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.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
28 * LibZFS_Core (lzc) is intended to replace most functionality in libzfs.
29 * It has the following characteristics:
31 * - Thread Safe. libzfs_core is accessible concurrently from multiple
32 * threads. This is accomplished primarily by avoiding global data
33 * (e.g. caching). Since it's thread-safe, there is no reason for a
34 * process to have multiple libzfs "instances". Therefore, we store
35 * our few pieces of data (e.g. the file descriptor) in global
36 * variables. The fd is reference-counted so that the libzfs_core
37 * library can be "initialized" multiple times (e.g. by different
38 * consumers within the same process).
40 * - Committed Interface. The libzfs_core interface will be committed,
41 * therefore consumers can compile against it and be confident that
42 * their code will continue to work on future releases of this code.
43 * Currently, the interface is Evolving (not Committed), but we intend
44 * to commit to it once it is more complete and we determine that it
45 * meets the needs of all consumers.
47 * - Programatic Error Handling. libzfs_core communicates errors with
48 * defined error numbers, and doesn't print anything to stdout/stderr.
50 * - Thin Layer. libzfs_core is a thin layer, marshaling arguments
51 * to/from the kernel ioctls. There is generally a 1:1 correspondence
52 * between libzfs_core functions and ioctls to /dev/zfs.
54 * - Clear Atomicity. Because libzfs_core functions are generally 1:1
55 * with kernel ioctls, and kernel ioctls are general atomic, each
56 * libzfs_core function is atomic. For example, creating multiple
57 * snapshots with a single call to lzc_snapshot() is atomic -- it
58 * can't fail with only some of the requested snapshots created, even
59 * in the event of power loss or system crash.
61 * - Continued libzfs Support. Some higher-level operations (e.g.
62 * support for "zfs send -R") are too complicated to fit the scope of
63 * libzfs_core. This functionality will continue to live in libzfs.
64 * Where appropriate, libzfs will use the underlying atomic operations
65 * of libzfs_core. For example, libzfs may implement "zfs send -R |
66 * zfs receive" by using individual "send one snapshot", rename,
67 * destroy, and "receive one snapshot" operations in libzfs_core.
68 * /sbin/zfs and /zbin/zpool will link with both libzfs and
69 * libzfs_core. Other consumers should aim to use only libzfs_core,
70 * since that will be the supported, stable interface going forwards.
73 #include <libzfs_core.h>
81 #include <sys/nvpair.h>
82 #include <sys/param.h>
83 #include <sys/types.h>
85 #include <sys/zfs_ioctl.h>
88 static pthread_mutex_t g_lock = PTHREAD_MUTEX_INITIALIZER;
89 static int g_refcount;
92 libzfs_core_init(void)
94 (void) pthread_mutex_lock(&g_lock);
95 if (g_refcount == 0) {
96 g_fd = open("/dev/zfs", O_RDWR);
98 (void) pthread_mutex_unlock(&g_lock);
103 (void) pthread_mutex_unlock(&g_lock);
108 libzfs_core_fini(void)
110 (void) pthread_mutex_lock(&g_lock);
111 ASSERT3S(g_refcount, >, 0);
115 (void) pthread_mutex_unlock(&g_lock);
119 lzc_ioctl(zfs_ioc_t ioc, const char *name,
120 nvlist_t *source, nvlist_t **resultp)
122 zfs_cmd_t zc = {"\0"};
127 ASSERT3S(g_refcount, >, 0);
129 (void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
131 packed = fnvlist_pack(source, &size);
132 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
133 zc.zc_nvlist_src_size = size;
135 if (resultp != NULL) {
137 zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
138 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
139 malloc(zc.zc_nvlist_dst_size);
140 if (zc.zc_nvlist_dst == (uint64_t)0) {
146 while (ioctl(g_fd, ioc, &zc) != 0) {
147 if (errno == ENOMEM && resultp != NULL) {
148 free((void *)(uintptr_t)zc.zc_nvlist_dst);
149 zc.zc_nvlist_dst_size *= 2;
150 zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
151 malloc(zc.zc_nvlist_dst_size);
152 if (zc.zc_nvlist_dst == (uint64_t)0) {
161 if (zc.zc_nvlist_dst_filled) {
162 *resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
163 zc.zc_nvlist_dst_size);
167 fnvlist_pack_free(packed, size);
168 free((void *)(uintptr_t)zc.zc_nvlist_dst);
173 lzc_create(const char *fsname, dmu_objset_type_t type, nvlist_t *props)
176 nvlist_t *args = fnvlist_alloc();
177 fnvlist_add_int32(args, "type", type);
179 fnvlist_add_nvlist(args, "props", props);
180 error = lzc_ioctl(ZFS_IOC_CREATE, fsname, args, NULL);
186 lzc_clone(const char *fsname, const char *origin,
190 nvlist_t *args = fnvlist_alloc();
191 fnvlist_add_string(args, "origin", origin);
193 fnvlist_add_nvlist(args, "props", props);
194 error = lzc_ioctl(ZFS_IOC_CLONE, fsname, args, NULL);
202 * The keys in the snaps nvlist are the snapshots to be created.
203 * They must all be in the same pool.
205 * The props nvlist is properties to set. Currently only user properties
206 * are supported. { user:prop_name -> string value }
208 * The returned results nvlist will have an entry for each snapshot that failed.
209 * The value will be the (int32) error code.
211 * The return value will be 0 if all snapshots were created, otherwise it will
212 * be the errno of a (unspecified) snapshot that failed.
215 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
220 char pool[MAXNAMELEN];
224 /* determine the pool name */
225 elem = nvlist_next_nvpair(snaps, NULL);
228 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
229 pool[strcspn(pool, "/@")] = '\0';
231 args = fnvlist_alloc();
232 fnvlist_add_nvlist(args, "snaps", snaps);
234 fnvlist_add_nvlist(args, "props", props);
236 error = lzc_ioctl(ZFS_IOC_SNAPSHOT, pool, args, errlist);
243 * Destroys snapshots.
245 * The keys in the snaps nvlist are the snapshots to be destroyed.
246 * They must all be in the same pool.
248 * Snapshots that do not exist will be silently ignored.
250 * If 'defer' is not set, and a snapshot has user holds or clones, the
251 * destroy operation will fail and none of the snapshots will be
254 * If 'defer' is set, and a snapshot has user holds or clones, it will be
255 * marked for deferred destruction, and will be destroyed when the last hold
256 * or clone is removed/destroyed.
258 * The return value will be ENOENT if none of the snapshots existed.
260 * The return value will be 0 if all snapshots were destroyed (or marked for
261 * later destruction if 'defer' is set) or didn't exist to begin with and
262 * at least one snapshot was destroyed.
264 * Otherwise the return value will be the errno of a (unspecified) snapshot
265 * that failed, no snapshots will be destroyed, and the errlist will have an
266 * entry for each snapshot that failed. The value in the errlist will be
267 * the (int32) error code.
270 lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist)
275 char pool[MAXNAMELEN];
277 /* determine the pool name */
278 elem = nvlist_next_nvpair(snaps, NULL);
281 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
282 pool[strcspn(pool, "/@")] = '\0';
284 args = fnvlist_alloc();
285 fnvlist_add_nvlist(args, "snaps", snaps);
287 fnvlist_add_boolean(args, "defer");
289 error = lzc_ioctl(ZFS_IOC_DESTROY_SNAPS, pool, args, errlist);
296 lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
305 /* determine the fs name */
306 (void) strlcpy(fs, firstsnap, sizeof (fs));
307 atp = strchr(fs, '@');
312 args = fnvlist_alloc();
313 fnvlist_add_string(args, "firstsnap", firstsnap);
315 err = lzc_ioctl(ZFS_IOC_SPACE_SNAPS, lastsnap, args, &result);
318 *usedp = fnvlist_lookup_uint64(result, "used");
319 fnvlist_free(result);
325 lzc_exists(const char *dataset)
328 * The objset_stats ioctl is still legacy, so we need to construct our
329 * own zfs_cmd_t rather than using zfsc_ioctl().
331 zfs_cmd_t zc = {"\0"};
333 (void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
334 return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
338 * Create "user holds" on snapshots. If there is a hold on a snapshot,
339 * the snapshot can not be destroyed. (However, it can be marked for deletion
340 * by lzc_destroy_snaps(defer=B_TRUE).)
342 * The keys in the nvlist are snapshot names.
343 * The snapshots must all be in the same pool.
344 * The value is the name of the hold (string type).
346 * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
347 * In this case, when the cleanup_fd is closed (including on process
348 * termination), the holds will be released. If the system is shut down
349 * uncleanly, the holds will be released when the pool is next opened
352 * Holds for snapshots which don't exist will be skipped and have an entry
353 * added to errlist, but will not cause an overall failure, except in the
354 * case that all holds where skipped.
356 * The return value will be ENOENT if none of the snapshots for the requested
359 * The return value will be 0 if the nvl holds was empty or all holds, for
360 * snapshots that existed, were succesfully created and at least one hold
363 * Otherwise the return value will be the errno of a (unspecified) hold that
364 * failed and no holds will be created.
366 * In all cases the errlist will have an entry for each hold that failed
367 * (name = snapshot), with its value being the error code (int32).
370 lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
372 char pool[MAXNAMELEN];
377 /* determine the pool name */
378 elem = nvlist_next_nvpair(holds, NULL);
381 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
382 pool[strcspn(pool, "/@")] = '\0';
384 args = fnvlist_alloc();
385 fnvlist_add_nvlist(args, "holds", holds);
386 if (cleanup_fd != -1)
387 fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
389 error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
395 * Release "user holds" on snapshots. If the snapshot has been marked for
396 * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
397 * any clones, and all the user holds are removed, then the snapshot will be
400 * The keys in the nvlist are snapshot names.
401 * The snapshots must all be in the same pool.
402 * The value is a nvlist whose keys are the holds to remove.
404 * Holds which failed to release because they didn't exist will have an entry
405 * added to errlist, but will not cause an overall failure, except in the
406 * case that all releases where skipped.
408 * The return value will be ENOENT if none of the specified holds existed.
410 * The return value will be 0 if the nvl holds was empty or all holds that
411 * existed, were successfully removed and at least one hold was removed.
413 * Otherwise the return value will be the errno of a (unspecified) hold that
414 * failed to release and no holds will be released.
416 * In all cases the errlist will have an entry for each hold that failed to
420 lzc_release(nvlist_t *holds, nvlist_t **errlist)
422 char pool[MAXNAMELEN];
425 /* determine the pool name */
426 elem = nvlist_next_nvpair(holds, NULL);
429 (void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
430 pool[strcspn(pool, "/@")] = '\0';
432 return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
436 * Retrieve list of user holds on the specified snapshot.
438 * On success, *holdsp will be set to a nvlist which the caller must free.
439 * The keys are the names of the holds, and the value is the creation time
440 * of the hold (uint64) in seconds since the epoch.
443 lzc_get_holds(const char *snapname, nvlist_t **holdsp)
446 nvlist_t *innvl = fnvlist_alloc();
447 error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp);
453 * If fromsnap is NULL, a full (non-incremental) stream will be sent.
456 lzc_send(const char *snapname, const char *fromsnap, int fd)
461 args = fnvlist_alloc();
462 fnvlist_add_int32(args, "fd", fd);
463 if (fromsnap != NULL)
464 fnvlist_add_string(args, "fromsnap", fromsnap);
465 err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
471 * If fromsnap is NULL, a full (non-incremental) stream will be estimated.
474 lzc_send_space(const char *snapname, const char *fromsnap, uint64_t *spacep)
480 args = fnvlist_alloc();
481 if (fromsnap != NULL)
482 fnvlist_add_string(args, "fromsnap", fromsnap);
483 err = lzc_ioctl(ZFS_IOC_SEND_SPACE, snapname, args, &result);
486 *spacep = fnvlist_lookup_uint64(result, "space");
492 recv_read(int fd, void *buf, int ilen)
499 rv = read(fd, cp, len);
504 if (rv < 0 || len != 0)
511 * The simplest receive case: receive from the specified fd, creating the
512 * specified snapshot. Apply the specified properties a "received" properties
513 * (which can be overridden by locally-set properties). If the stream is a
514 * clone, its origin snapshot must be specified by 'origin'. The 'force'
515 * flag will cause the target filesystem to be rolled back or destroyed if
516 * necessary to receive.
518 * Return 0 on success or an errno on failure.
520 * Note: this interface does not work on dedup'd streams
521 * (those with DMU_BACKUP_FEATURE_DEDUP).
524 lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
525 boolean_t force, int fd)
528 * The receive ioctl is still legacy, so we need to construct our own
529 * zfs_cmd_t rather than using zfsc_ioctl().
531 zfs_cmd_t zc = {"\0"};
535 dmu_replay_record_t drr;
538 ASSERT3S(g_refcount, >, 0);
540 /* zc_name is name of containing filesystem */
541 (void) strlcpy(zc.zc_name, snapname, sizeof (zc.zc_name));
542 atp = strchr(zc.zc_name, '@');
547 /* if the fs does not exist, try its parent. */
548 if (!lzc_exists(zc.zc_name)) {
549 char *slashp = strrchr(zc.zc_name, '/');
556 /* zc_value is full name of the snapshot to create */
557 (void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
560 /* zc_nvlist_src is props to set */
561 packed = fnvlist_pack(props, &size);
562 zc.zc_nvlist_src = (uint64_t)(uintptr_t)packed;
563 zc.zc_nvlist_src_size = size;
566 /* zc_string is name of clone origin (if DRR_FLAG_CLONE) */
568 (void) strlcpy(zc.zc_string, origin, sizeof (zc.zc_string));
570 /* zc_begin_record is non-byteswapped BEGIN record */
571 error = recv_read(fd, &drr, sizeof (drr));
574 zc.zc_begin_record = drr.drr_u.drr_begin;
576 /* zc_cookie is fd to read from */
579 /* zc guid is force flag */
582 /* zc_cleanup_fd is unused */
583 zc.zc_cleanup_fd = -1;
585 error = ioctl(g_fd, ZFS_IOC_RECV, &zc);
591 fnvlist_pack_free(packed, size);
592 free((void*)(uintptr_t)zc.zc_nvlist_dst);