Print this page
OS-1566 filesystem limits for ZFS datasets
| 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
|
↓ open down ↓ |
13 lines elided |
↑ open up ↑ |
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 + * Copyright (c) 2012, Joyent, Inc. All rights reserved.
24 25 */
25 26
26 27 /* Portions Copyright 2010 Robert Milkowski */
27 28
28 29 #include <sys/cred.h>
29 30 #include <sys/zfs_context.h>
30 31 #include <sys/dmu_objset.h>
31 32 #include <sys/dsl_dir.h>
32 33 #include <sys/dsl_dataset.h>
33 34 #include <sys/dsl_prop.h>
34 35 #include <sys/dsl_pool.h>
35 36 #include <sys/dsl_synctask.h>
36 37 #include <sys/dsl_deleg.h>
37 38 #include <sys/dnode.h>
38 39 #include <sys/dbuf.h>
39 40 #include <sys/zvol.h>
40 41 #include <sys/dmu_tx.h>
41 42 #include <sys/zap.h>
42 43 #include <sys/zil.h>
43 44 #include <sys/dmu_impl.h>
44 45 #include <sys/zfs_ioctl.h>
45 46 #include <sys/sa.h>
46 47 #include <sys/zfs_onexit.h>
47 48
48 49 /*
49 50 * Needed to close a window in dnode_move() that allows the objset to be freed
50 51 * before it can be safely accessed.
51 52 */
52 53 krwlock_t os_lock;
53 54
54 55 void
55 56 dmu_objset_init(void)
56 57 {
57 58 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
58 59 }
59 60
60 61 void
61 62 dmu_objset_fini(void)
62 63 {
63 64 rw_destroy(&os_lock);
64 65 }
65 66
66 67 spa_t *
67 68 dmu_objset_spa(objset_t *os)
68 69 {
69 70 return (os->os_spa);
70 71 }
71 72
72 73 zilog_t *
73 74 dmu_objset_zil(objset_t *os)
74 75 {
75 76 return (os->os_zil);
76 77 }
77 78
78 79 dsl_pool_t *
79 80 dmu_objset_pool(objset_t *os)
80 81 {
81 82 dsl_dataset_t *ds;
82 83
83 84 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
84 85 return (ds->ds_dir->dd_pool);
85 86 else
86 87 return (spa_get_dsl(os->os_spa));
87 88 }
88 89
89 90 dsl_dataset_t *
90 91 dmu_objset_ds(objset_t *os)
91 92 {
92 93 return (os->os_dsl_dataset);
93 94 }
94 95
95 96 dmu_objset_type_t
96 97 dmu_objset_type(objset_t *os)
97 98 {
98 99 return (os->os_phys->os_type);
99 100 }
100 101
101 102 void
102 103 dmu_objset_name(objset_t *os, char *buf)
103 104 {
104 105 dsl_dataset_name(os->os_dsl_dataset, buf);
105 106 }
106 107
107 108 uint64_t
108 109 dmu_objset_id(objset_t *os)
109 110 {
110 111 dsl_dataset_t *ds = os->os_dsl_dataset;
111 112
112 113 return (ds ? ds->ds_object : 0);
113 114 }
114 115
115 116 uint64_t
116 117 dmu_objset_syncprop(objset_t *os)
117 118 {
118 119 return (os->os_sync);
119 120 }
120 121
121 122 uint64_t
122 123 dmu_objset_logbias(objset_t *os)
123 124 {
124 125 return (os->os_logbias);
125 126 }
126 127
127 128 static void
128 129 checksum_changed_cb(void *arg, uint64_t newval)
129 130 {
130 131 objset_t *os = arg;
131 132
132 133 /*
133 134 * Inheritance should have been done by now.
134 135 */
135 136 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
136 137
137 138 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
138 139 }
139 140
140 141 static void
141 142 compression_changed_cb(void *arg, uint64_t newval)
142 143 {
143 144 objset_t *os = arg;
144 145
145 146 /*
146 147 * Inheritance and range checking should have been done by now.
147 148 */
148 149 ASSERT(newval != ZIO_COMPRESS_INHERIT);
149 150
150 151 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
151 152 }
152 153
153 154 static void
154 155 copies_changed_cb(void *arg, uint64_t newval)
155 156 {
156 157 objset_t *os = arg;
157 158
158 159 /*
159 160 * Inheritance and range checking should have been done by now.
160 161 */
161 162 ASSERT(newval > 0);
162 163 ASSERT(newval <= spa_max_replication(os->os_spa));
163 164
164 165 os->os_copies = newval;
165 166 }
166 167
167 168 static void
168 169 dedup_changed_cb(void *arg, uint64_t newval)
169 170 {
170 171 objset_t *os = arg;
171 172 spa_t *spa = os->os_spa;
172 173 enum zio_checksum checksum;
173 174
174 175 /*
175 176 * Inheritance should have been done by now.
176 177 */
177 178 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
178 179
179 180 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
180 181
181 182 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
182 183 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
183 184 }
184 185
185 186 static void
186 187 primary_cache_changed_cb(void *arg, uint64_t newval)
187 188 {
188 189 objset_t *os = arg;
189 190
190 191 /*
191 192 * Inheritance and range checking should have been done by now.
192 193 */
193 194 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
194 195 newval == ZFS_CACHE_METADATA);
195 196
196 197 os->os_primary_cache = newval;
197 198 }
198 199
199 200 static void
200 201 secondary_cache_changed_cb(void *arg, uint64_t newval)
201 202 {
202 203 objset_t *os = arg;
203 204
204 205 /*
205 206 * Inheritance and range checking should have been done by now.
206 207 */
207 208 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
208 209 newval == ZFS_CACHE_METADATA);
209 210
210 211 os->os_secondary_cache = newval;
211 212 }
212 213
213 214 static void
214 215 sync_changed_cb(void *arg, uint64_t newval)
215 216 {
216 217 objset_t *os = arg;
217 218
218 219 /*
219 220 * Inheritance and range checking should have been done by now.
220 221 */
221 222 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
222 223 newval == ZFS_SYNC_DISABLED);
223 224
224 225 os->os_sync = newval;
225 226 if (os->os_zil)
226 227 zil_set_sync(os->os_zil, newval);
227 228 }
228 229
229 230 static void
230 231 logbias_changed_cb(void *arg, uint64_t newval)
231 232 {
232 233 objset_t *os = arg;
233 234
234 235 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
235 236 newval == ZFS_LOGBIAS_THROUGHPUT);
236 237 os->os_logbias = newval;
237 238 if (os->os_zil)
238 239 zil_set_logbias(os->os_zil, newval);
239 240 }
240 241
241 242 void
242 243 dmu_objset_byteswap(void *buf, size_t size)
243 244 {
244 245 objset_phys_t *osp = buf;
245 246
246 247 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
247 248 dnode_byteswap(&osp->os_meta_dnode);
248 249 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
249 250 osp->os_type = BSWAP_64(osp->os_type);
250 251 osp->os_flags = BSWAP_64(osp->os_flags);
251 252 if (size == sizeof (objset_phys_t)) {
252 253 dnode_byteswap(&osp->os_userused_dnode);
253 254 dnode_byteswap(&osp->os_groupused_dnode);
254 255 }
255 256 }
256 257
257 258 int
258 259 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
259 260 objset_t **osp)
260 261 {
261 262 objset_t *os;
262 263 int i, err;
263 264
264 265 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
265 266
266 267 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
267 268 os->os_dsl_dataset = ds;
268 269 os->os_spa = spa;
269 270 os->os_rootbp = bp;
270 271 if (!BP_IS_HOLE(os->os_rootbp)) {
271 272 uint32_t aflags = ARC_WAIT;
272 273 zbookmark_t zb;
273 274 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
274 275 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
275 276
276 277 if (DMU_OS_IS_L2CACHEABLE(os))
277 278 aflags |= ARC_L2CACHE;
278 279
279 280 dprintf_bp(os->os_rootbp, "reading %s", "");
280 281 /*
281 282 * XXX when bprewrite scrub can change the bp,
282 283 * and this is called from dmu_objset_open_ds_os, the bp
283 284 * could change, and we'll need a lock.
284 285 */
285 286 err = dsl_read_nolock(NULL, spa, os->os_rootbp,
286 287 arc_getbuf_func, &os->os_phys_buf,
287 288 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
288 289 if (err) {
289 290 kmem_free(os, sizeof (objset_t));
290 291 /* convert checksum errors into IO errors */
291 292 if (err == ECKSUM)
292 293 err = EIO;
293 294 return (err);
294 295 }
295 296
296 297 /* Increase the blocksize if we are permitted. */
297 298 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
298 299 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
299 300 arc_buf_t *buf = arc_buf_alloc(spa,
300 301 sizeof (objset_phys_t), &os->os_phys_buf,
301 302 ARC_BUFC_METADATA);
302 303 bzero(buf->b_data, sizeof (objset_phys_t));
303 304 bcopy(os->os_phys_buf->b_data, buf->b_data,
304 305 arc_buf_size(os->os_phys_buf));
305 306 (void) arc_buf_remove_ref(os->os_phys_buf,
306 307 &os->os_phys_buf);
307 308 os->os_phys_buf = buf;
308 309 }
309 310
310 311 os->os_phys = os->os_phys_buf->b_data;
311 312 os->os_flags = os->os_phys->os_flags;
312 313 } else {
313 314 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
314 315 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
315 316 os->os_phys_buf = arc_buf_alloc(spa, size,
316 317 &os->os_phys_buf, ARC_BUFC_METADATA);
317 318 os->os_phys = os->os_phys_buf->b_data;
318 319 bzero(os->os_phys, size);
319 320 }
320 321
321 322 /*
322 323 * Note: the changed_cb will be called once before the register
323 324 * func returns, thus changing the checksum/compression from the
324 325 * default (fletcher2/off). Snapshots don't need to know about
325 326 * checksum/compression/copies.
326 327 */
327 328 if (ds) {
328 329 err = dsl_prop_register(ds, "primarycache",
329 330 primary_cache_changed_cb, os);
330 331 if (err == 0)
331 332 err = dsl_prop_register(ds, "secondarycache",
332 333 secondary_cache_changed_cb, os);
333 334 if (!dsl_dataset_is_snapshot(ds)) {
334 335 if (err == 0)
335 336 err = dsl_prop_register(ds, "checksum",
336 337 checksum_changed_cb, os);
337 338 if (err == 0)
338 339 err = dsl_prop_register(ds, "compression",
339 340 compression_changed_cb, os);
340 341 if (err == 0)
341 342 err = dsl_prop_register(ds, "copies",
342 343 copies_changed_cb, os);
343 344 if (err == 0)
344 345 err = dsl_prop_register(ds, "dedup",
345 346 dedup_changed_cb, os);
346 347 if (err == 0)
347 348 err = dsl_prop_register(ds, "logbias",
348 349 logbias_changed_cb, os);
349 350 if (err == 0)
350 351 err = dsl_prop_register(ds, "sync",
351 352 sync_changed_cb, os);
352 353 }
353 354 if (err) {
354 355 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
355 356 &os->os_phys_buf) == 1);
356 357 kmem_free(os, sizeof (objset_t));
357 358 return (err);
358 359 }
359 360 } else if (ds == NULL) {
360 361 /* It's the meta-objset. */
361 362 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
362 363 os->os_compress = ZIO_COMPRESS_LZJB;
363 364 os->os_copies = spa_max_replication(spa);
364 365 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
365 366 os->os_dedup_verify = 0;
366 367 os->os_logbias = 0;
367 368 os->os_sync = 0;
368 369 os->os_primary_cache = ZFS_CACHE_ALL;
369 370 os->os_secondary_cache = ZFS_CACHE_ALL;
370 371 }
371 372
372 373 if (ds == NULL || !dsl_dataset_is_snapshot(ds))
373 374 os->os_zil_header = os->os_phys->os_zil_header;
374 375 os->os_zil = zil_alloc(os, &os->os_zil_header);
375 376
376 377 for (i = 0; i < TXG_SIZE; i++) {
377 378 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
378 379 offsetof(dnode_t, dn_dirty_link[i]));
379 380 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
380 381 offsetof(dnode_t, dn_dirty_link[i]));
381 382 }
382 383 list_create(&os->os_dnodes, sizeof (dnode_t),
383 384 offsetof(dnode_t, dn_link));
384 385 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
385 386 offsetof(dmu_buf_impl_t, db_link));
386 387
387 388 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
388 389 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
389 390 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
390 391
391 392 DMU_META_DNODE(os) = dnode_special_open(os,
392 393 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
393 394 &os->os_meta_dnode);
394 395 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
395 396 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
396 397 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
397 398 &os->os_userused_dnode);
398 399 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
399 400 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
400 401 &os->os_groupused_dnode);
401 402 }
402 403
403 404 /*
404 405 * We should be the only thread trying to do this because we
405 406 * have ds_opening_lock
406 407 */
407 408 if (ds) {
408 409 mutex_enter(&ds->ds_lock);
409 410 ASSERT(ds->ds_objset == NULL);
410 411 ds->ds_objset = os;
411 412 mutex_exit(&ds->ds_lock);
412 413 }
413 414
414 415 *osp = os;
415 416 return (0);
416 417 }
417 418
418 419 int
419 420 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
420 421 {
421 422 int err = 0;
422 423
423 424 mutex_enter(&ds->ds_opening_lock);
424 425 *osp = ds->ds_objset;
425 426 if (*osp == NULL) {
426 427 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
427 428 ds, dsl_dataset_get_blkptr(ds), osp);
428 429 }
429 430 mutex_exit(&ds->ds_opening_lock);
430 431 return (err);
431 432 }
432 433
433 434 /* called from zpl */
434 435 int
435 436 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
436 437 {
437 438 dsl_dataset_t *ds;
438 439 int err;
439 440
440 441 err = dsl_dataset_hold(name, tag, &ds);
441 442 if (err)
442 443 return (err);
443 444
444 445 err = dmu_objset_from_ds(ds, osp);
445 446 if (err)
446 447 dsl_dataset_rele(ds, tag);
447 448
448 449 return (err);
449 450 }
450 451
451 452 /* called from zpl */
452 453 int
453 454 dmu_objset_own(const char *name, dmu_objset_type_t type,
454 455 boolean_t readonly, void *tag, objset_t **osp)
455 456 {
456 457 dsl_dataset_t *ds;
457 458 int err;
458 459
459 460 err = dsl_dataset_own(name, B_FALSE, tag, &ds);
460 461 if (err)
461 462 return (err);
462 463
463 464 err = dmu_objset_from_ds(ds, osp);
464 465 if (err) {
465 466 dsl_dataset_disown(ds, tag);
466 467 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
467 468 dmu_objset_disown(*osp, tag);
468 469 return (EINVAL);
469 470 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
470 471 dmu_objset_disown(*osp, tag);
471 472 return (EROFS);
472 473 }
473 474 return (err);
474 475 }
475 476
476 477 void
477 478 dmu_objset_rele(objset_t *os, void *tag)
478 479 {
479 480 dsl_dataset_rele(os->os_dsl_dataset, tag);
480 481 }
481 482
482 483 void
483 484 dmu_objset_disown(objset_t *os, void *tag)
484 485 {
485 486 dsl_dataset_disown(os->os_dsl_dataset, tag);
486 487 }
487 488
488 489 int
489 490 dmu_objset_evict_dbufs(objset_t *os)
490 491 {
491 492 dnode_t *dn;
492 493
493 494 mutex_enter(&os->os_lock);
494 495
495 496 /* process the mdn last, since the other dnodes have holds on it */
496 497 list_remove(&os->os_dnodes, DMU_META_DNODE(os));
497 498 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
498 499
499 500 /*
500 501 * Find the first dnode with holds. We have to do this dance
501 502 * because dnode_add_ref() only works if you already have a
502 503 * hold. If there are no holds then it has no dbufs so OK to
503 504 * skip.
504 505 */
505 506 for (dn = list_head(&os->os_dnodes);
506 507 dn && !dnode_add_ref(dn, FTAG);
507 508 dn = list_next(&os->os_dnodes, dn))
508 509 continue;
509 510
510 511 while (dn) {
511 512 dnode_t *next_dn = dn;
512 513
513 514 do {
514 515 next_dn = list_next(&os->os_dnodes, next_dn);
515 516 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
516 517
517 518 mutex_exit(&os->os_lock);
518 519 dnode_evict_dbufs(dn);
519 520 dnode_rele(dn, FTAG);
520 521 mutex_enter(&os->os_lock);
521 522 dn = next_dn;
522 523 }
523 524 dn = list_head(&os->os_dnodes);
524 525 mutex_exit(&os->os_lock);
525 526 return (dn != DMU_META_DNODE(os));
526 527 }
527 528
528 529 void
529 530 dmu_objset_evict(objset_t *os)
530 531 {
531 532 dsl_dataset_t *ds = os->os_dsl_dataset;
532 533
533 534 for (int t = 0; t < TXG_SIZE; t++)
534 535 ASSERT(!dmu_objset_is_dirty(os, t));
535 536
536 537 if (ds) {
537 538 if (!dsl_dataset_is_snapshot(ds)) {
538 539 VERIFY(0 == dsl_prop_unregister(ds, "checksum",
539 540 checksum_changed_cb, os));
540 541 VERIFY(0 == dsl_prop_unregister(ds, "compression",
541 542 compression_changed_cb, os));
542 543 VERIFY(0 == dsl_prop_unregister(ds, "copies",
543 544 copies_changed_cb, os));
544 545 VERIFY(0 == dsl_prop_unregister(ds, "dedup",
545 546 dedup_changed_cb, os));
546 547 VERIFY(0 == dsl_prop_unregister(ds, "logbias",
547 548 logbias_changed_cb, os));
548 549 VERIFY(0 == dsl_prop_unregister(ds, "sync",
549 550 sync_changed_cb, os));
550 551 }
551 552 VERIFY(0 == dsl_prop_unregister(ds, "primarycache",
552 553 primary_cache_changed_cb, os));
553 554 VERIFY(0 == dsl_prop_unregister(ds, "secondarycache",
554 555 secondary_cache_changed_cb, os));
555 556 }
556 557
557 558 if (os->os_sa)
558 559 sa_tear_down(os);
559 560
560 561 /*
561 562 * We should need only a single pass over the dnode list, since
562 563 * nothing can be added to the list at this point.
563 564 */
564 565 (void) dmu_objset_evict_dbufs(os);
565 566
566 567 dnode_special_close(&os->os_meta_dnode);
567 568 if (DMU_USERUSED_DNODE(os)) {
568 569 dnode_special_close(&os->os_userused_dnode);
569 570 dnode_special_close(&os->os_groupused_dnode);
570 571 }
571 572 zil_free(os->os_zil);
572 573
573 574 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
574 575
575 576 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf) == 1);
576 577
577 578 /*
578 579 * This is a barrier to prevent the objset from going away in
579 580 * dnode_move() until we can safely ensure that the objset is still in
580 581 * use. We consider the objset valid before the barrier and invalid
581 582 * after the barrier.
582 583 */
583 584 rw_enter(&os_lock, RW_READER);
584 585 rw_exit(&os_lock);
585 586
586 587 mutex_destroy(&os->os_lock);
587 588 mutex_destroy(&os->os_obj_lock);
588 589 mutex_destroy(&os->os_user_ptr_lock);
589 590 kmem_free(os, sizeof (objset_t));
590 591 }
591 592
592 593 timestruc_t
593 594 dmu_objset_snap_cmtime(objset_t *os)
594 595 {
595 596 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
596 597 }
597 598
598 599 /* called from dsl for meta-objset */
599 600 objset_t *
600 601 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
601 602 dmu_objset_type_t type, dmu_tx_t *tx)
602 603 {
603 604 objset_t *os;
604 605 dnode_t *mdn;
605 606
606 607 ASSERT(dmu_tx_is_syncing(tx));
607 608 if (ds != NULL)
608 609 VERIFY(0 == dmu_objset_from_ds(ds, &os));
609 610 else
610 611 VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os));
611 612
612 613 mdn = DMU_META_DNODE(os);
613 614
614 615 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
615 616 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
616 617
617 618 /*
618 619 * We don't want to have to increase the meta-dnode's nlevels
619 620 * later, because then we could do it in quescing context while
620 621 * we are also accessing it in open context.
621 622 *
622 623 * This precaution is not necessary for the MOS (ds == NULL),
623 624 * because the MOS is only updated in syncing context.
624 625 * This is most fortunate: the MOS is the only objset that
625 626 * needs to be synced multiple times as spa_sync() iterates
626 627 * to convergence, so minimizing its dn_nlevels matters.
627 628 */
628 629 if (ds != NULL) {
629 630 int levels = 1;
630 631
631 632 /*
632 633 * Determine the number of levels necessary for the meta-dnode
633 634 * to contain DN_MAX_OBJECT dnodes.
634 635 */
635 636 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
636 637 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
637 638 DN_MAX_OBJECT * sizeof (dnode_phys_t))
638 639 levels++;
639 640
640 641 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
641 642 mdn->dn_nlevels = levels;
642 643 }
643 644
644 645 ASSERT(type != DMU_OST_NONE);
645 646 ASSERT(type != DMU_OST_ANY);
646 647 ASSERT(type < DMU_OST_NUMTYPES);
647 648 os->os_phys->os_type = type;
648 649 if (dmu_objset_userused_enabled(os)) {
649 650 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
650 651 os->os_flags = os->os_phys->os_flags;
651 652 }
652 653
653 654 dsl_dataset_dirty(ds, tx);
654 655
655 656 return (os);
656 657 }
657 658
658 659 struct oscarg {
659 660 void (*userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
660 661 void *userarg;
661 662 dsl_dataset_t *clone_origin;
662 663 const char *lastname;
663 664 dmu_objset_type_t type;
664 665 uint64_t flags;
665 666 cred_t *cr;
666 667 };
667 668
668 669 /*ARGSUSED*/
669 670 static int
670 671 dmu_objset_create_check(void *arg1, void *arg2, dmu_tx_t *tx)
671 672 {
672 673 dsl_dir_t *dd = arg1;
673 674 struct oscarg *oa = arg2;
674 675 objset_t *mos = dd->dd_pool->dp_meta_objset;
675 676 int err;
676 677 uint64_t ddobj;
677 678
678 679 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
679 680 oa->lastname, sizeof (uint64_t), 1, &ddobj);
680 681 if (err != ENOENT)
681 682 return (err ? err : EEXIST);
682 683
|
↓ open down ↓ |
649 lines elided |
↑ open up ↑ |
683 684 if (oa->clone_origin != NULL) {
684 685 /* You can't clone across pools. */
685 686 if (oa->clone_origin->ds_dir->dd_pool != dd->dd_pool)
686 687 return (EXDEV);
687 688
688 689 /* You can only clone snapshots, not the head datasets. */
689 690 if (!dsl_dataset_is_snapshot(oa->clone_origin))
690 691 return (EINVAL);
691 692 }
692 693
693 - return (0);
694 + return (dsl_dir_fscount_check(dd, 1, NULL, oa->cr));
694 695 }
695 696
696 697 static void
697 698 dmu_objset_create_sync(void *arg1, void *arg2, dmu_tx_t *tx)
698 699 {
699 700 dsl_dir_t *dd = arg1;
700 701 spa_t *spa = dd->dd_pool->dp_spa;
701 702 struct oscarg *oa = arg2;
702 703 uint64_t obj;
703 704 dsl_dataset_t *ds;
704 705 blkptr_t *bp;
705 706
706 707 ASSERT(dmu_tx_is_syncing(tx));
707 708
709 + dsl_dir_fscount_adjust(dd, tx, 1, B_TRUE);
710 +
708 711 obj = dsl_dataset_create_sync(dd, oa->lastname,
709 712 oa->clone_origin, oa->flags, oa->cr, tx);
710 713
711 714 VERIFY3U(0, ==, dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds));
712 715 bp = dsl_dataset_get_blkptr(ds);
713 716 if (BP_IS_HOLE(bp)) {
714 717 objset_t *os =
715 718 dmu_objset_create_impl(spa, ds, bp, oa->type, tx);
716 719
717 720 if (oa->userfunc)
718 721 oa->userfunc(os, oa->userarg, oa->cr, tx);
719 722 }
720 723
721 724 if (oa->clone_origin == NULL) {
722 725 spa_history_log_internal_ds(ds, "create", tx, "");
723 726 } else {
724 727 char namebuf[MAXNAMELEN];
725 728 dsl_dataset_name(oa->clone_origin, namebuf);
726 729 spa_history_log_internal_ds(ds, "clone", tx,
727 730 "origin=%s (%llu)", namebuf, oa->clone_origin->ds_object);
728 731 }
729 732 dsl_dataset_rele(ds, FTAG);
730 733 }
731 734
732 735 int
733 736 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
734 737 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
735 738 {
736 739 dsl_dir_t *pdd;
737 740 const char *tail;
738 741 int err = 0;
739 742 struct oscarg oa = { 0 };
740 743
741 744 ASSERT(strchr(name, '@') == NULL);
742 745 err = dsl_dir_open(name, FTAG, &pdd, &tail);
743 746 if (err)
744 747 return (err);
745 748 if (tail == NULL) {
746 749 dsl_dir_close(pdd, FTAG);
747 750 return (EEXIST);
748 751 }
749 752
750 753 oa.userfunc = func;
751 754 oa.userarg = arg;
752 755 oa.lastname = tail;
753 756 oa.type = type;
754 757 oa.flags = flags;
755 758 oa.cr = CRED();
756 759
757 760 err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
758 761 dmu_objset_create_sync, pdd, &oa, 5);
759 762 dsl_dir_close(pdd, FTAG);
760 763 return (err);
761 764 }
762 765
763 766 int
764 767 dmu_objset_clone(const char *name, dsl_dataset_t *clone_origin, uint64_t flags)
765 768 {
766 769 dsl_dir_t *pdd;
767 770 const char *tail;
768 771 int err = 0;
769 772 struct oscarg oa = { 0 };
770 773
771 774 ASSERT(strchr(name, '@') == NULL);
772 775 err = dsl_dir_open(name, FTAG, &pdd, &tail);
773 776 if (err)
774 777 return (err);
775 778 if (tail == NULL) {
776 779 dsl_dir_close(pdd, FTAG);
777 780 return (EEXIST);
778 781 }
779 782
780 783 oa.lastname = tail;
781 784 oa.clone_origin = clone_origin;
782 785 oa.flags = flags;
783 786 oa.cr = CRED();
784 787
785 788 err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
786 789 dmu_objset_create_sync, pdd, &oa, 5);
787 790 dsl_dir_close(pdd, FTAG);
788 791 return (err);
789 792 }
790 793
791 794 int
792 795 dmu_objset_destroy(const char *name, boolean_t defer)
793 796 {
794 797 dsl_dataset_t *ds;
795 798 int error;
796 799
797 800 error = dsl_dataset_own(name, B_TRUE, FTAG, &ds);
798 801 if (error == 0) {
799 802 error = dsl_dataset_destroy(ds, FTAG, defer);
|
↓ open down ↓ |
82 lines elided |
↑ open up ↑ |
800 803 /* dsl_dataset_destroy() closes the ds. */
801 804 }
802 805
803 806 return (error);
804 807 }
805 808
806 809 typedef struct snapallarg {
807 810 dsl_sync_task_group_t *saa_dstg;
808 811 boolean_t saa_needsuspend;
809 812 nvlist_t *saa_props;
813 + cred_t *saa_cr;
810 814
811 815 /* the following are used only if 'temporary' is set: */
812 816 boolean_t saa_temporary;
813 817 const char *saa_htag;
814 818 struct dsl_ds_holdarg *saa_ha;
815 819 dsl_dataset_t *saa_newds;
816 820 } snapallarg_t;
817 821
818 822 typedef struct snaponearg {
819 823 const char *soa_longname; /* long snap name */
820 824 const char *soa_snapname; /* short snap name */
825 + uint64_t soa_tot_cnt;
821 826 snapallarg_t *soa_saa;
822 827 } snaponearg_t;
823 828
824 829 static int
825 830 snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
826 831 {
827 832 objset_t *os = arg1;
828 833 snaponearg_t *soa = arg2;
829 834 snapallarg_t *saa = soa->soa_saa;
830 835 int error;
831 836
832 837 /* The props have already been checked by zfs_check_userprops(). */
833 838
834 839 error = dsl_dataset_snapshot_check(os->os_dsl_dataset,
835 - soa->soa_snapname, tx);
840 + soa->soa_snapname, soa->soa_tot_cnt, tx, saa->saa_cr);
836 841 if (error)
837 842 return (error);
838 843
839 844 if (saa->saa_temporary) {
840 845 /*
841 846 * Ideally we would just call
842 847 * dsl_dataset_user_hold_check() and
843 848 * dsl_dataset_destroy_check() here. However the
844 849 * dataset we want to hold and destroy is the snapshot
845 850 * that we just confirmed we can create, but it won't
846 851 * exist until after these checks are run. Do any
847 852 * checks we can here and if more checks are added to
848 853 * those routines in the future, similar checks may be
849 854 * necessary here.
850 855 */
851 856 if (spa_version(os->os_spa) < SPA_VERSION_USERREFS)
852 857 return (ENOTSUP);
853 858 /*
854 859 * Not checking number of tags because the tag will be
855 860 * unique, as it will be the only tag.
856 861 */
857 862 if (strlen(saa->saa_htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
858 863 return (E2BIG);
859 864
860 865 saa->saa_ha = kmem_alloc(sizeof (struct dsl_ds_holdarg),
861 866 KM_SLEEP);
862 867 saa->saa_ha->temphold = B_TRUE;
863 868 saa->saa_ha->htag = saa->saa_htag;
864 869 }
865 870 return (error);
866 871 }
867 872
868 873 static void
869 874 snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
870 875 {
871 876 objset_t *os = arg1;
872 877 dsl_dataset_t *ds = os->os_dsl_dataset;
873 878 snaponearg_t *soa = arg2;
874 879 snapallarg_t *saa = soa->soa_saa;
875 880
876 881 dsl_dataset_snapshot_sync(ds, soa->soa_snapname, tx);
877 882
878 883 if (saa->saa_props != NULL) {
879 884 dsl_props_arg_t pa;
880 885 pa.pa_props = saa->saa_props;
881 886 pa.pa_source = ZPROP_SRC_LOCAL;
882 887 dsl_props_set_sync(ds->ds_prev, &pa, tx);
883 888 }
884 889
885 890 if (saa->saa_temporary) {
886 891 struct dsl_ds_destroyarg da;
887 892
888 893 dsl_dataset_user_hold_sync(ds->ds_prev, saa->saa_ha, tx);
889 894 kmem_free(saa->saa_ha, sizeof (struct dsl_ds_holdarg));
|
↓ open down ↓ |
44 lines elided |
↑ open up ↑ |
890 895 saa->saa_ha = NULL;
891 896 saa->saa_newds = ds->ds_prev;
892 897
893 898 da.ds = ds->ds_prev;
894 899 da.defer = B_TRUE;
895 900 dsl_dataset_destroy_sync(&da, FTAG, tx);
896 901 }
897 902 }
898 903
899 904 static int
900 -snapshot_one_impl(const char *snapname, void *arg)
905 +snapshot_one_impl(const char *snapname, void *arg, uint64_t cnt)
901 906 {
902 907 char fsname[MAXPATHLEN];
903 908 snapallarg_t *saa = arg;
904 909 snaponearg_t *soa;
905 910 objset_t *os;
906 911 int err;
907 912
908 913 (void) strlcpy(fsname, snapname, sizeof (fsname));
909 914 strchr(fsname, '@')[0] = '\0';
910 915
911 916 err = dmu_objset_hold(fsname, saa, &os);
912 917 if (err != 0)
913 918 return (err);
914 919
915 920 /*
916 921 * If the objset is in an inconsistent state (eg, in the process
917 922 * of being destroyed), don't snapshot it.
918 923 */
919 924 if (os->os_dsl_dataset->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) {
920 925 dmu_objset_rele(os, saa);
921 926 return (EBUSY);
922 927 }
923 928
924 929 if (saa->saa_needsuspend) {
925 930 err = zil_suspend(dmu_objset_zil(os));
|
↓ open down ↓ |
15 lines elided |
↑ open up ↑ |
926 931 if (err) {
927 932 dmu_objset_rele(os, saa);
928 933 return (err);
929 934 }
930 935 }
931 936
932 937 soa = kmem_zalloc(sizeof (*soa), KM_SLEEP);
933 938 soa->soa_saa = saa;
934 939 soa->soa_longname = snapname;
935 940 soa->soa_snapname = strchr(snapname, '@') + 1;
941 + soa->soa_tot_cnt = cnt;
936 942
937 943 dsl_sync_task_create(saa->saa_dstg, snapshot_check, snapshot_sync,
938 944 os, soa, 3);
939 945
940 946 return (0);
941 947 }
942 948
943 949 /*
944 950 * The snapshots must all be in the same pool.
945 951 */
946 952 int
947 953 dmu_objset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
948 954 {
949 955 dsl_sync_task_t *dst;
950 956 snapallarg_t saa = { 0 };
951 957 spa_t *spa;
952 958 int rv = 0;
953 959 int err;
954 960 nvpair_t *pair;
961 + nvlist_t *cnt_track = NULL;
962 + char *pdelim;
963 + uint64_t val;
964 + char nm[MAXPATHLEN];
955 965
956 966 pair = nvlist_next_nvpair(snaps, NULL);
957 967 if (pair == NULL)
958 968 return (0);
959 969
960 970 err = spa_open(nvpair_name(pair), &spa, FTAG);
961 971 if (err)
962 972 return (err);
963 973 saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
964 974 saa.saa_props = props;
965 975 saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
976 + saa.saa_cr = CRED();
966 977
978 + /*
979 + * Pre-compute how many total new snapshots will be created for each
980 + * level in the tree and below. This is needed for validating the
981 + * snapshot limit when taking a recursive snapshot.
982 + *
983 + * The problem is that the counts are not actually adjusted when
984 + * we are checking, only when we finally sync. For a single snapshot,
985 + * this is easy, the count will increase by 1 at each node up the tree,
986 + * but its more complicated for recursive snapshots. Since we are
987 + * validating each snapshot independently we need to be sure that we
988 + * are validating the complete count for the entire set of snapshots.
989 + * We do this by rolling up the counts for each component of the name
990 + * into an nvlist then we'll use that count in the validation of each
991 + * individual snapshot.
992 + *
993 + * We validated the snapshot names in zfs_ioc_snapshot so we know they
994 + * have a '@'.
995 + */
996 + cnt_track = fnvlist_alloc();
997 +
967 998 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
968 999 pair = nvlist_next_nvpair(snaps, pair)) {
969 - err = snapshot_one_impl(nvpair_name(pair), &saa);
1000 + (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1001 + pdelim = strchr(nm, '@');
1002 + *pdelim = '\0';
1003 +
1004 + do {
1005 + if (nvlist_lookup_uint64(cnt_track, nm, &val) == 0) {
1006 + /* update existing entry */
1007 + fnvlist_add_uint64(cnt_track, nm, val + 1);
1008 + } else {
1009 + /* add to list */
1010 + fnvlist_add_uint64(cnt_track, nm, 1);
1011 + }
1012 +
1013 + pdelim = strrchr(nm, '/');
1014 + if (pdelim != NULL)
1015 + *pdelim = '\0';
1016 + } while (pdelim != NULL);
1017 + }
1018 +
1019 + /*
1020 + * We've calculated the counts, now validate.
1021 + */
1022 + for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1023 + pair = nvlist_next_nvpair(snaps, pair)) {
1024 + (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1025 + pdelim = strchr(nm, '@');
1026 + *pdelim = '\0';
1027 +
1028 + val = fnvlist_lookup_uint64(cnt_track, nm);
1029 + err = snapshot_one_impl(nvpair_name(pair), &saa, val);
970 1030 if (err != 0) {
971 1031 if (errors != NULL) {
972 1032 fnvlist_add_int32(errors,
973 1033 nvpair_name(pair), err);
974 1034 }
975 1035 rv = err;
976 1036 }
977 1037 }
978 1038
1039 + nvlist_free(cnt_track);
1040 +
979 1041 /*
980 1042 * If any call to snapshot_one_impl() failed, don't execute the
981 1043 * sync task. The error handling code below will clean up the
982 1044 * snaponearg_t from any successful calls to
983 1045 * snapshot_one_impl().
984 1046 */
985 1047 if (rv == 0)
986 1048 err = dsl_sync_task_group_wait(saa.saa_dstg);
987 1049 if (err != 0)
988 1050 rv = err;
989 1051
990 1052 for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
991 1053 dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
992 1054 objset_t *os = dst->dst_arg1;
993 1055 snaponearg_t *soa = dst->dst_arg2;
994 1056 if (dst->dst_err != 0) {
995 1057 if (errors != NULL) {
996 1058 fnvlist_add_int32(errors,
997 1059 soa->soa_longname, dst->dst_err);
998 1060 }
999 1061 rv = dst->dst_err;
1000 1062 }
1001 1063
1002 1064 if (saa.saa_needsuspend)
1003 1065 zil_resume(dmu_objset_zil(os));
1004 1066 dmu_objset_rele(os, &saa);
1005 1067 kmem_free(soa, sizeof (*soa));
1006 1068 }
1007 1069
1008 1070 dsl_sync_task_group_destroy(saa.saa_dstg);
1009 1071 spa_close(spa, FTAG);
1010 1072 return (rv);
1011 1073 }
1012 1074
1013 1075 int
1014 1076 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1015 1077 {
1016 1078 int err;
1017 1079 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1018 1080 nvlist_t *snaps = fnvlist_alloc();
1019 1081
1020 1082 fnvlist_add_boolean(snaps, longsnap);
1021 1083 err = dmu_objset_snapshot(snaps, NULL, NULL);
1022 1084 fnvlist_free(snaps);
1023 1085 strfree(longsnap);
1024 1086 return (err);
1025 1087 }
1026 1088
1027 1089 int
1028 1090 dmu_objset_snapshot_tmp(const char *snapname, const char *tag, int cleanup_fd)
1029 1091 {
1030 1092 dsl_sync_task_t *dst;
1031 1093 snapallarg_t saa = { 0 };
1032 1094 spa_t *spa;
1033 1095 minor_t minor;
1034 1096 int err;
1035 1097
1036 1098 err = spa_open(snapname, &spa, FTAG);
1037 1099 if (err)
1038 1100 return (err);
1039 1101 saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
1040 1102 saa.saa_htag = tag;
1041 1103 saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1042 1104 saa.saa_temporary = B_TRUE;
|
↓ open down ↓ |
54 lines elided |
↑ open up ↑ |
1043 1105
1044 1106 if (cleanup_fd < 0) {
1045 1107 spa_close(spa, FTAG);
1046 1108 return (EINVAL);
1047 1109 }
1048 1110 if ((err = zfs_onexit_fd_hold(cleanup_fd, &minor)) != 0) {
1049 1111 spa_close(spa, FTAG);
1050 1112 return (err);
1051 1113 }
1052 1114
1053 - err = snapshot_one_impl(snapname, &saa);
1115 + err = snapshot_one_impl(snapname, &saa, 1);
1054 1116
1055 1117 if (err == 0)
1056 1118 err = dsl_sync_task_group_wait(saa.saa_dstg);
1057 1119
1058 1120 for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
1059 1121 dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
1060 1122 objset_t *os = dst->dst_arg1;
1061 1123 dsl_register_onexit_hold_cleanup(saa.saa_newds, tag, minor);
1062 1124 if (saa.saa_needsuspend)
1063 1125 zil_resume(dmu_objset_zil(os));
1064 1126 dmu_objset_rele(os, &saa);
1065 1127 }
1066 1128
1067 1129 zfs_onexit_fd_rele(cleanup_fd);
1068 1130 dsl_sync_task_group_destroy(saa.saa_dstg);
1069 1131 spa_close(spa, FTAG);
1070 1132 return (err);
1071 1133 }
1072 1134
1073 1135
1074 1136 static void
1075 1137 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1076 1138 {
1077 1139 dnode_t *dn;
1078 1140
1079 1141 while (dn = list_head(list)) {
1080 1142 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1081 1143 ASSERT(dn->dn_dbuf->db_data_pending);
1082 1144 /*
1083 1145 * Initialize dn_zio outside dnode_sync() because the
1084 1146 * meta-dnode needs to set it ouside dnode_sync().
1085 1147 */
1086 1148 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1087 1149 ASSERT(dn->dn_zio);
1088 1150
1089 1151 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1090 1152 list_remove(list, dn);
1091 1153
1092 1154 if (newlist) {
1093 1155 (void) dnode_add_ref(dn, newlist);
1094 1156 list_insert_tail(newlist, dn);
1095 1157 }
1096 1158
1097 1159 dnode_sync(dn, tx);
1098 1160 }
1099 1161 }
1100 1162
1101 1163 /* ARGSUSED */
1102 1164 static void
1103 1165 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1104 1166 {
1105 1167 blkptr_t *bp = zio->io_bp;
1106 1168 objset_t *os = arg;
1107 1169 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1108 1170
1109 1171 ASSERT(bp == os->os_rootbp);
1110 1172 ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
1111 1173 ASSERT(BP_GET_LEVEL(bp) == 0);
1112 1174
1113 1175 /*
1114 1176 * Update rootbp fill count: it should be the number of objects
1115 1177 * allocated in the object set (not counting the "special"
1116 1178 * objects that are stored in the objset_phys_t -- the meta
1117 1179 * dnode and user/group accounting objects).
