]>
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 | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
34dc7c2f BB |
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 | |
5405be03 | 35 | * Copyright (c) 2022 Hewlett Packard Enterprise Development LP. |
34dc7c2f BB |
36 | */ |
37 | ||
428870ff BB |
38 | /* Portions Copyright 2010 Robert Milkowski */ |
39 | ||
34dc7c2f BB |
40 | #include <sys/cred.h> |
41 | #include <sys/zfs_context.h> | |
42 | #include <sys/dmu_objset.h> | |
43 | #include <sys/dsl_dir.h> | |
44 | #include <sys/dsl_dataset.h> | |
45 | #include <sys/dsl_prop.h> | |
46 | #include <sys/dsl_pool.h> | |
47 | #include <sys/dsl_synctask.h> | |
48 | #include <sys/dsl_deleg.h> | |
49 | #include <sys/dnode.h> | |
50 | #include <sys/dbuf.h> | |
51 | #include <sys/zvol.h> | |
52 | #include <sys/dmu_tx.h> | |
34dc7c2f BB |
53 | #include <sys/zap.h> |
54 | #include <sys/zil.h> | |
55 | #include <sys/dmu_impl.h> | |
56 | #include <sys/zfs_ioctl.h> | |
428870ff | 57 | #include <sys/sa.h> |
572e2857 | 58 | #include <sys/zfs_onexit.h> |
13fe0198 | 59 | #include <sys/dsl_destroy.h> |
9c43027b | 60 | #include <sys/vdev.h> |
a1d477c2 | 61 | #include <sys/zfeature.h> |
f74b821a | 62 | #include <sys/policy.h> |
1de321e6 | 63 | #include <sys/spa_impl.h> |
03916905 | 64 | #include <sys/dmu_recv.h> |
9c5167d1 | 65 | #include <sys/zfs_project.h> |
a7ed98d8 | 66 | #include "zfs_namecheck.h" |
c9d62d13 GA |
67 | #include <sys/vdev_impl.h> |
68 | #include <sys/arc.h> | |
572e2857 BB |
69 | |
70 | /* | |
71 | * Needed to close a window in dnode_move() that allows the objset to be freed | |
72 | * before it can be safely accessed. | |
73 | */ | |
74 | krwlock_t os_lock; | |
75 | ||
9c43027b | 76 | /* |
4e33ba4c | 77 | * Tunable to overwrite the maximum number of threads for the parallelization |
9c43027b AJ |
78 | * of dmu_objset_find_dp, needed to speed up the import of pools with many |
79 | * datasets. | |
80 | * Default is 4 times the number of leaf vdevs. | |
81 | */ | |
18168da7 | 82 | static const int dmu_find_threads = 0; |
9c43027b | 83 | |
68cbd56e NB |
84 | /* |
85 | * Backfill lower metadnode objects after this many have been freed. | |
86 | * Backfilling negatively impacts object creation rates, so only do it | |
87 | * if there are enough holes to fill. | |
88 | */ | |
18168da7 | 89 | static const int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; |
68cbd56e | 90 | |
18168da7 | 91 | static const char *upgrade_tag = "upgrade_tag"; |
c0daec32 | 92 | |
9c43027b AJ |
93 | static void dmu_objset_find_dp_cb(void *arg); |
94 | ||
1de321e6 JX |
95 | static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); |
96 | static void dmu_objset_upgrade_stop(objset_t *os); | |
97 | ||
572e2857 BB |
98 | void |
99 | dmu_objset_init(void) | |
100 | { | |
101 | rw_init(&os_lock, NULL, RW_DEFAULT, NULL); | |
102 | } | |
103 | ||
104 | void | |
105 | dmu_objset_fini(void) | |
106 | { | |
107 | rw_destroy(&os_lock); | |
108 | } | |
34dc7c2f BB |
109 | |
110 | spa_t * | |
111 | dmu_objset_spa(objset_t *os) | |
112 | { | |
428870ff | 113 | return (os->os_spa); |
34dc7c2f BB |
114 | } |
115 | ||
116 | zilog_t * | |
117 | dmu_objset_zil(objset_t *os) | |
118 | { | |
428870ff | 119 | return (os->os_zil); |
34dc7c2f BB |
120 | } |
121 | ||
122 | dsl_pool_t * | |
123 | dmu_objset_pool(objset_t *os) | |
124 | { | |
125 | dsl_dataset_t *ds; | |
126 | ||
428870ff | 127 | if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) |
34dc7c2f BB |
128 | return (ds->ds_dir->dd_pool); |
129 | else | |
428870ff | 130 | return (spa_get_dsl(os->os_spa)); |
34dc7c2f BB |
131 | } |
132 | ||
133 | dsl_dataset_t * | |
134 | dmu_objset_ds(objset_t *os) | |
135 | { | |
428870ff | 136 | return (os->os_dsl_dataset); |
34dc7c2f BB |
137 | } |
138 | ||
139 | dmu_objset_type_t | |
140 | dmu_objset_type(objset_t *os) | |
141 | { | |
428870ff | 142 | return (os->os_phys->os_type); |
34dc7c2f BB |
143 | } |
144 | ||
145 | void | |
146 | dmu_objset_name(objset_t *os, char *buf) | |
147 | { | |
428870ff | 148 | dsl_dataset_name(os->os_dsl_dataset, buf); |
34dc7c2f BB |
149 | } |
150 | ||
151 | uint64_t | |
152 | dmu_objset_id(objset_t *os) | |
153 | { | |
428870ff | 154 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
155 | |
156 | return (ds ? ds->ds_object : 0); | |
157 | } | |
158 | ||
50c957f7 NB |
159 | uint64_t |
160 | dmu_objset_dnodesize(objset_t *os) | |
161 | { | |
162 | return (os->os_dnodesize); | |
163 | } | |
164 | ||
faf0f58c | 165 | zfs_sync_type_t |
428870ff BB |
166 | dmu_objset_syncprop(objset_t *os) |
167 | { | |
168 | return (os->os_sync); | |
169 | } | |
170 | ||
faf0f58c | 171 | zfs_logbias_op_t |
428870ff BB |
172 | dmu_objset_logbias(objset_t *os) |
173 | { | |
174 | return (os->os_logbias); | |
175 | } | |
176 | ||
34dc7c2f BB |
177 | static void |
178 | checksum_changed_cb(void *arg, uint64_t newval) | |
179 | { | |
428870ff | 180 | objset_t *os = arg; |
34dc7c2f BB |
181 | |
182 | /* | |
183 | * Inheritance should have been done by now. | |
184 | */ | |
185 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
186 | ||
428870ff | 187 | os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); |
34dc7c2f BB |
188 | } |
189 | ||
190 | static void | |
191 | compression_changed_cb(void *arg, uint64_t newval) | |
192 | { | |
428870ff | 193 | objset_t *os = arg; |
34dc7c2f BB |
194 | |
195 | /* | |
196 | * Inheritance and range checking should have been done by now. | |
197 | */ | |
198 | ASSERT(newval != ZIO_COMPRESS_INHERIT); | |
199 | ||
10b3c7f5 MN |
200 | os->os_compress = zio_compress_select(os->os_spa, |
201 | ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON); | |
202 | os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress, | |
203 | ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT); | |
34dc7c2f BB |
204 | } |
205 | ||
206 | static void | |
207 | copies_changed_cb(void *arg, uint64_t newval) | |
208 | { | |
428870ff | 209 | objset_t *os = arg; |
34dc7c2f BB |
210 | |
211 | /* | |
212 | * Inheritance and range checking should have been done by now. | |
213 | */ | |
214 | ASSERT(newval > 0); | |
428870ff | 215 | ASSERT(newval <= spa_max_replication(os->os_spa)); |
34dc7c2f | 216 | |
428870ff BB |
217 | os->os_copies = newval; |
218 | } | |
219 | ||
220 | static void | |
221 | dedup_changed_cb(void *arg, uint64_t newval) | |
222 | { | |
223 | objset_t *os = arg; | |
224 | spa_t *spa = os->os_spa; | |
225 | enum zio_checksum checksum; | |
226 | ||
227 | /* | |
228 | * Inheritance should have been done by now. | |
229 | */ | |
230 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
231 | ||
232 | checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); | |
233 | ||
234 | os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; | |
235 | os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); | |
34dc7c2f BB |
236 | } |
237 | ||
b128c09f BB |
238 | static void |
239 | primary_cache_changed_cb(void *arg, uint64_t newval) | |
240 | { | |
428870ff | 241 | objset_t *os = arg; |
b128c09f BB |
242 | |
243 | /* | |
244 | * Inheritance and range checking should have been done by now. | |
245 | */ | |
246 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
247 | newval == ZFS_CACHE_METADATA); | |
248 | ||
428870ff | 249 | os->os_primary_cache = newval; |
b128c09f BB |
250 | } |
251 | ||
252 | static void | |
253 | secondary_cache_changed_cb(void *arg, uint64_t newval) | |
254 | { | |
428870ff | 255 | objset_t *os = arg; |
b128c09f BB |
256 | |
257 | /* | |
258 | * Inheritance and range checking should have been done by now. | |
259 | */ | |
260 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
261 | newval == ZFS_CACHE_METADATA); | |
262 | ||
428870ff BB |
263 | os->os_secondary_cache = newval; |
264 | } | |
265 | ||
266 | static void | |
267 | sync_changed_cb(void *arg, uint64_t newval) | |
268 | { | |
269 | objset_t *os = arg; | |
270 | ||
271 | /* | |
272 | * Inheritance and range checking should have been done by now. | |
273 | */ | |
274 | ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || | |
275 | newval == ZFS_SYNC_DISABLED); | |
276 | ||
277 | os->os_sync = newval; | |
278 | if (os->os_zil) | |
279 | zil_set_sync(os->os_zil, newval); | |
280 | } | |
281 | ||
faf0f58c MA |
282 | static void |
283 | redundant_metadata_changed_cb(void *arg, uint64_t newval) | |
284 | { | |
285 | objset_t *os = arg; | |
286 | ||
287 | /* | |
288 | * Inheritance and range checking should have been done by now. | |
289 | */ | |
290 | ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || | |
5405be03 A |
291 | newval == ZFS_REDUNDANT_METADATA_MOST || |
292 | newval == ZFS_REDUNDANT_METADATA_SOME || | |
293 | newval == ZFS_REDUNDANT_METADATA_NONE); | |
faf0f58c MA |
294 | |
295 | os->os_redundant_metadata = newval; | |
296 | } | |
297 | ||
50c957f7 NB |
298 | static void |
299 | dnodesize_changed_cb(void *arg, uint64_t newval) | |
300 | { | |
301 | objset_t *os = arg; | |
302 | ||
303 | switch (newval) { | |
304 | case ZFS_DNSIZE_LEGACY: | |
305 | os->os_dnodesize = DNODE_MIN_SIZE; | |
306 | break; | |
307 | case ZFS_DNSIZE_AUTO: | |
308 | /* | |
309 | * Choose a dnode size that will work well for most | |
310 | * workloads if the user specified "auto". Future code | |
311 | * improvements could dynamically select a dnode size | |
312 | * based on observed workload patterns. | |
313 | */ | |
314 | os->os_dnodesize = DNODE_MIN_SIZE * 2; | |
315 | break; | |
316 | case ZFS_DNSIZE_1K: | |
317 | case ZFS_DNSIZE_2K: | |
318 | case ZFS_DNSIZE_4K: | |
319 | case ZFS_DNSIZE_8K: | |
320 | case ZFS_DNSIZE_16K: | |
321 | os->os_dnodesize = newval; | |
322 | break; | |
323 | } | |
324 | } | |
325 | ||
cc99f275 DB |
326 | static void |
327 | smallblk_changed_cb(void *arg, uint64_t newval) | |
328 | { | |
329 | objset_t *os = arg; | |
330 | ||
331 | /* | |
332 | * Inheritance and range checking should have been done by now. | |
333 | */ | |
5aa69a57 | 334 | ASSERT(newval <= SPA_MAXBLOCKSIZE); |
cc99f275 DB |
335 | ASSERT(ISP2(newval)); |
336 | ||
337 | os->os_zpl_special_smallblock = newval; | |
338 | } | |
339 | ||
428870ff BB |
340 | static void |
341 | logbias_changed_cb(void *arg, uint64_t newval) | |
342 | { | |
343 | objset_t *os = arg; | |
344 | ||
345 | ASSERT(newval == ZFS_LOGBIAS_LATENCY || | |
346 | newval == ZFS_LOGBIAS_THROUGHPUT); | |
347 | os->os_logbias = newval; | |
348 | if (os->os_zil) | |
349 | zil_set_logbias(os->os_zil, newval); | |
b128c09f BB |
350 | } |
351 | ||
f1512ee6 MA |
352 | static void |
353 | recordsize_changed_cb(void *arg, uint64_t newval) | |
354 | { | |
355 | objset_t *os = arg; | |
356 | ||
357 | os->os_recordsize = newval; | |
358 | } | |
359 | ||
34dc7c2f BB |
360 | void |
361 | dmu_objset_byteswap(void *buf, size_t size) | |
362 | { | |
363 | objset_phys_t *osp = buf; | |
364 | ||
9c5167d1 NF |
365 | ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || |
366 | size == sizeof (objset_phys_t)); | |
34dc7c2f BB |
367 | dnode_byteswap(&osp->os_meta_dnode); |
368 | byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); | |
369 | osp->os_type = BSWAP_64(osp->os_type); | |
9babb374 | 370 | osp->os_flags = BSWAP_64(osp->os_flags); |
9c5167d1 | 371 | if (size >= OBJSET_PHYS_SIZE_V2) { |
9babb374 BB |
372 | dnode_byteswap(&osp->os_userused_dnode); |
373 | dnode_byteswap(&osp->os_groupused_dnode); | |
9c5167d1 NF |
374 | if (size >= sizeof (objset_phys_t)) |
375 | dnode_byteswap(&osp->os_projectused_dnode); | |
9babb374 | 376 | } |
34dc7c2f BB |
377 | } |
378 | ||
64fc7762 MA |
379 | /* |
380 | * The hash is a CRC-based hash of the objset_t pointer and the object number. | |
381 | */ | |
382 | static uint64_t | |
383 | dnode_hash(const objset_t *os, uint64_t obj) | |
384 | { | |
385 | uintptr_t osv = (uintptr_t)os; | |
386 | uint64_t crc = -1ULL; | |
387 | ||
388 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
389 | /* | |
390 | * The low 6 bits of the pointer don't have much entropy, because | |
391 | * the objset_t is larger than 2^6 bytes long. | |
392 | */ | |
393 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; | |
394 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; | |
395 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; | |
396 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; | |
397 | ||
398 | crc ^= (osv>>14) ^ (obj>>24); | |
399 | ||
400 | return (crc); | |
401 | } | |
402 | ||
65c7cc49 | 403 | static unsigned int |
64fc7762 MA |
404 | dnode_multilist_index_func(multilist_t *ml, void *obj) |
405 | { | |
406 | dnode_t *dn = obj; | |
5b7053a9 AM |
407 | |
408 | /* | |
409 | * The low order bits of the hash value are thought to be | |
410 | * distributed evenly. Otherwise, in the case that the multilist | |
411 | * has a power of two number of sublists, each sublists' usage | |
412 | * would not be evenly distributed. In this context full 64bit | |
413 | * division would be a waste of time, so limit it to 32 bits. | |
414 | */ | |
415 | return ((unsigned int)dnode_hash(dn->dn_objset, dn->dn_object) % | |
64fc7762 MA |
416 | multilist_get_num_sublists(ml)); |
417 | } | |
418 | ||
c9d62d13 GA |
419 | static inline boolean_t |
420 | dmu_os_is_l2cacheable(objset_t *os) | |
421 | { | |
422 | vdev_t *vd = NULL; | |
423 | zfs_cache_type_t cache = os->os_secondary_cache; | |
424 | blkptr_t *bp = os->os_rootbp; | |
425 | ||
426 | if (bp != NULL && !BP_IS_HOLE(bp)) { | |
427 | uint64_t vdev = DVA_GET_VDEV(bp->blk_dva); | |
428 | vdev_t *rvd = os->os_spa->spa_root_vdev; | |
429 | ||
430 | if (vdev < rvd->vdev_children) | |
431 | vd = rvd->vdev_child[vdev]; | |
432 | ||
433 | if (cache == ZFS_CACHE_ALL || cache == ZFS_CACHE_METADATA) { | |
434 | if (vd == NULL) | |
435 | return (B_TRUE); | |
436 | ||
437 | if ((vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL && | |
438 | vd->vdev_alloc_bias != VDEV_BIAS_DEDUP) || | |
439 | l2arc_exclude_special == 0) | |
440 | return (B_TRUE); | |
441 | } | |
442 | } | |
443 | ||
444 | return (B_FALSE); | |
445 | } | |
446 | ||
a1d477c2 MA |
447 | /* |
448 | * Instantiates the objset_t in-memory structure corresponding to the | |
449 | * objset_phys_t that's pointed to by the specified blkptr_t. | |
450 | */ | |
34dc7c2f BB |
451 | int |
452 | dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
428870ff | 453 | objset_t **osp) |
34dc7c2f | 454 | { |
428870ff | 455 | objset_t *os; |
b128c09f | 456 | int i, err; |
34dc7c2f BB |
457 | |
458 | ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); | |
30af21b0 | 459 | ASSERT(!BP_IS_REDACTED(bp)); |
34dc7c2f | 460 | |
0fdd6106 MA |
461 | /* |
462 | * We need the pool config lock to get properties. | |
463 | */ | |
464 | ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool)); | |
465 | ||
a1d477c2 MA |
466 | /* |
467 | * The $ORIGIN dataset (if it exists) doesn't have an associated | |
468 | * objset, so there's no reason to open it. The $ORIGIN dataset | |
469 | * will not exist on pools older than SPA_VERSION_ORIGIN. | |
470 | */ | |
471 | if (ds != NULL && spa_get_dsl(spa) != NULL && | |
472 | spa_get_dsl(spa)->dp_origin_snap != NULL) { | |
