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, 2018 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>
87 #include <sys/nvpair.h>
88 #include <sys/param.h>
89 #include <sys/types.h>
91 #include <sys/zfs_ioctl.h>
94 static pthread_mutex_t g_lock
= PTHREAD_MUTEX_INITIALIZER
;
95 static int g_refcount
;
98 static zfs_ioc_t fail_ioc_cmd
;
99 static zfs_errno_t fail_ioc_err
;
102 libzfs_core_debug_ioc(void)
105 * To test running newer user space binaries with kernel's
106 * that don't yet support an ioctl or a new ioctl arg we
107 * provide an override to intentionally fail an ioctl.
110 * The override variable, ZFS_IOC_TEST, is of the form "cmd:err"
112 * For example, to fail a ZFS_IOC_POOL_CHECKPOINT with a
113 * ZFS_ERR_IOC_CMD_UNAVAIL, the string would be "0x5a4d:1029"
115 * $ sudo sh -c "ZFS_IOC_TEST=0x5a4d:1029 zpool checkpoint tank"
116 * cannot checkpoint 'tank': the loaded zfs module does not support
117 * this operation. A reboot may be required to enable this operation.
119 if (fail_ioc_cmd
== 0) {
120 char *ioc_test
= getenv("ZFS_IOC_TEST");
121 unsigned int ioc_num
= 0, ioc_err
= 0;
123 if (ioc_test
!= NULL
&&
124 sscanf(ioc_test
, "%i:%i", &ioc_num
, &ioc_err
) == 2 &&
125 ioc_num
< ZFS_IOC_LAST
) {
126 fail_ioc_cmd
= ioc_num
;
127 fail_ioc_err
= ioc_err
;
134 libzfs_core_init(void)
136 (void) pthread_mutex_lock(&g_lock
);
137 if (g_refcount
== 0) {
138 g_fd
= open("/dev/zfs", O_RDWR
);
140 (void) pthread_mutex_unlock(&g_lock
);
147 libzfs_core_debug_ioc();
149 (void) pthread_mutex_unlock(&g_lock
);
154 libzfs_core_fini(void)
156 (void) pthread_mutex_lock(&g_lock
);
157 ASSERT3S(g_refcount
, >, 0);
162 if (g_refcount
== 0 && g_fd
!= -1) {
166 (void) pthread_mutex_unlock(&g_lock
);
170 lzc_ioctl(zfs_ioc_t ioc
, const char *name
,
171 nvlist_t
*source
, nvlist_t
**resultp
)
173 zfs_cmd_t zc
= {"\0"};
178 ASSERT3S(g_refcount
, >, 0);
179 VERIFY3S(g_fd
, !=, -1);
182 if (ioc
== fail_ioc_cmd
)
183 return (fail_ioc_err
);
187 (void) strlcpy(zc
.zc_name
, name
, sizeof (zc
.zc_name
));
189 if (source
!= NULL
) {
190 packed
= fnvlist_pack(source
, &size
);
191 zc
.zc_nvlist_src
= (uint64_t)(uintptr_t)packed
;
192 zc
.zc_nvlist_src_size
= size
;
195 if (resultp
!= NULL
) {
197 if (ioc
== ZFS_IOC_CHANNEL_PROGRAM
) {
198 zc
.zc_nvlist_dst_size
= fnvlist_lookup_uint64(source
,
201 zc
.zc_nvlist_dst_size
= MAX(size
* 2, 128 * 1024);
203 zc
.zc_nvlist_dst
= (uint64_t)(uintptr_t)
204 malloc(zc
.zc_nvlist_dst_size
);
205 if (zc
.zc_nvlist_dst
== (uint64_t)0) {
211 while (ioctl(g_fd
, ioc
, &zc
) != 0) {
213 * If ioctl exited with ENOMEM, we retry the ioctl after
214 * increasing the size of the destination nvlist.
216 * Channel programs that exit with ENOMEM ran over the
217 * lua memory sandbox; they should not be retried.
219 if (errno
== ENOMEM
&& resultp
!= NULL
&&
220 ioc
!= ZFS_IOC_CHANNEL_PROGRAM
) {
221 free((void *)(uintptr_t)zc
.zc_nvlist_dst
);
222 zc
.zc_nvlist_dst_size
*= 2;
223 zc
.zc_nvlist_dst
= (uint64_t)(uintptr_t)
224 malloc(zc
.zc_nvlist_dst_size
);
225 if (zc
.zc_nvlist_dst
== (uint64_t)0) {
234 if (zc
.zc_nvlist_dst_filled
) {
235 *resultp
= fnvlist_unpack((void *)(uintptr_t)zc
.zc_nvlist_dst
,
236 zc
.zc_nvlist_dst_size
);
241 fnvlist_pack_free(packed
, size
);
242 free((void *)(uintptr_t)zc
.zc_nvlist_dst
);
247 lzc_create(const char *fsname
, enum lzc_dataset_type type
, nvlist_t
*props
,
248 uint8_t *wkeydata
, uint_t wkeylen
)
251 nvlist_t
*hidden_args
= NULL
;
252 nvlist_t
*args
= fnvlist_alloc();
254 fnvlist_add_int32(args
, "type", (dmu_objset_type_t
)type
);
256 fnvlist_add_nvlist(args
, "props", props
);
258 if (wkeydata
!= NULL
) {
259 hidden_args
= fnvlist_alloc();
260 fnvlist_add_uint8_array(hidden_args
, "wkeydata", wkeydata
,
262 fnvlist_add_nvlist(args
, ZPOOL_HIDDEN_ARGS
, hidden_args
);
265 error
= lzc_ioctl(ZFS_IOC_CREATE
, fsname
, args
, NULL
);
266 nvlist_free(hidden_args
);
272 lzc_clone(const char *fsname
, const char *origin
, nvlist_t
*props
)
275 nvlist_t
*hidden_args
= NULL
;
276 nvlist_t
*args
= fnvlist_alloc();
278 fnvlist_add_string(args
, "origin", origin
);
280 fnvlist_add_nvlist(args
, "props", props
);
281 error
= lzc_ioctl(ZFS_IOC_CLONE
, fsname
, args
, NULL
);
282 nvlist_free(hidden_args
);
288 lzc_promote(const char *fsname
, char *snapnamebuf
, int snapnamelen
)
291 * The promote ioctl is still legacy, so we need to construct our
292 * own zfs_cmd_t rather than using lzc_ioctl().
