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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
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(void *arg1, void *arg2,
44 cred_t *cr, dmu_tx_t *tx);
45
46
47 /* ARGSUSED */
48 static void
dsl_dir_evict(dmu_buf_t * db,void * arg)49 dsl_dir_evict(dmu_buf_t *db, void *arg)
50 {
51 dsl_dir_t *dd = arg;
52 dsl_pool_t *dp = dd->dd_pool;
53 int t;
54
55 for (t = 0; t < TXG_SIZE; t++) {
56 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
57 ASSERT(dd->dd_tempreserved[t] == 0);
58 ASSERT(dd->dd_space_towrite[t] == 0);
59 }
60
61 if (dd->dd_parent)
62 dsl_dir_close(dd->dd_parent, dd);
63
64 spa_close(dd->dd_pool->dp_spa, dd);
65
66 /*
67 * The props callback list should be empty since they hold the
68 * dir open.
69 */
70 list_destroy(&dd->dd_prop_cbs);
71 mutex_destroy(&dd->dd_lock);
72 kmem_free(dd, sizeof (dsl_dir_t));
73 }
74
75 int
dsl_dir_open_obj(dsl_pool_t * dp,uint64_t ddobj,const char * tail,void * tag,dsl_dir_t ** ddp)76 dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
77 const char *tail, void *tag, dsl_dir_t **ddp)
78 {
79 dmu_buf_t *dbuf;
80 dsl_dir_t *dd;
81 int err;
82
83 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
84 dsl_pool_sync_context(dp));
85
86 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
87 if (err)
88 return (err);
89 dd = dmu_buf_get_user(dbuf);
90 #ifdef ZFS_DEBUG
91 {
92 dmu_object_info_t doi;
93 dmu_object_info_from_db(dbuf, &doi);
94 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
95 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
96 }
97 #endif
98 if (dd == NULL) {
99 dsl_dir_t *winner;
100
101 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
102 dd->dd_object = ddobj;
103 dd->dd_dbuf = dbuf;
104 dd->dd_pool = dp;
105 dd->dd_phys = dbuf->db_data;
106 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
107
108 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
109 offsetof(dsl_prop_cb_record_t, cbr_node));
110
111 dsl_dir_snap_cmtime_update(dd);
112
113 if (dd->dd_phys->dd_parent_obj) {
114 err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
115 NULL, dd, &dd->dd_parent);
116 if (err)
117 goto errout;
118 if (tail) {
119 #ifdef ZFS_DEBUG
120 uint64_t foundobj;
121
122 err = zap_lookup(dp->dp_meta_objset,
123 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
124 tail, sizeof (foundobj), 1, &foundobj);
125 ASSERT(err || foundobj == ddobj);
126 #endif
127 (void) strcpy(dd->dd_myname, tail);
128 } else {
129 err = zap_value_search(dp->dp_meta_objset,
130 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
131 ddobj, 0, dd->dd_myname);
132 }
133 if (err)
134 goto errout;
135 } else {
136 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
137 }
138
139 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
140 dsl_dir_evict);
141 if (winner) {
142 if (dd->dd_parent)
143 dsl_dir_close(dd->dd_parent, dd);
144 mutex_destroy(&dd->dd_lock);
145 kmem_free(dd, sizeof (dsl_dir_t));
146 dd = winner;
147 } else {
148 spa_open_ref(dp->dp_spa, dd);
149 }
150 }
151
152 /*
153 * The dsl_dir_t has both open-to-close and instantiate-to-evict
154 * holds on the spa. We need the open-to-close holds because
155 * otherwise the spa_refcnt wouldn't change when we open a
156 * dir which the spa also has open, so we could incorrectly
157 * think it was OK to unload/export/destroy the pool. We need
158 * the instantiate-to-evict hold because the dsl_dir_t has a
159 * pointer to the dd_pool, which has a pointer to the spa_t.
160 */
161 spa_open_ref(dp->dp_spa, tag);
162 ASSERT3P(dd->dd_pool, ==, dp);
163 ASSERT3U(dd->dd_object, ==, ddobj);
164 ASSERT3P(dd->dd_dbuf, ==, dbuf);
165 *ddp = dd;
166 return (0);
167
168 errout:
169 if (dd->dd_parent)
170 dsl_dir_close(dd->dd_parent, dd);
171 mutex_destroy(&dd->dd_lock);
172 kmem_free(dd, sizeof (dsl_dir_t));
173 dmu_buf_rele(dbuf, tag);
174 return (err);
175
176 }
177
178 void
dsl_dir_close(dsl_dir_t * dd,void * tag)179 dsl_dir_close(dsl_dir_t *dd, void *tag)
180 {
181 dprintf_dd(dd, "%s\n", "");
182 spa_close(dd->dd_pool->dp_spa, tag);
183 dmu_buf_rele(dd->dd_dbuf, tag);
184 }
185
186 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
187 void
dsl_dir_name(dsl_dir_t * dd,char * buf)188 dsl_dir_name(dsl_dir_t *dd, char *buf)
189 {
190 if (dd->dd_parent) {
191 dsl_dir_name(dd->dd_parent, buf);
192 (void) strcat(buf, "/");
193 } else {
194 buf[0] = '\0';
195 }
196 if (!MUTEX_HELD(&dd->dd_lock)) {
197 /*
198 * recursive mutex so that we can use
199 * dprintf_dd() with dd_lock held
200 */
201 mutex_enter(&dd->dd_lock);
202 (void) strcat(buf, dd->dd_myname);
203 mutex_exit(&dd->dd_lock);
204 } else {
205 (void) strcat(buf, dd->dd_myname);
206 }
207 }
208
209 /* Calculate name legnth, avoiding all the strcat calls of dsl_dir_name */
210 int
dsl_dir_namelen(dsl_dir_t * dd)211 dsl_dir_namelen(dsl_dir_t *dd)
212 {
213 int result = 0;
214
215 if (dd->dd_parent) {
216 /* parent's name + 1 for the "/" */
217 result = dsl_dir_namelen(dd->dd_parent) + 1;
218 }
219
220 if (!MUTEX_HELD(&dd->dd_lock)) {
221 /* see dsl_dir_name */
222 mutex_enter(&dd->dd_lock);
223 result += strlen(dd->dd_myname);
224 mutex_exit(&dd->dd_lock);
225 } else {
226 result += strlen(dd->dd_myname);
227 }
228
229 return (result);
230 }
231
232 static int
getcomponent(const char * path,char * component,const char ** nextp)233 getcomponent(const char *path, char *component, const char **nextp)
234 {
235 char *p;
236 if ((path == NULL) || (path[0] == '\0'))
237 return (ENOENT);
238 /* This would be a good place to reserve some namespace... */
239 p = strpbrk(path, "/@");
240 if (p && (p[1] == '/' || p[1] == '@')) {
241 /* two separators in a row */
242 return (EINVAL);
243 }
244 if (p == NULL || p == path) {
245 /*
246 * if the first thing is an @ or /, it had better be an
247 * @ and it had better not have any more ats or slashes,
248 * and it had better have something after the @.
249 */
250 if (p != NULL &&
251 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
252 return (EINVAL);
253 if (strlen(path) >= MAXNAMELEN)
254 return (ENAMETOOLONG);
255 (void) strcpy(component, path);
256 p = NULL;
257 } else if (p[0] == '/') {
258 if (p-path >= MAXNAMELEN)
259 return (ENAMETOOLONG);
260 (void) strncpy(component, path, p - path);
261 component[p-path] = '\0';
262 p++;
263 } else if (p[0] == '@') {
264 /*
265 * if the next separator is an @, there better not be
266 * any more slashes.
267 */
268 if (strchr(path, '/'))
269 return (EINVAL);
270 if (p-path >= MAXNAMELEN)
271 return (ENAMETOOLONG);
272 (void) strncpy(component, path, p - path);
273 component[p-path] = '\0';
274 } else {
275 ASSERT(!"invalid p");
276 }
277 *nextp = p;
278 return (0);
279 }
280
281 /*
282 * same as dsl_open_dir, ignore the first component of name and use the
283 * spa instead
284 */
285 int
dsl_dir_open_spa(spa_t * spa,const char * name,void * tag,dsl_dir_t ** ddp,const char ** tailp)286 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
287 dsl_dir_t **ddp, const char **tailp)
288 {
289 char buf[MAXNAMELEN];
290 const char *next, *nextnext = NULL;
291 int err;
292 dsl_dir_t *dd;
293 dsl_pool_t *dp;
294 uint64_t ddobj;
295 int openedspa = FALSE;
296
297 dprintf("%s\n", name);
298
299 err = getcomponent(name, buf, &next);
300 if (err)
301 return (err);
302 if (spa == NULL) {
303 err = spa_open(buf, &spa, FTAG);
304 if (err) {
305 dprintf("spa_open(%s) failed\n", buf);
306 return (err);
307 }
308 openedspa = TRUE;
309
310 /* XXX this assertion belongs in spa_open */
311 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
312 }
313
314 dp = spa_get_dsl(spa);
315
316 rw_enter(&dp->dp_config_rwlock, RW_READER);
317 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
318 if (err) {
319 rw_exit(&dp->dp_config_rwlock);
320 if (openedspa)
321 spa_close(spa, FTAG);
322 return (err);
323 }
324
325 while (next != NULL) {
326 dsl_dir_t *child_ds;
327 err = getcomponent(next, buf, &nextnext);
328 if (err)
329 break;
330 ASSERT(next[0] != '\0');
331 if (next[0] == '@')
332 break;
333 dprintf("looking up %s in obj%lld\n",
334 buf, dd->dd_phys->dd_child_dir_zapobj);
335
336 err = zap_lookup(dp->dp_meta_objset,
337 dd->dd_phys->dd_child_dir_zapobj,
338 buf, sizeof (ddobj), 1, &ddobj);
339 if (err) {
340 if (err == ENOENT)
341 err = 0;
342 break;
343 }
344
345 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
346 if (err)
347 break;
348 dsl_dir_close(dd, tag);
349 dd = child_ds;
350 next = nextnext;
351 }
352 rw_exit(&dp->dp_config_rwlock);
353
354 if (err) {
355 dsl_dir_close(dd, tag);
356 if (openedspa)
357 spa_close(spa, FTAG);
358 return (err);
359 }
360
361 /*
362 * It's an error if there's more than one component left, or
363 * tailp==NULL and there's any component left.
364 */
365 if (next != NULL &&
366 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
367 /* bad path name */
368 dsl_dir_close(dd, tag);
369 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
370 err = ENOENT;
371 }
372 if (tailp)
373 *tailp = next;
374 if (openedspa)
375 spa_close(spa, FTAG);
376 *ddp = dd;
377 return (err);
378 }
379
380 /*
381 * Return the dsl_dir_t, and possibly the last component which couldn't
382 * be found in *tail. Return NULL if the path is bogus, or if
383 * tail==NULL and we couldn't parse the whole name. (*tail)[0] == '@'
384 * means that the last component is a snapshot.
385 */
386 int
dsl_dir_open(const char * name,void * tag,dsl_dir_t ** ddp,const char ** tailp)387 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
388 {
389 return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
390 }
391
392 uint64_t
dsl_dir_create_sync(dsl_pool_t * dp,dsl_dir_t * pds,const char * name,dmu_tx_t * tx)393 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
394 dmu_tx_t *tx)
395 {
396 objset_t *mos = dp->dp_meta_objset;
397 uint64_t ddobj;
398 dsl_dir_phys_t *dsphys;
399 dmu_buf_t *dbuf;
400
401 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
402 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
403 if (pds) {
404 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
405 name, sizeof (uint64_t), 1, &ddobj, tx));
406 } else {
407 /* it's the root dir */
408 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
409 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
410 }
411 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
412 dmu_buf_will_dirty(dbuf, tx);
413 dsphys = dbuf->db_data;
414
415 dsphys->dd_creation_time = gethrestime_sec();
416 if (pds)
417 dsphys->dd_parent_obj = pds->dd_object;
418 dsphys->dd_props_zapobj = zap_create(mos,
419 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
420 dsphys->dd_child_dir_zapobj = zap_create(mos,
421 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
422 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
423 dsphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
424 dmu_buf_rele(dbuf, FTAG);
425
426 return (ddobj);
427 }
428
429 /* ARGSUSED */
430 int
dsl_dir_destroy_check(void * arg1,void * arg2,dmu_tx_t * tx)431 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
432 {
433 dsl_dataset_t *ds = arg1;
434 dsl_dir_t *dd = ds->ds_dir;
435 dsl_pool_t *dp = dd->dd_pool;
436 objset_t *mos = dp->dp_meta_objset;
437 int err;
438 uint64_t count;
439
440 /*
441 * There should be exactly two holds, both from
442 * dsl_dataset_destroy: one on the dd directory, and one on its
443 * head ds. Otherwise, someone is trying to lookup something
444 * inside this dir while we want to destroy it. The
445 * config_rwlock ensures that nobody else opens it after we
446 * check.
447 */
448 if (dmu_buf_refcount(dd->dd_dbuf) > 2)
449 return (EBUSY);
450
451 err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
452 if (err)
453 return (err);
454 if (count != 0)
455 return (EEXIST);
456
457 return (0);
458 }
459
460 void
dsl_dir_destroy_sync(void * arg1,void * tag,cred_t * cr,dmu_tx_t * tx)461 dsl_dir_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
462 {
463 dsl_dataset_t *ds = arg1;
464 dsl_dir_t *dd = ds->ds_dir;
465 objset_t *mos = dd->dd_pool->dp_meta_objset;
466 dsl_prop_setarg_t psa;
467 uint64_t value = 0;
468 uint64_t obj;
469 dd_used_t t;
470
471 ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
472 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
473
474 /* Remove our reservation. */
475 dsl_prop_setarg_init_uint64(&psa, "reservation",
476 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
477 &value);
478 psa.psa_effective_value = 0; /* predict default value */
479
480 dsl_dir_set_reservation_sync(ds, &psa, cr, tx);
481
482 ASSERT3U(dd->dd_phys->dd_used_bytes, ==, 0);
483 ASSERT3U(dd->dd_phys->dd_reserved, ==, 0);
484 for (t = 0; t < DD_USED_NUM; t++)
485 ASSERT3U(dd->dd_phys->dd_used_breakdown[t], ==, 0);
486
487 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
488 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
489 VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
490 VERIFY(0 == zap_remove(mos,
491 dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
492
493 obj = dd->dd_object;
494 dsl_dir_close(dd, tag);
495 VERIFY(0 == dmu_object_free(mos, obj, tx));
496 }
497
498 boolean_t
dsl_dir_is_clone(dsl_dir_t * dd)499 dsl_dir_is_clone(dsl_dir_t *dd)
500 {
501 return (dd->dd_phys->dd_origin_obj &&
502 (dd->dd_pool->dp_origin_snap == NULL ||
503 dd->dd_phys->dd_origin_obj !=
504 dd->dd_pool->dp_origin_snap->ds_object));
505 }
506
507 void
dsl_dir_stats(dsl_dir_t * dd,nvlist_t * nv)508 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
509 {
510 mutex_enter(&dd->dd_lock);
511 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
512 dd->dd_phys->dd_used_bytes);
513 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
514 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
515 dd->dd_phys->dd_reserved);
516 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
517 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
518 (dd->dd_phys->dd_uncompressed_bytes * 100 /
519 dd->dd_phys->dd_compressed_bytes));
520 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
521 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
522 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
523 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
524 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
525 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
526 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
527 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
528 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
529 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
530 }
531 mutex_exit(&dd->dd_lock);
532
533 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
534 if (dsl_dir_is_clone(dd)) {
535 dsl_dataset_t *ds;
536 char buf[MAXNAMELEN];
537
538 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
539 dd->dd_phys->dd_origin_obj, FTAG, &ds));
540 dsl_dataset_name(ds, buf);
541 dsl_dataset_rele(ds, FTAG);
542 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
543 }
544 rw_exit(&dd->dd_pool->dp_config_rwlock);
545 }
546
547 void
dsl_dir_dirty(dsl_dir_t * dd,dmu_tx_t * tx)548 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
549 {
550 dsl_pool_t *dp = dd->dd_pool;
551
552 ASSERT(dd->dd_phys);
553
554 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
555 /* up the hold count until we can be written out */
556 dmu_buf_add_ref(dd->dd_dbuf, dd);
557 }
558 }
559
560 static int64_t
parent_delta(dsl_dir_t * dd,uint64_t used,int64_t delta)561 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
562 {
563 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
564 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
565 return (new_accounted - old_accounted);
566 }
567
568 void
dsl_dir_sync(dsl_dir_t * dd,dmu_tx_t * tx)569 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
570 {
571 ASSERT(dmu_tx_is_syncing(tx));
572
573 dmu_buf_will_dirty(dd->dd_dbuf, tx);
574
575 mutex_enter(&dd->dd_lock);
576 ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0);
577 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
578 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
579 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
580 mutex_exit(&dd->dd_lock);
581
582 /* release the hold from dsl_dir_dirty */
583 dmu_buf_rele(dd->dd_dbuf, dd);
584 }
585
586 static uint64_t
dsl_dir_space_towrite(dsl_dir_t * dd)587 dsl_dir_space_towrite(dsl_dir_t *dd)
588 {
589 uint64_t space = 0;
590 int i;
591
592 ASSERT(MUTEX_HELD(&dd->dd_lock));
593
594 for (i = 0; i < TXG_SIZE; i++) {
595 space += dd->dd_space_towrite[i&TXG_MASK];
596 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
597 }
598 return (space);
599 }
600
601 /*
602 * How much space would dd have available if ancestor had delta applied
603 * to it? If ondiskonly is set, we're only interested in what's
604 * on-disk, not estimated pending changes.
605 */
606 uint64_t
dsl_dir_space_available(dsl_dir_t * dd,dsl_dir_t * ancestor,int64_t delta,int ondiskonly)607 dsl_dir_space_available(dsl_dir_t *dd,
608 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
609 {
610 uint64_t parentspace, myspace, quota, used;
611
612 /*
613 * If there are no restrictions otherwise, assume we have
614 * unlimited space available.
615 */
616 quota = UINT64_MAX;
617 parentspace = UINT64_MAX;
618
619 if (dd->dd_parent != NULL) {
620 parentspace = dsl_dir_space_available(dd->dd_parent,
621 ancestor, delta, ondiskonly);
622 }
623
624 mutex_enter(&dd->dd_lock);
625 if (dd->dd_phys->dd_quota != 0)
626 quota = dd->dd_phys->dd_quota;
627 used = dd->dd_phys->dd_used_bytes;
628 if (!ondiskonly)
629 used += dsl_dir_space_towrite(dd);
630
631 if (dd->dd_parent == NULL) {
632 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
633 quota = MIN(quota, poolsize);
634 }
635
636 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
637 /*
638 * We have some space reserved, in addition to what our
639 * parent gave us.
640 */
641 parentspace += dd->dd_phys->dd_reserved - used;
642 }
643
644 if (dd == ancestor) {
645 ASSERT(delta <= 0);
646 ASSERT(used >= -delta);
647 used += delta;
648 if (parentspace != UINT64_MAX)
649 parentspace -= delta;
650 }
651
652 if (used > quota) {
653 /* over quota */
654 myspace = 0;
655
656 /*
657 * While it's OK to be a little over quota, if
658 * we think we are using more space than there
659 * is in the pool (which is already 1.6% more than
660 * dsl_pool_adjustedsize()), something is very
661 * wrong.
662 */
663 ASSERT3U(used, <=, spa_get_dspace(dd->dd_pool->dp_spa));
664 } else {
665 /*
666 * the lesser of the space provided by our parent and
667 * the space left in our quota
668 */
669 myspace = MIN(parentspace, quota - used);
670 }
671
672 mutex_exit(&dd->dd_lock);
673
674 return (myspace);
675 }
676
677 struct tempreserve {
678 list_node_t tr_node;
679 dsl_pool_t *tr_dp;
680 dsl_dir_t *tr_ds;
681 uint64_t tr_size;
682 };
683
684 static int
dsl_dir_tempreserve_impl(dsl_dir_t * dd,uint64_t asize,boolean_t netfree,boolean_t ignorequota,boolean_t checkrefquota,list_t * tr_list,dmu_tx_t * tx,boolean_t first)685 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
686 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
687 dmu_tx_t *tx, boolean_t first)
688 {
689 uint64_t txg = tx->tx_txg;
690 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
691 uint64_t deferred = 0;
692 struct tempreserve *tr;
693 int retval = EDQUOT;
694 int txgidx = txg & TXG_MASK;
695 int i;
696 uint64_t ref_rsrv = 0;
697
698 ASSERT3U(txg, !=, 0);
699 ASSERT3S(asize, >, 0);
700
701 mutex_enter(&dd->dd_lock);
702
703 /*
704 * Check against the dsl_dir's quota. We don't add in the delta
705 * when checking for over-quota because they get one free hit.
706 */
707 est_inflight = dsl_dir_space_towrite(dd);
708 for (i = 0; i < TXG_SIZE; i++)
709 est_inflight += dd->dd_tempreserved[i];
710 used_on_disk = dd->dd_phys->dd_used_bytes;
711
712 /*
713 * On the first iteration, fetch the dataset's used-on-disk and
714 * refreservation values. Also, if checkrefquota is set, test if
715 * allocating this space would exceed the dataset's refquota.
716 */
717 if (first && tx->tx_objset) {
718 int error;
719 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
720
721 error = dsl_dataset_check_quota(ds, checkrefquota,
722 asize, est_inflight, &used_on_disk, &ref_rsrv);
723 if (error) {
724 mutex_exit(&dd->dd_lock);
725 return (error);
726 }
727 }
728
729 /*
730 * If this transaction will result in a net free of space,
731 * we want to let it through.
732 */
733 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
734 quota = UINT64_MAX;
735 else
736 quota = dd->dd_phys->dd_quota;
737
738 /*
739 * Adjust the quota against the actual pool size at the root
740 * minus any outstanding deferred frees.
741 * To ensure that it's possible to remove files from a full
742 * pool without inducing transient overcommits, we throttle
743 * netfree transactions against a quota that is slightly larger,
744 * but still within the pool's allocation slop. In cases where
745 * we're very close to full, this will allow a steady trickle of
746 * removes to get through.
747 */
748 if (dd->dd_parent == NULL) {
749 spa_t *spa = dd->dd_pool->dp_spa;
750 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
751 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
752 if (poolsize - deferred < quota) {
753 quota = poolsize - deferred;
754 retval = ENOSPC;
755 }
756 }
757
758 /*
759 * If they are requesting more space, and our current estimate
760 * is over quota, they get to try again unless the actual
761 * on-disk is over quota and there are no pending changes (which
762 * may free up space for us).
763 */
764 if (used_on_disk + est_inflight >= quota) {
765 if (est_inflight > 0 || used_on_disk < quota ||
766 (retval == ENOSPC && used_on_disk < quota + deferred))
767 retval = ERESTART;
768 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
769 "quota=%lluK tr=%lluK err=%d\n",
770 used_on_disk>>10, est_inflight>>10,
771 quota>>10, asize>>10, retval);
772 mutex_exit(&dd->dd_lock);
773 return (retval);
774 }
775
776 /* We need to up our estimated delta before dropping dd_lock */
777 dd->dd_tempreserved[txgidx] += asize;
778
779 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
780 asize - ref_rsrv);
781 mutex_exit(&dd->dd_lock);
782
783 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
784 tr->tr_ds = dd;
785 tr->tr_size = asize;
786 list_insert_tail(tr_list, tr);
787
788 /* see if it's OK with our parent */
789 if (dd->dd_parent && parent_rsrv) {
790 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
791
792 return (dsl_dir_tempreserve_impl(dd->dd_parent,
793 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
794 } else {
795 return (0);
796 }
797 }
798
799 /*
800 * Reserve space in this dsl_dir, to be used in this tx's txg.
801 * After the space has been dirtied (and dsl_dir_willuse_space()
802 * has been called), the reservation should be canceled, using
803 * dsl_dir_tempreserve_clear().
804 */
805 int
dsl_dir_tempreserve_space(dsl_dir_t * dd,uint64_t lsize,uint64_t asize,uint64_t fsize,uint64_t usize,void ** tr_cookiep,dmu_tx_t * tx)806 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
807 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
808 {
809 int err;
810 list_t *tr_list;
811
812 if (asize == 0) {
813 *tr_cookiep = NULL;
814 return (0);
815 }
816
817 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
818 list_create(tr_list, sizeof (struct tempreserve),
819 offsetof(struct tempreserve, tr_node));
820 ASSERT3S(asize, >, 0);
821 ASSERT3S(fsize, >=, 0);
822
823 err = arc_tempreserve_space(lsize, tx->tx_txg);
824 if (err == 0) {
825 struct tempreserve *tr;
826
827 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
828 tr->tr_size = lsize;
829 list_insert_tail(tr_list, tr);
830
831 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
832 } else {
833 if (err == EAGAIN) {
834 txg_delay(dd->dd_pool, tx->tx_txg, 1);
835 err = ERESTART;
836 }
837 dsl_pool_memory_pressure(dd->dd_pool);
838 }
839
840 if (err == 0) {
841 struct tempreserve *tr;
842
843 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
844 tr->tr_dp = dd->dd_pool;
845 tr->tr_size = asize;
846 list_insert_tail(tr_list, tr);
847
848 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
849 FALSE, asize > usize, tr_list, tx, TRUE);
850 }
851
852 if (err)
853 dsl_dir_tempreserve_clear(tr_list, tx);
854 else
855 *tr_cookiep = tr_list;
856
857 return (err);
858 }
859
860 /*
861 * Clear a temporary reservation that we previously made with
862 * dsl_dir_tempreserve_space().
863 */
864 void
dsl_dir_tempreserve_clear(void * tr_cookie,dmu_tx_t * tx)865 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
866 {
867 int txgidx = tx->tx_txg & TXG_MASK;
868 list_t *tr_list = tr_cookie;
869 struct tempreserve *tr;
870
871 ASSERT3U(tx->tx_txg, !=, 0);
872
873 if (tr_cookie == NULL)
874 return;
875
876 while (tr = list_head(tr_list)) {
877 if (tr->tr_dp) {
878 dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
879 } else if (tr->tr_ds) {
880 mutex_enter(&tr->tr_ds->dd_lock);
881 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
882 tr->tr_size);
883 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
884 mutex_exit(&tr->tr_ds->dd_lock);
885 } else {
886 arc_tempreserve_clear(tr->tr_size);
887 }
888 list_remove(tr_list, tr);
889 kmem_free(tr, sizeof (struct tempreserve));
890 }
891
892 kmem_free(tr_list, sizeof (list_t));
893 }
894
895 static void
dsl_dir_willuse_space_impl(dsl_dir_t * dd,int64_t space,dmu_tx_t * tx)896 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
897 {
898 int64_t parent_space;
899 uint64_t est_used;
900
901 mutex_enter(&dd->dd_lock);
902 if (space > 0)
903 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
904
905 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
906 parent_space = parent_delta(dd, est_used, space);
907 mutex_exit(&dd->dd_lock);
908
909 /* Make sure that we clean up dd_space_to* */
910 dsl_dir_dirty(dd, tx);
911
912 /* XXX this is potentially expensive and unnecessary... */
913 if (parent_space && dd->dd_parent)
914 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
915 }
916
917 /*
918 * Call in open context when we think we're going to write/free space,
919 * eg. when dirtying data. Be conservative (ie. OK to write less than
920 * this or free more than this, but don't write more or free less).
921 */
922 void
dsl_dir_willuse_space(dsl_dir_t * dd,int64_t space,dmu_tx_t * tx)923 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
924 {
925 dsl_pool_willuse_space(dd->dd_pool, space, tx);
926 dsl_dir_willuse_space_impl(dd, space, tx);
927 }
928
929 /* call from syncing context when we actually write/free space for this dd */
930 void
dsl_dir_diduse_space(dsl_dir_t * dd,dd_used_t type,int64_t used,int64_t compressed,int64_t uncompressed,dmu_tx_t * tx)931 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
932 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
933 {
934 int64_t accounted_delta;
935 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
936
937 ASSERT(dmu_tx_is_syncing(tx));
938 ASSERT(type < DD_USED_NUM);
939
940 dsl_dir_dirty(dd, tx);
941
942 if (needlock)
943 mutex_enter(&dd->dd_lock);
944 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
945 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
946 ASSERT(compressed >= 0 ||
947 dd->dd_phys->dd_compressed_bytes >= -compressed);
948 ASSERT(uncompressed >= 0 ||
949 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
950 dd->dd_phys->dd_used_bytes += used;
951 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
952 dd->dd_phys->dd_compressed_bytes += compressed;
953
954 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
955 ASSERT(used > 0 ||
956 dd->dd_phys->dd_used_breakdown[type] >= -used);
957 dd->dd_phys->dd_used_breakdown[type] += used;
958 #ifdef DEBUG
959 dd_used_t t;
960 uint64_t u = 0;
961 for (t = 0; t < DD_USED_NUM; t++)
962 u += dd->dd_phys->dd_used_breakdown[t];
963 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
964 #endif
965 }
966 if (needlock)
967 mutex_exit(&dd->dd_lock);
968
969 if (dd->dd_parent != NULL) {
970 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
971 accounted_delta, compressed, uncompressed, tx);
972 dsl_dir_transfer_space(dd->dd_parent,
973 used - accounted_delta,
974 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
975 }
976 }
977
978 void
dsl_dir_transfer_space(dsl_dir_t * dd,int64_t delta,dd_used_t oldtype,dd_used_t newtype,dmu_tx_t * tx)979 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
980 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
981 {
982 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
983
984 ASSERT(dmu_tx_is_syncing(tx));
985 ASSERT(oldtype < DD_USED_NUM);
986 ASSERT(newtype < DD_USED_NUM);
987
988 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
989 return;
990
991 dsl_dir_dirty(dd, tx);
992 if (needlock)
993 mutex_enter(&dd->dd_lock);
994 ASSERT(delta > 0 ?
995 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
996 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
997 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
998 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
999 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1000 if (needlock)
1001 mutex_exit(&dd->dd_lock);
1002 }
1003
1004 static int
dsl_dir_set_quota_check(void * arg1,void * arg2,dmu_tx_t * tx)1005 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1006 {
1007 dsl_dataset_t *ds = arg1;
1008 dsl_dir_t *dd = ds->ds_dir;
1009 dsl_prop_setarg_t *psa = arg2;
1010 int err;
1011 uint64_t towrite;
1012
1013 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1014 return (err);
1015
1016 if (psa->psa_effective_value == 0)
1017 return (0);
1018
1019 mutex_enter(&dd->dd_lock);
1020 /*
1021 * If we are doing the preliminary check in open context, and
1022 * there are pending changes, then don't fail it, since the
1023 * pending changes could under-estimate the amount of space to be
1024 * freed up.
1025 */
1026 towrite = dsl_dir_space_towrite(dd);
1027 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1028 (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
1029 psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
1030 err = ENOSPC;
1031 }
1032 mutex_exit(&dd->dd_lock);
1033 return (err);
1034 }
1035
1036 extern void dsl_prop_set_sync(void *, void *, cred_t *, dmu_tx_t *);
1037
1038 /* ARGSUSED */
1039 static void
dsl_dir_set_quota_sync(void * arg1,void * arg2,cred_t * cr,dmu_tx_t * tx)1040 dsl_dir_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1041 {
1042 dsl_dataset_t *ds = arg1;
1043 dsl_dir_t *dd = ds->ds_dir;
1044 dsl_prop_setarg_t *psa = arg2;
1045 uint64_t effective_value = psa->psa_effective_value;
1046
1047 dsl_prop_set_sync(ds, psa, cr, tx);
1048 DSL_PROP_CHECK_PREDICTION(dd, psa);
1049
1050 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1051
1052 mutex_enter(&dd->dd_lock);
1053 dd->dd_phys->dd_quota = effective_value;
1054 mutex_exit(&dd->dd_lock);
1055
1056 spa_history_internal_log(LOG_DS_QUOTA, dd->dd_pool->dp_spa,
1057 tx, cr, "%lld dataset = %llu ",
1058 (longlong_t)effective_value, dd->dd_phys->dd_head_dataset_obj);
1059 }
1060
1061 int
dsl_dir_set_quota(const char * ddname,zprop_source_t source,uint64_t quota)1062 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1063 {
1064 dsl_dir_t *dd;
1065 dsl_dataset_t *ds;
1066 dsl_prop_setarg_t psa;
1067 int err;
1068
1069 dsl_prop_setarg_init_uint64(&psa, "quota", source, "a);
1070
1071 err = dsl_dataset_hold(ddname, FTAG, &ds);
1072 if (err)
1073 return (err);
1074
1075 err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1076 if (err) {
1077 dsl_dataset_rele(ds, FTAG);
1078 return (err);
1079 }
1080
1081 ASSERT(ds->ds_dir == dd);
1082
1083 /*
1084 * If someone removes a file, then tries to set the quota, we want to
1085 * make sure the file freeing takes effect.
1086 */
1087 txg_wait_open(dd->dd_pool, 0);
1088
1089 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1090 dsl_dir_set_quota_sync, ds, &psa, 0);
1091
1092 dsl_dir_close(dd, FTAG);
1093 dsl_dataset_rele(ds, FTAG);
1094 return (err);
1095 }
1096
1097 int
dsl_dir_set_reservation_check(void * arg1,void * arg2,dmu_tx_t * tx)1098 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1099 {
1100 dsl_dataset_t *ds = arg1;
1101 dsl_dir_t *dd = ds->ds_dir;
1102 dsl_prop_setarg_t *psa = arg2;
1103 uint64_t effective_value;
1104 uint64_t used, avail;
1105 int err;
1106
1107 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1108 return (err);
1109
1110 effective_value = psa->psa_effective_value;
1111
1112 /*
1113 * If we are doing the preliminary check in open context, the
1114 * space estimates may be inaccurate.
1115 */
1116 if (!dmu_tx_is_syncing(tx))
1117 return (0);
1118
1119 mutex_enter(&dd->dd_lock);
1120 used = dd->dd_phys->dd_used_bytes;
1121 mutex_exit(&dd->dd_lock);
1122
1123 if (dd->dd_parent) {
1124 avail = dsl_dir_space_available(dd->dd_parent,
1125 NULL, 0, FALSE);
1126 } else {
1127 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1128 }
1129
1130 if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
1131 uint64_t delta = MAX(used, effective_value) -
1132 MAX(used, dd->dd_phys->dd_reserved);
1133
1134 if (delta > avail)
1135 return (ENOSPC);
1136 if (dd->dd_phys->dd_quota > 0 &&
1137 effective_value > dd->dd_phys->dd_quota)
1138 return (ENOSPC);
1139 }
1140
1141 return (0);
1142 }
1143
1144 /* ARGSUSED */
1145 static void
dsl_dir_set_reservation_sync(void * arg1,void * arg2,cred_t * cr,dmu_tx_t * tx)1146 dsl_dir_set_reservation_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1147 {
1148 dsl_dataset_t *ds = arg1;
1149 dsl_dir_t *dd = ds->ds_dir;
1150 dsl_prop_setarg_t *psa = arg2;
1151 uint64_t effective_value = psa->psa_effective_value;
1152 uint64_t used;
1153 int64_t delta;
1154
1155 dsl_prop_set_sync(ds, psa, cr, tx);
1156 DSL_PROP_CHECK_PREDICTION(dd, psa);
1157
1158 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1159
1160 mutex_enter(&dd->dd_lock);
1161 used = dd->dd_phys->dd_used_bytes;
1162 delta = MAX(used, effective_value) -
1163 MAX(used, dd->dd_phys->dd_reserved);
1164 dd->dd_phys->dd_reserved = effective_value;
1165
1166 if (dd->dd_parent != NULL) {
1167 /* Roll up this additional usage into our ancestors */
1168 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1169 delta, 0, 0, tx);
1170 }
1171 mutex_exit(&dd->dd_lock);
1172
1173 spa_history_internal_log(LOG_DS_RESERVATION, dd->dd_pool->dp_spa,
1174 tx, cr, "%lld dataset = %llu",
1175 (longlong_t)effective_value, dd->dd_phys->dd_head_dataset_obj);
1176 }
1177
1178 int
dsl_dir_set_reservation(const char * ddname,zprop_source_t source,uint64_t reservation)1179 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1180 uint64_t reservation)
1181 {
1182 dsl_dir_t *dd;
1183 dsl_dataset_t *ds;
1184 dsl_prop_setarg_t psa;
1185 int err;
1186
1187 dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
1188
1189 err = dsl_dataset_hold(ddname, FTAG, &ds);
1190 if (err)
1191 return (err);
1192
1193 err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1194 if (err) {
1195 dsl_dataset_rele(ds, FTAG);
1196 return (err);
1197 }
1198
1199 ASSERT(ds->ds_dir == dd);
1200
1201 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1202 dsl_dir_set_reservation_sync, ds, &psa, 0);
1203
1204 dsl_dir_close(dd, FTAG);
1205 dsl_dataset_rele(ds, FTAG);
1206 return (err);
1207 }
1208
1209 static dsl_dir_t *
closest_common_ancestor(dsl_dir_t * ds1,dsl_dir_t * ds2)1210 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1211 {
1212 for (; ds1; ds1 = ds1->dd_parent) {
1213 dsl_dir_t *dd;
1214 for (dd = ds2; dd; dd = dd->dd_parent) {
1215 if (ds1 == dd)
1216 return (dd);
1217 }
1218 }
1219 return (NULL);
1220 }
1221
1222 /*
1223 * If delta is applied to dd, how much of that delta would be applied to
1224 * ancestor? Syncing context only.
1225 */
1226 static int64_t
would_change(dsl_dir_t * dd,int64_t delta,dsl_dir_t * ancestor)1227 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1228 {
1229 if (dd == ancestor)
1230 return (delta);
1231
1232 mutex_enter(&dd->dd_lock);
1233 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1234 mutex_exit(&dd->dd_lock);
1235 return (would_change(dd->dd_parent, delta, ancestor));
1236 }
1237
1238 struct renamearg {
1239 dsl_dir_t *newparent;
1240 const char *mynewname;
1241 };
1242
1243 /*ARGSUSED*/
1244 static int
dsl_dir_rename_check(void * arg1,void * arg2,dmu_tx_t * tx)1245 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1246 {
1247 dsl_dir_t *dd = arg1;
1248 struct renamearg *ra = arg2;
1249 dsl_pool_t *dp = dd->dd_pool;
1250 objset_t *mos = dp->dp_meta_objset;
1251 int err;
1252 uint64_t val;
1253
1254 /* There should be 2 references: the open and the dirty */
1255 if (dmu_buf_refcount(dd->dd_dbuf) > 2)
1256 return (EBUSY);
1257
1258 /* check for existing name */
1259 err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1260 ra->mynewname, 8, 1, &val);
1261 if (err == 0)
1262 return (EEXIST);
1263 if (err != ENOENT)
1264 return (err);
1265
1266 if (ra->newparent != dd->dd_parent) {
1267 /* is there enough space? */
1268 uint64_t myspace =
1269 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1270
1271 /* no rename into our descendant */
1272 if (closest_common_ancestor(dd, ra->newparent) == dd)
1273 return (EINVAL);
1274
1275 if (err = dsl_dir_transfer_possible(dd->dd_parent,
1276 ra->newparent, myspace))
1277 return (err);
1278 }
1279
1280 return (0);
1281 }
1282
1283 static void
dsl_dir_rename_sync(void * arg1,void * arg2,cred_t * cr,dmu_tx_t * tx)1284 dsl_dir_rename_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1285 {
1286 dsl_dir_t *dd = arg1;
1287 struct renamearg *ra = arg2;
1288 dsl_pool_t *dp = dd->dd_pool;
1289 objset_t *mos = dp->dp_meta_objset;
1290 int err;
1291
1292 ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2);
1293
1294 if (ra->newparent != dd->dd_parent) {
1295 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1296 -dd->dd_phys->dd_used_bytes,
1297 -dd->dd_phys->dd_compressed_bytes,
1298 -dd->dd_phys->dd_uncompressed_bytes, tx);
1299 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1300 dd->dd_phys->dd_used_bytes,
1301 dd->dd_phys->dd_compressed_bytes,
1302 dd->dd_phys->dd_uncompressed_bytes, tx);
1303
1304 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1305 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1306 dd->dd_phys->dd_used_bytes;
1307
1308 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1309 -unused_rsrv, 0, 0, tx);
1310 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1311 unused_rsrv, 0, 0, tx);
1312 }
1313 }
1314
1315 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1316
1317 /* remove from old parent zapobj */
1318 err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1319 dd->dd_myname, tx);
1320 ASSERT3U(err, ==, 0);
1321
1322 (void) strcpy(dd->dd_myname, ra->mynewname);
1323 dsl_dir_close(dd->dd_parent, dd);
1324 dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1325 VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1326 ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1327
1328 /* add to new parent zapobj */
1329 err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1330 dd->dd_myname, 8, 1, &dd->dd_object, tx);
1331 ASSERT3U(err, ==, 0);
1332
1333 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa,
1334 tx, cr, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj);
1335 }
1336
1337 int
dsl_dir_rename(dsl_dir_t * dd,const char * newname)1338 dsl_dir_rename(dsl_dir_t *dd, const char *newname)
1339 {
1340 struct renamearg ra;
1341 int err;
1342
1343 /* new parent should exist */
1344 err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1345 if (err)
1346 return (err);
1347
1348 /* can't rename to different pool */
1349 if (dd->dd_pool != ra.newparent->dd_pool) {
1350 err = ENXIO;
1351 goto out;
1352 }
1353
1354 /* new name should not already exist */
1355 if (ra.mynewname == NULL) {
1356 err = EEXIST;
1357 goto out;
1358 }
1359
1360 err = dsl_sync_task_do(dd->dd_pool,
1361 dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1362
1363 out:
1364 dsl_dir_close(ra.newparent, FTAG);
1365 return (err);
1366 }
1367
1368 int
dsl_dir_transfer_possible(dsl_dir_t * sdd,dsl_dir_t * tdd,uint64_t space)1369 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1370 {
1371 dsl_dir_t *ancestor;
1372 int64_t adelta;
1373 uint64_t avail;
1374
1375 ancestor = closest_common_ancestor(sdd, tdd);
1376 adelta = would_change(sdd, -space, ancestor);
1377 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1378 if (avail < space)
1379 return (ENOSPC);
1380
1381 return (0);
1382 }
1383
1384 timestruc_t
dsl_dir_snap_cmtime(dsl_dir_t * dd)1385 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1386 {
1387 timestruc_t t;
1388
1389 mutex_enter(&dd->dd_lock);
1390 t = dd->dd_snap_cmtime;
1391 mutex_exit(&dd->dd_lock);
1392
1393 return (t);
1394 }
1395
1396 void
dsl_dir_snap_cmtime_update(dsl_dir_t * dd)1397 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1398 {
1399 timestruc_t t;
1400
1401 gethrestime(&t);
1402 mutex_enter(&dd->dd_lock);
1403 dd->dd_snap_cmtime = t;
1404 mutex_exit(&dd->dd_lock);
1405 }
1406