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) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
29 */
30
31 #include <sys/dmu.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_deleg.h>
39 #include <sys/dmu_impl.h>
40 #include <sys/spa.h>
41 #include <sys/metaslab.h>
42 #include <sys/zap.h>
43 #include <sys/zio.h>
44 #include <sys/arc.h>
45 #include <sys/sunddi.h>
46 #include <sys/zvol.h>
47 #ifdef _KERNEL
48 #include <sys/zfs_vfsops.h>
49 #endif
50 #include <sys/zfeature.h>
51 #include <sys/policy.h>
52 #include <sys/zfs_znode.h>
53 #include "zfs_namecheck.h"
54 #include "zfs_prop.h"
55
56 /*
57 * Filesystem and Snapshot Limits
58 * ------------------------------
59 *
60 * These limits are used to restrict the number of filesystems and/or snapshots
61 * that can be created at a given level in the tree or below. A typical
62 * use-case is with a delegated dataset where the administrator wants to ensure
63 * that a user within the zone is not creating too many additional filesystems
64 * or snapshots, even though they're not exceeding their space quota.
65 *
66 * The filesystem and snapshot counts are stored as extensible properties. This
67 * capability is controlled by a feature flag and must be enabled to be used.
68 * Once enabled, the feature is not active until the first limit is set. At
69 * that point, future operations to create/destroy filesystems or snapshots
70 * will validate and update the counts.
71 *
72 * Because the count properties will not exist before the feature is active,
73 * the counts are updated when a limit is first set on an uninitialized
74 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
75 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
76 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
77 * snapshot count properties on a node indicate uninitialized counts on that
78 * node.) When first setting a limit on an uninitialized node, the code starts
79 * at the filesystem with the new limit and descends into all sub-filesystems
80 * to add the count properties.
81 *
82 * In practice this is lightweight since a limit is typically set when the
83 * filesystem is created and thus has no children. Once valid, changing the
84 * limit value won't require a re-traversal since the counts are already valid.
85 * When recursively fixing the counts, if a node with a limit is encountered
86 * during the descent, the counts are known to be valid and there is no need to
87 * descend into that filesystem's children. The counts on filesystems above the
88 * one with the new limit will still be uninitialized, unless a limit is
89 * eventually set on one of those filesystems. The counts are always recursively
90 * updated when a limit is set on a dataset, unless there is already a limit.
91 * When a new limit value is set on a filesystem with an existing limit, it is
92 * possible for the new limit to be less than the current count at that level
93 * since a user who can change the limit is also allowed to exceed the limit.
94 *
95 * Once the feature is active, then whenever a filesystem or snapshot is
96 * created, the code recurses up the tree, validating the new count against the
97 * limit at each initialized level. In practice, most levels will not have a
98 * limit set. If there is a limit at any initialized level up the tree, the
99 * check must pass or the creation will fail. Likewise, when a filesystem or
100 * snapshot is destroyed, the counts are recursively adjusted all the way up
101 * the initizized nodes in the tree. Renaming a filesystem into different point
102 * in the tree will first validate, then update the counts on each branch up to
103 * the common ancestor. A receive will also validate the counts and then update
104 * them.
105 *
106 * An exception to the above behavior is that the limit is not enforced if the
107 * user has permission to modify the limit. This is primarily so that
108 * recursive snapshots in the global zone always work. We want to prevent a
109 * denial-of-service in which a lower level delegated dataset could max out its
110 * limit and thus block recursive snapshots from being taken in the global zone.
111 * Because of this, it is possible for the snapshot count to be over the limit
112 * and snapshots taken in the global zone could cause a lower level dataset to
113 * hit or exceed its limit. The administrator taking the global zone recursive
114 * snapshot should be aware of this side-effect and behave accordingly.
115 * For consistency, the filesystem limit is also not enforced if the user can
116 * modify the limit.
117 *
118 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
119 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
120 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
121 * dsl_dir_init_fs_ss_count().
122 *
123 * There is a special case when we receive a filesystem that already exists. In
124 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
125 * never update the filesystem counts for temporary clones.
126 *
127 * Likewise, we do not update the snapshot counts for temporary snapshots,
128 * such as those created by zfs diff.
129 */
130
131 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
132
133 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
134
135 static void
dsl_dir_evict_async(void * dbu)136 dsl_dir_evict_async(void *dbu)
137 {
138 dsl_dir_t *dd = dbu;
139 dsl_pool_t *dp = dd->dd_pool;
140 int t;
141
142 dd->dd_dbuf = NULL;
143
144 for (t = 0; t < TXG_SIZE; t++) {
145 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
146 ASSERT(dd->dd_tempreserved[t] == 0);
147 ASSERT(dd->dd_space_towrite[t] == 0);
148 }
149
150 if (dd->dd_parent)
151 dsl_dir_async_rele(dd->dd_parent, dd);
152
153 spa_async_close(dd->dd_pool->dp_spa, dd);
154
155 dsl_prop_fini(dd);
156 mutex_destroy(&dd->dd_lock);
157 kmem_free(dd, sizeof (dsl_dir_t));
158 }
159
160 int
dsl_dir_hold_obj(dsl_pool_t * dp,uint64_t ddobj,const char * tail,void * tag,dsl_dir_t ** ddp)161 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
162 const char *tail, void *tag, dsl_dir_t **ddp)
163 {
164 dmu_buf_t *dbuf;
165 dsl_dir_t *dd;
166 int err;
167
168 ASSERT(dsl_pool_config_held(dp));
169
170 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
171 if (err != 0)
172 return (err);
173 dd = dmu_buf_get_user(dbuf);
174 #ifdef ZFS_DEBUG
175 {
176 dmu_object_info_t doi;
177 dmu_object_info_from_db(dbuf, &doi);
178 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
179 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
180 }
181 #endif
182 if (dd == NULL) {
183 dsl_dir_t *winner;
184
185 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
186 dd->dd_object = ddobj;
187 dd->dd_dbuf = dbuf;
188 dd->dd_pool = dp;
189 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
190 dsl_prop_init(dd);
191
192 dsl_dir_snap_cmtime_update(dd);
193
194 if (dsl_dir_phys(dd)->dd_parent_obj) {
195 err = dsl_dir_hold_obj(dp,
196 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
197 &dd->dd_parent);
198 if (err != 0)
199 goto errout;
200 if (tail) {
201 #ifdef ZFS_DEBUG
202 uint64_t foundobj;
203
204 err = zap_lookup(dp->dp_meta_objset,
205 dsl_dir_phys(dd->dd_parent)->
206 dd_child_dir_zapobj, tail,
207 sizeof (foundobj), 1, &foundobj);
208 ASSERT(err || foundobj == ddobj);
209 #endif
210 (void) strcpy(dd->dd_myname, tail);
211 } else {
212 err = zap_value_search(dp->dp_meta_objset,
213 dsl_dir_phys(dd->dd_parent)->
214 dd_child_dir_zapobj,
215 ddobj, 0, dd->dd_myname);
216 }
217 if (err != 0)
218 goto errout;
219 } else {
220 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
221 }
222
223 if (dsl_dir_is_clone(dd)) {
224 dmu_buf_t *origin_bonus;
225 dsl_dataset_phys_t *origin_phys;
226
227 /*
228 * We can't open the origin dataset, because
229 * that would require opening this dsl_dir.
230 * Just look at its phys directly instead.
231 */
232 err = dmu_bonus_hold(dp->dp_meta_objset,
233 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
234 &origin_bonus);
235 if (err != 0)
236 goto errout;
237 origin_phys = origin_bonus->db_data;
238 dd->dd_origin_txg =
239 origin_phys->ds_creation_txg;
240 dmu_buf_rele(origin_bonus, FTAG);
241 }
242
243 dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
244 &dd->dd_dbuf);
245 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
246 if (winner != NULL) {
247 if (dd->dd_parent)
248 dsl_dir_rele(dd->dd_parent, dd);
249 dsl_prop_fini(dd);
250 mutex_destroy(&dd->dd_lock);
251 kmem_free(dd, sizeof (dsl_dir_t));
252 dd = winner;
253 } else {
254 spa_open_ref(dp->dp_spa, dd);
255 }
256 }
257
258 /*
259 * The dsl_dir_t has both open-to-close and instantiate-to-evict
260 * holds on the spa. We need the open-to-close holds because
261 * otherwise the spa_refcnt wouldn't change when we open a
262 * dir which the spa also has open, so we could incorrectly
263 * think it was OK to unload/export/destroy the pool. We need
264 * the instantiate-to-evict hold because the dsl_dir_t has a
265 * pointer to the dd_pool, which has a pointer to the spa_t.
266 */
267 spa_open_ref(dp->dp_spa, tag);
268 ASSERT3P(dd->dd_pool, ==, dp);
269 ASSERT3U(dd->dd_object, ==, ddobj);
270 ASSERT3P(dd->dd_dbuf, ==, dbuf);
271 *ddp = dd;
272 return (0);
273
274 errout:
275 if (dd->dd_parent)
276 dsl_dir_rele(dd->dd_parent, dd);
277 dsl_prop_fini(dd);
278 mutex_destroy(&dd->dd_lock);
279 kmem_free(dd, sizeof (dsl_dir_t));
280 dmu_buf_rele(dbuf, tag);
281 return (err);
282 }
283
284 void
dsl_dir_rele(dsl_dir_t * dd,void * tag)285 dsl_dir_rele(dsl_dir_t *dd, void *tag)
286 {
287 dprintf_dd(dd, "%s\n", "");
288 spa_close(dd->dd_pool->dp_spa, tag);
289 dmu_buf_rele(dd->dd_dbuf, tag);
290 }
291
292 /*
293 * Remove a reference to the given dsl dir that is being asynchronously
294 * released. Async releases occur from a taskq performing eviction of
295 * dsl datasets and dirs. This process is identical to a normal release
296 * with the exception of using the async API for releasing the reference on
297 * the spa.
298 */
299 void
dsl_dir_async_rele(dsl_dir_t * dd,void * tag)300 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
301 {
302 dprintf_dd(dd, "%s\n", "");
303 spa_async_close(dd->dd_pool->dp_spa, tag);
304 dmu_buf_rele(dd->dd_dbuf, tag);
305 }
306
307 /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
308 void
dsl_dir_name(dsl_dir_t * dd,char * buf)309 dsl_dir_name(dsl_dir_t *dd, char *buf)
310 {
311 if (dd->dd_parent) {
312 dsl_dir_name(dd->dd_parent, buf);
313 VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
314 ZFS_MAX_DATASET_NAME_LEN);
315 } else {
316 buf[0] = '\0';
317 }
318 if (!MUTEX_HELD(&dd->dd_lock)) {
319 /*
320 * recursive mutex so that we can use
321 * dprintf_dd() with dd_lock held
322 */
323 mutex_enter(&dd->dd_lock);
324 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
325 <, ZFS_MAX_DATASET_NAME_LEN);
326 mutex_exit(&dd->dd_lock);
327 } else {
328 VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
329 <, ZFS_MAX_DATASET_NAME_LEN);
330 }
331 }
332
333 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
334 int
dsl_dir_namelen(dsl_dir_t * dd)335 dsl_dir_namelen(dsl_dir_t *dd)
336 {
337 int result = 0;
338
339 if (dd->dd_parent) {
340 /* parent's name + 1 for the "/" */
341 result = dsl_dir_namelen(dd->dd_parent) + 1;
342 }
343
344 if (!MUTEX_HELD(&dd->dd_lock)) {
345 /* see dsl_dir_name */
346 mutex_enter(&dd->dd_lock);
347 result += strlen(dd->dd_myname);
348 mutex_exit(&dd->dd_lock);
349 } else {
350 result += strlen(dd->dd_myname);
351 }
352
353 return (result);
354 }
355
356 static int
getcomponent(const char * path,char * component,const char ** nextp)357 getcomponent(const char *path, char *component, const char **nextp)
358 {
359 char *p;
360
361 if ((path == NULL) || (path[0] == '\0'))
362 return (SET_ERROR(ENOENT));
363 /* This would be a good place to reserve some namespace... */
364 p = strpbrk(path, "/@");
365 if (p && (p[1] == '/' || p[1] == '@')) {
366 /* two separators in a row */
367 return (SET_ERROR(EINVAL));
368 }
369 if (p == NULL || p == path) {
370 /*
371 * if the first thing is an @ or /, it had better be an
372 * @ and it had better not have any more ats or slashes,
373 * and it had better have something after the @.
374 */
375 if (p != NULL &&
376 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
377 return (SET_ERROR(EINVAL));
378 if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
379 return (SET_ERROR(ENAMETOOLONG));
380 (void) strcpy(component, path);
381 p = NULL;
382 } else if (p[0] == '/') {
383 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
384 return (SET_ERROR(ENAMETOOLONG));
385 (void) strncpy(component, path, p - path);
386 component[p - path] = '\0';
387 p++;
388 } else if (p[0] == '@') {
389 /*
390 * if the next separator is an @, there better not be
391 * any more slashes.
392 */
393 if (strchr(path, '/'))
394 return (SET_ERROR(EINVAL));
395 if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
396 return (SET_ERROR(ENAMETOOLONG));
397 (void) strncpy(component, path, p - path);
398 component[p - path] = '\0';
399 } else {
400 panic("invalid p=%p", (void *)p);
401 }
402 *nextp = p;
403 return (0);
404 }
405
406 /*
407 * Return the dsl_dir_t, and possibly the last component which couldn't
408 * be found in *tail. The name must be in the specified dsl_pool_t. This
409 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
410 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
411 * (*tail)[0] == '@' means that the last component is a snapshot.
412 */
413 int
dsl_dir_hold(dsl_pool_t * dp,const char * name,void * tag,dsl_dir_t ** ddp,const char ** tailp)414 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
415 dsl_dir_t **ddp, const char **tailp)
416 {
417 char buf[ZFS_MAX_DATASET_NAME_LEN];
418 const char *spaname, *next, *nextnext = NULL;
419 int err;
420 dsl_dir_t *dd;
421 uint64_t ddobj;
422
423 err = getcomponent(name, buf, &next);
424 if (err != 0)
425 return (err);
426
427 /* Make sure the name is in the specified pool. */
428 spaname = spa_name(dp->dp_spa);
429 if (strcmp(buf, spaname) != 0)
430 return (SET_ERROR(EXDEV));
431
432 ASSERT(dsl_pool_config_held(dp));
433
434 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
435 if (err != 0) {
436 return (err);
437 }
438
439 while (next != NULL) {
440 dsl_dir_t *child_dd;
441 err = getcomponent(next, buf, &nextnext);
442 if (err != 0)
443 break;
444 ASSERT(next[0] != '\0');
445 if (next[0] == '@')
446 break;
447 dprintf("looking up %s in obj%lld\n",
448 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
449
450 err = zap_lookup(dp->dp_meta_objset,
451 dsl_dir_phys(dd)->dd_child_dir_zapobj,
452 buf, sizeof (ddobj), 1, &ddobj);
453 if (err != 0) {
454 if (err == ENOENT)
455 err = 0;
456 break;
457 }
458
459 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
460 if (err != 0)
461 break;
462 dsl_dir_rele(dd, tag);
463 dd = child_dd;
464 next = nextnext;
465 }
466
467 if (err != 0) {
468 dsl_dir_rele(dd, tag);
469 return (err);
470 }
471
472 /*
473 * It's an error if there's more than one component left, or
474 * tailp==NULL and there's any component left.
475 */
476 if (next != NULL &&
477 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
478 /* bad path name */
479 dsl_dir_rele(dd, tag);
480 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
481 err = SET_ERROR(ENOENT);
482 }
483 if (tailp != NULL)
484 *tailp = next;
485 *ddp = dd;
486 return (err);
487 }
488
489 /*
490 * If the counts are already initialized for this filesystem and its
491 * descendants then do nothing, otherwise initialize the counts.
492 *
493 * The counts on this filesystem, and those below, may be uninitialized due to
494 * either the use of a pre-existing pool which did not support the
495 * filesystem/snapshot limit feature, or one in which the feature had not yet
496 * been enabled.
497 *
498 * Recursively descend the filesystem tree and update the filesystem/snapshot
499 * counts on each filesystem below, then update the cumulative count on the
500 * current filesystem. If the filesystem already has a count set on it,
501 * then we know that its counts, and the counts on the filesystems below it,
502 * are already correct, so we don't have to update this filesystem.
503 */
504 static void
dsl_dir_init_fs_ss_count(dsl_dir_t * dd,dmu_tx_t * tx)505 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
506 {
507 uint64_t my_fs_cnt = 0;
508 uint64_t my_ss_cnt = 0;
509 dsl_pool_t *dp = dd->dd_pool;
510 objset_t *os = dp->dp_meta_objset;
511 zap_cursor_t *zc;
512 zap_attribute_t *za;
513 dsl_dataset_t *ds;
514
515 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
516 ASSERT(dsl_pool_config_held(dp));
517 ASSERT(dmu_tx_is_syncing(tx));
518
519 dsl_dir_zapify(dd, tx);
520
521 /*
522 * If the filesystem count has already been initialized then we
523 * don't need to recurse down any further.
524 */
525 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
526 return;
527
528 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
529 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
530
531 /* Iterate my child dirs */
532 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
533 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
534 dsl_dir_t *chld_dd;
535 uint64_t count;
536
537 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
538 &chld_dd));
539
540 /*
541 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
542 * temporary datasets.
543 */
544 if (chld_dd->dd_myname[0] == '$' ||
545 chld_dd->dd_myname[0] == '%') {
546 dsl_dir_rele(chld_dd, FTAG);
547 continue;
548 }
549
550 my_fs_cnt++; /* count this child */
551
552 dsl_dir_init_fs_ss_count(chld_dd, tx);
553
554 VERIFY0(zap_lookup(os, chld_dd->dd_object,
555 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
556 my_fs_cnt += count;
557 VERIFY0(zap_lookup(os, chld_dd->dd_object,
558 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
559 my_ss_cnt += count;
560
561 dsl_dir_rele(chld_dd, FTAG);
562 }
563 zap_cursor_fini(zc);
564 /* Count my snapshots (we counted children's snapshots above) */
565 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
566 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
567
568 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
569 zap_cursor_retrieve(zc, za) == 0;
570 zap_cursor_advance(zc)) {
571 /* Don't count temporary snapshots */
572 if (za->za_name[0] != '%')
573 my_ss_cnt++;
574 }
575 zap_cursor_fini(zc);
576
577 dsl_dataset_rele(ds, FTAG);
578
579 kmem_free(zc, sizeof (zap_cursor_t));
580 kmem_free(za, sizeof (zap_attribute_t));
581
582 /* we're in a sync task, update counts */
583 dmu_buf_will_dirty(dd->dd_dbuf, tx);
584 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
585 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
586 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
587 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
588 }
589
590 static int
dsl_dir_actv_fs_ss_limit_check(void * arg,dmu_tx_t * tx)591 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
592 {
593 char *ddname = (char *)arg;
594 dsl_pool_t *dp = dmu_tx_pool(tx);
595 dsl_dataset_t *ds;
596 dsl_dir_t *dd;
597 int error;
598
599 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
600 if (error != 0)
601 return (error);
602
603 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
604 dsl_dataset_rele(ds, FTAG);
605 return (SET_ERROR(ENOTSUP));
606 }
607
608 dd = ds->ds_dir;
609 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
610 dsl_dir_is_zapified(dd) &&
611 zap_contains(dp->dp_meta_objset, dd->dd_object,
612 DD_FIELD_FILESYSTEM_COUNT) == 0) {
613 dsl_dataset_rele(ds, FTAG);
614 return (SET_ERROR(EALREADY));
615 }
616
617 dsl_dataset_rele(ds, FTAG);
618 return (0);
619 }
620
621 static void
dsl_dir_actv_fs_ss_limit_sync(void * arg,dmu_tx_t * tx)622 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
623 {
624 char *ddname = (char *)arg;
625 dsl_pool_t *dp = dmu_tx_pool(tx);
626 dsl_dataset_t *ds;
627 spa_t *spa;
628
629 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
630
631 spa = dsl_dataset_get_spa(ds);
632
633 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
634 /*
635 * Since the feature was not active and we're now setting a
636 * limit, increment the feature-active counter so that the
637 * feature becomes active for the first time.
638 *
639 * We are already in a sync task so we can update the MOS.
640 */
641 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
642 }
643
644 /*
645 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
646 * we need to ensure the counts are correct. Descend down the tree from
647 * this point and update all of the counts to be accurate.
648 */
649 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
650
651 dsl_dataset_rele(ds, FTAG);
652 }
653
654 /*
655 * Make sure the feature is enabled and activate it if necessary.
656 * Since we're setting a limit, ensure the on-disk counts are valid.
657 * This is only called by the ioctl path when setting a limit value.
658 *
659 * We do not need to validate the new limit, since users who can change the
660 * limit are also allowed to exceed the limit.
661 */
662 int
dsl_dir_activate_fs_ss_limit(const char * ddname)663 dsl_dir_activate_fs_ss_limit(const char *ddname)
664 {
665 int error;
666
667 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
668 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
669 ZFS_SPACE_CHECK_RESERVED);
670
671 if (error == EALREADY)
672 error = 0;
673
674 return (error);
675 }
676
677 /*
678 * Used to determine if the filesystem_limit or snapshot_limit should be
679 * enforced. We allow the limit to be exceeded if the user has permission to
680 * write the property value. We pass in the creds that we got in the open
681 * context since we will always be the GZ root in syncing context. We also have
682 * to handle the case where we are allowed to change the limit on the current
683 * dataset, but there may be another limit in the tree above.
684 *
685 * We can never modify these two properties within a non-global zone. In
686 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
687 * can't use that function since we are already holding the dp_config_rwlock.
688 * In addition, we already have the dd and dealing with snapshots is simplified
689 * in this code.
690 */
691
692 typedef enum {
693 ENFORCE_ALWAYS,
694 ENFORCE_NEVER,
695 ENFORCE_ABOVE
696 } enforce_res_t;
697
698 static enforce_res_t
dsl_enforce_ds_ss_limits(dsl_dir_t * dd,zfs_prop_t prop,cred_t * cr)699 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
700 {
701 enforce_res_t enforce = ENFORCE_ALWAYS;
702 uint64_t obj;
703 dsl_dataset_t *ds;
704 uint64_t zoned;
705
706 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
707 prop == ZFS_PROP_SNAPSHOT_LIMIT);
708
709 #ifdef _KERNEL
710 #ifdef illumos
711 if (crgetzoneid(cr) != GLOBAL_ZONEID)
712 #endif
713 #ifdef __FreeBSD__
714 if (jailed(cr))
715 #endif
716 #ifdef __NetBSD__
717 if (0)
718 #endif
719 return (ENFORCE_ALWAYS);
720
721 if (secpolicy_zfs(cr) == 0)
722 return (ENFORCE_NEVER);
723 #endif
724
725 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
726 return (ENFORCE_ALWAYS);
727
728 ASSERT(dsl_pool_config_held(dd->dd_pool));
729
730 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
731 return (ENFORCE_ALWAYS);
732
733 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
734 /* Only root can access zoned fs's from the GZ */
735 enforce = ENFORCE_ALWAYS;
736 } else {
737 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
738 enforce = ENFORCE_ABOVE;
739 }
740
741 dsl_dataset_rele(ds, FTAG);
742 return (enforce);
743 }
744
745 /*
746 * Check if adding additional child filesystem(s) would exceed any filesystem
747 * limits or adding additional snapshot(s) would exceed any snapshot limits.
748 * The prop argument indicates which limit to check.
749 *
750 * Note that all filesystem limits up to the root (or the highest
751 * initialized) filesystem or the given ancestor must be satisfied.
752 */
753 int
dsl_fs_ss_limit_check(dsl_dir_t * dd,uint64_t delta,zfs_prop_t prop,dsl_dir_t * ancestor,cred_t * cr)754 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
755 dsl_dir_t *ancestor, cred_t *cr)
756 {
757 objset_t *os = dd->dd_pool->dp_meta_objset;
758 uint64_t limit, count;
759 char *count_prop;
760 enforce_res_t enforce;
761 int err = 0;
762
763 ASSERT(dsl_pool_config_held(dd->dd_pool));
764 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
765 prop == ZFS_PROP_SNAPSHOT_LIMIT);
766
767 /*
768 * If we're allowed to change the limit, don't enforce the limit
769 * e.g. this can happen if a snapshot is taken by an administrative
770 * user in the global zone (i.e. a recursive snapshot by root).
771 * However, we must handle the case of delegated permissions where we
772 * are allowed to change the limit on the current dataset, but there
773 * is another limit in the tree above.
774 */
775 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
776 if (enforce == ENFORCE_NEVER)
777 return (0);
778
779 /*
780 * e.g. if renaming a dataset with no snapshots, count adjustment
781 * is 0.
782 */
783 if (delta == 0)
784 return (0);
785
786 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
787 /*
788 * We don't enforce the limit for temporary snapshots. This is
789 * indicated by a NULL cred_t argument.
790 */
791 if (cr == NULL)
792 return (0);
793
794 count_prop = DD_FIELD_SNAPSHOT_COUNT;
795 } else {
796 count_prop = DD_FIELD_FILESYSTEM_COUNT;
797 }
798
799 /*
800 * If an ancestor has been provided, stop checking the limit once we
801 * hit that dir. We need this during rename so that we don't overcount
802 * the check once we recurse up to the common ancestor.
803 */
804 if (ancestor == dd)
805 return (0);
806
807 /*
808 * If we hit an uninitialized node while recursing up the tree, we can
809 * stop since we know there is no limit here (or above). The counts are
810 * not valid on this node and we know we won't touch this node's counts.
811 */
812 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
813 count_prop, sizeof (count), 1, &count) == ENOENT)
814 return (0);
815
816 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
817 B_FALSE);
818 if (err != 0)
819 return (err);
820
821 /* Is there a limit which we've hit? */
822 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
823 return (SET_ERROR(EDQUOT));
824
825 if (dd->dd_parent != NULL)
826 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
827 ancestor, cr);
828
829 return (err);
830 }
831
832 /*
833 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
834 * parents. When a new filesystem/snapshot is created, increment the count on
835 * all parents, and when a filesystem/snapshot is destroyed, decrement the
836 * count.
837 */
838 void
dsl_fs_ss_count_adjust(dsl_dir_t * dd,int64_t delta,const char * prop,dmu_tx_t * tx)839 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
840 dmu_tx_t *tx)
841 {
842 int err;
843 objset_t *os = dd->dd_pool->dp_meta_objset;
844 uint64_t count;
845
846 ASSERT(dsl_pool_config_held(dd->dd_pool));
847 ASSERT(dmu_tx_is_syncing(tx));
848 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
849 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
850
851 /*
852 * When we receive an incremental stream into a filesystem that already
853 * exists, a temporary clone is created. We don't count this temporary
854 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
855 * $MOS & $ORIGIN) objsets.
856 */
857 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
858 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
859 return;
860
861 /*
862 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
863 */
864 if (delta == 0)
865 return;
866
867 /*
868 * If we hit an uninitialized node while recursing up the tree, we can
869 * stop since we know the counts are not valid on this node and we
870 * know we shouldn't touch this node's counts. An uninitialized count
871 * on the node indicates that either the feature has not yet been
872 * activated or there are no limits on this part of the tree.
873 */
874 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
875 prop, sizeof (count), 1, &count)) == ENOENT)
876 return;
877 VERIFY0(err);
878
879 count += delta;
880 /* Use a signed verify to make sure we're not neg. */
881 VERIFY3S(count, >=, 0);
882
883 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
884 tx));
885
886 /* Roll up this additional count into our ancestors */
887 if (dd->dd_parent != NULL)
888 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
889 }
890
891 uint64_t
dsl_dir_create_sync(dsl_pool_t * dp,dsl_dir_t * pds,const char * name,dmu_tx_t * tx)892 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
893 dmu_tx_t *tx)
894 {
895 objset_t *mos = dp->dp_meta_objset;
896 uint64_t ddobj;
897 dsl_dir_phys_t *ddphys;
898 dmu_buf_t *dbuf;
899
900 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
901 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
902 if (pds) {
903 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
904 name, sizeof (uint64_t), 1, &ddobj, tx));
905 } else {
906 /* it's the root dir */
907 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
908 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
909 }
910 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
911 dmu_buf_will_dirty(dbuf, tx);
912 ddphys = dbuf->db_data;
913
914 ddphys->dd_creation_time = gethrestime_sec();
915 if (pds) {
916 ddphys->dd_parent_obj = pds->dd_object;
917
918 /* update the filesystem counts */
919 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
920 }
921 ddphys->dd_props_zapobj = zap_create(mos,
922 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
923 ddphys->dd_child_dir_zapobj = zap_create(mos,
924 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
925 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
926 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
927 dmu_buf_rele(dbuf, FTAG);
928
929 return (ddobj);
930 }
931
932 boolean_t
dsl_dir_is_clone(dsl_dir_t * dd)933 dsl_dir_is_clone(dsl_dir_t *dd)
934 {
935 return (dsl_dir_phys(dd)->dd_origin_obj &&
936 (dd->dd_pool->dp_origin_snap == NULL ||
937 dsl_dir_phys(dd)->dd_origin_obj !=
938 dd->dd_pool->dp_origin_snap->ds_object));
939 }
940
941 void
dsl_dir_stats(dsl_dir_t * dd,nvlist_t * nv)942 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
943 {
944 mutex_enter(&dd->dd_lock);
945 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
946 dsl_dir_phys(dd)->dd_used_bytes);
947 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
948 dsl_dir_phys(dd)->dd_quota);
949 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
950 dsl_dir_phys(dd)->dd_reserved);
951 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
952 dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
953 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
954 dsl_dir_phys(dd)->dd_compressed_bytes));
955 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
956 dsl_dir_phys(dd)->dd_uncompressed_bytes);
957 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
958 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
959 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
960 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
961 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
962 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
963 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
964 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
965 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
966 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
967 }
968 mutex_exit(&dd->dd_lock);
969
970 if (dsl_dir_is_zapified(dd)) {
971 uint64_t count;
972 objset_t *os = dd->dd_pool->dp_meta_objset;
973
974 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
975 sizeof (count), 1, &count) == 0) {
976 dsl_prop_nvlist_add_uint64(nv,
977 ZFS_PROP_FILESYSTEM_COUNT, count);
978 }
979 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
980 sizeof (count), 1, &count) == 0) {
981 dsl_prop_nvlist_add_uint64(nv,
982 ZFS_PROP_SNAPSHOT_COUNT, count);
983 }
984 }
985
986 if (dsl_dir_is_clone(dd)) {
987 dsl_dataset_t *ds;
988 char buf[ZFS_MAX_DATASET_NAME_LEN];
989
990 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
991 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
992 dsl_dataset_name(ds, buf);
993 dsl_dataset_rele(ds, FTAG);
994 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
995 }
996 }
997
998 void
dsl_dir_dirty(dsl_dir_t * dd,dmu_tx_t * tx)999 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1000 {
1001 dsl_pool_t *dp = dd->dd_pool;
1002
1003 ASSERT(dsl_dir_phys(dd));
1004
1005 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
1006 /* up the hold count until we can be written out */
1007 dmu_buf_add_ref(dd->dd_dbuf, dd);
1008 }
1009 }
1010
1011 static int64_t
parent_delta(dsl_dir_t * dd,uint64_t used,int64_t delta)1012 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1013 {
1014 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1015 uint64_t new_accounted =
1016 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1017 return (new_accounted - old_accounted);
1018 }
1019
1020 void
dsl_dir_sync(dsl_dir_t * dd,dmu_tx_t * tx)1021 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1022 {
1023 ASSERT(dmu_tx_is_syncing(tx));
1024
1025 mutex_enter(&dd->dd_lock);
1026 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1027 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1028 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1029 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1030 mutex_exit(&dd->dd_lock);
1031
1032 /* release the hold from dsl_dir_dirty */
1033 dmu_buf_rele(dd->dd_dbuf, dd);
1034 }
1035
1036 static uint64_t
dsl_dir_space_towrite(dsl_dir_t * dd)1037 dsl_dir_space_towrite(dsl_dir_t *dd)
1038 {
1039 uint64_t space = 0;
1040 int i;
1041
1042 ASSERT(MUTEX_HELD(&dd->dd_lock));
1043
1044 for (i = 0; i < TXG_SIZE; i++) {
1045 space += dd->dd_space_towrite[i&TXG_MASK];
1046 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1047 }
1048 return (space);
1049 }
1050
1051 /*
1052 * How much space would dd have available if ancestor had delta applied
1053 * to it? If ondiskonly is set, we're only interested in what's
1054 * on-disk, not estimated pending changes.
1055 */
1056 uint64_t
dsl_dir_space_available(dsl_dir_t * dd,dsl_dir_t * ancestor,int64_t delta,int ondiskonly)1057 dsl_dir_space_available(dsl_dir_t *dd,
1058 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1059 {
1060 uint64_t parentspace, myspace, quota, used;
1061
1062 /*
1063 * If there are no restrictions otherwise, assume we have
1064 * unlimited space available.
1065 */
1066 quota = UINT64_MAX;
1067 parentspace = UINT64_MAX;
1068
1069 if (dd->dd_parent != NULL) {
1070 parentspace = dsl_dir_space_available(dd->dd_parent,
1071 ancestor, delta, ondiskonly);
1072 }
1073
1074 mutex_enter(&dd->dd_lock);
1075 if (dsl_dir_phys(dd)->dd_quota != 0)
1076 quota = dsl_dir_phys(dd)->dd_quota;
1077 used = dsl_dir_phys(dd)->dd_used_bytes;
1078 if (!ondiskonly)
1079 used += dsl_dir_space_towrite(dd);
1080
1081 if (dd->dd_parent == NULL) {
1082 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1083 quota = MIN(quota, poolsize);
1084 }
1085
1086 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1087 /*
1088 * We have some space reserved, in addition to what our
1089 * parent gave us.
1090 */
1091 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1092 }
1093
1094 if (dd == ancestor) {
1095 ASSERT(delta <= 0);
1096 ASSERT(used >= -delta);
1097 used += delta;
1098 if (parentspace != UINT64_MAX)
1099 parentspace -= delta;
1100 }
1101
1102 if (used > quota) {
1103 /* over quota */
1104 myspace = 0;
1105 } else {
1106 /*
1107 * the lesser of the space provided by our parent and
1108 * the space left in our quota
1109 */
1110 myspace = MIN(parentspace, quota - used);
1111 }
1112
1113 mutex_exit(&dd->dd_lock);
1114
1115 return (myspace);
1116 }
1117
1118 struct tempreserve {
1119 list_node_t tr_node;
1120 dsl_dir_t *tr_ds;
1121 uint64_t tr_size;
1122 };
1123
1124 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)1125 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1126 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1127 dmu_tx_t *tx, boolean_t first)
1128 {
1129 uint64_t txg = tx->tx_txg;
1130 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1131 uint64_t deferred = 0;
1132 struct tempreserve *tr;
1133 int retval = EDQUOT;
1134 int txgidx = txg & TXG_MASK;
1135 int i;
1136 uint64_t ref_rsrv = 0;
1137
1138 ASSERT3U(txg, !=, 0);
1139 ASSERT3S(asize, >, 0);
1140
1141 mutex_enter(&dd->dd_lock);
1142
1143 /*
1144 * Check against the dsl_dir's quota. We don't add in the delta
1145 * when checking for over-quota because they get one free hit.
1146 */
1147 est_inflight = dsl_dir_space_towrite(dd);
1148 for (i = 0; i < TXG_SIZE; i++)
1149 est_inflight += dd->dd_tempreserved[i];
1150 used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1151
1152 /*
1153 * On the first iteration, fetch the dataset's used-on-disk and
1154 * refreservation values. Also, if checkrefquota is set, test if
1155 * allocating this space would exceed the dataset's refquota.
1156 */
1157 if (first && tx->tx_objset) {
1158 int error;
1159 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1160
1161 error = dsl_dataset_check_quota(ds, checkrefquota,
1162 asize, est_inflight, &used_on_disk, &ref_rsrv);
1163 if (error) {
1164 mutex_exit(&dd->dd_lock);
1165 return (error);
1166 }
1167 }
1168
1169 /*
1170 * If this transaction will result in a net free of space,
1171 * we want to let it through.
1172 */
1173 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1174 quota = UINT64_MAX;
1175 else
1176 quota = dsl_dir_phys(dd)->dd_quota;
1177
1178 /*
1179 * Adjust the quota against the actual pool size at the root
1180 * minus any outstanding deferred frees.
1181 * To ensure that it's possible to remove files from a full
1182 * pool without inducing transient overcommits, we throttle
1183 * netfree transactions against a quota that is slightly larger,
1184 * but still within the pool's allocation slop. In cases where
1185 * we're very close to full, this will allow a steady trickle of
1186 * removes to get through.
1187 */
1188 if (dd->dd_parent == NULL) {
1189 spa_t *spa = dd->dd_pool->dp_spa;
1190 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1191 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1192 if (poolsize - deferred < quota) {
1193 quota = poolsize - deferred;
1194 retval = ENOSPC;
1195 }
1196 }
1197
1198 /*
1199 * If they are requesting more space, and our current estimate
1200 * is over quota, they get to try again unless the actual
1201 * on-disk is over quota and there are no pending changes (which
1202 * may free up space for us).
1203 */
1204 if (used_on_disk + est_inflight >= quota) {
1205 if (est_inflight > 0 || used_on_disk < quota ||
1206 (retval == ENOSPC && used_on_disk < quota + deferred))
1207 retval = ERESTART;
1208 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1209 "quota=%lluK tr=%lluK err=%d\n",
1210 used_on_disk>>10, est_inflight>>10,
1211 quota>>10, asize>>10, retval);
1212 mutex_exit(&dd->dd_lock);
1213 return (SET_ERROR(retval));
1214 }
1215
1216 /* We need to up our estimated delta before dropping dd_lock */
1217 dd->dd_tempreserved[txgidx] += asize;
1218
1219 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1220 asize - ref_rsrv);
1221 mutex_exit(&dd->dd_lock);
1222
1223 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1224 tr->tr_ds = dd;
1225 tr->tr_size = asize;
1226 list_insert_tail(tr_list, tr);
1227
1228 /* see if it's OK with our parent */
1229 if (dd->dd_parent && parent_rsrv) {
1230 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1231
1232 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1233 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1234 } else {
1235 return (0);
1236 }
1237 }
1238
1239 /*
1240 * Reserve space in this dsl_dir, to be used in this tx's txg.
1241 * After the space has been dirtied (and dsl_dir_willuse_space()
1242 * has been called), the reservation should be canceled, using
1243 * dsl_dir_tempreserve_clear().
1244 */
1245 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)1246 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1247 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1248 {
1249 int err;
1250 list_t *tr_list;
1251
1252 if (asize == 0) {
1253 *tr_cookiep = NULL;
1254 return (0);
1255 }
1256
1257 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1258 list_create(tr_list, sizeof (struct tempreserve),
1259 offsetof(struct tempreserve, tr_node));
1260 ASSERT3S(asize, >, 0);
1261 ASSERT3S(fsize, >=, 0);
1262
1263 err = arc_tempreserve_space(lsize, tx->tx_txg);
1264 if (err == 0) {
1265 struct tempreserve *tr;
1266
1267 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1268 tr->tr_size = lsize;
1269 list_insert_tail(tr_list, tr);
1270 } else {
1271 if (err == EAGAIN) {
1272 /*
1273 * If arc_memory_throttle() detected that pageout
1274 * is running and we are low on memory, we delay new
1275 * non-pageout transactions to give pageout an
1276 * advantage.
1277 *
1278 * It is unfortunate to be delaying while the caller's
1279 * locks are held.
1280 */
1281 txg_delay(dd->dd_pool, tx->tx_txg,
1282 MSEC2NSEC(10), MSEC2NSEC(10));
1283 err = SET_ERROR(ERESTART);
1284 }
1285 }
1286
1287 if (err == 0) {
1288 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1289 FALSE, asize > usize, tr_list, tx, TRUE);
1290 }
1291
1292 if (err != 0)
1293 dsl_dir_tempreserve_clear(tr_list, tx);
1294 else
1295 *tr_cookiep = tr_list;
1296
1297 return (err);
1298 }
1299
1300 /*
1301 * Clear a temporary reservation that we previously made with
1302 * dsl_dir_tempreserve_space().
1303 */
1304 void
dsl_dir_tempreserve_clear(void * tr_cookie,dmu_tx_t * tx)1305 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1306 {
1307 int txgidx = tx->tx_txg & TXG_MASK;
1308 list_t *tr_list = tr_cookie;
1309 struct tempreserve *tr;
1310
1311 ASSERT3U(tx->tx_txg, !=, 0);
1312
1313 if (tr_cookie == NULL)
1314 return;
1315
1316 while ((tr = list_head(tr_list)) != NULL) {
1317 if (tr->tr_ds) {
1318 mutex_enter(&tr->tr_ds->dd_lock);
1319 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1320 tr->tr_size);
1321 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1322 mutex_exit(&tr->tr_ds->dd_lock);
1323 } else {
1324 arc_tempreserve_clear(tr->tr_size);
1325 }
1326 list_remove(tr_list, tr);
1327 kmem_free(tr, sizeof (struct tempreserve));
1328 }
1329
1330 kmem_free(tr_list, sizeof (list_t));
1331 }
1332
1333 /*
1334 * This should be called from open context when we think we're going to write
1335 * or free space, for example when dirtying data. Be conservative; it's okay
1336 * to write less space or free more, but we don't want to write more or free
1337 * less than the amount specified.
1338 */
1339 void
dsl_dir_willuse_space(dsl_dir_t * dd,int64_t space,dmu_tx_t * tx)1340 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1341 {
1342 int64_t parent_space;
1343 uint64_t est_used;
1344
1345 mutex_enter(&dd->dd_lock);
1346 if (space > 0)
1347 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1348
1349 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1350 parent_space = parent_delta(dd, est_used, space);
1351 mutex_exit(&dd->dd_lock);
1352
1353 /* Make sure that we clean up dd_space_to* */
1354 dsl_dir_dirty(dd, tx);
1355
1356 /* XXX this is potentially expensive and unnecessary... */
1357 if (parent_space && dd->dd_parent)
1358 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1359 }
1360
1361 /* call from syncing context when we actually write/free space for this dd */
1362 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)1363 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1364 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1365 {
1366 int64_t accounted_delta;
1367
1368 /*
1369 * dsl_dataset_set_refreservation_sync_impl() calls this with
1370 * dd_lock held, so that it can atomically update
1371 * ds->ds_reserved and the dsl_dir accounting, so that
1372 * dsl_dataset_check_quota() can see dataset and dir accounting
1373 * consistently.
1374 */
1375 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1376
1377 ASSERT(dmu_tx_is_syncing(tx));
1378 ASSERT(type < DD_USED_NUM);
1379
1380 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1381
1382 if (needlock)
1383 mutex_enter(&dd->dd_lock);
1384 accounted_delta =
1385 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1386 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1387 ASSERT(compressed >= 0 ||
1388 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1389 ASSERT(uncompressed >= 0 ||
1390 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1391 dsl_dir_phys(dd)->dd_used_bytes += used;
1392 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1393 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1394
1395 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1396 ASSERT(used > 0 ||
1397 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1398 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1399 #ifdef DEBUG
1400 dd_used_t t;
1401 uint64_t u = 0;
1402 for (t = 0; t < DD_USED_NUM; t++)
1403 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1404 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1405 #endif
1406 }
1407 if (needlock)
1408 mutex_exit(&dd->dd_lock);
1409
1410 if (dd->dd_parent != NULL) {
1411 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1412 accounted_delta, compressed, uncompressed, tx);
1413 dsl_dir_transfer_space(dd->dd_parent,
1414 used - accounted_delta,
1415 DD_USED_CHILD_RSRV, DD_USED_CHILD, NULL);
1416 }
1417 }
1418
1419 void
dsl_dir_transfer_space(dsl_dir_t * dd,int64_t delta,dd_used_t oldtype,dd_used_t newtype,dmu_tx_t * tx)1420 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1421 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1422 {
1423 ASSERT(tx == NULL || dmu_tx_is_syncing(tx));
1424 ASSERT(oldtype < DD_USED_NUM);
1425 ASSERT(newtype < DD_USED_NUM);
1426
1427 if (delta == 0 ||
1428 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1429 return;
1430
1431 if (tx != NULL)
1432 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1433 mutex_enter(&dd->dd_lock);
1434 ASSERT(delta > 0 ?
1435 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1436 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1437 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1438 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1439 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1440 mutex_exit(&dd->dd_lock);
1441 }
1442
1443 typedef struct dsl_dir_set_qr_arg {
1444 const char *ddsqra_name;
1445 zprop_source_t ddsqra_source;
1446 uint64_t ddsqra_value;
1447 } dsl_dir_set_qr_arg_t;
1448
1449 static int
dsl_dir_set_quota_check(void * arg,dmu_tx_t * tx)1450 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1451 {
1452 dsl_dir_set_qr_arg_t *ddsqra = arg;
1453 dsl_pool_t *dp = dmu_tx_pool(tx);
1454 dsl_dataset_t *ds;
1455 int error;
1456 uint64_t towrite, newval;
1457
1458 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1459 if (error != 0)
1460 return (error);
1461
1462 error = dsl_prop_predict(ds->ds_dir, "quota",
1463 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1464 if (error != 0) {
1465 dsl_dataset_rele(ds, FTAG);
1466 return (error);
1467 }
1468
1469 if (newval == 0) {
1470 dsl_dataset_rele(ds, FTAG);
1471 return (0);
1472 }
1473
1474 mutex_enter(&ds->ds_dir->dd_lock);
1475 /*
1476 * If we are doing the preliminary check in open context, and
1477 * there are pending changes, then don't fail it, since the
1478 * pending changes could under-estimate the amount of space to be
1479 * freed up.
1480 */
1481 towrite = dsl_dir_space_towrite(ds->ds_dir);
1482 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1483 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1484 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1485 error = SET_ERROR(ENOSPC);
1486 }
1487 mutex_exit(&ds->ds_dir->dd_lock);
1488 dsl_dataset_rele(ds, FTAG);
1489 return (error);
1490 }
1491
1492 static void
dsl_dir_set_quota_sync(void * arg,dmu_tx_t * tx)1493 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1494 {
1495 dsl_dir_set_qr_arg_t *ddsqra = arg;
1496 dsl_pool_t *dp = dmu_tx_pool(tx);
1497 dsl_dataset_t *ds;
1498 uint64_t newval;
1499
1500 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1501
1502 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1503 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1504 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1505 &ddsqra->ddsqra_value, tx);
1506
1507 VERIFY0(dsl_prop_get_int_ds(ds,
1508 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1509 } else {
1510 newval = ddsqra->ddsqra_value;
1511 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1512 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1513 }
1514
1515 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1516 mutex_enter(&ds->ds_dir->dd_lock);
1517 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1518 mutex_exit(&ds->ds_dir->dd_lock);
1519 dsl_dataset_rele(ds, FTAG);
1520 }
1521
1522 int
dsl_dir_set_quota(const char * ddname,zprop_source_t source,uint64_t quota)1523 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1524 {
1525 dsl_dir_set_qr_arg_t ddsqra;
1526
1527 ddsqra.ddsqra_name = ddname;
1528 ddsqra.ddsqra_source = source;
1529 ddsqra.ddsqra_value = quota;
1530
1531 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1532 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1533 }
1534
1535 int
dsl_dir_set_reservation_check(void * arg,dmu_tx_t * tx)1536 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1537 {
1538 dsl_dir_set_qr_arg_t *ddsqra = arg;
1539 dsl_pool_t *dp = dmu_tx_pool(tx);
1540 dsl_dataset_t *ds;
1541 dsl_dir_t *dd;
1542 uint64_t newval, used, avail;
1543 int error;
1544
1545 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1546 if (error != 0)
1547 return (error);
1548 dd = ds->ds_dir;
1549
1550 /*
1551 * If we are doing the preliminary check in open context, the
1552 * space estimates may be inaccurate.
1553 */
1554 if (!dmu_tx_is_syncing(tx)) {
1555 dsl_dataset_rele(ds, FTAG);
1556 return (0);
1557 }
1558
1559 error = dsl_prop_predict(ds->ds_dir,
1560 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1561 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1562 if (error != 0) {
1563 dsl_dataset_rele(ds, FTAG);
1564 return (error);
1565 }
1566
1567 mutex_enter(&dd->dd_lock);
1568 used = dsl_dir_phys(dd)->dd_used_bytes;
1569 mutex_exit(&dd->dd_lock);
1570
1571 if (dd->dd_parent) {
1572 avail = dsl_dir_space_available(dd->dd_parent,
1573 NULL, 0, FALSE);
1574 } else {
1575 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1576 }
1577
1578 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1579 uint64_t delta = MAX(used, newval) -
1580 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1581
1582 if (delta > avail ||
1583 (dsl_dir_phys(dd)->dd_quota > 0 &&
1584 newval > dsl_dir_phys(dd)->dd_quota))
1585 error = SET_ERROR(ENOSPC);
1586 }
1587
1588 dsl_dataset_rele(ds, FTAG);
1589 return (error);
1590 }
1591
1592 void
dsl_dir_set_reservation_sync_impl(dsl_dir_t * dd,uint64_t value,dmu_tx_t * tx)1593 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1594 {
1595 uint64_t used;
1596 int64_t delta;
1597
1598 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1599
1600 mutex_enter(&dd->dd_lock);
1601 used = dsl_dir_phys(dd)->dd_used_bytes;
1602 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1603 dsl_dir_phys(dd)->dd_reserved = value;
1604
1605 if (dd->dd_parent != NULL) {
1606 /* Roll up this additional usage into our ancestors */
1607 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1608 delta, 0, 0, tx);
1609 }
1610 mutex_exit(&dd->dd_lock);
1611 }
1612
1613 static void
dsl_dir_set_reservation_sync(void * arg,dmu_tx_t * tx)1614 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1615 {
1616 dsl_dir_set_qr_arg_t *ddsqra = arg;
1617 dsl_pool_t *dp = dmu_tx_pool(tx);
1618 dsl_dataset_t *ds;
1619 uint64_t newval;
1620
1621 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1622
1623 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1624 dsl_prop_set_sync_impl(ds,
1625 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1626 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1627 &ddsqra->ddsqra_value, tx);
1628
1629 VERIFY0(dsl_prop_get_int_ds(ds,
1630 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1631 } else {
1632 newval = ddsqra->ddsqra_value;
1633 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1634 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1635 (longlong_t)newval);
1636 }
1637
1638 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1639 dsl_dataset_rele(ds, FTAG);
1640 }
1641
1642 int
dsl_dir_set_reservation(const char * ddname,zprop_source_t source,uint64_t reservation)1643 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1644 uint64_t reservation)
1645 {
1646 dsl_dir_set_qr_arg_t ddsqra;
1647
1648 ddsqra.ddsqra_name = ddname;
1649 ddsqra.ddsqra_source = source;
1650 ddsqra.ddsqra_value = reservation;
1651
1652 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1653 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1654 }
1655
1656 static dsl_dir_t *
closest_common_ancestor(dsl_dir_t * ds1,dsl_dir_t * ds2)1657 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1658 {
1659 for (; ds1; ds1 = ds1->dd_parent) {
1660 dsl_dir_t *dd;
1661 for (dd = ds2; dd; dd = dd->dd_parent) {
1662 if (ds1 == dd)
1663 return (dd);
1664 }
1665 }
1666 return (NULL);
1667 }
1668
1669 /*
1670 * If delta is applied to dd, how much of that delta would be applied to
1671 * ancestor? Syncing context only.
1672 */
1673 static int64_t
would_change(dsl_dir_t * dd,int64_t delta,dsl_dir_t * ancestor)1674 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1675 {
1676 if (dd == ancestor)
1677 return (delta);
1678
1679 mutex_enter(&dd->dd_lock);
1680 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1681 mutex_exit(&dd->dd_lock);
1682 return (would_change(dd->dd_parent, delta, ancestor));
1683 }
1684
1685 typedef struct dsl_dir_rename_arg {
1686 const char *ddra_oldname;
1687 const char *ddra_newname;
1688 cred_t *ddra_cred;
1689 } dsl_dir_rename_arg_t;
1690
1691 /* ARGSUSED */
1692 static int
dsl_valid_rename(dsl_pool_t * dp,dsl_dataset_t * ds,void * arg)1693 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1694 {
1695 int *deltap = arg;
1696 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1697
1698 dsl_dataset_name(ds, namebuf);
1699
1700 if (strlen(namebuf) + *deltap >= ZFS_MAX_DATASET_NAME_LEN)
1701 return (SET_ERROR(ENAMETOOLONG));
1702 return (0);
1703 }
1704
1705 static int
dsl_dir_rename_check(void * arg,dmu_tx_t * tx)1706 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1707 {
1708 dsl_dir_rename_arg_t *ddra = arg;
1709 dsl_pool_t *dp = dmu_tx_pool(tx);
1710 dsl_dir_t *dd, *newparent;
1711 const char *mynewname;
1712 int error;
1713 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1714
1715 /* target dir should exist */
1716 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1717 if (error != 0)
1718 return (error);
1719
1720 /* new parent should exist */
1721 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1722 &newparent, &mynewname);
1723 if (error != 0) {
1724 dsl_dir_rele(dd, FTAG);
1725 return (error);
1726 }
1727
1728 /* can't rename to different pool */
1729 if (dd->dd_pool != newparent->dd_pool) {
1730 dsl_dir_rele(newparent, FTAG);
1731 dsl_dir_rele(dd, FTAG);
1732 return (SET_ERROR(EXDEV));
1733 }
1734
1735 /* new name should not already exist */
1736 if (mynewname == NULL) {
1737 dsl_dir_rele(newparent, FTAG);
1738 dsl_dir_rele(dd, FTAG);
1739 return (SET_ERROR(EEXIST));
1740 }
1741
1742 /* if the name length is growing, validate child name lengths */
1743 if (delta > 0) {
1744 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1745 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1746 if (error != 0) {
1747 dsl_dir_rele(newparent, FTAG);
1748 dsl_dir_rele(dd, FTAG);
1749 return (error);
1750 }
1751 }
1752
1753 if (dmu_tx_is_syncing(tx)) {
1754 if (spa_feature_is_active(dp->dp_spa,
1755 SPA_FEATURE_FS_SS_LIMIT)) {
1756 /*
1757 * Although this is the check function and we don't
1758 * normally make on-disk changes in check functions,
1759 * we need to do that here.
1760 *
1761 * Ensure this portion of the tree's counts have been
1762 * initialized in case the new parent has limits set.
1763 */
1764 dsl_dir_init_fs_ss_count(dd, tx);
1765 }
1766 }
1767
1768 if (newparent != dd->dd_parent) {
1769 /* is there enough space? */
1770 uint64_t myspace =
1771 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1772 dsl_dir_phys(dd)->dd_reserved);
1773 objset_t *os = dd->dd_pool->dp_meta_objset;
1774 uint64_t fs_cnt = 0;
1775 uint64_t ss_cnt = 0;
1776
1777 if (dsl_dir_is_zapified(dd)) {
1778 int err;
1779
1780 err = zap_lookup(os, dd->dd_object,
1781 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1782 &fs_cnt);
1783 if (err != ENOENT && err != 0) {
1784 dsl_dir_rele(newparent, FTAG);
1785 dsl_dir_rele(dd, FTAG);
1786 return (err);
1787 }
1788
1789 /*
1790 * have to add 1 for the filesystem itself that we're
1791 * moving
1792 */
1793 fs_cnt++;
1794
1795 err = zap_lookup(os, dd->dd_object,
1796 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1797 &ss_cnt);
1798 if (err != ENOENT && err != 0) {
1799 dsl_dir_rele(newparent, FTAG);
1800 dsl_dir_rele(dd, FTAG);
1801 return (err);
1802 }
1803 }
1804
1805 /* no rename into our descendant */
1806 if (closest_common_ancestor(dd, newparent) == dd) {
1807 dsl_dir_rele(newparent, FTAG);
1808 dsl_dir_rele(dd, FTAG);
1809 return (SET_ERROR(EINVAL));
1810 }
1811
1812 error = dsl_dir_transfer_possible(dd->dd_parent,
1813 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1814 if (error != 0) {
1815 dsl_dir_rele(newparent, FTAG);
1816 dsl_dir_rele(dd, FTAG);
1817 return (error);
1818 }
1819 }
1820
1821 dsl_dir_rele(newparent, FTAG);
1822 dsl_dir_rele(dd, FTAG);
1823 return (0);
1824 }
1825
1826 static void
dsl_dir_rename_sync(void * arg,dmu_tx_t * tx)1827 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1828 {
1829 dsl_dir_rename_arg_t *ddra = arg;
1830 dsl_pool_t *dp = dmu_tx_pool(tx);
1831 dsl_dir_t *dd, *newparent;
1832 const char *mynewname;
1833 int error;
1834 objset_t *mos = dp->dp_meta_objset;
1835
1836 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1837 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1838 &mynewname));
1839
1840 /* Log this before we change the name. */
1841 spa_history_log_internal_dd(dd, "rename", tx,
1842 "-> %s", ddra->ddra_newname);
1843
1844 if (newparent != dd->dd_parent) {
1845 objset_t *os = dd->dd_pool->dp_meta_objset;
1846 uint64_t fs_cnt = 0;
1847 uint64_t ss_cnt = 0;
1848
1849 /*
1850 * We already made sure the dd counts were initialized in the
1851 * check function.
1852 */
1853 if (spa_feature_is_active(dp->dp_spa,
1854 SPA_FEATURE_FS_SS_LIMIT)) {
1855 VERIFY0(zap_lookup(os, dd->dd_object,
1856 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1857 &fs_cnt));
1858 /* add 1 for the filesystem itself that we're moving */
1859 fs_cnt++;
1860
1861 VERIFY0(zap_lookup(os, dd->dd_object,
1862 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1863 &ss_cnt));
1864 }
1865
1866 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1867 DD_FIELD_FILESYSTEM_COUNT, tx);
1868 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1869 DD_FIELD_FILESYSTEM_COUNT, tx);
1870
1871 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1872 DD_FIELD_SNAPSHOT_COUNT, tx);
1873 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1874 DD_FIELD_SNAPSHOT_COUNT, tx);
1875
1876 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1877 -dsl_dir_phys(dd)->dd_used_bytes,
1878 -dsl_dir_phys(dd)->dd_compressed_bytes,
1879 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1880 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1881 dsl_dir_phys(dd)->dd_used_bytes,
1882 dsl_dir_phys(dd)->dd_compressed_bytes,
1883 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1884
1885 if (dsl_dir_phys(dd)->dd_reserved >
1886 dsl_dir_phys(dd)->dd_used_bytes) {
1887 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
1888 dsl_dir_phys(dd)->dd_used_bytes;
1889
1890 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1891 -unused_rsrv, 0, 0, tx);
1892 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1893 unused_rsrv, 0, 0, tx);
1894 }
1895 }
1896
1897 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1898
1899 /* remove from old parent zapobj */
1900 error = zap_remove(mos,
1901 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
1902 dd->dd_myname, tx);
1903 ASSERT0(error);
1904
1905 (void) strcpy(dd->dd_myname, mynewname);
1906 dsl_dir_rele(dd->dd_parent, dd);
1907 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
1908 VERIFY0(dsl_dir_hold_obj(dp,
1909 newparent->dd_object, NULL, dd, &dd->dd_parent));
1910
1911 /* add to new parent zapobj */
1912 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
1913 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1914
1915 #if defined(__FreeBSD__) || defined(__NetBSD__)
1916 #ifdef _KERNEL
1917 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1918 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1919 #endif
1920 #endif
1921
1922 dsl_prop_notify_all(dd);
1923
1924 dsl_dir_rele(newparent, FTAG);
1925 dsl_dir_rele(dd, FTAG);
1926 }
1927
1928 int
dsl_dir_rename(const char * oldname,const char * newname)1929 dsl_dir_rename(const char *oldname, const char *newname)
1930 {
1931 dsl_dir_rename_arg_t ddra;
1932
1933 ddra.ddra_oldname = oldname;
1934 ddra.ddra_newname = newname;
1935 ddra.ddra_cred = CRED();
1936
1937 return (dsl_sync_task(oldname,
1938 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1939 3, ZFS_SPACE_CHECK_RESERVED));
1940 }
1941
1942 int
dsl_dir_transfer_possible(dsl_dir_t * sdd,dsl_dir_t * tdd,uint64_t fs_cnt,uint64_t ss_cnt,uint64_t space,cred_t * cr)1943 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1944 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1945 {
1946 dsl_dir_t *ancestor;
1947 int64_t adelta;
1948 uint64_t avail;
1949 int err;
1950
1951 ancestor = closest_common_ancestor(sdd, tdd);
1952 adelta = would_change(sdd, -space, ancestor);
1953 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1954 if (avail < space)
1955 return (SET_ERROR(ENOSPC));
1956
1957 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1958 ancestor, cr);
1959 if (err != 0)
1960 return (err);
1961 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1962 ancestor, cr);
1963 if (err != 0)
1964 return (err);
1965
1966 return (0);
1967 }
1968
1969 timestruc_t
dsl_dir_snap_cmtime(dsl_dir_t * dd)1970 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1971 {
1972 timestruc_t t;
1973
1974 mutex_enter(&dd->dd_lock);
1975 t = dd->dd_snap_cmtime;
1976 mutex_exit(&dd->dd_lock);
1977
1978 return (t);
1979 }
1980
1981 void
dsl_dir_snap_cmtime_update(dsl_dir_t * dd)1982 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1983 {
1984 timestruc_t t;
1985
1986 gethrestime(&t);
1987 mutex_enter(&dd->dd_lock);
1988 dd->dd_snap_cmtime = t;
1989 mutex_exit(&dd->dd_lock);
1990 }
1991
1992 void
dsl_dir_zapify(dsl_dir_t * dd,dmu_tx_t * tx)1993 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1994 {
1995 objset_t *mos = dd->dd_pool->dp_meta_objset;
1996 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1997 }
1998
1999 boolean_t
dsl_dir_is_zapified(dsl_dir_t * dd)2000 dsl_dir_is_zapified(dsl_dir_t *dd)
2001 {
2002 dmu_object_info_t doi;
2003
2004 dmu_object_info_from_db(dd->dd_dbuf, &doi);
2005 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
2006 }
2007