294 zfs_cmd_t zc
= { "\0" };
296 ASSERT3S(g_refcount
, >, 0);
297 VERIFY3S(g_fd
, !=, -1);
299 (void) strlcpy(zc
.zc_name
, fsname
, sizeof (zc
.zc_name
));
300 if (ioctl(g_fd
, ZFS_IOC_PROMOTE
, &zc
) != 0) {
302 if (error
== EEXIST
&& snapnamebuf
!= NULL
)
303 (void) strlcpy(snapnamebuf
, zc
.zc_string
, snapnamelen
);
310 lzc_remap(const char *fsname
)
313 nvlist_t
*args
= fnvlist_alloc();
314 error
= lzc_ioctl(ZFS_IOC_REMAP
, fsname
, args
, NULL
);
320 lzc_rename(const char *source
, const char *target
)
322 zfs_cmd_t zc
= { "\0" };
324 ASSERT3S(g_refcount
, >, 0);
325 VERIFY3S(g_fd
, !=, -1);
326 (void) strlcpy(zc
.zc_name
, source
, sizeof (zc
.zc_name
));
327 (void) strlcpy(zc
.zc_value
, target
, sizeof (zc
.zc_value
));
328 error
= ioctl(g_fd
, ZFS_IOC_RENAME
, &zc
);
334 lzc_destroy(const char *fsname
)
337 nvlist_t
*args
= fnvlist_alloc();
338 error
= lzc_ioctl(ZFS_IOC_DESTROY
, fsname
, args
, NULL
);
346 * The keys in the snaps nvlist are the snapshots to be created.
347 * They must all be in the same pool.
349 * The props nvlist is properties to set. Currently only user properties
350 * are supported. { user:prop_name -> string value }
352 * The returned results nvlist will have an entry for each snapshot that failed.
353 * The value will be the (int32) error code.
355 * The return value will be 0 if all snapshots were created, otherwise it will
356 * be the errno of a (unspecified) snapshot that failed.
359 lzc_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
**errlist
)
364 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
368 /* determine the pool name */
369 elem
= nvlist_next_nvpair(snaps
, NULL
);
372 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
373 pool
[strcspn(pool
, "/@")] = '\0';
375 args
= fnvlist_alloc();
376 fnvlist_add_nvlist(args
, "snaps", snaps
);
378 fnvlist_add_nvlist(args
, "props", props
);
380 error
= lzc_ioctl(ZFS_IOC_SNAPSHOT
, pool
, args
, errlist
);
387 * Destroys snapshots.
389 * The keys in the snaps nvlist are the snapshots to be destroyed.
390 * They must all be in the same pool.
392 * Snapshots that do not exist will be silently ignored.
394 * If 'defer' is not set, and a snapshot has user holds or clones, the
395 * destroy operation will fail and none of the snapshots will be
398 * If 'defer' is set, and a snapshot has user holds or clones, it will be
399 * marked for deferred destruction, and will be destroyed when the last hold
400 * or clone is removed/destroyed.
402 * The return value will be 0 if all snapshots were destroyed (or marked for
403 * later destruction if 'defer' is set) or didn't exist to begin with.
405 * Otherwise the return value will be the errno of a (unspecified) snapshot
406 * that failed, no snapshots will be destroyed, and the errlist will have an
407 * entry for each snapshot that failed. The value in the errlist will be
408 * the (int32) error code.
411 lzc_destroy_snaps(nvlist_t
*snaps
, boolean_t defer
, nvlist_t
**errlist
)
416 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
418 /* determine the pool name */
419 elem
= nvlist_next_nvpair(snaps
, NULL
);
422 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
423 pool
[strcspn(pool
, "/@")] = '\0';
425 args
= fnvlist_alloc();
426 fnvlist_add_nvlist(args
, "snaps", snaps
);
428 fnvlist_add_boolean(args
, "defer");
430 error
= lzc_ioctl(ZFS_IOC_DESTROY_SNAPS
, pool
, args
, errlist
);
437 lzc_snaprange_space(const char *firstsnap
, const char *lastsnap
,
443 char fs
[ZFS_MAX_DATASET_NAME_LEN
];
446 /* determine the fs name */
447 (void) strlcpy(fs
, firstsnap
, sizeof (fs
));
448 atp
= strchr(fs
, '@');
453 args
= fnvlist_alloc();
454 fnvlist_add_string(args
, "firstsnap", firstsnap
);
456 err
= lzc_ioctl(ZFS_IOC_SPACE_SNAPS
, lastsnap
, args
, &result
);
459 *usedp
= fnvlist_lookup_uint64(result
, "used");
460 fnvlist_free(result
);
466 lzc_exists(const char *dataset
)
469 * The objset_stats ioctl is still legacy, so we need to construct our
470 * own zfs_cmd_t rather than using lzc_ioctl().
