Print this page
patch first-pass
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/uts/common/fs/zfs/dsl_dir.c
+++ new/usr/src/uts/common/fs/zfs/dsl_dir.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
25 25 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
26 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 27 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
28 28 */
29 29
30 30 #include <sys/dmu.h>
31 31 #include <sys/dmu_objset.h>
32 32 #include <sys/dmu_tx.h>
33 33 #include <sys/dsl_dataset.h>
34 34 #include <sys/dsl_dir.h>
35 35 #include <sys/dsl_prop.h>
36 36 #include <sys/dsl_synctask.h>
37 37 #include <sys/dsl_deleg.h>
38 38 #include <sys/dmu_impl.h>
39 39 #include <sys/spa.h>
40 40 #include <sys/metaslab.h>
41 41 #include <sys/zap.h>
42 42 #include <sys/zio.h>
43 43 #include <sys/arc.h>
44 44 #include <sys/sunddi.h>
45 45 #include <sys/zfeature.h>
46 46 #include <sys/policy.h>
47 47 #include <sys/zfs_znode.h>
48 48 #include "zfs_namecheck.h"
49 49 #include "zfs_prop.h"
50 50
51 51 /*
52 52 * Filesystem and Snapshot Limits
53 53 * ------------------------------
54 54 *
55 55 * These limits are used to restrict the number of filesystems and/or snapshots
56 56 * that can be created at a given level in the tree or below. A typical
57 57 * use-case is with a delegated dataset where the administrator wants to ensure
58 58 * that a user within the zone is not creating too many additional filesystems
59 59 * or snapshots, even though they're not exceeding their space quota.
60 60 *
61 61 * The filesystem and snapshot counts are stored as extensible properties. This
62 62 * capability is controlled by a feature flag and must be enabled to be used.
63 63 * Once enabled, the feature is not active until the first limit is set. At
64 64 * that point, future operations to create/destroy filesystems or snapshots
65 65 * will validate and update the counts.
66 66 *
67 67 * Because the count properties will not exist before the feature is active,
68 68 * the counts are updated when a limit is first set on an uninitialized
69 69 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
70 70 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
71 71 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
72 72 * snapshot count properties on a node indicate uninitialized counts on that
73 73 * node.) When first setting a limit on an uninitialized node, the code starts
74 74 * at the filesystem with the new limit and descends into all sub-filesystems
75 75 * to add the count properties.
76 76 *
77 77 * In practice this is lightweight since a limit is typically set when the
78 78 * filesystem is created and thus has no children. Once valid, changing the
79 79 * limit value won't require a re-traversal since the counts are already valid.
80 80 * When recursively fixing the counts, if a node with a limit is encountered
81 81 * during the descent, the counts are known to be valid and there is no need to
82 82 * descend into that filesystem's children. The counts on filesystems above the
83 83 * one with the new limit will still be uninitialized, unless a limit is
84 84 * eventually set on one of those filesystems. The counts are always recursively
85 85 * updated when a limit is set on a dataset, unless there is already a limit.
86 86 * When a new limit value is set on a filesystem with an existing limit, it is
87 87 * possible for the new limit to be less than the current count at that level
88 88 * since a user who can change the limit is also allowed to exceed the limit.
89 89 *
90 90 * Once the feature is active, then whenever a filesystem or snapshot is
91 91 * created, the code recurses up the tree, validating the new count against the
92 92 * limit at each initialized level. In practice, most levels will not have a
93 93 * limit set. If there is a limit at any initialized level up the tree, the
94 94 * check must pass or the creation will fail. Likewise, when a filesystem or
95 95 * snapshot is destroyed, the counts are recursively adjusted all the way up
96 96 * the initizized nodes in the tree. Renaming a filesystem into different point
97 97 * in the tree will first validate, then update the counts on each branch up to
98 98 * the common ancestor. A receive will also validate the counts and then update
99 99 * them.
100 100 *
101 101 * An exception to the above behavior is that the limit is not enforced if the
102 102 * user has permission to modify the limit. This is primarily so that
103 103 * recursive snapshots in the global zone always work. We want to prevent a
104 104 * denial-of-service in which a lower level delegated dataset could max out its
105 105 * limit and thus block recursive snapshots from being taken in the global zone.
106 106 * Because of this, it is possible for the snapshot count to be over the limit
107 107 * and snapshots taken in the global zone could cause a lower level dataset to
108 108 * hit or exceed its limit. The administrator taking the global zone recursive
109 109 * snapshot should be aware of this side-effect and behave accordingly.
110 110 * For consistency, the filesystem limit is also not enforced if the user can
111 111 * modify the limit.
112 112 *
113 113 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
114 114 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
115 115 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
116 116 * dsl_dir_init_fs_ss_count().
117 117 *
118 118 * There is a special case when we receive a filesystem that already exists. In
119 119 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
120 120 * never update the filesystem counts for temporary clones.
121 121 *
122 122 * Likewise, we do not update the snapshot counts for temporary snapshots,
123 123 * such as those created by zfs diff.
124 124 */
125 125
126 126 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
127 127
128 128 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
129 129
130 130 static void
131 131 dsl_dir_evict(void *dbu)
132 132 {
133 133 dsl_dir_t *dd = dbu;
134 134 dsl_pool_t *dp = dd->dd_pool;
135 135 int t;
136 136
137 137 dd->dd_dbuf = NULL;
138 138
139 139 for (t = 0; t < TXG_SIZE; t++) {
140 140 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
141 141 ASSERT(dd->dd_tempreserved[t] == 0);
142 142 ASSERT(dd->dd_space_towrite[t] == 0);
143 143 }
144 144
145 145 if (dd->dd_parent)
146 146 dsl_dir_async_rele(dd->dd_parent, dd);
147 147
148 148 spa_async_close(dd->dd_pool->dp_spa, dd);
149 149
150 150 dsl_prop_fini(dd);
151 151 mutex_destroy(&dd->dd_lock);
152 152 kmem_free(dd, sizeof (dsl_dir_t));
153 153 }
154 154
155 155 int
156 156 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
157 157 const char *tail, void *tag, dsl_dir_t **ddp)
158 158 {
159 159 dmu_buf_t *dbuf;
160 160 dsl_dir_t *dd;
161 161 int err;
162 162
163 163 ASSERT(dsl_pool_config_held(dp));
164 164
165 165 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
166 166 if (err != 0)
167 167 return (err);
168 168 dd = dmu_buf_get_user(dbuf);
169 169 #ifdef ZFS_DEBUG
170 170 {
171 171 dmu_object_info_t doi;
172 172 dmu_object_info_from_db(dbuf, &doi);
173 173 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
174 174 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
175 175 }
176 176 #endif
177 177 if (dd == NULL) {
178 178 dsl_dir_t *winner;
179 179
180 180 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
181 181 dd->dd_object = ddobj;
182 182 dd->dd_dbuf = dbuf;
183 183 dd->dd_pool = dp;
184 184 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
185 185 dsl_prop_init(dd);
186 186
187 187 dsl_dir_snap_cmtime_update(dd);
188 188
189 189 if (dsl_dir_phys(dd)->dd_parent_obj) {
190 190 err = dsl_dir_hold_obj(dp,
191 191 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
192 192 &dd->dd_parent);
193 193 if (err != 0)
194 194 goto errout;
195 195 if (tail) {
196 196 #ifdef ZFS_DEBUG
197 197 uint64_t foundobj;
198 198
199 199 err = zap_lookup(dp->dp_meta_objset,
200 200 dsl_dir_phys(dd->dd_parent)->
201 201 dd_child_dir_zapobj, tail,
202 202 sizeof (foundobj), 1, &foundobj);
203 203 ASSERT(err || foundobj == ddobj);
204 204 #endif
205 205 (void) strcpy(dd->dd_myname, tail);
206 206 } else {
207 207 err = zap_value_search(dp->dp_meta_objset,
208 208 dsl_dir_phys(dd->dd_parent)->
209 209 dd_child_dir_zapobj,
210 210 ddobj, 0, dd->dd_myname);
211 211 }
212 212 if (err != 0)
213 213 goto errout;
214 214 } else {
215 215 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
216 216 }
217 217
218 218 if (dsl_dir_is_clone(dd)) {
219 219 dmu_buf_t *origin_bonus;
220 220 dsl_dataset_phys_t *origin_phys;
221 221
222 222 /*
223 223 * We can't open the origin dataset, because
224 224 * that would require opening this dsl_dir.
225 225 * Just look at its phys directly instead.
226 226 */
227 227 err = dmu_bonus_hold(dp->dp_meta_objset,
↓ open down ↓ |
227 lines elided |
↑ open up ↑ |
228 228 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
229 229 &origin_bonus);
230 230 if (err != 0)
231 231 goto errout;
232 232 origin_phys = origin_bonus->db_data;
233 233 dd->dd_origin_txg =
234 234 origin_phys->ds_creation_txg;
235 235 dmu_buf_rele(origin_bonus, FTAG);
236 236 }
237 237
238 - dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf);
238 + dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict,
239 + &dd->dd_dbuf);
239 240 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
240 241 if (winner != NULL) {
241 242 if (dd->dd_parent)
242 243 dsl_dir_rele(dd->dd_parent, dd);
243 244 dsl_prop_fini(dd);
244 245 mutex_destroy(&dd->dd_lock);
245 246 kmem_free(dd, sizeof (dsl_dir_t));
246 247 dd = winner;
247 248 } else {
248 249 spa_open_ref(dp->dp_spa, dd);
249 250 }
250 251 }
251 252
252 253 /*
253 254 * The dsl_dir_t has both open-to-close and instantiate-to-evict
254 255 * holds on the spa. We need the open-to-close holds because
255 256 * otherwise the spa_refcnt wouldn't change when we open a
256 257 * dir which the spa also has open, so we could incorrectly
257 258 * think it was OK to unload/export/destroy the pool. We need
258 259 * the instantiate-to-evict hold because the dsl_dir_t has a
259 260 * pointer to the dd_pool, which has a pointer to the spa_t.
260 261 */
261 262 spa_open_ref(dp->dp_spa, tag);
262 263 ASSERT3P(dd->dd_pool, ==, dp);
263 264 ASSERT3U(dd->dd_object, ==, ddobj);
264 265 ASSERT3P(dd->dd_dbuf, ==, dbuf);
265 266 *ddp = dd;
266 267 return (0);
267 268
268 269 errout:
269 270 if (dd->dd_parent)
270 271 dsl_dir_rele(dd->dd_parent, dd);
271 272 dsl_prop_fini(dd);
272 273 mutex_destroy(&dd->dd_lock);
273 274 kmem_free(dd, sizeof (dsl_dir_t));
274 275 dmu_buf_rele(dbuf, tag);
275 276 return (err);
276 277 }
277 278
278 279 void
279 280 dsl_dir_rele(dsl_dir_t *dd, void *tag)
280 281 {
281 282 dprintf_dd(dd, "%s\n", "");
282 283 spa_close(dd->dd_pool->dp_spa, tag);
283 284 dmu_buf_rele(dd->dd_dbuf, tag);
284 285 }
285 286
286 287 /*
287 288 * Remove a reference to the given dsl dir that is being asynchronously
288 289 * released. Async releases occur from a taskq performing eviction of
289 290 * dsl datasets and dirs. This process is identical to a normal release
290 291 * with the exception of using the async API for releasing the reference on
291 292 * the spa.
292 293 */
293 294 void
294 295 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
295 296 {
296 297 dprintf_dd(dd, "%s\n", "");
297 298 spa_async_close(dd->dd_pool->dp_spa, tag);
298 299 dmu_buf_rele(dd->dd_dbuf, tag);
299 300 }
300 301
301 302 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
302 303 void
303 304 dsl_dir_name(dsl_dir_t *dd, char *buf)
304 305 {
305 306 if (dd->dd_parent) {
306 307 dsl_dir_name(dd->dd_parent, buf);
307 308 (void) strcat(buf, "/");
308 309 } else {
309 310 buf[0] = '\0';
310 311 }
311 312 if (!MUTEX_HELD(&dd->dd_lock)) {
312 313 /*
313 314 * recursive mutex so that we can use
314 315 * dprintf_dd() with dd_lock held
315 316 */
316 317 mutex_enter(&dd->dd_lock);
317 318 (void) strcat(buf, dd->dd_myname);
318 319 mutex_exit(&dd->dd_lock);
319 320 } else {
320 321 (void) strcat(buf, dd->dd_myname);
321 322 }
322 323 }
323 324
324 325 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
325 326 int
326 327 dsl_dir_namelen(dsl_dir_t *dd)
327 328 {
328 329 int result = 0;
329 330
330 331 if (dd->dd_parent) {
331 332 /* parent's name + 1 for the "/" */
332 333 result = dsl_dir_namelen(dd->dd_parent) + 1;
333 334 }
334 335
335 336 if (!MUTEX_HELD(&dd->dd_lock)) {
336 337 /* see dsl_dir_name */
337 338 mutex_enter(&dd->dd_lock);
338 339 result += strlen(dd->dd_myname);
339 340 mutex_exit(&dd->dd_lock);
340 341 } else {
341 342 result += strlen(dd->dd_myname);
342 343 }
343 344
344 345 return (result);
345 346 }
346 347
347 348 static int
348 349 getcomponent(const char *path, char *component, const char **nextp)
349 350 {
350 351 char *p;
351 352
352 353 if ((path == NULL) || (path[0] == '\0'))
353 354 return (SET_ERROR(ENOENT));
354 355 /* This would be a good place to reserve some namespace... */
355 356 p = strpbrk(path, "/@");
356 357 if (p && (p[1] == '/' || p[1] == '@')) {
357 358 /* two separators in a row */
358 359 return (SET_ERROR(EINVAL));
359 360 }
360 361 if (p == NULL || p == path) {
361 362 /*
362 363 * if the first thing is an @ or /, it had better be an
363 364 * @ and it had better not have any more ats or slashes,
364 365 * and it had better have something after the @.
365 366 */
366 367 if (p != NULL &&
367 368 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
368 369 return (SET_ERROR(EINVAL));
369 370 if (strlen(path) >= MAXNAMELEN)
370 371 return (SET_ERROR(ENAMETOOLONG));
371 372 (void) strcpy(component, path);
372 373 p = NULL;
373 374 } else if (p[0] == '/') {
374 375 if (p - path >= MAXNAMELEN)
375 376 return (SET_ERROR(ENAMETOOLONG));
376 377 (void) strncpy(component, path, p - path);
377 378 component[p - path] = '\0';
378 379 p++;
379 380 } else if (p[0] == '@') {
380 381 /*
381 382 * if the next separator is an @, there better not be
382 383 * any more slashes.
383 384 */
384 385 if (strchr(path, '/'))
385 386 return (SET_ERROR(EINVAL));
386 387 if (p - path >= MAXNAMELEN)
387 388 return (SET_ERROR(ENAMETOOLONG));
388 389 (void) strncpy(component, path, p - path);
389 390 component[p - path] = '\0';
390 391 } else {
391 392 panic("invalid p=%p", (void *)p);
392 393 }
393 394 *nextp = p;
394 395 return (0);
395 396 }
396 397
397 398 /*
398 399 * Return the dsl_dir_t, and possibly the last component which couldn't
399 400 * be found in *tail. The name must be in the specified dsl_pool_t. This
400 401 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
401 402 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
402 403 * (*tail)[0] == '@' means that the last component is a snapshot.
403 404 */
404 405 int
405 406 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
406 407 dsl_dir_t **ddp, const char **tailp)
407 408 {
408 409 char buf[MAXNAMELEN];
409 410 const char *spaname, *next, *nextnext = NULL;
410 411 int err;
411 412 dsl_dir_t *dd;
412 413 uint64_t ddobj;
413 414
414 415 err = getcomponent(name, buf, &next);
415 416 if (err != 0)
416 417 return (err);
417 418
418 419 /* Make sure the name is in the specified pool. */
419 420 spaname = spa_name(dp->dp_spa);
420 421 if (strcmp(buf, spaname) != 0)
421 422 return (SET_ERROR(EXDEV));
422 423
423 424 ASSERT(dsl_pool_config_held(dp));
424 425
425 426 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
426 427 if (err != 0) {
427 428 return (err);
428 429 }
429 430
430 431 while (next != NULL) {
431 432 dsl_dir_t *child_dd;
432 433 err = getcomponent(next, buf, &nextnext);
433 434 if (err != 0)
434 435 break;
435 436 ASSERT(next[0] != '\0');
436 437 if (next[0] == '@')
437 438 break;
438 439 dprintf("looking up %s in obj%lld\n",
439 440 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
440 441
441 442 err = zap_lookup(dp->dp_meta_objset,
442 443 dsl_dir_phys(dd)->dd_child_dir_zapobj,
443 444 buf, sizeof (ddobj), 1, &ddobj);
444 445 if (err != 0) {
445 446 if (err == ENOENT)
446 447 err = 0;
447 448 break;
448 449 }
449 450
450 451 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
451 452 if (err != 0)
452 453 break;
453 454 dsl_dir_rele(dd, tag);
454 455 dd = child_dd;
455 456 next = nextnext;
456 457 }
457 458
458 459 if (err != 0) {
459 460 dsl_dir_rele(dd, tag);
460 461 return (err);
461 462 }
462 463
463 464 /*
464 465 * It's an error if there's more than one component left, or
465 466 * tailp==NULL and there's any component left.
466 467 */
467 468 if (next != NULL &&
468 469 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
469 470 /* bad path name */
470 471 dsl_dir_rele(dd, tag);
471 472 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
472 473 err = SET_ERROR(ENOENT);
473 474 }
474 475 if (tailp != NULL)
475 476 *tailp = next;
476 477 *ddp = dd;
477 478 return (err);
478 479 }
479 480
480 481 /*
481 482 * If the counts are already initialized for this filesystem and its
482 483 * descendants then do nothing, otherwise initialize the counts.
