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 by Delphix. All rights reserved.
27 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs.
28 * It has the following characteristics:
30 * - Thread Safe. libzfs_core is accessible concurrently from multiple
31 * threads. This is accomplished primarily by avoiding global data
32 * (e.g. caching). Since it's thread-safe, there is no reason for a
33 * process to have multiple libzfs "instances". Therefore, we store
34 * our few pieces of data (e.g. the file descriptor) in global
35 * variables. The fd is reference-counted so that the libzfs_core
36 * library can be "initialized" multiple times (e.g. by different
37 * consumers within the same process).
39 * - Committed Interface. The libzfs_core interface will be committed,
40 * therefore consumers can compile against it and be confident that
41 * their code will continue to work on future releases of this code.
42 * Currently, the interface is Evolving (not Committed), but we intend
43 * to commit to it once it is more complete and we determine that it
44 * meets the needs of all consumers.
46 * - Programatic Error Handling. libzfs_core communicates errors with
47 * defined error numbers, and doesn't print anything to stdout/stderr.
49 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments
50 * to/from the kernel ioctls. There is generally a 1:1 correspondence
51 * between libzfs_core functions and ioctls to /dev/zfs.
53 * - Clear Atomicity. Because libzfs_core functions are generally 1:1
54 * with kernel ioctls, and kernel ioctls are general atomic, each
55 * libzfs_core function is atomic. For example, creating multiple
56 * snapshots with a single call to lzc_snapshot() is atomic -- it
57 * can't fail with only some of the requested snapshots created, even
58 * in the event of power loss or system crash.
60 * - Continued libzfs Support. Some higher-level operations (e.g.
61 * support for "zfs send -R") are too complicated to fit the scope of
62 * libzfs_core. This functionality will continue to live in libzfs.
63 * Where appropriate, libzfs will use the underlying atomic operations
64 * of libzfs_core. For example, libzfs may implement "zfs send -R |
65 * zfs receive" by using individual "send one snapshot", rename,
66 * destroy, and "receive one snapshot" operations in libzfs_core.
67 * /sbin/zfs and /zbin/zpool will link with both libzfs and
68 * libzfs_core. Other consumers should aim to use only libzfs_core,
69 * since that will be the supported, stable interface going forwards.
72 #define _IN_LIBZFS_CORE_
74 #include <libzfs_core.h>
82 #include <sys/nvpair.h>
83 #include <sys/param.h>
84 #include <sys/types.h>
86 #include <sys/zfs_ioctl.h>
87 #include "libzfs_core_compat.h"
88 #include "libzfs_compat.h"
91 extern int zfs_ioctl_version;
95 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER;
96 static int g_refcount;
99 libzfs_core_init(void)
101 (void) pthread_mutex_lock(&g_lock);
102 if (g_refcount == 0) {
103 g_fd = open("/dev/zfs", O_RDWR);
105 (void) pthread_mutex_unlock(&g_lock);
110 (void) pthread_mutex_unlock(&g_lock);
116 libzfs_core_fini(void)
118 (void) pthread_mutex_lock(&g_lock);
119 ASSERT3S(g_refcount, >, 0);
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 };
138 ASSERT3S(g_refcount, >, 0);
140 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
143 if (zfs_ioctl_version == ZFS_IOCVER_UNDEF)
144 zfs_ioctl_version = get_zfs_ioctl_version();
146 if (zfs_ioctl_version < ZFS_IOCVER_LZC) {
148 error = lzc_compat_pre(&zc, &ioc, &source);
154 packed = fnvlist_pack(source, &size);
155 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
156 zc.zc_nvlist_src_size = size;
158 if (resultp != NULL) {
160 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
161 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
162 malloc(zc.zc_nvlist_dst_size);
164 if (zc.zc_nvlist_dst == NULL) {
166 if (zc.zc_nvlist_dst == 0) {
173 while (ioctl(g_fd, ioc, &zc) != 0) {
174 if (errno == ENOMEM && resultp != NULL) {
175 free((void *)(uintptr_t)zc.zc_nvlist_dst);
176 zc.zc_nvlist_dst_size *= 2;
177 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
178 malloc(zc.zc_nvlist_dst_size);
180 if (zc.zc_nvlist_dst == NULL) {
182 if (zc.zc_nvlist_dst == 0) {
194 if (zfs_ioctl_version < ZFS_IOCVER_LZC)
195 lzc_compat_post(&zc, ioc);
197 if (zc.zc_nvlist_dst_filled) {
198 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
199 zc.zc_nvlist_dst_size);
202 if (zfs_ioctl_version < ZFS_IOCVER_LZC)
203 lzc_compat_outnvl(&zc, ioc, resultp);
207 if (zfs_ioctl_version < ZFS_IOCVER_LZC) {
208 if (source != oldsource)
213 fnvlist_pack_free(packed, size);
214 free((void *)(uintptr_t)zc.zc_nvlist_dst);
219 lzc_create(const char *fsname, dmu_objset_type_t type, nvlist_t *props)
222 nvlist_t *args = fnvlist_alloc();
223 fnvlist_add_int32(args, "type", type);
225 fnvlist_add_nvlist(args, "props", props);
226 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
232 lzc_clone(const char *fsname, const char *origin,
236 nvlist_t *args = fnvlist_alloc();
237 fnvlist_add_string(args, "origin", origin);
239 fnvlist_add_nvlist(args, "props", props);
240 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
248 * The keys in the snaps nvlist are the snapshots to be created.
249 * They must all be in the same pool.
251 * The props nvlist is properties to set. Currently only user properties
252 * are supported. { user:prop_name -> string value }
254 * The returned results nvlist will have an entry for each snapshot that failed.
255 * The value will be the (int32) error code.
257 * The return value will be 0 if all snapshots were created, otherwise it will
258 * be the errno of a (unspecified) snapshot that failed.
261 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
266 char pool[MAXNAMELEN];
270 /* determine the pool name */
271 elem = nvlist_next_nvpair(snaps, NULL);
274 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
275 pool[strcspn(pool, "/@")] = '\0';
277 args = fnvlist_alloc();
278 fnvlist_add_nvlist(args, "snaps", snaps);
280 fnvlist_add_nvlist(args, "props", props);
282 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
289 * Destroys snapshots.
291 * The keys in the snaps nvlist are the snapshots to be destroyed.
292 * They must all be in the same pool.
294 * Snapshots that do not exist will be silently ignored.
296 * If 'defer' is not set, and a snapshot has user holds or clones, the
297 * destroy operation will fail and none of the snapshots will be
300 * If 'defer' is set, and a snapshot has user holds or clones, it will be
301 * marked for deferred destruction, and will be destroyed when the last hold
302 * or clone is removed/destroyed.
304 * The return value will be 0 if all snapshots were destroyed (or marked for
305 * later destruction if 'defer' is set) or didn't exist to begin with.
307 * Otherwise the return value will be the errno of a (unspecified) snapshot
308 * that failed, no snapshots will be destroyed, and the errlist will have an
309 * entry for each snapshot that failed. The value in the errlist will be
310 * the (int32) error code.
313 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
318 char pool[MAXNAMELEN];
320 /* determine the pool name */
321 elem = nvlist_next_nvpair(snaps, NULL);
324 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
325 pool[strcspn(pool, "/@")] = '\0';
327 args = fnvlist_alloc();
328 fnvlist_add_nvlist(args, "snaps", snaps);
330 fnvlist_add_boolean(args, "defer");
332 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
340 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
349 /* determine the fs name */
350 (void) strlcpy(fs, firstsnap, sizeof (fs));
351 atp = strchr(fs, '@');
356 args = fnvlist_alloc();
357 fnvlist_add_string(args, "firstsnap", firstsnap);
359 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
362 *usedp = fnvlist_lookup_uint64(result, "used");
363 fnvlist_free(result);
369 lzc_exists(const char *dataset)
372 * The objset_stats ioctl is still legacy, so we need to construct our
373 * own zfs_cmd_t rather than using zfsc_ioctl().
375 zfs_cmd_t zc = { 0 };
377 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
378 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
382 * Create "user holds" on snapshots. If there is a hold on a snapshot,
383 * the snapshot can not be destroyed. (However, it can be marked for deletion
384 * by lzc_destroy_snaps(defer=B_TRUE).)
386 * The keys in the nvlist are snapshot names.
387 * The snapshots must all be in the same pool.
388 * The value is the name of the hold (string type).
390 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
391 * In this case, when the cleanup_fd is closed (including on process
392 * termination), the holds will be released. If the system is shut down
393 * uncleanly, the holds will be released when the pool is next opened
396 * The return value will be 0 if all holds were created. Otherwise the return
397 * value will be the errno of a (unspecified) hold that failed, no holds will
398 * be created, and the errlist will have an entry for each hold that
399 * failed (name = snapshot). The value in the errlist will be the error
403 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
405 char pool[MAXNAMELEN];
410 /* determine the pool name */
411 elem = nvlist_next_nvpair(holds, NULL);
414 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
415 pool[strcspn(pool, "/@")] = '\0';
417 args = fnvlist_alloc();
418 fnvlist_add_nvlist(args, "holds", holds);
419 if (cleanup_fd != -1)
420 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
422 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
428 * Release "user holds" on snapshots. If the snapshot has been marked for
429 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
430 * any clones, and all the user holds are removed, then the snapshot will be
433 * The keys in the nvlist are snapshot names.
434 * The snapshots must all be in the same pool.
435 * The value is a nvlist whose keys are the holds to remove.
437 * The return value will be 0 if all holds were removed.
438 * Otherwise the return value will be the errno of a (unspecified) release
439 * that failed, no holds will be released, and the errlist will have an
440 * entry for each snapshot that has failed releases (name = snapshot).
441 * The value in the errlist will be the error code (int32) of a failed release.
444 lzc_release(nvlist_t *holds, nvlist_t **errlist)
446 char pool[MAXNAMELEN];
449 /* determine the pool name */
450 elem = nvlist_next_nvpair(holds, NULL);
453 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
454 pool[strcspn(pool, "/@")] = '\0';
456 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
460 * Retrieve list of user holds on the specified snapshot.
462 * On success, *holdsp will be set to a nvlist which the caller must free.
463 * The keys are the names of the holds, and the value is the creation time
464 * of the hold (uint64) in seconds since the epoch.
467 lzc_get_holds(const char *snapname, nvlist_t **holdsp)
470 nvlist_t *innvl = fnvlist_alloc();
471 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp);
477 * If fromsnap is NULL, a full (non-incremental) stream will be sent.
480 lzc_send(const char *snapname, const char *fromsnap, int fd)
485 args = fnvlist_alloc();
486 fnvlist_add_int32(args, "fd", fd);
487 if (fromsnap != NULL)
488 fnvlist_add_string(args, "fromsnap", fromsnap);
489 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
495 * If fromsnap is NULL, a full (non-incremental) stream will be estimated.
498 lzc_send_space(const char *snapname, const char *fromsnap, uint64_t *spacep)
504 args = fnvlist_alloc();
505 if (fromsnap != NULL)
506 fnvlist_add_string(args, "fromsnap", fromsnap);
507 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
510 *spacep = fnvlist_lookup_uint64(result, "space");
516 recv_read(int fd, void *buf, int ilen)
523 rv = read(fd, cp, len);
528 if (rv < 0 || len != 0)
535 * The simplest receive case: receive from the specified fd, creating the
536 * specified snapshot. Apply the specified properties a "received" properties
537 * (which can be overridden by locally-set properties). If the stream is a
538 * clone, its origin snapshot must be specified by 'origin'. The 'force'
539 * flag will cause the target filesystem to be rolled back or destroyed if
540 * necessary to receive.
542 * Return 0 on success or an errno on failure.
544 * Note: this interface does not work on dedup'd streams
545 * (those with DMU_BACKUP_FEATURE_DEDUP).
548 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
549 boolean_t force, int fd)
552 * The receive ioctl is still legacy, so we need to construct our own
553 * zfs_cmd_t rather than using zfsc_ioctl().
555 zfs_cmd_t zc = { 0 };
559 dmu_replay_record_t drr;
562 ASSERT3S(g_refcount, >, 0);
564 /* zc_name is name of containing filesystem */
565 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
566 atp = strchr(zc.zc_name, '@');
571 /* if the fs does not exist, try its parent. */
572 if (!lzc_exists(zc.zc_name)) {
573 char *slashp = strrchr(zc.zc_name, '/');
580 /* zc_value is full name of the snapshot to create */
581 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
584 /* zc_nvlist_src is props to set */
585 packed = fnvlist_pack(props, &size);
586 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
587 zc.zc_nvlist_src_size = size;
590 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
592 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
594 /* zc_begin_record is non-byteswapped BEGIN record */
595 error = recv_read(fd, &drr, sizeof (drr));
598 zc.zc_begin_record = drr.drr_u.drr_begin;
600 /* zc_cookie is fd to read from */
603 /* zc guid is force flag */
606 /* zc_cleanup_fd is unused */
607 zc.zc_cleanup_fd = -1;
609 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
615 fnvlist_pack_free(packed, size);
616 free((void*)(uintptr_t)zc.zc_nvlist_dst);