472 zfs_cmd_t zc
= {"\0"};
474 ASSERT3S(g_refcount
, >, 0);
475 VERIFY3S(g_fd
, !=, -1);
477 (void) strlcpy(zc
.zc_name
, dataset
, sizeof (zc
.zc_name
));
478 return (ioctl(g_fd
, ZFS_IOC_OBJSET_STATS
, &zc
) == 0);
483 * It was added to preserve the function signature in case it is
484 * needed in the future.
488 lzc_sync(const char *pool_name
, nvlist_t
*innvl
, nvlist_t
**outnvl
)
490 return (lzc_ioctl(ZFS_IOC_POOL_SYNC
, pool_name
, innvl
, NULL
));
494 * Create "user holds" on snapshots. If there is a hold on a snapshot,
495 * the snapshot can not be destroyed. (However, it can be marked for deletion
496 * by lzc_destroy_snaps(defer=B_TRUE).)
498 * The keys in the nvlist are snapshot names.
499 * The snapshots must all be in the same pool.
500 * The value is the name of the hold (string type).
502 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
503 * In this case, when the cleanup_fd is closed (including on process
504 * termination), the holds will be released. If the system is shut down
505 * uncleanly, the holds will be released when the pool is next opened
508 * Holds for snapshots which don't exist will be skipped and have an entry
509 * added to errlist, but will not cause an overall failure.
511 * The return value will be 0 if all holds, for snapshots that existed,
512 * were successfully created.
514 * Otherwise the return value will be the errno of a (unspecified) hold that
515 * failed and no holds will be created.
517 * In all cases the errlist will have an entry for each hold that failed
518 * (name = snapshot), with its value being the error code (int32).
521 lzc_hold(nvlist_t
*holds
, int cleanup_fd
, nvlist_t
**errlist
)
523 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
528 /* determine the pool name */
529 elem
= nvlist_next_nvpair(holds
, NULL
);
532 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
533 pool
[strcspn(pool
, "/@")] = '\0';
535 args
= fnvlist_alloc();
536 fnvlist_add_nvlist(args
, "holds", holds
);
537 if (cleanup_fd
!= -1)
538 fnvlist_add_int32(args
, "cleanup_fd", cleanup_fd
);
540 error
= lzc_ioctl(ZFS_IOC_HOLD
, pool
, args
, errlist
);
546 * Release "user holds" on snapshots. If the snapshot has been marked for
547 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
548 * any clones, and all the user holds are removed, then the snapshot will be
551 * The keys in the nvlist are snapshot names.
552 * The snapshots must all be in the same pool.
553 * The value is an nvlist whose keys are the holds to remove.
555 * Holds which failed to release because they didn't exist will have an entry
556 * added to errlist, but will not cause an overall failure.
558 * The return value will be 0 if the nvl holds was empty or all holds that
559 * existed, were successfully removed.
561 * Otherwise the return value will be the errno of a (unspecified) hold that
562 * failed to release and no holds will be released.
564 * In all cases the errlist will have an entry for each hold that failed to
568 lzc_release(nvlist_t
*holds
, nvlist_t
**errlist
)
570 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
573 /* determine the pool name */
574 elem
= nvlist_next_nvpair(holds
, NULL
);
577 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
578 pool
[strcspn(pool
, "/@")] = '\0';
580 return (lzc_ioctl(ZFS_IOC_RELEASE
, pool
, holds
, errlist
));
584 * Retrieve list of user holds on the specified snapshot.
586 * On success, *holdsp will be set to an nvlist which the caller must free.
587 * The keys are the names of the holds, and the value is the creation time
588 * of the hold (uint64) in seconds since the epoch.
591 lzc_get_holds(const char *snapname
, nvlist_t
**holdsp
)
593 return (lzc_ioctl(ZFS_IOC_GET_HOLDS
, snapname
, NULL
, holdsp
));
597 * Generate a zfs send stream for the specified snapshot and write it to
598 * the specified file descriptor.
600 * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap")
602 * If "from" is NULL, a full (non-incremental) stream will be sent.
603 * If "from" is non-NULL, it must be the full name of a snapshot or
604 * bookmark to send an incremental from (e.g. "pool/fs@earlier_snap" or
605 * "pool/fs#earlier_bmark"). If non-NULL, the specified snapshot or
606 * bookmark must represent an earlier point in the history of "snapname").
607 * It can be an earlier snapshot in the same filesystem or zvol as "snapname",
608 * or it can be the origin of "snapname"'s filesystem, or an earlier
609 * snapshot in the origin, etc.
611 * "fd" is the file descriptor to write the send stream to.
613 * If "flags" contains LZC_SEND_FLAG_LARGE_BLOCK, the stream is permitted
614 * to contain DRR_WRITE records with drr_length > 128K, and DRR_OBJECT
615 * records with drr_blksz > 128K.
617 * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted
618 * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA,
619 * which the receiving system must support (as indicated by support
620 * for the "embedded_data" feature).
622 * If "flags" contains LZC_SEND_FLAG_COMPRESS, the stream is generated by using
623 * compressed WRITE records for blocks which are compressed on disk and in
624 * memory. If the lz4_compress feature is active on the sending system, then
625 * the receiving system must have that feature enabled as well.
627 * If "flags" contains LZC_SEND_FLAG_RAW, the stream is generated, for encrypted
628 * datasets, by sending data exactly as it exists on disk. This allows backups
629 * to be taken even if encryption keys are not currently loaded.
632 lzc_send(const char *snapname
, const char *from
, int fd
,
633 enum lzc_send_flags flags
)
635 return (lzc_send_resume(snapname
, from
, fd
, flags
, 0, 0));
639 lzc_send_resume(const char *snapname
, const char *from
, int fd
,
640 enum lzc_send_flags flags
, uint64_t resumeobj
, uint64_t resumeoff
)
645 args
= fnvlist_alloc();
646 fnvlist_add_int32(args
, "fd", fd
);
648 fnvlist_add_string(args
, "fromsnap", from
);
649 if (flags
& LZC_SEND_FLAG_LARGE_BLOCK
)
650 fnvlist_add_boolean(args
, "largeblockok");
651 if (flags
& LZC_SEND_FLAG_EMBED_DATA
)
652 fnvlist_add_boolean(args
, "embedok");
653 if (flags
& LZC_SEND_FLAG_COMPRESS
)
654 fnvlist_add_boolean(args
, "compressok");
655 if (flags
& LZC_SEND_FLAG_RAW
)
656 fnvlist_add_boolean(args
, "rawok");
657 if (resumeobj
!= 0 || resumeoff
!= 0) {
658 fnvlist_add_uint64(args
, "resume_object", resumeobj
);
659 fnvlist_add_uint64(args
, "resume_offset", resumeoff
);
661 err
= lzc_ioctl(ZFS_IOC_SEND_NEW
, snapname
, args
, NULL
);
667 * "from" can be NULL, a snapshot, or a bookmark.
669 * If from is NULL, a full (non-incremental) stream will be estimated. This
670 * is calculated very efficiently.
672 * If from is a snapshot, lzc_send_space uses the deadlists attached to
673 * each snapshot to efficiently estimate the stream size.
675 * If from is a bookmark, the indirect blocks in the destination snapshot
676 * are traversed, looking for blocks with a birth time since the creation TXG of
677 * the snapshot this bookmark was created from. This will result in
678 * significantly more I/O and be less efficient than a send space estimation on
679 * an equivalent snapshot.
682 lzc_send_space(const char *snapname
, const char *from
,
683 enum lzc_send_flags flags
, uint64_t *spacep
)
689 args
= fnvlist_alloc();
691 fnvlist_add_string(args
, "from", from
);
692 if (flags
& LZC_SEND_FLAG_LARGE_BLOCK
)
693 fnvlist_add_boolean(args
, "largeblockok");
694 if (flags
& LZC_SEND_FLAG_EMBED_DATA
)
695 fnvlist_add_boolean(args
, "embedok");
696 if (flags
& LZC_SEND_FLAG_COMPRESS
)
697 fnvlist_add_boolean(args
, "compressok");
698 if (flags
& LZC_SEND_FLAG_RAW
)
699 fnvlist_add_boolean(args
, "rawok");
700 err
= lzc_ioctl(ZFS_IOC_SEND_SPACE
, snapname
, args
, &result
);
703 *spacep
= fnvlist_lookup_uint64(result
, "space");
709 recv_read(int fd
, void *buf
, int ilen
)
716 rv
= read(fd
, cp
, len
);
721 if (rv
< 0 || len
!= 0)
728 * Linux adds ZFS_IOC_RECV_NEW for resumable and raw streams and preserves the
729 * legacy ZFS_IOC_RECV user/kernel interface. The new interface supports all
730 * stream options but is currently only used for resumable streams. This way
731 * updated user space utilities will interoperate with older kernel modules.
733 * Non-Linux OpenZFS platforms have opted to modify the legacy interface.
736 recv_impl(const char *snapname
, nvlist_t
*recvdprops
, nvlist_t
*localprops
,
737 uint8_t *wkeydata
, uint_t wkeylen
, const char *origin
, boolean_t force
,
738 boolean_t resumable
, boolean_t raw
, int input_fd
,
739 const dmu_replay_record_t
*begin_record
, int cleanup_fd
,
740 uint64_t *read_bytes
, uint64_t *errflags
, uint64_t *action_handle
,
743 dmu_replay_record_t drr
;
744 char fsname
[MAXPATHLEN
];
748 ASSERT3S(g_refcount
, >, 0);
749 VERIFY3S(g_fd
, !=, -1);
751 /* Set 'fsname' to the name of containing filesystem */
752 (void) strlcpy(fsname
, snapname
, sizeof (fsname
));
753 atp
= strchr(fsname
, '@');
758 /* If the fs does not exist, try its parent. */
759 if (!lzc_exists(fsname
)) {
760 char *slashp
= strrchr(fsname
, '/');
767 * The begin_record is normally a non-byteswapped BEGIN record.
768 * For resumable streams it may be set to any non-byteswapped
769 * dmu_replay_record_t.
771 if (begin_record
== NULL
) {
772 error
= recv_read(input_fd
, &drr
, sizeof (drr
));
780 * Raw receives, resumable receives, and receives that include a
781 * wrapping key all use the new interface.
783 if (resumable
|| raw
|| wkeydata
!= NULL
) {
784 nvlist_t
*outnvl
= NULL
;
785 nvlist_t
*innvl
= fnvlist_alloc();
787 fnvlist_add_string(innvl
, "snapname", snapname
);
789 if (recvdprops
!= NULL
)
790 fnvlist_add_nvlist(innvl
, "props", recvdprops
);
792 if (localprops
!= NULL
)
793 fnvlist_add_nvlist(innvl
, "localprops", localprops
);
795 if (wkeydata
!= NULL
) {
797 * wkeydata must be placed in the special
798 * ZPOOL_HIDDEN_ARGS nvlist so that it
799 * will not be printed to the zpool history.
801 nvlist_t
*hidden_args
= fnvlist_alloc();
802 fnvlist_add_uint8_array(hidden_args
, "wkeydata",
804 fnvlist_add_nvlist(innvl
, ZPOOL_HIDDEN_ARGS
,
806 nvlist_free(hidden_args
);
809 if (origin
!= NULL
&& strlen(origin
))
810 fnvlist_add_string(innvl
, "origin", origin
);
812 fnvlist_add_byte_array(innvl
, "begin_record",
813 (uchar_t
*)&drr
, sizeof (drr
));
815 fnvlist_add_int32(innvl
, "input_fd", input_fd
);
818 fnvlist_add_boolean(innvl
, "force");
821 fnvlist_add_boolean(innvl
, "resumable");
824 fnvlist_add_int32(innvl
, "cleanup_fd", cleanup_fd
);
826 if (action_handle
!= NULL
)
827 fnvlist_add_uint64(innvl
, "action_handle",
830 error
= lzc_ioctl(ZFS_IOC_RECV_NEW
, fsname
, innvl
, &outnvl
);
832 if (error
== 0 && read_bytes
!= NULL
)
833 error
= nvlist_lookup_uint64(outnvl
, "read_bytes",
836 if (error
== 0 && errflags
!= NULL
)
837 error
= nvlist_lookup_uint64(outnvl
, "error_flags",
840 if (error
== 0 && action_handle
!= NULL
)
841 error
= nvlist_lookup_uint64(outnvl
, "action_handle",
844 if (error
== 0 && errors
!= NULL
) {
846 error
= nvlist_lookup_nvlist(outnvl
, "errors", &nvl
);
848 *errors
= fnvlist_dup(nvl
);
852 fnvlist_free(outnvl
);
854 zfs_cmd_t zc
= {"\0"};
858 ASSERT3S(g_refcount
, >, 0);
860 (void) strlcpy(zc
.zc_name
, fsname
, sizeof (zc
.zc_value
));
861 (void) strlcpy(zc
.zc_value
, snapname
, sizeof (zc
.zc_value
));
863 if (recvdprops
!= NULL
) {
864 packed
= fnvlist_pack(recvdprops
, &size
);
865 zc
.zc_nvlist_src
= (uint64_t)(uintptr_t)packed
;
866 zc
.zc_nvlist_src_size
= size
;
869 if (localprops
!= NULL
) {
870 packed
= fnvlist_pack(localprops
, &size
);
871 zc
.zc_nvlist_conf
= (uint64_t)(uintptr_t)packed
;
872 zc
.zc_nvlist_conf_size
= size
;
876 (void) strlcpy(zc
.zc_string
, origin
,
877 sizeof (zc
.zc_string
));
879 ASSERT3S(drr
.drr_type
, ==, DRR_BEGIN
);
880 zc
.zc_begin_record
= drr
.drr_u
.drr_begin
;
882 zc
.zc_cookie
= input_fd
;
883 zc
.zc_cleanup_fd
= -1;
884 zc
.zc_action_handle
= 0;
887 zc
.zc_cleanup_fd
= cleanup_fd
;
889 if (action_handle
!= NULL
)
890 zc
.zc_action_handle
= *action_handle
;
892 zc
.zc_nvlist_dst_size
= 128 * 1024;
893 zc
.zc_nvlist_dst
= (uint64_t)(uintptr_t)
894 malloc(zc
.zc_nvlist_dst_size
);
896 error
= ioctl(g_fd
, ZFS_IOC_RECV
, &zc
);
900 if (read_bytes
!= NULL
)
901 *read_bytes
= zc
.zc_cookie
;
903 if (errflags
!= NULL
)
904 *errflags
= zc
.zc_obj
;
906 if (action_handle
!= NULL
)
907 *action_handle
= zc
.zc_action_handle
;
910 VERIFY0(nvlist_unpack(
911 (void *)(uintptr_t)zc
.zc_nvlist_dst
,
912 zc
.zc_nvlist_dst_size
, errors
, KM_SLEEP
));
916 fnvlist_pack_free(packed
, size
);
917 free((void *)(uintptr_t)zc
.zc_nvlist_dst
);
924 * The simplest receive case: receive from the specified fd, creating the
925 * specified snapshot. Apply the specified properties as "received" properties
926 * (which can be overridden by locally-set properties). If the stream is a
927 * clone, its origin snapshot must be specified by 'origin'. The 'force'
928 * flag will cause the target filesystem to be rolled back or destroyed if
929 * necessary to receive.
931 * Return 0 on success or an errno on failure.
933 * Note: this interface does not work on dedup'd streams
934 * (those with DMU_BACKUP_FEATURE_DEDUP).
937 lzc_receive(const char *snapname
, nvlist_t
*props
, const char *origin
,
938 boolean_t force
, boolean_t raw
, int fd
)
940 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
941 B_FALSE
, raw
, fd
, NULL
, -1, NULL
, NULL
, NULL
, NULL
));
945 * Like lzc_receive, but if the receive fails due to premature stream
946 * termination, the intermediate state will be preserved on disk. In this
947 * case, ECKSUM will be returned. The receive may subsequently be resumed
948 * with a resuming send stream generated by lzc_send_resume().
951 lzc_receive_resumable(const char *snapname
, nvlist_t
*props
, const char *origin
,
952 boolean_t force
, boolean_t raw
, int fd
)
954 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
955 B_TRUE
, raw
, fd
, NULL
, -1, NULL
, NULL
, NULL
, NULL
));
959 * Like lzc_receive, but allows the caller to read the begin record and then to
960 * pass it in. That could be useful if the caller wants to derive, for example,
961 * the snapname or the origin parameters based on the information contained in
963 * The begin record must be in its original form as read from the stream,
964 * in other words, it should not be byteswapped.
966 * The 'resumable' parameter allows to obtain the same behavior as with
967 * lzc_receive_resumable.
970 lzc_receive_with_header(const char *snapname
, nvlist_t
*props
,
971 const char *origin
, boolean_t force
, boolean_t resumable
, boolean_t raw
,
972 int fd
, const dmu_replay_record_t
*begin_record
)
974 if (begin_record
== NULL
)
977 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
978 resumable
, raw
, fd
, begin_record
, -1, NULL
, NULL
, NULL
, NULL
));
982 * Like lzc_receive, but allows the caller to pass all supported arguments
983 * and retrieve all values returned. The only additional input parameter
984 * is 'cleanup_fd' which is used to set a cleanup-on-exit file descriptor.
986 * The following parameters all provide return values. Several may be set
987 * in the failure case and will contain additional information.
989 * The 'read_bytes' value will be set to the total number of bytes read.
991 * The 'errflags' value will contain zprop_errflags_t flags which are
992 * used to describe any failures.
994 * The 'action_handle' is used to pass the handle for this guid/ds mapping.
995 * It should be set to zero on first call and will contain an updated handle
996 * on success, it should be passed in subsequent calls.
998 * The 'errors' nvlist contains an entry for each unapplied received
999 * property. Callers are responsible for freeing this nvlist.
1001 int lzc_receive_one(const char *snapname
, nvlist_t
*props
,
1002 const char *origin
, boolean_t force
, boolean_t resumable
, boolean_t raw
,
1003 int input_fd
, const dmu_replay_record_t
*begin_record
, int cleanup_fd
,
1004 uint64_t *read_bytes
, uint64_t *errflags
, uint64_t *action_handle
,
1007 return (recv_impl(snapname
, props
, NULL
, NULL
, 0, origin
, force
,
1008 resumable
, raw
, input_fd
, begin_record
, cleanup_fd
, read_bytes
,
1009 errflags
, action_handle
, errors
));
1013 * Like lzc_receive_one, but allows the caller to pass an additional 'cmdprops'
1016 * The 'cmdprops' nvlist contains both override ('zfs receive -o') and
1017 * exclude ('zfs receive -x') properties. Callers are responsible for freeing
1020 int lzc_receive_with_cmdprops(const char *snapname
, nvlist_t
*props
,
1021 nvlist_t
*cmdprops
, uint8_t *wkeydata
, uint_t wkeylen
, const char *origin
,
1022 boolean_t force
, boolean_t resumable
, boolean_t raw
, int input_fd
,
1023 const dmu_replay_record_t
*begin_record
, int cleanup_fd
,
1024 uint64_t *read_bytes
, uint64_t *errflags
, uint64_t *action_handle
,
1027 return (recv_impl(snapname
, props
, cmdprops
, wkeydata
, wkeylen
, origin
,
1028 force
, resumable
, raw
, input_fd
, begin_record
, cleanup_fd
,
1029 read_bytes
, errflags
, action_handle
, errors
));
1033 * Roll back this filesystem or volume to its most recent snapshot.
1034 * If snapnamebuf is not NULL, it will be filled in with the name
1035 * of the most recent snapshot.
1036 * Note that the latest snapshot may change if a new one is concurrently
1037 * created or the current one is destroyed. lzc_rollback_to can be used
1038 * to roll back to a specific latest snapshot.
1040 * Return 0 on success or an errno on failure.
1043 lzc_rollback(const char *fsname
, char *snapnamebuf
, int snapnamelen
)
1049 args
= fnvlist_alloc();
1050 err
= lzc_ioctl(ZFS_IOC_ROLLBACK
, fsname
, args
, &result
);
1052 if (err
== 0 && snapnamebuf
!= NULL
) {
1053 const char *snapname
= fnvlist_lookup_string(result
, "target");
1054 (void) strlcpy(snapnamebuf
, snapname
, snapnamelen
);
1056 nvlist_free(result
);
1062 * Roll back this filesystem or volume to the specified snapshot,
1065 * Return 0 on success or an errno on failure.
1068 lzc_rollback_to(const char *fsname
, const char *snapname
)
1074 args
= fnvlist_alloc();
1075 fnvlist_add_string(args
, "target", snapname
);
1076 err
= lzc_ioctl(ZFS_IOC_ROLLBACK
, fsname
, args
, &result
);
1078 nvlist_free(result
);
1083 * Creates bookmarks.
1085 * The bookmarks nvlist maps from name of the bookmark (e.g. "pool/fs#bmark") to
1086 * the name of the snapshot (e.g. "pool/fs@snap"). All the bookmarks and
1087 * snapshots must be in the same pool.
1089 * The returned results nvlist will have an entry for each bookmark that failed.
1090 * The value will be the (int32) error code.
1092 * The return value will be 0 if all bookmarks were created, otherwise it will
1093 * be the errno of a (undetermined) bookmarks that failed.
1096 lzc_bookmark(nvlist_t
*bookmarks
, nvlist_t
**errlist
)
1100 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
1102 /* determine the pool name */
1103 elem
= nvlist_next_nvpair(bookmarks
, NULL
);
1106 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
1107 pool
[strcspn(pool
, "/#")] = '\0';
1109 error
= lzc_ioctl(ZFS_IOC_BOOKMARK
, pool
, bookmarks
, errlist
);
1115 * Retrieve bookmarks.
1117 * Retrieve the list of bookmarks for the given file system. The props
1118 * parameter is an nvlist of property names (with no values) that will be
1119 * returned for each bookmark.
1121 * The following are valid properties on bookmarks, all of which are numbers
1122 * (represented as uint64 in the nvlist)
1124 * "guid" - globally unique identifier of the snapshot it refers to
1125 * "createtxg" - txg when the snapshot it refers to was created
1126 * "creation" - timestamp when the snapshot it refers to was created
1128 * The format of the returned nvlist as follows:
1129 * <short name of bookmark> -> {
1130 * <name of property> -> {
1136 lzc_get_bookmarks(const char *fsname
, nvlist_t
*props
, nvlist_t
**bmarks
)
1138 return (lzc_ioctl(ZFS_IOC_GET_BOOKMARKS
, fsname
, props
, bmarks
));
1142 * Destroys bookmarks.
1144 * The keys in the bmarks nvlist are the bookmarks to be destroyed.
1145 * They must all be in the same pool. Bookmarks are specified as
1148 * Bookmarks that do not exist will be silently ignored.
1150 * The return value will be 0 if all bookmarks that existed were destroyed.
1152 * Otherwise the return value will be the errno of a (undetermined) bookmark
1153 * that failed, no bookmarks will be destroyed, and the errlist will have an
1154 * entry for each bookmarks that failed. The value in the errlist will be
1155 * the (int32) error code.
1158 lzc_destroy_bookmarks(nvlist_t
*bmarks
, nvlist_t
**errlist
)
1162 char pool
[ZFS_MAX_DATASET_NAME_LEN
];
1164 /* determine the pool name */
1165 elem
= nvlist_next_nvpair(bmarks
, NULL
);
1168 (void) strlcpy(pool
, nvpair_name(elem
), sizeof (pool
));
1169 pool
[strcspn(pool
, "/#")] = '\0';
1171 error
= lzc_ioctl(ZFS_IOC_DESTROY_BOOKMARKS
, pool
, bmarks
, errlist
);
1177 lzc_channel_program_impl(const char *pool
, const char *program
, boolean_t sync
,
1178 uint64_t instrlimit
, uint64_t memlimit
, nvlist_t
*argnvl
, nvlist_t
**outnvl
)
1183 args
= fnvlist_alloc();
1184 fnvlist_add_string(args
, ZCP_ARG_PROGRAM
, program
);
1185 fnvlist_add_nvlist(args
, ZCP_ARG_ARGLIST
, argnvl
);
1186 fnvlist_add_boolean_value(args
, ZCP_ARG_SYNC
, sync
);
1187 fnvlist_add_uint64(args
, ZCP_ARG_INSTRLIMIT
, instrlimit
);
1188 fnvlist_add_uint64(args
, ZCP_ARG_MEMLIMIT
, memlimit
);
1189 error
= lzc_ioctl(ZFS_IOC_CHANNEL_PROGRAM
, pool
, args
, outnvl
);
1196 * Executes a channel program.
1198 * If this function returns 0 the channel program was successfully loaded and
1199 * ran without failing. Note that individual commands the channel program ran
1200 * may have failed and the channel program is responsible for reporting such
1201 * errors through outnvl if they are important.
1203 * This method may also return:
1205 * EINVAL The program contains syntax errors, or an invalid memory or time
1206 * limit was given. No part of the channel program was executed.
1207 * If caused by syntax errors, 'outnvl' contains information about the
1210 * ECHRNG The program was executed, but encountered a runtime error, such as
1211 * calling a function with incorrect arguments, invoking the error()
1212 * function directly, failing an assert() command, etc. Some portion
1213 * of the channel program may have executed and committed changes.
1214 * Information about the failure can be found in 'outnvl'.
1216 * ENOMEM The program fully executed, but the output buffer was not large
1217 * enough to store the returned value. No output is returned through
1220 * ENOSPC The program was terminated because it exceeded its memory usage
1221 * limit. Some portion of the channel program may have executed and
1222 * committed changes to disk. No output is returned through 'outnvl'.
1224 * ETIME The program was terminated because it exceeded its Lua instruction
1225 * limit. Some portion of the channel program may have executed and
1226 * committed changes to disk. No output is returned through 'outnvl'.
1229 lzc_channel_program(const char *pool
, const char *program
, uint64_t instrlimit
,
1230 uint64_t memlimit
, nvlist_t
*argnvl
, nvlist_t
**outnvl
)
1232 return (lzc_channel_program_impl(pool
, program
, B_TRUE
, instrlimit
,
1233 memlimit
, argnvl
, outnvl
));
1237 * Creates a checkpoint for the specified pool.
1239 * If this function returns 0 the pool was successfully checkpointed.
1241 * This method may also return:
1243 * ZFS_ERR_CHECKPOINT_EXISTS
1244 * The pool already has a checkpoint. A pools can only have one
1245 * checkpoint at most, at any given time.
1247 * ZFS_ERR_DISCARDING_CHECKPOINT
1248 * ZFS is in the middle of discarding a checkpoint for this pool.
1249 * The pool can be checkpointed again once the discard is done.
1251 * ZFS_DEVRM_IN_PROGRESS
1252 * A vdev is currently being removed. The pool cannot be
1253 * checkpointed until the device removal is done.
1256 * One or more top-level vdevs exceed the maximum vdev size
1257 * supported for this feature.
1260 lzc_pool_checkpoint(const char *pool
)
1264 nvlist_t
*result
= NULL
;
1265 nvlist_t
*args
= fnvlist_alloc();
1267 error
= lzc_ioctl(ZFS_IOC_POOL_CHECKPOINT
, pool
, args
, &result
);
1270 fnvlist_free(result
);
1276 * Discard the checkpoint from the specified pool.
1278 * If this function returns 0 the checkpoint was successfully discarded.
1280 * This method may also return:
1282 * ZFS_ERR_NO_CHECKPOINT
1283 * The pool does not have a checkpoint.
1285 * ZFS_ERR_DISCARDING_CHECKPOINT
1286 * ZFS is already in the middle of discarding the checkpoint.
1289 lzc_pool_checkpoint_discard(const char *pool
)
1293 nvlist_t
*result
= NULL
;
1294 nvlist_t
*args
= fnvlist_alloc();
1296 error
= lzc_ioctl(ZFS_IOC_POOL_DISCARD_CHECKPOINT
, pool
, args
, &result
);
1299 fnvlist_free(result
);
1305 * Executes a read-only channel program.
1307 * A read-only channel program works programmatically the same way as a
1308 * normal channel program executed with lzc_channel_program(). The only
1309 * difference is it runs exclusively in open-context and therefore can
1310 * return faster. The downside to that, is that the program cannot change
1311 * on-disk state by calling functions from the zfs.sync submodule.
1313 * The return values of this function (and their meaning) are exactly the
1314 * same as the ones described in lzc_channel_program().
1317 lzc_channel_program_nosync(const char *pool
, const char *program
,
1318 uint64_t timeout
, uint64_t memlimit
, nvlist_t
*argnvl
, nvlist_t
**outnvl
)
1320 return (lzc_channel_program_impl(pool
, program
, B_FALSE
, timeout
,
1321 memlimit
, argnvl
, outnvl
));
1325 * Performs key management functions
1327 * crypto_cmd should be a value from dcp_cmd_t. If the command specifies to
1328 * load or change a wrapping key, the key should be specified in the
1329 * hidden_args nvlist so that it is not logged.
1332 lzc_load_key(const char *fsname
, boolean_t noop
, uint8_t *wkeydata
,
1337 nvlist_t
*hidden_args
;
1339 if (wkeydata
== NULL
)
1342 ioc_args
= fnvlist_alloc();
1343 hidden_args
= fnvlist_alloc();
1344 fnvlist_add_uint8_array(hidden_args
, "wkeydata", wkeydata
, wkeylen
);
1345 fnvlist_add_nvlist(ioc_args
, ZPOOL_HIDDEN_ARGS
, hidden_args
);
1347 fnvlist_add_boolean(ioc_args
, "noop");
1348 error
= lzc_ioctl(ZFS_IOC_LOAD_KEY
, fsname
, ioc_args
, NULL
);
1349 nvlist_free(hidden_args
);
1350 nvlist_free(ioc_args
);
1356 lzc_unload_key(const char *fsname
)
1358 return (lzc_ioctl(ZFS_IOC_UNLOAD_KEY
, fsname
, NULL
, NULL
));
1362 lzc_change_key(const char *fsname
, uint64_t crypt_cmd
, nvlist_t
*props
,
1363 uint8_t *wkeydata
, uint_t wkeylen
)
1366 nvlist_t
*ioc_args
= fnvlist_alloc();
1367 nvlist_t
*hidden_args
= NULL
;
1369 fnvlist_add_uint64(ioc_args
, "crypt_cmd", crypt_cmd
);
1371 if (wkeydata
!= NULL
) {
1372 hidden_args
= fnvlist_alloc();
1373 fnvlist_add_uint8_array(hidden_args
, "wkeydata", wkeydata
,
1375 fnvlist_add_nvlist(ioc_args
, ZPOOL_HIDDEN_ARGS
, hidden_args
);
1379 fnvlist_add_nvlist(ioc_args
, "props", props
);
1381 error
= lzc_ioctl(ZFS_IOC_CHANGE_KEY
, fsname
, ioc_args
, NULL
);
1382 nvlist_free(hidden_args
);
1383 nvlist_free(ioc_args
);
1389 lzc_reopen(const char *pool_name
, boolean_t scrub_restart
)
1391 nvlist_t
*args
= fnvlist_alloc();
1394 fnvlist_add_boolean_value(args
, "scrub_restart", scrub_restart
);
1396 error
= lzc_ioctl(ZFS_IOC_POOL_REOPEN
, pool_name
, args
, NULL
);