473 | ASSERT3P(ds->ds_dir, !=, | |
474 | spa_get_dsl(spa)->dp_origin_snap->ds_dir); | |
475 | } | |
476 | ||
79c76d5b | 477 | os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); |
428870ff BB |
478 | os->os_dsl_dataset = ds; |
479 | os->os_spa = spa; | |
480 | os->os_rootbp = bp; | |
481 | if (!BP_IS_HOLE(os->os_rootbp)) { | |
2a432414 | 482 | arc_flags_t aflags = ARC_FLAG_WAIT; |
5dbd68a3 | 483 | zbookmark_phys_t zb; |
9c5167d1 | 484 | int size; |
4938d01d | 485 | zio_flag_t zio_flags = ZIO_FLAG_CANFAIL; |
428870ff BB |
486 | SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, |
487 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
488 | ||
c9d62d13 | 489 | if (dmu_os_is_l2cacheable(os)) |
2a432414 | 490 | aflags |= ARC_FLAG_L2CACHE; |
34dc7c2f | 491 | |
b5256303 TC |
492 | if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { |
493 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
494 | ASSERT(BP_IS_AUTHENTICATED(bp)); | |
495 | zio_flags |= ZIO_FLAG_RAW; | |
496 | } | |
497 | ||
428870ff | 498 | dprintf_bp(os->os_rootbp, "reading %s", ""); |
294f6806 | 499 | err = arc_read(NULL, spa, os->os_rootbp, |
428870ff | 500 | arc_getbuf_func, &os->os_phys_buf, |
b5256303 | 501 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
13fe0198 | 502 | if (err != 0) { |
428870ff | 503 | kmem_free(os, sizeof (objset_t)); |
b128c09f BB |
504 | /* convert checksum errors into IO errors */ |
505 | if (err == ECKSUM) | |
2e528b49 | 506 | err = SET_ERROR(EIO); |
34dc7c2f BB |
507 | return (err); |
508 | } | |
9babb374 | 509 | |
9c5167d1 NF |
510 | if (spa_version(spa) < SPA_VERSION_USERSPACE) |
511 | size = OBJSET_PHYS_SIZE_V1; | |
512 | else if (!spa_feature_is_enabled(spa, | |
513 | SPA_FEATURE_PROJECT_QUOTA)) | |
514 | size = OBJSET_PHYS_SIZE_V2; | |
515 | else | |
516 | size = sizeof (objset_phys_t); | |
517 | ||
9babb374 | 518 | /* Increase the blocksize if we are permitted. */ |
9c5167d1 | 519 | if (arc_buf_size(os->os_phys_buf) < size) { |
2aa34383 | 520 | arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, |
9c5167d1 | 521 | ARC_BUFC_METADATA, size); |
861166b0 AZ |
522 | memset(buf->b_data, 0, size); |
523 | memcpy(buf->b_data, os->os_phys_buf->b_data, | |
428870ff | 524 | arc_buf_size(os->os_phys_buf)); |
d3c2ae1c | 525 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 526 | os->os_phys_buf = buf; |
9babb374 BB |
527 | } |
528 | ||
428870ff BB |
529 | os->os_phys = os->os_phys_buf->b_data; |
530 | os->os_flags = os->os_phys->os_flags; | |
34dc7c2f | 531 | } else { |
9babb374 | 532 | int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? |
9c5167d1 | 533 | sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; |
2aa34383 DK |
534 | os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, |
535 | ARC_BUFC_METADATA, size); | |
428870ff | 536 | os->os_phys = os->os_phys_buf->b_data; |
861166b0 | 537 | memset(os->os_phys, 0, size); |
34dc7c2f | 538 | } |
2e5dc449 MA |
539 | /* |
540 | * These properties will be filled in by the logic in zfs_get_zplprop() | |
541 | * when they are queried for the first time. | |
542 | */ | |
543 | os->os_version = OBJSET_PROP_UNINITIALIZED; | |
544 | os->os_normalization = OBJSET_PROP_UNINITIALIZED; | |
545 | os->os_utf8only = OBJSET_PROP_UNINITIALIZED; | |
546 | os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; | |
34dc7c2f BB |
547 | |
548 | /* | |
549 | * Note: the changed_cb will be called once before the register | |
550 | * func returns, thus changing the checksum/compression from the | |
b128c09f BB |
551 | * default (fletcher2/off). Snapshots don't need to know about |
552 | * checksum/compression/copies. | |
34dc7c2f | 553 | */ |
9b67f605 | 554 | if (ds != NULL) { |
b5256303 TC |
555 | os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); |
556 | ||
13fe0198 MA |
557 | err = dsl_prop_register(ds, |
558 | zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), | |
428870ff | 559 | primary_cache_changed_cb, os); |
13fe0198 MA |
560 | if (err == 0) { |
561 | err = dsl_prop_register(ds, | |
562 | zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), | |
428870ff | 563 | secondary_cache_changed_cb, os); |
13fe0198 | 564 | } |
0c66c32d | 565 | if (!ds->ds_is_snapshot) { |
13fe0198 MA |
566 | if (err == 0) { |
567 | err = dsl_prop_register(ds, | |
568 | zfs_prop_to_name(ZFS_PROP_CHECKSUM), | |
428870ff | 569 | checksum_changed_cb, os); |
13fe0198 MA |
570 | } |
571 | if (err == 0) { | |
572 | err = dsl_prop_register(ds, | |
573 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
428870ff | 574 | compression_changed_cb, os); |
13fe0198 MA |
575 | } |
576 | if (err == 0) { | |
577 | err = dsl_prop_register(ds, | |
578 | zfs_prop_to_name(ZFS_PROP_COPIES), | |
428870ff | 579 | copies_changed_cb, os); |
13fe0198 MA |
580 | } |
581 | if (err == 0) { | |
582 | err = dsl_prop_register(ds, | |
583 | zfs_prop_to_name(ZFS_PROP_DEDUP), | |
428870ff | 584 | dedup_changed_cb, os); |
13fe0198 MA |
585 | } |
586 | if (err == 0) { | |
587 | err = dsl_prop_register(ds, | |
588 | zfs_prop_to_name(ZFS_PROP_LOGBIAS), | |
428870ff | 589 | logbias_changed_cb, os); |
13fe0198 MA |
590 | } |
591 | if (err == 0) { | |
592 | err = dsl_prop_register(ds, | |
593 | zfs_prop_to_name(ZFS_PROP_SYNC), | |
428870ff | 594 | sync_changed_cb, os); |
13fe0198 | 595 | } |
faf0f58c MA |
596 | if (err == 0) { |
597 | err = dsl_prop_register(ds, | |
598 | zfs_prop_to_name( | |
599 | ZFS_PROP_REDUNDANT_METADATA), | |
600 | redundant_metadata_changed_cb, os); | |
601 | } | |
f1512ee6 MA |
602 | if (err == 0) { |
603 | err = dsl_prop_register(ds, | |
604 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
605 | recordsize_changed_cb, os); | |
606 | } | |
50c957f7 NB |
607 | if (err == 0) { |
608 | err = dsl_prop_register(ds, | |
609 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
610 | dnodesize_changed_cb, os); | |
611 | } | |
cc99f275 DB |
612 | if (err == 0) { |
613 | err = dsl_prop_register(ds, | |
614 | zfs_prop_to_name( | |
615 | ZFS_PROP_SPECIAL_SMALL_BLOCKS), | |
616 | smallblk_changed_cb, os); | |
617 | } | |
b128c09f | 618 | } |
13fe0198 | 619 | if (err != 0) { |
d3c2ae1c | 620 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 621 | kmem_free(os, sizeof (objset_t)); |
34dc7c2f BB |
622 | return (err); |
623 | } | |
9b67f605 | 624 | } else { |
34dc7c2f | 625 | /* It's the meta-objset. */ |
428870ff | 626 | os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; |
99197f03 | 627 | os->os_compress = ZIO_COMPRESS_ON; |
10b3c7f5 | 628 | os->os_complevel = ZIO_COMPLEVEL_DEFAULT; |
b5256303 | 629 | os->os_encrypted = B_FALSE; |
428870ff BB |
630 | os->os_copies = spa_max_replication(spa); |
631 | os->os_dedup_checksum = ZIO_CHECKSUM_OFF; | |
faf0f58c MA |
632 | os->os_dedup_verify = B_FALSE; |
633 | os->os_logbias = ZFS_LOGBIAS_LATENCY; | |
634 | os->os_sync = ZFS_SYNC_STANDARD; | |
428870ff BB |
635 | os->os_primary_cache = ZFS_CACHE_ALL; |
636 | os->os_secondary_cache = ZFS_CACHE_ALL; | |
50c957f7 | 637 | os->os_dnodesize = DNODE_MIN_SIZE; |
34dc7c2f BB |
638 | } |
639 | ||
0c66c32d | 640 | if (ds == NULL || !ds->ds_is_snapshot) |
572e2857 | 641 | os->os_zil_header = os->os_phys->os_zil_header; |
428870ff | 642 | os->os_zil = zil_alloc(os, &os->os_zil_header); |
34dc7c2f BB |
643 | |
644 | for (i = 0; i < TXG_SIZE; i++) { | |
ffdf019c | 645 | multilist_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), |
64fc7762 MA |
646 | offsetof(dnode_t, dn_dirty_link[i]), |
647 | dnode_multilist_index_func); | |
34dc7c2f | 648 | } |
428870ff | 649 | list_create(&os->os_dnodes, sizeof (dnode_t), |
34dc7c2f | 650 | offsetof(dnode_t, dn_link)); |
428870ff | 651 | list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), |
34dc7c2f BB |
652 | offsetof(dmu_buf_impl_t, db_link)); |
653 | ||
0c66c32d JG |
654 | list_link_init(&os->os_evicting_node); |
655 | ||
428870ff | 656 | mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); |
64fc7762 | 657 | mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); |
428870ff BB |
658 | mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); |
659 | mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); | |
dbeb8796 MA |
660 | os->os_obj_next_percpu_len = boot_ncpus; |
661 | os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * | |
662 | sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); | |
428870ff | 663 | |
0c66c32d JG |
664 | dnode_special_open(os, &os->os_phys->os_meta_dnode, |
665 | DMU_META_DNODE_OBJECT, &os->os_meta_dnode); | |
9c5167d1 | 666 | if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { |
0c66c32d JG |
667 | dnode_special_open(os, &os->os_phys->os_userused_dnode, |
668 | DMU_USERUSED_OBJECT, &os->os_userused_dnode); | |
669 | dnode_special_open(os, &os->os_phys->os_groupused_dnode, | |
670 | DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); | |
9c5167d1 NF |
671 | if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) |
672 | dnode_special_open(os, | |
673 | &os->os_phys->os_projectused_dnode, | |
674 | DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); | |
9babb374 | 675 | } |
34dc7c2f | 676 | |
1de321e6 JX |
677 | mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); |
678 | ||
428870ff | 679 | *osp = os; |
34dc7c2f BB |
680 | return (0); |
681 | } | |
682 | ||
428870ff BB |
683 | int |
684 | dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) | |
34dc7c2f | 685 | { |
428870ff | 686 | int err = 0; |
34dc7c2f | 687 | |
47dfff3b | 688 | /* |
0fdd6106 MA |
689 | * We need the pool_config lock to manipulate the dsl_dataset_t. |
690 | * Even if the dataset is long-held, we need the pool_config lock | |
691 | * to open the objset, as it needs to get properties. | |
47dfff3b | 692 | */ |
0fdd6106 | 693 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); |
47dfff3b | 694 | |
34dc7c2f | 695 | mutex_enter(&ds->ds_opening_lock); |
9b67f605 MA |
696 | if (ds->ds_objset == NULL) { |
697 | objset_t *os; | |
cc9bb3e5 | 698 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
34dc7c2f | 699 | err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), |
9b67f605 | 700 | ds, dsl_dataset_get_blkptr(ds), &os); |
cc9bb3e5 | 701 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
9b67f605 MA |
702 | |
703 | if (err == 0) { | |
704 | mutex_enter(&ds->ds_lock); | |
705 | ASSERT(ds->ds_objset == NULL); | |
706 | ds->ds_objset = os; | |
707 | mutex_exit(&ds->ds_lock); | |
708 | } | |
34dc7c2f | 709 | } |
9b67f605 | 710 | *osp = ds->ds_objset; |
34dc7c2f | 711 | mutex_exit(&ds->ds_opening_lock); |
428870ff | 712 | return (err); |
34dc7c2f BB |
713 | } |
714 | ||
13fe0198 MA |
715 | /* |
716 | * Holds the pool while the objset is held. Therefore only one objset | |
717 | * can be held at a time. | |
718 | */ | |
34dc7c2f | 719 | int |
a926aab9 | 720 | dmu_objset_hold_flags(const char *name, boolean_t decrypt, const void *tag, |
b5256303 | 721 | objset_t **osp) |
34dc7c2f | 722 | { |
13fe0198 | 723 | dsl_pool_t *dp; |
428870ff | 724 | dsl_dataset_t *ds; |
34dc7c2f | 725 | int err; |
40ab927a | 726 | ds_hold_flags_t flags; |
34dc7c2f | 727 | |
40ab927a | 728 | flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; |
13fe0198 MA |
729 | err = dsl_pool_hold(name, tag, &dp); |
730 | if (err != 0) | |
731 | return (err); | |
b5256303 | 732 | err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); |
13fe0198 MA |
733 | if (err != 0) { |
734 | dsl_pool_rele(dp, tag); | |
428870ff | 735 | return (err); |
13fe0198 | 736 | } |
428870ff BB |
737 | |
738 | err = dmu_objset_from_ds(ds, osp); | |
13fe0198 | 739 | if (err != 0) { |
428870ff | 740 | dsl_dataset_rele(ds, tag); |
13fe0198 MA |
741 | dsl_pool_rele(dp, tag); |
742 | } | |
428870ff | 743 | |
34dc7c2f BB |
744 | return (err); |
745 | } | |
746 | ||
b5256303 | 747 | int |
a926aab9 | 748 | dmu_objset_hold(const char *name, const void *tag, objset_t **osp) |
b5256303 TC |
749 | { |
750 | return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); | |
751 | } | |
752 | ||
9c43027b AJ |
753 | static int |
754 | dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, | |
a926aab9 | 755 | boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) |
9c43027b | 756 | { |
14e4e3cb | 757 | (void) tag; |
9c43027b | 758 | |
14e4e3cb | 759 | int err = dmu_objset_from_ds(ds, osp); |
9c43027b | 760 | if (err != 0) { |
b5256303 | 761 | return (err); |
9c43027b | 762 | } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { |
9c43027b AJ |
763 | return (SET_ERROR(EINVAL)); |
764 | } else if (!readonly && dsl_dataset_is_snapshot(ds)) { | |
9c43027b | 765 | return (SET_ERROR(EROFS)); |
ae76f45c TC |
766 | } else if (!readonly && decrypt && |
767 | dsl_dir_incompatible_encryption_version(ds->ds_dir)) { | |
768 | return (SET_ERROR(EROFS)); | |
9c43027b | 769 | } |
b5256303 TC |
770 | |
771 | /* if we are decrypting, we can now check MACs in os->os_phys_buf */ | |
772 | if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { | |
a2c2ed1b TC |
773 | zbookmark_phys_t zb; |
774 | ||
775 | SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, | |
776 | ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
b5256303 | 777 | err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, |
a2c2ed1b | 778 | &zb, B_FALSE); |
b5256303 TC |
779 | if (err != 0) |
780 | return (err); | |
781 | ||
782 | ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); | |
783 | } | |
784 | ||
785 | return (0); | |
9c43027b AJ |
786 | } |
787 | ||
13fe0198 MA |
788 | /* |
789 | * dsl_pool must not be held when this is called. | |
790 | * Upon successful return, there will be a longhold on the dataset, | |
791 | * and the dsl_pool will not be held. | |
792 | */ | |
34dc7c2f | 793 | int |
428870ff | 794 | dmu_objset_own(const char *name, dmu_objset_type_t type, |
a926aab9 | 795 | boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) |
34dc7c2f | 796 | { |
13fe0198 | 797 | dsl_pool_t *dp; |
34dc7c2f BB |
798 | dsl_dataset_t *ds; |
799 | int err; | |
40ab927a | 800 | ds_hold_flags_t flags; |
34dc7c2f | 801 | |
40ab927a | 802 | flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; |
13fe0198 MA |
803 | err = dsl_pool_hold(name, FTAG, &dp); |
804 | if (err != 0) | |
805 | return (err); | |
b5256303 | 806 | err = dsl_dataset_own(dp, name, flags, tag, &ds); |
13fe0198 MA |
807 | if (err != 0) { |
808 | dsl_pool_rele(dp, FTAG); | |
34dc7c2f | 809 | return (err); |
13fe0198 | 810 | } |
b5256303 TC |
811 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
812 | if (err != 0) { | |
813 | dsl_dataset_disown(ds, flags, tag); | |
814 | dsl_pool_rele(dp, FTAG); | |
815 | return (err); | |
816 | } | |
817 | ||
163a8c28 TC |
818 | /* |
819 | * User accounting requires the dataset to be decrypted and rw. | |
820 | * We also don't begin user accounting during claiming to help | |
821 | * speed up pool import times and to keep this txg reserved | |
822 | * completely for recovery work. | |
823 | */ | |
4072f465 | 824 | if (!readonly && !dp->dp_spa->spa_claiming && |
825 | (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) { | |
826 | if (dmu_objset_userobjspace_upgradable(*osp) || | |
827 | dmu_objset_projectquota_upgradable(*osp)) { | |
828 | dmu_objset_id_quota_upgrade(*osp); | |
829 | } else if (dmu_objset_userused_enabled(*osp)) { | |
830 | dmu_objset_userspace_upgrade(*osp); | |
831 | } | |
832 | } | |
1de321e6 | 833 | |
c0daec32 | 834 | dsl_pool_rele(dp, FTAG); |
b5256303 | 835 | return (0); |
34dc7c2f BB |
836 | } |
837 | ||
9c43027b AJ |
838 | int |
839 | dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, | |
a926aab9 | 840 | boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) |
9c43027b AJ |
841 | { |
842 | dsl_dataset_t *ds; | |
843 | int err; | |
40ab927a | 844 | ds_hold_flags_t flags; |
9c43027b | 845 | |
40ab927a | 846 | flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; |
b5256303 | 847 | err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); |
9c43027b AJ |
848 | if (err != 0) |
849 | return (err); | |
850 | ||
b5256303 TC |
851 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); |
852 | if (err != 0) { | |
853 | dsl_dataset_disown(ds, flags, tag); | |
854 | return (err); | |
855 | } | |
856 | ||
857 | return (0); | |
9c43027b AJ |
858 | } |
859 | ||
34dc7c2f | 860 | void |
a926aab9 | 861 | dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, const void *tag) |
34dc7c2f | 862 | { |
40ab927a | 863 | ds_hold_flags_t flags; |
13fe0198 | 864 | dsl_pool_t *dp = dmu_objset_pool(os); |
40ab927a TS |
865 | |
866 | flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; | |
b5256303 | 867 | dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); |
13fe0198 | 868 | dsl_pool_rele(dp, tag); |
428870ff | 869 | } |
b128c09f | 870 | |
b5256303 | 871 | void |
a926aab9 | 872 | dmu_objset_rele(objset_t *os, const void *tag) |
b5256303 TC |
873 | { |
874 | dmu_objset_rele_flags(os, B_FALSE, tag); | |
875 | } | |
876 | ||
831baf06 KW |
877 | /* |
878 | * When we are called, os MUST refer to an objset associated with a dataset | |
879 | * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner | |
880 | * == tag. We will then release and reacquire ownership of the dataset while | |
881 | * holding the pool config_rwlock to avoid intervening namespace or ownership | |
882 | * changes may occur. | |
883 | * | |
884 | * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to | |
885 | * release the hold on its dataset and acquire a new one on the dataset of the | |
886 | * same name so that it can be partially torn down and reconstructed. | |
887 | */ | |
888 | void | |
5e00213e | 889 | dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, |
a926aab9 | 890 | boolean_t decrypt, const void *tag) |
831baf06 KW |
891 | { |
892 | dsl_pool_t *dp; | |
eca7b760 | 893 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
40ab927a | 894 | ds_hold_flags_t flags; |
831baf06 | 895 | |
40ab927a | 896 | flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; |
831baf06 KW |
897 | VERIFY3P(ds, !=, NULL); |
898 | VERIFY3P(ds->ds_owner, ==, tag); | |
899 | VERIFY(dsl_dataset_long_held(ds)); | |
900 | ||
901 | dsl_dataset_name(ds, name); | |
5e00213e | 902 | dp = ds->ds_dir->dd_pool; |
831baf06 | 903 | dsl_pool_config_enter(dp, FTAG); |
40ab927a TS |
904 | dsl_dataset_disown(ds, flags, tag); |
905 | VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds)); | |
831baf06 KW |
906 | dsl_pool_config_exit(dp, FTAG); |
907 | } | |
908 | ||
428870ff | 909 | void |
a926aab9 | 910 | dmu_objset_disown(objset_t *os, boolean_t decrypt, const void *tag) |
428870ff | 911 | { |
40ab927a TS |
912 | ds_hold_flags_t flags; |
913 | ||
914 | flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; | |
1de321e6 JX |
915 | /* |
916 | * Stop upgrading thread | |
917 | */ | |
918 | dmu_objset_upgrade_stop(os); | |
40ab927a | 919 | dsl_dataset_disown(os->os_dsl_dataset, flags, tag); |
34dc7c2f BB |
920 | } |
921 | ||
13fe0198 | 922 | void |
34dc7c2f BB |
923 | dmu_objset_evict_dbufs(objset_t *os) |
924 | { | |
0c66c32d | 925 | dnode_t *dn_marker; |
34dc7c2f BB |
926 | dnode_t *dn; |
927 | ||
0c66c32d | 928 | dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); |
34dc7c2f | 929 | |
0c66c32d JG |
930 | mutex_enter(&os->os_lock); |
931 | dn = list_head(&os->os_dnodes); | |
932 | while (dn != NULL) { | |
933 | /* | |
934 | * Skip dnodes without holds. We have to do this dance | |
935 | * because dnode_add_ref() only works if there is already a | |
936 | * hold. If the dnode has no holds, then it has no dbufs. | |
937 | */ | |
938 | if (dnode_add_ref(dn, FTAG)) { | |
939 | list_insert_after(&os->os_dnodes, dn, dn_marker); | |
940 | mutex_exit(&os->os_lock); | |
34dc7c2f | 941 | |
0c66c32d JG |
942 | dnode_evict_dbufs(dn); |
943 | dnode_rele(dn, FTAG); | |
34dc7c2f | 944 | |
0c66c32d JG |
945 | mutex_enter(&os->os_lock); |
946 | dn = list_next(&os->os_dnodes, dn_marker); | |
947 | list_remove(&os->os_dnodes, dn_marker); | |
948 | } else { | |
949 | dn = list_next(&os->os_dnodes, dn); | |
950 | } | |
951 | } | |
952 | mutex_exit(&os->os_lock); | |
34dc7c2f | 953 | |
0c66c32d | 954 | kmem_free(dn_marker, sizeof (dnode_t)); |
34dc7c2f | 955 | |
0c66c32d | 956 | if (DMU_USERUSED_DNODE(os) != NULL) { |
9c5167d1 NF |
957 | if (DMU_PROJECTUSED_DNODE(os) != NULL) |
958 | dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); | |
0c66c32d JG |
959 | dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); |
960 | dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); | |
34dc7c2f | 961 | } |
0c66c32d | 962 | dnode_evict_dbufs(DMU_META_DNODE(os)); |
34dc7c2f BB |
963 | } |
964 | ||
0c66c32d JG |
965 | /* |
966 | * Objset eviction processing is split into into two pieces. | |
967 | * The first marks the objset as evicting, evicts any dbufs that | |
968 | * have a refcount of zero, and then queues up the objset for the | |
969 | * second phase of eviction. Once os->os_dnodes has been cleared by | |
970 | * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. | |
971 | * The second phase closes the special dnodes, dequeues the objset from | |
972 | * the list of those undergoing eviction, and finally frees the objset. | |
973 | * | |
974 | * NOTE: Due to asynchronous eviction processing (invocation of | |
975 | * dnode_buf_pageout()), it is possible for the meta dnode for the | |
976 | * objset to have no holds even though os->os_dnodes is not empty. | |
977 | */ | |
34dc7c2f | 978 | void |
428870ff | 979 | dmu_objset_evict(objset_t *os) |
34dc7c2f | 980 | { |
6f1ffb06 MA |
981 | dsl_dataset_t *ds = os->os_dsl_dataset; |
982 | ||
1c27024e | 983 | for (int t = 0; t < TXG_SIZE; t++) |
428870ff | 984 | ASSERT(!dmu_objset_is_dirty(os, t)); |
34dc7c2f | 985 | |
0eb21616 JG |
986 | if (ds) |
987 | dsl_prop_unregister_all(ds, os); | |
34dc7c2f | 988 | |
428870ff BB |
989 | if (os->os_sa) |
990 | sa_tear_down(os); | |
991 | ||
13fe0198 | 992 | dmu_objset_evict_dbufs(os); |
34dc7c2f | 993 | |
0c66c32d JG |
994 | mutex_enter(&os->os_lock); |
995 | spa_evicting_os_register(os->os_spa, os); | |
996 | if (list_is_empty(&os->os_dnodes)) { | |
997 | mutex_exit(&os->os_lock); | |
998 | dmu_objset_evict_done(os); | |
999 | } else { | |
1000 | mutex_exit(&os->os_lock); | |
1001 | } | |
b5256303 TC |
1002 | |
1003 | ||
0c66c32d JG |
1004 | } |
1005 | ||
1006 | void | |
1007 | dmu_objset_evict_done(objset_t *os) | |
1008 | { | |
1009 | ASSERT3P(list_head(&os->os_dnodes), ==, NULL); | |
1010 | ||
572e2857 BB |
1011 | dnode_special_close(&os->os_meta_dnode); |
1012 | if (DMU_USERUSED_DNODE(os)) { | |
9c5167d1 NF |
1013 | if (DMU_PROJECTUSED_DNODE(os)) |
1014 | dnode_special_close(&os->os_projectused_dnode); | |
572e2857 BB |
1015 | dnode_special_close(&os->os_userused_dnode); |
1016 | dnode_special_close(&os->os_groupused_dnode); | |
9babb374 | 1017 | } |
428870ff BB |
1018 | zil_free(os->os_zil); |
1019 | ||
d3c2ae1c | 1020 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
572e2857 BB |
1021 | |
1022 | /* | |
1023 | * This is a barrier to prevent the objset from going away in | |
1024 | * dnode_move() until we can safely ensure that the objset is still in | |
1025 | * use. We consider the objset valid before the barrier and invalid | |
1026 | * after the barrier. | |
1027 | */ | |
1028 | rw_enter(&os_lock, RW_READER); | |
1029 | rw_exit(&os_lock); | |
1030 | ||
dbeb8796 MA |
1031 | kmem_free(os->os_obj_next_percpu, |
1032 | os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); | |
1033 | ||
428870ff | 1034 | mutex_destroy(&os->os_lock); |
64fc7762 | 1035 | mutex_destroy(&os->os_userused_lock); |
428870ff BB |
1036 | mutex_destroy(&os->os_obj_lock); |
1037 | mutex_destroy(&os->os_user_ptr_lock); | |
c17486b2 | 1038 | mutex_destroy(&os->os_upgrade_lock); |
ffdf019c AM |
1039 | for (int i = 0; i < TXG_SIZE; i++) |
1040 | multilist_destroy(&os->os_dirty_dnodes[i]); | |
0c66c32d | 1041 | spa_evicting_os_deregister(os->os_spa, os); |
428870ff BB |
1042 | kmem_free(os, sizeof (objset_t)); |
1043 | } | |
9babb374 | 1044 | |
6413c95f | 1045 | inode_timespec_t |
428870ff BB |
1046 | dmu_objset_snap_cmtime(objset_t *os) |
1047 | { | |
1048 | return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); | |
34dc7c2f BB |
1049 | } |
1050 | ||
428870ff | 1051 | objset_t * |
b5256303 TC |
1052 | dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, |
1053 | dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) | |
34dc7c2f | 1054 | { |
428870ff | 1055 | objset_t *os; |
34dc7c2f BB |
1056 | dnode_t *mdn; |
1057 | ||
1058 | ASSERT(dmu_tx_is_syncing(tx)); | |
13fe0198 | 1059 | |
b5256303 TC |
1060 | if (blksz == 0) |
1061 | blksz = DNODE_BLOCK_SIZE; | |
4807c0ba | 1062 | if (ibs == 0) |
b5256303 TC |
1063 | ibs = DN_MAX_INDBLKSHIFT; |
1064 | ||
572e2857 | 1065 | if (ds != NULL) |
13fe0198 | 1066 | VERIFY0(dmu_objset_from_ds(ds, &os)); |
572e2857 | 1067 | else |
13fe0198 | 1068 | VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); |
572e2857 BB |
1069 | |
1070 | mdn = DMU_META_DNODE(os); | |
34dc7c2f | 1071 | |
b5256303 TC |
1072 | dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, |
1073 | DNODE_MIN_SLOTS, tx); | |
34dc7c2f BB |
1074 | |
1075 | /* | |
1076 | * We don't want to have to increase the meta-dnode's nlevels | |
e1cfd73f | 1077 | * later, because then we could do it in quiescing context while |
34dc7c2f BB |
1078 | * we are also accessing it in open context. |
1079 | * | |
1080 | * This precaution is not necessary for the MOS (ds == NULL), | |
1081 | * because the MOS is only updated in syncing context. | |
1082 | * This is most fortunate: the MOS is the only objset that | |
1083 | * needs to be synced multiple times as spa_sync() iterates | |
1084 | * to convergence, so minimizing its dn_nlevels matters. | |
1085 | */ | |
1086 | if (ds != NULL) { | |
b5256303 TC |
1087 | if (levels == 0) { |
1088 | levels = 1; | |
1089 | ||
1090 | /* | |
1091 | * Determine the number of levels necessary for the | |
1092 | * meta-dnode to contain DN_MAX_OBJECT dnodes. Note | |
1093 | * that in order to ensure that we do not overflow | |
1094 | * 64 bits, there has to be a nlevels that gives us a | |
1095 | * number of blocks > DN_MAX_OBJECT but < 2^64. | |
1096 | * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) | |
1097 | * (10) must be less than (64 - log2(DN_MAX_OBJECT)) | |
1098 | * (16). | |
1099 | */ | |
1100 | while ((uint64_t)mdn->dn_nblkptr << | |
1101 | (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * | |
1102 | (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < | |
1103 | DN_MAX_OBJECT) | |
1104 | levels++; | |
1105 | } | |
34dc7c2f BB |
1106 | |
1107 | mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = | |
1108 | mdn->dn_nlevels = levels; | |
1109 | } | |
1110 | ||
1111 | ASSERT(type != DMU_OST_NONE); | |
1112 | ASSERT(type != DMU_OST_ANY); | |
1113 | ASSERT(type < DMU_OST_NUMTYPES); | |
428870ff | 1114 | os->os_phys->os_type = type; |
b5256303 TC |
1115 | |
1116 | /* | |
1117 | * Enable user accounting if it is enabled and this is not an | |
1118 | * encrypted receive. | |
1119 | */ | |
1120 | if (dmu_objset_userused_enabled(os) && | |
1121 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
428870ff | 1122 | os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
1de321e6 | 1123 | if (dmu_objset_userobjused_enabled(os)) { |
d52d80b7 PD |
1124 | ds->ds_feature_activation[ |
1125 | SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; | |
1de321e6 JX |
1126 | os->os_phys->os_flags |= |
1127 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
1128 | } | |
9c5167d1 | 1129 | if (dmu_objset_projectquota_enabled(os)) { |
d52d80b7 PD |
1130 | ds->ds_feature_activation[ |
1131 | SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; | |
9c5167d1 NF |
1132 | os->os_phys->os_flags |= |
1133 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
1134 | } | |
428870ff | 1135 | os->os_flags = os->os_phys->os_flags; |
9babb374 | 1136 | } |
34dc7c2f BB |
1137 | |
1138 | dsl_dataset_dirty(ds, tx); | |
1139 | ||
428870ff | 1140 | return (os); |
34dc7c2f BB |
1141 | } |
1142 | ||
b5256303 TC |
1143 | /* called from dsl for meta-objset */ |
1144 | objset_t * | |
1145 | dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
1146 | dmu_objset_type_t type, dmu_tx_t *tx) | |
1147 | { | |
1148 | return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); | |
1149 | } | |
1150 | ||
13fe0198 MA |
1151 | typedef struct dmu_objset_create_arg { |
1152 | const char *doca_name; | |
1153 | cred_t *doca_cred; | |
e59a377a | 1154 | proc_t *doca_proc; |
13fe0198 MA |
1155 | void (*doca_userfunc)(objset_t *os, void *arg, |
1156 | cred_t *cr, dmu_tx_t *tx); | |
1157 | void *doca_userarg; | |
1158 | dmu_objset_type_t doca_type; | |
1159 | uint64_t doca_flags; | |
b5256303 | 1160 | dsl_crypto_params_t *doca_dcp; |
13fe0198 | 1161 | } dmu_objset_create_arg_t; |
34dc7c2f | 1162 | |
34dc7c2f | 1163 | static int |
13fe0198 | 1164 | dmu_objset_create_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1165 | { |
13fe0198 MA |
1166 | dmu_objset_create_arg_t *doca = arg; |
1167 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1168 | dsl_dir_t *pdd; | |
d8d418ff | 1169 | dsl_dataset_t *parentds; |
1170 | objset_t *parentos; | |
13fe0198 MA |
1171 | const char *tail; |
1172 | int error; | |
34dc7c2f | 1173 | |
13fe0198 | 1174 | if (strchr(doca->doca_name, '@') != NULL) |
2e528b49 | 1175 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 1176 | |
eca7b760 IK |
1177 | if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) |
1178 | return (SET_ERROR(ENAMETOOLONG)); | |
1179 | ||
a7ed98d8 SD |
1180 | if (dataset_nestcheck(doca->doca_name) != 0) |
1181 | return (SET_ERROR(ENAMETOOLONG)); | |
1182 | ||
13fe0198 MA |
1183 | error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); |
1184 | if (error != 0) | |
1185 | return (error); | |
1186 | if (tail == NULL) { | |
1187 | dsl_dir_rele(pdd, FTAG); | |
2e528b49 | 1188 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1189 | } |
b5256303 | 1190 | |
1fff937a | 1191 | error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); |
b5256303 TC |
1192 | if (error != 0) { |
1193 | dsl_dir_rele(pdd, FTAG); | |
1194 | return (error); | |
1195 | } | |
1196 | ||
788eb90c | 1197 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
e59a377a | 1198 | doca->doca_cred, doca->doca_proc); |
d8d418ff | 1199 | if (error != 0) { |
1200 | dsl_dir_rele(pdd, FTAG); | |
1201 | return (error); | |
1202 | } | |
b5256303 | 1203 | |
d8d418ff | 1204 | /* can't create below anything but filesystems (eg. no ZVOLs) */ |
1205 | error = dsl_dataset_hold_obj(pdd->dd_pool, | |
1206 | dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); | |
1207 | if (error != 0) { | |
1208 | dsl_dir_rele(pdd, FTAG); | |
1209 | return (error); | |
1210 | } | |
1211 | error = dmu_objset_from_ds(parentds, &parentos); | |
1212 | if (error != 0) { | |
1213 | dsl_dataset_rele(parentds, FTAG); | |
1214 | dsl_dir_rele(pdd, FTAG); | |
1215 | return (error); | |
1216 | } | |
1217 | if (dmu_objset_type(parentos) != DMU_OST_ZFS) { | |
1218 | dsl_dataset_rele(parentds, FTAG); | |
1219 | dsl_dir_rele(pdd, FTAG); | |
1220 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); | |
1221 | } | |
1222 | dsl_dataset_rele(parentds, FTAG); | |
13fe0198 | 1223 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f | 1224 | |
788eb90c | 1225 | return (error); |
34dc7c2f BB |
1226 | } |
1227 | ||
1228 | static void | |
13fe0198 | 1229 | dmu_objset_create_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 1230 | { |
13fe0198 MA |
1231 | dmu_objset_create_arg_t *doca = arg; |
1232 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
52ce99dd | 1233 | spa_t *spa = dp->dp_spa; |
13fe0198 MA |
1234 | dsl_dir_t *pdd; |
1235 | const char *tail; | |
6f1ffb06 | 1236 | dsl_dataset_t *ds; |
13fe0198 | 1237 | uint64_t obj; |
6f1ffb06 | 1238 | blkptr_t *bp; |
13fe0198 | 1239 | objset_t *os; |
b5256303 | 1240 | zio_t *rzio; |
34dc7c2f | 1241 | |
13fe0198 | 1242 | VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); |
34dc7c2f | 1243 | |
13fe0198 | 1244 | obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, |
b5256303 | 1245 | doca->doca_cred, doca->doca_dcp, tx); |
34dc7c2f | 1246 | |
b5256303 TC |
1247 | VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, |
1248 | DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); | |
cc9bb3e5 | 1249 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
6f1ffb06 | 1250 | bp = dsl_dataset_get_blkptr(ds); |
52ce99dd | 1251 | os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); |
cc9bb3e5 | 1252 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
34dc7c2f | 1253 | |
13fe0198 MA |
1254 | if (doca->doca_userfunc != NULL) { |
1255 | doca->doca_userfunc(os, doca->doca_userarg, | |
1256 | doca->doca_cred, tx); | |
34dc7c2f BB |
1257 | } |
1258 | ||
b5256303 | 1259 | /* |
4807c0ba | 1260 | * The doca_userfunc() may write out some data that needs to be |
b5256303 TC |
1261 | * encrypted if the dataset is encrypted (specifically the root |
1262 | * directory). This data must be written out before the encryption | |
1263 | * key mapping is removed by dsl_dataset_rele_flags(). Force the | |
1264 | * I/O to occur immediately by invoking the relevant sections of | |
1265 | * dsl_pool_sync(). | |
1266 | */ | |
1267 | if (os->os_encrypted) { | |
1268 | dsl_dataset_t *tmpds = NULL; | |
1269 | boolean_t need_sync_done = B_FALSE; | |
1270 | ||
4807c0ba TC |
1271 | mutex_enter(&ds->ds_lock); |
1272 | ds->ds_owner = FTAG; | |
1273 | mutex_exit(&ds->ds_lock); | |
1274 | ||
52ce99dd | 1275 | rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
4807c0ba TC |
1276 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, |
1277 | tx->tx_txg); | |
b5256303 | 1278 | if (tmpds != NULL) { |
b5256303 TC |
1279 | dsl_dataset_sync(ds, rzio, tx); |
1280 | need_sync_done = B_TRUE; | |
1281 | } | |
1282 | VERIFY0(zio_wait(rzio)); | |
1283 | ||
ba67d821 | 1284 | dmu_objset_sync_done(os, tx); |
b5256303 | 1285 | taskq_wait(dp->dp_sync_taskq); |
52ce99dd TC |
1286 | if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { |
1287 | ASSERT3P(ds->ds_key_mapping, !=, NULL); | |
1288 | key_mapping_rele(spa, ds->ds_key_mapping, ds); | |
1289 | } | |
b5256303 | 1290 | |
52ce99dd | 1291 | rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
4807c0ba TC |
1292 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, |
1293 | tx->tx_txg); | |
b5256303 | 1294 | if (tmpds != NULL) { |
b5256303 TC |
1295 | dmu_buf_rele(ds->ds_dbuf, ds); |
1296 | dsl_dataset_sync(ds, rzio, tx); | |
1297 | } | |
1298 | VERIFY0(zio_wait(rzio)); | |
1299 | ||
52ce99dd TC |
1300 | if (need_sync_done) { |
1301 | ASSERT3P(ds->ds_key_mapping, !=, NULL); | |
1302 | key_mapping_rele(spa, ds->ds_key_mapping, ds); | |
b5256303 | 1303 | dsl_dataset_sync_done(ds, tx); |
52ce99dd | 1304 | } |
4807c0ba TC |
1305 | |
1306 | mutex_enter(&ds->ds_lock); | |
1307 | ds->ds_owner = NULL; | |
1308 | mutex_exit(&ds->ds_lock); | |
b5256303 TC |
1309 | } |
1310 | ||
74756182 | 1311 | spa_history_log_internal_ds(ds, "create", tx, " "); |
a0bd735a | 1312 | |
b5256303 | 1313 | dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); |
13fe0198 | 1314 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f BB |
1315 | } |
1316 | ||
1317 | int | |
428870ff | 1318 | dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, |
b5256303 | 1319 | dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) |
34dc7c2f | 1320 | { |
13fe0198 | 1321 | dmu_objset_create_arg_t doca; |
b5256303 | 1322 | dsl_crypto_params_t tmp_dcp = { 0 }; |
34dc7c2f | 1323 | |
13fe0198 MA |
1324 | doca.doca_name = name; |
1325 | doca.doca_cred = CRED(); | |
e59a377a | 1326 | doca.doca_proc = curproc; |
13fe0198 MA |
1327 | doca.doca_flags = flags; |
1328 | doca.doca_userfunc = func; | |
1329 | doca.doca_userarg = arg; | |
1330 | doca.doca_type = type; | |
34dc7c2f | 1331 | |
b5256303 TC |
1332 | /* |
1333 | * Some callers (mostly for testing) do not provide a dcp on their | |
1334 | * own but various code inside the sync task will require it to be | |
1335 | * allocated. Rather than adding NULL checks throughout this code | |
1336 | * or adding dummy dcp's to all of the callers we simply create a | |
1337 | * dummy one here and use that. This zero dcp will have the same | |
85ce3f4f | 1338 | * effect as asking for inheritance of all encryption params. |
b5256303 TC |
1339 | */ |
1340 | doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; | |
1341 | ||
ec213971 | 1342 | int rv = dsl_sync_task(name, |
3d45fdd6 | 1343 | dmu_objset_create_check, dmu_objset_create_sync, &doca, |
ec213971 MA |
1344 | 6, ZFS_SPACE_CHECK_NORMAL); |
1345 | ||
1346 | if (rv == 0) | |
1347 | zvol_create_minor(name); | |
1348 | return (rv); | |
34dc7c2f BB |
1349 | } |
1350 | ||
13fe0198 MA |
1351 | typedef struct dmu_objset_clone_arg { |
1352 | const char *doca_clone; | |
1353 | const char *doca_origin; | |
1354 | cred_t *doca_cred; | |
e59a377a | 1355 | proc_t *doca_proc; |
13fe0198 MA |
1356 | } dmu_objset_clone_arg_t; |
1357 | ||
13fe0198 MA |
1358 | static int |
1359 | dmu_objset_clone_check(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1360 | { |
13fe0198 | 1361 | dmu_objset_clone_arg_t *doca = arg; |
428870ff BB |
1362 | dsl_dir_t *pdd; |
1363 | const char *tail; | |
13fe0198 MA |
1364 | int error; |
1365 | dsl_dataset_t *origin; | |
1366 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
34dc7c2f | 1367 | |
13fe0198 | 1368 | if (strchr(doca->doca_clone, '@') != NULL) |
2e528b49 | 1369 | return (SET_ERROR(EINVAL)); |
13fe0198 | 1370 | |
eca7b760 IK |
1371 | if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) |
1372 | return (SET_ERROR(ENAMETOOLONG)); | |
1373 | ||
13fe0198 MA |
1374 | error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); |
1375 | if (error != 0) | |
1376 | return (error); | |
428870ff | 1377 | if (tail == NULL) { |
13fe0198 | 1378 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1379 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 1380 | } |
1cddb8c9 | 1381 | |
788eb90c | 1382 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
e59a377a | 1383 | doca->doca_cred, doca->doca_proc); |
788eb90c JJ |
1384 | if (error != 0) { |
1385 | dsl_dir_rele(pdd, FTAG); | |
1386 | return (SET_ERROR(EDQUOT)); | |
1387 | } | |
428870ff | 1388 | |
13fe0198 | 1389 | error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); |
b5256303 TC |
1390 | if (error != 0) { |
1391 | dsl_dir_rele(pdd, FTAG); | |
572e2857 | 1392 | return (error); |
b5256303 | 1393 | } |
572e2857 | 1394 | |
13fe0198 | 1395 | /* You can only clone snapshots, not the head datasets. */ |
0c66c32d | 1396 | if (!origin->ds_is_snapshot) { |
13fe0198 | 1397 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1398 | dsl_dir_rele(pdd, FTAG); |
2e528b49 | 1399 | return (SET_ERROR(EINVAL)); |
428870ff | 1400 | } |
b5256303 | 1401 | |
13fe0198 | 1402 | dsl_dataset_rele(origin, FTAG); |
b5256303 | 1403 | dsl_dir_rele(pdd, FTAG); |
572e2857 | 1404 | |
13fe0198 | 1405 | return (0); |
9babb374 | 1406 | } |
34dc7c2f | 1407 | |
13fe0198 MA |
1408 | static void |
1409 | dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1410 | { |
13fe0198 MA |
1411 | dmu_objset_clone_arg_t *doca = arg; |
1412 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1413 | dsl_dir_t *pdd; | |
1414 | const char *tail; | |
1415 | dsl_dataset_t *origin, *ds; | |
1416 | uint64_t obj; | |
eca7b760 | 1417 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; |
6f1ffb06 | 1418 | |
13fe0198 MA |
1419 | VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); |
1420 | VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); | |
6f1ffb06 | 1421 | |
13fe0198 | 1422 | obj = dsl_dataset_create_sync(pdd, tail, origin, 0, |
b5256303 | 1423 | doca->doca_cred, NULL, tx); |
34dc7c2f | 1424 | |
13fe0198 MA |
1425 | VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); |
1426 | dsl_dataset_name(origin, namebuf); | |
1427 | spa_history_log_internal_ds(ds, "clone", tx, | |
74756182 | 1428 | "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object); |
13fe0198 MA |
1429 | dsl_dataset_rele(ds, FTAG); |
1430 | dsl_dataset_rele(origin, FTAG); | |
1431 | dsl_dir_rele(pdd, FTAG); | |
34dc7c2f BB |
1432 | } |
1433 | ||
1434 | int | |
13fe0198 | 1435 | dmu_objset_clone(const char *clone, const char *origin) |
34dc7c2f | 1436 | { |
13fe0198 | 1437 | dmu_objset_clone_arg_t doca; |
34dc7c2f | 1438 | |
13fe0198 MA |
1439 | doca.doca_clone = clone; |
1440 | doca.doca_origin = origin; | |
1441 | doca.doca_cred = CRED(); | |
e59a377a | 1442 | doca.doca_proc = curproc; |
572e2857 | 1443 | |
ec213971 | 1444 | int rv = dsl_sync_task(clone, |
3d45fdd6 | 1445 | dmu_objset_clone_check, dmu_objset_clone_sync, &doca, |
ec213971 MA |
1446 | 6, ZFS_SPACE_CHECK_NORMAL); |
1447 | ||
1448 | if (rv == 0) | |
1449 | zvol_create_minor(clone); | |
1450 | ||
1451 | return (rv); | |
6f1ffb06 MA |
1452 | } |
1453 | ||
1454 | int | |
1455 | dmu_objset_snapshot_one(const char *fsname, const char *snapname) | |
1456 | { | |
1457 | int err; | |
1458 | char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); | |
1459 | nvlist_t *snaps = fnvlist_alloc(); | |
1460 | ||
1461 | fnvlist_add_boolean(snaps, longsnap); | |
e4f5fa12 | 1462 | kmem_strfree(longsnap); |
13fe0198 MA |
1463 | err = dsl_dataset_snapshot(snaps, NULL, NULL); |
1464 | fnvlist_free(snaps); | |
6f1ffb06 MA |
1465 | return (err); |
1466 | } | |
1467 | ||
1de321e6 JX |
1468 | static void |
1469 | dmu_objset_upgrade_task_cb(void *data) | |
1470 | { | |
1471 | objset_t *os = data; | |
1472 | ||
1473 | mutex_enter(&os->os_upgrade_lock); | |
1474 | os->os_upgrade_status = EINTR; | |
1475 | if (!os->os_upgrade_exit) { | |
39372fa2 AF |
1476 | int status; |
1477 | ||
1de321e6 JX |
1478 | mutex_exit(&os->os_upgrade_lock); |
1479 | ||
39372fa2 AF |
1480 | status = os->os_upgrade_cb(os); |
1481 | ||
1de321e6 | 1482 | mutex_enter(&os->os_upgrade_lock); |
39372fa2 AF |
1483 | |
1484 | os->os_upgrade_status = status; | |
1de321e6 JX |
1485 | } |
1486 | os->os_upgrade_exit = B_TRUE; | |
1487 | os->os_upgrade_id = 0; | |
1488 | mutex_exit(&os->os_upgrade_lock); | |
c0daec32 | 1489 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); |
1de321e6 JX |
1490 | } |
1491 | ||
1492 | static void | |
1493 | dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) | |
1494 | { | |
1495 | if (os->os_upgrade_id != 0) | |
1496 | return; | |
1497 | ||
c0daec32 AB |
1498 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
1499 | dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); | |
1500 | ||
1de321e6 JX |
1501 | mutex_enter(&os->os_upgrade_lock); |
1502 | if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { | |
1503 | os->os_upgrade_exit = B_FALSE; | |
1504 | os->os_upgrade_cb = cb; | |
1505 | os->os_upgrade_id = taskq_dispatch( | |
1506 | os->os_spa->spa_upgrade_taskq, | |
1507 | dmu_objset_upgrade_task_cb, os, TQ_SLEEP); | |
c0daec32 AB |
1508 | if (os->os_upgrade_id == TASKQID_INVALID) { |
1509 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1de321e6 | 1510 | os->os_upgrade_status = ENOMEM; |
c0daec32 | 1511 | } |
39372fa2 AF |
1512 | } else { |
1513 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1de321e6 JX |
1514 | } |
1515 | mutex_exit(&os->os_upgrade_lock); | |
1516 | } | |
1517 | ||
1518 | static void | |
1519 | dmu_objset_upgrade_stop(objset_t *os) | |
1520 | { | |
1521 | mutex_enter(&os->os_upgrade_lock); | |
1522 | os->os_upgrade_exit = B_TRUE; | |
1523 | if (os->os_upgrade_id != 0) { | |
1524 | taskqid_t id = os->os_upgrade_id; | |
1525 | ||
1526 | os->os_upgrade_id = 0; | |
1527 | mutex_exit(&os->os_upgrade_lock); | |
1528 | ||
c0daec32 AB |
1529 | if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { |
1530 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1531 | } | |
4807c0ba | 1532 | txg_wait_synced(os->os_spa->spa_dsl_pool, 0); |
1de321e6 JX |
1533 | } else { |
1534 | mutex_exit(&os->os_upgrade_lock); | |
1535 | } | |
1536 | } | |
1537 | ||
34dc7c2f | 1538 | static void |
64fc7762 | 1539 | dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) |
34dc7c2f BB |
1540 | { |
1541 | dnode_t *dn; | |
1542 | ||
64fc7762 | 1543 | while ((dn = multilist_sublist_head(list)) != NULL) { |
34dc7c2f BB |
1544 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); |
1545 | ASSERT(dn->dn_dbuf->db_data_pending); | |
1546 | /* | |
9babb374 | 1547 | * Initialize dn_zio outside dnode_sync() because the |
93e28d66 | 1548 | * meta-dnode needs to set it outside dnode_sync(). |
34dc7c2f BB |
1549 | */ |
1550 | dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; | |
1551 | ASSERT(dn->dn_zio); | |
1552 | ||
1553 | ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); | |
64fc7762 | 1554 | multilist_sublist_remove(list, dn); |
9babb374 | 1555 | |
edc1e713 | 1556 | /* |
ba67d821 MA |
1557 | * See the comment above dnode_rele_task() for an explanation |
1558 | * of why this dnode hold is always needed (even when not | |
1559 | * doing user accounting). | |
edc1e713 | 1560 | */ |
ffdf019c | 1561 | multilist_t *newlist = &dn->dn_objset->os_synced_dnodes; |
ba67d821 MA |
1562 | (void) dnode_add_ref(dn, newlist); |
1563 | multilist_insert(newlist, dn); | |
9babb374 | 1564 | |
34dc7c2f BB |
1565 | dnode_sync(dn, tx); |
1566 | } | |
1567 | } | |
1568 | ||
34dc7c2f | 1569 | static void |
428870ff | 1570 | dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) |
34dc7c2f | 1571 | { |
14e4e3cb | 1572 | (void) abuf; |
b128c09f | 1573 | blkptr_t *bp = zio->io_bp; |
428870ff | 1574 | objset_t *os = arg; |
34dc7c2f | 1575 | dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; |
b5256303 | 1576 | uint64_t fill = 0; |
34dc7c2f | 1577 | |
9b67f605 | 1578 | ASSERT(!BP_IS_EMBEDDED(bp)); |
13fe0198 MA |
1579 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); |
1580 | ASSERT0(BP_GET_LEVEL(bp)); | |
34dc7c2f BB |
1581 | |
1582 | /* | |
9babb374 BB |
1583 | * Update rootbp fill count: it should be the number of objects |
1584 | * allocated in the object set (not counting the "special" | |
1585 | * objects that are stored in the objset_phys_t -- the meta | |
9c5167d1 | 1586 | * dnode and user/group/project accounting objects). |
34dc7c2f | 1587 | */ |
1c27024e | 1588 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
b5256303 TC |
1589 | fill += BP_GET_FILL(&dnp->dn_blkptr[i]); |
1590 | ||
1591 | BP_SET_FILL(bp, fill); | |
1592 | ||
cc9bb3e5 GM |
1593 | if (os->os_dsl_dataset != NULL) |
1594 | rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); | |
1595 | *os->os_rootbp = *bp; | |
1596 | if (os->os_dsl_dataset != NULL) | |
1597 | rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); | |
428870ff BB |
1598 | } |
1599 | ||
428870ff BB |
1600 | static void |
1601 | dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) | |
1602 | { | |
14e4e3cb | 1603 | (void) abuf; |
428870ff BB |
1604 | blkptr_t *bp = zio->io_bp; |
1605 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1606 | objset_t *os = arg; | |
34dc7c2f | 1607 | |
b128c09f | 1608 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
428870ff | 1609 | ASSERT(BP_EQUAL(bp, bp_orig)); |
b128c09f | 1610 | } else { |
428870ff BB |
1611 | dsl_dataset_t *ds = os->os_dsl_dataset; |
1612 | dmu_tx_t *tx = os->os_synctx; | |
1613 | ||
1614 | (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); | |
1615 | dsl_dataset_block_born(ds, bp, tx); | |
34dc7c2f | 1616 | } |
cc9bb3e5 | 1617 | kmem_free(bp, sizeof (*bp)); |
34dc7c2f BB |
1618 | } |
1619 | ||
64fc7762 MA |
1620 | typedef struct sync_dnodes_arg { |
1621 | multilist_t *sda_list; | |
1622 | int sda_sublist_idx; | |
1623 | multilist_t *sda_newlist; | |
1624 | dmu_tx_t *sda_tx; | |
1625 | } sync_dnodes_arg_t; | |
1626 | ||
1627 | static void | |
1628 | sync_dnodes_task(void *arg) | |
1629 | { | |
1630 | sync_dnodes_arg_t *sda = arg; | |
1631 | ||
1632 | multilist_sublist_t *ms = | |
1633 | multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); | |
1634 | ||
1635 | dmu_objset_sync_dnodes(ms, sda->sda_tx); | |
1636 | ||
1637 | multilist_sublist_unlock(ms); | |
1638 | ||
1639 | kmem_free(sda, sizeof (*sda)); | |
1640 | } | |
1641 | ||
1642 | ||
34dc7c2f BB |
1643 | /* called from dsl */ |
1644 | void | |
428870ff | 1645 | dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) |
34dc7c2f BB |
1646 | { |
1647 | int txgoff; | |
5dbd68a3 | 1648 | zbookmark_phys_t zb; |
428870ff | 1649 | zio_prop_t zp; |
34dc7c2f BB |
1650 | zio_t *zio; |
1651 | list_t *list; | |
1652 | dbuf_dirty_record_t *dr; | |
fc754677 AM |
1653 | int num_sublists; |
1654 | multilist_t *ml; | |
cc9bb3e5 GM |
1655 | blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); |
1656 | *blkptr_copy = *os->os_rootbp; | |
34dc7c2f | 1657 | |
8e739b2c | 1658 | dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", (u_longlong_t)tx->tx_txg); |
34dc7c2f BB |
1659 | |
1660 | ASSERT(dmu_tx_is_syncing(tx)); | |
1661 | /* XXX the write_done callback should really give us the tx... */ | |
1662 | os->os_synctx = tx; | |
1663 | ||
1664 | if (os->os_dsl_dataset == NULL) { | |
1665 | /* | |
1666 | * This is the MOS. If we have upgraded, | |
1667 | * spa_max_replication() could change, so reset | |
1668 | * os_copies here. | |
1669 | */ | |
1670 | os->os_copies = spa_max_replication(os->os_spa); | |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * Create the root block IO | |
1675 | */ | |
428870ff BB |
1676 | SET_BOOKMARK(&zb, os->os_dsl_dataset ? |
1677 | os->os_dsl_dataset->ds_object : DMU_META_OBJSET, | |
1678 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
294f6806 | 1679 | arc_release(os->os_phys_buf, &os->os_phys_buf); |
b128c09f | 1680 | |
82644107 | 1681 | dmu_write_policy(os, NULL, 0, 0, &zp); |
9babb374 | 1682 | |
b5256303 | 1683 | /* |
0c03d21a MA |
1684 | * If we are either claiming the ZIL or doing a raw receive, write |
1685 | * out the os_phys_buf raw. Neither of these actions will effect the | |
1686 | * MAC at this point. | |
b5256303 | 1687 | */ |
0c03d21a MA |
1688 | if (os->os_raw_receive || |
1689 | os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { | |
b5256303 | 1690 | ASSERT(os->os_encrypted); |
b5256303 TC |
1691 | arc_convert_to_raw(os->os_phys_buf, |
1692 | os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, | |
1693 | DMU_OT_OBJSET, NULL, NULL, NULL); | |
1694 | } | |
1695 | ||
428870ff | 1696 | zio = arc_write(pio, os->os_spa, tx->tx_txg, |
ed2f7ba0 | 1697 | blkptr_copy, os->os_phys_buf, B_FALSE, dmu_os_is_l2cacheable(os), |
bc77ba73 PD |
1698 | &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, |
1699 | os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); | |
34dc7c2f BB |
1700 | |
1701 | /* | |
9babb374 | 1702 | * Sync special dnodes - the parent IO for the sync is the root block |
34dc7c2f | 1703 | */ |
572e2857 BB |
1704 | DMU_META_DNODE(os)->dn_zio = zio; |
1705 | dnode_sync(DMU_META_DNODE(os), tx); | |
34dc7c2f | 1706 | |
9babb374 BB |
1707 | os->os_phys->os_flags = os->os_flags; |
1708 | ||
572e2857 BB |
1709 | if (DMU_USERUSED_DNODE(os) && |
1710 | DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1711 | DMU_USERUSED_DNODE(os)->dn_zio = zio; | |
1712 | dnode_sync(DMU_USERUSED_DNODE(os), tx); | |
1713 | DMU_GROUPUSED_DNODE(os)->dn_zio = zio; | |
1714 | dnode_sync(DMU_GROUPUSED_DNODE(os), tx); | |
9babb374 BB |
1715 | } |
1716 | ||
9c5167d1 NF |
1717 | if (DMU_PROJECTUSED_DNODE(os) && |
1718 | DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1719 | DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; | |
1720 | dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); | |
1721 | } | |
1722 | ||
34dc7c2f BB |
1723 | txgoff = tx->tx_txg & TXG_MASK; |
1724 | ||
ba67d821 MA |
1725 | /* |
1726 | * We must create the list here because it uses the | |
1727 | * dn_dirty_link[] of this txg. But it may already | |
1728 | * exist because we call dsl_dataset_sync() twice per txg. | |
1729 | */ | |
ffdf019c AM |
1730 | if (os->os_synced_dnodes.ml_sublists == NULL) { |
1731 | multilist_create(&os->os_synced_dnodes, sizeof (dnode_t), | |
ba67d821 MA |
1732 | offsetof(dnode_t, dn_dirty_link[txgoff]), |
1733 | dnode_multilist_index_func); | |
1734 | } else { | |
ffdf019c | 1735 | ASSERT3U(os->os_synced_dnodes.ml_offset, ==, |
ba67d821 | 1736 | offsetof(dnode_t, dn_dirty_link[txgoff])); |
9babb374 BB |
1737 | } |
1738 | ||
ffdf019c | 1739 | ml = &os->os_dirty_dnodes[txgoff]; |
fc754677 AM |
1740 | num_sublists = multilist_get_num_sublists(ml); |
1741 | for (int i = 0; i < num_sublists; i++) { | |
1742 | if (multilist_sublist_is_empty_idx(ml, i)) | |
1743 | continue; | |
64fc7762 | 1744 | sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); |
fc754677 | 1745 | sda->sda_list = ml; |
64fc7762 MA |
1746 | sda->sda_sublist_idx = i; |
1747 | sda->sda_tx = tx; | |
1748 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1749 | sync_dnodes_task, sda, 0); | |
1750 | /* callback frees sda */ | |
1751 | } | |
1752 | taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); | |
34dc7c2f | 1753 | |
572e2857 | 1754 | list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; |
64fc7762 | 1755 | while ((dr = list_head(list)) != NULL) { |
13fe0198 | 1756 | ASSERT0(dr->dr_dbuf->db_level); |
34dc7c2f | 1757 | list_remove(list, dr); |
9cdf7b1f | 1758 | zio_nowait(dr->dr_zio); |
34dc7c2f | 1759 | } |
68cbd56e NB |
1760 | |
1761 | /* Enable dnode backfill if enough objects have been freed. */ | |
1762 | if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { | |
1763 | os->os_rescan_dnodes = B_TRUE; | |
1764 | os->os_freed_dnodes = 0; | |
1765 | } | |
1766 | ||
34dc7c2f BB |
1767 | /* |
1768 | * Free intent log blocks up to this tx. | |
1769 | */ | |
1770 | zil_sync(os->os_zil, tx); | |
b128c09f | 1771 | os->os_phys->os_zil_header = os->os_zil_header; |
34dc7c2f BB |
1772 | zio_nowait(zio); |
1773 | } | |
1774 | ||
428870ff BB |
1775 | boolean_t |
1776 | dmu_objset_is_dirty(objset_t *os, uint64_t txg) | |
1777 | { | |
ffdf019c | 1778 | return (!multilist_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK])); |
428870ff BB |
1779 | } |
1780 | ||
7bcb7f08 | 1781 | static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES]; |
9babb374 BB |
1782 | |
1783 | void | |
7bcb7f08 | 1784 | dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb) |
9babb374 | 1785 | { |
7bcb7f08 MA |
1786 | file_cbs[ost] = cb; |
1787 | } | |
1788 | ||
1789 | int | |
1790 | dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data, | |
1791 | zfs_file_info_t *zfi) | |
1792 | { | |
1793 | file_info_cb_t *cb = file_cbs[os->os_phys->os_type]; | |
1794 | if (cb == NULL) | |
1795 | return (EINVAL); | |
1796 | return (cb(bonustype, data, zfi)); | |
9babb374 BB |
1797 | } |
1798 | ||
1799 | boolean_t | |
428870ff | 1800 | dmu_objset_userused_enabled(objset_t *os) |
9babb374 BB |
1801 | { |
1802 | return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && | |
7bcb7f08 | 1803 | file_cbs[os->os_phys->os_type] != NULL && |
572e2857 | 1804 | DMU_USERUSED_DNODE(os) != NULL); |
9babb374 BB |
1805 | } |
1806 | ||
1de321e6 JX |
1807 | boolean_t |
1808 | dmu_objset_userobjused_enabled(objset_t *os) | |
1809 | { | |
1810 | return (dmu_objset_userused_enabled(os) && | |
1811 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); | |
1812 | } | |
1813 | ||
9c5167d1 NF |
1814 | boolean_t |
1815 | dmu_objset_projectquota_enabled(objset_t *os) | |
1816 | { | |
7bcb7f08 | 1817 | return (file_cbs[os->os_phys->os_type] != NULL && |
9c5167d1 NF |
1818 | DMU_PROJECTUSED_DNODE(os) != NULL && |
1819 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); | |
1820 | } | |
1821 | ||
9b7a83cb JX |
1822 | typedef struct userquota_node { |
1823 | /* must be in the first filed, see userquota_update_cache() */ | |
1824 | char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1825 | int64_t uqn_delta; | |
1826 | avl_node_t uqn_node; | |
1827 | } userquota_node_t; | |
1828 | ||
1829 | typedef struct userquota_cache { | |
1830 | avl_tree_t uqc_user_deltas; | |
1831 | avl_tree_t uqc_group_deltas; | |
9c5167d1 | 1832 | avl_tree_t uqc_project_deltas; |
9b7a83cb JX |
1833 | } userquota_cache_t; |
1834 | ||
1835 | static int | |
1836 | userquota_compare(const void *l, const void *r) | |
1837 | { | |
1838 | const userquota_node_t *luqn = l; | |
1839 | const userquota_node_t *ruqn = r; | |
e4ffa98d | 1840 | int rv; |
9b7a83cb JX |
1841 | |
1842 | /* | |
1843 | * NB: can only access uqn_id because userquota_update_cache() doesn't | |
1844 | * pass in an entire userquota_node_t. | |
1845 | */ | |
e4ffa98d BB |
1846 | rv = strcmp(luqn->uqn_id, ruqn->uqn_id); |
1847 | ||
ca577779 | 1848 | return (TREE_ISIGN(rv)); |
9b7a83cb JX |
1849 | } |
1850 | ||
1851 | static void | |
1852 | do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) | |
1853 | { | |
1854 | void *cookie; | |
1855 | userquota_node_t *uqn; | |
1856 | ||
1857 | ASSERT(dmu_tx_is_syncing(tx)); | |
1858 | ||
1859 | cookie = NULL; | |
1860 | while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, | |
1861 | &cookie)) != NULL) { | |
64fc7762 MA |
1862 | /* |
1863 | * os_userused_lock protects against concurrent calls to | |
1864 | * zap_increment_int(). It's needed because zap_increment_int() | |
1865 | * is not thread-safe (i.e. not atomic). | |
1866 | */ | |
1867 | mutex_enter(&os->os_userused_lock); | |
9b7a83cb JX |
1868 | VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, |
1869 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1870 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1871 | kmem_free(uqn, sizeof (*uqn)); |
1872 | } | |
1873 | avl_destroy(&cache->uqc_user_deltas); | |
1874 | ||
1875 | cookie = NULL; | |
1876 | while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, | |
1877 | &cookie)) != NULL) { | |
64fc7762 | 1878 | mutex_enter(&os->os_userused_lock); |
9b7a83cb JX |
1879 | VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, |
1880 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
64fc7762 | 1881 | mutex_exit(&os->os_userused_lock); |
9b7a83cb JX |
1882 | kmem_free(uqn, sizeof (*uqn)); |
1883 | } | |
1884 | avl_destroy(&cache->uqc_group_deltas); | |
9c5167d1 NF |
1885 | |
1886 | if (dmu_objset_projectquota_enabled(os)) { | |
1887 | cookie = NULL; | |
1888 | while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, | |
1889 | &cookie)) != NULL) { | |
1890 | mutex_enter(&os->os_userused_lock); | |
1891 | VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, | |
1892 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1893 | mutex_exit(&os->os_userused_lock); | |
1894 | kmem_free(uqn, sizeof (*uqn)); | |
1895 | } | |
1896 | avl_destroy(&cache->uqc_project_deltas); | |
1897 | } | |
9b7a83cb JX |
1898 | } |
1899 | ||
1900 | static void | |
1901 | userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) | |
1902 | { | |
1903 | userquota_node_t *uqn; | |
1904 | avl_index_t idx; | |
1905 | ||
1906 | ASSERT(strlen(id) < sizeof (uqn->uqn_id)); | |
1907 | /* | |
1908 | * Use id directly for searching because uqn_id is the first field of | |
1909 | * userquota_node_t and fields after uqn_id won't be accessed in | |
1910 | * avl_find(). | |
1911 | */ | |
1912 | uqn = avl_find(avl, (const void *)id, &idx); | |
1913 | if (uqn == NULL) { | |
1914 | uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); | |
1f51b525 | 1915 | strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); |
9b7a83cb JX |
1916 | avl_insert(avl, uqn, idx); |
1917 | } | |
1918 | uqn->uqn_delta += delta; | |
1919 | } | |
1920 | ||
428870ff | 1921 | static void |
9c5167d1 NF |
1922 | do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, |
1923 | uint64_t flags, uint64_t user, uint64_t group, uint64_t project, | |
1924 | boolean_t subtract) | |
428870ff | 1925 | { |
9c5167d1 | 1926 | if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { |
50c957f7 | 1927 | int64_t delta = DNODE_MIN_SIZE + used; |
9b7a83cb JX |
1928 | char name[20]; |
1929 | ||
428870ff BB |
1930 | if (subtract) |
1931 | delta = -delta; | |
9b7a83cb | 1932 | |
c9e319fa | 1933 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user); |
9b7a83cb JX |
1934 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); |
1935 | ||
c9e319fa | 1936 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group); |
9b7a83cb | 1937 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); |
9c5167d1 NF |
1938 | |
1939 | if (dmu_objset_projectquota_enabled(os)) { | |
c9e319fa JL |
1940 | (void) snprintf(name, sizeof (name), "%llx", |
1941 | (longlong_t)project); | |
9c5167d1 NF |
1942 | userquota_update_cache(&cache->uqc_project_deltas, |
1943 | name, delta); | |
1944 | } | |
428870ff BB |
1945 | } |
1946 | } | |
1947 | ||
1de321e6 | 1948 | static void |
9c5167d1 NF |
1949 | do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, |
1950 | uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) | |
1de321e6 JX |
1951 | { |
1952 | if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { | |
1953 | char name[20 + DMU_OBJACCT_PREFIX_LEN]; | |
9b7a83cb | 1954 | int delta = subtract ? -1 : 1; |
1de321e6 JX |
1955 | |
1956 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1957 | (longlong_t)user); | |
9b7a83cb | 1958 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); |
1de321e6 JX |
1959 | |
1960 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1961 | (longlong_t)group); | |
9b7a83cb | 1962 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); |
9c5167d1 NF |
1963 | |
1964 | if (dmu_objset_projectquota_enabled(os)) { | |
1965 | (void) snprintf(name, sizeof (name), | |
1966 | DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); | |
1967 | userquota_update_cache(&cache->uqc_project_deltas, | |
1968 | name, delta); | |
1969 | } | |
1de321e6 JX |
1970 | } |
1971 | } | |
1972 | ||
64fc7762 MA |
1973 | typedef struct userquota_updates_arg { |
1974 | objset_t *uua_os; | |
1975 | int uua_sublist_idx; | |
1976 | dmu_tx_t *uua_tx; | |
1977 | } userquota_updates_arg_t; | |
1978 | ||
1979 | static void | |
1980 | userquota_updates_task(void *arg) | |
9babb374 | 1981 | { |
64fc7762 MA |
1982 | userquota_updates_arg_t *uua = arg; |
1983 | objset_t *os = uua->uua_os; | |
1984 | dmu_tx_t *tx = uua->uua_tx; | |
9babb374 | 1985 | dnode_t *dn; |
9b7a83cb | 1986 | userquota_cache_t cache = { { 0 } }; |
9babb374 | 1987 | |
64fc7762 | 1988 | multilist_sublist_t *list = |
ffdf019c | 1989 | multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx); |
9babb374 | 1990 | |
64fc7762 MA |
1991 | ASSERT(multilist_sublist_head(list) == NULL || |
1992 | dmu_objset_userused_enabled(os)); | |
9b7a83cb JX |
1993 | avl_create(&cache.uqc_user_deltas, userquota_compare, |
1994 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
1995 | avl_create(&cache.uqc_group_deltas, userquota_compare, | |
1996 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
9c5167d1 NF |
1997 | if (dmu_objset_projectquota_enabled(os)) |
1998 | avl_create(&cache.uqc_project_deltas, userquota_compare, | |
1999 | sizeof (userquota_node_t), offsetof(userquota_node_t, | |
2000 | uqn_node)); | |
9b7a83cb | 2001 | |
64fc7762 | 2002 | while ((dn = multilist_sublist_head(list)) != NULL) { |
572e2857 | 2003 | int flags; |
9babb374 | 2004 | ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); |
9babb374 BB |
2005 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || |
2006 | dn->dn_phys->dn_flags & | |
2007 | DNODE_FLAG_USERUSED_ACCOUNTED); | |
2008 | ||
572e2857 BB |
2009 | flags = dn->dn_id_flags; |
2010 | ASSERT(flags); | |
2011 | if (flags & DN_ID_OLD_EXIST) { | |
9c5167d1 NF |
2012 | do_userquota_update(os, &cache, dn->dn_oldused, |
2013 | dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, | |
2014 | dn->dn_oldprojid, B_TRUE); | |
2015 | do_userobjquota_update(os, &cache, dn->dn_oldflags, | |
2016 | dn->dn_olduid, dn->dn_oldgid, | |
2017 | dn->dn_oldprojid, B_TRUE); | |
428870ff | 2018 | } |
572e2857 | 2019 | if (flags & DN_ID_NEW_EXIST) { |
9c5167d1 | 2020 | do_userquota_update(os, &cache, |
9b7a83cb | 2021 | DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, |
9c5167d1 NF |
2022 | dn->dn_newuid, dn->dn_newgid, |
2023 | dn->dn_newprojid, B_FALSE); | |
2024 | do_userobjquota_update(os, &cache, | |
2025 | dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, | |
2026 | dn->dn_newprojid, B_FALSE); | |
428870ff BB |
2027 | } |
2028 | ||
572e2857 | 2029 | mutex_enter(&dn->dn_mtx); |
428870ff BB |
2030 | dn->dn_oldused = 0; |
2031 | dn->dn_oldflags = 0; | |
2032 | if (dn->dn_id_flags & DN_ID_NEW_EXIST) { | |
2033 | dn->dn_olduid = dn->dn_newuid; | |
2034 | dn->dn_oldgid = dn->dn_newgid; | |
9c5167d1 | 2035 | dn->dn_oldprojid = dn->dn_newprojid; |
428870ff BB |
2036 | dn->dn_id_flags |= DN_ID_OLD_EXIST; |
2037 | if (dn->dn_bonuslen == 0) | |
2038 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
2039 | else | |
2040 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2041 | } | |
2042 | dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); | |
9babb374 BB |
2043 | mutex_exit(&dn->dn_mtx); |
2044 | ||
64fc7762 | 2045 | multilist_sublist_remove(list, dn); |
ffdf019c | 2046 | dnode_rele(dn, &os->os_synced_dnodes); |
9babb374 | 2047 | } |
9b7a83cb | 2048 | do_userquota_cacheflush(os, &cache, tx); |
64fc7762 MA |
2049 | multilist_sublist_unlock(list); |
2050 | kmem_free(uua, sizeof (*uua)); | |
2051 | } | |
2052 | ||
ba67d821 MA |
2053 | /* |
2054 | * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being | |
2055 | * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be | |
2056 | * evicted because the block containing the dnode can't be evicted until it is | |
2057 | * written out. However, this hold is necessary to prevent the dnode_t from | |
2058 | * being moved (via dnode_move()) while it's still referenced by | |
2059 | * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for | |
2060 | * dirty_lightweight_leaf-type dirty records. | |
2061 | * | |
2062 | * If we are doing user-object accounting, the dnode_rele() happens from | |
2063 | * userquota_updates_task() instead. | |
2064 | */ | |
2065 | static void | |
2066 | dnode_rele_task(void *arg) | |
64fc7762 | 2067 | { |
ba67d821 MA |
2068 | userquota_updates_arg_t *uua = arg; |
2069 | objset_t *os = uua->uua_os; | |
2070 | ||
2071 | multilist_sublist_t *list = | |
ffdf019c | 2072 | multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx); |
fc754677 | 2073 | |
ba67d821 MA |
2074 | dnode_t *dn; |
2075 | while ((dn = multilist_sublist_head(list)) != NULL) { | |
2076 | multilist_sublist_remove(list, dn); | |
ffdf019c | 2077 | dnode_rele(dn, &os->os_synced_dnodes); |
ba67d821 MA |
2078 | } |
2079 | multilist_sublist_unlock(list); | |
2080 | kmem_free(uua, sizeof (*uua)); | |
2081 | } | |
2082 | ||
2083 | /* | |
2084 | * Return TRUE if userquota updates are needed. | |
2085 | */ | |
2086 | static boolean_t | |
2087 | dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx) | |
2088 | { | |
64fc7762 | 2089 | if (!dmu_objset_userused_enabled(os)) |
ba67d821 | 2090 | return (B_FALSE); |
64fc7762 | 2091 | |
163a8c28 TC |
2092 | /* |
2093 | * If this is a raw receive just return and handle accounting | |
2094 | * later when we have the keys loaded. We also don't do user | |
2095 | * accounting during claiming since the datasets are not owned | |
2096 | * for the duration of claiming and this txg should only be | |
2097 | * used for recovery. | |
2098 | */ | |
b5256303 | 2099 | if (os->os_encrypted && dmu_objset_is_receiving(os)) |
ba67d821 | 2100 | return (B_FALSE); |
b5256303 | 2101 | |
163a8c28 | 2102 | if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) |
ba67d821 | 2103 | return (B_FALSE); |
163a8c28 | 2104 | |
9c5167d1 | 2105 | /* Allocate the user/group/project used objects if necessary. */ |
64fc7762 MA |
2106 | if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { |
2107 | VERIFY0(zap_create_claim(os, | |
2108 | DMU_USERUSED_OBJECT, | |
2109 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2110 | VERIFY0(zap_create_claim(os, | |
2111 | DMU_GROUPUSED_OBJECT, | |
2112 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2113 | } | |
2114 | ||
9c5167d1 NF |
2115 | if (dmu_objset_projectquota_enabled(os) && |
2116 | DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { | |
2117 | VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, | |
2118 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2119 | } | |
ba67d821 MA |
2120 | return (B_TRUE); |
2121 | } | |
9c5167d1 | 2122 | |
ba67d821 MA |
2123 | /* |
2124 | * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and | |
2125 | * also release the holds on the dnodes from dmu_objset_sync_dnodes(). | |
2126 | * The caller must taskq_wait(dp_sync_taskq). | |
2127 | */ | |
2128 | void | |
2129 | dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx) | |
2130 | { | |
2131 | boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx); | |
2132 | ||
ffdf019c | 2133 | int num_sublists = multilist_get_num_sublists(&os->os_synced_dnodes); |
fc754677 | 2134 | for (int i = 0; i < num_sublists; i++) { |
64fc7762 MA |
2135 | userquota_updates_arg_t *uua = |
2136 | kmem_alloc(sizeof (*uua), KM_SLEEP); | |
2137 | uua->uua_os = os; | |
2138 | uua->uua_sublist_idx = i; | |
2139 | uua->uua_tx = tx; | |
ba67d821 MA |
2140 | |
2141 | /* | |
2142 | * If we don't need to update userquotas, use | |
2143 | * dnode_rele_task() to call dnode_rele() | |
2144 | */ | |
64fc7762 | 2145 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, |
ba67d821 MA |
2146 | need_userquota ? userquota_updates_task : dnode_rele_task, |
2147 | uua, 0); | |
64fc7762 MA |
2148 | /* callback frees uua */ |
2149 | } | |
9babb374 BB |
2150 | } |
2151 | ||
ba67d821 | 2152 | |
428870ff BB |
2153 | /* |
2154 | * Returns a pointer to data to find uid/gid from | |
2155 | * | |
2156 | * If a dirty record for transaction group that is syncing can't | |
2157 | * be found then NULL is returned. In the NULL case it is assumed | |
2158 | * the uid/gid aren't changing. | |
2159 | */ | |
2160 | static void * | |
2161 | dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) | |
2162 | { | |
cccbed9f | 2163 | dbuf_dirty_record_t *dr; |
428870ff BB |
2164 | void *data; |
2165 | ||
2166 | if (db->db_dirtycnt == 0) | |
2167 | return (db->db.db_data); /* Nothing is changing */ | |
2168 | ||
cccbed9f | 2169 | dr = dbuf_find_dirty_eq(db, tx->tx_txg); |
428870ff | 2170 | |
572e2857 | 2171 | if (dr == NULL) { |
428870ff | 2172 | data = NULL; |
572e2857 | 2173 | } else { |
ba67d821 | 2174 | if (dr->dr_dnode->dn_bonuslen == 0 && |
572e2857 BB |
2175 | dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) |
2176 | data = dr->dt.dl.dr_data->b_data; | |
2177 | else | |
2178 | data = dr->dt.dl.dr_data; | |
572e2857 BB |
2179 | } |
2180 | ||
428870ff BB |
2181 | return (data); |
2182 | } | |
2183 | ||
2184 | void | |
2185 | dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) | |
2186 | { | |
2187 | objset_t *os = dn->dn_objset; | |
2188 | void *data = NULL; | |
2189 | dmu_buf_impl_t *db = NULL; | |
428870ff BB |
2190 | int flags = dn->dn_id_flags; |
2191 | int error; | |
2192 | boolean_t have_spill = B_FALSE; | |
2193 | ||
2194 | if (!dmu_objset_userused_enabled(dn->dn_objset)) | |
2195 | return; | |
2196 | ||
b5256303 TC |
2197 | /* |
2198 | * Raw receives introduce a problem with user accounting. Raw | |
2199 | * receives cannot update the user accounting info because the | |
2200 | * user ids and the sizes are encrypted. To guarantee that we | |
2201 | * never end up with bad user accounting, we simply disable it | |
2202 | * during raw receives. We also disable this for normal receives | |
2203 | * so that an incremental raw receive may be done on top of an | |
2204 | * existing non-raw receive. | |
2205 | */ | |
2206 | if (os->os_encrypted && dmu_objset_is_receiving(os)) | |
2207 | return; | |
2208 | ||
428870ff BB |
2209 | if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| |
2210 | DN_ID_CHKED_SPILL))) | |
2211 | return; | |
2212 | ||
2213 | if (before && dn->dn_bonuslen != 0) | |
2214 | data = DN_BONUS(dn->dn_phys); | |
2215 | else if (!before && dn->dn_bonuslen != 0) { | |
a3000f93 BB |
2216 | if (dn->dn_bonus) { |
2217 | db = dn->dn_bonus; | |
428870ff BB |
2218 | mutex_enter(&db->db_mtx); |
2219 | data = dmu_objset_userquota_find_data(db, tx); | |
2220 | } else { | |
2221 | data = DN_BONUS(dn->dn_phys); | |
2222 | } | |
2223 | } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { | |
2224 | int rf = 0; | |
2225 | ||
2226 | if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) | |
2227 | rf |= DB_RF_HAVESTRUCT; | |
572e2857 BB |
2228 | error = dmu_spill_hold_by_dnode(dn, |
2229 | rf | DB_RF_MUST_SUCCEED, | |
428870ff BB |
2230 | FTAG, (dmu_buf_t **)&db); |
2231 | ASSERT(error == 0); | |
2232 | mutex_enter(&db->db_mtx); | |
2233 | data = (before) ? db->db.db_data : | |
2234 | dmu_objset_userquota_find_data(db, tx); | |
2235 | have_spill = B_TRUE; | |
2236 | } else { | |
2237 | mutex_enter(&dn->dn_mtx); | |
2238 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2239 | mutex_exit(&dn->dn_mtx); | |
2240 | return; | |
2241 | } | |
2242 | ||
428870ff BB |
2243 | /* |
2244 | * Must always call the callback in case the object | |
2245 | * type has changed and that type isn't an object type to track | |
2246 | */ | |
7bcb7f08 MA |
2247 | zfs_file_info_t zfi; |
2248 | error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi); | |
2249 | ||
2250 | if (before) { | |
2251 | ASSERT(data); | |
2252 | dn->dn_olduid = zfi.zfi_user; | |
2253 | dn->dn_oldgid = zfi.zfi_group; | |
2254 | dn->dn_oldprojid = zfi.zfi_project; | |
2255 | } else if (data) { | |
2256 | dn->dn_newuid = zfi.zfi_user; | |
2257 | dn->dn_newgid = zfi.zfi_group; | |
2258 | dn->dn_newprojid = zfi.zfi_project; | |
2259 | } | |
428870ff BB |
2260 | |
2261 | /* | |
2262 | * Preserve existing uid/gid when the callback can't determine | |
2263 | * what the new uid/gid are and the callback returned EEXIST. | |
2264 | * The EEXIST error tells us to just use the existing uid/gid. | |
2265 | * If we don't know what the old values are then just assign | |
2266 | * them to 0, since that is a new file being created. | |
2267 | */ | |
2268 | if (!before && data == NULL && error == EEXIST) { | |
2269 | if (flags & DN_ID_OLD_EXIST) { | |
2270 | dn->dn_newuid = dn->dn_olduid; | |
2271 | dn->dn_newgid = dn->dn_oldgid; | |
2705ebf0 | 2272 | dn->dn_newprojid = dn->dn_oldprojid; |
428870ff BB |
2273 | } else { |
2274 | dn->dn_newuid = 0; | |
2275 | dn->dn_newgid = 0; | |
9c5167d1 | 2276 | dn->dn_newprojid = ZFS_DEFAULT_PROJID; |
428870ff BB |
2277 | } |
2278 | error = 0; | |
2279 | } | |
2280 | ||
2281 | if (db) | |
2282 | mutex_exit(&db->db_mtx); | |
2283 | ||
2284 | mutex_enter(&dn->dn_mtx); | |
2285 | if (error == 0 && before) | |
2286 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
2287 | if (error == 0 && !before) | |
2288 | dn->dn_id_flags |= DN_ID_NEW_EXIST; | |
2289 | ||
2290 | if (have_spill) { | |
2291 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
2292 | } else { | |
2293 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2294 | } | |
2295 | mutex_exit(&dn->dn_mtx); | |
a3000f93 | 2296 | if (have_spill) |
428870ff BB |
2297 | dmu_buf_rele((dmu_buf_t *)db, FTAG); |
2298 | } | |
2299 | ||
9babb374 BB |
2300 | boolean_t |
2301 | dmu_objset_userspace_present(objset_t *os) | |
2302 | { | |
428870ff | 2303 | return (os->os_phys->os_flags & |
9babb374 BB |
2304 | OBJSET_FLAG_USERACCOUNTING_COMPLETE); |
2305 | } | |
2306 | ||
1de321e6 JX |
2307 | boolean_t |
2308 | dmu_objset_userobjspace_present(objset_t *os) | |
2309 | { | |
2310 | return (os->os_phys->os_flags & | |
2311 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); | |
2312 | } | |
2313 | ||
9c5167d1 NF |
2314 | boolean_t |
2315 | dmu_objset_projectquota_present(objset_t *os) | |
2316 | { | |
2317 | return (os->os_phys->os_flags & | |
2318 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE); | |
2319 | } | |
2320 | ||
1de321e6 JX |
2321 | static int |
2322 | dmu_objset_space_upgrade(objset_t *os) | |
9babb374 BB |
2323 | { |
2324 | uint64_t obj; | |
2325 | int err = 0; | |
2326 | ||
9babb374 BB |
2327 | /* |
2328 | * We simply need to mark every object dirty, so that it will be | |
2329 | * synced out and now accounted. If this is called | |
2330 | * concurrently, or if we already did some work before crashing, | |
2331 | * that's fine, since we track each object's accounted state | |
2332 | * independently. | |
2333 | */ | |
2334 | ||
2335 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { | |
45d1cae3 | 2336 | dmu_tx_t *tx; |
9babb374 BB |
2337 | dmu_buf_t *db; |
2338 | int objerr; | |
2339 | ||
1de321e6 JX |
2340 | mutex_enter(&os->os_upgrade_lock); |
2341 | if (os->os_upgrade_exit) | |
2342 | err = SET_ERROR(EINTR); | |
2343 | mutex_exit(&os->os_upgrade_lock); | |
2344 | if (err != 0) | |
2345 | return (err); | |
2346 | ||
9babb374 | 2347 | if (issig(JUSTLOOKING) && issig(FORREAL)) |
2e528b49 | 2348 | return (SET_ERROR(EINTR)); |
9babb374 BB |
2349 | |
2350 | objerr = dmu_bonus_hold(os, obj, FTAG, &db); | |
13fe0198 | 2351 | if (objerr != 0) |
9babb374 | 2352 | continue; |
45d1cae3 | 2353 | tx = dmu_tx_create(os); |
9babb374 BB |
2354 | dmu_tx_hold_bonus(tx, obj); |
2355 | objerr = dmu_tx_assign(tx, TXG_WAIT); | |
13fe0198 | 2356 | if (objerr != 0) { |
d22323e8 | 2357 | dmu_buf_rele(db, FTAG); |
9babb374 BB |
2358 | dmu_tx_abort(tx); |
2359 | continue; | |
2360 | } | |
2361 | dmu_buf_will_dirty(db, tx); | |
2362 | dmu_buf_rele(db, FTAG); | |
2363 | dmu_tx_commit(tx); | |
2364 | } | |
1de321e6 JX |
2365 | return (0); |
2366 | } | |
2367 | ||
4072f465 | 2368 | static int |
2369 | dmu_objset_userspace_upgrade_cb(objset_t *os) | |
1de321e6 JX |
2370 | { |
2371 | int err = 0; | |
2372 | ||
2373 | if (dmu_objset_userspace_present(os)) | |
2374 | return (0); | |
2375 | if (dmu_objset_is_snapshot(os)) | |
2376 | return (SET_ERROR(EINVAL)); | |
2377 | if (!dmu_objset_userused_enabled(os)) | |
2378 | return (SET_ERROR(ENOTSUP)); | |
2379 | ||
2380 | err = dmu_objset_space_upgrade(os); | |
2381 | if (err) | |
2382 | return (err); | |
9babb374 | 2383 | |
428870ff | 2384 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
9babb374 BB |
2385 | txg_wait_synced(dmu_objset_pool(os), 0); |
2386 | return (0); | |
2387 | } | |
2388 | ||
4072f465 | 2389 | void |
2390 | dmu_objset_userspace_upgrade(objset_t *os) | |
2391 | { | |
2392 | dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb); | |
2393 | } | |
2394 | ||
1de321e6 | 2395 | static int |
9c5167d1 | 2396 | dmu_objset_id_quota_upgrade_cb(objset_t *os) |
1de321e6 JX |
2397 | { |
2398 | int err = 0; | |
2399 | ||
9c5167d1 NF |
2400 | if (dmu_objset_userobjspace_present(os) && |
2401 | dmu_objset_projectquota_present(os)) | |
1de321e6 JX |
2402 | return (0); |
2403 | if (dmu_objset_is_snapshot(os)) | |
2404 | return (SET_ERROR(EINVAL)); | |
4072f465 | 2405 | if (!dmu_objset_userused_enabled(os)) |
1de321e6 | 2406 | return (SET_ERROR(ENOTSUP)); |
9c5167d1 NF |
2407 | if (!dmu_objset_projectquota_enabled(os) && |
2408 | dmu_objset_userobjspace_present(os)) | |
2409 | return (SET_ERROR(ENOTSUP)); | |
1de321e6 | 2410 | |
1de321e6 JX |
2411 | err = dmu_objset_space_upgrade(os); |
2412 | if (err) | |
2413 | return (err); | |
2414 | ||
4072f465 | 2415 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
2416 | if (dmu_objset_userobjused_enabled(os)) | |
2417 | os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
9c5167d1 NF |
2418 | if (dmu_objset_projectquota_enabled(os)) |
2419 | os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
2420 | ||
1de321e6 JX |
2421 | txg_wait_synced(dmu_objset_pool(os), 0); |
2422 | return (0); | |
2423 | } | |
2424 | ||
2425 | void | |
9c5167d1 | 2426 | dmu_objset_id_quota_upgrade(objset_t *os) |
1de321e6 | 2427 | { |
9c5167d1 | 2428 | dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); |
1de321e6 JX |
2429 | } |
2430 | ||
126ae9f4 JX |
2431 | boolean_t |
2432 | dmu_objset_userobjspace_upgradable(objset_t *os) | |
2433 | { | |
2434 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2435 | !dmu_objset_is_snapshot(os) && | |
2436 | dmu_objset_userobjused_enabled(os) && | |
bb1be77a | 2437 | !dmu_objset_userobjspace_present(os) && |
2438 | spa_writeable(dmu_objset_spa(os))); | |
126ae9f4 JX |
2439 | } |
2440 | ||
9c5167d1 NF |
2441 | boolean_t |
2442 | dmu_objset_projectquota_upgradable(objset_t *os) | |
2443 | { | |
2444 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2445 | !dmu_objset_is_snapshot(os) && | |
2446 | dmu_objset_projectquota_enabled(os) && | |
bb1be77a | 2447 | !dmu_objset_projectquota_present(os) && |
2448 | spa_writeable(dmu_objset_spa(os))); | |
9c5167d1 NF |
2449 | } |
2450 | ||
34dc7c2f BB |
2451 | void |
2452 | dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, | |
2453 | uint64_t *usedobjsp, uint64_t *availobjsp) | |
2454 | { | |
428870ff | 2455 | dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, |
34dc7c2f BB |
2456 | usedobjsp, availobjsp); |
2457 | } | |
2458 | ||
2459 | uint64_t | |
2460 | dmu_objset_fsid_guid(objset_t *os) | |
2461 | { | |
428870ff | 2462 | return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); |
34dc7c2f BB |
2463 | } |
2464 | ||
2465 | void | |
2466 | dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) | |
2467 | { | |
428870ff BB |
2468 | stat->dds_type = os->os_phys->os_type; |
2469 | if (os->os_dsl_dataset) | |
2470 | dsl_dataset_fast_stat(os->os_dsl_dataset, stat); | |
34dc7c2f BB |
2471 | } |
2472 | ||
2473 | void | |
2474 | dmu_objset_stats(objset_t *os, nvlist_t *nv) | |
2475 | { | |
428870ff BB |
2476 | ASSERT(os->os_dsl_dataset || |
2477 | os->os_phys->os_type == DMU_OST_META); | |
34dc7c2f | 2478 | |
428870ff BB |
2479 | if (os->os_dsl_dataset != NULL) |
2480 | dsl_dataset_stats(os->os_dsl_dataset, nv); | |
34dc7c2f BB |
2481 | |
2482 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, | |
428870ff | 2483 | os->os_phys->os_type); |
9babb374 BB |
2484 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, |
2485 | dmu_objset_userspace_present(os)); | |
34dc7c2f BB |
2486 | } |
2487 | ||
2488 | int | |
2489 | dmu_objset_is_snapshot(objset_t *os) | |
2490 | { | |
428870ff | 2491 | if (os->os_dsl_dataset != NULL) |
0c66c32d | 2492 | return (os->os_dsl_dataset->ds_is_snapshot); |
34dc7c2f BB |
2493 | else |
2494 | return (B_FALSE); | |
2495 | } | |
2496 | ||
2497 | int | |
4d55ea81 | 2498 | dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen, |
34dc7c2f BB |
2499 | boolean_t *conflict) |
2500 | { | |
428870ff | 2501 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2502 | uint64_t ignored; |
2503 | ||
d683ddbb | 2504 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2505 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2506 | |
2507 | return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2508 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, |
9b7b9cd3 | 2509 | MT_NORMALIZE, real, maxlen, conflict)); |
34dc7c2f BB |
2510 | } |
2511 | ||
2512 | int | |
2513 | dmu_snapshot_list_next(objset_t *os, int namelen, char *name, | |
2514 | uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) | |
2515 | { | |
428870ff | 2516 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
2517 | zap_cursor_t cursor; |
2518 | zap_attribute_t attr; | |
2519 | ||
13fe0198 MA |
2520 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
2521 | ||
d683ddbb | 2522 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
2e528b49 | 2523 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2524 | |
2525 | zap_cursor_init_serialized(&cursor, | |
2526 | ds->ds_dir->dd_pool->dp_meta_objset, | |
d683ddbb | 2527 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); |
34dc7c2f BB |
2528 | |
2529 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2530 | zap_cursor_fini(&cursor); | |
2e528b49 | 2531 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2532 | } |
2533 | ||
2534 | if (strlen(attr.za_name) + 1 > namelen) { | |
2535 | zap_cursor_fini(&cursor); | |
2e528b49 | 2536 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2537 | } |
2538 | ||
c9e319fa | 2539 | (void) strlcpy(name, attr.za_name, namelen); |
34dc7c2f BB |
2540 | if (idp) |
2541 | *idp = attr.za_first_integer; | |
2542 | if (case_conflict) | |
2543 | *case_conflict = attr.za_normalization_conflict; | |
2544 | zap_cursor_advance(&cursor); | |
2545 | *offp = zap_cursor_serialize(&cursor); | |
2546 | zap_cursor_fini(&cursor); | |
2547 | ||
2548 | return (0); | |
2549 | } | |
2550 | ||
ebe7e575 | 2551 | int |
6772fb67 | 2552 | dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) |
ebe7e575 | 2553 | { |
d1d7e268 | 2554 | return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); |
ebe7e575 BB |
2555 | } |
2556 | ||
34dc7c2f BB |
2557 | int |
2558 | dmu_dir_list_next(objset_t *os, int namelen, char *name, | |
2559 | uint64_t *idp, uint64_t *offp) | |
2560 | { | |
428870ff | 2561 | dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; |
34dc7c2f BB |
2562 | zap_cursor_t cursor; |
2563 | zap_attribute_t attr; | |
2564 | ||
2565 | /* there is no next dir on a snapshot! */ | |
428870ff | 2566 | if (os->os_dsl_dataset->ds_object != |
d683ddbb | 2567 | dsl_dir_phys(dd)->dd_head_dataset_obj) |
2e528b49 | 2568 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2569 | |
2570 | zap_cursor_init_serialized(&cursor, | |
2571 | dd->dd_pool->dp_meta_objset, | |
d683ddbb | 2572 | dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); |
34dc7c2f BB |
2573 | |
2574 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2575 | zap_cursor_fini(&cursor); | |
2e528b49 | 2576 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2577 | } |
2578 | ||
2579 | if (strlen(attr.za_name) + 1 > namelen) { | |
2580 | zap_cursor_fini(&cursor); | |
2e528b49 | 2581 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2582 | } |
2583 | ||
c9e319fa | 2584 | (void) strlcpy(name, attr.za_name, namelen); |
34dc7c2f BB |
2585 | if (idp) |
2586 | *idp = attr.za_first_integer; | |
2587 | zap_cursor_advance(&cursor); | |
2588 | *offp = zap_cursor_serialize(&cursor); | |
2589 | zap_cursor_fini(&cursor); | |
2590 | ||
2591 | return (0); | |
2592 | } | |
2593 | ||
9c43027b AJ |
2594 | typedef struct dmu_objset_find_ctx { |
2595 | taskq_t *dc_tq; | |
2596 | dsl_pool_t *dc_dp; | |
2597 | uint64_t dc_ddobj; | |
1149ba64 | 2598 | char *dc_ddname; /* last component of ddobj's name */ |
9c43027b AJ |
2599 | int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); |
2600 | void *dc_arg; | |
2601 | int dc_flags; | |
2602 | kmutex_t *dc_error_lock; | |
2603 | int *dc_error; | |
2604 | } dmu_objset_find_ctx_t; | |
2605 | ||
2606 | static void | |
2607 | dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) | |
b128c09f | 2608 | { |
9c43027b | 2609 | dsl_pool_t *dp = dcp->dc_dp; |
13fe0198 MA |
2610 | dsl_dir_t *dd; |
2611 | dsl_dataset_t *ds; | |
2612 | zap_cursor_t zc; | |
2613 | zap_attribute_t *attr; | |
2614 | uint64_t thisobj; | |
9c43027b | 2615 | int err = 0; |
13fe0198 | 2616 | |
9c43027b AJ |
2617 | /* don't process if there already was an error */ |
2618 | if (*dcp->dc_error != 0) | |
2619 | goto out; | |
13fe0198 | 2620 | |
1149ba64 GM |
2621 | /* |
2622 | * Note: passing the name (dc_ddname) here is optional, but it | |
2623 | * improves performance because we don't need to call | |
2624 | * zap_value_search() to determine the name. | |
2625 | */ | |
2626 | err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); | |
13fe0198 | 2627 | if (err != 0) |
9c43027b | 2628 | goto out; |
13fe0198 MA |
2629 | |
2630 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ | |
2631 | if (dd->dd_myname[0] == '$') { | |
2632 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2633 | goto out; |
13fe0198 MA |
2634 | } |
2635 | ||
d683ddbb | 2636 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2637 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
13fe0198 MA |
2638 | |
2639 | /* | |
2640 | * Iterate over all children. | |
2641 | */ | |
9c43027b | 2642 | if (dcp->dc_flags & DS_FIND_CHILDREN) { |
13fe0198 | 2643 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2644 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
13fe0198 MA |
2645 | zap_cursor_retrieve(&zc, attr) == 0; |
2646 | (void) zap_cursor_advance(&zc)) { | |
2647 | ASSERT3U(attr->za_integer_length, ==, | |
2648 | sizeof (uint64_t)); | |
2649 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2650 | ||
1c27024e | 2651 | dmu_objset_find_ctx_t *child_dcp = |
1149ba64 | 2652 | kmem_alloc(sizeof (*child_dcp), KM_SLEEP); |
9c43027b AJ |
2653 | *child_dcp = *dcp; |
2654 | child_dcp->dc_ddobj = attr->za_first_integer; | |
1149ba64 | 2655 | child_dcp->dc_ddname = spa_strdup(attr->za_name); |
9c43027b AJ |
2656 | if (dcp->dc_tq != NULL) |
2657 | (void) taskq_dispatch(dcp->dc_tq, | |
2658 | dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); | |
2659 | else | |
2660 | dmu_objset_find_dp_impl(child_dcp); | |
13fe0198 MA |
2661 | } |
2662 | zap_cursor_fini(&zc); | |
13fe0198 MA |
2663 | } |
2664 | ||
2665 | /* | |
2666 | * Iterate over all snapshots. | |
2667 | */ | |
9c43027b | 2668 | if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { |
13fe0198 MA |
2669 | dsl_dataset_t *ds; |
2670 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2671 | ||
2672 | if (err == 0) { | |
d683ddbb JG |
2673 | uint64_t snapobj; |
2674 | ||
2675 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
13fe0198 MA |
2676 | dsl_dataset_rele(ds, FTAG); |
2677 | ||
2678 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2679 | zap_cursor_retrieve(&zc, attr) == 0; | |
2680 | (void) zap_cursor_advance(&zc)) { | |
2681 | ASSERT3U(attr->za_integer_length, ==, | |
2682 | sizeof (uint64_t)); | |
2683 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2684 | ||
2685 | err = dsl_dataset_hold_obj(dp, | |
2686 | attr->za_first_integer, FTAG, &ds); | |
2687 | if (err != 0) | |
2688 | break; | |
9c43027b | 2689 | err = dcp->dc_func(dp, ds, dcp->dc_arg); |
13fe0198 MA |
2690 | dsl_dataset_rele(ds, FTAG); |
2691 | if (err != 0) | |
2692 | break; | |
2693 | } | |
2694 | zap_cursor_fini(&zc); | |
2695 | } | |
2696 | } | |
2697 | ||
13fe0198 MA |
2698 | kmem_free(attr, sizeof (zap_attribute_t)); |
2699 | ||
1149ba64 GM |
2700 | if (err != 0) { |
2701 | dsl_dir_rele(dd, FTAG); | |
9c43027b | 2702 | goto out; |
1149ba64 | 2703 | } |
13fe0198 MA |
2704 | |
2705 | /* | |
2706 | * Apply to self. | |
2707 | */ | |
2708 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
1149ba64 GM |
2709 | |
2710 | /* | |
2711 | * Note: we hold the dir while calling dsl_dataset_hold_obj() so | |
2712 | * that the dir will remain cached, and we won't have to re-instantiate | |
2713 | * it (which could be expensive due to finding its name via | |
2714 | * zap_value_search()). | |
2715 | */ | |
2716 | dsl_dir_rele(dd, FTAG); | |
13fe0198 | 2717 | if (err != 0) |
9c43027b AJ |
2718 | goto out; |
2719 | err = dcp->dc_func(dp, ds, dcp->dc_arg); | |
13fe0198 | 2720 | dsl_dataset_rele(ds, FTAG); |
9c43027b AJ |
2721 | |
2722 | out: | |
2723 | if (err != 0) { | |
2724 | mutex_enter(dcp->dc_error_lock); | |
2725 | /* only keep first error */ | |
2726 | if (*dcp->dc_error == 0) | |
2727 | *dcp->dc_error = err; | |
2728 | mutex_exit(dcp->dc_error_lock); | |
2729 | } | |
2730 | ||
1149ba64 GM |
2731 | if (dcp->dc_ddname != NULL) |
2732 | spa_strfree(dcp->dc_ddname); | |
9c43027b AJ |
2733 | kmem_free(dcp, sizeof (*dcp)); |
2734 | } | |
2735 | ||
2736 | static void | |
2737 | dmu_objset_find_dp_cb(void *arg) | |
2738 | { | |
2739 | dmu_objset_find_ctx_t *dcp = arg; | |
2740 | dsl_pool_t *dp = dcp->dc_dp; | |
2741 | ||
5e8cd5d1 AJ |
2742 | /* |
2743 | * We need to get a pool_config_lock here, as there are several | |
e1cfd73f | 2744 | * assert(pool_config_held) down the stack. Getting a lock via |
5e8cd5d1 AJ |
2745 | * dsl_pool_config_enter is risky, as it might be stalled by a |
2746 | * pending writer. This would deadlock, as the write lock can | |
2747 | * only be granted when our parent thread gives up the lock. | |
2748 | * The _prio interface gives us priority over a pending writer. | |
2749 | */ | |
2750 | dsl_pool_config_enter_prio(dp, FTAG); | |
9c43027b AJ |
2751 | |
2752 | dmu_objset_find_dp_impl(dcp); | |
2753 | ||
2754 | dsl_pool_config_exit(dp, FTAG); | |
2755 | } | |
2756 | ||
2757 | /* | |
2758 | * Find objsets under and including ddobj, call func(ds) on each. | |
2759 | * The order for the enumeration is completely undefined. | |
2760 | * func is called with dsl_pool_config held. | |
2761 | */ | |
2762 | int | |
2763 | dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, | |
2764 | int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) | |
2765 | { | |
2766 | int error = 0; | |
2767 | taskq_t *tq = NULL; | |
2768 | int ntasks; | |
2769 | dmu_objset_find_ctx_t *dcp; | |
2770 | kmutex_t err_lock; | |
2771 | ||
2772 | mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); | |
2773 | dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); | |
2774 | dcp->dc_tq = NULL; | |
2775 | dcp->dc_dp = dp; | |
2776 | dcp->dc_ddobj = ddobj; | |
1149ba64 | 2777 | dcp->dc_ddname = NULL; |
9c43027b AJ |
2778 | dcp->dc_func = func; |
2779 | dcp->dc_arg = arg; | |
2780 | dcp->dc_flags = flags; | |
2781 | dcp->dc_error_lock = &err_lock; | |
2782 | dcp->dc_error = &error; | |
2783 | ||
2784 | if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { | |
2785 | /* | |
2786 | * In case a write lock is held we can't make use of | |
2787 | * parallelism, as down the stack of the worker threads | |
2788 | * the lock is asserted via dsl_pool_config_held. | |
2789 | * In case of a read lock this is solved by getting a read | |
2790 | * lock in each worker thread, which isn't possible in case | |
2791 | * of a writer lock. So we fall back to the synchronous path | |
2792 | * here. | |
2793 | * In the future it might be possible to get some magic into | |
2794 | * dsl_pool_config_held in a way that it returns true for | |
2795 | * the worker threads so that a single lock held from this | |
2796 | * thread suffices. For now, stay single threaded. | |
2797 | */ | |
2798 | dmu_objset_find_dp_impl(dcp); | |
e5676636 | 2799 | mutex_destroy(&err_lock); |
9c43027b AJ |
2800 | |
2801 | return (error); | |
2802 | } | |
2803 | ||
2804 | ntasks = dmu_find_threads; | |
2805 | if (ntasks == 0) | |
2806 | ntasks = vdev_count_leaves(dp->dp_spa) * 4; | |
1229323d | 2807 | tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, |
9c43027b AJ |
2808 | INT_MAX, 0); |
2809 | if (tq == NULL) { | |
2810 | kmem_free(dcp, sizeof (*dcp)); | |
e5676636 BB |
2811 | mutex_destroy(&err_lock); |
2812 | ||
9c43027b AJ |
2813 | return (SET_ERROR(ENOMEM)); |
2814 | } | |
2815 | dcp->dc_tq = tq; | |
2816 | ||
2817 | /* dcp will be freed by task */ | |
2818 | (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); | |
2819 | ||
2820 | /* | |
2821 | * PORTING: this code relies on the property of taskq_wait to wait | |
2822 | * until no more tasks are queued and no more tasks are active. As | |
2823 | * we always queue new tasks from within other tasks, task_wait | |
2824 | * reliably waits for the full recursion to finish, even though we | |
2825 | * enqueue new tasks after taskq_wait has been called. | |
2826 | * On platforms other than illumos, taskq_wait may not have this | |
2827 | * property. | |
2828 | */ | |
2829 | taskq_wait(tq); | |
2830 | taskq_destroy(tq); | |
2831 | mutex_destroy(&err_lock); | |
2832 | ||
2833 | return (error); | |
b128c09f BB |
2834 | } |
2835 | ||
2836 | /* | |
13fe0198 MA |
2837 | * Find all objsets under name, and for each, call 'func(child_name, arg)'. |
2838 | * The dp_config_rwlock must not be held when this is called, and it | |
2839 | * will not be held when the callback is called. | |
2840 | * Therefore this function should only be used when the pool is not changing | |
2841 | * (e.g. in syncing context), or the callback can deal with the possible races. | |
b128c09f | 2842 | */ |
13fe0198 MA |
2843 | static int |
2844 | dmu_objset_find_impl(spa_t *spa, const char *name, | |
2845 | int func(const char *, void *), void *arg, int flags) | |
34dc7c2f BB |
2846 | { |
2847 | dsl_dir_t *dd; | |
13fe0198 | 2848 | dsl_pool_t *dp = spa_get_dsl(spa); |
b128c09f | 2849 | dsl_dataset_t *ds; |
34dc7c2f BB |
2850 | zap_cursor_t zc; |
2851 | zap_attribute_t *attr; | |
2852 | char *child; | |
b128c09f BB |
2853 | uint64_t thisobj; |
2854 | int err; | |
34dc7c2f | 2855 | |
13fe0198 MA |
2856 | dsl_pool_config_enter(dp, FTAG); |
2857 | ||
2858 | err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); | |
2859 | if (err != 0) { | |
2860 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f | 2861 | return (err); |
13fe0198 | 2862 | } |
34dc7c2f | 2863 | |
b128c09f BB |
2864 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ |
2865 | if (dd->dd_myname[0] == '$') { | |
13fe0198 MA |
2866 | dsl_dir_rele(dd, FTAG); |
2867 | dsl_pool_config_exit(dp, FTAG); | |
b128c09f BB |
2868 | return (0); |
2869 | } | |
2870 | ||
d683ddbb | 2871 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
79c76d5b | 2872 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
34dc7c2f BB |
2873 | |
2874 | /* | |
2875 | * Iterate over all children. | |
2876 | */ | |
2877 | if (flags & DS_FIND_CHILDREN) { | |
b128c09f | 2878 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
d683ddbb | 2879 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
2880 | zap_cursor_retrieve(&zc, attr) == 0; |
2881 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 MA |
2882 | ASSERT3U(attr->za_integer_length, ==, |
2883 | sizeof (uint64_t)); | |
2884 | ASSERT3U(attr->za_num_integers, ==, 1); | |
34dc7c2f | 2885 | |
428870ff | 2886 | child = kmem_asprintf("%s/%s", name, attr->za_name); |
13fe0198 MA |
2887 | dsl_pool_config_exit(dp, FTAG); |
2888 | err = dmu_objset_find_impl(spa, child, | |
2889 | func, arg, flags); | |
2890 | dsl_pool_config_enter(dp, FTAG); | |
e4f5fa12 | 2891 | kmem_strfree(child); |
13fe0198 | 2892 | if (err != 0) |
34dc7c2f BB |
2893 | break; |
2894 | } | |
2895 | zap_cursor_fini(&zc); | |
2896 | ||
13fe0198 MA |
2897 | if (err != 0) { |
2898 | dsl_dir_rele(dd, FTAG); | |
2899 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
2900 | kmem_free(attr, sizeof (zap_attribute_t)); |
2901 | return (err); | |
2902 | } | |
2903 | } | |
2904 | ||
2905 | /* | |
2906 | * Iterate over all snapshots. | |
2907 | */ | |
b128c09f | 2908 | if (flags & DS_FIND_SNAPSHOTS) { |
b128c09f | 2909 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); |
b128c09f BB |
2910 | |
2911 | if (err == 0) { | |
d683ddbb JG |
2912 | uint64_t snapobj; |
2913 | ||
2914 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
b128c09f BB |
2915 | dsl_dataset_rele(ds, FTAG); |
2916 | ||
2917 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2918 | zap_cursor_retrieve(&zc, attr) == 0; | |
2919 | (void) zap_cursor_advance(&zc)) { | |
13fe0198 | 2920 | ASSERT3U(attr->za_integer_length, ==, |
b128c09f | 2921 | sizeof (uint64_t)); |
13fe0198 | 2922 | ASSERT3U(attr->za_num_integers, ==, 1); |
b128c09f | 2923 | |
428870ff BB |
2924 | child = kmem_asprintf("%s@%s", |
2925 | name, attr->za_name); | |
13fe0198 MA |
2926 | dsl_pool_config_exit(dp, FTAG); |
2927 | err = func(child, arg); | |
2928 | dsl_pool_config_enter(dp, FTAG); | |
e4f5fa12 | 2929 | kmem_strfree(child); |
13fe0198 | 2930 | if (err != 0) |
b128c09f BB |
2931 | break; |
2932 | } | |
2933 | zap_cursor_fini(&zc); | |
34dc7c2f | 2934 | } |
34dc7c2f BB |
2935 | } |
2936 | ||
13fe0198 | 2937 | dsl_dir_rele(dd, FTAG); |
34dc7c2f | 2938 | kmem_free(attr, sizeof (zap_attribute_t)); |
13fe0198 | 2939 | dsl_pool_config_exit(dp, FTAG); |
34dc7c2f | 2940 | |
13fe0198 | 2941 | if (err != 0) |
34dc7c2f BB |
2942 | return (err); |
2943 | ||
13fe0198 MA |
2944 | /* Apply to self. */ |
2945 | return (func(name, arg)); | |
34dc7c2f BB |
2946 | } |
2947 | ||
13fe0198 MA |
2948 | /* |
2949 | * See comment above dmu_objset_find_impl(). | |
2950 | */ | |
d164b209 | 2951 | int |
5df7e9d8 | 2952 | dmu_objset_find(const char *name, int func(const char *, void *), void *arg, |
13fe0198 | 2953 | int flags) |
d164b209 | 2954 | { |
13fe0198 MA |
2955 | spa_t *spa; |
2956 | int error; | |
d164b209 | 2957 | |
13fe0198 MA |
2958 | error = spa_open(name, &spa, FTAG); |
2959 | if (error != 0) | |
2960 | return (error); | |
2961 | error = dmu_objset_find_impl(spa, name, func, arg, flags); | |
2962 | spa_close(spa, FTAG); | |
2963 | return (error); | |
d164b209 BB |
2964 | } |
2965 | ||
ae76f45c TC |
2966 | boolean_t |
2967 | dmu_objset_incompatible_encryption_version(objset_t *os) | |
2968 | { | |
2969 | return (dsl_dir_incompatible_encryption_version( | |
2970 | os->os_dsl_dataset->ds_dir)); | |
2971 | } | |
2972 | ||
34dc7c2f BB |
2973 | void |
2974 | dmu_objset_set_user(objset_t *os, void *user_ptr) | |
2975 | { | |
428870ff BB |
2976 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2977 | os->os_user_ptr = user_ptr; | |
34dc7c2f BB |
2978 | } |
2979 | ||
2980 | void * | |
2981 | dmu_objset_get_user(objset_t *os) | |
2982 | { | |
428870ff BB |
2983 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2984 | return (os->os_user_ptr); | |
34dc7c2f | 2985 | } |
c28b2279 | 2986 | |
13fe0198 MA |
2987 | /* |
2988 | * Determine name of filesystem, given name of snapshot. | |
eca7b760 | 2989 | * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes |
13fe0198 MA |
2990 | */ |
2991 | int | |
2992 | dmu_fsname(const char *snapname, char *buf) | |
2993 | { | |
2994 | char *atp = strchr(snapname, '@'); | |
2995 | if (atp == NULL) | |
2e528b49 | 2996 | return (SET_ERROR(EINVAL)); |
eca7b760 | 2997 | if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) |
2e528b49 | 2998 | return (SET_ERROR(ENAMETOOLONG)); |
13fe0198 MA |
2999 | (void) strlcpy(buf, snapname, atp - snapname + 1); |
3000 | return (0); | |
3001 | } | |
3002 | ||
3ec3bc21 | 3003 | /* |
0f8ff49e SD |
3004 | * Call when we think we're going to write/free space in open context |
3005 | * to track the amount of dirty data in the open txg, which is also the | |
3006 | * amount of memory that can not be evicted until this txg syncs. | |
3007 | * | |
3008 | * Note that there are two conditions where this can be called from | |
3009 | * syncing context: | |
3010 | * | |
3011 | * [1] When we just created the dataset, in which case we go on with | |
3012 | * updating any accounting of dirty data as usual. | |
3013 | * [2] When we are dirtying MOS data, in which case we only update the | |
3014 | * pool's accounting of dirty data. | |
3ec3bc21 BB |
3015 | */ |
3016 | void | |
3017 | dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) | |
3018 | { | |
3019 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
3020 | int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); | |
3021 | ||
3022 | if (ds != NULL) { | |
3023 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
3ec3bc21 | 3024 | } |
0f8ff49e SD |
3025 | |
3026 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
3ec3bc21 BB |
3027 | } |
3028 | ||
93ce2b4c | 3029 | #if defined(_KERNEL) |
f0fd83be | 3030 | EXPORT_SYMBOL(dmu_objset_zil); |
c28b2279 | 3031 | EXPORT_SYMBOL(dmu_objset_pool); |
f0fd83be BB |
3032 | EXPORT_SYMBOL(dmu_objset_ds); |
3033 | EXPORT_SYMBOL(dmu_objset_type); | |
c28b2279 BB |
3034 | EXPORT_SYMBOL(dmu_objset_name); |
3035 | EXPORT_SYMBOL(dmu_objset_hold); | |
b5256303 | 3036 | EXPORT_SYMBOL(dmu_objset_hold_flags); |
c28b2279 BB |
3037 | EXPORT_SYMBOL(dmu_objset_own); |
3038 | EXPORT_SYMBOL(dmu_objset_rele); | |
b5256303 | 3039 | EXPORT_SYMBOL(dmu_objset_rele_flags); |
c28b2279 BB |
3040 | EXPORT_SYMBOL(dmu_objset_disown); |
3041 | EXPORT_SYMBOL(dmu_objset_from_ds); | |
3042 | EXPORT_SYMBOL(dmu_objset_create); | |
3043 | EXPORT_SYMBOL(dmu_objset_clone); | |
c28b2279 BB |
3044 | EXPORT_SYMBOL(dmu_objset_stats); |
3045 | EXPORT_SYMBOL(dmu_objset_fast_stat); | |
54a179e7 | 3046 | EXPORT_SYMBOL(dmu_objset_spa); |
c28b2279 BB |
3047 | EXPORT_SYMBOL(dmu_objset_space); |
3048 | EXPORT_SYMBOL(dmu_objset_fsid_guid); | |
3049 | EXPORT_SYMBOL(dmu_objset_find); | |
c28b2279 BB |
3050 | EXPORT_SYMBOL(dmu_objset_byteswap); |
3051 | EXPORT_SYMBOL(dmu_objset_evict_dbufs); | |
3052 | EXPORT_SYMBOL(dmu_objset_snap_cmtime); | |
50c957f7 | 3053 | EXPORT_SYMBOL(dmu_objset_dnodesize); |
c28b2279 BB |
3054 | |
3055 | EXPORT_SYMBOL(dmu_objset_sync); | |
3056 | EXPORT_SYMBOL(dmu_objset_is_dirty); | |
b5256303 | 3057 | EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); |
c28b2279 BB |
3058 | EXPORT_SYMBOL(dmu_objset_create_impl); |
3059 | EXPORT_SYMBOL(dmu_objset_open_impl); | |
3060 | EXPORT_SYMBOL(dmu_objset_evict); | |
3061 | EXPORT_SYMBOL(dmu_objset_register_type); | |
ba67d821 | 3062 | EXPORT_SYMBOL(dmu_objset_sync_done); |
c28b2279 BB |
3063 | EXPORT_SYMBOL(dmu_objset_userquota_get_ids); |
3064 | EXPORT_SYMBOL(dmu_objset_userused_enabled); | |
3065 | EXPORT_SYMBOL(dmu_objset_userspace_upgrade); | |
3066 | EXPORT_SYMBOL(dmu_objset_userspace_present); | |
1de321e6 | 3067 | EXPORT_SYMBOL(dmu_objset_userobjused_enabled); |
126ae9f4 | 3068 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); |
1de321e6 | 3069 | EXPORT_SYMBOL(dmu_objset_userobjspace_present); |
9c5167d1 NF |
3070 | EXPORT_SYMBOL(dmu_objset_projectquota_enabled); |
3071 | EXPORT_SYMBOL(dmu_objset_projectquota_present); | |
3072 | EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); | |
3073 | EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); | |
c28b2279 | 3074 | #endif |