]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
37f03da8 | 23 | * Copyright (c) 2012, 2018 by Delphix. All rights reserved. |
b1118acb | 24 | * Copyright (c) 2013 Martin Matuska. All rights reserved. |
788eb90c | 25 | * Copyright (c) 2014 Joyent, Inc. All rights reserved. |
0c66c32d | 26 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
a0bd735a | 27 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
d8d418ff | 28 | * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved. |
34dc7c2f BB |
29 | */ |
30 | ||
34dc7c2f BB |
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> | |
fa86b5db | 39 | #include <sys/dmu_impl.h> |
34dc7c2f | 40 | #include <sys/spa.h> |
ae76f45c | 41 | #include <sys/spa_impl.h> |
428870ff | 42 | #include <sys/metaslab.h> |
34dc7c2f BB |
43 | #include <sys/zap.h> |
44 | #include <sys/zio.h> | |
45 | #include <sys/arc.h> | |
46 | #include <sys/sunddi.h> | |
788eb90c JJ |
47 | #include <sys/zfeature.h> |
48 | #include <sys/policy.h> | |
49 | #include <sys/zfs_znode.h> | |
ba6a2402 | 50 | #include <sys/zvol.h> |
37f03da8 | 51 | #include <sys/zthr.h> |
34dc7c2f | 52 | #include "zfs_namecheck.h" |
788eb90c JJ |
53 | #include "zfs_prop.h" |
54 | ||
55 | /* | |
56 | * Filesystem and Snapshot Limits | |
57 | * ------------------------------ | |
58 | * | |
59 | * These limits are used to restrict the number of filesystems and/or snapshots | |
60 | * that can be created at a given level in the tree or below. A typical | |
61 | * use-case is with a delegated dataset where the administrator wants to ensure | |
62 | * that a user within the zone is not creating too many additional filesystems | |
63 | * or snapshots, even though they're not exceeding their space quota. | |
64 | * | |
65 | * The filesystem and snapshot counts are stored as extensible properties. This | |
66 | * capability is controlled by a feature flag and must be enabled to be used. | |
67 | * Once enabled, the feature is not active until the first limit is set. At | |
68 | * that point, future operations to create/destroy filesystems or snapshots | |
69 | * will validate and update the counts. | |
70 | * | |
71 | * Because the count properties will not exist before the feature is active, | |
72 | * the counts are updated when a limit is first set on an uninitialized | |
73 | * dsl_dir node in the tree (The filesystem/snapshot count on a node includes | |
74 | * all of the nested filesystems/snapshots. Thus, a new leaf node has a | |
75 | * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and | |
76 | * snapshot count properties on a node indicate uninitialized counts on that | |
77 | * node.) When first setting a limit on an uninitialized node, the code starts | |
78 | * at the filesystem with the new limit and descends into all sub-filesystems | |
79 | * to add the count properties. | |
80 | * | |
81 | * In practice this is lightweight since a limit is typically set when the | |
82 | * filesystem is created and thus has no children. Once valid, changing the | |
83 | * limit value won't require a re-traversal since the counts are already valid. | |
84 | * When recursively fixing the counts, if a node with a limit is encountered | |
85 | * during the descent, the counts are known to be valid and there is no need to | |
86 | * descend into that filesystem's children. The counts on filesystems above the | |
87 | * one with the new limit will still be uninitialized, unless a limit is | |
88 | * eventually set on one of those filesystems. The counts are always recursively | |
89 | * updated when a limit is set on a dataset, unless there is already a limit. | |
90 | * When a new limit value is set on a filesystem with an existing limit, it is | |
91 | * possible for the new limit to be less than the current count at that level | |
92 | * since a user who can change the limit is also allowed to exceed the limit. | |
93 | * | |
94 | * Once the feature is active, then whenever a filesystem or snapshot is | |
95 | * created, the code recurses up the tree, validating the new count against the | |
96 | * limit at each initialized level. In practice, most levels will not have a | |
97 | * limit set. If there is a limit at any initialized level up the tree, the | |
98 | * check must pass or the creation will fail. Likewise, when a filesystem or | |
99 | * snapshot is destroyed, the counts are recursively adjusted all the way up | |
100 | * the initizized nodes in the tree. Renaming a filesystem into different point | |
101 | * in the tree will first validate, then update the counts on each branch up to | |
102 | * the common ancestor. A receive will also validate the counts and then update | |
103 | * them. | |
104 | * | |
105 | * An exception to the above behavior is that the limit is not enforced if the | |
106 | * user has permission to modify the limit. This is primarily so that | |
107 | * recursive snapshots in the global zone always work. We want to prevent a | |
108 | * denial-of-service in which a lower level delegated dataset could max out its | |
109 | * limit and thus block recursive snapshots from being taken in the global zone. | |
110 | * Because of this, it is possible for the snapshot count to be over the limit | |
111 | * and snapshots taken in the global zone could cause a lower level dataset to | |
112 | * hit or exceed its limit. The administrator taking the global zone recursive | |
113 | * snapshot should be aware of this side-effect and behave accordingly. | |
114 | * For consistency, the filesystem limit is also not enforced if the user can | |
115 | * modify the limit. | |
116 | * | |
117 | * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check() | |
118 | * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in | |
119 | * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by | |
120 | * dsl_dir_init_fs_ss_count(). | |
121 | * | |
122 | * There is a special case when we receive a filesystem that already exists. In | |
123 | * this case a temporary clone name of %X is created (see dmu_recv_begin). We | |
124 | * never update the filesystem counts for temporary clones. | |
125 | * | |
126 | * Likewise, we do not update the snapshot counts for temporary snapshots, | |
127 | * such as those created by zfs diff. | |
128 | */ | |
34dc7c2f | 129 | |
d683ddbb JG |
130 | extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd); |
131 | ||
34dc7c2f | 132 | static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); |
34dc7c2f | 133 | |
a1d477c2 MA |
134 | typedef struct ddulrt_arg { |
135 | dsl_dir_t *ddulrta_dd; | |
136 | uint64_t ddlrta_txg; | |
137 | } ddulrt_arg_t; | |
138 | ||
34dc7c2f | 139 | static void |
39efbde7 | 140 | dsl_dir_evict_async(void *dbu) |
34dc7c2f | 141 | { |
0c66c32d | 142 | dsl_dir_t *dd = dbu; |
34dc7c2f | 143 | int t; |
d1d7e268 | 144 | ASSERTV(dsl_pool_t *dp = dd->dd_pool); |
34dc7c2f | 145 | |
0c66c32d JG |
146 | dd->dd_dbuf = NULL; |
147 | ||
34dc7c2f BB |
148 | for (t = 0; t < TXG_SIZE; t++) { |
149 | ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); | |
150 | ASSERT(dd->dd_tempreserved[t] == 0); | |
151 | ASSERT(dd->dd_space_towrite[t] == 0); | |
152 | } | |
153 | ||
34dc7c2f | 154 | if (dd->dd_parent) |
0c66c32d | 155 | dsl_dir_async_rele(dd->dd_parent, dd); |
34dc7c2f | 156 | |
0c66c32d | 157 | spa_async_close(dd->dd_pool->dp_spa, dd); |
34dc7c2f | 158 | |
37f03da8 SH |
159 | if (dsl_deadlist_is_open(&dd->dd_livelist)) |
160 | dsl_dir_livelist_close(dd); | |
161 | ||
0eb21616 | 162 | dsl_prop_fini(dd); |
34dc7c2f BB |
163 | mutex_destroy(&dd->dd_lock); |
164 | kmem_free(dd, sizeof (dsl_dir_t)); | |
165 | } | |
166 | ||
167 | int | |
13fe0198 | 168 | dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, |
34dc7c2f BB |
169 | const char *tail, void *tag, dsl_dir_t **ddp) |
170 | { | |
171 | dmu_buf_t *dbuf; | |
172 | dsl_dir_t *dd; | |
b5256303 | 173 | dmu_object_info_t doi; |
34dc7c2f BB |
174 | int err; |
175 | ||
13fe0198 | 176 | ASSERT(dsl_pool_config_held(dp)); |
34dc7c2f BB |
177 | |
178 | err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); | |
13fe0198 | 179 | if (err != 0) |
34dc7c2f BB |
180 | return (err); |
181 | dd = dmu_buf_get_user(dbuf); | |
b5256303 TC |
182 | |
183 | dmu_object_info_from_db(dbuf, &doi); | |
184 | ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); | |
185 | ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); | |
186 | ||
34dc7c2f BB |
187 | if (dd == NULL) { |
188 | dsl_dir_t *winner; | |
34dc7c2f | 189 | |
79c76d5b | 190 | dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); |
34dc7c2f BB |
191 | dd->dd_object = ddobj; |
192 | dd->dd_dbuf = dbuf; | |
193 | dd->dd_pool = dp; | |
b5256303 TC |
194 | |
195 | if (dsl_dir_is_zapified(dd) && | |
196 | zap_contains(dp->dp_meta_objset, ddobj, | |
197 | DD_FIELD_CRYPTO_KEY_OBJ) == 0) { | |
198 | VERIFY0(zap_lookup(dp->dp_meta_objset, | |
199 | ddobj, DD_FIELD_CRYPTO_KEY_OBJ, | |
200 | sizeof (uint64_t), 1, &dd->dd_crypto_obj)); | |
ae76f45c TC |
201 | |
202 | /* check for on-disk format errata */ | |
203 | if (dsl_dir_incompatible_encryption_version(dd)) { | |
204 | dp->dp_spa->spa_errata = | |
205 | ZPOOL_ERRATA_ZOL_6845_ENCRYPTION; | |
206 | } | |
b5256303 TC |
207 | } |
208 | ||
34dc7c2f | 209 | mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); |
0eb21616 | 210 | dsl_prop_init(dd); |
34dc7c2f | 211 | |
428870ff BB |
212 | dsl_dir_snap_cmtime_update(dd); |
213 | ||
d683ddbb JG |
214 | if (dsl_dir_phys(dd)->dd_parent_obj) { |
215 | err = dsl_dir_hold_obj(dp, | |
216 | dsl_dir_phys(dd)->dd_parent_obj, NULL, dd, | |
217 | &dd->dd_parent); | |
13fe0198 | 218 | if (err != 0) |
b128c09f | 219 | goto errout; |
34dc7c2f BB |
220 | if (tail) { |
221 | #ifdef ZFS_DEBUG | |
222 | uint64_t foundobj; | |
223 | ||
224 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb JG |
225 | dsl_dir_phys(dd->dd_parent)-> |
226 | dd_child_dir_zapobj, tail, | |
227 | sizeof (foundobj), 1, &foundobj); | |
34dc7c2f BB |
228 | ASSERT(err || foundobj == ddobj); |
229 | #endif | |
680eada9 | 230 | (void) strlcpy(dd->dd_myname, tail, |
231 | sizeof (dd->dd_myname)); | |
34dc7c2f BB |
232 | } else { |
233 | err = zap_value_search(dp->dp_meta_objset, | |
d683ddbb JG |
234 | dsl_dir_phys(dd->dd_parent)-> |
235 | dd_child_dir_zapobj, | |
34dc7c2f BB |
236 | ddobj, 0, dd->dd_myname); |
237 | } | |
13fe0198 | 238 | if (err != 0) |
b128c09f | 239 | goto errout; |
34dc7c2f BB |
240 | } else { |
241 | (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); | |
242 | } | |
243 | ||
428870ff BB |
244 | if (dsl_dir_is_clone(dd)) { |
245 | dmu_buf_t *origin_bonus; | |
246 | dsl_dataset_phys_t *origin_phys; | |
247 | ||
248 | /* | |
249 | * We can't open the origin dataset, because | |
250 | * that would require opening this dsl_dir. | |
251 | * Just look at its phys directly instead. | |
252 | */ | |
253 | err = dmu_bonus_hold(dp->dp_meta_objset, | |
d683ddbb JG |
254 | dsl_dir_phys(dd)->dd_origin_obj, FTAG, |
255 | &origin_bonus); | |
13fe0198 | 256 | if (err != 0) |
428870ff BB |
257 | goto errout; |
258 | origin_phys = origin_bonus->db_data; | |
259 | dd->dd_origin_txg = | |
260 | origin_phys->ds_creation_txg; | |
261 | dmu_buf_rele(origin_bonus, FTAG); | |
37f03da8 SH |
262 | if (dsl_dir_is_zapified(dd)) { |
263 | uint64_t obj; | |
264 | err = zap_lookup(dp->dp_meta_objset, | |
265 | dd->dd_object, DD_FIELD_LIVELIST, | |
266 | sizeof (uint64_t), 1, &obj); | |
267 | if (err == 0) | |
268 | dsl_dir_livelist_open(dd, obj); | |
269 | else if (err != ENOENT) | |
270 | goto errout; | |
271 | } | |
428870ff BB |
272 | } |
273 | ||
39efbde7 GM |
274 | dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async, |
275 | &dd->dd_dbuf); | |
0c66c32d JG |
276 | winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu); |
277 | if (winner != NULL) { | |
34dc7c2f | 278 | if (dd->dd_parent) |
13fe0198 | 279 | dsl_dir_rele(dd->dd_parent, dd); |
37f03da8 SH |
280 | if (dsl_deadlist_is_open(&dd->dd_livelist)) |
281 | dsl_dir_livelist_close(dd); | |
0eb21616 | 282 | dsl_prop_fini(dd); |
34dc7c2f BB |
283 | mutex_destroy(&dd->dd_lock); |
284 | kmem_free(dd, sizeof (dsl_dir_t)); | |
285 | dd = winner; | |
286 | } else { | |
287 | spa_open_ref(dp->dp_spa, dd); | |
288 | } | |
289 | } | |
290 | ||
291 | /* | |
292 | * The dsl_dir_t has both open-to-close and instantiate-to-evict | |
293 | * holds on the spa. We need the open-to-close holds because | |
294 | * otherwise the spa_refcnt wouldn't change when we open a | |
295 | * dir which the spa also has open, so we could incorrectly | |
296 | * think it was OK to unload/export/destroy the pool. We need | |
297 | * the instantiate-to-evict hold because the dsl_dir_t has a | |
298 | * pointer to the dd_pool, which has a pointer to the spa_t. | |
299 | */ | |
300 | spa_open_ref(dp->dp_spa, tag); | |
301 | ASSERT3P(dd->dd_pool, ==, dp); | |
302 | ASSERT3U(dd->dd_object, ==, ddobj); | |
303 | ASSERT3P(dd->dd_dbuf, ==, dbuf); | |
304 | *ddp = dd; | |
305 | return (0); | |
b128c09f BB |
306 | |
307 | errout: | |
308 | if (dd->dd_parent) | |
13fe0198 | 309 | dsl_dir_rele(dd->dd_parent, dd); |
37f03da8 SH |
310 | if (dsl_deadlist_is_open(&dd->dd_livelist)) |
311 | dsl_dir_livelist_close(dd); | |
0eb21616 | 312 | dsl_prop_fini(dd); |
b128c09f BB |
313 | mutex_destroy(&dd->dd_lock); |
314 | kmem_free(dd, sizeof (dsl_dir_t)); | |
315 | dmu_buf_rele(dbuf, tag); | |
316 | return (err); | |
34dc7c2f BB |
317 | } |
318 | ||
319 | void | |
13fe0198 | 320 | dsl_dir_rele(dsl_dir_t *dd, void *tag) |
34dc7c2f BB |
321 | { |
322 | dprintf_dd(dd, "%s\n", ""); | |
323 | spa_close(dd->dd_pool->dp_spa, tag); | |
324 | dmu_buf_rele(dd->dd_dbuf, tag); | |
325 | } | |
326 | ||
0c66c32d JG |
327 | /* |
328 | * Remove a reference to the given dsl dir that is being asynchronously | |
329 | * released. Async releases occur from a taskq performing eviction of | |
330 | * dsl datasets and dirs. This process is identical to a normal release | |
331 | * with the exception of using the async API for releasing the reference on | |
332 | * the spa. | |
333 | */ | |
334 | void | |
335 | dsl_dir_async_rele(dsl_dir_t *dd, void *tag) | |
336 | { | |
337 | dprintf_dd(dd, "%s\n", ""); | |
338 | spa_async_close(dd->dd_pool->dp_spa, tag); | |
339 | dmu_buf_rele(dd->dd_dbuf, tag); | |
340 | } | |
341 | ||
eca7b760 | 342 | /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */ |
34dc7c2f BB |
343 | void |
344 | dsl_dir_name(dsl_dir_t *dd, char *buf) | |
345 | { | |
346 | if (dd->dd_parent) { | |
347 | dsl_dir_name(dd->dd_parent, buf); | |
eca7b760 IK |
348 | VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <, |
349 | ZFS_MAX_DATASET_NAME_LEN); | |
34dc7c2f BB |
350 | } else { |
351 | buf[0] = '\0'; | |
352 | } | |
353 | if (!MUTEX_HELD(&dd->dd_lock)) { | |
354 | /* | |
355 | * recursive mutex so that we can use | |
356 | * dprintf_dd() with dd_lock held | |
357 | */ | |
358 | mutex_enter(&dd->dd_lock); | |
eca7b760 IK |
359 | VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), |
360 | <, ZFS_MAX_DATASET_NAME_LEN); | |
34dc7c2f BB |
361 | mutex_exit(&dd->dd_lock); |
362 | } else { | |
eca7b760 IK |
363 | VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), |
364 | <, ZFS_MAX_DATASET_NAME_LEN); | |
34dc7c2f BB |
365 | } |
366 | } | |
367 | ||
29809a6c | 368 | /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ |
34dc7c2f BB |
369 | int |
370 | dsl_dir_namelen(dsl_dir_t *dd) | |
371 | { | |
372 | int result = 0; | |
373 | ||
374 | if (dd->dd_parent) { | |
375 | /* parent's name + 1 for the "/" */ | |
376 | result = dsl_dir_namelen(dd->dd_parent) + 1; | |
377 | } | |
378 | ||
379 | if (!MUTEX_HELD(&dd->dd_lock)) { | |
380 | /* see dsl_dir_name */ | |
381 | mutex_enter(&dd->dd_lock); | |
382 | result += strlen(dd->dd_myname); | |
383 | mutex_exit(&dd->dd_lock); | |
384 | } else { | |
385 | result += strlen(dd->dd_myname); | |
386 | } | |
387 | ||
388 | return (result); | |
389 | } | |
390 | ||
34dc7c2f BB |
391 | static int |
392 | getcomponent(const char *path, char *component, const char **nextp) | |
393 | { | |
394 | char *p; | |
13fe0198 | 395 | |
9babb374 | 396 | if ((path == NULL) || (path[0] == '\0')) |
2e528b49 | 397 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
398 | /* This would be a good place to reserve some namespace... */ |
399 | p = strpbrk(path, "/@"); | |
400 | if (p && (p[1] == '/' || p[1] == '@')) { | |
401 | /* two separators in a row */ | |
2e528b49 | 402 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
403 | } |
404 | if (p == NULL || p == path) { | |
405 | /* | |
406 | * if the first thing is an @ or /, it had better be an | |
407 | * @ and it had better not have any more ats or slashes, | |
408 | * and it had better have something after the @. | |
409 | */ | |
410 | if (p != NULL && | |
411 | (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) | |
2e528b49 | 412 | return (SET_ERROR(EINVAL)); |
eca7b760 | 413 | if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 414 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
415 | (void) strcpy(component, path); |
416 | p = NULL; | |
417 | } else if (p[0] == '/') { | |
eca7b760 | 418 | if (p - path >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 419 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f | 420 | (void) strncpy(component, path, p - path); |
13fe0198 | 421 | component[p - path] = '\0'; |
34dc7c2f BB |
422 | p++; |
423 | } else if (p[0] == '@') { | |
424 | /* | |
425 | * if the next separator is an @, there better not be | |
426 | * any more slashes. | |
427 | */ | |
428 | if (strchr(path, '/')) | |
2e528b49 | 429 | return (SET_ERROR(EINVAL)); |
eca7b760 | 430 | if (p - path >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 431 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f | 432 | (void) strncpy(component, path, p - path); |
13fe0198 | 433 | component[p - path] = '\0'; |
34dc7c2f | 434 | } else { |
13fe0198 | 435 | panic("invalid p=%p", (void *)p); |
34dc7c2f BB |
436 | } |
437 | *nextp = p; | |
438 | return (0); | |
439 | } | |
440 | ||
441 | /* | |
13fe0198 MA |
442 | * Return the dsl_dir_t, and possibly the last component which couldn't |
443 | * be found in *tail. The name must be in the specified dsl_pool_t. This | |
444 | * thread must hold the dp_config_rwlock for the pool. Returns NULL if the | |
445 | * path is bogus, or if tail==NULL and we couldn't parse the whole name. | |
446 | * (*tail)[0] == '@' means that the last component is a snapshot. | |
34dc7c2f BB |
447 | */ |
448 | int | |
13fe0198 | 449 | dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, |
34dc7c2f BB |
450 | dsl_dir_t **ddp, const char **tailp) |
451 | { | |
fcf37ec6 | 452 | char *buf; |
13fe0198 | 453 | const char *spaname, *next, *nextnext = NULL; |
34dc7c2f BB |
454 | int err; |
455 | dsl_dir_t *dd; | |
34dc7c2f | 456 | uint64_t ddobj; |
34dc7c2f | 457 | |
eca7b760 | 458 | buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
34dc7c2f | 459 | err = getcomponent(name, buf, &next); |
13fe0198 | 460 | if (err != 0) |
fcf37ec6 | 461 | goto error; |
34dc7c2f | 462 | |
13fe0198 MA |
463 | /* Make sure the name is in the specified pool. */ |
464 | spaname = spa_name(dp->dp_spa); | |
465 | if (strcmp(buf, spaname) != 0) { | |
9063f654 | 466 | err = SET_ERROR(EXDEV); |
13fe0198 | 467 | goto error; |
34dc7c2f BB |
468 | } |
469 | ||
13fe0198 | 470 | ASSERT(dsl_pool_config_held(dp)); |
34dc7c2f | 471 | |
13fe0198 MA |
472 | err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); |
473 | if (err != 0) { | |
fcf37ec6 | 474 | goto error; |
34dc7c2f BB |
475 | } |
476 | ||
477 | while (next != NULL) { | |
0c66c32d | 478 | dsl_dir_t *child_dd; |
34dc7c2f | 479 | err = getcomponent(next, buf, &nextnext); |
13fe0198 | 480 | if (err != 0) |
34dc7c2f BB |
481 | break; |
482 | ASSERT(next[0] != '\0'); | |
483 | if (next[0] == '@') | |
484 | break; | |
485 | dprintf("looking up %s in obj%lld\n", | |
d683ddbb | 486 | buf, dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
487 | |
488 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb | 489 | dsl_dir_phys(dd)->dd_child_dir_zapobj, |
34dc7c2f | 490 | buf, sizeof (ddobj), 1, &ddobj); |
13fe0198 | 491 | if (err != 0) { |
34dc7c2f BB |
492 | if (err == ENOENT) |
493 | err = 0; | |
494 | break; | |
495 | } | |
496 | ||
0c66c32d | 497 | err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd); |
13fe0198 | 498 | if (err != 0) |
34dc7c2f | 499 | break; |
13fe0198 | 500 | dsl_dir_rele(dd, tag); |
0c66c32d | 501 | dd = child_dd; |
34dc7c2f BB |
502 | next = nextnext; |
503 | } | |
34dc7c2f | 504 | |
13fe0198 MA |
505 | if (err != 0) { |
506 | dsl_dir_rele(dd, tag); | |
fcf37ec6 | 507 | goto error; |
34dc7c2f BB |
508 | } |
509 | ||
510 | /* | |
511 | * It's an error if there's more than one component left, or | |
512 | * tailp==NULL and there's any component left. | |
513 | */ | |
514 | if (next != NULL && | |
515 | (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { | |
516 | /* bad path name */ | |
13fe0198 | 517 | dsl_dir_rele(dd, tag); |
34dc7c2f | 518 | dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); |
2e528b49 | 519 | err = SET_ERROR(ENOENT); |
34dc7c2f | 520 | } |
13fe0198 | 521 | if (tailp != NULL) |
34dc7c2f | 522 | *tailp = next; |
fc1ecd16 DB |
523 | if (err == 0) |
524 | *ddp = dd; | |
fcf37ec6 | 525 | error: |
eca7b760 | 526 | kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN); |
34dc7c2f BB |
527 | return (err); |
528 | } | |
529 | ||
788eb90c JJ |
530 | /* |
531 | * If the counts are already initialized for this filesystem and its | |
532 | * descendants then do nothing, otherwise initialize the counts. | |
533 | * | |
534 | * The counts on this filesystem, and those below, may be uninitialized due to | |
535 | * either the use of a pre-existing pool which did not support the | |
536 | * filesystem/snapshot limit feature, or one in which the feature had not yet | |
537 | * been enabled. | |
538 | * | |
539 | * Recursively descend the filesystem tree and update the filesystem/snapshot | |
540 | * counts on each filesystem below, then update the cumulative count on the | |
541 | * current filesystem. If the filesystem already has a count set on it, | |
542 | * then we know that its counts, and the counts on the filesystems below it, | |
543 | * are already correct, so we don't have to update this filesystem. | |
544 | */ | |
545 | static void | |
546 | dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx) | |
547 | { | |
548 | uint64_t my_fs_cnt = 0; | |
549 | uint64_t my_ss_cnt = 0; | |
550 | dsl_pool_t *dp = dd->dd_pool; | |
551 | objset_t *os = dp->dp_meta_objset; | |
552 | zap_cursor_t *zc; | |
553 | zap_attribute_t *za; | |
554 | dsl_dataset_t *ds; | |
555 | ||
a0c9a17a | 556 | ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)); |
788eb90c JJ |
557 | ASSERT(dsl_pool_config_held(dp)); |
558 | ASSERT(dmu_tx_is_syncing(tx)); | |
559 | ||
560 | dsl_dir_zapify(dd, tx); | |
561 | ||
562 | /* | |
563 | * If the filesystem count has already been initialized then we | |
564 | * don't need to recurse down any further. | |
565 | */ | |
566 | if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0) | |
567 | return; | |
568 | ||
569 | zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); | |
570 | za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); | |
571 | ||
572 | /* Iterate my child dirs */ | |
d683ddbb | 573 | for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj); |
788eb90c JJ |
574 | zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) { |
575 | dsl_dir_t *chld_dd; | |
576 | uint64_t count; | |
577 | ||
578 | VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG, | |
579 | &chld_dd)); | |
580 | ||
581 | /* | |
582 | * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and | |
583 | * temporary datasets. | |
584 | */ | |
585 | if (chld_dd->dd_myname[0] == '$' || | |
586 | chld_dd->dd_myname[0] == '%') { | |
587 | dsl_dir_rele(chld_dd, FTAG); | |
588 | continue; | |
589 | } | |
590 | ||
591 | my_fs_cnt++; /* count this child */ | |
592 | ||
593 | dsl_dir_init_fs_ss_count(chld_dd, tx); | |
594 | ||
595 | VERIFY0(zap_lookup(os, chld_dd->dd_object, | |
596 | DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count)); | |
597 | my_fs_cnt += count; | |
598 | VERIFY0(zap_lookup(os, chld_dd->dd_object, | |
599 | DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count)); | |
600 | my_ss_cnt += count; | |
601 | ||
602 | dsl_dir_rele(chld_dd, FTAG); | |
603 | } | |
604 | zap_cursor_fini(zc); | |
605 | /* Count my snapshots (we counted children's snapshots above) */ | |
606 | VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, | |
d683ddbb | 607 | dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds)); |
788eb90c | 608 | |
d683ddbb | 609 | for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj); |
788eb90c JJ |
610 | zap_cursor_retrieve(zc, za) == 0; |
611 | zap_cursor_advance(zc)) { | |
612 | /* Don't count temporary snapshots */ | |
613 | if (za->za_name[0] != '%') | |
614 | my_ss_cnt++; | |
615 | } | |
ca227e54 | 616 | zap_cursor_fini(zc); |
788eb90c JJ |
617 | |
618 | dsl_dataset_rele(ds, FTAG); | |
619 | ||
620 | kmem_free(zc, sizeof (zap_cursor_t)); | |
621 | kmem_free(za, sizeof (zap_attribute_t)); | |
622 | ||
623 | /* we're in a sync task, update counts */ | |
624 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
625 | VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, | |
626 | sizeof (my_fs_cnt), 1, &my_fs_cnt, tx)); | |
627 | VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, | |
628 | sizeof (my_ss_cnt), 1, &my_ss_cnt, tx)); | |
629 | } | |
630 | ||
631 | static int | |
632 | dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx) | |
633 | { | |
634 | char *ddname = (char *)arg; | |
635 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
636 | dsl_dataset_t *ds; | |
637 | dsl_dir_t *dd; | |
638 | int error; | |
639 | ||
640 | error = dsl_dataset_hold(dp, ddname, FTAG, &ds); | |
641 | if (error != 0) | |
642 | return (error); | |
643 | ||
644 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { | |
645 | dsl_dataset_rele(ds, FTAG); | |
646 | return (SET_ERROR(ENOTSUP)); | |
647 | } | |
648 | ||
649 | dd = ds->ds_dir; | |
650 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) && | |
651 | dsl_dir_is_zapified(dd) && | |
652 | zap_contains(dp->dp_meta_objset, dd->dd_object, | |
653 | DD_FIELD_FILESYSTEM_COUNT) == 0) { | |
654 | dsl_dataset_rele(ds, FTAG); | |
655 | return (SET_ERROR(EALREADY)); | |
656 | } | |
657 | ||
658 | dsl_dataset_rele(ds, FTAG); | |
659 | return (0); | |
660 | } | |
661 | ||
662 | static void | |
663 | dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx) | |
664 | { | |
665 | char *ddname = (char *)arg; | |
666 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
667 | dsl_dataset_t *ds; | |
668 | spa_t *spa; | |
669 | ||
670 | VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds)); | |
671 | ||
672 | spa = dsl_dataset_get_spa(ds); | |
673 | ||
674 | if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) { | |
675 | /* | |
676 | * Since the feature was not active and we're now setting a | |
677 | * limit, increment the feature-active counter so that the | |
678 | * feature becomes active for the first time. | |
679 | * | |
680 | * We are already in a sync task so we can update the MOS. | |
681 | */ | |
682 | spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx); | |
683 | } | |
684 | ||
685 | /* | |
686 | * Since we are now setting a non-UINT64_MAX limit on the filesystem, | |
687 | * we need to ensure the counts are correct. Descend down the tree from | |
688 | * this point and update all of the counts to be accurate. | |
689 | */ | |
690 | dsl_dir_init_fs_ss_count(ds->ds_dir, tx); | |
691 | ||
692 | dsl_dataset_rele(ds, FTAG); | |
693 | } | |
694 | ||
695 | /* | |
696 | * Make sure the feature is enabled and activate it if necessary. | |
697 | * Since we're setting a limit, ensure the on-disk counts are valid. | |
698 | * This is only called by the ioctl path when setting a limit value. | |
699 | * | |
700 | * We do not need to validate the new limit, since users who can change the | |
701 | * limit are also allowed to exceed the limit. | |
702 | */ | |
703 | int | |
704 | dsl_dir_activate_fs_ss_limit(const char *ddname) | |
705 | { | |
706 | int error; | |
707 | ||
708 | error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check, | |
3d45fdd6 MA |
709 | dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0, |
710 | ZFS_SPACE_CHECK_RESERVED); | |
788eb90c JJ |
711 | |
712 | if (error == EALREADY) | |
713 | error = 0; | |
714 | ||
715 | return (error); | |
716 | } | |
717 | ||
718 | /* | |
719 | * Used to determine if the filesystem_limit or snapshot_limit should be | |
720 | * enforced. We allow the limit to be exceeded if the user has permission to | |
721 | * write the property value. We pass in the creds that we got in the open | |
722 | * context since we will always be the GZ root in syncing context. We also have | |
723 | * to handle the case where we are allowed to change the limit on the current | |
724 | * dataset, but there may be another limit in the tree above. | |
725 | * | |
726 | * We can never modify these two properties within a non-global zone. In | |
727 | * addition, the other checks are modeled on zfs_secpolicy_write_perms. We | |
728 | * can't use that function since we are already holding the dp_config_rwlock. | |
729 | * In addition, we already have the dd and dealing with snapshots is simplified | |
730 | * in this code. | |
731 | */ | |
732 | ||
733 | typedef enum { | |
734 | ENFORCE_ALWAYS, | |
735 | ENFORCE_NEVER, | |
736 | ENFORCE_ABOVE | |
737 | } enforce_res_t; | |
738 | ||
739 | static enforce_res_t | |
740 | dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr) | |
741 | { | |
742 | enforce_res_t enforce = ENFORCE_ALWAYS; | |
743 | uint64_t obj; | |
744 | dsl_dataset_t *ds; | |
745 | uint64_t zoned; | |
746 | ||
747 | ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || | |
748 | prop == ZFS_PROP_SNAPSHOT_LIMIT); | |
749 | ||
750 | #ifdef _KERNEL | |
751 | if (crgetzoneid(cr) != GLOBAL_ZONEID) | |
752 | return (ENFORCE_ALWAYS); | |
753 | ||
754 | if (secpolicy_zfs(cr) == 0) | |
755 | return (ENFORCE_NEVER); | |
756 | #endif | |
757 | ||
d683ddbb | 758 | if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0) |
788eb90c JJ |
759 | return (ENFORCE_ALWAYS); |
760 | ||
761 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
762 | ||
763 | if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0) | |
764 | return (ENFORCE_ALWAYS); | |
765 | ||
766 | if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) { | |
767 | /* Only root can access zoned fs's from the GZ */ | |
768 | enforce = ENFORCE_ALWAYS; | |
769 | } else { | |
770 | if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0) | |
771 | enforce = ENFORCE_ABOVE; | |
772 | } | |
773 | ||
774 | dsl_dataset_rele(ds, FTAG); | |
775 | return (enforce); | |
776 | } | |
777 | ||
778 | /* | |
779 | * Check if adding additional child filesystem(s) would exceed any filesystem | |
780 | * limits or adding additional snapshot(s) would exceed any snapshot limits. | |
781 | * The prop argument indicates which limit to check. | |
782 | * | |
783 | * Note that all filesystem limits up to the root (or the highest | |
784 | * initialized) filesystem or the given ancestor must be satisfied. | |
785 | */ | |
786 | int | |
787 | dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop, | |
788 | dsl_dir_t *ancestor, cred_t *cr) | |
789 | { | |
790 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
791 | uint64_t limit, count; | |
792 | char *count_prop; | |
793 | enforce_res_t enforce; | |
794 | int err = 0; | |
795 | ||
796 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
797 | ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || | |
798 | prop == ZFS_PROP_SNAPSHOT_LIMIT); | |
799 | ||
800 | /* | |
801 | * If we're allowed to change the limit, don't enforce the limit | |
802 | * e.g. this can happen if a snapshot is taken by an administrative | |
803 | * user in the global zone (i.e. a recursive snapshot by root). | |
804 | * However, we must handle the case of delegated permissions where we | |
805 | * are allowed to change the limit on the current dataset, but there | |
806 | * is another limit in the tree above. | |
807 | */ | |
808 | enforce = dsl_enforce_ds_ss_limits(dd, prop, cr); | |
809 | if (enforce == ENFORCE_NEVER) | |
810 | return (0); | |
811 | ||
812 | /* | |
813 | * e.g. if renaming a dataset with no snapshots, count adjustment | |
814 | * is 0. | |
815 | */ | |
816 | if (delta == 0) | |
817 | return (0); | |
818 | ||
819 | if (prop == ZFS_PROP_SNAPSHOT_LIMIT) { | |
820 | /* | |
821 | * We don't enforce the limit for temporary snapshots. This is | |
822 | * indicated by a NULL cred_t argument. | |
823 | */ | |
824 | if (cr == NULL) | |
825 | return (0); | |
826 | ||
827 | count_prop = DD_FIELD_SNAPSHOT_COUNT; | |
828 | } else { | |
829 | count_prop = DD_FIELD_FILESYSTEM_COUNT; | |
830 | } | |
831 | ||
832 | /* | |
833 | * If an ancestor has been provided, stop checking the limit once we | |
834 | * hit that dir. We need this during rename so that we don't overcount | |
835 | * the check once we recurse up to the common ancestor. | |
836 | */ | |
837 | if (ancestor == dd) | |
838 | return (0); | |
839 | ||
840 | /* | |
841 | * If we hit an uninitialized node while recursing up the tree, we can | |
842 | * stop since we know there is no limit here (or above). The counts are | |
843 | * not valid on this node and we know we won't touch this node's counts. | |
844 | */ | |
845 | if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object, | |
846 | count_prop, sizeof (count), 1, &count) == ENOENT) | |
847 | return (0); | |
848 | ||
849 | err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL, | |
850 | B_FALSE); | |
851 | if (err != 0) | |
852 | return (err); | |
853 | ||
854 | /* Is there a limit which we've hit? */ | |
855 | if (enforce == ENFORCE_ALWAYS && (count + delta) > limit) | |
856 | return (SET_ERROR(EDQUOT)); | |
857 | ||
858 | if (dd->dd_parent != NULL) | |
859 | err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop, | |
860 | ancestor, cr); | |
861 | ||
862 | return (err); | |
863 | } | |
864 | ||
865 | /* | |
866 | * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all | |
867 | * parents. When a new filesystem/snapshot is created, increment the count on | |
868 | * all parents, and when a filesystem/snapshot is destroyed, decrement the | |
869 | * count. | |
870 | */ | |
871 | void | |
872 | dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop, | |
873 | dmu_tx_t *tx) | |
874 | { | |
875 | int err; | |
876 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
877 | uint64_t count; | |
878 | ||
879 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
880 | ASSERT(dmu_tx_is_syncing(tx)); | |
881 | ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 || | |
882 | strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0); | |
883 | ||
884 | /* | |
885 | * When we receive an incremental stream into a filesystem that already | |
886 | * exists, a temporary clone is created. We don't count this temporary | |
887 | * clone, whose name begins with a '%'. We also ignore hidden ($FREE, | |
888 | * $MOS & $ORIGIN) objsets. | |
889 | */ | |
890 | if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') && | |
891 | strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0) | |
892 | return; | |
893 | ||
894 | /* | |
895 | * e.g. if renaming a dataset with no snapshots, count adjustment is 0 | |
896 | */ | |
897 | if (delta == 0) | |
898 | return; | |
899 | ||
900 | /* | |
901 | * If we hit an uninitialized node while recursing up the tree, we can | |
902 | * stop since we know the counts are not valid on this node and we | |
903 | * know we shouldn't touch this node's counts. An uninitialized count | |
904 | * on the node indicates that either the feature has not yet been | |
905 | * activated or there are no limits on this part of the tree. | |
906 | */ | |
907 | if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object, | |
908 | prop, sizeof (count), 1, &count)) == ENOENT) | |
909 | return; | |
910 | VERIFY0(err); | |
911 | ||
912 | count += delta; | |
913 | /* Use a signed verify to make sure we're not neg. */ | |
914 | VERIFY3S(count, >=, 0); | |
915 | ||
916 | VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count, | |
917 | tx)); | |
918 | ||
919 | /* Roll up this additional count into our ancestors */ | |
920 | if (dd->dd_parent != NULL) | |
921 | dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx); | |
922 | } | |
923 | ||
34dc7c2f | 924 | uint64_t |
b128c09f BB |
925 | dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, |
926 | dmu_tx_t *tx) | |
34dc7c2f | 927 | { |
b128c09f | 928 | objset_t *mos = dp->dp_meta_objset; |
34dc7c2f | 929 | uint64_t ddobj; |
428870ff | 930 | dsl_dir_phys_t *ddphys; |
34dc7c2f BB |
931 | dmu_buf_t *dbuf; |
932 | ||
933 | ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, | |
934 | DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); | |
b128c09f | 935 | if (pds) { |
d2734cce | 936 | VERIFY0(zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj, |
b128c09f BB |
937 | name, sizeof (uint64_t), 1, &ddobj, tx)); |
938 | } else { | |
939 | /* it's the root dir */ | |
d2734cce | 940 | VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, |
b128c09f BB |
941 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); |
942 | } | |
d2734cce | 943 | VERIFY0(dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); |
34dc7c2f | 944 | dmu_buf_will_dirty(dbuf, tx); |
428870ff | 945 | ddphys = dbuf->db_data; |
34dc7c2f | 946 | |
428870ff | 947 | ddphys->dd_creation_time = gethrestime_sec(); |
788eb90c | 948 | if (pds) { |
428870ff | 949 | ddphys->dd_parent_obj = pds->dd_object; |
788eb90c JJ |
950 | |
951 | /* update the filesystem counts */ | |
952 | dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx); | |
953 | } | |
428870ff | 954 | ddphys->dd_props_zapobj = zap_create(mos, |
34dc7c2f | 955 | DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); |
428870ff | 956 | ddphys->dd_child_dir_zapobj = zap_create(mos, |
34dc7c2f | 957 | DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); |
b128c09f | 958 | if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) |
428870ff | 959 | ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; |
b5256303 | 960 | |
34dc7c2f BB |
961 | dmu_buf_rele(dbuf, FTAG); |
962 | ||
963 | return (ddobj); | |
964 | } | |
965 | ||
b128c09f BB |
966 | boolean_t |
967 | dsl_dir_is_clone(dsl_dir_t *dd) | |
34dc7c2f | 968 | { |
d683ddbb | 969 | return (dsl_dir_phys(dd)->dd_origin_obj && |
b128c09f | 970 | (dd->dd_pool->dp_origin_snap == NULL || |
d683ddbb | 971 | dsl_dir_phys(dd)->dd_origin_obj != |
b128c09f | 972 | dd->dd_pool->dp_origin_snap->ds_object)); |
34dc7c2f BB |
973 | } |
974 | ||
d99a0153 CW |
975 | uint64_t |
976 | dsl_dir_get_used(dsl_dir_t *dd) | |
977 | { | |
978 | return (dsl_dir_phys(dd)->dd_used_bytes); | |
979 | } | |
980 | ||
d2734cce SD |
981 | uint64_t |
982 | dsl_dir_get_compressed(dsl_dir_t *dd) | |
983 | { | |
984 | return (dsl_dir_phys(dd)->dd_compressed_bytes); | |
985 | } | |
986 | ||
d99a0153 CW |
987 | uint64_t |
988 | dsl_dir_get_quota(dsl_dir_t *dd) | |
989 | { | |
990 | return (dsl_dir_phys(dd)->dd_quota); | |
991 | } | |
992 | ||
993 | uint64_t | |
994 | dsl_dir_get_reservation(dsl_dir_t *dd) | |
995 | { | |
996 | return (dsl_dir_phys(dd)->dd_reserved); | |
997 | } | |
998 | ||
999 | uint64_t | |
1000 | dsl_dir_get_compressratio(dsl_dir_t *dd) | |
1001 | { | |
1002 | /* a fixed point number, 100x the ratio */ | |
1003 | return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 : | |
1004 | (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 / | |
1005 | dsl_dir_phys(dd)->dd_compressed_bytes)); | |
1006 | } | |
1007 | ||
1008 | uint64_t | |
1009 | dsl_dir_get_logicalused(dsl_dir_t *dd) | |
1010 | { | |
1011 | return (dsl_dir_phys(dd)->dd_uncompressed_bytes); | |
1012 | } | |
1013 | ||
1014 | uint64_t | |
1015 | dsl_dir_get_usedsnap(dsl_dir_t *dd) | |
1016 | { | |
1017 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]); | |
1018 | } | |
1019 | ||
1020 | uint64_t | |
1021 | dsl_dir_get_usedds(dsl_dir_t *dd) | |
1022 | { | |
1023 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]); | |
1024 | } | |
1025 | ||
1026 | uint64_t | |
1027 | dsl_dir_get_usedrefreserv(dsl_dir_t *dd) | |
1028 | { | |
1029 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]); | |
1030 | } | |
1031 | ||
1032 | uint64_t | |
1033 | dsl_dir_get_usedchild(dsl_dir_t *dd) | |
1034 | { | |
1035 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] + | |
1036 | dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]); | |
1037 | } | |
1038 | ||
34dc7c2f | 1039 | void |
d99a0153 CW |
1040 | dsl_dir_get_origin(dsl_dir_t *dd, char *buf) |
1041 | { | |
1042 | dsl_dataset_t *ds; | |
1043 | VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, | |
1044 | dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds)); | |
1045 | ||
1046 | dsl_dataset_name(ds, buf); | |
1047 | ||
1048 | dsl_dataset_rele(ds, FTAG); | |
1049 | } | |
1050 | ||
1051 | int | |
1052 | dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count) | |
34dc7c2f | 1053 | { |
d99a0153 CW |
1054 | if (dsl_dir_is_zapified(dd)) { |
1055 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
1056 | return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, | |
1057 | sizeof (*count), 1, count)); | |
1058 | } else { | |
1059 | return (ENOENT); | |
1060 | } | |
1061 | } | |
1062 | ||
1063 | int | |
1064 | dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count) | |
1065 | { | |
1066 | if (dsl_dir_is_zapified(dd)) { | |
1067 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
1068 | return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, | |
1069 | sizeof (*count), 1, count)); | |
1070 | } else { | |
1071 | return (ENOENT); | |
1072 | } | |
1073 | } | |
b5256303 | 1074 | |
d99a0153 CW |
1075 | void |
1076 | dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) | |
1077 | { | |
34dc7c2f | 1078 | mutex_enter(&dd->dd_lock); |
d683ddbb | 1079 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, |
d99a0153 | 1080 | dsl_dir_get_quota(dd)); |
34dc7c2f | 1081 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, |
d99a0153 | 1082 | dsl_dir_get_reservation(dd)); |
24a64651 | 1083 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, |
d99a0153 | 1084 | dsl_dir_get_logicalused(dd)); |
d683ddbb | 1085 | if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { |
b128c09f | 1086 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, |
d99a0153 | 1087 | dsl_dir_get_usedsnap(dd)); |
b128c09f | 1088 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, |
d99a0153 | 1089 | dsl_dir_get_usedds(dd)); |
b128c09f | 1090 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, |
d99a0153 | 1091 | dsl_dir_get_usedrefreserv(dd)); |
b128c09f | 1092 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, |
d99a0153 | 1093 | dsl_dir_get_usedchild(dd)); |
b128c09f | 1094 | } |
34dc7c2f BB |
1095 | mutex_exit(&dd->dd_lock); |
1096 | ||
d99a0153 CW |
1097 | uint64_t count; |
1098 | if (dsl_dir_get_filesystem_count(dd, &count) == 0) { | |
1099 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT, | |
1100 | count); | |
1101 | } | |
1102 | if (dsl_dir_get_snapshot_count(dd, &count) == 0) { | |
1103 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT, | |
1104 | count); | |
788eb90c JJ |
1105 | } |
1106 | ||
b128c09f | 1107 | if (dsl_dir_is_clone(dd)) { |
eca7b760 | 1108 | char buf[ZFS_MAX_DATASET_NAME_LEN]; |
d99a0153 | 1109 | dsl_dir_get_origin(dd, buf); |
34dc7c2f BB |
1110 | dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); |
1111 | } | |
d99a0153 | 1112 | |
34dc7c2f BB |
1113 | } |
1114 | ||
1115 | void | |
1116 | dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) | |
1117 | { | |
1118 | dsl_pool_t *dp = dd->dd_pool; | |
1119 | ||
d683ddbb | 1120 | ASSERT(dsl_dir_phys(dd)); |
34dc7c2f | 1121 | |
13fe0198 | 1122 | if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { |
34dc7c2f BB |
1123 | /* up the hold count until we can be written out */ |
1124 | dmu_buf_add_ref(dd->dd_dbuf, dd); | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | static int64_t | |
1129 | parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) | |
1130 | { | |
d683ddbb JG |
1131 | uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved); |
1132 | uint64_t new_accounted = | |
1133 | MAX(used + delta, dsl_dir_phys(dd)->dd_reserved); | |
34dc7c2f BB |
1134 | return (new_accounted - old_accounted); |
1135 | } | |
1136 | ||
1137 | void | |
1138 | dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) | |
1139 | { | |
1140 | ASSERT(dmu_tx_is_syncing(tx)); | |
1141 | ||
34dc7c2f | 1142 | mutex_enter(&dd->dd_lock); |
3fa93bb8 | 1143 | ASSERT0(dd->dd_tempreserved[tx->tx_txg & TXG_MASK]); |
34dc7c2f | 1144 | dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, |
3fa93bb8 BB |
1145 | dd->dd_space_towrite[tx->tx_txg & TXG_MASK] / 1024); |
1146 | dd->dd_space_towrite[tx->tx_txg & TXG_MASK] = 0; | |
34dc7c2f BB |
1147 | mutex_exit(&dd->dd_lock); |
1148 | ||
1149 | /* release the hold from dsl_dir_dirty */ | |
1150 | dmu_buf_rele(dd->dd_dbuf, dd); | |
1151 | } | |
1152 | ||
1153 | static uint64_t | |
1154 | dsl_dir_space_towrite(dsl_dir_t *dd) | |
1155 | { | |
1156 | uint64_t space = 0; | |
34dc7c2f BB |
1157 | |
1158 | ASSERT(MUTEX_HELD(&dd->dd_lock)); | |
1159 | ||
3ec3bc21 BB |
1160 | for (int i = 0; i < TXG_SIZE; i++) { |
1161 | space += dd->dd_space_towrite[i & TXG_MASK]; | |
1162 | ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0); | |
34dc7c2f BB |
1163 | } |
1164 | return (space); | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * How much space would dd have available if ancestor had delta applied | |
1169 | * to it? If ondiskonly is set, we're only interested in what's | |
1170 | * on-disk, not estimated pending changes. | |
1171 | */ | |
1172 | uint64_t | |
1173 | dsl_dir_space_available(dsl_dir_t *dd, | |
1174 | dsl_dir_t *ancestor, int64_t delta, int ondiskonly) | |
1175 | { | |
1176 | uint64_t parentspace, myspace, quota, used; | |
1177 | ||
1178 | /* | |
1179 | * If there are no restrictions otherwise, assume we have | |
1180 | * unlimited space available. | |
1181 | */ | |
1182 | quota = UINT64_MAX; | |
1183 | parentspace = UINT64_MAX; | |
1184 | ||
1185 | if (dd->dd_parent != NULL) { | |
1186 | parentspace = dsl_dir_space_available(dd->dd_parent, | |
1187 | ancestor, delta, ondiskonly); | |
1188 | } | |
1189 | ||
1190 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1191 | if (dsl_dir_phys(dd)->dd_quota != 0) |
1192 | quota = dsl_dir_phys(dd)->dd_quota; | |
1193 | used = dsl_dir_phys(dd)->dd_used_bytes; | |
34dc7c2f BB |
1194 | if (!ondiskonly) |
1195 | used += dsl_dir_space_towrite(dd); | |
34dc7c2f BB |
1196 | |
1197 | if (dd->dd_parent == NULL) { | |
d2734cce SD |
1198 | uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, |
1199 | ZFS_SPACE_CHECK_NORMAL); | |
34dc7c2f BB |
1200 | quota = MIN(quota, poolsize); |
1201 | } | |
1202 | ||
d683ddbb | 1203 | if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) { |
34dc7c2f BB |
1204 | /* |
1205 | * We have some space reserved, in addition to what our | |
1206 | * parent gave us. | |
1207 | */ | |
d683ddbb | 1208 | parentspace += dsl_dir_phys(dd)->dd_reserved - used; |
34dc7c2f BB |
1209 | } |
1210 | ||
b128c09f BB |
1211 | if (dd == ancestor) { |
1212 | ASSERT(delta <= 0); | |
1213 | ASSERT(used >= -delta); | |
1214 | used += delta; | |
1215 | if (parentspace != UINT64_MAX) | |
1216 | parentspace -= delta; | |
1217 | } | |
1218 | ||
34dc7c2f BB |
1219 | if (used > quota) { |
1220 | /* over quota */ | |
1221 | myspace = 0; | |
34dc7c2f BB |
1222 | } else { |
1223 | /* | |
1224 | * the lesser of the space provided by our parent and | |
1225 | * the space left in our quota | |
1226 | */ | |
1227 | myspace = MIN(parentspace, quota - used); | |
1228 | } | |
1229 | ||
1230 | mutex_exit(&dd->dd_lock); | |
1231 | ||
1232 | return (myspace); | |
1233 | } | |
1234 | ||
1235 | struct tempreserve { | |
1236 | list_node_t tr_node; | |
34dc7c2f BB |
1237 | dsl_dir_t *tr_ds; |
1238 | uint64_t tr_size; | |
1239 | }; | |
1240 | ||
1241 | static int | |
1242 | dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, | |
3ec3bc21 | 1243 | boolean_t ignorequota, list_t *tr_list, |
34dc7c2f BB |
1244 | dmu_tx_t *tx, boolean_t first) |
1245 | { | |
419c80e6 | 1246 | uint64_t txg; |
3ec3bc21 | 1247 | uint64_t quota; |
34dc7c2f | 1248 | struct tempreserve *tr; |
419c80e6 D |
1249 | int retval; |
1250 | uint64_t ref_rsrv; | |
1251 | ||
1252 | top_of_function: | |
1253 | txg = tx->tx_txg; | |
1254 | retval = EDQUOT; | |
1255 | ref_rsrv = 0; | |
34dc7c2f BB |
1256 | |
1257 | ASSERT3U(txg, !=, 0); | |
1258 | ASSERT3S(asize, >, 0); | |
1259 | ||
1260 | mutex_enter(&dd->dd_lock); | |
1261 | ||
1262 | /* | |
1263 | * Check against the dsl_dir's quota. We don't add in the delta | |
1264 | * when checking for over-quota because they get one free hit. | |
1265 | */ | |
3ec3bc21 BB |
1266 | uint64_t est_inflight = dsl_dir_space_towrite(dd); |
1267 | for (int i = 0; i < TXG_SIZE; i++) | |
34dc7c2f | 1268 | est_inflight += dd->dd_tempreserved[i]; |
3ec3bc21 | 1269 | uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes; |
34dc7c2f BB |
1270 | |
1271 | /* | |
1272 | * On the first iteration, fetch the dataset's used-on-disk and | |
1273 | * refreservation values. Also, if checkrefquota is set, test if | |
1274 | * allocating this space would exceed the dataset's refquota. | |
1275 | */ | |
1276 | if (first && tx->tx_objset) { | |
1277 | int error; | |
428870ff | 1278 | dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; |
34dc7c2f | 1279 | |
3ec3bc21 | 1280 | error = dsl_dataset_check_quota(ds, !netfree, |
34dc7c2f | 1281 | asize, est_inflight, &used_on_disk, &ref_rsrv); |
3ec3bc21 | 1282 | if (error != 0) { |
34dc7c2f | 1283 | mutex_exit(&dd->dd_lock); |
3d920a15 | 1284 | DMU_TX_STAT_BUMP(dmu_tx_quota); |
34dc7c2f BB |
1285 | return (error); |
1286 | } | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * If this transaction will result in a net free of space, | |
1291 | * we want to let it through. | |
1292 | */ | |
d683ddbb | 1293 | if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0) |
34dc7c2f BB |
1294 | quota = UINT64_MAX; |
1295 | else | |
d683ddbb | 1296 | quota = dsl_dir_phys(dd)->dd_quota; |
34dc7c2f BB |
1297 | |
1298 | /* | |
428870ff BB |
1299 | * Adjust the quota against the actual pool size at the root |
1300 | * minus any outstanding deferred frees. | |
34dc7c2f BB |
1301 | * To ensure that it's possible to remove files from a full |
1302 | * pool without inducing transient overcommits, we throttle | |
1303 | * netfree transactions against a quota that is slightly larger, | |
1304 | * but still within the pool's allocation slop. In cases where | |
1305 | * we're very close to full, this will allow a steady trickle of | |
1306 | * removes to get through. | |
1307 | */ | |
3ec3bc21 | 1308 | uint64_t deferred = 0; |
34dc7c2f | 1309 | if (dd->dd_parent == NULL) { |
d2734cce SD |
1310 | uint64_t avail = dsl_pool_unreserved_space(dd->dd_pool, |
1311 | (netfree) ? | |
1312 | ZFS_SPACE_CHECK_RESERVED : ZFS_SPACE_CHECK_NORMAL); | |
1313 | ||
1314 | if (avail < quota) { | |
1315 | quota = avail; | |
428870ff | 1316 | retval = ENOSPC; |
34dc7c2f BB |
1317 | } |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * If they are requesting more space, and our current estimate | |
1322 | * is over quota, they get to try again unless the actual | |
1323 | * on-disk is over quota and there are no pending changes (which | |
1324 | * may free up space for us). | |
1325 | */ | |
428870ff BB |
1326 | if (used_on_disk + est_inflight >= quota) { |
1327 | if (est_inflight > 0 || used_on_disk < quota || | |
1328 | (retval == ENOSPC && used_on_disk < quota + deferred)) | |
1329 | retval = ERESTART; | |
34dc7c2f BB |
1330 | dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " |
1331 | "quota=%lluK tr=%lluK err=%d\n", | |
1332 | used_on_disk>>10, est_inflight>>10, | |
428870ff | 1333 | quota>>10, asize>>10, retval); |
34dc7c2f | 1334 | mutex_exit(&dd->dd_lock); |
3d920a15 | 1335 | DMU_TX_STAT_BUMP(dmu_tx_quota); |
2e528b49 | 1336 | return (SET_ERROR(retval)); |
34dc7c2f BB |
1337 | } |
1338 | ||
1339 | /* We need to up our estimated delta before dropping dd_lock */ | |
3ec3bc21 | 1340 | dd->dd_tempreserved[txg & TXG_MASK] += asize; |
34dc7c2f | 1341 | |
3ec3bc21 | 1342 | uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, |
34dc7c2f BB |
1343 | asize - ref_rsrv); |
1344 | mutex_exit(&dd->dd_lock); | |
1345 | ||
79c76d5b | 1346 | tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); |
34dc7c2f BB |
1347 | tr->tr_ds = dd; |
1348 | tr->tr_size = asize; | |
1349 | list_insert_tail(tr_list, tr); | |
1350 | ||
1351 | /* see if it's OK with our parent */ | |
3ec3bc21 | 1352 | if (dd->dd_parent != NULL && parent_rsrv != 0) { |
419c80e6 D |
1353 | /* |
1354 | * Recurse on our parent without recursion. This has been | |
1355 | * observed to be potentially large stack usage even within | |
1356 | * the test suite. Largest seen stack was 7632 bytes on linux. | |
1357 | */ | |
1358 | ||
1359 | dd = dd->dd_parent; | |
1360 | asize = parent_rsrv; | |
1361 | ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0); | |
1362 | first = B_FALSE; | |
1363 | goto top_of_function; | |
34dc7c2f | 1364 | |
34dc7c2f BB |
1365 | } else { |
1366 | return (0); | |
1367 | } | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Reserve space in this dsl_dir, to be used in this tx's txg. | |
1372 | * After the space has been dirtied (and dsl_dir_willuse_space() | |
1373 | * has been called), the reservation should be canceled, using | |
1374 | * dsl_dir_tempreserve_clear(). | |
1375 | */ | |
1376 | int | |
1377 | dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, | |
3ec3bc21 | 1378 | boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx) |
34dc7c2f BB |
1379 | { |
1380 | int err; | |
1381 | list_t *tr_list; | |
1382 | ||
1383 | if (asize == 0) { | |
1384 | *tr_cookiep = NULL; | |
1385 | return (0); | |
1386 | } | |
1387 | ||
79c76d5b | 1388 | tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); |
34dc7c2f BB |
1389 | list_create(tr_list, sizeof (struct tempreserve), |
1390 | offsetof(struct tempreserve, tr_node)); | |
1391 | ASSERT3S(asize, >, 0); | |
34dc7c2f | 1392 | |
dae3e9ea | 1393 | err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg); |
34dc7c2f BB |
1394 | if (err == 0) { |
1395 | struct tempreserve *tr; | |
1396 | ||
79c76d5b | 1397 | tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); |
34dc7c2f BB |
1398 | tr->tr_size = lsize; |
1399 | list_insert_tail(tr_list, tr); | |
34dc7c2f BB |
1400 | } else { |
1401 | if (err == EAGAIN) { | |
e8b96c60 MA |
1402 | /* |
1403 | * If arc_memory_throttle() detected that pageout | |
1404 | * is running and we are low on memory, we delay new | |
1405 | * non-pageout transactions to give pageout an | |
1406 | * advantage. | |
1407 | * | |
1408 | * It is unfortunate to be delaying while the caller's | |
1409 | * locks are held. | |
1410 | */ | |
63fd3c6c AL |
1411 | txg_delay(dd->dd_pool, tx->tx_txg, |
1412 | MSEC2NSEC(10), MSEC2NSEC(10)); | |
2e528b49 | 1413 | err = SET_ERROR(ERESTART); |
34dc7c2f | 1414 | } |
34dc7c2f BB |
1415 | } |
1416 | ||
1417 | if (err == 0) { | |
3ec3bc21 BB |
1418 | err = dsl_dir_tempreserve_impl(dd, asize, netfree, |
1419 | B_FALSE, tr_list, tx, B_TRUE); | |
34dc7c2f BB |
1420 | } |
1421 | ||
13fe0198 | 1422 | if (err != 0) |
34dc7c2f BB |
1423 | dsl_dir_tempreserve_clear(tr_list, tx); |
1424 | else | |
1425 | *tr_cookiep = tr_list; | |
1426 | ||
1427 | return (err); | |
1428 | } | |
1429 | ||
1430 | /* | |
1431 | * Clear a temporary reservation that we previously made with | |
1432 | * dsl_dir_tempreserve_space(). | |
1433 | */ | |
1434 | void | |
1435 | dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) | |
1436 | { | |
1437 | int txgidx = tx->tx_txg & TXG_MASK; | |
1438 | list_t *tr_list = tr_cookie; | |
1439 | struct tempreserve *tr; | |
1440 | ||
1441 | ASSERT3U(tx->tx_txg, !=, 0); | |
1442 | ||
1443 | if (tr_cookie == NULL) | |
1444 | return; | |
1445 | ||
e8b96c60 MA |
1446 | while ((tr = list_head(tr_list)) != NULL) { |
1447 | if (tr->tr_ds) { | |
34dc7c2f BB |
1448 | mutex_enter(&tr->tr_ds->dd_lock); |
1449 | ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, | |
1450 | tr->tr_size); | |
1451 | tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; | |
1452 | mutex_exit(&tr->tr_ds->dd_lock); | |
1453 | } else { | |
1454 | arc_tempreserve_clear(tr->tr_size); | |
1455 | } | |
1456 | list_remove(tr_list, tr); | |
1457 | kmem_free(tr, sizeof (struct tempreserve)); | |
1458 | } | |
1459 | ||
1460 | kmem_free(tr_list, sizeof (list_t)); | |
1461 | } | |
1462 | ||
e8b96c60 MA |
1463 | /* |
1464 | * This should be called from open context when we think we're going to write | |
1465 | * or free space, for example when dirtying data. Be conservative; it's okay | |
1466 | * to write less space or free more, but we don't want to write more or free | |
1467 | * less than the amount specified. | |
1ba16159 AB |
1468 | * |
1469 | * NOTE: The behavior of this function is identical to the Illumos / FreeBSD | |
1470 | * version however it has been adjusted to use an iterative rather then | |
1471 | * recursive algorithm to minimize stack usage. | |
e8b96c60 MA |
1472 | */ |
1473 | void | |
1474 | dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) | |
34dc7c2f BB |
1475 | { |
1476 | int64_t parent_space; | |
1477 | uint64_t est_used; | |
1478 | ||
1ba16159 AB |
1479 | do { |
1480 | mutex_enter(&dd->dd_lock); | |
1481 | if (space > 0) | |
1482 | dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; | |
34dc7c2f | 1483 | |
1ba16159 | 1484 | est_used = dsl_dir_space_towrite(dd) + |
d683ddbb | 1485 | dsl_dir_phys(dd)->dd_used_bytes; |
1ba16159 AB |
1486 | parent_space = parent_delta(dd, est_used, space); |
1487 | mutex_exit(&dd->dd_lock); | |
34dc7c2f | 1488 | |
1ba16159 AB |
1489 | /* Make sure that we clean up dd_space_to* */ |
1490 | dsl_dir_dirty(dd, tx); | |
34dc7c2f | 1491 | |
1ba16159 AB |
1492 | dd = dd->dd_parent; |
1493 | space = parent_space; | |
1494 | } while (space && dd); | |
34dc7c2f BB |
1495 | } |
1496 | ||
1497 | /* call from syncing context when we actually write/free space for this dd */ | |
1498 | void | |
b128c09f | 1499 | dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, |
34dc7c2f BB |
1500 | int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) |
1501 | { | |
1502 | int64_t accounted_delta; | |
a169a625 MA |
1503 | |
1504 | /* | |
1505 | * dsl_dataset_set_refreservation_sync_impl() calls this with | |
1506 | * dd_lock held, so that it can atomically update | |
1507 | * ds->ds_reserved and the dsl_dir accounting, so that | |
1508 | * dsl_dataset_check_quota() can see dataset and dir accounting | |
1509 | * consistently. | |
1510 | */ | |
b128c09f | 1511 | boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); |
34dc7c2f BB |
1512 | |
1513 | ASSERT(dmu_tx_is_syncing(tx)); | |
b128c09f | 1514 | ASSERT(type < DD_USED_NUM); |
34dc7c2f | 1515 | |
a169a625 MA |
1516 | dmu_buf_will_dirty(dd->dd_dbuf, tx); |
1517 | ||
b128c09f BB |
1518 | if (needlock) |
1519 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1520 | accounted_delta = |
1521 | parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used); | |
1522 | ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used); | |
34dc7c2f | 1523 | ASSERT(compressed >= 0 || |
d683ddbb | 1524 | dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed); |
34dc7c2f | 1525 | ASSERT(uncompressed >= 0 || |
d683ddbb JG |
1526 | dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed); |
1527 | dsl_dir_phys(dd)->dd_used_bytes += used; | |
1528 | dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed; | |
1529 | dsl_dir_phys(dd)->dd_compressed_bytes += compressed; | |
b128c09f | 1530 | |
d683ddbb | 1531 | if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { |
b128c09f | 1532 | ASSERT(used > 0 || |
d683ddbb JG |
1533 | dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used); |
1534 | dsl_dir_phys(dd)->dd_used_breakdown[type] += used; | |
b128c09f | 1535 | #ifdef DEBUG |
d6320ddb BB |
1536 | { |
1537 | dd_used_t t; | |
1538 | uint64_t u = 0; | |
1539 | for (t = 0; t < DD_USED_NUM; t++) | |
d683ddbb JG |
1540 | u += dsl_dir_phys(dd)->dd_used_breakdown[t]; |
1541 | ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes); | |
d6320ddb | 1542 | } |
b128c09f BB |
1543 | #endif |
1544 | } | |
1545 | if (needlock) | |
1546 | mutex_exit(&dd->dd_lock); | |
34dc7c2f BB |
1547 | |
1548 | if (dd->dd_parent != NULL) { | |
b128c09f | 1549 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, |
34dc7c2f | 1550 | accounted_delta, compressed, uncompressed, tx); |
b128c09f BB |
1551 | dsl_dir_transfer_space(dd->dd_parent, |
1552 | used - accounted_delta, | |
1553 | DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); | |
34dc7c2f BB |
1554 | } |
1555 | } | |
1556 | ||
b128c09f BB |
1557 | void |
1558 | dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, | |
1559 | dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) | |
1560 | { | |
b128c09f BB |
1561 | ASSERT(dmu_tx_is_syncing(tx)); |
1562 | ASSERT(oldtype < DD_USED_NUM); | |
1563 | ASSERT(newtype < DD_USED_NUM); | |
1564 | ||
d683ddbb JG |
1565 | if (delta == 0 || |
1566 | !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN)) | |
b128c09f BB |
1567 | return; |
1568 | ||
a169a625 MA |
1569 | dmu_buf_will_dirty(dd->dd_dbuf, tx); |
1570 | mutex_enter(&dd->dd_lock); | |
b128c09f | 1571 | ASSERT(delta > 0 ? |
d683ddbb JG |
1572 | dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta : |
1573 | dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta); | |
1574 | ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta)); | |
1575 | dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta; | |
1576 | dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta; | |
a169a625 | 1577 | mutex_exit(&dd->dd_lock); |
b128c09f BB |
1578 | } |
1579 | ||
13fe0198 MA |
1580 | typedef struct dsl_dir_set_qr_arg { |
1581 | const char *ddsqra_name; | |
1582 | zprop_source_t ddsqra_source; | |
1583 | uint64_t ddsqra_value; | |
1584 | } dsl_dir_set_qr_arg_t; | |
1585 | ||
34dc7c2f | 1586 | static int |
13fe0198 | 1587 | dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1588 | { |
13fe0198 MA |
1589 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1590 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1591 | dsl_dataset_t *ds; | |
1592 | int error; | |
1593 | uint64_t towrite, newval; | |
34dc7c2f | 1594 | |
13fe0198 MA |
1595 | error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); |
1596 | if (error != 0) | |
1597 | return (error); | |
1598 | ||
1599 | error = dsl_prop_predict(ds->ds_dir, "quota", | |
1600 | ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); | |
1601 | if (error != 0) { | |
1602 | dsl_dataset_rele(ds, FTAG); | |
1603 | return (error); | |
1604 | } | |
428870ff | 1605 | |
13fe0198 MA |
1606 | if (newval == 0) { |
1607 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f | 1608 | return (0); |
13fe0198 | 1609 | } |
34dc7c2f | 1610 | |
13fe0198 | 1611 | mutex_enter(&ds->ds_dir->dd_lock); |
34dc7c2f BB |
1612 | /* |
1613 | * If we are doing the preliminary check in open context, and | |
1614 | * there are pending changes, then don't fail it, since the | |
1615 | * pending changes could under-estimate the amount of space to be | |
1616 | * freed up. | |
1617 | */ | |
13fe0198 | 1618 | towrite = dsl_dir_space_towrite(ds->ds_dir); |
34dc7c2f | 1619 | if ((dmu_tx_is_syncing(tx) || towrite == 0) && |
d683ddbb JG |
1620 | (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved || |
1621 | newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) { | |
2e528b49 | 1622 | error = SET_ERROR(ENOSPC); |
34dc7c2f | 1623 | } |
13fe0198 MA |
1624 | mutex_exit(&ds->ds_dir->dd_lock); |
1625 | dsl_dataset_rele(ds, FTAG); | |
1626 | return (error); | |
34dc7c2f BB |
1627 | } |
1628 | ||
34dc7c2f | 1629 | static void |
13fe0198 | 1630 | dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1631 | { |
13fe0198 MA |
1632 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1633 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1634 | dsl_dataset_t *ds; | |
1635 | uint64_t newval; | |
428870ff | 1636 | |
13fe0198 | 1637 | VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); |
34dc7c2f | 1638 | |
b1118acb MM |
1639 | if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { |
1640 | dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), | |
1641 | ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, | |
1642 | &ddsqra->ddsqra_value, tx); | |
34dc7c2f | 1643 | |
b1118acb MM |
1644 | VERIFY0(dsl_prop_get_int_ds(ds, |
1645 | zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); | |
1646 | } else { | |
1647 | newval = ddsqra->ddsqra_value; | |
1648 | spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", | |
1649 | zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); | |
1650 | } | |
6f1ffb06 | 1651 | |
13fe0198 MA |
1652 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); |
1653 | mutex_enter(&ds->ds_dir->dd_lock); | |
d683ddbb | 1654 | dsl_dir_phys(ds->ds_dir)->dd_quota = newval; |
13fe0198 MA |
1655 | mutex_exit(&ds->ds_dir->dd_lock); |
1656 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
1657 | } |
1658 | ||
1659 | int | |
428870ff | 1660 | dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) |
34dc7c2f | 1661 | { |
13fe0198 | 1662 | dsl_dir_set_qr_arg_t ddsqra; |
428870ff | 1663 | |
13fe0198 MA |
1664 | ddsqra.ddsqra_name = ddname; |
1665 | ddsqra.ddsqra_source = source; | |
1666 | ddsqra.ddsqra_value = quota; | |
428870ff | 1667 | |
13fe0198 | 1668 | return (dsl_sync_task(ddname, dsl_dir_set_quota_check, |
d2734cce SD |
1669 | dsl_dir_set_quota_sync, &ddsqra, 0, |
1670 | ZFS_SPACE_CHECK_EXTRA_RESERVED)); | |
34dc7c2f BB |
1671 | } |
1672 | ||
1673 | int | |
13fe0198 | 1674 | dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1675 | { |
13fe0198 MA |
1676 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1677 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1678 | dsl_dataset_t *ds; | |
1679 | dsl_dir_t *dd; | |
1680 | uint64_t newval, used, avail; | |
1681 | int error; | |
428870ff | 1682 | |
13fe0198 MA |
1683 | error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); |
1684 | if (error != 0) | |
1685 | return (error); | |
1686 | dd = ds->ds_dir; | |
34dc7c2f BB |
1687 | |
1688 | /* | |
1689 | * If we are doing the preliminary check in open context, the | |
1690 | * space estimates may be inaccurate. | |
1691 | */ | |
13fe0198 MA |
1692 | if (!dmu_tx_is_syncing(tx)) { |
1693 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f | 1694 | return (0); |
13fe0198 MA |
1695 | } |
1696 | ||
1697 | error = dsl_prop_predict(ds->ds_dir, | |
1698 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1699 | ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); | |
1700 | if (error != 0) { | |
1701 | dsl_dataset_rele(ds, FTAG); | |
1702 | return (error); | |
1703 | } | |
34dc7c2f BB |
1704 | |
1705 | mutex_enter(&dd->dd_lock); | |
d683ddbb | 1706 | used = dsl_dir_phys(dd)->dd_used_bytes; |
34dc7c2f BB |
1707 | mutex_exit(&dd->dd_lock); |
1708 | ||
1709 | if (dd->dd_parent) { | |
1710 | avail = dsl_dir_space_available(dd->dd_parent, | |
1711 | NULL, 0, FALSE); | |
1712 | } else { | |
d2734cce SD |
1713 | avail = dsl_pool_adjustedsize(dd->dd_pool, |
1714 | ZFS_SPACE_CHECK_NORMAL) - used; | |
34dc7c2f BB |
1715 | } |
1716 | ||
d683ddbb | 1717 | if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) { |
13fe0198 | 1718 | uint64_t delta = MAX(used, newval) - |
d683ddbb | 1719 | MAX(used, dsl_dir_phys(dd)->dd_reserved); |
d164b209 | 1720 | |
13fe0198 | 1721 | if (delta > avail || |
d683ddbb JG |
1722 | (dsl_dir_phys(dd)->dd_quota > 0 && |
1723 | newval > dsl_dir_phys(dd)->dd_quota)) | |
2e528b49 | 1724 | error = SET_ERROR(ENOSPC); |
d164b209 BB |
1725 | } |
1726 | ||
13fe0198 MA |
1727 | dsl_dataset_rele(ds, FTAG); |
1728 | return (error); | |
34dc7c2f BB |
1729 | } |
1730 | ||
13fe0198 | 1731 | void |
6f1ffb06 | 1732 | dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) |
34dc7c2f | 1733 | { |
34dc7c2f BB |
1734 | uint64_t used; |
1735 | int64_t delta; | |
1736 | ||
1737 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
1738 | ||
1739 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1740 | used = dsl_dir_phys(dd)->dd_used_bytes; |
1741 | delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved); | |
1742 | dsl_dir_phys(dd)->dd_reserved = value; | |
34dc7c2f BB |
1743 | |
1744 | if (dd->dd_parent != NULL) { | |
1745 | /* Roll up this additional usage into our ancestors */ | |
b128c09f BB |
1746 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, |
1747 | delta, 0, 0, tx); | |
34dc7c2f | 1748 | } |
b128c09f | 1749 | mutex_exit(&dd->dd_lock); |
34dc7c2f BB |
1750 | } |
1751 | ||
6f1ffb06 | 1752 | static void |
13fe0198 | 1753 | dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) |
6f1ffb06 | 1754 | { |
13fe0198 MA |
1755 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1756 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1757 | dsl_dataset_t *ds; | |
1758 | uint64_t newval; | |
6f1ffb06 | 1759 | |
13fe0198 MA |
1760 | VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); |
1761 | ||
b1118acb MM |
1762 | if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { |
1763 | dsl_prop_set_sync_impl(ds, | |
1764 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1765 | ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, | |
1766 | &ddsqra->ddsqra_value, tx); | |
d1d7e268 | 1767 | |
b1118acb MM |
1768 | VERIFY0(dsl_prop_get_int_ds(ds, |
1769 | zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); | |
1770 | } else { | |
1771 | newval = ddsqra->ddsqra_value; | |
1772 | spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", | |
1773 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1774 | (longlong_t)newval); | |
1775 | } | |
1776 | ||
13fe0198 MA |
1777 | dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); |
1778 | dsl_dataset_rele(ds, FTAG); | |
d1d7e268 | 1779 | } |
6f1ffb06 | 1780 | |
34dc7c2f | 1781 | int |
428870ff BB |
1782 | dsl_dir_set_reservation(const char *ddname, zprop_source_t source, |
1783 | uint64_t reservation) | |
34dc7c2f | 1784 | { |
13fe0198 | 1785 | dsl_dir_set_qr_arg_t ddsqra; |
428870ff | 1786 | |
13fe0198 MA |
1787 | ddsqra.ddsqra_name = ddname; |
1788 | ddsqra.ddsqra_source = source; | |
1789 | ddsqra.ddsqra_value = reservation; | |
428870ff | 1790 | |
13fe0198 | 1791 | return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, |
d2734cce SD |
1792 | dsl_dir_set_reservation_sync, &ddsqra, 0, |
1793 | ZFS_SPACE_CHECK_EXTRA_RESERVED)); | |
34dc7c2f BB |
1794 | } |
1795 | ||
1796 | static dsl_dir_t * | |
1797 | closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) | |
1798 | { | |
1799 | for (; ds1; ds1 = ds1->dd_parent) { | |
1800 | dsl_dir_t *dd; | |
1801 | for (dd = ds2; dd; dd = dd->dd_parent) { | |
1802 | if (ds1 == dd) | |
1803 | return (dd); | |
1804 | } | |
1805 | } | |
1806 | return (NULL); | |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * If delta is applied to dd, how much of that delta would be applied to | |
1811 | * ancestor? Syncing context only. | |
1812 | */ | |
1813 | static int64_t | |
1814 | would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) | |
1815 | { | |
1816 | if (dd == ancestor) | |
1817 | return (delta); | |
1818 | ||
1819 | mutex_enter(&dd->dd_lock); | |
d683ddbb | 1820 | delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta); |
34dc7c2f BB |
1821 | mutex_exit(&dd->dd_lock); |
1822 | return (would_change(dd->dd_parent, delta, ancestor)); | |
1823 | } | |
1824 | ||
13fe0198 MA |
1825 | typedef struct dsl_dir_rename_arg { |
1826 | const char *ddra_oldname; | |
1827 | const char *ddra_newname; | |
788eb90c | 1828 | cred_t *ddra_cred; |
13fe0198 | 1829 | } dsl_dir_rename_arg_t; |
34dc7c2f | 1830 | |
a7ed98d8 SD |
1831 | typedef struct dsl_valid_rename_arg { |
1832 | int char_delta; | |
1833 | int nest_delta; | |
1834 | } dsl_valid_rename_arg_t; | |
1835 | ||
13fe0198 | 1836 | /* ARGSUSED */ |
34dc7c2f | 1837 | static int |
13fe0198 | 1838 | dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
34dc7c2f | 1839 | { |
a7ed98d8 | 1840 | dsl_valid_rename_arg_t *dvra = arg; |
eca7b760 | 1841 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; |
34dc7c2f | 1842 | |
13fe0198 MA |
1843 | dsl_dataset_name(ds, namebuf); |
1844 | ||
a7ed98d8 SD |
1845 | ASSERT3U(strnlen(namebuf, ZFS_MAX_DATASET_NAME_LEN), |
1846 | <, ZFS_MAX_DATASET_NAME_LEN); | |
1847 | int namelen = strlen(namebuf) + dvra->char_delta; | |
1848 | int depth = get_dataset_depth(namebuf) + dvra->nest_delta; | |
1849 | ||
1850 | if (namelen >= ZFS_MAX_DATASET_NAME_LEN) | |
1851 | return (SET_ERROR(ENAMETOOLONG)); | |
1852 | if (dvra->nest_delta > 0 && depth >= zfs_max_dataset_nesting) | |
2e528b49 | 1853 | return (SET_ERROR(ENAMETOOLONG)); |
13fe0198 MA |
1854 | return (0); |
1855 | } | |
1856 | ||
1857 | static int | |
1858 | dsl_dir_rename_check(void *arg, dmu_tx_t *tx) | |
1859 | { | |
1860 | dsl_dir_rename_arg_t *ddra = arg; | |
1861 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1862 | dsl_dir_t *dd, *newparent; | |
a7ed98d8 | 1863 | dsl_valid_rename_arg_t dvra; |
d8d418ff | 1864 | dsl_dataset_t *parentds; |
1865 | objset_t *parentos; | |
13fe0198 MA |
1866 | const char *mynewname; |
1867 | int error; | |
34dc7c2f | 1868 | |
13fe0198 MA |
1869 | /* target dir should exist */ |
1870 | error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); | |
1871 | if (error != 0) | |
1872 | return (error); | |
1873 | ||
1874 | /* new parent should exist */ | |
1875 | error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, | |
1876 | &newparent, &mynewname); | |
1877 | if (error != 0) { | |
1878 | dsl_dir_rele(dd, FTAG); | |
1879 | return (error); | |
1880 | } | |
1881 | ||
1882 | /* can't rename to different pool */ | |
1883 | if (dd->dd_pool != newparent->dd_pool) { | |
1884 | dsl_dir_rele(newparent, FTAG); | |
1885 | dsl_dir_rele(dd, FTAG); | |
9063f654 | 1886 | return (SET_ERROR(EXDEV)); |
13fe0198 MA |
1887 | } |
1888 | ||
1889 | /* new name should not already exist */ | |
1890 | if (mynewname == NULL) { | |
1891 | dsl_dir_rele(newparent, FTAG); | |
1892 | dsl_dir_rele(dd, FTAG); | |
2e528b49 | 1893 | return (SET_ERROR(EEXIST)); |
13fe0198 MA |
1894 | } |
1895 | ||
d8d418ff | 1896 | /* can't rename below anything but filesystems (eg. no ZVOLs) */ |
1897 | error = dsl_dataset_hold_obj(newparent->dd_pool, | |
1898 | dsl_dir_phys(newparent)->dd_head_dataset_obj, FTAG, &parentds); | |
1899 | if (error != 0) { | |
1900 | dsl_dir_rele(newparent, FTAG); | |
1901 | dsl_dir_rele(dd, FTAG); | |
1902 | return (error); | |
1903 | } | |
1904 | error = dmu_objset_from_ds(parentds, &parentos); | |
1905 | if (error != 0) { | |
1906 | dsl_dataset_rele(parentds, FTAG); | |
1907 | dsl_dir_rele(newparent, FTAG); | |
1908 | dsl_dir_rele(dd, FTAG); | |
1909 | return (error); | |
1910 | } | |
1911 | if (dmu_objset_type(parentos) != DMU_OST_ZFS) { | |
1912 | dsl_dataset_rele(parentds, FTAG); | |
1913 | dsl_dir_rele(newparent, FTAG); | |
1914 | dsl_dir_rele(dd, FTAG); | |
1915 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); | |
1916 | } | |
1917 | dsl_dataset_rele(parentds, FTAG); | |
1918 | ||
a7ed98d8 SD |
1919 | ASSERT3U(strnlen(ddra->ddra_newname, ZFS_MAX_DATASET_NAME_LEN), |
1920 | <, ZFS_MAX_DATASET_NAME_LEN); | |
1921 | ASSERT3U(strnlen(ddra->ddra_oldname, ZFS_MAX_DATASET_NAME_LEN), | |
1922 | <, ZFS_MAX_DATASET_NAME_LEN); | |
1923 | dvra.char_delta = strlen(ddra->ddra_newname) | |
1924 | - strlen(ddra->ddra_oldname); | |
1925 | dvra.nest_delta = get_dataset_depth(ddra->ddra_newname) | |
1926 | - get_dataset_depth(ddra->ddra_oldname); | |
1927 | ||
13fe0198 | 1928 | /* if the name length is growing, validate child name lengths */ |
a7ed98d8 | 1929 | if (dvra.char_delta > 0 || dvra.nest_delta > 0) { |
13fe0198 | 1930 | error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, |
a7ed98d8 | 1931 | &dvra, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); |
13fe0198 MA |
1932 | if (error != 0) { |
1933 | dsl_dir_rele(newparent, FTAG); | |
1934 | dsl_dir_rele(dd, FTAG); | |
1935 | return (error); | |
1936 | } | |
1937 | } | |
34dc7c2f | 1938 | |
788eb90c | 1939 | if (dmu_tx_is_syncing(tx)) { |
a0c9a17a | 1940 | if (spa_feature_is_active(dp->dp_spa, |
788eb90c JJ |
1941 | SPA_FEATURE_FS_SS_LIMIT)) { |
1942 | /* | |
1943 | * Although this is the check function and we don't | |
1944 | * normally make on-disk changes in check functions, | |
1945 | * we need to do that here. | |
1946 | * | |
1947 | * Ensure this portion of the tree's counts have been | |
1948 | * initialized in case the new parent has limits set. | |
1949 | */ | |
1950 | dsl_dir_init_fs_ss_count(dd, tx); | |
1951 | } | |
1952 | } | |
1953 | ||
13fe0198 | 1954 | if (newparent != dd->dd_parent) { |
34dc7c2f BB |
1955 | /* is there enough space? */ |
1956 | uint64_t myspace = | |
d683ddbb JG |
1957 | MAX(dsl_dir_phys(dd)->dd_used_bytes, |
1958 | dsl_dir_phys(dd)->dd_reserved); | |
788eb90c JJ |
1959 | objset_t *os = dd->dd_pool->dp_meta_objset; |
1960 | uint64_t fs_cnt = 0; | |
1961 | uint64_t ss_cnt = 0; | |
1962 | ||
1963 | if (dsl_dir_is_zapified(dd)) { | |
1964 | int err; | |
1965 | ||
1966 | err = zap_lookup(os, dd->dd_object, | |
1967 | DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, | |
1968 | &fs_cnt); | |
a0c9a17a JJ |
1969 | if (err != ENOENT && err != 0) { |
1970 | dsl_dir_rele(newparent, FTAG); | |
1971 | dsl_dir_rele(dd, FTAG); | |
788eb90c | 1972 | return (err); |
a0c9a17a | 1973 | } |
788eb90c JJ |
1974 | |
1975 | /* | |
1976 | * have to add 1 for the filesystem itself that we're | |
1977 | * moving | |
1978 | */ | |
1979 | fs_cnt++; | |
1980 | ||
1981 | err = zap_lookup(os, dd->dd_object, | |
1982 | DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, | |
1983 | &ss_cnt); | |
a0c9a17a JJ |
1984 | if (err != ENOENT && err != 0) { |
1985 | dsl_dir_rele(newparent, FTAG); | |
1986 | dsl_dir_rele(dd, FTAG); | |
788eb90c | 1987 | return (err); |
a0c9a17a | 1988 | } |
788eb90c | 1989 | } |
34dc7c2f | 1990 | |
b5256303 TC |
1991 | /* check for encryption errors */ |
1992 | error = dsl_dir_rename_crypt_check(dd, newparent); | |
1993 | if (error != 0) { | |
1994 | dsl_dir_rele(newparent, FTAG); | |
1995 | dsl_dir_rele(dd, FTAG); | |
1996 | return (SET_ERROR(EACCES)); | |
1997 | } | |
1998 | ||
34dc7c2f | 1999 | /* no rename into our descendant */ |
13fe0198 MA |
2000 | if (closest_common_ancestor(dd, newparent) == dd) { |
2001 | dsl_dir_rele(newparent, FTAG); | |
2002 | dsl_dir_rele(dd, FTAG); | |
2e528b49 | 2003 | return (SET_ERROR(EINVAL)); |
13fe0198 | 2004 | } |
34dc7c2f | 2005 | |
13fe0198 | 2006 | error = dsl_dir_transfer_possible(dd->dd_parent, |
788eb90c | 2007 | newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); |
13fe0198 MA |
2008 | if (error != 0) { |
2009 | dsl_dir_rele(newparent, FTAG); | |
2010 | dsl_dir_rele(dd, FTAG); | |
2011 | return (error); | |
2012 | } | |
34dc7c2f BB |
2013 | } |
2014 | ||
13fe0198 MA |
2015 | dsl_dir_rele(newparent, FTAG); |
2016 | dsl_dir_rele(dd, FTAG); | |
34dc7c2f BB |
2017 | return (0); |
2018 | } | |
2019 | ||
2020 | static void | |
13fe0198 | 2021 | dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 2022 | { |
13fe0198 MA |
2023 | dsl_dir_rename_arg_t *ddra = arg; |
2024 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2025 | dsl_dir_t *dd, *newparent; | |
2026 | const char *mynewname; | |
2027 | int error; | |
34dc7c2f | 2028 | objset_t *mos = dp->dp_meta_objset; |
34dc7c2f | 2029 | |
13fe0198 MA |
2030 | VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); |
2031 | VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, | |
2032 | &mynewname)); | |
34dc7c2f | 2033 | |
6f1ffb06 | 2034 | /* Log this before we change the name. */ |
6f1ffb06 | 2035 | spa_history_log_internal_dd(dd, "rename", tx, |
13fe0198 | 2036 | "-> %s", ddra->ddra_newname); |
6f1ffb06 | 2037 | |
13fe0198 | 2038 | if (newparent != dd->dd_parent) { |
788eb90c JJ |
2039 | objset_t *os = dd->dd_pool->dp_meta_objset; |
2040 | uint64_t fs_cnt = 0; | |
2041 | uint64_t ss_cnt = 0; | |
2042 | ||
2043 | /* | |
2044 | * We already made sure the dd counts were initialized in the | |
2045 | * check function. | |
2046 | */ | |
a0c9a17a | 2047 | if (spa_feature_is_active(dp->dp_spa, |
788eb90c JJ |
2048 | SPA_FEATURE_FS_SS_LIMIT)) { |
2049 | VERIFY0(zap_lookup(os, dd->dd_object, | |
2050 | DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, | |
2051 | &fs_cnt)); | |
2052 | /* add 1 for the filesystem itself that we're moving */ | |
2053 | fs_cnt++; | |
2054 | ||
2055 | VERIFY0(zap_lookup(os, dd->dd_object, | |
2056 | DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, | |
2057 | &ss_cnt)); | |
2058 | } | |
2059 | ||
2060 | dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, | |
2061 | DD_FIELD_FILESYSTEM_COUNT, tx); | |
2062 | dsl_fs_ss_count_adjust(newparent, fs_cnt, | |
2063 | DD_FIELD_FILESYSTEM_COUNT, tx); | |
2064 | ||
2065 | dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, | |
2066 | DD_FIELD_SNAPSHOT_COUNT, tx); | |
2067 | dsl_fs_ss_count_adjust(newparent, ss_cnt, | |
2068 | DD_FIELD_SNAPSHOT_COUNT, tx); | |
2069 | ||
b128c09f | 2070 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, |
d683ddbb JG |
2071 | -dsl_dir_phys(dd)->dd_used_bytes, |
2072 | -dsl_dir_phys(dd)->dd_compressed_bytes, | |
2073 | -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); | |
13fe0198 | 2074 | dsl_dir_diduse_space(newparent, DD_USED_CHILD, |
d683ddbb JG |
2075 | dsl_dir_phys(dd)->dd_used_bytes, |
2076 | dsl_dir_phys(dd)->dd_compressed_bytes, | |
2077 | dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); | |
b128c09f | 2078 | |
d683ddbb JG |
2079 | if (dsl_dir_phys(dd)->dd_reserved > |
2080 | dsl_dir_phys(dd)->dd_used_bytes) { | |
2081 | uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved - | |
2082 | dsl_dir_phys(dd)->dd_used_bytes; | |
b128c09f BB |
2083 | |
2084 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, | |
2085 | -unused_rsrv, 0, 0, tx); | |
13fe0198 | 2086 | dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, |
b128c09f BB |
2087 | unused_rsrv, 0, 0, tx); |
2088 | } | |
34dc7c2f BB |
2089 | } |
2090 | ||
2091 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
2092 | ||
2093 | /* remove from old parent zapobj */ | |
d683ddbb JG |
2094 | error = zap_remove(mos, |
2095 | dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj, | |
34dc7c2f | 2096 | dd->dd_myname, tx); |
13fe0198 | 2097 | ASSERT0(error); |
34dc7c2f | 2098 | |
c9d61adb | 2099 | (void) strlcpy(dd->dd_myname, mynewname, |
2100 | sizeof (dd->dd_myname)); | |
13fe0198 | 2101 | dsl_dir_rele(dd->dd_parent, dd); |
d683ddbb | 2102 | dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object; |
13fe0198 MA |
2103 | VERIFY0(dsl_dir_hold_obj(dp, |
2104 | newparent->dd_object, NULL, dd, &dd->dd_parent)); | |
34dc7c2f BB |
2105 | |
2106 | /* add to new parent zapobj */ | |
d683ddbb | 2107 | VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj, |
13fe0198 MA |
2108 | dd->dd_myname, 8, 1, &dd->dd_object, tx)); |
2109 | ||
a0bd735a BP |
2110 | zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname, |
2111 | ddra->ddra_newname, B_TRUE); | |
ba6a2402 | 2112 | |
13fe0198 | 2113 | dsl_prop_notify_all(dd); |
34dc7c2f | 2114 | |
13fe0198 MA |
2115 | dsl_dir_rele(newparent, FTAG); |
2116 | dsl_dir_rele(dd, FTAG); | |
34dc7c2f BB |
2117 | } |
2118 | ||
2119 | int | |
13fe0198 | 2120 | dsl_dir_rename(const char *oldname, const char *newname) |
34dc7c2f | 2121 | { |
13fe0198 | 2122 | dsl_dir_rename_arg_t ddra; |
34dc7c2f | 2123 | |
13fe0198 MA |
2124 | ddra.ddra_oldname = oldname; |
2125 | ddra.ddra_newname = newname; | |
788eb90c | 2126 | ddra.ddra_cred = CRED(); |
34dc7c2f | 2127 | |
13fe0198 | 2128 | return (dsl_sync_task(oldname, |
3d45fdd6 MA |
2129 | dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, |
2130 | 3, ZFS_SPACE_CHECK_RESERVED)); | |
34dc7c2f BB |
2131 | } |
2132 | ||
2133 | int | |
788eb90c JJ |
2134 | dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, |
2135 | uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) | |
34dc7c2f BB |
2136 | { |
2137 | dsl_dir_t *ancestor; | |
2138 | int64_t adelta; | |
2139 | uint64_t avail; | |
788eb90c | 2140 | int err; |
34dc7c2f BB |
2141 | |
2142 | ancestor = closest_common_ancestor(sdd, tdd); | |
2143 | adelta = would_change(sdd, -space, ancestor); | |
2144 | avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); | |
2145 | if (avail < space) | |
2e528b49 | 2146 | return (SET_ERROR(ENOSPC)); |
34dc7c2f | 2147 | |
788eb90c JJ |
2148 | err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, |
2149 | ancestor, cr); | |
2150 | if (err != 0) | |
2151 | return (err); | |
2152 | err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, | |
2153 | ancestor, cr); | |
2154 | if (err != 0) | |
2155 | return (err); | |
2156 | ||
34dc7c2f BB |
2157 | return (0); |
2158 | } | |
428870ff | 2159 | |
6413c95f | 2160 | inode_timespec_t |
428870ff BB |
2161 | dsl_dir_snap_cmtime(dsl_dir_t *dd) |
2162 | { | |
6413c95f | 2163 | inode_timespec_t t; |
428870ff BB |
2164 | |
2165 | mutex_enter(&dd->dd_lock); | |
2166 | t = dd->dd_snap_cmtime; | |
2167 | mutex_exit(&dd->dd_lock); | |
2168 | ||
2169 | return (t); | |
2170 | } | |
2171 | ||
2172 | void | |
2173 | dsl_dir_snap_cmtime_update(dsl_dir_t *dd) | |
2174 | { | |
6413c95f | 2175 | inode_timespec_t t; |
428870ff BB |
2176 | |
2177 | gethrestime(&t); | |
2178 | mutex_enter(&dd->dd_lock); | |
2179 | dd->dd_snap_cmtime = t; | |
2180 | mutex_exit(&dd->dd_lock); | |
2181 | } | |
c28b2279 | 2182 | |
fa86b5db MA |
2183 | void |
2184 | dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) | |
2185 | { | |
2186 | objset_t *mos = dd->dd_pool->dp_meta_objset; | |
2187 | dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); | |
2188 | } | |
2189 | ||
788eb90c JJ |
2190 | boolean_t |
2191 | dsl_dir_is_zapified(dsl_dir_t *dd) | |
2192 | { | |
2193 | dmu_object_info_t doi; | |
2194 | ||
2195 | dmu_object_info_from_db(dd->dd_dbuf, &doi); | |
2196 | return (doi.doi_type == DMU_OTN_ZAP_METADATA); | |
2197 | } | |
2198 | ||
37f03da8 SH |
2199 | void |
2200 | dsl_dir_livelist_open(dsl_dir_t *dd, uint64_t obj) | |
2201 | { | |
2202 | objset_t *mos = dd->dd_pool->dp_meta_objset; | |
2203 | ASSERT(spa_feature_is_active(dd->dd_pool->dp_spa, | |
2204 | SPA_FEATURE_LIVELIST)); | |
2205 | dsl_deadlist_open(&dd->dd_livelist, mos, obj); | |
2206 | bplist_create(&dd->dd_pending_allocs); | |
2207 | bplist_create(&dd->dd_pending_frees); | |
2208 | } | |
2209 | ||
2210 | void | |
2211 | dsl_dir_livelist_close(dsl_dir_t *dd) | |
2212 | { | |
2213 | dsl_deadlist_close(&dd->dd_livelist); | |
2214 | bplist_destroy(&dd->dd_pending_allocs); | |
2215 | bplist_destroy(&dd->dd_pending_frees); | |
2216 | } | |
2217 | ||
2218 | void | |
2219 | dsl_dir_remove_livelist(dsl_dir_t *dd, dmu_tx_t *tx, boolean_t total) | |
2220 | { | |
2221 | uint64_t obj; | |
2222 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2223 | spa_t *spa = dp->dp_spa; | |
2224 | livelist_condense_entry_t to_condense = spa->spa_to_condense; | |
2225 | ||
2226 | if (!dsl_deadlist_is_open(&dd->dd_livelist)) | |
2227 | return; | |
2228 | ||
2229 | /* | |
2230 | * If the livelist being removed is set to be condensed, stop the | |
2231 | * condense zthr and indicate the cancellation in the spa_to_condense | |
2232 | * struct in case the condense no-wait synctask has already started | |
2233 | */ | |
2234 | zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr; | |
2235 | if (ll_condense_thread != NULL && | |
2236 | (to_condense.ds != NULL) && (to_condense.ds->ds_dir == dd)) { | |
2237 | /* | |
2238 | * We use zthr_wait_cycle_done instead of zthr_cancel | |
2239 | * because we don't want to destroy the zthr, just have | |
2240 | * it skip its current task. | |
2241 | */ | |
2242 | spa->spa_to_condense.cancelled = B_TRUE; | |
2243 | zthr_wait_cycle_done(ll_condense_thread); | |
2244 | /* | |
2245 | * If we've returned from zthr_wait_cycle_done without | |
2246 | * clearing the to_condense data structure it's either | |
2247 | * because the no-wait synctask has started (which is | |
2248 | * indicated by 'syncing' field of to_condense) and we | |
2249 | * can expect it to clear to_condense on its own. | |
2250 | * Otherwise, we returned before the zthr ran. The | |
2251 | * checkfunc will now fail as cancelled == B_TRUE so we | |
2252 | * can safely NULL out ds, allowing a different dir's | |
2253 | * livelist to be condensed. | |
2254 | * | |
2255 | * We can be sure that the to_condense struct will not | |
2256 | * be repopulated at this stage because both this | |
2257 | * function and dsl_livelist_try_condense execute in | |
2258 | * syncing context. | |
2259 | */ | |
2260 | if ((spa->spa_to_condense.ds != NULL) && | |
2261 | !spa->spa_to_condense.syncing) { | |
2262 | dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, | |
2263 | spa); | |
2264 | spa->spa_to_condense.ds = NULL; | |
2265 | } | |
2266 | } | |
2267 | ||
2268 | dsl_dir_livelist_close(dd); | |
2269 | int err = zap_lookup(dp->dp_meta_objset, dd->dd_object, | |
2270 | DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &obj); | |
2271 | if (err == 0) { | |
2272 | VERIFY0(zap_remove(dp->dp_meta_objset, dd->dd_object, | |
2273 | DD_FIELD_LIVELIST, tx)); | |
2274 | if (total) { | |
2275 | dsl_deadlist_free(dp->dp_meta_objset, obj, tx); | |
2276 | spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx); | |
2277 | } | |
2278 | } else { | |
2279 | ASSERT3U(err, !=, ENOENT); | |
2280 | } | |
2281 | } | |
2282 | ||
93ce2b4c | 2283 | #if defined(_KERNEL) |
c28b2279 BB |
2284 | EXPORT_SYMBOL(dsl_dir_set_quota); |
2285 | EXPORT_SYMBOL(dsl_dir_set_reservation); | |
c28b2279 | 2286 | #endif |