]>
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. |
3d45fdd6 | 23 | * Copyright (c) 2012, 2014 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. |
34dc7c2f BB |
27 | */ |
28 | ||
34dc7c2f BB |
29 | #include <sys/dmu.h> |
30 | #include <sys/dmu_objset.h> | |
31 | #include <sys/dmu_tx.h> | |
32 | #include <sys/dsl_dataset.h> | |
33 | #include <sys/dsl_dir.h> | |
34 | #include <sys/dsl_prop.h> | |
35 | #include <sys/dsl_synctask.h> | |
36 | #include <sys/dsl_deleg.h> | |
fa86b5db | 37 | #include <sys/dmu_impl.h> |
34dc7c2f | 38 | #include <sys/spa.h> |
428870ff | 39 | #include <sys/metaslab.h> |
34dc7c2f BB |
40 | #include <sys/zap.h> |
41 | #include <sys/zio.h> | |
42 | #include <sys/arc.h> | |
43 | #include <sys/sunddi.h> | |
788eb90c JJ |
44 | #include <sys/zfeature.h> |
45 | #include <sys/policy.h> | |
46 | #include <sys/zfs_znode.h> | |
ba6a2402 | 47 | #include <sys/zvol.h> |
34dc7c2f | 48 | #include "zfs_namecheck.h" |
788eb90c JJ |
49 | #include "zfs_prop.h" |
50 | ||
51 | /* | |
52 | * Filesystem and Snapshot Limits | |
53 | * ------------------------------ | |
54 | * | |
55 | * These limits are used to restrict the number of filesystems and/or snapshots | |
56 | * that can be created at a given level in the tree or below. A typical | |
57 | * use-case is with a delegated dataset where the administrator wants to ensure | |
58 | * that a user within the zone is not creating too many additional filesystems | |
59 | * or snapshots, even though they're not exceeding their space quota. | |
60 | * | |
61 | * The filesystem and snapshot counts are stored as extensible properties. This | |
62 | * capability is controlled by a feature flag and must be enabled to be used. | |
63 | * Once enabled, the feature is not active until the first limit is set. At | |
64 | * that point, future operations to create/destroy filesystems or snapshots | |
65 | * will validate and update the counts. | |
66 | * | |
67 | * Because the count properties will not exist before the feature is active, | |
68 | * the counts are updated when a limit is first set on an uninitialized | |
69 | * dsl_dir node in the tree (The filesystem/snapshot count on a node includes | |
70 | * all of the nested filesystems/snapshots. Thus, a new leaf node has a | |
71 | * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and | |
72 | * snapshot count properties on a node indicate uninitialized counts on that | |
73 | * node.) When first setting a limit on an uninitialized node, the code starts | |
74 | * at the filesystem with the new limit and descends into all sub-filesystems | |
75 | * to add the count properties. | |
76 | * | |
77 | * In practice this is lightweight since a limit is typically set when the | |
78 | * filesystem is created and thus has no children. Once valid, changing the | |
79 | * limit value won't require a re-traversal since the counts are already valid. | |
80 | * When recursively fixing the counts, if a node with a limit is encountered | |
81 | * during the descent, the counts are known to be valid and there is no need to | |
82 | * descend into that filesystem's children. The counts on filesystems above the | |
83 | * one with the new limit will still be uninitialized, unless a limit is | |
84 | * eventually set on one of those filesystems. The counts are always recursively | |
85 | * updated when a limit is set on a dataset, unless there is already a limit. | |
86 | * When a new limit value is set on a filesystem with an existing limit, it is | |
87 | * possible for the new limit to be less than the current count at that level | |
88 | * since a user who can change the limit is also allowed to exceed the limit. | |
89 | * | |
90 | * Once the feature is active, then whenever a filesystem or snapshot is | |
91 | * created, the code recurses up the tree, validating the new count against the | |
92 | * limit at each initialized level. In practice, most levels will not have a | |
93 | * limit set. If there is a limit at any initialized level up the tree, the | |
94 | * check must pass or the creation will fail. Likewise, when a filesystem or | |
95 | * snapshot is destroyed, the counts are recursively adjusted all the way up | |
96 | * the initizized nodes in the tree. Renaming a filesystem into different point | |
97 | * in the tree will first validate, then update the counts on each branch up to | |
98 | * the common ancestor. A receive will also validate the counts and then update | |
99 | * them. | |
100 | * | |
101 | * An exception to the above behavior is that the limit is not enforced if the | |
102 | * user has permission to modify the limit. This is primarily so that | |
103 | * recursive snapshots in the global zone always work. We want to prevent a | |
104 | * denial-of-service in which a lower level delegated dataset could max out its | |
105 | * limit and thus block recursive snapshots from being taken in the global zone. | |
106 | * Because of this, it is possible for the snapshot count to be over the limit | |
107 | * and snapshots taken in the global zone could cause a lower level dataset to | |
108 | * hit or exceed its limit. The administrator taking the global zone recursive | |
109 | * snapshot should be aware of this side-effect and behave accordingly. | |
110 | * For consistency, the filesystem limit is also not enforced if the user can | |
111 | * modify the limit. | |
112 | * | |
113 | * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check() | |
114 | * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in | |
115 | * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by | |
116 | * dsl_dir_init_fs_ss_count(). | |
117 | * | |
118 | * There is a special case when we receive a filesystem that already exists. In | |
119 | * this case a temporary clone name of %X is created (see dmu_recv_begin). We | |
120 | * never update the filesystem counts for temporary clones. | |
121 | * | |
122 | * Likewise, we do not update the snapshot counts for temporary snapshots, | |
123 | * such as those created by zfs diff. | |
124 | */ | |
34dc7c2f | 125 | |
d683ddbb JG |
126 | extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd); |
127 | ||
34dc7c2f | 128 | static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); |
34dc7c2f | 129 | |
34dc7c2f | 130 | static void |
0c66c32d | 131 | dsl_dir_evict(void *dbu) |
34dc7c2f | 132 | { |
0c66c32d | 133 | dsl_dir_t *dd = dbu; |
34dc7c2f | 134 | int t; |
d1d7e268 | 135 | ASSERTV(dsl_pool_t *dp = dd->dd_pool); |
34dc7c2f | 136 | |
0c66c32d JG |
137 | dd->dd_dbuf = NULL; |
138 | ||
34dc7c2f BB |
139 | for (t = 0; t < TXG_SIZE; t++) { |
140 | ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); | |
141 | ASSERT(dd->dd_tempreserved[t] == 0); | |
142 | ASSERT(dd->dd_space_towrite[t] == 0); | |
143 | } | |
144 | ||
34dc7c2f | 145 | if (dd->dd_parent) |
0c66c32d | 146 | dsl_dir_async_rele(dd->dd_parent, dd); |
34dc7c2f | 147 | |
0c66c32d | 148 | spa_async_close(dd->dd_pool->dp_spa, dd); |
34dc7c2f BB |
149 | |
150 | /* | |
428870ff BB |
151 | * The props callback list should have been cleaned up by |
152 | * objset_evict(). | |
34dc7c2f BB |
153 | */ |
154 | list_destroy(&dd->dd_prop_cbs); | |
155 | mutex_destroy(&dd->dd_lock); | |
156 | kmem_free(dd, sizeof (dsl_dir_t)); | |
157 | } | |
158 | ||
159 | int | |
13fe0198 | 160 | dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj, |
34dc7c2f BB |
161 | const char *tail, void *tag, dsl_dir_t **ddp) |
162 | { | |
163 | dmu_buf_t *dbuf; | |
164 | dsl_dir_t *dd; | |
165 | int err; | |
166 | ||
13fe0198 | 167 | ASSERT(dsl_pool_config_held(dp)); |
34dc7c2f BB |
168 | |
169 | err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf); | |
13fe0198 | 170 | if (err != 0) |
34dc7c2f BB |
171 | return (err); |
172 | dd = dmu_buf_get_user(dbuf); | |
173 | #ifdef ZFS_DEBUG | |
174 | { | |
175 | dmu_object_info_t doi; | |
176 | dmu_object_info_from_db(dbuf, &doi); | |
fa86b5db | 177 | ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR); |
b128c09f | 178 | ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); |
34dc7c2f BB |
179 | } |
180 | #endif | |
34dc7c2f BB |
181 | if (dd == NULL) { |
182 | dsl_dir_t *winner; | |
34dc7c2f | 183 | |
79c76d5b | 184 | dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); |
34dc7c2f BB |
185 | dd->dd_object = ddobj; |
186 | dd->dd_dbuf = dbuf; | |
187 | dd->dd_pool = dp; | |
34dc7c2f BB |
188 | mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); |
189 | ||
190 | list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t), | |
191 | offsetof(dsl_prop_cb_record_t, cbr_node)); | |
192 | ||
428870ff BB |
193 | dsl_dir_snap_cmtime_update(dd); |
194 | ||
d683ddbb JG |
195 | if (dsl_dir_phys(dd)->dd_parent_obj) { |
196 | err = dsl_dir_hold_obj(dp, | |
197 | dsl_dir_phys(dd)->dd_parent_obj, NULL, dd, | |
198 | &dd->dd_parent); | |
13fe0198 | 199 | if (err != 0) |
b128c09f | 200 | goto errout; |
34dc7c2f BB |
201 | if (tail) { |
202 | #ifdef ZFS_DEBUG | |
203 | uint64_t foundobj; | |
204 | ||
205 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb JG |
206 | dsl_dir_phys(dd->dd_parent)-> |
207 | dd_child_dir_zapobj, tail, | |
208 | sizeof (foundobj), 1, &foundobj); | |
34dc7c2f BB |
209 | ASSERT(err || foundobj == ddobj); |
210 | #endif | |
211 | (void) strcpy(dd->dd_myname, tail); | |
212 | } else { | |
213 | err = zap_value_search(dp->dp_meta_objset, | |
d683ddbb JG |
214 | dsl_dir_phys(dd->dd_parent)-> |
215 | dd_child_dir_zapobj, | |
34dc7c2f BB |
216 | ddobj, 0, dd->dd_myname); |
217 | } | |
13fe0198 | 218 | if (err != 0) |
b128c09f | 219 | goto errout; |
34dc7c2f BB |
220 | } else { |
221 | (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); | |
222 | } | |
223 | ||
428870ff BB |
224 | if (dsl_dir_is_clone(dd)) { |
225 | dmu_buf_t *origin_bonus; | |
226 | dsl_dataset_phys_t *origin_phys; | |
227 | ||
228 | /* | |
229 | * We can't open the origin dataset, because | |
230 | * that would require opening this dsl_dir. | |
231 | * Just look at its phys directly instead. | |
232 | */ | |
233 | err = dmu_bonus_hold(dp->dp_meta_objset, | |
d683ddbb JG |
234 | dsl_dir_phys(dd)->dd_origin_obj, FTAG, |
235 | &origin_bonus); | |
13fe0198 | 236 | if (err != 0) |
428870ff BB |
237 | goto errout; |
238 | origin_phys = origin_bonus->db_data; | |
239 | dd->dd_origin_txg = | |
240 | origin_phys->ds_creation_txg; | |
241 | dmu_buf_rele(origin_bonus, FTAG); | |
242 | } | |
243 | ||
0c66c32d JG |
244 | dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf); |
245 | winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu); | |
246 | if (winner != NULL) { | |
34dc7c2f | 247 | if (dd->dd_parent) |
13fe0198 | 248 | dsl_dir_rele(dd->dd_parent, dd); |
34dc7c2f BB |
249 | mutex_destroy(&dd->dd_lock); |
250 | kmem_free(dd, sizeof (dsl_dir_t)); | |
251 | dd = winner; | |
252 | } else { | |
253 | spa_open_ref(dp->dp_spa, dd); | |
254 | } | |
255 | } | |
256 | ||
257 | /* | |
258 | * The dsl_dir_t has both open-to-close and instantiate-to-evict | |
259 | * holds on the spa. We need the open-to-close holds because | |
260 | * otherwise the spa_refcnt wouldn't change when we open a | |
261 | * dir which the spa also has open, so we could incorrectly | |
262 | * think it was OK to unload/export/destroy the pool. We need | |
263 | * the instantiate-to-evict hold because the dsl_dir_t has a | |
264 | * pointer to the dd_pool, which has a pointer to the spa_t. | |
265 | */ | |
266 | spa_open_ref(dp->dp_spa, tag); | |
267 | ASSERT3P(dd->dd_pool, ==, dp); | |
268 | ASSERT3U(dd->dd_object, ==, ddobj); | |
269 | ASSERT3P(dd->dd_dbuf, ==, dbuf); | |
270 | *ddp = dd; | |
271 | return (0); | |
b128c09f BB |
272 | |
273 | errout: | |
274 | if (dd->dd_parent) | |
13fe0198 | 275 | dsl_dir_rele(dd->dd_parent, dd); |
b128c09f BB |
276 | mutex_destroy(&dd->dd_lock); |
277 | kmem_free(dd, sizeof (dsl_dir_t)); | |
278 | dmu_buf_rele(dbuf, tag); | |
279 | return (err); | |
34dc7c2f BB |
280 | } |
281 | ||
282 | void | |
13fe0198 | 283 | dsl_dir_rele(dsl_dir_t *dd, void *tag) |
34dc7c2f BB |
284 | { |
285 | dprintf_dd(dd, "%s\n", ""); | |
286 | spa_close(dd->dd_pool->dp_spa, tag); | |
287 | dmu_buf_rele(dd->dd_dbuf, tag); | |
288 | } | |
289 | ||
0c66c32d JG |
290 | /* |
291 | * Remove a reference to the given dsl dir that is being asynchronously | |
292 | * released. Async releases occur from a taskq performing eviction of | |
293 | * dsl datasets and dirs. This process is identical to a normal release | |
294 | * with the exception of using the async API for releasing the reference on | |
295 | * the spa. | |
296 | */ | |
297 | void | |
298 | dsl_dir_async_rele(dsl_dir_t *dd, void *tag) | |
299 | { | |
300 | dprintf_dd(dd, "%s\n", ""); | |
301 | spa_async_close(dd->dd_pool->dp_spa, tag); | |
302 | dmu_buf_rele(dd->dd_dbuf, tag); | |
303 | } | |
304 | ||
34dc7c2f BB |
305 | /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */ |
306 | void | |
307 | dsl_dir_name(dsl_dir_t *dd, char *buf) | |
308 | { | |
309 | if (dd->dd_parent) { | |
310 | dsl_dir_name(dd->dd_parent, buf); | |
311 | (void) strcat(buf, "/"); | |
312 | } else { | |
313 | buf[0] = '\0'; | |
314 | } | |
315 | if (!MUTEX_HELD(&dd->dd_lock)) { | |
316 | /* | |
317 | * recursive mutex so that we can use | |
318 | * dprintf_dd() with dd_lock held | |
319 | */ | |
320 | mutex_enter(&dd->dd_lock); | |
321 | (void) strcat(buf, dd->dd_myname); | |
322 | mutex_exit(&dd->dd_lock); | |
323 | } else { | |
324 | (void) strcat(buf, dd->dd_myname); | |
325 | } | |
326 | } | |
327 | ||
29809a6c | 328 | /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */ |
34dc7c2f BB |
329 | int |
330 | dsl_dir_namelen(dsl_dir_t *dd) | |
331 | { | |
332 | int result = 0; | |
333 | ||
334 | if (dd->dd_parent) { | |
335 | /* parent's name + 1 for the "/" */ | |
336 | result = dsl_dir_namelen(dd->dd_parent) + 1; | |
337 | } | |
338 | ||
339 | if (!MUTEX_HELD(&dd->dd_lock)) { | |
340 | /* see dsl_dir_name */ | |
341 | mutex_enter(&dd->dd_lock); | |
342 | result += strlen(dd->dd_myname); | |
343 | mutex_exit(&dd->dd_lock); | |
344 | } else { | |
345 | result += strlen(dd->dd_myname); | |
346 | } | |
347 | ||
348 | return (result); | |
349 | } | |
350 | ||
34dc7c2f BB |
351 | static int |
352 | getcomponent(const char *path, char *component, const char **nextp) | |
353 | { | |
354 | char *p; | |
13fe0198 | 355 | |
9babb374 | 356 | if ((path == NULL) || (path[0] == '\0')) |
2e528b49 | 357 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
358 | /* This would be a good place to reserve some namespace... */ |
359 | p = strpbrk(path, "/@"); | |
360 | if (p && (p[1] == '/' || p[1] == '@')) { | |
361 | /* two separators in a row */ | |
2e528b49 | 362 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
363 | } |
364 | if (p == NULL || p == path) { | |
365 | /* | |
366 | * if the first thing is an @ or /, it had better be an | |
367 | * @ and it had better not have any more ats or slashes, | |
368 | * and it had better have something after the @. | |
369 | */ | |
370 | if (p != NULL && | |
371 | (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0')) | |
2e528b49 | 372 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 373 | if (strlen(path) >= MAXNAMELEN) |
2e528b49 | 374 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
375 | (void) strcpy(component, path); |
376 | p = NULL; | |
377 | } else if (p[0] == '/') { | |
13fe0198 | 378 | if (p - path >= MAXNAMELEN) |
2e528b49 | 379 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f | 380 | (void) strncpy(component, path, p - path); |
13fe0198 | 381 | component[p - path] = '\0'; |
34dc7c2f BB |
382 | p++; |
383 | } else if (p[0] == '@') { | |
384 | /* | |
385 | * if the next separator is an @, there better not be | |
386 | * any more slashes. | |
387 | */ | |
388 | if (strchr(path, '/')) | |
2e528b49 | 389 | return (SET_ERROR(EINVAL)); |
13fe0198 | 390 | if (p - path >= MAXNAMELEN) |
2e528b49 | 391 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f | 392 | (void) strncpy(component, path, p - path); |
13fe0198 | 393 | component[p - path] = '\0'; |
34dc7c2f | 394 | } else { |
13fe0198 | 395 | panic("invalid p=%p", (void *)p); |
34dc7c2f BB |
396 | } |
397 | *nextp = p; | |
398 | return (0); | |
399 | } | |
400 | ||
401 | /* | |
13fe0198 MA |
402 | * Return the dsl_dir_t, and possibly the last component which couldn't |
403 | * be found in *tail. The name must be in the specified dsl_pool_t. This | |
404 | * thread must hold the dp_config_rwlock for the pool. Returns NULL if the | |
405 | * path is bogus, or if tail==NULL and we couldn't parse the whole name. | |
406 | * (*tail)[0] == '@' means that the last component is a snapshot. | |
34dc7c2f BB |
407 | */ |
408 | int | |
13fe0198 | 409 | dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag, |
34dc7c2f BB |
410 | dsl_dir_t **ddp, const char **tailp) |
411 | { | |
fcf37ec6 | 412 | char *buf; |
13fe0198 | 413 | const char *spaname, *next, *nextnext = NULL; |
34dc7c2f BB |
414 | int err; |
415 | dsl_dir_t *dd; | |
34dc7c2f | 416 | uint64_t ddobj; |
34dc7c2f | 417 | |
79c76d5b | 418 | buf = kmem_alloc(MAXNAMELEN, KM_SLEEP); |
34dc7c2f | 419 | err = getcomponent(name, buf, &next); |
13fe0198 | 420 | if (err != 0) |
fcf37ec6 | 421 | goto error; |
34dc7c2f | 422 | |
13fe0198 MA |
423 | /* Make sure the name is in the specified pool. */ |
424 | spaname = spa_name(dp->dp_spa); | |
425 | if (strcmp(buf, spaname) != 0) { | |
9063f654 | 426 | err = SET_ERROR(EXDEV); |
13fe0198 | 427 | goto error; |
34dc7c2f BB |
428 | } |
429 | ||
13fe0198 | 430 | ASSERT(dsl_pool_config_held(dp)); |
34dc7c2f | 431 | |
13fe0198 MA |
432 | err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd); |
433 | if (err != 0) { | |
fcf37ec6 | 434 | goto error; |
34dc7c2f BB |
435 | } |
436 | ||
437 | while (next != NULL) { | |
0c66c32d | 438 | dsl_dir_t *child_dd; |
34dc7c2f | 439 | err = getcomponent(next, buf, &nextnext); |
13fe0198 | 440 | if (err != 0) |
34dc7c2f BB |
441 | break; |
442 | ASSERT(next[0] != '\0'); | |
443 | if (next[0] == '@') | |
444 | break; | |
445 | dprintf("looking up %s in obj%lld\n", | |
d683ddbb | 446 | buf, dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
447 | |
448 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb | 449 | dsl_dir_phys(dd)->dd_child_dir_zapobj, |
34dc7c2f | 450 | buf, sizeof (ddobj), 1, &ddobj); |
13fe0198 | 451 | if (err != 0) { |
34dc7c2f BB |
452 | if (err == ENOENT) |
453 | err = 0; | |
454 | break; | |
455 | } | |
456 | ||
0c66c32d | 457 | err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd); |
13fe0198 | 458 | if (err != 0) |
34dc7c2f | 459 | break; |
13fe0198 | 460 | dsl_dir_rele(dd, tag); |
0c66c32d | 461 | dd = child_dd; |
34dc7c2f BB |
462 | next = nextnext; |
463 | } | |
34dc7c2f | 464 | |
13fe0198 MA |
465 | if (err != 0) { |
466 | dsl_dir_rele(dd, tag); | |
fcf37ec6 | 467 | goto error; |
34dc7c2f BB |
468 | } |
469 | ||
470 | /* | |
471 | * It's an error if there's more than one component left, or | |
472 | * tailp==NULL and there's any component left. | |
473 | */ | |
474 | if (next != NULL && | |
475 | (tailp == NULL || (nextnext && nextnext[0] != '\0'))) { | |
476 | /* bad path name */ | |
13fe0198 | 477 | dsl_dir_rele(dd, tag); |
34dc7c2f | 478 | dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp); |
2e528b49 | 479 | err = SET_ERROR(ENOENT); |
34dc7c2f | 480 | } |
13fe0198 | 481 | if (tailp != NULL) |
34dc7c2f | 482 | *tailp = next; |
34dc7c2f | 483 | *ddp = dd; |
fcf37ec6 BB |
484 | error: |
485 | kmem_free(buf, MAXNAMELEN); | |
34dc7c2f BB |
486 | return (err); |
487 | } | |
488 | ||
788eb90c JJ |
489 | /* |
490 | * If the counts are already initialized for this filesystem and its | |
491 | * descendants then do nothing, otherwise initialize the counts. | |
492 | * | |
493 | * The counts on this filesystem, and those below, may be uninitialized due to | |
494 | * either the use of a pre-existing pool which did not support the | |
495 | * filesystem/snapshot limit feature, or one in which the feature had not yet | |
496 | * been enabled. | |
497 | * | |
498 | * Recursively descend the filesystem tree and update the filesystem/snapshot | |
499 | * counts on each filesystem below, then update the cumulative count on the | |
500 | * current filesystem. If the filesystem already has a count set on it, | |
501 | * then we know that its counts, and the counts on the filesystems below it, | |
502 | * are already correct, so we don't have to update this filesystem. | |
503 | */ | |
504 | static void | |
505 | dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx) | |
506 | { | |
507 | uint64_t my_fs_cnt = 0; | |
508 | uint64_t my_ss_cnt = 0; | |
509 | dsl_pool_t *dp = dd->dd_pool; | |
510 | objset_t *os = dp->dp_meta_objset; | |
511 | zap_cursor_t *zc; | |
512 | zap_attribute_t *za; | |
513 | dsl_dataset_t *ds; | |
514 | ||
a0c9a17a | 515 | ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)); |
788eb90c JJ |
516 | ASSERT(dsl_pool_config_held(dp)); |
517 | ASSERT(dmu_tx_is_syncing(tx)); | |
518 | ||
519 | dsl_dir_zapify(dd, tx); | |
520 | ||
521 | /* | |
522 | * If the filesystem count has already been initialized then we | |
523 | * don't need to recurse down any further. | |
524 | */ | |
525 | if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0) | |
526 | return; | |
527 | ||
528 | zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); | |
529 | za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); | |
530 | ||
531 | /* Iterate my child dirs */ | |
d683ddbb | 532 | for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj); |
788eb90c JJ |
533 | zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) { |
534 | dsl_dir_t *chld_dd; | |
535 | uint64_t count; | |
536 | ||
537 | VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG, | |
538 | &chld_dd)); | |
539 | ||
540 | /* | |
541 | * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and | |
542 | * temporary datasets. | |
543 | */ | |
544 | if (chld_dd->dd_myname[0] == '$' || | |
545 | chld_dd->dd_myname[0] == '%') { | |
546 | dsl_dir_rele(chld_dd, FTAG); | |
547 | continue; | |
548 | } | |
549 | ||
550 | my_fs_cnt++; /* count this child */ | |
551 | ||
552 | dsl_dir_init_fs_ss_count(chld_dd, tx); | |
553 | ||
554 | VERIFY0(zap_lookup(os, chld_dd->dd_object, | |
555 | DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count)); | |
556 | my_fs_cnt += count; | |
557 | VERIFY0(zap_lookup(os, chld_dd->dd_object, | |
558 | DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count)); | |
559 | my_ss_cnt += count; | |
560 | ||
561 | dsl_dir_rele(chld_dd, FTAG); | |
562 | } | |
563 | zap_cursor_fini(zc); | |
564 | /* Count my snapshots (we counted children's snapshots above) */ | |
565 | VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, | |
d683ddbb | 566 | dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds)); |
788eb90c | 567 | |
d683ddbb | 568 | for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj); |
788eb90c JJ |
569 | zap_cursor_retrieve(zc, za) == 0; |
570 | zap_cursor_advance(zc)) { | |
571 | /* Don't count temporary snapshots */ | |
572 | if (za->za_name[0] != '%') | |
573 | my_ss_cnt++; | |
574 | } | |
ca227e54 | 575 | zap_cursor_fini(zc); |
788eb90c JJ |
576 | |
577 | dsl_dataset_rele(ds, FTAG); | |
578 | ||
579 | kmem_free(zc, sizeof (zap_cursor_t)); | |
580 | kmem_free(za, sizeof (zap_attribute_t)); | |
581 | ||
582 | /* we're in a sync task, update counts */ | |
583 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
584 | VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, | |
585 | sizeof (my_fs_cnt), 1, &my_fs_cnt, tx)); | |
586 | VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, | |
587 | sizeof (my_ss_cnt), 1, &my_ss_cnt, tx)); | |
588 | } | |
589 | ||
590 | static int | |
591 | dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx) | |
592 | { | |
593 | char *ddname = (char *)arg; | |
594 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
595 | dsl_dataset_t *ds; | |
596 | dsl_dir_t *dd; | |
597 | int error; | |
598 | ||
599 | error = dsl_dataset_hold(dp, ddname, FTAG, &ds); | |
600 | if (error != 0) | |
601 | return (error); | |
602 | ||
603 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { | |
604 | dsl_dataset_rele(ds, FTAG); | |
605 | return (SET_ERROR(ENOTSUP)); | |
606 | } | |
607 | ||
608 | dd = ds->ds_dir; | |
609 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) && | |
610 | dsl_dir_is_zapified(dd) && | |
611 | zap_contains(dp->dp_meta_objset, dd->dd_object, | |
612 | DD_FIELD_FILESYSTEM_COUNT) == 0) { | |
613 | dsl_dataset_rele(ds, FTAG); | |
614 | return (SET_ERROR(EALREADY)); | |
615 | } | |
616 | ||
617 | dsl_dataset_rele(ds, FTAG); | |
618 | return (0); | |
619 | } | |
620 | ||
621 | static void | |
622 | dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx) | |
623 | { | |
624 | char *ddname = (char *)arg; | |
625 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
626 | dsl_dataset_t *ds; | |
627 | spa_t *spa; | |
628 | ||
629 | VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds)); | |
630 | ||
631 | spa = dsl_dataset_get_spa(ds); | |
632 | ||
633 | if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) { | |
634 | /* | |
635 | * Since the feature was not active and we're now setting a | |
636 | * limit, increment the feature-active counter so that the | |
637 | * feature becomes active for the first time. | |
638 | * | |
639 | * We are already in a sync task so we can update the MOS. | |
640 | */ | |
641 | spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx); | |
642 | } | |
643 | ||
644 | /* | |
645 | * Since we are now setting a non-UINT64_MAX limit on the filesystem, | |
646 | * we need to ensure the counts are correct. Descend down the tree from | |
647 | * this point and update all of the counts to be accurate. | |
648 | */ | |
649 | dsl_dir_init_fs_ss_count(ds->ds_dir, tx); | |
650 | ||
651 | dsl_dataset_rele(ds, FTAG); | |
652 | } | |
653 | ||
654 | /* | |
655 | * Make sure the feature is enabled and activate it if necessary. | |
656 | * Since we're setting a limit, ensure the on-disk counts are valid. | |
657 | * This is only called by the ioctl path when setting a limit value. | |
658 | * | |
659 | * We do not need to validate the new limit, since users who can change the | |
660 | * limit are also allowed to exceed the limit. | |
661 | */ | |
662 | int | |
663 | dsl_dir_activate_fs_ss_limit(const char *ddname) | |
664 | { | |
665 | int error; | |
666 | ||
667 | error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check, | |
3d45fdd6 MA |
668 | dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0, |
669 | ZFS_SPACE_CHECK_RESERVED); | |
788eb90c JJ |
670 | |
671 | if (error == EALREADY) | |
672 | error = 0; | |
673 | ||
674 | return (error); | |
675 | } | |
676 | ||
677 | /* | |
678 | * Used to determine if the filesystem_limit or snapshot_limit should be | |
679 | * enforced. We allow the limit to be exceeded if the user has permission to | |
680 | * write the property value. We pass in the creds that we got in the open | |
681 | * context since we will always be the GZ root in syncing context. We also have | |
682 | * to handle the case where we are allowed to change the limit on the current | |
683 | * dataset, but there may be another limit in the tree above. | |
684 | * | |
685 | * We can never modify these two properties within a non-global zone. In | |
686 | * addition, the other checks are modeled on zfs_secpolicy_write_perms. We | |
687 | * can't use that function since we are already holding the dp_config_rwlock. | |
688 | * In addition, we already have the dd and dealing with snapshots is simplified | |
689 | * in this code. | |
690 | */ | |
691 | ||
692 | typedef enum { | |
693 | ENFORCE_ALWAYS, | |
694 | ENFORCE_NEVER, | |
695 | ENFORCE_ABOVE | |
696 | } enforce_res_t; | |
697 | ||
698 | static enforce_res_t | |
699 | dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr) | |
700 | { | |
701 | enforce_res_t enforce = ENFORCE_ALWAYS; | |
702 | uint64_t obj; | |
703 | dsl_dataset_t *ds; | |
704 | uint64_t zoned; | |
705 | ||
706 | ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || | |
707 | prop == ZFS_PROP_SNAPSHOT_LIMIT); | |
708 | ||
709 | #ifdef _KERNEL | |
710 | if (crgetzoneid(cr) != GLOBAL_ZONEID) | |
711 | return (ENFORCE_ALWAYS); | |
712 | ||
713 | if (secpolicy_zfs(cr) == 0) | |
714 | return (ENFORCE_NEVER); | |
715 | #endif | |
716 | ||
d683ddbb | 717 | if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0) |
788eb90c JJ |
718 | return (ENFORCE_ALWAYS); |
719 | ||
720 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
721 | ||
722 | if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0) | |
723 | return (ENFORCE_ALWAYS); | |
724 | ||
725 | if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) { | |
726 | /* Only root can access zoned fs's from the GZ */ | |
727 | enforce = ENFORCE_ALWAYS; | |
728 | } else { | |
729 | if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0) | |
730 | enforce = ENFORCE_ABOVE; | |
731 | } | |
732 | ||
733 | dsl_dataset_rele(ds, FTAG); | |
734 | return (enforce); | |
735 | } | |
736 | ||
737 | /* | |
738 | * Check if adding additional child filesystem(s) would exceed any filesystem | |
739 | * limits or adding additional snapshot(s) would exceed any snapshot limits. | |
740 | * The prop argument indicates which limit to check. | |
741 | * | |
742 | * Note that all filesystem limits up to the root (or the highest | |
743 | * initialized) filesystem or the given ancestor must be satisfied. | |
744 | */ | |
745 | int | |
746 | dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop, | |
747 | dsl_dir_t *ancestor, cred_t *cr) | |
748 | { | |
749 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
750 | uint64_t limit, count; | |
751 | char *count_prop; | |
752 | enforce_res_t enforce; | |
753 | int err = 0; | |
754 | ||
755 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
756 | ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT || | |
757 | prop == ZFS_PROP_SNAPSHOT_LIMIT); | |
758 | ||
759 | /* | |
760 | * If we're allowed to change the limit, don't enforce the limit | |
761 | * e.g. this can happen if a snapshot is taken by an administrative | |
762 | * user in the global zone (i.e. a recursive snapshot by root). | |
763 | * However, we must handle the case of delegated permissions where we | |
764 | * are allowed to change the limit on the current dataset, but there | |
765 | * is another limit in the tree above. | |
766 | */ | |
767 | enforce = dsl_enforce_ds_ss_limits(dd, prop, cr); | |
768 | if (enforce == ENFORCE_NEVER) | |
769 | return (0); | |
770 | ||
771 | /* | |
772 | * e.g. if renaming a dataset with no snapshots, count adjustment | |
773 | * is 0. | |
774 | */ | |
775 | if (delta == 0) | |
776 | return (0); | |
777 | ||
778 | if (prop == ZFS_PROP_SNAPSHOT_LIMIT) { | |
779 | /* | |
780 | * We don't enforce the limit for temporary snapshots. This is | |
781 | * indicated by a NULL cred_t argument. | |
782 | */ | |
783 | if (cr == NULL) | |
784 | return (0); | |
785 | ||
786 | count_prop = DD_FIELD_SNAPSHOT_COUNT; | |
787 | } else { | |
788 | count_prop = DD_FIELD_FILESYSTEM_COUNT; | |
789 | } | |
790 | ||
791 | /* | |
792 | * If an ancestor has been provided, stop checking the limit once we | |
793 | * hit that dir. We need this during rename so that we don't overcount | |
794 | * the check once we recurse up to the common ancestor. | |
795 | */ | |
796 | if (ancestor == dd) | |
797 | return (0); | |
798 | ||
799 | /* | |
800 | * If we hit an uninitialized node while recursing up the tree, we can | |
801 | * stop since we know there is no limit here (or above). The counts are | |
802 | * not valid on this node and we know we won't touch this node's counts. | |
803 | */ | |
804 | if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object, | |
805 | count_prop, sizeof (count), 1, &count) == ENOENT) | |
806 | return (0); | |
807 | ||
808 | err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL, | |
809 | B_FALSE); | |
810 | if (err != 0) | |
811 | return (err); | |
812 | ||
813 | /* Is there a limit which we've hit? */ | |
814 | if (enforce == ENFORCE_ALWAYS && (count + delta) > limit) | |
815 | return (SET_ERROR(EDQUOT)); | |
816 | ||
817 | if (dd->dd_parent != NULL) | |
818 | err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop, | |
819 | ancestor, cr); | |
820 | ||
821 | return (err); | |
822 | } | |
823 | ||
824 | /* | |
825 | * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all | |
826 | * parents. When a new filesystem/snapshot is created, increment the count on | |
827 | * all parents, and when a filesystem/snapshot is destroyed, decrement the | |
828 | * count. | |
829 | */ | |
830 | void | |
831 | dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop, | |
832 | dmu_tx_t *tx) | |
833 | { | |
834 | int err; | |
835 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
836 | uint64_t count; | |
837 | ||
838 | ASSERT(dsl_pool_config_held(dd->dd_pool)); | |
839 | ASSERT(dmu_tx_is_syncing(tx)); | |
840 | ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 || | |
841 | strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0); | |
842 | ||
843 | /* | |
844 | * When we receive an incremental stream into a filesystem that already | |
845 | * exists, a temporary clone is created. We don't count this temporary | |
846 | * clone, whose name begins with a '%'. We also ignore hidden ($FREE, | |
847 | * $MOS & $ORIGIN) objsets. | |
848 | */ | |
849 | if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') && | |
850 | strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0) | |
851 | return; | |
852 | ||
853 | /* | |
854 | * e.g. if renaming a dataset with no snapshots, count adjustment is 0 | |
855 | */ | |
856 | if (delta == 0) | |
857 | return; | |
858 | ||
859 | /* | |
860 | * If we hit an uninitialized node while recursing up the tree, we can | |
861 | * stop since we know the counts are not valid on this node and we | |
862 | * know we shouldn't touch this node's counts. An uninitialized count | |
863 | * on the node indicates that either the feature has not yet been | |
864 | * activated or there are no limits on this part of the tree. | |
865 | */ | |
866 | if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object, | |
867 | prop, sizeof (count), 1, &count)) == ENOENT) | |
868 | return; | |
869 | VERIFY0(err); | |
870 | ||
871 | count += delta; | |
872 | /* Use a signed verify to make sure we're not neg. */ | |
873 | VERIFY3S(count, >=, 0); | |
874 | ||
875 | VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count, | |
876 | tx)); | |
877 | ||
878 | /* Roll up this additional count into our ancestors */ | |
879 | if (dd->dd_parent != NULL) | |
880 | dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx); | |
881 | } | |
882 | ||
34dc7c2f | 883 | uint64_t |
b128c09f BB |
884 | dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, |
885 | dmu_tx_t *tx) | |
34dc7c2f | 886 | { |
b128c09f | 887 | objset_t *mos = dp->dp_meta_objset; |
34dc7c2f | 888 | uint64_t ddobj; |
428870ff | 889 | dsl_dir_phys_t *ddphys; |
34dc7c2f BB |
890 | dmu_buf_t *dbuf; |
891 | ||
892 | ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, | |
893 | DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); | |
b128c09f | 894 | if (pds) { |
d683ddbb | 895 | VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj, |
b128c09f BB |
896 | name, sizeof (uint64_t), 1, &ddobj, tx)); |
897 | } else { | |
898 | /* it's the root dir */ | |
899 | VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, | |
900 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); | |
901 | } | |
34dc7c2f BB |
902 | VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); |
903 | dmu_buf_will_dirty(dbuf, tx); | |
428870ff | 904 | ddphys = dbuf->db_data; |
34dc7c2f | 905 | |
428870ff | 906 | ddphys->dd_creation_time = gethrestime_sec(); |
788eb90c | 907 | if (pds) { |
428870ff | 908 | ddphys->dd_parent_obj = pds->dd_object; |
788eb90c JJ |
909 | |
910 | /* update the filesystem counts */ | |
911 | dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx); | |
912 | } | |
428870ff | 913 | ddphys->dd_props_zapobj = zap_create(mos, |
34dc7c2f | 914 | DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); |
428870ff | 915 | ddphys->dd_child_dir_zapobj = zap_create(mos, |
34dc7c2f | 916 | DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); |
b128c09f | 917 | if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) |
428870ff | 918 | ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; |
34dc7c2f BB |
919 | dmu_buf_rele(dbuf, FTAG); |
920 | ||
921 | return (ddobj); | |
922 | } | |
923 | ||
b128c09f BB |
924 | boolean_t |
925 | dsl_dir_is_clone(dsl_dir_t *dd) | |
34dc7c2f | 926 | { |
d683ddbb | 927 | return (dsl_dir_phys(dd)->dd_origin_obj && |
b128c09f | 928 | (dd->dd_pool->dp_origin_snap == NULL || |
d683ddbb | 929 | dsl_dir_phys(dd)->dd_origin_obj != |
b128c09f | 930 | dd->dd_pool->dp_origin_snap->ds_object)); |
34dc7c2f BB |
931 | } |
932 | ||
933 | void | |
934 | dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) | |
935 | { | |
936 | mutex_enter(&dd->dd_lock); | |
b128c09f | 937 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, |
d683ddbb JG |
938 | dsl_dir_phys(dd)->dd_used_bytes); |
939 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, | |
940 | dsl_dir_phys(dd)->dd_quota); | |
34dc7c2f | 941 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, |
d683ddbb | 942 | dsl_dir_phys(dd)->dd_reserved); |
34dc7c2f | 943 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, |
d683ddbb JG |
944 | dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 : |
945 | (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 / | |
946 | dsl_dir_phys(dd)->dd_compressed_bytes)); | |
24a64651 | 947 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED, |
d683ddbb JG |
948 | dsl_dir_phys(dd)->dd_uncompressed_bytes); |
949 | if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { | |
b128c09f | 950 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, |
d683ddbb | 951 | dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]); |
b128c09f | 952 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, |
d683ddbb | 953 | dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]); |
b128c09f | 954 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, |
d683ddbb | 955 | dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]); |
b128c09f | 956 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, |
d683ddbb JG |
957 | dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] + |
958 | dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]); | |
b128c09f | 959 | } |
34dc7c2f BB |
960 | mutex_exit(&dd->dd_lock); |
961 | ||
788eb90c JJ |
962 | if (dsl_dir_is_zapified(dd)) { |
963 | uint64_t count; | |
964 | objset_t *os = dd->dd_pool->dp_meta_objset; | |
965 | ||
966 | if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, | |
967 | sizeof (count), 1, &count) == 0) { | |
968 | dsl_prop_nvlist_add_uint64(nv, | |
969 | ZFS_PROP_FILESYSTEM_COUNT, count); | |
970 | } | |
971 | if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, | |
972 | sizeof (count), 1, &count) == 0) { | |
973 | dsl_prop_nvlist_add_uint64(nv, | |
974 | ZFS_PROP_SNAPSHOT_COUNT, count); | |
975 | } | |
976 | } | |
977 | ||
b128c09f | 978 | if (dsl_dir_is_clone(dd)) { |
34dc7c2f BB |
979 | dsl_dataset_t *ds; |
980 | char buf[MAXNAMELEN]; | |
981 | ||
13fe0198 | 982 | VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, |
d683ddbb | 983 | dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds)); |
34dc7c2f | 984 | dsl_dataset_name(ds, buf); |
b128c09f | 985 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
986 | dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); |
987 | } | |
34dc7c2f BB |
988 | } |
989 | ||
990 | void | |
991 | dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) | |
992 | { | |
993 | dsl_pool_t *dp = dd->dd_pool; | |
994 | ||
d683ddbb | 995 | ASSERT(dsl_dir_phys(dd)); |
34dc7c2f | 996 | |
13fe0198 | 997 | if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) { |
34dc7c2f BB |
998 | /* up the hold count until we can be written out */ |
999 | dmu_buf_add_ref(dd->dd_dbuf, dd); | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | static int64_t | |
1004 | parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta) | |
1005 | { | |
d683ddbb JG |
1006 | uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved); |
1007 | uint64_t new_accounted = | |
1008 | MAX(used + delta, dsl_dir_phys(dd)->dd_reserved); | |
34dc7c2f BB |
1009 | return (new_accounted - old_accounted); |
1010 | } | |
1011 | ||
1012 | void | |
1013 | dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx) | |
1014 | { | |
1015 | ASSERT(dmu_tx_is_syncing(tx)); | |
1016 | ||
34dc7c2f | 1017 | mutex_enter(&dd->dd_lock); |
c99c9001 | 1018 | ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]); |
34dc7c2f BB |
1019 | dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, |
1020 | dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); | |
1021 | dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; | |
34dc7c2f BB |
1022 | mutex_exit(&dd->dd_lock); |
1023 | ||
1024 | /* release the hold from dsl_dir_dirty */ | |
1025 | dmu_buf_rele(dd->dd_dbuf, dd); | |
1026 | } | |
1027 | ||
1028 | static uint64_t | |
1029 | dsl_dir_space_towrite(dsl_dir_t *dd) | |
1030 | { | |
1031 | uint64_t space = 0; | |
1032 | int i; | |
1033 | ||
1034 | ASSERT(MUTEX_HELD(&dd->dd_lock)); | |
1035 | ||
1036 | for (i = 0; i < TXG_SIZE; i++) { | |
1037 | space += dd->dd_space_towrite[i&TXG_MASK]; | |
1038 | ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); | |
1039 | } | |
1040 | return (space); | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * How much space would dd have available if ancestor had delta applied | |
1045 | * to it? If ondiskonly is set, we're only interested in what's | |
1046 | * on-disk, not estimated pending changes. | |
1047 | */ | |
1048 | uint64_t | |
1049 | dsl_dir_space_available(dsl_dir_t *dd, | |
1050 | dsl_dir_t *ancestor, int64_t delta, int ondiskonly) | |
1051 | { | |
1052 | uint64_t parentspace, myspace, quota, used; | |
1053 | ||
1054 | /* | |
1055 | * If there are no restrictions otherwise, assume we have | |
1056 | * unlimited space available. | |
1057 | */ | |
1058 | quota = UINT64_MAX; | |
1059 | parentspace = UINT64_MAX; | |
1060 | ||
1061 | if (dd->dd_parent != NULL) { | |
1062 | parentspace = dsl_dir_space_available(dd->dd_parent, | |
1063 | ancestor, delta, ondiskonly); | |
1064 | } | |
1065 | ||
1066 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1067 | if (dsl_dir_phys(dd)->dd_quota != 0) |
1068 | quota = dsl_dir_phys(dd)->dd_quota; | |
1069 | used = dsl_dir_phys(dd)->dd_used_bytes; | |
34dc7c2f BB |
1070 | if (!ondiskonly) |
1071 | used += dsl_dir_space_towrite(dd); | |
34dc7c2f BB |
1072 | |
1073 | if (dd->dd_parent == NULL) { | |
1074 | uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); | |
1075 | quota = MIN(quota, poolsize); | |
1076 | } | |
1077 | ||
d683ddbb | 1078 | if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) { |
34dc7c2f BB |
1079 | /* |
1080 | * We have some space reserved, in addition to what our | |
1081 | * parent gave us. | |
1082 | */ | |
d683ddbb | 1083 | parentspace += dsl_dir_phys(dd)->dd_reserved - used; |
34dc7c2f BB |
1084 | } |
1085 | ||
b128c09f BB |
1086 | if (dd == ancestor) { |
1087 | ASSERT(delta <= 0); | |
1088 | ASSERT(used >= -delta); | |
1089 | used += delta; | |
1090 | if (parentspace != UINT64_MAX) | |
1091 | parentspace -= delta; | |
1092 | } | |
1093 | ||
34dc7c2f BB |
1094 | if (used > quota) { |
1095 | /* over quota */ | |
1096 | myspace = 0; | |
34dc7c2f BB |
1097 | } else { |
1098 | /* | |
1099 | * the lesser of the space provided by our parent and | |
1100 | * the space left in our quota | |
1101 | */ | |
1102 | myspace = MIN(parentspace, quota - used); | |
1103 | } | |
1104 | ||
1105 | mutex_exit(&dd->dd_lock); | |
1106 | ||
1107 | return (myspace); | |
1108 | } | |
1109 | ||
1110 | struct tempreserve { | |
1111 | list_node_t tr_node; | |
34dc7c2f BB |
1112 | dsl_dir_t *tr_ds; |
1113 | uint64_t tr_size; | |
1114 | }; | |
1115 | ||
1116 | static int | |
1117 | dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, | |
1118 | boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list, | |
1119 | dmu_tx_t *tx, boolean_t first) | |
1120 | { | |
1121 | uint64_t txg = tx->tx_txg; | |
1122 | uint64_t est_inflight, used_on_disk, quota, parent_rsrv; | |
428870ff | 1123 | uint64_t deferred = 0; |
34dc7c2f | 1124 | struct tempreserve *tr; |
428870ff | 1125 | int retval = EDQUOT; |
34dc7c2f BB |
1126 | int txgidx = txg & TXG_MASK; |
1127 | int i; | |
1128 | uint64_t ref_rsrv = 0; | |
1129 | ||
1130 | ASSERT3U(txg, !=, 0); | |
1131 | ASSERT3S(asize, >, 0); | |
1132 | ||
1133 | mutex_enter(&dd->dd_lock); | |
1134 | ||
1135 | /* | |
1136 | * Check against the dsl_dir's quota. We don't add in the delta | |
1137 | * when checking for over-quota because they get one free hit. | |
1138 | */ | |
1139 | est_inflight = dsl_dir_space_towrite(dd); | |
1140 | for (i = 0; i < TXG_SIZE; i++) | |
1141 | est_inflight += dd->dd_tempreserved[i]; | |
d683ddbb | 1142 | used_on_disk = dsl_dir_phys(dd)->dd_used_bytes; |
34dc7c2f BB |
1143 | |
1144 | /* | |
1145 | * On the first iteration, fetch the dataset's used-on-disk and | |
1146 | * refreservation values. Also, if checkrefquota is set, test if | |
1147 | * allocating this space would exceed the dataset's refquota. | |
1148 | */ | |
1149 | if (first && tx->tx_objset) { | |
1150 | int error; | |
428870ff | 1151 | dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset; |
34dc7c2f BB |
1152 | |
1153 | error = dsl_dataset_check_quota(ds, checkrefquota, | |
1154 | asize, est_inflight, &used_on_disk, &ref_rsrv); | |
1155 | if (error) { | |
1156 | mutex_exit(&dd->dd_lock); | |
3d920a15 | 1157 | DMU_TX_STAT_BUMP(dmu_tx_quota); |
34dc7c2f BB |
1158 | return (error); |
1159 | } | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * If this transaction will result in a net free of space, | |
1164 | * we want to let it through. | |
1165 | */ | |
d683ddbb | 1166 | if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0) |
34dc7c2f BB |
1167 | quota = UINT64_MAX; |
1168 | else | |
d683ddbb | 1169 | quota = dsl_dir_phys(dd)->dd_quota; |
34dc7c2f BB |
1170 | |
1171 | /* | |
428870ff BB |
1172 | * Adjust the quota against the actual pool size at the root |
1173 | * minus any outstanding deferred frees. | |
34dc7c2f BB |
1174 | * To ensure that it's possible to remove files from a full |
1175 | * pool without inducing transient overcommits, we throttle | |
1176 | * netfree transactions against a quota that is slightly larger, | |
1177 | * but still within the pool's allocation slop. In cases where | |
1178 | * we're very close to full, this will allow a steady trickle of | |
1179 | * removes to get through. | |
1180 | */ | |
1181 | if (dd->dd_parent == NULL) { | |
428870ff | 1182 | spa_t *spa = dd->dd_pool->dp_spa; |
34dc7c2f | 1183 | uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); |
428870ff BB |
1184 | deferred = metaslab_class_get_deferred(spa_normal_class(spa)); |
1185 | if (poolsize - deferred < quota) { | |
1186 | quota = poolsize - deferred; | |
1187 | retval = ENOSPC; | |
34dc7c2f BB |
1188 | } |
1189 | } | |
1190 | ||
1191 | /* | |
1192 | * If they are requesting more space, and our current estimate | |
1193 | * is over quota, they get to try again unless the actual | |
1194 | * on-disk is over quota and there are no pending changes (which | |
1195 | * may free up space for us). | |
1196 | */ | |
428870ff BB |
1197 | if (used_on_disk + est_inflight >= quota) { |
1198 | if (est_inflight > 0 || used_on_disk < quota || | |
1199 | (retval == ENOSPC && used_on_disk < quota + deferred)) | |
1200 | retval = ERESTART; | |
34dc7c2f BB |
1201 | dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " |
1202 | "quota=%lluK tr=%lluK err=%d\n", | |
1203 | used_on_disk>>10, est_inflight>>10, | |
428870ff | 1204 | quota>>10, asize>>10, retval); |
34dc7c2f | 1205 | mutex_exit(&dd->dd_lock); |
3d920a15 | 1206 | DMU_TX_STAT_BUMP(dmu_tx_quota); |
2e528b49 | 1207 | return (SET_ERROR(retval)); |
34dc7c2f BB |
1208 | } |
1209 | ||
1210 | /* We need to up our estimated delta before dropping dd_lock */ | |
1211 | dd->dd_tempreserved[txgidx] += asize; | |
1212 | ||
1213 | parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, | |
1214 | asize - ref_rsrv); | |
1215 | mutex_exit(&dd->dd_lock); | |
1216 | ||
79c76d5b | 1217 | tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); |
34dc7c2f BB |
1218 | tr->tr_ds = dd; |
1219 | tr->tr_size = asize; | |
1220 | list_insert_tail(tr_list, tr); | |
1221 | ||
1222 | /* see if it's OK with our parent */ | |
1223 | if (dd->dd_parent && parent_rsrv) { | |
d683ddbb | 1224 | boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0); |
34dc7c2f BB |
1225 | |
1226 | return (dsl_dir_tempreserve_impl(dd->dd_parent, | |
1227 | parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE)); | |
1228 | } else { | |
1229 | return (0); | |
1230 | } | |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Reserve space in this dsl_dir, to be used in this tx's txg. | |
1235 | * After the space has been dirtied (and dsl_dir_willuse_space() | |
1236 | * has been called), the reservation should be canceled, using | |
1237 | * dsl_dir_tempreserve_clear(). | |
1238 | */ | |
1239 | int | |
1240 | dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, | |
1241 | uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx) | |
1242 | { | |
1243 | int err; | |
1244 | list_t *tr_list; | |
1245 | ||
1246 | if (asize == 0) { | |
1247 | *tr_cookiep = NULL; | |
1248 | return (0); | |
1249 | } | |
