]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
9b7b9cd3 | 21 | |
34dc7c2f | 22 | /* |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
64fc7762 | 24 | * Copyright (c) 2012, 2017 by Delphix. All rights reserved. |
3a17a7a9 | 25 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
788eb90c | 26 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
0c66c32d | 27 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
9c43027b | 28 | * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. |
a0bd735a | 29 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
9b7b9cd3 | 30 | * Copyright 2017 Nexenta Systems, Inc. |
34dc7c2f BB |
31 | */ |
32 | ||
428870ff BB |
33 | /* Portions Copyright 2010 Robert Milkowski */ |
34 | ||
1de321e6 | 35 | #include <sys/zfeature.h> |
34dc7c2f BB |
36 | #include <sys/cred.h> |
37 | #include <sys/zfs_context.h> | |
38 | #include <sys/dmu_objset.h> | |
39 | #include <sys/dsl_dir.h> | |
40 | #include <sys/dsl_dataset.h> | |
41 | #include <sys/dsl_prop.h> | |
42 | #include <sys/dsl_pool.h> | |
43 | #include <sys/dsl_synctask.h> | |
44 | #include <sys/dsl_deleg.h> | |
45 | #include <sys/dnode.h> | |
46 | #include <sys/dbuf.h> | |
47 | #include <sys/zvol.h> | |
48 | #include <sys/dmu_tx.h> | |
34dc7c2f BB |
49 | #include <sys/zap.h> |
50 | #include <sys/zil.h> | |
51 | #include <sys/dmu_impl.h> | |
52 | #include <sys/zfs_ioctl.h> | |
428870ff | 53 | #include <sys/sa.h> |
572e2857 | 54 | #include <sys/zfs_onexit.h> |
13fe0198 | 55 | #include <sys/dsl_destroy.h> |
9c43027b | 56 | #include <sys/vdev.h> |
f74b821a | 57 | #include <sys/policy.h> |
1de321e6 | 58 | #include <sys/spa_impl.h> |
b5256303 | 59 | #include <sys/dmu_send.h> |
572e2857 BB |
60 | |
61 | /* | |
62 | * Needed to close a window in dnode_move() that allows the objset to be freed | |
63 | * before it can be safely accessed. | |
64 | */ | |
65 | krwlock_t os_lock; | |
66 | ||
9c43027b | 67 | /* |
4e33ba4c | 68 | * Tunable to overwrite the maximum number of threads for the parallelization |
9c43027b AJ |
69 | * of dmu_objset_find_dp, needed to speed up the import of pools with many |
70 | * datasets. | |
71 | * Default is 4 times the number of leaf vdevs. | |
72 | */ | |
73 | int dmu_find_threads = 0; | |
74 | ||
68cbd56e NB |
75 | /* |
76 | * Backfill lower metadnode objects after this many have been freed. | |
77 | * Backfilling negatively impacts object creation rates, so only do it | |
78 | * if there are enough holes to fill. | |
79 | */ | |
80 | int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; | |
81 | ||
9c43027b AJ |
82 | static void dmu_objset_find_dp_cb(void *arg); |
83 | ||
1de321e6 JX |
84 | static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); |
85 | static void dmu_objset_upgrade_stop(objset_t *os); | |
86 | ||
572e2857 BB |
87 | void |
88 | dmu_objset_init(void) | |
89 | { | |
90 | rw_init(&os_lock, NULL, RW_DEFAULT, NULL); | |
91 | } | |
92 | ||
93 | void | |
94 | dmu_objset_fini(void) | |
95 | { | |
96 | rw_destroy(&os_lock); | |
97 | } | |
34dc7c2f BB |
98 | |
99 | spa_t * | |
100 | dmu_objset_spa(objset_t *os) | |
101 | { | |
428870ff | 102 | return (os->os_spa); |
34dc7c2f BB |
103 | } |
104 | ||
105 | zilog_t * | |
106 | dmu_objset_zil(objset_t *os) | |
107 | { | |
428870ff | 108 | return (os->os_zil); |
34dc7c2f BB |
109 | } |
110 | ||
111 | dsl_pool_t * | |
112 | dmu_objset_pool(objset_t *os) | |
113 | { | |
114 | dsl_dataset_t *ds; | |
115 | ||
428870ff | 116 | if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) |
34dc7c2f BB |
117 | return (ds->ds_dir->dd_pool); |
118 | else | |
428870ff | 119 | return (spa_get_dsl(os->os_spa)); |
34dc7c2f BB |
120 | } |
121 | ||
122 | dsl_dataset_t * | |
123 | dmu_objset_ds(objset_t *os) | |
124 | { | |
428870ff | 125 | return (os->os_dsl_dataset); |
34dc7c2f BB |
126 | } |
127 | ||
128 | dmu_objset_type_t | |
129 | dmu_objset_type(objset_t *os) | |
130 | { | |
428870ff | 131 | return (os->os_phys->os_type); |
34dc7c2f BB |
132 | } |
133 | ||
134 | void | |
135 | dmu_objset_name(objset_t *os, char *buf) | |
136 | { | |
428870ff | 137 | dsl_dataset_name(os->os_dsl_dataset, buf); |
34dc7c2f BB |
138 | } |
139 | ||
140 | uint64_t | |
141 | dmu_objset_id(objset_t *os) | |
142 | { | |
428870ff | 143 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
144 | |
145 | return (ds ? ds->ds_object : 0); | |
146 | } | |
147 | ||
50c957f7 NB |
148 | uint64_t |
149 | dmu_objset_dnodesize(objset_t *os) | |
150 | { | |
151 | return (os->os_dnodesize); | |
152 | } | |
153 | ||
faf0f58c | 154 | zfs_sync_type_t |
428870ff BB |
155 | dmu_objset_syncprop(objset_t *os) |
156 | { | |
157 | return (os->os_sync); | |
158 | } | |
159 | ||
faf0f58c | 160 | zfs_logbias_op_t |
428870ff BB |
161 | dmu_objset_logbias(objset_t *os) |
162 | { | |
163 | return (os->os_logbias); | |
164 | } | |
165 | ||
34dc7c2f BB |
166 | static void |
167 | checksum_changed_cb(void *arg, uint64_t newval) | |
168 | { | |
428870ff | 169 | objset_t *os = arg; |
34dc7c2f BB |
170 | |
171 | /* | |
172 | * Inheritance should have been done by now. | |
173 | */ | |
174 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
175 | ||
428870ff | 176 | os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); |
34dc7c2f BB |
177 | } |
178 | ||
179 | static void | |
180 | compression_changed_cb(void *arg, uint64_t newval) | |
181 | { | |
428870ff | 182 | objset_t *os = arg; |
34dc7c2f BB |
183 | |
184 | /* | |
185 | * Inheritance and range checking should have been done by now. | |
186 | */ | |
187 | ASSERT(newval != ZIO_COMPRESS_INHERIT); | |
188 | ||
99197f03 JG |
189 | os->os_compress = zio_compress_select(os->os_spa, newval, |
190 | ZIO_COMPRESS_ON); | |
34dc7c2f BB |
191 | } |
192 | ||
193 | static void | |
194 | copies_changed_cb(void *arg, uint64_t newval) | |
195 | { | |
428870ff | 196 | objset_t *os = arg; |
34dc7c2f BB |
197 | |
198 | /* | |
199 | * Inheritance and range checking should have been done by now. | |
200 | */ | |
201 | ASSERT(newval > 0); | |
428870ff | 202 | ASSERT(newval <= spa_max_replication(os->os_spa)); |
34dc7c2f | 203 | |
428870ff BB |
204 | os->os_copies = newval; |
205 | } | |
206 | ||
207 | static void | |
208 | dedup_changed_cb(void *arg, uint64_t newval) | |
209 | { | |
210 | objset_t *os = arg; | |
211 | spa_t *spa = os->os_spa; | |
212 | enum zio_checksum checksum; | |
213 | ||
214 | /* | |
215 | * Inheritance should have been done by now. | |
216 | */ | |
217 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
218 | ||
219 | checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); | |
220 | ||
221 | os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; | |
222 | os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); | |
34dc7c2f BB |
223 | } |
224 | ||
b128c09f BB |
225 | static void |
226 | primary_cache_changed_cb(void *arg, uint64_t newval) | |
227 | { | |
428870ff | 228 | objset_t *os = arg; |
b128c09f BB |
229 | |
230 | /* | |
231 | * Inheritance and range checking should have been done by now. | |
232 | */ | |
233 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
234 | newval == ZFS_CACHE_METADATA); | |
235 | ||
428870ff | 236 | os->os_primary_cache = newval; |
b128c09f BB |
237 | } |
238 | ||
239 | static void | |
240 | secondary_cache_changed_cb(void *arg, uint64_t newval) | |
241 | { | |
428870ff | 242 | objset_t *os = arg; |
b128c09f BB |
243 | |
244 | /* | |
245 | * Inheritance and range checking should have been done by now. | |
246 | */ | |
247 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
248 | newval == ZFS_CACHE_METADATA); | |
249 | ||
428870ff BB |
250 | os->os_secondary_cache = newval; |
251 | } | |
252 | ||
253 | static void | |
254 | sync_changed_cb(void *arg, uint64_t newval) | |
255 | { | |
256 | objset_t *os = arg; | |
257 | ||
258 | /* | |
259 | * Inheritance and range checking should have been done by now. | |
260 | */ | |
261 | ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || | |
262 | newval == ZFS_SYNC_DISABLED); | |
263 | ||
264 | os->os_sync = newval; | |
265 | if (os->os_zil) | |
266 | zil_set_sync(os->os_zil, newval); | |
267 | } | |
268 | ||
faf0f58c MA |
269 | static void |
270 | redundant_metadata_changed_cb(void *arg, uint64_t newval) | |
271 | { | |
272 | objset_t *os = arg; | |
273 | ||
274 | /* | |
275 | * Inheritance and range checking should have been done by now. | |
276 | */ | |
277 | ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || | |
278 | newval == ZFS_REDUNDANT_METADATA_MOST); | |
279 | ||
280 | os->os_redundant_metadata = newval; | |
281 | } | |
282 | ||
50c957f7 NB |
283 | static void |
284 | dnodesize_changed_cb(void *arg, uint64_t newval) | |
285 | { | |
286 | objset_t *os = arg; | |
287 | ||
288 | switch (newval) { | |
289 | case ZFS_DNSIZE_LEGACY: | |
290 | os->os_dnodesize = DNODE_MIN_SIZE; | |
291 | break; | |
292 | case ZFS_DNSIZE_AUTO: | |
293 | /* | |
294 | * Choose a dnode size that will work well for most | |
295 | * workloads if the user specified "auto". Future code | |
296 | * improvements could dynamically select a dnode size | |
297 | * based on observed workload patterns. | |
298 | */ | |
299 | os->os_dnodesize = DNODE_MIN_SIZE * 2; | |
300 | break; | |
301 | case ZFS_DNSIZE_1K: | |
302 | case ZFS_DNSIZE_2K: | |
303 | case ZFS_DNSIZE_4K: | |
304 | case ZFS_DNSIZE_8K: | |
305 | case ZFS_DNSIZE_16K: | |
306 | os->os_dnodesize = newval; | |
307 | break; | |
308 | } | |
309 | } | |
310 | ||
428870ff BB |
311 | static void |
312 | logbias_changed_cb(void *arg, uint64_t newval) | |
313 | { | |
314 | objset_t *os = arg; | |
315 | ||
316 | ASSERT(newval == ZFS_LOGBIAS_LATENCY || | |
317 | newval == ZFS_LOGBIAS_THROUGHPUT); | |
318 | os->os_logbias = newval; | |
319 | if (os->os_zil) | |
320 | zil_set_logbias(os->os_zil, newval); | |
b128c09f BB |
321 | } |
322 | ||
f1512ee6 MA |
323 | static void |
324 | recordsize_changed_cb(void *arg, uint64_t newval) | |
325 | { | |
326 | objset_t *os = arg; | |
327 | ||
328 | os->os_recordsize = newval; | |
329 | } | |
330 | ||
34dc7c2f BB |
331 | void |
332 | dmu_objset_byteswap(void *buf, size_t size) | |
333 | { | |
334 | objset_phys_t *osp = buf; | |
335 | ||
9babb374 | 336 | ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); |
34dc7c2f BB |
337 | dnode_byteswap(&osp->os_meta_dnode); |
338 | byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); | |
339 | osp->os_type = BSWAP_64(osp->os_type); | |
9babb374 BB |
340 | osp->os_flags = BSWAP_64(osp->os_flags); |
341 | if (size == sizeof (objset_phys_t)) { | |
342 | dnode_byteswap(&osp->os_userused_dnode); | |
343 | dnode_byteswap(&osp->os_groupused_dnode); | |
344 | } | |
34dc7c2f BB |
345 | } |
346 | ||
64fc7762 MA |
347 | /* |
348 | * The hash is a CRC-based hash of the objset_t pointer and the object number. | |
349 | */ | |
350 | static uint64_t | |
351 | dnode_hash(const objset_t *os, uint64_t obj) | |
352 | { | |
353 | uintptr_t osv = (uintptr_t)os; | |
354 | uint64_t crc = -1ULL; | |
355 | ||
356 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
357 | /* | |
358 | * The low 6 bits of the pointer don't have much entropy, because | |
359 | * the objset_t is larger than 2^6 bytes long. | |
360 | */ | |
361 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; | |
362 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; | |
363 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; | |
364 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; | |
365 | ||
366 | crc ^= (osv>>14) ^ (obj>>24); | |
367 | ||
368 | return (crc); | |
369 | } | |
370 | ||
371 | unsigned int | |
372 | dnode_multilist_index_func(multilist_t *ml, void *obj) | |
373 | { | |
374 | dnode_t *dn = obj; | |
375 | return (dnode_hash(dn->dn_objset, dn->dn_object) % | |
376 | multilist_get_num_sublists(ml)); | |
377 | } | |
378 | ||
34dc7c2f BB |
379 | int |
380 | dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
428870ff | 381 | objset_t **osp) |
34dc7c2f | 382 | { |
428870ff | 383 | objset_t *os; |
b128c09f | 384 | int i, err; |
34dc7c2f BB |
385 | |
386 | ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); | |
387 | ||
79c76d5b | 388 | os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); |
428870ff BB |
389 | os->os_dsl_dataset = ds; |
390 | os->os_spa = spa; | |
391 | os->os_rootbp = bp; | |
392 | if (!BP_IS_HOLE(os->os_rootbp)) { | |
2a432414 | 393 | arc_flags_t aflags = ARC_FLAG_WAIT; |
5dbd68a3 | 394 | zbookmark_phys_t zb; |
b5256303 | 395 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; |
428870ff BB |
396 | SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, |
397 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
398 | ||
399 | if (DMU_OS_IS_L2CACHEABLE(os)) | |
2a432414 | 400 | aflags |= ARC_FLAG_L2CACHE; |
34dc7c2f | 401 | |
b5256303 TC |
402 | if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { |
403 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
404 | ASSERT(BP_IS_AUTHENTICATED(bp)); | |
405 | zio_flags |= ZIO_FLAG_RAW; | |
406 | } | |
407 | ||
428870ff | 408 | dprintf_bp(os->os_rootbp, "reading %s", ""); |
294f6806 | 409 | err = arc_read(NULL, spa, os->os_rootbp, |
428870ff | 410 | arc_getbuf_func, &os->os_phys_buf, |
b5256303 | 411 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
13fe0198 | 412 | if (err != 0) { |
428870ff | 413 | kmem_free(os, sizeof (objset_t)); |
b128c09f BB |
414 | /* convert checksum errors into IO errors */ |
415 | if (err == ECKSUM) | |
2e528b49 | 416 | err = SET_ERROR(EIO); |
34dc7c2f BB |
417 | return (err); |
418 | } | |
9babb374 BB |
419 | |
420 | /* Increase the blocksize if we are permitted. */ | |
421 | if (spa_version(spa) >= SPA_VERSION_USERSPACE && | |
428870ff | 422 | arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { |
2aa34383 DK |
423 | arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, |
424 | ARC_BUFC_METADATA, sizeof (objset_phys_t)); | |
9babb374 | 425 | bzero(buf->b_data, sizeof (objset_phys_t)); |
428870ff BB |
426 | bcopy(os->os_phys_buf->b_data, buf->b_data, |
427 | arc_buf_size(os->os_phys_buf)); | |
d3c2ae1c | 428 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 429 | os->os_phys_buf = buf; |
9babb374 BB |
430 | } |
431 | ||
428870ff BB |
432 | os->os_phys = os->os_phys_buf->b_data; |
433 | os->os_flags = os->os_phys->os_flags; | |
34dc7c2f | 434 | } else { |
9babb374 BB |
435 | int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? |
436 | sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; | |
2aa34383 DK |
437 | os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, |
438 | ARC_BUFC_METADATA, size); | |
428870ff BB |
439 | os->os_phys = os->os_phys_buf->b_data; |
440 | bzero(os->os_phys, size); | |
34dc7c2f BB |
441 | } |
442 | ||
443 | /* | |
444 | * Note: the changed_cb will be called once before the register | |
445 | * func returns, thus changing the checksum/compression from the | |
b128c09f BB |
446 | * default (fletcher2/off). Snapshots don't need to know about |
447 | * checksum/compression/copies. | |
34dc7c2f | 448 | */ |
9b67f605 | 449 | if (ds != NULL) { |
47dfff3b MA |
450 | boolean_t needlock = B_FALSE; |
451 | ||
b5256303 TC |
452 | os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); |
453 | ||
47dfff3b MA |
454 | /* |
455 | * Note: it's valid to open the objset if the dataset is | |
456 | * long-held, in which case the pool_config lock will not | |
457 | * be held. | |
458 | */ | |
459 | if (!dsl_pool_config_held(dmu_objset_pool(os))) { | |
460 | needlock = B_TRUE; | |
461 | dsl_pool_config_enter(dmu_objset_pool(os), FTAG); | |
462 | } | |
b5256303 | 463 | |
13fe0198 MA |
464 | err = dsl_prop_register(ds, |
465 | zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), | |
428870ff | 466 | primary_cache_changed_cb, os); |
13fe0198 MA |
467 | if (err == 0) { |
468 | err = dsl_prop_register(ds, | |
469 | zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), | |
428870ff | 470 | secondary_cache_changed_cb, os); |
13fe0198 | 471 | } |
0c66c32d | 472 | if (!ds->ds_is_snapshot) { |
13fe0198 MA |
473 | if (err == 0) { |
474 | err = dsl_prop_register(ds, | |
475 | zfs_prop_to_name(ZFS_PROP_CHECKSUM), | |
428870ff | 476 | checksum_changed_cb, os); |
13fe0198 MA |
477 | } |
478 | if (err == 0) { | |
479 | err = dsl_prop_register(ds, | |
480 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
428870ff | 481 | compression_changed_cb, os); |
13fe0198 MA |
482 | } |
483 | if (err == 0) { | |
484 | err = dsl_prop_register(ds, | |
485 | zfs_prop_to_name(ZFS_PROP_COPIES), | |
428870ff | 486 | copies_changed_cb, os); |
13fe0198 MA |
487 | } |
488 | if (err == 0) { | |
489 | err = dsl_prop_register(ds, | |
490 | zfs_prop_to_name(ZFS_PROP_DEDUP), | |
428870ff | 491 | dedup_changed_cb, os); |
13fe0198 MA |
492 | } |
493 | if (err == 0) { | |
494 | err = dsl_prop_register(ds, | |
495 | zfs_prop_to_name(ZFS_PROP_LOGBIAS), | |
428870ff | 496 | logbias_changed_cb, os); |
13fe0198 MA |
497 | } |
498 | if (err == 0) { | |
499 | err = dsl_prop_register(ds, | |
500 | zfs_prop_to_name(ZFS_PROP_SYNC), | |
428870ff | 501 | sync_changed_cb, os); |
13fe0198 | 502 | } |
faf0f58c MA |
503 | if (err == 0) { |
504 | err = dsl_prop_register(ds, | |
505 | zfs_prop_to_name( | |
506 | ZFS_PROP_REDUNDANT_METADATA), | |
507 | redundant_metadata_changed_cb, os); | |
508 | } | |
f1512ee6 MA |
509 | if (err == 0) { |
510 | err = dsl_prop_register(ds, | |
511 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
512 | recordsize_changed_cb, os); | |
513 | } | |
50c957f7 NB |
514 | if (err == 0) { |
515 | err = dsl_prop_register(ds, | |
516 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
517 | dnodesize_changed_cb, os); | |
518 | } | |
b128c09f | 519 | } |
47dfff3b MA |
520 | if (needlock) |
521 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); | |
13fe0198 | 522 | if (err != 0) { |
d3c2ae1c | 523 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 524 | kmem_free(os, sizeof (objset_t)); |
34dc7c2f BB |
525 | return (err); |
526 | } | |
9b67f605 | 527 | } else { |
34dc7c2f | 528 | /* It's the meta-objset. */ |
428870ff | 529 | os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; |
99197f03 | 530 | os->os_compress = ZIO_COMPRESS_ON; |
b5256303 | 531 | os->os_encrypted = B_FALSE; |
428870ff BB |
532 | os->os_copies = spa_max_replication(spa); |
533 | os->os_dedup_checksum = ZIO_CHECKSUM_OFF; | |
faf0f58c MA |
534 | os->os_dedup_verify = B_FALSE; |
535 | os->os_logbias = ZFS_LOGBIAS_LATENCY; | |
536 | os->os_sync = ZFS_SYNC_STANDARD; | |
428870ff BB |
537 | os->os_primary_cache = ZFS_CACHE_ALL; |
538 | os->os_secondary_cache = ZFS_CACHE_ALL; | |
50c957f7 | 539 | os->os_dnodesize = DNODE_MIN_SIZE; |
34dc7c2f BB |
540 | } |
541 | ||
0c66c32d | 542 | if (ds == NULL || !ds->ds_is_snapshot) |
572e2857 | 543 | os->os_zil_header = os->os_phys->os_zil_header; |
428870ff | 544 | os->os_zil = zil_alloc(os, &os->os_zil_header); |
34dc7c2f BB |
545 | |
546 | for (i = 0; i < TXG_SIZE; i++) { | |
64fc7762 MA |
547 | os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), |
548 | offsetof(dnode_t, dn_dirty_link[i]), | |
549 | dnode_multilist_index_func); | |
34dc7c2f | 550 | } |
428870ff | 551 | list_create(&os->os_dnodes, sizeof (dnode_t), |
34dc7c2f | 552 | offsetof(dnode_t, dn_link)); |
428870ff | 553 | list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), |
34dc7c2f BB |
554 | offsetof(dmu_buf_impl_t, db_link)); |
555 | ||
0c66c32d JG |
556 | list_link_init(&os->os_evicting_node); |
557 | ||
428870ff | 558 | mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); |
64fc7762 | 559 | mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); |
428870ff BB |
560 | mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); |
561 | mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); | |
dbeb8796 MA |
562 | os->os_obj_next_percpu_len = boot_ncpus; |
563 | os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * | |
564 | sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); | |
428870ff | 565 | |
0c66c32d JG |
566 | dnode_special_open(os, &os->os_phys->os_meta_dnode, |
567 | DMU_META_DNODE_OBJECT, &os->os_meta_dnode); | |
428870ff | 568 | if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { |
0c66c32d JG |
569 | dnode_special_open(os, &os->os_phys->os_userused_dnode, |
570 | DMU_USERUSED_OBJECT, &os->os_userused_dnode); | |
571 | dnode_special_open(os, &os->os_phys->os_groupused_dnode, | |
572 | DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); | |
9babb374 | 573 | } |
34dc7c2f | 574 | |
1de321e6 JX |
575 | mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); |
576 | ||
428870ff | 577 | *osp = os; |
34dc7c2f BB |
578 | return (0); |
579 | } | |
580 | ||
428870ff BB |
581 | int |
582 | dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) | |
34dc7c2f | 583 | { |
428870ff | 584 | int err = 0; |
34dc7c2f | 585 | |
47dfff3b MA |
586 | /* |
587 | * We shouldn't be doing anything with dsl_dataset_t's unless the | |
588 | * pool_config lock is held, or the dataset is long-held. | |
589 | */ | |
590 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || | |
591 | dsl_dataset_long_held(ds)); | |
592 | ||
34dc7c2f | 593 | mutex_enter(&ds->ds_opening_lock); |
9b67f605 MA |
594 | if (ds->ds_objset == NULL) { |
595 | objset_t *os; | |
cc9bb3e5 | 596 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
34dc7c2f | 597 | err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), |
9b67f605 | 598 | ds, dsl_dataset_get_blkptr(ds), &os); |
cc9bb3e5 | 599 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
9b67f605 MA |
600 | |
601 | if (err == 0) { | |
602 | mutex_enter(&ds->ds_lock); | |
603 | ASSERT(ds->ds_objset == NULL); | |
604 | ds->ds_objset = os; | |
605 | mutex_exit(&ds->ds_lock); | |
606 | } | |
34dc7c2f | 607 | } |
9b67f605 | 608 | *osp = ds->ds_objset; |
34dc7c2f | 609 | mutex_exit(&ds->ds_opening_lock); |
428870ff | 610 | return (err); |
34dc7c2f BB |
611 | } |
612 | ||
13fe0198 MA |
613 | /* |
614 | * Holds the pool while the objset is held. Therefore only one objset | |
615 | * can be held at a time. | |
616 | */ | |
34dc7c2f | 617 | int |
b5256303 TC |
618 | dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag, |
619 | objset_t **osp) | |
34dc7c2f | 620 | { |
13fe0198 | 621 | dsl_pool_t *dp; |
428870ff | 622 | dsl_dataset_t *ds; |
34dc7c2f | 623 | int err; |
b5256303 | 624 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
34dc7c2f | 625 | |
13fe0198 MA |
626 | err = dsl_pool_hold(name, tag, &dp); |
627 | if (err != 0) | |
628 | return (err); | |
b5256303 | 629 | err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); |
13fe0198 MA |
630 | if (err != 0) { |
631 | dsl_pool_rele(dp, tag); | |
428870ff | 632 | return (err); |
13fe0198 | 633 | } |
428870ff BB |
634 | |
635 | err = dmu_objset_from_ds(ds, osp); | |
13fe0198 | 636 | if (err != 0) { |
428870ff | 637 | dsl_dataset_rele(ds, tag); |
13fe0198 MA |
638 | dsl_pool_rele(dp, tag); |
639 | } | |
428870ff | 640 | |
34dc7c2f BB |
641 | return (err); |
642 | } | |
643 | ||
b5256303 TC |
644 | int |
645 | dmu_objset_hold(const char *name, void *tag, objset_t **osp) | |
646 | { | |
647 | return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); | |
648 | } | |
649 | ||
9c43027b AJ |
650 | static int |
651 | dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, | |
b5256303 | 652 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
9c43027b AJ |
653 | { |
654 | int err; | |
655 | ||
656 | err = dmu_objset_from_ds(ds, osp); | |
657 | if (err != 0) { | |
b5256303 | 658 | return (err); |
9c43027b | 659 | } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { |
9c43027b AJ |
660 | return (SET_ERROR(EINVAL)); |
661 | } else if (!readonly && dsl_dataset_is_snapshot(ds)) { | |
9c43027b AJ |
662 | return (SET_ERROR(EROFS)); |
663 | } | |
b5256303 TC |
664 | |
665 | /* if we are decrypting, we can now check MACs in os->os_phys_buf */ | |
666 | if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { | |
667 | err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, | |
668 | ds->ds_object, B_FALSE); | |
669 | if (err != 0) | |
670 | return (err); | |
671 | ||
672 | ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); | |
673 | } | |
674 | ||
675 | return (0); | |
9c43027b AJ |
676 | } |
677 | ||
13fe0198 MA |
678 | /* |
679 | * dsl_pool must not be held when this is called. | |
680 | * Upon successful return, there will be a longhold on the dataset, | |
681 | * and the dsl_pool will not be held. | |
682 | */ | |
34dc7c2f | 683 | int |
428870ff | 684 | dmu_objset_own(const char *name, dmu_objset_type_t type, |
b5256303 | 685 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
34dc7c2f | 686 | { |
13fe0198 | 687 | dsl_pool_t *dp; |
34dc7c2f BB |
688 | dsl_dataset_t *ds; |
689 | int err; | |
b5256303 | 690 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
34dc7c2f | 691 | |
13fe0198 MA |
692 | err = dsl_pool_hold(name, FTAG, &dp); |
693 | if (err != 0) | |
694 | return (err); | |
b5256303 | 695 | err = dsl_dataset_own(dp, name, flags, tag, &ds); |
13fe0198 MA |
696 | if (err != 0) { |
697 | dsl_pool_rele(dp, FTAG); | |
34dc7c2f | 698 | return (err); |
13fe0198 | 699 | } |
b5256303 TC |
700 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
701 | if (err != 0) { | |
702 | dsl_dataset_disown(ds, flags, tag); | |
703 | dsl_pool_rele(dp, FTAG); | |
704 | return (err); | |
705 | } | |
706 | ||
13fe0198 | 707 | dsl_pool_rele(dp, FTAG); |
9c43027b | 708 | |
b5256303 | 709 | if (dmu_objset_userobjspace_upgradable(*osp)) |
1de321e6 JX |
710 | dmu_objset_userobjspace_upgrade(*osp); |
711 | ||
b5256303 | 712 | return (0); |
34dc7c2f BB |
713 | } |
714 | ||
9c43027b AJ |
715 | int |
716 | dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, | |
b5256303 | 717 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
9c43027b AJ |
718 | { |
719 | dsl_dataset_t *ds; | |
720 | int err; | |
b5256303 | 721 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
9c43027b | 722 | |
b5256303 | 723 | err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); |
9c43027b AJ |
724 | if (err != 0) |
725 | return (err); | |
726 | ||
b5256303 TC |
727 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
728 | if (err != 0) { | |
729 | dsl_dataset_disown(ds, flags, tag); | |
730 | return (err); | |
731 | } | |
732 | ||
733 | return (0); | |
9c43027b AJ |
734 | } |
735 | ||
34dc7c2f | 736 | void |
b5256303 | 737 | dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag) |
34dc7c2f | 738 | { |
b5256303 TC |
739 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
740 | ||
13fe0198 | 741 | dsl_pool_t *dp = dmu_objset_pool(os); |
b5256303 | 742 | dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); |
13fe0198 | 743 | dsl_pool_rele(dp, tag); |
428870ff | 744 | } |
b128c09f | 745 | |
b5256303 TC |
746 | void |
747 | dmu_objset_rele(objset_t *os, void *tag) | |
748 | { | |
749 | dmu_objset_rele_flags(os, B_FALSE, tag); | |
750 | } | |
751 | ||
831baf06 KW |
752 | /* |
753 | * When we are called, os MUST refer to an objset associated with a dataset | |
754 | * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner | |
755 | * == tag. We will then release and reacquire ownership of the dataset while | |
756 | * holding the pool config_rwlock to avoid intervening namespace or ownership | |
757 | * changes may occur. | |
758 | * | |
759 | * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to | |
760 | * release the hold on its dataset and acquire a new one on the dataset of the | |
761 | * same name so that it can be partially torn down and reconstructed. | |
762 | */ | |
763 | void | |
b5256303 | 764 | dmu_objset_refresh_ownership(objset_t *os, boolean_t decrypt, void *tag) |
831baf06 KW |
765 | { |
766 | dsl_pool_t *dp; | |
767 | dsl_dataset_t *ds, *newds; | |
eca7b760 | 768 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
831baf06 KW |
769 | |
770 | ds = os->os_dsl_dataset; | |
771 | VERIFY3P(ds, !=, NULL); | |
772 | VERIFY3P(ds->ds_owner, ==, tag); | |
773 | VERIFY(dsl_dataset_long_held(ds)); | |
774 | ||
775 | dsl_dataset_name(ds, name); | |
776 | dp = dmu_objset_pool(os); | |
777 | dsl_pool_config_enter(dp, FTAG); | |
b5256303 TC |
778 | dmu_objset_disown(os, decrypt, tag); |
779 | VERIFY0(dsl_dataset_own(dp, name, | |
780 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag, &newds)); | |
831baf06 KW |
781 | VERIFY3P(newds, ==, os->os_dsl_dataset); |
782 | dsl_pool_config_exit(dp, FTAG); | |
783 | } | |
784 | ||
428870ff | 785 | void |
b5256303 | 786 | dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag) |
428870ff | 787 | { |
1de321e6 JX |
788 | /* |
789 | * Stop upgrading thread | |
790 | */ | |
791 | dmu_objset_upgrade_stop(os); | |
b5256303 TC |
792 | dsl_dataset_disown(os->os_dsl_dataset, |
793 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag); | |
34dc7c2f BB |
794 | } |
795 | ||
13fe0198 | 796 | void |
34dc7c2f BB |
797 | dmu_objset_evict_dbufs(objset_t *os) |
798 | { | |
0c66c32d | 799 | dnode_t *dn_marker; |
34dc7c2f BB |
800 | dnode_t *dn; |
801 | ||
0c66c32d | 802 | dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); |
34dc7c2f | 803 | |
0c66c32d JG |
804 | mutex_enter(&os->os_lock); |
805 | dn = list_head(&os->os_dnodes); | |
806 | while (dn != NULL) { | |
807 | /* | |
808 | * Skip dnodes without holds. We have to do this dance | |
809 | * because dnode_add_ref() only works if there is already a | |
810 | * hold. If the dnode has no holds, then it has no dbufs. | |
811 | */ | |
812 | if (dnode_add_ref(dn, FTAG)) { | |
813 | list_insert_after(&os->os_dnodes, dn, dn_marker); | |
814 | mutex_exit(&os->os_lock); | |
34dc7c2f | 815 | |
0c66c32d JG |
816 | dnode_evict_dbufs(dn); |
817 | dnode_rele(dn, FTAG); | |
34dc7c2f | 818 | |
0c66c32d JG |
819 | mutex_enter(&os->os_lock); |
820 | dn = list_next(&os->os_dnodes, dn_marker); | |
821 | list_remove(&os->os_dnodes, dn_marker); | |
822 | } else { | |
823 | dn = list_next(&os->os_dnodes, dn); | |
824 | } | |
825 | } | |
826 | mutex_exit(&os->os_lock); | |
34dc7c2f | 827 | |
0c66c32d | 828 | kmem_free(dn_marker, sizeof (dnode_t)); |
34dc7c2f | 829 | |
0c66c32d JG |
830 | if (DMU_USERUSED_DNODE(os) != NULL) { |
831 | dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); | |
832 | dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); | |
34dc7c2f | 833 | } |
0c66c32d | 834 | dnode_evict_dbufs(DMU_META_DNODE(os)); |
34dc7c2f BB |
835 | } |
836 | ||
0c66c32d JG |
837 | /* |
838 | * Objset eviction processing is split into into two pieces. | |
839 | * The first marks the objset as evicting, evicts any dbufs that | |
840 | * have a refcount of zero, and then queues up the objset for the | |
841 | * second phase of eviction. Once os->os_dnodes has been cleared by | |
842 | * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. | |
843 | * The second phase closes the special dnodes, dequeues the objset from | |
844 | * the list of those undergoing eviction, and finally frees the objset. | |
845 | * | |
846 | * NOTE: Due to asynchronous eviction processing (invocation of | |
847 | * dnode_buf_pageout()), it is possible for the meta dnode for the | |
848 | * objset to have no holds even though os->os_dnodes is not empty. | |
849 | */ | |
34dc7c2f | 850 | void |
428870ff | 851 | dmu_objset_evict(objset_t *os) |
34dc7c2f | 852 | { |
d6320ddb | 853 | int t; |
34dc7c2f | 854 | |
6f1ffb06 MA |
855 | dsl_dataset_t *ds = os->os_dsl_dataset; |
856 | ||
d6320ddb | 857 | for (t = 0; t < TXG_SIZE; t++) |
428870ff | 858 | ASSERT(!dmu_objset_is_dirty(os, t)); |
34dc7c2f | 859 | |
0eb21616 JG |
860 | if (ds) |
861 | dsl_prop_unregister_all(ds, os); | |
34dc7c2f | 862 | |
428870ff BB |
863 | if (os->os_sa) |
864 | sa_tear_down(os); | |
865 | ||
13fe0198 | 866 | dmu_objset_evict_dbufs(os); |
34dc7c2f | 867 | |
0c66c32d JG |
868 | mutex_enter(&os->os_lock); |
869 | spa_evicting_os_register(os->os_spa, os); | |
870 | if (list_is_empty(&os->os_dnodes)) { | |
871 | mutex_exit(&os->os_lock); | |
872 | dmu_objset_evict_done(os); | |
873 | } else { | |
874 | mutex_exit(&os->os_lock); | |
875 | } | |
b5256303 TC |
876 | |
877 | ||
0c66c32d JG |
878 | } |
879 | ||
880 | void | |
881 | dmu_objset_evict_done(objset_t *os) | |
882 | { | |
883 | ASSERT3P(list_head(&os->os_dnodes), ==, NULL); | |
884 | ||
572e2857 BB |
885 | dnode_special_close(&os->os_meta_dnode); |
886 | if (DMU_USERUSED_DNODE(os)) { | |
887 | dnode_special_close(&os->os_userused_dnode); | |
888 | dnode_special_close(&os->os_groupused_dnode); | |
9babb374 | 889 | } |
428870ff BB |
890 | zil_free(os->os_zil); |
891 | ||
d3c2ae1c | 892 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
572e2857 BB |
893 | |
894 | /* | |
895 | * This is a barrier to prevent the objset from going away in | |
896 | * dnode_move() until we can safely ensure that the objset is still in | |
897 | * use. We consider the objset valid before the barrier and invalid | |
898 | * after the barrier. | |
899 | */ | |
900 | rw_enter(&os_lock, RW_READER); | |
901 | rw_exit(&os_lock); | |
902 | ||
dbeb8796 MA |
903 | kmem_free(os->os_obj_next_percpu, |
904 | os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); | |
905 | ||
428870ff | 906 | mutex_destroy(&os->os_lock); |
64fc7762 | 907 | mutex_destroy(&os->os_userused_lock); |
428870ff BB |
908 | mutex_destroy(&os->os_obj_lock); |
909 | mutex_destroy(&os->os_user_ptr_lock); | |
c17486b2 | 910 | mutex_destroy(&os->os_upgrade_lock); |
64fc7762 MA |
911 | for (int i = 0; i < TXG_SIZE; i++) { |
912 | multilist_destroy(os->os_dirty_dnodes[i]); | |
913 | } | |
0c66c32d | 914 | spa_evicting_os_deregister(os->os_spa, os); |
428870ff BB |
915 | kmem_free(os, sizeof (objset_t)); |
916 | } | |
9babb374 | 917 | |
428870ff BB |
918 | timestruc_t |
919 | dmu_objset_snap_cmtime(objset_t *os) | |
920 | { | |
921 | return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); | |
34dc7c2f BB |
922 | } |
923 | ||
428870ff | 924 | objset_t * |
b5256303 TC |
925 | dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, |
926 | dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) | |
34dc7c2f | 927 | { |
428870ff | 928 | objset_t *os; |
34dc7c2f BB |
929 | dnode_t *mdn; |
930 | ||
931 | ASSERT(dmu_tx_is_syncing(tx)); | |
13fe0198 | 932 | |
b5256303 TC |
933 | if (blksz == 0) |
934 | blksz = DNODE_BLOCK_SIZE; | |
935 | if (blksz == 0) | |
936 | ibs = DN_MAX_INDBLKSHIFT; | |
937 | ||
572e2857 | 938 | if (ds != NULL) |
13fe0198 | 939 | VERIFY0(dmu_objset_from_ds(ds, &os)); |
572e2857 | 940 | else |
13fe0198 | 941 | VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); |
572e2857 BB |
942 | |
943 | mdn = DMU_META_DNODE(os); | |
34dc7c2f | 944 | |
b5256303 TC |
945 | dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, |
946 | DNODE_MIN_SLOTS, tx); | |
34dc7c2f BB |
947 | |
948 | /* | |
949 | * We don't want to have to increase the meta-dnode's nlevels | |
950 | * later, because then we could do it in quescing context while | |
951 | * we are also accessing it in open context. | |
952 | * | |
953 | * This precaution is not necessary for the MOS (ds == NULL), | |
954 | * because the MOS is only updated in syncing context. | |
955 | * This is most fortunate: the MOS is the only objset that | |
956 | * needs to be synced multiple times as spa_sync() iterates | |
957 | * to convergence, so minimizing its dn_nlevels matters. | |
958 | */ | |
959 | if (ds != NULL) { | |
b5256303 TC |
960 | if (levels == 0) { |
961 | levels = 1; | |
962 | ||
963 | /* | |
964 | * Determine the number of levels necessary for the | |
965 | * meta-dnode to contain DN_MAX_OBJECT dnodes. Note | |
966 | * that in order to ensure that we do not overflow | |
967 | * 64 bits, there has to be a nlevels that gives us a | |
968 | * number of blocks > DN_MAX_OBJECT but < 2^64. | |
969 | * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) | |
970 | * (10) must be less than (64 - log2(DN_MAX_OBJECT)) | |
971 | * (16). | |
972 | */ | |
973 | while ((uint64_t)mdn->dn_nblkptr << | |
974 | (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * | |
975 | (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < | |
976 | DN_MAX_OBJECT) | |
977 | levels++; | |
978 | } | |
34dc7c2f BB |
979 | |
980 | mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = | |
981 | mdn->dn_nlevels = levels; | |
982 | } | |
983 | ||
984 | ASSERT(type != DMU_OST_NONE); | |
985 | ASSERT(type != DMU_OST_ANY); | |
986 | ASSERT(type < DMU_OST_NUMTYPES); | |
428870ff | 987 | os->os_phys->os_type = type; |
b5256303 TC |
988 | |
989 | /* | |
990 | * Enable user accounting if it is enabled and this is not an | |
991 | * encrypted receive. | |
992 | */ | |
993 | if (dmu_objset_userused_enabled(os) && | |
994 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
428870ff | 995 | os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
1de321e6 JX |
996 | if (dmu_objset_userobjused_enabled(os)) { |
997 | ds->ds_feature_activation_needed[ | |
998 | SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE; | |
999 | os->os_phys->os_flags |= | |
1000 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
1001 | } | |
428870ff | 1002 | os->os_flags = os->os_phys->os_flags; |
9babb374 | 1003 | } |
34dc7c2f BB |
1004 | |
1005 | dsl_dataset_dirty(ds, tx); | |
1006 | ||
428870ff | 1007 | return (os); |
34dc7c2f BB |
1008 | } |
1009 | ||
b5256303 TC |
1010 | /* called from dsl for meta-objset */ |
1011 | objset_t * | |
1012 | dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
1013 | dmu_objset_type_t type, dmu_tx_t *tx) | |
1014 | { | |
1015 | return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); | |
1016 | } | |
1017 | ||
13fe0198 MA |
1018 | typedef struct dmu_objset_create_arg { |
1019 | const char *doca_name; | |
1020 | cred_t *doca_cred; | |
1021 | void (*doca_userfunc)(objset_t *os, void *arg, | |
1022 | cred_t *cr, dmu_tx_t *tx); | |
1023 | void *doca_userarg; | |
1024 | dmu_objset_type_t doca_type; | |
1025 | uint64_t doca_flags; | |
b5256303 | 1026 | dsl_crypto_params_t *doca_dcp; |
13fe0198 | 1027 | } dmu_objset_create_arg_t; |
34dc7c2f BB |
1028 | |
1029 | /*ARGSUSED*/ | |
1030 | static int | |
13fe0198 | 1031 | dmu_objset_create_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1032 | { |
13fe0198 MA |
1033 | dmu_objset_create_arg_t *doca = arg; |
1034 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1035 | dsl_dir_t *pdd; | |
1036 | const char *tail; | |
1037 | int error; | |
34dc7c2f | 1038 | |
13fe0198 | 1039 | if (strchr(doca->doca_name, '@') != NULL) |
2e528b49 | 1040 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 1041 | |
eca7b760 IK |
1042 | if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) |
1043 | return (SET_ERROR(ENAMETOOLONG)); | |
1044 | ||
13fe0198 MA |
1045 | error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); |
1046 | if (error != 0) | |
1047 | return (error); | |
1048 | if (tail == NULL) { | |
1049 | dsl_dir_rele(pdd, FTAG); | |
2e528b49 | 1050 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1051 | } |
b5256303 TC |
1052 | |
1053 | error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp); | |
1054 | if (error != 0) { | |
1055 | dsl_dir_rele(pdd, FTAG); | |
1056 | return (error); | |
1057 | } | |
1058 | ||
788eb90c JJ |
1059 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
1060 | doca->doca_cred); | |
b5256303 | 1061 | |
13fe0198 | 1062 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f | 1063 | |
788eb90c | 1064 | return (error); |
34dc7c2f BB |
1065 | } |
1066 | ||
1067 | static void | |
13fe0198 | 1068 | dmu_objset_create_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1069 | { |
13fe0198 MA |
1070 | dmu_objset_create_arg_t *doca = arg; |
1071 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1072 | dsl_dir_t *pdd; | |
1073 | const char *tail; | |
6f1ffb06 | 1074 | dsl_dataset_t *ds; |
13fe0198 | 1075 | uint64_t obj; |
6f1ffb06 | 1076 | blkptr_t *bp; |
13fe0198 | 1077 | objset_t *os; |
b5256303 | 1078 | zio_t *rzio; |
34dc7c2f | 1079 | |
13fe0198 | 1080 | VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); |
34dc7c2f | 1081 | |
13fe0198 | 1082 | obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, |
b5256303 | 1083 | doca->doca_cred, doca->doca_dcp, tx); |
34dc7c2f | 1084 | |
b5256303 TC |
1085 | VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, |
1086 | DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); | |
cc9bb3e5 | 1087 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
6f1ffb06 | 1088 | bp = dsl_dataset_get_blkptr(ds); |
13fe0198 MA |
1089 | os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, |
1090 | ds, bp, doca->doca_type, tx); | |
cc9bb3e5 | 1091 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
34dc7c2f | 1092 | |
13fe0198 MA |
1093 | if (doca->doca_userfunc != NULL) { |
1094 | doca->doca_userfunc(os, doca->doca_userarg, | |
1095 | doca->doca_cred, tx); | |
34dc7c2f BB |
1096 | } |
1097 | ||
b5256303 TC |
1098 | /* |
1099 | * The doca_userfunc() will write out some data that needs to be | |
1100 | * encrypted if the dataset is encrypted (specifically the root | |
1101 | * directory). This data must be written out before the encryption | |
1102 | * key mapping is removed by dsl_dataset_rele_flags(). Force the | |
1103 | * I/O to occur immediately by invoking the relevant sections of | |
1104 | * dsl_pool_sync(). | |
1105 | */ | |
1106 | if (os->os_encrypted) { | |
1107 | dsl_dataset_t *tmpds = NULL; | |
1108 | boolean_t need_sync_done = B_FALSE; | |
1109 | ||
1110 | rzio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
1111 | tmpds = txg_list_remove(&dp->dp_dirty_datasets, tx->tx_txg); | |
1112 | if (tmpds != NULL) { | |
1113 | ASSERT3P(ds, ==, tmpds); | |
1114 | dsl_dataset_sync(ds, rzio, tx); | |
1115 | need_sync_done = B_TRUE; | |
1116 | } | |
1117 | VERIFY0(zio_wait(rzio)); | |
1118 | ||
1119 | dmu_objset_do_userquota_updates(os, tx); | |
1120 | taskq_wait(dp->dp_sync_taskq); | |
1121 | ||
1122 | rzio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
1123 | tmpds = txg_list_remove(&dp->dp_dirty_datasets, tx->tx_txg); | |
1124 | if (tmpds != NULL) { | |
1125 | ASSERT3P(ds, ==, tmpds); | |
1126 | dmu_buf_rele(ds->ds_dbuf, ds); | |
1127 | dsl_dataset_sync(ds, rzio, tx); | |
1128 | } | |
1129 | VERIFY0(zio_wait(rzio)); | |
1130 | ||
1131 | if (need_sync_done) | |
1132 | dsl_dataset_sync_done(ds, tx); | |
1133 | } | |
1134 | ||
13fe0198 | 1135 | spa_history_log_internal_ds(ds, "create", tx, ""); |
a0bd735a BP |
1136 | zvol_create_minors(dp->dp_spa, doca->doca_name, B_TRUE); |
1137 | ||
b5256303 | 1138 | dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); |
13fe0198 | 1139 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f BB |
1140 | } |
1141 | ||
1142 | int | |
428870ff | 1143 | dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, |
b5256303 | 1144 | dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) |
34dc7c2f | 1145 | { |
13fe0198 | 1146 | dmu_objset_create_arg_t doca; |
b5256303 | 1147 | dsl_crypto_params_t tmp_dcp = { 0 }; |
34dc7c2f | 1148 | |
13fe0198 MA |
1149 | doca.doca_name = name; |
1150 | doca.doca_cred = CRED(); | |
1151 | doca.doca_flags = flags; | |
1152 | doca.doca_userfunc = func; | |
1153 | doca.doca_userarg = arg; | |
1154 | doca.doca_type = type; | |
34dc7c2f | 1155 | |
b5256303 TC |
1156 | /* |
1157 | * Some callers (mostly for testing) do not provide a dcp on their | |
1158 | * own but various code inside the sync task will require it to be | |
1159 | * allocated. Rather than adding NULL checks throughout this code | |
1160 | * or adding dummy dcp's to all of the callers we simply create a | |
1161 | * dummy one here and use that. This zero dcp will have the same | |
1162 | * effect as asking for inheritence of all encryption params. | |
1163 | */ | |
1164 | doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; | |
1165 | ||
13fe0198 | 1166 | return (dsl_sync_task(name, |
3d45fdd6 | 1167 | dmu_objset_create_check, dmu_objset_create_sync, &doca, |
b5256303 | 1168 | 6, ZFS_SPACE_CHECK_NORMAL)); |
34dc7c2f BB |
1169 | } |
1170 | ||
13fe0198 MA |
1171 | typedef struct dmu_objset_clone_arg { |
1172 | const char *doca_clone; | |
1173 | const char *doca_origin; | |
1174 | cred_t *doca_cred; | |
1175 | } dmu_objset_clone_arg_t; | |
1176 | ||
1177 | /*ARGSUSED*/ | |
1178 | static int | |
1179 | dmu_objset_clone_check(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1180 | { |
13fe0198 | 1181 | dmu_objset_clone_arg_t *doca = arg; |
428870ff BB |
1182 | dsl_dir_t *pdd; |
1183 | const char *tail; | |
13fe0198 MA |
1184 | int error; |
1185 | dsl_dataset_t *origin; | |
1186 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
34dc7c2f | 1187 | |
13fe0198 | 1188 | if (strchr(doca->doca_clone, '@') != NULL) |
2e528b49 | 1189 | return (SET_ERROR(EINVAL)); |
13fe0198 | 1190 | |
eca7b760 IK |
1191 | if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) |
1192 | return (SET_ERROR(ENAMETOOLONG)); | |
1193 | ||
13fe0198 MA |
1194 | error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); |
1195 | if (error != 0) | |
1196 | return (error); | |
428870ff | 1197 | if (tail == NULL) { |
13fe0198 | 1198 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1199 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1200 | } |
1cddb8c9 | 1201 | |
788eb90c JJ |
1202 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
1203 | doca->doca_cred); | |
1204 | if (error != 0) { | |
1205 | dsl_dir_rele(pdd, FTAG); | |
1206 | return (SET_ERROR(EDQUOT)); | |
1207 | } | |
428870ff | 1208 | |
13fe0198 | 1209 | error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); |
b5256303 TC |
1210 | if (error != 0) { |
1211 | dsl_dir_rele(pdd, FTAG); | |
572e2857 | 1212 | return (error); |
b5256303 | 1213 | } |
572e2857 | 1214 | |
13fe0198 | 1215 | /* You can only clone snapshots, not the head datasets. */ |
0c66c32d | 1216 | if (!origin->ds_is_snapshot) { |
13fe0198 | 1217 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1218 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1219 | return (SET_ERROR(EINVAL)); |
428870ff | 1220 | } |
b5256303 TC |
1221 | |
1222 | error = dmu_objset_clone_crypt_check(pdd, origin->ds_dir); | |
1223 | if (error != 0) { | |
1224 | dsl_dataset_rele(origin, FTAG); | |
1225 | dsl_dir_rele(pdd, FTAG); | |
1226 | return (error); | |
1227 | } | |
1228 | ||
13fe0198 | 1229 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1230 | dsl_dir_rele(pdd, FTAG); |
572e2857 | 1231 | |
13fe0198 | 1232 | return (0); |
9babb374 | 1233 | } |
34dc7c2f | 1234 | |
13fe0198 MA |
1235 | static void |
1236 | dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1237 | { |
13fe0198 MA |
1238 | dmu_objset_clone_arg_t *doca = arg; |
1239 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1240 | dsl_dir_t *pdd; | |
1241 | const char *tail; | |
1242 | dsl_dataset_t *origin, *ds; | |
1243 | uint64_t obj; | |
eca7b760 | 1244 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; |
6f1ffb06 | 1245 | |
13fe0198 MA |
1246 | VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); |
1247 | VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); | |
6f1ffb06 | 1248 | |
13fe0198 | 1249 | obj = dsl_dataset_create_sync(pdd, tail, origin, 0, |
b5256303 | 1250 | doca->doca_cred, NULL, tx); |
34dc7c2f | 1251 | |
13fe0198 MA |
1252 | VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); |
1253 | dsl_dataset_name(origin, namebuf); | |
1254 | spa_history_log_internal_ds(ds, "clone", tx, | |
1255 | "origin=%s (%llu)", namebuf, origin->ds_object); | |
a0bd735a | 1256 | zvol_create_minors(dp->dp_spa, doca->doca_clone, B_TRUE); |
13fe0198 MA |
1257 | dsl_dataset_rele(ds, FTAG); |
1258 | dsl_dataset_rele(origin, FTAG); | |
1259 | dsl_dir_rele(pdd, FTAG); | |
34dc7c2f BB |
1260 | } |
1261 | ||
1262 | int | |
13fe0198 | 1263 | dmu_objset_clone(const char *clone, const char *origin) |
34dc7c2f | 1264 | { |
13fe0198 | 1265 | dmu_objset_clone_arg_t doca; |
34dc7c2f | 1266 | |
13fe0198 MA |
1267 | doca.doca_clone = clone; |
1268 | doca.doca_origin = origin; | |
1269 | doca.doca_cred = CRED(); | |
572e2857 | 1270 | |
13fe0198 | 1271 | return (dsl_sync_task(clone, |
3d45fdd6 | 1272 | dmu_objset_clone_check, dmu_objset_clone_sync, &doca, |
b5256303 | 1273 | 6, ZFS_SPACE_CHECK_NORMAL)); |
6f1ffb06 MA |
1274 | } |
1275 | ||
1276 | int | |
1277 | dmu_objset_snapshot_one(const char *fsname, const char *snapname) | |
1278 | { | |
1279 | int err; | |
1280 | char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); | |
1281 | nvlist_t *snaps = fnvlist_alloc(); | |
1282 | ||
1283 | fnvlist_add_boolean(snaps, longsnap); | |
6f1ffb06 | 1284 | strfree(longsnap); |
13fe0198 MA |
1285 | err = dsl_dataset_snapshot(snaps, NULL, NULL); |
1286 | fnvlist_free(snaps); | |
6f1ffb06 MA |
1287 | return (err); |
1288 | } | |
1289 | ||
1de321e6 JX |
1290 | static void |
1291 | dmu_objset_upgrade_task_cb(void *data) | |
1292 | { | |
1293 | objset_t *os = data; | |
1294 | ||
1295 | mutex_enter(&os->os_upgrade_lock); | |
1296 | os->os_upgrade_status = EINTR; | |
1297 | if (!os->os_upgrade_exit) { | |
1298 | mutex_exit(&os->os_upgrade_lock); | |
1299 | ||
1300 | os->os_upgrade_status = os->os_upgrade_cb(os); | |
1301 | mutex_enter(&os->os_upgrade_lock); | |
1302 | } | |
1303 | os->os_upgrade_exit = B_TRUE; | |
1304 | os->os_upgrade_id = 0; | |
1305 | mutex_exit(&os->os_upgrade_lock); | |
1306 | } | |
1307 | ||
1308 | static void | |
1309 | dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) | |
1310 | { | |
1311 | if (os->os_upgrade_id != 0) | |
1312 | return; | |
1313 | ||
1314 | mutex_enter(&os->os_upgrade_lock); | |
1315 | if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { | |
1316 | os->os_upgrade_exit = B_FALSE; | |
1317 | os->os_upgrade_cb = cb; | |
1318 | os->os_upgrade_id = taskq_dispatch( | |
1319 | os->os_spa->spa_upgrade_taskq, | |
1320 | dmu_objset_upgrade_task_cb, os, TQ_SLEEP); | |
48d3eb40 | 1321 | if (os->os_upgrade_id == TASKQID_INVALID) |
1de321e6 JX |
1322 | os->os_upgrade_status = ENOMEM; |
1323 | } | |
1324 | mutex_exit(&os->os_upgrade_lock); | |
1325 | } | |
1326 | ||
1327 | static void | |
1328 | dmu_objset_upgrade_stop(objset_t *os) | |
1329 | { | |
1330 | mutex_enter(&os->os_upgrade_lock); | |
1331 | os->os_upgrade_exit = B_TRUE; | |
1332 | if (os->os_upgrade_id != 0) { | |
1333 | taskqid_t id = os->os_upgrade_id; | |
1334 | ||
1335 | os->os_upgrade_id = 0; | |
1336 | mutex_exit(&os->os_upgrade_lock); | |
1337 | ||
1338 | taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id); | |
1339 | } else { | |
1340 | mutex_exit(&os->os_upgrade_lock); | |
1341 | } | |
1342 | } | |
1343 | ||
34dc7c2f | 1344 | static void |
64fc7762 | 1345 | dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) |
34dc7c2f BB |
1346 | { |
1347 | dnode_t *dn; | |
1348 | ||
64fc7762 | 1349 | while ((dn = multilist_sublist_head(list)) != NULL) { |
34dc7c2f BB |
1350 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); |
1351 | ASSERT(dn->dn_dbuf->db_data_pending); | |
1352 | /* | |
9babb374 BB |
1353 | * Initialize dn_zio outside dnode_sync() because the |
1354 | * meta-dnode needs to set it ouside dnode_sync(). | |
34dc7c2f BB |
1355 | */ |
1356 | dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; | |
1357 | ASSERT(dn->dn_zio); | |
1358 | ||
1359 | ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); | |
64fc7762 | 1360 | multilist_sublist_remove(list, dn); |
9babb374 | 1361 | |
64fc7762 MA |
1362 | multilist_t *newlist = dn->dn_objset->os_synced_dnodes; |
1363 | if (newlist != NULL) { | |
9babb374 | 1364 | (void) dnode_add_ref(dn, newlist); |
64fc7762 | 1365 | multilist_insert(newlist, dn); |
9babb374 BB |
1366 | } |
1367 | ||
34dc7c2f BB |
1368 | dnode_sync(dn, tx); |
1369 | } | |
1370 | } | |
1371 | ||
1372 | /* ARGSUSED */ | |
1373 | static void | |
428870ff | 1374 | dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) |
34dc7c2f | 1375 | { |
d6320ddb BB |
1376 | int i; |
1377 | ||
b128c09f | 1378 | blkptr_t *bp = zio->io_bp; |
428870ff | 1379 | objset_t *os = arg; |
34dc7c2f | 1380 | dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; |
b5256303 | 1381 | uint64_t fill = 0; |
34dc7c2f | 1382 | |
9b67f605 | 1383 | ASSERT(!BP_IS_EMBEDDED(bp)); |
13fe0198 MA |
1384 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); |
1385 | ASSERT0(BP_GET_LEVEL(bp)); | |
34dc7c2f BB |
1386 | |
1387 | /* | |
9babb374 BB |
1388 | * Update rootbp fill count: it should be the number of objects |
1389 | * allocated in the object set (not counting the "special" | |
1390 | * objects that are stored in the objset_phys_t -- the meta | |
1391 | * dnode and user/group accounting objects). | |
34dc7c2f | 1392 | */ |
d6320ddb | 1393 | for (i = 0; i < dnp->dn_nblkptr; i++) |
b5256303 TC |
1394 | fill += BP_GET_FILL(&dnp->dn_blkptr[i]); |
1395 | ||
1396 | BP_SET_FILL(bp, fill); | |
1397 | ||
cc9bb3e5 GM |
1398 | if (os->os_dsl_dataset != NULL) |
1399 | rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); | |
1400 | *os->os_rootbp = *bp; | |
1401 | if (os->os_dsl_dataset != NULL) | |
1402 | rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); | |
428870ff BB |
1403 | } |
1404 | ||
1405 | /* ARGSUSED */ | |
1406 | static void | |
1407 | dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) | |
1408 | { | |
1409 | blkptr_t *bp = zio->io_bp; | |
1410 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1411 | objset_t *os = arg; | |
34dc7c2f | 1412 | |
b128c09f | 1413 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
428870ff | 1414 | ASSERT(BP_EQUAL(bp, bp_orig)); |
b128c09f | 1415 | } else { |
428870ff BB |
1416 | dsl_dataset_t *ds = os->os_dsl_dataset; |
1417 | dmu_tx_t *tx = os->os_synctx; | |
1418 | ||
1419 | (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); | |
1420 | dsl_dataset_block_born(ds, bp, tx); | |
34dc7c2f | 1421 | } |
cc9bb3e5 | 1422 | kmem_free(bp, sizeof (*bp)); |
34dc7c2f BB |
1423 | } |
1424 | ||
64fc7762 MA |
1425 | typedef struct sync_dnodes_arg { |
1426 | multilist_t *sda_list; | |
1427 | int sda_sublist_idx; | |
1428 | multilist_t *sda_newlist; | |
1429 | dmu_tx_t *sda_tx; | |
1430 | } sync_dnodes_arg_t; | |
1431 | ||
1432 | static void | |
1433 | sync_dnodes_task(void *arg) | |
1434 | { | |
1435 | sync_dnodes_arg_t *sda = arg; | |
1436 | ||
1437 | multilist_sublist_t *ms = | |
1438 | multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); | |
1439 | ||
1440 | dmu_objset_sync_dnodes(ms, sda->sda_tx); | |
1441 | ||
1442 | multilist_sublist_unlock(ms); | |
1443 | ||
1444 | kmem_free(sda, sizeof (*sda)); | |
1445 | } | |
1446 | ||
1447 | ||
34dc7c2f BB |
1448 | /* called from dsl */ |
1449 | void | |
428870ff | 1450 | dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) |
34dc7c2f BB |
1451 | { |
1452 | int txgoff; | |
5dbd68a3 | 1453 | zbookmark_phys_t zb; |
428870ff | 1454 | zio_prop_t zp; |
34dc7c2f BB |
1455 | zio_t *zio; |
1456 | list_t *list; | |
1457 | dbuf_dirty_record_t *dr; | |
cc9bb3e5 GM |
1458 | blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); |
1459 | *blkptr_copy = *os->os_rootbp; | |
34dc7c2f BB |
1460 | |
1461 | dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); | |
1462 | ||
1463 | ASSERT(dmu_tx_is_syncing(tx)); | |
1464 | /* XXX the write_done callback should really give us the tx... */ | |
1465 | os->os_synctx = tx; | |
1466 | ||
1467 | if (os->os_dsl_dataset == NULL) { | |
1468 | /* | |
1469 | * This is the MOS. If we have upgraded, | |
1470 | * spa_max_replication() could change, so reset | |
1471 | * os_copies here. | |
1472 | */ | |
1473 | os->os_copies = spa_max_replication(os->os_spa); | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * Create the root block IO | |
1478 | */ | |
428870ff BB |
1479 | SET_BOOKMARK(&zb, os->os_dsl_dataset ? |
1480 | os->os_dsl_dataset->ds_object : DMU_META_OBJSET, | |
1481 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
294f6806 | 1482 | arc_release(os->os_phys_buf, &os->os_phys_buf); |
b128c09f | 1483 | |
82644107 | 1484 | dmu_write_policy(os, NULL, 0, 0, &zp); |
9babb374 | 1485 | |
b5256303 TC |
1486 | /* |
1487 | * If we are either claiming the ZIL or doing a raw receive write out | |
1488 | * the os_phys_buf raw. Neither of these actions will effect the MAC | |
1489 | * at this point. | |
1490 | */ | |
1491 | if (arc_is_unauthenticated(os->os_phys_buf) || os->os_next_write_raw) { | |
1492 | ASSERT(os->os_encrypted); | |
1493 | os->os_next_write_raw = B_FALSE; | |
1494 | arc_convert_to_raw(os->os_phys_buf, | |
1495 | os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, | |
1496 | DMU_OT_OBJSET, NULL, NULL, NULL); | |
1497 | } | |
1498 | ||
428870ff | 1499 | zio = arc_write(pio, os->os_spa, tx->tx_txg, |
cc9bb3e5 | 1500 | blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), |
bc77ba73 PD |
1501 | &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, |
1502 | os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); | |
34dc7c2f BB |
1503 | |
1504 | /* | |
9babb374 | 1505 | * Sync special dnodes - the parent IO for the sync is the root block |
34dc7c2f | 1506 | */ |
572e2857 BB |
1507 | DMU_META_DNODE(os)->dn_zio = zio; |
1508 | dnode_sync(DMU_META_DNODE(os), tx); | |
34dc7c2f | 1509 | |
9babb374 BB |
1510 | os->os_phys->os_flags = os->os_flags; |
1511 | ||
572e2857 BB |
1512 | if (DMU_USERUSED_DNODE(os) && |
1513 | DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1514 | DMU_USERUSED_DNODE(os)->dn_zio = zio; | |
1515 | dnode_sync(DMU_USERUSED_DNODE(os), tx); | |
1516 | DMU_GROUPUSED_DNODE(os)->dn_zio = zio; | |
1517 | dnode_sync(DMU_GROUPUSED_DNODE(os), tx); | |
9babb374 BB |
1518 | } |
1519 | ||
34dc7c2f BB |
1520 | txgoff = tx->tx_txg & TXG_MASK; |
1521 | ||
b5256303 TC |
1522 | if (dmu_objset_userused_enabled(os) && |
1523 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
9babb374 BB |
1524 | /* |
1525 | * We must create the list here because it uses the | |
64fc7762 MA |
1526 | * dn_dirty_link[] of this txg. But it may already |
1527 | * exist because we call dsl_dataset_sync() twice per txg. | |
9babb374 | 1528 | */ |
64fc7762 MA |
1529 | if (os->os_synced_dnodes == NULL) { |
1530 | os->os_synced_dnodes = | |
1531 | multilist_create(sizeof (dnode_t), | |
1532 | offsetof(dnode_t, dn_dirty_link[txgoff]), | |
1533 | dnode_multilist_index_func); | |
1534 | } else { | |
1535 | ASSERT3U(os->os_synced_dnodes->ml_offset, ==, | |
1536 | offsetof(dnode_t, dn_dirty_link[txgoff])); | |
1537 | } | |
9babb374 BB |
1538 | } |
1539 | ||
64fc7762 MA |
1540 | for (int i = 0; |
1541 | i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { | |
1542 | sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); | |
1543 | sda->sda_list = os->os_dirty_dnodes[txgoff]; | |
1544 | sda->sda_sublist_idx = i; | |
1545 | sda->sda_tx = tx; | |
1546 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1547 | sync_dnodes_task, sda, 0); | |
1548 | /* callback frees sda */ | |
1549 | } | |
1550 | taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); | |
34dc7c2f | 1551 | |
572e2857 | 1552 | list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; |
64fc7762 | 1553 | while ((dr = list_head(list)) != NULL) { |
13fe0198 | 1554 | ASSERT0(dr->dr_dbuf->db_level); |
34dc7c2f BB |
1555 | list_remove(list, dr); |
1556 | if (dr->dr_zio) | |
1557 | zio_nowait(dr->dr_zio); | |
1558 | } | |
68cbd56e NB |
1559 | |
1560 | /* Enable dnode backfill if enough objects have been freed. */ | |
1561 | if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { | |
1562 | os->os_rescan_dnodes = B_TRUE; | |
1563 | os->os_freed_dnodes = 0; | |
1564 | } | |
1565 | ||
34dc7c2f BB |
1566 | /* |
1567 | * Free intent log blocks up to this tx. | |
1568 | */ | |
1569 | zil_sync(os->os_zil, tx); | |
b128c09f | 1570 | os->os_phys->os_zil_header = os->os_zil_header; |
34dc7c2f BB |
1571 | zio_nowait(zio); |
1572 | } | |
1573 | ||
428870ff BB |
1574 | boolean_t |
1575 | dmu_objset_is_dirty(objset_t *os, uint64_t txg) | |
1576 | { | |
64fc7762 | 1577 | return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); |
428870ff BB |
1578 | } |
1579 | ||
572e2857 | 1580 | static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; |
9babb374 BB |
1581 | |
1582 | void | |
1583 | dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) | |
1584 | { | |
1585 | used_cbs[ost] = cb; | |
1586 | } | |
1587 | ||
1588 | boolean_t | |
428870ff | 1589 | dmu_objset_userused_enabled(objset_t *os) |
9babb374 BB |
1590 | { |
1591 | return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && | |
572e2857 BB |
1592 | used_cbs[os->os_phys->os_type] != NULL && |
1593 | DMU_USERUSED_DNODE(os) != NULL); | |
9babb374 BB |
1594 | } |
1595 | ||
1de321e6 JX |
1596 | boolean_t |
1597 | dmu_objset_userobjused_enabled(objset_t *os) | |
1598 | { | |
1599 | return (dmu_objset_userused_enabled(os) && | |
1600 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); | |
1601 | } | |
1602 | ||
9b7a83cb JX |
1603 | typedef struct userquota_node { |
1604 | /* must be in the first filed, see userquota_update_cache() */ | |
1605 | char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1606 | int64_t uqn_delta; | |
1607 | avl_node_t uqn_node; | |
1608 | } userquota_node_t; | |
1609 | ||
1610 | typedef struct userquota_cache { | |
1611 | avl_tree_t uqc_user_deltas; | |
1612 | avl_tree_t uqc_group_deltas; | |
1613 | } userquota_cache_t; | |
1614 | ||
1615 | static int | |
1616 | userquota_compare(const void *l, const void *r) | |
1617 | { | |
1618 | const userquota_node_t *luqn = l; | |
1619 | const userquota_node_t *ruqn = r; | |
e4ffa98d | 1620 | int rv; |
9b7a83cb JX |
1621 | |
1622 | /* | |
1623 | * NB: can only access uqn_id because userquota_update_cache() doesn't | |
1624 | * pass in an entire userquota_node_t. | |
1625 | */ | |
e4ffa98d BB |
1626 | rv = strcmp(luqn->uqn_id, ruqn->uqn_id); |
1627 | ||
1628 | return (AVL_ISIGN(rv)); | |
9b7a83cb JX |
1629 | } |
1630 | ||
1631 | static void | |
1632 | do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) | |
1633 | { | |
1634 | void *cookie; | |
1635 | userquota_node_t *uqn; | |
1636 | ||
1637 | ASSERT(dmu_tx_is_syncing(tx)); | |
1638 | ||
1639 | cookie = NULL; | |
1640 | while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, | |
1641 | &cookie)) != NULL) { | |
64fc7762 MA |
1642 | /* |
1643 | * os_userused_lock protects against concurrent calls to | |
1644 | * zap_increment_int(). It's needed because zap_increment_int() | |
1645 | * is not thread-safe (i.e. not atomic). | |
1646 | */ | |
1647 | mutex_enter(&os->os_userused_lock); | |
9b7a83cb JX |
1648 | VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, |
1649 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1650 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1651 | kmem_free(uqn, sizeof (*uqn)); |
1652 | } | |
1653 | avl_destroy(&cache->uqc_user_deltas); | |
1654 | ||
1655 | cookie = NULL; | |
1656 | while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, | |
1657 | &cookie)) != NULL) { | |
64fc7762 | 1658 | mutex_enter(&os->os_userused_lock); |
9b7a83cb JX |
1659 | VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, |
1660 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1661 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1662 | kmem_free(uqn, sizeof (*uqn)); |
1663 | } | |
1664 | avl_destroy(&cache->uqc_group_deltas); | |
1665 | } | |
1666 | ||
1667 | static void | |
1668 | userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) | |
1669 | { | |
1670 | userquota_node_t *uqn; | |
1671 | avl_index_t idx; | |
1672 | ||
1673 | ASSERT(strlen(id) < sizeof (uqn->uqn_id)); | |
1674 | /* | |
1675 | * Use id directly for searching because uqn_id is the first field of | |
1676 | * userquota_node_t and fields after uqn_id won't be accessed in | |
1677 | * avl_find(). | |
1678 | */ | |
1679 | uqn = avl_find(avl, (const void *)id, &idx); | |
1680 | if (uqn == NULL) { | |
1681 | uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); | |
1f51b525 | 1682 | strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); |
9b7a83cb JX |
1683 | avl_insert(avl, uqn, idx); |
1684 | } | |
1685 | uqn->uqn_delta += delta; | |
1686 | } | |
1687 | ||
428870ff | 1688 | static void |
9b7a83cb JX |
1689 | do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags, |
1690 | uint64_t user, uint64_t group, boolean_t subtract) | |
428870ff BB |
1691 | { |
1692 | if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { | |
50c957f7 | 1693 | int64_t delta = DNODE_MIN_SIZE + used; |
9b7a83cb JX |
1694 | char name[20]; |
1695 | ||
428870ff BB |
1696 | if (subtract) |
1697 | delta = -delta; | |
9b7a83cb JX |
1698 | |
1699 | (void) sprintf(name, "%llx", (longlong_t)user); | |
1700 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); | |
1701 | ||
1702 | (void) sprintf(name, "%llx", (longlong_t)group); | |
1703 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); | |
428870ff BB |
1704 | } |
1705 | } | |
1706 | ||
1de321e6 | 1707 | static void |
9b7a83cb JX |
1708 | do_userobjquota_update(userquota_cache_t *cache, uint64_t flags, |
1709 | uint64_t user, uint64_t group, boolean_t subtract) | |
1de321e6 JX |
1710 | { |
1711 | if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { | |
1712 | char name[20 + DMU_OBJACCT_PREFIX_LEN]; | |
9b7a83cb | 1713 | int delta = subtract ? -1 : 1; |
1de321e6 JX |
1714 | |
1715 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1716 | (longlong_t)user); | |
9b7a83cb | 1717 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); |
1de321e6 JX |
1718 | |
1719 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1720 | (longlong_t)group); | |
9b7a83cb | 1721 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); |
1de321e6 JX |
1722 | } |
1723 | } | |
1724 | ||
64fc7762 MA |
1725 | typedef struct userquota_updates_arg { |
1726 | objset_t *uua_os; | |
1727 | int uua_sublist_idx; | |
1728 | dmu_tx_t *uua_tx; | |
1729 | } userquota_updates_arg_t; | |
1730 | ||
1731 | static void | |
1732 | userquota_updates_task(void *arg) | |
9babb374 | 1733 | { |
64fc7762 MA |
1734 | userquota_updates_arg_t *uua = arg; |
1735 | objset_t *os = uua->uua_os; | |
1736 | dmu_tx_t *tx = uua->uua_tx; | |
9babb374 | 1737 | dnode_t *dn; |
9b7a83cb | 1738 | userquota_cache_t cache = { { 0 } }; |
9babb374 | 1739 | |
64fc7762 MA |
1740 | multilist_sublist_t *list = |
1741 | multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); | |
9babb374 | 1742 | |
64fc7762 MA |
1743 | ASSERT(multilist_sublist_head(list) == NULL || |
1744 | dmu_objset_userused_enabled(os)); | |
9b7a83cb JX |
1745 | avl_create(&cache.uqc_user_deltas, userquota_compare, |
1746 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
1747 | avl_create(&cache.uqc_group_deltas, userquota_compare, | |
1748 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
1749 | ||
64fc7762 | 1750 | while ((dn = multilist_sublist_head(list)) != NULL) { |
572e2857 | 1751 | int flags; |
9babb374 | 1752 | ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); |
9babb374 BB |
1753 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || |
1754 | dn->dn_phys->dn_flags & | |
1755 | DNODE_FLAG_USERUSED_ACCOUNTED); | |
1756 | ||
572e2857 BB |
1757 | flags = dn->dn_id_flags; |
1758 | ASSERT(flags); | |
1759 | if (flags & DN_ID_OLD_EXIST) { | |
9b7a83cb JX |
1760 | do_userquota_update(&cache, |
1761 | dn->dn_oldused, dn->dn_oldflags, | |
1762 | dn->dn_olduid, dn->dn_oldgid, B_TRUE); | |
1763 | do_userobjquota_update(&cache, dn->dn_oldflags, | |
1764 | dn->dn_olduid, dn->dn_oldgid, B_TRUE); | |
428870ff | 1765 | } |
572e2857 | 1766 | if (flags & DN_ID_NEW_EXIST) { |
9b7a83cb JX |
1767 | do_userquota_update(&cache, |
1768 | DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, | |
1769 | dn->dn_newuid, dn->dn_newgid, B_FALSE); | |
1770 | do_userobjquota_update(&cache, dn->dn_phys->dn_flags, | |
1771 | dn->dn_newuid, dn->dn_newgid, B_FALSE); | |
428870ff BB |
1772 | } |
1773 | ||
572e2857 | 1774 | mutex_enter(&dn->dn_mtx); |
428870ff BB |
1775 | dn->dn_oldused = 0; |
1776 | dn->dn_oldflags = 0; | |
1777 | if (dn->dn_id_flags & DN_ID_NEW_EXIST) { | |
1778 | dn->dn_olduid = dn->dn_newuid; | |
1779 | dn->dn_oldgid = dn->dn_newgid; | |
1780 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
1781 | if (dn->dn_bonuslen == 0) | |
1782 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
1783 | else | |
1784 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
1785 | } | |
1786 | dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); | |
9babb374 BB |
1787 | mutex_exit(&dn->dn_mtx); |
1788 | ||
64fc7762 MA |
1789 | multilist_sublist_remove(list, dn); |
1790 | dnode_rele(dn, os->os_synced_dnodes); | |
9babb374 | 1791 | } |
9b7a83cb | 1792 | do_userquota_cacheflush(os, &cache, tx); |
64fc7762 MA |
1793 | multilist_sublist_unlock(list); |
1794 | kmem_free(uua, sizeof (*uua)); | |
1795 | } | |
1796 | ||
1797 | void | |
1798 | dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) | |
1799 | { | |
1800 | if (!dmu_objset_userused_enabled(os)) | |
1801 | return; | |
1802 | ||
b5256303 TC |
1803 | /* if this is a raw receive just return and handle accounting later */ |
1804 | if (os->os_encrypted && dmu_objset_is_receiving(os)) | |
1805 | return; | |
1806 | ||
64fc7762 MA |
1807 | /* Allocate the user/groupused objects if necessary. */ |
1808 | if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { | |
1809 | VERIFY0(zap_create_claim(os, | |
1810 | DMU_USERUSED_OBJECT, | |
1811 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
1812 | VERIFY0(zap_create_claim(os, | |
1813 | DMU_GROUPUSED_OBJECT, | |
1814 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
1815 | } | |
1816 | ||
1817 | for (int i = 0; | |
1818 | i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { | |
1819 | userquota_updates_arg_t *uua = | |
1820 | kmem_alloc(sizeof (*uua), KM_SLEEP); | |
1821 | uua->uua_os = os; | |
1822 | uua->uua_sublist_idx = i; | |
1823 | uua->uua_tx = tx; | |
1824 | /* note: caller does taskq_wait() */ | |
1825 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1826 | userquota_updates_task, uua, 0); | |
1827 | /* callback frees uua */ | |
1828 | } | |
9babb374 BB |
1829 | } |
1830 | ||
428870ff BB |
1831 | /* |
1832 | * Returns a pointer to data to find uid/gid from | |
1833 | * | |
1834 | * If a dirty record for transaction group that is syncing can't | |
1835 | * be found then NULL is returned. In the NULL case it is assumed | |
1836 | * the uid/gid aren't changing. | |
1837 | */ | |
1838 | static void * | |
1839 | dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) | |
1840 | { | |
1841 | dbuf_dirty_record_t *dr, **drp; | |
1842 | void *data; | |
1843 | ||
1844 | if (db->db_dirtycnt == 0) | |
1845 | return (db->db.db_data); /* Nothing is changing */ | |
1846 | ||
1847 | for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) | |
1848 | if (dr->dr_txg == tx->tx_txg) | |
1849 | break; | |
1850 | ||
572e2857 | 1851 | if (dr == NULL) { |
428870ff | 1852 | data = NULL; |
572e2857 BB |
1853 | } else { |
1854 | dnode_t *dn; | |
1855 | ||
1856 | DB_DNODE_ENTER(dr->dr_dbuf); | |
1857 | dn = DB_DNODE(dr->dr_dbuf); | |
1858 | ||
1859 | if (dn->dn_bonuslen == 0 && | |
1860 | dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) | |
1861 | data = dr->dt.dl.dr_data->b_data; | |
1862 | else | |
1863 | data = dr->dt.dl.dr_data; | |
1864 | ||
1865 | DB_DNODE_EXIT(dr->dr_dbuf); | |
1866 | } | |
1867 | ||
428870ff BB |
1868 | return (data); |
1869 | } | |
1870 | ||
1871 | void | |
1872 | dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) | |
1873 | { | |
1874 | objset_t *os = dn->dn_objset; | |
1875 | void *data = NULL; | |
1876 | dmu_buf_impl_t *db = NULL; | |
a117a6d6 GW |
1877 | uint64_t *user = NULL; |
1878 | uint64_t *group = NULL; | |
428870ff BB |
1879 | int flags = dn->dn_id_flags; |
1880 | int error; | |
1881 | boolean_t have_spill = B_FALSE; | |
1882 | ||
1883 | if (!dmu_objset_userused_enabled(dn->dn_objset)) | |
1884 | return; | |
1885 | ||
b5256303 TC |
1886 | /* |
1887 | * Raw receives introduce a problem with user accounting. Raw | |
1888 | * receives cannot update the user accounting info because the | |
1889 | * user ids and the sizes are encrypted. To guarantee that we | |
1890 | * never end up with bad user accounting, we simply disable it | |
1891 | * during raw receives. We also disable this for normal receives | |
1892 | * so that an incremental raw receive may be done on top of an | |
1893 | * existing non-raw receive. | |
1894 | */ | |
1895 | if (os->os_encrypted && dmu_objset_is_receiving(os)) | |
1896 | return; | |
1897 | ||
428870ff BB |
1898 | if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| |
1899 | DN_ID_CHKED_SPILL))) | |
1900 | return; | |
1901 | ||
1902 | if (before && dn->dn_bonuslen != 0) | |
1903 | data = DN_BONUS(dn->dn_phys); | |
1904 | else if (!before && dn->dn_bonuslen != 0) { | |
a3000f93 BB |
1905 | if (dn->dn_bonus) { |
1906 | db = dn->dn_bonus; | |
428870ff BB |
1907 | mutex_enter(&db->db_mtx); |
1908 | data = dmu_objset_userquota_find_data(db, tx); | |
1909 | } else { | |
1910 | data = DN_BONUS(dn->dn_phys); | |
1911 | } | |
1912 | } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { | |
1913 | int rf = 0; | |
1914 | ||
1915 | if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) | |
1916 | rf |= DB_RF_HAVESTRUCT; | |
572e2857 BB |
1917 | error = dmu_spill_hold_by_dnode(dn, |
1918 | rf | DB_RF_MUST_SUCCEED, | |
428870ff BB |
1919 | FTAG, (dmu_buf_t **)&db); |
1920 | ASSERT(error == 0); | |
1921 | mutex_enter(&db->db_mtx); | |
1922 | data = (before) ? db->db.db_data : | |
1923 | dmu_objset_userquota_find_data(db, tx); | |
1924 | have_spill = B_TRUE; | |
1925 | } else { | |
1926 | mutex_enter(&dn->dn_mtx); | |
1927 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
1928 | mutex_exit(&dn->dn_mtx); | |
1929 | return; | |
1930 | } | |
1931 | ||
1932 | if (before) { | |
1933 | ASSERT(data); | |
1934 | user = &dn->dn_olduid; | |
1935 | group = &dn->dn_oldgid; | |
1936 | } else if (data) { | |
1937 | user = &dn->dn_newuid; | |
1938 | group = &dn->dn_newgid; | |
1939 | } | |
1940 | ||
1941 | /* | |
1942 | * Must always call the callback in case the object | |
1943 | * type has changed and that type isn't an object type to track | |
1944 | */ | |
1945 | error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, | |
1946 | user, group); | |
1947 | ||
1948 | /* | |
1949 | * Preserve existing uid/gid when the callback can't determine | |
1950 | * what the new uid/gid are and the callback returned EEXIST. | |
1951 | * The EEXIST error tells us to just use the existing uid/gid. | |
1952 | * If we don't know what the old values are then just assign | |
1953 | * them to 0, since that is a new file being created. | |
1954 | */ | |
1955 | if (!before && data == NULL && error == EEXIST) { | |
1956 | if (flags & DN_ID_OLD_EXIST) { | |
1957 | dn->dn_newuid = dn->dn_olduid; | |
1958 | dn->dn_newgid = dn->dn_oldgid; | |
1959 | } else { | |
1960 | dn->dn_newuid = 0; | |
1961 | dn->dn_newgid = 0; | |
1962 | } | |
1963 | error = 0; | |
1964 | } | |
1965 | ||
1966 | if (db) | |
1967 | mutex_exit(&db->db_mtx); | |
1968 | ||
1969 | mutex_enter(&dn->dn_mtx); | |
1970 | if (error == 0 && before) | |
1971 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
1972 | if (error == 0 && !before) | |
1973 | dn->dn_id_flags |= DN_ID_NEW_EXIST; | |
1974 | ||
1975 | if (have_spill) { | |
1976 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
1977 | } else { | |
1978 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
1979 | } | |
1980 | mutex_exit(&dn->dn_mtx); | |
a3000f93 | 1981 | if (have_spill) |
428870ff BB |
1982 | dmu_buf_rele((dmu_buf_t *)db, FTAG); |
1983 | } | |
1984 | ||
9babb374 BB |
1985 | boolean_t |
1986 | dmu_objset_userspace_present(objset_t *os) | |
1987 | { | |
428870ff | 1988 | return (os->os_phys->os_flags & |
9babb374 BB |
1989 | OBJSET_FLAG_USERACCOUNTING_COMPLETE); |
1990 | } | |
1991 | ||
1de321e6 JX |
1992 | boolean_t |
1993 | dmu_objset_userobjspace_present(objset_t *os) | |
1994 | { | |
1995 | return (os->os_phys->os_flags & | |
1996 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); | |
1997 | } | |
1998 | ||
1999 | static int | |
2000 | dmu_objset_space_upgrade(objset_t *os) | |
9babb374 BB |
2001 | { |
2002 | uint64_t obj; | |
2003 | int err = 0; | |
2004 | ||
9babb374 BB |
2005 | /* |
2006 | * We simply need to mark every object dirty, so that it will be | |
2007 | * synced out and now accounted. If this is called | |
2008 | * concurrently, or if we already did some work before crashing, | |
2009 | * that's fine, since we track each object's accounted state | |
2010 | * independently. | |
2011 | */ | |
2012 | ||
2013 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { | |
45d1cae3 | 2014 | dmu_tx_t *tx; |
9babb374 BB |
2015 | dmu_buf_t *db; |
2016 | int objerr; | |
2017 | ||
1de321e6 JX |
2018 | mutex_enter(&os->os_upgrade_lock); |
2019 | if (os->os_upgrade_exit) | |
2020 | err = SET_ERROR(EINTR); | |
2021 | mutex_exit(&os->os_upgrade_lock); | |
2022 | if (err != 0) | |
2023 | return (err); | |
2024 | ||
9babb374 | 2025 | if (issig(JUSTLOOKING) && issig(FORREAL)) |
2e528b49 | 2026 | return (SET_ERROR(EINTR)); |
9babb374 BB |
2027 | |
2028 | objerr = dmu_bonus_hold(os, obj, FTAG, &db); | |
13fe0198 | 2029 | if (objerr != 0) |
9babb374 | 2030 | continue; |
45d1cae3 | 2031 | tx = dmu_tx_create(os); |
9babb374 BB |
2032 | dmu_tx_hold_bonus(tx, obj); |
2033 | objerr = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 2034 | if (objerr != 0) { |
9babb374 BB |
2035 | dmu_tx_abort(tx); |
2036 | continue; | |
2037 | } | |
2038 | dmu_buf_will_dirty(db, tx); | |
2039 | dmu_buf_rele(db, FTAG); | |
2040 | dmu_tx_commit(tx); | |
2041 | } | |
1de321e6 JX |
2042 | return (0); |
2043 | } | |
2044 | ||
2045 | int | |
2046 | dmu_objset_userspace_upgrade(objset_t *os) | |
2047 | { | |
2048 | int err = 0; | |
2049 | ||
2050 | if (dmu_objset_userspace_present(os)) | |
2051 | return (0); | |
2052 | if (dmu_objset_is_snapshot(os)) | |
2053 | return (SET_ERROR(EINVAL)); | |
2054 | if (!dmu_objset_userused_enabled(os)) | |
2055 | return (SET_ERROR(ENOTSUP)); | |
2056 | ||
2057 | err = dmu_objset_space_upgrade(os); | |
2058 | if (err) | |
2059 | return (err); | |
9babb374 | 2060 | |
428870ff | 2061 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
9babb374 BB |
2062 | txg_wait_synced(dmu_objset_pool(os), 0); |
2063 | return (0); | |
2064 | } | |
2065 | ||
1de321e6 JX |
2066 | static int |
2067 | dmu_objset_userobjspace_upgrade_cb(objset_t *os) | |
2068 | { | |
2069 | int err = 0; | |
2070 | ||
2071 | if (dmu_objset_userobjspace_present(os)) | |
2072 | return (0); | |
2073 | if (dmu_objset_is_snapshot(os)) | |
2074 | return (SET_ERROR(EINVAL)); | |
2075 | if (!dmu_objset_userobjused_enabled(os)) | |
2076 | return (SET_ERROR(ENOTSUP)); | |
2077 | ||
2078 | dmu_objset_ds(os)->ds_feature_activation_needed[ | |
2079 | SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE; | |
2080 | ||
2081 | err = dmu_objset_space_upgrade(os); | |
2082 | if (err) | |
2083 | return (err); | |
2084 | ||
2085 | os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
2086 | txg_wait_synced(dmu_objset_pool(os), 0); | |
2087 | return (0); | |
2088 | } | |
2089 | ||
2090 | void | |
2091 | dmu_objset_userobjspace_upgrade(objset_t *os) | |
2092 | { | |
2093 | dmu_objset_upgrade(os, dmu_objset_userobjspace_upgrade_cb); | |
2094 | } | |
2095 | ||
126ae9f4 JX |
2096 | boolean_t |
2097 | dmu_objset_userobjspace_upgradable(objset_t *os) | |
2098 | { | |
2099 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2100 | !dmu_objset_is_snapshot(os) && | |
2101 | dmu_objset_userobjused_enabled(os) && | |
2102 | !dmu_objset_userobjspace_present(os)); | |
2103 | } | |
2104 | ||
34dc7c2f BB |
2105 | void |
2106 | dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, | |
2107 | uint64_t *usedobjsp, uint64_t *availobjsp) | |
2108 | { | |
428870ff | 2109 | dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, |
34dc7c2f BB |
2110 | usedobjsp, availobjsp); |
2111 | } | |
2112 | ||
2113 | uint64_t | |
2114 | dmu_objset_fsid_guid(objset_t *os) | |
2115 | { | |
428870ff | 2116 | return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); |
34dc7c2f BB |
2117 | } |
2118 | ||
2119 | void | |
2120 | dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) | |
2121 | { | |
428870ff BB |
2122 | stat->dds_type = os->os_phys->os_type; |
2123 | if (os->os_dsl_dataset) | |
2124 | dsl_dataset_fast_stat(os->os_dsl_dataset, stat); | |
34dc7c2f BB |
2125 | } |
2126 | ||
2127 | void | |
2128 | dmu_objset_stats(objset_t *os, nvlist_t *nv) | |
2129 | { | |
428870ff BB |
2130 | ASSERT(os->os_dsl_dataset || |
2131 | os->os_phys->os_type == DMU_OST_META); | |
34dc7c2f | 2132 | |
428870ff BB |
2133 | if (os->os_dsl_dataset != NULL) |
2134 | dsl_dataset_stats(os->os_dsl_dataset, nv); | |
34dc7c2f BB |
2135 | |
2136 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, | |
428870ff | 2137 | os->os_phys->os_type); |
9babb374 BB |
2138 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, |
2139 | dmu_objset_userspace_present(os)); | |
34dc7c2f BB |
2140 | } |
2141 | ||
2142 | int | |
2143 | dmu_objset_is_snapshot(objset_t *os) | |
2144 | { | |
428870ff | 2145 | if (os->os_dsl_dataset != NULL) |
0c66c32d | 2146 | return (os->os_dsl_dataset->ds_is_snapshot); |
34dc7c2f BB |
2147 | else |
2148 | return (B_FALSE); | |
2149 | } | |
2150 | ||
2151 | int | |
2152 | dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, | |
2153 | boolean_t *conflict) | |
2154 | { | |
428870ff | 2155 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2156 | uint64_t ignored; |
2157 | ||
d683ddbb | 2158 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2159 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2160 | |
2161 | return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2162 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, |
9b7b9cd3 | 2163 | MT_NORMALIZE, real, maxlen, conflict)); |
34dc7c2f BB |
2164 | } |
2165 | ||
2166 | int | |
2167 | dmu_snapshot_list_next(objset_t *os, int namelen, char *name, | |
2168 | uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) | |
2169 | { | |
428870ff | 2170 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2171 | zap_cursor_t cursor; |
2172 | zap_attribute_t attr; | |
2173 | ||
13fe0198 MA |
2174 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
2175 | ||
d683ddbb | 2176 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2177 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2178 | |
2179 | zap_cursor_init_serialized(&cursor, | |
2180 | ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2181 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); |
34dc7c2f BB |
2182 | |
2183 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2184 | zap_cursor_fini(&cursor); | |
2e528b49 | 2185 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2186 | } |
2187 | ||
2188 | if (strlen(attr.za_name) + 1 > namelen) { | |
2189 | zap_cursor_fini(&cursor); | |
2e528b49 | 2190 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2191 | } |
2192 | ||
2193 | (void) strcpy(name, attr.za_name); | |
2194 | if (idp) | |
2195 | *idp = attr.za_first_integer; | |
2196 | if (case_conflict) | |
2197 | *case_conflict = attr.za_normalization_conflict; | |
2198 | zap_cursor_advance(&cursor); | |
2199 | *offp = zap_cursor_serialize(&cursor); | |
2200 | zap_cursor_fini(&cursor); | |
2201 | ||
2202 | return (0); | |
2203 | } | |
2204 | ||
ebe7e575 | 2205 | int |
6772fb67 | 2206 | dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) |
ebe7e575 | 2207 | { |
d1d7e268 | 2208 | return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); |
ebe7e575 BB |
2209 | } |
2210 | ||
34dc7c2f BB |
2211 | int |
2212 | dmu_dir_list_next(objset_t *os, int namelen, char *name, | |
2213 | uint64_t *idp, uint64_t *offp) | |
2214 | { | |
428870ff | 2215 | dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; |
34dc7c2f BB |
2216 | zap_cursor_t cursor; |
2217 | zap_attribute_t attr; | |
2218 | ||
2219 | /* there is no next dir on a snapshot! */ | |
428870ff | 2220 | if (os->os_dsl_dataset->ds_object != |
d683ddbb | 2221 | dsl_dir_phys(dd)->dd_head_dataset_obj) |
2e528b49 | 2222 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2223 | |
2224 | zap_cursor_init_serialized(&cursor, | |
2225 | dd->dd_pool->dp_meta_objset, | |
d683ddbb | 2226 | dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); |
34dc7c2f BB |
2227 | |
2228 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2229 | zap_cursor_fini(&cursor); | |
2e528b49 | 2230 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2231 | } |
2232 | ||
2233 | if (strlen(attr.za_name) + 1 > namelen) { | |
2234 | zap_cursor_fini(&cursor); | |
2e528b49 | 2235 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2236 | } |
2237 | ||
2238 | (void) strcpy(name, attr.za_name); | |
2239 | if (idp) | |
2240 | *idp = attr.za_first_integer; | |
2241 | zap_cursor_advance(&cursor); | |
2242 | *offp = zap_cursor_serialize(&cursor); | |
2243 | zap_cursor_fini(&cursor); | |
2244 | ||
2245 | return (0); | |
2246 | } | |
2247 | ||
9c43027b AJ |
2248 | typedef struct dmu_objset_find_ctx { |
2249 | taskq_t *dc_tq; | |
2250 | dsl_pool_t *dc_dp; | |
2251 | uint64_t dc_ddobj; | |
1149ba64 | 2252 | char *dc_ddname; /* last component of ddobj's name */ |
9c43027b AJ |
2253 | int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); |
2254 | void *dc_arg; | |
2255 | int dc_flags; | |
2256 | kmutex_t *dc_error_lock; | |
2257 | int *dc_error; | |
2258 | } dmu_objset_find_ctx_t; | |
2259 | ||
2260 | static void | |
2261 | dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) | |
b128c09f | 2262 | { |
9c43027b AJ |
2263 | dsl_pool_t *dp = dcp->dc_dp; |
2264 | dmu_objset_find_ctx_t *child_dcp; | |
13fe0198 MA |
2265 | dsl_dir_t *dd; |
2266 | dsl_dataset_t *ds; | |
2267 | zap_cursor_t zc; | |
2268 | zap_attribute_t *attr; | |
2269 | uint64_t thisobj; | |
9c43027b | 2270 | int err = 0; |
13fe0198 | 2271 | |
9c43027b AJ |
2272 | /* don't process if there already was an error */ |
2273 | if (*dcp->dc_error != 0) | |
2274 | goto out; | |
13fe0198 | 2275 | |
1149ba64 GM |
2276 | /* |
2277 | * Note: passing the name (dc_ddname) here is optional, but it | |
2278 | * improves performance because we don't need to call | |
2279 | * zap_value_search() to determine the name. | |
2280 | */ | |
2281 | err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); | |
13fe0198 | 2282 | if (err != 0) |
9c43027b | 2283 | goto out; |
13fe0198 MA |
2284 | |
2285 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ | |
2286 | if (dd->dd_myname[0] == '$') { | |
2287 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2288 | goto out; |
13fe0198 MA |
2289 | } |
2290 | ||
d683ddbb | 2291 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2292 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
13fe0198 MA |
2293 | |
2294 | /* | |
2295 | * Iterate over all children. | |
2296 | */ | |
9c43027b | 2297 | if (dcp->dc_flags & DS_FIND_CHILDREN) { |
13fe0198 | 2298 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2299 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
13fe0198 MA |
2300 | zap_cursor_retrieve(&zc, attr) == 0; |
2301 | (void) zap_cursor_advance(&zc)) { | |
2302 | ASSERT3U(attr->za_integer_length, ==, | |
2303 | sizeof (uint64_t)); | |
2304 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2305 | ||
1149ba64 GM |
2306 | child_dcp = |
2307 | kmem_alloc(sizeof (*child_dcp), KM_SLEEP); | |
9c43027b AJ |
2308 | *child_dcp = *dcp; |
2309 | child_dcp->dc_ddobj = attr->za_first_integer; | |
1149ba64 | 2310 | child_dcp->dc_ddname = spa_strdup(attr->za_name); |
9c43027b AJ |
2311 | if (dcp->dc_tq != NULL) |
2312 | (void) taskq_dispatch(dcp->dc_tq, | |
2313 | dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); | |
2314 | else | |
2315 | dmu_objset_find_dp_impl(child_dcp); | |
13fe0198 MA |
2316 | } |
2317 | zap_cursor_fini(&zc); | |
13fe0198 MA |
2318 | } |
2319 | ||
2320 | /* | |
2321 | * Iterate over all snapshots. | |
2322 | */ | |
9c43027b | 2323 | if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { |
13fe0198 MA |
2324 | dsl_dataset_t *ds; |
2325 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2326 | ||
2327 | if (err == 0) { | |
d683ddbb JG |
2328 | uint64_t snapobj; |
2329 | ||
2330 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
13fe0198 MA |
2331 | dsl_dataset_rele(ds, FTAG); |
2332 | ||
2333 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2334 | zap_cursor_retrieve(&zc, attr) == 0; | |
2335 | (void) zap_cursor_advance(&zc)) { | |
2336 | ASSERT3U(attr->za_integer_length, ==, | |
2337 | sizeof (uint64_t)); | |
2338 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2339 | ||
2340 | err = dsl_dataset_hold_obj(dp, | |
2341 | attr->za_first_integer, FTAG, &ds); | |
2342 | if (err != 0) | |
2343 | break; | |
9c43027b | 2344 | err = dcp->dc_func(dp, ds, dcp->dc_arg); |
13fe0198 MA |
2345 | dsl_dataset_rele(ds, FTAG); |
2346 | if (err != 0) | |
2347 | break; | |
2348 | } | |
2349 | zap_cursor_fini(&zc); | |
2350 | } | |
2351 | } | |
2352 | ||
13fe0198 MA |
2353 | kmem_free(attr, sizeof (zap_attribute_t)); |
2354 | ||
1149ba64 GM |
2355 | if (err != 0) { |
2356 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2357 | goto out; |
1149ba64 | 2358 | } |
13fe0198 MA |
2359 | |
2360 | /* | |
2361 | * Apply to self. | |
2362 | */ | |
2363 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
1149ba64 GM |
2364 | |
2365 | /* | |
2366 | * Note: we hold the dir while calling dsl_dataset_hold_obj() so | |
2367 | * that the dir will remain cached, and we won't have to re-instantiate | |
2368 | * it (which could be expensive due to finding its name via | |
2369 | * zap_value_search()). | |
2370 | */ | |
2371 | dsl_dir_rele(dd, FTAG); | |
13fe0198 | 2372 | if (err != 0) |
9c43027b AJ |
2373 | goto out; |
2374 | err = dcp->dc_func(dp, ds, dcp->dc_arg); | |
13fe0198 | 2375 | dsl_dataset_rele(ds, FTAG); |
9c43027b AJ |
2376 | |
2377 | out: | |
2378 | if (err != 0) { | |
2379 | mutex_enter(dcp->dc_error_lock); | |
2380 | /* only keep first error */ | |
2381 | if (*dcp->dc_error == 0) | |
2382 | *dcp->dc_error = err; | |
2383 | mutex_exit(dcp->dc_error_lock); | |
2384 | } | |
2385 | ||
1149ba64 GM |
2386 | if (dcp->dc_ddname != NULL) |
2387 | spa_strfree(dcp->dc_ddname); | |
9c43027b AJ |
2388 | kmem_free(dcp, sizeof (*dcp)); |
2389 | } | |
2390 | ||
2391 | static void | |
2392 | dmu_objset_find_dp_cb(void *arg) | |
2393 | { | |
2394 | dmu_objset_find_ctx_t *dcp = arg; | |
2395 | dsl_pool_t *dp = dcp->dc_dp; | |
2396 | ||
5e8cd5d1 AJ |
2397 | /* |
2398 | * We need to get a pool_config_lock here, as there are several | |
2399 | * asssert(pool_config_held) down the stack. Getting a lock via | |
2400 | * dsl_pool_config_enter is risky, as it might be stalled by a | |
2401 | * pending writer. This would deadlock, as the write lock can | |
2402 | * only be granted when our parent thread gives up the lock. | |
2403 | * The _prio interface gives us priority over a pending writer. | |
2404 | */ | |
2405 | dsl_pool_config_enter_prio(dp, FTAG); | |
9c43027b AJ |
2406 | |
2407 | dmu_objset_find_dp_impl(dcp); | |
2408 | ||
2409 | dsl_pool_config_exit(dp, FTAG); | |
2410 | } | |
2411 | ||
2412 | /* | |
2413 | * Find objsets under and including ddobj, call func(ds) on each. | |
2414 | * The order for the enumeration is completely undefined. | |
2415 | * func is called with dsl_pool_config held. | |
2416 | */ | |
2417 | int | |
2418 | dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, | |
2419 | int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) | |
2420 | { | |
2421 | int error = 0; | |
2422 | taskq_t *tq = NULL; | |
2423 | int ntasks; | |
2424 | dmu_objset_find_ctx_t *dcp; | |
2425 | kmutex_t err_lock; | |
2426 | ||
2427 | mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); | |
2428 | dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); | |
2429 | dcp->dc_tq = NULL; | |
2430 | dcp->dc_dp = dp; | |
2431 | dcp->dc_ddobj = ddobj; | |
1149ba64 | 2432 | dcp->dc_ddname = NULL; |
9c43027b AJ |
2433 | dcp->dc_func = func; |
2434 | dcp->dc_arg = arg; | |
2435 | dcp->dc_flags = flags; | |
2436 | dcp->dc_error_lock = &err_lock; | |
2437 | dcp->dc_error = &error; | |
2438 | ||
2439 | if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { | |
2440 | /* | |
2441 | * In case a write lock is held we can't make use of | |
2442 | * parallelism, as down the stack of the worker threads | |
2443 | * the lock is asserted via dsl_pool_config_held. | |
2444 | * In case of a read lock this is solved by getting a read | |
2445 | * lock in each worker thread, which isn't possible in case | |
2446 | * of a writer lock. So we fall back to the synchronous path | |
2447 | * here. | |
2448 | * In the future it might be possible to get some magic into | |
2449 | * dsl_pool_config_held in a way that it returns true for | |
2450 | * the worker threads so that a single lock held from this | |
2451 | * thread suffices. For now, stay single threaded. | |
2452 | */ | |
2453 | dmu_objset_find_dp_impl(dcp); | |
e5676636 | 2454 | mutex_destroy(&err_lock); |
9c43027b AJ |
2455 | |
2456 | return (error); | |
2457 | } | |
2458 | ||
2459 | ntasks = dmu_find_threads; | |
2460 | if (ntasks == 0) | |
2461 | ntasks = vdev_count_leaves(dp->dp_spa) * 4; | |
1229323d | 2462 | tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, |
9c43027b AJ |
2463 | INT_MAX, 0); |
2464 | if (tq == NULL) { | |
2465 | kmem_free(dcp, sizeof (*dcp)); | |
e5676636 BB |
2466 | mutex_destroy(&err_lock); |
2467 | ||
9c43027b AJ |
2468 | return (SET_ERROR(ENOMEM)); |
2469 | } | |
2470 | dcp->dc_tq = tq; | |
2471 | ||
2472 | /* dcp will be freed by task */ | |
2473 | (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); | |
2474 | ||
2475 | /* | |
2476 | * PORTING: this code relies on the property of taskq_wait to wait | |
2477 | * until no more tasks are queued and no more tasks are active. As | |
2478 | * we always queue new tasks from within other tasks, task_wait | |
2479 | * reliably waits for the full recursion to finish, even though we | |
2480 | * enqueue new tasks after taskq_wait has been called. | |
2481 | * On platforms other than illumos, taskq_wait may not have this | |
2482 | * property. | |
2483 | */ | |
2484 | taskq_wait(tq); | |
2485 | taskq_destroy(tq); | |
2486 | mutex_destroy(&err_lock); | |
2487 | ||
2488 | return (error); | |
b128c09f BB |
2489 | } |
2490 | ||
2491 | /* | |
13fe0198 MA |
2492 | * Find all objsets under name, and for each, call 'func(child_name, arg)'. |
2493 | * The dp_config_rwlock must not be held when this is called, and it | |
2494 | * will not be held when the callback is called. | |
2495 | * Therefore this function should only be used when the pool is not changing | |
2496 | * (e.g. in syncing context), or the callback can deal with the possible races. | |
b128c09f | 2497 | */ |
13fe0198 MA |
2498 | static int |
2499 | dmu_objset_find_impl(spa_t *spa, const char *name, | |
2500 | int func(const char *, void *), void *arg, int flags) | |
34dc7c2f BB |
2501 | { |
2502 | dsl_dir_t *dd; | |
13fe0198 | 2503 | dsl_pool_t *dp = spa_get_dsl(spa); |
b128c09f | 2504 | dsl_dataset_t *ds; |
34dc7c2f BB |
2505 | zap_cursor_t zc; |
2506 | zap_attribute_t *attr; | |
2507 | char *child; | |
b128c09f BB |
2508 | uint64_t thisobj; |
2509 | int err; | |
34dc7c2f | 2510 | |
13fe0198 MA |
2511 | dsl_pool_config_enter(dp, FTAG); |
2512 | ||
2513 | err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); | |
2514 | if (err != 0) { | |
2515 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f | 2516 | return (err); |
13fe0198 | 2517 | } |
34dc7c2f | 2518 | |
b128c09f BB |
2519 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ |
2520 | if (dd->dd_myname[0] == '$') { | |
13fe0198 MA |
2521 | dsl_dir_rele(dd, FTAG); |
2522 | dsl_pool_config_exit(dp, FTAG); | |
b128c09f BB |
2523 | return (0); |
2524 | } | |
2525 | ||
d683ddbb | 2526 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2527 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
34dc7c2f BB |
2528 | |
2529 | /* | |
2530 | * Iterate over all children. | |
2531 | */ | |
2532 | if (flags & DS_FIND_CHILDREN) { | |
b128c09f | 2533 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2534 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
2535 | zap_cursor_retrieve(&zc, attr) == 0; |
2536 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 MA |
2537 | ASSERT3U(attr->za_integer_length, ==, |
2538 | sizeof (uint64_t)); | |
2539 | ASSERT3U(attr->za_num_integers, ==, 1); | |
34dc7c2f | 2540 | |
428870ff | 2541 | child = kmem_asprintf("%s/%s", name, attr->za_name); |
13fe0198 MA |
2542 | dsl_pool_config_exit(dp, FTAG); |
2543 | err = dmu_objset_find_impl(spa, child, | |
2544 | func, arg, flags); | |
2545 | dsl_pool_config_enter(dp, FTAG); | |
428870ff | 2546 | strfree(child); |
13fe0198 | 2547 | if (err != 0) |
34dc7c2f BB |
2548 | break; |
2549 | } | |
2550 | zap_cursor_fini(&zc); | |
2551 | ||
13fe0198 MA |
2552 | if (err != 0) { |
2553 | dsl_dir_rele(dd, FTAG); | |
2554 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
2555 | kmem_free(attr, sizeof (zap_attribute_t)); |
2556 | return (err); | |
2557 | } | |
2558 | } | |
2559 | ||
2560 | /* | |
2561 | * Iterate over all snapshots. | |
2562 | */ | |
b128c09f | 2563 | if (flags & DS_FIND_SNAPSHOTS) { |
b128c09f | 2564 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); |
b128c09f BB |
2565 | |
2566 | if (err == 0) { | |
d683ddbb JG |
2567 | uint64_t snapobj; |
2568 | ||
2569 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
b128c09f BB |
2570 | dsl_dataset_rele(ds, FTAG); |
2571 | ||
2572 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2573 | zap_cursor_retrieve(&zc, attr) == 0; | |
2574 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 | 2575 | ASSERT3U(attr->za_integer_length, ==, |
b128c09f | 2576 | sizeof (uint64_t)); |
13fe0198 | 2577 | ASSERT3U(attr->za_num_integers, ==, 1); |
b128c09f | 2578 | |
428870ff BB |
2579 | child = kmem_asprintf("%s@%s", |
2580 | name, attr->za_name); | |
13fe0198 MA |
2581 | dsl_pool_config_exit(dp, FTAG); |
2582 | err = func(child, arg); | |
2583 | dsl_pool_config_enter(dp, FTAG); | |
428870ff | 2584 | strfree(child); |
13fe0198 | 2585 | if (err != 0) |
b128c09f BB |
2586 | break; |
2587 | } | |
2588 | zap_cursor_fini(&zc); | |
34dc7c2f | 2589 | } |
34dc7c2f BB |
2590 | } |
2591 | ||
13fe0198 | 2592 | dsl_dir_rele(dd, FTAG); |
34dc7c2f | 2593 | kmem_free(attr, sizeof (zap_attribute_t)); |
13fe0198 | 2594 | dsl_pool_config_exit(dp, FTAG); |
34dc7c2f | 2595 | |
13fe0198 | 2596 | if (err != 0) |
34dc7c2f BB |
2597 | return (err); |
2598 | ||
13fe0198 MA |
2599 | /* Apply to self. */ |
2600 | return (func(name, arg)); | |
34dc7c2f BB |
2601 | } |
2602 | ||
13fe0198 MA |
2603 | /* |
2604 | * See comment above dmu_objset_find_impl(). | |
2605 | */ | |
d164b209 | 2606 | int |
13fe0198 MA |
2607 | dmu_objset_find(char *name, int func(const char *, void *), void *arg, |
2608 | int flags) | |
d164b209 | 2609 | { |
13fe0198 MA |
2610 | spa_t *spa; |
2611 | int error; | |
d164b209 | 2612 | |
13fe0198 MA |
2613 | error = spa_open(name, &spa, FTAG); |
2614 | if (error != 0) | |
2615 | return (error); | |
2616 | error = dmu_objset_find_impl(spa, name, func, arg, flags); | |
2617 | spa_close(spa, FTAG); | |
2618 | return (error); | |
d164b209 BB |
2619 | } |
2620 | ||
34dc7c2f BB |
2621 | void |
2622 | dmu_objset_set_user(objset_t *os, void *user_ptr) | |
2623 | { | |
428870ff BB |
2624 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2625 | os->os_user_ptr = user_ptr; | |
34dc7c2f BB |
2626 | } |
2627 | ||
2628 | void * | |
2629 | dmu_objset_get_user(objset_t *os) | |
2630 | { | |
428870ff BB |
2631 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2632 | return (os->os_user_ptr); | |
34dc7c2f | 2633 | } |
c28b2279 | 2634 | |
13fe0198 MA |
2635 | /* |
2636 | * Determine name of filesystem, given name of snapshot. | |
eca7b760 | 2637 | * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes |
13fe0198 MA |
2638 | */ |
2639 | int | |
2640 | dmu_fsname(const char *snapname, char *buf) | |
2641 | { | |
2642 | char *atp = strchr(snapname, '@'); | |
2643 | if (atp == NULL) | |
2e528b49 | 2644 | return (SET_ERROR(EINVAL)); |
eca7b760 | 2645 | if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 2646 | return (SET_ERROR(ENAMETOOLONG)); |
13fe0198 MA |
2647 | (void) strlcpy(buf, snapname, atp - snapname + 1); |
2648 | return (0); | |
2649 | } | |
2650 | ||
3ec3bc21 BB |
2651 | /* |
2652 | * Call when we think we're going to write/free space in open context to track | |
2653 | * the amount of dirty data in the open txg, which is also the amount | |
2654 | * of memory that can not be evicted until this txg syncs. | |
2655 | */ | |
2656 | void | |
2657 | dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) | |
2658 | { | |
2659 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
2660 | int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); | |
2661 | ||
2662 | if (ds != NULL) { | |
2663 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
2664 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
2665 | } | |
2666 | } | |
2667 | ||
c28b2279 | 2668 | #if defined(_KERNEL) && defined(HAVE_SPL) |
f0fd83be | 2669 | EXPORT_SYMBOL(dmu_objset_zil); |
c28b2279 | 2670 | EXPORT_SYMBOL(dmu_objset_pool); |
f0fd83be BB |
2671 | EXPORT_SYMBOL(dmu_objset_ds); |
2672 | EXPORT_SYMBOL(dmu_objset_type); | |
c28b2279 BB |
2673 | EXPORT_SYMBOL(dmu_objset_name); |
2674 | EXPORT_SYMBOL(dmu_objset_hold); | |
b5256303 | 2675 | EXPORT_SYMBOL(dmu_objset_hold_flags); |
c28b2279 BB |
2676 | EXPORT_SYMBOL(dmu_objset_own); |
2677 | EXPORT_SYMBOL(dmu_objset_rele); | |
b5256303 | 2678 | EXPORT_SYMBOL(dmu_objset_rele_flags); |
c28b2279 BB |
2679 | EXPORT_SYMBOL(dmu_objset_disown); |
2680 | EXPORT_SYMBOL(dmu_objset_from_ds); | |
2681 | EXPORT_SYMBOL(dmu_objset_create); | |
2682 | EXPORT_SYMBOL(dmu_objset_clone); | |
c28b2279 BB |
2683 | EXPORT_SYMBOL(dmu_objset_stats); |
2684 | EXPORT_SYMBOL(dmu_objset_fast_stat); | |
54a179e7 | 2685 | EXPORT_SYMBOL(dmu_objset_spa); |
c28b2279 BB |
2686 | EXPORT_SYMBOL(dmu_objset_space); |
2687 | EXPORT_SYMBOL(dmu_objset_fsid_guid); | |
2688 | EXPORT_SYMBOL(dmu_objset_find); | |
c28b2279 BB |
2689 | EXPORT_SYMBOL(dmu_objset_byteswap); |
2690 | EXPORT_SYMBOL(dmu_objset_evict_dbufs); | |
2691 | EXPORT_SYMBOL(dmu_objset_snap_cmtime); | |
50c957f7 | 2692 | EXPORT_SYMBOL(dmu_objset_dnodesize); |
c28b2279 BB |
2693 | |
2694 | EXPORT_SYMBOL(dmu_objset_sync); | |
2695 | EXPORT_SYMBOL(dmu_objset_is_dirty); | |
b5256303 | 2696 | EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); |
c28b2279 BB |
2697 | EXPORT_SYMBOL(dmu_objset_create_impl); |
2698 | EXPORT_SYMBOL(dmu_objset_open_impl); | |
2699 | EXPORT_SYMBOL(dmu_objset_evict); | |
2700 | EXPORT_SYMBOL(dmu_objset_register_type); | |
2701 | EXPORT_SYMBOL(dmu_objset_do_userquota_updates); | |
2702 | EXPORT_SYMBOL(dmu_objset_userquota_get_ids); | |
2703 | EXPORT_SYMBOL(dmu_objset_userused_enabled); | |
2704 | EXPORT_SYMBOL(dmu_objset_userspace_upgrade); | |
2705 | EXPORT_SYMBOL(dmu_objset_userspace_present); | |
1de321e6 JX |
2706 | EXPORT_SYMBOL(dmu_objset_userobjused_enabled); |
2707 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgrade); | |
126ae9f4 | 2708 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); |
1de321e6 | 2709 | EXPORT_SYMBOL(dmu_objset_userobjspace_present); |
c28b2279 | 2710 | #endif |