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) 2012 by Delphix. All rights reserved.
24 */
25
26 #include <sys/dmu.h>
27 #include <sys/dmu_objset.h>
28 #include <sys/dmu_tx.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dsl_deleg.h>
34 #include <sys/spa.h>
35 #include <sys/metaslab.h>
36 #include <sys/zap.h>
37 #include <sys/zio.h>
38 #include <sys/arc.h>
39 #include <sys/sunddi.h>
40 #include "zfs_namecheck.h"
41
42 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
43 static void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd,
44 uint64_t value, dmu_tx_t *tx);
45
46 /* ARGSUSED */
47 static void
48 dsl_dir_evict(dmu_buf_t *db, void *arg)
49 {
50 dsl_dir_t *dd = arg;
51 dsl_pool_t *dp = dd->dd_pool;
52 int t;
53
54 for (t = 0; t < TXG_SIZE; t++) {
55 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
56 ASSERT(dd->dd_tempreserved[t] == 0);
57 ASSERT(dd->dd_space_towrite[t] == 0);
58 }
59
60 if (dd->dd_parent)
61 dsl_dir_close(dd->dd_parent, dd);
62
63 spa_close(dd->dd_pool->dp_spa, dd);
64
65 /*
66 * The props callback list should have been cleaned up by
67 * objset_evict().
68 */
69 list_destroy(&dd->dd_prop_cbs);
70 mutex_destroy(&dd->dd_lock);
71 kmem_free(dd, sizeof (dsl_dir_t));
72 }
73
74 int
75 dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
76 const char *tail, void *tag, dsl_dir_t **ddp)
77 {
78 dmu_buf_t *dbuf;
79 dsl_dir_t *dd;
80 int err;
81
82 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
83 dsl_pool_sync_context(dp));
84
85 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
86 if (err)
87 return (err);
88 dd = dmu_buf_get_user(dbuf);
89 #ifdef ZFS_DEBUG
90 {
91 dmu_object_info_t doi;
92 dmu_object_info_from_db(dbuf, &doi);
93 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
94 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
95 }
96 #endif
97 if (dd == NULL) {
98 dsl_dir_t *winner;
99
100 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
101 dd->dd_object = ddobj;
102 dd->dd_dbuf = dbuf;
103 dd->dd_pool = dp;
104 dd->dd_phys = dbuf->db_data;
105 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
106
107 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
108 offsetof(dsl_prop_cb_record_t, cbr_node));
109
110 dsl_dir_snap_cmtime_update(dd);
111
112 if (dd->dd_phys->dd_parent_obj) {
113 err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
114 NULL, dd, &dd->dd_parent);
115 if (err)
116 goto errout;
117 if (tail) {
118 #ifdef ZFS_DEBUG
119 uint64_t foundobj;
120
121 err = zap_lookup(dp->dp_meta_objset,
122 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
123 tail, sizeof (foundobj), 1, &foundobj);
124 ASSERT(err || foundobj == ddobj);
125 #endif
126 (void) strcpy(dd->dd_myname, tail);
127 } else {
128 err = zap_value_search(dp->dp_meta_objset,
129 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
130 ddobj, 0, dd->dd_myname);
131 }
132 if (err)
133 goto errout;
134 } else {
135 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
136 }
137
138 if (dsl_dir_is_clone(dd)) {
139 dmu_buf_t *origin_bonus;
140 dsl_dataset_phys_t *origin_phys;
141
142 /*
143 * We can't open the origin dataset, because
144 * that would require opening this dsl_dir.
145 * Just look at its phys directly instead.
146 */
147 err = dmu_bonus_hold(dp->dp_meta_objset,
148 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
149 if (err)
150 goto errout;
151 origin_phys = origin_bonus->db_data;
152 dd->dd_origin_txg =
153 origin_phys->ds_creation_txg;
154 dmu_buf_rele(origin_bonus, FTAG);
155 }
156
157 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
158 dsl_dir_evict);
159 if (winner) {
160 if (dd->dd_parent)
161 dsl_dir_close(dd->dd_parent, dd);
162 mutex_destroy(&dd->dd_lock);
163 kmem_free(dd, sizeof (dsl_dir_t));
164 dd = winner;
165 } else {
166 spa_open_ref(dp->dp_spa, dd);
167 }
168 }
169
170 /*
171 * The dsl_dir_t has both open-to-close and instantiate-to-evict
172 * holds on the spa. We need the open-to-close holds because
173 * otherwise the spa_refcnt wouldn't change when we open a
174 * dir which the spa also has open, so we could incorrectly
175 * think it was OK to unload/export/destroy the pool. We need
176 * the instantiate-to-evict hold because the dsl_dir_t has a
177 * pointer to the dd_pool, which has a pointer to the spa_t.
178 */
179 spa_open_ref(dp->dp_spa, tag);
180 ASSERT3P(dd->dd_pool, ==, dp);
181 ASSERT3U(dd->dd_object, ==, ddobj);
182 ASSERT3P(dd->dd_dbuf, ==, dbuf);
183 *ddp = dd;
184 return (0);
185
186 errout:
187 if (dd->dd_parent)
188 dsl_dir_close(dd->dd_parent, dd);
189 mutex_destroy(&dd->dd_lock);
190 kmem_free(dd, sizeof (dsl_dir_t));
191 dmu_buf_rele(dbuf, tag);
192 return (err);
193
194 }
195
196 void
197 dsl_dir_close(dsl_dir_t *dd, void *tag)
198 {
199 dprintf_dd(dd, "%s\n", "");
200 spa_close(dd->dd_pool->dp_spa, tag);
201 dmu_buf_rele(dd->dd_dbuf, tag);
202 }
203
204 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
205 void
206 dsl_dir_name(dsl_dir_t *dd, char *buf)
207 {
208 if (dd->dd_parent) {
209 dsl_dir_name(dd->dd_parent, buf);
210 (void) strcat(buf, "/");
211 } else {
212 buf[0] = '\0';
213 }
214 if (!MUTEX_HELD(&dd->dd_lock)) {
215 /*
216 * recursive mutex so that we can use
217 * dprintf_dd() with dd_lock held
218 */
219 mutex_enter(&dd->dd_lock);
220 (void) strcat(buf, dd->dd_myname);
221 mutex_exit(&dd->dd_lock);
222 } else {
223 (void) strcat(buf, dd->dd_myname);
224 }
225 }
226
227 /* Calculate name legnth, avoiding all the strcat calls of dsl_dir_name */
228 int
229 dsl_dir_namelen(dsl_dir_t *dd)
230 {
231 int result = 0;
232
233 if (dd->dd_parent) {
234 /* parent's name + 1 for the "/" */
235 result = dsl_dir_namelen(dd->dd_parent) + 1;
236 }
237
238 if (!MUTEX_HELD(&dd->dd_lock)) {
239 /* see dsl_dir_name */
240 mutex_enter(&dd->dd_lock);
241 result += strlen(dd->dd_myname);
242 mutex_exit(&dd->dd_lock);
243 } else {
244 result += strlen(dd->dd_myname);
245 }
246
247 return (result);
248 }
249
250 static int
251 getcomponent(const char *path, char *component, const char **nextp)
252 {
253 char *p;
254 if ((path == NULL) || (path[0] == '\0'))
255 return (ENOENT);
256 /* This would be a good place to reserve some namespace... */
257 p = strpbrk(path, "/@");
258 if (p && (p[1] == '/' || p[1] == '@')) {
259 /* two separators in a row */
260 return (EINVAL);
261 }
262 if (p == NULL || p == path) {
263 /*
264 * if the first thing is an @ or /, it had better be an
265 * @ and it had better not have any more ats or slashes,
266 * and it had better have something after the @.
267 */
268 if (p != NULL &&
269 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
270 return (EINVAL);
271 if (strlen(path) >= MAXNAMELEN)
272 return (ENAMETOOLONG);
273 (void) strcpy(component, path);
274 p = NULL;
275 } else if (p[0] == '/') {
276 if (p-path >= MAXNAMELEN)
277 return (ENAMETOOLONG);
278 (void) strncpy(component, path, p - path);
279 component[p-path] = '\0';
280 p++;
281 } else if (p[0] == '@') {
282 /*
283 * if the next separator is an @, there better not be
284 * any more slashes.
285 */
286 if (strchr(path, '/'))
287 return (EINVAL);
288 if (p-path >= MAXNAMELEN)
289 return (ENAMETOOLONG);
290 (void) strncpy(component, path, p - path);
291 component[p-path] = '\0';
292 } else {
293 ASSERT(!"invalid p");
294 }
295 *nextp = p;
296 return (0);
297 }
298
299 /*
300 * same as dsl_open_dir, ignore the first component of name and use the
301 * spa instead
302 */
303 int
304 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
305 dsl_dir_t **ddp, const char **tailp)
306 {
307 char buf[MAXNAMELEN];
308 const char *next, *nextnext = NULL;
309 int err;
310 dsl_dir_t *dd;
311 dsl_pool_t *dp;
312 uint64_t ddobj;
313 int openedspa = FALSE;
314
315 dprintf("%s\n", name);
316
317 err = getcomponent(name, buf, &next);
318 if (err)
319 return (err);
320 if (spa == NULL) {
321 err = spa_open(buf, &spa, FTAG);
322 if (err) {
323 dprintf("spa_open(%s) failed\n", buf);
324 return (err);
325 }
326 openedspa = TRUE;
327
328 /* XXX this assertion belongs in spa_open */
329 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
330 }
331
332 dp = spa_get_dsl(spa);
333
334 rw_enter(&dp->dp_config_rwlock, RW_READER);
335 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
336 if (err) {
337 rw_exit(&dp->dp_config_rwlock);
338 if (openedspa)
339 spa_close(spa, FTAG);
340 return (err);
341 }
342
343 while (next != NULL) {
344 dsl_dir_t *child_ds;
345 err = getcomponent(next, buf, &nextnext);
346 if (err)
347 break;
348 ASSERT(next[0] != '\0');
349 if (next[0] == '@')
350 break;
351 dprintf("looking up %s in obj%lld\n",
352 buf, dd->dd_phys->dd_child_dir_zapobj);
353
354 err = zap_lookup(dp->dp_meta_objset,
355 dd->dd_phys->dd_child_dir_zapobj,
356 buf, sizeof (ddobj), 1, &ddobj);
357 if (err) {
358 if (err == ENOENT)
359 err = 0;
360 break;
361 }
362
363 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
364 if (err)
365 break;
366 dsl_dir_close(dd, tag);
367 dd = child_ds;
368 next = nextnext;
369 }
370 rw_exit(&dp->dp_config_rwlock);
371
372 if (err) {
373 dsl_dir_close(dd, tag);
374 if (openedspa)
375 spa_close(spa, FTAG);
376 return (err);
377 }
378
379 /*
380 * It's an error if there's more than one component left, or
381 * tailp==NULL and there's any component left.
382 */
383 if (next != NULL &&
384 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
385 /* bad path name */
386 dsl_dir_close(dd, tag);
387 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
388 err = ENOENT;
389 }
390 if (tailp)
391 *tailp = next;
392 if (openedspa)
393 spa_close(spa, FTAG);
394 *ddp = dd;
395 return (err);
396 }
397
398 /*
399 * Return the dsl_dir_t, and possibly the last component which couldn't
400 * be found in *tail. Return NULL if the path is bogus, or if
401 * tail==NULL and we couldn't parse the whole name. (*tail)[0] == '@'
402 * means that the last component is a snapshot.
403 */
404 int
405 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
406 {
407 return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
408 }
409
410 uint64_t
411 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
412 dmu_tx_t *tx)
413 {
414 objset_t *mos = dp->dp_meta_objset;
415 uint64_t ddobj;
416 dsl_dir_phys_t *ddphys;
417 dmu_buf_t *dbuf;
418
419 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
420 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
421 if (pds) {
422 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
423 name, sizeof (uint64_t), 1, &ddobj, tx));
424 } else {
425 /* it's the root dir */
426 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
427 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
428 }
429 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
430 dmu_buf_will_dirty(dbuf, tx);
431 ddphys = dbuf->db_data;
432
433 ddphys->dd_creation_time = gethrestime_sec();
434 if (pds)
435 ddphys->dd_parent_obj = pds->dd_object;
436 ddphys->dd_props_zapobj = zap_create(mos,
437 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
438 ddphys->dd_child_dir_zapobj = zap_create(mos,
439 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
440 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
441 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
442 dmu_buf_rele(dbuf, FTAG);
443
444 return (ddobj);
445 }
446
447 /* ARGSUSED */
448 int
449 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
450 {
451 dsl_dir_t *dd = arg1;
452 dsl_pool_t *dp = dd->dd_pool;
453 objset_t *mos = dp->dp_meta_objset;
454 int err;
455 uint64_t count;
456
457 /*
458 * There should be exactly two holds, both from
459 * dsl_dataset_destroy: one on the dd directory, and one on its
460 * head ds. Otherwise, someone is trying to lookup something
461 * inside this dir while we want to destroy it. The
462 * config_rwlock ensures that nobody else opens it after we
463 * check.
464 */
465 if (dmu_buf_refcount(dd->dd_dbuf) > 2)
466 return (EBUSY);
467
468 err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
469 if (err)
470 return (err);
471 if (count != 0)
472 return (EEXIST);
473
474 return (0);
475 }
476
477 void
478 dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
479 {
480 dsl_dir_t *dd = arg1;
481 objset_t *mos = dd->dd_pool->dp_meta_objset;
482 uint64_t obj;
483 dd_used_t t;
484
485 ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
486 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
487
488 /*
489 * Remove our reservation. The impl() routine avoids setting the
490 * actual property, which would require the (already destroyed) ds.
491 */
492 dsl_dir_set_reservation_sync_impl(dd, 0, tx);
493
494 ASSERT0(dd->dd_phys->dd_used_bytes);
495 ASSERT0(dd->dd_phys->dd_reserved);
496 for (t = 0; t < DD_USED_NUM; t++)
497 ASSERT0(dd->dd_phys->dd_used_breakdown[t]);
498
499 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
500 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
501 VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
502 VERIFY(0 == zap_remove(mos,
503 dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
504
505 obj = dd->dd_object;
506 dsl_dir_close(dd, tag);
507 VERIFY(0 == dmu_object_free(mos, obj, tx));
508 }
509
510 boolean_t
511 dsl_dir_is_clone(dsl_dir_t *dd)
512 {
513 return (dd->dd_phys->dd_origin_obj &&
514 (dd->dd_pool->dp_origin_snap == NULL ||
515 dd->dd_phys->dd_origin_obj !=
516 dd->dd_pool->dp_origin_snap->ds_object));
517 }
518
519 void
520 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
521 {
522 mutex_enter(&dd->dd_lock);
523 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
524 dd->dd_phys->dd_used_bytes);
525 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
526 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
527 dd->dd_phys->dd_reserved);
528 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
529 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
530 (dd->dd_phys->dd_uncompressed_bytes * 100 /
531 dd->dd_phys->dd_compressed_bytes));
532 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
533 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
534 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
535 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
536 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
537 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
538 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
539 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
540 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
541 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
542 }
543 mutex_exit(&dd->dd_lock);
544
545 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
546 if (dsl_dir_is_clone(dd)) {
547 dsl_dataset_t *ds;
548 char buf[MAXNAMELEN];
549
550 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
551 dd->dd_phys->dd_origin_obj, FTAG, &ds));
552 dsl_dataset_name(ds, buf);
553 dsl_dataset_rele(ds, FTAG);
554 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
555 }
556 rw_exit(&dd->dd_pool->dp_config_rwlock);
557 }
558
559 void
560 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
561 {
562 dsl_pool_t *dp = dd->dd_pool;
563
564 ASSERT(dd->dd_phys);
565
566 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
567 /* up the hold count until we can be written out */
568 dmu_buf_add_ref(dd->dd_dbuf, dd);
569 }
570 }
571
572 static int64_t
573 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
574 {
575 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
576 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
577 return (new_accounted - old_accounted);
578 }
579
580 void
581 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
582 {
583 ASSERT(dmu_tx_is_syncing(tx));
584
585 dmu_buf_will_dirty(dd->dd_dbuf, tx);
586
587 mutex_enter(&dd->dd_lock);
588 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
589 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
590 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
591 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
592 mutex_exit(&dd->dd_lock);
593
594 /* release the hold from dsl_dir_dirty */
595 dmu_buf_rele(dd->dd_dbuf, dd);
596 }
597
598 static uint64_t
599 dsl_dir_space_towrite(dsl_dir_t *dd)
600 {
601 uint64_t space = 0;
602 int i;
603
604 ASSERT(MUTEX_HELD(&dd->dd_lock));
605
606 for (i = 0; i < TXG_SIZE; i++) {
607 space += dd->dd_space_towrite[i&TXG_MASK];
608 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
609 }
610 return (space);
611 }
612
613 /*
614 * How much space would dd have available if ancestor had delta applied
615 * to it? If ondiskonly is set, we're only interested in what's
616 * on-disk, not estimated pending changes.
617 */
618 uint64_t
619 dsl_dir_space_available(dsl_dir_t *dd,
620 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
621 {
622 uint64_t parentspace, myspace, quota, used;
623
624 /*
625 * If there are no restrictions otherwise, assume we have
626 * unlimited space available.
627 */
628 quota = UINT64_MAX;
629 parentspace = UINT64_MAX;
630
631 if (dd->dd_parent != NULL) {
632 parentspace = dsl_dir_space_available(dd->dd_parent,
633 ancestor, delta, ondiskonly);
634 }
635
636 mutex_enter(&dd->dd_lock);
637 if (dd->dd_phys->dd_quota != 0)
638 quota = dd->dd_phys->dd_quota;
639 used = dd->dd_phys->dd_used_bytes;
640 if (!ondiskonly)
641 used += dsl_dir_space_towrite(dd);
642
643 if (dd->dd_parent == NULL) {
644 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
645 quota = MIN(quota, poolsize);
646 }
647
648 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
649 /*
650 * We have some space reserved, in addition to what our
651 * parent gave us.
652 */
653 parentspace += dd->dd_phys->dd_reserved - used;
654 }
655
656 if (dd == ancestor) {
657 ASSERT(delta <= 0);
658 ASSERT(used >= -delta);
659 used += delta;
660 if (parentspace != UINT64_MAX)
661 parentspace -= delta;
662 }
663
664 if (used > quota) {
665 /* over quota */
666 myspace = 0;
667 } else {
668 /*
669 * the lesser of the space provided by our parent and
670 * the space left in our quota
671 */
672 myspace = MIN(parentspace, quota - used);
673 }
674
675 mutex_exit(&dd->dd_lock);
676
677 return (myspace);
678 }
679
680 struct tempreserve {
681 list_node_t tr_node;
682 dsl_pool_t *tr_dp;
683 dsl_dir_t *tr_ds;
684 uint64_t tr_size;
685 };
686
687 static int
688 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
689 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
690 dmu_tx_t *tx, boolean_t first)
691 {
692 uint64_t txg = tx->tx_txg;
693 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
694 uint64_t deferred = 0;
695 struct tempreserve *tr;
696 int retval = EDQUOT;
697 int txgidx = txg & TXG_MASK;
698 int i;
699 uint64_t ref_rsrv = 0;
700
701 ASSERT3U(txg, !=, 0);
702 ASSERT3S(asize, >, 0);
703
704 mutex_enter(&dd->dd_lock);
705
706 /*
707 * Check against the dsl_dir's quota. We don't add in the delta
708 * when checking for over-quota because they get one free hit.
709 */
710 est_inflight = dsl_dir_space_towrite(dd);
711 for (i = 0; i < TXG_SIZE; i++)
712 est_inflight += dd->dd_tempreserved[i];
713 used_on_disk = dd->dd_phys->dd_used_bytes;
714
715 /*
716 * On the first iteration, fetch the dataset's used-on-disk and
717 * refreservation values. Also, if checkrefquota is set, test if
718 * allocating this space would exceed the dataset's refquota.
719 */
720 if (first && tx->tx_objset) {
721 int error;
722 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
723
724 error = dsl_dataset_check_quota(ds, checkrefquota,
725 asize, est_inflight, &used_on_disk, &ref_rsrv);
726 if (error) {
727 mutex_exit(&dd->dd_lock);
728 return (error);
729 }
730 }
731
732 /*
733 * If this transaction will result in a net free of space,
734 * we want to let it through.
735 */
736 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
737 quota = UINT64_MAX;
738 else
739 quota = dd->dd_phys->dd_quota;
740
741 /*
742 * Adjust the quota against the actual pool size at the root
743 * minus any outstanding deferred frees.
744 * To ensure that it's possible to remove files from a full
745 * pool without inducing transient overcommits, we throttle
746 * netfree transactions against a quota that is slightly larger,
747 * but still within the pool's allocation slop. In cases where
748 * we're very close to full, this will allow a steady trickle of
749 * removes to get through.
750 */
751 if (dd->dd_parent == NULL) {
752 spa_t *spa = dd->dd_pool->dp_spa;
753 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
754 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
755 if (poolsize - deferred < quota) {
756 quota = poolsize - deferred;
757 retval = ENOSPC;
758 }
759 }
760
761 /*
762 * If they are requesting more space, and our current estimate
763 * is over quota, they get to try again unless the actual
764 * on-disk is over quota and there are no pending changes (which
765 * may free up space for us).
766 */
767 if (used_on_disk + est_inflight >= quota) {
768 if (est_inflight > 0 || used_on_disk < quota ||
769 (retval == ENOSPC && used_on_disk < quota + deferred))
770 retval = ERESTART;
771 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
772 "quota=%lluK tr=%lluK err=%d\n",
773 used_on_disk>>10, est_inflight>>10,
774 quota>>10, asize>>10, retval);
775 mutex_exit(&dd->dd_lock);
776 return (retval);
777 }
778
779 /* We need to up our estimated delta before dropping dd_lock */
780 dd->dd_tempreserved[txgidx] += asize;
781
782 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
783 asize - ref_rsrv);
784 mutex_exit(&dd->dd_lock);
785
786 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
787 tr->tr_ds = dd;
788 tr->tr_size = asize;
789 list_insert_tail(tr_list, tr);
790
791 /* see if it's OK with our parent */
792 if (dd->dd_parent && parent_rsrv) {
793 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
794
795 return (dsl_dir_tempreserve_impl(dd->dd_parent,
796 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
797 } else {
798 return (0);
799 }
800 }
801
802 /*
803 * Reserve space in this dsl_dir, to be used in this tx's txg.
804 * After the space has been dirtied (and dsl_dir_willuse_space()
805 * has been called), the reservation should be canceled, using
806 * dsl_dir_tempreserve_clear().
807 */
808 int
809 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
810 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
811 {
812 int err;
813 list_t *tr_list;
814
815 if (asize == 0) {
816 *tr_cookiep = NULL;
817 return (0);
818 }
819
820 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
821 list_create(tr_list, sizeof (struct tempreserve),
822 offsetof(struct tempreserve, tr_node));
823 ASSERT3S(asize, >, 0);
824 ASSERT3S(fsize, >=, 0);
825
826 err = arc_tempreserve_space(lsize, tx->tx_txg);
827 if (err == 0) {
828 struct tempreserve *tr;
829
830 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
831 tr->tr_size = lsize;
832 list_insert_tail(tr_list, tr);
833
834 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
835 } else {
836 if (err == EAGAIN) {
837 txg_delay(dd->dd_pool, tx->tx_txg, 1);
838 err = ERESTART;
839 }
840 dsl_pool_memory_pressure(dd->dd_pool);
841 }
842
843 if (err == 0) {
844 struct tempreserve *tr;
845
846 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
847 tr->tr_dp = dd->dd_pool;
848 tr->tr_size = asize;
849 list_insert_tail(tr_list, tr);
850
851 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
852 FALSE, asize > usize, tr_list, tx, TRUE);
853 }
854
855 if (err)
856 dsl_dir_tempreserve_clear(tr_list, tx);
857 else
858 *tr_cookiep = tr_list;
859
860 return (err);
861 }
862
863 /*
864 * Clear a temporary reservation that we previously made with
865 * dsl_dir_tempreserve_space().
866 */
867 void
868 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
869 {
870 int txgidx = tx->tx_txg & TXG_MASK;
871 list_t *tr_list = tr_cookie;
872 struct tempreserve *tr;
873
874 ASSERT3U(tx->tx_txg, !=, 0);
875
876 if (tr_cookie == NULL)
877 return;
878
879 while (tr = list_head(tr_list)) {
880 if (tr->tr_dp) {
881 dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
882 } else if (tr->tr_ds) {
883 mutex_enter(&tr->tr_ds->dd_lock);
884 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
885 tr->tr_size);
886 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
887 mutex_exit(&tr->tr_ds->dd_lock);
888 } else {
889 arc_tempreserve_clear(tr->tr_size);
890 }
891 list_remove(tr_list, tr);
892 kmem_free(tr, sizeof (struct tempreserve));
893 }
894
895 kmem_free(tr_list, sizeof (list_t));
896 }
897
898 static void
899 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
900 {
901 int64_t parent_space;
902 uint64_t est_used;
903
904 mutex_enter(&dd->dd_lock);
905 if (space > 0)
906 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
907
908 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
909 parent_space = parent_delta(dd, est_used, space);
910 mutex_exit(&dd->dd_lock);
911
912 /* Make sure that we clean up dd_space_to* */
913 dsl_dir_dirty(dd, tx);
914
915 /* XXX this is potentially expensive and unnecessary... */
916 if (parent_space && dd->dd_parent)
917 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
918 }
919
920 /*
921 * Call in open context when we think we're going to write/free space,
922 * eg. when dirtying data. Be conservative (ie. OK to write less than
923 * this or free more than this, but don't write more or free less).
924 */
925 void
926 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
927 {
928 dsl_pool_willuse_space(dd->dd_pool, space, tx);
929 dsl_dir_willuse_space_impl(dd, space, tx);
930 }
931
932 /* call from syncing context when we actually write/free space for this dd */
933 void
934 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
935 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
936 {
937 int64_t accounted_delta;
938 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
939
940 ASSERT(dmu_tx_is_syncing(tx));
941 ASSERT(type < DD_USED_NUM);
942
943 dsl_dir_dirty(dd, tx);
944
945 if (needlock)
946 mutex_enter(&dd->dd_lock);
947 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
948 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
949 ASSERT(compressed >= 0 ||
950 dd->dd_phys->dd_compressed_bytes >= -compressed);
951 ASSERT(uncompressed >= 0 ||
952 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
953 dd->dd_phys->dd_used_bytes += used;
954 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
955 dd->dd_phys->dd_compressed_bytes += compressed;
956
957 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
958 ASSERT(used > 0 ||
959 dd->dd_phys->dd_used_breakdown[type] >= -used);
960 dd->dd_phys->dd_used_breakdown[type] += used;
961 #ifdef DEBUG
962 dd_used_t t;
963 uint64_t u = 0;
964 for (t = 0; t < DD_USED_NUM; t++)
965 u += dd->dd_phys->dd_used_breakdown[t];
966 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
967 #endif
968 }
969 if (needlock)
970 mutex_exit(&dd->dd_lock);
971
972 if (dd->dd_parent != NULL) {
973 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
974 accounted_delta, compressed, uncompressed, tx);
975 dsl_dir_transfer_space(dd->dd_parent,
976 used - accounted_delta,
977 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
978 }
979 }
980
981 void
982 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
983 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
984 {
985 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
986
987 ASSERT(dmu_tx_is_syncing(tx));
988 ASSERT(oldtype < DD_USED_NUM);
989 ASSERT(newtype < DD_USED_NUM);
990
991 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
992 return;
993
994 dsl_dir_dirty(dd, tx);
995 if (needlock)
996 mutex_enter(&dd->dd_lock);
997 ASSERT(delta > 0 ?
998 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
999 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1000 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1001 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1002 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1003 if (needlock)
1004 mutex_exit(&dd->dd_lock);
1005 }
1006
1007 static int
1008 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1009 {
1010 dsl_dataset_t *ds = arg1;
1011 dsl_dir_t *dd = ds->ds_dir;
1012 dsl_prop_setarg_t *psa = arg2;
1013 int err;
1014 uint64_t towrite;
1015
1016 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1017 return (err);
1018
1019 if (psa->psa_effective_value == 0)
1020 return (0);
1021
1022 mutex_enter(&dd->dd_lock);
1023 /*
1024 * If we are doing the preliminary check in open context, and
1025 * there are pending changes, then don't fail it, since the
1026 * pending changes could under-estimate the amount of space to be
1027 * freed up.
1028 */
1029 towrite = dsl_dir_space_towrite(dd);
1030 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1031 (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
1032 psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
1033 err = ENOSPC;
1034 }
1035 mutex_exit(&dd->dd_lock);
1036 return (err);
1037 }
1038
1039 extern dsl_syncfunc_t dsl_prop_set_sync;
1040
1041 static void
1042 dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1043 {
1044 dsl_dataset_t *ds = arg1;
1045 dsl_dir_t *dd = ds->ds_dir;
1046 dsl_prop_setarg_t *psa = arg2;
1047 uint64_t effective_value = psa->psa_effective_value;
1048
1049 dsl_prop_set_sync(ds, psa, tx);
1050 DSL_PROP_CHECK_PREDICTION(dd, psa);
1051
1052 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1053
1054 mutex_enter(&dd->dd_lock);
1055 dd->dd_phys->dd_quota = effective_value;
1056 mutex_exit(&dd->dd_lock);
1057
1058 spa_history_log_internal_dd(dd, "set quota", tx,
1059 "quota=%lld", (longlong_t)effective_value);
1060 }
1061
1062 int
1063 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1064 {
1065 dsl_dir_t *dd;
1066 dsl_dataset_t *ds;
1067 dsl_prop_setarg_t psa;
1068 int err;
1069
1070 dsl_prop_setarg_init_uint64(&psa, "quota", source, "a);
1071
1072 err = dsl_dataset_hold(ddname, FTAG, &ds);
1073 if (err)
1074 return (err);
1075
1076 err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1077 if (err) {
1078 dsl_dataset_rele(ds, FTAG);
1079 return (err);
1080 }
1081
1082 ASSERT(ds->ds_dir == dd);
1083
1084 /*
1085 * If someone removes a file, then tries to set the quota, we want to
1086 * make sure the file freeing takes effect.
1087 */
1088 txg_wait_open(dd->dd_pool, 0);
1089
1090 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1091 dsl_dir_set_quota_sync, ds, &psa, 0);
1092
1093 dsl_dir_close(dd, FTAG);
1094 dsl_dataset_rele(ds, FTAG);
1095 return (err);
1096 }
1097
1098 int
1099 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1100 {
1101 dsl_dataset_t *ds = arg1;
1102 dsl_dir_t *dd = ds->ds_dir;
1103 dsl_prop_setarg_t *psa = arg2;
1104 uint64_t effective_value;
1105 uint64_t used, avail;
1106 int err;
1107
1108 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1109 return (err);
1110
1111 effective_value = psa->psa_effective_value;
1112
1113 /*
1114 * If we are doing the preliminary check in open context, the
1115 * space estimates may be inaccurate.
1116 */
1117 if (!dmu_tx_is_syncing(tx))
1118 return (0);
1119
1120 mutex_enter(&dd->dd_lock);
1121 used = dd->dd_phys->dd_used_bytes;
1122 mutex_exit(&dd->dd_lock);
1123
1124 if (dd->dd_parent) {
1125 avail = dsl_dir_space_available(dd->dd_parent,
1126 NULL, 0, FALSE);
1127 } else {
1128 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1129 }
1130
1131 if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
1132 uint64_t delta = MAX(used, effective_value) -
1133 MAX(used, dd->dd_phys->dd_reserved);
1134
1135 if (delta > avail)
1136 return (ENOSPC);
1137 if (dd->dd_phys->dd_quota > 0 &&
1138 effective_value > dd->dd_phys->dd_quota)
1139 return (ENOSPC);
1140 }
1141
1142 return (0);
1143 }
1144
1145 static void
1146 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1147 {
1148 uint64_t used;
1149 int64_t delta;
1150
1151 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1152
1153 mutex_enter(&dd->dd_lock);
1154 used = dd->dd_phys->dd_used_bytes;
1155 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1156 dd->dd_phys->dd_reserved = value;
1157
1158 if (dd->dd_parent != NULL) {
1159 /* Roll up this additional usage into our ancestors */
1160 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1161 delta, 0, 0, tx);
1162 }
1163 mutex_exit(&dd->dd_lock);
1164 }
1165
1166
1167 static void
1168 dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1169 {
1170 dsl_dataset_t *ds = arg1;
1171 dsl_dir_t *dd = ds->ds_dir;
1172 dsl_prop_setarg_t *psa = arg2;
1173 uint64_t value = psa->psa_effective_value;
1174
1175 dsl_prop_set_sync(ds, psa, tx);
1176 DSL_PROP_CHECK_PREDICTION(dd, psa);
1177
1178 dsl_dir_set_reservation_sync_impl(dd, value, tx);
1179
1180 spa_history_log_internal_dd(dd, "set reservation", tx,
1181 "reservation=%lld", (longlong_t)value);
1182 }
1183
1184 int
1185 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1186 uint64_t reservation)
1187 {
1188 dsl_dir_t *dd;
1189 dsl_dataset_t *ds;
1190 dsl_prop_setarg_t psa;
1191 int err;
1192
1193 dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
1194
1195 err = dsl_dataset_hold(ddname, FTAG, &ds);
1196 if (err)
1197 return (err);
1198
1199 err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1200 if (err) {
1201 dsl_dataset_rele(ds, FTAG);
1202 return (err);
1203 }
1204
1205 ASSERT(ds->ds_dir == dd);
1206
1207 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1208 dsl_dir_set_reservation_sync, ds, &psa, 0);
1209
1210 dsl_dir_close(dd, FTAG);
1211 dsl_dataset_rele(ds, FTAG);
1212 return (err);
1213 }
1214
1215 static dsl_dir_t *
1216 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1217 {
1218 for (; ds1; ds1 = ds1->dd_parent) {
1219 dsl_dir_t *dd;
1220 for (dd = ds2; dd; dd = dd->dd_parent) {
1221 if (ds1 == dd)
1222 return (dd);
1223 }
1224 }
1225 return (NULL);
1226 }
1227
1228 /*
1229 * If delta is applied to dd, how much of that delta would be applied to
1230 * ancestor? Syncing context only.
1231 */
1232 static int64_t
1233 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1234 {
1235 if (dd == ancestor)
1236 return (delta);
1237
1238 mutex_enter(&dd->dd_lock);
1239 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1240 mutex_exit(&dd->dd_lock);
1241 return (would_change(dd->dd_parent, delta, ancestor));
1242 }
1243
1244 struct renamearg {
1245 dsl_dir_t *newparent;
1246 const char *mynewname;
1247 };
1248
1249 static int
1250 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1251 {
1252 dsl_dir_t *dd = arg1;
1253 struct renamearg *ra = arg2;
1254 dsl_pool_t *dp = dd->dd_pool;
1255 objset_t *mos = dp->dp_meta_objset;
1256 int err;
1257 uint64_t val;
1258
1259 /*
1260 * There should only be one reference, from dmu_objset_rename().
1261 * Fleeting holds are also possible (eg, from "zfs list" getting
1262 * stats), but any that are present in open context will likely
1263 * be gone by syncing context, so only fail from syncing
1264 * context.
1265 */
1266 if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 1)
1267 return (EBUSY);
1268
1269 /* check for existing name */
1270 err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1271 ra->mynewname, 8, 1, &val);
1272 if (err == 0)
1273 return (EEXIST);
1274 if (err != ENOENT)
1275 return (err);
1276
1277 if (ra->newparent != dd->dd_parent) {
1278 /* is there enough space? */
1279 uint64_t myspace =
1280 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1281
1282 /* no rename into our descendant */
1283 if (closest_common_ancestor(dd, ra->newparent) == dd)
1284 return (EINVAL);
1285
1286 if (err = dsl_dir_transfer_possible(dd->dd_parent,
1287 ra->newparent, myspace))
1288 return (err);
1289 }
1290
1291 return (0);
1292 }
1293
1294 static void
1295 dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1296 {
1297 dsl_dir_t *dd = arg1;
1298 struct renamearg *ra = arg2;
1299 dsl_pool_t *dp = dd->dd_pool;
1300 objset_t *mos = dp->dp_meta_objset;
1301 int err;
1302 char namebuf[MAXNAMELEN];
1303
1304 ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2);
1305
1306 /* Log this before we change the name. */
1307 dsl_dir_name(ra->newparent, namebuf);
1308 spa_history_log_internal_dd(dd, "rename", tx,
1309 "-> %s/%s", namebuf, ra->mynewname);
1310
1311 if (ra->newparent != dd->dd_parent) {
1312 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1313 -dd->dd_phys->dd_used_bytes,
1314 -dd->dd_phys->dd_compressed_bytes,
1315 -dd->dd_phys->dd_uncompressed_bytes, tx);
1316 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1317 dd->dd_phys->dd_used_bytes,
1318 dd->dd_phys->dd_compressed_bytes,
1319 dd->dd_phys->dd_uncompressed_bytes, tx);
1320
1321 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1322 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1323 dd->dd_phys->dd_used_bytes;
1324
1325 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1326 -unused_rsrv, 0, 0, tx);
1327 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1328 unused_rsrv, 0, 0, tx);
1329 }
1330 }
1331
1332 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1333
1334 /* remove from old parent zapobj */
1335 err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1336 dd->dd_myname, tx);
1337 ASSERT0(err);
1338
1339 (void) strcpy(dd->dd_myname, ra->mynewname);
1340 dsl_dir_close(dd->dd_parent, dd);
1341 dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1342 VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1343 ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1344
1345 /* add to new parent zapobj */
1346 err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1347 dd->dd_myname, 8, 1, &dd->dd_object, tx);
1348 ASSERT0(err);
1349
1350 }
1351
1352 int
1353 dsl_dir_rename(dsl_dir_t *dd, const char *newname)
1354 {
1355 struct renamearg ra;
1356 int err;
1357
1358 /* new parent should exist */
1359 err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1360 if (err)
1361 return (err);
1362
1363 /* can't rename to different pool */
1364 if (dd->dd_pool != ra.newparent->dd_pool) {
1365 err = ENXIO;
1366 goto out;
1367 }
1368
1369 /* new name should not already exist */
1370 if (ra.mynewname == NULL) {
1371 err = EEXIST;
1372 goto out;
1373 }
1374
1375 err = dsl_sync_task_do(dd->dd_pool,
1376 dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1377
1378 out:
1379 dsl_dir_close(ra.newparent, FTAG);
1380 return (err);
1381 }
1382
1383 int
1384 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1385 {
1386 dsl_dir_t *ancestor;
1387 int64_t adelta;
1388 uint64_t avail;
1389
1390 ancestor = closest_common_ancestor(sdd, tdd);
1391 adelta = would_change(sdd, -space, ancestor);
1392 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1393 if (avail < space)
1394 return (ENOSPC);
1395
1396 return (0);
1397 }
1398
1399 timestruc_t
1400 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1401 {
1402 timestruc_t t;
1403
1404 mutex_enter(&dd->dd_lock);
1405 t = dd->dd_snap_cmtime;
1406 mutex_exit(&dd->dd_lock);
1407
1408 return (t);
1409 }
1410
1411 void
1412 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1413 {
1414 timestruc_t t;
1415
1416 gethrestime(&t);
1417 mutex_enter(&dd->dd_lock);
1418 dd->dd_snap_cmtime = t;
1419 mutex_exit(&dd->dd_lock);
1420 }