Print this page
5610 zfs clone from different source and target pools produces coredump
| 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
|
↓ open down ↓ |
16 lines elided |
↑ open up ↑ |
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 + * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
27 28 */
28 29
29 30 #include <sys/dmu.h>
30 31 #include <sys/dmu_objset.h>
31 32 #include <sys/dmu_tx.h>
32 33 #include <sys/dsl_dataset.h>
33 34 #include <sys/dsl_dir.h>
34 35 #include <sys/dsl_prop.h>
35 36 #include <sys/dsl_synctask.h>
36 37 #include <sys/dsl_deleg.h>
37 38 #include <sys/dmu_impl.h>
38 39 #include <sys/spa.h>
39 40 #include <sys/metaslab.h>
40 41 #include <sys/zap.h>
41 42 #include <sys/zio.h>
42 43 #include <sys/arc.h>
43 44 #include <sys/sunddi.h>
44 45 #include <sys/zfeature.h>
45 46 #include <sys/policy.h>
46 47 #include <sys/zfs_znode.h>
47 48 #include "zfs_namecheck.h"
48 49 #include "zfs_prop.h"
49 50
50 51 /*
51 52 * Filesystem and Snapshot Limits
52 53 * ------------------------------
53 54 *
54 55 * These limits are used to restrict the number of filesystems and/or snapshots
55 56 * that can be created at a given level in the tree or below. A typical
56 57 * use-case is with a delegated dataset where the administrator wants to ensure
57 58 * that a user within the zone is not creating too many additional filesystems
58 59 * or snapshots, even though they're not exceeding their space quota.
59 60 *
60 61 * The filesystem and snapshot counts are stored as extensible properties. This
61 62 * capability is controlled by a feature flag and must be enabled to be used.
62 63 * Once enabled, the feature is not active until the first limit is set. At
63 64 * that point, future operations to create/destroy filesystems or snapshots
64 65 * will validate and update the counts.
65 66 *
66 67 * Because the count properties will not exist before the feature is active,
67 68 * the counts are updated when a limit is first set on an uninitialized
68 69 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
69 70 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
70 71 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
71 72 * snapshot count properties on a node indicate uninitialized counts on that
72 73 * node.) When first setting a limit on an uninitialized node, the code starts
73 74 * at the filesystem with the new limit and descends into all sub-filesystems
74 75 * to add the count properties.
75 76 *
76 77 * In practice this is lightweight since a limit is typically set when the
77 78 * filesystem is created and thus has no children. Once valid, changing the
78 79 * limit value won't require a re-traversal since the counts are already valid.
79 80 * When recursively fixing the counts, if a node with a limit is encountered
80 81 * during the descent, the counts are known to be valid and there is no need to
81 82 * descend into that filesystem's children. The counts on filesystems above the
82 83 * one with the new limit will still be uninitialized, unless a limit is
83 84 * eventually set on one of those filesystems. The counts are always recursively
84 85 * updated when a limit is set on a dataset, unless there is already a limit.
85 86 * When a new limit value is set on a filesystem with an existing limit, it is
86 87 * possible for the new limit to be less than the current count at that level
87 88 * since a user who can change the limit is also allowed to exceed the limit.
88 89 *
89 90 * Once the feature is active, then whenever a filesystem or snapshot is
90 91 * created, the code recurses up the tree, validating the new count against the
91 92 * limit at each initialized level. In practice, most levels will not have a
92 93 * limit set. If there is a limit at any initialized level up the tree, the
93 94 * check must pass or the creation will fail. Likewise, when a filesystem or
94 95 * snapshot is destroyed, the counts are recursively adjusted all the way up
95 96 * the initizized nodes in the tree. Renaming a filesystem into different point
96 97 * in the tree will first validate, then update the counts on each branch up to
97 98 * the common ancestor. A receive will also validate the counts and then update
98 99 * them.
99 100 *
100 101 * An exception to the above behavior is that the limit is not enforced if the
101 102 * user has permission to modify the limit. This is primarily so that
102 103 * recursive snapshots in the global zone always work. We want to prevent a
103 104 * denial-of-service in which a lower level delegated dataset could max out its
104 105 * limit and thus block recursive snapshots from being taken in the global zone.
105 106 * Because of this, it is possible for the snapshot count to be over the limit
106 107 * and snapshots taken in the global zone could cause a lower level dataset to
107 108 * hit or exceed its limit. The administrator taking the global zone recursive
108 109 * snapshot should be aware of this side-effect and behave accordingly.
109 110 * For consistency, the filesystem limit is also not enforced if the user can
110 111 * modify the limit.
111 112 *
112 113 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
113 114 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
114 115 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
115 116 * dsl_dir_init_fs_ss_count().
116 117 *
117 118 * There is a special case when we receive a filesystem that already exists. In
118 119 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
119 120 * never update the filesystem counts for temporary clones.
120 121 *
121 122 * Likewise, we do not update the snapshot counts for temporary snapshots,
122 123 * such as those created by zfs diff.
123 124 */
124 125
125 126 extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
126 127
127 128 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
128 129
129 130 static void
130 131 dsl_dir_evict(void *dbu)
131 132 {
132 133 dsl_dir_t *dd = dbu;
133 134 dsl_pool_t *dp = dd->dd_pool;
134 135 int t;
135 136
136 137 dd->dd_dbuf = NULL;
137 138
138 139 for (t = 0; t < TXG_SIZE; t++) {
139 140 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
140 141 ASSERT(dd->dd_tempreserved[t] == 0);
141 142 ASSERT(dd->dd_space_towrite[t] == 0);
142 143 }
143 144
144 145 if (dd->dd_parent)
145 146 dsl_dir_async_rele(dd->dd_parent, dd);
146 147
147 148 spa_async_close(dd->dd_pool->dp_spa, dd);
148 149
149 150 /*
150 151 * The props callback list should have been cleaned up by
151 152 * objset_evict().
152 153 */
153 154 list_destroy(&dd->dd_prop_cbs);
154 155 mutex_destroy(&dd->dd_lock);
155 156 kmem_free(dd, sizeof (dsl_dir_t));
156 157 }
157 158
158 159 int
159 160 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
160 161 const char *tail, void *tag, dsl_dir_t **ddp)
161 162 {
162 163 dmu_buf_t *dbuf;
163 164 dsl_dir_t *dd;
164 165 int err;
165 166
166 167 ASSERT(dsl_pool_config_held(dp));
167 168
168 169 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
169 170 if (err != 0)
170 171 return (err);
171 172 dd = dmu_buf_get_user(dbuf);
172 173 #ifdef ZFS_DEBUG
173 174 {
174 175 dmu_object_info_t doi;
175 176 dmu_object_info_from_db(dbuf, &doi);
176 177 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
177 178 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
178 179 }
179 180 #endif
180 181 if (dd == NULL) {
181 182 dsl_dir_t *winner;
182 183
183 184 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
184 185 dd->dd_object = ddobj;
185 186 dd->dd_dbuf = dbuf;
186 187 dd->dd_pool = dp;
187 188 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
188 189
189 190 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
190 191 offsetof(dsl_prop_cb_record_t, cbr_node));
191 192
192 193 dsl_dir_snap_cmtime_update(dd);
193 194
194 195 if (dsl_dir_phys(dd)->dd_parent_obj) {
195 196 err = dsl_dir_hold_obj(dp,
196 197 dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
197 198 &dd->dd_parent);
198 199 if (err != 0)
199 200 goto errout;
200 201 if (tail) {
201 202 #ifdef ZFS_DEBUG
202 203 uint64_t foundobj;
203 204
204 205 err = zap_lookup(dp->dp_meta_objset,
205 206 dsl_dir_phys(dd->dd_parent)->
206 207 dd_child_dir_zapobj, tail,
207 208 sizeof (foundobj), 1, &foundobj);
208 209 ASSERT(err || foundobj == ddobj);
209 210 #endif
210 211 (void) strcpy(dd->dd_myname, tail);
211 212 } else {
212 213 err = zap_value_search(dp->dp_meta_objset,
213 214 dsl_dir_phys(dd->dd_parent)->
214 215 dd_child_dir_zapobj,
215 216 ddobj, 0, dd->dd_myname);
216 217 }
217 218 if (err != 0)
218 219 goto errout;
219 220 } else {
220 221 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
221 222 }
222 223
223 224 if (dsl_dir_is_clone(dd)) {
224 225 dmu_buf_t *origin_bonus;
225 226 dsl_dataset_phys_t *origin_phys;
226 227
227 228 /*
228 229 * We can't open the origin dataset, because
229 230 * that would require opening this dsl_dir.
230 231 * Just look at its phys directly instead.
231 232 */
232 233 err = dmu_bonus_hold(dp->dp_meta_objset,
233 234 dsl_dir_phys(dd)->dd_origin_obj, FTAG,
234 235 &origin_bonus);
235 236 if (err != 0)
236 237 goto errout;
237 238 origin_phys = origin_bonus->db_data;
238 239 dd->dd_origin_txg =
239 240 origin_phys->ds_creation_txg;
240 241 dmu_buf_rele(origin_bonus, FTAG);
241 242 }
242 243
243 244 dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf);
244 245 winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
245 246 if (winner != NULL) {
246 247 if (dd->dd_parent)
247 248 dsl_dir_rele(dd->dd_parent, dd);
248 249 mutex_destroy(&dd->dd_lock);
249 250 kmem_free(dd, sizeof (dsl_dir_t));
250 251 dd = winner;
251 252 } else {
252 253 spa_open_ref(dp->dp_spa, dd);
253 254 }
254 255 }
255 256
256 257 /*
257 258 * The dsl_dir_t has both open-to-close and instantiate-to-evict
258 259 * holds on the spa. We need the open-to-close holds because
259 260 * otherwise the spa_refcnt wouldn't change when we open a
260 261 * dir which the spa also has open, so we could incorrectly
261 262 * think it was OK to unload/export/destroy the pool. We need
262 263 * the instantiate-to-evict hold because the dsl_dir_t has a
263 264 * pointer to the dd_pool, which has a pointer to the spa_t.
264 265 */
265 266 spa_open_ref(dp->dp_spa, tag);
266 267 ASSERT3P(dd->dd_pool, ==, dp);
267 268 ASSERT3U(dd->dd_object, ==, ddobj);
268 269 ASSERT3P(dd->dd_dbuf, ==, dbuf);
269 270 *ddp = dd;
270 271 return (0);
271 272
272 273 errout:
273 274 if (dd->dd_parent)
274 275 dsl_dir_rele(dd->dd_parent, dd);
275 276 mutex_destroy(&dd->dd_lock);
276 277 kmem_free(dd, sizeof (dsl_dir_t));
277 278 dmu_buf_rele(dbuf, tag);
278 279 return (err);
279 280 }
280 281
281 282 void
282 283 dsl_dir_rele(dsl_dir_t *dd, void *tag)
283 284 {
284 285 dprintf_dd(dd, "%s\n", "");
285 286 spa_close(dd->dd_pool->dp_spa, tag);
286 287 dmu_buf_rele(dd->dd_dbuf, tag);
287 288 }
288 289
289 290 /*
290 291 * Remove a reference to the given dsl dir that is being asynchronously
291 292 * released. Async releases occur from a taskq performing eviction of
292 293 * dsl datasets and dirs. This process is identical to a normal release
293 294 * with the exception of using the async API for releasing the reference on
294 295 * the spa.
295 296 */
296 297 void
297 298 dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
298 299 {
299 300 dprintf_dd(dd, "%s\n", "");
300 301 spa_async_close(dd->dd_pool->dp_spa, tag);
301 302 dmu_buf_rele(dd->dd_dbuf, tag);
302 303 }
303 304
304 305 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
305 306 void
306 307 dsl_dir_name(dsl_dir_t *dd, char *buf)
307 308 {
308 309 if (dd->dd_parent) {
309 310 dsl_dir_name(dd->dd_parent, buf);
310 311 (void) strcat(buf, "/");
311 312 } else {
312 313 buf[0] = '\0';
313 314 }
314 315 if (!MUTEX_HELD(&dd->dd_lock)) {
315 316 /*
316 317 * recursive mutex so that we can use
317 318 * dprintf_dd() with dd_lock held
318 319 */
319 320 mutex_enter(&dd->dd_lock);
320 321 (void) strcat(buf, dd->dd_myname);
321 322 mutex_exit(&dd->dd_lock);
322 323 } else {
323 324 (void) strcat(buf, dd->dd_myname);
324 325 }
325 326 }
326 327
327 328 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
328 329 int
329 330 dsl_dir_namelen(dsl_dir_t *dd)
330 331 {
331 332 int result = 0;
332 333
333 334 if (dd->dd_parent) {
334 335 /* parent's name + 1 for the "/" */
335 336 result = dsl_dir_namelen(dd->dd_parent) + 1;
336 337 }
337 338
338 339 if (!MUTEX_HELD(&dd->dd_lock)) {
339 340 /* see dsl_dir_name */
340 341 mutex_enter(&dd->dd_lock);
341 342 result += strlen(dd->dd_myname);
342 343 mutex_exit(&dd->dd_lock);
343 344 } else {
344 345 result += strlen(dd->dd_myname);
345 346 }
346 347
347 348 return (result);
348 349 }
349 350
350 351 static int
351 352 getcomponent(const char *path, char *component, const char **nextp)
352 353 {
353 354 char *p;
354 355
355 356 if ((path == NULL) || (path[0] == '\0'))
356 357 return (SET_ERROR(ENOENT));
357 358 /* This would be a good place to reserve some namespace... */
358 359 p = strpbrk(path, "/@");
359 360 if (p && (p[1] == '/' || p[1] == '@')) {
360 361 /* two separators in a row */
361 362 return (SET_ERROR(EINVAL));
362 363 }
363 364 if (p == NULL || p == path) {
364 365 /*
365 366 * if the first thing is an @ or /, it had better be an
366 367 * @ and it had better not have any more ats or slashes,
367 368 * and it had better have something after the @.
368 369 */
369 370 if (p != NULL &&
370 371 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
371 372 return (SET_ERROR(EINVAL));
372 373 if (strlen(path) >= MAXNAMELEN)
373 374 return (SET_ERROR(ENAMETOOLONG));
374 375 (void) strcpy(component, path);
375 376 p = NULL;
376 377 } else if (p[0] == '/') {
377 378 if (p - path >= MAXNAMELEN)
378 379 return (SET_ERROR(ENAMETOOLONG));
379 380 (void) strncpy(component, path, p - path);
380 381 component[p - path] = '\0';
381 382 p++;
382 383 } else if (p[0] == '@') {
383 384 /*
384 385 * if the next separator is an @, there better not be
385 386 * any more slashes.
386 387 */
387 388 if (strchr(path, '/'))
388 389 return (SET_ERROR(EINVAL));
389 390 if (p - path >= MAXNAMELEN)
390 391 return (SET_ERROR(ENAMETOOLONG));
391 392 (void) strncpy(component, path, p - path);
392 393 component[p - path] = '\0';
393 394 } else {
394 395 panic("invalid p=%p", (void *)p);
395 396 }
396 397 *nextp = p;
397 398 return (0);
398 399 }
399 400
400 401 /*
401 402 * Return the dsl_dir_t, and possibly the last component which couldn't
402 403 * be found in *tail. The name must be in the specified dsl_pool_t. This
403 404 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
404 405 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
405 406 * (*tail)[0] == '@' means that the last component is a snapshot.
406 407 */
407 408 int
408 409 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
409 410 dsl_dir_t **ddp, const char **tailp)
410 411 {
411 412 char buf[MAXNAMELEN];
412 413 const char *spaname, *next, *nextnext = NULL;
413 414 int err;
|
↓ open down ↓ |
377 lines elided |
↑ open up ↑ |
414 415 dsl_dir_t *dd;
415 416 uint64_t ddobj;
416 417
417 418 err = getcomponent(name, buf, &next);
418 419 if (err != 0)
419 420 return (err);
420 421
421 422 /* Make sure the name is in the specified pool. */
422 423 spaname = spa_name(dp->dp_spa);
423 424 if (strcmp(buf, spaname) != 0)
424 - return (SET_ERROR(EINVAL));
425 + return (SET_ERROR(EXDEV));
425 426
426 427 ASSERT(dsl_pool_config_held(dp));
427 428
428 429 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
429 430 if (err != 0) {
430 431 return (err);
431 432 }
432 433
433 434 while (next != NULL) {
434 435 dsl_dir_t *child_dd;
435 436 err = getcomponent(next, buf, &nextnext);
436 437 if (err != 0)
437 438 break;
438 439 ASSERT(next[0] != '\0');
439 440 if (next[0] == '@')
440 441 break;
441 442 dprintf("looking up %s in obj%lld\n",
442 443 buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
443 444
444 445 err = zap_lookup(dp->dp_meta_objset,
445 446 dsl_dir_phys(dd)->dd_child_dir_zapobj,
446 447 buf, sizeof (ddobj), 1, &ddobj);
447 448 if (err != 0) {
448 449 if (err == ENOENT)
449 450 err = 0;
450 451 break;
451 452 }
452 453
453 454 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
454 455 if (err != 0)
455 456 break;
456 457 dsl_dir_rele(dd, tag);
457 458 dd = child_dd;
458 459 next = nextnext;
459 460 }
460 461
461 462 if (err != 0) {
462 463 dsl_dir_rele(dd, tag);
463 464 return (err);
464 465 }
465 466
466 467 /*
467 468 * It's an error if there's more than one component left, or
468 469 * tailp==NULL and there's any component left.
469 470 */
470 471 if (next != NULL &&
471 472 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
472 473 /* bad path name */
473 474 dsl_dir_rele(dd, tag);
474 475 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
475 476 err = SET_ERROR(ENOENT);
476 477 }
477 478 if (tailp != NULL)
478 479 *tailp = next;
479 480 *ddp = dd;
480 481 return (err);
481 482 }
482 483
483 484 /*
484 485 * If the counts are already initialized for this filesystem and its
485 486 * descendants then do nothing, otherwise initialize the counts.
486 487 *
487 488 * The counts on this filesystem, and those below, may be uninitialized due to
488 489 * either the use of a pre-existing pool which did not support the
489 490 * filesystem/snapshot limit feature, or one in which the feature had not yet
490 491 * been enabled.
491 492 *
492 493 * Recursively descend the filesystem tree and update the filesystem/snapshot
493 494 * counts on each filesystem below, then update the cumulative count on the
494 495 * current filesystem. If the filesystem already has a count set on it,
495 496 * then we know that its counts, and the counts on the filesystems below it,
496 497 * are already correct, so we don't have to update this filesystem.
497 498 */
498 499 static void
499 500 dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
500 501 {
501 502 uint64_t my_fs_cnt = 0;
502 503 uint64_t my_ss_cnt = 0;
503 504 dsl_pool_t *dp = dd->dd_pool;
504 505 objset_t *os = dp->dp_meta_objset;
505 506 zap_cursor_t *zc;
506 507 zap_attribute_t *za;
507 508 dsl_dataset_t *ds;
508 509
509 510 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
510 511 ASSERT(dsl_pool_config_held(dp));
511 512 ASSERT(dmu_tx_is_syncing(tx));
512 513
513 514 dsl_dir_zapify(dd, tx);
514 515
515 516 /*
516 517 * If the filesystem count has already been initialized then we
517 518 * don't need to recurse down any further.
518 519 */
519 520 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
520 521 return;
521 522
522 523 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
523 524 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
524 525
525 526 /* Iterate my child dirs */
526 527 for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
527 528 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
528 529 dsl_dir_t *chld_dd;
529 530 uint64_t count;
530 531
531 532 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
532 533 &chld_dd));
533 534
534 535 /*
535 536 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
536 537 * temporary datasets.
537 538 */
538 539 if (chld_dd->dd_myname[0] == '$' ||
539 540 chld_dd->dd_myname[0] == '%') {
540 541 dsl_dir_rele(chld_dd, FTAG);
541 542 continue;
542 543 }
543 544
544 545 my_fs_cnt++; /* count this child */
545 546
546 547 dsl_dir_init_fs_ss_count(chld_dd, tx);
547 548
548 549 VERIFY0(zap_lookup(os, chld_dd->dd_object,
549 550 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
550 551 my_fs_cnt += count;
551 552 VERIFY0(zap_lookup(os, chld_dd->dd_object,
552 553 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
553 554 my_ss_cnt += count;
554 555
555 556 dsl_dir_rele(chld_dd, FTAG);
556 557 }
557 558 zap_cursor_fini(zc);
558 559 /* Count my snapshots (we counted children's snapshots above) */
559 560 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
560 561 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
561 562
562 563 for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
563 564 zap_cursor_retrieve(zc, za) == 0;
564 565 zap_cursor_advance(zc)) {
565 566 /* Don't count temporary snapshots */
566 567 if (za->za_name[0] != '%')
567 568 my_ss_cnt++;
568 569 }
569 570 zap_cursor_fini(zc);
570 571
571 572 dsl_dataset_rele(ds, FTAG);
572 573
573 574 kmem_free(zc, sizeof (zap_cursor_t));
574 575 kmem_free(za, sizeof (zap_attribute_t));
575 576
576 577 /* we're in a sync task, update counts */
577 578 dmu_buf_will_dirty(dd->dd_dbuf, tx);
578 579 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
579 580 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
580 581 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
581 582 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
582 583 }
583 584
584 585 static int
585 586 dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
586 587 {
587 588 char *ddname = (char *)arg;
588 589 dsl_pool_t *dp = dmu_tx_pool(tx);
589 590 dsl_dataset_t *ds;
590 591 dsl_dir_t *dd;
591 592 int error;
592 593
593 594 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
594 595 if (error != 0)
595 596 return (error);
596 597
597 598 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
598 599 dsl_dataset_rele(ds, FTAG);
599 600 return (SET_ERROR(ENOTSUP));
600 601 }
601 602
602 603 dd = ds->ds_dir;
603 604 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
604 605 dsl_dir_is_zapified(dd) &&
605 606 zap_contains(dp->dp_meta_objset, dd->dd_object,
606 607 DD_FIELD_FILESYSTEM_COUNT) == 0) {
607 608 dsl_dataset_rele(ds, FTAG);
608 609 return (SET_ERROR(EALREADY));
609 610 }
610 611
611 612 dsl_dataset_rele(ds, FTAG);
612 613 return (0);
613 614 }
614 615
615 616 static void
616 617 dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
617 618 {
618 619 char *ddname = (char *)arg;
619 620 dsl_pool_t *dp = dmu_tx_pool(tx);
620 621 dsl_dataset_t *ds;
621 622 spa_t *spa;
622 623
623 624 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
624 625
625 626 spa = dsl_dataset_get_spa(ds);
626 627
627 628 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
628 629 /*
629 630 * Since the feature was not active and we're now setting a
630 631 * limit, increment the feature-active counter so that the
631 632 * feature becomes active for the first time.
632 633 *
633 634 * We are already in a sync task so we can update the MOS.
634 635 */
635 636 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
636 637 }
637 638
638 639 /*
639 640 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
640 641 * we need to ensure the counts are correct. Descend down the tree from
641 642 * this point and update all of the counts to be accurate.
642 643 */
643 644 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
644 645
645 646 dsl_dataset_rele(ds, FTAG);
646 647 }
647 648
648 649 /*
649 650 * Make sure the feature is enabled and activate it if necessary.
650 651 * Since we're setting a limit, ensure the on-disk counts are valid.
651 652 * This is only called by the ioctl path when setting a limit value.
652 653 *
653 654 * We do not need to validate the new limit, since users who can change the
654 655 * limit are also allowed to exceed the limit.
655 656 */
656 657 int
657 658 dsl_dir_activate_fs_ss_limit(const char *ddname)
658 659 {
659 660 int error;
660 661
661 662 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
662 663 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
663 664 ZFS_SPACE_CHECK_RESERVED);
664 665
665 666 if (error == EALREADY)
666 667 error = 0;
667 668
668 669 return (error);
669 670 }
670 671
671 672 /*
672 673 * Used to determine if the filesystem_limit or snapshot_limit should be
673 674 * enforced. We allow the limit to be exceeded if the user has permission to
674 675 * write the property value. We pass in the creds that we got in the open
675 676 * context since we will always be the GZ root in syncing context. We also have
676 677 * to handle the case where we are allowed to change the limit on the current
677 678 * dataset, but there may be another limit in the tree above.
678 679 *
679 680 * We can never modify these two properties within a non-global zone. In
680 681 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
681 682 * can't use that function since we are already holding the dp_config_rwlock.
682 683 * In addition, we already have the dd and dealing with snapshots is simplified
683 684 * in this code.
684 685 */
685 686
686 687 typedef enum {
687 688 ENFORCE_ALWAYS,
688 689 ENFORCE_NEVER,
689 690 ENFORCE_ABOVE
690 691 } enforce_res_t;
691 692
692 693 static enforce_res_t
693 694 dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
694 695 {
695 696 enforce_res_t enforce = ENFORCE_ALWAYS;
696 697 uint64_t obj;
697 698 dsl_dataset_t *ds;
698 699 uint64_t zoned;
699 700
700 701 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
701 702 prop == ZFS_PROP_SNAPSHOT_LIMIT);
702 703
703 704 #ifdef _KERNEL
704 705 if (crgetzoneid(cr) != GLOBAL_ZONEID)
705 706 return (ENFORCE_ALWAYS);
706 707
707 708 if (secpolicy_zfs(cr) == 0)
708 709 return (ENFORCE_NEVER);
709 710 #endif
710 711
711 712 if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
712 713 return (ENFORCE_ALWAYS);
713 714
714 715 ASSERT(dsl_pool_config_held(dd->dd_pool));
715 716
716 717 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
717 718 return (ENFORCE_ALWAYS);
718 719
719 720 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
720 721 /* Only root can access zoned fs's from the GZ */
721 722 enforce = ENFORCE_ALWAYS;
722 723 } else {
723 724 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
724 725 enforce = ENFORCE_ABOVE;
725 726 }
726 727
727 728 dsl_dataset_rele(ds, FTAG);
728 729 return (enforce);
729 730 }
730 731
731 732 /*
732 733 * Check if adding additional child filesystem(s) would exceed any filesystem
733 734 * limits or adding additional snapshot(s) would exceed any snapshot limits.
734 735 * The prop argument indicates which limit to check.
735 736 *
736 737 * Note that all filesystem limits up to the root (or the highest
737 738 * initialized) filesystem or the given ancestor must be satisfied.
738 739 */
739 740 int
740 741 dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
741 742 dsl_dir_t *ancestor, cred_t *cr)
742 743 {
743 744 objset_t *os = dd->dd_pool->dp_meta_objset;
744 745 uint64_t limit, count;
745 746 char *count_prop;
746 747 enforce_res_t enforce;
747 748 int err = 0;
748 749
749 750 ASSERT(dsl_pool_config_held(dd->dd_pool));
750 751 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
751 752 prop == ZFS_PROP_SNAPSHOT_LIMIT);
752 753
753 754 /*
754 755 * If we're allowed to change the limit, don't enforce the limit
755 756 * e.g. this can happen if a snapshot is taken by an administrative
756 757 * user in the global zone (i.e. a recursive snapshot by root).
757 758 * However, we must handle the case of delegated permissions where we
758 759 * are allowed to change the limit on the current dataset, but there
759 760 * is another limit in the tree above.
760 761 */
761 762 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
762 763 if (enforce == ENFORCE_NEVER)
763 764 return (0);
764 765
765 766 /*
766 767 * e.g. if renaming a dataset with no snapshots, count adjustment
767 768 * is 0.
768 769 */
769 770 if (delta == 0)
770 771 return (0);
771 772
772 773 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
773 774 /*
774 775 * We don't enforce the limit for temporary snapshots. This is
775 776 * indicated by a NULL cred_t argument.
776 777 */
777 778 if (cr == NULL)
778 779 return (0);
779 780
780 781 count_prop = DD_FIELD_SNAPSHOT_COUNT;
781 782 } else {
782 783 count_prop = DD_FIELD_FILESYSTEM_COUNT;
783 784 }
784 785
785 786 /*
786 787 * If an ancestor has been provided, stop checking the limit once we
787 788 * hit that dir. We need this during rename so that we don't overcount
788 789 * the check once we recurse up to the common ancestor.
789 790 */
790 791 if (ancestor == dd)
791 792 return (0);
792 793
793 794 /*
794 795 * If we hit an uninitialized node while recursing up the tree, we can
795 796 * stop since we know there is no limit here (or above). The counts are
796 797 * not valid on this node and we know we won't touch this node's counts.
797 798 */
798 799 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
799 800 count_prop, sizeof (count), 1, &count) == ENOENT)
800 801 return (0);
801 802
802 803 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
803 804 B_FALSE);
804 805 if (err != 0)
805 806 return (err);
806 807
807 808 /* Is there a limit which we've hit? */
808 809 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
809 810 return (SET_ERROR(EDQUOT));
810 811
811 812 if (dd->dd_parent != NULL)
812 813 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
813 814 ancestor, cr);
814 815
815 816 return (err);
816 817 }
817 818
818 819 /*
819 820 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
820 821 * parents. When a new filesystem/snapshot is created, increment the count on
821 822 * all parents, and when a filesystem/snapshot is destroyed, decrement the
822 823 * count.
823 824 */
824 825 void
825 826 dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
826 827 dmu_tx_t *tx)
827 828 {
828 829 int err;
829 830 objset_t *os = dd->dd_pool->dp_meta_objset;
830 831 uint64_t count;
831 832
832 833 ASSERT(dsl_pool_config_held(dd->dd_pool));
833 834 ASSERT(dmu_tx_is_syncing(tx));
834 835 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
835 836 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
836 837
837 838 /*
838 839 * When we receive an incremental stream into a filesystem that already
839 840 * exists, a temporary clone is created. We don't count this temporary
840 841 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
841 842 * $MOS & $ORIGIN) objsets.
842 843 */
843 844 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
844 845 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
845 846 return;
846 847
847 848 /*
848 849 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
849 850 */
850 851 if (delta == 0)
851 852 return;
852 853
853 854 /*
854 855 * If we hit an uninitialized node while recursing up the tree, we can
855 856 * stop since we know the counts are not valid on this node and we
856 857 * know we shouldn't touch this node's counts. An uninitialized count
857 858 * on the node indicates that either the feature has not yet been
858 859 * activated or there are no limits on this part of the tree.
859 860 */
860 861 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
861 862 prop, sizeof (count), 1, &count)) == ENOENT)
862 863 return;
863 864 VERIFY0(err);
864 865
865 866 count += delta;
866 867 /* Use a signed verify to make sure we're not neg. */
867 868 VERIFY3S(count, >=, 0);
868 869
869 870 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
870 871 tx));
871 872
872 873 /* Roll up this additional count into our ancestors */
873 874 if (dd->dd_parent != NULL)
874 875 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
875 876 }
876 877
877 878 uint64_t
878 879 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
879 880 dmu_tx_t *tx)
880 881 {
881 882 objset_t *mos = dp->dp_meta_objset;
882 883 uint64_t ddobj;
883 884 dsl_dir_phys_t *ddphys;
884 885 dmu_buf_t *dbuf;
885 886
886 887 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
887 888 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
888 889 if (pds) {
889 890 VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
890 891 name, sizeof (uint64_t), 1, &ddobj, tx));
891 892 } else {
892 893 /* it's the root dir */
893 894 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
894 895 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
895 896 }
896 897 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
897 898 dmu_buf_will_dirty(dbuf, tx);
898 899 ddphys = dbuf->db_data;
899 900
900 901 ddphys->dd_creation_time = gethrestime_sec();
901 902 if (pds) {
902 903 ddphys->dd_parent_obj = pds->dd_object;
903 904
904 905 /* update the filesystem counts */
905 906 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
906 907 }
907 908 ddphys->dd_props_zapobj = zap_create(mos,
908 909 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
909 910 ddphys->dd_child_dir_zapobj = zap_create(mos,
910 911 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
911 912 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
912 913 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
913 914 dmu_buf_rele(dbuf, FTAG);
914 915
915 916 return (ddobj);
916 917 }
917 918
918 919 boolean_t
919 920 dsl_dir_is_clone(dsl_dir_t *dd)
920 921 {
921 922 return (dsl_dir_phys(dd)->dd_origin_obj &&
922 923 (dd->dd_pool->dp_origin_snap == NULL ||
923 924 dsl_dir_phys(dd)->dd_origin_obj !=
924 925 dd->dd_pool->dp_origin_snap->ds_object));
925 926 }
926 927
927 928 void
928 929 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
929 930 {
930 931 mutex_enter(&dd->dd_lock);
931 932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
932 933 dsl_dir_phys(dd)->dd_used_bytes);
933 934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
934 935 dsl_dir_phys(dd)->dd_quota);
935 936 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
936 937 dsl_dir_phys(dd)->dd_reserved);
937 938 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
938 939 dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
939 940 (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
940 941 dsl_dir_phys(dd)->dd_compressed_bytes));
941 942 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
942 943 dsl_dir_phys(dd)->dd_uncompressed_bytes);
943 944 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
944 945 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
945 946 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
946 947 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
947 948 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
948 949 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
949 950 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
950 951 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
951 952 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
952 953 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
953 954 }
954 955 mutex_exit(&dd->dd_lock);
955 956
956 957 if (dsl_dir_is_zapified(dd)) {
957 958 uint64_t count;
958 959 objset_t *os = dd->dd_pool->dp_meta_objset;
959 960
960 961 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
961 962 sizeof (count), 1, &count) == 0) {
962 963 dsl_prop_nvlist_add_uint64(nv,
963 964 ZFS_PROP_FILESYSTEM_COUNT, count);
964 965 }
965 966 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
966 967 sizeof (count), 1, &count) == 0) {
967 968 dsl_prop_nvlist_add_uint64(nv,
968 969 ZFS_PROP_SNAPSHOT_COUNT, count);
969 970 }
970 971 }
971 972
972 973 if (dsl_dir_is_clone(dd)) {
973 974 dsl_dataset_t *ds;
974 975 char buf[MAXNAMELEN];
975 976
976 977 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
977 978 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
978 979 dsl_dataset_name(ds, buf);
979 980 dsl_dataset_rele(ds, FTAG);
980 981 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
981 982 }
982 983 }
983 984
984 985 void
985 986 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
986 987 {
987 988 dsl_pool_t *dp = dd->dd_pool;
988 989
989 990 ASSERT(dsl_dir_phys(dd));
990 991
991 992 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
992 993 /* up the hold count until we can be written out */
993 994 dmu_buf_add_ref(dd->dd_dbuf, dd);
994 995 }
995 996 }
996 997
997 998 static int64_t
998 999 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
999 1000 {
1000 1001 uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1001 1002 uint64_t new_accounted =
1002 1003 MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1003 1004 return (new_accounted - old_accounted);
1004 1005 }
1005 1006
1006 1007 void
1007 1008 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1008 1009 {
1009 1010 ASSERT(dmu_tx_is_syncing(tx));
1010 1011
1011 1012 mutex_enter(&dd->dd_lock);
1012 1013 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1013 1014 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1014 1015 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1015 1016 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1016 1017 mutex_exit(&dd->dd_lock);
1017 1018
1018 1019 /* release the hold from dsl_dir_dirty */
1019 1020 dmu_buf_rele(dd->dd_dbuf, dd);
1020 1021 }
1021 1022
1022 1023 static uint64_t
1023 1024 dsl_dir_space_towrite(dsl_dir_t *dd)
1024 1025 {
1025 1026 uint64_t space = 0;
1026 1027 int i;
1027 1028
1028 1029 ASSERT(MUTEX_HELD(&dd->dd_lock));
1029 1030
1030 1031 for (i = 0; i < TXG_SIZE; i++) {
1031 1032 space += dd->dd_space_towrite[i&TXG_MASK];
1032 1033 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1033 1034 }
1034 1035 return (space);
1035 1036 }
1036 1037
1037 1038 /*
1038 1039 * How much space would dd have available if ancestor had delta applied
1039 1040 * to it? If ondiskonly is set, we're only interested in what's
1040 1041 * on-disk, not estimated pending changes.
1041 1042 */
1042 1043 uint64_t
1043 1044 dsl_dir_space_available(dsl_dir_t *dd,
1044 1045 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1045 1046 {
1046 1047 uint64_t parentspace, myspace, quota, used;
1047 1048
1048 1049 /*
1049 1050 * If there are no restrictions otherwise, assume we have
1050 1051 * unlimited space available.
1051 1052 */
1052 1053 quota = UINT64_MAX;
1053 1054 parentspace = UINT64_MAX;
1054 1055
1055 1056 if (dd->dd_parent != NULL) {
1056 1057 parentspace = dsl_dir_space_available(dd->dd_parent,
1057 1058 ancestor, delta, ondiskonly);
1058 1059 }
1059 1060
1060 1061 mutex_enter(&dd->dd_lock);
1061 1062 if (dsl_dir_phys(dd)->dd_quota != 0)
1062 1063 quota = dsl_dir_phys(dd)->dd_quota;
1063 1064 used = dsl_dir_phys(dd)->dd_used_bytes;
1064 1065 if (!ondiskonly)
1065 1066 used += dsl_dir_space_towrite(dd);
1066 1067
1067 1068 if (dd->dd_parent == NULL) {
1068 1069 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1069 1070 quota = MIN(quota, poolsize);
1070 1071 }
1071 1072
1072 1073 if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1073 1074 /*
1074 1075 * We have some space reserved, in addition to what our
1075 1076 * parent gave us.
1076 1077 */
1077 1078 parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1078 1079 }
1079 1080
1080 1081 if (dd == ancestor) {
1081 1082 ASSERT(delta <= 0);
1082 1083 ASSERT(used >= -delta);
1083 1084 used += delta;
1084 1085 if (parentspace != UINT64_MAX)
1085 1086 parentspace -= delta;
1086 1087 }
1087 1088
1088 1089 if (used > quota) {
1089 1090 /* over quota */
1090 1091 myspace = 0;
1091 1092 } else {
1092 1093 /*
1093 1094 * the lesser of the space provided by our parent and
1094 1095 * the space left in our quota
1095 1096 */
1096 1097 myspace = MIN(parentspace, quota - used);
1097 1098 }
1098 1099
1099 1100 mutex_exit(&dd->dd_lock);
1100 1101
1101 1102 return (myspace);
1102 1103 }
1103 1104
1104 1105 struct tempreserve {
1105 1106 list_node_t tr_node;
1106 1107 dsl_dir_t *tr_ds;
1107 1108 uint64_t tr_size;
1108 1109 };
1109 1110
1110 1111 static int
1111 1112 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1112 1113 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1113 1114 dmu_tx_t *tx, boolean_t first)
1114 1115 {
1115 1116 uint64_t txg = tx->tx_txg;
1116 1117 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1117 1118 uint64_t deferred = 0;
1118 1119 struct tempreserve *tr;
1119 1120 int retval = EDQUOT;
1120 1121 int txgidx = txg & TXG_MASK;
1121 1122 int i;
1122 1123 uint64_t ref_rsrv = 0;
1123 1124
1124 1125 ASSERT3U(txg, !=, 0);
1125 1126 ASSERT3S(asize, >, 0);
1126 1127
1127 1128 mutex_enter(&dd->dd_lock);
1128 1129
1129 1130 /*
1130 1131 * Check against the dsl_dir's quota. We don't add in the delta
1131 1132 * when checking for over-quota because they get one free hit.
1132 1133 */
1133 1134 est_inflight = dsl_dir_space_towrite(dd);
1134 1135 for (i = 0; i < TXG_SIZE; i++)
1135 1136 est_inflight += dd->dd_tempreserved[i];
1136 1137 used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1137 1138
1138 1139 /*
1139 1140 * On the first iteration, fetch the dataset's used-on-disk and
1140 1141 * refreservation values. Also, if checkrefquota is set, test if
1141 1142 * allocating this space would exceed the dataset's refquota.
1142 1143 */
1143 1144 if (first && tx->tx_objset) {
1144 1145 int error;
1145 1146 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1146 1147
1147 1148 error = dsl_dataset_check_quota(ds, checkrefquota,
1148 1149 asize, est_inflight, &used_on_disk, &ref_rsrv);
1149 1150 if (error) {
1150 1151 mutex_exit(&dd->dd_lock);
1151 1152 return (error);
1152 1153 }
1153 1154 }
1154 1155
1155 1156 /*
1156 1157 * If this transaction will result in a net free of space,
1157 1158 * we want to let it through.
1158 1159 */
1159 1160 if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1160 1161 quota = UINT64_MAX;
1161 1162 else
1162 1163 quota = dsl_dir_phys(dd)->dd_quota;
1163 1164
1164 1165 /*
1165 1166 * Adjust the quota against the actual pool size at the root
1166 1167 * minus any outstanding deferred frees.
1167 1168 * To ensure that it's possible to remove files from a full
1168 1169 * pool without inducing transient overcommits, we throttle
1169 1170 * netfree transactions against a quota that is slightly larger,
1170 1171 * but still within the pool's allocation slop. In cases where
1171 1172 * we're very close to full, this will allow a steady trickle of
1172 1173 * removes to get through.
1173 1174 */
1174 1175 if (dd->dd_parent == NULL) {
1175 1176 spa_t *spa = dd->dd_pool->dp_spa;
1176 1177 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1177 1178 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1178 1179 if (poolsize - deferred < quota) {
1179 1180 quota = poolsize - deferred;
1180 1181 retval = ENOSPC;
1181 1182 }
1182 1183 }
1183 1184
1184 1185 /*
1185 1186 * If they are requesting more space, and our current estimate
1186 1187 * is over quota, they get to try again unless the actual
1187 1188 * on-disk is over quota and there are no pending changes (which
1188 1189 * may free up space for us).
1189 1190 */
1190 1191 if (used_on_disk + est_inflight >= quota) {
1191 1192 if (est_inflight > 0 || used_on_disk < quota ||
1192 1193 (retval == ENOSPC && used_on_disk < quota + deferred))
1193 1194 retval = ERESTART;
1194 1195 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1195 1196 "quota=%lluK tr=%lluK err=%d\n",
1196 1197 used_on_disk>>10, est_inflight>>10,
1197 1198 quota>>10, asize>>10, retval);
1198 1199 mutex_exit(&dd->dd_lock);
1199 1200 return (SET_ERROR(retval));
1200 1201 }
1201 1202
1202 1203 /* We need to up our estimated delta before dropping dd_lock */
1203 1204 dd->dd_tempreserved[txgidx] += asize;
1204 1205
1205 1206 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1206 1207 asize - ref_rsrv);
1207 1208 mutex_exit(&dd->dd_lock);
1208 1209
1209 1210 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1210 1211 tr->tr_ds = dd;
1211 1212 tr->tr_size = asize;
1212 1213 list_insert_tail(tr_list, tr);
1213 1214
1214 1215 /* see if it's OK with our parent */
1215 1216 if (dd->dd_parent && parent_rsrv) {
1216 1217 boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1217 1218
1218 1219 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1219 1220 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1220 1221 } else {
1221 1222 return (0);
1222 1223 }
1223 1224 }
1224 1225
1225 1226 /*
1226 1227 * Reserve space in this dsl_dir, to be used in this tx's txg.
1227 1228 * After the space has been dirtied (and dsl_dir_willuse_space()
1228 1229 * has been called), the reservation should be canceled, using
1229 1230 * dsl_dir_tempreserve_clear().
1230 1231 */
1231 1232 int
1232 1233 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1233 1234 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1234 1235 {
1235 1236 int err;
1236 1237 list_t *tr_list;
1237 1238
1238 1239 if (asize == 0) {
1239 1240 *tr_cookiep = NULL;
1240 1241 return (0);
1241 1242 }
1242 1243
1243 1244 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1244 1245 list_create(tr_list, sizeof (struct tempreserve),
1245 1246 offsetof(struct tempreserve, tr_node));
1246 1247 ASSERT3S(asize, >, 0);
1247 1248 ASSERT3S(fsize, >=, 0);
1248 1249
1249 1250 err = arc_tempreserve_space(lsize, tx->tx_txg);
1250 1251 if (err == 0) {
1251 1252 struct tempreserve *tr;
1252 1253
1253 1254 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1254 1255 tr->tr_size = lsize;
1255 1256 list_insert_tail(tr_list, tr);
1256 1257 } else {
1257 1258 if (err == EAGAIN) {
1258 1259 /*
1259 1260 * If arc_memory_throttle() detected that pageout
1260 1261 * is running and we are low on memory, we delay new
1261 1262 * non-pageout transactions to give pageout an
1262 1263 * advantage.
1263 1264 *
1264 1265 * It is unfortunate to be delaying while the caller's
1265 1266 * locks are held.
1266 1267 */
1267 1268 txg_delay(dd->dd_pool, tx->tx_txg,
1268 1269 MSEC2NSEC(10), MSEC2NSEC(10));
1269 1270 err = SET_ERROR(ERESTART);
1270 1271 }
1271 1272 }
1272 1273
1273 1274 if (err == 0) {
1274 1275 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1275 1276 FALSE, asize > usize, tr_list, tx, TRUE);
1276 1277 }
1277 1278
1278 1279 if (err != 0)
1279 1280 dsl_dir_tempreserve_clear(tr_list, tx);
1280 1281 else
1281 1282 *tr_cookiep = tr_list;
1282 1283
1283 1284 return (err);
1284 1285 }
1285 1286
1286 1287 /*
1287 1288 * Clear a temporary reservation that we previously made with
1288 1289 * dsl_dir_tempreserve_space().
1289 1290 */
1290 1291 void
1291 1292 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1292 1293 {
1293 1294 int txgidx = tx->tx_txg & TXG_MASK;
1294 1295 list_t *tr_list = tr_cookie;
1295 1296 struct tempreserve *tr;
1296 1297
1297 1298 ASSERT3U(tx->tx_txg, !=, 0);
1298 1299
1299 1300 if (tr_cookie == NULL)
1300 1301 return;
1301 1302
1302 1303 while ((tr = list_head(tr_list)) != NULL) {
1303 1304 if (tr->tr_ds) {
1304 1305 mutex_enter(&tr->tr_ds->dd_lock);
1305 1306 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1306 1307 tr->tr_size);
1307 1308 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1308 1309 mutex_exit(&tr->tr_ds->dd_lock);
1309 1310 } else {
1310 1311 arc_tempreserve_clear(tr->tr_size);
1311 1312 }
1312 1313 list_remove(tr_list, tr);
1313 1314 kmem_free(tr, sizeof (struct tempreserve));
1314 1315 }
1315 1316
1316 1317 kmem_free(tr_list, sizeof (list_t));
1317 1318 }
1318 1319
1319 1320 /*
1320 1321 * This should be called from open context when we think we're going to write
1321 1322 * or free space, for example when dirtying data. Be conservative; it's okay
1322 1323 * to write less space or free more, but we don't want to write more or free
1323 1324 * less than the amount specified.
1324 1325 */
1325 1326 void
1326 1327 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1327 1328 {
1328 1329 int64_t parent_space;
1329 1330 uint64_t est_used;
1330 1331
1331 1332 mutex_enter(&dd->dd_lock);
1332 1333 if (space > 0)
1333 1334 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1334 1335
1335 1336 est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1336 1337 parent_space = parent_delta(dd, est_used, space);
1337 1338 mutex_exit(&dd->dd_lock);
1338 1339
1339 1340 /* Make sure that we clean up dd_space_to* */
1340 1341 dsl_dir_dirty(dd, tx);
1341 1342
1342 1343 /* XXX this is potentially expensive and unnecessary... */
1343 1344 if (parent_space && dd->dd_parent)
1344 1345 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1345 1346 }
1346 1347
1347 1348 /* call from syncing context when we actually write/free space for this dd */
1348 1349 void
1349 1350 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1350 1351 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1351 1352 {
1352 1353 int64_t accounted_delta;
1353 1354
1354 1355 /*
1355 1356 * dsl_dataset_set_refreservation_sync_impl() calls this with
1356 1357 * dd_lock held, so that it can atomically update
1357 1358 * ds->ds_reserved and the dsl_dir accounting, so that
1358 1359 * dsl_dataset_check_quota() can see dataset and dir accounting
1359 1360 * consistently.
1360 1361 */
1361 1362 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1362 1363
1363 1364 ASSERT(dmu_tx_is_syncing(tx));
1364 1365 ASSERT(type < DD_USED_NUM);
1365 1366
1366 1367 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1367 1368
1368 1369 if (needlock)
1369 1370 mutex_enter(&dd->dd_lock);
1370 1371 accounted_delta =
1371 1372 parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1372 1373 ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1373 1374 ASSERT(compressed >= 0 ||
1374 1375 dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1375 1376 ASSERT(uncompressed >= 0 ||
1376 1377 dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1377 1378 dsl_dir_phys(dd)->dd_used_bytes += used;
1378 1379 dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1379 1380 dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1380 1381
1381 1382 if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1382 1383 ASSERT(used > 0 ||
1383 1384 dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1384 1385 dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1385 1386 #ifdef DEBUG
1386 1387 dd_used_t t;
1387 1388 uint64_t u = 0;
1388 1389 for (t = 0; t < DD_USED_NUM; t++)
1389 1390 u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1390 1391 ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1391 1392 #endif
1392 1393 }
1393 1394 if (needlock)
1394 1395 mutex_exit(&dd->dd_lock);
1395 1396
1396 1397 if (dd->dd_parent != NULL) {
1397 1398 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1398 1399 accounted_delta, compressed, uncompressed, tx);
1399 1400 dsl_dir_transfer_space(dd->dd_parent,
1400 1401 used - accounted_delta,
1401 1402 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1402 1403 }
1403 1404 }
1404 1405
1405 1406 void
1406 1407 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1407 1408 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1408 1409 {
1409 1410 ASSERT(dmu_tx_is_syncing(tx));
1410 1411 ASSERT(oldtype < DD_USED_NUM);
1411 1412 ASSERT(newtype < DD_USED_NUM);
1412 1413
1413 1414 if (delta == 0 ||
1414 1415 !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1415 1416 return;
1416 1417
1417 1418 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1418 1419 mutex_enter(&dd->dd_lock);
1419 1420 ASSERT(delta > 0 ?
1420 1421 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1421 1422 dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1422 1423 ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1423 1424 dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1424 1425 dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1425 1426 mutex_exit(&dd->dd_lock);
1426 1427 }
1427 1428
1428 1429 typedef struct dsl_dir_set_qr_arg {
1429 1430 const char *ddsqra_name;
1430 1431 zprop_source_t ddsqra_source;
1431 1432 uint64_t ddsqra_value;
1432 1433 } dsl_dir_set_qr_arg_t;
1433 1434
1434 1435 static int
1435 1436 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1436 1437 {
1437 1438 dsl_dir_set_qr_arg_t *ddsqra = arg;
1438 1439 dsl_pool_t *dp = dmu_tx_pool(tx);
1439 1440 dsl_dataset_t *ds;
1440 1441 int error;
1441 1442 uint64_t towrite, newval;
1442 1443
1443 1444 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1444 1445 if (error != 0)
1445 1446 return (error);
1446 1447
1447 1448 error = dsl_prop_predict(ds->ds_dir, "quota",
1448 1449 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1449 1450 if (error != 0) {
1450 1451 dsl_dataset_rele(ds, FTAG);
1451 1452 return (error);
1452 1453 }
1453 1454
1454 1455 if (newval == 0) {
1455 1456 dsl_dataset_rele(ds, FTAG);
1456 1457 return (0);
1457 1458 }
1458 1459
1459 1460 mutex_enter(&ds->ds_dir->dd_lock);
1460 1461 /*
1461 1462 * If we are doing the preliminary check in open context, and
1462 1463 * there are pending changes, then don't fail it, since the
1463 1464 * pending changes could under-estimate the amount of space to be
1464 1465 * freed up.
1465 1466 */
1466 1467 towrite = dsl_dir_space_towrite(ds->ds_dir);
1467 1468 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1468 1469 (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1469 1470 newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1470 1471 error = SET_ERROR(ENOSPC);
1471 1472 }
1472 1473 mutex_exit(&ds->ds_dir->dd_lock);
1473 1474 dsl_dataset_rele(ds, FTAG);
1474 1475 return (error);
1475 1476 }
1476 1477
1477 1478 static void
1478 1479 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1479 1480 {
1480 1481 dsl_dir_set_qr_arg_t *ddsqra = arg;
1481 1482 dsl_pool_t *dp = dmu_tx_pool(tx);
1482 1483 dsl_dataset_t *ds;
1483 1484 uint64_t newval;
1484 1485
1485 1486 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1486 1487
1487 1488 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1488 1489 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1489 1490 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1490 1491 &ddsqra->ddsqra_value, tx);
1491 1492
1492 1493 VERIFY0(dsl_prop_get_int_ds(ds,
1493 1494 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1494 1495 } else {
1495 1496 newval = ddsqra->ddsqra_value;
1496 1497 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1497 1498 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1498 1499 }
1499 1500
1500 1501 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1501 1502 mutex_enter(&ds->ds_dir->dd_lock);
1502 1503 dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1503 1504 mutex_exit(&ds->ds_dir->dd_lock);
1504 1505 dsl_dataset_rele(ds, FTAG);
1505 1506 }
1506 1507
1507 1508 int
1508 1509 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1509 1510 {
1510 1511 dsl_dir_set_qr_arg_t ddsqra;
1511 1512
1512 1513 ddsqra.ddsqra_name = ddname;
1513 1514 ddsqra.ddsqra_source = source;
1514 1515 ddsqra.ddsqra_value = quota;
1515 1516
1516 1517 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1517 1518 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1518 1519 }
1519 1520
1520 1521 int
1521 1522 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1522 1523 {
1523 1524 dsl_dir_set_qr_arg_t *ddsqra = arg;
1524 1525 dsl_pool_t *dp = dmu_tx_pool(tx);
1525 1526 dsl_dataset_t *ds;
1526 1527 dsl_dir_t *dd;
1527 1528 uint64_t newval, used, avail;
1528 1529 int error;
1529 1530
1530 1531 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1531 1532 if (error != 0)
1532 1533 return (error);
1533 1534 dd = ds->ds_dir;
1534 1535
1535 1536 /*
1536 1537 * If we are doing the preliminary check in open context, the
1537 1538 * space estimates may be inaccurate.
1538 1539 */
1539 1540 if (!dmu_tx_is_syncing(tx)) {
1540 1541 dsl_dataset_rele(ds, FTAG);
1541 1542 return (0);
1542 1543 }
1543 1544
1544 1545 error = dsl_prop_predict(ds->ds_dir,
1545 1546 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1546 1547 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1547 1548 if (error != 0) {
1548 1549 dsl_dataset_rele(ds, FTAG);
1549 1550 return (error);
1550 1551 }
1551 1552
1552 1553 mutex_enter(&dd->dd_lock);
1553 1554 used = dsl_dir_phys(dd)->dd_used_bytes;
1554 1555 mutex_exit(&dd->dd_lock);
1555 1556
1556 1557 if (dd->dd_parent) {
1557 1558 avail = dsl_dir_space_available(dd->dd_parent,
1558 1559 NULL, 0, FALSE);
1559 1560 } else {
1560 1561 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1561 1562 }
1562 1563
1563 1564 if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1564 1565 uint64_t delta = MAX(used, newval) -
1565 1566 MAX(used, dsl_dir_phys(dd)->dd_reserved);
1566 1567
1567 1568 if (delta > avail ||
1568 1569 (dsl_dir_phys(dd)->dd_quota > 0 &&
1569 1570 newval > dsl_dir_phys(dd)->dd_quota))
1570 1571 error = SET_ERROR(ENOSPC);
1571 1572 }
1572 1573
1573 1574 dsl_dataset_rele(ds, FTAG);
1574 1575 return (error);
1575 1576 }
1576 1577
1577 1578 void
1578 1579 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1579 1580 {
1580 1581 uint64_t used;
1581 1582 int64_t delta;
1582 1583
1583 1584 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1584 1585
1585 1586 mutex_enter(&dd->dd_lock);
1586 1587 used = dsl_dir_phys(dd)->dd_used_bytes;
1587 1588 delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1588 1589 dsl_dir_phys(dd)->dd_reserved = value;
1589 1590
1590 1591 if (dd->dd_parent != NULL) {
1591 1592 /* Roll up this additional usage into our ancestors */
1592 1593 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1593 1594 delta, 0, 0, tx);
1594 1595 }
1595 1596 mutex_exit(&dd->dd_lock);
1596 1597 }
1597 1598
1598 1599
1599 1600 static void
1600 1601 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1601 1602 {
1602 1603 dsl_dir_set_qr_arg_t *ddsqra = arg;
1603 1604 dsl_pool_t *dp = dmu_tx_pool(tx);
1604 1605 dsl_dataset_t *ds;
1605 1606 uint64_t newval;
1606 1607
1607 1608 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1608 1609
1609 1610 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1610 1611 dsl_prop_set_sync_impl(ds,
1611 1612 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1612 1613 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1613 1614 &ddsqra->ddsqra_value, tx);
1614 1615
1615 1616 VERIFY0(dsl_prop_get_int_ds(ds,
1616 1617 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1617 1618 } else {
1618 1619 newval = ddsqra->ddsqra_value;
1619 1620 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1620 1621 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1621 1622 (longlong_t)newval);
1622 1623 }
1623 1624
1624 1625 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1625 1626 dsl_dataset_rele(ds, FTAG);
1626 1627 }
1627 1628
1628 1629 int
1629 1630 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1630 1631 uint64_t reservation)
1631 1632 {
1632 1633 dsl_dir_set_qr_arg_t ddsqra;
1633 1634
1634 1635 ddsqra.ddsqra_name = ddname;
1635 1636 ddsqra.ddsqra_source = source;
1636 1637 ddsqra.ddsqra_value = reservation;
1637 1638
1638 1639 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1639 1640 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1640 1641 }
1641 1642
1642 1643 static dsl_dir_t *
1643 1644 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1644 1645 {
1645 1646 for (; ds1; ds1 = ds1->dd_parent) {
1646 1647 dsl_dir_t *dd;
1647 1648 for (dd = ds2; dd; dd = dd->dd_parent) {
1648 1649 if (ds1 == dd)
1649 1650 return (dd);
1650 1651 }
1651 1652 }
1652 1653 return (NULL);
1653 1654 }
1654 1655
1655 1656 /*
1656 1657 * If delta is applied to dd, how much of that delta would be applied to
1657 1658 * ancestor? Syncing context only.
1658 1659 */
1659 1660 static int64_t
1660 1661 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1661 1662 {
1662 1663 if (dd == ancestor)
1663 1664 return (delta);
1664 1665
1665 1666 mutex_enter(&dd->dd_lock);
1666 1667 delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1667 1668 mutex_exit(&dd->dd_lock);
1668 1669 return (would_change(dd->dd_parent, delta, ancestor));
1669 1670 }
1670 1671
1671 1672 typedef struct dsl_dir_rename_arg {
1672 1673 const char *ddra_oldname;
1673 1674 const char *ddra_newname;
1674 1675 cred_t *ddra_cred;
1675 1676 } dsl_dir_rename_arg_t;
1676 1677
1677 1678 /* ARGSUSED */
1678 1679 static int
1679 1680 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1680 1681 {
1681 1682 int *deltap = arg;
1682 1683 char namebuf[MAXNAMELEN];
1683 1684
1684 1685 dsl_dataset_name(ds, namebuf);
1685 1686
1686 1687 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1687 1688 return (SET_ERROR(ENAMETOOLONG));
1688 1689 return (0);
1689 1690 }
1690 1691
1691 1692 static int
1692 1693 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1693 1694 {
1694 1695 dsl_dir_rename_arg_t *ddra = arg;
1695 1696 dsl_pool_t *dp = dmu_tx_pool(tx);
1696 1697 dsl_dir_t *dd, *newparent;
1697 1698 const char *mynewname;
1698 1699 int error;
1699 1700 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1700 1701
1701 1702 /* target dir should exist */
1702 1703 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1703 1704 if (error != 0)
1704 1705 return (error);
1705 1706
1706 1707 /* new parent should exist */
1707 1708 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1708 1709 &newparent, &mynewname);
1709 1710 if (error != 0) {
1710 1711 dsl_dir_rele(dd, FTAG);
1711 1712 return (error);
1712 1713 }
1713 1714
1714 1715 /* can't rename to different pool */
1715 1716 if (dd->dd_pool != newparent->dd_pool) {
1716 1717 dsl_dir_rele(newparent, FTAG);
1717 1718 dsl_dir_rele(dd, FTAG);
1718 1719 return (SET_ERROR(ENXIO));
1719 1720 }
1720 1721
1721 1722 /* new name should not already exist */
1722 1723 if (mynewname == NULL) {
1723 1724 dsl_dir_rele(newparent, FTAG);
1724 1725 dsl_dir_rele(dd, FTAG);
1725 1726 return (SET_ERROR(EEXIST));
1726 1727 }
1727 1728
1728 1729 /* if the name length is growing, validate child name lengths */
1729 1730 if (delta > 0) {
1730 1731 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1731 1732 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1732 1733 if (error != 0) {
1733 1734 dsl_dir_rele(newparent, FTAG);
1734 1735 dsl_dir_rele(dd, FTAG);
1735 1736 return (error);
1736 1737 }
1737 1738 }
1738 1739
1739 1740 if (dmu_tx_is_syncing(tx)) {
1740 1741 if (spa_feature_is_active(dp->dp_spa,
1741 1742 SPA_FEATURE_FS_SS_LIMIT)) {
1742 1743 /*
1743 1744 * Although this is the check function and we don't
1744 1745 * normally make on-disk changes in check functions,
1745 1746 * we need to do that here.
1746 1747 *
1747 1748 * Ensure this portion of the tree's counts have been
1748 1749 * initialized in case the new parent has limits set.
1749 1750 */
1750 1751 dsl_dir_init_fs_ss_count(dd, tx);
1751 1752 }
1752 1753 }
1753 1754
1754 1755 if (newparent != dd->dd_parent) {
1755 1756 /* is there enough space? */
1756 1757 uint64_t myspace =
1757 1758 MAX(dsl_dir_phys(dd)->dd_used_bytes,
1758 1759 dsl_dir_phys(dd)->dd_reserved);
1759 1760 objset_t *os = dd->dd_pool->dp_meta_objset;
1760 1761 uint64_t fs_cnt = 0;
1761 1762 uint64_t ss_cnt = 0;
1762 1763
1763 1764 if (dsl_dir_is_zapified(dd)) {
1764 1765 int err;
1765 1766
1766 1767 err = zap_lookup(os, dd->dd_object,
1767 1768 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1768 1769 &fs_cnt);
1769 1770 if (err != ENOENT && err != 0) {
1770 1771 dsl_dir_rele(newparent, FTAG);
1771 1772 dsl_dir_rele(dd, FTAG);
1772 1773 return (err);
1773 1774 }
1774 1775
1775 1776 /*
1776 1777 * have to add 1 for the filesystem itself that we're
1777 1778 * moving
1778 1779 */
1779 1780 fs_cnt++;
1780 1781
1781 1782 err = zap_lookup(os, dd->dd_object,
1782 1783 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1783 1784 &ss_cnt);
1784 1785 if (err != ENOENT && err != 0) {
1785 1786 dsl_dir_rele(newparent, FTAG);
1786 1787 dsl_dir_rele(dd, FTAG);
1787 1788 return (err);
1788 1789 }
1789 1790 }
1790 1791
1791 1792 /* no rename into our descendant */
1792 1793 if (closest_common_ancestor(dd, newparent) == dd) {
1793 1794 dsl_dir_rele(newparent, FTAG);
1794 1795 dsl_dir_rele(dd, FTAG);
1795 1796 return (SET_ERROR(EINVAL));
1796 1797 }
1797 1798
1798 1799 error = dsl_dir_transfer_possible(dd->dd_parent,
1799 1800 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1800 1801 if (error != 0) {
1801 1802 dsl_dir_rele(newparent, FTAG);
1802 1803 dsl_dir_rele(dd, FTAG);
1803 1804 return (error);
1804 1805 }
1805 1806 }
1806 1807
1807 1808 dsl_dir_rele(newparent, FTAG);
1808 1809 dsl_dir_rele(dd, FTAG);
1809 1810 return (0);
1810 1811 }
1811 1812
1812 1813 static void
1813 1814 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1814 1815 {
1815 1816 dsl_dir_rename_arg_t *ddra = arg;
1816 1817 dsl_pool_t *dp = dmu_tx_pool(tx);
1817 1818 dsl_dir_t *dd, *newparent;
1818 1819 const char *mynewname;
1819 1820 int error;
1820 1821 objset_t *mos = dp->dp_meta_objset;
1821 1822
1822 1823 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1823 1824 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1824 1825 &mynewname));
1825 1826
1826 1827 /* Log this before we change the name. */
1827 1828 spa_history_log_internal_dd(dd, "rename", tx,
1828 1829 "-> %s", ddra->ddra_newname);
1829 1830
1830 1831 if (newparent != dd->dd_parent) {
1831 1832 objset_t *os = dd->dd_pool->dp_meta_objset;
1832 1833 uint64_t fs_cnt = 0;
1833 1834 uint64_t ss_cnt = 0;
1834 1835
1835 1836 /*
1836 1837 * We already made sure the dd counts were initialized in the
1837 1838 * check function.
1838 1839 */
1839 1840 if (spa_feature_is_active(dp->dp_spa,
1840 1841 SPA_FEATURE_FS_SS_LIMIT)) {
1841 1842 VERIFY0(zap_lookup(os, dd->dd_object,
1842 1843 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1843 1844 &fs_cnt));
1844 1845 /* add 1 for the filesystem itself that we're moving */
1845 1846 fs_cnt++;
1846 1847
1847 1848 VERIFY0(zap_lookup(os, dd->dd_object,
1848 1849 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1849 1850 &ss_cnt));
1850 1851 }
1851 1852
1852 1853 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1853 1854 DD_FIELD_FILESYSTEM_COUNT, tx);
1854 1855 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1855 1856 DD_FIELD_FILESYSTEM_COUNT, tx);
1856 1857
1857 1858 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1858 1859 DD_FIELD_SNAPSHOT_COUNT, tx);
1859 1860 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1860 1861 DD_FIELD_SNAPSHOT_COUNT, tx);
1861 1862
1862 1863 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1863 1864 -dsl_dir_phys(dd)->dd_used_bytes,
1864 1865 -dsl_dir_phys(dd)->dd_compressed_bytes,
1865 1866 -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1866 1867 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1867 1868 dsl_dir_phys(dd)->dd_used_bytes,
1868 1869 dsl_dir_phys(dd)->dd_compressed_bytes,
1869 1870 dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1870 1871
1871 1872 if (dsl_dir_phys(dd)->dd_reserved >
1872 1873 dsl_dir_phys(dd)->dd_used_bytes) {
1873 1874 uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
1874 1875 dsl_dir_phys(dd)->dd_used_bytes;
1875 1876
1876 1877 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1877 1878 -unused_rsrv, 0, 0, tx);
1878 1879 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1879 1880 unused_rsrv, 0, 0, tx);
1880 1881 }
1881 1882 }
1882 1883
1883 1884 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1884 1885
1885 1886 /* remove from old parent zapobj */
1886 1887 error = zap_remove(mos,
1887 1888 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
1888 1889 dd->dd_myname, tx);
1889 1890 ASSERT0(error);
1890 1891
1891 1892 (void) strcpy(dd->dd_myname, mynewname);
1892 1893 dsl_dir_rele(dd->dd_parent, dd);
1893 1894 dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
1894 1895 VERIFY0(dsl_dir_hold_obj(dp,
1895 1896 newparent->dd_object, NULL, dd, &dd->dd_parent));
1896 1897
1897 1898 /* add to new parent zapobj */
1898 1899 VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
1899 1900 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1900 1901
1901 1902 dsl_prop_notify_all(dd);
1902 1903
1903 1904 dsl_dir_rele(newparent, FTAG);
1904 1905 dsl_dir_rele(dd, FTAG);
1905 1906 }
1906 1907
1907 1908 int
1908 1909 dsl_dir_rename(const char *oldname, const char *newname)
1909 1910 {
1910 1911 dsl_dir_rename_arg_t ddra;
1911 1912
1912 1913 ddra.ddra_oldname = oldname;
1913 1914 ddra.ddra_newname = newname;
1914 1915 ddra.ddra_cred = CRED();
1915 1916
1916 1917 return (dsl_sync_task(oldname,
1917 1918 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1918 1919 3, ZFS_SPACE_CHECK_RESERVED));
1919 1920 }
1920 1921
1921 1922 int
1922 1923 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1923 1924 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1924 1925 {
1925 1926 dsl_dir_t *ancestor;
1926 1927 int64_t adelta;
1927 1928 uint64_t avail;
1928 1929 int err;
1929 1930
1930 1931 ancestor = closest_common_ancestor(sdd, tdd);
1931 1932 adelta = would_change(sdd, -space, ancestor);
1932 1933 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1933 1934 if (avail < space)
1934 1935 return (SET_ERROR(ENOSPC));
1935 1936
1936 1937 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1937 1938 ancestor, cr);
1938 1939 if (err != 0)
1939 1940 return (err);
1940 1941 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1941 1942 ancestor, cr);
1942 1943 if (err != 0)
1943 1944 return (err);
1944 1945
1945 1946 return (0);
1946 1947 }
1947 1948
1948 1949 timestruc_t
1949 1950 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1950 1951 {
1951 1952 timestruc_t t;
1952 1953
1953 1954 mutex_enter(&dd->dd_lock);
1954 1955 t = dd->dd_snap_cmtime;
1955 1956 mutex_exit(&dd->dd_lock);
1956 1957
1957 1958 return (t);
1958 1959 }
1959 1960
1960 1961 void
1961 1962 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1962 1963 {
1963 1964 timestruc_t t;
1964 1965
1965 1966 gethrestime(&t);
1966 1967 mutex_enter(&dd->dd_lock);
1967 1968 dd->dd_snap_cmtime = t;
1968 1969 mutex_exit(&dd->dd_lock);
1969 1970 }
1970 1971
1971 1972 void
1972 1973 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1973 1974 {
1974 1975 objset_t *mos = dd->dd_pool->dp_meta_objset;
1975 1976 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1976 1977 }
1977 1978
1978 1979 boolean_t
1979 1980 dsl_dir_is_zapified(dsl_dir_t *dd)
1980 1981 {
1981 1982 dmu_object_info_t doi;
1982 1983
1983 1984 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1984 1985 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1985 1986 }
|
↓ open down ↓ |
1551 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX