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 int lzc_ioctl_version = -1;
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 (lzc_ioctl_version == -1)
144 lzc_ioctl_version = get_zfs_ioctl_version();
146 if (lzc_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) {
159 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
160 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
161 malloc(zc.zc_nvlist_dst_size);
163 if (zc.zc_nvlist_dst == NULL) {
165 if (zc.zc_nvlist_dst == 0) {
172 while (ioctl(g_fd, ioc, &zc) != 0) {
173 if (errno == ENOMEM && resultp != NULL) {
174 free((void *)(uintptr_t)zc.zc_nvlist_dst);
175 zc.zc_nvlist_dst_size *= 2;
176 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
177 malloc(zc.zc_nvlist_dst_size);
179 if (zc.zc_nvlist_dst == NULL) {
181 if (zc.zc_nvlist_dst == 0) {
193 if (lzc_ioctl_version < ZFS_IOCVER_LZC)
194 lzc_compat_post(&zc, ioc);
196 if (zc.zc_nvlist_dst_filled) {
197 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
198 zc.zc_nvlist_dst_size);
199 } else if (resultp != NULL) {
203 if (lzc_ioctl_version < ZFS_IOCVER_LZC)
204 lzc_compat_outnvl(&zc, ioc, resultp);
208 if (lzc_ioctl_version < ZFS_IOCVER_LZC) {
209 if (source != oldsource)
214 fnvlist_pack_free(packed, size);
215 free((void *)(uintptr_t)zc.zc_nvlist_dst);
220 lzc_create(const char *fsname, dmu_objset_type_t type, nvlist_t *props)
223 nvlist_t *args = fnvlist_alloc();
224 fnvlist_add_int32(args, "type", type);
226 fnvlist_add_nvlist(args, "props", props);
227 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
233 lzc_clone(const char *fsname, const char *origin,
237 nvlist_t *args = fnvlist_alloc();
238 fnvlist_add_string(args, "origin", origin);
240 fnvlist_add_nvlist(args, "props", props);
241 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
249 * The keys in the snaps nvlist are the snapshots to be created.
250 * They must all be in the same pool.
252 * The props nvlist is properties to set. Currently only user properties
253 * are supported. { user:prop_name -> string value }
255 * The returned results nvlist will have an entry for each snapshot that failed.
256 * The value will be the (int32) error code.
258 * The return value will be 0 if all snapshots were created, otherwise it will
259 * be the errno of a (undetermined) snapshot that failed.
262 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
267 char pool[MAXNAMELEN];
271 /* determine the pool name */
272 elem = nvlist_next_nvpair(snaps, NULL);
275 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
276 pool[strcspn(pool, "/@")] = '\0';
278 args = fnvlist_alloc();
279 fnvlist_add_nvlist(args, "snaps", snaps);
281 fnvlist_add_nvlist(args, "props", props);
283 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
290 * Destroys snapshots.
292 * The keys in the snaps nvlist are the snapshots to be destroyed.
293 * They must all be in the same pool.
295 * Snapshots that do not exist will be silently ignored.
297 * If 'defer' is not set, and a snapshot has user holds or clones, the
298 * destroy operation will fail and none of the snapshots will be
301 * If 'defer' is set, and a snapshot has user holds or clones, it will be
302 * marked for deferred destruction, and will be destroyed when the last hold
303 * or clone is removed/destroyed.
305 * The return value will be 0 if all snapshots were destroyed (or marked for
306 * later destruction if 'defer' is set) or didn't exist to begin with.
308 * Otherwise the return value will be the errno of a (undetermined) snapshot
309 * that failed, no snapshots will be destroyed, and the errlist will have an
310 * entry for each snapshot that failed. The value in the errlist will be
311 * the (int32) error code.
314 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
319 char pool[MAXNAMELEN];
321 /* determine the pool name */
322 elem = nvlist_next_nvpair(snaps, NULL);
325 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
326 pool[strcspn(pool, "/@")] = '\0';
328 args = fnvlist_alloc();
329 fnvlist_add_nvlist(args, "snaps", snaps);
331 fnvlist_add_boolean(args, "defer");
333 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
341 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
350 /* determine the fs name */
351 (void) strlcpy(fs, firstsnap, sizeof (fs));
352 atp = strchr(fs, '@');
357 args = fnvlist_alloc();
358 fnvlist_add_string(args, "firstsnap", firstsnap);
360 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
363 *usedp = fnvlist_lookup_uint64(result, "used");
364 fnvlist_free(result);
370 lzc_exists(const char *dataset)
373 * The objset_stats ioctl is still legacy, so we need to construct our
374 * own zfs_cmd_t rather than using zfsc_ioctl().
376 zfs_cmd_t zc = { 0 };
378 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
379 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
383 * If fromsnap is NULL, a full (non-incremental) stream will be sent.
386 lzc_send(const char *snapname, const char *fromsnap, int fd)
391 args = fnvlist_alloc();
392 fnvlist_add_int32(args, "fd", fd);
393 if (fromsnap != NULL)
394 fnvlist_add_string(args, "fromsnap", fromsnap);
395 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
401 * If fromsnap is NULL, a full (non-incremental) stream will be estimated.
404 lzc_send_space(const char *snapname, const char *fromsnap, uint64_t *spacep)
410 args = fnvlist_alloc();
411 if (fromsnap != NULL)
412 fnvlist_add_string(args, "fromsnap", fromsnap);
413 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
416 *spacep = fnvlist_lookup_uint64(result, "space");
422 recv_read(int fd, void *buf, int ilen)
429 rv = read(fd, cp, len);
434 if (rv < 0 || len != 0)
441 * The simplest receive case: receive from the specified fd, creating the
442 * specified snapshot. Apply the specified properties a "received" properties
443 * (which can be overridden by locally-set properties). If the stream is a
444 * clone, its origin snapshot must be specified by 'origin'. The 'force'
445 * flag will cause the target filesystem to be rolled back or destroyed if
446 * necessary to receive.
448 * Return 0 on success or an errno on failure.
450 * Note: this interface does not work on dedup'd streams
451 * (those with DMU_BACKUP_FEATURE_DEDUP).
454 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
455 boolean_t force, int fd)
458 * The receive ioctl is still legacy, so we need to construct our own
459 * zfs_cmd_t rather than using zfsc_ioctl().
461 zfs_cmd_t zc = { 0 };
465 dmu_replay_record_t drr;
468 ASSERT3S(g_refcount, >, 0);
470 /* zc_name is name of containing filesystem */
471 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
472 atp = strchr(zc.zc_name, '@');
477 /* if the fs does not exist, try its parent. */
478 if (!lzc_exists(zc.zc_name)) {
479 char *slashp = strrchr(zc.zc_name, '/');
486 /* zc_value is full name of the snapshot to create */
487 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
490 /* zc_nvlist_src is props to set */
491 packed = fnvlist_pack(props, &size);
492 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
493 zc.zc_nvlist_src_size = size;
496 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
498 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
500 /* zc_begin_record is non-byteswapped BEGIN record */
501 error = recv_read(fd, &drr, sizeof (drr));
504 zc.zc_begin_record = drr.drr_u.drr_begin;
506 /* zc_cookie is fd to read from */
509 /* zc guid is force flag */
512 /* zc_cleanup_fd is unused */
513 zc.zc_cleanup_fd = -1;
515 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
521 fnvlist_pack_free(packed, size);
522 free((void*)(uintptr_t)zc.zc_nvlist_dst);