1118 1180 */
1119 1181 bp->blk_fill = 0;
1120 1182 for (int i = 0; i < dnp->dn_nblkptr; i++)
1121 1183 bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
1122 1184 }
1123 1185
1124 1186 /* ARGSUSED */
1125 1187 static void
1126 1188 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1127 1189 {
1128 1190 blkptr_t *bp = zio->io_bp;
1129 1191 blkptr_t *bp_orig = &zio->io_bp_orig;
1130 1192 objset_t *os = arg;
1131 1193
1132 1194 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1133 1195 ASSERT(BP_EQUAL(bp, bp_orig));
1134 1196 } else {
1135 1197 dsl_dataset_t *ds = os->os_dsl_dataset;
1136 1198 dmu_tx_t *tx = os->os_synctx;
1137 1199
1138 1200 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1139 1201 dsl_dataset_block_born(ds, bp, tx);
1140 1202 }
1141 1203 }
1142 1204
1143 1205 /* called from dsl */
1144 1206 void
1145 1207 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1146 1208 {
1147 1209 int txgoff;
1148 1210 zbookmark_t zb;
1149 1211 zio_prop_t zp;
1150 1212 zio_t *zio;
1151 1213 list_t *list;
1152 1214 list_t *newlist = NULL;
1153 1215 dbuf_dirty_record_t *dr;
1154 1216
1155 1217 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1156 1218
1157 1219 ASSERT(dmu_tx_is_syncing(tx));
1158 1220 /* XXX the write_done callback should really give us the tx... */
1159 1221 os->os_synctx = tx;
1160 1222
1161 1223 if (os->os_dsl_dataset == NULL) {
1162 1224 /*
1163 1225 * This is the MOS. If we have upgraded,
1164 1226 * spa_max_replication() could change, so reset
1165 1227 * os_copies here.
1166 1228 */
1167 1229 os->os_copies = spa_max_replication(os->os_spa);
1168 1230 }
1169 1231
1170 1232 /*
1171 1233 * Create the root block IO
1172 1234 */
1173 1235 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1174 1236 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1175 1237 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1176 1238 VERIFY3U(0, ==, arc_release_bp(os->os_phys_buf, &os->os_phys_buf,
1177 1239 os->os_rootbp, os->os_spa, &zb));
1178 1240
1179 1241 dmu_write_policy(os, NULL, 0, 0, &zp);
1180 1242
1181 1243 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1182 1244 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
1183 1245 dmu_objset_write_ready, dmu_objset_write_done, os,
1184 1246 ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1185 1247
1186 1248 /*
1187 1249 * Sync special dnodes - the parent IO for the sync is the root block
1188 1250 */
1189 1251 DMU_META_DNODE(os)->dn_zio = zio;
1190 1252 dnode_sync(DMU_META_DNODE(os), tx);
1191 1253
1192 1254 os->os_phys->os_flags = os->os_flags;
1193 1255
1194 1256 if (DMU_USERUSED_DNODE(os) &&
1195 1257 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1196 1258 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1197 1259 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1198 1260 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1199 1261 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1200 1262 }
1201 1263
1202 1264 txgoff = tx->tx_txg & TXG_MASK;
1203 1265
1204 1266 if (dmu_objset_userused_enabled(os)) {
1205 1267 newlist = &os->os_synced_dnodes;
1206 1268 /*
1207 1269 * We must create the list here because it uses the
1208 1270 * dn_dirty_link[] of this txg.
1209 1271 */
1210 1272 list_create(newlist, sizeof (dnode_t),
1211 1273 offsetof(dnode_t, dn_dirty_link[txgoff]));
1212 1274 }
1213 1275
1214 1276 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1215 1277 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1216 1278
1217 1279 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1218 1280 while (dr = list_head(list)) {
1219 1281 ASSERT(dr->dr_dbuf->db_level == 0);
1220 1282 list_remove(list, dr);
1221 1283 if (dr->dr_zio)
1222 1284 zio_nowait(dr->dr_zio);
1223 1285 }
1224 1286 /*
1225 1287 * Free intent log blocks up to this tx.
1226 1288 */
1227 1289 zil_sync(os->os_zil, tx);
1228 1290 os->os_phys->os_zil_header = os->os_zil_header;
1229 1291 zio_nowait(zio);
1230 1292 }
1231 1293
1232 1294 boolean_t
1233 1295 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1234 1296 {
1235 1297 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1236 1298 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1237 1299 }
1238 1300
1239 1301 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1240 1302
1241 1303 void
1242 1304 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1243 1305 {
1244 1306 used_cbs[ost] = cb;
1245 1307 }
1246 1308
1247 1309 boolean_t
1248 1310 dmu_objset_userused_enabled(objset_t *os)
1249 1311 {
1250 1312 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1251 1313 used_cbs[os->os_phys->os_type] != NULL &&
1252 1314 DMU_USERUSED_DNODE(os) != NULL);
1253 1315 }
1254 1316
1255 1317 static void
1256 1318 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1257 1319 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1258 1320 {
1259 1321 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1260 1322 int64_t delta = DNODE_SIZE + used;
1261 1323 if (subtract)
1262 1324 delta = -delta;
1263 1325 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1264 1326 user, delta, tx));
1265 1327 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1266 1328 group, delta, tx));
1267 1329 }
1268 1330 }
1269 1331
1270 1332 void
1271 1333 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1272 1334 {
1273 1335 dnode_t *dn;
1274 1336 list_t *list = &os->os_synced_dnodes;
1275 1337
1276 1338 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1277 1339
1278 1340 while (dn = list_head(list)) {
1279 1341 int flags;
1280 1342 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1281 1343 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1282 1344 dn->dn_phys->dn_flags &
1283 1345 DNODE_FLAG_USERUSED_ACCOUNTED);
1284 1346
1285 1347 /* Allocate the user/groupused objects if necessary. */
1286 1348 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1287 1349 VERIFY(0 == zap_create_claim(os,
1288 1350 DMU_USERUSED_OBJECT,
1289 1351 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1290 1352 VERIFY(0 == zap_create_claim(os,
1291 1353 DMU_GROUPUSED_OBJECT,
1292 1354 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1293 1355 }
1294 1356
1295 1357 /*
1296 1358 * We intentionally modify the zap object even if the
1297 1359 * net delta is zero. Otherwise
1298 1360 * the block of the zap obj could be shared between
1299 1361 * datasets but need to be different between them after
1300 1362 * a bprewrite.
1301 1363 */
1302 1364
1303 1365 flags = dn->dn_id_flags;
1304 1366 ASSERT(flags);
1305 1367 if (flags & DN_ID_OLD_EXIST) {
1306 1368 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1307 1369 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1308 1370 }
1309 1371 if (flags & DN_ID_NEW_EXIST) {
1310 1372 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1311 1373 dn->dn_phys->dn_flags, dn->dn_newuid,
1312 1374 dn->dn_newgid, B_FALSE, tx);
1313 1375 }
1314 1376
1315 1377 mutex_enter(&dn->dn_mtx);
1316 1378 dn->dn_oldused = 0;
1317 1379 dn->dn_oldflags = 0;
1318 1380 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1319 1381 dn->dn_olduid = dn->dn_newuid;
1320 1382 dn->dn_oldgid = dn->dn_newgid;
1321 1383 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1322 1384 if (dn->dn_bonuslen == 0)
1323 1385 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1324 1386 else
1325 1387 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1326 1388 }
1327 1389 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1328 1390 mutex_exit(&dn->dn_mtx);
1329 1391
1330 1392 list_remove(list, dn);
1331 1393 dnode_rele(dn, list);
1332 1394 }
1333 1395 }
1334 1396
1335 1397 /*
1336 1398 * Returns a pointer to data to find uid/gid from
1337 1399 *
1338 1400 * If a dirty record for transaction group that is syncing can't
1339 1401 * be found then NULL is returned. In the NULL case it is assumed
1340 1402 * the uid/gid aren't changing.
1341 1403 */
1342 1404 static void *
1343 1405 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1344 1406 {
1345 1407 dbuf_dirty_record_t *dr, **drp;
1346 1408 void *data;
1347 1409
1348 1410 if (db->db_dirtycnt == 0)
1349 1411 return (db->db.db_data); /* Nothing is changing */
1350 1412
1351 1413 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1352 1414 if (dr->dr_txg == tx->tx_txg)
1353 1415 break;
1354 1416
1355 1417 if (dr == NULL) {
1356 1418 data = NULL;
1357 1419 } else {
1358 1420 dnode_t *dn;
1359 1421
1360 1422 DB_DNODE_ENTER(dr->dr_dbuf);
1361 1423 dn = DB_DNODE(dr->dr_dbuf);
1362 1424
1363 1425 if (dn->dn_bonuslen == 0 &&
1364 1426 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1365 1427 data = dr->dt.dl.dr_data->b_data;
1366 1428 else
1367 1429 data = dr->dt.dl.dr_data;
1368 1430
1369 1431 DB_DNODE_EXIT(dr->dr_dbuf);
1370 1432 }
1371 1433
1372 1434 return (data);
1373 1435 }
1374 1436
1375 1437 void
1376 1438 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1377 1439 {
1378 1440 objset_t *os = dn->dn_objset;
1379 1441 void *data = NULL;
1380 1442 dmu_buf_impl_t *db = NULL;
1381 1443 uint64_t *user, *group;
1382 1444 int flags = dn->dn_id_flags;
1383 1445 int error;
1384 1446 boolean_t have_spill = B_FALSE;
1385 1447
1386 1448 if (!dmu_objset_userused_enabled(dn->dn_objset))
1387 1449 return;
1388 1450
1389 1451 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1390 1452 DN_ID_CHKED_SPILL)))
1391 1453 return;
1392 1454
1393 1455 if (before && dn->dn_bonuslen != 0)
1394 1456 data = DN_BONUS(dn->dn_phys);
1395 1457 else if (!before && dn->dn_bonuslen != 0) {
1396 1458 if (dn->dn_bonus) {
1397 1459 db = dn->dn_bonus;
1398 1460 mutex_enter(&db->db_mtx);
1399 1461 data = dmu_objset_userquota_find_data(db, tx);
1400 1462 } else {
1401 1463 data = DN_BONUS(dn->dn_phys);
1402 1464 }
1403 1465 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1404 1466 int rf = 0;
1405 1467
1406 1468 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1407 1469 rf |= DB_RF_HAVESTRUCT;
1408 1470 error = dmu_spill_hold_by_dnode(dn,
1409 1471 rf | DB_RF_MUST_SUCCEED,
1410 1472 FTAG, (dmu_buf_t **)&db);
1411 1473 ASSERT(error == 0);
1412 1474 mutex_enter(&db->db_mtx);
1413 1475 data = (before) ? db->db.db_data :
1414 1476 dmu_objset_userquota_find_data(db, tx);
1415 1477 have_spill = B_TRUE;
1416 1478 } else {
1417 1479 mutex_enter(&dn->dn_mtx);
1418 1480 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1419 1481 mutex_exit(&dn->dn_mtx);
1420 1482 return;
1421 1483 }
1422 1484
1423 1485 if (before) {
1424 1486 ASSERT(data);
1425 1487 user = &dn->dn_olduid;
1426 1488 group = &dn->dn_oldgid;
1427 1489 } else if (data) {
1428 1490 user = &dn->dn_newuid;
1429 1491 group = &dn->dn_newgid;
1430 1492 }
1431 1493
1432 1494 /*
1433 1495 * Must always call the callback in case the object
1434 1496 * type has changed and that type isn't an object type to track
1435 1497 */
1436 1498 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1437 1499 user, group);
1438 1500
1439 1501 /*
1440 1502 * Preserve existing uid/gid when the callback can't determine
1441 1503 * what the new uid/gid are and the callback returned EEXIST.
1442 1504 * The EEXIST error tells us to just use the existing uid/gid.
1443 1505 * If we don't know what the old values are then just assign
1444 1506 * them to 0, since that is a new file being created.
1445 1507 */
1446 1508 if (!before && data == NULL && error == EEXIST) {
1447 1509 if (flags & DN_ID_OLD_EXIST) {
1448 1510 dn->dn_newuid = dn->dn_olduid;
1449 1511 dn->dn_newgid = dn->dn_oldgid;
1450 1512 } else {
1451 1513 dn->dn_newuid = 0;
1452 1514 dn->dn_newgid = 0;
1453 1515 }
1454 1516 error = 0;
1455 1517 }
1456 1518
1457 1519 if (db)
1458 1520 mutex_exit(&db->db_mtx);
1459 1521
1460 1522 mutex_enter(&dn->dn_mtx);
1461 1523 if (error == 0 && before)
1462 1524 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1463 1525 if (error == 0 && !before)
1464 1526 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1465 1527
1466 1528 if (have_spill) {
1467 1529 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1468 1530 } else {
1469 1531 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1470 1532 }
1471 1533 mutex_exit(&dn->dn_mtx);
1472 1534 if (have_spill)
1473 1535 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1474 1536 }
1475 1537
1476 1538 boolean_t
1477 1539 dmu_objset_userspace_present(objset_t *os)
1478 1540 {
1479 1541 return (os->os_phys->os_flags &
1480 1542 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1481 1543 }
1482 1544
1483 1545 int
1484 1546 dmu_objset_userspace_upgrade(objset_t *os)
1485 1547 {
1486 1548 uint64_t obj;
1487 1549 int err = 0;
1488 1550
1489 1551 if (dmu_objset_userspace_present(os))
1490 1552 return (0);
1491 1553 if (!dmu_objset_userused_enabled(os))
1492 1554 return (ENOTSUP);
1493 1555 if (dmu_objset_is_snapshot(os))
1494 1556 return (EINVAL);
1495 1557
1496 1558 /*
1497 1559 * We simply need to mark every object dirty, so that it will be
1498 1560 * synced out and now accounted. If this is called
1499 1561 * concurrently, or if we already did some work before crashing,
1500 1562 * that's fine, since we track each object's accounted state
1501 1563 * independently.
1502 1564 */
1503 1565
1504 1566 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1505 1567 dmu_tx_t *tx;
1506 1568 dmu_buf_t *db;
1507 1569 int objerr;
1508 1570
1509 1571 if (issig(JUSTLOOKING) && issig(FORREAL))
1510 1572 return (EINTR);
1511 1573
1512 1574 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1513 1575 if (objerr)
1514 1576 continue;
1515 1577 tx = dmu_tx_create(os);
1516 1578 dmu_tx_hold_bonus(tx, obj);
1517 1579 objerr = dmu_tx_assign(tx, TXG_WAIT);
1518 1580 if (objerr) {
1519 1581 dmu_tx_abort(tx);
1520 1582 continue;
1521 1583 }
1522 1584 dmu_buf_will_dirty(db, tx);
1523 1585 dmu_buf_rele(db, FTAG);
1524 1586 dmu_tx_commit(tx);
1525 1587 }
1526 1588
1527 1589 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1528 1590 txg_wait_synced(dmu_objset_pool(os), 0);
1529 1591 return (0);
1530 1592 }
1531 1593
1532 1594 void
1533 1595 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1534 1596 uint64_t *usedobjsp, uint64_t *availobjsp)
1535 1597 {
1536 1598 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1537 1599 usedobjsp, availobjsp);
1538 1600 }
1539 1601
1540 1602 uint64_t
1541 1603 dmu_objset_fsid_guid(objset_t *os)
1542 1604 {
1543 1605 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1544 1606 }
1545 1607
1546 1608 void
1547 1609 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1548 1610 {
1549 1611 stat->dds_type = os->os_phys->os_type;
1550 1612 if (os->os_dsl_dataset)
1551 1613 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1552 1614 }
1553 1615
1554 1616 void
1555 1617 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1556 1618 {
1557 1619 ASSERT(os->os_dsl_dataset ||
1558 1620 os->os_phys->os_type == DMU_OST_META);
1559 1621
1560 1622 if (os->os_dsl_dataset != NULL)
1561 1623 dsl_dataset_stats(os->os_dsl_dataset, nv);
1562 1624
1563 1625 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1564 1626 os->os_phys->os_type);
1565 1627 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1566 1628 dmu_objset_userspace_present(os));
1567 1629 }
1568 1630
1569 1631 int
1570 1632 dmu_objset_is_snapshot(objset_t *os)
1571 1633 {
1572 1634 if (os->os_dsl_dataset != NULL)
1573 1635 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1574 1636 else
1575 1637 return (B_FALSE);
1576 1638 }
1577 1639
1578 1640 int
1579 1641 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1580 1642 boolean_t *conflict)
1581 1643 {
1582 1644 dsl_dataset_t *ds = os->os_dsl_dataset;
1583 1645 uint64_t ignored;
1584 1646
1585 1647 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1586 1648 return (ENOENT);
1587 1649
1588 1650 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1589 1651 ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1590 1652 real, maxlen, conflict));
1591 1653 }
1592 1654
1593 1655 int
1594 1656 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1595 1657 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1596 1658 {
1597 1659 dsl_dataset_t *ds = os->os_dsl_dataset;
1598 1660 zap_cursor_t cursor;
1599 1661 zap_attribute_t attr;
1600 1662
1601 1663 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1602 1664 return (ENOENT);
1603 1665
1604 1666 zap_cursor_init_serialized(&cursor,
1605 1667 ds->ds_dir->dd_pool->dp_meta_objset,
1606 1668 ds->ds_phys->ds_snapnames_zapobj, *offp);
1607 1669
1608 1670 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1609 1671 zap_cursor_fini(&cursor);
1610 1672 return (ENOENT);
1611 1673 }
1612 1674
1613 1675 if (strlen(attr.za_name) + 1 > namelen) {
1614 1676 zap_cursor_fini(&cursor);
1615 1677 return (ENAMETOOLONG);
1616 1678 }
1617 1679
1618 1680 (void) strcpy(name, attr.za_name);
1619 1681 if (idp)
1620 1682 *idp = attr.za_first_integer;
1621 1683 if (case_conflict)
1622 1684 *case_conflict = attr.za_normalization_conflict;
1623 1685 zap_cursor_advance(&cursor);
1624 1686 *offp = zap_cursor_serialize(&cursor);
1625 1687 zap_cursor_fini(&cursor);
1626 1688
1627 1689 return (0);
1628 1690 }
1629 1691
1630 1692 int
1631 1693 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1632 1694 uint64_t *idp, uint64_t *offp)
1633 1695 {
1634 1696 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1635 1697 zap_cursor_t cursor;
1636 1698 zap_attribute_t attr;
1637 1699
1638 1700 /* there is no next dir on a snapshot! */
1639 1701 if (os->os_dsl_dataset->ds_object !=
1640 1702 dd->dd_phys->dd_head_dataset_obj)
1641 1703 return (ENOENT);
1642 1704
1643 1705 zap_cursor_init_serialized(&cursor,
1644 1706 dd->dd_pool->dp_meta_objset,
1645 1707 dd->dd_phys->dd_child_dir_zapobj, *offp);
1646 1708
1647 1709 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1648 1710 zap_cursor_fini(&cursor);
1649 1711 return (ENOENT);
1650 1712 }
1651 1713
1652 1714 if (strlen(attr.za_name) + 1 > namelen) {
1653 1715 zap_cursor_fini(&cursor);
1654 1716 return (ENAMETOOLONG);
1655 1717 }
1656 1718
1657 1719 (void) strcpy(name, attr.za_name);
1658 1720 if (idp)
1659 1721 *idp = attr.za_first_integer;
1660 1722 zap_cursor_advance(&cursor);
1661 1723 *offp = zap_cursor_serialize(&cursor);
1662 1724 zap_cursor_fini(&cursor);
1663 1725
1664 1726 return (0);
1665 1727 }
1666 1728
1667 1729 struct findarg {
1668 1730 int (*func)(const char *, void *);
1669 1731 void *arg;
1670 1732 };
1671 1733
1672 1734 /* ARGSUSED */
1673 1735 static int
1674 1736 findfunc(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
1675 1737 {
1676 1738 struct findarg *fa = arg;
1677 1739 return (fa->func(dsname, fa->arg));
1678 1740 }
1679 1741
1680 1742 /*
1681 1743 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1682 1744 * Perhaps change all callers to use dmu_objset_find_spa()?
1683 1745 */
1684 1746 int
1685 1747 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1686 1748 int flags)
1687 1749 {
1688 1750 struct findarg fa;
1689 1751 fa.func = func;
1690 1752 fa.arg = arg;
1691 1753 return (dmu_objset_find_spa(NULL, name, findfunc, &fa, flags));
1692 1754 }
1693 1755
1694 1756 /*
1695 1757 * Find all objsets under name, call func on each
1696 1758 */
1697 1759 int
1698 1760 dmu_objset_find_spa(spa_t *spa, const char *name,
1699 1761 int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
1700 1762 {
1701 1763 dsl_dir_t *dd;
1702 1764 dsl_pool_t *dp;
1703 1765 dsl_dataset_t *ds;
1704 1766 zap_cursor_t zc;
1705 1767 zap_attribute_t *attr;
1706 1768 char *child;
1707 1769 uint64_t thisobj;
1708 1770 int err;
1709 1771
1710 1772 if (name == NULL)
1711 1773 name = spa_name(spa);
1712 1774 err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
1713 1775 if (err)
1714 1776 return (err);
1715 1777
1716 1778 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1717 1779 if (dd->dd_myname[0] == '$') {
1718 1780 dsl_dir_close(dd, FTAG);
1719 1781 return (0);
1720 1782 }
1721 1783
1722 1784 thisobj = dd->dd_phys->dd_head_dataset_obj;
1723 1785 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1724 1786 dp = dd->dd_pool;
1725 1787
1726 1788 /*
1727 1789 * Iterate over all children.
1728 1790 */
1729 1791 if (flags & DS_FIND_CHILDREN) {
1730 1792 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1731 1793 dd->dd_phys->dd_child_dir_zapobj);
1732 1794 zap_cursor_retrieve(&zc, attr) == 0;
1733 1795 (void) zap_cursor_advance(&zc)) {
1734 1796 ASSERT(attr->za_integer_length == sizeof (uint64_t));
1735 1797 ASSERT(attr->za_num_integers == 1);
1736 1798
1737 1799 child = kmem_asprintf("%s/%s", name, attr->za_name);
1738 1800 err = dmu_objset_find_spa(spa, child, func, arg, flags);
1739 1801 strfree(child);
1740 1802 if (err)
1741 1803 break;
1742 1804 }
1743 1805 zap_cursor_fini(&zc);
1744 1806
1745 1807 if (err) {
1746 1808 dsl_dir_close(dd, FTAG);
1747 1809 kmem_free(attr, sizeof (zap_attribute_t));
1748 1810 return (err);
1749 1811 }
1750 1812 }
1751 1813
1752 1814 /*
1753 1815 * Iterate over all snapshots.
1754 1816 */
1755 1817 if (flags & DS_FIND_SNAPSHOTS) {
1756 1818 if (!dsl_pool_sync_context(dp))
1757 1819 rw_enter(&dp->dp_config_rwlock, RW_READER);
1758 1820 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1759 1821 if (!dsl_pool_sync_context(dp))
1760 1822 rw_exit(&dp->dp_config_rwlock);
1761 1823
1762 1824 if (err == 0) {
1763 1825 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1764 1826 dsl_dataset_rele(ds, FTAG);
1765 1827
1766 1828 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1767 1829 zap_cursor_retrieve(&zc, attr) == 0;
1768 1830 (void) zap_cursor_advance(&zc)) {
1769 1831 ASSERT(attr->za_integer_length ==
1770 1832 sizeof (uint64_t));
1771 1833 ASSERT(attr->za_num_integers == 1);
1772 1834
1773 1835 child = kmem_asprintf("%s@%s",
1774 1836 name, attr->za_name);
1775 1837 err = func(spa, attr->za_first_integer,
1776 1838 child, arg);
1777 1839 strfree(child);
1778 1840 if (err)
1779 1841 break;
1780 1842 }
1781 1843 zap_cursor_fini(&zc);
1782 1844 }
1783 1845 }
1784 1846
1785 1847 dsl_dir_close(dd, FTAG);
1786 1848 kmem_free(attr, sizeof (zap_attribute_t));
1787 1849
1788 1850 if (err)
1789 1851 return (err);
1790 1852
1791 1853 /*
1792 1854 * Apply to self if appropriate.
1793 1855 */
1794 1856 err = func(spa, thisobj, name, arg);
1795 1857 return (err);
1796 1858 }
1797 1859
1798 1860 /* ARGSUSED */
1799 1861 int
1800 1862 dmu_objset_prefetch(const char *name, void *arg)
1801 1863 {
1802 1864 dsl_dataset_t *ds;
1803 1865
1804 1866 if (dsl_dataset_hold(name, FTAG, &ds))
1805 1867 return (0);
1806 1868
1807 1869 if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) {
1808 1870 mutex_enter(&ds->ds_opening_lock);
1809 1871 if (ds->ds_objset == NULL) {
1810 1872 uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
1811 1873 zbookmark_t zb;
1812 1874
1813 1875 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
1814 1876 ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1815 1877
1816 1878 (void) dsl_read_nolock(NULL, dsl_dataset_get_spa(ds),
1817 1879 &ds->ds_phys->ds_bp, NULL, NULL,
1818 1880 ZIO_PRIORITY_ASYNC_READ,
1819 1881 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
1820 1882 &aflags, &zb);
1821 1883 }
1822 1884 mutex_exit(&ds->ds_opening_lock);
1823 1885 }
1824 1886
1825 1887 dsl_dataset_rele(ds, FTAG);
1826 1888 return (0);
1827 1889 }
1828 1890
1829 1891 void
1830 1892 dmu_objset_set_user(objset_t *os, void *user_ptr)
1831 1893 {
1832 1894 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1833 1895 os->os_user_ptr = user_ptr;
1834 1896 }
1835 1897
1836 1898 void *
1837 1899 dmu_objset_get_user(objset_t *os)
1838 1900 {
1839 1901 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1840 1902 return (os->os_user_ptr);
1841 1903 }
|
↓ open down ↓ |
778 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX