Print this page
3006 VERIFY[S,U,P] and ASSERT[S,U,P] frequently check if first argument is zero
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/dmu_objset.c
+++ new/usr/src/uts/common/fs/zfs/dmu_objset.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 by Delphix. All rights reserved.
24 24 */
25 25
26 26 /* Portions Copyright 2010 Robert Milkowski */
27 27
28 28 #include <sys/cred.h>
29 29 #include <sys/zfs_context.h>
30 30 #include <sys/dmu_objset.h>
31 31 #include <sys/dsl_dir.h>
32 32 #include <sys/dsl_dataset.h>
33 33 #include <sys/dsl_prop.h>
34 34 #include <sys/dsl_pool.h>
35 35 #include <sys/dsl_synctask.h>
36 36 #include <sys/dsl_deleg.h>
37 37 #include <sys/dnode.h>
38 38 #include <sys/dbuf.h>
39 39 #include <sys/zvol.h>
40 40 #include <sys/dmu_tx.h>
41 41 #include <sys/zap.h>
42 42 #include <sys/zil.h>
43 43 #include <sys/dmu_impl.h>
44 44 #include <sys/zfs_ioctl.h>
45 45 #include <sys/sa.h>
46 46 #include <sys/zfs_onexit.h>
47 47
48 48 /*
49 49 * Needed to close a window in dnode_move() that allows the objset to be freed
50 50 * before it can be safely accessed.
51 51 */
52 52 krwlock_t os_lock;
53 53
54 54 void
55 55 dmu_objset_init(void)
56 56 {
57 57 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
58 58 }
59 59
60 60 void
61 61 dmu_objset_fini(void)
62 62 {
63 63 rw_destroy(&os_lock);
64 64 }
65 65
66 66 spa_t *
67 67 dmu_objset_spa(objset_t *os)
68 68 {
69 69 return (os->os_spa);
70 70 }
71 71
72 72 zilog_t *
73 73 dmu_objset_zil(objset_t *os)
74 74 {
75 75 return (os->os_zil);
76 76 }
77 77
78 78 dsl_pool_t *
79 79 dmu_objset_pool(objset_t *os)
80 80 {
81 81 dsl_dataset_t *ds;
82 82
83 83 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
84 84 return (ds->ds_dir->dd_pool);
85 85 else
86 86 return (spa_get_dsl(os->os_spa));
87 87 }
88 88
89 89 dsl_dataset_t *
90 90 dmu_objset_ds(objset_t *os)
91 91 {
92 92 return (os->os_dsl_dataset);
93 93 }
94 94
95 95 dmu_objset_type_t
96 96 dmu_objset_type(objset_t *os)
97 97 {
98 98 return (os->os_phys->os_type);
99 99 }
100 100
101 101 void
102 102 dmu_objset_name(objset_t *os, char *buf)
103 103 {
104 104 dsl_dataset_name(os->os_dsl_dataset, buf);
105 105 }
106 106
107 107 uint64_t
108 108 dmu_objset_id(objset_t *os)
109 109 {
110 110 dsl_dataset_t *ds = os->os_dsl_dataset;
111 111
112 112 return (ds ? ds->ds_object : 0);
113 113 }
114 114
115 115 uint64_t
116 116 dmu_objset_syncprop(objset_t *os)
117 117 {
118 118 return (os->os_sync);
119 119 }
120 120
121 121 uint64_t
122 122 dmu_objset_logbias(objset_t *os)
123 123 {
124 124 return (os->os_logbias);
125 125 }
126 126
127 127 static void
128 128 checksum_changed_cb(void *arg, uint64_t newval)
129 129 {
130 130 objset_t *os = arg;
131 131
132 132 /*
133 133 * Inheritance should have been done by now.
134 134 */
135 135 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
136 136
137 137 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
138 138 }
139 139
140 140 static void
141 141 compression_changed_cb(void *arg, uint64_t newval)
142 142 {
143 143 objset_t *os = arg;
144 144
145 145 /*
146 146 * Inheritance and range checking should have been done by now.
147 147 */
148 148 ASSERT(newval != ZIO_COMPRESS_INHERIT);
149 149
150 150 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
151 151 }
152 152
153 153 static void
154 154 copies_changed_cb(void *arg, uint64_t newval)
155 155 {
156 156 objset_t *os = arg;
157 157
158 158 /*
159 159 * Inheritance and range checking should have been done by now.
160 160 */
161 161 ASSERT(newval > 0);
162 162 ASSERT(newval <= spa_max_replication(os->os_spa));
163 163
164 164 os->os_copies = newval;
165 165 }
166 166
167 167 static void
168 168 dedup_changed_cb(void *arg, uint64_t newval)
169 169 {
170 170 objset_t *os = arg;
171 171 spa_t *spa = os->os_spa;
172 172 enum zio_checksum checksum;
173 173
174 174 /*
175 175 * Inheritance should have been done by now.
176 176 */
177 177 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
178 178
179 179 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
180 180
181 181 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
182 182 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
183 183 }
184 184
185 185 static void
186 186 primary_cache_changed_cb(void *arg, uint64_t newval)
187 187 {
188 188 objset_t *os = arg;
189 189
190 190 /*
191 191 * Inheritance and range checking should have been done by now.
192 192 */
193 193 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
194 194 newval == ZFS_CACHE_METADATA);
195 195
196 196 os->os_primary_cache = newval;
197 197 }
198 198
199 199 static void
200 200 secondary_cache_changed_cb(void *arg, uint64_t newval)
201 201 {
202 202 objset_t *os = arg;
203 203
204 204 /*
205 205 * Inheritance and range checking should have been done by now.
206 206 */
207 207 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
208 208 newval == ZFS_CACHE_METADATA);
209 209
210 210 os->os_secondary_cache = newval;
211 211 }
212 212
213 213 static void
214 214 sync_changed_cb(void *arg, uint64_t newval)
215 215 {
216 216 objset_t *os = arg;
217 217
218 218 /*
219 219 * Inheritance and range checking should have been done by now.
220 220 */
221 221 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
222 222 newval == ZFS_SYNC_DISABLED);
223 223
224 224 os->os_sync = newval;
225 225 if (os->os_zil)
226 226 zil_set_sync(os->os_zil, newval);
227 227 }
228 228
229 229 static void
230 230 logbias_changed_cb(void *arg, uint64_t newval)
231 231 {
232 232 objset_t *os = arg;
233 233
234 234 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
235 235 newval == ZFS_LOGBIAS_THROUGHPUT);
236 236 os->os_logbias = newval;
237 237 if (os->os_zil)
238 238 zil_set_logbias(os->os_zil, newval);
239 239 }
240 240
241 241 void
242 242 dmu_objset_byteswap(void *buf, size_t size)
243 243 {
244 244 objset_phys_t *osp = buf;
245 245
246 246 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
247 247 dnode_byteswap(&osp->os_meta_dnode);
248 248 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
249 249 osp->os_type = BSWAP_64(osp->os_type);
250 250 osp->os_flags = BSWAP_64(osp->os_flags);
251 251 if (size == sizeof (objset_phys_t)) {
252 252 dnode_byteswap(&osp->os_userused_dnode);
253 253 dnode_byteswap(&osp->os_groupused_dnode);
254 254 }
255 255 }
256 256
257 257 int
258 258 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
259 259 objset_t **osp)
260 260 {
261 261 objset_t *os;
262 262 int i, err;
263 263
264 264 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
265 265
266 266 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
267 267 os->os_dsl_dataset = ds;
268 268 os->os_spa = spa;
269 269 os->os_rootbp = bp;
270 270 if (!BP_IS_HOLE(os->os_rootbp)) {
271 271 uint32_t aflags = ARC_WAIT;
272 272 zbookmark_t zb;
273 273 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
274 274 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
275 275
276 276 if (DMU_OS_IS_L2CACHEABLE(os))
277 277 aflags |= ARC_L2CACHE;
278 278
279 279 dprintf_bp(os->os_rootbp, "reading %s", "");
280 280 /*
281 281 * XXX when bprewrite scrub can change the bp,
282 282 * and this is called from dmu_objset_open_ds_os, the bp
283 283 * could change, and we'll need a lock.
284 284 */
285 285 err = dsl_read_nolock(NULL, spa, os->os_rootbp,
286 286 arc_getbuf_func, &os->os_phys_buf,
287 287 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
288 288 if (err) {
289 289 kmem_free(os, sizeof (objset_t));
290 290 /* convert checksum errors into IO errors */
291 291 if (err == ECKSUM)
292 292 err = EIO;
293 293 return (err);
294 294 }
295 295
296 296 /* Increase the blocksize if we are permitted. */
297 297 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
298 298 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
299 299 arc_buf_t *buf = arc_buf_alloc(spa,
300 300 sizeof (objset_phys_t), &os->os_phys_buf,
301 301 ARC_BUFC_METADATA);
302 302 bzero(buf->b_data, sizeof (objset_phys_t));
303 303 bcopy(os->os_phys_buf->b_data, buf->b_data,
304 304 arc_buf_size(os->os_phys_buf));
305 305 (void) arc_buf_remove_ref(os->os_phys_buf,
306 306 &os->os_phys_buf);
307 307 os->os_phys_buf = buf;
308 308 }
309 309
310 310 os->os_phys = os->os_phys_buf->b_data;
311 311 os->os_flags = os->os_phys->os_flags;
312 312 } else {
313 313 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
314 314 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
315 315 os->os_phys_buf = arc_buf_alloc(spa, size,
316 316 &os->os_phys_buf, ARC_BUFC_METADATA);
317 317 os->os_phys = os->os_phys_buf->b_data;
318 318 bzero(os->os_phys, size);
319 319 }
320 320
321 321 /*
322 322 * Note: the changed_cb will be called once before the register
323 323 * func returns, thus changing the checksum/compression from the
324 324 * default (fletcher2/off). Snapshots don't need to know about
325 325 * checksum/compression/copies.
326 326 */
327 327 if (ds) {
328 328 err = dsl_prop_register(ds, "primarycache",
329 329 primary_cache_changed_cb, os);
330 330 if (err == 0)
331 331 err = dsl_prop_register(ds, "secondarycache",
332 332 secondary_cache_changed_cb, os);
333 333 if (!dsl_dataset_is_snapshot(ds)) {
334 334 if (err == 0)
335 335 err = dsl_prop_register(ds, "checksum",
336 336 checksum_changed_cb, os);
337 337 if (err == 0)
338 338 err = dsl_prop_register(ds, "compression",
339 339 compression_changed_cb, os);
340 340 if (err == 0)
341 341 err = dsl_prop_register(ds, "copies",
342 342 copies_changed_cb, os);
343 343 if (err == 0)
344 344 err = dsl_prop_register(ds, "dedup",
345 345 dedup_changed_cb, os);
346 346 if (err == 0)
347 347 err = dsl_prop_register(ds, "logbias",
348 348 logbias_changed_cb, os);
349 349 if (err == 0)
350 350 err = dsl_prop_register(ds, "sync",
351 351 sync_changed_cb, os);
352 352 }
353 353 if (err) {
354 354 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
355 355 &os->os_phys_buf) == 1);
356 356 kmem_free(os, sizeof (objset_t));
357 357 return (err);
358 358 }
359 359 } else if (ds == NULL) {
360 360 /* It's the meta-objset. */
361 361 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
362 362 os->os_compress = ZIO_COMPRESS_LZJB;
363 363 os->os_copies = spa_max_replication(spa);
364 364 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
365 365 os->os_dedup_verify = 0;
366 366 os->os_logbias = 0;
367 367 os->os_sync = 0;
368 368 os->os_primary_cache = ZFS_CACHE_ALL;
369 369 os->os_secondary_cache = ZFS_CACHE_ALL;
370 370 }
371 371
372 372 if (ds == NULL || !dsl_dataset_is_snapshot(ds))
373 373 os->os_zil_header = os->os_phys->os_zil_header;
374 374 os->os_zil = zil_alloc(os, &os->os_zil_header);
375 375
376 376 for (i = 0; i < TXG_SIZE; i++) {
377 377 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
378 378 offsetof(dnode_t, dn_dirty_link[i]));
379 379 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
380 380 offsetof(dnode_t, dn_dirty_link[i]));
381 381 }
382 382 list_create(&os->os_dnodes, sizeof (dnode_t),
383 383 offsetof(dnode_t, dn_link));
384 384 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
385 385 offsetof(dmu_buf_impl_t, db_link));
386 386
387 387 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
388 388 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
389 389 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
390 390
391 391 DMU_META_DNODE(os) = dnode_special_open(os,
392 392 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
393 393 &os->os_meta_dnode);
394 394 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
395 395 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
396 396 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
397 397 &os->os_userused_dnode);
398 398 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
399 399 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
400 400 &os->os_groupused_dnode);
401 401 }
402 402
403 403 /*
404 404 * We should be the only thread trying to do this because we
405 405 * have ds_opening_lock
406 406 */
407 407 if (ds) {
408 408 mutex_enter(&ds->ds_lock);
409 409 ASSERT(ds->ds_objset == NULL);
410 410 ds->ds_objset = os;
411 411 mutex_exit(&ds->ds_lock);
412 412 }
413 413
414 414 *osp = os;
415 415 return (0);
416 416 }
417 417
418 418 int
419 419 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
420 420 {
421 421 int err = 0;
422 422
423 423 mutex_enter(&ds->ds_opening_lock);
424 424 *osp = ds->ds_objset;
425 425 if (*osp == NULL) {
426 426 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
427 427 ds, dsl_dataset_get_blkptr(ds), osp);
428 428 }
429 429 mutex_exit(&ds->ds_opening_lock);
430 430 return (err);
431 431 }
432 432
433 433 /* called from zpl */
434 434 int
435 435 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
436 436 {
437 437 dsl_dataset_t *ds;
438 438 int err;
439 439
440 440 err = dsl_dataset_hold(name, tag, &ds);
441 441 if (err)
442 442 return (err);
443 443
444 444 err = dmu_objset_from_ds(ds, osp);
445 445 if (err)
446 446 dsl_dataset_rele(ds, tag);
447 447
448 448 return (err);
449 449 }
450 450
451 451 /* called from zpl */
452 452 int
453 453 dmu_objset_own(const char *name, dmu_objset_type_t type,
454 454 boolean_t readonly, void *tag, objset_t **osp)
455 455 {
456 456 dsl_dataset_t *ds;
457 457 int err;
458 458
459 459 err = dsl_dataset_own(name, B_FALSE, tag, &ds);
460 460 if (err)
461 461 return (err);
462 462
463 463 err = dmu_objset_from_ds(ds, osp);
464 464 if (err) {
465 465 dsl_dataset_disown(ds, tag);
466 466 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
467 467 dmu_objset_disown(*osp, tag);
468 468 return (EINVAL);
469 469 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
470 470 dmu_objset_disown(*osp, tag);
471 471 return (EROFS);
472 472 }
473 473 return (err);
474 474 }
475 475
476 476 void
477 477 dmu_objset_rele(objset_t *os, void *tag)
478 478 {
479 479 dsl_dataset_rele(os->os_dsl_dataset, tag);
480 480 }
481 481
482 482 void
483 483 dmu_objset_disown(objset_t *os, void *tag)
484 484 {
485 485 dsl_dataset_disown(os->os_dsl_dataset, tag);
486 486 }
487 487
488 488 int
489 489 dmu_objset_evict_dbufs(objset_t *os)
490 490 {
491 491 dnode_t *dn;
492 492
493 493 mutex_enter(&os->os_lock);
494 494
495 495 /* process the mdn last, since the other dnodes have holds on it */
496 496 list_remove(&os->os_dnodes, DMU_META_DNODE(os));
497 497 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
498 498
499 499 /*
500 500 * Find the first dnode with holds. We have to do this dance
501 501 * because dnode_add_ref() only works if you already have a
502 502 * hold. If there are no holds then it has no dbufs so OK to
503 503 * skip.
504 504 */
505 505 for (dn = list_head(&os->os_dnodes);
506 506 dn && !dnode_add_ref(dn, FTAG);
507 507 dn = list_next(&os->os_dnodes, dn))
508 508 continue;
509 509
510 510 while (dn) {
511 511 dnode_t *next_dn = dn;
512 512
513 513 do {
514 514 next_dn = list_next(&os->os_dnodes, next_dn);
515 515 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
516 516
517 517 mutex_exit(&os->os_lock);
518 518 dnode_evict_dbufs(dn);
519 519 dnode_rele(dn, FTAG);
520 520 mutex_enter(&os->os_lock);
521 521 dn = next_dn;
522 522 }
523 523 dn = list_head(&os->os_dnodes);
524 524 mutex_exit(&os->os_lock);
525 525 return (dn != DMU_META_DNODE(os));
526 526 }
527 527
528 528 void
529 529 dmu_objset_evict(objset_t *os)
530 530 {
531 531 dsl_dataset_t *ds = os->os_dsl_dataset;
532 532
533 533 for (int t = 0; t < TXG_SIZE; t++)
534 534 ASSERT(!dmu_objset_is_dirty(os, t));
535 535
536 536 if (ds) {
537 537 if (!dsl_dataset_is_snapshot(ds)) {
538 538 VERIFY(0 == dsl_prop_unregister(ds, "checksum",
539 539 checksum_changed_cb, os));
540 540 VERIFY(0 == dsl_prop_unregister(ds, "compression",
541 541 compression_changed_cb, os));
542 542 VERIFY(0 == dsl_prop_unregister(ds, "copies",
543 543 copies_changed_cb, os));
544 544 VERIFY(0 == dsl_prop_unregister(ds, "dedup",
545 545 dedup_changed_cb, os));
546 546 VERIFY(0 == dsl_prop_unregister(ds, "logbias",
547 547 logbias_changed_cb, os));
548 548 VERIFY(0 == dsl_prop_unregister(ds, "sync",
549 549 sync_changed_cb, os));
550 550 }
551 551 VERIFY(0 == dsl_prop_unregister(ds, "primarycache",
552 552 primary_cache_changed_cb, os));
553 553 VERIFY(0 == dsl_prop_unregister(ds, "secondarycache",
554 554 secondary_cache_changed_cb, os));
555 555 }
556 556
557 557 if (os->os_sa)
558 558 sa_tear_down(os);
559 559
560 560 /*
561 561 * We should need only a single pass over the dnode list, since
562 562 * nothing can be added to the list at this point.
563 563 */
564 564 (void) dmu_objset_evict_dbufs(os);
565 565
566 566 dnode_special_close(&os->os_meta_dnode);
567 567 if (DMU_USERUSED_DNODE(os)) {
568 568 dnode_special_close(&os->os_userused_dnode);
569 569 dnode_special_close(&os->os_groupused_dnode);
570 570 }
571 571 zil_free(os->os_zil);
572 572
573 573 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
574 574
575 575 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf) == 1);
576 576
577 577 /*
578 578 * This is a barrier to prevent the objset from going away in
579 579 * dnode_move() until we can safely ensure that the objset is still in
580 580 * use. We consider the objset valid before the barrier and invalid
581 581 * after the barrier.
582 582 */
583 583 rw_enter(&os_lock, RW_READER);
584 584 rw_exit(&os_lock);
585 585
586 586 mutex_destroy(&os->os_lock);
587 587 mutex_destroy(&os->os_obj_lock);
588 588 mutex_destroy(&os->os_user_ptr_lock);
589 589 kmem_free(os, sizeof (objset_t));
590 590 }
591 591
592 592 timestruc_t
593 593 dmu_objset_snap_cmtime(objset_t *os)
594 594 {
595 595 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
596 596 }
597 597
598 598 /* called from dsl for meta-objset */
599 599 objset_t *
600 600 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
601 601 dmu_objset_type_t type, dmu_tx_t *tx)
602 602 {
603 603 objset_t *os;
604 604 dnode_t *mdn;
605 605
606 606 ASSERT(dmu_tx_is_syncing(tx));
607 607 if (ds != NULL)
608 608 VERIFY(0 == dmu_objset_from_ds(ds, &os));
609 609 else
610 610 VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os));
611 611
612 612 mdn = DMU_META_DNODE(os);
613 613
614 614 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
615 615 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
616 616
617 617 /*
618 618 * We don't want to have to increase the meta-dnode's nlevels
619 619 * later, because then we could do it in quescing context while
620 620 * we are also accessing it in open context.
621 621 *
622 622 * This precaution is not necessary for the MOS (ds == NULL),
623 623 * because the MOS is only updated in syncing context.
624 624 * This is most fortunate: the MOS is the only objset that
625 625 * needs to be synced multiple times as spa_sync() iterates
626 626 * to convergence, so minimizing its dn_nlevels matters.
627 627 */
628 628 if (ds != NULL) {
629 629 int levels = 1;
630 630
631 631 /*
632 632 * Determine the number of levels necessary for the meta-dnode
633 633 * to contain DN_MAX_OBJECT dnodes.
634 634 */
635 635 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
636 636 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
637 637 DN_MAX_OBJECT * sizeof (dnode_phys_t))
638 638 levels++;
639 639
640 640 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
641 641 mdn->dn_nlevels = levels;
642 642 }
643 643
644 644 ASSERT(type != DMU_OST_NONE);
645 645 ASSERT(type != DMU_OST_ANY);
646 646 ASSERT(type < DMU_OST_NUMTYPES);
647 647 os->os_phys->os_type = type;
648 648 if (dmu_objset_userused_enabled(os)) {
649 649 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
650 650 os->os_flags = os->os_phys->os_flags;
651 651 }
652 652
653 653 dsl_dataset_dirty(ds, tx);
654 654
655 655 return (os);
656 656 }
657 657
658 658 struct oscarg {
659 659 void (*userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
660 660 void *userarg;
661 661 dsl_dataset_t *clone_origin;
662 662 const char *lastname;
663 663 dmu_objset_type_t type;
664 664 uint64_t flags;
665 665 cred_t *cr;
666 666 };
667 667
668 668 /*ARGSUSED*/
669 669 static int
670 670 dmu_objset_create_check(void *arg1, void *arg2, dmu_tx_t *tx)
671 671 {
672 672 dsl_dir_t *dd = arg1;
673 673 struct oscarg *oa = arg2;
674 674 objset_t *mos = dd->dd_pool->dp_meta_objset;
675 675 int err;
676 676 uint64_t ddobj;
677 677
678 678 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
679 679 oa->lastname, sizeof (uint64_t), 1, &ddobj);
680 680 if (err != ENOENT)
681 681 return (err ? err : EEXIST);
682 682
683 683 if (oa->clone_origin != NULL) {
684 684 /* You can't clone across pools. */
685 685 if (oa->clone_origin->ds_dir->dd_pool != dd->dd_pool)
686 686 return (EXDEV);
687 687
688 688 /* You can only clone snapshots, not the head datasets. */
689 689 if (!dsl_dataset_is_snapshot(oa->clone_origin))
690 690 return (EINVAL);
691 691 }
692 692
693 693 return (0);
694 694 }
695 695
696 696 static void
697 697 dmu_objset_create_sync(void *arg1, void *arg2, dmu_tx_t *tx)
698 698 {
699 699 dsl_dir_t *dd = arg1;
700 700 spa_t *spa = dd->dd_pool->dp_spa;
|
↓ open down ↓ |
700 lines elided |
↑ open up ↑ |
701 701 struct oscarg *oa = arg2;
702 702 uint64_t obj;
703 703 dsl_dataset_t *ds;
704 704 blkptr_t *bp;
705 705
706 706 ASSERT(dmu_tx_is_syncing(tx));
707 707
708 708 obj = dsl_dataset_create_sync(dd, oa->lastname,
709 709 oa->clone_origin, oa->flags, oa->cr, tx);
710 710
711 - VERIFY3U(0, ==, dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds));
711 + VERIFY0(dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds));
712 712 bp = dsl_dataset_get_blkptr(ds);
713 713 if (BP_IS_HOLE(bp)) {
714 714 objset_t *os =
715 715 dmu_objset_create_impl(spa, ds, bp, oa->type, tx);
716 716
717 717 if (oa->userfunc)
718 718 oa->userfunc(os, oa->userarg, oa->cr, tx);
719 719 }
720 720
721 721 if (oa->clone_origin == NULL) {
722 722 spa_history_log_internal_ds(ds, "create", tx, "");
723 723 } else {
724 724 char namebuf[MAXNAMELEN];
725 725 dsl_dataset_name(oa->clone_origin, namebuf);
726 726 spa_history_log_internal_ds(ds, "clone", tx,
727 727 "origin=%s (%llu)", namebuf, oa->clone_origin->ds_object);
728 728 }
729 729 dsl_dataset_rele(ds, FTAG);
730 730 }
731 731
732 732 int
733 733 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
734 734 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
735 735 {
736 736 dsl_dir_t *pdd;
737 737 const char *tail;
738 738 int err = 0;
739 739 struct oscarg oa = { 0 };
740 740
741 741 ASSERT(strchr(name, '@') == NULL);
742 742 err = dsl_dir_open(name, FTAG, &pdd, &tail);
743 743 if (err)
744 744 return (err);
745 745 if (tail == NULL) {
746 746 dsl_dir_close(pdd, FTAG);
747 747 return (EEXIST);
748 748 }
749 749
750 750 oa.userfunc = func;
751 751 oa.userarg = arg;
752 752 oa.lastname = tail;
753 753 oa.type = type;
754 754 oa.flags = flags;
755 755 oa.cr = CRED();
756 756
757 757 err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
758 758 dmu_objset_create_sync, pdd, &oa, 5);
759 759 dsl_dir_close(pdd, FTAG);
760 760 return (err);
761 761 }
762 762
763 763 int
764 764 dmu_objset_clone(const char *name, dsl_dataset_t *clone_origin, uint64_t flags)
765 765 {
766 766 dsl_dir_t *pdd;
767 767 const char *tail;
768 768 int err = 0;
769 769 struct oscarg oa = { 0 };
770 770
771 771 ASSERT(strchr(name, '@') == NULL);
772 772 err = dsl_dir_open(name, FTAG, &pdd, &tail);
773 773 if (err)
774 774 return (err);
775 775 if (tail == NULL) {
776 776 dsl_dir_close(pdd, FTAG);
777 777 return (EEXIST);
778 778 }
779 779
780 780 oa.lastname = tail;
781 781 oa.clone_origin = clone_origin;
782 782 oa.flags = flags;
783 783 oa.cr = CRED();
784 784
785 785 err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
786 786 dmu_objset_create_sync, pdd, &oa, 5);
787 787 dsl_dir_close(pdd, FTAG);
788 788 return (err);
789 789 }
790 790
791 791 int
792 792 dmu_objset_destroy(const char *name, boolean_t defer)
793 793 {
794 794 dsl_dataset_t *ds;
795 795 int error;
796 796
797 797 error = dsl_dataset_own(name, B_TRUE, FTAG, &ds);
798 798 if (error == 0) {
799 799 error = dsl_dataset_destroy(ds, FTAG, defer);
800 800 /* dsl_dataset_destroy() closes the ds. */
801 801 }
802 802
803 803 return (error);
804 804 }
805 805
806 806 typedef struct snapallarg {
807 807 dsl_sync_task_group_t *saa_dstg;
808 808 boolean_t saa_needsuspend;
809 809 nvlist_t *saa_props;
810 810
811 811 /* the following are used only if 'temporary' is set: */
812 812 boolean_t saa_temporary;
813 813 const char *saa_htag;
814 814 struct dsl_ds_holdarg *saa_ha;
815 815 dsl_dataset_t *saa_newds;
816 816 } snapallarg_t;
817 817
818 818 typedef struct snaponearg {
819 819 const char *soa_longname; /* long snap name */
820 820 const char *soa_snapname; /* short snap name */
821 821 snapallarg_t *soa_saa;
822 822 } snaponearg_t;
823 823
824 824 static int
825 825 snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
826 826 {
827 827 objset_t *os = arg1;
828 828 snaponearg_t *soa = arg2;
829 829 snapallarg_t *saa = soa->soa_saa;
830 830 int error;
831 831
832 832 /* The props have already been checked by zfs_check_userprops(). */
833 833
834 834 error = dsl_dataset_snapshot_check(os->os_dsl_dataset,
835 835 soa->soa_snapname, tx);
836 836 if (error)
837 837 return (error);
838 838
839 839 if (saa->saa_temporary) {
840 840 /*
841 841 * Ideally we would just call
842 842 * dsl_dataset_user_hold_check() and
843 843 * dsl_dataset_destroy_check() here. However the
844 844 * dataset we want to hold and destroy is the snapshot
845 845 * that we just confirmed we can create, but it won't
846 846 * exist until after these checks are run. Do any
847 847 * checks we can here and if more checks are added to
848 848 * those routines in the future, similar checks may be
849 849 * necessary here.
850 850 */
851 851 if (spa_version(os->os_spa) < SPA_VERSION_USERREFS)
852 852 return (ENOTSUP);
853 853 /*
854 854 * Not checking number of tags because the tag will be
855 855 * unique, as it will be the only tag.
856 856 */
857 857 if (strlen(saa->saa_htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
858 858 return (E2BIG);
859 859
860 860 saa->saa_ha = kmem_alloc(sizeof (struct dsl_ds_holdarg),
861 861 KM_SLEEP);
862 862 saa->saa_ha->temphold = B_TRUE;
863 863 saa->saa_ha->htag = saa->saa_htag;
864 864 }
865 865 return (error);
866 866 }
867 867
868 868 static void
869 869 snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
870 870 {
871 871 objset_t *os = arg1;
872 872 dsl_dataset_t *ds = os->os_dsl_dataset;
873 873 snaponearg_t *soa = arg2;
874 874 snapallarg_t *saa = soa->soa_saa;
875 875
876 876 dsl_dataset_snapshot_sync(ds, soa->soa_snapname, tx);
877 877
878 878 if (saa->saa_props != NULL) {
879 879 dsl_props_arg_t pa;
880 880 pa.pa_props = saa->saa_props;
881 881 pa.pa_source = ZPROP_SRC_LOCAL;
882 882 dsl_props_set_sync(ds->ds_prev, &pa, tx);
883 883 }
884 884
885 885 if (saa->saa_temporary) {
886 886 struct dsl_ds_destroyarg da;
887 887
888 888 dsl_dataset_user_hold_sync(ds->ds_prev, saa->saa_ha, tx);
889 889 kmem_free(saa->saa_ha, sizeof (struct dsl_ds_holdarg));
890 890 saa->saa_ha = NULL;
891 891 saa->saa_newds = ds->ds_prev;
892 892
893 893 da.ds = ds->ds_prev;
894 894 da.defer = B_TRUE;
895 895 dsl_dataset_destroy_sync(&da, FTAG, tx);
896 896 }
897 897 }
898 898
899 899 static int
900 900 snapshot_one_impl(const char *snapname, void *arg)
901 901 {
902 902 char fsname[MAXPATHLEN];
903 903 snapallarg_t *saa = arg;
904 904 snaponearg_t *soa;
905 905 objset_t *os;
906 906 int err;
907 907
908 908 (void) strlcpy(fsname, snapname, sizeof (fsname));
909 909 strchr(fsname, '@')[0] = '\0';
910 910
911 911 err = dmu_objset_hold(fsname, saa, &os);
912 912 if (err != 0)
913 913 return (err);
914 914
915 915 /*
916 916 * If the objset is in an inconsistent state (eg, in the process
917 917 * of being destroyed), don't snapshot it.
918 918 */
919 919 if (os->os_dsl_dataset->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) {
920 920 dmu_objset_rele(os, saa);
921 921 return (EBUSY);
922 922 }
923 923
924 924 if (saa->saa_needsuspend) {
925 925 err = zil_suspend(dmu_objset_zil(os));
926 926 if (err) {
927 927 dmu_objset_rele(os, saa);
928 928 return (err);
929 929 }
930 930 }
931 931
932 932 soa = kmem_zalloc(sizeof (*soa), KM_SLEEP);
933 933 soa->soa_saa = saa;
934 934 soa->soa_longname = snapname;
935 935 soa->soa_snapname = strchr(snapname, '@') + 1;
936 936
937 937 dsl_sync_task_create(saa->saa_dstg, snapshot_check, snapshot_sync,
938 938 os, soa, 3);
939 939
940 940 return (0);
941 941 }
942 942
943 943 /*
944 944 * The snapshots must all be in the same pool.
945 945 */
946 946 int
947 947 dmu_objset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
948 948 {
949 949 dsl_sync_task_t *dst;
950 950 snapallarg_t saa = { 0 };
951 951 spa_t *spa;
952 952 int rv = 0;
953 953 int err;
954 954 nvpair_t *pair;
955 955
956 956 pair = nvlist_next_nvpair(snaps, NULL);
957 957 if (pair == NULL)
958 958 return (0);
959 959
960 960 err = spa_open(nvpair_name(pair), &spa, FTAG);
961 961 if (err)
962 962 return (err);
963 963 saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
964 964 saa.saa_props = props;
965 965 saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
966 966
967 967 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
968 968 pair = nvlist_next_nvpair(snaps, pair)) {
969 969 err = snapshot_one_impl(nvpair_name(pair), &saa);
970 970 if (err != 0) {
971 971 if (errors != NULL) {
972 972 fnvlist_add_int32(errors,
973 973 nvpair_name(pair), err);
974 974 }
975 975 rv = err;
976 976 }
977 977 }
978 978
979 979 /*
980 980 * If any call to snapshot_one_impl() failed, don't execute the
981 981 * sync task. The error handling code below will clean up the
982 982 * snaponearg_t from any successful calls to
983 983 * snapshot_one_impl().
984 984 */
985 985 if (rv == 0)
986 986 err = dsl_sync_task_group_wait(saa.saa_dstg);
987 987 if (err != 0)
988 988 rv = err;
989 989
990 990 for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
991 991 dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
992 992 objset_t *os = dst->dst_arg1;
993 993 snaponearg_t *soa = dst->dst_arg2;
994 994 if (dst->dst_err != 0) {
995 995 if (errors != NULL) {
996 996 fnvlist_add_int32(errors,
997 997 soa->soa_longname, dst->dst_err);
998 998 }
999 999 rv = dst->dst_err;
1000 1000 }
1001 1001
1002 1002 if (saa.saa_needsuspend)
1003 1003 zil_resume(dmu_objset_zil(os));
1004 1004 dmu_objset_rele(os, &saa);
1005 1005 kmem_free(soa, sizeof (*soa));
1006 1006 }
1007 1007
1008 1008 dsl_sync_task_group_destroy(saa.saa_dstg);
1009 1009 spa_close(spa, FTAG);
1010 1010 return (rv);
1011 1011 }
1012 1012
1013 1013 int
1014 1014 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1015 1015 {
1016 1016 int err;
1017 1017 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1018 1018 nvlist_t *snaps = fnvlist_alloc();
1019 1019
1020 1020 fnvlist_add_boolean(snaps, longsnap);
1021 1021 err = dmu_objset_snapshot(snaps, NULL, NULL);
1022 1022 fnvlist_free(snaps);
1023 1023 strfree(longsnap);
1024 1024 return (err);
1025 1025 }
1026 1026
1027 1027 int
1028 1028 dmu_objset_snapshot_tmp(const char *snapname, const char *tag, int cleanup_fd)
1029 1029 {
1030 1030 dsl_sync_task_t *dst;
1031 1031 snapallarg_t saa = { 0 };
1032 1032 spa_t *spa;
1033 1033 minor_t minor;
1034 1034 int err;
1035 1035
1036 1036 err = spa_open(snapname, &spa, FTAG);
1037 1037 if (err)
1038 1038 return (err);
1039 1039 saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
1040 1040 saa.saa_htag = tag;
1041 1041 saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1042 1042 saa.saa_temporary = B_TRUE;
1043 1043
1044 1044 if (cleanup_fd < 0) {
1045 1045 spa_close(spa, FTAG);
1046 1046 return (EINVAL);
1047 1047 }
1048 1048 if ((err = zfs_onexit_fd_hold(cleanup_fd, &minor)) != 0) {
1049 1049 spa_close(spa, FTAG);
1050 1050 return (err);
1051 1051 }
1052 1052
1053 1053 err = snapshot_one_impl(snapname, &saa);
1054 1054
1055 1055 if (err == 0)
1056 1056 err = dsl_sync_task_group_wait(saa.saa_dstg);
1057 1057
1058 1058 for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
1059 1059 dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
1060 1060 objset_t *os = dst->dst_arg1;
1061 1061 dsl_register_onexit_hold_cleanup(saa.saa_newds, tag, minor);
1062 1062 if (saa.saa_needsuspend)
1063 1063 zil_resume(dmu_objset_zil(os));
1064 1064 dmu_objset_rele(os, &saa);
1065 1065 }
1066 1066
1067 1067 zfs_onexit_fd_rele(cleanup_fd);
1068 1068 dsl_sync_task_group_destroy(saa.saa_dstg);
1069 1069 spa_close(spa, FTAG);
1070 1070 return (err);
1071 1071 }
1072 1072
1073 1073
1074 1074 static void
1075 1075 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1076 1076 {
1077 1077 dnode_t *dn;
1078 1078
1079 1079 while (dn = list_head(list)) {
1080 1080 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1081 1081 ASSERT(dn->dn_dbuf->db_data_pending);
1082 1082 /*
1083 1083 * Initialize dn_zio outside dnode_sync() because the
1084 1084 * meta-dnode needs to set it ouside dnode_sync().
1085 1085 */
1086 1086 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1087 1087 ASSERT(dn->dn_zio);
1088 1088
1089 1089 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1090 1090 list_remove(list, dn);
1091 1091
1092 1092 if (newlist) {
1093 1093 (void) dnode_add_ref(dn, newlist);
1094 1094 list_insert_tail(newlist, dn);
1095 1095 }
1096 1096
1097 1097 dnode_sync(dn, tx);
1098 1098 }
1099 1099 }
1100 1100
1101 1101 /* ARGSUSED */
1102 1102 static void
1103 1103 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1104 1104 {
1105 1105 blkptr_t *bp = zio->io_bp;
1106 1106 objset_t *os = arg;
1107 1107 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1108 1108
1109 1109 ASSERT(bp == os->os_rootbp);
1110 1110 ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
1111 1111 ASSERT(BP_GET_LEVEL(bp) == 0);
1112 1112
1113 1113 /*
1114 1114 * Update rootbp fill count: it should be the number of objects
1115 1115 * allocated in the object set (not counting the "special"
1116 1116 * objects that are stored in the objset_phys_t -- the meta
1117 1117 * dnode and user/group accounting objects).
1118 1118 */
1119 1119 bp->blk_fill = 0;
1120 1120 for (int i = 0; i < dnp->dn_nblkptr; i++)
1121 1121 bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
1122 1122 }
1123 1123
1124 1124 /* ARGSUSED */
1125 1125 static void
1126 1126 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1127 1127 {
1128 1128 blkptr_t *bp = zio->io_bp;
1129 1129 blkptr_t *bp_orig = &zio->io_bp_orig;
1130 1130 objset_t *os = arg;
1131 1131
1132 1132 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1133 1133 ASSERT(BP_EQUAL(bp, bp_orig));
1134 1134 } else {
1135 1135 dsl_dataset_t *ds = os->os_dsl_dataset;
1136 1136 dmu_tx_t *tx = os->os_synctx;
1137 1137
1138 1138 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1139 1139 dsl_dataset_block_born(ds, bp, tx);
1140 1140 }
1141 1141 }
1142 1142
1143 1143 /* called from dsl */
1144 1144 void
1145 1145 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1146 1146 {
1147 1147 int txgoff;
1148 1148 zbookmark_t zb;
1149 1149 zio_prop_t zp;
1150 1150 zio_t *zio;
1151 1151 list_t *list;
1152 1152 list_t *newlist = NULL;
1153 1153 dbuf_dirty_record_t *dr;
1154 1154
1155 1155 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1156 1156
1157 1157 ASSERT(dmu_tx_is_syncing(tx));
1158 1158 /* XXX the write_done callback should really give us the tx... */
1159 1159 os->os_synctx = tx;
1160 1160
1161 1161 if (os->os_dsl_dataset == NULL) {
1162 1162 /*
1163 1163 * This is the MOS. If we have upgraded,
1164 1164 * spa_max_replication() could change, so reset
1165 1165 * os_copies here.
|
↓ open down ↓ |
444 lines elided |
↑ open up ↑ |
1166 1166 */
1167 1167 os->os_copies = spa_max_replication(os->os_spa);
1168 1168 }
1169 1169
1170 1170 /*
1171 1171 * Create the root block IO
1172 1172 */
1173 1173 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1174 1174 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1175 1175 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1176 - VERIFY3U(0, ==, arc_release_bp(os->os_phys_buf, &os->os_phys_buf,
1176 + VERIFY0(arc_release_bp(os->os_phys_buf, &os->os_phys_buf,
1177 1177 os->os_rootbp, os->os_spa, &zb));
1178 1178
1179 1179 dmu_write_policy(os, NULL, 0, 0, &zp);
1180 1180
1181 1181 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1182 1182 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
1183 1183 dmu_objset_write_ready, dmu_objset_write_done, os,
1184 1184 ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1185 1185
1186 1186 /*
1187 1187 * Sync special dnodes - the parent IO for the sync is the root block
1188 1188 */
1189 1189 DMU_META_DNODE(os)->dn_zio = zio;
1190 1190 dnode_sync(DMU_META_DNODE(os), tx);
1191 1191
1192 1192 os->os_phys->os_flags = os->os_flags;
1193 1193
1194 1194 if (DMU_USERUSED_DNODE(os) &&
1195 1195 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1196 1196 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1197 1197 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1198 1198 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1199 1199 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1200 1200 }
1201 1201
1202 1202 txgoff = tx->tx_txg & TXG_MASK;
1203 1203
1204 1204 if (dmu_objset_userused_enabled(os)) {
1205 1205 newlist = &os->os_synced_dnodes;
1206 1206 /*
1207 1207 * We must create the list here because it uses the
1208 1208 * dn_dirty_link[] of this txg.
1209 1209 */
1210 1210 list_create(newlist, sizeof (dnode_t),
1211 1211 offsetof(dnode_t, dn_dirty_link[txgoff]));
1212 1212 }
1213 1213
1214 1214 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1215 1215 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1216 1216
1217 1217 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1218 1218 while (dr = list_head(list)) {
1219 1219 ASSERT(dr->dr_dbuf->db_level == 0);
1220 1220 list_remove(list, dr);
1221 1221 if (dr->dr_zio)
1222 1222 zio_nowait(dr->dr_zio);
1223 1223 }
1224 1224 /*
1225 1225 * Free intent log blocks up to this tx.
1226 1226 */
1227 1227 zil_sync(os->os_zil, tx);
1228 1228 os->os_phys->os_zil_header = os->os_zil_header;
1229 1229 zio_nowait(zio);
1230 1230 }
1231 1231
1232 1232 boolean_t
1233 1233 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1234 1234 {
1235 1235 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1236 1236 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1237 1237 }
1238 1238
1239 1239 boolean_t
1240 1240 dmu_objset_is_dirty_anywhere(objset_t *os)
1241 1241 {
1242 1242 for (int t = 0; t < TXG_SIZE; t++)
1243 1243 if (dmu_objset_is_dirty(os, t))
1244 1244 return (B_TRUE);
1245 1245 return (B_FALSE);
1246 1246 }
1247 1247
1248 1248 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1249 1249
1250 1250 void
1251 1251 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1252 1252 {
1253 1253 used_cbs[ost] = cb;
1254 1254 }
1255 1255
1256 1256 boolean_t
1257 1257 dmu_objset_userused_enabled(objset_t *os)
1258 1258 {
1259 1259 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1260 1260 used_cbs[os->os_phys->os_type] != NULL &&
1261 1261 DMU_USERUSED_DNODE(os) != NULL);
|
↓ open down ↓ |
75 lines elided |
↑ open up ↑ |
1262 1262 }
1263 1263
1264 1264 static void
1265 1265 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1266 1266 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1267 1267 {
1268 1268 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1269 1269 int64_t delta = DNODE_SIZE + used;
1270 1270 if (subtract)
1271 1271 delta = -delta;
1272 - VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1272 + VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT,
1273 1273 user, delta, tx));
1274 - VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1274 + VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1275 1275 group, delta, tx));
1276 1276 }
1277 1277 }
1278 1278
1279 1279 void
1280 1280 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1281 1281 {
1282 1282 dnode_t *dn;
1283 1283 list_t *list = &os->os_synced_dnodes;
1284 1284
1285 1285 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1286 1286
1287 1287 while (dn = list_head(list)) {
1288 1288 int flags;
1289 1289 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1290 1290 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1291 1291 dn->dn_phys->dn_flags &
1292 1292 DNODE_FLAG_USERUSED_ACCOUNTED);
1293 1293
1294 1294 /* Allocate the user/groupused objects if necessary. */
1295 1295 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1296 1296 VERIFY(0 == zap_create_claim(os,
1297 1297 DMU_USERUSED_OBJECT,
1298 1298 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1299 1299 VERIFY(0 == zap_create_claim(os,
1300 1300 DMU_GROUPUSED_OBJECT,
1301 1301 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1302 1302 }
1303 1303
1304 1304 /*
1305 1305 * We intentionally modify the zap object even if the
1306 1306 * net delta is zero. Otherwise
1307 1307 * the block of the zap obj could be shared between
1308 1308 * datasets but need to be different between them after
1309 1309 * a bprewrite.
1310 1310 */
1311 1311
1312 1312 flags = dn->dn_id_flags;
1313 1313 ASSERT(flags);
1314 1314 if (flags & DN_ID_OLD_EXIST) {
1315 1315 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1316 1316 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1317 1317 }
1318 1318 if (flags & DN_ID_NEW_EXIST) {
1319 1319 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1320 1320 dn->dn_phys->dn_flags, dn->dn_newuid,
1321 1321 dn->dn_newgid, B_FALSE, tx);
1322 1322 }
1323 1323
1324 1324 mutex_enter(&dn->dn_mtx);
1325 1325 dn->dn_oldused = 0;
1326 1326 dn->dn_oldflags = 0;
1327 1327 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1328 1328 dn->dn_olduid = dn->dn_newuid;
1329 1329 dn->dn_oldgid = dn->dn_newgid;
1330 1330 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1331 1331 if (dn->dn_bonuslen == 0)
1332 1332 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1333 1333 else
1334 1334 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1335 1335 }
1336 1336 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1337 1337 mutex_exit(&dn->dn_mtx);
1338 1338
1339 1339 list_remove(list, dn);
1340 1340 dnode_rele(dn, list);
1341 1341 }
1342 1342 }
1343 1343
1344 1344 /*
1345 1345 * Returns a pointer to data to find uid/gid from
1346 1346 *
1347 1347 * If a dirty record for transaction group that is syncing can't
1348 1348 * be found then NULL is returned. In the NULL case it is assumed
1349 1349 * the uid/gid aren't changing.
1350 1350 */
1351 1351 static void *
1352 1352 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1353 1353 {
1354 1354 dbuf_dirty_record_t *dr, **drp;
1355 1355 void *data;
1356 1356
1357 1357 if (db->db_dirtycnt == 0)
1358 1358 return (db->db.db_data); /* Nothing is changing */
1359 1359
1360 1360 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1361 1361 if (dr->dr_txg == tx->tx_txg)
1362 1362 break;
1363 1363
1364 1364 if (dr == NULL) {
1365 1365 data = NULL;
1366 1366 } else {
1367 1367 dnode_t *dn;
1368 1368
1369 1369 DB_DNODE_ENTER(dr->dr_dbuf);
1370 1370 dn = DB_DNODE(dr->dr_dbuf);
1371 1371
1372 1372 if (dn->dn_bonuslen == 0 &&
1373 1373 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1374 1374 data = dr->dt.dl.dr_data->b_data;
1375 1375 else
1376 1376 data = dr->dt.dl.dr_data;
1377 1377
1378 1378 DB_DNODE_EXIT(dr->dr_dbuf);
1379 1379 }
1380 1380
1381 1381 return (data);
1382 1382 }
1383 1383
1384 1384 void
1385 1385 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1386 1386 {
1387 1387 objset_t *os = dn->dn_objset;
1388 1388 void *data = NULL;
1389 1389 dmu_buf_impl_t *db = NULL;
1390 1390 uint64_t *user, *group;
1391 1391 int flags = dn->dn_id_flags;
1392 1392 int error;
1393 1393 boolean_t have_spill = B_FALSE;
1394 1394
1395 1395 if (!dmu_objset_userused_enabled(dn->dn_objset))
1396 1396 return;
1397 1397
1398 1398 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1399 1399 DN_ID_CHKED_SPILL)))
1400 1400 return;
1401 1401
1402 1402 if (before && dn->dn_bonuslen != 0)
1403 1403 data = DN_BONUS(dn->dn_phys);
1404 1404 else if (!before && dn->dn_bonuslen != 0) {
1405 1405 if (dn->dn_bonus) {
1406 1406 db = dn->dn_bonus;
1407 1407 mutex_enter(&db->db_mtx);
1408 1408 data = dmu_objset_userquota_find_data(db, tx);
1409 1409 } else {
1410 1410 data = DN_BONUS(dn->dn_phys);
1411 1411 }
1412 1412 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1413 1413 int rf = 0;
1414 1414
1415 1415 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1416 1416 rf |= DB_RF_HAVESTRUCT;
1417 1417 error = dmu_spill_hold_by_dnode(dn,
1418 1418 rf | DB_RF_MUST_SUCCEED,
1419 1419 FTAG, (dmu_buf_t **)&db);
1420 1420 ASSERT(error == 0);
1421 1421 mutex_enter(&db->db_mtx);
1422 1422 data = (before) ? db->db.db_data :
1423 1423 dmu_objset_userquota_find_data(db, tx);
1424 1424 have_spill = B_TRUE;
1425 1425 } else {
1426 1426 mutex_enter(&dn->dn_mtx);
1427 1427 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1428 1428 mutex_exit(&dn->dn_mtx);
1429 1429 return;
1430 1430 }
1431 1431
1432 1432 if (before) {
1433 1433 ASSERT(data);
1434 1434 user = &dn->dn_olduid;
1435 1435 group = &dn->dn_oldgid;
1436 1436 } else if (data) {
1437 1437 user = &dn->dn_newuid;
1438 1438 group = &dn->dn_newgid;
1439 1439 }
1440 1440
1441 1441 /*
1442 1442 * Must always call the callback in case the object
1443 1443 * type has changed and that type isn't an object type to track
1444 1444 */
1445 1445 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1446 1446 user, group);
1447 1447
1448 1448 /*
1449 1449 * Preserve existing uid/gid when the callback can't determine
1450 1450 * what the new uid/gid are and the callback returned EEXIST.
1451 1451 * The EEXIST error tells us to just use the existing uid/gid.
1452 1452 * If we don't know what the old values are then just assign
1453 1453 * them to 0, since that is a new file being created.
1454 1454 */
1455 1455 if (!before && data == NULL && error == EEXIST) {
1456 1456 if (flags & DN_ID_OLD_EXIST) {
1457 1457 dn->dn_newuid = dn->dn_olduid;
1458 1458 dn->dn_newgid = dn->dn_oldgid;
1459 1459 } else {
1460 1460 dn->dn_newuid = 0;
1461 1461 dn->dn_newgid = 0;
1462 1462 }
1463 1463 error = 0;
1464 1464 }
1465 1465
1466 1466 if (db)
1467 1467 mutex_exit(&db->db_mtx);
1468 1468
1469 1469 mutex_enter(&dn->dn_mtx);
1470 1470 if (error == 0 && before)
1471 1471 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1472 1472 if (error == 0 && !before)
1473 1473 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1474 1474
1475 1475 if (have_spill) {
1476 1476 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1477 1477 } else {
1478 1478 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1479 1479 }
1480 1480 mutex_exit(&dn->dn_mtx);
1481 1481 if (have_spill)
1482 1482 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1483 1483 }
1484 1484
1485 1485 boolean_t
1486 1486 dmu_objset_userspace_present(objset_t *os)
1487 1487 {
1488 1488 return (os->os_phys->os_flags &
1489 1489 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1490 1490 }
1491 1491
1492 1492 int
1493 1493 dmu_objset_userspace_upgrade(objset_t *os)
1494 1494 {
1495 1495 uint64_t obj;
1496 1496 int err = 0;
1497 1497
1498 1498 if (dmu_objset_userspace_present(os))
1499 1499 return (0);
1500 1500 if (!dmu_objset_userused_enabled(os))
1501 1501 return (ENOTSUP);
1502 1502 if (dmu_objset_is_snapshot(os))
1503 1503 return (EINVAL);
1504 1504
1505 1505 /*
1506 1506 * We simply need to mark every object dirty, so that it will be
1507 1507 * synced out and now accounted. If this is called
1508 1508 * concurrently, or if we already did some work before crashing,
1509 1509 * that's fine, since we track each object's accounted state
1510 1510 * independently.
1511 1511 */
1512 1512
1513 1513 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1514 1514 dmu_tx_t *tx;
1515 1515 dmu_buf_t *db;
1516 1516 int objerr;
1517 1517
1518 1518 if (issig(JUSTLOOKING) && issig(FORREAL))
1519 1519 return (EINTR);
1520 1520
1521 1521 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1522 1522 if (objerr)
1523 1523 continue;
1524 1524 tx = dmu_tx_create(os);
1525 1525 dmu_tx_hold_bonus(tx, obj);
1526 1526 objerr = dmu_tx_assign(tx, TXG_WAIT);
1527 1527 if (objerr) {
1528 1528 dmu_tx_abort(tx);
1529 1529 continue;
1530 1530 }
1531 1531 dmu_buf_will_dirty(db, tx);
1532 1532 dmu_buf_rele(db, FTAG);
1533 1533 dmu_tx_commit(tx);
1534 1534 }
1535 1535
1536 1536 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1537 1537 txg_wait_synced(dmu_objset_pool(os), 0);
1538 1538 return (0);
1539 1539 }
1540 1540
1541 1541 void
1542 1542 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1543 1543 uint64_t *usedobjsp, uint64_t *availobjsp)
1544 1544 {
1545 1545 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1546 1546 usedobjsp, availobjsp);
1547 1547 }
1548 1548
1549 1549 uint64_t
1550 1550 dmu_objset_fsid_guid(objset_t *os)
1551 1551 {
1552 1552 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1553 1553 }
1554 1554
1555 1555 void
1556 1556 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1557 1557 {
1558 1558 stat->dds_type = os->os_phys->os_type;
1559 1559 if (os->os_dsl_dataset)
1560 1560 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1561 1561 }
1562 1562
1563 1563 void
1564 1564 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1565 1565 {
1566 1566 ASSERT(os->os_dsl_dataset ||
1567 1567 os->os_phys->os_type == DMU_OST_META);
1568 1568
1569 1569 if (os->os_dsl_dataset != NULL)
1570 1570 dsl_dataset_stats(os->os_dsl_dataset, nv);
1571 1571
1572 1572 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1573 1573 os->os_phys->os_type);
1574 1574 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1575 1575 dmu_objset_userspace_present(os));
1576 1576 }
1577 1577
1578 1578 int
1579 1579 dmu_objset_is_snapshot(objset_t *os)
1580 1580 {
1581 1581 if (os->os_dsl_dataset != NULL)
1582 1582 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1583 1583 else
1584 1584 return (B_FALSE);
1585 1585 }
1586 1586
1587 1587 int
1588 1588 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1589 1589 boolean_t *conflict)
1590 1590 {
1591 1591 dsl_dataset_t *ds = os->os_dsl_dataset;
1592 1592 uint64_t ignored;
1593 1593
1594 1594 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1595 1595 return (ENOENT);
1596 1596
1597 1597 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1598 1598 ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1599 1599 real, maxlen, conflict));
1600 1600 }
1601 1601
1602 1602 int
1603 1603 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1604 1604 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1605 1605 {
1606 1606 dsl_dataset_t *ds = os->os_dsl_dataset;
1607 1607 zap_cursor_t cursor;
1608 1608 zap_attribute_t attr;
1609 1609
1610 1610 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1611 1611 return (ENOENT);
1612 1612
1613 1613 zap_cursor_init_serialized(&cursor,
1614 1614 ds->ds_dir->dd_pool->dp_meta_objset,
1615 1615 ds->ds_phys->ds_snapnames_zapobj, *offp);
1616 1616
1617 1617 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1618 1618 zap_cursor_fini(&cursor);
1619 1619 return (ENOENT);
1620 1620 }
1621 1621
1622 1622 if (strlen(attr.za_name) + 1 > namelen) {
1623 1623 zap_cursor_fini(&cursor);
1624 1624 return (ENAMETOOLONG);
1625 1625 }
1626 1626
1627 1627 (void) strcpy(name, attr.za_name);
1628 1628 if (idp)
1629 1629 *idp = attr.za_first_integer;
1630 1630 if (case_conflict)
1631 1631 *case_conflict = attr.za_normalization_conflict;
1632 1632 zap_cursor_advance(&cursor);
1633 1633 *offp = zap_cursor_serialize(&cursor);
1634 1634 zap_cursor_fini(&cursor);
1635 1635
1636 1636 return (0);
1637 1637 }
1638 1638
1639 1639 int
1640 1640 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1641 1641 uint64_t *idp, uint64_t *offp)
1642 1642 {
1643 1643 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1644 1644 zap_cursor_t cursor;
1645 1645 zap_attribute_t attr;
1646 1646
1647 1647 /* there is no next dir on a snapshot! */
1648 1648 if (os->os_dsl_dataset->ds_object !=
1649 1649 dd->dd_phys->dd_head_dataset_obj)
1650 1650 return (ENOENT);
1651 1651
1652 1652 zap_cursor_init_serialized(&cursor,
1653 1653 dd->dd_pool->dp_meta_objset,
1654 1654 dd->dd_phys->dd_child_dir_zapobj, *offp);
1655 1655
1656 1656 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1657 1657 zap_cursor_fini(&cursor);
1658 1658 return (ENOENT);
1659 1659 }
1660 1660
1661 1661 if (strlen(attr.za_name) + 1 > namelen) {
1662 1662 zap_cursor_fini(&cursor);
1663 1663 return (ENAMETOOLONG);
1664 1664 }
1665 1665
1666 1666 (void) strcpy(name, attr.za_name);
1667 1667 if (idp)
1668 1668 *idp = attr.za_first_integer;
1669 1669 zap_cursor_advance(&cursor);
1670 1670 *offp = zap_cursor_serialize(&cursor);
1671 1671 zap_cursor_fini(&cursor);
1672 1672
1673 1673 return (0);
1674 1674 }
1675 1675
1676 1676 struct findarg {
1677 1677 int (*func)(const char *, void *);
1678 1678 void *arg;
1679 1679 };
1680 1680
1681 1681 /* ARGSUSED */
1682 1682 static int
1683 1683 findfunc(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
1684 1684 {
1685 1685 struct findarg *fa = arg;
1686 1686 return (fa->func(dsname, fa->arg));
1687 1687 }
1688 1688
1689 1689 /*
1690 1690 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1691 1691 * Perhaps change all callers to use dmu_objset_find_spa()?
1692 1692 */
1693 1693 int
1694 1694 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1695 1695 int flags)
1696 1696 {
1697 1697 struct findarg fa;
1698 1698 fa.func = func;
1699 1699 fa.arg = arg;
1700 1700 return (dmu_objset_find_spa(NULL, name, findfunc, &fa, flags));
1701 1701 }
1702 1702
1703 1703 /*
1704 1704 * Find all objsets under name, call func on each
1705 1705 */
1706 1706 int
1707 1707 dmu_objset_find_spa(spa_t *spa, const char *name,
1708 1708 int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
1709 1709 {
1710 1710 dsl_dir_t *dd;
1711 1711 dsl_pool_t *dp;
1712 1712 dsl_dataset_t *ds;
1713 1713 zap_cursor_t zc;
1714 1714 zap_attribute_t *attr;
1715 1715 char *child;
1716 1716 uint64_t thisobj;
1717 1717 int err;
1718 1718
1719 1719 if (name == NULL)
1720 1720 name = spa_name(spa);
1721 1721 err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
1722 1722 if (err)
1723 1723 return (err);
1724 1724
1725 1725 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1726 1726 if (dd->dd_myname[0] == '$') {
1727 1727 dsl_dir_close(dd, FTAG);
1728 1728 return (0);
1729 1729 }
1730 1730
1731 1731 thisobj = dd->dd_phys->dd_head_dataset_obj;
1732 1732 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1733 1733 dp = dd->dd_pool;
1734 1734
1735 1735 /*
1736 1736 * Iterate over all children.
1737 1737 */
1738 1738 if (flags & DS_FIND_CHILDREN) {
1739 1739 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1740 1740 dd->dd_phys->dd_child_dir_zapobj);
1741 1741 zap_cursor_retrieve(&zc, attr) == 0;
1742 1742 (void) zap_cursor_advance(&zc)) {
1743 1743 ASSERT(attr->za_integer_length == sizeof (uint64_t));
1744 1744 ASSERT(attr->za_num_integers == 1);
1745 1745
1746 1746 child = kmem_asprintf("%s/%s", name, attr->za_name);
1747 1747 err = dmu_objset_find_spa(spa, child, func, arg, flags);
1748 1748 strfree(child);
1749 1749 if (err)
1750 1750 break;
1751 1751 }
1752 1752 zap_cursor_fini(&zc);
1753 1753
1754 1754 if (err) {
1755 1755 dsl_dir_close(dd, FTAG);
1756 1756 kmem_free(attr, sizeof (zap_attribute_t));
1757 1757 return (err);
1758 1758 }
1759 1759 }
1760 1760
1761 1761 /*
1762 1762 * Iterate over all snapshots.
1763 1763 */
1764 1764 if (flags & DS_FIND_SNAPSHOTS) {
1765 1765 if (!dsl_pool_sync_context(dp))
1766 1766 rw_enter(&dp->dp_config_rwlock, RW_READER);
1767 1767 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1768 1768 if (!dsl_pool_sync_context(dp))
1769 1769 rw_exit(&dp->dp_config_rwlock);
1770 1770
1771 1771 if (err == 0) {
1772 1772 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1773 1773 dsl_dataset_rele(ds, FTAG);
1774 1774
1775 1775 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1776 1776 zap_cursor_retrieve(&zc, attr) == 0;
1777 1777 (void) zap_cursor_advance(&zc)) {
1778 1778 ASSERT(attr->za_integer_length ==
1779 1779 sizeof (uint64_t));
1780 1780 ASSERT(attr->za_num_integers == 1);
1781 1781
1782 1782 child = kmem_asprintf("%s@%s",
1783 1783 name, attr->za_name);
1784 1784 err = func(spa, attr->za_first_integer,
1785 1785 child, arg);
1786 1786 strfree(child);
1787 1787 if (err)
1788 1788 break;
1789 1789 }
1790 1790 zap_cursor_fini(&zc);
1791 1791 }
1792 1792 }
1793 1793
1794 1794 dsl_dir_close(dd, FTAG);
1795 1795 kmem_free(attr, sizeof (zap_attribute_t));
1796 1796
1797 1797 if (err)
1798 1798 return (err);
1799 1799
1800 1800 /*
1801 1801 * Apply to self if appropriate.
1802 1802 */
1803 1803 err = func(spa, thisobj, name, arg);
1804 1804 return (err);
1805 1805 }
1806 1806
1807 1807 /* ARGSUSED */
1808 1808 int
1809 1809 dmu_objset_prefetch(const char *name, void *arg)
1810 1810 {
1811 1811 dsl_dataset_t *ds;
1812 1812
1813 1813 if (dsl_dataset_hold(name, FTAG, &ds))
1814 1814 return (0);
1815 1815
1816 1816 if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) {
1817 1817 mutex_enter(&ds->ds_opening_lock);
1818 1818 if (ds->ds_objset == NULL) {
1819 1819 uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
1820 1820 zbookmark_t zb;
1821 1821
1822 1822 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
1823 1823 ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1824 1824
1825 1825 (void) dsl_read_nolock(NULL, dsl_dataset_get_spa(ds),
1826 1826 &ds->ds_phys->ds_bp, NULL, NULL,
1827 1827 ZIO_PRIORITY_ASYNC_READ,
1828 1828 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
1829 1829 &aflags, &zb);
1830 1830 }
1831 1831 mutex_exit(&ds->ds_opening_lock);
1832 1832 }
1833 1833
1834 1834 dsl_dataset_rele(ds, FTAG);
1835 1835 return (0);
1836 1836 }
1837 1837
1838 1838 void
1839 1839 dmu_objset_set_user(objset_t *os, void *user_ptr)
1840 1840 {
1841 1841 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1842 1842 os->os_user_ptr = user_ptr;
1843 1843 }
1844 1844
1845 1845 void *
1846 1846 dmu_objset_get_user(objset_t *os)
1847 1847 {
1848 1848 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1849 1849 return (os->os_user_ptr);
1850 1850 }
|
↓ open down ↓ |
566 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX