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, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 * Copyright (c) 2017 Datto Inc.
26 * Copyright 2017 RackTop Systems.
27 * Copyright (c) 2017 Open-E, Inc. All Rights Reserved.
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 * - Programmatic 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 #include <libzfs_core.h>
84 #include <sys/nvpair.h>
85 #include <sys/param.h>
86 #include <sys/types.h>
88 #include <sys/zfs_ioctl.h>
91 static pthread_mutex_t g_lock
= PTHREAD_MUTEX_INITIALIZER
;
92 static int g_refcount
;
95 libzfs_core_init(void)
97 (void) pthread_mutex_lock(&g_lock
);
98 if (g_refcount
== 0) {
99 g_fd
= open("/dev/zfs", O_RDWR
);
101 (void) pthread_mutex_unlock(&g_lock
);
106 (void) pthread_mutex_unlock(&g_lock
);
111 libzfs_core_fini(void)
113 (void) pthread_mutex_lock(&g_lock
);
114 ASSERT3S(g_refcount
, >, 0);
119 if (g_refcount
== 0 && g_fd
!= -1) {
123 (void) pthread_mutex_unlock(&g_lock
);
127 lzc_ioctl(zfs_ioc_t ioc
, const char *name
,
128 nvlist_t
*source
, nvlist_t
**resultp
)
130 zfs_cmd_t zc
= {"\0"};
135 ASSERT3S(g_refcount
, >, 0);
136 VERIFY3S(g_fd
, !=, -1);
139 (void) strlcpy(zc
.zc_name
, name
, sizeof (zc
.zc_name
));
141 if (source
!= NULL
) {
142 packed
= fnvlist_pack(source
, &size
);
143 zc
.zc_nvlist_src
= (uint64_t)(uintptr_t)packed
;
144 zc
.zc_nvlist_src_size
= size
;
147 if (resultp
!= NULL
) {
149 if (ioc
== ZFS_IOC_CHANNEL_PROGRAM
) {
150 zc
.zc_nvlist_dst_size
= fnvlist_lookup_uint64(source
,
153 zc
.zc_nvlist_dst_size
= MAX(size
* 2, 128 * 1024);
155 zc
.zc_nvlist_dst
= (uint64_t)(uintptr_t)
156 malloc(zc
.zc_nvlist_dst_size
);
157 if (zc
.zc_nvlist_dst
== (uint64_t)0) {
163 while (ioctl(g_fd
, ioc
, &zc
) != 0) {
165 * If ioctl exited with ENOMEM, we retry the ioctl after
166 * increasing the size of the destination nvlist.
168 * Channel programs that exit with ENOMEM ran over the
169 * lua memory sandbox; they should not be retried.
171 if (errno
== ENOMEM
&& resultp
!= NULL
&&
172 ioc
!= ZFS_IOC_CHANNEL_PROGRAM
) {
173 free((void *)(uintptr_t)zc
.zc_nvlist_dst
);
174 zc
.zc_nvlist_dst_size
*= 2;
175 zc
.zc_nvlist_dst
= (uint64_t)(uintptr_t)
176 malloc(zc
.zc_nvlist_dst_size
);
177 if (zc
.zc_nvlist_dst
== (uint64_t)0) {
186 if (zc
.zc_nvlist_dst_filled
) {
187 *resultp
= fnvlist_unpack((void *)(uintptr_t)zc
.zc_nvlist_dst
,
188 zc
.zc_nvlist_dst_size
);
193 fnvlist_pack_free(packed
, size
);
194 free((void *)(uintptr_t)zc
.zc_nvlist_dst
);
199 lzc_create(const char *fsname
, enum lzc_dataset_type type
, nvlist_t
*props
,
200 uint8_t *wkeydata
, uint_t wkeylen
)
203 nvlist_t
*hidden_args
= NULL
;
204 nvlist_t
*args
= fnvlist_alloc();
206 fnvlist_add_int32(args
, "type", (dmu_objset_type_t
)type
);
208 fnvlist_add_nvlist(args
, "props", props
);
210 if (wkeydata
!= NULL
) {
211 hidden_args
= fnvlist_alloc();
212 fnvlist_add_uint8_array(hidden_args
, "wkeydata", wkeydata
,
214 fnvlist_add_nvlist(args
, ZPOOL_HIDDEN_ARGS
, hidden_args
);
217 error
= lzc_ioctl(ZFS_IOC_CREATE
, fsname
, args
, NULL
);
218 nvlist_free(hidden_args
);
224 lzc_clone(const char *fsname
, const char *origin
, nvlist_t
*props
)
227 nvlist_t
*hidden_args
= NULL
;
228 nvlist_t
*args
= fnvlist_alloc();
230 fnvlist_add_string(args
, "origin", origin
);
232 fnvlist_add_nvlist(args
, "props", props
);
233 error
= lzc_ioctl(ZFS_IOC_CLONE
, fsname
, args
, NULL
);
234 nvlist_free(hidden_args
);
240 lzc_promote(const char *fsname
, char *snapnamebuf
, int snapnamelen
)
243 * The promote ioctl is still legacy, so we need to construct our
244 * own zfs_cmd_t rather than using lzc_ioctl().
246 zfs_cmd_t zc
= { "\0" };
248 ASSERT3S(g_refcount
, >, 0);
249 VERIFY3S(g_fd
, !=, -1);
251 (void) strlcpy(zc
.zc_name
, fsname
, sizeof (zc
.zc_name
));
252 if (ioctl(g_fd
, ZFS_IOC_PROMOTE
, &zc
) != 0) {
254 if (error
== EEXIST
&& snapnamebuf
!= NULL
)
255 (void) strlcpy(snapnamebuf
, zc
.zc_string
, snapnamelen
);
262 lzc_remap(const char *fsname
)
265 nvlist_t
*args
= fnvlist_alloc();
266 error
= lzc_ioctl(ZFS_IOC_REMAP
, fsname
, args
, NULL
);
274 * The keys in the snaps nvlist are the snapshots to be created.
275 * They must all be in the same pool.
277 * The props nvlist is properties to set. Currently only user properties
278 * are supported. { user:prop_name -> string value }
280 * The returned results nvlist will have an entry for each snapshot that failed.
281 * The value will be the (int32) error code.
283 * The return value will be 0 if all snapshots were created, otherwise it will
284 * be the errno of a (unspecified) snapshot that failed.
287 lzc_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
**errlist
)
292 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
296 /* determine the pool name */
297 elem
= nvlist_next_nvpair(snaps
, NULL
);
300 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
301 pool
[strcspn(pool
, "/@")] = '\0';
303 args
= fnvlist_alloc();
304 fnvlist_add_nvlist(args
, "snaps", snaps
);
306 fnvlist_add_nvlist(args
, "props", props
);
308 error
= lzc_ioctl(ZFS_IOC_SNAPSHOT
, pool
, args
, errlist
);
315 * Destroys snapshots.
317 * The keys in the snaps nvlist are the snapshots to be destroyed.
318 * They must all be in the same pool.
320 * Snapshots that do not exist will be silently ignored.
322 * If 'defer' is not set, and a snapshot has user holds or clones, the
323 * destroy operation will fail and none of the snapshots will be
326 * If 'defer' is set, and a snapshot has user holds or clones, it will be
327 * marked for deferred destruction, and will be destroyed when the last hold
328 * or clone is removed/destroyed.
330 * The return value will be 0 if all snapshots were destroyed (or marked for
331 * later destruction if 'defer' is set) or didn't exist to begin with.
333 * Otherwise the return value will be the errno of a (unspecified) snapshot
334 * that failed, no snapshots will be destroyed, and the errlist will have an
335 * entry for each snapshot that failed. The value in the errlist will be
336 * the (int32) error code.
339 lzc_destroy_snaps(nvlist_t
*snaps
, boolean_t defer
, nvlist_t
**errlist
)
344 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
346 /* determine the pool name */
347 elem
= nvlist_next_nvpair(snaps
, NULL
);
350 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
351 pool
[strcspn(pool
, "/@")] = '\0';
353 args
= fnvlist_alloc();
354 fnvlist_add_nvlist(args
, "snaps", snaps
);
356 fnvlist_add_boolean(args
, "defer");
358 error
= lzc_ioctl(ZFS_IOC_DESTROY_SNAPS
, pool
, args
, errlist
);
365 lzc_snaprange_space(const char *firstsnap
, const char *lastsnap
,
371 char fs
[ZFS_MAX_DATASET_NAME_LEN
];
374 /* determine the fs name */
375 (void) strlcpy(fs
, firstsnap
, sizeof (fs
));
376 atp
= strchr(fs
, '@');
381 args
= fnvlist_alloc();
382 fnvlist_add_string(args
, "firstsnap", firstsnap
);
384 err
= lzc_ioctl(ZFS_IOC_SPACE_SNAPS
, lastsnap
, args
, &result
);
387 *usedp
= fnvlist_lookup_uint64(result
, "used");
388 fnvlist_free(result
);
394 lzc_exists(const char *dataset
)
397 * The objset_stats ioctl is still legacy, so we need to construct our
398 * own zfs_cmd_t rather than using lzc_ioctl().
400 zfs_cmd_t zc
= {"\0"};
402 ASSERT3S(g_refcount
, >, 0);
403 VERIFY3S(g_fd
, !=, -1);
405 (void) strlcpy(zc
.zc_name
, dataset
, sizeof (zc
.zc_name
));
406 return (ioctl(g_fd
, ZFS_IOC_OBJSET_STATS
, &zc
) == 0);
411 * It was added to preserve the function signature in case it is
412 * needed in the future.
416 lzc_sync(const char *pool_name
, nvlist_t
*innvl
, nvlist_t
**outnvl
)
418 return (lzc_ioctl(ZFS_IOC_POOL_SYNC
, pool_name
, innvl
, NULL
));
422 * Create "user holds" on snapshots. If there is a hold on a snapshot,
423 * the snapshot can not be destroyed. (However, it can be marked for deletion
424 * by lzc_destroy_snaps(defer=B_TRUE).)
426 * The keys in the nvlist are snapshot names.
427 * The snapshots must all be in the same pool.
428 * The value is the name of the hold (string type).
430 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
431 * In this case, when the cleanup_fd is closed (including on process
432 * termination), the holds will be released. If the system is shut down
433 * uncleanly, the holds will be released when the pool is next opened
436 * Holds for snapshots which don't exist will be skipped and have an entry
437 * added to errlist, but will not cause an overall failure.
439 * The return value will be 0 if all holds, for snapshots that existed,
440 * were successfully created.
442 * Otherwise the return value will be the errno of a (unspecified) hold that
443 * failed and no holds will be created.
445 * In all cases the errlist will have an entry for each hold that failed
446 * (name = snapshot), with its value being the error code (int32).
449 lzc_hold(nvlist_t
*holds
, int cleanup_fd
, nvlist_t
**errlist
)
451 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
456 /* determine the pool name */
457 elem
= nvlist_next_nvpair(holds
, NULL
);
460 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
461 pool
[strcspn(pool
, "/@")] = '\0';
463 args
= fnvlist_alloc();
464 fnvlist_add_nvlist(args
, "holds", holds
);
465 if (cleanup_fd
!= -1)
466 fnvlist_add_int32(args
, "cleanup_fd", cleanup_fd
);
468 error
= lzc_ioctl(ZFS_IOC_HOLD
, pool
, args
, errlist
);
474 * Release "user holds" on snapshots. If the snapshot has been marked for
475 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
476 * any clones, and all the user holds are removed, then the snapshot will be
479 * The keys in the nvlist are snapshot names.
480 * The snapshots must all be in the same pool.
481 * The value is an nvlist whose keys are the holds to remove.
483 * Holds which failed to release because they didn't exist will have an entry
484 * added to errlist, but will not cause an overall failure.
486 * The return value will be 0 if the nvl holds was empty or all holds that
487 * existed, were successfully removed.
489 * Otherwise the return value will be the errno of a (unspecified) hold that
490 * failed to release and no holds will be released.
492 * In all cases the errlist will have an entry for each hold that failed to
496 lzc_release(nvlist_t
*holds
, nvlist_t
**errlist
)
498 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
501 /* determine the pool name */
502 elem
= nvlist_next_nvpair(holds
, NULL
);
505 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
506 pool
[strcspn(pool
, "/@")] = '\0';
508 return (lzc_ioctl(ZFS_IOC_RELEASE
, pool
, holds
, errlist
));
512 * Retrieve list of user holds on the specified snapshot.
514 * On success, *holdsp will be set to an nvlist which the caller must free.
515 * The keys are the names of the holds, and the value is the creation time
516 * of the hold (uint64) in seconds since the epoch.
519 lzc_get_holds(const char *snapname
, nvlist_t
**holdsp
)
521 return (lzc_ioctl(ZFS_IOC_GET_HOLDS
, snapname
, NULL
, holdsp
));
525 * Generate a zfs send stream for the specified snapshot and write it to
526 * the specified file descriptor.
528 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap")
530 * If "from" is NULL, a full (non-incremental) stream will be sent.
531 * If "from" is non-NULL, it must be the full name of a snapshot or
532 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or
533 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or
534 * bookmark must represent an earlier point in the history of "snapname").
535 * It can be an earlier snapshot in the same filesystem or zvol as "snapname",
536 * or it can be the origin of "snapname"'s filesystem, or an earlier
537 * snapshot in the origin, etc.
539 * "fd" is the file descriptor to write the send stream to.
541 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted
542 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT
543 * records with drr_blksz > 128K.
545 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted
546 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA,
547 * which the receiving system must support (as indicated by support
548 * for the "embedded_data" feature).
550 * If "flags" contains LZC_SEND_FLAG_COMPRESS, the stream is generated by using
551 * compressed WRITE records for blocks which are compressed on disk and in
552 * memory. If the lz4_compress feature is active on the sending system, then
553 * the receiving system must have that feature enabled as well.
555 * If "flags" contains LZC_SEND_FLAG_RAW, the stream is generated, for encrypted
556 * datasets, by sending data exactly as it exists on disk. This allows backups
557 * to be taken even if encryption keys are not currently loaded.
560 lzc_send(const char *snapname
, const char *from
, int fd
,
561 enum lzc_send_flags flags
)
563 return (lzc_send_resume(snapname
, from
, fd
, flags
, 0, 0));
567 lzc_send_resume(const char *snapname
, const char *from
, int fd
,
568 enum lzc_send_flags flags
, uint64_t resumeobj
, uint64_t resumeoff
)
573 args
= fnvlist_alloc();
574 fnvlist_add_int32(args
, "fd", fd
);
576 fnvlist_add_string(args
, "fromsnap", from
);
577 if (flags
& LZC_SEND_FLAG_LARGE_BLOCK
)
578 fnvlist_add_boolean(args
, "largeblockok");
579 if (flags
& LZC_SEND_FLAG_EMBED_DATA
)
580 fnvlist_add_boolean(args
, "embedok");
581 if (flags
& LZC_SEND_FLAG_COMPRESS
)
582 fnvlist_add_boolean(args
, "compressok");
583 if (flags
& LZC_SEND_FLAG_RAW
)
584 fnvlist_add_boolean(args
, "rawok");
585 if (resumeobj
!= 0 || resumeoff
!= 0) {
586 fnvlist_add_uint64(args
, "resume_object", resumeobj
);
587 fnvlist_add_uint64(args
, "resume_offset", resumeoff
);
589 err
= lzc_ioctl(ZFS_IOC_SEND_NEW
, snapname
, args
, NULL
);
595 * "from" can be NULL, a snapshot, or a bookmark.
597 * If from is NULL, a full (non-incremental) stream will be estimated. This
598 * is calculated very efficiently.
600 * If from is a snapshot, lzc_send_space uses the deadlists attached to
601 * each snapshot to efficiently estimate the stream size.
603 * If from is a bookmark, the indirect blocks in the destination snapshot
604 * are traversed, looking for blocks with a birth time since the creation TXG of
605 * the snapshot this bookmark was created from. This will result in
606 * significantly more I/O and be less efficient than a send space estimation on
607 * an equivalent snapshot.
610 lzc_send_space(const char *snapname
, const char *from
,
611 enum lzc_send_flags flags
, uint64_t *spacep
)
617 args
= fnvlist_alloc();
619 fnvlist_add_string(args
, "from", from
);
620 if (flags
& LZC_SEND_FLAG_LARGE_BLOCK
)
621 fnvlist_add_boolean(args
, "largeblockok");
622 if (flags
& LZC_SEND_FLAG_EMBED_DATA
)
623 fnvlist_add_boolean(args
, "embedok");
624 if (flags
& LZC_SEND_FLAG_COMPRESS
)
625 fnvlist_add_boolean(args
, "compressok");
626 if (flags
& LZC_SEND_FLAG_RAW
)
627 fnvlist_add_boolean(args
, "rawok");
628 err
= lzc_ioctl(ZFS_IOC_SEND_SPACE
, snapname
, args
, &result
);
631 *spacep
= fnvlist_lookup_uint64(result
, "space");
637 recv_read(int fd
, void *buf
, int ilen
)
644 rv
= read(fd
, cp
, len
);
649 if (rv
< 0 || len
!= 0)
656 * Linux adds ZFS_IOC_RECV_NEW for resumable and raw streams and preserves the
657 * legacy ZFS_IOC_RECV user/kernel interface. The new interface supports all
658 * stream options but is currently only used for resumable streams. This way
659 * updated user space utilities will interoperate with older kernel modules.
661 * Non-Linux OpenZFS platforms have opted to modify the legacy interface.
664 recv_impl(const char *snapname
, nvlist_t
*recvdprops
, nvlist_t
*localprops
,
665 uint8_t *wkeydata
, uint_t wkeylen
, const char *origin
, boolean_t force
,
666 boolean_t resumable
, boolean_t raw
, int input_fd
,
667 const dmu_replay_record_t
*begin_record
, int cleanup_fd
,
668 uint64_t *read_bytes
, uint64_t *errflags
, uint64_t *action_handle
,
671 dmu_replay_record_t drr
;
672 char fsname
[MAXPATHLEN
];
676 ASSERT3S(g_refcount
, >, 0);
677 VERIFY3S(g_fd
, !=, -1);
679 /* Set 'fsname' to the name of containing filesystem */
680 (void) strlcpy(fsname
, snapname
, sizeof (fsname
));
681 atp
= strchr(fsname
, '@');
686 /* If the fs does not exist, try its parent. */
687 if (!lzc_exists(fsname
)) {
688 char *slashp
= strrchr(fsname
, '/');
695 * The begin_record is normally a non-byteswapped BEGIN record.
696 * For resumable streams it may be set to any non-byteswapped
697 * dmu_replay_record_t.
699 if (begin_record
== NULL
) {
700 error
= recv_read(input_fd
, &drr
, sizeof (drr
));
708 * Raw receives, resumable receives, and receives that include a
709 * wrapping key all use the new interface.
711 if (resumable
|| raw
|| wkeydata
!= NULL
) {
712 nvlist_t
*outnvl
= NULL
;
713 nvlist_t
*innvl
= fnvlist_alloc();
715 fnvlist_add_string(innvl
, "snapname", snapname
);
717 if (recvdprops
!= NULL
)
718 fnvlist_add_nvlist(innvl
, "props", recvdprops
);
720 if (localprops
!= NULL
)
721 fnvlist_add_nvlist(innvl
, "localprops", localprops
);
723 if (wkeydata
!= NULL
) {
725 * wkeydata must be placed in the special
726 * ZPOOL_HIDDEN_ARGS nvlist so that it
727 * will not be printed to the zpool history.
729 nvlist_t
*hidden_args
= fnvlist_alloc();
730 fnvlist_add_uint8_array(hidden_args
, "wkeydata",
732 fnvlist_add_nvlist(innvl
, ZPOOL_HIDDEN_ARGS
,
734 nvlist_free(hidden_args
);
737 if (origin
!= NULL
&& strlen(origin
))
738 fnvlist_add_string(innvl
, "origin", origin
);
740 fnvlist_add_byte_array(innvl
, "begin_record",
741 (uchar_t
*)&drr
, sizeof (drr
));
743 fnvlist_add_int32(innvl
, "input_fd", input_fd
);
746 fnvlist_add_boolean(innvl
, "force");
749 fnvlist_add_boolean(innvl
, "resumable");
752 fnvlist_add_int32(innvl
, "cleanup_fd", cleanup_fd
);
754 if (action_handle
!= NULL
)
755 fnvlist_add_uint64(innvl
, "action_handle",
758 error
= lzc_ioctl(ZFS_IOC_RECV_NEW
, fsname
, innvl
, &outnvl
);
760 if (error
== 0 && read_bytes
!= NULL
)
761 error
= nvlist_lookup_uint64(outnvl
, "read_bytes",
764 if (error
== 0 && errflags
!= NULL
)
765 error
= nvlist_lookup_uint64(outnvl
, "error_flags",
768 if (error
== 0 && action_handle
!= NULL
)
769 error
= nvlist_lookup_uint64(outnvl
, "action_handle",
772 if (error
== 0 && errors
!= NULL
) {
774 error
= nvlist_lookup_nvlist(outnvl
, "errors", &nvl
);
776 *errors
= fnvlist_dup(nvl
);
780 fnvlist_free(outnvl
);
782 zfs_cmd_t zc
= {"\0"};
786 ASSERT3S(g_refcount
, >, 0);
788 (void) strlcpy(zc
.zc_name
, fsname
, sizeof (zc
.zc_value
));
789 (void) strlcpy(zc
.zc_value
, snapname
, sizeof (zc
.zc_value
));
791 if (recvdprops
!= NULL
) {
792 packed
= fnvlist_pack(recvdprops
, &size
);
793 zc
.zc_nvlist_src
= (uint64_t)(uintptr_t)packed
;
794 zc
.zc_nvlist_src_size
= size
;
797 if (localprops
!= NULL
) {
798 packed
= fnvlist_pack(localprops
, &size
);
799 zc
.zc_nvlist_conf
= (uint64_t)(uintptr_t)packed
;
800 zc
.zc_nvlist_conf_size
= size
;
804 (void) strlcpy(zc
.zc_string
, origin
,
805 sizeof (zc
.zc_string
));
807 ASSERT3S(drr
.drr_type
, ==, DRR_BEGIN
);
808 zc
.zc_begin_record
= drr
.drr_u
.drr_begin
;
810 zc
.zc_cookie
= input_fd
;
811 zc
.zc_cleanup_fd
= -1;
812 zc
.zc_action_handle
= 0;
815 zc
.zc_cleanup_fd
= cleanup_fd
;
817 if (action_handle
!= NULL
)
818 zc
.zc_action_handle
= *action_handle
;
820 zc
.zc_nvlist_dst_size
= 128 * 1024;
821 zc
.zc_nvlist_dst
= (uint64_t)(uintptr_t)
822 malloc(zc
.zc_nvlist_dst_size
);
824 error
= ioctl(g_fd
, ZFS_IOC_RECV
, &zc
);
828 if (read_bytes
!= NULL
)
829 *read_bytes
= zc
.zc_cookie
;
831 if (errflags
!= NULL
)
832 *errflags
= zc
.zc_obj
;
834 if (action_handle
!= NULL
)
835 *action_handle
= zc
.zc_action_handle
;
838 VERIFY0(nvlist_unpack(
839 (void *)(uintptr_t)zc
.zc_nvlist_dst
,
840 zc
.zc_nvlist_dst_size
, errors
, KM_SLEEP
));
844 fnvlist_pack_free(packed
, size
);
845 free((void *)(uintptr_t)zc
.zc_nvlist_dst
);
852 * The simplest receive case: receive from the specified fd, creating the
853 * specified snapshot. Apply the specified properties as "received" properties
854 * (which can be overridden by locally-set properties). If the stream is a
855 * clone, its origin snapshot must be specified by 'origin'. The 'force'
856 * flag will cause the target filesystem to be rolled back or destroyed if
857 * necessary to receive.
859 * Return 0 on success or an errno on failure.
861 * Note: this interface does not work on dedup'd streams
862 * (those with DMU_BACKUP_FEATURE_DEDUP).
865 lzc_receive(const char *snapname
, nvlist_t
*props
, const char *origin
,
866 boolean_t force
, boolean_t raw
, int fd
)
868 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
869 B_FALSE
, raw
, fd
, NULL
, -1, NULL
, NULL
, NULL
, NULL
));
873 * Like lzc_receive, but if the receive fails due to premature stream
874 * termination, the intermediate state will be preserved on disk. In this
875 * case, ECKSUM will be returned. The receive may subsequently be resumed
876 * with a resuming send stream generated by lzc_send_resume().
879 lzc_receive_resumable(const char *snapname
, nvlist_t
*props
, const char *origin
,
880 boolean_t force
, boolean_t raw
, int fd
)
882 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
883 B_TRUE
, raw
, fd
, NULL
, -1, NULL
, NULL
, NULL
, NULL
));
887 * Like lzc_receive, but allows the caller to read the begin record and then to
888 * pass it in. That could be useful if the caller wants to derive, for example,
889 * the snapname or the origin parameters based on the information contained in
891 * The begin record must be in its original form as read from the stream,
892 * in other words, it should not be byteswapped.
894 * The 'resumable' parameter allows to obtain the same behavior as with
895 * lzc_receive_resumable.
898 lzc_receive_with_header(const char *snapname
, nvlist_t
*props
,
899 const char *origin
, boolean_t force
, boolean_t resumable
, boolean_t raw
,
900 int fd
, const dmu_replay_record_t
*begin_record
)
902 if (begin_record
== NULL
)
905 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
906 resumable
, raw
, fd
, begin_record
, -1, NULL
, NULL
, NULL
, NULL
));
910 * Like lzc_receive, but allows the caller to pass all supported arguments
911 * and retrieve all values returned. The only additional input parameter
912 * is 'cleanup_fd' which is used to set a cleanup-on-exit file descriptor.
914 * The following parameters all provide return values. Several may be set
915 * in the failure case and will contain additional information.
917 * The 'read_bytes' value will be set to the total number of bytes read.
919 * The 'errflags' value will contain zprop_errflags_t flags which are
920 * used to describe any failures.
922 * The 'action_handle' is used to pass the handle for this guid/ds mapping.
923 * It should be set to zero on first call and will contain an updated handle
924 * on success, it should be passed in subsequent calls.
926 * The 'errors' nvlist contains an entry for each unapplied received
927 * property. Callers are responsible for freeing this nvlist.
929 int lzc_receive_one(const char *snapname
, nvlist_t
*props
,
930 const char *origin
, boolean_t force
, boolean_t resumable
, boolean_t raw
,
931 int input_fd
, const dmu_replay_record_t
*begin_record
, int cleanup_fd
,
932 uint64_t *read_bytes
, uint64_t *errflags
, uint64_t *action_handle
,
935 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
936 resumable
, raw
, input_fd
, begin_record
, cleanup_fd
, read_bytes
,
937 errflags
, action_handle
, errors
));
941 * Like lzc_receive_one, but allows the caller to pass an additional 'cmdprops'
944 * The 'cmdprops' nvlist contains both override ('zfs receive -o') and
945 * exclude ('zfs receive -x') properties. Callers are responsible for freeing
948 int lzc_receive_with_cmdprops(const char *snapname
, nvlist_t
*props
,
949 nvlist_t
*cmdprops
, uint8_t *wkeydata
, uint_t wkeylen
, const char *origin
,
950 boolean_t force
, boolean_t resumable
, boolean_t raw
, int input_fd
,
951 const dmu_replay_record_t
*begin_record
, int cleanup_fd
,
952 uint64_t *read_bytes
, uint64_t *errflags
, uint64_t *action_handle
,
955 return (recv_impl(snapname
, props
, cmdprops
, wkeydata
, wkeylen
, origin
,
956 force
, resumable
, raw
, input_fd
, begin_record
, cleanup_fd
,
957 read_bytes
, errflags
, action_handle
, errors
));
961 * Roll back this filesystem or volume to its most recent snapshot.
962 * If snapnamebuf is not NULL, it will be filled in with the name
963 * of the most recent snapshot.
964 * Note that the latest snapshot may change if a new one is concurrently
965 * created or the current one is destroyed. lzc_rollback_to can be used
966 * to roll back to a specific latest snapshot.
968 * Return 0 on success or an errno on failure.
971 lzc_rollback(const char *fsname
, char *snapnamebuf
, int snapnamelen
)
977 args
= fnvlist_alloc();
978 err
= lzc_ioctl(ZFS_IOC_ROLLBACK
, fsname
, args
, &result
);
980 if (err
== 0 && snapnamebuf
!= NULL
) {
981 const char *snapname
= fnvlist_lookup_string(result
, "target");
982 (void) strlcpy(snapnamebuf
, snapname
, snapnamelen
);
990 * Roll back this filesystem or volume to the specified snapshot,
993 * Return 0 on success or an errno on failure.
996 lzc_rollback_to(const char *fsname
, const char *snapname
)
1002 args
= fnvlist_alloc();
1003 fnvlist_add_string(args
, "target", snapname
);
1004 err
= lzc_ioctl(ZFS_IOC_ROLLBACK
, fsname
, args
, &result
);
1006 nvlist_free(result
);
1011 * Creates bookmarks.
1013 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to
1014 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and
1015 * snapshots must be in the same pool.
1017 * The returned results nvlist will have an entry for each bookmark that failed.
1018 * The value will be the (int32) error code.
1020 * The return value will be 0 if all bookmarks were created, otherwise it will
1021 * be the errno of a (undetermined) bookmarks that failed.
1024 lzc_bookmark(nvlist_t
*bookmarks
, nvlist_t
**errlist
)
1028 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
1030 /* determine the pool name */
1031 elem
= nvlist_next_nvpair(bookmarks
, NULL
);
1034 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
1035 pool
[strcspn(pool
, "/#")] = '\0';
1037 error
= lzc_ioctl(ZFS_IOC_BOOKMARK
, pool
, bookmarks
, errlist
);
1043 * Retrieve bookmarks.
1045 * Retrieve the list of bookmarks for the given file system. The props
1046 * parameter is an nvlist of property names (with no values) that will be
1047 * returned for each bookmark.
1049 * The following are valid properties on bookmarks, all of which are numbers
1050 * (represented as uint64 in the nvlist)
1052 * "guid" - globally unique identifier of the snapshot it refers to
1053 * "createtxg" - txg when the snapshot it refers to was created
1054 * "creation" - timestamp when the snapshot it refers to was created
1056 * The format of the returned nvlist as follows:
1057 * <short name of bookmark> -> {
1058 * <name of property> -> {
1064 lzc_get_bookmarks(const char *fsname
, nvlist_t
*props
, nvlist_t
**bmarks
)
1066 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS
, fsname
, props
, bmarks
));
1070 * Destroys bookmarks.
1072 * The keys in the bmarks nvlist are the bookmarks to be destroyed.
1073 * They must all be in the same pool. Bookmarks are specified as
1076 * Bookmarks that do not exist will be silently ignored.
1078 * The return value will be 0 if all bookmarks that existed were destroyed.
1080 * Otherwise the return value will be the errno of a (undetermined) bookmark
1081 * that failed, no bookmarks will be destroyed, and the errlist will have an
1082 * entry for each bookmarks that failed. The value in the errlist will be
1083 * the (int32) error code.
1086 lzc_destroy_bookmarks(nvlist_t
*bmarks
, nvlist_t
**errlist
)
1090 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
1092 /* determine the pool name */
1093 elem
= nvlist_next_nvpair(bmarks
, NULL
);
1096 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
1097 pool
[strcspn(pool
, "/#")] = '\0';
1099 error
= lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS
, pool
, bmarks
, errlist
);
1105 lzc_channel_program_impl(const char *pool
, const char *program
, boolean_t sync
,
1106 uint64_t instrlimit
, uint64_t memlimit
, nvlist_t
*argnvl
, nvlist_t
**outnvl
)
1111 args
= fnvlist_alloc();
1112 fnvlist_add_string(args
, ZCP_ARG_PROGRAM
, program
);
1113 fnvlist_add_nvlist(args
, ZCP_ARG_ARGLIST
, argnvl
);
1114 fnvlist_add_boolean_value(args
, ZCP_ARG_SYNC
, sync
);
1115 fnvlist_add_uint64(args
, ZCP_ARG_INSTRLIMIT
, instrlimit
);
1116 fnvlist_add_uint64(args
, ZCP_ARG_MEMLIMIT
, memlimit
);
1117 error
= lzc_ioctl(ZFS_IOC_CHANNEL_PROGRAM
, pool
, args
, outnvl
);
1124 * Executes a channel program.
1126 * If this function returns 0 the channel program was successfully loaded and
1127 * ran without failing. Note that individual commands the channel program ran
1128 * may have failed and the channel program is responsible for reporting such
1129 * errors through outnvl if they are important.
1131 * This method may also return:
1133 * EINVAL The program contains syntax errors, or an invalid memory or time
1134 * limit was given. No part of the channel program was executed.
1135 * If caused by syntax errors, 'outnvl' contains information about the
1138 * ECHRNG The program was executed, but encountered a runtime error, such as
1139 * calling a function with incorrect arguments, invoking the error()
1140 * function directly, failing an assert() command, etc. Some portion
1141 * of the channel program may have executed and committed changes.
1142 * Information about the failure can be found in 'outnvl'.
1144 * ENOMEM The program fully executed, but the output buffer was not large
1145 * enough to store the returned value. No output is returned through
1148 * ENOSPC The program was terminated because it exceeded its memory usage
1149 * limit. Some portion of the channel program may have executed and
1150 * committed changes to disk. No output is returned through 'outnvl'.
1152 * ETIME The program was terminated because it exceeded its Lua instruction
1153 * limit. Some portion of the channel program may have executed and
1154 * committed changes to disk. No output is returned through 'outnvl'.
1157 lzc_channel_program(const char *pool
, const char *program
, uint64_t instrlimit
,
1158 uint64_t memlimit
, nvlist_t
*argnvl
, nvlist_t
**outnvl
)
1160 return (lzc_channel_program_impl(pool
, program
, B_TRUE
, instrlimit
,
1161 memlimit
, argnvl
, outnvl
));
1165 * Creates a checkpoint for the specified pool.
1167 * If this function returns 0 the pool was successfully checkpointed.
1169 * This method may also return:
1171 * ZFS_ERR_CHECKPOINT_EXISTS
1172 * The pool already has a checkpoint. A pools can only have one
1173 * checkpoint at most, at any given time.
1175 * ZFS_ERR_DISCARDING_CHECKPOINT
1176 * ZFS is in the middle of discarding a checkpoint for this pool.
1177 * The pool can be checkpointed again once the discard is done.
1179 * ZFS_DEVRM_IN_PROGRESS
1180 * A vdev is currently being removed. The pool cannot be
1181 * checkpointed until the device removal is done.
1184 * One or more top-level vdevs exceed the maximum vdev size
1185 * supported for this feature.
1188 lzc_pool_checkpoint(const char *pool
)
1192 nvlist_t
*result
= NULL
;
1193 nvlist_t
*args
= fnvlist_alloc();
1195 error
= lzc_ioctl(ZFS_IOC_POOL_CHECKPOINT
, pool
, args
, &result
);
1198 fnvlist_free(result
);
1204 * Discard the checkpoint from the specified pool.
1206 * If this function returns 0 the checkpoint was successfully discarded.
1208 * This method may also return:
1210 * ZFS_ERR_NO_CHECKPOINT
1211 * The pool does not have a checkpoint.
1213 * ZFS_ERR_DISCARDING_CHECKPOINT
1214 * ZFS is already in the middle of discarding the checkpoint.
1217 lzc_pool_checkpoint_discard(const char *pool
)
1221 nvlist_t
*result
= NULL
;
1222 nvlist_t
*args
= fnvlist_alloc();
1224 error
= lzc_ioctl(ZFS_IOC_POOL_DISCARD_CHECKPOINT
, pool
, args
, &result
);
1227 fnvlist_free(result
);
1233 * Executes a read-only channel program.
1235 * A read-only channel program works programmatically the same way as a
1236 * normal channel program executed with lzc_channel_program(). The only
1237 * difference is it runs exclusively in open-context and therefore can
1238 * return faster. The downside to that, is that the program cannot change
1239 * on-disk state by calling functions from the zfs.sync submodule.
1241 * The return values of this function (and their meaning) are exactly the
1242 * same as the ones described in lzc_channel_program().
1245 lzc_channel_program_nosync(const char *pool
, const char *program
,
1246 uint64_t timeout
, uint64_t memlimit
, nvlist_t
*argnvl
, nvlist_t
**outnvl
)
1248 return (lzc_channel_program_impl(pool
, program
, B_FALSE
, timeout
,
1249 memlimit
, argnvl
, outnvl
));
1253 * Performs key management functions
1255 * crypto_cmd should be a value from dcp_cmd_t. If the command specifies to
1256 * load or change a wrapping key, the key should be specified in the
1257 * hidden_args nvlist so that it is not logged.
1260 lzc_load_key(const char *fsname
, boolean_t noop
, uint8_t *wkeydata
,
1265 nvlist_t
*hidden_args
;
1267 if (wkeydata
== NULL
)
1270 ioc_args
= fnvlist_alloc();
1271 hidden_args
= fnvlist_alloc();
1272 fnvlist_add_uint8_array(hidden_args
, "wkeydata", wkeydata
, wkeylen
);
1273 fnvlist_add_nvlist(ioc_args
, ZPOOL_HIDDEN_ARGS
, hidden_args
);
1275 fnvlist_add_boolean(ioc_args
, "noop");
1276 error
= lzc_ioctl(ZFS_IOC_LOAD_KEY
, fsname
, ioc_args
, NULL
);
1277 nvlist_free(hidden_args
);
1278 nvlist_free(ioc_args
);
1284 lzc_unload_key(const char *fsname
)
1286 return (lzc_ioctl(ZFS_IOC_UNLOAD_KEY
, fsname
, NULL
, NULL
));
1290 lzc_change_key(const char *fsname
, uint64_t crypt_cmd
, nvlist_t
*props
,
1291 uint8_t *wkeydata
, uint_t wkeylen
)
1294 nvlist_t
*ioc_args
= fnvlist_alloc();
1295 nvlist_t
*hidden_args
= NULL
;
1297 fnvlist_add_uint64(ioc_args
, "crypt_cmd", crypt_cmd
);
1299 if (wkeydata
!= NULL
) {
1300 hidden_args
= fnvlist_alloc();
1301 fnvlist_add_uint8_array(hidden_args
, "wkeydata", wkeydata
,
1303 fnvlist_add_nvlist(ioc_args
, ZPOOL_HIDDEN_ARGS
, hidden_args
);
1307 fnvlist_add_nvlist(ioc_args
, "props", props
);
1309 error
= lzc_ioctl(ZFS_IOC_CHANGE_KEY
, fsname
, ioc_args
, NULL
);
1310 nvlist_free(hidden_args
);
1311 nvlist_free(ioc_args
);
1317 lzc_reopen(const char *pool_name
, boolean_t scrub_restart
)
1319 nvlist_t
*args
= fnvlist_alloc();
1322 fnvlist_add_boolean_value(args
, "scrub_restart", scrub_restart
);
1324 error
= lzc_ioctl(ZFS_IOC_POOL_REOPEN
, pool_name
, args
, NULL
);