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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
27 #include <sys/zfs_context.h>
28 #include <sys/dsl_userhold.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_destroy.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/zfs_onexit.h>
34 #include <sys/dsl_pool.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/zfs_ioctl.h>
39 typedef struct dsl_dataset_user_hold_arg {
40 nvlist_t *dduha_holds;
41 nvlist_t *dduha_chkholds;
42 nvlist_t *dduha_errlist;
44 } dsl_dataset_user_hold_arg_t;
47 * If you add new checks here, you may need to add additional checks to the
48 * "temporary" case in snapshot_check() in dmu_objset.c.
51 dsl_dataset_user_hold_check_one(dsl_dataset_t *ds, const char *htag,
52 boolean_t temphold, dmu_tx_t *tx)
54 dsl_pool_t *dp = dmu_tx_pool(tx);
55 objset_t *mos = dp->dp_meta_objset;
58 ASSERT(dsl_pool_config_held(dp));
60 if (strlen(htag) > MAXNAMELEN)
61 return (SET_ERROR(E2BIG));
62 /* Tempholds have a more restricted length */
63 if (temphold && strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
64 return (SET_ERROR(E2BIG));
66 /* tags must be unique (if ds already exists) */
67 if (ds != NULL && dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
70 error = zap_lookup(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
73 error = SET_ERROR(EEXIST);
74 else if (error == ENOENT)
82 dsl_dataset_user_hold_check(void *arg, dmu_tx_t *tx)
84 dsl_dataset_user_hold_arg_t *dduha = arg;
85 dsl_pool_t *dp = dmu_tx_pool(tx);
87 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS)
88 return (SET_ERROR(ENOTSUP));
90 if (!dmu_tx_is_syncing(tx))
93 for (nvpair_t *pair = nvlist_next_nvpair(dduha->dduha_holds, NULL);
94 pair != NULL; pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
99 /* must be a snapshot */
100 name = nvpair_name(pair);
101 if (strchr(name, '@') == NULL)
102 error = SET_ERROR(EINVAL);
105 error = nvpair_value_string(pair, &htag);
108 error = dsl_dataset_hold(dp, name, FTAG, &ds);
111 error = dsl_dataset_user_hold_check_one(ds, htag,
112 dduha->dduha_minor != 0, tx);
113 dsl_dataset_rele(ds, FTAG);
117 fnvlist_add_string(dduha->dduha_chkholds, name, htag);
120 * We register ENOENT errors so they can be correctly
121 * reported if needed, such as when all holds fail.
123 fnvlist_add_int32(dduha->dduha_errlist, name, error);
134 dsl_dataset_user_hold_sync_one_impl(nvlist_t *tmpholds, dsl_dataset_t *ds,
135 const char *htag, minor_t minor, uint64_t now, dmu_tx_t *tx)
137 dsl_pool_t *dp = ds->ds_dir->dd_pool;
138 objset_t *mos = dp->dp_meta_objset;
141 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
143 if (dsl_dataset_phys(ds)->ds_userrefs_obj == 0) {
145 * This is the first user hold for this dataset. Create
146 * the userrefs zap object.
148 dmu_buf_will_dirty(ds->ds_dbuf, tx);
149 zapobj = dsl_dataset_phys(ds)->ds_userrefs_obj =
150 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
152 zapobj = dsl_dataset_phys(ds)->ds_userrefs_obj;
156 VERIFY0(zap_add(mos, zapobj, htag, 8, 1, &now, tx));
159 char name[MAXNAMELEN];
162 VERIFY0(dsl_pool_user_hold(dp, ds->ds_object,
164 (void) snprintf(name, sizeof (name), "%llx",
165 (u_longlong_t)ds->ds_object);
167 if (nvlist_lookup_nvlist(tmpholds, name, &tags) != 0) {
168 tags = fnvlist_alloc();
169 fnvlist_add_boolean(tags, htag);
170 fnvlist_add_nvlist(tmpholds, name, tags);
173 fnvlist_add_boolean(tags, htag);
177 spa_history_log_internal_ds(ds, "hold", tx,
178 "tag=%s temp=%d refs=%llu",
179 htag, minor != 0, ds->ds_userrefs);
182 typedef struct zfs_hold_cleanup_arg {
183 char zhca_spaname[MAXNAMELEN];
184 uint64_t zhca_spa_load_guid;
185 nvlist_t *zhca_holds;
186 } zfs_hold_cleanup_arg_t;
189 dsl_dataset_user_release_onexit(void *arg)
191 zfs_hold_cleanup_arg_t *ca = arg;
195 error = spa_open(ca->zhca_spaname, &spa, FTAG);
197 zfs_dbgmsg("couldn't release holds on pool=%s "
198 "because pool is no longer loaded",
202 if (spa_load_guid(spa) != ca->zhca_spa_load_guid) {
203 zfs_dbgmsg("couldn't release holds on pool=%s "
204 "because pool is no longer loaded (guid doesn't match)",
206 spa_close(spa, FTAG);
210 (void) dsl_dataset_user_release_tmp(spa_get_dsl(spa), ca->zhca_holds);
211 fnvlist_free(ca->zhca_holds);
212 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
213 spa_close(spa, FTAG);
217 dsl_onexit_hold_cleanup(spa_t *spa, nvlist_t *holds, minor_t minor)
219 zfs_hold_cleanup_arg_t *ca;
221 if (minor == 0 || nvlist_empty(holds)) {
227 ca = kmem_alloc(sizeof (*ca), KM_SLEEP);
229 (void) strlcpy(ca->zhca_spaname, spa_name(spa),
230 sizeof (ca->zhca_spaname));
231 ca->zhca_spa_load_guid = spa_load_guid(spa);
232 ca->zhca_holds = holds;
233 VERIFY0(zfs_onexit_add_cb(minor,
234 dsl_dataset_user_release_onexit, ca, NULL));
238 dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag,
239 minor_t minor, uint64_t now, dmu_tx_t *tx)
244 tmpholds = fnvlist_alloc();
247 dsl_dataset_user_hold_sync_one_impl(tmpholds, ds, htag, minor, now, tx);
248 dsl_onexit_hold_cleanup(dsl_dataset_get_spa(ds), tmpholds, minor);
252 dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx)
254 dsl_dataset_user_hold_arg_t *dduha = arg;
255 dsl_pool_t *dp = dmu_tx_pool(tx);
257 uint64_t now = gethrestime_sec();
259 if (dduha->dduha_minor != 0)
260 tmpholds = fnvlist_alloc();
263 for (nvpair_t *pair = nvlist_next_nvpair(dduha->dduha_chkholds, NULL);
265 pair = nvlist_next_nvpair(dduha->dduha_chkholds, pair)) {
268 VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
269 dsl_dataset_user_hold_sync_one_impl(tmpholds, ds,
270 fnvpair_value_string(pair), dduha->dduha_minor, now, tx);
271 dsl_dataset_rele(ds, FTAG);
273 dsl_onexit_hold_cleanup(dp->dp_spa, tmpholds, dduha->dduha_minor);
277 * The full semantics of this function are described in the comment above
281 * holds is nvl of snapname -> holdname
282 * errlist will be filled in with snapname -> error
284 * The snaphosts must all be in the same pool.
286 * Holds for snapshots that don't exist will be skipped.
288 * If none of the snapshots for requested holds exist then ENOENT will be
291 * If cleanup_minor is not 0, the holds will be temporary, which will be cleaned
292 * up when the process exits.
294 * On success all the holds, for snapshots that existed, will be created and 0
297 * On failure no holds will be created, the errlist will be filled in,
298 * and an errno will returned.
300 * In all cases the errlist will contain entries for holds where the snapshot
304 dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor, nvlist_t *errlist)
306 dsl_dataset_user_hold_arg_t dduha;
310 pair = nvlist_next_nvpair(holds, NULL);
314 dduha.dduha_holds = holds;
315 dduha.dduha_chkholds = fnvlist_alloc();
316 dduha.dduha_errlist = errlist;
317 dduha.dduha_minor = cleanup_minor;
319 ret = dsl_sync_task(nvpair_name(pair), dsl_dataset_user_hold_check,
320 dsl_dataset_user_hold_sync, &dduha,
321 fnvlist_num_pairs(holds), ZFS_SPACE_CHECK_RESERVED);
322 fnvlist_free(dduha.dduha_chkholds);
327 typedef int (dsl_holdfunc_t)(dsl_pool_t *dp, const char *name, void *tag,
328 dsl_dataset_t **dsp);
330 typedef struct dsl_dataset_user_release_arg {
331 dsl_holdfunc_t *ddura_holdfunc;
332 nvlist_t *ddura_holds;
333 nvlist_t *ddura_todelete;
334 nvlist_t *ddura_errlist;
335 nvlist_t *ddura_chkholds;
336 } dsl_dataset_user_release_arg_t;
338 /* Place a dataset hold on the snapshot identified by passed dsobj string */
340 dsl_dataset_hold_obj_string(dsl_pool_t *dp, const char *dsobj, void *tag,
343 return (dsl_dataset_hold_obj(dp, strtonum(dsobj, NULL), tag, dsp));
347 dsl_dataset_user_release_check_one(dsl_dataset_user_release_arg_t *ddura,
348 dsl_dataset_t *ds, nvlist_t *holds, const char *snapname)
351 nvlist_t *holds_found;
355 if (!dsl_dataset_is_snapshot(ds))
356 return (SET_ERROR(EINVAL));
358 if (nvlist_empty(holds))
362 mos = ds->ds_dir->dd_pool->dp_meta_objset;
363 zapobj = dsl_dataset_phys(ds)->ds_userrefs_obj;
364 holds_found = fnvlist_alloc();
366 for (nvpair_t *pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
367 pair = nvlist_next_nvpair(holds, pair)) {
370 const char *holdname = nvpair_name(pair);
373 error = zap_lookup(mos, zapobj, holdname, 8, 1, &tmp);
375 error = SET_ERROR(ENOENT);
378 * Non-existent holds are put on the errlist, but don't
379 * cause an overall failure.
381 if (error == ENOENT) {
382 if (ddura->ddura_errlist != NULL) {
383 char *errtag = kmem_asprintf("%s#%s",
385 fnvlist_add_int32(ddura->ddura_errlist, errtag,
393 fnvlist_free(holds_found);
397 fnvlist_add_boolean(holds_found, holdname);
401 if (DS_IS_DEFER_DESTROY(ds) &&
402 dsl_dataset_phys(ds)->ds_num_children == 1 &&
403 ds->ds_userrefs == numholds) {
404 /* we need to destroy the snapshot as well */
405 if (dsl_dataset_long_held(ds)) {
406 fnvlist_free(holds_found);
407 return (SET_ERROR(EBUSY));
409 fnvlist_add_boolean(ddura->ddura_todelete, snapname);
413 fnvlist_add_nvlist(ddura->ddura_chkholds, snapname,
416 fnvlist_free(holds_found);
422 dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx)
424 dsl_dataset_user_release_arg_t *ddura;
425 dsl_holdfunc_t *holdfunc;
428 if (!dmu_tx_is_syncing(tx))
431 dp = dmu_tx_pool(tx);
433 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
436 holdfunc = ddura->ddura_holdfunc;
438 for (nvpair_t *pair = nvlist_next_nvpair(ddura->ddura_holds, NULL);
439 pair != NULL; pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
443 const char *snapname = nvpair_name(pair);
445 error = nvpair_value_nvlist(pair, &holds);
447 error = (SET_ERROR(EINVAL));
449 error = holdfunc(dp, snapname, FTAG, &ds);
451 error = dsl_dataset_user_release_check_one(ddura, ds,
453 dsl_dataset_rele(ds, FTAG);
456 if (ddura->ddura_errlist != NULL) {
457 fnvlist_add_int32(ddura->ddura_errlist,
461 * Non-existent snapshots are put on the errlist,
462 * but don't cause an overall failure.
473 dsl_dataset_user_release_sync_one(dsl_dataset_t *ds, nvlist_t *holds,
476 dsl_pool_t *dp = ds->ds_dir->dd_pool;
477 objset_t *mos = dp->dp_meta_objset;
479 for (nvpair_t *pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
480 pair = nvlist_next_nvpair(holds, pair)) {
482 const char *holdname = nvpair_name(pair);
484 /* Remove temporary hold if one exists. */
485 error = dsl_pool_user_release(dp, ds->ds_object, holdname, tx);
486 VERIFY(error == 0 || error == ENOENT);
488 VERIFY0(zap_remove(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
492 spa_history_log_internal_ds(ds, "release", tx,
493 "tag=%s refs=%lld", holdname, (longlong_t)ds->ds_userrefs);
498 dsl_dataset_user_release_sync(void *arg, dmu_tx_t *tx)
500 dsl_dataset_user_release_arg_t *ddura = arg;
501 dsl_holdfunc_t *holdfunc = ddura->ddura_holdfunc;
502 dsl_pool_t *dp = dmu_tx_pool(tx);
504 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
506 for (nvpair_t *pair = nvlist_next_nvpair(ddura->ddura_chkholds, NULL);
507 pair != NULL; pair = nvlist_next_nvpair(ddura->ddura_chkholds,
510 const char *name = nvpair_name(pair);
512 VERIFY0(holdfunc(dp, name, FTAG, &ds));
514 dsl_dataset_user_release_sync_one(ds,
515 fnvpair_value_nvlist(pair), tx);
516 if (nvlist_exists(ddura->ddura_todelete, name)) {
517 ASSERT(ds->ds_userrefs == 0 &&
518 dsl_dataset_phys(ds)->ds_num_children == 1 &&
519 DS_IS_DEFER_DESTROY(ds));
520 dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
522 dsl_dataset_rele(ds, FTAG);
527 * The full semantics of this function are described in the comment above
531 * Releases holds specified in the nvl holds.
533 * holds is nvl of snapname -> { holdname, ... }
534 * errlist will be filled in with snapname -> error
536 * If tmpdp is not NULL the names for holds should be the dsobj's of snapshots,
537 * otherwise they should be the names of shapshots.
539 * As a release may cause snapshots to be destroyed this trys to ensure they
542 * The release of non-existent holds are skipped.
544 * At least one hold must have been released for the this function to succeed
548 dsl_dataset_user_release_impl(nvlist_t *holds, nvlist_t *errlist,
551 dsl_dataset_user_release_arg_t ddura;
556 pair = nvlist_next_nvpair(holds, NULL);
561 * The release may cause snapshots to be destroyed; make sure they
565 /* Temporary holds are specified by dsobj string. */
566 ddura.ddura_holdfunc = dsl_dataset_hold_obj_string;
567 pool = spa_name(tmpdp->dp_spa);
569 for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
570 pair = nvlist_next_nvpair(holds, pair)) {
573 dsl_pool_config_enter(tmpdp, FTAG);
574 error = dsl_dataset_hold_obj_string(tmpdp,
575 nvpair_name(pair), FTAG, &ds);
577 char name[MAXNAMELEN];
578 dsl_dataset_name(ds, name);
579 dsl_pool_config_exit(tmpdp, FTAG);
580 dsl_dataset_rele(ds, FTAG);
581 (void) zfs_unmount_snap(name);
583 dsl_pool_config_exit(tmpdp, FTAG);
588 /* Non-temporary holds are specified by name. */
589 ddura.ddura_holdfunc = dsl_dataset_hold;
590 pool = nvpair_name(pair);
592 for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
593 pair = nvlist_next_nvpair(holds, pair)) {
594 (void) zfs_unmount_snap(nvpair_name(pair));
599 ddura.ddura_holds = holds;
600 ddura.ddura_errlist = errlist;
601 ddura.ddura_todelete = fnvlist_alloc();
602 ddura.ddura_chkholds = fnvlist_alloc();
604 error = dsl_sync_task(pool, dsl_dataset_user_release_check,
605 dsl_dataset_user_release_sync, &ddura, 0, ZFS_SPACE_CHECK_NONE);
606 fnvlist_free(ddura.ddura_todelete);
607 fnvlist_free(ddura.ddura_chkholds);
613 * holds is nvl of snapname -> { holdname, ... }
614 * errlist will be filled in with snapname -> error
617 dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist)
619 return (dsl_dataset_user_release_impl(holds, errlist, NULL));
623 * holds is nvl of snapdsobj -> { holdname, ... }
626 dsl_dataset_user_release_tmp(struct dsl_pool *dp, nvlist_t *holds)
629 (void) dsl_dataset_user_release_impl(holds, NULL, dp);
633 dsl_dataset_get_holds(const char *dsname, nvlist_t *nvl)
639 err = dsl_pool_hold(dsname, FTAG, &dp);
642 err = dsl_dataset_hold(dp, dsname, FTAG, &ds);
644 dsl_pool_rele(dp, FTAG);
648 if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
652 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
653 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
654 dsl_dataset_phys(ds)->ds_userrefs_obj);
655 zap_cursor_retrieve(&zc, za) == 0;
656 zap_cursor_advance(&zc)) {
657 fnvlist_add_uint64(nvl, za->za_name,
658 za->za_first_integer);
660 zap_cursor_fini(&zc);
661 kmem_free(za, sizeof (zap_attribute_t));
663 dsl_dataset_rele(ds, FTAG);
664 dsl_pool_rele(dp, FTAG);