]>
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. |
d2734cce | 23 | * Copyright (c) 2012, 2017 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. |
34dc7c2f BB |
28 | */ |
29 | ||
34dc7c2f BB |
30 | #include <sys/dmu.h> |
31 | #include <sys/dmu_objset.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/dsl_dataset.h> | |
34 | #include <sys/dsl_dir.h> | |
35 | #include <sys/dsl_prop.h> | |
36 | #include <sys/dsl_synctask.h> | |
37 | #include <sys/dsl_deleg.h> | |
fa86b5db | 38 | #include <sys/dmu_impl.h> |
34dc7c2f | 39 | #include <sys/spa.h> |
ae76f45c | 40 | #include <sys/spa_impl.h> |
428870ff | 41 | #include <sys/metaslab.h> |
34dc7c2f BB |
42 | #include <sys/zap.h> |
43 | #include <sys/zio.h> | |
44 | #include <sys/arc.h> | |
45 | #include <sys/sunddi.h> | |
788eb90c JJ |
46 | #include <sys/zfeature.h> |
47 | #include <sys/policy.h> | |
48 | #include <sys/zfs_znode.h> | |
ba6a2402 | 49 | #include <sys/zvol.h> |
34dc7c2f | 50 | #include "zfs_namecheck.h" |
788eb90c JJ |
51 | #include "zfs_prop.h" |
52 | ||
53 | /* | |
54 | * Filesystem and Snapshot Limits | |
55 | * ------------------------------ | |
56 | * | |
57 | * These limits are used to restrict the number of filesystems and/or snapshots | |
58 | * that can be created at a given level in the tree or below. A typical | |
59 | * use-case is with a delegated dataset where the administrator wants to ensure | |
60 | * that a user within the zone is not creating too many additional filesystems | |
61 | * or snapshots, even though they're not exceeding their space quota. | |
62 | * | |
63 | * The filesystem and snapshot counts are stored as extensible properties. This | |
64 | * capability is controlled by a feature flag and must be enabled to be used. | |
65 | * Once enabled, the feature is not active until the first limit is set. At | |
66 | * that point, future operations to create/destroy filesystems or snapshots | |
67 | * will validate and update the counts. | |
68 | * | |
69 | * Because the count properties will not exist before the feature is active, | |
70 | * the counts are updated when a limit is first set on an uninitialized | |
71 | * dsl_dir node in the tree (The filesystem/snapshot count on a node includes | |
72 | * all of the nested filesystems/snapshots. Thus, a new leaf node has a | |
73 | * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and | |
74 | * snapshot count properties on a node indicate uninitialized counts on that | |
75 | * node.) When first setting a limit on an uninitialized node, the code starts | |
76 | * at the filesystem with the new limit and descends into all sub-filesystems | |
77 | * to add the count properties. | |
78 | * | |
79 | * In practice this is lightweight since a limit is typically set when the | |
80 | * filesystem is created and thus has no children. Once valid, changing the | |
81 | * limit value won't require a re-traversal since the counts are already valid. | |
82 | * When recursively fixing the counts, if a node with a limit is encountered | |
83 | * during the descent, the counts are known to be valid and there is no need to | |
84 | * descend into that filesystem's children. The counts on filesystems above the | |
85 | * one with the new limit will still be uninitialized, unless a limit is | |
86 | * eventually set on one of those filesystems. The counts are always recursively | |
87 | * updated when a limit is set on a dataset, unless there is already a limit. | |
88 | * When a new limit value is set on a filesystem with an existing limit, it is | |
89 | * possible for the new limit to be less than the current count at that level | |
90 | * since a user who can change the limit is also allowed to exceed the limit. | |
91 | * | |
92 | * Once the feature is active, then whenever a filesystem or snapshot is | |
93 | * created, the code recurses up the tree, validating the new count against the | |
94 | * limit at each initialized level. In practice, most levels will not have a | |
95 | * limit set. If there is a limit at any initialized level up the tree, the | |
96 | * check must pass or the creation will fail. Likewise, when a filesystem or | |
97 | * snapshot is destroyed, the counts are recursively adjusted all the way up | |
98 | * the initizized nodes in the tree. Renaming a filesystem into different point | |
99 | * in the tree will first validate, then update the counts on each branch up to | |
100 | * the common ancestor. A receive will also validate the counts and then update | |
101 | * them. | |
102 | * | |
103 | * An exception to the above behavior is that the limit is not enforced if the | |
104 | * user has permission to modify the limit. This is primarily so that | |
105 | * recursive snapshots in the global zone always work. We want to prevent a | |
106 | * denial-of-service in which a lower level delegated dataset could max out its | |
107 | * limit and thus block recursive snapshots from being taken in the global zone. | |
108 | * Because of this, it is possible for the snapshot count to be over the limit | |
109 | * and snapshots taken in the global zone could cause a lower level dataset to | |
110 | * hit or exceed its limit. The administrator taking the global zone recursive | |
111 | * snapshot should be aware of this side-effect and behave accordingly. | |
112 | * For consistency, the filesystem limit is also not enforced if the user can | |
113 | * modify the limit. | |
114 | * | |
115 | * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check() | |
116 | * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in | |
117 | * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by | |
118 | * dsl_dir_init_fs_ss_count(). | |
119 | * | |
120 | * There is a special case when we receive a filesystem that already exists. In | |
121 | * this case a temporary clone name of %X is created (see dmu_recv_begin). We | |
122 | * never update the filesystem counts for temporary clones. | |
123 | * | |
124 | * Likewise, we do not update the snapshot counts for temporary snapshots, | |
125 | * such as those created by zfs diff. | |
126 | */ | |
34dc7c2f | 127 | |
d683ddbb JG |
128 | extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd); |
129 | ||
34dc7c2f | 130 | static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); |
34dc7c2f | 131 | |
a1d477c2 MA |
132 | typedef struct ddulrt_arg { |
133 | dsl_dir_t *ddulrta_dd; | |
134 | uint64_t ddlrta_txg; | |
135 | } ddulrt_arg_t; | |
136 | ||
34dc7c2f | 137 | static void |
39efbde7 | 138 | dsl_dir_evict_async(void *dbu) |
34dc7c2f | 139 | { |
0c66c32d | 140 | dsl_dir_t *dd = dbu; |
34dc7c2f | 141 | int t; |
d1d7e268 | 142 | ASSERTV(dsl_pool_t *dp = dd->dd_pool); |
34dc7c2f | 143 | |
0c66c32d JG |
144 | dd->dd_dbuf = NULL; |
145 | ||
34dc7c2f BB |
146 | for (t = 0; t < TXG_SIZE; t++) { |
147 | ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); | |
148 | ASSERT(dd->dd_tempreserved[t] == 0); | |
149 | ASSERT(dd->dd_space_towrite[t] == 0); | |
150 | } | |
151 | ||
34dc7c2f | 152 | if (dd->dd_parent) |
0c66c32d | 153 | dsl_dir_async_rele(dd->dd_parent, dd); |
34dc7c2f | 154 | |
0c66c32d | 155 | spa_async_close(dd->dd_pool->dp_spa, dd); |
34dc7c2f | 156 | |
0eb21616 | 157 | dsl_prop_fini(dd); |
34dc7c2f BB |
158 | mutex_destroy(&dd->dd_lock); |
159 | kmem_free(dd, sizeof (dsl_dir_t)); | |
160 | } | |
161 | ||
162 | int | |
13fe0198 | 163 | dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, |
34dc7c2f BB |
164 | const char *tail, void *tag, dsl_dir_t **ddp) |
165 | { | |
166 | dmu_buf_t *dbuf; | |
167 | dsl_dir_t *dd; | |
b5256303 | 168 | dmu_object_info_t doi; |
34dc7c2f BB |
169 | int err; |
170 | ||
13fe0198 | 171 | ASSERT(dsl_pool_config_held(dp)); |
34dc7c2f BB |
172 | |
173 | err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); | |
13fe0198 | 174 | if (err != 0) |
34dc7c2f BB |
175 | return (err); |
176 | dd = dmu_buf_get_user(dbuf); | |
b5256303 TC |
177 | |
178 | dmu_object_info_from_db(dbuf, &doi); | |
179 | ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); | |
180 | ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); | |
181 | ||
34dc7c2f BB |
182 | if (dd == NULL) { |
183 | dsl_dir_t *winner; | |
34dc7c2f | 184 | |
79c76d5b | 185 | dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); |
34dc7c2f BB |
186 | dd->dd_object = ddobj; |
187 | dd->dd_dbuf = dbuf; | |
188 | dd->dd_pool = dp; | |
b5256303 TC |
189 | |
190 | if (dsl_dir_is_zapified(dd) && | |
191 | zap_contains(dp->dp_meta_objset, ddobj, | |
192 | DD_FIELD_CRYPTO_KEY_OBJ) == 0) { | |
193 | VERIFY0(zap_lookup(dp->dp_meta_objset, | |
194 | ddobj, DD_FIELD_CRYPTO_KEY_OBJ, | |
195 | sizeof (uint64_t), 1, &dd->dd_crypto_obj)); | |
ae76f45c TC |
196 | |
197 | /* check for on-disk format errata */ | |
198 | if (dsl_dir_incompatible_encryption_version(dd)) { | |
199 | dp->dp_spa->spa_errata = | |
200 | ZPOOL_ERRATA_ZOL_6845_ENCRYPTION; | |
201 | } | |
b5256303 TC |
202 | } |
203 | ||
34dc7c2f | 204 | mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); |
0eb21616 | 205 | dsl_prop_init(dd); |
34dc7c2f | 206 | |
428870ff BB |
207 | dsl_dir_snap_cmtime_update(dd); |
208 | ||
d683ddbb JG |
209 | if (dsl_dir_phys(dd)->dd_parent_obj) { |
210 | err = dsl_dir_hold_obj(dp, | |
211 | dsl_dir_phys(dd)->dd_parent_obj, NULL, dd, | |
212 | &dd->dd_parent); | |
13fe0198 | 213 | if (err != 0) |
b128c09f | 214 | goto errout; |
34dc7c2f BB |
215 | if (tail) { |
216 | #ifdef ZFS_DEBUG | |
217 | uint64_t foundobj; | |
218 | ||
219 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb JG |
220 | dsl_dir_phys(dd->dd_parent)-> |
221 | dd_child_dir_zapobj, tail, | |
222 | sizeof (foundobj), 1, &foundobj); | |
34dc7c2f BB |
223 | ASSERT(err || foundobj == ddobj); |
224 | #endif | |
680eada9 | 225 | (void) strlcpy(dd->dd_myname, tail, |
226 | sizeof (dd->dd_myname)); | |
34dc7c2f BB |
227 | } else { |
228 | err = zap_value_search(dp->dp_meta_objset, | |
d683ddbb JG |
229 | dsl_dir_phys(dd->dd_parent)-> |
230 | dd_child_dir_zapobj, | |
34dc7c2f BB |
231 | ddobj, 0, dd->dd_myname); |
232 | } | |
13fe0198 | 233 | if (err != 0) |
b128c09f | 234 | goto errout; |
34dc7c2f BB |
235 | } else { |
236 | (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); | |
237 | } | |
238 | ||
428870ff BB |
239 | if (dsl_dir_is_clone(dd)) { |
240 | dmu_buf_t *origin_bonus; | |
241 | dsl_dataset_phys_t *origin_phys; | |
242 | ||
243 | /* | |
244 | * We can't open the origin dataset, because | |
245 | * that would require opening this dsl_dir. | |
246 | * Just look at its phys directly instead. | |
247 | */ | |
248 | err = dmu_bonus_hold(dp->dp_meta_objset, | |
d683ddbb JG |
249 | dsl_dir_phys(dd)->dd_origin_obj, FTAG, |
250 | &origin_bonus); | |
13fe0198 | 251 | if (err != 0) |
428870ff BB |
252 | goto errout; |
253 | origin_phys = origin_bonus->db_data; | |
254 | dd->dd_origin_txg = | |
255 | origin_phys->ds_creation_txg; | |
256 | dmu_buf_rele(origin_bonus, FTAG); | |
257 | } | |
258 | ||
39efbde7 GM |
259 | dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async, |
260 | &dd->dd_dbuf); | |
0c66c32d JG |
261 | winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu); |
262 | if (winner != NULL) { | |
34dc7c2f | 263 | if (dd->dd_parent) |
13fe0198 | 264 | dsl_dir_rele(dd->dd_parent, dd); |
0eb21616 | 265 | dsl_prop_fini(dd); |
34dc7c2f BB |
266 | mutex_destroy(&dd->dd_lock); |
267 | kmem_free(dd, sizeof (dsl_dir_t)); | |
268 | dd = winner; | |
269 | } else { | |
270 | spa_open_ref(dp->dp_spa, dd); | |
271 | } | |
272 | } | |
273 | ||
274 | /* | |
275 | * The dsl_dir_t has both open-to-close and instantiate-to-evict | |
276 | * holds on the spa. We need the open-to-close holds because | |
277 | * otherwise the spa_refcnt wouldn't change when we open a | |
278 | * dir which the spa also has open, so we could incorrectly | |
279 | * think it was OK to unload/export/destroy the pool. We need | |
280 | * the instantiate-to-evict hold because the dsl_dir_t has a | |
281 | * pointer to the dd_pool, which has a pointer to the spa_t. | |
282 | */ | |
283 | spa_open_ref(dp->dp_spa, tag); | |
284 | ASSERT3P(dd->dd_pool, ==, dp); | |
285 | ASSERT3U(dd->dd_object, ==, ddobj); | |
286 | ASSERT3P(dd->dd_dbuf, ==, dbuf); | |
287 | *ddp = dd; | |
288 | return (0); | |
b128c09f BB |
289 | |
290 | errout: | |
291 | if (dd->dd_parent) | |
13fe0198 | 292 | dsl_dir_rele(dd->dd_parent, dd); |
0eb21616 | 293 | dsl_prop_fini(dd); |
b128c09f BB |
294 | mutex_destroy(&dd->dd_lock); |
295 | kmem_free(dd, sizeof (dsl_dir_t)); | |
296 | dmu_buf_rele(dbuf, tag); | |
297 | return (err); | |
34dc7c2f BB |
298 | } |
299 | ||
300 | void | |
13fe0198 | 301 | dsl_dir_rele(dsl_dir_t *dd, void *tag) |
34dc7c2f BB |
302 | { |
303 | dprintf_dd(dd, "%s\n", ""); | |
304 | spa_close(dd->dd_pool->dp_spa, tag); | |
305 | dmu_buf_rele(dd->dd_dbuf, tag); | |
306 | } | |
307 | ||
0c66c32d JG |
308 | /* |
309 | * Remove a reference to the given dsl dir that is being asynchronously | |
310 | * released. Async releases occur from a taskq performing eviction of | |
311 | * dsl datasets and dirs. This process is identical to a normal release | |
312 | * with the exception of using the async API for releasing the reference on | |
313 | * the spa. | |
314 | */ | |
315 | void | |
316 | dsl_dir_async_rele(dsl_dir_t *dd, void *tag) | |
317 | { | |
318 | dprintf_dd(dd, "%s\n", ""); | |
319 | spa_async_close(dd->dd_pool->dp_spa, tag); | |
320 | dmu_buf_rele(dd->dd_dbuf, tag); | |
321 | } | |
322 | ||
eca7b760 | 323 | /* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */ |
34dc7c2f BB |
324 | void |
325 | dsl_dir_name(dsl_dir_t *dd, char *buf) | |
326 | { | |
327 | if (dd->dd_parent) { | |
328 | dsl_dir_name(dd->dd_parent, buf); | |
eca7b760 IK |
329 | VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <, |
330 | ZFS_MAX_DATASET_NAME_LEN); | |
34dc7c2f BB |
331 | } else { |
332 | buf[0] = '\0'; | |
333 | } | |
334 | if (!MUTEX_HELD(&dd->dd_lock)) { | |
335 | /* | |
336 | * recursive mutex so that we can use | |
337 | * dprintf_dd() with dd_lock held | |
338 | */ | |
339 | mutex_enter(&dd->dd_lock); | |
eca7b760 IK |
340 | VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), |
341 | <, ZFS_MAX_DATASET_NAME_LEN); | |
34dc7c2f BB |
342 | mutex_exit(&dd->dd_lock); |
343 | } else { | |
eca7b760 IK |
344 | VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN), |
345 | <, ZFS_MAX_DATASET_NAME_LEN); | |
34dc7c2f BB |
346 | } |
347 | } | |
348 | ||
29809a6c | 349 | /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ |
34dc7c2f BB |
350 | int |
351 | dsl_dir_namelen(dsl_dir_t *dd) | |
352 | { | |
353 | int result = 0; | |
354 | ||
355 | if (dd->dd_parent) { | |
356 | /* parent's name + 1 for the "/" */ | |
357 | result = dsl_dir_namelen(dd->dd_parent) + 1; | |
358 | } | |
359 | ||
360 | if (!MUTEX_HELD(&dd->dd_lock)) { | |
361 | /* see dsl_dir_name */ | |
362 | mutex_enter(&dd->dd_lock); | |
363 | result += strlen(dd->dd_myname); | |
364 | mutex_exit(&dd->dd_lock); | |
365 | } else { | |
366 | result += strlen(dd->dd_myname); | |
367 | } | |
368 | ||
369 | return (result); | |
370 | } | |
371 | ||
34dc7c2f BB |
372 | static int |
373 | getcomponent(const char *path, char *component, const char **nextp) | |
374 | { | |
375 | char *p; | |
13fe0198 | 376 | |
9babb374 | 377 | if ((path == NULL) || (path[0] == '\0')) |
2e528b49 | 378 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
379 | /* This would be a good place to reserve some namespace... */ |
380 | p = strpbrk(path, "/@"); | |
381 | if (p && (p[1] == '/' || p[1] == '@')) { | |
382 | /* two separators in a row */ | |
2e528b49 | 383 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
384 | } |
385 | if (p == NULL || p == path) { | |
386 | /* | |
387 | * if the first thing is an @ or /, it had better be an | |
388 | * @ and it had better not have any more ats or slashes, | |
389 | * and it had better have something after the @. | |
390 | */ | |
391 | if (p != NULL && | |
392 | (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) | |
2e528b49 | 393 | return (SET_ERROR(EINVAL)); |
eca7b760 | 394 | if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 395 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
396 | (void) strcpy(component, path); |
397 | p = NULL; | |
398 | } else if (p[0] == '/') { | |
eca7b760 | 399 | if (p - path >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 400 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f | 401 | (void) strncpy(component, path, p - path); |
13fe0198 | 402 | component[p - path] = '\0'; |
34dc7c2f BB |
403 | p++; |
404 | } else if (p[0] == '@') { | |
405 | /* | |
406 | * if the next separator is an @, there better not be | |
407 | * any more slashes. | |
408 | */ | |
409 | if (strchr(path, '/')) | |
2e528b49 | 410 | return (SET_ERROR(EINVAL)); |
eca7b760 | 411 | if (p - path >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 412 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f | 413 | (void) strncpy(component, path, p - path); |
13fe0198 | 414 | component[p - path] = '\0'; |
34dc7c2f | 415 | } else { |
13fe0198 | 416 | panic("invalid p=%p", (void *)p); |
34dc7c2f BB |
417 | } |
418 | *nextp = p; | |
419 | return (0); | |
420 | } | |
421 | ||
422 | /* | |
13fe0198 MA |
423 | * Return the dsl_dir_t, and possibly the last component which couldn't |
424 | * be found in *tail. The name must be in the specified dsl_pool_t. This | |
425 | * thread must hold the dp_config_rwlock for the pool. Returns NULL if the | |
426 | * path is bogus, or if tail==NULL and we couldn't parse the whole name. | |
427 | * (*tail)[0] == '@' means that the last component is a snapshot. | |
34dc7c2f BB |
428 | */ |
429 | int | |
13fe0198 | 430 | dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, |
34dc7c2f BB |
431 | dsl_dir_t **ddp, const char **tailp) |
432 | { | |
fcf37ec6 | 433 | char *buf; |
13fe0198 | 434 | const char *spaname, *next, *nextnext = NULL; |
34dc7c2f BB |
435 | int err; |
436 | dsl_dir_t *dd; | |
34dc7c2f | 437 | uint64_t ddobj; |
34dc7c2f | 438 | |
eca7b760 | 439 | buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
34dc7c2f | 440 | err = getcomponent(name, buf, &next); |
13fe0198 | 441 | if (err != 0) |
fcf37ec6 | 442 | goto error; |
34dc7c2f | 443 | |
13fe0198 MA |
444 | /* Make sure the name is in the specified pool. */ |
445 | spaname = spa_name(dp->dp_spa); | |
446 | if (strcmp(buf, spaname) != 0) { | |
9063f654 | 447 | err = SET_ERROR(EXDEV); |
13fe0198 | 448 | goto error; |
34dc7c2f BB |
449 | } |
450 | ||
13fe0198 | 451 | ASSERT(dsl_pool_config_held(dp)); |
34dc7c2f | 452 | |
13fe0198 MA |
453 | err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); |
454 | if (err != 0) { | |
fcf37ec6 | 455 | goto error; |
34dc7c2f BB |
456 | } |
457 | ||
458 | while (next != NULL) { | |
0c66c32d | 459 | dsl_dir_t *child_dd; |
34dc7c2f | 460 | err = getcomponent(next, buf, &nextnext); |
13fe0198 | 461 | if (err != 0) |
34dc7c2f BB |
462 | break; |
463 | ASSERT(next[0] != '\0'); | |
464 | if (next[0] == '@') | |
465 | break; | |
466 | dprintf("looking up %s in obj%lld\n", | |
d683ddbb | 467 | buf, dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
468 | |
469 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb | 470 | dsl_dir_phys(dd)->dd_child_dir_zapobj, |
34dc7c2f | 471 | buf, sizeof (ddobj), 1, &ddobj); |
13fe0198 | 472 | if (err != 0) { |
34dc7c2f BB |
473 | if (err == ENOENT) |
474 | err = 0; | |
475 | break; | |
476 | } | |
477 | ||
0c66c32d | 478 | err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd); |
13fe0198 | 479 | if (err != 0) |
34dc7c2f | 480 | break; |
13fe0198 | 481 | dsl_dir_rele(dd, tag); |
0c66c32d | 482 | dd = child_dd; |
34dc7c2f BB |
483 | next = nextnext; |
484 | } | |
34dc7c2f | 485 | |
13fe0198 MA |
486 | if (err != 0) { |
487 | dsl_dir_rele(dd, tag); | |
fcf37ec6 | 488 | goto error; |
34dc7c2f BB |
489 | } |
490 | ||
491 | /* | |
492 | * It's an error if there's more than one component left, or | |
493 | * tailp==NULL and there's any component left. | |
494 | */ | |
495 | if (next != NULL && | |
496 | (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { | |
497 | /* bad path name */ | |
13fe0198 | 498 | dsl_dir_rele(dd, tag); |
34dc7c2f | 499 | dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); |
2e528b49 | 500 | err = SET_ERROR(ENOENT); |
34dc7c2f | 501 | } |
13fe0198 | 502 | if (tailp != NULL) |
34dc7c2f | 503 | *tailp = next; |
fc1ecd16 DB |
504 | if (err == 0) |
505 | *ddp = dd; | |
fcf37ec6 | 506 | error: |
eca7b760 | 507 | kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN); |
34dc7c2f BB |
508 | return (err); |
509 | } | |
510 | ||
788eb90c JJ |
511 | /* |
512 | * If the counts are already initialized for this filesystem and its | |
513 | * descendants then do nothing, otherwise initialize the counts. | |
514 | * | |
515 | * The counts on this filesystem, and those below, may be uninitialized due to | |
516 | * either the use of a pre-existing pool which did not support the | |
517 | * filesystem/snapshot limit feature, or one in which the feature had not yet | |
518 | * been enabled. | |
519 | * | |
520 | * Recursively descend the filesystem tree and update the filesystem/snapshot | |
521 | * counts on each filesystem below, then update the cumulative count on the | |
522 | * current filesystem. If the filesystem already has a count set on it, | |
523 | * then we know that its counts, and the counts on the filesystems below it, | |
524 | * are already correct, so we don't have to update this filesystem. | |
525 | */ | |
526 | static void | |
527 | dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx) | |
528 | { | |
529 | uint64_t my_fs_cnt = 0; | |
530 | uint64_t my_ss_cnt = 0; | |
531 | dsl_pool_t *dp = dd->dd_pool; | |
532 | objset_t *os = dp->dp_meta_objset; | |
533 | zap_cursor_t *zc; | |
534 | zap_attribute_t *za; | |
535 | dsl_dataset_t *ds; | |
536 | ||
a0c9a17a | 537 | ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)); |
788eb90c JJ |
538 | ASSERT(dsl_pool_config_held(dp)); |
539 | ASSERT(dmu_tx_is_syncing(tx)); | |
540 | ||
541 | dsl_dir_zapify(dd, tx); | |
542 | ||
543 | /* | |
544 | * If the filesystem count has already been initialized then we | |
545 | * don't need to recurse down any further. | |
546 | */ | |
547 | if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0) | |
548 | return; | |
549 | ||
550 | zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); | |
551 | za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); | |
552 | ||
553 | /* Iterate my child dirs */ | |
d683ddbb | 554 | for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj); |
788eb90c JJ |
555 | zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) { |
556 | dsl_dir_t *chld_dd; | |
557 | uint64_t count; | |
558 | ||
559 | VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG, | |
560 | &chld_dd)); | |
561 | ||
562 | /* | |
563 | * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and | |
564 | * temporary datasets. | |
565 | */ | |
566 | if (chld_dd->dd_myname[0] == '$' || | |
567 | chld_dd->dd_myname[0] == '%') { | |
568 | dsl_dir_rele(chld_dd, FTAG); | |
569 | continue; | |
570 | } | |
571 | ||
572 | my_fs_cnt++; /* count this child */ | |
573 | ||
574 | dsl_dir_init_fs_ss_count(chld_dd, tx); | |
575 | ||
576 | VERIFY0(zap_lookup(os, chld_dd->dd_object, | |
577 | DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count)); | |
578 | my_fs_cnt += count; | |
579 | VERIFY0(zap_lookup(os, chld_dd->dd_object, | |
580 | DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count)); | |
581 | my_ss_cnt += count; | |
582 | ||
583 | dsl_dir_rele(chld_dd, FTAG); | |
584 | } | |
585 | zap_cursor_fini(zc); | |
586 | /* Count my snapshots (we counted children's snapshots above) */ | |
587 | VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, | |
d683ddbb | 588 | dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds)); |
788eb90c | 589 | |
d683ddbb | 590 | for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj); |
788eb90c JJ |
591 | zap_cursor_retrieve(zc, za) == 0; |
592 | zap_cursor_advance(zc)) { | |
593 | /* Don't count temporary snapshots */ | |
594 | if (za->za_name[0] != '%') | |
595 | my_ss_cnt++; | |
596 | } | |
ca227e54 | 597 | zap_cursor_fini(zc); |
788eb90c JJ |
598 | |
599 | dsl_dataset_rele(ds, FTAG); | |
600 | ||
601 | kmem_free(zc, sizeof (zap_cursor_t)); | |
602 | kmem_free(za, sizeof (zap_attribute_t)); | |
603 | ||
604 | /* we're in a sync task, update counts */ | |
605 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
606 | VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, | |
607 | sizeof (my_fs_cnt), 1, &my_fs_cnt, tx)); | |
608 | VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, | |
609 | sizeof (my_ss_cnt), 1, &my_ss_cnt, tx)); | |
610 | } | |
611 | ||
612 | static int | |
613 | dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx) | |
614 | { | |
615 | char *ddname = (char *)arg; | |
616 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
617 | dsl_dataset_t *ds; | |
618 | dsl_dir_t *dd; | |
619 | int error; | |
620 | ||
621 | error = dsl_dataset_hold(dp, ddname, FTAG, &ds); | |
622 | if (error != 0) | |
623 | return (error); | |
624 | ||
625 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { | |
626 | dsl_dataset_rele(ds, FTAG); | |
627 | return (SET_ERROR(ENOTSUP)); | |
628 | } | |
629 | ||
630 | dd = ds->ds_dir; | |
631 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) && | |
632 | dsl_dir_is_zapified(dd) && | |
633 | zap_contains(dp->dp_meta_objset, dd->dd_object, | |
634 | DD_FIELD_FILESYSTEM_COUNT) == 0) { | |
635 | dsl_dataset_rele(ds, FTAG); | |
636 | return (SET_ERROR(EALREADY)); | |
637 | } | |
638 | ||
639 | dsl_dataset_rele(ds, FTAG); | |
640 | return (0); | |
641 | } | |
642 | ||
643 | static void | |
644 | dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx) | |
645 | { | |
646 | char *ddname = (char *)arg; | |
647 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
648 | dsl_dataset_t *ds; | |
649 | spa_t *spa; | |
650 | ||
651 | VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds)); | |
652 | ||
653 | spa = dsl_dataset_get_spa(ds); | |
654 | ||
655 | if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) { | |
656 | /* | |
657 | * Since the feature was not active and we're now setting a | |
658 | * limit, increment the feature-active counter so that the | |
659 | * feature becomes active for the first time. | |
660 | * | |
661 | * We are already in a sync task so we can update the MOS. | |
662 | */ | |
663 | spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx); | |
664 | } | |
665 | ||
666 | /* | |
667 | * Since we are now setting a non-UINT64_MAX limit on the filesystem, | |
668 | * we need to ensure the counts are correct. Descend down the tree from | |
669 | * this point and update all of the counts to be accurate. | |
670 | */ | |
671 | dsl_dir_init_fs_ss_count(ds->ds_dir, tx); | |
672 | ||
673 | dsl_dataset_rele(ds, FTAG); | |
674 | } | |
675 | ||
676 | /* | |
677 | * Make sure the feature is enabled and activate it if necessary. | |
678 | * Since we're setting a limit, ensure the on-disk counts are valid. | |
679 | * This is only called by the ioctl path when setting a limit value. | |
680 | * | |
681 | * We do not need to validate the new limit, since users who can change the | |
682 | * limit are also allowed to exceed the limit. | |
683 | */ | |
684 | int | |
685 | dsl_dir_activate_fs_ss_limit(const char *ddname) | |
686 | { | |
687 | int error; | |
688 | ||
689 | error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check, | |
3d45fdd6 MA |
690 | dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0, |
691 | ZFS_SPACE_CHECK_RESERVED); | |
788eb90c JJ |
692 | |
693 | if (error == EALREADY) | |
694 | error = 0; | |
695 | ||
696 | return (error); | |
697 | } | |
698 | ||
699 | /* | |
700 | * Used to determine if the filesystem_limit or snapshot_limit should be | |
701 | * enforced. We allow the limit to be exceeded if the user has permission to | |
702 | * write the property value. We pass in the creds that we got in the open | |
703 | * context since we will always be the GZ root in syncing context. We also have | |
704 | * to handle the case where we are allowed to change the limit on the current | |
705 | * dataset, but there may be another limit in the tree above. | |
706 | * | |
707 | * We can never modify these two properties within a non-global zone. In | |
708 | * addition, the other checks are modeled on zfs_secpolicy_write_perms. We | |
709 | * can't use that function since we are already holding the dp_config_rwlock. | |
710 | * In addition, we already have the dd and dealing with snapshots is simplified | |
711 | * in this code. | |
712 | */ | |
713 | ||
714 | typedef enum { | |
715 | ENFORCE_ALWAYS, | |
716 | ENFORCE_NEVER, | |
717 | ENFORCE_ABOVE | |
718 | } enforce_res_t; | |
719 | ||
720 | static enforce_res_t | |
721 | dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr) | |
722 | { | |
723 | enforce_res_t enforce = ENFORCE_ALWAYS; | |
724 | uint64_t obj; | |
725 | dsl_dataset_t *ds; | |
726 | uint64_t zoned; | |
727 | ||
728 | ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || | |
729 | prop == ZFS_PROP_SNAPSHOT_LIMIT); | |
730 | ||
731 | #ifdef _KERNEL | |
732 | if (crgetzoneid(cr) != GLOBAL_ZONEID) | |
733 | return (ENFORCE_ALWAYS); | |
734 | ||
735 | if (secpolicy_zfs(cr) == 0) | |
736 | return (ENFORCE_NEVER); | |
737 | #endif | |
738 | ||
d683ddbb | 739 | if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0) |
788eb90c JJ |
740 | return (ENFORCE_ALWAYS); |
741 | ||
742 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
743 | ||
744 | if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0) | |
745 | return (ENFORCE_ALWAYS); | |
746 | ||
747 | if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) { | |
748 | /* Only root can access zoned fs's from the GZ */ | |
749 | enforce = ENFORCE_ALWAYS; | |
750 | } else { | |
751 | if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0) | |
752 | enforce = ENFORCE_ABOVE; | |
753 | } | |
754 | ||
755 | dsl_dataset_rele(ds, FTAG); | |
756 | return (enforce); | |
757 | } | |
758 | ||
a1d477c2 MA |
759 | static void |
760 | dsl_dir_update_last_remap_txg_sync(void *varg, dmu_tx_t *tx) | |
761 | { | |
762 | ddulrt_arg_t *arg = varg; | |
763 | uint64_t last_remap_txg; | |
764 | dsl_dir_t *dd = arg->ddulrta_dd; | |
765 | objset_t *mos = dd->dd_pool->dp_meta_objset; | |
766 | ||
767 | dsl_dir_zapify(dd, tx); | |
768 | if (zap_lookup(mos, dd->dd_object, DD_FIELD_LAST_REMAP_TXG, | |
769 | sizeof (last_remap_txg), 1, &last_remap_txg) != 0 || | |
770 | last_remap_txg < arg->ddlrta_txg) { | |
771 | VERIFY0(zap_update(mos, dd->dd_object, DD_FIELD_LAST_REMAP_TXG, | |
772 | sizeof (arg->ddlrta_txg), 1, &arg->ddlrta_txg, tx)); | |
773 | } | |
774 | } | |
775 | ||
776 | int | |
777 | dsl_dir_update_last_remap_txg(dsl_dir_t *dd, uint64_t txg) | |
778 | { | |
779 | ddulrt_arg_t arg; | |
780 | arg.ddulrta_dd = dd; | |
781 | arg.ddlrta_txg = txg; | |
782 | ||
783 | return (dsl_sync_task(spa_name(dd->dd_pool->dp_spa), | |
784 | NULL, dsl_dir_update_last_remap_txg_sync, &arg, | |
785 | 1, ZFS_SPACE_CHECK_RESERVED)); | |
786 | } | |
787 | ||
788eb90c JJ |
788 | /* |
789 | * Check if adding additional child filesystem(s) would exceed any filesystem | |
790 | * limits or adding additional snapshot(s) would exceed any snapshot limits. | |
791 | * The prop argument indicates which limit to check. | |
792 | * | |
793 | * Note that all filesystem limits up to the root (or the highest | |
794 | * initialized) filesystem or the given ancestor must be satisfied. | |
795 | */ | |
796 | int | |
797 | dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop, | |
798 | dsl_dir_t *ancestor, cred_t *cr) | |
799 | { | |
800 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
801 | uint64_t limit, count; | |
802 | char *count_prop; | |
803 | enforce_res_t enforce; | |
804 | int err = 0; | |
805 | ||
806 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
807 | ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || | |
808 | prop == ZFS_PROP_SNAPSHOT_LIMIT); | |
809 | ||
810 | /* | |
811 | * If we're allowed to change the limit, don't enforce the limit | |
812 | * e.g. this can happen if a snapshot is taken by an administrative | |
813 | * user in the global zone (i.e. a recursive snapshot by root). | |
814 | * However, we must handle the case of delegated permissions where we | |
815 | * are allowed to change the limit on the current dataset, but there | |
816 | * is another limit in the tree above. | |
817 | */ | |
818 | enforce = dsl_enforce_ds_ss_limits(dd, prop, cr); | |
819 | if (enforce == ENFORCE_NEVER) | |
820 | return (0); | |
821 | ||
822 | /* | |
823 | * e.g. if renaming a dataset with no snapshots, count adjustment | |
824 | * is 0. | |
825 | */ | |
826 | if (delta == 0) | |
827 | return (0); | |
828 | ||
829 | if (prop == ZFS_PROP_SNAPSHOT_LIMIT) { | |
830 | /* | |
831 | * We don't enforce the limit for temporary snapshots. This is | |
832 | * indicated by a NULL cred_t argument. | |
833 | */ | |
834 | if (cr == NULL) | |
835 | return (0); | |
836 | ||
837 | count_prop = DD_FIELD_SNAPSHOT_COUNT; | |
838 | } else { | |
839 | count_prop = DD_FIELD_FILESYSTEM_COUNT; | |
840 | } | |
841 | ||
842 | /* | |
843 | * If an ancestor has been provided, stop checking the limit once we | |
844 | * hit that dir. We need this during rename so that we don't overcount | |
845 | * the check once we recurse up to the common ancestor. | |
846 | */ | |
847 | if (ancestor == dd) | |
848 | return (0); | |
849 | ||
850 | /* | |
851 | * If we hit an uninitialized node while recursing up the tree, we can | |
852 | * stop since we know there is no limit here (or above). The counts are | |
853 | * not valid on this node and we know we won't touch this node's counts. | |
854 | */ | |
855 | if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object, | |
856 | count_prop, sizeof (count), 1, &count) == ENOENT) | |
857 | return (0); | |
858 | ||
859 | err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL, | |
860 | B_FALSE); | |
861 | if (err != 0) | |
862 | return (err); | |
863 | ||
864 | /* Is there a limit which we've hit? */ | |
865 | if (enforce == ENFORCE_ALWAYS && (count + delta) > limit) | |
866 | return (SET_ERROR(EDQUOT)); | |
867 | ||
868 | if (dd->dd_parent != NULL) | |
869 | err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop, | |
870 | ancestor, cr); | |
871 | ||
872 | return (err); | |
873 | } | |
874 | ||
875 | /* | |
876 | * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all | |
877 | * parents. When a new filesystem/snapshot is created, increment the count on | |
878 | * all parents, and when a filesystem/snapshot is destroyed, decrement the | |
879 | * count. | |
880 | */ | |
881 | void | |
882 | dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop, | |
883 | dmu_tx_t *tx) | |
884 | { | |
885 | int err; | |
886 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
887 | uint64_t count; | |
888 | ||
889 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
890 | ASSERT(dmu_tx_is_syncing(tx)); | |
891 | ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 || | |
892 | strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0); | |
893 | ||
894 | /* | |
895 | * When we receive an incremental stream into a filesystem that already | |
896 | * exists, a temporary clone is created. We don't count this temporary | |
897 | * clone, whose name begins with a '%'. We also ignore hidden ($FREE, | |
898 | * $MOS & $ORIGIN) objsets. | |
899 | */ | |
900 | if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') && | |
901 | strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0) | |
902 | return; | |
903 | ||
904 | /* | |
905 | * e.g. if renaming a dataset with no snapshots, count adjustment is 0 | |
906 | */ | |
907 | if (delta == 0) | |
908 | return; | |
909 | ||
910 | /* | |
911 | * If we hit an uninitialized node while recursing up the tree, we can | |
912 | * stop since we know the counts are not valid on this node and we | |
913 | * know we shouldn't touch this node's counts. An uninitialized count | |
914 | * on the node indicates that either the feature has not yet been | |
915 | * activated or there are no limits on this part of the tree. | |
916 | */ | |
917 | if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object, | |
918 | prop, sizeof (count), 1, &count)) == ENOENT) | |
919 | return; | |
920 | VERIFY0(err); | |
921 | ||
922 | count += delta; | |
923 | /* Use a signed verify to make sure we're not neg. */ | |
924 | VERIFY3S(count, >=, 0); | |
925 | ||
926 | VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count, | |
927 | tx)); | |
928 | ||
929 | /* Roll up this additional count into our ancestors */ | |
930 | if (dd->dd_parent != NULL) | |
931 | dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx); | |
932 | } | |
933 | ||
34dc7c2f | 934 | uint64_t |
b128c09f BB |
935 | dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, |
936 | dmu_tx_t *tx) | |
34dc7c2f | 937 | { |
b128c09f | 938 | objset_t *mos = dp->dp_meta_objset; |
34dc7c2f | 939 | uint64_t ddobj; |
428870ff | 940 | dsl_dir_phys_t *ddphys; |
34dc7c2f BB |
941 | dmu_buf_t *dbuf; |
942 | ||
943 | ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, | |
944 | DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); | |
b128c09f | 945 | if (pds) { |
d2734cce | 946 | VERIFY0(zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj, |
b128c09f BB |
947 | name, sizeof (uint64_t), 1, &ddobj, tx)); |
948 | } else { | |
949 | /* it's the root dir */ | |
d2734cce | 950 | VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, |
b128c09f BB |
951 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); |
952 | } | |
d2734cce | 953 | VERIFY0(dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); |
34dc7c2f | 954 | dmu_buf_will_dirty(dbuf, tx); |
428870ff | 955 | ddphys = dbuf->db_data; |
34dc7c2f | 956 | |
428870ff | 957 | ddphys->dd_creation_time = gethrestime_sec(); |
788eb90c | 958 | if (pds) { |
428870ff | 959 | ddphys->dd_parent_obj = pds->dd_object; |
788eb90c JJ |
960 | |
961 | /* update the filesystem counts */ | |
962 | dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx); | |
963 | } | |
428870ff | 964 | ddphys->dd_props_zapobj = zap_create(mos, |
34dc7c2f | 965 | DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); |
428870ff | 966 | ddphys->dd_child_dir_zapobj = zap_create(mos, |
34dc7c2f | 967 | DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); |
b128c09f | 968 | if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) |
428870ff | 969 | ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; |
b5256303 | 970 | |
34dc7c2f BB |
971 | dmu_buf_rele(dbuf, FTAG); |
972 | ||
973 | return (ddobj); | |
974 | } | |
975 | ||
b128c09f BB |
976 | boolean_t |
977 | dsl_dir_is_clone(dsl_dir_t *dd) | |
34dc7c2f | 978 | { |
d683ddbb | 979 | return (dsl_dir_phys(dd)->dd_origin_obj && |
b128c09f | 980 | (dd->dd_pool->dp_origin_snap == NULL || |
d683ddbb | 981 | dsl_dir_phys(dd)->dd_origin_obj != |
b128c09f | 982 | dd->dd_pool->dp_origin_snap->ds_object)); |
34dc7c2f BB |
983 | } |
984 | ||
d99a0153 CW |
985 | uint64_t |
986 | dsl_dir_get_used(dsl_dir_t *dd) | |
987 | { | |
988 | return (dsl_dir_phys(dd)->dd_used_bytes); | |
989 | } | |
990 | ||
d2734cce SD |
991 | uint64_t |
992 | dsl_dir_get_compressed(dsl_dir_t *dd) | |
993 | { | |
994 | return (dsl_dir_phys(dd)->dd_compressed_bytes); | |
995 | } | |
996 | ||
d99a0153 CW |
997 | uint64_t |
998 | dsl_dir_get_quota(dsl_dir_t *dd) | |
999 | { | |
1000 | return (dsl_dir_phys(dd)->dd_quota); | |
1001 | } | |
1002 | ||
1003 | uint64_t | |
1004 | dsl_dir_get_reservation(dsl_dir_t *dd) | |
1005 | { | |
1006 | return (dsl_dir_phys(dd)->dd_reserved); | |
1007 | } | |
1008 | ||
1009 | uint64_t | |
1010 | dsl_dir_get_compressratio(dsl_dir_t *dd) | |
1011 | { | |
1012 | /* a fixed point number, 100x the ratio */ | |
1013 | return (dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 : | |
1014 | (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 / | |
1015 | dsl_dir_phys(dd)->dd_compressed_bytes)); | |
1016 | } | |
1017 | ||
1018 | uint64_t | |
1019 | dsl_dir_get_logicalused(dsl_dir_t *dd) | |
1020 | { | |
1021 | return (dsl_dir_phys(dd)->dd_uncompressed_bytes); | |
1022 | } | |
1023 | ||
1024 | uint64_t | |
1025 | dsl_dir_get_usedsnap(dsl_dir_t *dd) | |
1026 | { | |
1027 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]); | |
1028 | } | |
1029 | ||
1030 | uint64_t | |
1031 | dsl_dir_get_usedds(dsl_dir_t *dd) | |
1032 | { | |
1033 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]); | |
1034 | } | |
1035 | ||
1036 | uint64_t | |
1037 | dsl_dir_get_usedrefreserv(dsl_dir_t *dd) | |
1038 | { | |
1039 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]); | |
1040 | } | |
1041 | ||
1042 | uint64_t | |
1043 | dsl_dir_get_usedchild(dsl_dir_t *dd) | |
1044 | { | |
1045 | return (dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] + | |
1046 | dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]); | |
1047 | } | |
1048 | ||
34dc7c2f | 1049 | void |
d99a0153 CW |
1050 | dsl_dir_get_origin(dsl_dir_t *dd, char *buf) |
1051 | { | |
1052 | dsl_dataset_t *ds; | |
1053 | VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, | |
1054 | dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds)); | |
1055 | ||
1056 | dsl_dataset_name(ds, buf); | |
1057 | ||
1058 | dsl_dataset_rele(ds, FTAG); | |
1059 | } | |
1060 | ||
1061 | int | |
1062 | dsl_dir_get_filesystem_count(dsl_dir_t *dd, uint64_t *count) | |
34dc7c2f | 1063 | { |
d99a0153 CW |
1064 | if (dsl_dir_is_zapified(dd)) { |
1065 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
1066 | return (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, | |
1067 | sizeof (*count), 1, count)); | |
1068 | } else { | |
1069 | return (ENOENT); | |
1070 | } | |
1071 | } | |
1072 | ||
1073 | int | |
1074 | dsl_dir_get_snapshot_count(dsl_dir_t *dd, uint64_t *count) | |
1075 | { | |
1076 | if (dsl_dir_is_zapified(dd)) { | |
1077 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
1078 | return (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, | |
1079 | sizeof (*count), 1, count)); | |
1080 | } else { | |
1081 | return (ENOENT); | |
1082 | } | |
1083 | } | |
b5256303 | 1084 | |
a1d477c2 MA |
1085 | int |
1086 | dsl_dir_get_remaptxg(dsl_dir_t *dd, uint64_t *count) | |
1087 | { | |
1088 | if (dsl_dir_is_zapified(dd)) { | |
1089 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
1090 | return (zap_lookup(os, dd->dd_object, DD_FIELD_LAST_REMAP_TXG, | |
1091 | sizeof (*count), 1, count)); | |
1092 | } else { | |
1093 | return (ENOENT); | |
1094 | } | |
1095 | ||
1096 | } | |
1097 | ||
d99a0153 CW |
1098 | void |
1099 | dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) | |
1100 | { | |
34dc7c2f | 1101 | mutex_enter(&dd->dd_lock); |
d683ddbb | 1102 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, |
d99a0153 | 1103 | dsl_dir_get_quota(dd)); |
34dc7c2f | 1104 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, |
d99a0153 | 1105 | dsl_dir_get_reservation(dd)); |
24a64651 | 1106 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, |
d99a0153 | 1107 | dsl_dir_get_logicalused(dd)); |
d683ddbb | 1108 | if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { |
b128c09f | 1109 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, |
d99a0153 | 1110 | dsl_dir_get_usedsnap(dd)); |
b128c09f | 1111 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, |
d99a0153 | 1112 | dsl_dir_get_usedds(dd)); |
b128c09f | 1113 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, |
d99a0153 | 1114 | dsl_dir_get_usedrefreserv(dd)); |
b128c09f | 1115 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, |
d99a0153 | 1116 | dsl_dir_get_usedchild(dd)); |
b128c09f | 1117 | } |
34dc7c2f BB |
1118 | mutex_exit(&dd->dd_lock); |
1119 | ||
d99a0153 CW |
1120 | uint64_t count; |
1121 | if (dsl_dir_get_filesystem_count(dd, &count) == 0) { | |
1122 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FILESYSTEM_COUNT, | |
1123 | count); | |
1124 | } | |
1125 | if (dsl_dir_get_snapshot_count(dd, &count) == 0) { | |
1126 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOT_COUNT, | |
1127 | count); | |
788eb90c | 1128 | } |
a1d477c2 MA |
1129 | if (dsl_dir_get_remaptxg(dd, &count) == 0) { |
1130 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REMAPTXG, | |
1131 | count); | |
1132 | } | |
788eb90c | 1133 | |
b128c09f | 1134 | if (dsl_dir_is_clone(dd)) { |
eca7b760 | 1135 | char buf[ZFS_MAX_DATASET_NAME_LEN]; |
d99a0153 | 1136 | dsl_dir_get_origin(dd, buf); |
34dc7c2f BB |
1137 | dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); |
1138 | } | |
d99a0153 | 1139 | |
34dc7c2f BB |
1140 | } |
1141 | ||
1142 | void | |
1143 | dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) | |
1144 | { | |
1145 | dsl_pool_t *dp = dd->dd_pool; | |
1146 | ||
d683ddbb | 1147 | ASSERT(dsl_dir_phys(dd)); |
34dc7c2f | 1148 | |
13fe0198 | 1149 | if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { |
34dc7c2f BB |
1150 | /* up the hold count until we can be written out */ |
1151 | dmu_buf_add_ref(dd->dd_dbuf, dd); | |
1152 | } | |
1153 | } | |
1154 | ||
1155 | static int64_t | |
1156 | parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) | |
1157 | { | |
d683ddbb JG |
1158 | uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved); |
1159 | uint64_t new_accounted = | |
1160 | MAX(used + delta, dsl_dir_phys(dd)->dd_reserved); | |
34dc7c2f BB |
1161 | return (new_accounted - old_accounted); |
1162 | } | |
1163 | ||
1164 | void | |
1165 | dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) | |
1166 | { | |
1167 | ASSERT(dmu_tx_is_syncing(tx)); | |
1168 | ||
34dc7c2f | 1169 | mutex_enter(&dd->dd_lock); |
c99c9001 | 1170 | ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); |
34dc7c2f BB |
1171 | dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, |
1172 | dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); | |
1173 | dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; | |
34dc7c2f BB |
1174 | mutex_exit(&dd->dd_lock); |
1175 | ||
1176 | /* release the hold from dsl_dir_dirty */ | |
1177 | dmu_buf_rele(dd->dd_dbuf, dd); | |
1178 | } | |
1179 | ||
1180 | static uint64_t | |
1181 | dsl_dir_space_towrite(dsl_dir_t *dd) | |
1182 | { | |
1183 | uint64_t space = 0; | |
34dc7c2f BB |
1184 | |
1185 | ASSERT(MUTEX_HELD(&dd->dd_lock)); | |
1186 | ||
3ec3bc21 BB |
1187 | for (int i = 0; i < TXG_SIZE; i++) { |
1188 | space += dd->dd_space_towrite[i & TXG_MASK]; | |
1189 | ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0); | |
34dc7c2f BB |
1190 | } |
1191 | return (space); | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * How much space would dd have available if ancestor had delta applied | |
1196 | * to it? If ondiskonly is set, we're only interested in what's | |
1197 | * on-disk, not estimated pending changes. | |
1198 | */ | |
1199 | uint64_t | |
1200 | dsl_dir_space_available(dsl_dir_t *dd, | |
1201 | dsl_dir_t *ancestor, int64_t delta, int ondiskonly) | |
1202 | { | |
1203 | uint64_t parentspace, myspace, quota, used; | |
1204 | ||
1205 | /* | |
1206 | * If there are no restrictions otherwise, assume we have | |
1207 | * unlimited space available. | |
1208 | */ | |
1209 | quota = UINT64_MAX; | |
1210 | parentspace = UINT64_MAX; | |
1211 | ||
1212 | if (dd->dd_parent != NULL) { | |
1213 | parentspace = dsl_dir_space_available(dd->dd_parent, | |
1214 | ancestor, delta, ondiskonly); | |
1215 | } | |
1216 | ||
1217 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1218 | if (dsl_dir_phys(dd)->dd_quota != 0) |
1219 | quota = dsl_dir_phys(dd)->dd_quota; | |
1220 | used = dsl_dir_phys(dd)->dd_used_bytes; | |
34dc7c2f BB |
1221 | if (!ondiskonly) |
1222 | used += dsl_dir_space_towrite(dd); | |
34dc7c2f BB |
1223 | |
1224 | if (dd->dd_parent == NULL) { | |
d2734cce SD |
1225 | uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, |
1226 | ZFS_SPACE_CHECK_NORMAL); | |
34dc7c2f BB |
1227 | quota = MIN(quota, poolsize); |
1228 | } | |
1229 | ||
d683ddbb | 1230 | if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) { |
34dc7c2f BB |
1231 | /* |
1232 | * We have some space reserved, in addition to what our | |
1233 | * parent gave us. | |
1234 | */ | |
d683ddbb | 1235 | parentspace += dsl_dir_phys(dd)->dd_reserved - used; |
34dc7c2f BB |
1236 | } |
1237 | ||
b128c09f BB |
1238 | if (dd == ancestor) { |
1239 | ASSERT(delta <= 0); | |
1240 | ASSERT(used >= -delta); | |
1241 | used += delta; | |
1242 | if (parentspace != UINT64_MAX) | |
1243 | parentspace -= delta; | |
1244 | } | |
1245 | ||
34dc7c2f BB |
1246 | if (used > quota) { |
1247 | /* over quota */ | |
1248 | myspace = 0; | |
34dc7c2f BB |
1249 | } else { |
1250 | /* | |
1251 | * the lesser of the space provided by our parent and | |
1252 | * the space left in our quota | |
1253 | */ | |
1254 | myspace = MIN(parentspace, quota - used); | |
1255 | } | |
1256 | ||
1257 | mutex_exit(&dd->dd_lock); | |
1258 | ||
1259 | return (myspace); | |
1260 | } | |
1261 | ||
1262 | struct tempreserve { | |
1263 | list_node_t tr_node; | |
34dc7c2f BB |
1264 | dsl_dir_t *tr_ds; |
1265 | uint64_t tr_size; | |
1266 | }; | |
1267 | ||
1268 | static int | |
1269 | dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, | |
3ec3bc21 | 1270 | boolean_t ignorequota, list_t *tr_list, |
34dc7c2f BB |
1271 | dmu_tx_t *tx, boolean_t first) |
1272 | { | |
419c80e6 | 1273 | uint64_t txg; |
3ec3bc21 | 1274 | uint64_t quota; |
34dc7c2f | 1275 | struct tempreserve *tr; |
419c80e6 D |
1276 | int retval; |
1277 | uint64_t ref_rsrv; | |
1278 | ||
1279 | top_of_function: | |
1280 | txg = tx->tx_txg; | |
1281 | retval = EDQUOT; | |
1282 | ref_rsrv = 0; | |
34dc7c2f BB |
1283 | |
1284 | ASSERT3U(txg, !=, 0); | |
1285 | ASSERT3S(asize, >, 0); | |
1286 | ||
1287 | mutex_enter(&dd->dd_lock); | |
1288 | ||
1289 | /* | |
1290 | * Check against the dsl_dir's quota. We don't add in the delta | |
1291 | * when checking for over-quota because they get one free hit. | |
1292 | */ | |
3ec3bc21 BB |
1293 | uint64_t est_inflight = dsl_dir_space_towrite(dd); |
1294 | for (int i = 0; i < TXG_SIZE; i++) | |
34dc7c2f | 1295 | est_inflight += dd->dd_tempreserved[i]; |
3ec3bc21 | 1296 | uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes; |
34dc7c2f BB |
1297 | |
1298 | /* | |
1299 | * On the first iteration, fetch the dataset's used-on-disk and | |
1300 | * refreservation values. Also, if checkrefquota is set, test if | |
1301 | * allocating this space would exceed the dataset's refquota. | |
1302 | */ | |
1303 | if (first && tx->tx_objset) { | |
1304 | int error; | |
428870ff | 1305 | dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; |
34dc7c2f | 1306 | |
3ec3bc21 | 1307 | error = dsl_dataset_check_quota(ds, !netfree, |
34dc7c2f | 1308 | asize, est_inflight, &used_on_disk, &ref_rsrv); |
3ec3bc21 | 1309 | if (error != 0) { |
34dc7c2f | 1310 | mutex_exit(&dd->dd_lock); |
3d920a15 | 1311 | DMU_TX_STAT_BUMP(dmu_tx_quota); |
34dc7c2f BB |
1312 | return (error); |
1313 | } | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * If this transaction will result in a net free of space, | |
1318 | * we want to let it through. | |
1319 | */ | |
d683ddbb | 1320 | if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0) |
34dc7c2f BB |
1321 | quota = UINT64_MAX; |
1322 | else | |
d683ddbb | 1323 | quota = dsl_dir_phys(dd)->dd_quota; |
34dc7c2f BB |
1324 | |
1325 | /* | |
428870ff BB |
1326 | * Adjust the quota against the actual pool size at the root |
1327 | * minus any outstanding deferred frees. | |
34dc7c2f BB |
1328 | * To ensure that it's possible to remove files from a full |
1329 | * pool without inducing transient overcommits, we throttle | |
1330 | * netfree transactions against a quota that is slightly larger, | |
1331 | * but still within the pool's allocation slop. In cases where | |
1332 | * we're very close to full, this will allow a steady trickle of | |
1333 | * removes to get through. | |
1334 | */ | |
3ec3bc21 | 1335 | uint64_t deferred = 0; |
34dc7c2f | 1336 | if (dd->dd_parent == NULL) { |
d2734cce SD |
1337 | uint64_t avail = dsl_pool_unreserved_space(dd->dd_pool, |
1338 | (netfree) ? | |
1339 | ZFS_SPACE_CHECK_RESERVED : ZFS_SPACE_CHECK_NORMAL); | |
1340 | ||
1341 | if (avail < quota) { | |
1342 | quota = avail; | |
428870ff | 1343 | retval = ENOSPC; |
34dc7c2f BB |
1344 | } |
1345 | } | |
1346 | ||
1347 | /* | |
1348 | * If they are requesting more space, and our current estimate | |
1349 | * is over quota, they get to try again unless the actual | |
1350 | * on-disk is over quota and there are no pending changes (which | |
1351 | * may free up space for us). | |
1352 | */ | |
428870ff BB |
1353 | if (used_on_disk + est_inflight >= quota) { |
1354 | if (est_inflight > 0 || used_on_disk < quota || | |
1355 | (retval == ENOSPC && used_on_disk < quota + deferred)) | |
1356 | retval = ERESTART; | |
34dc7c2f BB |
1357 | dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " |
1358 | "quota=%lluK tr=%lluK err=%d\n", | |
1359 | used_on_disk>>10, est_inflight>>10, | |
428870ff | 1360 | quota>>10, asize>>10, retval); |
34dc7c2f | 1361 | mutex_exit(&dd->dd_lock); |
3d920a15 | 1362 | DMU_TX_STAT_BUMP(dmu_tx_quota); |
2e528b49 | 1363 | return (SET_ERROR(retval)); |
34dc7c2f BB |
1364 | } |
1365 | ||
1366 | /* We need to up our estimated delta before dropping dd_lock */ | |
3ec3bc21 | 1367 | dd->dd_tempreserved[txg & TXG_MASK] += asize; |
34dc7c2f | 1368 | |
3ec3bc21 | 1369 | uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, |
34dc7c2f BB |
1370 | asize - ref_rsrv); |
1371 | mutex_exit(&dd->dd_lock); | |
1372 | ||
79c76d5b | 1373 | tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); |
34dc7c2f BB |
1374 | tr->tr_ds = dd; |
1375 | tr->tr_size = asize; | |
1376 | list_insert_tail(tr_list, tr); | |
1377 | ||
1378 | /* see if it's OK with our parent */ | |
3ec3bc21 | 1379 | if (dd->dd_parent != NULL && parent_rsrv != 0) { |
419c80e6 D |
1380 | /* |
1381 | * Recurse on our parent without recursion. This has been | |
1382 | * observed to be potentially large stack usage even within | |
1383 | * the test suite. Largest seen stack was 7632 bytes on linux. | |
1384 | */ | |
1385 | ||
1386 | dd = dd->dd_parent; | |
1387 | asize = parent_rsrv; | |
1388 | ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0); | |
1389 | first = B_FALSE; | |
1390 | goto top_of_function; | |
34dc7c2f | 1391 | |
34dc7c2f BB |
1392 | } else { |
1393 | return (0); | |
1394 | } | |
1395 | } | |
1396 | ||
1397 | /* | |
1398 | * Reserve space in this dsl_dir, to be used in this tx's txg. | |
1399 | * After the space has been dirtied (and dsl_dir_willuse_space() | |
1400 | * has been called), the reservation should be canceled, using | |
1401 | * dsl_dir_tempreserve_clear(). | |
1402 | */ | |
1403 | int | |
1404 | dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, | |
3ec3bc21 | 1405 | boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx) |
34dc7c2f BB |
1406 | { |
1407 | int err; | |
1408 | list_t *tr_list; | |
1409 | ||
1410 | if (asize == 0) { | |
1411 | *tr_cookiep = NULL; | |
1412 | return (0); | |
1413 | } | |
1414 | ||
79c76d5b | 1415 | tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); |
34dc7c2f BB |
1416 | list_create(tr_list, sizeof (struct tempreserve), |
1417 | offsetof(struct tempreserve, tr_node)); | |
1418 | ASSERT3S(asize, >, 0); | |
34dc7c2f | 1419 | |
dae3e9ea | 1420 | err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg); |
34dc7c2f BB |
1421 | if (err == 0) { |
1422 | struct tempreserve *tr; | |
1423 | ||
79c76d5b | 1424 | tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); |
34dc7c2f BB |
1425 | tr->tr_size = lsize; |
1426 | list_insert_tail(tr_list, tr); | |
34dc7c2f BB |
1427 | } else { |
1428 | if (err == EAGAIN) { | |
e8b96c60 MA |
1429 | /* |
1430 | * If arc_memory_throttle() detected that pageout | |
1431 | * is running and we are low on memory, we delay new | |
1432 | * non-pageout transactions to give pageout an | |
1433 | * advantage. | |
1434 | * | |
1435 | * It is unfortunate to be delaying while the caller's | |
1436 | * locks are held. | |
1437 | */ | |
63fd3c6c AL |
1438 | txg_delay(dd->dd_pool, tx->tx_txg, |
1439 | MSEC2NSEC(10), MSEC2NSEC(10)); | |
2e528b49 | 1440 | err = SET_ERROR(ERESTART); |
34dc7c2f | 1441 | } |
34dc7c2f BB |
1442 | } |
1443 | ||
1444 | if (err == 0) { | |
3ec3bc21 BB |
1445 | err = dsl_dir_tempreserve_impl(dd, asize, netfree, |
1446 | B_FALSE, tr_list, tx, B_TRUE); | |
34dc7c2f BB |
1447 | } |
1448 | ||
13fe0198 | 1449 | if (err != 0) |
34dc7c2f BB |
1450 | dsl_dir_tempreserve_clear(tr_list, tx); |
1451 | else | |
1452 | *tr_cookiep = tr_list; | |
1453 | ||
1454 | return (err); | |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * Clear a temporary reservation that we previously made with | |
1459 | * dsl_dir_tempreserve_space(). | |
1460 | */ | |
1461 | void | |
1462 | dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) | |
1463 | { | |
1464 | int txgidx = tx->tx_txg & TXG_MASK; | |
1465 | list_t *tr_list = tr_cookie; | |
1466 | struct tempreserve *tr; | |
1467 | ||
1468 | ASSERT3U(tx->tx_txg, !=, 0); | |
1469 | ||
1470 | if (tr_cookie == NULL) | |
1471 | return; | |
1472 | ||
e8b96c60 MA |
1473 | while ((tr = list_head(tr_list)) != NULL) { |
1474 | if (tr->tr_ds) { | |
34dc7c2f BB |
1475 | mutex_enter(&tr->tr_ds->dd_lock); |
1476 | ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, | |
1477 | tr->tr_size); | |
1478 | tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; | |
1479 | mutex_exit(&tr->tr_ds->dd_lock); | |
1480 | } else { | |
1481 | arc_tempreserve_clear(tr->tr_size); | |
1482 | } | |
1483 | list_remove(tr_list, tr); | |
1484 | kmem_free(tr, sizeof (struct tempreserve)); | |
1485 | } | |
1486 | ||
1487 | kmem_free(tr_list, sizeof (list_t)); | |
1488 | } | |
1489 | ||
e8b96c60 MA |
1490 | /* |
1491 | * This should be called from open context when we think we're going to write | |
1492 | * or free space, for example when dirtying data. Be conservative; it's okay | |
1493 | * to write less space or free more, but we don't want to write more or free | |
1494 | * less than the amount specified. | |
1ba16159 AB |
1495 | * |
1496 | * NOTE: The behavior of this function is identical to the Illumos / FreeBSD | |
1497 | * version however it has been adjusted to use an iterative rather then | |
1498 | * recursive algorithm to minimize stack usage. | |
e8b96c60 MA |
1499 | */ |
1500 | void | |
1501 | dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) | |
34dc7c2f BB |
1502 | { |
1503 | int64_t parent_space; | |
1504 | uint64_t est_used; | |
1505 | ||
1ba16159 AB |
1506 | do { |
1507 | mutex_enter(&dd->dd_lock); | |
1508 | if (space > 0) | |
1509 | dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; | |
34dc7c2f | 1510 | |
1ba16159 | 1511 | est_used = dsl_dir_space_towrite(dd) + |
d683ddbb | 1512 | dsl_dir_phys(dd)->dd_used_bytes; |
1ba16159 AB |
1513 | parent_space = parent_delta(dd, est_used, space); |
1514 | mutex_exit(&dd->dd_lock); | |
34dc7c2f | 1515 | |
1ba16159 AB |
1516 | /* Make sure that we clean up dd_space_to* */ |
1517 | dsl_dir_dirty(dd, tx); | |
34dc7c2f | 1518 | |
1ba16159 AB |
1519 | dd = dd->dd_parent; |
1520 | space = parent_space; | |
1521 | } while (space && dd); | |
34dc7c2f BB |
1522 | } |
1523 | ||
1524 | /* call from syncing context when we actually write/free space for this dd */ | |
1525 | void | |
b128c09f | 1526 | dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, |
34dc7c2f BB |
1527 | int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) |
1528 | { | |
1529 | int64_t accounted_delta; | |
a169a625 MA |
1530 | |
1531 | /* | |
1532 | * dsl_dataset_set_refreservation_sync_impl() calls this with | |
1533 | * dd_lock held, so that it can atomically update | |
1534 | * ds->ds_reserved and the dsl_dir accounting, so that | |
1535 | * dsl_dataset_check_quota() can see dataset and dir accounting | |
1536 | * consistently. | |
1537 | */ | |
b128c09f | 1538 | boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); |
34dc7c2f BB |
1539 | |
1540 | ASSERT(dmu_tx_is_syncing(tx)); | |
b128c09f | 1541 | ASSERT(type < DD_USED_NUM); |
34dc7c2f | 1542 | |
a169a625 MA |
1543 | dmu_buf_will_dirty(dd->dd_dbuf, tx); |
1544 | ||
b128c09f BB |
1545 | if (needlock) |
1546 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1547 | accounted_delta = |
1548 | parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used); | |
1549 | ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used); | |
34dc7c2f | 1550 | ASSERT(compressed >= 0 || |
d683ddbb | 1551 | dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed); |
34dc7c2f | 1552 | ASSERT(uncompressed >= 0 || |
d683ddbb JG |
1553 | dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed); |
1554 | dsl_dir_phys(dd)->dd_used_bytes += used; | |
1555 | dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed; | |
1556 | dsl_dir_phys(dd)->dd_compressed_bytes += compressed; | |
b128c09f | 1557 | |
d683ddbb | 1558 | if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { |
b128c09f | 1559 | ASSERT(used > 0 || |
d683ddbb JG |
1560 | dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used); |
1561 | dsl_dir_phys(dd)->dd_used_breakdown[type] += used; | |
b128c09f | 1562 | #ifdef DEBUG |
d6320ddb BB |
1563 | { |
1564 | dd_used_t t; | |
1565 | uint64_t u = 0; | |
1566 | for (t = 0; t < DD_USED_NUM; t++) | |
d683ddbb JG |
1567 | u += dsl_dir_phys(dd)->dd_used_breakdown[t]; |
1568 | ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes); | |
d6320ddb | 1569 | } |
b128c09f BB |
1570 | #endif |
1571 | } | |
1572 | if (needlock) | |
1573 | mutex_exit(&dd->dd_lock); | |
34dc7c2f BB |
1574 | |
1575 | if (dd->dd_parent != NULL) { | |
b128c09f | 1576 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, |
34dc7c2f | 1577 | accounted_delta, compressed, uncompressed, tx); |
b128c09f BB |
1578 | dsl_dir_transfer_space(dd->dd_parent, |
1579 | used - accounted_delta, | |
1580 | DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); | |
34dc7c2f BB |
1581 | } |
1582 | } | |
1583 | ||
b128c09f BB |
1584 | void |
1585 | dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, | |
1586 | dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) | |
1587 | { | |
b128c09f BB |
1588 | ASSERT(dmu_tx_is_syncing(tx)); |
1589 | ASSERT(oldtype < DD_USED_NUM); | |
1590 | ASSERT(newtype < DD_USED_NUM); | |
1591 | ||
d683ddbb JG |
1592 | if (delta == 0 || |
1593 | !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN)) | |
b128c09f BB |
1594 | return; |
1595 | ||
a169a625 MA |
1596 | dmu_buf_will_dirty(dd->dd_dbuf, tx); |
1597 | mutex_enter(&dd->dd_lock); | |
b128c09f | 1598 | ASSERT(delta > 0 ? |
d683ddbb JG |
1599 | dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta : |
1600 | dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta); | |
1601 | ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta)); | |
1602 | dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta; | |
1603 | dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta; | |
a169a625 | 1604 | mutex_exit(&dd->dd_lock); |
b128c09f BB |
1605 | } |
1606 | ||
13fe0198 MA |
1607 | typedef struct dsl_dir_set_qr_arg { |
1608 | const char *ddsqra_name; | |
1609 | zprop_source_t ddsqra_source; | |
1610 | uint64_t ddsqra_value; | |
1611 | } dsl_dir_set_qr_arg_t; | |
1612 | ||
34dc7c2f | 1613 | static int |
13fe0198 | 1614 | dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1615 | { |
13fe0198 MA |
1616 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1617 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1618 | dsl_dataset_t *ds; | |
1619 | int error; | |
1620 | uint64_t towrite, newval; | |
34dc7c2f | 1621 | |
13fe0198 MA |
1622 | error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); |
1623 | if (error != 0) | |
1624 | return (error); | |
1625 | ||
1626 | error = dsl_prop_predict(ds->ds_dir, "quota", | |
1627 | ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); | |
1628 | if (error != 0) { | |
1629 | dsl_dataset_rele(ds, FTAG); | |
1630 | return (error); | |
1631 | } | |
428870ff | 1632 | |
13fe0198 MA |
1633 | if (newval == 0) { |
1634 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f | 1635 | return (0); |
13fe0198 | 1636 | } |
34dc7c2f | 1637 | |
13fe0198 | 1638 | mutex_enter(&ds->ds_dir->dd_lock); |
34dc7c2f BB |
1639 | /* |
1640 | * If we are doing the preliminary check in open context, and | |
1641 | * there are pending changes, then don't fail it, since the | |
1642 | * pending changes could under-estimate the amount of space to be | |
1643 | * freed up. | |
1644 | */ | |
13fe0198 | 1645 | towrite = dsl_dir_space_towrite(ds->ds_dir); |
34dc7c2f | 1646 | if ((dmu_tx_is_syncing(tx) || towrite == 0) && |
d683ddbb JG |
1647 | (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved || |
1648 | newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) { | |
2e528b49 | 1649 | error = SET_ERROR(ENOSPC); |
34dc7c2f | 1650 | } |
13fe0198 MA |
1651 | mutex_exit(&ds->ds_dir->dd_lock); |
1652 | dsl_dataset_rele(ds, FTAG); | |
1653 | return (error); | |
34dc7c2f BB |
1654 | } |
1655 | ||
34dc7c2f | 1656 | static void |
13fe0198 | 1657 | dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1658 | { |
13fe0198 MA |
1659 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1660 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1661 | dsl_dataset_t *ds; | |
1662 | uint64_t newval; | |
428870ff | 1663 | |
13fe0198 | 1664 | VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); |
34dc7c2f | 1665 | |
b1118acb MM |
1666 | if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { |
1667 | dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), | |
1668 | ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, | |
1669 | &ddsqra->ddsqra_value, tx); | |
34dc7c2f | 1670 | |
b1118acb MM |
1671 | VERIFY0(dsl_prop_get_int_ds(ds, |
1672 | zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); | |
1673 | } else { | |
1674 | newval = ddsqra->ddsqra_value; | |
1675 | spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", | |
1676 | zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); | |
1677 | } | |
6f1ffb06 | 1678 | |
13fe0198 MA |
1679 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); |
1680 | mutex_enter(&ds->ds_dir->dd_lock); | |
d683ddbb | 1681 | dsl_dir_phys(ds->ds_dir)->dd_quota = newval; |
13fe0198 MA |
1682 | mutex_exit(&ds->ds_dir->dd_lock); |
1683 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
1684 | } |
1685 | ||
1686 | int | |
428870ff | 1687 | dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) |
34dc7c2f | 1688 | { |
13fe0198 | 1689 | dsl_dir_set_qr_arg_t ddsqra; |
428870ff | 1690 | |
13fe0198 MA |
1691 | ddsqra.ddsqra_name = ddname; |
1692 | ddsqra.ddsqra_source = source; | |
1693 | ddsqra.ddsqra_value = quota; | |
428870ff | 1694 | |
13fe0198 | 1695 | return (dsl_sync_task(ddname, dsl_dir_set_quota_check, |
d2734cce SD |
1696 | dsl_dir_set_quota_sync, &ddsqra, 0, |
1697 | ZFS_SPACE_CHECK_EXTRA_RESERVED)); | |
34dc7c2f BB |
1698 | } |
1699 | ||
1700 | int | |
13fe0198 | 1701 | dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1702 | { |
13fe0198 MA |
1703 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1704 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1705 | dsl_dataset_t *ds; | |
1706 | dsl_dir_t *dd; | |
1707 | uint64_t newval, used, avail; | |
1708 | int error; | |
428870ff | 1709 | |
13fe0198 MA |
1710 | error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); |
1711 | if (error != 0) | |
1712 | return (error); | |
1713 | dd = ds->ds_dir; | |
34dc7c2f BB |
1714 | |
1715 | /* | |
1716 | * If we are doing the preliminary check in open context, the | |
1717 | * space estimates may be inaccurate. | |
1718 | */ | |
13fe0198 MA |
1719 | if (!dmu_tx_is_syncing(tx)) { |
1720 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f | 1721 | return (0); |
13fe0198 MA |
1722 | } |
1723 | ||
1724 | error = dsl_prop_predict(ds->ds_dir, | |
1725 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1726 | ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); | |
1727 | if (error != 0) { | |
1728 | dsl_dataset_rele(ds, FTAG); | |
1729 | return (error); | |
1730 | } | |
34dc7c2f BB |
1731 | |
1732 | mutex_enter(&dd->dd_lock); | |
d683ddbb | 1733 | used = dsl_dir_phys(dd)->dd_used_bytes; |
34dc7c2f BB |
1734 | mutex_exit(&dd->dd_lock); |
1735 | ||
1736 | if (dd->dd_parent) { | |
1737 | avail = dsl_dir_space_available(dd->dd_parent, | |
1738 | NULL, 0, FALSE); | |
1739 | } else { | |
d2734cce SD |
1740 | avail = dsl_pool_adjustedsize(dd->dd_pool, |
1741 | ZFS_SPACE_CHECK_NORMAL) - used; | |
34dc7c2f BB |
1742 | } |
1743 | ||
d683ddbb | 1744 | if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) { |
13fe0198 | 1745 | uint64_t delta = MAX(used, newval) - |
d683ddbb | 1746 | MAX(used, dsl_dir_phys(dd)->dd_reserved); |
d164b209 | 1747 | |
13fe0198 | 1748 | if (delta > avail || |
d683ddbb JG |
1749 | (dsl_dir_phys(dd)->dd_quota > 0 && |
1750 | newval > dsl_dir_phys(dd)->dd_quota)) | |
2e528b49 | 1751 | error = SET_ERROR(ENOSPC); |
d164b209 BB |
1752 | } |
1753 | ||
13fe0198 MA |
1754 | dsl_dataset_rele(ds, FTAG); |
1755 | return (error); | |
34dc7c2f BB |
1756 | } |
1757 | ||
13fe0198 | 1758 | void |
6f1ffb06 | 1759 | dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) |
34dc7c2f | 1760 | { |
34dc7c2f BB |
1761 | uint64_t used; |
1762 | int64_t delta; | |
1763 | ||
1764 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
1765 | ||
1766 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1767 | used = dsl_dir_phys(dd)->dd_used_bytes; |
1768 | delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved); | |
1769 | dsl_dir_phys(dd)->dd_reserved = value; | |
34dc7c2f BB |
1770 | |
1771 | if (dd->dd_parent != NULL) { | |
1772 | /* Roll up this additional usage into our ancestors */ | |
b128c09f BB |
1773 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, |
1774 | delta, 0, 0, tx); | |
34dc7c2f | 1775 | } |
b128c09f | 1776 | mutex_exit(&dd->dd_lock); |
34dc7c2f BB |
1777 | } |
1778 | ||
6f1ffb06 | 1779 | static void |
13fe0198 | 1780 | dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) |
6f1ffb06 | 1781 | { |
13fe0198 MA |
1782 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1783 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1784 | dsl_dataset_t *ds; | |
1785 | uint64_t newval; | |
6f1ffb06 | 1786 | |
13fe0198 MA |
1787 | VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); |
1788 | ||
b1118acb MM |
1789 | if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { |
1790 | dsl_prop_set_sync_impl(ds, | |
1791 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1792 | ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, | |
1793 | &ddsqra->ddsqra_value, tx); | |
d1d7e268 | 1794 | |
b1118acb MM |
1795 | VERIFY0(dsl_prop_get_int_ds(ds, |
1796 | zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); | |
1797 | } else { | |
1798 | newval = ddsqra->ddsqra_value; | |
1799 | spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", | |
1800 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1801 | (longlong_t)newval); | |
1802 | } | |
1803 | ||
13fe0198 MA |
1804 | dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); |
1805 | dsl_dataset_rele(ds, FTAG); | |
d1d7e268 | 1806 | } |
6f1ffb06 | 1807 | |
34dc7c2f | 1808 | int |
428870ff BB |
1809 | dsl_dir_set_reservation(const char *ddname, zprop_source_t source, |
1810 | uint64_t reservation) | |
34dc7c2f | 1811 | { |
13fe0198 | 1812 | dsl_dir_set_qr_arg_t ddsqra; |
428870ff | 1813 | |
13fe0198 MA |
1814 | ddsqra.ddsqra_name = ddname; |
1815 | ddsqra.ddsqra_source = source; | |
1816 | ddsqra.ddsqra_value = reservation; | |
428870ff | 1817 | |
13fe0198 | 1818 | return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, |
d2734cce SD |
1819 | dsl_dir_set_reservation_sync, &ddsqra, 0, |
1820 | ZFS_SPACE_CHECK_EXTRA_RESERVED)); | |
34dc7c2f BB |
1821 | } |
1822 | ||
1823 | static dsl_dir_t * | |
1824 | closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) | |
1825 | { | |
1826 | for (; ds1; ds1 = ds1->dd_parent) { | |
1827 | dsl_dir_t *dd; | |
1828 | for (dd = ds2; dd; dd = dd->dd_parent) { | |
1829 | if (ds1 == dd) | |
1830 | return (dd); | |
1831 | } | |
1832 | } | |
1833 | return (NULL); | |
1834 | } | |
1835 | ||
1836 | /* | |
1837 | * If delta is applied to dd, how much of that delta would be applied to | |
1838 | * ancestor? Syncing context only. | |
1839 | */ | |
1840 | static int64_t | |
1841 | would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) | |
1842 | { | |
1843 | if (dd == ancestor) | |
1844 | return (delta); | |
1845 | ||
1846 | mutex_enter(&dd->dd_lock); | |
d683ddbb | 1847 | delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta); |
34dc7c2f BB |
1848 | mutex_exit(&dd->dd_lock); |
1849 | return (would_change(dd->dd_parent, delta, ancestor)); | |
1850 | } | |
1851 | ||
13fe0198 MA |
1852 | typedef struct dsl_dir_rename_arg { |
1853 | const char *ddra_oldname; | |
1854 | const char *ddra_newname; | |
788eb90c | 1855 | cred_t *ddra_cred; |
13fe0198 | 1856 | } dsl_dir_rename_arg_t; |
34dc7c2f | 1857 | |
a7ed98d8 SD |
1858 | typedef struct dsl_valid_rename_arg { |
1859 | int char_delta; | |
1860 | int nest_delta; | |
1861 | } dsl_valid_rename_arg_t; | |
1862 | ||
13fe0198 | 1863 | /* ARGSUSED */ |
34dc7c2f | 1864 | static int |
13fe0198 | 1865 | dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
34dc7c2f | 1866 | { |
a7ed98d8 | 1867 | dsl_valid_rename_arg_t *dvra = arg; |
eca7b760 | 1868 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; |
34dc7c2f | 1869 | |
13fe0198 MA |
1870 | dsl_dataset_name(ds, namebuf); |
1871 | ||
a7ed98d8 SD |
1872 | ASSERT3U(strnlen(namebuf, ZFS_MAX_DATASET_NAME_LEN), |
1873 | <, ZFS_MAX_DATASET_NAME_LEN); | |
1874 | int namelen = strlen(namebuf) + dvra->char_delta; | |
1875 | int depth = get_dataset_depth(namebuf) + dvra->nest_delta; | |
1876 | ||
1877 | if (namelen >= ZFS_MAX_DATASET_NAME_LEN) | |
1878 | return (SET_ERROR(ENAMETOOLONG)); | |
1879 | if (dvra->nest_delta > 0 && depth >= zfs_max_dataset_nesting) | |
2e528b49 | 1880 | return (SET_ERROR(ENAMETOOLONG)); |
13fe0198 MA |
1881 | return (0); |
1882 | } | |
1883 | ||
1884 | static int | |
1885 | dsl_dir_rename_check(void *arg, dmu_tx_t *tx) | |
1886 | { | |
1887 | dsl_dir_rename_arg_t *ddra = arg; | |
1888 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1889 | dsl_dir_t *dd, *newparent; | |
a7ed98d8 | 1890 | dsl_valid_rename_arg_t dvra; |
13fe0198 MA |
1891 | const char *mynewname; |
1892 | int error; | |
34dc7c2f | 1893 | |
13fe0198 MA |
1894 | /* target dir should exist */ |
1895 | error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); | |
1896 | if (error != 0) | |
1897 | return (error); | |
1898 | ||
1899 | /* new parent should exist */ | |
1900 | error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, | |
1901 | &newparent, &mynewname); | |
1902 | if (error != 0) { | |
1903 | dsl_dir_rele(dd, FTAG); | |
1904 | return (error); | |
1905 | } | |
1906 | ||
1907 | /* can't rename to different pool */ | |
1908 | if (dd->dd_pool != newparent->dd_pool) { | |
1909 | dsl_dir_rele(newparent, FTAG); | |
1910 | dsl_dir_rele(dd, FTAG); | |
9063f654 | 1911 | return (SET_ERROR(EXDEV)); |
13fe0198 MA |
1912 | } |
1913 | ||
1914 | /* new name should not already exist */ | |
1915 | if (mynewname == NULL) { | |
1916 | dsl_dir_rele(newparent, FTAG); | |
1917 | dsl_dir_rele(dd, FTAG); | |
2e528b49 | 1918 | return (SET_ERROR(EEXIST)); |
13fe0198 MA |
1919 | } |
1920 | ||
a7ed98d8 SD |
1921 | ASSERT3U(strnlen(ddra->ddra_newname, ZFS_MAX_DATASET_NAME_LEN), |
1922 | <, ZFS_MAX_DATASET_NAME_LEN); | |
1923 | ASSERT3U(strnlen(ddra->ddra_oldname, ZFS_MAX_DATASET_NAME_LEN), | |
1924 | <, ZFS_MAX_DATASET_NAME_LEN); | |
1925 | dvra.char_delta = strlen(ddra->ddra_newname) | |
1926 | - strlen(ddra->ddra_oldname); | |
1927 | dvra.nest_delta = get_dataset_depth(ddra->ddra_newname) | |
1928 | - get_dataset_depth(ddra->ddra_oldname); | |
1929 | ||
13fe0198 | 1930 | /* if the name length is growing, validate child name lengths */ |
a7ed98d8 | 1931 | if (dvra.char_delta > 0 || dvra.nest_delta > 0) { |
13fe0198 | 1932 | error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, |
a7ed98d8 | 1933 | &dvra, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); |
13fe0198 MA |
1934 | if (error != 0) { |
1935 | dsl_dir_rele(newparent, FTAG); | |
1936 | dsl_dir_rele(dd, FTAG); | |
1937 | return (error); | |
1938 | } | |
1939 | } | |
34dc7c2f | 1940 | |
788eb90c | 1941 | if (dmu_tx_is_syncing(tx)) { |
a0c9a17a | 1942 | if (spa_feature_is_active(dp->dp_spa, |
788eb90c JJ |
1943 | SPA_FEATURE_FS_SS_LIMIT)) { |
1944 | /* | |
1945 | * Although this is the check function and we don't | |
1946 | * normally make on-disk changes in check functions, | |
1947 | * we need to do that here. | |
1948 | * | |
1949 | * Ensure this portion of the tree's counts have been | |
1950 | * initialized in case the new parent has limits set. | |
1951 | */ | |
1952 | dsl_dir_init_fs_ss_count(dd, tx); | |
1953 | } | |
1954 | } | |
1955 | ||
13fe0198 | 1956 | if (newparent != dd->dd_parent) { |
34dc7c2f BB |
1957 | /* is there enough space? */ |
1958 | uint64_t myspace = | |
d683ddbb JG |
1959 | MAX(dsl_dir_phys(dd)->dd_used_bytes, |
1960 | dsl_dir_phys(dd)->dd_reserved); | |
788eb90c JJ |
1961 | objset_t *os = dd->dd_pool->dp_meta_objset; |
1962 | uint64_t fs_cnt = 0; | |
1963 | uint64_t ss_cnt = 0; | |
1964 | ||
1965 | if (dsl_dir_is_zapified(dd)) { | |
1966 | int err; | |
1967 | ||
1968 | err = zap_lookup(os, dd->dd_object, | |
1969 | DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, | |
1970 | &fs_cnt); | |
a0c9a17a JJ |
1971 | if (err != ENOENT && err != 0) { |
1972 | dsl_dir_rele(newparent, FTAG); | |
1973 | dsl_dir_rele(dd, FTAG); | |
788eb90c | 1974 | return (err); |
a0c9a17a | 1975 | } |
788eb90c JJ |
1976 | |
1977 | /* | |
1978 | * have to add 1 for the filesystem itself that we're | |
1979 | * moving | |
1980 | */ | |
1981 | fs_cnt++; | |
1982 | ||
1983 | err = zap_lookup(os, dd->dd_object, | |
1984 | DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, | |
1985 | &ss_cnt); | |
a0c9a17a JJ |
1986 | if (err != ENOENT && err != 0) { |
1987 | dsl_dir_rele(newparent, FTAG); | |
1988 | dsl_dir_rele(dd, FTAG); | |
788eb90c | 1989 | return (err); |
a0c9a17a | 1990 | } |
788eb90c | 1991 | } |
34dc7c2f | 1992 | |
b5256303 TC |
1993 | /* check for encryption errors */ |
1994 | error = dsl_dir_rename_crypt_check(dd, newparent); | |
1995 | if (error != 0) { | |
1996 | dsl_dir_rele(newparent, FTAG); | |
1997 | dsl_dir_rele(dd, FTAG); | |
1998 | return (SET_ERROR(EACCES)); | |
1999 | } | |
2000 | ||
34dc7c2f | 2001 | /* no rename into our descendant */ |
13fe0198 MA |
2002 | if (closest_common_ancestor(dd, newparent) == dd) { |
2003 | dsl_dir_rele(newparent, FTAG); | |
2004 | dsl_dir_rele(dd, FTAG); | |
2e528b49 | 2005 | return (SET_ERROR(EINVAL)); |
13fe0198 | 2006 | } |
34dc7c2f | 2007 | |
13fe0198 | 2008 | error = dsl_dir_transfer_possible(dd->dd_parent, |
788eb90c | 2009 | newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); |
13fe0198 MA |
2010 | if (error != 0) { |
2011 | dsl_dir_rele(newparent, FTAG); | |
2012 | dsl_dir_rele(dd, FTAG); | |
2013 | return (error); | |
2014 | } | |
34dc7c2f BB |
2015 | } |
2016 | ||
13fe0198 MA |
2017 | dsl_dir_rele(newparent, FTAG); |
2018 | dsl_dir_rele(dd, FTAG); | |
34dc7c2f BB |
2019 | return (0); |
2020 | } | |
2021 | ||
2022 | static void | |
13fe0198 | 2023 | dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 2024 | { |
13fe0198 MA |
2025 | dsl_dir_rename_arg_t *ddra = arg; |
2026 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
2027 | dsl_dir_t *dd, *newparent; | |
2028 | const char *mynewname; | |
2029 | int error; | |
34dc7c2f | 2030 | objset_t *mos = dp->dp_meta_objset; |
34dc7c2f | 2031 | |
13fe0198 MA |
2032 | VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); |
2033 | VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, | |
2034 | &mynewname)); | |
34dc7c2f | 2035 | |
6f1ffb06 | 2036 | /* Log this before we change the name. */ |
6f1ffb06 | 2037 | spa_history_log_internal_dd(dd, "rename", tx, |
13fe0198 | 2038 | "-> %s", ddra->ddra_newname); |
6f1ffb06 | 2039 | |
13fe0198 | 2040 | if (newparent != dd->dd_parent) { |
788eb90c JJ |
2041 | objset_t *os = dd->dd_pool->dp_meta_objset; |
2042 | uint64_t fs_cnt = 0; | |
2043 | uint64_t ss_cnt = 0; | |
2044 | ||
2045 | /* | |
2046 | * We already made sure the dd counts were initialized in the | |
2047 | * check function. | |
2048 | */ | |
a0c9a17a | 2049 | if (spa_feature_is_active(dp->dp_spa, |
788eb90c JJ |
2050 | SPA_FEATURE_FS_SS_LIMIT)) { |
2051 | VERIFY0(zap_lookup(os, dd->dd_object, | |
2052 | DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, | |
2053 | &fs_cnt)); | |
2054 | /* add 1 for the filesystem itself that we're moving */ | |
2055 | fs_cnt++; | |
2056 | ||
2057 | VERIFY0(zap_lookup(os, dd->dd_object, | |
2058 | DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, | |
2059 | &ss_cnt)); | |
2060 | } | |
2061 | ||
2062 | dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, | |
2063 | DD_FIELD_FILESYSTEM_COUNT, tx); | |
2064 | dsl_fs_ss_count_adjust(newparent, fs_cnt, | |
2065 | DD_FIELD_FILESYSTEM_COUNT, tx); | |
2066 | ||
2067 | dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, | |
2068 | DD_FIELD_SNAPSHOT_COUNT, tx); | |
2069 | dsl_fs_ss_count_adjust(newparent, ss_cnt, | |
2070 | DD_FIELD_SNAPSHOT_COUNT, tx); | |
2071 | ||
b128c09f | 2072 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, |
d683ddbb JG |
2073 | -dsl_dir_phys(dd)->dd_used_bytes, |
2074 | -dsl_dir_phys(dd)->dd_compressed_bytes, | |
2075 | -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); | |
13fe0198 | 2076 | dsl_dir_diduse_space(newparent, DD_USED_CHILD, |
d683ddbb JG |
2077 | dsl_dir_phys(dd)->dd_used_bytes, |
2078 | dsl_dir_phys(dd)->dd_compressed_bytes, | |
2079 | dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); | |
b128c09f | 2080 | |
d683ddbb JG |
2081 | if (dsl_dir_phys(dd)->dd_reserved > |
2082 | dsl_dir_phys(dd)->dd_used_bytes) { | |
2083 | uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved - | |
2084 | dsl_dir_phys(dd)->dd_used_bytes; | |
b128c09f BB |
2085 | |
2086 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, | |
2087 | -unused_rsrv, 0, 0, tx); | |
13fe0198 | 2088 | dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, |
b128c09f BB |
2089 | unused_rsrv, 0, 0, tx); |
2090 | } | |
34dc7c2f BB |
2091 | } |
2092 | ||
2093 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
2094 | ||
2095 | /* remove from old parent zapobj */ | |
d683ddbb JG |
2096 | error = zap_remove(mos, |
2097 | dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj, | |
34dc7c2f | 2098 | dd->dd_myname, tx); |
13fe0198 | 2099 | ASSERT0(error); |
34dc7c2f | 2100 | |
c9d61adb | 2101 | (void) strlcpy(dd->dd_myname, mynewname, |
2102 | sizeof (dd->dd_myname)); | |
13fe0198 | 2103 | dsl_dir_rele(dd->dd_parent, dd); |
d683ddbb | 2104 | dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object; |
13fe0198 MA |
2105 | VERIFY0(dsl_dir_hold_obj(dp, |
2106 | newparent->dd_object, NULL, dd, &dd->dd_parent)); | |
34dc7c2f BB |
2107 | |
2108 | /* add to new parent zapobj */ | |
d683ddbb | 2109 | VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj, |
13fe0198 MA |
2110 | dd->dd_myname, 8, 1, &dd->dd_object, tx)); |
2111 | ||
a0bd735a BP |
2112 | zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname, |
2113 | ddra->ddra_newname, B_TRUE); | |
ba6a2402 | 2114 | |
13fe0198 | 2115 | dsl_prop_notify_all(dd); |
34dc7c2f | 2116 | |
13fe0198 MA |
2117 | dsl_dir_rele(newparent, FTAG); |
2118 | dsl_dir_rele(dd, FTAG); | |
34dc7c2f BB |
2119 | } |
2120 | ||
2121 | int | |
13fe0198 | 2122 | dsl_dir_rename(const char *oldname, const char *newname) |
34dc7c2f | 2123 | { |
13fe0198 | 2124 | dsl_dir_rename_arg_t ddra; |
34dc7c2f | 2125 | |
13fe0198 MA |
2126 | ddra.ddra_oldname = oldname; |
2127 | ddra.ddra_newname = newname; | |
788eb90c | 2128 | ddra.ddra_cred = CRED(); |
34dc7c2f | 2129 | |
13fe0198 | 2130 | return (dsl_sync_task(oldname, |
3d45fdd6 MA |
2131 | dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, |
2132 | 3, ZFS_SPACE_CHECK_RESERVED)); | |
34dc7c2f BB |
2133 | } |
2134 | ||
2135 | int | |
788eb90c JJ |
2136 | dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, |
2137 | uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) | |
34dc7c2f BB |
2138 | { |
2139 | dsl_dir_t *ancestor; | |
2140 | int64_t adelta; | |
2141 | uint64_t avail; | |
788eb90c | 2142 | int err; |
34dc7c2f BB |
2143 | |
2144 | ancestor = closest_common_ancestor(sdd, tdd); | |
2145 | adelta = would_change(sdd, -space, ancestor); | |
2146 | avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); | |
2147 | if (avail < space) | |
2e528b49 | 2148 | return (SET_ERROR(ENOSPC)); |
34dc7c2f | 2149 | |
788eb90c JJ |
2150 | err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, |
2151 | ancestor, cr); | |
2152 | if (err != 0) | |
2153 | return (err); | |
2154 | err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, | |
2155 | ancestor, cr); | |
2156 | if (err != 0) | |
2157 | return (err); | |
2158 | ||
34dc7c2f BB |
2159 | return (0); |
2160 | } | |
428870ff | 2161 | |
6413c95f | 2162 | inode_timespec_t |
428870ff BB |
2163 | dsl_dir_snap_cmtime(dsl_dir_t *dd) |
2164 | { | |
6413c95f | 2165 | inode_timespec_t t; |
428870ff BB |
2166 | |
2167 | mutex_enter(&dd->dd_lock); | |
2168 | t = dd->dd_snap_cmtime; | |
2169 | mutex_exit(&dd->dd_lock); | |
2170 | ||
2171 | return (t); | |
2172 | } | |
2173 | ||
2174 | void | |
2175 | dsl_dir_snap_cmtime_update(dsl_dir_t *dd) | |
2176 | { | |
6413c95f | 2177 | inode_timespec_t t; |
428870ff BB |
2178 | |
2179 | gethrestime(&t); | |
2180 | mutex_enter(&dd->dd_lock); | |
2181 | dd->dd_snap_cmtime = t; | |
2182 | mutex_exit(&dd->dd_lock); | |
2183 | } | |
c28b2279 | 2184 | |
fa86b5db MA |
2185 | void |
2186 | dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) | |
2187 | { | |
2188 | objset_t *mos = dd->dd_pool->dp_meta_objset; | |
2189 | dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); | |
2190 | } | |
2191 | ||
788eb90c JJ |
2192 | boolean_t |
2193 | dsl_dir_is_zapified(dsl_dir_t *dd) | |
2194 | { | |
2195 | dmu_object_info_t doi; | |
2196 | ||
2197 | dmu_object_info_from_db(dd->dd_dbuf, &doi); | |
2198 | return (doi.doi_type == DMU_OTN_ZAP_METADATA); | |
2199 | } | |
2200 | ||
93ce2b4c | 2201 | #if defined(_KERNEL) |
c28b2279 BB |
2202 | EXPORT_SYMBOL(dsl_dir_set_quota); |
2203 | EXPORT_SYMBOL(dsl_dir_set_reservation); | |
c28b2279 | 2204 | #endif |