1250 | ||
79c76d5b | 1251 | tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); |
34dc7c2f BB |
1252 | list_create(tr_list, sizeof (struct tempreserve), |
1253 | offsetof(struct tempreserve, tr_node)); | |
1254 | ASSERT3S(asize, >, 0); | |
1255 | ASSERT3S(fsize, >=, 0); | |
1256 | ||
1257 | err = arc_tempreserve_space(lsize, tx->tx_txg); | |
1258 | if (err == 0) { | |
1259 | struct tempreserve *tr; | |
1260 | ||
79c76d5b | 1261 | tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); |
34dc7c2f BB |
1262 | tr->tr_size = lsize; |
1263 | list_insert_tail(tr_list, tr); | |
34dc7c2f BB |
1264 | } else { |
1265 | if (err == EAGAIN) { | |
e8b96c60 MA |
1266 | /* |
1267 | * If arc_memory_throttle() detected that pageout | |
1268 | * is running and we are low on memory, we delay new | |
1269 | * non-pageout transactions to give pageout an | |
1270 | * advantage. | |
1271 | * | |
1272 | * It is unfortunate to be delaying while the caller's | |
1273 | * locks are held. | |
1274 | */ | |
63fd3c6c AL |
1275 | txg_delay(dd->dd_pool, tx->tx_txg, |
1276 | MSEC2NSEC(10), MSEC2NSEC(10)); | |
2e528b49 | 1277 | err = SET_ERROR(ERESTART); |
34dc7c2f | 1278 | } |
34dc7c2f BB |
1279 | } |
1280 | ||
1281 | if (err == 0) { | |
34dc7c2f BB |
1282 | err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, |
1283 | FALSE, asize > usize, tr_list, tx, TRUE); | |
1284 | } | |
1285 | ||
13fe0198 | 1286 | if (err != 0) |
34dc7c2f BB |
1287 | dsl_dir_tempreserve_clear(tr_list, tx); |
1288 | else | |
1289 | *tr_cookiep = tr_list; | |
1290 | ||
1291 | return (err); | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Clear a temporary reservation that we previously made with | |
1296 | * dsl_dir_tempreserve_space(). | |
1297 | */ | |
1298 | void | |
1299 | dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) | |
1300 | { | |
1301 | int txgidx = tx->tx_txg & TXG_MASK; | |
1302 | list_t *tr_list = tr_cookie; | |
1303 | struct tempreserve *tr; | |
1304 | ||
1305 | ASSERT3U(tx->tx_txg, !=, 0); | |
1306 | ||
1307 | if (tr_cookie == NULL) | |
1308 | return; | |
1309 | ||
e8b96c60 MA |
1310 | while ((tr = list_head(tr_list)) != NULL) { |
1311 | if (tr->tr_ds) { | |
34dc7c2f BB |
1312 | mutex_enter(&tr->tr_ds->dd_lock); |
1313 | ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, | |
1314 | tr->tr_size); | |
1315 | tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; | |
1316 | mutex_exit(&tr->tr_ds->dd_lock); | |
1317 | } else { | |
1318 | arc_tempreserve_clear(tr->tr_size); | |
1319 | } | |
1320 | list_remove(tr_list, tr); | |
1321 | kmem_free(tr, sizeof (struct tempreserve)); | |
1322 | } | |
1323 | ||
1324 | kmem_free(tr_list, sizeof (list_t)); | |
1325 | } | |
1326 | ||
e8b96c60 MA |
1327 | /* |
1328 | * This should be called from open context when we think we're going to write | |
1329 | * or free space, for example when dirtying data. Be conservative; it's okay | |
1330 | * to write less space or free more, but we don't want to write more or free | |
1331 | * less than the amount specified. | |
1ba16159 AB |
1332 | * |
1333 | * NOTE: The behavior of this function is identical to the Illumos / FreeBSD | |
1334 | * version however it has been adjusted to use an iterative rather then | |
1335 | * recursive algorithm to minimize stack usage. | |
e8b96c60 MA |
1336 | */ |
1337 | void | |
1338 | dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) | |
34dc7c2f BB |
1339 | { |
1340 | int64_t parent_space; | |
1341 | uint64_t est_used; | |
1342 | ||
1ba16159 AB |
1343 | do { |
1344 | mutex_enter(&dd->dd_lock); | |
1345 | if (space > 0) | |
1346 | dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; | |
34dc7c2f | 1347 | |
1ba16159 | 1348 | est_used = dsl_dir_space_towrite(dd) + |
d683ddbb | 1349 | dsl_dir_phys(dd)->dd_used_bytes; |
1ba16159 AB |
1350 | parent_space = parent_delta(dd, est_used, space); |
1351 | mutex_exit(&dd->dd_lock); | |
34dc7c2f | 1352 | |
1ba16159 AB |
1353 | /* Make sure that we clean up dd_space_to* */ |
1354 | dsl_dir_dirty(dd, tx); | |
34dc7c2f | 1355 | |
1ba16159 AB |
1356 | dd = dd->dd_parent; |
1357 | space = parent_space; | |
1358 | } while (space && dd); | |
34dc7c2f BB |
1359 | } |
1360 | ||
1361 | /* call from syncing context when we actually write/free space for this dd */ | |
1362 | void | |
b128c09f | 1363 | dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, |
34dc7c2f BB |
1364 | int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) |
1365 | { | |
1366 | int64_t accounted_delta; | |
a169a625 MA |
1367 | |
1368 | /* | |
1369 | * dsl_dataset_set_refreservation_sync_impl() calls this with | |
1370 | * dd_lock held, so that it can atomically update | |
1371 | * ds->ds_reserved and the dsl_dir accounting, so that | |
1372 | * dsl_dataset_check_quota() can see dataset and dir accounting | |
1373 | * consistently. | |
1374 | */ | |
b128c09f | 1375 | boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); |
34dc7c2f BB |
1376 | |
1377 | ASSERT(dmu_tx_is_syncing(tx)); | |
b128c09f | 1378 | ASSERT(type < DD_USED_NUM); |
34dc7c2f | 1379 | |
a169a625 MA |
1380 | dmu_buf_will_dirty(dd->dd_dbuf, tx); |
1381 | ||
b128c09f BB |
1382 | if (needlock) |
1383 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1384 | accounted_delta = |
1385 | parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used); | |
1386 | ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used); | |
34dc7c2f | 1387 | ASSERT(compressed >= 0 || |
d683ddbb | 1388 | dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed); |
34dc7c2f | 1389 | ASSERT(uncompressed >= 0 || |
d683ddbb JG |
1390 | dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed); |
1391 | dsl_dir_phys(dd)->dd_used_bytes += used; | |
1392 | dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed; | |
1393 | dsl_dir_phys(dd)->dd_compressed_bytes += compressed; | |
b128c09f | 1394 | |
d683ddbb | 1395 | if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) { |
b128c09f | 1396 | ASSERT(used > 0 || |
d683ddbb JG |
1397 | dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used); |
1398 | dsl_dir_phys(dd)->dd_used_breakdown[type] += used; | |
b128c09f | 1399 | #ifdef DEBUG |
d6320ddb BB |
1400 | { |
1401 | dd_used_t t; | |
1402 | uint64_t u = 0; | |
1403 | for (t = 0; t < DD_USED_NUM; t++) | |
d683ddbb JG |
1404 | u += dsl_dir_phys(dd)->dd_used_breakdown[t]; |
1405 | ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes); | |
d6320ddb | 1406 | } |
b128c09f BB |
1407 | #endif |
1408 | } | |
1409 | if (needlock) | |
1410 | mutex_exit(&dd->dd_lock); | |
34dc7c2f BB |
1411 | |
1412 | if (dd->dd_parent != NULL) { | |
b128c09f | 1413 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, |
34dc7c2f | 1414 | accounted_delta, compressed, uncompressed, tx); |
b128c09f BB |
1415 | dsl_dir_transfer_space(dd->dd_parent, |
1416 | used - accounted_delta, | |
1417 | DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); | |
34dc7c2f BB |
1418 | } |
1419 | } | |
1420 | ||
b128c09f BB |
1421 | void |
1422 | dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, | |
1423 | dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) | |
1424 | { | |
b128c09f BB |
1425 | ASSERT(dmu_tx_is_syncing(tx)); |
1426 | ASSERT(oldtype < DD_USED_NUM); | |
1427 | ASSERT(newtype < DD_USED_NUM); | |
1428 | ||
d683ddbb JG |
1429 | if (delta == 0 || |
1430 | !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN)) | |
b128c09f BB |
1431 | return; |
1432 | ||
a169a625 MA |
1433 | dmu_buf_will_dirty(dd->dd_dbuf, tx); |
1434 | mutex_enter(&dd->dd_lock); | |
b128c09f | 1435 | ASSERT(delta > 0 ? |
d683ddbb JG |
1436 | dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta : |
1437 | dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta); | |
1438 | ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta)); | |
1439 | dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta; | |
1440 | dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta; | |
a169a625 | 1441 | mutex_exit(&dd->dd_lock); |
b128c09f BB |
1442 | } |
1443 | ||
13fe0198 MA |
1444 | typedef struct dsl_dir_set_qr_arg { |
1445 | const char *ddsqra_name; | |
1446 | zprop_source_t ddsqra_source; | |
1447 | uint64_t ddsqra_value; | |
1448 | } dsl_dir_set_qr_arg_t; | |
1449 | ||
34dc7c2f | 1450 | static int |
13fe0198 | 1451 | dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1452 | { |
13fe0198 MA |
1453 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1454 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1455 | dsl_dataset_t *ds; | |
1456 | int error; | |
1457 | uint64_t towrite, newval; | |
34dc7c2f | 1458 | |
13fe0198 MA |
1459 | error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); |
1460 | if (error != 0) | |
1461 | return (error); | |
1462 | ||
1463 | error = dsl_prop_predict(ds->ds_dir, "quota", | |
1464 | ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); | |
1465 | if (error != 0) { | |
1466 | dsl_dataset_rele(ds, FTAG); | |
1467 | return (error); | |
1468 | } | |
428870ff | 1469 | |
13fe0198 MA |
1470 | if (newval == 0) { |
1471 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f | 1472 | return (0); |
13fe0198 | 1473 | } |
34dc7c2f | 1474 | |
13fe0198 | 1475 | mutex_enter(&ds->ds_dir->dd_lock); |
34dc7c2f BB |
1476 | /* |
1477 | * If we are doing the preliminary check in open context, and | |
1478 | * there are pending changes, then don't fail it, since the | |
1479 | * pending changes could under-estimate the amount of space to be | |
1480 | * freed up. | |
1481 | */ | |
13fe0198 | 1482 | towrite = dsl_dir_space_towrite(ds->ds_dir); |
34dc7c2f | 1483 | if ((dmu_tx_is_syncing(tx) || towrite == 0) && |
d683ddbb JG |
1484 | (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved || |
1485 | newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) { | |
2e528b49 | 1486 | error = SET_ERROR(ENOSPC); |
34dc7c2f | 1487 | } |
13fe0198 MA |
1488 | mutex_exit(&ds->ds_dir->dd_lock); |
1489 | dsl_dataset_rele(ds, FTAG); | |
1490 | return (error); | |
34dc7c2f BB |
1491 | } |
1492 | ||
34dc7c2f | 1493 | static void |
13fe0198 | 1494 | dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1495 | { |
13fe0198 MA |
1496 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1497 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1498 | dsl_dataset_t *ds; | |
1499 | uint64_t newval; | |
428870ff | 1500 | |
13fe0198 | 1501 | VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); |
34dc7c2f | 1502 | |
b1118acb MM |
1503 | if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { |
1504 | dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA), | |
1505 | ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, | |
1506 | &ddsqra->ddsqra_value, tx); | |
34dc7c2f | 1507 | |
b1118acb MM |
1508 | VERIFY0(dsl_prop_get_int_ds(ds, |
1509 | zfs_prop_to_name(ZFS_PROP_QUOTA), &newval)); | |
1510 | } else { | |
1511 | newval = ddsqra->ddsqra_value; | |
1512 | spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", | |
1513 | zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval); | |
1514 | } | |
6f1ffb06 | 1515 | |
13fe0198 MA |
1516 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); |
1517 | mutex_enter(&ds->ds_dir->dd_lock); | |
d683ddbb | 1518 | dsl_dir_phys(ds->ds_dir)->dd_quota = newval; |
13fe0198 MA |
1519 | mutex_exit(&ds->ds_dir->dd_lock); |
1520 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
1521 | } |
1522 | ||
1523 | int | |
428870ff | 1524 | dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota) |
34dc7c2f | 1525 | { |
13fe0198 | 1526 | dsl_dir_set_qr_arg_t ddsqra; |
428870ff | 1527 | |
13fe0198 MA |
1528 | ddsqra.ddsqra_name = ddname; |
1529 | ddsqra.ddsqra_source = source; | |
1530 | ddsqra.ddsqra_value = quota; | |
428870ff | 1531 | |
13fe0198 | 1532 | return (dsl_sync_task(ddname, dsl_dir_set_quota_check, |
3d45fdd6 | 1533 | dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); |
34dc7c2f BB |
1534 | } |
1535 | ||
1536 | int | |
13fe0198 | 1537 | dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1538 | { |
13fe0198 MA |
1539 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1540 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1541 | dsl_dataset_t *ds; | |
1542 | dsl_dir_t *dd; | |
1543 | uint64_t newval, used, avail; | |
1544 | int error; | |
428870ff | 1545 | |
13fe0198 MA |
1546 | error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); |
1547 | if (error != 0) | |
1548 | return (error); | |
1549 | dd = ds->ds_dir; | |
34dc7c2f BB |
1550 | |
1551 | /* | |
1552 | * If we are doing the preliminary check in open context, the | |
1553 | * space estimates may be inaccurate. | |
1554 | */ | |
13fe0198 MA |
1555 | if (!dmu_tx_is_syncing(tx)) { |
1556 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f | 1557 | return (0); |
13fe0198 MA |
1558 | } |
1559 | ||
1560 | error = dsl_prop_predict(ds->ds_dir, | |
1561 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1562 | ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); | |
1563 | if (error != 0) { | |
1564 | dsl_dataset_rele(ds, FTAG); | |
1565 | return (error); | |
1566 | } | |
34dc7c2f BB |
1567 | |
1568 | mutex_enter(&dd->dd_lock); | |
d683ddbb | 1569 | used = dsl_dir_phys(dd)->dd_used_bytes; |
34dc7c2f BB |
1570 | mutex_exit(&dd->dd_lock); |
1571 | ||
1572 | if (dd->dd_parent) { | |
1573 | avail = dsl_dir_space_available(dd->dd_parent, | |
1574 | NULL, 0, FALSE); | |
1575 | } else { | |
1576 | avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; | |
1577 | } | |
1578 | ||
d683ddbb | 1579 | if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) { |
13fe0198 | 1580 | uint64_t delta = MAX(used, newval) - |
d683ddbb | 1581 | MAX(used, dsl_dir_phys(dd)->dd_reserved); |
d164b209 | 1582 | |
13fe0198 | 1583 | if (delta > avail || |
d683ddbb JG |
1584 | (dsl_dir_phys(dd)->dd_quota > 0 && |
1585 | newval > dsl_dir_phys(dd)->dd_quota)) | |
2e528b49 | 1586 | error = SET_ERROR(ENOSPC); |
d164b209 BB |
1587 | } |
1588 | ||
13fe0198 MA |
1589 | dsl_dataset_rele(ds, FTAG); |
1590 | return (error); | |
34dc7c2f BB |
1591 | } |
1592 | ||
13fe0198 | 1593 | void |
6f1ffb06 | 1594 | dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx) |
34dc7c2f | 1595 | { |
34dc7c2f BB |
1596 | uint64_t used; |
1597 | int64_t delta; | |
1598 | ||
1599 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
1600 | ||
1601 | mutex_enter(&dd->dd_lock); | |
d683ddbb JG |
1602 | used = dsl_dir_phys(dd)->dd_used_bytes; |
1603 | delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved); | |
1604 | dsl_dir_phys(dd)->dd_reserved = value; | |
34dc7c2f BB |
1605 | |
1606 | if (dd->dd_parent != NULL) { | |
1607 | /* Roll up this additional usage into our ancestors */ | |
b128c09f BB |
1608 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, |
1609 | delta, 0, 0, tx); | |
34dc7c2f | 1610 | } |
b128c09f | 1611 | mutex_exit(&dd->dd_lock); |
34dc7c2f BB |
1612 | } |
1613 | ||
6f1ffb06 | 1614 | static void |
13fe0198 | 1615 | dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx) |
6f1ffb06 | 1616 | { |
13fe0198 MA |
1617 | dsl_dir_set_qr_arg_t *ddsqra = arg; |
1618 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1619 | dsl_dataset_t *ds; | |
1620 | uint64_t newval; | |
6f1ffb06 | 1621 | |
13fe0198 MA |
1622 | VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); |
1623 | ||
b1118acb MM |
1624 | if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) { |
1625 | dsl_prop_set_sync_impl(ds, | |
1626 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1627 | ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, | |
1628 | &ddsqra->ddsqra_value, tx); | |
d1d7e268 | 1629 | |
b1118acb MM |
1630 | VERIFY0(dsl_prop_get_int_ds(ds, |
1631 | zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval)); | |
1632 | } else { | |
1633 | newval = ddsqra->ddsqra_value; | |
1634 | spa_history_log_internal_ds(ds, "set", tx, "%s=%lld", | |
1635 | zfs_prop_to_name(ZFS_PROP_RESERVATION), | |
1636 | (longlong_t)newval); | |
1637 | } | |
1638 | ||
13fe0198 MA |
1639 | dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx); |
1640 | dsl_dataset_rele(ds, FTAG); | |
d1d7e268 | 1641 | } |
6f1ffb06 | 1642 | |
34dc7c2f | 1643 | int |
428870ff BB |
1644 | dsl_dir_set_reservation(const char *ddname, zprop_source_t source, |
1645 | uint64_t reservation) | |
34dc7c2f | 1646 | { |
13fe0198 | 1647 | dsl_dir_set_qr_arg_t ddsqra; |
428870ff | 1648 | |
13fe0198 MA |
1649 | ddsqra.ddsqra_name = ddname; |
1650 | ddsqra.ddsqra_source = source; | |
1651 | ddsqra.ddsqra_value = reservation; | |
428870ff | 1652 | |
13fe0198 | 1653 | return (dsl_sync_task(ddname, dsl_dir_set_reservation_check, |
3d45fdd6 | 1654 | dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); |
34dc7c2f BB |
1655 | } |
1656 | ||
1657 | static dsl_dir_t * | |
1658 | closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2) | |
1659 | { | |
1660 | for (; ds1; ds1 = ds1->dd_parent) { | |
1661 | dsl_dir_t *dd; | |
1662 | for (dd = ds2; dd; dd = dd->dd_parent) { | |
1663 | if (ds1 == dd) | |
1664 | return (dd); | |
1665 | } | |
1666 | } | |
1667 | return (NULL); | |
1668 | } | |
1669 | ||
1670 | /* | |
1671 | * If delta is applied to dd, how much of that delta would be applied to | |
1672 | * ancestor? Syncing context only. | |
1673 | */ | |
1674 | static int64_t | |
1675 | would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) | |
1676 | { | |
1677 | if (dd == ancestor) | |
1678 | return (delta); | |
1679 | ||
1680 | mutex_enter(&dd->dd_lock); | |
d683ddbb | 1681 | delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta); |
34dc7c2f BB |
1682 | mutex_exit(&dd->dd_lock); |
1683 | return (would_change(dd->dd_parent, delta, ancestor)); | |
1684 | } | |
1685 | ||
13fe0198 MA |
1686 | typedef struct dsl_dir_rename_arg { |
1687 | const char *ddra_oldname; | |
1688 | const char *ddra_newname; | |
788eb90c | 1689 | cred_t *ddra_cred; |
13fe0198 | 1690 | } dsl_dir_rename_arg_t; |
34dc7c2f | 1691 | |
13fe0198 | 1692 | /* ARGSUSED */ |
34dc7c2f | 1693 | static int |
13fe0198 | 1694 | dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
34dc7c2f | 1695 | { |
13fe0198 MA |
1696 | int *deltap = arg; |
1697 | char namebuf[MAXNAMELEN]; | |
34dc7c2f | 1698 | |
13fe0198 MA |
1699 | dsl_dataset_name(ds, namebuf); |
1700 | ||
1701 | if (strlen(namebuf) + *deltap >= MAXNAMELEN) | |
2e528b49 | 1702 | return (SET_ERROR(ENAMETOOLONG)); |
13fe0198 MA |
1703 | return (0); |
1704 | } | |
1705 | ||
1706 | static int | |
1707 | dsl_dir_rename_check(void *arg, dmu_tx_t *tx) | |
1708 | { | |
1709 | dsl_dir_rename_arg_t *ddra = arg; | |
1710 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1711 | dsl_dir_t *dd, *newparent; | |
1712 | const char *mynewname; | |
1713 | int error; | |
1714 | int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname); | |
34dc7c2f | 1715 | |
13fe0198 MA |
1716 | /* target dir should exist */ |
1717 | error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL); | |
1718 | if (error != 0) | |
1719 | return (error); | |
1720 | ||
1721 | /* new parent should exist */ | |
1722 | error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG, | |
1723 | &newparent, &mynewname); | |
1724 | if (error != 0) { | |
1725 | dsl_dir_rele(dd, FTAG); | |
1726 | return (error); | |
1727 | } | |
1728 | ||
1729 | /* can't rename to different pool */ | |
1730 | if (dd->dd_pool != newparent->dd_pool) { | |
1731 | dsl_dir_rele(newparent, FTAG); | |
1732 | dsl_dir_rele(dd, FTAG); | |
9063f654 | 1733 | return (SET_ERROR(EXDEV)); |
13fe0198 MA |
1734 | } |
1735 | ||
1736 | /* new name should not already exist */ | |
1737 | if (mynewname == NULL) { | |
1738 | dsl_dir_rele(newparent, FTAG); | |
1739 | dsl_dir_rele(dd, FTAG); | |
2e528b49 | 1740 | return (SET_ERROR(EEXIST)); |
13fe0198 MA |
1741 | } |
1742 | ||
1743 | /* if the name length is growing, validate child name lengths */ | |
1744 | if (delta > 0) { | |
1745 | error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename, | |
1746 | &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); | |
1747 | if (error != 0) { | |
1748 | dsl_dir_rele(newparent, FTAG); | |
1749 | dsl_dir_rele(dd, FTAG); | |
1750 | return (error); | |
1751 | } | |
1752 | } | |
34dc7c2f | 1753 | |
788eb90c | 1754 | if (dmu_tx_is_syncing(tx)) { |
a0c9a17a | 1755 | if (spa_feature_is_active(dp->dp_spa, |
788eb90c JJ |
1756 | SPA_FEATURE_FS_SS_LIMIT)) { |
1757 | /* | |
1758 | * Although this is the check function and we don't | |
1759 | * normally make on-disk changes in check functions, | |
1760 | * we need to do that here. | |
1761 | * | |
1762 | * Ensure this portion of the tree's counts have been | |
1763 | * initialized in case the new parent has limits set. | |
1764 | */ | |
1765 | dsl_dir_init_fs_ss_count(dd, tx); | |
1766 | } | |
1767 | } | |
1768 | ||
13fe0198 | 1769 | if (newparent != dd->dd_parent) { |
34dc7c2f BB |
1770 | /* is there enough space? */ |
1771 | uint64_t myspace = | |
d683ddbb JG |
1772 | MAX(dsl_dir_phys(dd)->dd_used_bytes, |
1773 | dsl_dir_phys(dd)->dd_reserved); | |
788eb90c JJ |
1774 | objset_t *os = dd->dd_pool->dp_meta_objset; |
1775 | uint64_t fs_cnt = 0; | |
1776 | uint64_t ss_cnt = 0; | |
1777 | ||
1778 | if (dsl_dir_is_zapified(dd)) { | |
1779 | int err; | |
1780 | ||
1781 | err = zap_lookup(os, dd->dd_object, | |
1782 | DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, | |
1783 | &fs_cnt); | |
a0c9a17a JJ |
1784 | if (err != ENOENT && err != 0) { |
1785 | dsl_dir_rele(newparent, FTAG); | |
1786 | dsl_dir_rele(dd, FTAG); | |
788eb90c | 1787 | return (err); |
a0c9a17a | 1788 | } |
788eb90c JJ |
1789 | |
1790 | /* | |
1791 | * have to add 1 for the filesystem itself that we're | |
1792 | * moving | |
1793 | */ | |
1794 | fs_cnt++; | |
1795 | ||
1796 | err = zap_lookup(os, dd->dd_object, | |
1797 | DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, | |
1798 | &ss_cnt); | |
a0c9a17a JJ |
1799 | if (err != ENOENT && err != 0) { |
1800 | dsl_dir_rele(newparent, FTAG); | |
1801 | dsl_dir_rele(dd, FTAG); | |
788eb90c | 1802 | return (err); |
a0c9a17a | 1803 | } |
788eb90c | 1804 | } |
34dc7c2f BB |
1805 | |
1806 | /* no rename into our descendant */ | |
13fe0198 MA |
1807 | if (closest_common_ancestor(dd, newparent) == dd) { |
1808 | dsl_dir_rele(newparent, FTAG); | |
1809 | dsl_dir_rele(dd, FTAG); | |
2e528b49 | 1810 | return (SET_ERROR(EINVAL)); |
13fe0198 | 1811 | } |
34dc7c2f | 1812 | |
13fe0198 | 1813 | error = dsl_dir_transfer_possible(dd->dd_parent, |
788eb90c | 1814 | newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred); |
13fe0198 MA |
1815 | if (error != 0) { |
1816 | dsl_dir_rele(newparent, FTAG); | |
1817 | dsl_dir_rele(dd, FTAG); | |
1818 | return (error); | |
1819 | } | |
34dc7c2f BB |
1820 | } |
1821 | ||
13fe0198 MA |
1822 | dsl_dir_rele(newparent, FTAG); |
1823 | dsl_dir_rele(dd, FTAG); | |
34dc7c2f BB |
1824 | return (0); |
1825 | } | |
1826 | ||
1827 | static void | |
13fe0198 | 1828 | dsl_dir_rename_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1829 | { |
13fe0198 MA |
1830 | dsl_dir_rename_arg_t *ddra = arg; |
1831 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1832 | dsl_dir_t *dd, *newparent; | |
1833 | const char *mynewname; | |
1834 | int error; | |
34dc7c2f | 1835 | objset_t *mos = dp->dp_meta_objset; |
34dc7c2f | 1836 | |
13fe0198 MA |
1837 | VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL)); |
1838 | VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent, | |
1839 | &mynewname)); | |
34dc7c2f | 1840 | |
6f1ffb06 | 1841 | /* Log this before we change the name. */ |
6f1ffb06 | 1842 | spa_history_log_internal_dd(dd, "rename", tx, |
13fe0198 | 1843 | "-> %s", ddra->ddra_newname); |
6f1ffb06 | 1844 | |
13fe0198 | 1845 | if (newparent != dd->dd_parent) { |
788eb90c JJ |
1846 | objset_t *os = dd->dd_pool->dp_meta_objset; |
1847 | uint64_t fs_cnt = 0; | |
1848 | uint64_t ss_cnt = 0; | |
1849 | ||
1850 | /* | |
1851 | * We already made sure the dd counts were initialized in the | |
1852 | * check function. | |
1853 | */ | |
a0c9a17a | 1854 | if (spa_feature_is_active(dp->dp_spa, |
788eb90c JJ |
1855 | SPA_FEATURE_FS_SS_LIMIT)) { |
1856 | VERIFY0(zap_lookup(os, dd->dd_object, | |
1857 | DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1, | |
1858 | &fs_cnt)); | |
1859 | /* add 1 for the filesystem itself that we're moving */ | |
1860 | fs_cnt++; | |
1861 | ||
1862 | VERIFY0(zap_lookup(os, dd->dd_object, | |
1863 | DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1, | |
1864 | &ss_cnt)); | |
1865 | } | |
1866 | ||
1867 | dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt, | |
1868 | DD_FIELD_FILESYSTEM_COUNT, tx); | |
1869 | dsl_fs_ss_count_adjust(newparent, fs_cnt, | |
1870 | DD_FIELD_FILESYSTEM_COUNT, tx); | |
1871 | ||
1872 | dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt, | |
1873 | DD_FIELD_SNAPSHOT_COUNT, tx); | |
1874 | dsl_fs_ss_count_adjust(newparent, ss_cnt, | |
1875 | DD_FIELD_SNAPSHOT_COUNT, tx); | |
1876 | ||
b128c09f | 1877 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, |
d683ddbb JG |
1878 | -dsl_dir_phys(dd)->dd_used_bytes, |
1879 | -dsl_dir_phys(dd)->dd_compressed_bytes, | |
1880 | -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); | |
13fe0198 | 1881 | dsl_dir_diduse_space(newparent, DD_USED_CHILD, |
d683ddbb JG |
1882 | dsl_dir_phys(dd)->dd_used_bytes, |
1883 | dsl_dir_phys(dd)->dd_compressed_bytes, | |
1884 | dsl_dir_phys(dd)->dd_uncompressed_bytes, tx); | |
b128c09f | 1885 | |
d683ddbb JG |
1886 | if (dsl_dir_phys(dd)->dd_reserved > |
1887 | dsl_dir_phys(dd)->dd_used_bytes) { | |
1888 | uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved - | |
1889 | dsl_dir_phys(dd)->dd_used_bytes; | |
b128c09f BB |
1890 | |
1891 | dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, | |
1892 | -unused_rsrv, 0, 0, tx); | |
13fe0198 | 1893 | dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV, |
b128c09f BB |
1894 | unused_rsrv, 0, 0, tx); |
1895 | } | |
34dc7c2f BB |
1896 | } |
1897 | ||
1898 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
1899 | ||
1900 | /* remove from old parent zapobj */ | |
d683ddbb JG |
1901 | error = zap_remove(mos, |
1902 | dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj, | |
34dc7c2f | 1903 | dd->dd_myname, tx); |
13fe0198 | 1904 | ASSERT0(error); |
34dc7c2f | 1905 | |
13fe0198 MA |
1906 | (void) strcpy(dd->dd_myname, mynewname); |
1907 | dsl_dir_rele(dd->dd_parent, dd); | |
d683ddbb | 1908 | dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object; |
13fe0198 MA |
1909 | VERIFY0(dsl_dir_hold_obj(dp, |
1910 | newparent->dd_object, NULL, dd, &dd->dd_parent)); | |
34dc7c2f BB |
1911 | |
1912 | /* add to new parent zapobj */ | |
d683ddbb | 1913 | VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj, |
13fe0198 MA |
1914 | dd->dd_myname, 8, 1, &dd->dd_object, tx)); |
1915 | ||
ba6a2402 BB |
1916 | #ifdef _KERNEL |
1917 | zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname); | |
1918 | #endif | |
1919 | ||
13fe0198 | 1920 | dsl_prop_notify_all(dd); |
34dc7c2f | 1921 | |
13fe0198 MA |
1922 | dsl_dir_rele(newparent, FTAG); |
1923 | dsl_dir_rele(dd, FTAG); | |
34dc7c2f BB |
1924 | } |
1925 | ||
1926 | int | |
13fe0198 | 1927 | dsl_dir_rename(const char *oldname, const char *newname) |
34dc7c2f | 1928 | { |
13fe0198 | 1929 | dsl_dir_rename_arg_t ddra; |
34dc7c2f | 1930 | |
13fe0198 MA |
1931 | ddra.ddra_oldname = oldname; |
1932 | ddra.ddra_newname = newname; | |
788eb90c | 1933 | ddra.ddra_cred = CRED(); |
34dc7c2f | 1934 | |
13fe0198 | 1935 | return (dsl_sync_task(oldname, |
3d45fdd6 MA |
1936 | dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, |
1937 | 3, ZFS_SPACE_CHECK_RESERVED)); | |
34dc7c2f BB |
1938 | } |
1939 | ||
1940 | int | |
788eb90c JJ |
1941 | dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, |
1942 | uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr) | |
34dc7c2f BB |
1943 | { |
1944 | dsl_dir_t *ancestor; | |
1945 | int64_t adelta; | |
1946 | uint64_t avail; | |
788eb90c | 1947 | int err; |
34dc7c2f BB |
1948 | |
1949 | ancestor = closest_common_ancestor(sdd, tdd); | |
1950 | adelta = would_change(sdd, -space, ancestor); | |
1951 | avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE); | |
1952 | if (avail < space) | |
2e528b49 | 1953 | return (SET_ERROR(ENOSPC)); |
34dc7c2f | 1954 | |
788eb90c JJ |
1955 | err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT, |
1956 | ancestor, cr); | |
1957 | if (err != 0) | |
1958 | return (err); | |
1959 | err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT, | |
1960 | ancestor, cr); | |
1961 | if (err != 0) | |
1962 | return (err); | |
1963 | ||
34dc7c2f BB |
1964 | return (0); |
1965 | } | |
428870ff BB |
1966 | |
1967 | timestruc_t | |
1968 | dsl_dir_snap_cmtime(dsl_dir_t *dd) | |
1969 | { | |
1970 | timestruc_t t; | |
1971 | ||
1972 | mutex_enter(&dd->dd_lock); | |
1973 | t = dd->dd_snap_cmtime; | |
1974 | mutex_exit(&dd->dd_lock); | |
1975 | ||
1976 | return (t); | |
1977 | } | |
1978 | ||
1979 | void | |
1980 | dsl_dir_snap_cmtime_update(dsl_dir_t *dd) | |
1981 | { | |
1982 | timestruc_t t; | |
1983 | ||
1984 | gethrestime(&t); | |
1985 | mutex_enter(&dd->dd_lock); | |
1986 | dd->dd_snap_cmtime = t; | |
1987 | mutex_exit(&dd->dd_lock); | |
1988 | } | |
c28b2279 | 1989 | |
fa86b5db MA |
1990 | void |
1991 | dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx) | |
1992 | { | |
1993 | objset_t *mos = dd->dd_pool->dp_meta_objset; | |
1994 | dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx); | |
1995 | } | |
1996 | ||
788eb90c JJ |
1997 | boolean_t |
1998 | dsl_dir_is_zapified(dsl_dir_t *dd) | |
1999 | { | |
2000 | dmu_object_info_t doi; | |
2001 | ||
2002 | dmu_object_info_from_db(dd->dd_dbuf, &doi); | |
2003 | return (doi.doi_type == DMU_OTN_ZAP_METADATA); | |
2004 | } | |
2005 | ||
c28b2279 BB |
2006 | #if defined(_KERNEL) && defined(HAVE_SPL) |
2007 | EXPORT_SYMBOL(dsl_dir_set_quota); | |
2008 | EXPORT_SYMBOL(dsl_dir_set_reservation); | |
c28b2279 | 2009 | #endif |