for the "embedded_data" feature).
659 lzc_send(const char *snapname, const char *from, int fd,
660 enum lzc_send_flags flags)
662 return (lzc_send_resume(snapname, from, fd, flags, 0, 0));
666 lzc_send_resume(const char *snapname, const char *from, int fd,
667 enum lzc_send_flags flags, uint64_t resumeobj, uint64_t resumeoff)
672 args = fnvlist_alloc();
673 fnvlist_add_int32(args, "fd", fd);
675 fnvlist_add_string(args, "fromsnap", from);
676 if (flags & LZC_SEND_FLAG_LARGE_BLOCK)
677 fnvlist_add_boolean(args, "largeblockok");
678 if (flags & LZC_SEND_FLAG_EMBED_DATA)
679 fnvlist_add_boolean(args, "embedok");
680 if (flags & LZC_SEND_FLAG_COMPRESS)
681 fnvlist_add_boolean(args, "compressok");
682 if (resumeobj != 0 || resumeoff != 0) {
683 fnvlist_add_uint64(args, "resume_object", resumeobj);
684 fnvlist_add_uint64(args, "resume_offset", resumeoff);
686 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
692 * "from" can be NULL, a snapshot, or a bookmark.
694 * If from is NULL, a full (non-incremental) stream will be estimated. This
695 * is calculated very efficiently.
697 * If from is a snapshot, lzc_send_space uses the deadlists attached to
698 * each snapshot to efficiently estimate the stream size.
700 * If from is a bookmark, the indirect blocks in the destination snapshot
701 * are traversed, looking for blocks with a birth time since the creation TXG of
702 * the snapshot this bookmark was created from. This will result in
703 * significantly more I/O and be less efficient than a send space estimation on
704 * an equivalent snapshot.
707 lzc_send_space(const char *snapname, const char *from,
708 enum lzc_send_flags flags, uint64_t *spacep)
714 args = fnvlist_alloc();
716 fnvlist_add_string(args, "from", from);
717 if (flags & LZC_SEND_FLAG_LARGE_BLOCK)
718 fnvlist_add_boolean(args, "largeblockok");
719 if (flags & LZC_SEND_FLAG_EMBED_DATA)
720 fnvlist_add_boolean(args, "embedok");
721 if (flags & LZC_SEND_FLAG_COMPRESS)
722 fnvlist_add_boolean(args, "compressok");
723 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
726 *spacep = fnvlist_lookup_uint64(result, "space");
732 recv_read(int fd, void *buf, int ilen)
739 rv = read(fd, cp, len);
744 if (rv < 0 || len != 0)
751 recv_impl(const char *snapname, nvlist_t *props, const char *origin,
752 boolean_t force, boolean_t resumable, int fd,
753 const dmu_replay_record_t *begin_record)
756 * The receive ioctl is still legacy, so we need to construct our own
757 * zfs_cmd_t rather than using zfsc_ioctl().
759 zfs_cmd_t zc = { 0 };
765 ASSERT3S(g_refcount, >, 0);
766 VERIFY3S(g_fd, !=, -1);
768 /* zc_name is name of containing filesystem */
769 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
770 atp = strchr(zc.zc_name, '@');
775 /* if the fs does not exist, try its parent. */
776 if (!lzc_exists(zc.zc_name)) {
777 char *slashp = strrchr(zc.zc_name, '/');
784 /* zc_value is full name of the snapshot to create */
785 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
788 /* zc_nvlist_src is props to set */
789 packed = fnvlist_pack(props, &size);
790 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
791 zc.zc_nvlist_src_size = size;
794 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
796 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
798 /* zc_begin_record is non-byteswapped BEGIN record */
799 if (begin_record == NULL) {
800 error = recv_read(fd, &zc.zc_begin_record,
801 sizeof (zc.zc_begin_record));
805 zc.zc_begin_record = *begin_record;
808 /* zc_cookie is fd to read from */
811 /* zc guid is force flag */
814 zc.zc_resumable = resumable;
816 /* zc_cleanup_fd is unused */
817 zc.zc_cleanup_fd = -1;
819 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
825 fnvlist_pack_free(packed, size);
826 free((void*)(uintptr_t)zc.zc_nvlist_dst);
831 * The simplest receive case: receive from the specified fd, creating the
832 * specified snapshot. Apply the specified properties as "received" properties
833 * (which can be overridden by locally-set properties). If the stream is a
834 * clone, its origin snapshot must be specified by 'origin'. The 'force'
835 * flag will cause the target filesystem to be rolled back or destroyed if
836 * necessary to receive.
838 * Return 0 on success or an errno on failure.
840 * Note: this interface does not work on dedup'd streams
841 * (those with DMU_BACKUP_FEATURE_DEDUP).
844 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
845 boolean_t force, int fd)
847 return (recv_impl(snapname, props, origin, force, B_FALSE, fd, NULL));
851 * Like lzc_receive, but if the receive fails due to premature stream
852 * termination, the intermediate state will be preserved on disk. In this
853 * case, ECKSUM will be returned. The receive may subsequently be resumed
854 * with a resuming send stream generated by lzc_send_resume().
857 lzc_receive_resumable(const char *snapname, nvlist_t *props, const char *origin,
858 boolean_t force, int fd)
860 return (recv_impl(snapname, props, origin, force, B_TRUE, fd, NULL));
864 * Like lzc_receive, but allows the caller to read the begin record and then to
865 * pass it in. That could be useful if the caller wants to derive, for example,
866 * the snapname or the origin parameters based on the information contained in
868 * The begin record must be in its original form as read from the stream,
869 * in other words, it should not be byteswapped.
871 * The 'resumable' parameter allows to obtain the same behavior as with
872 * lzc_receive_resumable.
875 lzc_receive_with_header(const char *snapname, nvlist_t *props,
876 const char *origin, boolean_t force, boolean_t resumable, int fd,
877 const dmu_replay_record_t *begin_record)
879 if (begin_record == NULL)
881 return (recv_impl(snapname, props, origin, force, resumable, fd,
886 * Roll back this filesystem or volume to its most recent snapshot.
887 * If snapnamebuf is not NULL, it will be filled in with the name
888 * of the most recent snapshot.
889 * Note that the latest snapshot may change if a new one is concurrently
890 * created or the current one is destroyed. lzc_rollback_to can be used
891 * to roll back to a specific latest snapshot.
893 * Return 0 on success or an errno on failure.
896 lzc_rollback(const char *fsname, char *snapnamebuf, int snapnamelen)
902 args = fnvlist_alloc();
903 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result);
905 if (err == 0 && snapnamebuf != NULL) {
906 const char *snapname = fnvlist_lookup_string(result, "target");
907 (void) strlcpy(snapnamebuf, snapname, snapnamelen);
915 * Roll back this filesystem or volume to the specified snapshot,
918 * Return 0 on success or an errno on failure.
921 lzc_rollback_to(const char *fsname, const char *snapname)
927 args = fnvlist_alloc();
928 fnvlist_add_string(args, "target", snapname);
929 err = lzc_ioctl(ZFS_IOC_ROLLBACK, fsname, args, &result);
938 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to
939 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and
940 * snapshots must be in the same pool.
942 * The returned results nvlist will have an entry for each bookmark that failed.
943 * The value will be the (int32) error code.
945 * The return value will be 0 if all bookmarks were created, otherwise it will
946 * be the errno of a (undetermined) bookmarks that failed.
949 lzc_bookmark(nvlist_t *bookmarks, nvlist_t **errlist)
953 char pool[ZFS_MAX_DATASET_NAME_LEN];
955 /* determine the pool name */
956 elem = nvlist_next_nvpair(bookmarks, NULL);
959 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
960 pool[strcspn(pool, "/#")] = '\0';
962 error = lzc_ioctl(ZFS_IOC_BOOKMARK, pool, bookmarks, errlist);
968 * Retrieve bookmarks.
970 * Retrieve the list of bookmarks for the given file system. The props
971 * parameter is an nvlist of property names (with no values) that will be
972 * returned for each bookmark.
974 * The following are valid properties on bookmarks, all of which are numbers
975 * (represented as uint64 in the nvlist)
977 * "guid" - globally unique identifier of the snapshot it refers to
978 * "createtxg" - txg when the snapshot it refers to was created
979 * "creation" - timestamp when the snapshot it refers to was created
981 * The format of the returned nvlist as follows:
982 * <short name of bookmark> -> {
983 * <name of property> -> {
989 lzc_get_bookmarks(const char *fsname, nvlist_t *props, nvlist_t **bmarks)
991 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS, fsname, props, bmarks));
995 * Destroys bookmarks.
997 * The keys in the bmarks nvlist are the bookmarks to be destroyed.
998 * They must all be in the same pool. Bookmarks are specified as
1001 * Bookmarks that do not exist will be silently ignored.
1003 * The return value will be 0 if all bookmarks that existed were destroyed.
1005 * Otherwise the return value will be the errno of a (undetermined) bookmark
1006 * that failed, no bookmarks will be destroyed, and the errlist will have an
1007 * entry for each bookmarks that failed. The value in the errlist will be
1008 * the (int32) error code.
1011 lzc_destroy_bookmarks(nvlist_t *bmarks, nvlist_t **errlist)
1015 char pool[ZFS_MAX_DATASET_NAME_LEN];
1017 /* determine the pool name */
1018 elem = nvlist_next_nvpair(bmarks, NULL);
1021 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
1022 pool[strcspn(pool, "/#")] = '\0';
1024 error = lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS, pool, bmarks, errlist);
1030 lzc_channel_program_impl(const char *pool, const char *program, boolean_t sync,
1031 uint64_t instrlimit, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl)
1036 args = fnvlist_alloc();
1037 fnvlist_add_string(args, ZCP_ARG_PROGRAM, program);
1038 fnvlist_add_nvlist(args, ZCP_ARG_ARGLIST, argnvl);
1039 fnvlist_add_boolean_value(args, ZCP_ARG_SYNC, sync);
1040 fnvlist_add_uint64(args, ZCP_ARG_INSTRLIMIT, instrlimit);
1041 fnvlist_add_uint64(args, ZCP_ARG_MEMLIMIT, memlimit);
1042 error = lzc_ioctl(ZFS_IOC_CHANNEL_PROGRAM, pool, args, outnvl);
1049 * Executes a channel program.
1051 * If this function returns 0 the channel program was successfully loaded and
1052 * ran without failing. Note that individual commands the channel program ran
1053 * may have failed and the channel program is responsible for reporting such
1054 * errors through outnvl if they are important.
1056 * This method may also return:
1058 * EINVAL The program contains syntax errors, or an invalid memory or time
1059 * limit was given. No part of the channel program was executed.
1060 * If caused by syntax errors, 'outnvl' contains information about the
1063 * EDOM The program was executed, but encountered a runtime error, such as
1064 * calling a function with incorrect arguments, invoking the error()
1065 * function directly, failing an assert() command, etc. Some portion
1066 * of the channel program may have executed and committed changes.
1067 * Information about the failure can be found in 'outnvl'.
1069 * ENOMEM The program fully executed, but the output buffer was not large
1070 * enough to store the returned value. No output is returned through
1073 * ENOSPC The program was terminated because it exceeded its memory usage
1074 * limit. Some portion of the channel program may have executed and
1075 * committed changes to disk. No output is returned through 'outnvl'.
1077 * ETIMEDOUT The program was terminated because it exceeded its Lua instruction
1078 * limit. Some portion of the channel program may have executed and
1079 * committed changes to disk. No output is returned through 'outnvl'.
1082 lzc_channel_program(const char *pool, const char *program, uint64_t instrlimit,
1083 uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl)
1085 return (lzc_channel_program_impl(pool, program, B_TRUE, instrlimit,
1086 memlimit, argnvl, outnvl));
1090 * Creates a checkpoint for the specified pool.
1092 * If this function returns 0 the pool was successfully checkpointed.
1094 * This method may also return:
1096 * ZFS_ERR_CHECKPOINT_EXISTS
1097 * The pool already has a checkpoint. A pools can only have one
1098 * checkpoint at most, at any given time.
1100 * ZFS_ERR_DISCARDING_CHECKPOINT
1101 * ZFS is in the middle of discarding a checkpoint for this pool.
1102 * The pool can be checkpointed again once the discard is done.
1104 * ZFS_DEVRM_IN_PROGRESS
1105 * A vdev is currently being removed. The pool cannot be
1106 * checkpointed until the device removal is done.
1109 * One or more top-level vdevs exceed the maximum vdev size
1110 * supported for this feature.
1113 lzc_pool_checkpoint(const char *pool)
1117 nvlist_t *result = NULL;
1118 nvlist_t *args = fnvlist_alloc();
1120 error = lzc_ioctl(ZFS_IOC_POOL_CHECKPOINT, pool, args, &result);
1123 fnvlist_free(result);
1129 * Discard the checkpoint from the specified pool.
1131 * If this function returns 0 the checkpoint was successfully discarded.
1133 * This method may also return:
1135 * ZFS_ERR_NO_CHECKPOINT
1136 * The pool does not have a checkpoint.
1138 * ZFS_ERR_DISCARDING_CHECKPOINT
1139 * ZFS is already in the middle of discarding the checkpoint.
1142 lzc_pool_checkpoint_discard(const char *pool)
1146 nvlist_t *result = NULL;
1147 nvlist_t *args = fnvlist_alloc();
1149 error = lzc_ioctl(ZFS_IOC_POOL_DISCARD_CHECKPOINT, pool, args, &result);
1152 fnvlist_free(result);
1158 * Executes a read-only channel program.
1160 * A read-only channel program works programmatically the same way as a
1161 * normal channel program executed with lzc_channel_program(). The only
1162 * difference is it runs exclusively in open-context and therefore can
1163 * return faster. The downside to that, is that the program cannot change
1164 * on-disk state by calling functions from the zfs.sync submodule.
1166 * The return values of this function (and their meaning) are exactly the
1167 * same as the ones described in lzc_channel_program().
1170 lzc_channel_program_nosync(const char *pool, const char *program,
1171 uint64_t timeout, uint64_t memlimit, nvlist_t *argnvl, nvlist_t **outnvl)
1173 return (lzc_channel_program_impl(pool, program, B_FALSE, timeout,
1174 memlimit, argnvl, outnvl));
1178 * Changes initializing state.
1180 * vdevs should be a list of (<key>, guid) where guid is a uint64 vdev GUID.
1181 * The key is ignored.
1183 * If there are errors related to vdev arguments, per-vdev errors are returned
1184 * in an nvlist with the key "vdevs". Each error is a (guid, errno) pair where
1185 * guid is stringified with PRIu64, and errno is one of the following as
1187 * - ENODEV if the device was not found
1188 * - EINVAL if the devices is not a leaf or is not concrete (e.g. missing)
1189 * - EROFS if the device is not writeable
1190 * - EBUSY start requested but the device is already being initialized
1191 * - ESRCH cancel/suspend requested but device is not being initialized
1193 * If the errlist is empty, then return value will be:
1194 * - EINVAL if one or more arguments was invalid
1195 * - Other spa_open failures
1196 * - 0 if the operation succeeded
1199 lzc_initialize(const char *poolname, pool_initialize_func_t cmd_type,
1200 nvlist_t *vdevs, nvlist_t **errlist)
1203 nvlist_t *args = fnvlist_alloc();
1204 fnvlist_add_uint64(args, ZPOOL_INITIALIZE_COMMAND, (uint64_t)cmd_type);
1205 fnvlist_add_nvlist(args, ZPOOL_INITIALIZE_VDEVS, vdevs);
1207 error = lzc_ioctl(ZFS_IOC_POOL_INITIALIZE, poolname, args, errlist);
1215 * Set the bootenv contents for the given pool.
1218 lzc_set_bootenv(const char *pool, const char *env)
1220 nvlist_t *args = fnvlist_alloc();
1221 fnvlist_add_string(args, "envmap", env);
1222 int error = lzc_ioctl(ZFS_IOC_SET_BOOTENV, pool, args, NULL);
1228 * Get the contents of the bootenv of the given pool.
1231 lzc_get_bootenv(const char *pool, nvlist_t **outnvl)
1233 return (lzc_ioctl(ZFS_IOC_GET_BOOTENV, pool, NULL, outnvl));