]>
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. |
c0daec32 | 31 | * Copyright (c) 2017 Open-E, Inc. All Rights Reserved. |
34dc7c2f BB |
32 | */ |
33 | ||
428870ff BB |
34 | /* Portions Copyright 2010 Robert Milkowski */ |
35 | ||
1de321e6 | 36 | #include <sys/zfeature.h> |
34dc7c2f BB |
37 | #include <sys/cred.h> |
38 | #include <sys/zfs_context.h> | |
39 | #include <sys/dmu_objset.h> | |
40 | #include <sys/dsl_dir.h> | |
41 | #include <sys/dsl_dataset.h> | |
42 | #include <sys/dsl_prop.h> | |
43 | #include <sys/dsl_pool.h> | |
44 | #include <sys/dsl_synctask.h> | |
45 | #include <sys/dsl_deleg.h> | |
46 | #include <sys/dnode.h> | |
47 | #include <sys/dbuf.h> | |
48 | #include <sys/zvol.h> | |
49 | #include <sys/dmu_tx.h> | |
34dc7c2f BB |
50 | #include <sys/zap.h> |
51 | #include <sys/zil.h> | |
52 | #include <sys/dmu_impl.h> | |
53 | #include <sys/zfs_ioctl.h> | |
428870ff | 54 | #include <sys/sa.h> |
572e2857 | 55 | #include <sys/zfs_onexit.h> |
13fe0198 | 56 | #include <sys/dsl_destroy.h> |
9c43027b | 57 | #include <sys/vdev.h> |
a1d477c2 | 58 | #include <sys/zfeature.h> |
f74b821a | 59 | #include <sys/policy.h> |
1de321e6 | 60 | #include <sys/spa_impl.h> |
b5256303 | 61 | #include <sys/dmu_send.h> |
9c5167d1 | 62 | #include <sys/zfs_project.h> |
572e2857 BB |
63 | |
64 | /* | |
65 | * Needed to close a window in dnode_move() that allows the objset to be freed | |
66 | * before it can be safely accessed. | |
67 | */ | |
68 | krwlock_t os_lock; | |
69 | ||
9c43027b | 70 | /* |
4e33ba4c | 71 | * Tunable to overwrite the maximum number of threads for the parallelization |
9c43027b AJ |
72 | * of dmu_objset_find_dp, needed to speed up the import of pools with many |
73 | * datasets. | |
74 | * Default is 4 times the number of leaf vdevs. | |
75 | */ | |
76 | int dmu_find_threads = 0; | |
77 | ||
68cbd56e NB |
78 | /* |
79 | * Backfill lower metadnode objects after this many have been freed. | |
80 | * Backfilling negatively impacts object creation rates, so only do it | |
81 | * if there are enough holes to fill. | |
82 | */ | |
83 | int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; | |
84 | ||
c0daec32 AB |
85 | static char *upgrade_tag = "upgrade_tag"; |
86 | ||
9c43027b AJ |
87 | static void dmu_objset_find_dp_cb(void *arg); |
88 | ||
1de321e6 JX |
89 | static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); |
90 | static void dmu_objset_upgrade_stop(objset_t *os); | |
91 | ||
572e2857 BB |
92 | void |
93 | dmu_objset_init(void) | |
94 | { | |
95 | rw_init(&os_lock, NULL, RW_DEFAULT, NULL); | |
96 | } | |
97 | ||
98 | void | |
99 | dmu_objset_fini(void) | |
100 | { | |
101 | rw_destroy(&os_lock); | |
102 | } | |
34dc7c2f BB |
103 | |
104 | spa_t * | |
105 | dmu_objset_spa(objset_t *os) | |
106 | { | |
428870ff | 107 | return (os->os_spa); |
34dc7c2f BB |
108 | } |
109 | ||
110 | zilog_t * | |
111 | dmu_objset_zil(objset_t *os) | |
112 | { | |
428870ff | 113 | return (os->os_zil); |
34dc7c2f BB |
114 | } |
115 | ||
116 | dsl_pool_t * | |
117 | dmu_objset_pool(objset_t *os) | |
118 | { | |
119 | dsl_dataset_t *ds; | |
120 | ||
428870ff | 121 | if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) |
34dc7c2f BB |
122 | return (ds->ds_dir->dd_pool); |
123 | else | |
428870ff | 124 | return (spa_get_dsl(os->os_spa)); |
34dc7c2f BB |
125 | } |
126 | ||
127 | dsl_dataset_t * | |
128 | dmu_objset_ds(objset_t *os) | |
129 | { | |
428870ff | 130 | return (os->os_dsl_dataset); |
34dc7c2f BB |
131 | } |
132 | ||
133 | dmu_objset_type_t | |
134 | dmu_objset_type(objset_t *os) | |
135 | { | |
428870ff | 136 | return (os->os_phys->os_type); |
34dc7c2f BB |
137 | } |
138 | ||
139 | void | |
140 | dmu_objset_name(objset_t *os, char *buf) | |
141 | { | |
428870ff | 142 | dsl_dataset_name(os->os_dsl_dataset, buf); |
34dc7c2f BB |
143 | } |
144 | ||
145 | uint64_t | |
146 | dmu_objset_id(objset_t *os) | |
147 | { | |
428870ff | 148 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
149 | |
150 | return (ds ? ds->ds_object : 0); | |
151 | } | |
152 | ||
50c957f7 NB |
153 | uint64_t |
154 | dmu_objset_dnodesize(objset_t *os) | |
155 | { | |
156 | return (os->os_dnodesize); | |
157 | } | |
158 | ||
faf0f58c | 159 | zfs_sync_type_t |
428870ff BB |
160 | dmu_objset_syncprop(objset_t *os) |
161 | { | |
162 | return (os->os_sync); | |
163 | } | |
164 | ||
faf0f58c | 165 | zfs_logbias_op_t |
428870ff BB |
166 | dmu_objset_logbias(objset_t *os) |
167 | { | |
168 | return (os->os_logbias); | |
169 | } | |
170 | ||
34dc7c2f BB |
171 | static void |
172 | checksum_changed_cb(void *arg, uint64_t newval) | |
173 | { | |
428870ff | 174 | objset_t *os = arg; |
34dc7c2f BB |
175 | |
176 | /* | |
177 | * Inheritance should have been done by now. | |
178 | */ | |
179 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
180 | ||
428870ff | 181 | os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); |
34dc7c2f BB |
182 | } |
183 | ||
184 | static void | |
185 | compression_changed_cb(void *arg, uint64_t newval) | |
186 | { | |
428870ff | 187 | objset_t *os = arg; |
34dc7c2f BB |
188 | |
189 | /* | |
190 | * Inheritance and range checking should have been done by now. | |
191 | */ | |
192 | ASSERT(newval != ZIO_COMPRESS_INHERIT); | |
193 | ||
99197f03 JG |
194 | os->os_compress = zio_compress_select(os->os_spa, newval, |
195 | ZIO_COMPRESS_ON); | |
34dc7c2f BB |
196 | } |
197 | ||
198 | static void | |
199 | copies_changed_cb(void *arg, uint64_t newval) | |
200 | { | |
428870ff | 201 | objset_t *os = arg; |
34dc7c2f BB |
202 | |
203 | /* | |
204 | * Inheritance and range checking should have been done by now. | |
205 | */ | |
206 | ASSERT(newval > 0); | |
428870ff | 207 | ASSERT(newval <= spa_max_replication(os->os_spa)); |
34dc7c2f | 208 | |
428870ff BB |
209 | os->os_copies = newval; |
210 | } | |
211 | ||
212 | static void | |
213 | dedup_changed_cb(void *arg, uint64_t newval) | |
214 | { | |
215 | objset_t *os = arg; | |
216 | spa_t *spa = os->os_spa; | |
217 | enum zio_checksum checksum; | |
218 | ||
219 | /* | |
220 | * Inheritance should have been done by now. | |
221 | */ | |
222 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
223 | ||
224 | checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); | |
225 | ||
226 | os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; | |
227 | os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); | |
34dc7c2f BB |
228 | } |
229 | ||
b128c09f BB |
230 | static void |
231 | primary_cache_changed_cb(void *arg, uint64_t newval) | |
232 | { | |
428870ff | 233 | objset_t *os = arg; |
b128c09f BB |
234 | |
235 | /* | |
236 | * Inheritance and range checking should have been done by now. | |
237 | */ | |
238 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
239 | newval == ZFS_CACHE_METADATA); | |
240 | ||
428870ff | 241 | os->os_primary_cache = newval; |
b128c09f BB |
242 | } |
243 | ||
244 | static void | |
245 | secondary_cache_changed_cb(void *arg, uint64_t newval) | |
246 | { | |
428870ff | 247 | objset_t *os = arg; |
b128c09f BB |
248 | |
249 | /* | |
250 | * Inheritance and range checking should have been done by now. | |
251 | */ | |
252 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
253 | newval == ZFS_CACHE_METADATA); | |
254 | ||
428870ff BB |
255 | os->os_secondary_cache = newval; |
256 | } | |
257 | ||
258 | static void | |
259 | sync_changed_cb(void *arg, uint64_t newval) | |
260 | { | |
261 | objset_t *os = arg; | |
262 | ||
263 | /* | |
264 | * Inheritance and range checking should have been done by now. | |
265 | */ | |
266 | ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || | |
267 | newval == ZFS_SYNC_DISABLED); | |
268 | ||
269 | os->os_sync = newval; | |
270 | if (os->os_zil) | |
271 | zil_set_sync(os->os_zil, newval); | |
272 | } | |
273 | ||
faf0f58c MA |
274 | static void |
275 | redundant_metadata_changed_cb(void *arg, uint64_t newval) | |
276 | { | |
277 | objset_t *os = arg; | |
278 | ||
279 | /* | |
280 | * Inheritance and range checking should have been done by now. | |
281 | */ | |
282 | ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || | |
283 | newval == ZFS_REDUNDANT_METADATA_MOST); | |
284 | ||
285 | os->os_redundant_metadata = newval; | |
286 | } | |
287 | ||
50c957f7 NB |
288 | static void |
289 | dnodesize_changed_cb(void *arg, uint64_t newval) | |
290 | { | |
291 | objset_t *os = arg; | |
292 | ||
293 | switch (newval) { | |
294 | case ZFS_DNSIZE_LEGACY: | |
295 | os->os_dnodesize = DNODE_MIN_SIZE; | |
296 | break; | |
297 | case ZFS_DNSIZE_AUTO: | |
298 | /* | |
299 | * Choose a dnode size that will work well for most | |
300 | * workloads if the user specified "auto". Future code | |
301 | * improvements could dynamically select a dnode size | |
302 | * based on observed workload patterns. | |
303 | */ | |
304 | os->os_dnodesize = DNODE_MIN_SIZE * 2; | |
305 | break; | |
306 | case ZFS_DNSIZE_1K: | |
307 | case ZFS_DNSIZE_2K: | |
308 | case ZFS_DNSIZE_4K: | |
309 | case ZFS_DNSIZE_8K: | |
310 | case ZFS_DNSIZE_16K: | |
311 | os->os_dnodesize = newval; | |
312 | break; | |
313 | } | |
314 | } | |
315 | ||
428870ff BB |
316 | static void |
317 | logbias_changed_cb(void *arg, uint64_t newval) | |
318 | { | |
319 | objset_t *os = arg; | |
320 | ||
321 | ASSERT(newval == ZFS_LOGBIAS_LATENCY || | |
322 | newval == ZFS_LOGBIAS_THROUGHPUT); | |
323 | os->os_logbias = newval; | |
324 | if (os->os_zil) | |
325 | zil_set_logbias(os->os_zil, newval); | |
b128c09f BB |
326 | } |
327 | ||
f1512ee6 MA |
328 | static void |
329 | recordsize_changed_cb(void *arg, uint64_t newval) | |
330 | { | |
331 | objset_t *os = arg; | |
332 | ||
333 | os->os_recordsize = newval; | |
334 | } | |
335 | ||
34dc7c2f BB |
336 | void |
337 | dmu_objset_byteswap(void *buf, size_t size) | |
338 | { | |
339 | objset_phys_t *osp = buf; | |
340 | ||
9c5167d1 NF |
341 | ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || |
342 | size == sizeof (objset_phys_t)); | |
34dc7c2f BB |
343 | dnode_byteswap(&osp->os_meta_dnode); |
344 | byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); | |
345 | osp->os_type = BSWAP_64(osp->os_type); | |
9babb374 | 346 | osp->os_flags = BSWAP_64(osp->os_flags); |
9c5167d1 | 347 | if (size >= OBJSET_PHYS_SIZE_V2) { |
9babb374 BB |
348 | dnode_byteswap(&osp->os_userused_dnode); |
349 | dnode_byteswap(&osp->os_groupused_dnode); | |
9c5167d1 NF |
350 | if (size >= sizeof (objset_phys_t)) |
351 | dnode_byteswap(&osp->os_projectused_dnode); | |
9babb374 | 352 | } |
34dc7c2f BB |
353 | } |
354 | ||
64fc7762 MA |
355 | /* |
356 | * The hash is a CRC-based hash of the objset_t pointer and the object number. | |
357 | */ | |
358 | static uint64_t | |
359 | dnode_hash(const objset_t *os, uint64_t obj) | |
360 | { | |
361 | uintptr_t osv = (uintptr_t)os; | |
362 | uint64_t crc = -1ULL; | |
363 | ||
364 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
365 | /* | |
366 | * The low 6 bits of the pointer don't have much entropy, because | |
367 | * the objset_t is larger than 2^6 bytes long. | |
368 | */ | |
369 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; | |
370 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; | |
371 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; | |
372 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; | |
373 | ||
374 | crc ^= (osv>>14) ^ (obj>>24); | |
375 | ||
376 | return (crc); | |
377 | } | |
378 | ||
379 | unsigned int | |
380 | dnode_multilist_index_func(multilist_t *ml, void *obj) | |
381 | { | |
382 | dnode_t *dn = obj; | |
383 | return (dnode_hash(dn->dn_objset, dn->dn_object) % | |
384 | multilist_get_num_sublists(ml)); | |
385 | } | |
386 | ||
a1d477c2 MA |
387 | /* |
388 | * Instantiates the objset_t in-memory structure corresponding to the | |
389 | * objset_phys_t that's pointed to by the specified blkptr_t. | |
390 | */ | |
34dc7c2f BB |
391 | int |
392 | dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
428870ff | 393 | objset_t **osp) |
34dc7c2f | 394 | { |
428870ff | 395 | objset_t *os; |
b128c09f | 396 | int i, err; |
34dc7c2f BB |
397 | |
398 | ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); | |
399 | ||
a1d477c2 MA |
400 | /* |
401 | * The $ORIGIN dataset (if it exists) doesn't have an associated | |
402 | * objset, so there's no reason to open it. The $ORIGIN dataset | |
403 | * will not exist on pools older than SPA_VERSION_ORIGIN. | |
404 | */ | |
405 | if (ds != NULL && spa_get_dsl(spa) != NULL && | |
406 | spa_get_dsl(spa)->dp_origin_snap != NULL) { | |
407 | ASSERT3P(ds->ds_dir, !=, | |
408 | spa_get_dsl(spa)->dp_origin_snap->ds_dir); | |
409 | } | |
410 | ||
79c76d5b | 411 | os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); |
428870ff BB |
412 | os->os_dsl_dataset = ds; |
413 | os->os_spa = spa; | |
414 | os->os_rootbp = bp; | |
415 | if (!BP_IS_HOLE(os->os_rootbp)) { | |
2a432414 | 416 | arc_flags_t aflags = ARC_FLAG_WAIT; |
5dbd68a3 | 417 | zbookmark_phys_t zb; |
9c5167d1 | 418 | int size; |
b5256303 | 419 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; |
428870ff BB |
420 | SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, |
421 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
422 | ||
423 | if (DMU_OS_IS_L2CACHEABLE(os)) | |
2a432414 | 424 | aflags |= ARC_FLAG_L2CACHE; |
34dc7c2f | 425 | |
b5256303 TC |
426 | if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { |
427 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
428 | ASSERT(BP_IS_AUTHENTICATED(bp)); | |
429 | zio_flags |= ZIO_FLAG_RAW; | |
430 | } | |
431 | ||
428870ff | 432 | dprintf_bp(os->os_rootbp, "reading %s", ""); |
294f6806 | 433 | err = arc_read(NULL, spa, os->os_rootbp, |
428870ff | 434 | arc_getbuf_func, &os->os_phys_buf, |
b5256303 | 435 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
13fe0198 | 436 | if (err != 0) { |
428870ff | 437 | kmem_free(os, sizeof (objset_t)); |
b128c09f BB |
438 | /* convert checksum errors into IO errors */ |
439 | if (err == ECKSUM) | |
2e528b49 | 440 | err = SET_ERROR(EIO); |
34dc7c2f BB |
441 | return (err); |
442 | } | |
9babb374 | 443 | |
9c5167d1 NF |
444 | if (spa_version(spa) < SPA_VERSION_USERSPACE) |
445 | size = OBJSET_PHYS_SIZE_V1; | |
446 | else if (!spa_feature_is_enabled(spa, | |
447 | SPA_FEATURE_PROJECT_QUOTA)) | |
448 | size = OBJSET_PHYS_SIZE_V2; | |
449 | else | |
450 | size = sizeof (objset_phys_t); | |
451 | ||
9babb374 | 452 | /* Increase the blocksize if we are permitted. */ |
9c5167d1 | 453 | if (arc_buf_size(os->os_phys_buf) < size) { |
2aa34383 | 454 | arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, |
9c5167d1 NF |
455 | ARC_BUFC_METADATA, size); |
456 | bzero(buf->b_data, size); | |
428870ff BB |
457 | bcopy(os->os_phys_buf->b_data, buf->b_data, |
458 | arc_buf_size(os->os_phys_buf)); | |
d3c2ae1c | 459 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 460 | os->os_phys_buf = buf; |
9babb374 BB |
461 | } |
462 | ||
428870ff BB |
463 | os->os_phys = os->os_phys_buf->b_data; |
464 | os->os_flags = os->os_phys->os_flags; | |
34dc7c2f | 465 | } else { |
9babb374 | 466 | int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? |
9c5167d1 | 467 | sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; |
2aa34383 DK |
468 | os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, |
469 | ARC_BUFC_METADATA, size); | |
428870ff BB |
470 | os->os_phys = os->os_phys_buf->b_data; |
471 | bzero(os->os_phys, size); | |
34dc7c2f BB |
472 | } |
473 | ||
474 | /* | |
475 | * Note: the changed_cb will be called once before the register | |
476 | * func returns, thus changing the checksum/compression from the | |
b128c09f BB |
477 | * default (fletcher2/off). Snapshots don't need to know about |
478 | * checksum/compression/copies. | |
34dc7c2f | 479 | */ |
9b67f605 | 480 | if (ds != NULL) { |
47dfff3b MA |
481 | boolean_t needlock = B_FALSE; |
482 | ||
b5256303 TC |
483 | os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); |
484 | ||
47dfff3b MA |
485 | /* |
486 | * Note: it's valid to open the objset if the dataset is | |
487 | * long-held, in which case the pool_config lock will not | |
488 | * be held. | |
489 | */ | |
490 | if (!dsl_pool_config_held(dmu_objset_pool(os))) { | |
491 | needlock = B_TRUE; | |
492 | dsl_pool_config_enter(dmu_objset_pool(os), FTAG); | |
493 | } | |
b5256303 | 494 | |
13fe0198 MA |
495 | err = dsl_prop_register(ds, |
496 | zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), | |
428870ff | 497 | primary_cache_changed_cb, os); |
13fe0198 MA |
498 | if (err == 0) { |
499 | err = dsl_prop_register(ds, | |
500 | zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), | |
428870ff | 501 | secondary_cache_changed_cb, os); |
13fe0198 | 502 | } |
0c66c32d | 503 | if (!ds->ds_is_snapshot) { |
13fe0198 MA |
504 | if (err == 0) { |
505 | err = dsl_prop_register(ds, | |
506 | zfs_prop_to_name(ZFS_PROP_CHECKSUM), | |
428870ff | 507 | checksum_changed_cb, os); |
13fe0198 MA |
508 | } |
509 | if (err == 0) { | |
510 | err = dsl_prop_register(ds, | |
511 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
428870ff | 512 | compression_changed_cb, os); |
13fe0198 MA |
513 | } |
514 | if (err == 0) { | |
515 | err = dsl_prop_register(ds, | |
516 | zfs_prop_to_name(ZFS_PROP_COPIES), | |
428870ff | 517 | copies_changed_cb, os); |
13fe0198 MA |
518 | } |
519 | if (err == 0) { | |
520 | err = dsl_prop_register(ds, | |
521 | zfs_prop_to_name(ZFS_PROP_DEDUP), | |
428870ff | 522 | dedup_changed_cb, os); |
13fe0198 MA |
523 | } |
524 | if (err == 0) { | |
525 | err = dsl_prop_register(ds, | |
526 | zfs_prop_to_name(ZFS_PROP_LOGBIAS), | |
428870ff | 527 | logbias_changed_cb, os); |
13fe0198 MA |
528 | } |
529 | if (err == 0) { | |
530 | err = dsl_prop_register(ds, | |
531 | zfs_prop_to_name(ZFS_PROP_SYNC), | |
428870ff | 532 | sync_changed_cb, os); |
13fe0198 | 533 | } |
faf0f58c MA |
534 | if (err == 0) { |
535 | err = dsl_prop_register(ds, | |
536 | zfs_prop_to_name( | |
537 | ZFS_PROP_REDUNDANT_METADATA), | |
538 | redundant_metadata_changed_cb, os); | |
539 | } | |
f1512ee6 MA |
540 | if (err == 0) { |
541 | err = dsl_prop_register(ds, | |
542 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
543 | recordsize_changed_cb, os); | |
544 | } | |
50c957f7 NB |
545 | if (err == 0) { |
546 | err = dsl_prop_register(ds, | |
547 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
548 | dnodesize_changed_cb, os); | |
549 | } | |
b128c09f | 550 | } |
47dfff3b MA |
551 | if (needlock) |
552 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); | |
13fe0198 | 553 | if (err != 0) { |
d3c2ae1c | 554 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 555 | kmem_free(os, sizeof (objset_t)); |
34dc7c2f BB |
556 | return (err); |
557 | } | |
9b67f605 | 558 | } else { |
34dc7c2f | 559 | /* It's the meta-objset. */ |
428870ff | 560 | os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; |
99197f03 | 561 | os->os_compress = ZIO_COMPRESS_ON; |
b5256303 | 562 | os->os_encrypted = B_FALSE; |
428870ff BB |
563 | os->os_copies = spa_max_replication(spa); |
564 | os->os_dedup_checksum = ZIO_CHECKSUM_OFF; | |
faf0f58c MA |
565 | os->os_dedup_verify = B_FALSE; |
566 | os->os_logbias = ZFS_LOGBIAS_LATENCY; | |
567 | os->os_sync = ZFS_SYNC_STANDARD; | |
428870ff BB |
568 | os->os_primary_cache = ZFS_CACHE_ALL; |
569 | os->os_secondary_cache = ZFS_CACHE_ALL; | |
50c957f7 | 570 | os->os_dnodesize = DNODE_MIN_SIZE; |
34dc7c2f BB |
571 | } |
572 | ||
0c66c32d | 573 | if (ds == NULL || !ds->ds_is_snapshot) |
572e2857 | 574 | os->os_zil_header = os->os_phys->os_zil_header; |
428870ff | 575 | os->os_zil = zil_alloc(os, &os->os_zil_header); |
34dc7c2f BB |
576 | |
577 | for (i = 0; i < TXG_SIZE; i++) { | |
64fc7762 MA |
578 | os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), |
579 | offsetof(dnode_t, dn_dirty_link[i]), | |
580 | dnode_multilist_index_func); | |
34dc7c2f | 581 | } |
428870ff | 582 | list_create(&os->os_dnodes, sizeof (dnode_t), |
34dc7c2f | 583 | offsetof(dnode_t, dn_link)); |
428870ff | 584 | list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), |
34dc7c2f BB |
585 | offsetof(dmu_buf_impl_t, db_link)); |
586 | ||
0c66c32d JG |
587 | list_link_init(&os->os_evicting_node); |
588 | ||
428870ff | 589 | mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); |
64fc7762 | 590 | mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); |
428870ff BB |
591 | mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); |
592 | mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); | |
dbeb8796 MA |
593 | os->os_obj_next_percpu_len = boot_ncpus; |
594 | os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * | |
595 | sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); | |
428870ff | 596 | |
0c66c32d JG |
597 | dnode_special_open(os, &os->os_phys->os_meta_dnode, |
598 | DMU_META_DNODE_OBJECT, &os->os_meta_dnode); | |
9c5167d1 | 599 | if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { |
0c66c32d JG |
600 | dnode_special_open(os, &os->os_phys->os_userused_dnode, |
601 | DMU_USERUSED_OBJECT, &os->os_userused_dnode); | |
602 | dnode_special_open(os, &os->os_phys->os_groupused_dnode, | |
603 | DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); | |
9c5167d1 NF |
604 | if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) |
605 | dnode_special_open(os, | |
606 | &os->os_phys->os_projectused_dnode, | |
607 | DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); | |
9babb374 | 608 | } |
34dc7c2f | 609 | |
1de321e6 JX |
610 | mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); |
611 | ||
428870ff | 612 | *osp = os; |
34dc7c2f BB |
613 | return (0); |
614 | } | |
615 | ||
428870ff BB |
616 | int |
617 | dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) | |
34dc7c2f | 618 | { |
428870ff | 619 | int err = 0; |
34dc7c2f | 620 | |
47dfff3b MA |
621 | /* |
622 | * We shouldn't be doing anything with dsl_dataset_t's unless the | |
623 | * pool_config lock is held, or the dataset is long-held. | |
624 | */ | |
625 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || | |
626 | dsl_dataset_long_held(ds)); | |
627 | ||
34dc7c2f | 628 | mutex_enter(&ds->ds_opening_lock); |
9b67f605 MA |
629 | if (ds->ds_objset == NULL) { |
630 | objset_t *os; | |
cc9bb3e5 | 631 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
34dc7c2f | 632 | err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), |
9b67f605 | 633 | ds, dsl_dataset_get_blkptr(ds), &os); |
cc9bb3e5 | 634 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
9b67f605 MA |
635 | |
636 | if (err == 0) { | |
637 | mutex_enter(&ds->ds_lock); | |
638 | ASSERT(ds->ds_objset == NULL); | |
639 | ds->ds_objset = os; | |
640 | mutex_exit(&ds->ds_lock); | |
641 | } | |
34dc7c2f | 642 | } |
9b67f605 | 643 | *osp = ds->ds_objset; |
34dc7c2f | 644 | mutex_exit(&ds->ds_opening_lock); |
428870ff | 645 | return (err); |
34dc7c2f BB |
646 | } |
647 | ||
13fe0198 MA |
648 | /* |
649 | * Holds the pool while the objset is held. Therefore only one objset | |
650 | * can be held at a time. | |
651 | */ | |
34dc7c2f | 652 | int |
b5256303 TC |
653 | dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag, |
654 | objset_t **osp) | |
34dc7c2f | 655 | { |
13fe0198 | 656 | dsl_pool_t *dp; |
428870ff | 657 | dsl_dataset_t *ds; |
34dc7c2f | 658 | int err; |
b5256303 | 659 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
34dc7c2f | 660 | |
13fe0198 MA |
661 | err = dsl_pool_hold(name, tag, &dp); |
662 | if (err != 0) | |
663 | return (err); | |
b5256303 | 664 | err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); |
13fe0198 MA |
665 | if (err != 0) { |
666 | dsl_pool_rele(dp, tag); | |
428870ff | 667 | return (err); |
13fe0198 | 668 | } |
428870ff BB |
669 | |
670 | err = dmu_objset_from_ds(ds, osp); | |
13fe0198 | 671 | if (err != 0) { |
428870ff | 672 | dsl_dataset_rele(ds, tag); |
13fe0198 MA |
673 | dsl_pool_rele(dp, tag); |
674 | } | |
428870ff | 675 | |
34dc7c2f BB |
676 | return (err); |
677 | } | |
678 | ||
b5256303 TC |
679 | int |
680 | dmu_objset_hold(const char *name, void *tag, objset_t **osp) | |
681 | { | |
682 | return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); | |
683 | } | |
684 | ||
9c43027b AJ |
685 | static int |
686 | dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, | |
b5256303 | 687 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
9c43027b AJ |
688 | { |
689 | int err; | |
690 | ||
691 | err = dmu_objset_from_ds(ds, osp); | |
692 | if (err != 0) { | |
b5256303 | 693 | return (err); |
9c43027b | 694 | } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { |
9c43027b AJ |
695 | return (SET_ERROR(EINVAL)); |
696 | } else if (!readonly && dsl_dataset_is_snapshot(ds)) { | |
9c43027b | 697 | return (SET_ERROR(EROFS)); |
ae76f45c TC |
698 | } else if (!readonly && decrypt && |
699 | dsl_dir_incompatible_encryption_version(ds->ds_dir)) { | |
700 | return (SET_ERROR(EROFS)); | |
9c43027b | 701 | } |
b5256303 TC |
702 | |
703 | /* if we are decrypting, we can now check MACs in os->os_phys_buf */ | |
704 | if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { | |
a2c2ed1b TC |
705 | zbookmark_phys_t zb; |
706 | ||
707 | SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, | |
708 | ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
b5256303 | 709 | err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, |
a2c2ed1b | 710 | &zb, B_FALSE); |
b5256303 TC |
711 | if (err != 0) |
712 | return (err); | |
713 | ||
714 | ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); | |
715 | } | |
716 | ||
717 | return (0); | |
9c43027b AJ |
718 | } |
719 | ||
13fe0198 MA |
720 | /* |
721 | * dsl_pool must not be held when this is called. | |
722 | * Upon successful return, there will be a longhold on the dataset, | |
723 | * and the dsl_pool will not be held. | |
724 | */ | |
34dc7c2f | 725 | int |
428870ff | 726 | dmu_objset_own(const char *name, dmu_objset_type_t type, |
b5256303 | 727 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
34dc7c2f | 728 | { |
13fe0198 | 729 | dsl_pool_t *dp; |
34dc7c2f BB |
730 | dsl_dataset_t *ds; |
731 | int err; | |
b5256303 | 732 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
34dc7c2f | 733 | |
13fe0198 MA |
734 | err = dsl_pool_hold(name, FTAG, &dp); |
735 | if (err != 0) | |
736 | return (err); | |
b5256303 | 737 | err = dsl_dataset_own(dp, name, flags, tag, &ds); |
13fe0198 MA |
738 | if (err != 0) { |
739 | dsl_pool_rele(dp, FTAG); | |
34dc7c2f | 740 | return (err); |
13fe0198 | 741 | } |
b5256303 TC |
742 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
743 | if (err != 0) { | |
744 | dsl_dataset_disown(ds, flags, tag); | |
745 | dsl_pool_rele(dp, FTAG); | |
746 | return (err); | |
747 | } | |
748 | ||
163a8c28 TC |
749 | /* |
750 | * User accounting requires the dataset to be decrypted and rw. | |
751 | * We also don't begin user accounting during claiming to help | |
752 | * speed up pool import times and to keep this txg reserved | |
753 | * completely for recovery work. | |
754 | */ | |
9c5167d1 NF |
755 | if ((dmu_objset_userobjspace_upgradable(*osp) || |
756 | dmu_objset_projectquota_upgradable(*osp)) && | |
163a8c28 | 757 | !readonly && !dp->dp_spa->spa_claiming && |
4807c0ba | 758 | (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) |
9c5167d1 | 759 | dmu_objset_id_quota_upgrade(*osp); |
1de321e6 | 760 | |
c0daec32 | 761 | dsl_pool_rele(dp, FTAG); |
b5256303 | 762 | return (0); |
34dc7c2f BB |
763 | } |
764 | ||
9c43027b AJ |
765 | int |
766 | dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, | |
b5256303 | 767 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
9c43027b AJ |
768 | { |
769 | dsl_dataset_t *ds; | |
770 | int err; | |
b5256303 | 771 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
9c43027b | 772 | |
b5256303 | 773 | err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); |
9c43027b AJ |
774 | if (err != 0) |
775 | return (err); | |
776 | ||
b5256303 TC |
777 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
778 | if (err != 0) { | |
779 | dsl_dataset_disown(ds, flags, tag); | |
780 | return (err); | |
781 | } | |
782 | ||
783 | return (0); | |
9c43027b AJ |
784 | } |
785 | ||
34dc7c2f | 786 | void |
b5256303 | 787 | dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag) |
34dc7c2f | 788 | { |
b5256303 TC |
789 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
790 | ||
13fe0198 | 791 | dsl_pool_t *dp = dmu_objset_pool(os); |
b5256303 | 792 | dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); |
13fe0198 | 793 | dsl_pool_rele(dp, tag); |
428870ff | 794 | } |
b128c09f | 795 | |
b5256303 TC |
796 | void |
797 | dmu_objset_rele(objset_t *os, void *tag) | |
798 | { | |
799 | dmu_objset_rele_flags(os, B_FALSE, tag); | |
800 | } | |
801 | ||
831baf06 KW |
802 | /* |
803 | * When we are called, os MUST refer to an objset associated with a dataset | |
804 | * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner | |
805 | * == tag. We will then release and reacquire ownership of the dataset while | |
806 | * holding the pool config_rwlock to avoid intervening namespace or ownership | |
807 | * changes may occur. | |
808 | * | |
809 | * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to | |
810 | * release the hold on its dataset and acquire a new one on the dataset of the | |
811 | * same name so that it can be partially torn down and reconstructed. | |
812 | */ | |
813 | void | |
5e00213e AG |
814 | dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, |
815 | boolean_t decrypt, void *tag) | |
831baf06 KW |
816 | { |
817 | dsl_pool_t *dp; | |
eca7b760 | 818 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
831baf06 | 819 | |
831baf06 KW |
820 | VERIFY3P(ds, !=, NULL); |
821 | VERIFY3P(ds->ds_owner, ==, tag); | |
822 | VERIFY(dsl_dataset_long_held(ds)); | |
823 | ||
824 | dsl_dataset_name(ds, name); | |
5e00213e | 825 | dp = ds->ds_dir->dd_pool; |
831baf06 | 826 | dsl_pool_config_enter(dp, FTAG); |
5e00213e | 827 | dsl_dataset_disown(ds, decrypt, tag); |
b5256303 | 828 | VERIFY0(dsl_dataset_own(dp, name, |
5e00213e | 829 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag, newds)); |
831baf06 KW |
830 | dsl_pool_config_exit(dp, FTAG); |
831 | } | |
832 | ||
428870ff | 833 | void |
b5256303 | 834 | dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag) |
428870ff | 835 | { |
1de321e6 JX |
836 | /* |
837 | * Stop upgrading thread | |
838 | */ | |
839 | dmu_objset_upgrade_stop(os); | |
b5256303 TC |
840 | dsl_dataset_disown(os->os_dsl_dataset, |
841 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag); | |
34dc7c2f BB |
842 | } |
843 | ||
13fe0198 | 844 | void |
34dc7c2f BB |
845 | dmu_objset_evict_dbufs(objset_t *os) |
846 | { | |
0c66c32d | 847 | dnode_t *dn_marker; |
34dc7c2f BB |
848 | dnode_t *dn; |
849 | ||
0c66c32d | 850 | dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); |
34dc7c2f | 851 | |
0c66c32d JG |
852 | mutex_enter(&os->os_lock); |
853 | dn = list_head(&os->os_dnodes); | |
854 | while (dn != NULL) { | |
855 | /* | |
856 | * Skip dnodes without holds. We have to do this dance | |
857 | * because dnode_add_ref() only works if there is already a | |
858 | * hold. If the dnode has no holds, then it has no dbufs. | |
859 | */ | |
860 | if (dnode_add_ref(dn, FTAG)) { | |
861 | list_insert_after(&os->os_dnodes, dn, dn_marker); | |
862 | mutex_exit(&os->os_lock); | |
34dc7c2f | 863 | |
0c66c32d JG |
864 | dnode_evict_dbufs(dn); |
865 | dnode_rele(dn, FTAG); | |
34dc7c2f | 866 | |
0c66c32d JG |
867 | mutex_enter(&os->os_lock); |
868 | dn = list_next(&os->os_dnodes, dn_marker); | |
869 | list_remove(&os->os_dnodes, dn_marker); | |
870 | } else { | |
871 | dn = list_next(&os->os_dnodes, dn); | |
872 | } | |
873 | } | |
874 | mutex_exit(&os->os_lock); | |
34dc7c2f | 875 | |
0c66c32d | 876 | kmem_free(dn_marker, sizeof (dnode_t)); |
34dc7c2f | 877 | |
0c66c32d | 878 | if (DMU_USERUSED_DNODE(os) != NULL) { |
9c5167d1 NF |
879 | if (DMU_PROJECTUSED_DNODE(os) != NULL) |
880 | dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); | |
0c66c32d JG |
881 | dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); |
882 | dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); | |
34dc7c2f | 883 | } |
0c66c32d | 884 | dnode_evict_dbufs(DMU_META_DNODE(os)); |
34dc7c2f BB |
885 | } |
886 | ||
0c66c32d JG |
887 | /* |
888 | * Objset eviction processing is split into into two pieces. | |
889 | * The first marks the objset as evicting, evicts any dbufs that | |
890 | * have a refcount of zero, and then queues up the objset for the | |
891 | * second phase of eviction. Once os->os_dnodes has been cleared by | |
892 | * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. | |
893 | * The second phase closes the special dnodes, dequeues the objset from | |
894 | * the list of those undergoing eviction, and finally frees the objset. | |
895 | * | |
896 | * NOTE: Due to asynchronous eviction processing (invocation of | |
897 | * dnode_buf_pageout()), it is possible for the meta dnode for the | |
898 | * objset to have no holds even though os->os_dnodes is not empty. | |
899 | */ | |
34dc7c2f | 900 | void |
428870ff | 901 | dmu_objset_evict(objset_t *os) |
34dc7c2f | 902 | { |
6f1ffb06 MA |
903 | dsl_dataset_t *ds = os->os_dsl_dataset; |
904 | ||
1c27024e | 905 | for (int t = 0; t < TXG_SIZE; t++) |
428870ff | 906 | ASSERT(!dmu_objset_is_dirty(os, t)); |
34dc7c2f | 907 | |
0eb21616 JG |
908 | if (ds) |
909 | dsl_prop_unregister_all(ds, os); | |
34dc7c2f | 910 | |
428870ff BB |
911 | if (os->os_sa) |
912 | sa_tear_down(os); | |
913 | ||
13fe0198 | 914 | dmu_objset_evict_dbufs(os); |
34dc7c2f | 915 | |
0c66c32d JG |
916 | mutex_enter(&os->os_lock); |
917 | spa_evicting_os_register(os->os_spa, os); | |
918 | if (list_is_empty(&os->os_dnodes)) { | |
919 | mutex_exit(&os->os_lock); | |
920 | dmu_objset_evict_done(os); | |
921 | } else { | |
922 | mutex_exit(&os->os_lock); | |
923 | } | |
b5256303 TC |
924 | |
925 | ||
0c66c32d JG |
926 | } |
927 | ||
928 | void | |
929 | dmu_objset_evict_done(objset_t *os) | |
930 | { | |
931 | ASSERT3P(list_head(&os->os_dnodes), ==, NULL); | |
932 | ||
572e2857 BB |
933 | dnode_special_close(&os->os_meta_dnode); |
934 | if (DMU_USERUSED_DNODE(os)) { | |
9c5167d1 NF |
935 | if (DMU_PROJECTUSED_DNODE(os)) |
936 | dnode_special_close(&os->os_projectused_dnode); | |
572e2857 BB |
937 | dnode_special_close(&os->os_userused_dnode); |
938 | dnode_special_close(&os->os_groupused_dnode); | |
9babb374 | 939 | } |
428870ff BB |
940 | zil_free(os->os_zil); |
941 | ||
d3c2ae1c | 942 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
572e2857 BB |
943 | |
944 | /* | |
945 | * This is a barrier to prevent the objset from going away in | |
946 | * dnode_move() until we can safely ensure that the objset is still in | |
947 | * use. We consider the objset valid before the barrier and invalid | |
948 | * after the barrier. | |
949 | */ | |
950 | rw_enter(&os_lock, RW_READER); | |
951 | rw_exit(&os_lock); | |
952 | ||
dbeb8796 MA |
953 | kmem_free(os->os_obj_next_percpu, |
954 | os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); | |
955 | ||
428870ff | 956 | mutex_destroy(&os->os_lock); |
64fc7762 | 957 | mutex_destroy(&os->os_userused_lock); |
428870ff BB |
958 | mutex_destroy(&os->os_obj_lock); |
959 | mutex_destroy(&os->os_user_ptr_lock); | |
c17486b2 | 960 | mutex_destroy(&os->os_upgrade_lock); |
64fc7762 MA |
961 | for (int i = 0; i < TXG_SIZE; i++) { |
962 | multilist_destroy(os->os_dirty_dnodes[i]); | |
963 | } | |
0c66c32d | 964 | spa_evicting_os_deregister(os->os_spa, os); |
428870ff BB |
965 | kmem_free(os, sizeof (objset_t)); |
966 | } | |
9babb374 | 967 | |
428870ff BB |
968 | timestruc_t |
969 | dmu_objset_snap_cmtime(objset_t *os) | |
970 | { | |
971 | return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); | |
34dc7c2f BB |
972 | } |
973 | ||
428870ff | 974 | objset_t * |
b5256303 TC |
975 | dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, |
976 | dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) | |
34dc7c2f | 977 | { |
428870ff | 978 | objset_t *os; |
34dc7c2f BB |
979 | dnode_t *mdn; |
980 | ||
981 | ASSERT(dmu_tx_is_syncing(tx)); | |
13fe0198 | 982 | |
b5256303 TC |
983 | if (blksz == 0) |
984 | blksz = DNODE_BLOCK_SIZE; | |
4807c0ba | 985 | if (ibs == 0) |
b5256303 TC |
986 | ibs = DN_MAX_INDBLKSHIFT; |
987 | ||
572e2857 | 988 | if (ds != NULL) |
13fe0198 | 989 | VERIFY0(dmu_objset_from_ds(ds, &os)); |
572e2857 | 990 | else |
13fe0198 | 991 | VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); |
572e2857 BB |
992 | |
993 | mdn = DMU_META_DNODE(os); | |
34dc7c2f | 994 | |
b5256303 TC |
995 | dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, |
996 | DNODE_MIN_SLOTS, tx); | |
34dc7c2f BB |
997 | |
998 | /* | |
999 | * We don't want to have to increase the meta-dnode's nlevels | |
1000 | * later, because then we could do it in quescing context while | |
1001 | * we are also accessing it in open context. | |
1002 | * | |
1003 | * This precaution is not necessary for the MOS (ds == NULL), | |
1004 | * because the MOS is only updated in syncing context. | |
1005 | * This is most fortunate: the MOS is the only objset that | |
1006 | * needs to be synced multiple times as spa_sync() iterates | |
1007 | * to convergence, so minimizing its dn_nlevels matters. | |
1008 | */ | |
1009 | if (ds != NULL) { | |
b5256303 TC |
1010 | if (levels == 0) { |
1011 | levels = 1; | |
1012 | ||
1013 | /* | |
1014 | * Determine the number of levels necessary for the | |
1015 | * meta-dnode to contain DN_MAX_OBJECT dnodes. Note | |
1016 | * that in order to ensure that we do not overflow | |
1017 | * 64 bits, there has to be a nlevels that gives us a | |
1018 | * number of blocks > DN_MAX_OBJECT but < 2^64. | |
1019 | * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) | |
1020 | * (10) must be less than (64 - log2(DN_MAX_OBJECT)) | |
1021 | * (16). | |
1022 | */ | |
1023 | while ((uint64_t)mdn->dn_nblkptr << | |
1024 | (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * | |
1025 | (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < | |
1026 | DN_MAX_OBJECT) | |
1027 | levels++; | |
1028 | } | |
34dc7c2f BB |
1029 | |
1030 | mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = | |
1031 | mdn->dn_nlevels = levels; | |
1032 | } | |
1033 | ||
1034 | ASSERT(type != DMU_OST_NONE); | |
1035 | ASSERT(type != DMU_OST_ANY); | |
1036 | ASSERT(type < DMU_OST_NUMTYPES); | |
428870ff | 1037 | os->os_phys->os_type = type; |
b5256303 TC |
1038 | |
1039 | /* | |
1040 | * Enable user accounting if it is enabled and this is not an | |
1041 | * encrypted receive. | |
1042 | */ | |
1043 | if (dmu_objset_userused_enabled(os) && | |
1044 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
428870ff | 1045 | os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
1de321e6 JX |
1046 | if (dmu_objset_userobjused_enabled(os)) { |
1047 | ds->ds_feature_activation_needed[ | |
1048 | SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE; | |
1049 | os->os_phys->os_flags |= | |
1050 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
1051 | } | |
9c5167d1 NF |
1052 | if (dmu_objset_projectquota_enabled(os)) { |
1053 | ds->ds_feature_activation_needed[ | |
1054 | SPA_FEATURE_PROJECT_QUOTA] = B_TRUE; | |
1055 | os->os_phys->os_flags |= | |
1056 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
1057 | } | |
428870ff | 1058 | os->os_flags = os->os_phys->os_flags; |
9babb374 | 1059 | } |
34dc7c2f BB |
1060 | |
1061 | dsl_dataset_dirty(ds, tx); | |
1062 | ||
428870ff | 1063 | return (os); |
34dc7c2f BB |
1064 | } |
1065 | ||
b5256303 TC |
1066 | /* called from dsl for meta-objset */ |
1067 | objset_t * | |
1068 | dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
1069 | dmu_objset_type_t type, dmu_tx_t *tx) | |
1070 | { | |
1071 | return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); | |
1072 | } | |
1073 | ||
13fe0198 MA |
1074 | typedef struct dmu_objset_create_arg { |
1075 | const char *doca_name; | |
1076 | cred_t *doca_cred; | |
1077 | void (*doca_userfunc)(objset_t *os, void *arg, | |
1078 | cred_t *cr, dmu_tx_t *tx); | |
1079 | void *doca_userarg; | |
1080 | dmu_objset_type_t doca_type; | |
1081 | uint64_t doca_flags; | |
b5256303 | 1082 | dsl_crypto_params_t *doca_dcp; |
13fe0198 | 1083 | } dmu_objset_create_arg_t; |
34dc7c2f BB |
1084 | |
1085 | /*ARGSUSED*/ | |
1086 | static int | |
13fe0198 | 1087 | dmu_objset_create_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1088 | { |
13fe0198 MA |
1089 | dmu_objset_create_arg_t *doca = arg; |
1090 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1091 | dsl_dir_t *pdd; | |
1092 | const char *tail; | |
1093 | int error; | |
34dc7c2f | 1094 | |
13fe0198 | 1095 | if (strchr(doca->doca_name, '@') != NULL) |
2e528b49 | 1096 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 1097 | |
eca7b760 IK |
1098 | if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) |
1099 | return (SET_ERROR(ENAMETOOLONG)); | |
1100 | ||
13fe0198 MA |
1101 | error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); |
1102 | if (error != 0) | |
1103 | return (error); | |
1104 | if (tail == NULL) { | |
1105 | dsl_dir_rele(pdd, FTAG); | |
2e528b49 | 1106 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1107 | } |
b5256303 TC |
1108 | |
1109 | error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp); | |
1110 | if (error != 0) { | |
1111 | dsl_dir_rele(pdd, FTAG); | |
1112 | return (error); | |
1113 | } | |
1114 | ||
788eb90c JJ |
1115 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
1116 | doca->doca_cred); | |
b5256303 | 1117 | |
13fe0198 | 1118 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f | 1119 | |
788eb90c | 1120 | return (error); |
34dc7c2f BB |
1121 | } |
1122 | ||
1123 | static void | |
13fe0198 | 1124 | dmu_objset_create_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1125 | { |
13fe0198 MA |
1126 | dmu_objset_create_arg_t *doca = arg; |
1127 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1128 | dsl_dir_t *pdd; | |
1129 | const char *tail; | |
6f1ffb06 | 1130 | dsl_dataset_t *ds; |
13fe0198 | 1131 | uint64_t obj; |
6f1ffb06 | 1132 | blkptr_t *bp; |
13fe0198 | 1133 | objset_t *os; |
b5256303 | 1134 | zio_t *rzio; |
34dc7c2f | 1135 | |
13fe0198 | 1136 | VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); |
34dc7c2f | 1137 | |
13fe0198 | 1138 | obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, |
b5256303 | 1139 | doca->doca_cred, doca->doca_dcp, tx); |
34dc7c2f | 1140 | |
b5256303 TC |
1141 | VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, |
1142 | DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); | |
cc9bb3e5 | 1143 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
6f1ffb06 | 1144 | bp = dsl_dataset_get_blkptr(ds); |
13fe0198 MA |
1145 | os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, |
1146 | ds, bp, doca->doca_type, tx); | |
cc9bb3e5 | 1147 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
34dc7c2f | 1148 | |
13fe0198 MA |
1149 | if (doca->doca_userfunc != NULL) { |
1150 | doca->doca_userfunc(os, doca->doca_userarg, | |
1151 | doca->doca_cred, tx); | |
34dc7c2f BB |
1152 | } |
1153 | ||
b5256303 | 1154 | /* |
4807c0ba | 1155 | * The doca_userfunc() may write out some data that needs to be |
b5256303 TC |
1156 | * encrypted if the dataset is encrypted (specifically the root |
1157 | * directory). This data must be written out before the encryption | |
1158 | * key mapping is removed by dsl_dataset_rele_flags(). Force the | |
1159 | * I/O to occur immediately by invoking the relevant sections of | |
1160 | * dsl_pool_sync(). | |
1161 | */ | |
1162 | if (os->os_encrypted) { | |
1163 | dsl_dataset_t *tmpds = NULL; | |
1164 | boolean_t need_sync_done = B_FALSE; | |
1165 | ||
4807c0ba TC |
1166 | mutex_enter(&ds->ds_lock); |
1167 | ds->ds_owner = FTAG; | |
1168 | mutex_exit(&ds->ds_lock); | |
1169 | ||
b5256303 | 1170 | rzio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
4807c0ba TC |
1171 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, |
1172 | tx->tx_txg); | |
b5256303 | 1173 | if (tmpds != NULL) { |
b5256303 TC |
1174 | dsl_dataset_sync(ds, rzio, tx); |
1175 | need_sync_done = B_TRUE; | |
1176 | } | |
1177 | VERIFY0(zio_wait(rzio)); | |
1178 | ||
1179 | dmu_objset_do_userquota_updates(os, tx); | |
1180 | taskq_wait(dp->dp_sync_taskq); | |
1181 | ||
1182 | rzio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
4807c0ba TC |
1183 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, |
1184 | tx->tx_txg); | |
b5256303 | 1185 | if (tmpds != NULL) { |
b5256303 TC |
1186 | dmu_buf_rele(ds->ds_dbuf, ds); |
1187 | dsl_dataset_sync(ds, rzio, tx); | |
1188 | } | |
1189 | VERIFY0(zio_wait(rzio)); | |
1190 | ||
1191 | if (need_sync_done) | |
1192 | dsl_dataset_sync_done(ds, tx); | |
4807c0ba TC |
1193 | |
1194 | mutex_enter(&ds->ds_lock); | |
1195 | ds->ds_owner = NULL; | |
1196 | mutex_exit(&ds->ds_lock); | |
b5256303 TC |
1197 | } |
1198 | ||
13fe0198 | 1199 | spa_history_log_internal_ds(ds, "create", tx, ""); |
a0bd735a BP |
1200 | zvol_create_minors(dp->dp_spa, doca->doca_name, B_TRUE); |
1201 | ||
b5256303 | 1202 | dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); |
13fe0198 | 1203 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f BB |
1204 | } |
1205 | ||
1206 | int | |
428870ff | 1207 | dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, |
b5256303 | 1208 | dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) |
34dc7c2f | 1209 | { |
13fe0198 | 1210 | dmu_objset_create_arg_t doca; |
b5256303 | 1211 | dsl_crypto_params_t tmp_dcp = { 0 }; |
34dc7c2f | 1212 | |
13fe0198 MA |
1213 | doca.doca_name = name; |
1214 | doca.doca_cred = CRED(); | |
1215 | doca.doca_flags = flags; | |
1216 | doca.doca_userfunc = func; | |
1217 | doca.doca_userarg = arg; | |
1218 | doca.doca_type = type; | |
34dc7c2f | 1219 | |
b5256303 TC |
1220 | /* |
1221 | * Some callers (mostly for testing) do not provide a dcp on their | |
1222 | * own but various code inside the sync task will require it to be | |
1223 | * allocated. Rather than adding NULL checks throughout this code | |
1224 | * or adding dummy dcp's to all of the callers we simply create a | |
1225 | * dummy one here and use that. This zero dcp will have the same | |
1226 | * effect as asking for inheritence of all encryption params. | |
1227 | */ | |
1228 | doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; | |
1229 | ||
13fe0198 | 1230 | return (dsl_sync_task(name, |
3d45fdd6 | 1231 | dmu_objset_create_check, dmu_objset_create_sync, &doca, |
b5256303 | 1232 | 6, ZFS_SPACE_CHECK_NORMAL)); |
34dc7c2f BB |
1233 | } |
1234 | ||
13fe0198 MA |
1235 | typedef struct dmu_objset_clone_arg { |
1236 | const char *doca_clone; | |
1237 | const char *doca_origin; | |
1238 | cred_t *doca_cred; | |
1239 | } dmu_objset_clone_arg_t; | |
1240 | ||
1241 | /*ARGSUSED*/ | |
1242 | static int | |
1243 | dmu_objset_clone_check(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1244 | { |
13fe0198 | 1245 | dmu_objset_clone_arg_t *doca = arg; |
428870ff BB |
1246 | dsl_dir_t *pdd; |
1247 | const char *tail; | |
13fe0198 MA |
1248 | int error; |
1249 | dsl_dataset_t *origin; | |
1250 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
34dc7c2f | 1251 | |
13fe0198 | 1252 | if (strchr(doca->doca_clone, '@') != NULL) |
2e528b49 | 1253 | return (SET_ERROR(EINVAL)); |
13fe0198 | 1254 | |
eca7b760 IK |
1255 | if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) |
1256 | return (SET_ERROR(ENAMETOOLONG)); | |
1257 | ||
13fe0198 MA |
1258 | error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); |
1259 | if (error != 0) | |
1260 | return (error); | |
428870ff | 1261 | if (tail == NULL) { |
13fe0198 | 1262 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1263 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1264 | } |
1cddb8c9 | 1265 | |
788eb90c JJ |
1266 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
1267 | doca->doca_cred); | |
1268 | if (error != 0) { | |
1269 | dsl_dir_rele(pdd, FTAG); | |
1270 | return (SET_ERROR(EDQUOT)); | |
1271 | } | |
428870ff | 1272 | |
13fe0198 | 1273 | error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); |
b5256303 TC |
1274 | if (error != 0) { |
1275 | dsl_dir_rele(pdd, FTAG); | |
572e2857 | 1276 | return (error); |
b5256303 | 1277 | } |
572e2857 | 1278 | |
13fe0198 | 1279 | /* You can only clone snapshots, not the head datasets. */ |
0c66c32d | 1280 | if (!origin->ds_is_snapshot) { |
13fe0198 | 1281 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1282 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1283 | return (SET_ERROR(EINVAL)); |
428870ff | 1284 | } |
b5256303 TC |
1285 | |
1286 | error = dmu_objset_clone_crypt_check(pdd, origin->ds_dir); | |
1287 | if (error != 0) { | |
1288 | dsl_dataset_rele(origin, FTAG); | |
1289 | dsl_dir_rele(pdd, FTAG); | |
1290 | return (error); | |
1291 | } | |
1292 | ||
13fe0198 | 1293 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1294 | dsl_dir_rele(pdd, FTAG); |
572e2857 | 1295 | |
13fe0198 | 1296 | return (0); |
9babb374 | 1297 | } |
34dc7c2f | 1298 | |
13fe0198 MA |
1299 | static void |
1300 | dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1301 | { |
13fe0198 MA |
1302 | dmu_objset_clone_arg_t *doca = arg; |
1303 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1304 | dsl_dir_t *pdd; | |
1305 | const char *tail; | |
1306 | dsl_dataset_t *origin, *ds; | |
1307 | uint64_t obj; | |
eca7b760 | 1308 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; |
6f1ffb06 | 1309 | |
13fe0198 MA |
1310 | VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); |
1311 | VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); | |
6f1ffb06 | 1312 | |
13fe0198 | 1313 | obj = dsl_dataset_create_sync(pdd, tail, origin, 0, |
b5256303 | 1314 | doca->doca_cred, NULL, tx); |
34dc7c2f | 1315 | |
13fe0198 MA |
1316 | VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); |
1317 | dsl_dataset_name(origin, namebuf); | |
1318 | spa_history_log_internal_ds(ds, "clone", tx, | |
1319 | "origin=%s (%llu)", namebuf, origin->ds_object); | |
a0bd735a | 1320 | zvol_create_minors(dp->dp_spa, doca->doca_clone, B_TRUE); |
13fe0198 MA |
1321 | dsl_dataset_rele(ds, FTAG); |
1322 | dsl_dataset_rele(origin, FTAG); | |
1323 | dsl_dir_rele(pdd, FTAG); | |
34dc7c2f BB |
1324 | } |
1325 | ||
1326 | int | |
13fe0198 | 1327 | dmu_objset_clone(const char *clone, const char *origin) |
34dc7c2f | 1328 | { |
13fe0198 | 1329 | dmu_objset_clone_arg_t doca; |
34dc7c2f | 1330 | |
13fe0198 MA |
1331 | doca.doca_clone = clone; |
1332 | doca.doca_origin = origin; | |
1333 | doca.doca_cred = CRED(); | |
572e2857 | 1334 | |
13fe0198 | 1335 | return (dsl_sync_task(clone, |
3d45fdd6 | 1336 | dmu_objset_clone_check, dmu_objset_clone_sync, &doca, |
b5256303 | 1337 | 6, ZFS_SPACE_CHECK_NORMAL)); |
6f1ffb06 MA |
1338 | } |
1339 | ||
a1d477c2 MA |
1340 | static int |
1341 | dmu_objset_remap_indirects_impl(objset_t *os, uint64_t last_removed_txg) | |
1342 | { | |
1343 | int error = 0; | |
1344 | uint64_t object = 0; | |
1345 | while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) { | |
1346 | error = dmu_object_remap_indirects(os, object, | |
1347 | last_removed_txg); | |
1348 | /* | |
1349 | * If the ZPL removed the object before we managed to dnode_hold | |
1350 | * it, we would get an ENOENT. If the ZPL declares its intent | |
1351 | * to remove the object (dnode_free) before we manage to | |
1352 | * dnode_hold it, we would get an EEXIST. In either case, we | |
1353 | * want to continue remapping the other objects in the objset; | |
1354 | * in all other cases, we want to break early. | |
1355 | */ | |
1356 | if (error != 0 && error != ENOENT && error != EEXIST) { | |
1357 | break; | |
1358 | } | |
1359 | } | |
1360 | if (error == ESRCH) { | |
1361 | error = 0; | |
1362 | } | |
1363 | return (error); | |
1364 | } | |
1365 | ||
1366 | int | |
1367 | dmu_objset_remap_indirects(const char *fsname) | |
1368 | { | |
1369 | int error = 0; | |
1370 | objset_t *os = NULL; | |
1371 | uint64_t last_removed_txg; | |
1372 | uint64_t remap_start_txg; | |
1373 | dsl_dir_t *dd; | |
1374 | ||
1375 | error = dmu_objset_hold(fsname, FTAG, &os); | |
1376 | if (error != 0) { | |
1377 | return (error); | |
1378 | } | |
1379 | dd = dmu_objset_ds(os)->ds_dir; | |
1380 | ||
1381 | if (!spa_feature_is_enabled(dmu_objset_spa(os), | |
1382 | SPA_FEATURE_OBSOLETE_COUNTS)) { | |
1383 | dmu_objset_rele(os, FTAG); | |
1384 | return (SET_ERROR(ENOTSUP)); | |
1385 | } | |
1386 | ||
1387 | if (dsl_dataset_is_snapshot(dmu_objset_ds(os))) { | |
1388 | dmu_objset_rele(os, FTAG); | |
1389 | return (SET_ERROR(EINVAL)); | |
1390 | } | |
1391 | ||
1392 | /* | |
1393 | * If there has not been a removal, we're done. | |
1394 | */ | |
1395 | last_removed_txg = spa_get_last_removal_txg(dmu_objset_spa(os)); | |
1396 | if (last_removed_txg == -1ULL) { | |
1397 | dmu_objset_rele(os, FTAG); | |
1398 | return (0); | |
1399 | } | |
1400 | ||
1401 | /* | |
1402 | * If we have remapped since the last removal, we're done. | |
1403 | */ | |
1404 | if (dsl_dir_is_zapified(dd)) { | |
1405 | uint64_t last_remap_txg; | |
1406 | if (zap_lookup(spa_meta_objset(dmu_objset_spa(os)), | |
1407 | dd->dd_object, DD_FIELD_LAST_REMAP_TXG, | |
1408 | sizeof (last_remap_txg), 1, &last_remap_txg) == 0 && | |
1409 | last_remap_txg > last_removed_txg) { | |
1410 | dmu_objset_rele(os, FTAG); | |
1411 | return (0); | |
1412 | } | |
1413 | } | |
1414 | ||
1415 | dsl_dataset_long_hold(dmu_objset_ds(os), FTAG); | |
1416 | dsl_pool_rele(dmu_objset_pool(os), FTAG); | |
1417 | ||
1418 | remap_start_txg = spa_last_synced_txg(dmu_objset_spa(os)); | |
1419 | error = dmu_objset_remap_indirects_impl(os, last_removed_txg); | |
1420 | if (error == 0) { | |
1421 | /* | |
1422 | * We update the last_remap_txg to be the start txg so that | |
1423 | * we can guarantee that every block older than last_remap_txg | |
1424 | * that can be remapped has been remapped. | |
1425 | */ | |
1426 | error = dsl_dir_update_last_remap_txg(dd, remap_start_txg); | |
1427 | } | |
1428 | ||
1429 | dsl_dataset_long_rele(dmu_objset_ds(os), FTAG); | |
1430 | dsl_dataset_rele(dmu_objset_ds(os), FTAG); | |
1431 | ||
1432 | return (error); | |
1433 | } | |
1434 | ||
6f1ffb06 MA |
1435 | int |
1436 | dmu_objset_snapshot_one(const char *fsname, const char *snapname) | |
1437 | { | |
1438 | int err; | |
1439 | char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); | |
1440 | nvlist_t *snaps = fnvlist_alloc(); | |
1441 | ||
1442 | fnvlist_add_boolean(snaps, longsnap); | |
6f1ffb06 | 1443 | strfree(longsnap); |
13fe0198 MA |
1444 | err = dsl_dataset_snapshot(snaps, NULL, NULL); |
1445 | fnvlist_free(snaps); | |
6f1ffb06 MA |
1446 | return (err); |
1447 | } | |
1448 | ||
1de321e6 JX |
1449 | static void |
1450 | dmu_objset_upgrade_task_cb(void *data) | |
1451 | { | |
1452 | objset_t *os = data; | |
1453 | ||
1454 | mutex_enter(&os->os_upgrade_lock); | |
1455 | os->os_upgrade_status = EINTR; | |
1456 | if (!os->os_upgrade_exit) { | |
1457 | mutex_exit(&os->os_upgrade_lock); | |
1458 | ||
1459 | os->os_upgrade_status = os->os_upgrade_cb(os); | |
1460 | mutex_enter(&os->os_upgrade_lock); | |
1461 | } | |
1462 | os->os_upgrade_exit = B_TRUE; | |
1463 | os->os_upgrade_id = 0; | |
1464 | mutex_exit(&os->os_upgrade_lock); | |
c0daec32 | 1465 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); |
1de321e6 JX |
1466 | } |
1467 | ||
1468 | static void | |
1469 | dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) | |
1470 | { | |
1471 | if (os->os_upgrade_id != 0) | |
1472 | return; | |
1473 | ||
c0daec32 AB |
1474 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
1475 | dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); | |
1476 | ||
1de321e6 JX |
1477 | mutex_enter(&os->os_upgrade_lock); |
1478 | if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { | |
1479 | os->os_upgrade_exit = B_FALSE; | |
1480 | os->os_upgrade_cb = cb; | |
1481 | os->os_upgrade_id = taskq_dispatch( | |
1482 | os->os_spa->spa_upgrade_taskq, | |
1483 | dmu_objset_upgrade_task_cb, os, TQ_SLEEP); | |
c0daec32 AB |
1484 | if (os->os_upgrade_id == TASKQID_INVALID) { |
1485 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1de321e6 | 1486 | os->os_upgrade_status = ENOMEM; |
c0daec32 | 1487 | } |
1de321e6 JX |
1488 | } |
1489 | mutex_exit(&os->os_upgrade_lock); | |
1490 | } | |
1491 | ||
1492 | static void | |
1493 | dmu_objset_upgrade_stop(objset_t *os) | |
1494 | { | |
1495 | mutex_enter(&os->os_upgrade_lock); | |
1496 | os->os_upgrade_exit = B_TRUE; | |
1497 | if (os->os_upgrade_id != 0) { | |
1498 | taskqid_t id = os->os_upgrade_id; | |
1499 | ||
1500 | os->os_upgrade_id = 0; | |
1501 | mutex_exit(&os->os_upgrade_lock); | |
1502 | ||
c0daec32 AB |
1503 | if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { |
1504 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1505 | } | |
4807c0ba | 1506 | txg_wait_synced(os->os_spa->spa_dsl_pool, 0); |
1de321e6 JX |
1507 | } else { |
1508 | mutex_exit(&os->os_upgrade_lock); | |
1509 | } | |
1510 | } | |
1511 | ||
34dc7c2f | 1512 | static void |
64fc7762 | 1513 | dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) |
34dc7c2f BB |
1514 | { |
1515 | dnode_t *dn; | |
1516 | ||
64fc7762 | 1517 | while ((dn = multilist_sublist_head(list)) != NULL) { |
34dc7c2f BB |
1518 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); |
1519 | ASSERT(dn->dn_dbuf->db_data_pending); | |
1520 | /* | |
9babb374 BB |
1521 | * Initialize dn_zio outside dnode_sync() because the |
1522 | * meta-dnode needs to set it ouside dnode_sync(). | |
34dc7c2f BB |
1523 | */ |
1524 | dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; | |
1525 | ASSERT(dn->dn_zio); | |
1526 | ||
1527 | ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); | |
64fc7762 | 1528 | multilist_sublist_remove(list, dn); |
9babb374 | 1529 | |
edc1e713 TC |
1530 | /* |
1531 | * If we are not doing useraccounting (os_synced_dnodes == NULL) | |
1532 | * we are done with this dnode for this txg. Unset dn_dirty_txg | |
1533 | * if later txgs aren't dirtying it so that future holders do | |
1534 | * not get a stale value. Otherwise, we will do this in | |
1535 | * userquota_updates_task() when processing has completely | |
1536 | * finished for this txg. | |
1537 | */ | |
64fc7762 MA |
1538 | multilist_t *newlist = dn->dn_objset->os_synced_dnodes; |
1539 | if (newlist != NULL) { | |
9babb374 | 1540 | (void) dnode_add_ref(dn, newlist); |
64fc7762 | 1541 | multilist_insert(newlist, dn); |
edc1e713 TC |
1542 | } else { |
1543 | mutex_enter(&dn->dn_mtx); | |
1544 | if (dn->dn_dirty_txg == tx->tx_txg) | |
1545 | dn->dn_dirty_txg = 0; | |
1546 | mutex_exit(&dn->dn_mtx); | |
9babb374 BB |
1547 | } |
1548 | ||
34dc7c2f BB |
1549 | dnode_sync(dn, tx); |
1550 | } | |
1551 | } | |
1552 | ||
1553 | /* ARGSUSED */ | |
1554 | static void | |
428870ff | 1555 | dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) |
34dc7c2f | 1556 | { |
b128c09f | 1557 | blkptr_t *bp = zio->io_bp; |
428870ff | 1558 | objset_t *os = arg; |
34dc7c2f | 1559 | dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; |
b5256303 | 1560 | uint64_t fill = 0; |
34dc7c2f | 1561 | |
9b67f605 | 1562 | ASSERT(!BP_IS_EMBEDDED(bp)); |
13fe0198 MA |
1563 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); |
1564 | ASSERT0(BP_GET_LEVEL(bp)); | |
34dc7c2f BB |
1565 | |
1566 | /* | |
9babb374 BB |
1567 | * Update rootbp fill count: it should be the number of objects |
1568 | * allocated in the object set (not counting the "special" | |
1569 | * objects that are stored in the objset_phys_t -- the meta | |
9c5167d1 | 1570 | * dnode and user/group/project accounting objects). |
34dc7c2f | 1571 | */ |
1c27024e | 1572 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
b5256303 TC |
1573 | fill += BP_GET_FILL(&dnp->dn_blkptr[i]); |
1574 | ||
1575 | BP_SET_FILL(bp, fill); | |
1576 | ||
cc9bb3e5 GM |
1577 | if (os->os_dsl_dataset != NULL) |
1578 | rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); | |
1579 | *os->os_rootbp = *bp; | |
1580 | if (os->os_dsl_dataset != NULL) | |
1581 | rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); | |
428870ff BB |
1582 | } |
1583 | ||
1584 | /* ARGSUSED */ | |
1585 | static void | |
1586 | dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) | |
1587 | { | |
1588 | blkptr_t *bp = zio->io_bp; | |
1589 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1590 | objset_t *os = arg; | |
34dc7c2f | 1591 | |
b128c09f | 1592 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
428870ff | 1593 | ASSERT(BP_EQUAL(bp, bp_orig)); |
b128c09f | 1594 | } else { |
428870ff BB |
1595 | dsl_dataset_t *ds = os->os_dsl_dataset; |
1596 | dmu_tx_t *tx = os->os_synctx; | |
1597 | ||
1598 | (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); | |
1599 | dsl_dataset_block_born(ds, bp, tx); | |
34dc7c2f | 1600 | } |
cc9bb3e5 | 1601 | kmem_free(bp, sizeof (*bp)); |
34dc7c2f BB |
1602 | } |
1603 | ||
64fc7762 MA |
1604 | typedef struct sync_dnodes_arg { |
1605 | multilist_t *sda_list; | |
1606 | int sda_sublist_idx; | |
1607 | multilist_t *sda_newlist; | |
1608 | dmu_tx_t *sda_tx; | |
1609 | } sync_dnodes_arg_t; | |
1610 | ||
1611 | static void | |
1612 | sync_dnodes_task(void *arg) | |
1613 | { | |
1614 | sync_dnodes_arg_t *sda = arg; | |
1615 | ||
1616 | multilist_sublist_t *ms = | |
1617 | multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); | |
1618 | ||
1619 | dmu_objset_sync_dnodes(ms, sda->sda_tx); | |
1620 | ||
1621 | multilist_sublist_unlock(ms); | |
1622 | ||
1623 | kmem_free(sda, sizeof (*sda)); | |
1624 | } | |
1625 | ||
1626 | ||
34dc7c2f BB |
1627 | /* called from dsl */ |
1628 | void | |
428870ff | 1629 | dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) |
34dc7c2f BB |
1630 | { |
1631 | int txgoff; | |
5dbd68a3 | 1632 | zbookmark_phys_t zb; |
428870ff | 1633 | zio_prop_t zp; |
34dc7c2f BB |
1634 | zio_t *zio; |
1635 | list_t *list; | |
1636 | dbuf_dirty_record_t *dr; | |
cc9bb3e5 GM |
1637 | blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); |
1638 | *blkptr_copy = *os->os_rootbp; | |
34dc7c2f BB |
1639 | |
1640 | dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); | |
1641 | ||
1642 | ASSERT(dmu_tx_is_syncing(tx)); | |
1643 | /* XXX the write_done callback should really give us the tx... */ | |
1644 | os->os_synctx = tx; | |
1645 | ||
1646 | if (os->os_dsl_dataset == NULL) { | |
1647 | /* | |
1648 | * This is the MOS. If we have upgraded, | |
1649 | * spa_max_replication() could change, so reset | |
1650 | * os_copies here. | |
1651 | */ | |
1652 | os->os_copies = spa_max_replication(os->os_spa); | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * Create the root block IO | |
1657 | */ | |
428870ff BB |
1658 | SET_BOOKMARK(&zb, os->os_dsl_dataset ? |
1659 | os->os_dsl_dataset->ds_object : DMU_META_OBJSET, | |
1660 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
294f6806 | 1661 | arc_release(os->os_phys_buf, &os->os_phys_buf); |
b128c09f | 1662 | |
82644107 | 1663 | dmu_write_policy(os, NULL, 0, 0, &zp); |
9babb374 | 1664 | |
b5256303 TC |
1665 | /* |
1666 | * If we are either claiming the ZIL or doing a raw receive write out | |
1667 | * the os_phys_buf raw. Neither of these actions will effect the MAC | |
1668 | * at this point. | |
1669 | */ | |
1b66810b | 1670 | if (os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { |
b5256303 | 1671 | ASSERT(os->os_encrypted); |
1b66810b | 1672 | os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE; |
b5256303 TC |
1673 | arc_convert_to_raw(os->os_phys_buf, |
1674 | os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, | |
1675 | DMU_OT_OBJSET, NULL, NULL, NULL); | |
1676 | } | |
1677 | ||
428870ff | 1678 | zio = arc_write(pio, os->os_spa, tx->tx_txg, |
cc9bb3e5 | 1679 | blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), |
bc77ba73 PD |
1680 | &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, |
1681 | os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); | |
34dc7c2f BB |
1682 | |
1683 | /* | |
9babb374 | 1684 | * Sync special dnodes - the parent IO for the sync is the root block |
34dc7c2f | 1685 | */ |
572e2857 BB |
1686 | DMU_META_DNODE(os)->dn_zio = zio; |
1687 | dnode_sync(DMU_META_DNODE(os), tx); | |
34dc7c2f | 1688 | |
9babb374 BB |
1689 | os->os_phys->os_flags = os->os_flags; |
1690 | ||
572e2857 BB |
1691 | if (DMU_USERUSED_DNODE(os) && |
1692 | DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1693 | DMU_USERUSED_DNODE(os)->dn_zio = zio; | |
1694 | dnode_sync(DMU_USERUSED_DNODE(os), tx); | |
1695 | DMU_GROUPUSED_DNODE(os)->dn_zio = zio; | |
1696 | dnode_sync(DMU_GROUPUSED_DNODE(os), tx); | |
9babb374 BB |
1697 | } |
1698 | ||
9c5167d1 NF |
1699 | if (DMU_PROJECTUSED_DNODE(os) && |
1700 | DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1701 | DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; | |
1702 | dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); | |
1703 | } | |
1704 | ||
34dc7c2f BB |
1705 | txgoff = tx->tx_txg & TXG_MASK; |
1706 | ||
b5256303 TC |
1707 | if (dmu_objset_userused_enabled(os) && |
1708 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
9babb374 BB |
1709 | /* |
1710 | * We must create the list here because it uses the | |
64fc7762 MA |
1711 | * dn_dirty_link[] of this txg. But it may already |
1712 | * exist because we call dsl_dataset_sync() twice per txg. | |
9babb374 | 1713 | */ |
64fc7762 MA |
1714 | if (os->os_synced_dnodes == NULL) { |
1715 | os->os_synced_dnodes = | |
1716 | multilist_create(sizeof (dnode_t), | |
1717 | offsetof(dnode_t, dn_dirty_link[txgoff]), | |
1718 | dnode_multilist_index_func); | |
1719 | } else { | |
1720 | ASSERT3U(os->os_synced_dnodes->ml_offset, ==, | |
1721 | offsetof(dnode_t, dn_dirty_link[txgoff])); | |
1722 | } | |
9babb374 BB |
1723 | } |
1724 | ||
64fc7762 MA |
1725 | for (int i = 0; |
1726 | i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { | |
1727 | sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); | |
1728 | sda->sda_list = os->os_dirty_dnodes[txgoff]; | |
1729 | sda->sda_sublist_idx = i; | |
1730 | sda->sda_tx = tx; | |
1731 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1732 | sync_dnodes_task, sda, 0); | |
1733 | /* callback frees sda */ | |
1734 | } | |
1735 | taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); | |
34dc7c2f | 1736 | |
572e2857 | 1737 | list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; |
64fc7762 | 1738 | while ((dr = list_head(list)) != NULL) { |
13fe0198 | 1739 | ASSERT0(dr->dr_dbuf->db_level); |
34dc7c2f BB |
1740 | list_remove(list, dr); |
1741 | if (dr->dr_zio) | |
1742 | zio_nowait(dr->dr_zio); | |
1743 | } | |
68cbd56e NB |
1744 | |
1745 | /* Enable dnode backfill if enough objects have been freed. */ | |
1746 | if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { | |
1747 | os->os_rescan_dnodes = B_TRUE; | |
1748 | os->os_freed_dnodes = 0; | |
1749 | } | |
1750 | ||
34dc7c2f BB |
1751 | /* |
1752 | * Free intent log blocks up to this tx. | |
1753 | */ | |
1754 | zil_sync(os->os_zil, tx); | |
b128c09f | 1755 | os->os_phys->os_zil_header = os->os_zil_header; |
34dc7c2f BB |
1756 | zio_nowait(zio); |
1757 | } | |
1758 | ||
428870ff BB |
1759 | boolean_t |
1760 | dmu_objset_is_dirty(objset_t *os, uint64_t txg) | |
1761 | { | |
64fc7762 | 1762 | return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); |
428870ff BB |
1763 | } |
1764 | ||
572e2857 | 1765 | static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; |
9babb374 BB |
1766 | |
1767 | void | |
1768 | dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) | |
1769 | { | |
1770 | used_cbs[ost] = cb; | |
1771 | } | |
1772 | ||
1773 | boolean_t | |
428870ff | 1774 | dmu_objset_userused_enabled(objset_t *os) |
9babb374 BB |
1775 | { |
1776 | return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && | |
572e2857 BB |
1777 | used_cbs[os->os_phys->os_type] != NULL && |
1778 | DMU_USERUSED_DNODE(os) != NULL); | |
9babb374 BB |
1779 | } |
1780 | ||
1de321e6 JX |
1781 | boolean_t |
1782 | dmu_objset_userobjused_enabled(objset_t *os) | |
1783 | { | |
1784 | return (dmu_objset_userused_enabled(os) && | |
1785 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); | |
1786 | } | |
1787 | ||
9c5167d1 NF |
1788 | boolean_t |
1789 | dmu_objset_projectquota_enabled(objset_t *os) | |
1790 | { | |
1791 | return (used_cbs[os->os_phys->os_type] != NULL && | |
1792 | DMU_PROJECTUSED_DNODE(os) != NULL && | |
1793 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); | |
1794 | } | |
1795 | ||
9b7a83cb JX |
1796 | typedef struct userquota_node { |
1797 | /* must be in the first filed, see userquota_update_cache() */ | |
1798 | char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1799 | int64_t uqn_delta; | |
1800 | avl_node_t uqn_node; | |
1801 | } userquota_node_t; | |
1802 | ||
1803 | typedef struct userquota_cache { | |
1804 | avl_tree_t uqc_user_deltas; | |
1805 | avl_tree_t uqc_group_deltas; | |
9c5167d1 | 1806 | avl_tree_t uqc_project_deltas; |
9b7a83cb JX |
1807 | } userquota_cache_t; |
1808 | ||
1809 | static int | |
1810 | userquota_compare(const void *l, const void *r) | |
1811 | { | |
1812 | const userquota_node_t *luqn = l; | |
1813 | const userquota_node_t *ruqn = r; | |
e4ffa98d | 1814 | int rv; |
9b7a83cb JX |
1815 | |
1816 | /* | |
1817 | * NB: can only access uqn_id because userquota_update_cache() doesn't | |
1818 | * pass in an entire userquota_node_t. | |
1819 | */ | |
e4ffa98d BB |
1820 | rv = strcmp(luqn->uqn_id, ruqn->uqn_id); |
1821 | ||
1822 | return (AVL_ISIGN(rv)); | |
9b7a83cb JX |
1823 | } |
1824 | ||
1825 | static void | |
1826 | do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) | |
1827 | { | |
1828 | void *cookie; | |
1829 | userquota_node_t *uqn; | |
1830 | ||
1831 | ASSERT(dmu_tx_is_syncing(tx)); | |
1832 | ||
1833 | cookie = NULL; | |
1834 | while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, | |
1835 | &cookie)) != NULL) { | |
64fc7762 MA |
1836 | /* |
1837 | * os_userused_lock protects against concurrent calls to | |
1838 | * zap_increment_int(). It's needed because zap_increment_int() | |
1839 | * is not thread-safe (i.e. not atomic). | |
1840 | */ | |
1841 | mutex_enter(&os->os_userused_lock); | |
9b7a83cb JX |
1842 | VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, |
1843 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1844 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1845 | kmem_free(uqn, sizeof (*uqn)); |
1846 | } | |
1847 | avl_destroy(&cache->uqc_user_deltas); | |
1848 | ||
1849 | cookie = NULL; | |
1850 | while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, | |
1851 | &cookie)) != NULL) { | |
64fc7762 | 1852 | mutex_enter(&os->os_userused_lock); |
9b7a83cb JX |
1853 | VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, |
1854 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1855 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1856 | kmem_free(uqn, sizeof (*uqn)); |
1857 | } | |
1858 | avl_destroy(&cache->uqc_group_deltas); | |
9c5167d1 NF |
1859 | |
1860 | if (dmu_objset_projectquota_enabled(os)) { | |
1861 | cookie = NULL; | |
1862 | while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, | |
1863 | &cookie)) != NULL) { | |
1864 | mutex_enter(&os->os_userused_lock); | |
1865 | VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, | |
1866 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1867 | mutex_exit(&os->os_userused_lock); | |
1868 | kmem_free(uqn, sizeof (*uqn)); | |
1869 | } | |
1870 | avl_destroy(&cache->uqc_project_deltas); | |
1871 | } | |
9b7a83cb JX |
1872 | } |
1873 | ||
1874 | static void | |
1875 | userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) | |
1876 | { | |
1877 | userquota_node_t *uqn; | |
1878 | avl_index_t idx; | |
1879 | ||
1880 | ASSERT(strlen(id) < sizeof (uqn->uqn_id)); | |
1881 | /* | |
1882 | * Use id directly for searching because uqn_id is the first field of | |
1883 | * userquota_node_t and fields after uqn_id won't be accessed in | |
1884 | * avl_find(). | |
1885 | */ | |
1886 | uqn = avl_find(avl, (const void *)id, &idx); | |
1887 | if (uqn == NULL) { | |
1888 | uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); | |
1f51b525 | 1889 | strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); |
9b7a83cb JX |
1890 | avl_insert(avl, uqn, idx); |
1891 | } | |
1892 | uqn->uqn_delta += delta; | |
1893 | } | |
1894 | ||
428870ff | 1895 | static void |
9c5167d1 NF |
1896 | do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, |
1897 | uint64_t flags, uint64_t user, uint64_t group, uint64_t project, | |
1898 | boolean_t subtract) | |
428870ff | 1899 | { |
9c5167d1 | 1900 | if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { |
50c957f7 | 1901 | int64_t delta = DNODE_MIN_SIZE + used; |
9b7a83cb JX |
1902 | char name[20]; |
1903 | ||
428870ff BB |
1904 | if (subtract) |
1905 | delta = -delta; | |
9b7a83cb JX |
1906 | |
1907 | (void) sprintf(name, "%llx", (longlong_t)user); | |
1908 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); | |
1909 | ||
1910 | (void) sprintf(name, "%llx", (longlong_t)group); | |
1911 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); | |
9c5167d1 NF |
1912 | |
1913 | if (dmu_objset_projectquota_enabled(os)) { | |
1914 | (void) sprintf(name, "%llx", (longlong_t)project); | |
1915 | userquota_update_cache(&cache->uqc_project_deltas, | |
1916 | name, delta); | |
1917 | } | |
428870ff BB |
1918 | } |
1919 | } | |
1920 | ||
1de321e6 | 1921 | static void |
9c5167d1 NF |
1922 | do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, |
1923 | uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) | |
1de321e6 JX |
1924 | { |
1925 | if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { | |
1926 | char name[20 + DMU_OBJACCT_PREFIX_LEN]; | |
9b7a83cb | 1927 | int delta = subtract ? -1 : 1; |
1de321e6 JX |
1928 | |
1929 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1930 | (longlong_t)user); | |
9b7a83cb | 1931 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); |
1de321e6 JX |
1932 | |
1933 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1934 | (longlong_t)group); | |
9b7a83cb | 1935 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); |
9c5167d1 NF |
1936 | |
1937 | if (dmu_objset_projectquota_enabled(os)) { | |
1938 | (void) snprintf(name, sizeof (name), | |
1939 | DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); | |
1940 | userquota_update_cache(&cache->uqc_project_deltas, | |
1941 | name, delta); | |
1942 | } | |
1de321e6 JX |
1943 | } |
1944 | } | |
1945 | ||
64fc7762 MA |
1946 | typedef struct userquota_updates_arg { |
1947 | objset_t *uua_os; | |
1948 | int uua_sublist_idx; | |
1949 | dmu_tx_t *uua_tx; | |
1950 | } userquota_updates_arg_t; | |
1951 | ||
1952 | static void | |
1953 | userquota_updates_task(void *arg) | |
9babb374 | 1954 | { |
64fc7762 MA |
1955 | userquota_updates_arg_t *uua = arg; |
1956 | objset_t *os = uua->uua_os; | |
1957 | dmu_tx_t *tx = uua->uua_tx; | |
9babb374 | 1958 | dnode_t *dn; |
9b7a83cb | 1959 | userquota_cache_t cache = { { 0 } }; |
9babb374 | 1960 | |
64fc7762 MA |
1961 | multilist_sublist_t *list = |
1962 | multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); | |
9babb374 | 1963 | |
64fc7762 MA |
1964 | ASSERT(multilist_sublist_head(list) == NULL || |
1965 | dmu_objset_userused_enabled(os)); | |
9b7a83cb JX |
1966 | avl_create(&cache.uqc_user_deltas, userquota_compare, |
1967 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
1968 | avl_create(&cache.uqc_group_deltas, userquota_compare, | |
1969 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
9c5167d1 NF |
1970 | if (dmu_objset_projectquota_enabled(os)) |
1971 | avl_create(&cache.uqc_project_deltas, userquota_compare, | |
1972 | sizeof (userquota_node_t), offsetof(userquota_node_t, | |
1973 | uqn_node)); | |
9b7a83cb | 1974 | |
64fc7762 | 1975 | while ((dn = multilist_sublist_head(list)) != NULL) { |
572e2857 | 1976 | int flags; |
9babb374 | 1977 | ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); |
9babb374 BB |
1978 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || |
1979 | dn->dn_phys->dn_flags & | |
1980 | DNODE_FLAG_USERUSED_ACCOUNTED); | |
1981 | ||
572e2857 BB |
1982 | flags = dn->dn_id_flags; |
1983 | ASSERT(flags); | |
1984 | if (flags & DN_ID_OLD_EXIST) { | |
9c5167d1 NF |
1985 | do_userquota_update(os, &cache, dn->dn_oldused, |
1986 | dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, | |
1987 | dn->dn_oldprojid, B_TRUE); | |
1988 | do_userobjquota_update(os, &cache, dn->dn_oldflags, | |
1989 | dn->dn_olduid, dn->dn_oldgid, | |
1990 | dn->dn_oldprojid, B_TRUE); | |
428870ff | 1991 | } |
572e2857 | 1992 | if (flags & DN_ID_NEW_EXIST) { |
9c5167d1 | 1993 | do_userquota_update(os, &cache, |
9b7a83cb | 1994 | DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, |
9c5167d1 NF |
1995 | dn->dn_newuid, dn->dn_newgid, |
1996 | dn->dn_newprojid, B_FALSE); | |
1997 | do_userobjquota_update(os, &cache, | |
1998 | dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, | |
1999 | dn->dn_newprojid, B_FALSE); | |
428870ff BB |
2000 | } |
2001 | ||
572e2857 | 2002 | mutex_enter(&dn->dn_mtx); |
428870ff BB |
2003 | dn->dn_oldused = 0; |
2004 | dn->dn_oldflags = 0; | |
2005 | if (dn->dn_id_flags & DN_ID_NEW_EXIST) { | |
2006 | dn->dn_olduid = dn->dn_newuid; | |
2007 | dn->dn_oldgid = dn->dn_newgid; | |
9c5167d1 | 2008 | dn->dn_oldprojid = dn->dn_newprojid; |
428870ff BB |
2009 | dn->dn_id_flags |= DN_ID_OLD_EXIST; |
2010 | if (dn->dn_bonuslen == 0) | |
2011 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
2012 | else | |
2013 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2014 | } | |
2015 | dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); | |
edc1e713 TC |
2016 | if (dn->dn_dirty_txg == spa_syncing_txg(os->os_spa)) |
2017 | dn->dn_dirty_txg = 0; | |
9babb374 BB |
2018 | mutex_exit(&dn->dn_mtx); |
2019 | ||
64fc7762 MA |
2020 | multilist_sublist_remove(list, dn); |
2021 | dnode_rele(dn, os->os_synced_dnodes); | |
9babb374 | 2022 | } |
9b7a83cb | 2023 | do_userquota_cacheflush(os, &cache, tx); |
64fc7762 MA |
2024 | multilist_sublist_unlock(list); |
2025 | kmem_free(uua, sizeof (*uua)); | |
2026 | } | |
2027 | ||
2028 | void | |
2029 | dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) | |
2030 | { | |
2031 | if (!dmu_objset_userused_enabled(os)) | |
2032 | return; | |
2033 | ||
163a8c28 TC |
2034 | /* |
2035 | * If this is a raw receive just return and handle accounting | |
2036 | * later when we have the keys loaded. We also don't do user | |
2037 | * accounting during claiming since the datasets are not owned | |
2038 | * for the duration of claiming and this txg should only be | |
2039 | * used for recovery. | |
2040 | */ | |
b5256303 TC |
2041 | if (os->os_encrypted && dmu_objset_is_receiving(os)) |
2042 | return; | |
2043 | ||
163a8c28 TC |
2044 | if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) |
2045 | return; | |
2046 | ||
9c5167d1 | 2047 | /* Allocate the user/group/project used objects if necessary. */ |
64fc7762 MA |
2048 | if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { |
2049 | VERIFY0(zap_create_claim(os, | |
2050 | DMU_USERUSED_OBJECT, | |
2051 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2052 | VERIFY0(zap_create_claim(os, | |
2053 | DMU_GROUPUSED_OBJECT, | |
2054 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2055 | } | |
2056 | ||
9c5167d1 NF |
2057 | if (dmu_objset_projectquota_enabled(os) && |
2058 | DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { | |
2059 | VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, | |
2060 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2061 | } | |
2062 | ||
64fc7762 MA |
2063 | for (int i = 0; |
2064 | i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { | |
2065 | userquota_updates_arg_t *uua = | |
2066 | kmem_alloc(sizeof (*uua), KM_SLEEP); | |
2067 | uua->uua_os = os; | |
2068 | uua->uua_sublist_idx = i; | |
2069 | uua->uua_tx = tx; | |
2070 | /* note: caller does taskq_wait() */ | |
2071 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
2072 | userquota_updates_task, uua, 0); | |
2073 | /* callback frees uua */ | |
2074 | } | |
9babb374 BB |
2075 | } |
2076 | ||
428870ff BB |
2077 | /* |
2078 | * Returns a pointer to data to find uid/gid from | |
2079 | * | |
2080 | * If a dirty record for transaction group that is syncing can't | |
2081 | * be found then NULL is returned. In the NULL case it is assumed | |
2082 | * the uid/gid aren't changing. | |
2083 | */ | |
2084 | static void * | |
2085 | dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) | |
2086 | { | |
2087 | dbuf_dirty_record_t *dr, **drp; | |
2088 | void *data; | |
2089 | ||
2090 | if (db->db_dirtycnt == 0) | |
2091 | return (db->db.db_data); /* Nothing is changing */ | |
2092 | ||
2093 | for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) | |
2094 | if (dr->dr_txg == tx->tx_txg) | |
2095 | break; | |
2096 | ||
572e2857 | 2097 | if (dr == NULL) { |
428870ff | 2098 | data = NULL; |
572e2857 BB |
2099 | } else { |
2100 | dnode_t *dn; | |
2101 | ||
2102 | DB_DNODE_ENTER(dr->dr_dbuf); | |
2103 | dn = DB_DNODE(dr->dr_dbuf); | |
2104 | ||
2105 | if (dn->dn_bonuslen == 0 && | |
2106 | dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) | |
2107 | data = dr->dt.dl.dr_data->b_data; | |
2108 | else | |
2109 | data = dr->dt.dl.dr_data; | |
2110 | ||
2111 | DB_DNODE_EXIT(dr->dr_dbuf); | |
2112 | } | |
2113 | ||
428870ff BB |
2114 | return (data); |
2115 | } | |
2116 | ||
2117 | void | |
2118 | dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) | |
2119 | { | |
2120 | objset_t *os = dn->dn_objset; | |
2121 | void *data = NULL; | |
2122 | dmu_buf_impl_t *db = NULL; | |
a117a6d6 GW |
2123 | uint64_t *user = NULL; |
2124 | uint64_t *group = NULL; | |
9c5167d1 | 2125 | uint64_t *project = NULL; |
428870ff BB |
2126 | int flags = dn->dn_id_flags; |
2127 | int error; | |
2128 | boolean_t have_spill = B_FALSE; | |
2129 | ||
2130 | if (!dmu_objset_userused_enabled(dn->dn_objset)) | |
2131 | return; | |
2132 | ||
b5256303 TC |
2133 | /* |
2134 | * Raw receives introduce a problem with user accounting. Raw | |
2135 | * receives cannot update the user accounting info because the | |
2136 | * user ids and the sizes are encrypted. To guarantee that we | |
2137 | * never end up with bad user accounting, we simply disable it | |
2138 | * during raw receives. We also disable this for normal receives | |
2139 | * so that an incremental raw receive may be done on top of an | |
2140 | * existing non-raw receive. | |
2141 | */ | |
2142 | if (os->os_encrypted && dmu_objset_is_receiving(os)) | |
2143 | return; | |
2144 | ||
428870ff BB |
2145 | if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| |
2146 | DN_ID_CHKED_SPILL))) | |
2147 | return; | |
2148 | ||
2149 | if (before && dn->dn_bonuslen != 0) | |
2150 | data = DN_BONUS(dn->dn_phys); | |
2151 | else if (!before && dn->dn_bonuslen != 0) { | |
a3000f93 BB |
2152 | if (dn->dn_bonus) { |
2153 | db = dn->dn_bonus; | |
428870ff BB |
2154 | mutex_enter(&db->db_mtx); |
2155 | data = dmu_objset_userquota_find_data(db, tx); | |
2156 | } else { | |
2157 | data = DN_BONUS(dn->dn_phys); | |
2158 | } | |
2159 | } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { | |
2160 | int rf = 0; | |
2161 | ||
2162 | if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) | |
2163 | rf |= DB_RF_HAVESTRUCT; | |
572e2857 BB |
2164 | error = dmu_spill_hold_by_dnode(dn, |
2165 | rf | DB_RF_MUST_SUCCEED, | |
428870ff BB |
2166 | FTAG, (dmu_buf_t **)&db); |
2167 | ASSERT(error == 0); | |
2168 | mutex_enter(&db->db_mtx); | |
2169 | data = (before) ? db->db.db_data : | |
2170 | dmu_objset_userquota_find_data(db, tx); | |
2171 | have_spill = B_TRUE; | |
2172 | } else { | |
2173 | mutex_enter(&dn->dn_mtx); | |
2174 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2175 | mutex_exit(&dn->dn_mtx); | |
2176 | return; | |
2177 | } | |
2178 | ||
2179 | if (before) { | |
2180 | ASSERT(data); | |
2181 | user = &dn->dn_olduid; | |
2182 | group = &dn->dn_oldgid; | |
9c5167d1 | 2183 | project = &dn->dn_oldprojid; |
428870ff BB |
2184 | } else if (data) { |
2185 | user = &dn->dn_newuid; | |
2186 | group = &dn->dn_newgid; | |
9c5167d1 | 2187 | project = &dn->dn_newprojid; |
428870ff BB |
2188 | } |
2189 | ||
2190 | /* | |
2191 | * Must always call the callback in case the object | |
2192 | * type has changed and that type isn't an object type to track | |
2193 | */ | |
2194 | error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, | |
9c5167d1 | 2195 | user, group, project); |
428870ff BB |
2196 | |
2197 | /* | |
2198 | * Preserve existing uid/gid when the callback can't determine | |
2199 | * what the new uid/gid are and the callback returned EEXIST. | |
2200 | * The EEXIST error tells us to just use the existing uid/gid. | |
2201 | * If we don't know what the old values are then just assign | |
2202 | * them to 0, since that is a new file being created. | |
2203 | */ | |
2204 | if (!before && data == NULL && error == EEXIST) { | |
2205 | if (flags & DN_ID_OLD_EXIST) { | |
2206 | dn->dn_newuid = dn->dn_olduid; | |
2207 | dn->dn_newgid = dn->dn_oldgid; | |
2705ebf0 | 2208 | dn->dn_newprojid = dn->dn_oldprojid; |
428870ff BB |
2209 | } else { |
2210 | dn->dn_newuid = 0; | |
2211 | dn->dn_newgid = 0; | |
9c5167d1 | 2212 | dn->dn_newprojid = ZFS_DEFAULT_PROJID; |
428870ff BB |
2213 | } |
2214 | error = 0; | |
2215 | } | |
2216 | ||
2217 | if (db) | |
2218 | mutex_exit(&db->db_mtx); | |
2219 | ||
2220 | mutex_enter(&dn->dn_mtx); | |
2221 | if (error == 0 && before) | |
2222 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
2223 | if (error == 0 && !before) | |
2224 | dn->dn_id_flags |= DN_ID_NEW_EXIST; | |
2225 | ||
2226 | if (have_spill) { | |
2227 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
2228 | } else { | |
2229 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2230 | } | |
2231 | mutex_exit(&dn->dn_mtx); | |
a3000f93 | 2232 | if (have_spill) |
428870ff BB |
2233 | dmu_buf_rele((dmu_buf_t *)db, FTAG); |
2234 | } | |
2235 | ||
9babb374 BB |
2236 | boolean_t |
2237 | dmu_objset_userspace_present(objset_t *os) | |
2238 | { | |
428870ff | 2239 | return (os->os_phys->os_flags & |
9babb374 BB |
2240 | OBJSET_FLAG_USERACCOUNTING_COMPLETE); |
2241 | } | |
2242 | ||
1de321e6 JX |
2243 | boolean_t |
2244 | dmu_objset_userobjspace_present(objset_t *os) | |
2245 | { | |
2246 | return (os->os_phys->os_flags & | |
2247 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); | |
2248 | } | |
2249 | ||
9c5167d1 NF |
2250 | boolean_t |
2251 | dmu_objset_projectquota_present(objset_t *os) | |
2252 | { | |
2253 | return (os->os_phys->os_flags & | |
2254 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE); | |
2255 | } | |
2256 | ||
1de321e6 JX |
2257 | static int |
2258 | dmu_objset_space_upgrade(objset_t *os) | |
9babb374 BB |
2259 | { |
2260 | uint64_t obj; | |
2261 | int err = 0; | |
2262 | ||
9babb374 BB |
2263 | /* |
2264 | * We simply need to mark every object dirty, so that it will be | |
2265 | * synced out and now accounted. If this is called | |
2266 | * concurrently, or if we already did some work before crashing, | |
2267 | * that's fine, since we track each object's accounted state | |
2268 | * independently. | |
2269 | */ | |
2270 | ||
2271 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { | |
45d1cae3 | 2272 | dmu_tx_t *tx; |
9babb374 BB |
2273 | dmu_buf_t *db; |
2274 | int objerr; | |
2275 | ||
1de321e6 JX |
2276 | mutex_enter(&os->os_upgrade_lock); |
2277 | if (os->os_upgrade_exit) | |
2278 | err = SET_ERROR(EINTR); | |
2279 | mutex_exit(&os->os_upgrade_lock); | |
2280 | if (err != 0) | |
2281 | return (err); | |
2282 | ||
9babb374 | 2283 | if (issig(JUSTLOOKING) && issig(FORREAL)) |
2e528b49 | 2284 | return (SET_ERROR(EINTR)); |
9babb374 BB |
2285 | |
2286 | objerr = dmu_bonus_hold(os, obj, FTAG, &db); | |
13fe0198 | 2287 | if (objerr != 0) |
9babb374 | 2288 | continue; |
45d1cae3 | 2289 | tx = dmu_tx_create(os); |
9babb374 BB |
2290 | dmu_tx_hold_bonus(tx, obj); |
2291 | objerr = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 2292 | if (objerr != 0) { |
d22323e8 | 2293 | dmu_buf_rele(db, FTAG); |
9babb374 BB |
2294 | dmu_tx_abort(tx); |
2295 | continue; | |
2296 | } | |
2297 | dmu_buf_will_dirty(db, tx); | |
2298 | dmu_buf_rele(db, FTAG); | |
2299 | dmu_tx_commit(tx); | |
2300 | } | |
1de321e6 JX |
2301 | return (0); |
2302 | } | |
2303 | ||
2304 | int | |
2305 | dmu_objset_userspace_upgrade(objset_t *os) | |
2306 | { | |
2307 | int err = 0; | |
2308 | ||
2309 | if (dmu_objset_userspace_present(os)) | |
2310 | return (0); | |
2311 | if (dmu_objset_is_snapshot(os)) | |
2312 | return (SET_ERROR(EINVAL)); | |
2313 | if (!dmu_objset_userused_enabled(os)) | |
2314 | return (SET_ERROR(ENOTSUP)); | |
2315 | ||
2316 | err = dmu_objset_space_upgrade(os); | |
2317 | if (err) | |
2318 | return (err); | |
9babb374 | 2319 | |
428870ff | 2320 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
9babb374 BB |
2321 | txg_wait_synced(dmu_objset_pool(os), 0); |
2322 | return (0); | |
2323 | } | |
2324 | ||
1de321e6 | 2325 | static int |
9c5167d1 | 2326 | dmu_objset_id_quota_upgrade_cb(objset_t *os) |
1de321e6 JX |
2327 | { |
2328 | int err = 0; | |
2329 | ||
9c5167d1 NF |
2330 | if (dmu_objset_userobjspace_present(os) && |
2331 | dmu_objset_projectquota_present(os)) | |
1de321e6 JX |
2332 | return (0); |
2333 | if (dmu_objset_is_snapshot(os)) | |
2334 | return (SET_ERROR(EINVAL)); | |
2335 | if (!dmu_objset_userobjused_enabled(os)) | |
2336 | return (SET_ERROR(ENOTSUP)); | |
9c5167d1 NF |
2337 | if (!dmu_objset_projectquota_enabled(os) && |
2338 | dmu_objset_userobjspace_present(os)) | |
2339 | return (SET_ERROR(ENOTSUP)); | |
1de321e6 JX |
2340 | |
2341 | dmu_objset_ds(os)->ds_feature_activation_needed[ | |
2342 | SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE; | |
9c5167d1 NF |
2343 | if (dmu_objset_projectquota_enabled(os)) |
2344 | dmu_objset_ds(os)->ds_feature_activation_needed[ | |
2345 | SPA_FEATURE_PROJECT_QUOTA] = B_TRUE; | |
1de321e6 JX |
2346 | |
2347 | err = dmu_objset_space_upgrade(os); | |
2348 | if (err) | |
2349 | return (err); | |
2350 | ||
2351 | os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
9c5167d1 NF |
2352 | if (dmu_objset_projectquota_enabled(os)) |
2353 | os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
2354 | ||
1de321e6 JX |
2355 | txg_wait_synced(dmu_objset_pool(os), 0); |
2356 | return (0); | |
2357 | } | |
2358 | ||
2359 | void | |
9c5167d1 | 2360 | dmu_objset_id_quota_upgrade(objset_t *os) |
1de321e6 | 2361 | { |
9c5167d1 | 2362 | dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); |
1de321e6 JX |
2363 | } |
2364 | ||
126ae9f4 JX |
2365 | boolean_t |
2366 | dmu_objset_userobjspace_upgradable(objset_t *os) | |
2367 | { | |
2368 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2369 | !dmu_objset_is_snapshot(os) && | |
2370 | dmu_objset_userobjused_enabled(os) && | |
2371 | !dmu_objset_userobjspace_present(os)); | |
2372 | } | |
2373 | ||
9c5167d1 NF |
2374 | boolean_t |
2375 | dmu_objset_projectquota_upgradable(objset_t *os) | |
2376 | { | |
2377 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2378 | !dmu_objset_is_snapshot(os) && | |
2379 | dmu_objset_projectquota_enabled(os) && | |
2380 | !dmu_objset_projectquota_present(os)); | |
2381 | } | |
2382 | ||
34dc7c2f BB |
2383 | void |
2384 | dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, | |
2385 | uint64_t *usedobjsp, uint64_t *availobjsp) | |
2386 | { | |
428870ff | 2387 | dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, |
34dc7c2f BB |
2388 | usedobjsp, availobjsp); |
2389 | } | |
2390 | ||
2391 | uint64_t | |
2392 | dmu_objset_fsid_guid(objset_t *os) | |
2393 | { | |
428870ff | 2394 | return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); |
34dc7c2f BB |
2395 | } |
2396 | ||
2397 | void | |
2398 | dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) | |
2399 | { | |
428870ff BB |
2400 | stat->dds_type = os->os_phys->os_type; |
2401 | if (os->os_dsl_dataset) | |
2402 | dsl_dataset_fast_stat(os->os_dsl_dataset, stat); | |
34dc7c2f BB |
2403 | } |
2404 | ||
2405 | void | |
2406 | dmu_objset_stats(objset_t *os, nvlist_t *nv) | |
2407 | { | |
428870ff BB |
2408 | ASSERT(os->os_dsl_dataset || |
2409 | os->os_phys->os_type == DMU_OST_META); | |
34dc7c2f | 2410 | |
428870ff BB |
2411 | if (os->os_dsl_dataset != NULL) |
2412 | dsl_dataset_stats(os->os_dsl_dataset, nv); | |
34dc7c2f BB |
2413 | |
2414 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, | |
428870ff | 2415 | os->os_phys->os_type); |
9babb374 BB |
2416 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, |
2417 | dmu_objset_userspace_present(os)); | |
34dc7c2f BB |
2418 | } |
2419 | ||
2420 | int | |
2421 | dmu_objset_is_snapshot(objset_t *os) | |
2422 | { | |
428870ff | 2423 | if (os->os_dsl_dataset != NULL) |
0c66c32d | 2424 | return (os->os_dsl_dataset->ds_is_snapshot); |
34dc7c2f BB |
2425 | else |
2426 | return (B_FALSE); | |
2427 | } | |
2428 | ||
2429 | int | |
2430 | dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, | |
2431 | boolean_t *conflict) | |
2432 | { | |
428870ff | 2433 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2434 | uint64_t ignored; |
2435 | ||
d683ddbb | 2436 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2437 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2438 | |
2439 | return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2440 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, |
9b7b9cd3 | 2441 | MT_NORMALIZE, real, maxlen, conflict)); |
34dc7c2f BB |
2442 | } |
2443 | ||
2444 | int | |
2445 | dmu_snapshot_list_next(objset_t *os, int namelen, char *name, | |
2446 | uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) | |
2447 | { | |
428870ff | 2448 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2449 | zap_cursor_t cursor; |
2450 | zap_attribute_t attr; | |
2451 | ||
13fe0198 MA |
2452 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
2453 | ||
d683ddbb | 2454 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2455 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2456 | |
2457 | zap_cursor_init_serialized(&cursor, | |
2458 | ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2459 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); |
34dc7c2f BB |
2460 | |
2461 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2462 | zap_cursor_fini(&cursor); | |
2e528b49 | 2463 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2464 | } |
2465 | ||
2466 | if (strlen(attr.za_name) + 1 > namelen) { | |
2467 | zap_cursor_fini(&cursor); | |
2e528b49 | 2468 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2469 | } |
2470 | ||
2471 | (void) strcpy(name, attr.za_name); | |
2472 | if (idp) | |
2473 | *idp = attr.za_first_integer; | |
2474 | if (case_conflict) | |
2475 | *case_conflict = attr.za_normalization_conflict; | |
2476 | zap_cursor_advance(&cursor); | |
2477 | *offp = zap_cursor_serialize(&cursor); | |
2478 | zap_cursor_fini(&cursor); | |
2479 | ||
2480 | return (0); | |
2481 | } | |
2482 | ||
ebe7e575 | 2483 | int |
6772fb67 | 2484 | dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) |
ebe7e575 | 2485 | { |
d1d7e268 | 2486 | return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); |
ebe7e575 BB |
2487 | } |
2488 | ||
34dc7c2f BB |
2489 | int |
2490 | dmu_dir_list_next(objset_t *os, int namelen, char *name, | |
2491 | uint64_t *idp, uint64_t *offp) | |
2492 | { | |
428870ff | 2493 | dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; |
34dc7c2f BB |
2494 | zap_cursor_t cursor; |
2495 | zap_attribute_t attr; | |
2496 | ||
2497 | /* there is no next dir on a snapshot! */ | |
428870ff | 2498 | if (os->os_dsl_dataset->ds_object != |
d683ddbb | 2499 | dsl_dir_phys(dd)->dd_head_dataset_obj) |
2e528b49 | 2500 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2501 | |
2502 | zap_cursor_init_serialized(&cursor, | |
2503 | dd->dd_pool->dp_meta_objset, | |
d683ddbb | 2504 | dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); |
34dc7c2f BB |
2505 | |
2506 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2507 | zap_cursor_fini(&cursor); | |
2e528b49 | 2508 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2509 | } |
2510 | ||
2511 | if (strlen(attr.za_name) + 1 > namelen) { | |
2512 | zap_cursor_fini(&cursor); | |
2e528b49 | 2513 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2514 | } |
2515 | ||
2516 | (void) strcpy(name, attr.za_name); | |
2517 | if (idp) | |
2518 | *idp = attr.za_first_integer; | |
2519 | zap_cursor_advance(&cursor); | |
2520 | *offp = zap_cursor_serialize(&cursor); | |
2521 | zap_cursor_fini(&cursor); | |
2522 | ||
2523 | return (0); | |
2524 | } | |
2525 | ||
9c43027b AJ |
2526 | typedef struct dmu_objset_find_ctx { |
2527 | taskq_t *dc_tq; | |
2528 | dsl_pool_t *dc_dp; | |
2529 | uint64_t dc_ddobj; | |
1149ba64 | 2530 | char *dc_ddname; /* last component of ddobj's name */ |
9c43027b AJ |
2531 | int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); |
2532 | void *dc_arg; | |
2533 | int dc_flags; | |
2534 | kmutex_t *dc_error_lock; | |
2535 | int *dc_error; | |
2536 | } dmu_objset_find_ctx_t; | |
2537 | ||
2538 | static void | |
2539 | dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) | |
b128c09f | 2540 | { |
9c43027b | 2541 | dsl_pool_t *dp = dcp->dc_dp; |
13fe0198 MA |
2542 | dsl_dir_t *dd; |
2543 | dsl_dataset_t *ds; | |
2544 | zap_cursor_t zc; | |
2545 | zap_attribute_t *attr; | |
2546 | uint64_t thisobj; | |
9c43027b | 2547 | int err = 0; |
13fe0198 | 2548 | |
9c43027b AJ |
2549 | /* don't process if there already was an error */ |
2550 | if (*dcp->dc_error != 0) | |
2551 | goto out; | |
13fe0198 | 2552 | |
1149ba64 GM |
2553 | /* |
2554 | * Note: passing the name (dc_ddname) here is optional, but it | |
2555 | * improves performance because we don't need to call | |
2556 | * zap_value_search() to determine the name. | |
2557 | */ | |
2558 | err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); | |
13fe0198 | 2559 | if (err != 0) |
9c43027b | 2560 | goto out; |
13fe0198 MA |
2561 | |
2562 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ | |
2563 | if (dd->dd_myname[0] == '$') { | |
2564 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2565 | goto out; |
13fe0198 MA |
2566 | } |
2567 | ||
d683ddbb | 2568 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2569 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
13fe0198 MA |
2570 | |
2571 | /* | |
2572 | * Iterate over all children. | |
2573 | */ | |
9c43027b | 2574 | if (dcp->dc_flags & DS_FIND_CHILDREN) { |
13fe0198 | 2575 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2576 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
13fe0198 MA |
2577 | zap_cursor_retrieve(&zc, attr) == 0; |
2578 | (void) zap_cursor_advance(&zc)) { | |
2579 | ASSERT3U(attr->za_integer_length, ==, | |
2580 | sizeof (uint64_t)); | |
2581 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2582 | ||
1c27024e | 2583 | dmu_objset_find_ctx_t *child_dcp = |
1149ba64 | 2584 | kmem_alloc(sizeof (*child_dcp), KM_SLEEP); |
9c43027b AJ |
2585 | *child_dcp = *dcp; |
2586 | child_dcp->dc_ddobj = attr->za_first_integer; | |
1149ba64 | 2587 | child_dcp->dc_ddname = spa_strdup(attr->za_name); |
9c43027b AJ |
2588 | if (dcp->dc_tq != NULL) |
2589 | (void) taskq_dispatch(dcp->dc_tq, | |
2590 | dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); | |
2591 | else | |
2592 | dmu_objset_find_dp_impl(child_dcp); | |
13fe0198 MA |
2593 | } |
2594 | zap_cursor_fini(&zc); | |
13fe0198 MA |
2595 | } |
2596 | ||
2597 | /* | |
2598 | * Iterate over all snapshots. | |
2599 | */ | |
9c43027b | 2600 | if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { |
13fe0198 MA |
2601 | dsl_dataset_t *ds; |
2602 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2603 | ||
2604 | if (err == 0) { | |
d683ddbb JG |
2605 | uint64_t snapobj; |
2606 | ||
2607 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
13fe0198 MA |
2608 | dsl_dataset_rele(ds, FTAG); |
2609 | ||
2610 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2611 | zap_cursor_retrieve(&zc, attr) == 0; | |
2612 | (void) zap_cursor_advance(&zc)) { | |
2613 | ASSERT3U(attr->za_integer_length, ==, | |
2614 | sizeof (uint64_t)); | |
2615 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2616 | ||
2617 | err = dsl_dataset_hold_obj(dp, | |
2618 | attr->za_first_integer, FTAG, &ds); | |
2619 | if (err != 0) | |
2620 | break; | |
9c43027b | 2621 | err = dcp->dc_func(dp, ds, dcp->dc_arg); |
13fe0198 MA |
2622 | dsl_dataset_rele(ds, FTAG); |
2623 | if (err != 0) | |
2624 | break; | |
2625 | } | |
2626 | zap_cursor_fini(&zc); | |
2627 | } | |
2628 | } | |
2629 | ||
13fe0198 MA |
2630 | kmem_free(attr, sizeof (zap_attribute_t)); |
2631 | ||
1149ba64 GM |
2632 | if (err != 0) { |
2633 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2634 | goto out; |
1149ba64 | 2635 | } |
13fe0198 MA |
2636 | |
2637 | /* | |
2638 | * Apply to self. | |
2639 | */ | |
2640 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
1149ba64 GM |
2641 | |
2642 | /* | |
2643 | * Note: we hold the dir while calling dsl_dataset_hold_obj() so | |
2644 | * that the dir will remain cached, and we won't have to re-instantiate | |
2645 | * it (which could be expensive due to finding its name via | |
2646 | * zap_value_search()). | |
2647 | */ | |
2648 | dsl_dir_rele(dd, FTAG); | |
13fe0198 | 2649 | if (err != 0) |
9c43027b AJ |
2650 | goto out; |
2651 | err = dcp->dc_func(dp, ds, dcp->dc_arg); | |
13fe0198 | 2652 | dsl_dataset_rele(ds, FTAG); |
9c43027b AJ |
2653 | |
2654 | out: | |
2655 | if (err != 0) { | |
2656 | mutex_enter(dcp->dc_error_lock); | |
2657 | /* only keep first error */ | |
2658 | if (*dcp->dc_error == 0) | |
2659 | *dcp->dc_error = err; | |
2660 | mutex_exit(dcp->dc_error_lock); | |
2661 | } | |
2662 | ||
1149ba64 GM |
2663 | if (dcp->dc_ddname != NULL) |
2664 | spa_strfree(dcp->dc_ddname); | |
9c43027b AJ |
2665 | kmem_free(dcp, sizeof (*dcp)); |
2666 | } | |
2667 | ||
2668 | static void | |
2669 | dmu_objset_find_dp_cb(void *arg) | |
2670 | { | |
2671 | dmu_objset_find_ctx_t *dcp = arg; | |
2672 | dsl_pool_t *dp = dcp->dc_dp; | |
2673 | ||
5e8cd5d1 AJ |
2674 | /* |
2675 | * We need to get a pool_config_lock here, as there are several | |
2676 | * asssert(pool_config_held) down the stack. Getting a lock via | |
2677 | * dsl_pool_config_enter is risky, as it might be stalled by a | |
2678 | * pending writer. This would deadlock, as the write lock can | |
2679 | * only be granted when our parent thread gives up the lock. | |
2680 | * The _prio interface gives us priority over a pending writer. | |
2681 | */ | |
2682 | dsl_pool_config_enter_prio(dp, FTAG); | |
9c43027b AJ |
2683 | |
2684 | dmu_objset_find_dp_impl(dcp); | |
2685 | ||
2686 | dsl_pool_config_exit(dp, FTAG); | |
2687 | } | |
2688 | ||
2689 | /* | |
2690 | * Find objsets under and including ddobj, call func(ds) on each. | |
2691 | * The order for the enumeration is completely undefined. | |
2692 | * func is called with dsl_pool_config held. | |
2693 | */ | |
2694 | int | |
2695 | dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, | |
2696 | int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) | |
2697 | { | |
2698 | int error = 0; | |
2699 | taskq_t *tq = NULL; | |
2700 | int ntasks; | |
2701 | dmu_objset_find_ctx_t *dcp; | |
2702 | kmutex_t err_lock; | |
2703 | ||
2704 | mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); | |
2705 | dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); | |
2706 | dcp->dc_tq = NULL; | |
2707 | dcp->dc_dp = dp; | |
2708 | dcp->dc_ddobj = ddobj; | |
1149ba64 | 2709 | dcp->dc_ddname = NULL; |
9c43027b AJ |
2710 | dcp->dc_func = func; |
2711 | dcp->dc_arg = arg; | |
2712 | dcp->dc_flags = flags; | |
2713 | dcp->dc_error_lock = &err_lock; | |
2714 | dcp->dc_error = &error; | |
2715 | ||
2716 | if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { | |
2717 | /* | |
2718 | * In case a write lock is held we can't make use of | |
2719 | * parallelism, as down the stack of the worker threads | |
2720 | * the lock is asserted via dsl_pool_config_held. | |
2721 | * In case of a read lock this is solved by getting a read | |
2722 | * lock in each worker thread, which isn't possible in case | |
2723 | * of a writer lock. So we fall back to the synchronous path | |
2724 | * here. | |
2725 | * In the future it might be possible to get some magic into | |
2726 | * dsl_pool_config_held in a way that it returns true for | |
2727 | * the worker threads so that a single lock held from this | |
2728 | * thread suffices. For now, stay single threaded. | |
2729 | */ | |
2730 | dmu_objset_find_dp_impl(dcp); | |
e5676636 | 2731 | mutex_destroy(&err_lock); |
9c43027b AJ |
2732 | |
2733 | return (error); | |
2734 | } | |
2735 | ||
2736 | ntasks = dmu_find_threads; | |
2737 | if (ntasks == 0) | |
2738 | ntasks = vdev_count_leaves(dp->dp_spa) * 4; | |
1229323d | 2739 | tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, |
9c43027b AJ |
2740 | INT_MAX, 0); |
2741 | if (tq == NULL) { | |
2742 | kmem_free(dcp, sizeof (*dcp)); | |
e5676636 BB |
2743 | mutex_destroy(&err_lock); |
2744 | ||
9c43027b AJ |
2745 | return (SET_ERROR(ENOMEM)); |
2746 | } | |
2747 | dcp->dc_tq = tq; | |
2748 | ||
2749 | /* dcp will be freed by task */ | |
2750 | (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); | |
2751 | ||
2752 | /* | |
2753 | * PORTING: this code relies on the property of taskq_wait to wait | |
2754 | * until no more tasks are queued and no more tasks are active. As | |
2755 | * we always queue new tasks from within other tasks, task_wait | |
2756 | * reliably waits for the full recursion to finish, even though we | |
2757 | * enqueue new tasks after taskq_wait has been called. | |
2758 | * On platforms other than illumos, taskq_wait may not have this | |
2759 | * property. | |
2760 | */ | |
2761 | taskq_wait(tq); | |
2762 | taskq_destroy(tq); | |
2763 | mutex_destroy(&err_lock); | |
2764 | ||
2765 | return (error); | |
b128c09f BB |
2766 | } |
2767 | ||
2768 | /* | |
13fe0198 MA |
2769 | * Find all objsets under name, and for each, call 'func(child_name, arg)'. |
2770 | * The dp_config_rwlock must not be held when this is called, and it | |
2771 | * will not be held when the callback is called. | |
2772 | * Therefore this function should only be used when the pool is not changing | |
2773 | * (e.g. in syncing context), or the callback can deal with the possible races. | |
b128c09f | 2774 | */ |
13fe0198 MA |
2775 | static int |
2776 | dmu_objset_find_impl(spa_t *spa, const char *name, | |
2777 | int func(const char *, void *), void *arg, int flags) | |
34dc7c2f BB |
2778 | { |
2779 | dsl_dir_t *dd; | |
13fe0198 | 2780 | dsl_pool_t *dp = spa_get_dsl(spa); |
b128c09f | 2781 | dsl_dataset_t *ds; |
34dc7c2f BB |
2782 | zap_cursor_t zc; |
2783 | zap_attribute_t *attr; | |
2784 | char *child; | |
b128c09f BB |
2785 | uint64_t thisobj; |
2786 | int err; | |
34dc7c2f | 2787 | |
13fe0198 MA |
2788 | dsl_pool_config_enter(dp, FTAG); |
2789 | ||
2790 | err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); | |
2791 | if (err != 0) { | |
2792 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f | 2793 | return (err); |
13fe0198 | 2794 | } |
34dc7c2f | 2795 | |
b128c09f BB |
2796 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ |
2797 | if (dd->dd_myname[0] == '$') { | |
13fe0198 MA |
2798 | dsl_dir_rele(dd, FTAG); |
2799 | dsl_pool_config_exit(dp, FTAG); | |
b128c09f BB |
2800 | return (0); |
2801 | } | |
2802 | ||
d683ddbb | 2803 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2804 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
34dc7c2f BB |
2805 | |
2806 | /* | |
2807 | * Iterate over all children. | |
2808 | */ | |
2809 | if (flags & DS_FIND_CHILDREN) { | |
b128c09f | 2810 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2811 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
2812 | zap_cursor_retrieve(&zc, attr) == 0; |
2813 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 MA |
2814 | ASSERT3U(attr->za_integer_length, ==, |
2815 | sizeof (uint64_t)); | |
2816 | ASSERT3U(attr->za_num_integers, ==, 1); | |
34dc7c2f | 2817 | |
428870ff | 2818 | child = kmem_asprintf("%s/%s", name, attr->za_name); |
13fe0198 MA |
2819 | dsl_pool_config_exit(dp, FTAG); |
2820 | err = dmu_objset_find_impl(spa, child, | |
2821 | func, arg, flags); | |
2822 | dsl_pool_config_enter(dp, FTAG); | |
428870ff | 2823 | strfree(child); |
13fe0198 | 2824 | if (err != 0) |
34dc7c2f BB |
2825 | break; |
2826 | } | |
2827 | zap_cursor_fini(&zc); | |
2828 | ||
13fe0198 MA |
2829 | if (err != 0) { |
2830 | dsl_dir_rele(dd, FTAG); | |
2831 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
2832 | kmem_free(attr, sizeof (zap_attribute_t)); |
2833 | return (err); | |
2834 | } | |
2835 | } | |
2836 | ||
2837 | /* | |
2838 | * Iterate over all snapshots. | |
2839 | */ | |
b128c09f | 2840 | if (flags & DS_FIND_SNAPSHOTS) { |
b128c09f | 2841 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); |
b128c09f BB |
2842 | |
2843 | if (err == 0) { | |
d683ddbb JG |
2844 | uint64_t snapobj; |
2845 | ||
2846 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
b128c09f BB |
2847 | dsl_dataset_rele(ds, FTAG); |
2848 | ||
2849 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2850 | zap_cursor_retrieve(&zc, attr) == 0; | |
2851 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 | 2852 | ASSERT3U(attr->za_integer_length, ==, |
b128c09f | 2853 | sizeof (uint64_t)); |
13fe0198 | 2854 | ASSERT3U(attr->za_num_integers, ==, 1); |
b128c09f | 2855 | |
428870ff BB |
2856 | child = kmem_asprintf("%s@%s", |
2857 | name, attr->za_name); | |
13fe0198 MA |
2858 | dsl_pool_config_exit(dp, FTAG); |
2859 | err = func(child, arg); | |
2860 | dsl_pool_config_enter(dp, FTAG); | |
428870ff | 2861 | strfree(child); |
13fe0198 | 2862 | if (err != 0) |
b128c09f BB |
2863 | break; |
2864 | } | |
2865 | zap_cursor_fini(&zc); | |
34dc7c2f | 2866 | } |
34dc7c2f BB |
2867 | } |
2868 | ||
13fe0198 | 2869 | dsl_dir_rele(dd, FTAG); |
34dc7c2f | 2870 | kmem_free(attr, sizeof (zap_attribute_t)); |
13fe0198 | 2871 | dsl_pool_config_exit(dp, FTAG); |
34dc7c2f | 2872 | |
13fe0198 | 2873 | if (err != 0) |
34dc7c2f BB |
2874 | return (err); |
2875 | ||
13fe0198 MA |
2876 | /* Apply to self. */ |
2877 | return (func(name, arg)); | |
34dc7c2f BB |
2878 | } |
2879 | ||
13fe0198 MA |
2880 | /* |
2881 | * See comment above dmu_objset_find_impl(). | |
2882 | */ | |
d164b209 | 2883 | int |
13fe0198 MA |
2884 | dmu_objset_find(char *name, int func(const char *, void *), void *arg, |
2885 | int flags) | |
d164b209 | 2886 | { |
13fe0198 MA |
2887 | spa_t *spa; |
2888 | int error; | |
d164b209 | 2889 | |
13fe0198 MA |
2890 | error = spa_open(name, &spa, FTAG); |
2891 | if (error != 0) | |
2892 | return (error); | |
2893 | error = dmu_objset_find_impl(spa, name, func, arg, flags); | |
2894 | spa_close(spa, FTAG); | |
2895 | return (error); | |
d164b209 BB |
2896 | } |
2897 | ||
ae76f45c TC |
2898 | boolean_t |
2899 | dmu_objset_incompatible_encryption_version(objset_t *os) | |
2900 | { | |
2901 | return (dsl_dir_incompatible_encryption_version( | |
2902 | os->os_dsl_dataset->ds_dir)); | |
2903 | } | |
2904 | ||
34dc7c2f BB |
2905 | void |
2906 | dmu_objset_set_user(objset_t *os, void *user_ptr) | |
2907 | { | |
428870ff BB |
2908 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2909 | os->os_user_ptr = user_ptr; | |
34dc7c2f BB |
2910 | } |
2911 | ||
2912 | void * | |
2913 | dmu_objset_get_user(objset_t *os) | |
2914 | { | |
428870ff BB |
2915 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2916 | return (os->os_user_ptr); | |
34dc7c2f | 2917 | } |
c28b2279 | 2918 | |
13fe0198 MA |
2919 | /* |
2920 | * Determine name of filesystem, given name of snapshot. | |
eca7b760 | 2921 | * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes |
13fe0198 MA |
2922 | */ |
2923 | int | |
2924 | dmu_fsname(const char *snapname, char *buf) | |
2925 | { | |
2926 | char *atp = strchr(snapname, '@'); | |
2927 | if (atp == NULL) | |
2e528b49 | 2928 | return (SET_ERROR(EINVAL)); |
eca7b760 | 2929 | if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 2930 | return (SET_ERROR(ENAMETOOLONG)); |
13fe0198 MA |
2931 | (void) strlcpy(buf, snapname, atp - snapname + 1); |
2932 | return (0); | |
2933 | } | |
2934 | ||
3ec3bc21 BB |
2935 | /* |
2936 | * Call when we think we're going to write/free space in open context to track | |
2937 | * the amount of dirty data in the open txg, which is also the amount | |
2938 | * of memory that can not be evicted until this txg syncs. | |
2939 | */ | |
2940 | void | |
2941 | dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) | |
2942 | { | |
2943 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
2944 | int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); | |
2945 | ||
2946 | if (ds != NULL) { | |
2947 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
2948 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
2949 | } | |
2950 | } | |
2951 | ||
c28b2279 | 2952 | #if defined(_KERNEL) && defined(HAVE_SPL) |
f0fd83be | 2953 | EXPORT_SYMBOL(dmu_objset_zil); |
c28b2279 | 2954 | EXPORT_SYMBOL(dmu_objset_pool); |
f0fd83be BB |
2955 | EXPORT_SYMBOL(dmu_objset_ds); |
2956 | EXPORT_SYMBOL(dmu_objset_type); | |
c28b2279 BB |
2957 | EXPORT_SYMBOL(dmu_objset_name); |
2958 | EXPORT_SYMBOL(dmu_objset_hold); | |
b5256303 | 2959 | EXPORT_SYMBOL(dmu_objset_hold_flags); |
c28b2279 BB |
2960 | EXPORT_SYMBOL(dmu_objset_own); |
2961 | EXPORT_SYMBOL(dmu_objset_rele); | |
b5256303 | 2962 | EXPORT_SYMBOL(dmu_objset_rele_flags); |
c28b2279 BB |
2963 | EXPORT_SYMBOL(dmu_objset_disown); |
2964 | EXPORT_SYMBOL(dmu_objset_from_ds); | |
2965 | EXPORT_SYMBOL(dmu_objset_create); | |
2966 | EXPORT_SYMBOL(dmu_objset_clone); | |
c28b2279 BB |
2967 | EXPORT_SYMBOL(dmu_objset_stats); |
2968 | EXPORT_SYMBOL(dmu_objset_fast_stat); | |
54a179e7 | 2969 | EXPORT_SYMBOL(dmu_objset_spa); |
c28b2279 BB |
2970 | EXPORT_SYMBOL(dmu_objset_space); |
2971 | EXPORT_SYMBOL(dmu_objset_fsid_guid); | |
2972 | EXPORT_SYMBOL(dmu_objset_find); | |
c28b2279 BB |
2973 | EXPORT_SYMBOL(dmu_objset_byteswap); |
2974 | EXPORT_SYMBOL(dmu_objset_evict_dbufs); | |
2975 | EXPORT_SYMBOL(dmu_objset_snap_cmtime); | |
50c957f7 | 2976 | EXPORT_SYMBOL(dmu_objset_dnodesize); |
c28b2279 BB |
2977 | |
2978 | EXPORT_SYMBOL(dmu_objset_sync); | |
2979 | EXPORT_SYMBOL(dmu_objset_is_dirty); | |
b5256303 | 2980 | EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); |
c28b2279 BB |
2981 | EXPORT_SYMBOL(dmu_objset_create_impl); |
2982 | EXPORT_SYMBOL(dmu_objset_open_impl); | |
2983 | EXPORT_SYMBOL(dmu_objset_evict); | |
2984 | EXPORT_SYMBOL(dmu_objset_register_type); | |
2985 | EXPORT_SYMBOL(dmu_objset_do_userquota_updates); | |
2986 | EXPORT_SYMBOL(dmu_objset_userquota_get_ids); | |
2987 | EXPORT_SYMBOL(dmu_objset_userused_enabled); | |
2988 | EXPORT_SYMBOL(dmu_objset_userspace_upgrade); | |
2989 | EXPORT_SYMBOL(dmu_objset_userspace_present); | |
1de321e6 | 2990 | EXPORT_SYMBOL(dmu_objset_userobjused_enabled); |
126ae9f4 | 2991 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); |
1de321e6 | 2992 | EXPORT_SYMBOL(dmu_objset_userobjspace_present); |
9c5167d1 NF |
2993 | EXPORT_SYMBOL(dmu_objset_projectquota_enabled); |
2994 | EXPORT_SYMBOL(dmu_objset_projectquota_present); | |
2995 | EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); | |
2996 | EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); | |
c28b2279 | 2997 | #endif |