483 484 *
484 485 * The counts on this filesystem, and those below, may be uninitialized due to
485 486 * either the use of a pre-existing pool which did not support the
486 487 * filesystem/snapshot limit feature, or one in which the feature had not yet
487 488 * been enabled.
488 489 *
489 490 * Recursively descend the filesystem tree and update the filesystem/snapshot
490 491 * counts on each filesystem below, then update the cumulative count on the
491 492 * current filesystem. If the filesystem already has a count set on it,
492 493 * then we know that its counts, and the counts on the filesystems below it,
493 494 * are already correct, so we don't have to update this filesystem.
494 495 */
495 496 static void
496 497 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
497 498 {
498 499 uint64_t my_fs_cnt = 0;
499 500 uint64_t my_ss_cnt = 0;
500 501 dsl_pool_t *dp = dd->dd_pool;
501 502 objset_t *os = dp->dp_meta_objset;
502 503 zap_cursor_t *zc;
503 504 zap_attribute_t *za;
504 505 dsl_dataset_t *ds;
505 506
506 507 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
507 508 ASSERT(dsl_pool_config_held(dp));
508 509 ASSERT(dmu_tx_is_syncing(tx));
509 510
510 511 dsl_dir_zapify(dd, tx);
511 512
512 513 /*
513 514 * If the filesystem count has already been initialized then we
514 515 * don't need to recurse down any further.
515 516 */
516 517 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
517 518 return;
518 519
519 520 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
520 521 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
521 522
522 523 /* Iterate my child dirs */
523 524 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
524 525 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
525 526 dsl_dir_t *chld_dd;
526 527 uint64_t count;
527 528
528 529 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
529 530 &chld_dd));
530 531
531 532 /*
532 533 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
533 534 * temporary datasets.
534 535 */
535 536 if (chld_dd->dd_myname[0] == '$' ||
536 537 chld_dd->dd_myname[0] == '%') {
537 538 dsl_dir_rele(chld_dd, FTAG);
538 539 continue;
539 540 }
540 541
541 542 my_fs_cnt++; /* count this child */
542 543
543 544 dsl_dir_init_fs_ss_count(chld_dd, tx);
544 545
545 546 VERIFY0(zap_lookup(os, chld_dd->dd_object,
546 547 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
547 548 my_fs_cnt += count;
548 549 VERIFY0(zap_lookup(os, chld_dd->dd_object,
549 550 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
550 551 my_ss_cnt += count;
551 552
552 553 dsl_dir_rele(chld_dd, FTAG);
553 554 }
554 555 zap_cursor_fini(zc);
555 556 /* Count my snapshots (we counted children's snapshots above) */
556 557 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
557 558 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
558 559
559 560 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
560 561 zap_cursor_retrieve(zc, za) == 0;
561 562 zap_cursor_advance(zc)) {
562 563 /* Don't count temporary snapshots */
563 564 if (za->za_name[0] != '%')
564 565 my_ss_cnt++;
565 566 }
566 567 zap_cursor_fini(zc);
567 568
568 569 dsl_dataset_rele(ds, FTAG);
569 570
570 571 kmem_free(zc, sizeof (zap_cursor_t));
571 572 kmem_free(za, sizeof (zap_attribute_t));
572 573
573 574 /* we're in a sync task, update counts */
574 575 dmu_buf_will_dirty(dd->dd_dbuf, tx);
575 576 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
576 577 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
577 578 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
578 579 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
579 580 }
580 581
581 582 static int
582 583 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
583 584 {
584 585 char *ddname = (char *)arg;
585 586 dsl_pool_t *dp = dmu_tx_pool(tx);
586 587 dsl_dataset_t *ds;
587 588 dsl_dir_t *dd;
588 589 int error;
589 590
590 591 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
591 592 if (error != 0)
592 593 return (error);
593 594
594 595 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
595 596 dsl_dataset_rele(ds, FTAG);
596 597 return (SET_ERROR(ENOTSUP));
597 598 }
598 599
599 600 dd = ds->ds_dir;
600 601 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
601 602 dsl_dir_is_zapified(dd) &&
602 603 zap_contains(dp->dp_meta_objset, dd->dd_object,
603 604 DD_FIELD_FILESYSTEM_COUNT) == 0) {
604 605 dsl_dataset_rele(ds, FTAG);
605 606 return (SET_ERROR(EALREADY));
606 607 }
607 608
608 609 dsl_dataset_rele(ds, FTAG);
609 610 return (0);
610 611 }
611 612
612 613 static void
613 614 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
614 615 {
615 616 char *ddname = (char *)arg;
616 617 dsl_pool_t *dp = dmu_tx_pool(tx);
617 618 dsl_dataset_t *ds;
618 619 spa_t *spa;
619 620
620 621 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
621 622
622 623 spa = dsl_dataset_get_spa(ds);
623 624
624 625 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
625 626 /*
626 627 * Since the feature was not active and we're now setting a
627 628 * limit, increment the feature-active counter so that the
628 629 * feature becomes active for the first time.
629 630 *
630 631 * We are already in a sync task so we can update the MOS.
631 632 */
632 633 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
633 634 }
634 635
635 636 /*
636 637 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
637 638 * we need to ensure the counts are correct. Descend down the tree from
638 639 * this point and update all of the counts to be accurate.
639 640 */
640 641 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
641 642
642 643 dsl_dataset_rele(ds, FTAG);
643 644 }
644 645
645 646 /*
646 647 * Make sure the feature is enabled and activate it if necessary.
647 648 * Since we're setting a limit, ensure the on-disk counts are valid.
648 649 * This is only called by the ioctl path when setting a limit value.
649 650 *
650 651 * We do not need to validate the new limit, since users who can change the
651 652 * limit are also allowed to exceed the limit.
652 653 */
653 654 int
654 655 dsl_dir_activate_fs_ss_limit(const char *ddname)
655 656 {
656 657 int error;
657 658
658 659 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
659 660 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
660 661 ZFS_SPACE_CHECK_RESERVED);
661 662
662 663 if (error == EALREADY)
663 664 error = 0;
664 665
665 666 return (error);
666 667 }
667 668
668 669 /*
669 670 * Used to determine if the filesystem_limit or snapshot_limit should be
670 671 * enforced. We allow the limit to be exceeded if the user has permission to
671 672 * write the property value. We pass in the creds that we got in the open
672 673 * context since we will always be the GZ root in syncing context. We also have
673 674 * to handle the case where we are allowed to change the limit on the current
674 675 * dataset, but there may be another limit in the tree above.
675 676 *
676 677 * We can never modify these two properties within a non-global zone. In
677 678 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
678 679 * can't use that function since we are already holding the dp_config_rwlock.
679 680 * In addition, we already have the dd and dealing with snapshots is simplified
680 681 * in this code.
681 682 */
682 683
683 684 typedef enum {
684 685 ENFORCE_ALWAYS,
685 686 ENFORCE_NEVER,
686 687 ENFORCE_ABOVE
687 688 } enforce_res_t;
688 689
689 690 static enforce_res_t
690 691 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
691 692 {
692 693 enforce_res_t enforce = ENFORCE_ALWAYS;
693 694 uint64_t obj;
694 695 dsl_dataset_t *ds;
695 696 uint64_t zoned;
696 697
697 698 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
698 699 prop == ZFS_PROP_SNAPSHOT_LIMIT);
699 700
700 701 #ifdef _KERNEL
701 702 if (crgetzoneid(cr) != GLOBAL_ZONEID)
702 703 return (ENFORCE_ALWAYS);
703 704
704 705 if (secpolicy_zfs(cr) == 0)
705 706 return (ENFORCE_NEVER);
706 707 #endif
707 708
708 709 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
709 710 return (ENFORCE_ALWAYS);
710 711
711 712 ASSERT(dsl_pool_config_held(dd->dd_pool));
712 713
713 714 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
714 715 return (ENFORCE_ALWAYS);
715 716
716 717 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
717 718 /* Only root can access zoned fs's from the GZ */
718 719 enforce = ENFORCE_ALWAYS;
719 720 } else {
720 721 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
721 722 enforce = ENFORCE_ABOVE;
722 723 }
723 724
724 725 dsl_dataset_rele(ds, FTAG);
725 726 return (enforce);
726 727 }
727 728
728 729 /*
729 730 * Check if adding additional child filesystem(s) would exceed any filesystem
730 731 * limits or adding additional snapshot(s) would exceed any snapshot limits.
731 732 * The prop argument indicates which limit to check.
732 733 *
733 734 * Note that all filesystem limits up to the root (or the highest
734 735 * initialized) filesystem or the given ancestor must be satisfied.
735 736 */
736 737 int
737 738 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
738 739 dsl_dir_t *ancestor, cred_t *cr)
739 740 {
740 741 objset_t *os = dd->dd_pool->dp_meta_objset;
741 742 uint64_t limit, count;
742 743 char *count_prop;
743 744 enforce_res_t enforce;
744 745 int err = 0;
745 746
746 747 ASSERT(dsl_pool_config_held(dd->dd_pool));
747 748 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
748 749 prop == ZFS_PROP_SNAPSHOT_LIMIT);
749 750
750 751 /*
751 752 * If we're allowed to change the limit, don't enforce the limit
752 753 * e.g. this can happen if a snapshot is taken by an administrative
753 754 * user in the global zone (i.e. a recursive snapshot by root).
754 755 * However, we must handle the case of delegated permissions where we
755 756 * are allowed to change the limit on the current dataset, but there
756 757 * is another limit in the tree above.
757 758 */
758 759 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
759 760 if (enforce == ENFORCE_NEVER)
760 761 return (0);
761 762
762 763 /*
763 764 * e.g. if renaming a dataset with no snapshots, count adjustment
764 765 * is 0.
765 766 */
766 767 if (delta == 0)
767 768 return (0);
768 769
769 770 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
770 771 /*
771 772 * We don't enforce the limit for temporary snapshots. This is
772 773 * indicated by a NULL cred_t argument.
773 774 */
774 775 if (cr == NULL)
775 776 return (0);
776 777
777 778 count_prop = DD_FIELD_SNAPSHOT_COUNT;
778 779 } else {
779 780 count_prop = DD_FIELD_FILESYSTEM_COUNT;
780 781 }
781 782
782 783 /*
783 784 * If an ancestor has been provided, stop checking the limit once we
784 785 * hit that dir. We need this during rename so that we don't overcount
785 786 * the check once we recurse up to the common ancestor.
786 787 */
787 788 if (ancestor == dd)
788 789 return (0);
789 790
790 791 /*
791 792 * If we hit an uninitialized node while recursing up the tree, we can
792 793 * stop since we know there is no limit here (or above). The counts are
793 794 * not valid on this node and we know we won't touch this node's counts.
794 795 */
795 796 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
796 797 count_prop, sizeof (count), 1, &count) == ENOENT)
797 798 return (0);
798 799
799 800 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
800 801 B_FALSE);
801 802 if (err != 0)
802 803 return (err);
803 804
804 805 /* Is there a limit which we've hit? */
805 806 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
806 807 return (SET_ERROR(EDQUOT));
807 808
808 809 if (dd->dd_parent != NULL)
809 810 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
810 811 ancestor, cr);
811 812
812 813 return (err);
813 814 }
814 815
815 816 /*
816 817 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
817 818 * parents. When a new filesystem/snapshot is created, increment the count on
818 819 * all parents, and when a filesystem/snapshot is destroyed, decrement the
819 820 * count.
820 821 */
821 822 void
822 823 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
823 824 dmu_tx_t *tx)
824 825 {
825 826 int err;
826 827 objset_t *os = dd->dd_pool->dp_meta_objset;
827 828 uint64_t count;
828 829
829 830 ASSERT(dsl_pool_config_held(dd->dd_pool));
830 831 ASSERT(dmu_tx_is_syncing(tx));
831 832 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
832 833 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
833 834
834 835 /*
835 836 * When we receive an incremental stream into a filesystem that already
836 837 * exists, a temporary clone is created. We don't count this temporary
837 838 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
838 839 * $MOS & $ORIGIN) objsets.
839 840 */
840 841 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
841 842 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
842 843 return;
843 844
844 845 /*
845 846 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
846 847 */
847 848 if (delta == 0)
848 849 return;
849 850
850 851 /*
851 852 * If we hit an uninitialized node while recursing up the tree, we can
852 853 * stop since we know the counts are not valid on this node and we
853 854 * know we shouldn't touch this node's counts. An uninitialized count
854 855 * on the node indicates that either the feature has not yet been
855 856 * activated or there are no limits on this part of the tree.
856 857 */
857 858 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
858 859 prop, sizeof (count), 1, &count)) == ENOENT)
859 860 return;
860 861 VERIFY0(err);
861 862
862 863 count += delta;
863 864 /* Use a signed verify to make sure we're not neg. */
864 865 VERIFY3S(count, >=, 0);
865 866
866 867 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
867 868 tx));
868 869
869 870 /* Roll up this additional count into our ancestors */
870 871 if (dd->dd_parent != NULL)
871 872 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
872 873 }
873 874
874 875 uint64_t
875 876 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
876 877 dmu_tx_t *tx)
877 878 {
878 879 objset_t *mos = dp->dp_meta_objset;
879 880 uint64_t ddobj;
880 881 dsl_dir_phys_t *ddphys;
881 882 dmu_buf_t *dbuf;
882 883
883 884 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
884 885 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
885 886 if (pds) {
886 887 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
887 888 name, sizeof (uint64_t), 1, &ddobj, tx));
888 889 } else {
889 890 /* it's the root dir */
890 891 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
891 892 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
892 893 }
893 894 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
894 895 dmu_buf_will_dirty(dbuf, tx);
895 896 ddphys = dbuf->db_data;
896 897
897 898 ddphys->dd_creation_time = gethrestime_sec();
898 899 if (pds) {
899 900 ddphys->dd_parent_obj = pds->dd_object;
900 901
901 902 /* update the filesystem counts */
902 903 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
903 904 }
904 905 ddphys->dd_props_zapobj = zap_create(mos,
905 906 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
906 907 ddphys->dd_child_dir_zapobj = zap_create(mos,
907 908 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
908 909 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
909 910 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
910 911 dmu_buf_rele(dbuf, FTAG);
911 912
912 913 return (ddobj);
913 914 }
914 915
915 916 boolean_t
916 917 dsl_dir_is_clone(dsl_dir_t *dd)
917 918 {
918 919 return (dsl_dir_phys(dd)->dd_origin_obj &&
919 920 (dd->dd_pool->dp_origin_snap == NULL ||
920 921 dsl_dir_phys(dd)->dd_origin_obj !=
921 922 dd->dd_pool->dp_origin_snap->ds_object));
922 923 }
923 924
924 925 void
925 926 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
926 927 {
927 928 mutex_enter(&dd->dd_lock);
928 929 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
929 930 dsl_dir_phys(dd)->dd_used_bytes);
930 931 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
931 932 dsl_dir_phys(dd)->dd_quota);
932 933 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
933 934 dsl_dir_phys(dd)->dd_reserved);
934 935 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
935 936 dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
936 937 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
937 938 dsl_dir_phys(dd)->dd_compressed_bytes));
938 939 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
939 940 dsl_dir_phys(dd)->dd_uncompressed_bytes);
940 941 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
941 942 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
942 943 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
943 944 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
944 945 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
945 946 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
946 947 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
947 948 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
948 949 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
949 950 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
950 951 }
951 952 mutex_exit(&dd->dd_lock);
952 953
953 954 if (dsl_dir_is_zapified(dd)) {
954 955 uint64_t count;
955 956 objset_t *os = dd->dd_pool->dp_meta_objset;
956 957
957 958 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
958 959 sizeof (count), 1, &count) == 0) {
959 960 dsl_prop_nvlist_add_uint64(nv,
960 961 ZFS_PROP_FILESYSTEM_COUNT, count);
961 962 }
962 963 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
963 964 sizeof (count), 1, &count) == 0) {
964 965 dsl_prop_nvlist_add_uint64(nv,
965 966 ZFS_PROP_SNAPSHOT_COUNT, count);
966 967 }
967 968 }
968 969
969 970 if (dsl_dir_is_clone(dd)) {
970 971 dsl_dataset_t *ds;
971 972 char buf[MAXNAMELEN];
972 973
973 974 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
974 975 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
975 976 dsl_dataset_name(ds, buf);
976 977 dsl_dataset_rele(ds, FTAG);
977 978 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
978 979 }
979 980 }
980 981
981 982 void
982 983 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
983 984 {
984 985 dsl_pool_t *dp = dd->dd_pool;
985 986
986 987 ASSERT(dsl_dir_phys(dd));
987 988
988 989 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
989 990 /* up the hold count until we can be written out */
990 991 dmu_buf_add_ref(dd->dd_dbuf, dd);
991 992 }
992 993 }
993 994
994 995 static int64_t
995 996 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
996 997 {
997 998 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
998 999 uint64_t new_accounted =
999 1000 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1000 1001 return (new_accounted - old_accounted);
1001 1002 }
1002 1003
1003 1004 void
1004 1005 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1005 1006 {
1006 1007 ASSERT(dmu_tx_is_syncing(tx));
1007 1008
1008 1009 mutex_enter(&dd->dd_lock);
1009 1010 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1010 1011 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1011 1012 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1012 1013 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1013 1014 mutex_exit(&dd->dd_lock);
1014 1015
1015 1016 /* release the hold from dsl_dir_dirty */
1016 1017 dmu_buf_rele(dd->dd_dbuf, dd);
1017 1018 }
1018 1019
1019 1020 static uint64_t
1020 1021 dsl_dir_space_towrite(dsl_dir_t *dd)
1021 1022 {
1022 1023 uint64_t space = 0;
1023 1024 int i;
1024 1025
1025 1026 ASSERT(MUTEX_HELD(&dd->dd_lock));
1026 1027
1027 1028 for (i = 0; i < TXG_SIZE; i++) {
1028 1029 space += dd->dd_space_towrite[i&TXG_MASK];
1029 1030 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1030 1031 }
1031 1032 return (space);
1032 1033 }
1033 1034
1034 1035 /*
1035 1036 * How much space would dd have available if ancestor had delta applied
1036 1037 * to it? If ondiskonly is set, we're only interested in what's
1037 1038 * on-disk, not estimated pending changes.
1038 1039 */
1039 1040 uint64_t
1040 1041 dsl_dir_space_available(dsl_dir_t *dd,
1041 1042 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1042 1043 {
1043 1044 uint64_t parentspace, myspace, quota, used;
1044 1045
1045 1046 /*
1046 1047 * If there are no restrictions otherwise, assume we have
1047 1048 * unlimited space available.
1048 1049 */
1049 1050 quota = UINT64_MAX;
1050 1051 parentspace = UINT64_MAX;
1051 1052
1052 1053 if (dd->dd_parent != NULL) {
1053 1054 parentspace = dsl_dir_space_available(dd->dd_parent,
1054 1055 ancestor, delta, ondiskonly);
1055 1056 }
1056 1057
1057 1058 mutex_enter(&dd->dd_lock);
1058 1059 if (dsl_dir_phys(dd)->dd_quota != 0)
1059 1060 quota = dsl_dir_phys(dd)->dd_quota;
1060 1061 used = dsl_dir_phys(dd)->dd_used_bytes;
1061 1062 if (!ondiskonly)
1062 1063 used += dsl_dir_space_towrite(dd);
1063 1064
1064 1065 if (dd->dd_parent == NULL) {
1065 1066 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1066 1067 quota = MIN(quota, poolsize);
1067 1068 }
1068 1069
1069 1070 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1070 1071 /*
1071 1072 * We have some space reserved, in addition to what our
1072 1073 * parent gave us.
1073 1074 */
1074 1075 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1075 1076 }
1076 1077
1077 1078 if (dd == ancestor) {
1078 1079 ASSERT(delta <= 0);
1079 1080 ASSERT(used >= -delta);
1080 1081 used += delta;
1081 1082 if (parentspace != UINT64_MAX)
1082 1083 parentspace -= delta;
1083 1084 }
1084 1085
1085 1086 if (used > quota) {
1086 1087 /* over quota */
1087 1088 myspace = 0;
1088 1089 } else {
1089 1090 /*
1090 1091 * the lesser of the space provided by our parent and
1091 1092 * the space left in our quota
1092 1093 */
1093 1094 myspace = MIN(parentspace, quota - used);
1094 1095 }
1095 1096
1096 1097 mutex_exit(&dd->dd_lock);
1097 1098
1098 1099 return (myspace);
1099 1100 }
1100 1101
1101 1102 struct tempreserve {
1102 1103 list_node_t tr_node;
1103 1104 dsl_dir_t *tr_ds;
1104 1105 uint64_t tr_size;
1105 1106 };
1106 1107
1107 1108 static int
1108 1109 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1109 1110 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1110 1111 dmu_tx_t *tx, boolean_t first)
1111 1112 {
1112 1113 uint64_t txg = tx->tx_txg;
1113 1114 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1114 1115 uint64_t deferred = 0;
1115 1116 struct tempreserve *tr;
1116 1117 int retval = EDQUOT;
1117 1118 int txgidx = txg & TXG_MASK;
1118 1119 int i;
1119 1120 uint64_t ref_rsrv = 0;
1120 1121
1121 1122 ASSERT3U(txg, !=, 0);
1122 1123 ASSERT3S(asize, >, 0);
1123 1124
1124 1125 mutex_enter(&dd->dd_lock);
1125 1126
1126 1127 /*
1127 1128 * Check against the dsl_dir's quota. We don't add in the delta
1128 1129 * when checking for over-quota because they get one free hit.
1129 1130 */
1130 1131 est_inflight = dsl_dir_space_towrite(dd);
1131 1132 for (i = 0; i < TXG_SIZE; i++)
1132 1133 est_inflight += dd->dd_tempreserved[i];
1133 1134 used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1134 1135
1135 1136 /*
1136 1137 * On the first iteration, fetch the dataset's used-on-disk and
1137 1138 * refreservation values. Also, if checkrefquota is set, test if
1138 1139 * allocating this space would exceed the dataset's refquota.
1139 1140 */
1140 1141 if (first && tx->tx_objset) {
1141 1142 int error;
1142 1143 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1143 1144
1144 1145 error = dsl_dataset_check_quota(ds, checkrefquota,
1145 1146 asize, est_inflight, &used_on_disk, &ref_rsrv);
1146 1147 if (error) {
1147 1148 mutex_exit(&dd->dd_lock);
1148 1149 return (error);
1149 1150 }
1150 1151 }
1151 1152
1152 1153 /*
1153 1154 * If this transaction will result in a net free of space,
1154 1155 * we want to let it through.
1155 1156 */
1156 1157 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1157 1158 quota = UINT64_MAX;
1158 1159 else
1159 1160 quota = dsl_dir_phys(dd)->dd_quota;
1160 1161
1161 1162 /*
1162 1163 * Adjust the quota against the actual pool size at the root
1163 1164 * minus any outstanding deferred frees.
1164 1165 * To ensure that it's possible to remove files from a full
1165 1166 * pool without inducing transient overcommits, we throttle
1166 1167 * netfree transactions against a quota that is slightly larger,
1167 1168 * but still within the pool's allocation slop. In cases where
1168 1169 * we're very close to full, this will allow a steady trickle of
1169 1170 * removes to get through.
1170 1171 */
1171 1172 if (dd->dd_parent == NULL) {
1172 1173 spa_t *spa = dd->dd_pool->dp_spa;
1173 1174 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1174 1175 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1175 1176 if (poolsize - deferred < quota) {
1176 1177 quota = poolsize - deferred;
1177 1178 retval = ENOSPC;
1178 1179 }
1179 1180 }
1180 1181
1181 1182 /*
1182 1183 * If they are requesting more space, and our current estimate
1183 1184 * is over quota, they get to try again unless the actual
1184 1185 * on-disk is over quota and there are no pending changes (which
1185 1186 * may free up space for us).
1186 1187 */
1187 1188 if (used_on_disk + est_inflight >= quota) {
1188 1189 if (est_inflight > 0 || used_on_disk < quota ||
1189 1190 (retval == ENOSPC && used_on_disk < quota + deferred))
1190 1191 retval = ERESTART;
1191 1192 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1192 1193 "quota=%lluK tr=%lluK err=%d\n",
1193 1194 used_on_disk>>10, est_inflight>>10,
1194 1195 quota>>10, asize>>10, retval);
1195 1196 mutex_exit(&dd->dd_lock);
1196 1197 return (SET_ERROR(retval));
1197 1198 }
1198 1199
1199 1200 /* We need to up our estimated delta before dropping dd_lock */
1200 1201 dd->dd_tempreserved[txgidx] += asize;
1201 1202
1202 1203 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1203 1204 asize - ref_rsrv);
1204 1205 mutex_exit(&dd->dd_lock);
1205 1206
1206 1207 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1207 1208 tr->tr_ds = dd;
1208 1209 tr->tr_size = asize;
1209 1210 list_insert_tail(tr_list, tr);
1210 1211
1211 1212 /* see if it's OK with our parent */
1212 1213 if (dd->dd_parent && parent_rsrv) {
1213 1214 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1214 1215
1215 1216 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1216 1217 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1217 1218 } else {
1218 1219 return (0);
1219 1220 }
1220 1221 }
1221 1222
1222 1223 /*
1223 1224 * Reserve space in this dsl_dir, to be used in this tx's txg.
1224 1225 * After the space has been dirtied (and dsl_dir_willuse_space()
1225 1226 * has been called), the reservation should be canceled, using
1226 1227 * dsl_dir_tempreserve_clear().
1227 1228 */
1228 1229 int
1229 1230 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1230 1231 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1231 1232 {
1232 1233 int err;
1233 1234 list_t *tr_list;
1234 1235
1235 1236 if (asize == 0) {
1236 1237 *tr_cookiep = NULL;
1237 1238 return (0);
1238 1239 }
1239 1240
1240 1241 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1241 1242 list_create(tr_list, sizeof (struct tempreserve),
1242 1243 offsetof(struct tempreserve, tr_node));
1243 1244 ASSERT3S(asize, >, 0);
1244 1245 ASSERT3S(fsize, >=, 0);
1245 1246
1246 1247 err = arc_tempreserve_space(lsize, tx->tx_txg);
1247 1248 if (err == 0) {
1248 1249 struct tempreserve *tr;
1249 1250
1250 1251 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1251 1252 tr->tr_size = lsize;
1252 1253 list_insert_tail(tr_list, tr);
1253 1254 } else {
1254 1255 if (err == EAGAIN) {
1255 1256 /*
1256 1257 * If arc_memory_throttle() detected that pageout
1257 1258 * is running and we are low on memory, we delay new
1258 1259 * non-pageout transactions to give pageout an
1259 1260 * advantage.
1260 1261 *
1261 1262 * It is unfortunate to be delaying while the caller's
1262 1263 * locks are held.
1263 1264 */
1264 1265 txg_delay(dd->dd_pool, tx->tx_txg,
1265 1266 MSEC2NSEC(10), MSEC2NSEC(10));
1266 1267 err = SET_ERROR(ERESTART);
1267 1268 }
1268 1269 }
1269 1270
1270 1271 if (err == 0) {
1271 1272 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1272 1273 FALSE, asize > usize, tr_list, tx, TRUE);
1273 1274 }
1274 1275
1275 1276 if (err != 0)
1276 1277 dsl_dir_tempreserve_clear(tr_list, tx);
1277 1278 else
1278 1279 *tr_cookiep = tr_list;
1279 1280
1280 1281 return (err);
1281 1282 }
1282 1283
1283 1284 /*
1284 1285 * Clear a temporary reservation that we previously made with
1285 1286 * dsl_dir_tempreserve_space().
1286 1287 */
1287 1288 void
1288 1289 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1289 1290 {
1290 1291 int txgidx = tx->tx_txg & TXG_MASK;
1291 1292 list_t *tr_list = tr_cookie;
1292 1293 struct tempreserve *tr;
1293 1294
1294 1295 ASSERT3U(tx->tx_txg, !=, 0);
1295 1296
1296 1297 if (tr_cookie == NULL)
1297 1298 return;
1298 1299
1299 1300 while ((tr = list_head(tr_list)) != NULL) {
1300 1301 if (tr->tr_ds) {
1301 1302 mutex_enter(&tr->tr_ds->dd_lock);
1302 1303 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1303 1304 tr->tr_size);
1304 1305 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1305 1306 mutex_exit(&tr->tr_ds->dd_lock);
1306 1307 } else {
1307 1308 arc_tempreserve_clear(tr->tr_size);
1308 1309 }
1309 1310 list_remove(tr_list, tr);
1310 1311 kmem_free(tr, sizeof (struct tempreserve));
1311 1312 }
1312 1313
1313 1314 kmem_free(tr_list, sizeof (list_t));
1314 1315 }
1315 1316
1316 1317 /*
1317 1318 * This should be called from open context when we think we're going to write
1318 1319 * or free space, for example when dirtying data. Be conservative; it's okay
1319 1320 * to write less space or free more, but we don't want to write more or free
1320 1321 * less than the amount specified.
1321 1322 */
1322 1323 void
1323 1324 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1324 1325 {
1325 1326 int64_t parent_space;
1326 1327 uint64_t est_used;
1327 1328
1328 1329 mutex_enter(&dd->dd_lock);
1329 1330 if (space > 0)
1330 1331 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1331 1332
1332 1333 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1333 1334 parent_space = parent_delta(dd, est_used, space);
1334 1335 mutex_exit(&dd->dd_lock);
1335 1336
1336 1337 /* Make sure that we clean up dd_space_to* */
1337 1338 dsl_dir_dirty(dd, tx);
1338 1339
1339 1340 /* XXX this is potentially expensive and unnecessary... */
1340 1341 if (parent_space && dd->dd_parent)
1341 1342 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1342 1343 }
1343 1344
1344 1345 /* call from syncing context when we actually write/free space for this dd */
1345 1346 void
1346 1347 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1347 1348 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1348 1349 {
1349 1350 int64_t accounted_delta;
1350 1351
1351 1352 /*
1352 1353 * dsl_dataset_set_refreservation_sync_impl() calls this with
1353 1354 * dd_lock held, so that it can atomically update
1354 1355 * ds->ds_reserved and the dsl_dir accounting, so that
1355 1356 * dsl_dataset_check_quota() can see dataset and dir accounting
1356 1357 * consistently.
1357 1358 */
1358 1359 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1359 1360
1360 1361 ASSERT(dmu_tx_is_syncing(tx));
1361 1362 ASSERT(type < DD_USED_NUM);
1362 1363
1363 1364 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1364 1365
1365 1366 if (needlock)
1366 1367 mutex_enter(&dd->dd_lock);
1367 1368 accounted_delta =
1368 1369 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1369 1370 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1370 1371 ASSERT(compressed >= 0 ||
1371 1372 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1372 1373 ASSERT(uncompressed >= 0 ||
1373 1374 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1374 1375 dsl_dir_phys(dd)->dd_used_bytes += used;
1375 1376 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1376 1377 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1377 1378
1378 1379 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1379 1380 ASSERT(used > 0 ||
1380 1381 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1381 1382 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1382 1383 #ifdef DEBUG
1383 1384 dd_used_t t;
1384 1385 uint64_t u = 0;
1385 1386 for (t = 0; t < DD_USED_NUM; t++)
1386 1387 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1387 1388 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1388 1389 #endif
1389 1390 }
1390 1391 if (needlock)
1391 1392 mutex_exit(&dd->dd_lock);
1392 1393
1393 1394 if (dd->dd_parent != NULL) {
1394 1395 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1395 1396 accounted_delta, compressed, uncompressed, tx);
1396 1397 dsl_dir_transfer_space(dd->dd_parent,
1397 1398 used - accounted_delta,
1398 1399 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1399 1400 }
1400 1401 }
1401 1402
1402 1403 void
1403 1404 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1404 1405 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1405 1406 {
1406 1407 ASSERT(dmu_tx_is_syncing(tx));
1407 1408 ASSERT(oldtype < DD_USED_NUM);
1408 1409 ASSERT(newtype < DD_USED_NUM);
1409 1410
1410 1411 if (delta == 0 ||
1411 1412 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1412 1413 return;
1413 1414
1414 1415 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1415 1416 mutex_enter(&dd->dd_lock);
1416 1417 ASSERT(delta > 0 ?
1417 1418 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1418 1419 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1419 1420 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1420 1421 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1421 1422 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1422 1423 mutex_exit(&dd->dd_lock);
1423 1424 }
1424 1425
1425 1426 typedef struct dsl_dir_set_qr_arg {
1426 1427 const char *ddsqra_name;
1427 1428 zprop_source_t ddsqra_source;
1428 1429 uint64_t ddsqra_value;
1429 1430 } dsl_dir_set_qr_arg_t;
1430 1431
1431 1432 static int
1432 1433 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1433 1434 {
1434 1435 dsl_dir_set_qr_arg_t *ddsqra = arg;
1435 1436 dsl_pool_t *dp = dmu_tx_pool(tx);
1436 1437 dsl_dataset_t *ds;
1437 1438 int error;
1438 1439 uint64_t towrite, newval;
1439 1440
1440 1441 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1441 1442 if (error != 0)
1442 1443 return (error);
1443 1444
1444 1445 error = dsl_prop_predict(ds->ds_dir, "quota",
1445 1446 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1446 1447 if (error != 0) {
1447 1448 dsl_dataset_rele(ds, FTAG);
1448 1449 return (error);
1449 1450 }
1450 1451
1451 1452 if (newval == 0) {
1452 1453 dsl_dataset_rele(ds, FTAG);
1453 1454 return (0);
1454 1455 }
1455 1456
1456 1457 mutex_enter(&ds->ds_dir->dd_lock);
1457 1458 /*
1458 1459 * If we are doing the preliminary check in open context, and
1459 1460 * there are pending changes, then don't fail it, since the
1460 1461 * pending changes could under-estimate the amount of space to be
1461 1462 * freed up.
