]>
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. |
ba67d821 | 24 | * Copyright (c) 2012, 2020 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. |
d8d418ff | 32 | * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved. |
10b3c7f5 MN |
33 | * Copyright (c) 2019, Klara Inc. |
34 | * Copyright (c) 2019, Allan Jude | |
34dc7c2f BB |
35 | */ |
36 | ||
428870ff BB |
37 | /* Portions Copyright 2010 Robert Milkowski */ |
38 | ||
34dc7c2f BB |
39 | #include <sys/cred.h> |
40 | #include <sys/zfs_context.h> | |
41 | #include <sys/dmu_objset.h> | |
42 | #include <sys/dsl_dir.h> | |
43 | #include <sys/dsl_dataset.h> | |
44 | #include <sys/dsl_prop.h> | |
45 | #include <sys/dsl_pool.h> | |
46 | #include <sys/dsl_synctask.h> | |
47 | #include <sys/dsl_deleg.h> | |
48 | #include <sys/dnode.h> | |
49 | #include <sys/dbuf.h> | |
50 | #include <sys/zvol.h> | |
51 | #include <sys/dmu_tx.h> | |
34dc7c2f BB |
52 | #include <sys/zap.h> |
53 | #include <sys/zil.h> | |
54 | #include <sys/dmu_impl.h> | |
55 | #include <sys/zfs_ioctl.h> | |
428870ff | 56 | #include <sys/sa.h> |
572e2857 | 57 | #include <sys/zfs_onexit.h> |
13fe0198 | 58 | #include <sys/dsl_destroy.h> |
9c43027b | 59 | #include <sys/vdev.h> |
a1d477c2 | 60 | #include <sys/zfeature.h> |
f74b821a | 61 | #include <sys/policy.h> |
1de321e6 | 62 | #include <sys/spa_impl.h> |
03916905 | 63 | #include <sys/dmu_recv.h> |
9c5167d1 | 64 | #include <sys/zfs_project.h> |
a7ed98d8 | 65 | #include "zfs_namecheck.h" |
572e2857 BB |
66 | |
67 | /* | |
68 | * Needed to close a window in dnode_move() that allows the objset to be freed | |
69 | * before it can be safely accessed. | |
70 | */ | |
71 | krwlock_t os_lock; | |
72 | ||
9c43027b | 73 | /* |
4e33ba4c | 74 | * Tunable to overwrite the maximum number of threads for the parallelization |
9c43027b AJ |
75 | * of dmu_objset_find_dp, needed to speed up the import of pools with many |
76 | * datasets. | |
77 | * Default is 4 times the number of leaf vdevs. | |
78 | */ | |
79 | int dmu_find_threads = 0; | |
80 | ||
68cbd56e NB |
81 | /* |
82 | * Backfill lower metadnode objects after this many have been freed. | |
83 | * Backfilling negatively impacts object creation rates, so only do it | |
84 | * if there are enough holes to fill. | |
85 | */ | |
86 | int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; | |
87 | ||
c0daec32 AB |
88 | static char *upgrade_tag = "upgrade_tag"; |
89 | ||
9c43027b AJ |
90 | static void dmu_objset_find_dp_cb(void *arg); |
91 | ||
1de321e6 JX |
92 | static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); |
93 | static void dmu_objset_upgrade_stop(objset_t *os); | |
94 | ||
572e2857 BB |
95 | void |
96 | dmu_objset_init(void) | |
97 | { | |
98 | rw_init(&os_lock, NULL, RW_DEFAULT, NULL); | |
99 | } | |
100 | ||
101 | void | |
102 | dmu_objset_fini(void) | |
103 | { | |
104 | rw_destroy(&os_lock); | |
105 | } | |
34dc7c2f BB |
106 | |
107 | spa_t * | |
108 | dmu_objset_spa(objset_t *os) | |
109 | { | |
428870ff | 110 | return (os->os_spa); |
34dc7c2f BB |
111 | } |
112 | ||
113 | zilog_t * | |
114 | dmu_objset_zil(objset_t *os) | |
115 | { | |
428870ff | 116 | return (os->os_zil); |
34dc7c2f BB |
117 | } |
118 | ||
119 | dsl_pool_t * | |
120 | dmu_objset_pool(objset_t *os) | |
121 | { | |
122 | dsl_dataset_t *ds; | |
123 | ||
428870ff | 124 | if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) |
34dc7c2f BB |
125 | return (ds->ds_dir->dd_pool); |
126 | else | |
428870ff | 127 | return (spa_get_dsl(os->os_spa)); |
34dc7c2f BB |
128 | } |
129 | ||
130 | dsl_dataset_t * | |
131 | dmu_objset_ds(objset_t *os) | |
132 | { | |
428870ff | 133 | return (os->os_dsl_dataset); |
34dc7c2f BB |
134 | } |
135 | ||
136 | dmu_objset_type_t | |
137 | dmu_objset_type(objset_t *os) | |
138 | { | |
428870ff | 139 | return (os->os_phys->os_type); |
34dc7c2f BB |
140 | } |
141 | ||
142 | void | |
143 | dmu_objset_name(objset_t *os, char *buf) | |
144 | { | |
428870ff | 145 | dsl_dataset_name(os->os_dsl_dataset, buf); |
34dc7c2f BB |
146 | } |
147 | ||
148 | uint64_t | |
149 | dmu_objset_id(objset_t *os) | |
150 | { | |
428870ff | 151 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
152 | |
153 | return (ds ? ds->ds_object : 0); | |
154 | } | |
155 | ||
50c957f7 NB |
156 | uint64_t |
157 | dmu_objset_dnodesize(objset_t *os) | |
158 | { | |
159 | return (os->os_dnodesize); | |
160 | } | |
161 | ||
faf0f58c | 162 | zfs_sync_type_t |
428870ff BB |
163 | dmu_objset_syncprop(objset_t *os) |
164 | { | |
165 | return (os->os_sync); | |
166 | } | |
167 | ||
faf0f58c | 168 | zfs_logbias_op_t |
428870ff BB |
169 | dmu_objset_logbias(objset_t *os) |
170 | { | |
171 | return (os->os_logbias); | |
172 | } | |
173 | ||
34dc7c2f BB |
174 | static void |
175 | checksum_changed_cb(void *arg, uint64_t newval) | |
176 | { | |
428870ff | 177 | objset_t *os = arg; |
34dc7c2f BB |
178 | |
179 | /* | |
180 | * Inheritance should have been done by now. | |
181 | */ | |
182 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
183 | ||
428870ff | 184 | os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); |
34dc7c2f BB |
185 | } |
186 | ||
187 | static void | |
188 | compression_changed_cb(void *arg, uint64_t newval) | |
189 | { | |
428870ff | 190 | objset_t *os = arg; |
34dc7c2f BB |
191 | |
192 | /* | |
193 | * Inheritance and range checking should have been done by now. | |
194 | */ | |
195 | ASSERT(newval != ZIO_COMPRESS_INHERIT); | |
196 | ||
10b3c7f5 MN |
197 | os->os_compress = zio_compress_select(os->os_spa, |
198 | ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON); | |
199 | os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress, | |
200 | ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT); | |
34dc7c2f BB |
201 | } |
202 | ||
203 | static void | |
204 | copies_changed_cb(void *arg, uint64_t newval) | |
205 | { | |
428870ff | 206 | objset_t *os = arg; |
34dc7c2f BB |
207 | |
208 | /* | |
209 | * Inheritance and range checking should have been done by now. | |
210 | */ | |
211 | ASSERT(newval > 0); | |
428870ff | 212 | ASSERT(newval <= spa_max_replication(os->os_spa)); |
34dc7c2f | 213 | |
428870ff BB |
214 | os->os_copies = newval; |
215 | } | |
216 | ||
217 | static void | |
218 | dedup_changed_cb(void *arg, uint64_t newval) | |
219 | { | |
220 | objset_t *os = arg; | |
221 | spa_t *spa = os->os_spa; | |
222 | enum zio_checksum checksum; | |
223 | ||
224 | /* | |
225 | * Inheritance should have been done by now. | |
226 | */ | |
227 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
228 | ||
229 | checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); | |
230 | ||
231 | os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; | |
232 | os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); | |
34dc7c2f BB |
233 | } |
234 | ||
b128c09f BB |
235 | static void |
236 | primary_cache_changed_cb(void *arg, uint64_t newval) | |
237 | { | |
428870ff | 238 | objset_t *os = arg; |
b128c09f BB |
239 | |
240 | /* | |
241 | * Inheritance and range checking should have been done by now. | |
242 | */ | |
243 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
244 | newval == ZFS_CACHE_METADATA); | |
245 | ||
428870ff | 246 | os->os_primary_cache = newval; |
b128c09f BB |
247 | } |
248 | ||
249 | static void | |
250 | secondary_cache_changed_cb(void *arg, uint64_t newval) | |
251 | { | |
428870ff | 252 | objset_t *os = arg; |
b128c09f BB |
253 | |
254 | /* | |
255 | * Inheritance and range checking should have been done by now. | |
256 | */ | |
257 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
258 | newval == ZFS_CACHE_METADATA); | |
259 | ||
428870ff BB |
260 | os->os_secondary_cache = newval; |
261 | } | |
262 | ||
263 | static void | |
264 | sync_changed_cb(void *arg, uint64_t newval) | |
265 | { | |
266 | objset_t *os = arg; | |
267 | ||
268 | /* | |
269 | * Inheritance and range checking should have been done by now. | |
270 | */ | |
271 | ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || | |
272 | newval == ZFS_SYNC_DISABLED); | |
273 | ||
274 | os->os_sync = newval; | |
275 | if (os->os_zil) | |
276 | zil_set_sync(os->os_zil, newval); | |
277 | } | |
278 | ||
faf0f58c MA |
279 | static void |
280 | redundant_metadata_changed_cb(void *arg, uint64_t newval) | |
281 | { | |
282 | objset_t *os = arg; | |
283 | ||
284 | /* | |
285 | * Inheritance and range checking should have been done by now. | |
286 | */ | |
287 | ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || | |
288 | newval == ZFS_REDUNDANT_METADATA_MOST); | |
289 | ||
290 | os->os_redundant_metadata = newval; | |
291 | } | |
292 | ||
50c957f7 NB |
293 | static void |
294 | dnodesize_changed_cb(void *arg, uint64_t newval) | |
295 | { | |
296 | objset_t *os = arg; | |
297 | ||
298 | switch (newval) { | |
299 | case ZFS_DNSIZE_LEGACY: | |
300 | os->os_dnodesize = DNODE_MIN_SIZE; | |
301 | break; | |
302 | case ZFS_DNSIZE_AUTO: | |
303 | /* | |
304 | * Choose a dnode size that will work well for most | |
305 | * workloads if the user specified "auto". Future code | |
306 | * improvements could dynamically select a dnode size | |
307 | * based on observed workload patterns. | |
308 | */ | |
309 | os->os_dnodesize = DNODE_MIN_SIZE * 2; | |
310 | break; | |
311 | case ZFS_DNSIZE_1K: | |
312 | case ZFS_DNSIZE_2K: | |
313 | case ZFS_DNSIZE_4K: | |
314 | case ZFS_DNSIZE_8K: | |
315 | case ZFS_DNSIZE_16K: | |
316 | os->os_dnodesize = newval; | |
317 | break; | |
318 | } | |
319 | } | |
320 | ||
cc99f275 DB |
321 | static void |
322 | smallblk_changed_cb(void *arg, uint64_t newval) | |
323 | { | |
324 | objset_t *os = arg; | |
325 | ||
326 | /* | |
327 | * Inheritance and range checking should have been done by now. | |
328 | */ | |
329 | ASSERT(newval <= SPA_OLD_MAXBLOCKSIZE); | |
330 | ASSERT(ISP2(newval)); | |
331 | ||
332 | os->os_zpl_special_smallblock = newval; | |
333 | } | |
334 | ||
428870ff BB |
335 | static void |
336 | logbias_changed_cb(void *arg, uint64_t newval) | |
337 | { | |
338 | objset_t *os = arg; | |
339 | ||
340 | ASSERT(newval == ZFS_LOGBIAS_LATENCY || | |
341 | newval == ZFS_LOGBIAS_THROUGHPUT); | |
342 | os->os_logbias = newval; | |
343 | if (os->os_zil) | |
344 | zil_set_logbias(os->os_zil, newval); | |
b128c09f BB |
345 | } |
346 | ||
f1512ee6 MA |
347 | static void |
348 | recordsize_changed_cb(void *arg, uint64_t newval) | |
349 | { | |
350 | objset_t *os = arg; | |
351 | ||
352 | os->os_recordsize = newval; | |
353 | } | |
354 | ||
34dc7c2f BB |
355 | void |
356 | dmu_objset_byteswap(void *buf, size_t size) | |
357 | { | |
358 | objset_phys_t *osp = buf; | |
359 | ||
9c5167d1 NF |
360 | ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || |
361 | size == sizeof (objset_phys_t)); | |
34dc7c2f BB |
362 | dnode_byteswap(&osp->os_meta_dnode); |
363 | byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); | |
364 | osp->os_type = BSWAP_64(osp->os_type); | |
9babb374 | 365 | osp->os_flags = BSWAP_64(osp->os_flags); |
9c5167d1 | 366 | if (size >= OBJSET_PHYS_SIZE_V2) { |
9babb374 BB |
367 | dnode_byteswap(&osp->os_userused_dnode); |
368 | dnode_byteswap(&osp->os_groupused_dnode); | |
9c5167d1 NF |
369 | if (size >= sizeof (objset_phys_t)) |
370 | dnode_byteswap(&osp->os_projectused_dnode); | |
9babb374 | 371 | } |
34dc7c2f BB |
372 | } |
373 | ||
64fc7762 MA |
374 | /* |
375 | * The hash is a CRC-based hash of the objset_t pointer and the object number. | |
376 | */ | |
377 | static uint64_t | |
378 | dnode_hash(const objset_t *os, uint64_t obj) | |
379 | { | |
380 | uintptr_t osv = (uintptr_t)os; | |
381 | uint64_t crc = -1ULL; | |
382 | ||
383 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
384 | /* | |
385 | * The low 6 bits of the pointer don't have much entropy, because | |
386 | * the objset_t is larger than 2^6 bytes long. | |
387 | */ | |
388 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; | |
389 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; | |
390 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; | |
391 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; | |
392 | ||
393 | crc ^= (osv>>14) ^ (obj>>24); | |
394 | ||
395 | return (crc); | |
396 | } | |
397 | ||
65c7cc49 | 398 | static unsigned int |
64fc7762 MA |
399 | dnode_multilist_index_func(multilist_t *ml, void *obj) |
400 | { | |
401 | dnode_t *dn = obj; | |
402 | return (dnode_hash(dn->dn_objset, dn->dn_object) % | |
403 | multilist_get_num_sublists(ml)); | |
404 | } | |
405 | ||
a1d477c2 MA |
406 | /* |
407 | * Instantiates the objset_t in-memory structure corresponding to the | |
408 | * objset_phys_t that's pointed to by the specified blkptr_t. | |
409 | */ | |
34dc7c2f BB |
410 | int |
411 | dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
428870ff | 412 | objset_t **osp) |
34dc7c2f | 413 | { |
428870ff | 414 | objset_t *os; |
b128c09f | 415 | int i, err; |
34dc7c2f BB |
416 | |
417 | ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); | |
30af21b0 | 418 | ASSERT(!BP_IS_REDACTED(bp)); |
34dc7c2f | 419 | |
0fdd6106 MA |
420 | /* |
421 | * We need the pool config lock to get properties. | |
422 | */ | |
423 | ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool)); | |
424 | ||
a1d477c2 MA |
425 | /* |
426 | * The $ORIGIN dataset (if it exists) doesn't have an associated | |
427 | * objset, so there's no reason to open it. The $ORIGIN dataset | |
428 | * will not exist on pools older than SPA_VERSION_ORIGIN. | |
429 | */ | |
430 | if (ds != NULL && spa_get_dsl(spa) != NULL && | |
431 | spa_get_dsl(spa)->dp_origin_snap != NULL) { | |
432 | ASSERT3P(ds->ds_dir, !=, | |
433 | spa_get_dsl(spa)->dp_origin_snap->ds_dir); | |
434 | } | |
435 | ||
79c76d5b | 436 | os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); |
428870ff BB |
437 | os->os_dsl_dataset = ds; |
438 | os->os_spa = spa; | |
439 | os->os_rootbp = bp; | |
440 | if (!BP_IS_HOLE(os->os_rootbp)) { | |
2a432414 | 441 | arc_flags_t aflags = ARC_FLAG_WAIT; |
5dbd68a3 | 442 | zbookmark_phys_t zb; |
9c5167d1 | 443 | int size; |
b5256303 | 444 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; |
428870ff BB |
445 | SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, |
446 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
447 | ||
448 | if (DMU_OS_IS_L2CACHEABLE(os)) | |
2a432414 | 449 | aflags |= ARC_FLAG_L2CACHE; |
34dc7c2f | 450 | |
b5256303 TC |
451 | if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { |
452 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
453 | ASSERT(BP_IS_AUTHENTICATED(bp)); | |
454 | zio_flags |= ZIO_FLAG_RAW; | |
455 | } | |
456 | ||
428870ff | 457 | dprintf_bp(os->os_rootbp, "reading %s", ""); |
294f6806 | 458 | err = arc_read(NULL, spa, os->os_rootbp, |
428870ff | 459 | arc_getbuf_func, &os->os_phys_buf, |
b5256303 | 460 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
13fe0198 | 461 | if (err != 0) { |
428870ff | 462 | kmem_free(os, sizeof (objset_t)); |
b128c09f BB |
463 | /* convert checksum errors into IO errors */ |
464 | if (err == ECKSUM) | |
2e528b49 | 465 | err = SET_ERROR(EIO); |
34dc7c2f BB |
466 | return (err); |
467 | } | |
9babb374 | 468 | |
9c5167d1 NF |
469 | if (spa_version(spa) < SPA_VERSION_USERSPACE) |
470 | size = OBJSET_PHYS_SIZE_V1; | |
471 | else if (!spa_feature_is_enabled(spa, | |
472 | SPA_FEATURE_PROJECT_QUOTA)) | |
473 | size = OBJSET_PHYS_SIZE_V2; | |
474 | else | |
475 | size = sizeof (objset_phys_t); | |
476 | ||
9babb374 | 477 | /* Increase the blocksize if we are permitted. */ |
9c5167d1 | 478 | if (arc_buf_size(os->os_phys_buf) < size) { |
2aa34383 | 479 | arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, |
9c5167d1 NF |
480 | ARC_BUFC_METADATA, size); |
481 | bzero(buf->b_data, size); | |
428870ff BB |
482 | bcopy(os->os_phys_buf->b_data, buf->b_data, |
483 | arc_buf_size(os->os_phys_buf)); | |
d3c2ae1c | 484 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 485 | os->os_phys_buf = buf; |
9babb374 BB |
486 | } |
487 | ||
428870ff BB |
488 | os->os_phys = os->os_phys_buf->b_data; |
489 | os->os_flags = os->os_phys->os_flags; | |
34dc7c2f | 490 | } else { |
9babb374 | 491 | int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? |
9c5167d1 | 492 | sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; |
2aa34383 DK |
493 | os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, |
494 | ARC_BUFC_METADATA, size); | |
428870ff BB |
495 | os->os_phys = os->os_phys_buf->b_data; |
496 | bzero(os->os_phys, size); | |
34dc7c2f | 497 | } |
2e5dc449 MA |
498 | /* |
499 | * These properties will be filled in by the logic in zfs_get_zplprop() | |
500 | * when they are queried for the first time. | |
501 | */ | |
502 | os->os_version = OBJSET_PROP_UNINITIALIZED; | |
503 | os->os_normalization = OBJSET_PROP_UNINITIALIZED; | |
504 | os->os_utf8only = OBJSET_PROP_UNINITIALIZED; | |
505 | os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; | |
34dc7c2f BB |
506 | |
507 | /* | |
508 | * Note: the changed_cb will be called once before the register | |
509 | * func returns, thus changing the checksum/compression from the | |
b128c09f BB |
510 | * default (fletcher2/off). Snapshots don't need to know about |
511 | * checksum/compression/copies. | |
34dc7c2f | 512 | */ |
9b67f605 | 513 | if (ds != NULL) { |
b5256303 TC |
514 | os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); |
515 | ||
13fe0198 MA |
516 | err = dsl_prop_register(ds, |
517 | zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), | |
428870ff | 518 | primary_cache_changed_cb, os); |
13fe0198 MA |
519 | if (err == 0) { |
520 | err = dsl_prop_register(ds, | |
521 | zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), | |
428870ff | 522 | secondary_cache_changed_cb, os); |
13fe0198 | 523 | } |
0c66c32d | 524 | if (!ds->ds_is_snapshot) { |
13fe0198 MA |
525 | if (err == 0) { |
526 | err = dsl_prop_register(ds, | |
527 | zfs_prop_to_name(ZFS_PROP_CHECKSUM), | |
428870ff | 528 | checksum_changed_cb, os); |
13fe0198 MA |
529 | } |
530 | if (err == 0) { | |
531 | err = dsl_prop_register(ds, | |
532 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
428870ff | 533 | compression_changed_cb, os); |
13fe0198 MA |
534 | } |
535 | if (err == 0) { | |
536 | err = dsl_prop_register(ds, | |
537 | zfs_prop_to_name(ZFS_PROP_COPIES), | |
428870ff | 538 | copies_changed_cb, os); |
13fe0198 MA |
539 | } |
540 | if (err == 0) { | |
541 | err = dsl_prop_register(ds, | |
542 | zfs_prop_to_name(ZFS_PROP_DEDUP), | |
428870ff | 543 | dedup_changed_cb, os); |
13fe0198 MA |
544 | } |
545 | if (err == 0) { | |
546 | err = dsl_prop_register(ds, | |
547 | zfs_prop_to_name(ZFS_PROP_LOGBIAS), | |
428870ff | 548 | logbias_changed_cb, os); |
13fe0198 MA |
549 | } |
550 | if (err == 0) { | |
551 | err = dsl_prop_register(ds, | |
552 | zfs_prop_to_name(ZFS_PROP_SYNC), | |
428870ff | 553 | sync_changed_cb, os); |
13fe0198 | 554 | } |
faf0f58c MA |
555 | if (err == 0) { |
556 | err = dsl_prop_register(ds, | |
557 | zfs_prop_to_name( | |
558 | ZFS_PROP_REDUNDANT_METADATA), | |
559 | redundant_metadata_changed_cb, os); | |
560 | } | |
f1512ee6 MA |
561 | if (err == 0) { |
562 | err = dsl_prop_register(ds, | |
563 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
564 | recordsize_changed_cb, os); | |
565 | } | |
50c957f7 NB |
566 | if (err == 0) { |
567 | err = dsl_prop_register(ds, | |
568 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
569 | dnodesize_changed_cb, os); | |
570 | } | |
cc99f275 DB |
571 | if (err == 0) { |
572 | err = dsl_prop_register(ds, | |
573 | zfs_prop_to_name( | |
574 | ZFS_PROP_SPECIAL_SMALL_BLOCKS), | |
575 | smallblk_changed_cb, os); | |
576 | } | |
b128c09f | 577 | } |
13fe0198 | 578 | if (err != 0) { |
d3c2ae1c | 579 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 580 | kmem_free(os, sizeof (objset_t)); |
34dc7c2f BB |
581 | return (err); |
582 | } | |
9b67f605 | 583 | } else { |
34dc7c2f | 584 | /* It's the meta-objset. */ |
428870ff | 585 | os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; |
99197f03 | 586 | os->os_compress = ZIO_COMPRESS_ON; |
10b3c7f5 | 587 | os->os_complevel = ZIO_COMPLEVEL_DEFAULT; |
b5256303 | 588 | os->os_encrypted = B_FALSE; |
428870ff BB |
589 | os->os_copies = spa_max_replication(spa); |
590 | os->os_dedup_checksum = ZIO_CHECKSUM_OFF; | |
faf0f58c MA |
591 | os->os_dedup_verify = B_FALSE; |
592 | os->os_logbias = ZFS_LOGBIAS_LATENCY; | |
593 | os->os_sync = ZFS_SYNC_STANDARD; | |
428870ff BB |
594 | os->os_primary_cache = ZFS_CACHE_ALL; |
595 | os->os_secondary_cache = ZFS_CACHE_ALL; | |
50c957f7 | 596 | os->os_dnodesize = DNODE_MIN_SIZE; |
34dc7c2f BB |
597 | } |
598 | ||
0c66c32d | 599 | if (ds == NULL || !ds->ds_is_snapshot) |
572e2857 | 600 | os->os_zil_header = os->os_phys->os_zil_header; |
428870ff | 601 | os->os_zil = zil_alloc(os, &os->os_zil_header); |
34dc7c2f BB |
602 | |
603 | for (i = 0; i < TXG_SIZE; i++) { | |
64fc7762 MA |
604 | os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), |
605 | offsetof(dnode_t, dn_dirty_link[i]), | |
606 | dnode_multilist_index_func); | |
34dc7c2f | 607 | } |
428870ff | 608 | list_create(&os->os_dnodes, sizeof (dnode_t), |
34dc7c2f | 609 | offsetof(dnode_t, dn_link)); |
428870ff | 610 | list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), |
34dc7c2f BB |
611 | offsetof(dmu_buf_impl_t, db_link)); |
612 | ||
0c66c32d JG |
613 | list_link_init(&os->os_evicting_node); |
614 | ||
428870ff | 615 | mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); |
64fc7762 | 616 | mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); |
428870ff BB |
617 | mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); |
618 | mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); | |
dbeb8796 MA |
619 | os->os_obj_next_percpu_len = boot_ncpus; |
620 | os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * | |
621 | sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); | |
428870ff | 622 | |
0c66c32d JG |
623 | dnode_special_open(os, &os->os_phys->os_meta_dnode, |
624 | DMU_META_DNODE_OBJECT, &os->os_meta_dnode); | |
9c5167d1 | 625 | if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { |
0c66c32d JG |
626 | dnode_special_open(os, &os->os_phys->os_userused_dnode, |
627 | DMU_USERUSED_OBJECT, &os->os_userused_dnode); | |
628 | dnode_special_open(os, &os->os_phys->os_groupused_dnode, | |
629 | DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); | |
9c5167d1 NF |
630 | if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) |
631 | dnode_special_open(os, | |
632 | &os->os_phys->os_projectused_dnode, | |
633 | DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); | |
9babb374 | 634 | } |
34dc7c2f | 635 | |
1de321e6 JX |
636 | mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); |
637 | ||
428870ff | 638 | *osp = os; |
34dc7c2f BB |
639 | return (0); |
640 | } | |
641 | ||
428870ff BB |
642 | int |
643 | dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) | |
34dc7c2f | 644 | { |
428870ff | 645 | int err = 0; |
34dc7c2f | 646 | |
47dfff3b | 647 | /* |
0fdd6106 MA |
648 | * We need the pool_config lock to manipulate the dsl_dataset_t. |
649 | * Even if the dataset is long-held, we need the pool_config lock | |
650 | * to open the objset, as it needs to get properties. | |
47dfff3b | 651 | */ |
0fdd6106 | 652 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); |
47dfff3b | 653 | |
34dc7c2f | 654 | mutex_enter(&ds->ds_opening_lock); |
9b67f605 MA |
655 | if (ds->ds_objset == NULL) { |
656 | objset_t *os; | |
cc9bb3e5 | 657 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
34dc7c2f | 658 | err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), |
9b67f605 | 659 | ds, dsl_dataset_get_blkptr(ds), &os); |
cc9bb3e5 | 660 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
9b67f605 MA |
661 | |
662 | if (err == 0) { | |
663 | mutex_enter(&ds->ds_lock); | |
664 | ASSERT(ds->ds_objset == NULL); | |
665 | ds->ds_objset = os; | |
666 | mutex_exit(&ds->ds_lock); | |
667 | } | |
34dc7c2f | 668 | } |
9b67f605 | 669 | *osp = ds->ds_objset; |
34dc7c2f | 670 | mutex_exit(&ds->ds_opening_lock); |
428870ff | 671 | return (err); |
34dc7c2f BB |
672 | } |
673 | ||
13fe0198 MA |
674 | /* |
675 | * Holds the pool while the objset is held. Therefore only one objset | |
676 | * can be held at a time. | |
677 | */ | |
34dc7c2f | 678 | int |
b5256303 TC |
679 | dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag, |
680 | objset_t **osp) | |
34dc7c2f | 681 | { |
13fe0198 | 682 | dsl_pool_t *dp; |
428870ff | 683 | dsl_dataset_t *ds; |
34dc7c2f | 684 | int err; |
b5256303 | 685 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
34dc7c2f | 686 | |
13fe0198 MA |
687 | err = dsl_pool_hold(name, tag, &dp); |
688 | if (err != 0) | |
689 | return (err); | |
b5256303 | 690 | err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); |
13fe0198 MA |
691 | if (err != 0) { |
692 | dsl_pool_rele(dp, tag); | |
428870ff | 693 | return (err); |
13fe0198 | 694 | } |
428870ff BB |
695 | |
696 | err = dmu_objset_from_ds(ds, osp); | |
13fe0198 | 697 | if (err != 0) { |
428870ff | 698 | dsl_dataset_rele(ds, tag); |
13fe0198 MA |
699 | dsl_pool_rele(dp, tag); |
700 | } | |
428870ff | 701 | |
34dc7c2f BB |
702 | return (err); |
703 | } | |
704 | ||
b5256303 TC |
705 | int |
706 | dmu_objset_hold(const char *name, void *tag, objset_t **osp) | |
707 | { | |
708 | return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); | |
709 | } | |
710 | ||
9c43027b AJ |
711 | static int |
712 | dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, | |
b5256303 | 713 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
9c43027b AJ |
714 | { |
715 | int err; | |
716 | ||
717 | err = dmu_objset_from_ds(ds, osp); | |
718 | if (err != 0) { | |
b5256303 | 719 | return (err); |
9c43027b | 720 | } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { |
9c43027b AJ |
721 | return (SET_ERROR(EINVAL)); |
722 | } else if (!readonly && dsl_dataset_is_snapshot(ds)) { | |
9c43027b | 723 | return (SET_ERROR(EROFS)); |
ae76f45c TC |
724 | } else if (!readonly && decrypt && |
725 | dsl_dir_incompatible_encryption_version(ds->ds_dir)) { | |
726 | return (SET_ERROR(EROFS)); | |
9c43027b | 727 | } |
b5256303 TC |
728 | |
729 | /* if we are decrypting, we can now check MACs in os->os_phys_buf */ | |
730 | if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { | |
a2c2ed1b TC |
731 | zbookmark_phys_t zb; |
732 | ||
733 | SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, | |
734 | ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
b5256303 | 735 | err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, |
a2c2ed1b | 736 | &zb, B_FALSE); |
b5256303 TC |
737 | if (err != 0) |
738 | return (err); | |
739 | ||
740 | ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); | |
741 | } | |
742 | ||
743 | return (0); | |
9c43027b AJ |
744 | } |
745 | ||
13fe0198 MA |
746 | /* |
747 | * dsl_pool must not be held when this is called. | |
748 | * Upon successful return, there will be a longhold on the dataset, | |
749 | * and the dsl_pool will not be held. | |
750 | */ | |
34dc7c2f | 751 | int |
428870ff | 752 | dmu_objset_own(const char *name, dmu_objset_type_t type, |
b5256303 | 753 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
34dc7c2f | 754 | { |
13fe0198 | 755 | dsl_pool_t *dp; |
34dc7c2f BB |
756 | dsl_dataset_t *ds; |
757 | int err; | |
b5256303 | 758 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
34dc7c2f | 759 | |
13fe0198 MA |
760 | err = dsl_pool_hold(name, FTAG, &dp); |
761 | if (err != 0) | |
762 | return (err); | |
b5256303 | 763 | err = dsl_dataset_own(dp, name, flags, tag, &ds); |
13fe0198 MA |
764 | if (err != 0) { |
765 | dsl_pool_rele(dp, FTAG); | |
34dc7c2f | 766 | return (err); |
13fe0198 | 767 | } |
b5256303 TC |
768 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
769 | if (err != 0) { | |
770 | dsl_dataset_disown(ds, flags, tag); | |
771 | dsl_pool_rele(dp, FTAG); | |
772 | return (err); | |
773 | } | |
774 | ||
163a8c28 TC |
775 | /* |
776 | * User accounting requires the dataset to be decrypted and rw. | |
777 | * We also don't begin user accounting during claiming to help | |
778 | * speed up pool import times and to keep this txg reserved | |
779 | * completely for recovery work. | |
780 | */ | |
4072f465 | 781 | if (!readonly && !dp->dp_spa->spa_claiming && |
782 | (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) { | |
783 | if (dmu_objset_userobjspace_upgradable(*osp) || | |
784 | dmu_objset_projectquota_upgradable(*osp)) { | |
785 | dmu_objset_id_quota_upgrade(*osp); | |
786 | } else if (dmu_objset_userused_enabled(*osp)) { | |
787 | dmu_objset_userspace_upgrade(*osp); | |
788 | } | |
789 | } | |
1de321e6 | 790 | |
c0daec32 | 791 | dsl_pool_rele(dp, FTAG); |
b5256303 | 792 | return (0); |
34dc7c2f BB |
793 | } |
794 | ||
9c43027b AJ |
795 | int |
796 | dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, | |
b5256303 | 797 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) |
9c43027b AJ |
798 | { |
799 | dsl_dataset_t *ds; | |
800 | int err; | |
b5256303 | 801 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
9c43027b | 802 | |
b5256303 | 803 | err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); |
9c43027b AJ |
804 | if (err != 0) |
805 | return (err); | |
806 | ||
b5256303 TC |
807 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
808 | if (err != 0) { | |
809 | dsl_dataset_disown(ds, flags, tag); | |
810 | return (err); | |
811 | } | |
812 | ||
813 | return (0); | |
9c43027b AJ |
814 | } |
815 | ||
34dc7c2f | 816 | void |
b5256303 | 817 | dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag) |
34dc7c2f | 818 | { |
b5256303 TC |
819 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; |
820 | ||
13fe0198 | 821 | dsl_pool_t *dp = dmu_objset_pool(os); |
b5256303 | 822 | dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); |
13fe0198 | 823 | dsl_pool_rele(dp, tag); |
428870ff | 824 | } |
b128c09f | 825 | |
b5256303 TC |
826 | void |
827 | dmu_objset_rele(objset_t *os, void *tag) | |
828 | { | |
829 | dmu_objset_rele_flags(os, B_FALSE, tag); | |
830 | } | |
831 | ||
831baf06 KW |
832 | /* |
833 | * When we are called, os MUST refer to an objset associated with a dataset | |
834 | * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner | |
835 | * == tag. We will then release and reacquire ownership of the dataset while | |
836 | * holding the pool config_rwlock to avoid intervening namespace or ownership | |
837 | * changes may occur. | |
838 | * | |
839 | * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to | |
840 | * release the hold on its dataset and acquire a new one on the dataset of the | |
841 | * same name so that it can be partially torn down and reconstructed. | |
842 | */ | |
843 | void | |
5e00213e AG |
844 | dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, |
845 | boolean_t decrypt, void *tag) | |
831baf06 KW |
846 | { |
847 | dsl_pool_t *dp; | |
eca7b760 | 848 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
831baf06 | 849 | |
831baf06 KW |
850 | VERIFY3P(ds, !=, NULL); |
851 | VERIFY3P(ds->ds_owner, ==, tag); | |
852 | VERIFY(dsl_dataset_long_held(ds)); | |
853 | ||
854 | dsl_dataset_name(ds, name); | |
5e00213e | 855 | dp = ds->ds_dir->dd_pool; |
831baf06 | 856 | dsl_pool_config_enter(dp, FTAG); |
5e00213e | 857 | dsl_dataset_disown(ds, decrypt, tag); |
b5256303 | 858 | VERIFY0(dsl_dataset_own(dp, name, |
5e00213e | 859 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag, newds)); |
831baf06 KW |
860 | dsl_pool_config_exit(dp, FTAG); |
861 | } | |
862 | ||
428870ff | 863 | void |
b5256303 | 864 | dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag) |
428870ff | 865 | { |
1de321e6 JX |
866 | /* |
867 | * Stop upgrading thread | |
868 | */ | |
869 | dmu_objset_upgrade_stop(os); | |
b5256303 TC |
870 | dsl_dataset_disown(os->os_dsl_dataset, |
871 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag); | |
34dc7c2f BB |
872 | } |
873 | ||
13fe0198 | 874 | void |
34dc7c2f BB |
875 | dmu_objset_evict_dbufs(objset_t *os) |
876 | { | |
0c66c32d | 877 | dnode_t *dn_marker; |
34dc7c2f BB |
878 | dnode_t *dn; |
879 | ||
0c66c32d | 880 | dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); |
34dc7c2f | 881 | |
0c66c32d JG |
882 | mutex_enter(&os->os_lock); |
883 | dn = list_head(&os->os_dnodes); | |
884 | while (dn != NULL) { | |
885 | /* | |
886 | * Skip dnodes without holds. We have to do this dance | |
887 | * because dnode_add_ref() only works if there is already a | |
888 | * hold. If the dnode has no holds, then it has no dbufs. | |
889 | */ | |
890 | if (dnode_add_ref(dn, FTAG)) { | |
891 | list_insert_after(&os->os_dnodes, dn, dn_marker); | |
892 | mutex_exit(&os->os_lock); | |
34dc7c2f | 893 | |
0c66c32d JG |
894 | dnode_evict_dbufs(dn); |
895 | dnode_rele(dn, FTAG); | |
34dc7c2f | 896 | |
0c66c32d JG |
897 | mutex_enter(&os->os_lock); |
898 | dn = list_next(&os->os_dnodes, dn_marker); | |
899 | list_remove(&os->os_dnodes, dn_marker); | |
900 | } else { | |
901 | dn = list_next(&os->os_dnodes, dn); | |
902 | } | |
903 | } | |
904 | mutex_exit(&os->os_lock); | |
34dc7c2f | 905 | |
0c66c32d | 906 | kmem_free(dn_marker, sizeof (dnode_t)); |
34dc7c2f | 907 | |
0c66c32d | 908 | if (DMU_USERUSED_DNODE(os) != NULL) { |
9c5167d1 NF |
909 | if (DMU_PROJECTUSED_DNODE(os) != NULL) |
910 | dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); | |
0c66c32d JG |
911 | dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); |
912 | dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); | |
34dc7c2f | 913 | } |
0c66c32d | 914 | dnode_evict_dbufs(DMU_META_DNODE(os)); |
34dc7c2f BB |
915 | } |
916 | ||
0c66c32d JG |
917 | /* |
918 | * Objset eviction processing is split into into two pieces. | |
919 | * The first marks the objset as evicting, evicts any dbufs that | |
920 | * have a refcount of zero, and then queues up the objset for the | |
921 | * second phase of eviction. Once os->os_dnodes has been cleared by | |
922 | * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. | |
923 | * The second phase closes the special dnodes, dequeues the objset from | |
924 | * the list of those undergoing eviction, and finally frees the objset. | |
925 | * | |
926 | * NOTE: Due to asynchronous eviction processing (invocation of | |
927 | * dnode_buf_pageout()), it is possible for the meta dnode for the | |
928 | * objset to have no holds even though os->os_dnodes is not empty. | |
929 | */ | |
34dc7c2f | 930 | void |
428870ff | 931 | dmu_objset_evict(objset_t *os) |
34dc7c2f | 932 | { |
6f1ffb06 MA |
933 | dsl_dataset_t *ds = os->os_dsl_dataset; |
934 | ||
1c27024e | 935 | for (int t = 0; t < TXG_SIZE; t++) |
428870ff | 936 | ASSERT(!dmu_objset_is_dirty(os, t)); |
34dc7c2f | 937 | |
0eb21616 JG |
938 | if (ds) |
939 | dsl_prop_unregister_all(ds, os); | |
34dc7c2f | 940 | |
428870ff BB |
941 | if (os->os_sa) |
942 | sa_tear_down(os); | |
943 | ||
13fe0198 | 944 | dmu_objset_evict_dbufs(os); |
34dc7c2f | 945 | |
0c66c32d JG |
946 | mutex_enter(&os->os_lock); |
947 | spa_evicting_os_register(os->os_spa, os); | |
948 | if (list_is_empty(&os->os_dnodes)) { | |
949 | mutex_exit(&os->os_lock); | |
950 | dmu_objset_evict_done(os); | |
951 | } else { | |
952 | mutex_exit(&os->os_lock); | |
953 | } | |
b5256303 TC |
954 | |
955 | ||
0c66c32d JG |
956 | } |
957 | ||
958 | void | |
959 | dmu_objset_evict_done(objset_t *os) | |
960 | { | |
961 | ASSERT3P(list_head(&os->os_dnodes), ==, NULL); | |
962 | ||
572e2857 BB |
963 | dnode_special_close(&os->os_meta_dnode); |
964 | if (DMU_USERUSED_DNODE(os)) { | |
9c5167d1 NF |
965 | if (DMU_PROJECTUSED_DNODE(os)) |
966 | dnode_special_close(&os->os_projectused_dnode); | |
572e2857 BB |
967 | dnode_special_close(&os->os_userused_dnode); |
968 | dnode_special_close(&os->os_groupused_dnode); | |
9babb374 | 969 | } |
428870ff BB |
970 | zil_free(os->os_zil); |
971 | ||
d3c2ae1c | 972 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
572e2857 BB |
973 | |
974 | /* | |
975 | * This is a barrier to prevent the objset from going away in | |
976 | * dnode_move() until we can safely ensure that the objset is still in | |
977 | * use. We consider the objset valid before the barrier and invalid | |
978 | * after the barrier. | |
979 | */ | |
980 | rw_enter(&os_lock, RW_READER); | |
981 | rw_exit(&os_lock); | |
982 | ||
dbeb8796 MA |
983 | kmem_free(os->os_obj_next_percpu, |
984 | os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); | |
985 | ||
428870ff | 986 | mutex_destroy(&os->os_lock); |
64fc7762 | 987 | mutex_destroy(&os->os_userused_lock); |
428870ff BB |
988 | mutex_destroy(&os->os_obj_lock); |
989 | mutex_destroy(&os->os_user_ptr_lock); | |
c17486b2 | 990 | mutex_destroy(&os->os_upgrade_lock); |
64fc7762 MA |
991 | for (int i = 0; i < TXG_SIZE; i++) { |
992 | multilist_destroy(os->os_dirty_dnodes[i]); | |
993 | } | |
0c66c32d | 994 | spa_evicting_os_deregister(os->os_spa, os); |
428870ff BB |
995 | kmem_free(os, sizeof (objset_t)); |
996 | } | |
9babb374 | 997 | |
6413c95f | 998 | inode_timespec_t |
428870ff BB |
999 | dmu_objset_snap_cmtime(objset_t *os) |
1000 | { | |
1001 | return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); | |
34dc7c2f BB |
1002 | } |
1003 | ||
428870ff | 1004 | objset_t * |
b5256303 TC |
1005 | dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, |
1006 | dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) | |
34dc7c2f | 1007 | { |
428870ff | 1008 | objset_t *os; |
34dc7c2f BB |
1009 | dnode_t *mdn; |
1010 | ||
1011 | ASSERT(dmu_tx_is_syncing(tx)); | |
13fe0198 | 1012 | |
b5256303 TC |
1013 | if (blksz == 0) |
1014 | blksz = DNODE_BLOCK_SIZE; | |
4807c0ba | 1015 | if (ibs == 0) |
b5256303 TC |
1016 | ibs = DN_MAX_INDBLKSHIFT; |
1017 | ||
572e2857 | 1018 | if (ds != NULL) |
13fe0198 | 1019 | VERIFY0(dmu_objset_from_ds(ds, &os)); |
572e2857 | 1020 | else |
13fe0198 | 1021 | VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); |
572e2857 BB |
1022 | |
1023 | mdn = DMU_META_DNODE(os); | |
34dc7c2f | 1024 | |
b5256303 TC |
1025 | dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, |
1026 | DNODE_MIN_SLOTS, tx); | |
34dc7c2f BB |
1027 | |
1028 | /* | |
1029 | * We don't want to have to increase the meta-dnode's nlevels | |
e1cfd73f | 1030 | * later, because then we could do it in quiescing context while |
34dc7c2f BB |
1031 | * we are also accessing it in open context. |
1032 | * | |
1033 | * This precaution is not necessary for the MOS (ds == NULL), | |
1034 | * because the MOS is only updated in syncing context. | |
1035 | * This is most fortunate: the MOS is the only objset that | |
1036 | * needs to be synced multiple times as spa_sync() iterates | |
1037 | * to convergence, so minimizing its dn_nlevels matters. | |
1038 | */ | |
1039 | if (ds != NULL) { | |
b5256303 TC |
1040 | if (levels == 0) { |
1041 | levels = 1; | |
1042 | ||
1043 | /* | |
1044 | * Determine the number of levels necessary for the | |
1045 | * meta-dnode to contain DN_MAX_OBJECT dnodes. Note | |
1046 | * that in order to ensure that we do not overflow | |
1047 | * 64 bits, there has to be a nlevels that gives us a | |
1048 | * number of blocks > DN_MAX_OBJECT but < 2^64. | |
1049 | * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) | |
1050 | * (10) must be less than (64 - log2(DN_MAX_OBJECT)) | |
1051 | * (16). | |
1052 | */ | |
1053 | while ((uint64_t)mdn->dn_nblkptr << | |
1054 | (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * | |
1055 | (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < | |
1056 | DN_MAX_OBJECT) | |
1057 | levels++; | |
1058 | } | |
34dc7c2f BB |
1059 | |
1060 | mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = | |
1061 | mdn->dn_nlevels = levels; | |
1062 | } | |
1063 | ||
1064 | ASSERT(type != DMU_OST_NONE); | |
1065 | ASSERT(type != DMU_OST_ANY); | |
1066 | ASSERT(type < DMU_OST_NUMTYPES); | |
428870ff | 1067 | os->os_phys->os_type = type; |
b5256303 TC |
1068 | |
1069 | /* | |
1070 | * Enable user accounting if it is enabled and this is not an | |
1071 | * encrypted receive. | |
1072 | */ | |
1073 | if (dmu_objset_userused_enabled(os) && | |
1074 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
428870ff | 1075 | os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
1de321e6 | 1076 | if (dmu_objset_userobjused_enabled(os)) { |
d52d80b7 PD |
1077 | ds->ds_feature_activation[ |
1078 | SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; | |
1de321e6 JX |
1079 | os->os_phys->os_flags |= |
1080 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
1081 | } | |
9c5167d1 | 1082 | if (dmu_objset_projectquota_enabled(os)) { |
d52d80b7 PD |
1083 | ds->ds_feature_activation[ |
1084 | SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; | |
9c5167d1 NF |
1085 | os->os_phys->os_flags |= |
1086 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
1087 | } | |
428870ff | 1088 | os->os_flags = os->os_phys->os_flags; |
9babb374 | 1089 | } |
34dc7c2f BB |
1090 | |
1091 | dsl_dataset_dirty(ds, tx); | |
1092 | ||
428870ff | 1093 | return (os); |
34dc7c2f BB |
1094 | } |
1095 | ||
b5256303 TC |
1096 | /* called from dsl for meta-objset */ |
1097 | objset_t * | |
1098 | dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
1099 | dmu_objset_type_t type, dmu_tx_t *tx) | |
1100 | { | |
1101 | return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); | |
1102 | } | |
1103 | ||
13fe0198 MA |
1104 | typedef struct dmu_objset_create_arg { |
1105 | const char *doca_name; | |
1106 | cred_t *doca_cred; | |
e59a377a | 1107 | proc_t *doca_proc; |
13fe0198 MA |
1108 | void (*doca_userfunc)(objset_t *os, void *arg, |
1109 | cred_t *cr, dmu_tx_t *tx); | |
1110 | void *doca_userarg; | |
1111 | dmu_objset_type_t doca_type; | |
1112 | uint64_t doca_flags; | |
b5256303 | 1113 | dsl_crypto_params_t *doca_dcp; |
13fe0198 | 1114 | } dmu_objset_create_arg_t; |
34dc7c2f BB |
1115 | |
1116 | /*ARGSUSED*/ | |
1117 | static int | |
13fe0198 | 1118 | dmu_objset_create_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1119 | { |
13fe0198 MA |
1120 | dmu_objset_create_arg_t *doca = arg; |
1121 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1122 | dsl_dir_t *pdd; | |
d8d418ff | 1123 | dsl_dataset_t *parentds; |
1124 | objset_t *parentos; | |
13fe0198 MA |
1125 | const char *tail; |
1126 | int error; | |
34dc7c2f | 1127 | |
13fe0198 | 1128 | if (strchr(doca->doca_name, '@') != NULL) |
2e528b49 | 1129 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 1130 | |
eca7b760 IK |
1131 | if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) |
1132 | return (SET_ERROR(ENAMETOOLONG)); | |
1133 | ||
a7ed98d8 SD |
1134 | if (dataset_nestcheck(doca->doca_name) != 0) |
1135 | return (SET_ERROR(ENAMETOOLONG)); | |
1136 | ||
13fe0198 MA |
1137 | error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); |
1138 | if (error != 0) | |
1139 | return (error); | |
1140 | if (tail == NULL) { | |
1141 | dsl_dir_rele(pdd, FTAG); | |
2e528b49 | 1142 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1143 | } |
b5256303 | 1144 | |
1fff937a | 1145 | error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); |
b5256303 TC |
1146 | if (error != 0) { |
1147 | dsl_dir_rele(pdd, FTAG); | |
1148 | return (error); | |
1149 | } | |
1150 | ||
788eb90c | 1151 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
e59a377a | 1152 | doca->doca_cred, doca->doca_proc); |
d8d418ff | 1153 | if (error != 0) { |
1154 | dsl_dir_rele(pdd, FTAG); | |
1155 | return (error); | |
1156 | } | |
b5256303 | 1157 | |
d8d418ff | 1158 | /* can't create below anything but filesystems (eg. no ZVOLs) */ |
1159 | error = dsl_dataset_hold_obj(pdd->dd_pool, | |
1160 | dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); | |
1161 | if (error != 0) { | |
1162 | dsl_dir_rele(pdd, FTAG); | |
1163 | return (error); | |
1164 | } | |
1165 | error = dmu_objset_from_ds(parentds, &parentos); | |
1166 | if (error != 0) { | |
1167 | dsl_dataset_rele(parentds, FTAG); | |
1168 | dsl_dir_rele(pdd, FTAG); | |
1169 | return (error); | |
1170 | } | |
1171 | if (dmu_objset_type(parentos) != DMU_OST_ZFS) { | |
1172 | dsl_dataset_rele(parentds, FTAG); | |
1173 | dsl_dir_rele(pdd, FTAG); | |
1174 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); | |
1175 | } | |
1176 | dsl_dataset_rele(parentds, FTAG); | |
13fe0198 | 1177 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f | 1178 | |
788eb90c | 1179 | return (error); |
34dc7c2f BB |
1180 | } |
1181 | ||
1182 | static void | |
13fe0198 | 1183 | dmu_objset_create_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1184 | { |
13fe0198 MA |
1185 | dmu_objset_create_arg_t *doca = arg; |
1186 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
52ce99dd | 1187 | spa_t *spa = dp->dp_spa; |
13fe0198 MA |
1188 | dsl_dir_t *pdd; |
1189 | const char *tail; | |
6f1ffb06 | 1190 | dsl_dataset_t *ds; |
13fe0198 | 1191 | uint64_t obj; |
6f1ffb06 | 1192 | blkptr_t *bp; |
13fe0198 | 1193 | objset_t *os; |
b5256303 | 1194 | zio_t *rzio; |
34dc7c2f | 1195 | |
13fe0198 | 1196 | VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); |
34dc7c2f | 1197 | |
13fe0198 | 1198 | obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, |
b5256303 | 1199 | doca->doca_cred, doca->doca_dcp, tx); |
34dc7c2f | 1200 | |
b5256303 TC |
1201 | VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, |
1202 | DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); | |
cc9bb3e5 | 1203 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
6f1ffb06 | 1204 | bp = dsl_dataset_get_blkptr(ds); |
52ce99dd | 1205 | os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); |
cc9bb3e5 | 1206 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
34dc7c2f | 1207 | |
13fe0198 MA |
1208 | if (doca->doca_userfunc != NULL) { |
1209 | doca->doca_userfunc(os, doca->doca_userarg, | |
1210 | doca->doca_cred, tx); | |
34dc7c2f BB |
1211 | } |
1212 | ||
b5256303 | 1213 | /* |
4807c0ba | 1214 | * The doca_userfunc() may write out some data that needs to be |
b5256303 TC |
1215 | * encrypted if the dataset is encrypted (specifically the root |
1216 | * directory). This data must be written out before the encryption | |
1217 | * key mapping is removed by dsl_dataset_rele_flags(). Force the | |
1218 | * I/O to occur immediately by invoking the relevant sections of | |
1219 | * dsl_pool_sync(). | |
1220 | */ | |
1221 | if (os->os_encrypted) { | |
1222 | dsl_dataset_t *tmpds = NULL; | |
1223 | boolean_t need_sync_done = B_FALSE; | |
1224 | ||
4807c0ba TC |
1225 | mutex_enter(&ds->ds_lock); |
1226 | ds->ds_owner = FTAG; | |
1227 | mutex_exit(&ds->ds_lock); | |
1228 | ||
52ce99dd | 1229 | rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
4807c0ba TC |
1230 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, |
1231 | tx->tx_txg); | |
b5256303 | 1232 | if (tmpds != NULL) { |
b5256303 TC |
1233 | dsl_dataset_sync(ds, rzio, tx); |
1234 | need_sync_done = B_TRUE; | |
1235 | } | |
1236 | VERIFY0(zio_wait(rzio)); | |
1237 | ||
ba67d821 | 1238 | dmu_objset_sync_done(os, tx); |
b5256303 | 1239 | taskq_wait(dp->dp_sync_taskq); |
52ce99dd TC |
1240 | if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { |
1241 | ASSERT3P(ds->ds_key_mapping, !=, NULL); | |
1242 | key_mapping_rele(spa, ds->ds_key_mapping, ds); | |
1243 | } | |
b5256303 | 1244 | |
52ce99dd | 1245 | rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
4807c0ba TC |
1246 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, |
1247 | tx->tx_txg); | |
b5256303 | 1248 | if (tmpds != NULL) { |
b5256303 TC |
1249 | dmu_buf_rele(ds->ds_dbuf, ds); |
1250 | dsl_dataset_sync(ds, rzio, tx); | |
1251 | } | |
1252 | VERIFY0(zio_wait(rzio)); | |
1253 | ||
52ce99dd TC |
1254 | if (need_sync_done) { |
1255 | ASSERT3P(ds->ds_key_mapping, !=, NULL); | |
1256 | key_mapping_rele(spa, ds->ds_key_mapping, ds); | |
b5256303 | 1257 | dsl_dataset_sync_done(ds, tx); |
52ce99dd | 1258 | } |
4807c0ba TC |
1259 | |
1260 | mutex_enter(&ds->ds_lock); | |
1261 | ds->ds_owner = NULL; | |
1262 | mutex_exit(&ds->ds_lock); | |
b5256303 TC |
1263 | } |
1264 | ||
74756182 | 1265 | spa_history_log_internal_ds(ds, "create", tx, " "); |
a0bd735a | 1266 | |
b5256303 | 1267 | dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); |
13fe0198 | 1268 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f BB |
1269 | } |
1270 | ||
1271 | int | |
428870ff | 1272 | dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, |
b5256303 | 1273 | dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) |
34dc7c2f | 1274 | { |
13fe0198 | 1275 | dmu_objset_create_arg_t doca; |
b5256303 | 1276 | dsl_crypto_params_t tmp_dcp = { 0 }; |
34dc7c2f | 1277 | |
13fe0198 MA |
1278 | doca.doca_name = name; |
1279 | doca.doca_cred = CRED(); | |
e59a377a | 1280 | doca.doca_proc = curproc; |
13fe0198 MA |
1281 | doca.doca_flags = flags; |
1282 | doca.doca_userfunc = func; | |
1283 | doca.doca_userarg = arg; | |
1284 | doca.doca_type = type; | |
34dc7c2f | 1285 | |
b5256303 TC |
1286 | /* |
1287 | * Some callers (mostly for testing) do not provide a dcp on their | |
1288 | * own but various code inside the sync task will require it to be | |
1289 | * allocated. Rather than adding NULL checks throughout this code | |
1290 | * or adding dummy dcp's to all of the callers we simply create a | |
1291 | * dummy one here and use that. This zero dcp will have the same | |
85ce3f4f | 1292 | * effect as asking for inheritance of all encryption params. |
b5256303 TC |
1293 | */ |
1294 | doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; | |
1295 | ||
ec213971 | 1296 | int rv = dsl_sync_task(name, |
3d45fdd6 | 1297 | dmu_objset_create_check, dmu_objset_create_sync, &doca, |
ec213971 MA |
1298 | 6, ZFS_SPACE_CHECK_NORMAL); |
1299 | ||
1300 | if (rv == 0) | |
1301 | zvol_create_minor(name); | |
1302 | return (rv); | |
34dc7c2f BB |
1303 | } |
1304 | ||
13fe0198 MA |
1305 | typedef struct dmu_objset_clone_arg { |
1306 | const char *doca_clone; | |
1307 | const char *doca_origin; | |
1308 | cred_t *doca_cred; | |
e59a377a | 1309 | proc_t *doca_proc; |
13fe0198 MA |
1310 | } dmu_objset_clone_arg_t; |
1311 | ||
1312 | /*ARGSUSED*/ | |
1313 | static int | |
1314 | dmu_objset_clone_check(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1315 | { |
13fe0198 | 1316 | dmu_objset_clone_arg_t *doca = arg; |
428870ff BB |
1317 | dsl_dir_t *pdd; |
1318 | const char *tail; | |
13fe0198 MA |
1319 | int error; |
1320 | dsl_dataset_t *origin; | |
1321 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
34dc7c2f | 1322 | |
13fe0198 | 1323 | if (strchr(doca->doca_clone, '@') != NULL) |
2e528b49 | 1324 | return (SET_ERROR(EINVAL)); |
13fe0198 | 1325 | |
eca7b760 IK |
1326 | if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) |
1327 | return (SET_ERROR(ENAMETOOLONG)); | |
1328 | ||
13fe0198 MA |
1329 | error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); |
1330 | if (error != 0) | |
1331 | return (error); | |
428870ff | 1332 | if (tail == NULL) { |
13fe0198 | 1333 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1334 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1335 | } |
1cddb8c9 | 1336 | |
788eb90c | 1337 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
e59a377a | 1338 | doca->doca_cred, doca->doca_proc); |
788eb90c JJ |
1339 | if (error != 0) { |
1340 | dsl_dir_rele(pdd, FTAG); | |
1341 | return (SET_ERROR(EDQUOT)); | |
1342 | } | |
428870ff | 1343 | |
13fe0198 | 1344 | error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); |
b5256303 TC |
1345 | if (error != 0) { |
1346 | dsl_dir_rele(pdd, FTAG); | |
572e2857 | 1347 | return (error); |
b5256303 | 1348 | } |
572e2857 | 1349 | |
13fe0198 | 1350 | /* You can only clone snapshots, not the head datasets. */ |
0c66c32d | 1351 | if (!origin->ds_is_snapshot) { |
13fe0198 | 1352 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1353 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1354 | return (SET_ERROR(EINVAL)); |
428870ff | 1355 | } |
b5256303 | 1356 | |
13fe0198 | 1357 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1358 | dsl_dir_rele(pdd, FTAG); |
572e2857 | 1359 | |
13fe0198 | 1360 | return (0); |
9babb374 | 1361 | } |
34dc7c2f | 1362 | |
13fe0198 MA |
1363 | static void |
1364 | dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1365 | { |
13fe0198 MA |
1366 | dmu_objset_clone_arg_t *doca = arg; |
1367 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1368 | dsl_dir_t *pdd; | |
1369 | const char *tail; | |
1370 | dsl_dataset_t *origin, *ds; | |
1371 | uint64_t obj; | |
eca7b760 | 1372 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; |
6f1ffb06 | 1373 | |
13fe0198 MA |
1374 | VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); |
1375 | VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); | |
6f1ffb06 | 1376 | |
13fe0198 | 1377 | obj = dsl_dataset_create_sync(pdd, tail, origin, 0, |
b5256303 | 1378 | doca->doca_cred, NULL, tx); |
34dc7c2f | 1379 | |
13fe0198 MA |
1380 | VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); |
1381 | dsl_dataset_name(origin, namebuf); | |
1382 | spa_history_log_internal_ds(ds, "clone", tx, | |
74756182 | 1383 | "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object); |
13fe0198 MA |
1384 | dsl_dataset_rele(ds, FTAG); |
1385 | dsl_dataset_rele(origin, FTAG); | |
1386 | dsl_dir_rele(pdd, FTAG); | |
34dc7c2f BB |
1387 | } |
1388 | ||
1389 | int | |
13fe0198 | 1390 | dmu_objset_clone(const char *clone, const char *origin) |
34dc7c2f | 1391 | { |
13fe0198 | 1392 | dmu_objset_clone_arg_t doca; |
34dc7c2f | 1393 | |
13fe0198 MA |
1394 | doca.doca_clone = clone; |
1395 | doca.doca_origin = origin; | |
1396 | doca.doca_cred = CRED(); | |
e59a377a | 1397 | doca.doca_proc = curproc; |
572e2857 | 1398 | |
ec213971 | 1399 | int rv = dsl_sync_task(clone, |
3d45fdd6 | 1400 | dmu_objset_clone_check, dmu_objset_clone_sync, &doca, |
ec213971 MA |
1401 | 6, ZFS_SPACE_CHECK_NORMAL); |
1402 | ||
1403 | if (rv == 0) | |
1404 | zvol_create_minor(clone); | |
1405 | ||
1406 | return (rv); | |
6f1ffb06 MA |
1407 | } |
1408 | ||
1409 | int | |
1410 | dmu_objset_snapshot_one(const char *fsname, const char *snapname) | |
1411 | { | |
1412 | int err; | |
1413 | char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); | |
1414 | nvlist_t *snaps = fnvlist_alloc(); | |
1415 | ||
1416 | fnvlist_add_boolean(snaps, longsnap); | |
e4f5fa12 | 1417 | kmem_strfree(longsnap); |
13fe0198 MA |
1418 | err = dsl_dataset_snapshot(snaps, NULL, NULL); |
1419 | fnvlist_free(snaps); | |
6f1ffb06 MA |
1420 | return (err); |
1421 | } | |
1422 | ||
1de321e6 JX |
1423 | static void |
1424 | dmu_objset_upgrade_task_cb(void *data) | |
1425 | { | |
1426 | objset_t *os = data; | |
1427 | ||
1428 | mutex_enter(&os->os_upgrade_lock); | |
1429 | os->os_upgrade_status = EINTR; | |
1430 | if (!os->os_upgrade_exit) { | |
39372fa2 AF |
1431 | int status; |
1432 | ||
1de321e6 JX |
1433 | mutex_exit(&os->os_upgrade_lock); |
1434 | ||
39372fa2 AF |
1435 | status = os->os_upgrade_cb(os); |
1436 | ||
1de321e6 | 1437 | mutex_enter(&os->os_upgrade_lock); |
39372fa2 AF |
1438 | |
1439 | os->os_upgrade_status = status; | |
1de321e6 JX |
1440 | } |
1441 | os->os_upgrade_exit = B_TRUE; | |
1442 | os->os_upgrade_id = 0; | |
1443 | mutex_exit(&os->os_upgrade_lock); | |
c0daec32 | 1444 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); |
1de321e6 JX |
1445 | } |
1446 | ||
1447 | static void | |
1448 | dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) | |
1449 | { | |
1450 | if (os->os_upgrade_id != 0) | |
1451 | return; | |
1452 | ||
c0daec32 AB |
1453 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
1454 | dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); | |
1455 | ||
1de321e6 JX |
1456 | mutex_enter(&os->os_upgrade_lock); |
1457 | if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { | |
1458 | os->os_upgrade_exit = B_FALSE; | |
1459 | os->os_upgrade_cb = cb; | |
1460 | os->os_upgrade_id = taskq_dispatch( | |
1461 | os->os_spa->spa_upgrade_taskq, | |
1462 | dmu_objset_upgrade_task_cb, os, TQ_SLEEP); | |
c0daec32 AB |
1463 | if (os->os_upgrade_id == TASKQID_INVALID) { |
1464 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1de321e6 | 1465 | os->os_upgrade_status = ENOMEM; |
c0daec32 | 1466 | } |
39372fa2 AF |
1467 | } else { |
1468 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1de321e6 JX |
1469 | } |
1470 | mutex_exit(&os->os_upgrade_lock); | |
1471 | } | |
1472 | ||
1473 | static void | |
1474 | dmu_objset_upgrade_stop(objset_t *os) | |
1475 | { | |
1476 | mutex_enter(&os->os_upgrade_lock); | |
1477 | os->os_upgrade_exit = B_TRUE; | |
1478 | if (os->os_upgrade_id != 0) { | |
1479 | taskqid_t id = os->os_upgrade_id; | |
1480 | ||
1481 | os->os_upgrade_id = 0; | |
1482 | mutex_exit(&os->os_upgrade_lock); | |
1483 | ||
c0daec32 AB |
1484 | if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { |
1485 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1486 | } | |
4807c0ba | 1487 | txg_wait_synced(os->os_spa->spa_dsl_pool, 0); |
1de321e6 JX |
1488 | } else { |
1489 | mutex_exit(&os->os_upgrade_lock); | |
1490 | } | |
1491 | } | |
1492 | ||
34dc7c2f | 1493 | static void |
64fc7762 | 1494 | dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) |
34dc7c2f BB |
1495 | { |
1496 | dnode_t *dn; | |
1497 | ||
64fc7762 | 1498 | while ((dn = multilist_sublist_head(list)) != NULL) { |
34dc7c2f BB |
1499 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); |
1500 | ASSERT(dn->dn_dbuf->db_data_pending); | |
1501 | /* | |
9babb374 | 1502 | * Initialize dn_zio outside dnode_sync() because the |
93e28d66 | 1503 | * meta-dnode needs to set it outside dnode_sync(). |
34dc7c2f BB |
1504 | */ |
1505 | dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; | |
1506 | ASSERT(dn->dn_zio); | |
1507 | ||
1508 | ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); | |
64fc7762 | 1509 | multilist_sublist_remove(list, dn); |
9babb374 | 1510 | |
edc1e713 | 1511 | /* |
ba67d821 MA |
1512 | * See the comment above dnode_rele_task() for an explanation |
1513 | * of why this dnode hold is always needed (even when not | |
1514 | * doing user accounting). | |
edc1e713 | 1515 | */ |
64fc7762 | 1516 | multilist_t *newlist = dn->dn_objset->os_synced_dnodes; |
ba67d821 MA |
1517 | (void) dnode_add_ref(dn, newlist); |
1518 | multilist_insert(newlist, dn); | |
9babb374 | 1519 | |
34dc7c2f BB |
1520 | dnode_sync(dn, tx); |
1521 | } | |
1522 | } | |
1523 | ||
1524 | /* ARGSUSED */ | |
1525 | static void | |
428870ff | 1526 | dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) |
34dc7c2f | 1527 | { |
b128c09f | 1528 | blkptr_t *bp = zio->io_bp; |
428870ff | 1529 | objset_t *os = arg; |
34dc7c2f | 1530 | dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; |
b5256303 | 1531 | uint64_t fill = 0; |
34dc7c2f | 1532 | |
9b67f605 | 1533 | ASSERT(!BP_IS_EMBEDDED(bp)); |
13fe0198 MA |
1534 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); |
1535 | ASSERT0(BP_GET_LEVEL(bp)); | |
34dc7c2f BB |
1536 | |
1537 | /* | |
9babb374 BB |
1538 | * Update rootbp fill count: it should be the number of objects |
1539 | * allocated in the object set (not counting the "special" | |
1540 | * objects that are stored in the objset_phys_t -- the meta | |
9c5167d1 | 1541 | * dnode and user/group/project accounting objects). |
34dc7c2f | 1542 | */ |
1c27024e | 1543 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
b5256303 TC |
1544 | fill += BP_GET_FILL(&dnp->dn_blkptr[i]); |
1545 | ||
1546 | BP_SET_FILL(bp, fill); | |
1547 | ||
cc9bb3e5 GM |
1548 | if (os->os_dsl_dataset != NULL) |
1549 | rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); | |
1550 | *os->os_rootbp = *bp; | |
1551 | if (os->os_dsl_dataset != NULL) | |
1552 | rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); | |
428870ff BB |
1553 | } |
1554 | ||
1555 | /* ARGSUSED */ | |
1556 | static void | |
1557 | dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) | |
1558 | { | |
1559 | blkptr_t *bp = zio->io_bp; | |
1560 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1561 | objset_t *os = arg; | |
34dc7c2f | 1562 | |
b128c09f | 1563 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
428870ff | 1564 | ASSERT(BP_EQUAL(bp, bp_orig)); |
b128c09f | 1565 | } else { |
428870ff BB |
1566 | dsl_dataset_t *ds = os->os_dsl_dataset; |
1567 | dmu_tx_t *tx = os->os_synctx; | |
1568 | ||
1569 | (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); | |
1570 | dsl_dataset_block_born(ds, bp, tx); | |
34dc7c2f | 1571 | } |
cc9bb3e5 | 1572 | kmem_free(bp, sizeof (*bp)); |
34dc7c2f BB |
1573 | } |
1574 | ||
64fc7762 MA |
1575 | typedef struct sync_dnodes_arg { |
1576 | multilist_t *sda_list; | |
1577 | int sda_sublist_idx; | |
1578 | multilist_t *sda_newlist; | |
1579 | dmu_tx_t *sda_tx; | |
1580 | } sync_dnodes_arg_t; | |
1581 | ||
1582 | static void | |
1583 | sync_dnodes_task(void *arg) | |
1584 | { | |
1585 | sync_dnodes_arg_t *sda = arg; | |
1586 | ||
1587 | multilist_sublist_t *ms = | |
1588 | multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); | |
1589 | ||
1590 | dmu_objset_sync_dnodes(ms, sda->sda_tx); | |
1591 | ||
1592 | multilist_sublist_unlock(ms); | |
1593 | ||
1594 | kmem_free(sda, sizeof (*sda)); | |
1595 | } | |
1596 | ||
1597 | ||
34dc7c2f BB |
1598 | /* called from dsl */ |
1599 | void | |
428870ff | 1600 | dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) |
34dc7c2f BB |
1601 | { |
1602 | int txgoff; | |
5dbd68a3 | 1603 | zbookmark_phys_t zb; |
428870ff | 1604 | zio_prop_t zp; |
34dc7c2f BB |
1605 | zio_t *zio; |
1606 | list_t *list; | |
1607 | dbuf_dirty_record_t *dr; | |
fc754677 AM |
1608 | int num_sublists; |
1609 | multilist_t *ml; | |
cc9bb3e5 GM |
1610 | blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); |
1611 | *blkptr_copy = *os->os_rootbp; | |
34dc7c2f BB |
1612 | |
1613 | dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); | |
1614 | ||
1615 | ASSERT(dmu_tx_is_syncing(tx)); | |
1616 | /* XXX the write_done callback should really give us the tx... */ | |
1617 | os->os_synctx = tx; | |
1618 | ||
1619 | if (os->os_dsl_dataset == NULL) { | |
1620 | /* | |
1621 | * This is the MOS. If we have upgraded, | |
1622 | * spa_max_replication() could change, so reset | |
1623 | * os_copies here. | |
1624 | */ | |
1625 | os->os_copies = spa_max_replication(os->os_spa); | |
1626 | } | |
1627 | ||
1628 | /* | |
1629 | * Create the root block IO | |
1630 | */ | |
428870ff BB |
1631 | SET_BOOKMARK(&zb, os->os_dsl_dataset ? |
1632 | os->os_dsl_dataset->ds_object : DMU_META_OBJSET, | |
1633 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
294f6806 | 1634 | arc_release(os->os_phys_buf, &os->os_phys_buf); |
b128c09f | 1635 | |
82644107 | 1636 | dmu_write_policy(os, NULL, 0, 0, &zp); |
9babb374 | 1637 | |
b5256303 | 1638 | /* |
0c03d21a MA |
1639 | * If we are either claiming the ZIL or doing a raw receive, write |
1640 | * out the os_phys_buf raw. Neither of these actions will effect the | |
1641 | * MAC at this point. | |
b5256303 | 1642 | */ |
0c03d21a MA |
1643 | if (os->os_raw_receive || |
1644 | os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { | |
b5256303 | 1645 | ASSERT(os->os_encrypted); |
b5256303 TC |
1646 | arc_convert_to_raw(os->os_phys_buf, |
1647 | os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, | |
1648 | DMU_OT_OBJSET, NULL, NULL, NULL); | |
1649 | } | |
1650 | ||
428870ff | 1651 | zio = arc_write(pio, os->os_spa, tx->tx_txg, |
cc9bb3e5 | 1652 | blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), |
bc77ba73 PD |
1653 | &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, |
1654 | os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); | |
34dc7c2f BB |
1655 | |
1656 | /* | |
9babb374 | 1657 | * Sync special dnodes - the parent IO for the sync is the root block |
34dc7c2f | 1658 | */ |
572e2857 BB |
1659 | DMU_META_DNODE(os)->dn_zio = zio; |
1660 | dnode_sync(DMU_META_DNODE(os), tx); | |
34dc7c2f | 1661 | |
9babb374 BB |
1662 | os->os_phys->os_flags = os->os_flags; |
1663 | ||
572e2857 BB |
1664 | if (DMU_USERUSED_DNODE(os) && |
1665 | DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1666 | DMU_USERUSED_DNODE(os)->dn_zio = zio; | |
1667 | dnode_sync(DMU_USERUSED_DNODE(os), tx); | |
1668 | DMU_GROUPUSED_DNODE(os)->dn_zio = zio; | |
1669 | dnode_sync(DMU_GROUPUSED_DNODE(os), tx); | |
9babb374 BB |
1670 | } |
1671 | ||
9c5167d1 NF |
1672 | if (DMU_PROJECTUSED_DNODE(os) && |
1673 | DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1674 | DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; | |
1675 | dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); | |
1676 | } | |
1677 | ||
34dc7c2f BB |
1678 | txgoff = tx->tx_txg & TXG_MASK; |
1679 | ||
ba67d821 MA |
1680 | /* |
1681 | * We must create the list here because it uses the | |
1682 | * dn_dirty_link[] of this txg. But it may already | |
1683 | * exist because we call dsl_dataset_sync() twice per txg. | |
1684 | */ | |
1685 | if (os->os_synced_dnodes == NULL) { | |
1686 | os->os_synced_dnodes = | |
1687 | multilist_create(sizeof (dnode_t), | |
1688 | offsetof(dnode_t, dn_dirty_link[txgoff]), | |
1689 | dnode_multilist_index_func); | |
1690 | } else { | |
1691 | ASSERT3U(os->os_synced_dnodes->ml_offset, ==, | |
1692 | offsetof(dnode_t, dn_dirty_link[txgoff])); | |
9babb374 BB |
1693 | } |
1694 | ||
fc754677 AM |
1695 | ml = os->os_dirty_dnodes[txgoff]; |
1696 | num_sublists = multilist_get_num_sublists(ml); | |
1697 | for (int i = 0; i < num_sublists; i++) { | |
1698 | if (multilist_sublist_is_empty_idx(ml, i)) | |
1699 | continue; | |
64fc7762 | 1700 | sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); |
fc754677 | 1701 | sda->sda_list = ml; |
64fc7762 MA |
1702 | sda->sda_sublist_idx = i; |
1703 | sda->sda_tx = tx; | |
1704 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1705 | sync_dnodes_task, sda, 0); | |
1706 | /* callback frees sda */ | |
1707 | } | |
1708 | taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); | |
34dc7c2f | 1709 | |
572e2857 | 1710 | list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; |
64fc7762 | 1711 | while ((dr = list_head(list)) != NULL) { |
13fe0198 | 1712 | ASSERT0(dr->dr_dbuf->db_level); |
34dc7c2f | 1713 | list_remove(list, dr); |
9cdf7b1f | 1714 | zio_nowait(dr->dr_zio); |
34dc7c2f | 1715 | } |
68cbd56e NB |
1716 | |
1717 | /* Enable dnode backfill if enough objects have been freed. */ | |
1718 | if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { | |
1719 | os->os_rescan_dnodes = B_TRUE; | |
1720 | os->os_freed_dnodes = 0; | |
1721 | } | |
1722 | ||
34dc7c2f BB |
1723 | /* |
1724 | * Free intent log blocks up to this tx. | |
1725 | */ | |
1726 | zil_sync(os->os_zil, tx); | |
b128c09f | 1727 | os->os_phys->os_zil_header = os->os_zil_header; |
34dc7c2f BB |
1728 | zio_nowait(zio); |
1729 | } | |
1730 | ||
428870ff BB |
1731 | boolean_t |
1732 | dmu_objset_is_dirty(objset_t *os, uint64_t txg) | |
1733 | { | |
64fc7762 | 1734 | return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); |
428870ff BB |
1735 | } |
1736 | ||
7bcb7f08 | 1737 | static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES]; |
9babb374 BB |
1738 | |
1739 | void | |
7bcb7f08 | 1740 | dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb) |
9babb374 | 1741 | { |
7bcb7f08 MA |
1742 | file_cbs[ost] = cb; |
1743 | } | |
1744 | ||
1745 | int | |
1746 | dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data, | |
1747 | zfs_file_info_t *zfi) | |
1748 | { | |
1749 | file_info_cb_t *cb = file_cbs[os->os_phys->os_type]; | |
1750 | if (cb == NULL) | |
1751 | return (EINVAL); | |
1752 | return (cb(bonustype, data, zfi)); | |
9babb374 BB |
1753 | } |
1754 | ||
1755 | boolean_t | |
428870ff | 1756 | dmu_objset_userused_enabled(objset_t *os) |
9babb374 BB |
1757 | { |
1758 | return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && | |
7bcb7f08 | 1759 | file_cbs[os->os_phys->os_type] != NULL && |
572e2857 | 1760 | DMU_USERUSED_DNODE(os) != NULL); |
9babb374 BB |
1761 | } |
1762 | ||
1de321e6 JX |
1763 | boolean_t |
1764 | dmu_objset_userobjused_enabled(objset_t *os) | |
1765 | { | |
1766 | return (dmu_objset_userused_enabled(os) && | |
1767 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); | |
1768 | } | |
1769 | ||
9c5167d1 NF |
1770 | boolean_t |
1771 | dmu_objset_projectquota_enabled(objset_t *os) | |
1772 | { | |
7bcb7f08 | 1773 | return (file_cbs[os->os_phys->os_type] != NULL && |
9c5167d1 NF |
1774 | DMU_PROJECTUSED_DNODE(os) != NULL && |
1775 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); | |
1776 | } | |
1777 | ||
9b7a83cb JX |
1778 | typedef struct userquota_node { |
1779 | /* must be in the first filed, see userquota_update_cache() */ | |
1780 | char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1781 | int64_t uqn_delta; | |
1782 | avl_node_t uqn_node; | |
1783 | } userquota_node_t; | |
1784 | ||
1785 | typedef struct userquota_cache { | |
1786 | avl_tree_t uqc_user_deltas; | |
1787 | avl_tree_t uqc_group_deltas; | |
9c5167d1 | 1788 | avl_tree_t uqc_project_deltas; |
9b7a83cb JX |
1789 | } userquota_cache_t; |
1790 | ||
1791 | static int | |
1792 | userquota_compare(const void *l, const void *r) | |
1793 | { | |
1794 | const userquota_node_t *luqn = l; | |
1795 | const userquota_node_t *ruqn = r; | |
e4ffa98d | 1796 | int rv; |
9b7a83cb JX |
1797 | |
1798 | /* | |
1799 | * NB: can only access uqn_id because userquota_update_cache() doesn't | |
1800 | * pass in an entire userquota_node_t. | |
1801 | */ | |
e4ffa98d BB |
1802 | rv = strcmp(luqn->uqn_id, ruqn->uqn_id); |
1803 | ||
ca577779 | 1804 | return (TREE_ISIGN(rv)); |
9b7a83cb JX |
1805 | } |
1806 | ||
1807 | static void | |
1808 | do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) | |
1809 | { | |
1810 | void *cookie; | |
1811 | userquota_node_t *uqn; | |
1812 | ||
1813 | ASSERT(dmu_tx_is_syncing(tx)); | |
1814 | ||
1815 | cookie = NULL; | |
1816 | while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, | |
1817 | &cookie)) != NULL) { | |
64fc7762 MA |
1818 | /* |
1819 | * os_userused_lock protects against concurrent calls to | |
1820 | * zap_increment_int(). It's needed because zap_increment_int() | |
1821 | * is not thread-safe (i.e. not atomic). | |
1822 | */ | |
1823 | mutex_enter(&os->os_userused_lock); | |
9b7a83cb JX |
1824 | VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, |
1825 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1826 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1827 | kmem_free(uqn, sizeof (*uqn)); |
1828 | } | |
1829 | avl_destroy(&cache->uqc_user_deltas); | |
1830 | ||
1831 | cookie = NULL; | |
1832 | while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, | |
1833 | &cookie)) != NULL) { | |
64fc7762 | 1834 | mutex_enter(&os->os_userused_lock); |
9b7a83cb JX |
1835 | VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, |
1836 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1837 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1838 | kmem_free(uqn, sizeof (*uqn)); |
1839 | } | |
1840 | avl_destroy(&cache->uqc_group_deltas); | |
9c5167d1 NF |
1841 | |
1842 | if (dmu_objset_projectquota_enabled(os)) { | |
1843 | cookie = NULL; | |
1844 | while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, | |
1845 | &cookie)) != NULL) { | |
1846 | mutex_enter(&os->os_userused_lock); | |
1847 | VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, | |
1848 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1849 | mutex_exit(&os->os_userused_lock); | |
1850 | kmem_free(uqn, sizeof (*uqn)); | |
1851 | } | |
1852 | avl_destroy(&cache->uqc_project_deltas); | |
1853 | } | |
9b7a83cb JX |
1854 | } |
1855 | ||
1856 | static void | |
1857 | userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) | |
1858 | { | |
1859 | userquota_node_t *uqn; | |
1860 | avl_index_t idx; | |
1861 | ||
1862 | ASSERT(strlen(id) < sizeof (uqn->uqn_id)); | |
1863 | /* | |
1864 | * Use id directly for searching because uqn_id is the first field of | |
1865 | * userquota_node_t and fields after uqn_id won't be accessed in | |
1866 | * avl_find(). | |
1867 | */ | |
1868 | uqn = avl_find(avl, (const void *)id, &idx); | |
1869 | if (uqn == NULL) { | |
1870 | uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); | |
1f51b525 | 1871 | strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); |
9b7a83cb JX |
1872 | avl_insert(avl, uqn, idx); |
1873 | } | |
1874 | uqn->uqn_delta += delta; | |
1875 | } | |
1876 | ||
428870ff | 1877 | static void |
9c5167d1 NF |
1878 | do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, |
1879 | uint64_t flags, uint64_t user, uint64_t group, uint64_t project, | |
1880 | boolean_t subtract) | |
428870ff | 1881 | { |
9c5167d1 | 1882 | if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { |
50c957f7 | 1883 | int64_t delta = DNODE_MIN_SIZE + used; |
9b7a83cb JX |
1884 | char name[20]; |
1885 | ||
428870ff BB |
1886 | if (subtract) |
1887 | delta = -delta; | |
9b7a83cb | 1888 | |
c9e319fa | 1889 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user); |
9b7a83cb JX |
1890 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); |
1891 | ||
c9e319fa | 1892 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group); |
9b7a83cb | 1893 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); |
9c5167d1 NF |
1894 | |
1895 | if (dmu_objset_projectquota_enabled(os)) { | |
c9e319fa JL |
1896 | (void) snprintf(name, sizeof (name), "%llx", |
1897 | (longlong_t)project); | |
9c5167d1 NF |
1898 | userquota_update_cache(&cache->uqc_project_deltas, |
1899 | name, delta); | |
1900 | } | |
428870ff BB |
1901 | } |
1902 | } | |
1903 | ||
1de321e6 | 1904 | static void |
9c5167d1 NF |
1905 | do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, |
1906 | uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) | |
1de321e6 JX |
1907 | { |
1908 | if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { | |
1909 | char name[20 + DMU_OBJACCT_PREFIX_LEN]; | |
9b7a83cb | 1910 | int delta = subtract ? -1 : 1; |
1de321e6 JX |
1911 | |
1912 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1913 | (longlong_t)user); | |
9b7a83cb | 1914 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); |
1de321e6 JX |
1915 | |
1916 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1917 | (longlong_t)group); | |
9b7a83cb | 1918 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); |
9c5167d1 NF |
1919 | |
1920 | if (dmu_objset_projectquota_enabled(os)) { | |
1921 | (void) snprintf(name, sizeof (name), | |
1922 | DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); | |
1923 | userquota_update_cache(&cache->uqc_project_deltas, | |
1924 | name, delta); | |
1925 | } | |
1de321e6 JX |
1926 | } |
1927 | } | |
1928 | ||
64fc7762 MA |
1929 | typedef struct userquota_updates_arg { |
1930 | objset_t *uua_os; | |
1931 | int uua_sublist_idx; | |
1932 | dmu_tx_t *uua_tx; | |
1933 | } userquota_updates_arg_t; | |
1934 | ||
1935 | static void | |
1936 | userquota_updates_task(void *arg) | |
9babb374 | 1937 | { |
64fc7762 MA |
1938 | userquota_updates_arg_t *uua = arg; |
1939 | objset_t *os = uua->uua_os; | |
1940 | dmu_tx_t *tx = uua->uua_tx; | |
9babb374 | 1941 | dnode_t *dn; |
9b7a83cb | 1942 | userquota_cache_t cache = { { 0 } }; |
9babb374 | 1943 | |
64fc7762 MA |
1944 | multilist_sublist_t *list = |
1945 | multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); | |
9babb374 | 1946 | |
64fc7762 MA |
1947 | ASSERT(multilist_sublist_head(list) == NULL || |
1948 | dmu_objset_userused_enabled(os)); | |
9b7a83cb JX |
1949 | avl_create(&cache.uqc_user_deltas, userquota_compare, |
1950 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
1951 | avl_create(&cache.uqc_group_deltas, userquota_compare, | |
1952 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
9c5167d1 NF |
1953 | if (dmu_objset_projectquota_enabled(os)) |
1954 | avl_create(&cache.uqc_project_deltas, userquota_compare, | |
1955 | sizeof (userquota_node_t), offsetof(userquota_node_t, | |
1956 | uqn_node)); | |
9b7a83cb | 1957 | |
64fc7762 | 1958 | while ((dn = multilist_sublist_head(list)) != NULL) { |
572e2857 | 1959 | int flags; |
9babb374 | 1960 | ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); |
9babb374 BB |
1961 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || |
1962 | dn->dn_phys->dn_flags & | |
1963 | DNODE_FLAG_USERUSED_ACCOUNTED); | |
1964 | ||
572e2857 BB |
1965 | flags = dn->dn_id_flags; |
1966 | ASSERT(flags); | |
1967 | if (flags & DN_ID_OLD_EXIST) { | |
9c5167d1 NF |
1968 | do_userquota_update(os, &cache, dn->dn_oldused, |
1969 | dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, | |
1970 | dn->dn_oldprojid, B_TRUE); | |
1971 | do_userobjquota_update(os, &cache, dn->dn_oldflags, | |
1972 | dn->dn_olduid, dn->dn_oldgid, | |
1973 | dn->dn_oldprojid, B_TRUE); | |
428870ff | 1974 | } |
572e2857 | 1975 | if (flags & DN_ID_NEW_EXIST) { |
9c5167d1 | 1976 | do_userquota_update(os, &cache, |
9b7a83cb | 1977 | DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, |
9c5167d1 NF |
1978 | dn->dn_newuid, dn->dn_newgid, |
1979 | dn->dn_newprojid, B_FALSE); | |
1980 | do_userobjquota_update(os, &cache, | |
1981 | dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, | |
1982 | dn->dn_newprojid, B_FALSE); | |
428870ff BB |
1983 | } |
1984 | ||
572e2857 | 1985 | mutex_enter(&dn->dn_mtx); |
428870ff BB |
1986 | dn->dn_oldused = 0; |
1987 | dn->dn_oldflags = 0; | |
1988 | if (dn->dn_id_flags & DN_ID_NEW_EXIST) { | |
1989 | dn->dn_olduid = dn->dn_newuid; | |
1990 | dn->dn_oldgid = dn->dn_newgid; | |
9c5167d1 | 1991 | dn->dn_oldprojid = dn->dn_newprojid; |
428870ff BB |
1992 | dn->dn_id_flags |= DN_ID_OLD_EXIST; |
1993 | if (dn->dn_bonuslen == 0) | |
1994 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
1995 | else | |
1996 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
1997 | } | |
1998 | dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); | |
9babb374 BB |
1999 | mutex_exit(&dn->dn_mtx); |
2000 | ||
64fc7762 MA |
2001 | multilist_sublist_remove(list, dn); |
2002 | dnode_rele(dn, os->os_synced_dnodes); | |
9babb374 | 2003 | } |
9b7a83cb | 2004 | do_userquota_cacheflush(os, &cache, tx); |
64fc7762 MA |
2005 | multilist_sublist_unlock(list); |
2006 | kmem_free(uua, sizeof (*uua)); | |
2007 | } | |
2008 | ||
ba67d821 MA |
2009 | /* |
2010 | * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being | |
2011 | * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be | |
2012 | * evicted because the block containing the dnode can't be evicted until it is | |
2013 | * written out. However, this hold is necessary to prevent the dnode_t from | |
2014 | * being moved (via dnode_move()) while it's still referenced by | |
2015 | * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for | |
2016 | * dirty_lightweight_leaf-type dirty records. | |
2017 | * | |
2018 | * If we are doing user-object accounting, the dnode_rele() happens from | |
2019 | * userquota_updates_task() instead. | |
2020 | */ | |
2021 | static void | |
2022 | dnode_rele_task(void *arg) | |
64fc7762 | 2023 | { |
ba67d821 MA |
2024 | userquota_updates_arg_t *uua = arg; |
2025 | objset_t *os = uua->uua_os; | |
2026 | ||
2027 | multilist_sublist_t *list = | |
2028 | multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); | |
fc754677 | 2029 | |
ba67d821 MA |
2030 | dnode_t *dn; |
2031 | while ((dn = multilist_sublist_head(list)) != NULL) { | |
2032 | multilist_sublist_remove(list, dn); | |
2033 | dnode_rele(dn, os->os_synced_dnodes); | |
2034 | } | |
2035 | multilist_sublist_unlock(list); | |
2036 | kmem_free(uua, sizeof (*uua)); | |
2037 | } | |
2038 | ||
2039 | /* | |
2040 | * Return TRUE if userquota updates are needed. | |
2041 | */ | |
2042 | static boolean_t | |
2043 | dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx) | |
2044 | { | |
64fc7762 | 2045 | if (!dmu_objset_userused_enabled(os)) |
ba67d821 | 2046 | return (B_FALSE); |
64fc7762 | 2047 | |
163a8c28 TC |
2048 | /* |
2049 | * If this is a raw receive just return and handle accounting | |
2050 | * later when we have the keys loaded. We also don't do user | |
2051 | * accounting during claiming since the datasets are not owned | |
2052 | * for the duration of claiming and this txg should only be | |
2053 | * used for recovery. | |
2054 | */ | |
b5256303 | 2055 | if (os->os_encrypted && dmu_objset_is_receiving(os)) |
ba67d821 | 2056 | return (B_FALSE); |
b5256303 | 2057 | |
163a8c28 | 2058 | if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) |
ba67d821 | 2059 | return (B_FALSE); |
163a8c28 | 2060 | |
9c5167d1 | 2061 | /* Allocate the user/group/project used objects if necessary. */ |
64fc7762 MA |
2062 | if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { |
2063 | VERIFY0(zap_create_claim(os, | |
2064 | DMU_USERUSED_OBJECT, | |
2065 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2066 | VERIFY0(zap_create_claim(os, | |
2067 | DMU_GROUPUSED_OBJECT, | |
2068 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2069 | } | |
2070 | ||
9c5167d1 NF |
2071 | if (dmu_objset_projectquota_enabled(os) && |
2072 | DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { | |
2073 | VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, | |
2074 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2075 | } | |
ba67d821 MA |
2076 | return (B_TRUE); |
2077 | } | |
9c5167d1 | 2078 | |
ba67d821 MA |
2079 | /* |
2080 | * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and | |
2081 | * also release the holds on the dnodes from dmu_objset_sync_dnodes(). | |
2082 | * The caller must taskq_wait(dp_sync_taskq). | |
2083 | */ | |
2084 | void | |
2085 | dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx) | |
2086 | { | |
2087 | boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx); | |
2088 | ||
2089 | int num_sublists = multilist_get_num_sublists(os->os_synced_dnodes); | |
fc754677 | 2090 | for (int i = 0; i < num_sublists; i++) { |
64fc7762 MA |
2091 | userquota_updates_arg_t *uua = |
2092 | kmem_alloc(sizeof (*uua), KM_SLEEP); | |
2093 | uua->uua_os = os; | |
2094 | uua->uua_sublist_idx = i; | |
2095 | uua->uua_tx = tx; | |
ba67d821 MA |
2096 | |
2097 | /* | |
2098 | * If we don't need to update userquotas, use | |
2099 | * dnode_rele_task() to call dnode_rele() | |
2100 | */ | |
64fc7762 | 2101 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, |
ba67d821 MA |
2102 | need_userquota ? userquota_updates_task : dnode_rele_task, |
2103 | uua, 0); | |
64fc7762 MA |
2104 | /* callback frees uua */ |
2105 | } | |
9babb374 BB |
2106 | } |
2107 | ||
ba67d821 | 2108 | |
428870ff BB |
2109 | /* |
2110 | * Returns a pointer to data to find uid/gid from | |
2111 | * | |
2112 | * If a dirty record for transaction group that is syncing can't | |
2113 | * be found then NULL is returned. In the NULL case it is assumed | |
2114 | * the uid/gid aren't changing. | |
2115 | */ | |
2116 | static void * | |
2117 | dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) | |
2118 | { | |
cccbed9f | 2119 | dbuf_dirty_record_t *dr; |
428870ff BB |
2120 | void *data; |
2121 | ||
2122 | if (db->db_dirtycnt == 0) | |
2123 | return (db->db.db_data); /* Nothing is changing */ | |
2124 | ||
cccbed9f | 2125 | dr = dbuf_find_dirty_eq(db, tx->tx_txg); |
428870ff | 2126 | |
572e2857 | 2127 | if (dr == NULL) { |
428870ff | 2128 | data = NULL; |
572e2857 | 2129 | } else { |
ba67d821 | 2130 | if (dr->dr_dnode->dn_bonuslen == 0 && |
572e2857 BB |
2131 | dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) |
2132 | data = dr->dt.dl.dr_data->b_data; | |
2133 | else | |
2134 | data = dr->dt.dl.dr_data; | |
572e2857 BB |
2135 | } |
2136 | ||
428870ff BB |
2137 | return (data); |
2138 | } | |
2139 | ||
2140 | void | |
2141 | dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) | |
2142 | { | |
2143 | objset_t *os = dn->dn_objset; | |
2144 | void *data = NULL; | |
2145 | dmu_buf_impl_t *db = NULL; | |
428870ff BB |
2146 | int flags = dn->dn_id_flags; |
2147 | int error; | |
2148 | boolean_t have_spill = B_FALSE; | |
2149 | ||
2150 | if (!dmu_objset_userused_enabled(dn->dn_objset)) | |
2151 | return; | |
2152 | ||
b5256303 TC |
2153 | /* |
2154 | * Raw receives introduce a problem with user accounting. Raw | |
2155 | * receives cannot update the user accounting info because the | |
2156 | * user ids and the sizes are encrypted. To guarantee that we | |
2157 | * never end up with bad user accounting, we simply disable it | |
2158 | * during raw receives. We also disable this for normal receives | |
2159 | * so that an incremental raw receive may be done on top of an | |
2160 | * existing non-raw receive. | |
2161 | */ | |
2162 | if (os->os_encrypted && dmu_objset_is_receiving(os)) | |
2163 | return; | |
2164 | ||
428870ff BB |
2165 | if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| |
2166 | DN_ID_CHKED_SPILL))) | |
2167 | return; | |
2168 | ||
2169 | if (before && dn->dn_bonuslen != 0) | |
2170 | data = DN_BONUS(dn->dn_phys); | |
2171 | else if (!before && dn->dn_bonuslen != 0) { | |
a3000f93 BB |
2172 | if (dn->dn_bonus) { |
2173 | db = dn->dn_bonus; | |
428870ff BB |
2174 | mutex_enter(&db->db_mtx); |
2175 | data = dmu_objset_userquota_find_data(db, tx); | |
2176 | } else { | |
2177 | data = DN_BONUS(dn->dn_phys); | |
2178 | } | |
2179 | } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { | |
2180 | int rf = 0; | |
2181 | ||
2182 | if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) | |
2183 | rf |= DB_RF_HAVESTRUCT; | |
572e2857 BB |
2184 | error = dmu_spill_hold_by_dnode(dn, |
2185 | rf | DB_RF_MUST_SUCCEED, | |
428870ff BB |
2186 | FTAG, (dmu_buf_t **)&db); |
2187 | ASSERT(error == 0); | |
2188 | mutex_enter(&db->db_mtx); | |
2189 | data = (before) ? db->db.db_data : | |
2190 | dmu_objset_userquota_find_data(db, tx); | |
2191 | have_spill = B_TRUE; | |
2192 | } else { | |
2193 | mutex_enter(&dn->dn_mtx); | |
2194 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2195 | mutex_exit(&dn->dn_mtx); | |
2196 | return; | |
2197 | } | |
2198 | ||
428870ff BB |
2199 | /* |
2200 | * Must always call the callback in case the object | |
2201 | * type has changed and that type isn't an object type to track | |
2202 | */ | |
7bcb7f08 MA |
2203 | zfs_file_info_t zfi; |
2204 | error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi); | |
2205 | ||
2206 | if (before) { | |
2207 | ASSERT(data); | |
2208 | dn->dn_olduid = zfi.zfi_user; | |
2209 | dn->dn_oldgid = zfi.zfi_group; | |
2210 | dn->dn_oldprojid = zfi.zfi_project; | |
2211 | } else if (data) { | |
2212 | dn->dn_newuid = zfi.zfi_user; | |
2213 | dn->dn_newgid = zfi.zfi_group; | |
2214 | dn->dn_newprojid = zfi.zfi_project; | |
2215 | } | |
428870ff BB |
2216 | |
2217 | /* | |
2218 | * Preserve existing uid/gid when the callback can't determine | |
2219 | * what the new uid/gid are and the callback returned EEXIST. | |
2220 | * The EEXIST error tells us to just use the existing uid/gid. | |
2221 | * If we don't know what the old values are then just assign | |
2222 | * them to 0, since that is a new file being created. | |
2223 | */ | |
2224 | if (!before && data == NULL && error == EEXIST) { | |
2225 | if (flags & DN_ID_OLD_EXIST) { | |
2226 | dn->dn_newuid = dn->dn_olduid; | |
2227 | dn->dn_newgid = dn->dn_oldgid; | |
2705ebf0 | 2228 | dn->dn_newprojid = dn->dn_oldprojid; |
428870ff BB |
2229 | } else { |
2230 | dn->dn_newuid = 0; | |
2231 | dn->dn_newgid = 0; | |
9c5167d1 | 2232 | dn->dn_newprojid = ZFS_DEFAULT_PROJID; |
428870ff BB |
2233 | } |
2234 | error = 0; | |
2235 | } | |
2236 | ||
2237 | if (db) | |
2238 | mutex_exit(&db->db_mtx); | |
2239 | ||
2240 | mutex_enter(&dn->dn_mtx); | |
2241 | if (error == 0 && before) | |
2242 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
2243 | if (error == 0 && !before) | |
2244 | dn->dn_id_flags |= DN_ID_NEW_EXIST; | |
2245 | ||
2246 | if (have_spill) { | |
2247 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
2248 | } else { | |
2249 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2250 | } | |
2251 | mutex_exit(&dn->dn_mtx); | |
a3000f93 | 2252 | if (have_spill) |
428870ff BB |
2253 | dmu_buf_rele((dmu_buf_t *)db, FTAG); |
2254 | } | |
2255 | ||
9babb374 BB |
2256 | boolean_t |
2257 | dmu_objset_userspace_present(objset_t *os) | |
2258 | { | |
428870ff | 2259 | return (os->os_phys->os_flags & |
9babb374 BB |
2260 | OBJSET_FLAG_USERACCOUNTING_COMPLETE); |
2261 | } | |
2262 | ||
1de321e6 JX |
2263 | boolean_t |
2264 | dmu_objset_userobjspace_present(objset_t *os) | |
2265 | { | |
2266 | return (os->os_phys->os_flags & | |
2267 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); | |
2268 | } | |
2269 | ||
9c5167d1 NF |
2270 | boolean_t |
2271 | dmu_objset_projectquota_present(objset_t *os) | |
2272 | { | |
2273 | return (os->os_phys->os_flags & | |
2274 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE); | |
2275 | } | |
2276 | ||
1de321e6 JX |
2277 | static int |
2278 | dmu_objset_space_upgrade(objset_t *os) | |
9babb374 BB |
2279 | { |
2280 | uint64_t obj; | |
2281 | int err = 0; | |
2282 | ||
9babb374 BB |
2283 | /* |
2284 | * We simply need to mark every object dirty, so that it will be | |
2285 | * synced out and now accounted. If this is called | |
2286 | * concurrently, or if we already did some work before crashing, | |
2287 | * that's fine, since we track each object's accounted state | |
2288 | * independently. | |
2289 | */ | |
2290 | ||
2291 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { | |
45d1cae3 | 2292 | dmu_tx_t *tx; |
9babb374 BB |
2293 | dmu_buf_t *db; |
2294 | int objerr; | |
2295 | ||
1de321e6 JX |
2296 | mutex_enter(&os->os_upgrade_lock); |
2297 | if (os->os_upgrade_exit) | |
2298 | err = SET_ERROR(EINTR); | |
2299 | mutex_exit(&os->os_upgrade_lock); | |
2300 | if (err != 0) | |
2301 | return (err); | |
2302 | ||
9babb374 | 2303 | if (issig(JUSTLOOKING) && issig(FORREAL)) |
2e528b49 | 2304 | return (SET_ERROR(EINTR)); |
9babb374 BB |
2305 | |
2306 | objerr = dmu_bonus_hold(os, obj, FTAG, &db); | |
13fe0198 | 2307 | if (objerr != 0) |
9babb374 | 2308 | continue; |
45d1cae3 | 2309 | tx = dmu_tx_create(os); |
9babb374 BB |
2310 | dmu_tx_hold_bonus(tx, obj); |
2311 | objerr = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 2312 | if (objerr != 0) { |
d22323e8 | 2313 | dmu_buf_rele(db, FTAG); |
9babb374 BB |
2314 | dmu_tx_abort(tx); |
2315 | continue; | |
2316 | } | |
2317 | dmu_buf_will_dirty(db, tx); | |
2318 | dmu_buf_rele(db, FTAG); | |
2319 | dmu_tx_commit(tx); | |
2320 | } | |
1de321e6 JX |
2321 | return (0); |
2322 | } | |
2323 | ||
4072f465 | 2324 | static int |
2325 | dmu_objset_userspace_upgrade_cb(objset_t *os) | |
1de321e6 JX |
2326 | { |
2327 | int err = 0; | |
2328 | ||
2329 | if (dmu_objset_userspace_present(os)) | |
2330 | return (0); | |
2331 | if (dmu_objset_is_snapshot(os)) | |
2332 | return (SET_ERROR(EINVAL)); | |
2333 | if (!dmu_objset_userused_enabled(os)) | |
2334 | return (SET_ERROR(ENOTSUP)); | |
2335 | ||
2336 | err = dmu_objset_space_upgrade(os); | |
2337 | if (err) | |
2338 | return (err); | |
9babb374 | 2339 | |
428870ff | 2340 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
9babb374 BB |
2341 | txg_wait_synced(dmu_objset_pool(os), 0); |
2342 | return (0); | |
2343 | } | |
2344 | ||
4072f465 | 2345 | void |
2346 | dmu_objset_userspace_upgrade(objset_t *os) | |
2347 | { | |
2348 | dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb); | |
2349 | } | |
2350 | ||
1de321e6 | 2351 | static int |
9c5167d1 | 2352 | dmu_objset_id_quota_upgrade_cb(objset_t *os) |
1de321e6 JX |
2353 | { |
2354 | int err = 0; | |
2355 | ||
9c5167d1 NF |
2356 | if (dmu_objset_userobjspace_present(os) && |
2357 | dmu_objset_projectquota_present(os)) | |
1de321e6 JX |
2358 | return (0); |
2359 | if (dmu_objset_is_snapshot(os)) | |
2360 | return (SET_ERROR(EINVAL)); | |
4072f465 | 2361 | if (!dmu_objset_userused_enabled(os)) |
1de321e6 | 2362 | return (SET_ERROR(ENOTSUP)); |
9c5167d1 NF |
2363 | if (!dmu_objset_projectquota_enabled(os) && |
2364 | dmu_objset_userobjspace_present(os)) | |
2365 | return (SET_ERROR(ENOTSUP)); | |
1de321e6 | 2366 | |
4072f465 | 2367 | if (dmu_objset_userobjused_enabled(os)) |
2368 | dmu_objset_ds(os)->ds_feature_activation[ | |
2369 | SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; | |
9c5167d1 | 2370 | if (dmu_objset_projectquota_enabled(os)) |
d52d80b7 PD |
2371 | dmu_objset_ds(os)->ds_feature_activation[ |
2372 | SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; | |
1de321e6 JX |
2373 | |
2374 | err = dmu_objset_space_upgrade(os); | |
2375 | if (err) | |
2376 | return (err); | |
2377 | ||
4072f465 | 2378 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
2379 | if (dmu_objset_userobjused_enabled(os)) | |
2380 | os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
9c5167d1 NF |
2381 | if (dmu_objset_projectquota_enabled(os)) |
2382 | os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
2383 | ||
1de321e6 JX |
2384 | txg_wait_synced(dmu_objset_pool(os), 0); |
2385 | return (0); | |
2386 | } | |
2387 | ||
2388 | void | |
9c5167d1 | 2389 | dmu_objset_id_quota_upgrade(objset_t *os) |
1de321e6 | 2390 | { |
9c5167d1 | 2391 | dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); |
1de321e6 JX |
2392 | } |
2393 | ||
126ae9f4 JX |
2394 | boolean_t |
2395 | dmu_objset_userobjspace_upgradable(objset_t *os) | |
2396 | { | |
2397 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2398 | !dmu_objset_is_snapshot(os) && | |
2399 | dmu_objset_userobjused_enabled(os) && | |
bb1be77a | 2400 | !dmu_objset_userobjspace_present(os) && |
2401 | spa_writeable(dmu_objset_spa(os))); | |
126ae9f4 JX |
2402 | } |
2403 | ||
9c5167d1 NF |
2404 | boolean_t |
2405 | dmu_objset_projectquota_upgradable(objset_t *os) | |
2406 | { | |
2407 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2408 | !dmu_objset_is_snapshot(os) && | |
2409 | dmu_objset_projectquota_enabled(os) && | |
bb1be77a | 2410 | !dmu_objset_projectquota_present(os) && |
2411 | spa_writeable(dmu_objset_spa(os))); | |
9c5167d1 NF |
2412 | } |
2413 | ||
34dc7c2f BB |
2414 | void |
2415 | dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, | |
2416 | uint64_t *usedobjsp, uint64_t *availobjsp) | |
2417 | { | |
428870ff | 2418 | dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, |
34dc7c2f BB |
2419 | usedobjsp, availobjsp); |
2420 | } | |
2421 | ||
2422 | uint64_t | |
2423 | dmu_objset_fsid_guid(objset_t *os) | |
2424 | { | |
428870ff | 2425 | return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); |
34dc7c2f BB |
2426 | } |
2427 | ||
2428 | void | |
2429 | dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) | |
2430 | { | |
428870ff BB |
2431 | stat->dds_type = os->os_phys->os_type; |
2432 | if (os->os_dsl_dataset) | |
2433 | dsl_dataset_fast_stat(os->os_dsl_dataset, stat); | |
34dc7c2f BB |
2434 | } |
2435 | ||
2436 | void | |
2437 | dmu_objset_stats(objset_t *os, nvlist_t *nv) | |
2438 | { | |
428870ff BB |
2439 | ASSERT(os->os_dsl_dataset || |
2440 | os->os_phys->os_type == DMU_OST_META); | |
34dc7c2f | 2441 | |
428870ff BB |
2442 | if (os->os_dsl_dataset != NULL) |
2443 | dsl_dataset_stats(os->os_dsl_dataset, nv); | |
34dc7c2f BB |
2444 | |
2445 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, | |
428870ff | 2446 | os->os_phys->os_type); |
9babb374 BB |
2447 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, |
2448 | dmu_objset_userspace_present(os)); | |
34dc7c2f BB |
2449 | } |
2450 | ||
2451 | int | |
2452 | dmu_objset_is_snapshot(objset_t *os) | |
2453 | { | |
428870ff | 2454 | if (os->os_dsl_dataset != NULL) |
0c66c32d | 2455 | return (os->os_dsl_dataset->ds_is_snapshot); |
34dc7c2f BB |
2456 | else |
2457 | return (B_FALSE); | |
2458 | } | |
2459 | ||
2460 | int | |
4d55ea81 | 2461 | dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen, |
34dc7c2f BB |
2462 | boolean_t *conflict) |
2463 | { | |
428870ff | 2464 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2465 | uint64_t ignored; |
2466 | ||
d683ddbb | 2467 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2468 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2469 | |
2470 | return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2471 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, |
9b7b9cd3 | 2472 | MT_NORMALIZE, real, maxlen, conflict)); |
34dc7c2f BB |
2473 | } |
2474 | ||
2475 | int | |
2476 | dmu_snapshot_list_next(objset_t *os, int namelen, char *name, | |
2477 | uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) | |
2478 | { | |
428870ff | 2479 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2480 | zap_cursor_t cursor; |
2481 | zap_attribute_t attr; | |
2482 | ||
13fe0198 MA |
2483 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
2484 | ||
d683ddbb | 2485 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2486 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2487 | |
2488 | zap_cursor_init_serialized(&cursor, | |
2489 | ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2490 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); |
34dc7c2f BB |
2491 | |
2492 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2493 | zap_cursor_fini(&cursor); | |
2e528b49 | 2494 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2495 | } |
2496 | ||
2497 | if (strlen(attr.za_name) + 1 > namelen) { | |
2498 | zap_cursor_fini(&cursor); | |
2e528b49 | 2499 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2500 | } |
2501 | ||
c9e319fa | 2502 | (void) strlcpy(name, attr.za_name, namelen); |
34dc7c2f BB |
2503 | if (idp) |
2504 | *idp = attr.za_first_integer; | |
2505 | if (case_conflict) | |
2506 | *case_conflict = attr.za_normalization_conflict; | |
2507 | zap_cursor_advance(&cursor); | |
2508 | *offp = zap_cursor_serialize(&cursor); | |
2509 | zap_cursor_fini(&cursor); | |
2510 | ||
2511 | return (0); | |
2512 | } | |
2513 | ||
ebe7e575 | 2514 | int |
6772fb67 | 2515 | dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) |
ebe7e575 | 2516 | { |
d1d7e268 | 2517 | return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); |
ebe7e575 BB |
2518 | } |
2519 | ||
34dc7c2f BB |
2520 | int |
2521 | dmu_dir_list_next(objset_t *os, int namelen, char *name, | |
2522 | uint64_t *idp, uint64_t *offp) | |
2523 | { | |
428870ff | 2524 | dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; |
34dc7c2f BB |
2525 | zap_cursor_t cursor; |
2526 | zap_attribute_t attr; | |
2527 | ||
2528 | /* there is no next dir on a snapshot! */ | |
428870ff | 2529 | if (os->os_dsl_dataset->ds_object != |
d683ddbb | 2530 | dsl_dir_phys(dd)->dd_head_dataset_obj) |
2e528b49 | 2531 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2532 | |
2533 | zap_cursor_init_serialized(&cursor, | |
2534 | dd->dd_pool->dp_meta_objset, | |
d683ddbb | 2535 | dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); |
34dc7c2f BB |
2536 | |
2537 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2538 | zap_cursor_fini(&cursor); | |
2e528b49 | 2539 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2540 | } |
2541 | ||
2542 | if (strlen(attr.za_name) + 1 > namelen) { | |
2543 | zap_cursor_fini(&cursor); | |
2e528b49 | 2544 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2545 | } |
2546 | ||
c9e319fa | 2547 | (void) strlcpy(name, attr.za_name, namelen); |
34dc7c2f BB |
2548 | if (idp) |
2549 | *idp = attr.za_first_integer; | |
2550 | zap_cursor_advance(&cursor); | |
2551 | *offp = zap_cursor_serialize(&cursor); | |
2552 | zap_cursor_fini(&cursor); | |
2553 | ||
2554 | return (0); | |
2555 | } | |
2556 | ||
9c43027b AJ |
2557 | typedef struct dmu_objset_find_ctx { |
2558 | taskq_t *dc_tq; | |
2559 | dsl_pool_t *dc_dp; | |
2560 | uint64_t dc_ddobj; | |
1149ba64 | 2561 | char *dc_ddname; /* last component of ddobj's name */ |
9c43027b AJ |
2562 | int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); |
2563 | void *dc_arg; | |
2564 | int dc_flags; | |
2565 | kmutex_t *dc_error_lock; | |
2566 | int *dc_error; | |
2567 | } dmu_objset_find_ctx_t; | |
2568 | ||
2569 | static void | |
2570 | dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) | |
b128c09f | 2571 | { |
9c43027b | 2572 | dsl_pool_t *dp = dcp->dc_dp; |
13fe0198 MA |
2573 | dsl_dir_t *dd; |
2574 | dsl_dataset_t *ds; | |
2575 | zap_cursor_t zc; | |
2576 | zap_attribute_t *attr; | |
2577 | uint64_t thisobj; | |
9c43027b | 2578 | int err = 0; |
13fe0198 | 2579 | |
9c43027b AJ |
2580 | /* don't process if there already was an error */ |
2581 | if (*dcp->dc_error != 0) | |
2582 | goto out; | |
13fe0198 | 2583 | |
1149ba64 GM |
2584 | /* |
2585 | * Note: passing the name (dc_ddname) here is optional, but it | |
2586 | * improves performance because we don't need to call | |
2587 | * zap_value_search() to determine the name. | |
2588 | */ | |
2589 | err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); | |
13fe0198 | 2590 | if (err != 0) |
9c43027b | 2591 | goto out; |
13fe0198 MA |
2592 | |
2593 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ | |
2594 | if (dd->dd_myname[0] == '$') { | |
2595 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2596 | goto out; |
13fe0198 MA |
2597 | } |
2598 | ||
d683ddbb | 2599 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2600 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
13fe0198 MA |
2601 | |
2602 | /* | |
2603 | * Iterate over all children. | |
2604 | */ | |
9c43027b | 2605 | if (dcp->dc_flags & DS_FIND_CHILDREN) { |
13fe0198 | 2606 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2607 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
13fe0198 MA |
2608 | zap_cursor_retrieve(&zc, attr) == 0; |
2609 | (void) zap_cursor_advance(&zc)) { | |
2610 | ASSERT3U(attr->za_integer_length, ==, | |
2611 | sizeof (uint64_t)); | |
2612 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2613 | ||
1c27024e | 2614 | dmu_objset_find_ctx_t *child_dcp = |
1149ba64 | 2615 | kmem_alloc(sizeof (*child_dcp), KM_SLEEP); |
9c43027b AJ |
2616 | *child_dcp = *dcp; |
2617 | child_dcp->dc_ddobj = attr->za_first_integer; | |
1149ba64 | 2618 | child_dcp->dc_ddname = spa_strdup(attr->za_name); |
9c43027b AJ |
2619 | if (dcp->dc_tq != NULL) |
2620 | (void) taskq_dispatch(dcp->dc_tq, | |
2621 | dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); | |
2622 | else | |
2623 | dmu_objset_find_dp_impl(child_dcp); | |
13fe0198 MA |
2624 | } |
2625 | zap_cursor_fini(&zc); | |
13fe0198 MA |
2626 | } |
2627 | ||
2628 | /* | |
2629 | * Iterate over all snapshots. | |
2630 | */ | |
9c43027b | 2631 | if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { |
13fe0198 MA |
2632 | dsl_dataset_t *ds; |
2633 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2634 | ||
2635 | if (err == 0) { | |
d683ddbb JG |
2636 | uint64_t snapobj; |
2637 | ||
2638 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
13fe0198 MA |
2639 | dsl_dataset_rele(ds, FTAG); |
2640 | ||
2641 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2642 | zap_cursor_retrieve(&zc, attr) == 0; | |
2643 | (void) zap_cursor_advance(&zc)) { | |
2644 | ASSERT3U(attr->za_integer_length, ==, | |
2645 | sizeof (uint64_t)); | |
2646 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2647 | ||
2648 | err = dsl_dataset_hold_obj(dp, | |
2649 | attr->za_first_integer, FTAG, &ds); | |
2650 | if (err != 0) | |
2651 | break; | |
9c43027b | 2652 | err = dcp->dc_func(dp, ds, dcp->dc_arg); |
13fe0198 MA |
2653 | dsl_dataset_rele(ds, FTAG); |
2654 | if (err != 0) | |
2655 | break; | |
2656 | } | |
2657 | zap_cursor_fini(&zc); | |
2658 | } | |
2659 | } | |
2660 | ||
13fe0198 MA |
2661 | kmem_free(attr, sizeof (zap_attribute_t)); |
2662 | ||
1149ba64 GM |
2663 | if (err != 0) { |
2664 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2665 | goto out; |
1149ba64 | 2666 | } |
13fe0198 MA |
2667 | |
2668 | /* | |
2669 | * Apply to self. | |
2670 | */ | |
2671 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
1149ba64 GM |
2672 | |
2673 | /* | |
2674 | * Note: we hold the dir while calling dsl_dataset_hold_obj() so | |
2675 | * that the dir will remain cached, and we won't have to re-instantiate | |
2676 | * it (which could be expensive due to finding its name via | |
2677 | * zap_value_search()). | |
2678 | */ | |
2679 | dsl_dir_rele(dd, FTAG); | |
13fe0198 | 2680 | if (err != 0) |
9c43027b AJ |
2681 | goto out; |
2682 | err = dcp->dc_func(dp, ds, dcp->dc_arg); | |
13fe0198 | 2683 | dsl_dataset_rele(ds, FTAG); |
9c43027b AJ |
2684 | |
2685 | out: | |
2686 | if (err != 0) { | |
2687 | mutex_enter(dcp->dc_error_lock); | |
2688 | /* only keep first error */ | |
2689 | if (*dcp->dc_error == 0) | |
2690 | *dcp->dc_error = err; | |
2691 | mutex_exit(dcp->dc_error_lock); | |
2692 | } | |
2693 | ||
1149ba64 GM |
2694 | if (dcp->dc_ddname != NULL) |
2695 | spa_strfree(dcp->dc_ddname); | |
9c43027b AJ |
2696 | kmem_free(dcp, sizeof (*dcp)); |
2697 | } | |
2698 | ||
2699 | static void | |
2700 | dmu_objset_find_dp_cb(void *arg) | |
2701 | { | |
2702 | dmu_objset_find_ctx_t *dcp = arg; | |
2703 | dsl_pool_t *dp = dcp->dc_dp; | |
2704 | ||
5e8cd5d1 AJ |
2705 | /* |
2706 | * We need to get a pool_config_lock here, as there are several | |
e1cfd73f | 2707 | * assert(pool_config_held) down the stack. Getting a lock via |
5e8cd5d1 AJ |
2708 | * dsl_pool_config_enter is risky, as it might be stalled by a |
2709 | * pending writer. This would deadlock, as the write lock can | |
2710 | * only be granted when our parent thread gives up the lock. | |
2711 | * The _prio interface gives us priority over a pending writer. | |
2712 | */ | |
2713 | dsl_pool_config_enter_prio(dp, FTAG); | |
9c43027b AJ |
2714 | |
2715 | dmu_objset_find_dp_impl(dcp); | |
2716 | ||
2717 | dsl_pool_config_exit(dp, FTAG); | |
2718 | } | |
2719 | ||
2720 | /* | |
2721 | * Find objsets under and including ddobj, call func(ds) on each. | |
2722 | * The order for the enumeration is completely undefined. | |
2723 | * func is called with dsl_pool_config held. | |
2724 | */ | |
2725 | int | |
2726 | dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, | |
2727 | int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) | |
2728 | { | |
2729 | int error = 0; | |
2730 | taskq_t *tq = NULL; | |
2731 | int ntasks; | |
2732 | dmu_objset_find_ctx_t *dcp; | |
2733 | kmutex_t err_lock; | |
2734 | ||
2735 | mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); | |
2736 | dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); | |
2737 | dcp->dc_tq = NULL; | |
2738 | dcp->dc_dp = dp; | |
2739 | dcp->dc_ddobj = ddobj; | |
1149ba64 | 2740 | dcp->dc_ddname = NULL; |
9c43027b AJ |
2741 | dcp->dc_func = func; |
2742 | dcp->dc_arg = arg; | |
2743 | dcp->dc_flags = flags; | |
2744 | dcp->dc_error_lock = &err_lock; | |
2745 | dcp->dc_error = &error; | |
2746 | ||
2747 | if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { | |
2748 | /* | |
2749 | * In case a write lock is held we can't make use of | |
2750 | * parallelism, as down the stack of the worker threads | |
2751 | * the lock is asserted via dsl_pool_config_held. | |
2752 | * In case of a read lock this is solved by getting a read | |
2753 | * lock in each worker thread, which isn't possible in case | |
2754 | * of a writer lock. So we fall back to the synchronous path | |
2755 | * here. | |
2756 | * In the future it might be possible to get some magic into | |
2757 | * dsl_pool_config_held in a way that it returns true for | |
2758 | * the worker threads so that a single lock held from this | |
2759 | * thread suffices. For now, stay single threaded. | |
2760 | */ | |
2761 | dmu_objset_find_dp_impl(dcp); | |
e5676636 | 2762 | mutex_destroy(&err_lock); |
9c43027b AJ |
2763 | |
2764 | return (error); | |
2765 | } | |
2766 | ||
2767 | ntasks = dmu_find_threads; | |
2768 | if (ntasks == 0) | |
2769 | ntasks = vdev_count_leaves(dp->dp_spa) * 4; | |
1229323d | 2770 | tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, |
9c43027b AJ |
2771 | INT_MAX, 0); |
2772 | if (tq == NULL) { | |
2773 | kmem_free(dcp, sizeof (*dcp)); | |
e5676636 BB |
2774 | mutex_destroy(&err_lock); |
2775 | ||
9c43027b AJ |
2776 | return (SET_ERROR(ENOMEM)); |
2777 | } | |
2778 | dcp->dc_tq = tq; | |
2779 | ||
2780 | /* dcp will be freed by task */ | |
2781 | (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); | |
2782 | ||
2783 | /* | |
2784 | * PORTING: this code relies on the property of taskq_wait to wait | |
2785 | * until no more tasks are queued and no more tasks are active. As | |
2786 | * we always queue new tasks from within other tasks, task_wait | |
2787 | * reliably waits for the full recursion to finish, even though we | |
2788 | * enqueue new tasks after taskq_wait has been called. | |
2789 | * On platforms other than illumos, taskq_wait may not have this | |
2790 | * property. | |
2791 | */ | |
2792 | taskq_wait(tq); | |
2793 | taskq_destroy(tq); | |
2794 | mutex_destroy(&err_lock); | |
2795 | ||
2796 | return (error); | |
b128c09f BB |
2797 | } |
2798 | ||
2799 | /* | |
13fe0198 MA |
2800 | * Find all objsets under name, and for each, call 'func(child_name, arg)'. |
2801 | * The dp_config_rwlock must not be held when this is called, and it | |
2802 | * will not be held when the callback is called. | |
2803 | * Therefore this function should only be used when the pool is not changing | |
2804 | * (e.g. in syncing context), or the callback can deal with the possible races. | |
b128c09f | 2805 | */ |
13fe0198 MA |
2806 | static int |
2807 | dmu_objset_find_impl(spa_t *spa, const char *name, | |
2808 | int func(const char *, void *), void *arg, int flags) | |
34dc7c2f BB |
2809 | { |
2810 | dsl_dir_t *dd; | |
13fe0198 | 2811 | dsl_pool_t *dp = spa_get_dsl(spa); |
b128c09f | 2812 | dsl_dataset_t *ds; |
34dc7c2f BB |
2813 | zap_cursor_t zc; |
2814 | zap_attribute_t *attr; | |
2815 | char *child; | |
b128c09f BB |
2816 | uint64_t thisobj; |
2817 | int err; | |
34dc7c2f | 2818 | |
13fe0198 MA |
2819 | dsl_pool_config_enter(dp, FTAG); |
2820 | ||
2821 | err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); | |
2822 | if (err != 0) { | |
2823 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f | 2824 | return (err); |
13fe0198 | 2825 | } |
34dc7c2f | 2826 | |
b128c09f BB |
2827 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ |
2828 | if (dd->dd_myname[0] == '$') { | |
13fe0198 MA |
2829 | dsl_dir_rele(dd, FTAG); |
2830 | dsl_pool_config_exit(dp, FTAG); | |
b128c09f BB |
2831 | return (0); |
2832 | } | |
2833 | ||
d683ddbb | 2834 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2835 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
34dc7c2f BB |
2836 | |
2837 | /* | |
2838 | * Iterate over all children. | |
2839 | */ | |
2840 | if (flags & DS_FIND_CHILDREN) { | |
b128c09f | 2841 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2842 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
2843 | zap_cursor_retrieve(&zc, attr) == 0; |
2844 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 MA |
2845 | ASSERT3U(attr->za_integer_length, ==, |
2846 | sizeof (uint64_t)); | |
2847 | ASSERT3U(attr->za_num_integers, ==, 1); | |
34dc7c2f | 2848 | |
428870ff | 2849 | child = kmem_asprintf("%s/%s", name, attr->za_name); |
13fe0198 MA |
2850 | dsl_pool_config_exit(dp, FTAG); |
2851 | err = dmu_objset_find_impl(spa, child, | |
2852 | func, arg, flags); | |
2853 | dsl_pool_config_enter(dp, FTAG); | |
e4f5fa12 | 2854 | kmem_strfree(child); |
13fe0198 | 2855 | if (err != 0) |
34dc7c2f BB |
2856 | break; |
2857 | } | |
2858 | zap_cursor_fini(&zc); | |
2859 | ||
13fe0198 MA |
2860 | if (err != 0) { |
2861 | dsl_dir_rele(dd, FTAG); | |
2862 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
2863 | kmem_free(attr, sizeof (zap_attribute_t)); |
2864 | return (err); | |
2865 | } | |
2866 | } | |
2867 | ||
2868 | /* | |
2869 | * Iterate over all snapshots. | |
2870 | */ | |
b128c09f | 2871 | if (flags & DS_FIND_SNAPSHOTS) { |
b128c09f | 2872 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); |
b128c09f BB |
2873 | |
2874 | if (err == 0) { | |
d683ddbb JG |
2875 | uint64_t snapobj; |
2876 | ||
2877 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
b128c09f BB |
2878 | dsl_dataset_rele(ds, FTAG); |
2879 | ||
2880 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2881 | zap_cursor_retrieve(&zc, attr) == 0; | |
2882 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 | 2883 | ASSERT3U(attr->za_integer_length, ==, |
b128c09f | 2884 | sizeof (uint64_t)); |
13fe0198 | 2885 | ASSERT3U(attr->za_num_integers, ==, 1); |
b128c09f | 2886 | |
428870ff BB |
2887 | child = kmem_asprintf("%s@%s", |
2888 | name, attr->za_name); | |
13fe0198 MA |
2889 | dsl_pool_config_exit(dp, FTAG); |
2890 | err = func(child, arg); | |
2891 | dsl_pool_config_enter(dp, FTAG); | |
e4f5fa12 | 2892 | kmem_strfree(child); |
13fe0198 | 2893 | if (err != 0) |
b128c09f BB |
2894 | break; |
2895 | } | |
2896 | zap_cursor_fini(&zc); | |
34dc7c2f | 2897 | } |
34dc7c2f BB |
2898 | } |
2899 | ||
13fe0198 | 2900 | dsl_dir_rele(dd, FTAG); |
34dc7c2f | 2901 | kmem_free(attr, sizeof (zap_attribute_t)); |
13fe0198 | 2902 | dsl_pool_config_exit(dp, FTAG); |
34dc7c2f | 2903 | |
13fe0198 | 2904 | if (err != 0) |
34dc7c2f BB |
2905 | return (err); |
2906 | ||
13fe0198 MA |
2907 | /* Apply to self. */ |
2908 | return (func(name, arg)); | |
34dc7c2f BB |
2909 | } |
2910 | ||
13fe0198 MA |
2911 | /* |
2912 | * See comment above dmu_objset_find_impl(). | |
2913 | */ | |
d164b209 | 2914 | int |
5df7e9d8 | 2915 | dmu_objset_find(const char *name, int func(const char *, void *), void *arg, |
13fe0198 | 2916 | int flags) |
d164b209 | 2917 | { |
13fe0198 MA |
2918 | spa_t *spa; |
2919 | int error; | |
d164b209 | 2920 | |
13fe0198 MA |
2921 | error = spa_open(name, &spa, FTAG); |
2922 | if (error != 0) | |
2923 | return (error); | |
2924 | error = dmu_objset_find_impl(spa, name, func, arg, flags); | |
2925 | spa_close(spa, FTAG); | |
2926 | return (error); | |
d164b209 BB |
2927 | } |
2928 | ||
ae76f45c TC |
2929 | boolean_t |
2930 | dmu_objset_incompatible_encryption_version(objset_t *os) | |
2931 | { | |
2932 | return (dsl_dir_incompatible_encryption_version( | |
2933 | os->os_dsl_dataset->ds_dir)); | |
2934 | } | |
2935 | ||
34dc7c2f BB |
2936 | void |
2937 | dmu_objset_set_user(objset_t *os, void *user_ptr) | |
2938 | { | |
428870ff BB |
2939 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2940 | os->os_user_ptr = user_ptr; | |
34dc7c2f BB |
2941 | } |
2942 | ||
2943 | void * | |
2944 | dmu_objset_get_user(objset_t *os) | |
2945 | { | |
428870ff BB |
2946 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2947 | return (os->os_user_ptr); | |
34dc7c2f | 2948 | } |
c28b2279 | 2949 | |
13fe0198 MA |
2950 | /* |
2951 | * Determine name of filesystem, given name of snapshot. | |
eca7b760 | 2952 | * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes |
13fe0198 MA |
2953 | */ |
2954 | int | |
2955 | dmu_fsname(const char *snapname, char *buf) | |
2956 | { | |
2957 | char *atp = strchr(snapname, '@'); | |
2958 | if (atp == NULL) | |
2e528b49 | 2959 | return (SET_ERROR(EINVAL)); |
eca7b760 | 2960 | if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 2961 | return (SET_ERROR(ENAMETOOLONG)); |
13fe0198 MA |
2962 | (void) strlcpy(buf, snapname, atp - snapname + 1); |
2963 | return (0); | |
2964 | } | |
2965 | ||
3ec3bc21 | 2966 | /* |
0f8ff49e SD |
2967 | * Call when we think we're going to write/free space in open context |
2968 | * to track the amount of dirty data in the open txg, which is also the | |
2969 | * amount of memory that can not be evicted until this txg syncs. | |
2970 | * | |
2971 | * Note that there are two conditions where this can be called from | |
2972 | * syncing context: | |
2973 | * | |
2974 | * [1] When we just created the dataset, in which case we go on with | |
2975 | * updating any accounting of dirty data as usual. | |
2976 | * [2] When we are dirtying MOS data, in which case we only update the | |
2977 | * pool's accounting of dirty data. | |
3ec3bc21 BB |
2978 | */ |
2979 | void | |
2980 | dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) | |
2981 | { | |
2982 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
2983 | int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); | |
2984 | ||
2985 | if (ds != NULL) { | |
2986 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
3ec3bc21 | 2987 | } |
0f8ff49e SD |
2988 | |
2989 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
3ec3bc21 BB |
2990 | } |
2991 | ||
93ce2b4c | 2992 | #if defined(_KERNEL) |
f0fd83be | 2993 | EXPORT_SYMBOL(dmu_objset_zil); |
c28b2279 | 2994 | EXPORT_SYMBOL(dmu_objset_pool); |
f0fd83be BB |
2995 | EXPORT_SYMBOL(dmu_objset_ds); |
2996 | EXPORT_SYMBOL(dmu_objset_type); | |
c28b2279 BB |
2997 | EXPORT_SYMBOL(dmu_objset_name); |
2998 | EXPORT_SYMBOL(dmu_objset_hold); | |
b5256303 | 2999 | EXPORT_SYMBOL(dmu_objset_hold_flags); |
c28b2279 BB |
3000 | EXPORT_SYMBOL(dmu_objset_own); |
3001 | EXPORT_SYMBOL(dmu_objset_rele); | |
b5256303 | 3002 | EXPORT_SYMBOL(dmu_objset_rele_flags); |
c28b2279 BB |
3003 | EXPORT_SYMBOL(dmu_objset_disown); |
3004 | EXPORT_SYMBOL(dmu_objset_from_ds); | |
3005 | EXPORT_SYMBOL(dmu_objset_create); | |
3006 | EXPORT_SYMBOL(dmu_objset_clone); | |
c28b2279 BB |
3007 | EXPORT_SYMBOL(dmu_objset_stats); |
3008 | EXPORT_SYMBOL(dmu_objset_fast_stat); | |
54a179e7 | 3009 | EXPORT_SYMBOL(dmu_objset_spa); |
c28b2279 BB |
3010 | EXPORT_SYMBOL(dmu_objset_space); |
3011 | EXPORT_SYMBOL(dmu_objset_fsid_guid); | |
3012 | EXPORT_SYMBOL(dmu_objset_find); | |
c28b2279 BB |
3013 | EXPORT_SYMBOL(dmu_objset_byteswap); |
3014 | EXPORT_SYMBOL(dmu_objset_evict_dbufs); | |
3015 | EXPORT_SYMBOL(dmu_objset_snap_cmtime); | |
50c957f7 | 3016 | EXPORT_SYMBOL(dmu_objset_dnodesize); |
c28b2279 BB |
3017 | |
3018 | EXPORT_SYMBOL(dmu_objset_sync); | |
3019 | EXPORT_SYMBOL(dmu_objset_is_dirty); | |
b5256303 | 3020 | EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); |
c28b2279 BB |
3021 | EXPORT_SYMBOL(dmu_objset_create_impl); |
3022 | EXPORT_SYMBOL(dmu_objset_open_impl); | |
3023 | EXPORT_SYMBOL(dmu_objset_evict); | |
3024 | EXPORT_SYMBOL(dmu_objset_register_type); | |
ba67d821 | 3025 | EXPORT_SYMBOL(dmu_objset_sync_done); |
c28b2279 BB |
3026 | EXPORT_SYMBOL(dmu_objset_userquota_get_ids); |
3027 | EXPORT_SYMBOL(dmu_objset_userused_enabled); | |
3028 | EXPORT_SYMBOL(dmu_objset_userspace_upgrade); | |
3029 | EXPORT_SYMBOL(dmu_objset_userspace_present); | |
1de321e6 | 3030 | EXPORT_SYMBOL(dmu_objset_userobjused_enabled); |
126ae9f4 | 3031 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); |
1de321e6 | 3032 | EXPORT_SYMBOL(dmu_objset_userobjspace_present); |
9c5167d1 NF |
3033 | EXPORT_SYMBOL(dmu_objset_projectquota_enabled); |
3034 | EXPORT_SYMBOL(dmu_objset_projectquota_present); | |
3035 | EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); | |
3036 | EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); | |
c28b2279 | 3037 | #endif |