1 /*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
25 */
26
27 #include <sys/dmu.h>
28 #include <sys/dmu_objset.h>
29 #include <sys/dmu_tx.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dsl_deleg.h>
35 #include <sys/spa.h>
36 #include <sys/metaslab.h>
37 #include <sys/zap.h>
38 #include <sys/zio.h>
39 #include <sys/arc.h>
40 #include <sys/sunddi.h>
41 #include "zfs_namecheck.h"
42
43 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
44
45 /* ARGSUSED */
46 static void
47 dsl_dir_evict(dmu_buf_t *db, void *arg)
48 {
49 dsl_dir_t *dd = arg;
50 dsl_pool_t *dp = dd->dd_pool;
51 int t;
52
53 for (t = 0; t < TXG_SIZE; t++) {
54 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
55 ASSERT(dd->dd_tempreserved[t] == 0);
56 ASSERT(dd->dd_space_towrite[t] == 0);
57 }
58
59 if (dd->dd_parent)
60 dsl_dir_rele(dd->dd_parent, dd);
61
62 spa_close(dd->dd_pool->dp_spa, dd);
63
64 /*
65 * The props callback list should have been cleaned up by
66 * objset_evict().
67 */
68 list_destroy(&dd->dd_prop_cbs);
69 mutex_destroy(&dd->dd_lock);
70 kmem_free(dd, sizeof (dsl_dir_t));
71 }
72
73 int
74 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
75 const char *tail, void *tag, dsl_dir_t **ddp)
76 {
77 dmu_buf_t *dbuf;
78 dsl_dir_t *dd;
79 int err;
80
81 ASSERT(dsl_pool_config_held(dp));
82
83 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
84 if (err != 0)
85 return (err);
86 dd = dmu_buf_get_user(dbuf);
87 #ifdef ZFS_DEBUG
88 {
89 dmu_object_info_t doi;
90 dmu_object_info_from_db(dbuf, &doi);
91 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
92 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
93 }
94 #endif
95 if (dd == NULL) {
96 dsl_dir_t *winner;
97
98 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
99 dd->dd_object = ddobj;
100 dd->dd_dbuf = dbuf;
101 dd->dd_pool = dp;
102 dd->dd_phys = dbuf->db_data;
103 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
104
105 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
106 offsetof(dsl_prop_cb_record_t, cbr_node));
107
108 dsl_dir_snap_cmtime_update(dd);
109
110 if (dd->dd_phys->dd_parent_obj) {
111 err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
112 NULL, dd, &dd->dd_parent);
113 if (err != 0)
114 goto errout;
115 if (tail) {
116 #ifdef ZFS_DEBUG
117 uint64_t foundobj;
118
119 err = zap_lookup(dp->dp_meta_objset,
120 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
121 tail, sizeof (foundobj), 1, &foundobj);
122 ASSERT(err || foundobj == ddobj);
123 #endif
124 (void) strcpy(dd->dd_myname, tail);
125 } else {
126 err = zap_value_search(dp->dp_meta_objset,
127 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
128 ddobj, 0, dd->dd_myname);
129 }
130 if (err != 0)
131 goto errout;
132 } else {
133 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
134 }
135
136 if (dsl_dir_is_clone(dd)) {
137 dmu_buf_t *origin_bonus;
138 dsl_dataset_phys_t *origin_phys;
139
140 /*
141 * We can't open the origin dataset, because
142 * that would require opening this dsl_dir.
143 * Just look at its phys directly instead.
144 */
145 err = dmu_bonus_hold(dp->dp_meta_objset,
146 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
147 if (err != 0)
148 goto errout;
149 origin_phys = origin_bonus->db_data;
150 dd->dd_origin_txg =
151 origin_phys->ds_creation_txg;
152 dmu_buf_rele(origin_bonus, FTAG);
153 }
154
155 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
156 dsl_dir_evict);
157 if (winner) {
158 if (dd->dd_parent)
159 dsl_dir_rele(dd->dd_parent, dd);
160 mutex_destroy(&dd->dd_lock);
161 kmem_free(dd, sizeof (dsl_dir_t));
162 dd = winner;
163 } else {
164 spa_open_ref(dp->dp_spa, dd);
165 }
166 }
167
168 /*
169 * The dsl_dir_t has both open-to-close and instantiate-to-evict
170 * holds on the spa. We need the open-to-close holds because
171 * otherwise the spa_refcnt wouldn't change when we open a
172 * dir which the spa also has open, so we could incorrectly
173 * think it was OK to unload/export/destroy the pool. We need
174 * the instantiate-to-evict hold because the dsl_dir_t has a
175 * pointer to the dd_pool, which has a pointer to the spa_t.
176 */
177 spa_open_ref(dp->dp_spa, tag);
178 ASSERT3P(dd->dd_pool, ==, dp);
179 ASSERT3U(dd->dd_object, ==, ddobj);
180 ASSERT3P(dd->dd_dbuf, ==, dbuf);
181 *ddp = dd;
182 return (0);
183
184 errout:
185 if (dd->dd_parent)
186 dsl_dir_rele(dd->dd_parent, dd);
187 mutex_destroy(&dd->dd_lock);
188 kmem_free(dd, sizeof (dsl_dir_t));
189 dmu_buf_rele(dbuf, tag);
190 return (err);
191 }
192
193 void
194 dsl_dir_rele(dsl_dir_t *dd, void *tag)
195 {
196 dprintf_dd(dd, "%s\n", "");
197 spa_close(dd->dd_pool->dp_spa, tag);
198 dmu_buf_rele(dd->dd_dbuf, tag);
199 }
200
201 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
202 void
203 dsl_dir_name(dsl_dir_t *dd, char *buf)
204 {
205 if (dd->dd_parent) {
206 dsl_dir_name(dd->dd_parent, buf);
207 (void) strcat(buf, "/");
208 } else {
209 buf[0] = '\0';
210 }
211 if (!MUTEX_HELD(&dd->dd_lock)) {
212 /*
213 * recursive mutex so that we can use
214 * dprintf_dd() with dd_lock held
215 */
216 mutex_enter(&dd->dd_lock);
217 (void) strcat(buf, dd->dd_myname);
218 mutex_exit(&dd->dd_lock);
219 } else {
220 (void) strcat(buf, dd->dd_myname);
221 }
222 }
223
224 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
225 int
226 dsl_dir_namelen(dsl_dir_t *dd)
227 {
228 int result = 0;
229
230 if (dd->dd_parent) {
231 /* parent's name + 1 for the "/" */
232 result = dsl_dir_namelen(dd->dd_parent) + 1;
233 }
234
235 if (!MUTEX_HELD(&dd->dd_lock)) {
236 /* see dsl_dir_name */
237 mutex_enter(&dd->dd_lock);
238 result += strlen(dd->dd_myname);
239 mutex_exit(&dd->dd_lock);
240 } else {
241 result += strlen(dd->dd_myname);
242 }
243
244 return (result);
245 }
246
247 static int
248 getcomponent(const char *path, char *component, const char **nextp)
249 {
250 char *p;
251
252 if ((path == NULL) || (path[0] == '\0'))
253 return (SET_ERROR(ENOENT));
254 /* This would be a good place to reserve some namespace... */
255 p = strpbrk(path, "/@");
256 if (p && (p[1] == '/' || p[1] == '@')) {
257 /* two separators in a row */
258 return (SET_ERROR(EINVAL));
259 }
260 if (p == NULL || p == path) {
261 /*
262 * if the first thing is an @ or /, it had better be an
263 * @ and it had better not have any more ats or slashes,
264 * and it had better have something after the @.
265 */
266 if (p != NULL &&
267 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
268 return (SET_ERROR(EINVAL));
269 if (strlen(path) >= MAXNAMELEN)
270 return (SET_ERROR(ENAMETOOLONG));
271 (void) strcpy(component, path);
272 p = NULL;
273 } else if (p[0] == '/') {
274 if (p - path >= MAXNAMELEN)
275 return (SET_ERROR(ENAMETOOLONG));
276 (void) strncpy(component, path, p - path);
277 component[p - path] = '\0';
278 p++;
279 } else if (p[0] == '@') {
280 /*
281 * if the next separator is an @, there better not be
282 * any more slashes.
283 */
284 if (strchr(path, '/'))
285 return (SET_ERROR(EINVAL));
286 if (p - path >= MAXNAMELEN)
287 return (SET_ERROR(ENAMETOOLONG));
288 (void) strncpy(component, path, p - path);
289 component[p - path] = '\0';
290 } else {
291 panic("invalid p=%p", (void *)p);
292 }
293 *nextp = p;
294 return (0);
295 }
296
297 /*
298 * Return the dsl_dir_t, and possibly the last component which couldn't
299 * be found in *tail. The name must be in the specified dsl_pool_t. This
300 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
301 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
302 * (*tail)[0] == '@' means that the last component is a snapshot.
303 */
304 int
305 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
306 dsl_dir_t **ddp, const char **tailp)
307 {
308 char buf[MAXNAMELEN];
309 const char *spaname, *next, *nextnext = NULL;
310 int err;
311 dsl_dir_t *dd;
312 uint64_t ddobj;
313
314 err = getcomponent(name, buf, &next);
315 if (err != 0)
316 return (err);
317
318 /* Make sure the name is in the specified pool. */
319 spaname = spa_name(dp->dp_spa);
320 if (strcmp(buf, spaname) != 0)
321 return (SET_ERROR(EINVAL));
322
323 ASSERT(dsl_pool_config_held(dp));
324
325 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
326 if (err != 0) {
327 return (err);
328 }
329
330 while (next != NULL) {
331 dsl_dir_t *child_ds;
332 err = getcomponent(next, buf, &nextnext);
333 if (err != 0)
334 break;
335 ASSERT(next[0] != '\0');
336 if (next[0] == '@')
337 break;
338 dprintf("looking up %s in obj%lld\n",
339 buf, dd->dd_phys->dd_child_dir_zapobj);
340
341 err = zap_lookup(dp->dp_meta_objset,
342 dd->dd_phys->dd_child_dir_zapobj,
343 buf, sizeof (ddobj), 1, &ddobj);
344 if (err != 0) {
345 if (err == ENOENT)
346 err = 0;
347 break;
348 }
349
350 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
351 if (err != 0)
352 break;
353 dsl_dir_rele(dd, tag);
354 dd = child_ds;
355 next = nextnext;
356 }
357
358 if (err != 0) {
359 dsl_dir_rele(dd, tag);
360 return (err);
361 }
362
363 /*
364 * It's an error if there's more than one component left, or
365 * tailp==NULL and there's any component left.
366 */
367 if (next != NULL &&
368 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
369 /* bad path name */
370 dsl_dir_rele(dd, tag);
371 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
372 err = SET_ERROR(ENOENT);
373 }
374 if (tailp != NULL)
375 *tailp = next;
376 *ddp = dd;
377 return (err);
378 }
379
380 uint64_t
381 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
382 dmu_tx_t *tx)
383 {
384 objset_t *mos = dp->dp_meta_objset;
385 uint64_t ddobj;
386 dsl_dir_phys_t *ddphys;
387 dmu_buf_t *dbuf;
388
389 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
390 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
391 if (pds) {
392 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
393 name, sizeof (uint64_t), 1, &ddobj, tx));
394 } else {
395 /* it's the root dir */
396 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
397 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
398 }
399 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
400 dmu_buf_will_dirty(dbuf, tx);
401 ddphys = dbuf->db_data;
402
403 ddphys->dd_creation_time = gethrestime_sec();
404 if (pds)
405 ddphys->dd_parent_obj = pds->dd_object;
406 ddphys->dd_props_zapobj = zap_create(mos,
407 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
408 ddphys->dd_child_dir_zapobj = zap_create(mos,
409 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
410 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
411 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
412 dmu_buf_rele(dbuf, FTAG);
413
414 return (ddobj);
415 }
416
417 boolean_t
418 dsl_dir_is_clone(dsl_dir_t *dd)
419 {
420 return (dd->dd_phys->dd_origin_obj &&
421 (dd->dd_pool->dp_origin_snap == NULL ||
422 dd->dd_phys->dd_origin_obj !=
423 dd->dd_pool->dp_origin_snap->ds_object));
424 }
425
426 void
427 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
428 {
429 mutex_enter(&dd->dd_lock);
430 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
431 dd->dd_phys->dd_used_bytes);
432 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
433 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
434 dd->dd_phys->dd_reserved);
435 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
436 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
437 (dd->dd_phys->dd_uncompressed_bytes * 100 /
438 dd->dd_phys->dd_compressed_bytes));
439 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
440 dd->dd_phys->dd_uncompressed_bytes);
441 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
442 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
443 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
444 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
445 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
446 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
447 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
448 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
449 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
450 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
451 }
452 mutex_exit(&dd->dd_lock);
453
454 if (dsl_dir_is_clone(dd)) {
455 dsl_dataset_t *ds;
456 char buf[MAXNAMELEN];
457
458 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
459 dd->dd_phys->dd_origin_obj, FTAG, &ds));
460 dsl_dataset_name(ds, buf);
461 dsl_dataset_rele(ds, FTAG);
462 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
463 }
464 }
465
466 void
467 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
468 {
469 dsl_pool_t *dp = dd->dd_pool;
470
471 ASSERT(dd->dd_phys);
472
473 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
474 /* up the hold count until we can be written out */
475 dmu_buf_add_ref(dd->dd_dbuf, dd);
476 }
477 }
478
479 static int64_t
480 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
481 {
482 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
483 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
484 return (new_accounted - old_accounted);
485 }
486
487 void
488 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
489 {
490 ASSERT(dmu_tx_is_syncing(tx));
491
492 mutex_enter(&dd->dd_lock);
493 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
494 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
495 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
496 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
497 mutex_exit(&dd->dd_lock);
498
499 /* release the hold from dsl_dir_dirty */
500 dmu_buf_rele(dd->dd_dbuf, dd);
501 }
502
503 static uint64_t
504 dsl_dir_space_towrite(dsl_dir_t *dd)
505 {
506 uint64_t space = 0;
507 int i;
508
509 ASSERT(MUTEX_HELD(&dd->dd_lock));
510
511 for (i = 0; i < TXG_SIZE; i++) {
512 space += dd->dd_space_towrite[i&TXG_MASK];
513 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
514 }
515 return (space);
516 }
517
518 /*
519 * How much space would dd have available if ancestor had delta applied
520 * to it? If ondiskonly is set, we're only interested in what's
521 * on-disk, not estimated pending changes.
522 */
523 uint64_t
524 dsl_dir_space_available(dsl_dir_t *dd,
525 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
526 {
527 uint64_t parentspace, myspace, quota, used;
528
529 /*
530 * If there are no restrictions otherwise, assume we have
531 * unlimited space available.
532 */
533 quota = UINT64_MAX;
534 parentspace = UINT64_MAX;
535
536 if (dd->dd_parent != NULL) {
537 parentspace = dsl_dir_space_available(dd->dd_parent,
538 ancestor, delta, ondiskonly);
539 }
540
541 mutex_enter(&dd->dd_lock);
542 if (dd->dd_phys->dd_quota != 0)
543 quota = dd->dd_phys->dd_quota;
544 used = dd->dd_phys->dd_used_bytes;
545 if (!ondiskonly)
546 used += dsl_dir_space_towrite(dd);
547
548 if (dd->dd_parent == NULL) {
549 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
550 quota = MIN(quota, poolsize);
551 }
552
553 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
554 /*
555 * We have some space reserved, in addition to what our
556 * parent gave us.
557 */
558 parentspace += dd->dd_phys->dd_reserved - used;
559 }
560
561 if (dd == ancestor) {
562 ASSERT(delta <= 0);
563 ASSERT(used >= -delta);
564 used += delta;
565 if (parentspace != UINT64_MAX)
566 parentspace -= delta;
567 }
568
569 if (used > quota) {
570 /* over quota */
571 myspace = 0;
572 } else {
573 /*
574 * the lesser of the space provided by our parent and
575 * the space left in our quota
576 */
577 myspace = MIN(parentspace, quota - used);
578 }
579
580 mutex_exit(&dd->dd_lock);
581
582 return (myspace);
583 }
584
585 struct tempreserve {
586 list_node_t tr_node;
587 dsl_pool_t *tr_dp;
588 dsl_dir_t *tr_ds;
589 uint64_t tr_size;
590 };
591
592 static int
593 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
594 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
595 dmu_tx_t *tx, boolean_t first)
596 {
597 uint64_t txg = tx->tx_txg;
598 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
599 uint64_t deferred = 0;
600 struct tempreserve *tr;
601 int retval = EDQUOT;
602 int txgidx = txg & TXG_MASK;
603 int i;
604 uint64_t ref_rsrv = 0;
605
606 ASSERT3U(txg, !=, 0);
607 ASSERT3S(asize, >, 0);
608
609 mutex_enter(&dd->dd_lock);
610
611 /*
612 * Check against the dsl_dir's quota. We don't add in the delta
613 * when checking for over-quota because they get one free hit.
614 */
615 est_inflight = dsl_dir_space_towrite(dd);
616 for (i = 0; i < TXG_SIZE; i++)
617 est_inflight += dd->dd_tempreserved[i];
618 used_on_disk = dd->dd_phys->dd_used_bytes;
619
620 /*
621 * On the first iteration, fetch the dataset's used-on-disk and
622 * refreservation values. Also, if checkrefquota is set, test if
623 * allocating this space would exceed the dataset's refquota.
624 */
625 if (first && tx->tx_objset) {
626 int error;
627 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
628
629 error = dsl_dataset_check_quota(ds, checkrefquota,
630 asize, est_inflight, &used_on_disk, &ref_rsrv);
631 if (error) {
632 mutex_exit(&dd->dd_lock);
633 return (error);
634 }
635 }
636
637 /*
638 * If this transaction will result in a net free of space,
639 * we want to let it through.
640 */
641 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
642 quota = UINT64_MAX;
643 else
644 quota = dd->dd_phys->dd_quota;
645
646 /*
647 * Adjust the quota against the actual pool size at the root
648 * minus any outstanding deferred frees.
649 * To ensure that it's possible to remove files from a full
650 * pool without inducing transient overcommits, we throttle
651 * netfree transactions against a quota that is slightly larger,
652 * but still within the pool's allocation slop. In cases where
653 * we're very close to full, this will allow a steady trickle of
654 * removes to get through.
655 */
656 if (dd->dd_parent == NULL) {
657 spa_t *spa = dd->dd_pool->dp_spa;
658 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
659 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
660 if (poolsize - deferred < quota) {
661 quota = poolsize - deferred;
662 retval = ENOSPC;
663 }
664 }
665
666 /*
667 * If they are requesting more space, and our current estimate
668 * is over quota, they get to try again unless the actual
669 * on-disk is over quota and there are no pending changes (which
670 * may free up space for us).
671 */
672 if (used_on_disk + est_inflight >= quota) {
673 if (est_inflight > 0 || used_on_disk < quota ||
674 (retval == ENOSPC && used_on_disk < quota + deferred))
675 retval = ERESTART;
676 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
677 "quota=%lluK tr=%lluK err=%d\n",
678 used_on_disk>>10, est_inflight>>10,
679 quota>>10, asize>>10, retval);
680 mutex_exit(&dd->dd_lock);
681 return (SET_ERROR(retval));
682 }
683
684 /* We need to up our estimated delta before dropping dd_lock */
685 dd->dd_tempreserved[txgidx] += asize;
686
687 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
688 asize - ref_rsrv);
689 mutex_exit(&dd->dd_lock);
690
691 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
692 tr->tr_ds = dd;
693 tr->tr_size = asize;
694 list_insert_tail(tr_list, tr);
695
696 /* see if it's OK with our parent */
697 if (dd->dd_parent && parent_rsrv) {
698 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
699
700 return (dsl_dir_tempreserve_impl(dd->dd_parent,
701 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
702 } else {
703 return (0);
704 }
705 }
706
707 /*
708 * Reserve space in this dsl_dir, to be used in this tx's txg.
709 * After the space has been dirtied (and dsl_dir_willuse_space()
710 * has been called), the reservation should be canceled, using
711 * dsl_dir_tempreserve_clear().
712 */
713 int
714 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
715 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
716 {
717 int err;
718 list_t *tr_list;
719
720 if (asize == 0) {
721 *tr_cookiep = NULL;
722 return (0);
723 }
724
725 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
726 list_create(tr_list, sizeof (struct tempreserve),
727 offsetof(struct tempreserve, tr_node));
728 ASSERT3S(asize, >, 0);
729 ASSERT3S(fsize, >=, 0);
730
731 err = arc_tempreserve_space(lsize, tx->tx_txg);
732 if (err == 0) {
733 struct tempreserve *tr;
734
735 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
736 tr->tr_size = lsize;
737 list_insert_tail(tr_list, tr);
738
739 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
740 } else {
741 if (err == EAGAIN) {
742 txg_delay(dd->dd_pool, tx->tx_txg,
743 MSEC2NSEC(10), MSEC2NSEC(10));
744 err = SET_ERROR(ERESTART);
745 }
746 dsl_pool_memory_pressure(dd->dd_pool);
747 }
748
749 if (err == 0) {
750 struct tempreserve *tr;
751
752 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
753 tr->tr_dp = dd->dd_pool;
754 tr->tr_size = asize;
755 list_insert_tail(tr_list, tr);
756
757 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
758 FALSE, asize > usize, tr_list, tx, TRUE);
759 }
760
761 if (err != 0)
762 dsl_dir_tempreserve_clear(tr_list, tx);
763 else
764 *tr_cookiep = tr_list;
765
766 return (err);
767 }
768
769 /*
770 * Clear a temporary reservation that we previously made with
771 * dsl_dir_tempreserve_space().
772 */
773 void
774 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
775 {
776 int txgidx = tx->tx_txg & TXG_MASK;
777 list_t *tr_list = tr_cookie;
778 struct tempreserve *tr;
779
780 ASSERT3U(tx->tx_txg, !=, 0);
781
782 if (tr_cookie == NULL)
783 return;
784
785 while (tr = list_head(tr_list)) {
786 if (tr->tr_dp) {
787 dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
788 } else if (tr->tr_ds) {
789 mutex_enter(&tr->tr_ds->dd_lock);
790 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
791 tr->tr_size);
792 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
793 mutex_exit(&tr->tr_ds->dd_lock);
794 } else {
795 arc_tempreserve_clear(tr->tr_size);
796 }
797 list_remove(tr_list, tr);
798 kmem_free(tr, sizeof (struct tempreserve));
799 }
800
801 kmem_free(tr_list, sizeof (list_t));
802 }
803
804 static void
805 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
806 {
807 int64_t parent_space;
808 uint64_t est_used;
809
810 mutex_enter(&dd->dd_lock);
811 if (space > 0)
812 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
813
814 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
815 parent_space = parent_delta(dd, est_used, space);
816 mutex_exit(&dd->dd_lock);
817
818 /* Make sure that we clean up dd_space_to* */
819 dsl_dir_dirty(dd, tx);
820
821 /* XXX this is potentially expensive and unnecessary... */
822 if (parent_space && dd->dd_parent)
823 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
824 }
825
826 /*
827 * Call in open context when we think we're going to write/free space,
828 * eg. when dirtying data. Be conservative (ie. OK to write less than
829 * this or free more than this, but don't write more or free less).
830 */
831 void
832 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
833 {
834 dsl_pool_willuse_space(dd->dd_pool, space, tx);
835 dsl_dir_willuse_space_impl(dd, space, tx);
836 }
837
838 /* call from syncing context when we actually write/free space for this dd */
839 void
840 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
841 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
842 {
843 int64_t accounted_delta;
844
845 /*
846 * dsl_dataset_set_refreservation_sync_impl() calls this with
847 * dd_lock held, so that it can atomically update
848 * ds->ds_reserved and the dsl_dir accounting, so that
849 * dsl_dataset_check_quota() can see dataset and dir accounting
850 * consistently.
851 */
852 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
853
854 ASSERT(dmu_tx_is_syncing(tx));
855 ASSERT(type < DD_USED_NUM);
856
857 dmu_buf_will_dirty(dd->dd_dbuf, tx);
858
859 if (needlock)
860 mutex_enter(&dd->dd_lock);
861 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
862 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
863 ASSERT(compressed >= 0 ||
864 dd->dd_phys->dd_compressed_bytes >= -compressed);
865 ASSERT(uncompressed >= 0 ||
866 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
867 dd->dd_phys->dd_used_bytes += used;
868 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
869 dd->dd_phys->dd_compressed_bytes += compressed;
870
871 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
872 ASSERT(used > 0 ||
873 dd->dd_phys->dd_used_breakdown[type] >= -used);
874 dd->dd_phys->dd_used_breakdown[type] += used;
875 #ifdef DEBUG
876 dd_used_t t;
877 uint64_t u = 0;
878 for (t = 0; t < DD_USED_NUM; t++)
879 u += dd->dd_phys->dd_used_breakdown[t];
880 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
881 #endif
882 }
883 if (needlock)
884 mutex_exit(&dd->dd_lock);
885
886 if (dd->dd_parent != NULL) {
887 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
888 accounted_delta, compressed, uncompressed, tx);
889 dsl_dir_transfer_space(dd->dd_parent,
890 used - accounted_delta,
891 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
892 }
893 }
894
895 void
896 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
897 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
898 {
899 ASSERT(dmu_tx_is_syncing(tx));
900 ASSERT(oldtype < DD_USED_NUM);
901 ASSERT(newtype < DD_USED_NUM);
902
903 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
904 return;
905
906 dmu_buf_will_dirty(dd->dd_dbuf, tx);
907 mutex_enter(&dd->dd_lock);
908 ASSERT(delta > 0 ?
909 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
910 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
911 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
912 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
913 dd->dd_phys->dd_used_breakdown[newtype] += delta;
914 mutex_exit(&dd->dd_lock);
915 }
916
917 typedef struct dsl_dir_set_qr_arg {
918 const char *ddsqra_name;
919 zprop_source_t ddsqra_source;
920 uint64_t ddsqra_value;
921 } dsl_dir_set_qr_arg_t;
922
923 static int
924 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
925 {
926 dsl_dir_set_qr_arg_t *ddsqra = arg;
927 dsl_pool_t *dp = dmu_tx_pool(tx);
928 dsl_dataset_t *ds;
929 int error;
930 uint64_t towrite, newval;
931
932 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
933 if (error != 0)
934 return (error);
935
936 error = dsl_prop_predict(ds->ds_dir, "quota",
937 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
938 if (error != 0) {
939 dsl_dataset_rele(ds, FTAG);
940 return (error);
941 }
942
943 if (newval == 0) {
944 dsl_dataset_rele(ds, FTAG);
945 return (0);
946 }
947
948 mutex_enter(&ds->ds_dir->dd_lock);
949 /*
950 * If we are doing the preliminary check in open context, and
951 * there are pending changes, then don't fail it, since the
952 * pending changes could under-estimate the amount of space to be
953 * freed up.
954 */
955 towrite = dsl_dir_space_towrite(ds->ds_dir);
956 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
957 (newval < ds->ds_dir->dd_phys->dd_reserved ||
958 newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
959 error = SET_ERROR(ENOSPC);
960 }
961 mutex_exit(&ds->ds_dir->dd_lock);
962 dsl_dataset_rele(ds, FTAG);
963 return (error);
964 }
965
966 static void
967 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
968 {
969 dsl_dir_set_qr_arg_t *ddsqra = arg;
970 dsl_pool_t *dp = dmu_tx_pool(tx);
971 dsl_dataset_t *ds;
972 uint64_t newval;
973
974 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
975
976 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
977 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
978 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
979 &ddsqra->ddsqra_value, tx);
980
981 VERIFY0(dsl_prop_get_int_ds(ds,
982 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
983 } else {
984 newval = ddsqra->ddsqra_value;
985 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
986 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
987 }
988
989 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
990 mutex_enter(&ds->ds_dir->dd_lock);
991 ds->ds_dir->dd_phys->dd_quota = newval;
992 mutex_exit(&ds->ds_dir->dd_lock);
993 dsl_dataset_rele(ds, FTAG);
994 }
995
996 int
997 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
998 {
999 dsl_dir_set_qr_arg_t ddsqra;
1000
1001 ddsqra.ddsqra_name = ddname;
1002 ddsqra.ddsqra_source = source;
1003 ddsqra.ddsqra_value = quota;
1004
1005 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1006 dsl_dir_set_quota_sync, &ddsqra, 0));
1007 }
1008
1009 int
1010 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1011 {
1012 dsl_dir_set_qr_arg_t *ddsqra = arg;
1013 dsl_pool_t *dp = dmu_tx_pool(tx);
1014 dsl_dataset_t *ds;
1015 dsl_dir_t *dd;
1016 uint64_t newval, used, avail;
1017 int error;
1018
1019 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1020 if (error != 0)
1021 return (error);
1022 dd = ds->ds_dir;
1023
1024 /*
1025 * If we are doing the preliminary check in open context, the
1026 * space estimates may be inaccurate.
1027 */
1028 if (!dmu_tx_is_syncing(tx)) {
1029 dsl_dataset_rele(ds, FTAG);
1030 return (0);
1031 }
1032
1033 error = dsl_prop_predict(ds->ds_dir,
1034 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1035 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1036 if (error != 0) {
1037 dsl_dataset_rele(ds, FTAG);
1038 return (error);
1039 }
1040
1041 mutex_enter(&dd->dd_lock);
1042 used = dd->dd_phys->dd_used_bytes;
1043 mutex_exit(&dd->dd_lock);
1044
1045 if (dd->dd_parent) {
1046 avail = dsl_dir_space_available(dd->dd_parent,
1047 NULL, 0, FALSE);
1048 } else {
1049 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1050 }
1051
1052 if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1053 uint64_t delta = MAX(used, newval) -
1054 MAX(used, dd->dd_phys->dd_reserved);
1055
1056 if (delta > avail ||
1057 (dd->dd_phys->dd_quota > 0 &&
1058 newval > dd->dd_phys->dd_quota))
1059 error = SET_ERROR(ENOSPC);
1060 }
1061
1062 dsl_dataset_rele(ds, FTAG);
1063 return (error);
1064 }
1065
1066 void
1067 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1068 {
1069 uint64_t used;
1070 int64_t delta;
1071
1072 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1073
1074 mutex_enter(&dd->dd_lock);
1075 used = dd->dd_phys->dd_used_bytes;
1076 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1077 dd->dd_phys->dd_reserved = value;
1078
1079 if (dd->dd_parent != NULL) {
1080 /* Roll up this additional usage into our ancestors */
1081 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1082 delta, 0, 0, tx);
1083 }
1084 mutex_exit(&dd->dd_lock);
1085 }
1086
1087
1088 static void
1089 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1090 {
1091 dsl_dir_set_qr_arg_t *ddsqra = arg;
1092 dsl_pool_t *dp = dmu_tx_pool(tx);
1093 dsl_dataset_t *ds;
1094 uint64_t newval;
1095
1096 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1097
1098 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1099 dsl_prop_set_sync_impl(ds,
1100 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1101 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1102 &ddsqra->ddsqra_value, tx);
1103
1104 VERIFY0(dsl_prop_get_int_ds(ds,
1105 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1106 } else {
1107 newval = ddsqra->ddsqra_value;
1108 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1109 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1110 (longlong_t)newval);
1111 }
1112
1113 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1114 dsl_dataset_rele(ds, FTAG);
1115 }
1116
1117 int
1118 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1119 uint64_t reservation)
1120 {
1121 dsl_dir_set_qr_arg_t ddsqra;
1122
1123 ddsqra.ddsqra_name = ddname;
1124 ddsqra.ddsqra_source = source;
1125 ddsqra.ddsqra_value = reservation;
1126
1127 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1128 dsl_dir_set_reservation_sync, &ddsqra, 0));
1129 }
1130
1131 static dsl_dir_t *
1132 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1133 {
1134 for (; ds1; ds1 = ds1->dd_parent) {
1135 dsl_dir_t *dd;
1136 for (dd = ds2; dd; dd = dd->dd_parent) {
1137 if (ds1 == dd)
1138 return (dd);
1139 }
1140 }
1141 return (NULL);
1142 }
1143
1144 /*
1145 * If delta is applied to dd, how much of that delta would be applied to
1146 * ancestor? Syncing context only.
1147 */
1148 static int64_t
1149 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1150 {
1151 if (dd == ancestor)
1152 return (delta);
1153
1154 mutex_enter(&dd->dd_lock);
1155 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1156 mutex_exit(&dd->dd_lock);
1157 return (would_change(dd->dd_parent, delta, ancestor));
1158 }
1159
1160 typedef struct dsl_dir_rename_arg {
1161 const char *ddra_oldname;
1162 const char *ddra_newname;
1163 } dsl_dir_rename_arg_t;
1164
1165 /* ARGSUSED */
1166 static int
1167 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1168 {
1169 int *deltap = arg;
1170 char namebuf[MAXNAMELEN];
1171
1172 dsl_dataset_name(ds, namebuf);
1173
1174 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1175 return (SET_ERROR(ENAMETOOLONG));
1176 return (0);
1177 }
1178
1179 static int
1180 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1181 {
1182 dsl_dir_rename_arg_t *ddra = arg;
1183 dsl_pool_t *dp = dmu_tx_pool(tx);
1184 dsl_dir_t *dd, *newparent;
1185 const char *mynewname;
1186 int error;
1187 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1188
1189 /* target dir should exist */
1190 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1191 if (error != 0)
1192 return (error);
1193
1194 /* new parent should exist */
1195 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1196 &newparent, &mynewname);
1197 if (error != 0) {
1198 dsl_dir_rele(dd, FTAG);
1199 return (error);
1200 }
1201
1202 /* can't rename to different pool */
1203 if (dd->dd_pool != newparent->dd_pool) {
1204 dsl_dir_rele(newparent, FTAG);
1205 dsl_dir_rele(dd, FTAG);
1206 return (SET_ERROR(ENXIO));
1207 }
1208
1209 /* new name should not already exist */
1210 if (mynewname == NULL) {
1211 dsl_dir_rele(newparent, FTAG);
1212 dsl_dir_rele(dd, FTAG);
1213 return (SET_ERROR(EEXIST));
1214 }
1215
1216 /* if the name length is growing, validate child name lengths */
1217 if (delta > 0) {
1218 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1219 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1220 if (error != 0) {
1221 dsl_dir_rele(newparent, FTAG);
1222 dsl_dir_rele(dd, FTAG);
1223 return (error);
1224 }
1225 }
1226
1227 if (newparent != dd->dd_parent) {
1228 /* is there enough space? */
1229 uint64_t myspace =
1230 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1231
1232 /* no rename into our descendant */
1233 if (closest_common_ancestor(dd, newparent) == dd) {
1234 dsl_dir_rele(newparent, FTAG);
1235 dsl_dir_rele(dd, FTAG);
1236 return (SET_ERROR(EINVAL));
1237 }
1238
1239 error = dsl_dir_transfer_possible(dd->dd_parent,
1240 newparent, myspace);
1241 if (error != 0) {
1242 dsl_dir_rele(newparent, FTAG);
1243 dsl_dir_rele(dd, FTAG);
1244 return (error);
1245 }
1246 }
1247
1248 dsl_dir_rele(newparent, FTAG);
1249 dsl_dir_rele(dd, FTAG);
1250 return (0);
1251 }
1252
1253 static void
1254 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1255 {
1256 dsl_dir_rename_arg_t *ddra = arg;
1257 dsl_pool_t *dp = dmu_tx_pool(tx);
1258 dsl_dir_t *dd, *newparent;
1259 const char *mynewname;
1260 int error;
1261 objset_t *mos = dp->dp_meta_objset;
1262
1263 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1264 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1265 &mynewname));
1266
1267 /* Log this before we change the name. */
1268 spa_history_log_internal_dd(dd, "rename", tx,
1269 "-> %s", ddra->ddra_newname);
1270
1271 if (newparent != dd->dd_parent) {
1272 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1273 -dd->dd_phys->dd_used_bytes,
1274 -dd->dd_phys->dd_compressed_bytes,
1275 -dd->dd_phys->dd_uncompressed_bytes, tx);
1276 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1277 dd->dd_phys->dd_used_bytes,
1278 dd->dd_phys->dd_compressed_bytes,
1279 dd->dd_phys->dd_uncompressed_bytes, tx);
1280
1281 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1282 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1283 dd->dd_phys->dd_used_bytes;
1284
1285 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1286 -unused_rsrv, 0, 0, tx);
1287 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1288 unused_rsrv, 0, 0, tx);
1289 }
1290 }
1291
1292 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1293
1294 /* remove from old parent zapobj */
1295 error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1296 dd->dd_myname, tx);
1297 ASSERT0(error);
1298
1299 (void) strcpy(dd->dd_myname, mynewname);
1300 dsl_dir_rele(dd->dd_parent, dd);
1301 dd->dd_phys->dd_parent_obj = newparent->dd_object;
1302 VERIFY0(dsl_dir_hold_obj(dp,
1303 newparent->dd_object, NULL, dd, &dd->dd_parent));
1304
1305 /* add to new parent zapobj */
1306 VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1307 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1308
1309 dsl_prop_notify_all(dd);
1310
1311 dsl_dir_rele(newparent, FTAG);
1312 dsl_dir_rele(dd, FTAG);
1313 }
1314
1315 int
1316 dsl_dir_rename(const char *oldname, const char *newname)
1317 {
1318 dsl_dir_rename_arg_t ddra;
1319
1320 ddra.ddra_oldname = oldname;
1321 ddra.ddra_newname = newname;
1322
1323 return (dsl_sync_task(oldname,
1324 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1325 }
1326
1327 int
1328 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1329 {
1330 dsl_dir_t *ancestor;
1331 int64_t adelta;
1332 uint64_t avail;
1333
1334 ancestor = closest_common_ancestor(sdd, tdd);
1335 adelta = would_change(sdd, -space, ancestor);
1336 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1337 if (avail < space)
1338 return (SET_ERROR(ENOSPC));
1339
1340 return (0);
1341 }
1342
1343 timestruc_t
1344 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1345 {
1346 timestruc_t t;
1347
1348 mutex_enter(&dd->dd_lock);
1349 t = dd->dd_snap_cmtime;
1350 mutex_exit(&dd->dd_lock);
1351
1352 return (t);
1353 }
1354
1355 void
1356 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1357 {
1358 timestruc_t t;
1359
1360 gethrestime(&t);
1361 mutex_enter(&dd->dd_lock);
1362 dd->dd_snap_cmtime = t;
1363 mutex_exit(&dd->dd_lock);
1364 }