1462 1463 */
1463 1464 towrite = dsl_dir_space_towrite(ds->ds_dir);
1464 1465 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1465 1466 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1466 1467 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1467 1468 error = SET_ERROR(ENOSPC);
1468 1469 }
1469 1470 mutex_exit(&ds->ds_dir->dd_lock);
1470 1471 dsl_dataset_rele(ds, FTAG);
1471 1472 return (error);
1472 1473 }
1473 1474
1474 1475 static void
1475 1476 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1476 1477 {
1477 1478 dsl_dir_set_qr_arg_t *ddsqra = arg;
1478 1479 dsl_pool_t *dp = dmu_tx_pool(tx);
1479 1480 dsl_dataset_t *ds;
1480 1481 uint64_t newval;
1481 1482
1482 1483 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1483 1484
1484 1485 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1485 1486 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1486 1487 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1487 1488 &ddsqra->ddsqra_value, tx);
1488 1489
1489 1490 VERIFY0(dsl_prop_get_int_ds(ds,
1490 1491 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1491 1492 } else {
1492 1493 newval = ddsqra->ddsqra_value;
1493 1494 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1494 1495 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1495 1496 }
1496 1497
1497 1498 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1498 1499 mutex_enter(&ds->ds_dir->dd_lock);
1499 1500 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1500 1501 mutex_exit(&ds->ds_dir->dd_lock);
1501 1502 dsl_dataset_rele(ds, FTAG);
1502 1503 }
1503 1504
1504 1505 int
1505 1506 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1506 1507 {
1507 1508 dsl_dir_set_qr_arg_t ddsqra;
1508 1509
1509 1510 ddsqra.ddsqra_name = ddname;
1510 1511 ddsqra.ddsqra_source = source;
1511 1512 ddsqra.ddsqra_value = quota;
1512 1513
1513 1514 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1514 1515 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1515 1516 }
1516 1517
1517 1518 int
1518 1519 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1519 1520 {
1520 1521 dsl_dir_set_qr_arg_t *ddsqra = arg;
1521 1522 dsl_pool_t *dp = dmu_tx_pool(tx);
1522 1523 dsl_dataset_t *ds;
1523 1524 dsl_dir_t *dd;
1524 1525 uint64_t newval, used, avail;
1525 1526 int error;
1526 1527
1527 1528 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1528 1529 if (error != 0)
1529 1530 return (error);
1530 1531 dd = ds->ds_dir;
1531 1532
1532 1533 /*
1533 1534 * If we are doing the preliminary check in open context, the
1534 1535 * space estimates may be inaccurate.
1535 1536 */
1536 1537 if (!dmu_tx_is_syncing(tx)) {
1537 1538 dsl_dataset_rele(ds, FTAG);
1538 1539 return (0);
1539 1540 }
1540 1541
1541 1542 error = dsl_prop_predict(ds->ds_dir,
1542 1543 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1543 1544 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1544 1545 if (error != 0) {
1545 1546 dsl_dataset_rele(ds, FTAG);
1546 1547 return (error);
1547 1548 }
1548 1549
1549 1550 mutex_enter(&dd->dd_lock);
1550 1551 used = dsl_dir_phys(dd)->dd_used_bytes;
1551 1552 mutex_exit(&dd->dd_lock);
1552 1553
1553 1554 if (dd->dd_parent) {
1554 1555 avail = dsl_dir_space_available(dd->dd_parent,
1555 1556 NULL, 0, FALSE);
1556 1557 } else {
1557 1558 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1558 1559 }
1559 1560
1560 1561 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1561 1562 uint64_t delta = MAX(used, newval) -
1562 1563 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1563 1564
1564 1565 if (delta > avail ||
1565 1566 (dsl_dir_phys(dd)->dd_quota > 0 &&
1566 1567 newval > dsl_dir_phys(dd)->dd_quota))
1567 1568 error = SET_ERROR(ENOSPC);
1568 1569 }
1569 1570
1570 1571 dsl_dataset_rele(ds, FTAG);
1571 1572 return (error);
1572 1573 }
1573 1574
1574 1575 void
1575 1576 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1576 1577 {
1577 1578 uint64_t used;
1578 1579 int64_t delta;
1579 1580
1580 1581 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1581 1582
1582 1583 mutex_enter(&dd->dd_lock);
1583 1584 used = dsl_dir_phys(dd)->dd_used_bytes;
1584 1585 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1585 1586 dsl_dir_phys(dd)->dd_reserved = value;
1586 1587
1587 1588 if (dd->dd_parent != NULL) {
1588 1589 /* Roll up this additional usage into our ancestors */
1589 1590 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1590 1591 delta, 0, 0, tx);
1591 1592 }
1592 1593 mutex_exit(&dd->dd_lock);
1593 1594 }
1594 1595
1595 1596
1596 1597 static void
1597 1598 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1598 1599 {
1599 1600 dsl_dir_set_qr_arg_t *ddsqra = arg;
1600 1601 dsl_pool_t *dp = dmu_tx_pool(tx);
1601 1602 dsl_dataset_t *ds;
1602 1603 uint64_t newval;
1603 1604
1604 1605 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1605 1606
1606 1607 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1607 1608 dsl_prop_set_sync_impl(ds,
1608 1609 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1609 1610 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1610 1611 &ddsqra->ddsqra_value, tx);
1611 1612
1612 1613 VERIFY0(dsl_prop_get_int_ds(ds,
1613 1614 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1614 1615 } else {
1615 1616 newval = ddsqra->ddsqra_value;
1616 1617 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1617 1618 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1618 1619 (longlong_t)newval);
1619 1620 }
1620 1621
1621 1622 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1622 1623 dsl_dataset_rele(ds, FTAG);
1623 1624 }
1624 1625
1625 1626 int
1626 1627 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1627 1628 uint64_t reservation)
1628 1629 {
1629 1630 dsl_dir_set_qr_arg_t ddsqra;
1630 1631
1631 1632 ddsqra.ddsqra_name = ddname;
1632 1633 ddsqra.ddsqra_source = source;
1633 1634 ddsqra.ddsqra_value = reservation;
1634 1635
1635 1636 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1636 1637 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1637 1638 }
1638 1639
1639 1640 static dsl_dir_t *
1640 1641 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1641 1642 {
1642 1643 for (; ds1; ds1 = ds1->dd_parent) {
1643 1644 dsl_dir_t *dd;
1644 1645 for (dd = ds2; dd; dd = dd->dd_parent) {
1645 1646 if (ds1 == dd)
1646 1647 return (dd);
1647 1648 }
1648 1649 }
1649 1650 return (NULL);
1650 1651 }
1651 1652
1652 1653 /*
1653 1654 * If delta is applied to dd, how much of that delta would be applied to
1654 1655 * ancestor? Syncing context only.
1655 1656 */
1656 1657 static int64_t
1657 1658 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1658 1659 {
1659 1660 if (dd == ancestor)
1660 1661 return (delta);
1661 1662
1662 1663 mutex_enter(&dd->dd_lock);
1663 1664 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1664 1665 mutex_exit(&dd->dd_lock);
1665 1666 return (would_change(dd->dd_parent, delta, ancestor));
1666 1667 }
1667 1668
1668 1669 typedef struct dsl_dir_rename_arg {
1669 1670 const char *ddra_oldname;
1670 1671 const char *ddra_newname;
1671 1672 cred_t *ddra_cred;
1672 1673 } dsl_dir_rename_arg_t;
1673 1674
1674 1675 /* ARGSUSED */
1675 1676 static int
1676 1677 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1677 1678 {
1678 1679 int *deltap = arg;
1679 1680 char namebuf[MAXNAMELEN];
1680 1681
1681 1682 dsl_dataset_name(ds, namebuf);
1682 1683
1683 1684 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1684 1685 return (SET_ERROR(ENAMETOOLONG));
1685 1686 return (0);
1686 1687 }
1687 1688
1688 1689 static int
1689 1690 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1690 1691 {
1691 1692 dsl_dir_rename_arg_t *ddra = arg;
1692 1693 dsl_pool_t *dp = dmu_tx_pool(tx);
1693 1694 dsl_dir_t *dd, *newparent;
1694 1695 const char *mynewname;
1695 1696 int error;
1696 1697 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1697 1698
1698 1699 /* target dir should exist */
1699 1700 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1700 1701 if (error != 0)
1701 1702 return (error);
1702 1703
1703 1704 /* new parent should exist */
1704 1705 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1705 1706 &newparent, &mynewname);
1706 1707 if (error != 0) {
1707 1708 dsl_dir_rele(dd, FTAG);
1708 1709 return (error);
1709 1710 }
1710 1711
1711 1712 /* can't rename to different pool */
1712 1713 if (dd->dd_pool != newparent->dd_pool) {
1713 1714 dsl_dir_rele(newparent, FTAG);
1714 1715 dsl_dir_rele(dd, FTAG);
1715 1716 return (SET_ERROR(ENXIO));
1716 1717 }
1717 1718
1718 1719 /* new name should not already exist */
1719 1720 if (mynewname == NULL) {
1720 1721 dsl_dir_rele(newparent, FTAG);
1721 1722 dsl_dir_rele(dd, FTAG);
1722 1723 return (SET_ERROR(EEXIST));
1723 1724 }
1724 1725
1725 1726 /* if the name length is growing, validate child name lengths */
1726 1727 if (delta > 0) {
1727 1728 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1728 1729 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1729 1730 if (error != 0) {
1730 1731 dsl_dir_rele(newparent, FTAG);
1731 1732 dsl_dir_rele(dd, FTAG);
1732 1733 return (error);
1733 1734 }
1734 1735 }
1735 1736
1736 1737 if (dmu_tx_is_syncing(tx)) {
1737 1738 if (spa_feature_is_active(dp->dp_spa,
1738 1739 SPA_FEATURE_FS_SS_LIMIT)) {
1739 1740 /*
1740 1741 * Although this is the check function and we don't
1741 1742 * normally make on-disk changes in check functions,
1742 1743 * we need to do that here.
1743 1744 *
1744 1745 * Ensure this portion of the tree's counts have been
1745 1746 * initialized in case the new parent has limits set.
1746 1747 */
1747 1748 dsl_dir_init_fs_ss_count(dd, tx);
1748 1749 }
1749 1750 }
1750 1751
1751 1752 if (newparent != dd->dd_parent) {
1752 1753 /* is there enough space? */
1753 1754 uint64_t myspace =
1754 1755 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1755 1756 dsl_dir_phys(dd)->dd_reserved);
1756 1757 objset_t *os = dd->dd_pool->dp_meta_objset;
1757 1758 uint64_t fs_cnt = 0;
1758 1759 uint64_t ss_cnt = 0;
1759 1760
1760 1761 if (dsl_dir_is_zapified(dd)) {
1761 1762 int err;
1762 1763
1763 1764 err = zap_lookup(os, dd->dd_object,
1764 1765 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1765 1766 &fs_cnt);
1766 1767 if (err != ENOENT && err != 0) {
1767 1768 dsl_dir_rele(newparent, FTAG);
1768 1769 dsl_dir_rele(dd, FTAG);
1769 1770 return (err);
1770 1771 }
1771 1772
1772 1773 /*
1773 1774 * have to add 1 for the filesystem itself that we're
1774 1775 * moving
1775 1776 */
1776 1777 fs_cnt++;
1777 1778
1778 1779 err = zap_lookup(os, dd->dd_object,
1779 1780 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1780 1781 &ss_cnt);
1781 1782 if (err != ENOENT && err != 0) {
1782 1783 dsl_dir_rele(newparent, FTAG);
1783 1784 dsl_dir_rele(dd, FTAG);
1784 1785 return (err);
1785 1786 }
1786 1787 }
1787 1788
1788 1789 /* no rename into our descendant */
1789 1790 if (closest_common_ancestor(dd, newparent) == dd) {
1790 1791 dsl_dir_rele(newparent, FTAG);
1791 1792 dsl_dir_rele(dd, FTAG);
1792 1793 return (SET_ERROR(EINVAL));
1793 1794 }
1794 1795
1795 1796 error = dsl_dir_transfer_possible(dd->dd_parent,
1796 1797 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1797 1798 if (error != 0) {
1798 1799 dsl_dir_rele(newparent, FTAG);
1799 1800 dsl_dir_rele(dd, FTAG);
1800 1801 return (error);
1801 1802 }
1802 1803 }
1803 1804
1804 1805 dsl_dir_rele(newparent, FTAG);
1805 1806 dsl_dir_rele(dd, FTAG);
1806 1807 return (0);
1807 1808 }
1808 1809
1809 1810 static void
1810 1811 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1811 1812 {
1812 1813 dsl_dir_rename_arg_t *ddra = arg;
1813 1814 dsl_pool_t *dp = dmu_tx_pool(tx);
1814 1815 dsl_dir_t *dd, *newparent;
1815 1816 const char *mynewname;
1816 1817 int error;
1817 1818 objset_t *mos = dp->dp_meta_objset;
1818 1819
1819 1820 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1820 1821 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1821 1822 &mynewname));
1822 1823
1823 1824 /* Log this before we change the name. */
1824 1825 spa_history_log_internal_dd(dd, "rename", tx,
1825 1826 "-> %s", ddra->ddra_newname);
1826 1827
1827 1828 if (newparent != dd->dd_parent) {
1828 1829 objset_t *os = dd->dd_pool->dp_meta_objset;
1829 1830 uint64_t fs_cnt = 0;
1830 1831 uint64_t ss_cnt = 0;
1831 1832
1832 1833 /*
1833 1834 * We already made sure the dd counts were initialized in the
1834 1835 * check function.
1835 1836 */
1836 1837 if (spa_feature_is_active(dp->dp_spa,
1837 1838 SPA_FEATURE_FS_SS_LIMIT)) {
1838 1839 VERIFY0(zap_lookup(os, dd->dd_object,
1839 1840 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1840 1841 &fs_cnt));
1841 1842 /* add 1 for the filesystem itself that we're moving */
1842 1843 fs_cnt++;
1843 1844
1844 1845 VERIFY0(zap_lookup(os, dd->dd_object,
1845 1846 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1846 1847 &ss_cnt));
1847 1848 }
1848 1849
1849 1850 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1850 1851 DD_FIELD_FILESYSTEM_COUNT, tx);
1851 1852 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1852 1853 DD_FIELD_FILESYSTEM_COUNT, tx);
1853 1854
1854 1855 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1855 1856 DD_FIELD_SNAPSHOT_COUNT, tx);
1856 1857 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1857 1858 DD_FIELD_SNAPSHOT_COUNT, tx);
1858 1859
1859 1860 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1860 1861 -dsl_dir_phys(dd)->dd_used_bytes,
1861 1862 -dsl_dir_phys(dd)->dd_compressed_bytes,
1862 1863 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1863 1864 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1864 1865 dsl_dir_phys(dd)->dd_used_bytes,
1865 1866 dsl_dir_phys(dd)->dd_compressed_bytes,
1866 1867 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1867 1868
1868 1869 if (dsl_dir_phys(dd)->dd_reserved >
1869 1870 dsl_dir_phys(dd)->dd_used_bytes) {
1870 1871 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
1871 1872 dsl_dir_phys(dd)->dd_used_bytes;
1872 1873
1873 1874 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1874 1875 -unused_rsrv, 0, 0, tx);
1875 1876 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1876 1877 unused_rsrv, 0, 0, tx);
1877 1878 }
1878 1879 }
1879 1880
1880 1881 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1881 1882
1882 1883 /* remove from old parent zapobj */
1883 1884 error = zap_remove(mos,
1884 1885 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
1885 1886 dd->dd_myname, tx);
1886 1887 ASSERT0(error);
1887 1888
1888 1889 (void) strcpy(dd->dd_myname, mynewname);
1889 1890 dsl_dir_rele(dd->dd_parent, dd);
1890 1891 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
1891 1892 VERIFY0(dsl_dir_hold_obj(dp,
1892 1893 newparent->dd_object, NULL, dd, &dd->dd_parent));
1893 1894
1894 1895 /* add to new parent zapobj */
1895 1896 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
1896 1897 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1897 1898
1898 1899 dsl_prop_notify_all(dd);
1899 1900
1900 1901 dsl_dir_rele(newparent, FTAG);
1901 1902 dsl_dir_rele(dd, FTAG);
1902 1903 }
1903 1904
1904 1905 int
1905 1906 dsl_dir_rename(const char *oldname, const char *newname)
1906 1907 {
1907 1908 dsl_dir_rename_arg_t ddra;
1908 1909
1909 1910 ddra.ddra_oldname = oldname;
1910 1911 ddra.ddra_newname = newname;
1911 1912 ddra.ddra_cred = CRED();
1912 1913
1913 1914 return (dsl_sync_task(oldname,
1914 1915 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1915 1916 3, ZFS_SPACE_CHECK_RESERVED));
1916 1917 }
1917 1918
1918 1919 int
1919 1920 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1920 1921 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1921 1922 {
1922 1923 dsl_dir_t *ancestor;
1923 1924 int64_t adelta;
1924 1925 uint64_t avail;
1925 1926 int err;
1926 1927
1927 1928 ancestor = closest_common_ancestor(sdd, tdd);
1928 1929 adelta = would_change(sdd, -space, ancestor);
1929 1930 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1930 1931 if (avail < space)
1931 1932 return (SET_ERROR(ENOSPC));
1932 1933
1933 1934 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1934 1935 ancestor, cr);
1935 1936 if (err != 0)
1936 1937 return (err);
1937 1938 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1938 1939 ancestor, cr);
1939 1940 if (err != 0)
1940 1941 return (err);
1941 1942
1942 1943 return (0);
1943 1944 }
1944 1945
1945 1946 timestruc_t
1946 1947 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1947 1948 {
1948 1949 timestruc_t t;
1949 1950
1950 1951 mutex_enter(&dd->dd_lock);
1951 1952 t = dd->dd_snap_cmtime;
1952 1953 mutex_exit(&dd->dd_lock);
1953 1954
1954 1955 return (t);
1955 1956 }
1956 1957
1957 1958 void
1958 1959 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1959 1960 {
1960 1961 timestruc_t t;
1961 1962
1962 1963 gethrestime(&t);
1963 1964 mutex_enter(&dd->dd_lock);
1964 1965 dd->dd_snap_cmtime = t;
1965 1966 mutex_exit(&dd->dd_lock);
1966 1967 }
1967 1968
1968 1969 void
1969 1970 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1970 1971 {
1971 1972 objset_t *mos = dd->dd_pool->dp_meta_objset;
1972 1973 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1973 1974 }
1974 1975
1975 1976 boolean_t
1976 1977 dsl_dir_is_zapified(dsl_dir_t *dd)
1977 1978 {
1978 1979 dmu_object_info_t doi;
1979 1980
1980 1981 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1981 1982 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1982 1983 }
↓ open down ↓ |
1734 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX