]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Copyright (c) 2012, 2018 by Delphix. All rights reserved. | |
25 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. | |
26 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. | |
27 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. | |
28 | * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. | |
29 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. | |
30 | * Copyright 2017 Nexenta Systems, Inc. | |
31 | * Copyright (c) 2017 Open-E, Inc. All Rights Reserved. | |
32 | * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved. | |
33 | */ | |
34 | ||
35 | /* Portions Copyright 2010 Robert Milkowski */ | |
36 | ||
37 | #include <sys/zfeature.h> | |
38 | #include <sys/cred.h> | |
39 | #include <sys/zfs_context.h> | |
40 | #include <sys/dmu_objset.h> | |
41 | #include <sys/dsl_dir.h> | |
42 | #include <sys/dsl_dataset.h> | |
43 | #include <sys/dsl_prop.h> | |
44 | #include <sys/dsl_pool.h> | |
45 | #include <sys/dsl_synctask.h> | |
46 | #include <sys/dsl_deleg.h> | |
47 | #include <sys/dnode.h> | |
48 | #include <sys/dbuf.h> | |
49 | #include <sys/zvol.h> | |
50 | #include <sys/dmu_tx.h> | |
51 | #include <sys/zap.h> | |
52 | #include <sys/zil.h> | |
53 | #include <sys/dmu_impl.h> | |
54 | #include <sys/zfs_ioctl.h> | |
55 | #include <sys/sa.h> | |
56 | #include <sys/zfs_onexit.h> | |
57 | #include <sys/dsl_destroy.h> | |
58 | #include <sys/vdev.h> | |
59 | #include <sys/zfeature.h> | |
60 | #include <sys/policy.h> | |
61 | #include <sys/spa_impl.h> | |
62 | #include <sys/dmu_recv.h> | |
63 | #include <sys/zfs_project.h> | |
64 | #include "zfs_namecheck.h" | |
65 | ||
66 | /* | |
67 | * Needed to close a window in dnode_move() that allows the objset to be freed | |
68 | * before it can be safely accessed. | |
69 | */ | |
70 | krwlock_t os_lock; | |
71 | ||
72 | /* | |
73 | * Tunable to overwrite the maximum number of threads for the parallelization | |
74 | * of dmu_objset_find_dp, needed to speed up the import of pools with many | |
75 | * datasets. | |
76 | * Default is 4 times the number of leaf vdevs. | |
77 | */ | |
78 | int dmu_find_threads = 0; | |
79 | ||
80 | /* | |
81 | * Backfill lower metadnode objects after this many have been freed. | |
82 | * Backfilling negatively impacts object creation rates, so only do it | |
83 | * if there are enough holes to fill. | |
84 | */ | |
85 | int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; | |
86 | ||
87 | static char *upgrade_tag = "upgrade_tag"; | |
88 | ||
89 | static void dmu_objset_find_dp_cb(void *arg); | |
90 | ||
91 | static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); | |
92 | static void dmu_objset_upgrade_stop(objset_t *os); | |
93 | ||
94 | void | |
95 | dmu_objset_init(void) | |
96 | { | |
97 | rw_init(&os_lock, NULL, RW_DEFAULT, NULL); | |
98 | } | |
99 | ||
100 | void | |
101 | dmu_objset_fini(void) | |
102 | { | |
103 | rw_destroy(&os_lock); | |
104 | } | |
105 | ||
106 | spa_t * | |
107 | dmu_objset_spa(objset_t *os) | |
108 | { | |
109 | return (os->os_spa); | |
110 | } | |
111 | ||
112 | zilog_t * | |
113 | dmu_objset_zil(objset_t *os) | |
114 | { | |
115 | return (os->os_zil); | |
116 | } | |
117 | ||
118 | dsl_pool_t * | |
119 | dmu_objset_pool(objset_t *os) | |
120 | { | |
121 | dsl_dataset_t *ds; | |
122 | ||
123 | if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) | |
124 | return (ds->ds_dir->dd_pool); | |
125 | else | |
126 | return (spa_get_dsl(os->os_spa)); | |
127 | } | |
128 | ||
129 | dsl_dataset_t * | |
130 | dmu_objset_ds(objset_t *os) | |
131 | { | |
132 | return (os->os_dsl_dataset); | |
133 | } | |
134 | ||
135 | dmu_objset_type_t | |
136 | dmu_objset_type(objset_t *os) | |
137 | { | |
138 | return (os->os_phys->os_type); | |
139 | } | |
140 | ||
141 | void | |
142 | dmu_objset_name(objset_t *os, char *buf) | |
143 | { | |
144 | dsl_dataset_name(os->os_dsl_dataset, buf); | |
145 | } | |
146 | ||
147 | uint64_t | |
148 | dmu_objset_id(objset_t *os) | |
149 | { | |
150 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
151 | ||
152 | return (ds ? ds->ds_object : 0); | |
153 | } | |
154 | ||
155 | uint64_t | |
156 | dmu_objset_dnodesize(objset_t *os) | |
157 | { | |
158 | return (os->os_dnodesize); | |
159 | } | |
160 | ||
161 | zfs_sync_type_t | |
162 | dmu_objset_syncprop(objset_t *os) | |
163 | { | |
164 | return (os->os_sync); | |
165 | } | |
166 | ||
167 | zfs_logbias_op_t | |
168 | dmu_objset_logbias(objset_t *os) | |
169 | { | |
170 | return (os->os_logbias); | |
171 | } | |
172 | ||
173 | static void | |
174 | checksum_changed_cb(void *arg, uint64_t newval) | |
175 | { | |
176 | objset_t *os = arg; | |
177 | ||
178 | /* | |
179 | * Inheritance should have been done by now. | |
180 | */ | |
181 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
182 | ||
183 | os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); | |
184 | } | |
185 | ||
186 | static void | |
187 | compression_changed_cb(void *arg, uint64_t newval) | |
188 | { | |
189 | objset_t *os = arg; | |
190 | ||
191 | /* | |
192 | * Inheritance and range checking should have been done by now. | |
193 | */ | |
194 | ASSERT(newval != ZIO_COMPRESS_INHERIT); | |
195 | ||
196 | os->os_compress = zio_compress_select(os->os_spa, newval, | |
197 | ZIO_COMPRESS_ON); | |
198 | } | |
199 | ||
200 | static void | |
201 | copies_changed_cb(void *arg, uint64_t newval) | |
202 | { | |
203 | objset_t *os = arg; | |
204 | ||
205 | /* | |
206 | * Inheritance and range checking should have been done by now. | |
207 | */ | |
208 | ASSERT(newval > 0); | |
209 | ASSERT(newval <= spa_max_replication(os->os_spa)); | |
210 | ||
211 | os->os_copies = newval; | |
212 | } | |
213 | ||
214 | static void | |
215 | dedup_changed_cb(void *arg, uint64_t newval) | |
216 | { | |
217 | objset_t *os = arg; | |
218 | spa_t *spa = os->os_spa; | |
219 | enum zio_checksum checksum; | |
220 | ||
221 | /* | |
222 | * Inheritance should have been done by now. | |
223 | */ | |
224 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
225 | ||
226 | checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); | |
227 | ||
228 | os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; | |
229 | os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); | |
230 | } | |
231 | ||
232 | static void | |
233 | primary_cache_changed_cb(void *arg, uint64_t newval) | |
234 | { | |
235 | objset_t *os = arg; | |
236 | ||
237 | /* | |
238 | * Inheritance and range checking should have been done by now. | |
239 | */ | |
240 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
241 | newval == ZFS_CACHE_METADATA); | |
242 | ||
243 | os->os_primary_cache = newval; | |
244 | } | |
245 | ||
246 | static void | |
247 | secondary_cache_changed_cb(void *arg, uint64_t newval) | |
248 | { | |
249 | objset_t *os = arg; | |
250 | ||
251 | /* | |
252 | * Inheritance and range checking should have been done by now. | |
253 | */ | |
254 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
255 | newval == ZFS_CACHE_METADATA); | |
256 | ||
257 | os->os_secondary_cache = newval; | |
258 | } | |
259 | ||
260 | static void | |
261 | sync_changed_cb(void *arg, uint64_t newval) | |
262 | { | |
263 | objset_t *os = arg; | |
264 | ||
265 | /* | |
266 | * Inheritance and range checking should have been done by now. | |
267 | */ | |
268 | ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || | |
269 | newval == ZFS_SYNC_DISABLED); | |
270 | ||
271 | os->os_sync = newval; | |
272 | if (os->os_zil) | |
273 | zil_set_sync(os->os_zil, newval); | |
274 | } | |
275 | ||
276 | static void | |
277 | redundant_metadata_changed_cb(void *arg, uint64_t newval) | |
278 | { | |
279 | objset_t *os = arg; | |
280 | ||
281 | /* | |
282 | * Inheritance and range checking should have been done by now. | |
283 | */ | |
284 | ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || | |
285 | newval == ZFS_REDUNDANT_METADATA_MOST); | |
286 | ||
287 | os->os_redundant_metadata = newval; | |
288 | } | |
289 | ||
290 | static void | |
291 | dnodesize_changed_cb(void *arg, uint64_t newval) | |
292 | { | |
293 | objset_t *os = arg; | |
294 | ||
295 | switch (newval) { | |
296 | case ZFS_DNSIZE_LEGACY: | |
297 | os->os_dnodesize = DNODE_MIN_SIZE; | |
298 | break; | |
299 | case ZFS_DNSIZE_AUTO: | |
300 | /* | |
301 | * Choose a dnode size that will work well for most | |
302 | * workloads if the user specified "auto". Future code | |
303 | * improvements could dynamically select a dnode size | |
304 | * based on observed workload patterns. | |
305 | */ | |
306 | os->os_dnodesize = DNODE_MIN_SIZE * 2; | |
307 | break; | |
308 | case ZFS_DNSIZE_1K: | |
309 | case ZFS_DNSIZE_2K: | |
310 | case ZFS_DNSIZE_4K: | |
311 | case ZFS_DNSIZE_8K: | |
312 | case ZFS_DNSIZE_16K: | |
313 | os->os_dnodesize = newval; | |
314 | break; | |
315 | } | |
316 | } | |
317 | ||
318 | static void | |
319 | smallblk_changed_cb(void *arg, uint64_t newval) | |
320 | { | |
321 | objset_t *os = arg; | |
322 | ||
323 | /* | |
324 | * Inheritance and range checking should have been done by now. | |
325 | */ | |
326 | ASSERT(newval <= SPA_OLD_MAXBLOCKSIZE); | |
327 | ASSERT(ISP2(newval)); | |
328 | ||
329 | os->os_zpl_special_smallblock = newval; | |
330 | } | |
331 | ||
332 | static void | |
333 | logbias_changed_cb(void *arg, uint64_t newval) | |
334 | { | |
335 | objset_t *os = arg; | |
336 | ||
337 | ASSERT(newval == ZFS_LOGBIAS_LATENCY || | |
338 | newval == ZFS_LOGBIAS_THROUGHPUT); | |
339 | os->os_logbias = newval; | |
340 | if (os->os_zil) | |
341 | zil_set_logbias(os->os_zil, newval); | |
342 | } | |
343 | ||
344 | static void | |
345 | recordsize_changed_cb(void *arg, uint64_t newval) | |
346 | { | |
347 | objset_t *os = arg; | |
348 | ||
349 | os->os_recordsize = newval; | |
350 | } | |
351 | ||
352 | void | |
353 | dmu_objset_byteswap(void *buf, size_t size) | |
354 | { | |
355 | objset_phys_t *osp = buf; | |
356 | ||
357 | ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || | |
358 | size == sizeof (objset_phys_t)); | |
359 | dnode_byteswap(&osp->os_meta_dnode); | |
360 | byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); | |
361 | osp->os_type = BSWAP_64(osp->os_type); | |
362 | osp->os_flags = BSWAP_64(osp->os_flags); | |
363 | if (size >= OBJSET_PHYS_SIZE_V2) { | |
364 | dnode_byteswap(&osp->os_userused_dnode); | |
365 | dnode_byteswap(&osp->os_groupused_dnode); | |
366 | if (size >= sizeof (objset_phys_t)) | |
367 | dnode_byteswap(&osp->os_projectused_dnode); | |
368 | } | |
369 | } | |
370 | ||
371 | /* | |
372 | * The hash is a CRC-based hash of the objset_t pointer and the object number. | |
373 | */ | |
374 | static uint64_t | |
375 | dnode_hash(const objset_t *os, uint64_t obj) | |
376 | { | |
377 | uintptr_t osv = (uintptr_t)os; | |
378 | uint64_t crc = -1ULL; | |
379 | ||
380 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
381 | /* | |
382 | * The low 6 bits of the pointer don't have much entropy, because | |
383 | * the objset_t is larger than 2^6 bytes long. | |
384 | */ | |
385 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; | |
386 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; | |
387 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; | |
388 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; | |
389 | ||
390 | crc ^= (osv>>14) ^ (obj>>24); | |
391 | ||
392 | return (crc); | |
393 | } | |
394 | ||
395 | unsigned int | |
396 | dnode_multilist_index_func(multilist_t *ml, void *obj) | |
397 | { | |
398 | dnode_t *dn = obj; | |
399 | return (dnode_hash(dn->dn_objset, dn->dn_object) % | |
400 | multilist_get_num_sublists(ml)); | |
401 | } | |
402 | ||
403 | /* | |
404 | * Instantiates the objset_t in-memory structure corresponding to the | |
405 | * objset_phys_t that's pointed to by the specified blkptr_t. | |
406 | */ | |
407 | int | |
408 | dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
409 | objset_t **osp) | |
410 | { | |
411 | objset_t *os; | |
412 | int i, err; | |
413 | ||
414 | ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); | |
415 | ||
416 | /* | |
417 | * The $ORIGIN dataset (if it exists) doesn't have an associated | |
418 | * objset, so there's no reason to open it. The $ORIGIN dataset | |
419 | * will not exist on pools older than SPA_VERSION_ORIGIN. | |
420 | */ | |
421 | if (ds != NULL && spa_get_dsl(spa) != NULL && | |
422 | spa_get_dsl(spa)->dp_origin_snap != NULL) { | |
423 | ASSERT3P(ds->ds_dir, !=, | |
424 | spa_get_dsl(spa)->dp_origin_snap->ds_dir); | |
425 | } | |
426 | ||
427 | os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); | |
428 | os->os_dsl_dataset = ds; | |
429 | os->os_spa = spa; | |
430 | os->os_rootbp = bp; | |
431 | if (!BP_IS_HOLE(os->os_rootbp)) { | |
432 | arc_flags_t aflags = ARC_FLAG_WAIT; | |
433 | zbookmark_phys_t zb; | |
434 | int size; | |
435 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; | |
436 | SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, | |
437 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
438 | ||
439 | if (DMU_OS_IS_L2CACHEABLE(os)) | |
440 | aflags |= ARC_FLAG_L2CACHE; | |
441 | ||
442 | if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { | |
443 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
444 | ASSERT(BP_IS_AUTHENTICATED(bp)); | |
445 | zio_flags |= ZIO_FLAG_RAW; | |
446 | } | |
447 | ||
448 | dprintf_bp(os->os_rootbp, "reading %s", ""); | |
449 | err = arc_read(NULL, spa, os->os_rootbp, | |
450 | arc_getbuf_func, &os->os_phys_buf, | |
451 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); | |
452 | if (err != 0) { | |
453 | kmem_free(os, sizeof (objset_t)); | |
454 | /* convert checksum errors into IO errors */ | |
455 | if (err == ECKSUM) | |
456 | err = SET_ERROR(EIO); | |
457 | return (err); | |
458 | } | |
459 | ||
460 | if (spa_version(spa) < SPA_VERSION_USERSPACE) | |
461 | size = OBJSET_PHYS_SIZE_V1; | |
462 | else if (!spa_feature_is_enabled(spa, | |
463 | SPA_FEATURE_PROJECT_QUOTA)) | |
464 | size = OBJSET_PHYS_SIZE_V2; | |
465 | else | |
466 | size = sizeof (objset_phys_t); | |
467 | ||
468 | /* Increase the blocksize if we are permitted. */ | |
469 | if (arc_buf_size(os->os_phys_buf) < size) { | |
470 | arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, | |
471 | ARC_BUFC_METADATA, size); | |
472 | bzero(buf->b_data, size); | |
473 | bcopy(os->os_phys_buf->b_data, buf->b_data, | |
474 | arc_buf_size(os->os_phys_buf)); | |
475 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); | |
476 | os->os_phys_buf = buf; | |
477 | } | |
478 | ||
479 | os->os_phys = os->os_phys_buf->b_data; | |
480 | os->os_flags = os->os_phys->os_flags; | |
481 | } else { | |
482 | int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? | |
483 | sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; | |
484 | os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, | |
485 | ARC_BUFC_METADATA, size); | |
486 | os->os_phys = os->os_phys_buf->b_data; | |
487 | bzero(os->os_phys, size); | |
488 | } | |
489 | /* | |
490 | * These properties will be filled in by the logic in zfs_get_zplprop() | |
491 | * when they are queried for the first time. | |
492 | */ | |
493 | os->os_version = OBJSET_PROP_UNINITIALIZED; | |
494 | os->os_normalization = OBJSET_PROP_UNINITIALIZED; | |
495 | os->os_utf8only = OBJSET_PROP_UNINITIALIZED; | |
496 | os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; | |
497 | ||
498 | /* | |
499 | * Note: the changed_cb will be called once before the register | |
500 | * func returns, thus changing the checksum/compression from the | |
501 | * default (fletcher2/off). Snapshots don't need to know about | |
502 | * checksum/compression/copies. | |
503 | */ | |
504 | if (ds != NULL) { | |
505 | boolean_t needlock = B_FALSE; | |
506 | ||
507 | os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); | |
508 | ||
509 | /* | |
510 | * Note: it's valid to open the objset if the dataset is | |
511 | * long-held, in which case the pool_config lock will not | |
512 | * be held. | |
513 | */ | |
514 | if (!dsl_pool_config_held(dmu_objset_pool(os))) { | |
515 | needlock = B_TRUE; | |
516 | dsl_pool_config_enter(dmu_objset_pool(os), FTAG); | |
517 | } | |
518 | ||
519 | err = dsl_prop_register(ds, | |
520 | zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), | |
521 | primary_cache_changed_cb, os); | |
522 | if (err == 0) { | |
523 | err = dsl_prop_register(ds, | |
524 | zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), | |
525 | secondary_cache_changed_cb, os); | |
526 | } | |
527 | if (!ds->ds_is_snapshot) { | |
528 | if (err == 0) { | |
529 | err = dsl_prop_register(ds, | |
530 | zfs_prop_to_name(ZFS_PROP_CHECKSUM), | |
531 | checksum_changed_cb, os); | |
532 | } | |
533 | if (err == 0) { | |
534 | err = dsl_prop_register(ds, | |
535 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
536 | compression_changed_cb, os); | |
537 | } | |
538 | if (err == 0) { | |
539 | err = dsl_prop_register(ds, | |
540 | zfs_prop_to_name(ZFS_PROP_COPIES), | |
541 | copies_changed_cb, os); | |
542 | } | |
543 | if (err == 0) { | |
544 | err = dsl_prop_register(ds, | |
545 | zfs_prop_to_name(ZFS_PROP_DEDUP), | |
546 | dedup_changed_cb, os); | |
547 | } | |
548 | if (err == 0) { | |
549 | err = dsl_prop_register(ds, | |
550 | zfs_prop_to_name(ZFS_PROP_LOGBIAS), | |
551 | logbias_changed_cb, os); | |
552 | } | |
553 | if (err == 0) { | |
554 | err = dsl_prop_register(ds, | |
555 | zfs_prop_to_name(ZFS_PROP_SYNC), | |
556 | sync_changed_cb, os); | |
557 | } | |
558 | if (err == 0) { | |
559 | err = dsl_prop_register(ds, | |
560 | zfs_prop_to_name( | |
561 | ZFS_PROP_REDUNDANT_METADATA), | |
562 | redundant_metadata_changed_cb, os); | |
563 | } | |
564 | if (err == 0) { | |
565 | err = dsl_prop_register(ds, | |
566 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
567 | recordsize_changed_cb, os); | |
568 | } | |
569 | if (err == 0) { | |
570 | err = dsl_prop_register(ds, | |
571 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
572 | dnodesize_changed_cb, os); | |
573 | } | |
574 | if (err == 0) { | |
575 | err = dsl_prop_register(ds, | |
576 | zfs_prop_to_name( | |
577 | ZFS_PROP_SPECIAL_SMALL_BLOCKS), | |
578 | smallblk_changed_cb, os); | |
579 | } | |
580 | } | |
581 | if (needlock) | |
582 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); | |
583 | if (err != 0) { | |
584 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); | |
585 | kmem_free(os, sizeof (objset_t)); | |
586 | return (err); | |
587 | } | |
588 | } else { | |
589 | /* It's the meta-objset. */ | |
590 | os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; | |
591 | os->os_compress = ZIO_COMPRESS_ON; | |
592 | os->os_encrypted = B_FALSE; | |
593 | os->os_copies = spa_max_replication(spa); | |
594 | os->os_dedup_checksum = ZIO_CHECKSUM_OFF; | |
595 | os->os_dedup_verify = B_FALSE; | |
596 | os->os_logbias = ZFS_LOGBIAS_LATENCY; | |
597 | os->os_sync = ZFS_SYNC_STANDARD; | |
598 | os->os_primary_cache = ZFS_CACHE_ALL; | |
599 | os->os_secondary_cache = ZFS_CACHE_ALL; | |
600 | os->os_dnodesize = DNODE_MIN_SIZE; | |
601 | } | |
602 | ||
603 | if (ds == NULL || !ds->ds_is_snapshot) | |
604 | os->os_zil_header = os->os_phys->os_zil_header; | |
605 | os->os_zil = zil_alloc(os, &os->os_zil_header); | |
606 | ||
607 | for (i = 0; i < TXG_SIZE; i++) { | |
608 | os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), | |
609 | offsetof(dnode_t, dn_dirty_link[i]), | |
610 | dnode_multilist_index_func); | |
611 | } | |
612 | list_create(&os->os_dnodes, sizeof (dnode_t), | |
613 | offsetof(dnode_t, dn_link)); | |
614 | list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), | |
615 | offsetof(dmu_buf_impl_t, db_link)); | |
616 | ||
617 | list_link_init(&os->os_evicting_node); | |
618 | ||
619 | mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); | |
620 | mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); | |
621 | mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); | |
622 | mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); | |
623 | os->os_obj_next_percpu_len = boot_ncpus; | |
624 | os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * | |
625 | sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); | |
626 | ||
627 | dnode_special_open(os, &os->os_phys->os_meta_dnode, | |
628 | DMU_META_DNODE_OBJECT, &os->os_meta_dnode); | |
629 | if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { | |
630 | dnode_special_open(os, &os->os_phys->os_userused_dnode, | |
631 | DMU_USERUSED_OBJECT, &os->os_userused_dnode); | |
632 | dnode_special_open(os, &os->os_phys->os_groupused_dnode, | |
633 | DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); | |
634 | if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) | |
635 | dnode_special_open(os, | |
636 | &os->os_phys->os_projectused_dnode, | |
637 | DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); | |
638 | } | |
639 | ||
640 | mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); | |
641 | ||
642 | *osp = os; | |
643 | return (0); | |
644 | } | |
645 | ||
646 | int | |
647 | dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) | |
648 | { | |
649 | int err = 0; | |
650 | ||
651 | /* | |
652 | * We shouldn't be doing anything with dsl_dataset_t's unless the | |
653 | * pool_config lock is held, or the dataset is long-held. | |
654 | */ | |
655 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || | |
656 | dsl_dataset_long_held(ds)); | |
657 | ||
658 | mutex_enter(&ds->ds_opening_lock); | |
659 | if (ds->ds_objset == NULL) { | |
660 | objset_t *os; | |
661 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); | |
662 | err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), | |
663 | ds, dsl_dataset_get_blkptr(ds), &os); | |
664 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
665 | ||
666 | if (err == 0) { | |
667 | mutex_enter(&ds->ds_lock); | |
668 | ASSERT(ds->ds_objset == NULL); | |
669 | ds->ds_objset = os; | |
670 | mutex_exit(&ds->ds_lock); | |
671 | } | |
672 | } | |
673 | *osp = ds->ds_objset; | |
674 | mutex_exit(&ds->ds_opening_lock); | |
675 | return (err); | |
676 | } | |
677 | ||
678 | /* | |
679 | * Holds the pool while the objset is held. Therefore only one objset | |
680 | * can be held at a time. | |
681 | */ | |
682 | int | |
683 | dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag, | |
684 | objset_t **osp) | |
685 | { | |
686 | dsl_pool_t *dp; | |
687 | dsl_dataset_t *ds; | |
688 | int err; | |
689 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; | |
690 | ||
691 | err = dsl_pool_hold(name, tag, &dp); | |
692 | if (err != 0) | |
693 | return (err); | |
694 | err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); | |
695 | if (err != 0) { | |
696 | dsl_pool_rele(dp, tag); | |
697 | return (err); | |
698 | } | |
699 | ||
700 | err = dmu_objset_from_ds(ds, osp); | |
701 | if (err != 0) { | |
702 | dsl_dataset_rele(ds, tag); | |
703 | dsl_pool_rele(dp, tag); | |
704 | } | |
705 | ||
706 | return (err); | |
707 | } | |
708 | ||
709 | int | |
710 | dmu_objset_hold(const char *name, void *tag, objset_t **osp) | |
711 | { | |
712 | return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); | |
713 | } | |
714 | ||
715 | static int | |
716 | dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, | |
717 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) | |
718 | { | |
719 | int err; | |
720 | ||
721 | err = dmu_objset_from_ds(ds, osp); | |
722 | if (err != 0) { | |
723 | return (err); | |
724 | } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { | |
725 | return (SET_ERROR(EINVAL)); | |
726 | } else if (!readonly && dsl_dataset_is_snapshot(ds)) { | |
727 | return (SET_ERROR(EROFS)); | |
728 | } else if (!readonly && decrypt && | |
729 | dsl_dir_incompatible_encryption_version(ds->ds_dir)) { | |
730 | return (SET_ERROR(EROFS)); | |
731 | } | |
732 | ||
733 | /* if we are decrypting, we can now check MACs in os->os_phys_buf */ | |
734 | if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { | |
735 | zbookmark_phys_t zb; | |
736 | ||
737 | SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, | |
738 | ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
739 | err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, | |
740 | &zb, B_FALSE); | |
741 | if (err != 0) | |
742 | return (err); | |
743 | ||
744 | ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); | |
745 | } | |
746 | ||
747 | return (0); | |
748 | } | |
749 | ||
750 | /* | |
751 | * dsl_pool must not be held when this is called. | |
752 | * Upon successful return, there will be a longhold on the dataset, | |
753 | * and the dsl_pool will not be held. | |
754 | */ | |
755 | int | |
756 | dmu_objset_own(const char *name, dmu_objset_type_t type, | |
757 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) | |
758 | { | |
759 | dsl_pool_t *dp; | |
760 | dsl_dataset_t *ds; | |
761 | int err; | |
762 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; | |
763 | ||
764 | err = dsl_pool_hold(name, FTAG, &dp); | |
765 | if (err != 0) | |
766 | return (err); | |
767 | err = dsl_dataset_own(dp, name, flags, tag, &ds); | |
768 | if (err != 0) { | |
769 | dsl_pool_rele(dp, FTAG); | |
770 | return (err); | |
771 | } | |
772 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); | |
773 | if (err != 0) { | |
774 | dsl_dataset_disown(ds, flags, tag); | |
775 | dsl_pool_rele(dp, FTAG); | |
776 | return (err); | |
777 | } | |
778 | ||
779 | /* | |
780 | * User accounting requires the dataset to be decrypted and rw. | |
781 | * We also don't begin user accounting during claiming to help | |
782 | * speed up pool import times and to keep this txg reserved | |
783 | * completely for recovery work. | |
784 | */ | |
785 | if ((dmu_objset_userobjspace_upgradable(*osp) || | |
786 | dmu_objset_projectquota_upgradable(*osp)) && | |
787 | !readonly && !dp->dp_spa->spa_claiming && | |
788 | (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) | |
789 | dmu_objset_id_quota_upgrade(*osp); | |
790 | ||
791 | dsl_pool_rele(dp, FTAG); | |
792 | return (0); | |
793 | } | |
794 | ||
795 | int | |
796 | dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, | |
797 | boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) | |
798 | { | |
799 | dsl_dataset_t *ds; | |
800 | int err; | |
801 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; | |
802 | ||
803 | err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); | |
804 | if (err != 0) | |
805 | return (err); | |
806 | ||
807 | err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); | |
808 | if (err != 0) { | |
809 | dsl_dataset_disown(ds, flags, tag); | |
810 | return (err); | |
811 | } | |
812 | ||
813 | return (0); | |
814 | } | |
815 | ||
816 | void | |
817 | dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag) | |
818 | { | |
819 | ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; | |
820 | ||
821 | dsl_pool_t *dp = dmu_objset_pool(os); | |
822 | dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); | |
823 | dsl_pool_rele(dp, tag); | |
824 | } | |
825 | ||
826 | void | |
827 | dmu_objset_rele(objset_t *os, void *tag) | |
828 | { | |
829 | dmu_objset_rele_flags(os, B_FALSE, tag); | |
830 | } | |
831 | ||
832 | /* | |
833 | * When we are called, os MUST refer to an objset associated with a dataset | |
834 | * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner | |
835 | * == tag. We will then release and reacquire ownership of the dataset while | |
836 | * holding the pool config_rwlock to avoid intervening namespace or ownership | |
837 | * changes may occur. | |
838 | * | |
839 | * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to | |
840 | * release the hold on its dataset and acquire a new one on the dataset of the | |
841 | * same name so that it can be partially torn down and reconstructed. | |
842 | */ | |
843 | void | |
844 | dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, | |
845 | boolean_t decrypt, void *tag) | |
846 | { | |
847 | dsl_pool_t *dp; | |
848 | char name[ZFS_MAX_DATASET_NAME_LEN]; | |
849 | ||
850 | VERIFY3P(ds, !=, NULL); | |
851 | VERIFY3P(ds->ds_owner, ==, tag); | |
852 | VERIFY(dsl_dataset_long_held(ds)); | |
853 | ||
854 | dsl_dataset_name(ds, name); | |
855 | dp = ds->ds_dir->dd_pool; | |
856 | dsl_pool_config_enter(dp, FTAG); | |
857 | dsl_dataset_disown(ds, decrypt, tag); | |
858 | VERIFY0(dsl_dataset_own(dp, name, | |
859 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag, newds)); | |
860 | dsl_pool_config_exit(dp, FTAG); | |
861 | } | |
862 | ||
863 | void | |
864 | dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag) | |
865 | { | |
866 | /* | |
867 | * Stop upgrading thread | |
868 | */ | |
869 | dmu_objset_upgrade_stop(os); | |
870 | dsl_dataset_disown(os->os_dsl_dataset, | |
871 | (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag); | |
872 | } | |
873 | ||
874 | void | |
875 | dmu_objset_evict_dbufs(objset_t *os) | |
876 | { | |
877 | dnode_t *dn_marker; | |
878 | dnode_t *dn; | |
879 | ||
880 | dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); | |
881 | ||
882 | mutex_enter(&os->os_lock); | |
883 | dn = list_head(&os->os_dnodes); | |
884 | while (dn != NULL) { | |
885 | /* | |
886 | * Skip dnodes without holds. We have to do this dance | |
887 | * because dnode_add_ref() only works if there is already a | |
888 | * hold. If the dnode has no holds, then it has no dbufs. | |
889 | */ | |
890 | if (dnode_add_ref(dn, FTAG)) { | |
891 | list_insert_after(&os->os_dnodes, dn, dn_marker); | |
892 | mutex_exit(&os->os_lock); | |
893 | ||
894 | dnode_evict_dbufs(dn); | |
895 | dnode_rele(dn, FTAG); | |
896 | ||
897 | mutex_enter(&os->os_lock); | |
898 | dn = list_next(&os->os_dnodes, dn_marker); | |
899 | list_remove(&os->os_dnodes, dn_marker); | |
900 | } else { | |
901 | dn = list_next(&os->os_dnodes, dn); | |
902 | } | |
903 | } | |
904 | mutex_exit(&os->os_lock); | |
905 | ||
906 | kmem_free(dn_marker, sizeof (dnode_t)); | |
907 | ||
908 | if (DMU_USERUSED_DNODE(os) != NULL) { | |
909 | if (DMU_PROJECTUSED_DNODE(os) != NULL) | |
910 | dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); | |
911 | dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); | |
912 | dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); | |
913 | } | |
914 | dnode_evict_dbufs(DMU_META_DNODE(os)); | |
915 | } | |
916 | ||
917 | /* | |
918 | * Objset eviction processing is split into into two pieces. | |
919 | * The first marks the objset as evicting, evicts any dbufs that | |
920 | * have a refcount of zero, and then queues up the objset for the | |
921 | * second phase of eviction. Once os->os_dnodes has been cleared by | |
922 | * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. | |
923 | * The second phase closes the special dnodes, dequeues the objset from | |
924 | * the list of those undergoing eviction, and finally frees the objset. | |
925 | * | |
926 | * NOTE: Due to asynchronous eviction processing (invocation of | |
927 | * dnode_buf_pageout()), it is possible for the meta dnode for the | |
928 | * objset to have no holds even though os->os_dnodes is not empty. | |
929 | */ | |
930 | void | |
931 | dmu_objset_evict(objset_t *os) | |
932 | { | |
933 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
934 | ||
935 | for (int t = 0; t < TXG_SIZE; t++) | |
936 | ASSERT(!dmu_objset_is_dirty(os, t)); | |
937 | ||
938 | if (ds) | |
939 | dsl_prop_unregister_all(ds, os); | |
940 | ||
941 | if (os->os_sa) | |
942 | sa_tear_down(os); | |
943 | ||
944 | dmu_objset_evict_dbufs(os); | |
945 | ||
946 | mutex_enter(&os->os_lock); | |
947 | spa_evicting_os_register(os->os_spa, os); | |
948 | if (list_is_empty(&os->os_dnodes)) { | |
949 | mutex_exit(&os->os_lock); | |
950 | dmu_objset_evict_done(os); | |
951 | } else { | |
952 | mutex_exit(&os->os_lock); | |
953 | } | |
954 | ||
955 | ||
956 | } | |
957 | ||
958 | void | |
959 | dmu_objset_evict_done(objset_t *os) | |
960 | { | |
961 | ASSERT3P(list_head(&os->os_dnodes), ==, NULL); | |
962 | ||
963 | dnode_special_close(&os->os_meta_dnode); | |
964 | if (DMU_USERUSED_DNODE(os)) { | |
965 | if (DMU_PROJECTUSED_DNODE(os)) | |
966 | dnode_special_close(&os->os_projectused_dnode); | |
967 | dnode_special_close(&os->os_userused_dnode); | |
968 | dnode_special_close(&os->os_groupused_dnode); | |
969 | } | |
970 | zil_free(os->os_zil); | |
971 | ||
972 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); | |
973 | ||
974 | /* | |
975 | * This is a barrier to prevent the objset from going away in | |
976 | * dnode_move() until we can safely ensure that the objset is still in | |
977 | * use. We consider the objset valid before the barrier and invalid | |
978 | * after the barrier. | |
979 | */ | |
980 | rw_enter(&os_lock, RW_READER); | |
981 | rw_exit(&os_lock); | |
982 | ||
983 | kmem_free(os->os_obj_next_percpu, | |
984 | os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); | |
985 | ||
986 | mutex_destroy(&os->os_lock); | |
987 | mutex_destroy(&os->os_userused_lock); | |
988 | mutex_destroy(&os->os_obj_lock); | |
989 | mutex_destroy(&os->os_user_ptr_lock); | |
990 | mutex_destroy(&os->os_upgrade_lock); | |
991 | for (int i = 0; i < TXG_SIZE; i++) { | |
992 | multilist_destroy(os->os_dirty_dnodes[i]); | |
993 | } | |
994 | spa_evicting_os_deregister(os->os_spa, os); | |
995 | kmem_free(os, sizeof (objset_t)); | |
996 | } | |
997 | ||
998 | inode_timespec_t | |
999 | dmu_objset_snap_cmtime(objset_t *os) | |
1000 | { | |
1001 | return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); | |
1002 | } | |
1003 | ||
1004 | objset_t * | |
1005 | dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
1006 | dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) | |
1007 | { | |
1008 | objset_t *os; | |
1009 | dnode_t *mdn; | |
1010 | ||
1011 | ASSERT(dmu_tx_is_syncing(tx)); | |
1012 | ||
1013 | if (blksz == 0) | |
1014 | blksz = DNODE_BLOCK_SIZE; | |
1015 | if (ibs == 0) | |
1016 | ibs = DN_MAX_INDBLKSHIFT; | |
1017 | ||
1018 | if (ds != NULL) | |
1019 | VERIFY0(dmu_objset_from_ds(ds, &os)); | |
1020 | else | |
1021 | VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); | |
1022 | ||
1023 | mdn = DMU_META_DNODE(os); | |
1024 | ||
1025 | dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, | |
1026 | DNODE_MIN_SLOTS, tx); | |
1027 | ||
1028 | /* | |
1029 | * We don't want to have to increase the meta-dnode's nlevels | |
1030 | * later, because then we could do it in quescing context while | |
1031 | * we are also accessing it in open context. | |
1032 | * | |
1033 | * This precaution is not necessary for the MOS (ds == NULL), | |
1034 | * because the MOS is only updated in syncing context. | |
1035 | * This is most fortunate: the MOS is the only objset that | |
1036 | * needs to be synced multiple times as spa_sync() iterates | |
1037 | * to convergence, so minimizing its dn_nlevels matters. | |
1038 | */ | |
1039 | if (ds != NULL) { | |
1040 | if (levels == 0) { | |
1041 | levels = 1; | |
1042 | ||
1043 | /* | |
1044 | * Determine the number of levels necessary for the | |
1045 | * meta-dnode to contain DN_MAX_OBJECT dnodes. Note | |
1046 | * that in order to ensure that we do not overflow | |
1047 | * 64 bits, there has to be a nlevels that gives us a | |
1048 | * number of blocks > DN_MAX_OBJECT but < 2^64. | |
1049 | * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) | |
1050 | * (10) must be less than (64 - log2(DN_MAX_OBJECT)) | |
1051 | * (16). | |
1052 | */ | |
1053 | while ((uint64_t)mdn->dn_nblkptr << | |
1054 | (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * | |
1055 | (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < | |
1056 | DN_MAX_OBJECT) | |
1057 | levels++; | |
1058 | } | |
1059 | ||
1060 | mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = | |
1061 | mdn->dn_nlevels = levels; | |
1062 | } | |
1063 | ||
1064 | ASSERT(type != DMU_OST_NONE); | |
1065 | ASSERT(type != DMU_OST_ANY); | |
1066 | ASSERT(type < DMU_OST_NUMTYPES); | |
1067 | os->os_phys->os_type = type; | |
1068 | ||
1069 | /* | |
1070 | * Enable user accounting if it is enabled and this is not an | |
1071 | * encrypted receive. | |
1072 | */ | |
1073 | if (dmu_objset_userused_enabled(os) && | |
1074 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
1075 | os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; | |
1076 | if (dmu_objset_userobjused_enabled(os)) { | |
1077 | ds->ds_feature_activation[ | |
1078 | SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; | |
1079 | os->os_phys->os_flags |= | |
1080 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
1081 | } | |
1082 | if (dmu_objset_projectquota_enabled(os)) { | |
1083 | ds->ds_feature_activation[ | |
1084 | SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; | |
1085 | os->os_phys->os_flags |= | |
1086 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
1087 | } | |
1088 | os->os_flags = os->os_phys->os_flags; | |
1089 | } | |
1090 | ||
1091 | dsl_dataset_dirty(ds, tx); | |
1092 | ||
1093 | return (os); | |
1094 | } | |
1095 | ||
1096 | /* called from dsl for meta-objset */ | |
1097 | objset_t * | |
1098 | dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
1099 | dmu_objset_type_t type, dmu_tx_t *tx) | |
1100 | { | |
1101 | return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); | |
1102 | } | |
1103 | ||
1104 | typedef struct dmu_objset_create_arg { | |
1105 | const char *doca_name; | |
1106 | cred_t *doca_cred; | |
1107 | void (*doca_userfunc)(objset_t *os, void *arg, | |
1108 | cred_t *cr, dmu_tx_t *tx); | |
1109 | void *doca_userarg; | |
1110 | dmu_objset_type_t doca_type; | |
1111 | uint64_t doca_flags; | |
1112 | dsl_crypto_params_t *doca_dcp; | |
1113 | } dmu_objset_create_arg_t; | |
1114 | ||
1115 | /*ARGSUSED*/ | |
1116 | static int | |
1117 | dmu_objset_create_check(void *arg, dmu_tx_t *tx) | |
1118 | { | |
1119 | dmu_objset_create_arg_t *doca = arg; | |
1120 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1121 | dsl_dir_t *pdd; | |
1122 | dsl_dataset_t *parentds; | |
1123 | objset_t *parentos; | |
1124 | const char *tail; | |
1125 | int error; | |
1126 | ||
1127 | if (strchr(doca->doca_name, '@') != NULL) | |
1128 | return (SET_ERROR(EINVAL)); | |
1129 | ||
1130 | if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) | |
1131 | return (SET_ERROR(ENAMETOOLONG)); | |
1132 | ||
1133 | if (dataset_nestcheck(doca->doca_name) != 0) | |
1134 | return (SET_ERROR(ENAMETOOLONG)); | |
1135 | ||
1136 | error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); | |
1137 | if (error != 0) | |
1138 | return (error); | |
1139 | if (tail == NULL) { | |
1140 | dsl_dir_rele(pdd, FTAG); | |
1141 | return (SET_ERROR(EEXIST)); | |
1142 | } | |
1143 | ||
1144 | error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); | |
1145 | if (error != 0) { | |
1146 | dsl_dir_rele(pdd, FTAG); | |
1147 | return (error); | |
1148 | } | |
1149 | ||
1150 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, | |
1151 | doca->doca_cred); | |
1152 | if (error != 0) { | |
1153 | dsl_dir_rele(pdd, FTAG); | |
1154 | return (error); | |
1155 | } | |
1156 | ||
1157 | /* can't create below anything but filesystems (eg. no ZVOLs) */ | |
1158 | error = dsl_dataset_hold_obj(pdd->dd_pool, | |
1159 | dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); | |
1160 | if (error != 0) { | |
1161 | dsl_dir_rele(pdd, FTAG); | |
1162 | return (error); | |
1163 | } | |
1164 | error = dmu_objset_from_ds(parentds, &parentos); | |
1165 | if (error != 0) { | |
1166 | dsl_dataset_rele(parentds, FTAG); | |
1167 | dsl_dir_rele(pdd, FTAG); | |
1168 | return (error); | |
1169 | } | |
1170 | if (dmu_objset_type(parentos) != DMU_OST_ZFS) { | |
1171 | dsl_dataset_rele(parentds, FTAG); | |
1172 | dsl_dir_rele(pdd, FTAG); | |
1173 | return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); | |
1174 | } | |
1175 | dsl_dataset_rele(parentds, FTAG); | |
1176 | dsl_dir_rele(pdd, FTAG); | |
1177 | ||
1178 | return (error); | |
1179 | } | |
1180 | ||
1181 | static void | |
1182 | dmu_objset_create_sync(void *arg, dmu_tx_t *tx) | |
1183 | { | |
1184 | dmu_objset_create_arg_t *doca = arg; | |
1185 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1186 | spa_t *spa = dp->dp_spa; | |
1187 | dsl_dir_t *pdd; | |
1188 | const char *tail; | |
1189 | dsl_dataset_t *ds; | |
1190 | uint64_t obj; | |
1191 | blkptr_t *bp; | |
1192 | objset_t *os; | |
1193 | zio_t *rzio; | |
1194 | ||
1195 | VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); | |
1196 | ||
1197 | obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, | |
1198 | doca->doca_cred, doca->doca_dcp, tx); | |
1199 | ||
1200 | VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, | |
1201 | DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); | |
1202 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); | |
1203 | bp = dsl_dataset_get_blkptr(ds); | |
1204 | os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); | |
1205 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
1206 | ||
1207 | if (doca->doca_userfunc != NULL) { | |
1208 | doca->doca_userfunc(os, doca->doca_userarg, | |
1209 | doca->doca_cred, tx); | |
1210 | } | |
1211 | ||
1212 | /* | |
1213 | * The doca_userfunc() may write out some data that needs to be | |
1214 | * encrypted if the dataset is encrypted (specifically the root | |
1215 | * directory). This data must be written out before the encryption | |
1216 | * key mapping is removed by dsl_dataset_rele_flags(). Force the | |
1217 | * I/O to occur immediately by invoking the relevant sections of | |
1218 | * dsl_pool_sync(). | |
1219 | */ | |
1220 | if (os->os_encrypted) { | |
1221 | dsl_dataset_t *tmpds = NULL; | |
1222 | boolean_t need_sync_done = B_FALSE; | |
1223 | ||
1224 | mutex_enter(&ds->ds_lock); | |
1225 | ds->ds_owner = FTAG; | |
1226 | mutex_exit(&ds->ds_lock); | |
1227 | ||
1228 | rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
1229 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, | |
1230 | tx->tx_txg); | |
1231 | if (tmpds != NULL) { | |
1232 | dsl_dataset_sync(ds, rzio, tx); | |
1233 | need_sync_done = B_TRUE; | |
1234 | } | |
1235 | VERIFY0(zio_wait(rzio)); | |
1236 | ||
1237 | dmu_objset_do_userquota_updates(os, tx); | |
1238 | taskq_wait(dp->dp_sync_taskq); | |
1239 | if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { | |
1240 | ASSERT3P(ds->ds_key_mapping, !=, NULL); | |
1241 | key_mapping_rele(spa, ds->ds_key_mapping, ds); | |
1242 | } | |
1243 | ||
1244 | rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
1245 | tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, | |
1246 | tx->tx_txg); | |
1247 | if (tmpds != NULL) { | |
1248 | dmu_buf_rele(ds->ds_dbuf, ds); | |
1249 | dsl_dataset_sync(ds, rzio, tx); | |
1250 | } | |
1251 | VERIFY0(zio_wait(rzio)); | |
1252 | ||
1253 | if (need_sync_done) { | |
1254 | ASSERT3P(ds->ds_key_mapping, !=, NULL); | |
1255 | key_mapping_rele(spa, ds->ds_key_mapping, ds); | |
1256 | dsl_dataset_sync_done(ds, tx); | |
1257 | } | |
1258 | ||
1259 | mutex_enter(&ds->ds_lock); | |
1260 | ds->ds_owner = NULL; | |
1261 | mutex_exit(&ds->ds_lock); | |
1262 | } | |
1263 | ||
1264 | spa_history_log_internal_ds(ds, "create", tx, ""); | |
1265 | zvol_create_minors(spa, doca->doca_name, B_TRUE); | |
1266 | ||
1267 | dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); | |
1268 | dsl_dir_rele(pdd, FTAG); | |
1269 | } | |
1270 | ||
1271 | int | |
1272 | dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, | |
1273 | dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) | |
1274 | { | |
1275 | dmu_objset_create_arg_t doca; | |
1276 | dsl_crypto_params_t tmp_dcp = { 0 }; | |
1277 | ||
1278 | doca.doca_name = name; | |
1279 | doca.doca_cred = CRED(); | |
1280 | doca.doca_flags = flags; | |
1281 | doca.doca_userfunc = func; | |
1282 | doca.doca_userarg = arg; | |
1283 | doca.doca_type = type; | |
1284 | ||
1285 | /* | |
1286 | * Some callers (mostly for testing) do not provide a dcp on their | |
1287 | * own but various code inside the sync task will require it to be | |
1288 | * allocated. Rather than adding NULL checks throughout this code | |
1289 | * or adding dummy dcp's to all of the callers we simply create a | |
1290 | * dummy one here and use that. This zero dcp will have the same | |
1291 | * effect as asking for inheritance of all encryption params. | |
1292 | */ | |
1293 | doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; | |
1294 | ||
1295 | return (dsl_sync_task(name, | |
1296 | dmu_objset_create_check, dmu_objset_create_sync, &doca, | |
1297 | 6, ZFS_SPACE_CHECK_NORMAL)); | |
1298 | } | |
1299 | ||
1300 | typedef struct dmu_objset_clone_arg { | |
1301 | const char *doca_clone; | |
1302 | const char *doca_origin; | |
1303 | cred_t *doca_cred; | |
1304 | } dmu_objset_clone_arg_t; | |
1305 | ||
1306 | /*ARGSUSED*/ | |
1307 | static int | |
1308 | dmu_objset_clone_check(void *arg, dmu_tx_t *tx) | |
1309 | { | |
1310 | dmu_objset_clone_arg_t *doca = arg; | |
1311 | dsl_dir_t *pdd; | |
1312 | const char *tail; | |
1313 | int error; | |
1314 | dsl_dataset_t *origin; | |
1315 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1316 | ||
1317 | if (strchr(doca->doca_clone, '@') != NULL) | |
1318 | return (SET_ERROR(EINVAL)); | |
1319 | ||
1320 | if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) | |
1321 | return (SET_ERROR(ENAMETOOLONG)); | |
1322 | ||
1323 | error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); | |
1324 | if (error != 0) | |
1325 | return (error); | |
1326 | if (tail == NULL) { | |
1327 | dsl_dir_rele(pdd, FTAG); | |
1328 | return (SET_ERROR(EEXIST)); | |
1329 | } | |
1330 | ||
1331 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, | |
1332 | doca->doca_cred); | |
1333 | if (error != 0) { | |
1334 | dsl_dir_rele(pdd, FTAG); | |
1335 | return (SET_ERROR(EDQUOT)); | |
1336 | } | |
1337 | ||
1338 | error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); | |
1339 | if (error != 0) { | |
1340 | dsl_dir_rele(pdd, FTAG); | |
1341 | return (error); | |
1342 | } | |
1343 | ||
1344 | /* You can only clone snapshots, not the head datasets. */ | |
1345 | if (!origin->ds_is_snapshot) { | |
1346 | dsl_dataset_rele(origin, FTAG); | |
1347 | dsl_dir_rele(pdd, FTAG); | |
1348 | return (SET_ERROR(EINVAL)); | |
1349 | } | |
1350 | ||
1351 | error = dmu_objset_clone_crypt_check(pdd, origin->ds_dir); | |
1352 | if (error != 0) { | |
1353 | dsl_dataset_rele(origin, FTAG); | |
1354 | dsl_dir_rele(pdd, FTAG); | |
1355 | return (error); | |
1356 | } | |
1357 | ||
1358 | dsl_dataset_rele(origin, FTAG); | |
1359 | dsl_dir_rele(pdd, FTAG); | |
1360 | ||
1361 | return (0); | |
1362 | } | |
1363 | ||
1364 | static void | |
1365 | dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) | |
1366 | { | |
1367 | dmu_objset_clone_arg_t *doca = arg; | |
1368 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1369 | dsl_dir_t *pdd; | |
1370 | const char *tail; | |
1371 | dsl_dataset_t *origin, *ds; | |
1372 | uint64_t obj; | |
1373 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; | |
1374 | ||
1375 | VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); | |
1376 | VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); | |
1377 | ||
1378 | obj = dsl_dataset_create_sync(pdd, tail, origin, 0, | |
1379 | doca->doca_cred, NULL, tx); | |
1380 | ||
1381 | VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); | |
1382 | dsl_dataset_name(origin, namebuf); | |
1383 | spa_history_log_internal_ds(ds, "clone", tx, | |
1384 | "origin=%s (%llu)", namebuf, origin->ds_object); | |
1385 | zvol_create_minors(dp->dp_spa, doca->doca_clone, B_TRUE); | |
1386 | dsl_dataset_rele(ds, FTAG); | |
1387 | dsl_dataset_rele(origin, FTAG); | |
1388 | dsl_dir_rele(pdd, FTAG); | |
1389 | } | |
1390 | ||
1391 | int | |
1392 | dmu_objset_clone(const char *clone, const char *origin) | |
1393 | { | |
1394 | dmu_objset_clone_arg_t doca; | |
1395 | ||
1396 | doca.doca_clone = clone; | |
1397 | doca.doca_origin = origin; | |
1398 | doca.doca_cred = CRED(); | |
1399 | ||
1400 | return (dsl_sync_task(clone, | |
1401 | dmu_objset_clone_check, dmu_objset_clone_sync, &doca, | |
1402 | 6, ZFS_SPACE_CHECK_NORMAL)); | |
1403 | } | |
1404 | ||
1405 | static int | |
1406 | dmu_objset_remap_indirects_impl(objset_t *os, uint64_t last_removed_txg) | |
1407 | { | |
1408 | int error = 0; | |
1409 | uint64_t object = 0; | |
1410 | while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) { | |
1411 | error = dmu_object_remap_indirects(os, object, | |
1412 | last_removed_txg); | |
1413 | /* | |
1414 | * If the ZPL removed the object before we managed to dnode_hold | |
1415 | * it, we would get an ENOENT. If the ZPL declares its intent | |
1416 | * to remove the object (dnode_free) before we manage to | |
1417 | * dnode_hold it, we would get an EEXIST. In either case, we | |
1418 | * want to continue remapping the other objects in the objset; | |
1419 | * in all other cases, we want to break early. | |
1420 | */ | |
1421 | if (error != 0 && error != ENOENT && error != EEXIST) { | |
1422 | break; | |
1423 | } | |
1424 | } | |
1425 | if (error == ESRCH) { | |
1426 | error = 0; | |
1427 | } | |
1428 | return (error); | |
1429 | } | |
1430 | ||
1431 | int | |
1432 | dmu_objset_remap_indirects(const char *fsname) | |
1433 | { | |
1434 | int error = 0; | |
1435 | objset_t *os = NULL; | |
1436 | uint64_t last_removed_txg; | |
1437 | uint64_t remap_start_txg; | |
1438 | dsl_dir_t *dd; | |
1439 | ||
1440 | error = dmu_objset_hold(fsname, FTAG, &os); | |
1441 | if (error != 0) { | |
1442 | return (error); | |
1443 | } | |
1444 | dd = dmu_objset_ds(os)->ds_dir; | |
1445 | ||
1446 | if (!spa_feature_is_enabled(dmu_objset_spa(os), | |
1447 | SPA_FEATURE_OBSOLETE_COUNTS)) { | |
1448 | dmu_objset_rele(os, FTAG); | |
1449 | return (SET_ERROR(ENOTSUP)); | |
1450 | } | |
1451 | ||
1452 | if (dsl_dataset_is_snapshot(dmu_objset_ds(os))) { | |
1453 | dmu_objset_rele(os, FTAG); | |
1454 | return (SET_ERROR(EINVAL)); | |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * If there has not been a removal, we're done. | |
1459 | */ | |
1460 | last_removed_txg = spa_get_last_removal_txg(dmu_objset_spa(os)); | |
1461 | if (last_removed_txg == -1ULL) { | |
1462 | dmu_objset_rele(os, FTAG); | |
1463 | return (0); | |
1464 | } | |
1465 | ||
1466 | /* | |
1467 | * If we have remapped since the last removal, we're done. | |
1468 | */ | |
1469 | if (dsl_dir_is_zapified(dd)) { | |
1470 | uint64_t last_remap_txg; | |
1471 | if (zap_lookup(spa_meta_objset(dmu_objset_spa(os)), | |
1472 | dd->dd_object, DD_FIELD_LAST_REMAP_TXG, | |
1473 | sizeof (last_remap_txg), 1, &last_remap_txg) == 0 && | |
1474 | last_remap_txg > last_removed_txg) { | |
1475 | dmu_objset_rele(os, FTAG); | |
1476 | return (0); | |
1477 | } | |
1478 | } | |
1479 | ||
1480 | dsl_dataset_long_hold(dmu_objset_ds(os), FTAG); | |
1481 | dsl_pool_rele(dmu_objset_pool(os), FTAG); | |
1482 | ||
1483 | remap_start_txg = spa_last_synced_txg(dmu_objset_spa(os)); | |
1484 | error = dmu_objset_remap_indirects_impl(os, last_removed_txg); | |
1485 | if (error == 0) { | |
1486 | /* | |
1487 | * We update the last_remap_txg to be the start txg so that | |
1488 | * we can guarantee that every block older than last_remap_txg | |
1489 | * that can be remapped has been remapped. | |
1490 | */ | |
1491 | error = dsl_dir_update_last_remap_txg(dd, remap_start_txg); | |
1492 | } | |
1493 | ||
1494 | dsl_dataset_long_rele(dmu_objset_ds(os), FTAG); | |
1495 | dsl_dataset_rele(dmu_objset_ds(os), FTAG); | |
1496 | ||
1497 | return (error); | |
1498 | } | |
1499 | ||
1500 | int | |
1501 | dmu_objset_snapshot_one(const char *fsname, const char *snapname) | |
1502 | { | |
1503 | int err; | |
1504 | char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); | |
1505 | nvlist_t *snaps = fnvlist_alloc(); | |
1506 | ||
1507 | fnvlist_add_boolean(snaps, longsnap); | |
1508 | strfree(longsnap); | |
1509 | err = dsl_dataset_snapshot(snaps, NULL, NULL); | |
1510 | fnvlist_free(snaps); | |
1511 | return (err); | |
1512 | } | |
1513 | ||
1514 | static void | |
1515 | dmu_objset_upgrade_task_cb(void *data) | |
1516 | { | |
1517 | objset_t *os = data; | |
1518 | ||
1519 | mutex_enter(&os->os_upgrade_lock); | |
1520 | os->os_upgrade_status = EINTR; | |
1521 | if (!os->os_upgrade_exit) { | |
1522 | mutex_exit(&os->os_upgrade_lock); | |
1523 | ||
1524 | os->os_upgrade_status = os->os_upgrade_cb(os); | |
1525 | mutex_enter(&os->os_upgrade_lock); | |
1526 | } | |
1527 | os->os_upgrade_exit = B_TRUE; | |
1528 | os->os_upgrade_id = 0; | |
1529 | mutex_exit(&os->os_upgrade_lock); | |
1530 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1531 | } | |
1532 | ||
1533 | static void | |
1534 | dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) | |
1535 | { | |
1536 | if (os->os_upgrade_id != 0) | |
1537 | return; | |
1538 | ||
1539 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); | |
1540 | dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); | |
1541 | ||
1542 | mutex_enter(&os->os_upgrade_lock); | |
1543 | if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { | |
1544 | os->os_upgrade_exit = B_FALSE; | |
1545 | os->os_upgrade_cb = cb; | |
1546 | os->os_upgrade_id = taskq_dispatch( | |
1547 | os->os_spa->spa_upgrade_taskq, | |
1548 | dmu_objset_upgrade_task_cb, os, TQ_SLEEP); | |
1549 | if (os->os_upgrade_id == TASKQID_INVALID) { | |
1550 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1551 | os->os_upgrade_status = ENOMEM; | |
1552 | } | |
1553 | } | |
1554 | mutex_exit(&os->os_upgrade_lock); | |
1555 | } | |
1556 | ||
1557 | static void | |
1558 | dmu_objset_upgrade_stop(objset_t *os) | |
1559 | { | |
1560 | mutex_enter(&os->os_upgrade_lock); | |
1561 | os->os_upgrade_exit = B_TRUE; | |
1562 | if (os->os_upgrade_id != 0) { | |
1563 | taskqid_t id = os->os_upgrade_id; | |
1564 | ||
1565 | os->os_upgrade_id = 0; | |
1566 | mutex_exit(&os->os_upgrade_lock); | |
1567 | ||
1568 | if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { | |
1569 | dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); | |
1570 | } | |
1571 | txg_wait_synced(os->os_spa->spa_dsl_pool, 0); | |
1572 | } else { | |
1573 | mutex_exit(&os->os_upgrade_lock); | |
1574 | } | |
1575 | } | |
1576 | ||
1577 | static void | |
1578 | dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) | |
1579 | { | |
1580 | dnode_t *dn; | |
1581 | ||
1582 | while ((dn = multilist_sublist_head(list)) != NULL) { | |
1583 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); | |
1584 | ASSERT(dn->dn_dbuf->db_data_pending); | |
1585 | /* | |
1586 | * Initialize dn_zio outside dnode_sync() because the | |
1587 | * meta-dnode needs to set it ouside dnode_sync(). | |
1588 | */ | |
1589 | dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; | |
1590 | ASSERT(dn->dn_zio); | |
1591 | ||
1592 | ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); | |
1593 | multilist_sublist_remove(list, dn); | |
1594 | ||
1595 | /* | |
1596 | * If we are not doing useraccounting (os_synced_dnodes == NULL) | |
1597 | * we are done with this dnode for this txg. Unset dn_dirty_txg | |
1598 | * if later txgs aren't dirtying it so that future holders do | |
1599 | * not get a stale value. Otherwise, we will do this in | |
1600 | * userquota_updates_task() when processing has completely | |
1601 | * finished for this txg. | |
1602 | */ | |
1603 | multilist_t *newlist = dn->dn_objset->os_synced_dnodes; | |
1604 | if (newlist != NULL) { | |
1605 | (void) dnode_add_ref(dn, newlist); | |
1606 | multilist_insert(newlist, dn); | |
1607 | } else { | |
1608 | mutex_enter(&dn->dn_mtx); | |
1609 | if (dn->dn_dirty_txg == tx->tx_txg) | |
1610 | dn->dn_dirty_txg = 0; | |
1611 | mutex_exit(&dn->dn_mtx); | |
1612 | } | |
1613 | ||
1614 | dnode_sync(dn, tx); | |
1615 | } | |
1616 | } | |
1617 | ||
1618 | /* ARGSUSED */ | |
1619 | static void | |
1620 | dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) | |
1621 | { | |
1622 | blkptr_t *bp = zio->io_bp; | |
1623 | objset_t *os = arg; | |
1624 | dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; | |
1625 | uint64_t fill = 0; | |
1626 | ||
1627 | ASSERT(!BP_IS_EMBEDDED(bp)); | |
1628 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); | |
1629 | ASSERT0(BP_GET_LEVEL(bp)); | |
1630 | ||
1631 | /* | |
1632 | * Update rootbp fill count: it should be the number of objects | |
1633 | * allocated in the object set (not counting the "special" | |
1634 | * objects that are stored in the objset_phys_t -- the meta | |
1635 | * dnode and user/group/project accounting objects). | |
1636 | */ | |
1637 | for (int i = 0; i < dnp->dn_nblkptr; i++) | |
1638 | fill += BP_GET_FILL(&dnp->dn_blkptr[i]); | |
1639 | ||
1640 | BP_SET_FILL(bp, fill); | |
1641 | ||
1642 | if (os->os_dsl_dataset != NULL) | |
1643 | rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); | |
1644 | *os->os_rootbp = *bp; | |
1645 | if (os->os_dsl_dataset != NULL) | |
1646 | rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); | |
1647 | } | |
1648 | ||
1649 | /* ARGSUSED */ | |
1650 | static void | |
1651 | dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) | |
1652 | { | |
1653 | blkptr_t *bp = zio->io_bp; | |
1654 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1655 | objset_t *os = arg; | |
1656 | ||
1657 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { | |
1658 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
1659 | } else { | |
1660 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
1661 | dmu_tx_t *tx = os->os_synctx; | |
1662 | ||
1663 | (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); | |
1664 | dsl_dataset_block_born(ds, bp, tx); | |
1665 | } | |
1666 | kmem_free(bp, sizeof (*bp)); | |
1667 | } | |
1668 | ||
1669 | typedef struct sync_dnodes_arg { | |
1670 | multilist_t *sda_list; | |
1671 | int sda_sublist_idx; | |
1672 | multilist_t *sda_newlist; | |
1673 | dmu_tx_t *sda_tx; | |
1674 | } sync_dnodes_arg_t; | |
1675 | ||
1676 | static void | |
1677 | sync_dnodes_task(void *arg) | |
1678 | { | |
1679 | sync_dnodes_arg_t *sda = arg; | |
1680 | ||
1681 | multilist_sublist_t *ms = | |
1682 | multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); | |
1683 | ||
1684 | dmu_objset_sync_dnodes(ms, sda->sda_tx); | |
1685 | ||
1686 | multilist_sublist_unlock(ms); | |
1687 | ||
1688 | kmem_free(sda, sizeof (*sda)); | |
1689 | } | |
1690 | ||
1691 | ||
1692 | /* called from dsl */ | |
1693 | void | |
1694 | dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) | |
1695 | { | |
1696 | int txgoff; | |
1697 | zbookmark_phys_t zb; | |
1698 | zio_prop_t zp; | |
1699 | zio_t *zio; | |
1700 | list_t *list; | |
1701 | dbuf_dirty_record_t *dr; | |
1702 | blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); | |
1703 | *blkptr_copy = *os->os_rootbp; | |
1704 | ||
1705 | dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); | |
1706 | ||
1707 | ASSERT(dmu_tx_is_syncing(tx)); | |
1708 | /* XXX the write_done callback should really give us the tx... */ | |
1709 | os->os_synctx = tx; | |
1710 | ||
1711 | if (os->os_dsl_dataset == NULL) { | |
1712 | /* | |
1713 | * This is the MOS. If we have upgraded, | |
1714 | * spa_max_replication() could change, so reset | |
1715 | * os_copies here. | |
1716 | */ | |
1717 | os->os_copies = spa_max_replication(os->os_spa); | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * Create the root block IO | |
1722 | */ | |
1723 | SET_BOOKMARK(&zb, os->os_dsl_dataset ? | |
1724 | os->os_dsl_dataset->ds_object : DMU_META_OBJSET, | |
1725 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
1726 | arc_release(os->os_phys_buf, &os->os_phys_buf); | |
1727 | ||
1728 | dmu_write_policy(os, NULL, 0, 0, &zp); | |
1729 | ||
1730 | /* | |
1731 | * If we are either claiming the ZIL or doing a raw receive, write | |
1732 | * out the os_phys_buf raw. Neither of these actions will effect the | |
1733 | * MAC at this point. | |
1734 | */ | |
1735 | if (os->os_raw_receive || | |
1736 | os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { | |
1737 | ASSERT(os->os_encrypted); | |
1738 | arc_convert_to_raw(os->os_phys_buf, | |
1739 | os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, | |
1740 | DMU_OT_OBJSET, NULL, NULL, NULL); | |
1741 | } | |
1742 | ||
1743 | zio = arc_write(pio, os->os_spa, tx->tx_txg, | |
1744 | blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), | |
1745 | &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, | |
1746 | os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); | |
1747 | ||
1748 | /* | |
1749 | * Sync special dnodes - the parent IO for the sync is the root block | |
1750 | */ | |
1751 | DMU_META_DNODE(os)->dn_zio = zio; | |
1752 | dnode_sync(DMU_META_DNODE(os), tx); | |
1753 | ||
1754 | os->os_phys->os_flags = os->os_flags; | |
1755 | ||
1756 | if (DMU_USERUSED_DNODE(os) && | |
1757 | DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1758 | DMU_USERUSED_DNODE(os)->dn_zio = zio; | |
1759 | dnode_sync(DMU_USERUSED_DNODE(os), tx); | |
1760 | DMU_GROUPUSED_DNODE(os)->dn_zio = zio; | |
1761 | dnode_sync(DMU_GROUPUSED_DNODE(os), tx); | |
1762 | } | |
1763 | ||
1764 | if (DMU_PROJECTUSED_DNODE(os) && | |
1765 | DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1766 | DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; | |
1767 | dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); | |
1768 | } | |
1769 | ||
1770 | txgoff = tx->tx_txg & TXG_MASK; | |
1771 | ||
1772 | if (dmu_objset_userused_enabled(os) && | |
1773 | (!os->os_encrypted || !dmu_objset_is_receiving(os))) { | |
1774 | /* | |
1775 | * We must create the list here because it uses the | |
1776 | * dn_dirty_link[] of this txg. But it may already | |
1777 | * exist because we call dsl_dataset_sync() twice per txg. | |
1778 | */ | |
1779 | if (os->os_synced_dnodes == NULL) { | |
1780 | os->os_synced_dnodes = | |
1781 | multilist_create(sizeof (dnode_t), | |
1782 | offsetof(dnode_t, dn_dirty_link[txgoff]), | |
1783 | dnode_multilist_index_func); | |
1784 | } else { | |
1785 | ASSERT3U(os->os_synced_dnodes->ml_offset, ==, | |
1786 | offsetof(dnode_t, dn_dirty_link[txgoff])); | |
1787 | } | |
1788 | } | |
1789 | ||
1790 | for (int i = 0; | |
1791 | i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { | |
1792 | sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); | |
1793 | sda->sda_list = os->os_dirty_dnodes[txgoff]; | |
1794 | sda->sda_sublist_idx = i; | |
1795 | sda->sda_tx = tx; | |
1796 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1797 | sync_dnodes_task, sda, 0); | |
1798 | /* callback frees sda */ | |
1799 | } | |
1800 | taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); | |
1801 | ||
1802 | list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; | |
1803 | while ((dr = list_head(list)) != NULL) { | |
1804 | ASSERT0(dr->dr_dbuf->db_level); | |
1805 | list_remove(list, dr); | |
1806 | if (dr->dr_zio) | |
1807 | zio_nowait(dr->dr_zio); | |
1808 | } | |
1809 | ||
1810 | /* Enable dnode backfill if enough objects have been freed. */ | |
1811 | if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { | |
1812 | os->os_rescan_dnodes = B_TRUE; | |
1813 | os->os_freed_dnodes = 0; | |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * Free intent log blocks up to this tx. | |
1818 | */ | |
1819 | zil_sync(os->os_zil, tx); | |
1820 | os->os_phys->os_zil_header = os->os_zil_header; | |
1821 | zio_nowait(zio); | |
1822 | } | |
1823 | ||
1824 | boolean_t | |
1825 | dmu_objset_is_dirty(objset_t *os, uint64_t txg) | |
1826 | { | |
1827 | return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); | |
1828 | } | |
1829 | ||
1830 | static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; | |
1831 | ||
1832 | void | |
1833 | dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) | |
1834 | { | |
1835 | used_cbs[ost] = cb; | |
1836 | } | |
1837 | ||
1838 | boolean_t | |
1839 | dmu_objset_userused_enabled(objset_t *os) | |
1840 | { | |
1841 | return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && | |
1842 | used_cbs[os->os_phys->os_type] != NULL && | |
1843 | DMU_USERUSED_DNODE(os) != NULL); | |
1844 | } | |
1845 | ||
1846 | boolean_t | |
1847 | dmu_objset_userobjused_enabled(objset_t *os) | |
1848 | { | |
1849 | return (dmu_objset_userused_enabled(os) && | |
1850 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); | |
1851 | } | |
1852 | ||
1853 | boolean_t | |
1854 | dmu_objset_projectquota_enabled(objset_t *os) | |
1855 | { | |
1856 | return (used_cbs[os->os_phys->os_type] != NULL && | |
1857 | DMU_PROJECTUSED_DNODE(os) != NULL && | |
1858 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); | |
1859 | } | |
1860 | ||
1861 | typedef struct userquota_node { | |
1862 | /* must be in the first filed, see userquota_update_cache() */ | |
1863 | char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1864 | int64_t uqn_delta; | |
1865 | avl_node_t uqn_node; | |
1866 | } userquota_node_t; | |
1867 | ||
1868 | typedef struct userquota_cache { | |
1869 | avl_tree_t uqc_user_deltas; | |
1870 | avl_tree_t uqc_group_deltas; | |
1871 | avl_tree_t uqc_project_deltas; | |
1872 | } userquota_cache_t; | |
1873 | ||
1874 | static int | |
1875 | userquota_compare(const void *l, const void *r) | |
1876 | { | |
1877 | const userquota_node_t *luqn = l; | |
1878 | const userquota_node_t *ruqn = r; | |
1879 | int rv; | |
1880 | ||
1881 | /* | |
1882 | * NB: can only access uqn_id because userquota_update_cache() doesn't | |
1883 | * pass in an entire userquota_node_t. | |
1884 | */ | |
1885 | rv = strcmp(luqn->uqn_id, ruqn->uqn_id); | |
1886 | ||
1887 | return (AVL_ISIGN(rv)); | |
1888 | } | |
1889 | ||
1890 | static void | |
1891 | do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) | |
1892 | { | |
1893 | void *cookie; | |
1894 | userquota_node_t *uqn; | |
1895 | ||
1896 | ASSERT(dmu_tx_is_syncing(tx)); | |
1897 | ||
1898 | cookie = NULL; | |
1899 | while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, | |
1900 | &cookie)) != NULL) { | |
1901 | /* | |
1902 | * os_userused_lock protects against concurrent calls to | |
1903 | * zap_increment_int(). It's needed because zap_increment_int() | |
1904 | * is not thread-safe (i.e. not atomic). | |
1905 | */ | |
1906 | mutex_enter(&os->os_userused_lock); | |
1907 | VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, | |
1908 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1909 | mutex_exit(&os->os_userused_lock); | |
1910 | kmem_free(uqn, sizeof (*uqn)); | |
1911 | } | |
1912 | avl_destroy(&cache->uqc_user_deltas); | |
1913 | ||
1914 | cookie = NULL; | |
1915 | while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, | |
1916 | &cookie)) != NULL) { | |
1917 | mutex_enter(&os->os_userused_lock); | |
1918 | VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, | |
1919 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1920 | mutex_exit(&os->os_userused_lock); | |
1921 | kmem_free(uqn, sizeof (*uqn)); | |
1922 | } | |
1923 | avl_destroy(&cache->uqc_group_deltas); | |
1924 | ||
1925 | if (dmu_objset_projectquota_enabled(os)) { | |
1926 | cookie = NULL; | |
1927 | while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, | |
1928 | &cookie)) != NULL) { | |
1929 | mutex_enter(&os->os_userused_lock); | |
1930 | VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, | |
1931 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1932 | mutex_exit(&os->os_userused_lock); | |
1933 | kmem_free(uqn, sizeof (*uqn)); | |
1934 | } | |
1935 | avl_destroy(&cache->uqc_project_deltas); | |
1936 | } | |
1937 | } | |
1938 | ||
1939 | static void | |
1940 | userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) | |
1941 | { | |
1942 | userquota_node_t *uqn; | |
1943 | avl_index_t idx; | |
1944 | ||
1945 | ASSERT(strlen(id) < sizeof (uqn->uqn_id)); | |
1946 | /* | |
1947 | * Use id directly for searching because uqn_id is the first field of | |
1948 | * userquota_node_t and fields after uqn_id won't be accessed in | |
1949 | * avl_find(). | |
1950 | */ | |
1951 | uqn = avl_find(avl, (const void *)id, &idx); | |
1952 | if (uqn == NULL) { | |
1953 | uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); | |
1954 | strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); | |
1955 | avl_insert(avl, uqn, idx); | |
1956 | } | |
1957 | uqn->uqn_delta += delta; | |
1958 | } | |
1959 | ||
1960 | static void | |
1961 | do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, | |
1962 | uint64_t flags, uint64_t user, uint64_t group, uint64_t project, | |
1963 | boolean_t subtract) | |
1964 | { | |
1965 | if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { | |
1966 | int64_t delta = DNODE_MIN_SIZE + used; | |
1967 | char name[20]; | |
1968 | ||
1969 | if (subtract) | |
1970 | delta = -delta; | |
1971 | ||
1972 | (void) sprintf(name, "%llx", (longlong_t)user); | |
1973 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); | |
1974 | ||
1975 | (void) sprintf(name, "%llx", (longlong_t)group); | |
1976 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); | |
1977 | ||
1978 | if (dmu_objset_projectquota_enabled(os)) { | |
1979 | (void) sprintf(name, "%llx", (longlong_t)project); | |
1980 | userquota_update_cache(&cache->uqc_project_deltas, | |
1981 | name, delta); | |
1982 | } | |
1983 | } | |
1984 | } | |
1985 | ||
1986 | static void | |
1987 | do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, | |
1988 | uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) | |
1989 | { | |
1990 | if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { | |
1991 | char name[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1992 | int delta = subtract ? -1 : 1; | |
1993 | ||
1994 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1995 | (longlong_t)user); | |
1996 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); | |
1997 | ||
1998 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1999 | (longlong_t)group); | |
2000 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); | |
2001 | ||
2002 | if (dmu_objset_projectquota_enabled(os)) { | |
2003 | (void) snprintf(name, sizeof (name), | |
2004 | DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); | |
2005 | userquota_update_cache(&cache->uqc_project_deltas, | |
2006 | name, delta); | |
2007 | } | |
2008 | } | |
2009 | } | |
2010 | ||
2011 | typedef struct userquota_updates_arg { | |
2012 | objset_t *uua_os; | |
2013 | int uua_sublist_idx; | |
2014 | dmu_tx_t *uua_tx; | |
2015 | } userquota_updates_arg_t; | |
2016 | ||
2017 | static void | |
2018 | userquota_updates_task(void *arg) | |
2019 | { | |
2020 | userquota_updates_arg_t *uua = arg; | |
2021 | objset_t *os = uua->uua_os; | |
2022 | dmu_tx_t *tx = uua->uua_tx; | |
2023 | dnode_t *dn; | |
2024 | userquota_cache_t cache = { { 0 } }; | |
2025 | ||
2026 | multilist_sublist_t *list = | |
2027 | multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); | |
2028 | ||
2029 | ASSERT(multilist_sublist_head(list) == NULL || | |
2030 | dmu_objset_userused_enabled(os)); | |
2031 | avl_create(&cache.uqc_user_deltas, userquota_compare, | |
2032 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
2033 | avl_create(&cache.uqc_group_deltas, userquota_compare, | |
2034 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
2035 | if (dmu_objset_projectquota_enabled(os)) | |
2036 | avl_create(&cache.uqc_project_deltas, userquota_compare, | |
2037 | sizeof (userquota_node_t), offsetof(userquota_node_t, | |
2038 | uqn_node)); | |
2039 | ||
2040 | while ((dn = multilist_sublist_head(list)) != NULL) { | |
2041 | int flags; | |
2042 | ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); | |
2043 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || | |
2044 | dn->dn_phys->dn_flags & | |
2045 | DNODE_FLAG_USERUSED_ACCOUNTED); | |
2046 | ||
2047 | flags = dn->dn_id_flags; | |
2048 | ASSERT(flags); | |
2049 | if (flags & DN_ID_OLD_EXIST) { | |
2050 | do_userquota_update(os, &cache, dn->dn_oldused, | |
2051 | dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, | |
2052 | dn->dn_oldprojid, B_TRUE); | |
2053 | do_userobjquota_update(os, &cache, dn->dn_oldflags, | |
2054 | dn->dn_olduid, dn->dn_oldgid, | |
2055 | dn->dn_oldprojid, B_TRUE); | |
2056 | } | |
2057 | if (flags & DN_ID_NEW_EXIST) { | |
2058 | do_userquota_update(os, &cache, | |
2059 | DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, | |
2060 | dn->dn_newuid, dn->dn_newgid, | |
2061 | dn->dn_newprojid, B_FALSE); | |
2062 | do_userobjquota_update(os, &cache, | |
2063 | dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, | |
2064 | dn->dn_newprojid, B_FALSE); | |
2065 | } | |
2066 | ||
2067 | mutex_enter(&dn->dn_mtx); | |
2068 | dn->dn_oldused = 0; | |
2069 | dn->dn_oldflags = 0; | |
2070 | if (dn->dn_id_flags & DN_ID_NEW_EXIST) { | |
2071 | dn->dn_olduid = dn->dn_newuid; | |
2072 | dn->dn_oldgid = dn->dn_newgid; | |
2073 | dn->dn_oldprojid = dn->dn_newprojid; | |
2074 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
2075 | if (dn->dn_bonuslen == 0) | |
2076 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
2077 | else | |
2078 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2079 | } | |
2080 | dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); | |
2081 | if (dn->dn_dirty_txg == spa_syncing_txg(os->os_spa)) | |
2082 | dn->dn_dirty_txg = 0; | |
2083 | mutex_exit(&dn->dn_mtx); | |
2084 | ||
2085 | multilist_sublist_remove(list, dn); | |
2086 | dnode_rele(dn, os->os_synced_dnodes); | |
2087 | } | |
2088 | do_userquota_cacheflush(os, &cache, tx); | |
2089 | multilist_sublist_unlock(list); | |
2090 | kmem_free(uua, sizeof (*uua)); | |
2091 | } | |
2092 | ||
2093 | void | |
2094 | dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) | |
2095 | { | |
2096 | if (!dmu_objset_userused_enabled(os)) | |
2097 | return; | |
2098 | ||
2099 | /* | |
2100 | * If this is a raw receive just return and handle accounting | |
2101 | * later when we have the keys loaded. We also don't do user | |
2102 | * accounting during claiming since the datasets are not owned | |
2103 | * for the duration of claiming and this txg should only be | |
2104 | * used for recovery. | |
2105 | */ | |
2106 | if (os->os_encrypted && dmu_objset_is_receiving(os)) | |
2107 | return; | |
2108 | ||
2109 | if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) | |
2110 | return; | |
2111 | ||
2112 | /* Allocate the user/group/project used objects if necessary. */ | |
2113 | if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { | |
2114 | VERIFY0(zap_create_claim(os, | |
2115 | DMU_USERUSED_OBJECT, | |
2116 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2117 | VERIFY0(zap_create_claim(os, | |
2118 | DMU_GROUPUSED_OBJECT, | |
2119 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2120 | } | |
2121 | ||
2122 | if (dmu_objset_projectquota_enabled(os) && | |
2123 | DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { | |
2124 | VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, | |
2125 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
2126 | } | |
2127 | ||
2128 | for (int i = 0; | |
2129 | i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { | |
2130 | userquota_updates_arg_t *uua = | |
2131 | kmem_alloc(sizeof (*uua), KM_SLEEP); | |
2132 | uua->uua_os = os; | |
2133 | uua->uua_sublist_idx = i; | |
2134 | uua->uua_tx = tx; | |
2135 | /* note: caller does taskq_wait() */ | |
2136 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
2137 | userquota_updates_task, uua, 0); | |
2138 | /* callback frees uua */ | |
2139 | } | |
2140 | } | |
2141 | ||
2142 | /* | |
2143 | * Returns a pointer to data to find uid/gid from | |
2144 | * | |
2145 | * If a dirty record for transaction group that is syncing can't | |
2146 | * be found then NULL is returned. In the NULL case it is assumed | |
2147 | * the uid/gid aren't changing. | |
2148 | */ | |
2149 | static void * | |
2150 | dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) | |
2151 | { | |
2152 | dbuf_dirty_record_t *dr, **drp; | |
2153 | void *data; | |
2154 | ||
2155 | if (db->db_dirtycnt == 0) | |
2156 | return (db->db.db_data); /* Nothing is changing */ | |
2157 | ||
2158 | for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) | |
2159 | if (dr->dr_txg == tx->tx_txg) | |
2160 | break; | |
2161 | ||
2162 | if (dr == NULL) { | |
2163 | data = NULL; | |
2164 | } else { | |
2165 | dnode_t *dn; | |
2166 | ||
2167 | DB_DNODE_ENTER(dr->dr_dbuf); | |
2168 | dn = DB_DNODE(dr->dr_dbuf); | |
2169 | ||
2170 | if (dn->dn_bonuslen == 0 && | |
2171 | dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) | |
2172 | data = dr->dt.dl.dr_data->b_data; | |
2173 | else | |
2174 | data = dr->dt.dl.dr_data; | |
2175 | ||
2176 | DB_DNODE_EXIT(dr->dr_dbuf); | |
2177 | } | |
2178 | ||
2179 | return (data); | |
2180 | } | |
2181 | ||
2182 | void | |
2183 | dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) | |
2184 | { | |
2185 | objset_t *os = dn->dn_objset; | |
2186 | void *data = NULL; | |
2187 | dmu_buf_impl_t *db = NULL; | |
2188 | uint64_t *user = NULL; | |
2189 | uint64_t *group = NULL; | |
2190 | uint64_t *project = NULL; | |
2191 | int flags = dn->dn_id_flags; | |
2192 | int error; | |
2193 | boolean_t have_spill = B_FALSE; | |
2194 | ||
2195 | if (!dmu_objset_userused_enabled(dn->dn_objset)) | |
2196 | return; | |
2197 | ||
2198 | /* | |
2199 | * Raw receives introduce a problem with user accounting. Raw | |
2200 | * receives cannot update the user accounting info because the | |
2201 | * user ids and the sizes are encrypted. To guarantee that we | |
2202 | * never end up with bad user accounting, we simply disable it | |
2203 | * during raw receives. We also disable this for normal receives | |
2204 | * so that an incremental raw receive may be done on top of an | |
2205 | * existing non-raw receive. | |
2206 | */ | |
2207 | if (os->os_encrypted && dmu_objset_is_receiving(os)) | |
2208 | return; | |
2209 | ||
2210 | if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| | |
2211 | DN_ID_CHKED_SPILL))) | |
2212 | return; | |
2213 | ||
2214 | if (before && dn->dn_bonuslen != 0) | |
2215 | data = DN_BONUS(dn->dn_phys); | |
2216 | else if (!before && dn->dn_bonuslen != 0) { | |
2217 | if (dn->dn_bonus) { | |
2218 | db = dn->dn_bonus; | |
2219 | mutex_enter(&db->db_mtx); | |
2220 | data = dmu_objset_userquota_find_data(db, tx); | |
2221 | } else { | |
2222 | data = DN_BONUS(dn->dn_phys); | |
2223 | } | |
2224 | } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { | |
2225 | int rf = 0; | |
2226 | ||
2227 | if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) | |
2228 | rf |= DB_RF_HAVESTRUCT; | |
2229 | error = dmu_spill_hold_by_dnode(dn, | |
2230 | rf | DB_RF_MUST_SUCCEED, | |
2231 | FTAG, (dmu_buf_t **)&db); | |
2232 | ASSERT(error == 0); | |
2233 | mutex_enter(&db->db_mtx); | |
2234 | data = (before) ? db->db.db_data : | |
2235 | dmu_objset_userquota_find_data(db, tx); | |
2236 | have_spill = B_TRUE; | |
2237 | } else { | |
2238 | mutex_enter(&dn->dn_mtx); | |
2239 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2240 | mutex_exit(&dn->dn_mtx); | |
2241 | return; | |
2242 | } | |
2243 | ||
2244 | if (before) { | |
2245 | ASSERT(data); | |
2246 | user = &dn->dn_olduid; | |
2247 | group = &dn->dn_oldgid; | |
2248 | project = &dn->dn_oldprojid; | |
2249 | } else if (data) { | |
2250 | user = &dn->dn_newuid; | |
2251 | group = &dn->dn_newgid; | |
2252 | project = &dn->dn_newprojid; | |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * Must always call the callback in case the object | |
2257 | * type has changed and that type isn't an object type to track | |
2258 | */ | |
2259 | error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, | |
2260 | user, group, project); | |
2261 | ||
2262 | /* | |
2263 | * Preserve existing uid/gid when the callback can't determine | |
2264 | * what the new uid/gid are and the callback returned EEXIST. | |
2265 | * The EEXIST error tells us to just use the existing uid/gid. | |
2266 | * If we don't know what the old values are then just assign | |
2267 | * them to 0, since that is a new file being created. | |
2268 | */ | |
2269 | if (!before && data == NULL && error == EEXIST) { | |
2270 | if (flags & DN_ID_OLD_EXIST) { | |
2271 | dn->dn_newuid = dn->dn_olduid; | |
2272 | dn->dn_newgid = dn->dn_oldgid; | |
2273 | dn->dn_newprojid = dn->dn_oldprojid; | |
2274 | } else { | |
2275 | dn->dn_newuid = 0; | |
2276 | dn->dn_newgid = 0; | |
2277 | dn->dn_newprojid = ZFS_DEFAULT_PROJID; | |
2278 | } | |
2279 | error = 0; | |
2280 | } | |
2281 | ||
2282 | if (db) | |
2283 | mutex_exit(&db->db_mtx); | |
2284 | ||
2285 | mutex_enter(&dn->dn_mtx); | |
2286 | if (error == 0 && before) | |
2287 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
2288 | if (error == 0 && !before) | |
2289 | dn->dn_id_flags |= DN_ID_NEW_EXIST; | |
2290 | ||
2291 | if (have_spill) { | |
2292 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
2293 | } else { | |
2294 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
2295 | } | |
2296 | mutex_exit(&dn->dn_mtx); | |
2297 | if (have_spill) | |
2298 | dmu_buf_rele((dmu_buf_t *)db, FTAG); | |
2299 | } | |
2300 | ||
2301 | boolean_t | |
2302 | dmu_objset_userspace_present(objset_t *os) | |
2303 | { | |
2304 | return (os->os_phys->os_flags & | |
2305 | OBJSET_FLAG_USERACCOUNTING_COMPLETE); | |
2306 | } | |
2307 | ||
2308 | boolean_t | |
2309 | dmu_objset_userobjspace_present(objset_t *os) | |
2310 | { | |
2311 | return (os->os_phys->os_flags & | |
2312 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); | |
2313 | } | |
2314 | ||
2315 | boolean_t | |
2316 | dmu_objset_projectquota_present(objset_t *os) | |
2317 | { | |
2318 | return (os->os_phys->os_flags & | |
2319 | OBJSET_FLAG_PROJECTQUOTA_COMPLETE); | |
2320 | } | |
2321 | ||
2322 | static int | |
2323 | dmu_objset_space_upgrade(objset_t *os) | |
2324 | { | |
2325 | uint64_t obj; | |
2326 | int err = 0; | |
2327 | ||
2328 | /* | |
2329 | * We simply need to mark every object dirty, so that it will be | |
2330 | * synced out and now accounted. If this is called | |
2331 | * concurrently, or if we already did some work before crashing, | |
2332 | * that's fine, since we track each object's accounted state | |
2333 | * independently. | |
2334 | */ | |
2335 | ||
2336 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { | |
2337 | dmu_tx_t *tx; | |
2338 | dmu_buf_t *db; | |
2339 | int objerr; | |
2340 | ||
2341 | mutex_enter(&os->os_upgrade_lock); | |
2342 | if (os->os_upgrade_exit) | |
2343 | err = SET_ERROR(EINTR); | |
2344 | mutex_exit(&os->os_upgrade_lock); | |
2345 | if (err != 0) | |
2346 | return (err); | |
2347 | ||
2348 | if (issig(JUSTLOOKING) && issig(FORREAL)) | |
2349 | return (SET_ERROR(EINTR)); | |
2350 | ||
2351 | objerr = dmu_bonus_hold(os, obj, FTAG, &db); | |
2352 | if (objerr != 0) | |
2353 | continue; | |
2354 | tx = dmu_tx_create(os); | |
2355 | dmu_tx_hold_bonus(tx, obj); | |
2356 | objerr = dmu_tx_assign(tx, TXG_WAIT); | |
2357 | if (objerr != 0) { | |
2358 | dmu_buf_rele(db, FTAG); | |
2359 | dmu_tx_abort(tx); | |
2360 | continue; | |
2361 | } | |
2362 | dmu_buf_will_dirty(db, tx); | |
2363 | dmu_buf_rele(db, FTAG); | |
2364 | dmu_tx_commit(tx); | |
2365 | } | |
2366 | return (0); | |
2367 | } | |
2368 | ||
2369 | int | |
2370 | dmu_objset_userspace_upgrade(objset_t *os) | |
2371 | { | |
2372 | int err = 0; | |
2373 | ||
2374 | if (dmu_objset_userspace_present(os)) | |
2375 | return (0); | |
2376 | if (dmu_objset_is_snapshot(os)) | |
2377 | return (SET_ERROR(EINVAL)); | |
2378 | if (!dmu_objset_userused_enabled(os)) | |
2379 | return (SET_ERROR(ENOTSUP)); | |
2380 | ||
2381 | err = dmu_objset_space_upgrade(os); | |
2382 | if (err) | |
2383 | return (err); | |
2384 | ||
2385 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; | |
2386 | txg_wait_synced(dmu_objset_pool(os), 0); | |
2387 | return (0); | |
2388 | } | |
2389 | ||
2390 | static int | |
2391 | dmu_objset_id_quota_upgrade_cb(objset_t *os) | |
2392 | { | |
2393 | int err = 0; | |
2394 | ||
2395 | if (dmu_objset_userobjspace_present(os) && | |
2396 | dmu_objset_projectquota_present(os)) | |
2397 | return (0); | |
2398 | if (dmu_objset_is_snapshot(os)) | |
2399 | return (SET_ERROR(EINVAL)); | |
2400 | if (!dmu_objset_userobjused_enabled(os)) | |
2401 | return (SET_ERROR(ENOTSUP)); | |
2402 | if (!dmu_objset_projectquota_enabled(os) && | |
2403 | dmu_objset_userobjspace_present(os)) | |
2404 | return (SET_ERROR(ENOTSUP)); | |
2405 | ||
2406 | dmu_objset_ds(os)->ds_feature_activation[ | |
2407 | SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; | |
2408 | if (dmu_objset_projectquota_enabled(os)) | |
2409 | dmu_objset_ds(os)->ds_feature_activation[ | |
2410 | SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; | |
2411 | ||
2412 | err = dmu_objset_space_upgrade(os); | |
2413 | if (err) | |
2414 | return (err); | |
2415 | ||
2416 | os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
2417 | if (dmu_objset_projectquota_enabled(os)) | |
2418 | os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; | |
2419 | ||
2420 | txg_wait_synced(dmu_objset_pool(os), 0); | |
2421 | return (0); | |
2422 | } | |
2423 | ||
2424 | void | |
2425 | dmu_objset_id_quota_upgrade(objset_t *os) | |
2426 | { | |
2427 | dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); | |
2428 | } | |
2429 | ||
2430 | boolean_t | |
2431 | dmu_objset_userobjspace_upgradable(objset_t *os) | |
2432 | { | |
2433 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2434 | !dmu_objset_is_snapshot(os) && | |
2435 | dmu_objset_userobjused_enabled(os) && | |
2436 | !dmu_objset_userobjspace_present(os) && | |
2437 | spa_writeable(dmu_objset_spa(os))); | |
2438 | } | |
2439 | ||
2440 | boolean_t | |
2441 | dmu_objset_projectquota_upgradable(objset_t *os) | |
2442 | { | |
2443 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
2444 | !dmu_objset_is_snapshot(os) && | |
2445 | dmu_objset_projectquota_enabled(os) && | |
2446 | !dmu_objset_projectquota_present(os) && | |
2447 | spa_writeable(dmu_objset_spa(os))); | |
2448 | } | |
2449 | ||
2450 | void | |
2451 | dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, | |
2452 | uint64_t *usedobjsp, uint64_t *availobjsp) | |
2453 | { | |
2454 | dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, | |
2455 | usedobjsp, availobjsp); | |
2456 | } | |
2457 | ||
2458 | uint64_t | |
2459 | dmu_objset_fsid_guid(objset_t *os) | |
2460 | { | |
2461 | return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); | |
2462 | } | |
2463 | ||
2464 | void | |
2465 | dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) | |
2466 | { | |
2467 | stat->dds_type = os->os_phys->os_type; | |
2468 | if (os->os_dsl_dataset) | |
2469 | dsl_dataset_fast_stat(os->os_dsl_dataset, stat); | |
2470 | } | |
2471 | ||
2472 | void | |
2473 | dmu_objset_stats(objset_t *os, nvlist_t *nv) | |
2474 | { | |
2475 | ASSERT(os->os_dsl_dataset || | |
2476 | os->os_phys->os_type == DMU_OST_META); | |
2477 | ||
2478 | if (os->os_dsl_dataset != NULL) | |
2479 | dsl_dataset_stats(os->os_dsl_dataset, nv); | |
2480 | ||
2481 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, | |
2482 | os->os_phys->os_type); | |
2483 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, | |
2484 | dmu_objset_userspace_present(os)); | |
2485 | } | |
2486 | ||
2487 | int | |
2488 | dmu_objset_is_snapshot(objset_t *os) | |
2489 | { | |
2490 | if (os->os_dsl_dataset != NULL) | |
2491 | return (os->os_dsl_dataset->ds_is_snapshot); | |
2492 | else | |
2493 | return (B_FALSE); | |
2494 | } | |
2495 | ||
2496 | int | |
2497 | dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, | |
2498 | boolean_t *conflict) | |
2499 | { | |
2500 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
2501 | uint64_t ignored; | |
2502 | ||
2503 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) | |
2504 | return (SET_ERROR(ENOENT)); | |
2505 | ||
2506 | return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, | |
2507 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, | |
2508 | MT_NORMALIZE, real, maxlen, conflict)); | |
2509 | } | |
2510 | ||
2511 | int | |
2512 | dmu_snapshot_list_next(objset_t *os, int namelen, char *name, | |
2513 | uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) | |
2514 | { | |
2515 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
2516 | zap_cursor_t cursor; | |
2517 | zap_attribute_t attr; | |
2518 | ||
2519 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); | |
2520 | ||
2521 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) | |
2522 | return (SET_ERROR(ENOENT)); | |
2523 | ||
2524 | zap_cursor_init_serialized(&cursor, | |
2525 | ds->ds_dir->dd_pool->dp_meta_objset, | |
2526 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); | |
2527 | ||
2528 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2529 | zap_cursor_fini(&cursor); | |
2530 | return (SET_ERROR(ENOENT)); | |
2531 | } | |
2532 | ||
2533 | if (strlen(attr.za_name) + 1 > namelen) { | |
2534 | zap_cursor_fini(&cursor); | |
2535 | return (SET_ERROR(ENAMETOOLONG)); | |
2536 | } | |
2537 | ||
2538 | (void) strcpy(name, attr.za_name); | |
2539 | if (idp) | |
2540 | *idp = attr.za_first_integer; | |
2541 | if (case_conflict) | |
2542 | *case_conflict = attr.za_normalization_conflict; | |
2543 | zap_cursor_advance(&cursor); | |
2544 | *offp = zap_cursor_serialize(&cursor); | |
2545 | zap_cursor_fini(&cursor); | |
2546 | ||
2547 | return (0); | |
2548 | } | |
2549 | ||
2550 | int | |
2551 | dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) | |
2552 | { | |
2553 | return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); | |
2554 | } | |
2555 | ||
2556 | int | |
2557 | dmu_dir_list_next(objset_t *os, int namelen, char *name, | |
2558 | uint64_t *idp, uint64_t *offp) | |
2559 | { | |
2560 | dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; | |
2561 | zap_cursor_t cursor; | |
2562 | zap_attribute_t attr; | |
2563 | ||
2564 | /* there is no next dir on a snapshot! */ | |
2565 | if (os->os_dsl_dataset->ds_object != | |
2566 | dsl_dir_phys(dd)->dd_head_dataset_obj) | |
2567 | return (SET_ERROR(ENOENT)); | |
2568 | ||
2569 | zap_cursor_init_serialized(&cursor, | |
2570 | dd->dd_pool->dp_meta_objset, | |
2571 | dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); | |
2572 | ||
2573 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2574 | zap_cursor_fini(&cursor); | |
2575 | return (SET_ERROR(ENOENT)); | |
2576 | } | |
2577 | ||
2578 | if (strlen(attr.za_name) + 1 > namelen) { | |
2579 | zap_cursor_fini(&cursor); | |
2580 | return (SET_ERROR(ENAMETOOLONG)); | |
2581 | } | |
2582 | ||
2583 | (void) strcpy(name, attr.za_name); | |
2584 | if (idp) | |
2585 | *idp = attr.za_first_integer; | |
2586 | zap_cursor_advance(&cursor); | |
2587 | *offp = zap_cursor_serialize(&cursor); | |
2588 | zap_cursor_fini(&cursor); | |
2589 | ||
2590 | return (0); | |
2591 | } | |
2592 | ||
2593 | typedef struct dmu_objset_find_ctx { | |
2594 | taskq_t *dc_tq; | |
2595 | dsl_pool_t *dc_dp; | |
2596 | uint64_t dc_ddobj; | |
2597 | char *dc_ddname; /* last component of ddobj's name */ | |
2598 | int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); | |
2599 | void *dc_arg; | |
2600 | int dc_flags; | |
2601 | kmutex_t *dc_error_lock; | |
2602 | int *dc_error; | |
2603 | } dmu_objset_find_ctx_t; | |
2604 | ||
2605 | static void | |
2606 | dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) | |
2607 | { | |
2608 | dsl_pool_t *dp = dcp->dc_dp; | |
2609 | dsl_dir_t *dd; | |
2610 | dsl_dataset_t *ds; | |
2611 | zap_cursor_t zc; | |
2612 | zap_attribute_t *attr; | |
2613 | uint64_t thisobj; | |
2614 | int err = 0; | |
2615 | ||
2616 | /* don't process if there already was an error */ | |
2617 | if (*dcp->dc_error != 0) | |
2618 | goto out; | |
2619 | ||
2620 | /* | |
2621 | * Note: passing the name (dc_ddname) here is optional, but it | |
2622 | * improves performance because we don't need to call | |
2623 | * zap_value_search() to determine the name. | |
2624 | */ | |
2625 | err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); | |
2626 | if (err != 0) | |
2627 | goto out; | |
2628 | ||
2629 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ | |
2630 | if (dd->dd_myname[0] == '$') { | |
2631 | dsl_dir_rele(dd, FTAG); | |
2632 | goto out; | |
2633 | } | |
2634 | ||
2635 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; | |
2636 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); | |
2637 | ||
2638 | /* | |
2639 | * Iterate over all children. | |
2640 | */ | |
2641 | if (dcp->dc_flags & DS_FIND_CHILDREN) { | |
2642 | for (zap_cursor_init(&zc, dp->dp_meta_objset, | |
2643 | dsl_dir_phys(dd)->dd_child_dir_zapobj); | |
2644 | zap_cursor_retrieve(&zc, attr) == 0; | |
2645 | (void) zap_cursor_advance(&zc)) { | |
2646 | ASSERT3U(attr->za_integer_length, ==, | |
2647 | sizeof (uint64_t)); | |
2648 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2649 | ||
2650 | dmu_objset_find_ctx_t *child_dcp = | |
2651 | kmem_alloc(sizeof (*child_dcp), KM_SLEEP); | |
2652 | *child_dcp = *dcp; | |
2653 | child_dcp->dc_ddobj = attr->za_first_integer; | |
2654 | child_dcp->dc_ddname = spa_strdup(attr->za_name); | |
2655 | if (dcp->dc_tq != NULL) | |
2656 | (void) taskq_dispatch(dcp->dc_tq, | |
2657 | dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); | |
2658 | else | |
2659 | dmu_objset_find_dp_impl(child_dcp); | |
2660 | } | |
2661 | zap_cursor_fini(&zc); | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * Iterate over all snapshots. | |
2666 | */ | |
2667 | if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { | |
2668 | dsl_dataset_t *ds; | |
2669 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2670 | ||
2671 | if (err == 0) { | |
2672 | uint64_t snapobj; | |
2673 | ||
2674 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
2675 | dsl_dataset_rele(ds, FTAG); | |
2676 | ||
2677 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2678 | zap_cursor_retrieve(&zc, attr) == 0; | |
2679 | (void) zap_cursor_advance(&zc)) { | |
2680 | ASSERT3U(attr->za_integer_length, ==, | |
2681 | sizeof (uint64_t)); | |
2682 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2683 | ||
2684 | err = dsl_dataset_hold_obj(dp, | |
2685 | attr->za_first_integer, FTAG, &ds); | |
2686 | if (err != 0) | |
2687 | break; | |
2688 | err = dcp->dc_func(dp, ds, dcp->dc_arg); | |
2689 | dsl_dataset_rele(ds, FTAG); | |
2690 | if (err != 0) | |
2691 | break; | |
2692 | } | |
2693 | zap_cursor_fini(&zc); | |
2694 | } | |
2695 | } | |
2696 | ||
2697 | kmem_free(attr, sizeof (zap_attribute_t)); | |
2698 | ||
2699 | if (err != 0) { | |
2700 | dsl_dir_rele(dd, FTAG); | |
2701 | goto out; | |
2702 | } | |
2703 | ||
2704 | /* | |
2705 | * Apply to self. | |
2706 | */ | |
2707 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2708 | ||
2709 | /* | |
2710 | * Note: we hold the dir while calling dsl_dataset_hold_obj() so | |
2711 | * that the dir will remain cached, and we won't have to re-instantiate | |
2712 | * it (which could be expensive due to finding its name via | |
2713 | * zap_value_search()). | |
2714 | */ | |
2715 | dsl_dir_rele(dd, FTAG); | |
2716 | if (err != 0) | |
2717 | goto out; | |
2718 | err = dcp->dc_func(dp, ds, dcp->dc_arg); | |
2719 | dsl_dataset_rele(ds, FTAG); | |
2720 | ||
2721 | out: | |
2722 | if (err != 0) { | |
2723 | mutex_enter(dcp->dc_error_lock); | |
2724 | /* only keep first error */ | |
2725 | if (*dcp->dc_error == 0) | |
2726 | *dcp->dc_error = err; | |
2727 | mutex_exit(dcp->dc_error_lock); | |
2728 | } | |
2729 | ||
2730 | if (dcp->dc_ddname != NULL) | |
2731 | spa_strfree(dcp->dc_ddname); | |
2732 | kmem_free(dcp, sizeof (*dcp)); | |
2733 | } | |
2734 | ||
2735 | static void | |
2736 | dmu_objset_find_dp_cb(void *arg) | |
2737 | { | |
2738 | dmu_objset_find_ctx_t *dcp = arg; | |
2739 | dsl_pool_t *dp = dcp->dc_dp; | |
2740 | ||
2741 | /* | |
2742 | * We need to get a pool_config_lock here, as there are several | |
2743 | * asssert(pool_config_held) down the stack. Getting a lock via | |
2744 | * dsl_pool_config_enter is risky, as it might be stalled by a | |
2745 | * pending writer. This would deadlock, as the write lock can | |
2746 | * only be granted when our parent thread gives up the lock. | |
2747 | * The _prio interface gives us priority over a pending writer. | |
2748 | */ | |
2749 | dsl_pool_config_enter_prio(dp, FTAG); | |
2750 | ||
2751 | dmu_objset_find_dp_impl(dcp); | |
2752 | ||
2753 | dsl_pool_config_exit(dp, FTAG); | |
2754 | } | |
2755 | ||
2756 | /* | |
2757 | * Find objsets under and including ddobj, call func(ds) on each. | |
2758 | * The order for the enumeration is completely undefined. | |
2759 | * func is called with dsl_pool_config held. | |
2760 | */ | |
2761 | int | |
2762 | dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, | |
2763 | int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) | |
2764 | { | |
2765 | int error = 0; | |
2766 | taskq_t *tq = NULL; | |
2767 | int ntasks; | |
2768 | dmu_objset_find_ctx_t *dcp; | |
2769 | kmutex_t err_lock; | |
2770 | ||
2771 | mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); | |
2772 | dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); | |
2773 | dcp->dc_tq = NULL; | |
2774 | dcp->dc_dp = dp; | |
2775 | dcp->dc_ddobj = ddobj; | |
2776 | dcp->dc_ddname = NULL; | |
2777 | dcp->dc_func = func; | |
2778 | dcp->dc_arg = arg; | |
2779 | dcp->dc_flags = flags; | |
2780 | dcp->dc_error_lock = &err_lock; | |
2781 | dcp->dc_error = &error; | |
2782 | ||
2783 | if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { | |
2784 | /* | |
2785 | * In case a write lock is held we can't make use of | |
2786 | * parallelism, as down the stack of the worker threads | |
2787 | * the lock is asserted via dsl_pool_config_held. | |
2788 | * In case of a read lock this is solved by getting a read | |
2789 | * lock in each worker thread, which isn't possible in case | |
2790 | * of a writer lock. So we fall back to the synchronous path | |
2791 | * here. | |
2792 | * In the future it might be possible to get some magic into | |
2793 | * dsl_pool_config_held in a way that it returns true for | |
2794 | * the worker threads so that a single lock held from this | |
2795 | * thread suffices. For now, stay single threaded. | |
2796 | */ | |
2797 | dmu_objset_find_dp_impl(dcp); | |
2798 | mutex_destroy(&err_lock); | |
2799 | ||
2800 | return (error); | |
2801 | } | |
2802 | ||
2803 | ntasks = dmu_find_threads; | |
2804 | if (ntasks == 0) | |
2805 | ntasks = vdev_count_leaves(dp->dp_spa) * 4; | |
2806 | tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, | |
2807 | INT_MAX, 0); | |
2808 | if (tq == NULL) { | |
2809 | kmem_free(dcp, sizeof (*dcp)); | |
2810 | mutex_destroy(&err_lock); | |
2811 | ||
2812 | return (SET_ERROR(ENOMEM)); | |
2813 | } | |
2814 | dcp->dc_tq = tq; | |
2815 | ||
2816 | /* dcp will be freed by task */ | |
2817 | (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); | |
2818 | ||
2819 | /* | |
2820 | * PORTING: this code relies on the property of taskq_wait to wait | |
2821 | * until no more tasks are queued and no more tasks are active. As | |
2822 | * we always queue new tasks from within other tasks, task_wait | |
2823 | * reliably waits for the full recursion to finish, even though we | |
2824 | * enqueue new tasks after taskq_wait has been called. | |
2825 | * On platforms other than illumos, taskq_wait may not have this | |
2826 | * property. | |
2827 | */ | |
2828 | taskq_wait(tq); | |
2829 | taskq_destroy(tq); | |
2830 | mutex_destroy(&err_lock); | |
2831 | ||
2832 | return (error); | |
2833 | } | |
2834 | ||
2835 | /* | |
2836 | * Find all objsets under name, and for each, call 'func(child_name, arg)'. | |
2837 | * The dp_config_rwlock must not be held when this is called, and it | |
2838 | * will not be held when the callback is called. | |
2839 | * Therefore this function should only be used when the pool is not changing | |
2840 | * (e.g. in syncing context), or the callback can deal with the possible races. | |
2841 | */ | |
2842 | static int | |
2843 | dmu_objset_find_impl(spa_t *spa, const char *name, | |
2844 | int func(const char *, void *), void *arg, int flags) | |
2845 | { | |
2846 | dsl_dir_t *dd; | |
2847 | dsl_pool_t *dp = spa_get_dsl(spa); | |
2848 | dsl_dataset_t *ds; | |
2849 | zap_cursor_t zc; | |
2850 | zap_attribute_t *attr; | |
2851 | char *child; | |
2852 | uint64_t thisobj; | |
2853 | int err; | |
2854 | ||
2855 | dsl_pool_config_enter(dp, FTAG); | |
2856 | ||
2857 | err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); | |
2858 | if (err != 0) { | |
2859 | dsl_pool_config_exit(dp, FTAG); | |
2860 | return (err); | |
2861 | } | |
2862 | ||
2863 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ | |
2864 | if (dd->dd_myname[0] == '$') { | |
2865 | dsl_dir_rele(dd, FTAG); | |
2866 | dsl_pool_config_exit(dp, FTAG); | |
2867 | return (0); | |
2868 | } | |
2869 | ||
2870 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; | |
2871 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); | |
2872 | ||
2873 | /* | |
2874 | * Iterate over all children. | |
2875 | */ | |
2876 | if (flags & DS_FIND_CHILDREN) { | |
2877 | for (zap_cursor_init(&zc, dp->dp_meta_objset, | |
2878 | dsl_dir_phys(dd)->dd_child_dir_zapobj); | |
2879 | zap_cursor_retrieve(&zc, attr) == 0; | |
2880 | (void) zap_cursor_advance(&zc)) { | |
2881 | ASSERT3U(attr->za_integer_length, ==, | |
2882 | sizeof (uint64_t)); | |
2883 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2884 | ||
2885 | child = kmem_asprintf("%s/%s", name, attr->za_name); | |
2886 | dsl_pool_config_exit(dp, FTAG); | |
2887 | err = dmu_objset_find_impl(spa, child, | |
2888 | func, arg, flags); | |
2889 | dsl_pool_config_enter(dp, FTAG); | |
2890 | strfree(child); | |
2891 | if (err != 0) | |
2892 | break; | |
2893 | } | |
2894 | zap_cursor_fini(&zc); | |
2895 | ||
2896 | if (err != 0) { | |
2897 | dsl_dir_rele(dd, FTAG); | |
2898 | dsl_pool_config_exit(dp, FTAG); | |
2899 | kmem_free(attr, sizeof (zap_attribute_t)); | |
2900 | return (err); | |
2901 | } | |
2902 | } | |
2903 | ||
2904 | /* | |
2905 | * Iterate over all snapshots. | |
2906 | */ | |
2907 | if (flags & DS_FIND_SNAPSHOTS) { | |
2908 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2909 | ||
2910 | if (err == 0) { | |
2911 | uint64_t snapobj; | |
2912 | ||
2913 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
2914 | dsl_dataset_rele(ds, FTAG); | |
2915 | ||
2916 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2917 | zap_cursor_retrieve(&zc, attr) == 0; | |
2918 | (void) zap_cursor_advance(&zc)) { | |
2919 | ASSERT3U(attr->za_integer_length, ==, | |
2920 | sizeof (uint64_t)); | |
2921 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2922 | ||
2923 | child = kmem_asprintf("%s@%s", | |
2924 | name, attr->za_name); | |
2925 | dsl_pool_config_exit(dp, FTAG); | |
2926 | err = func(child, arg); | |
2927 | dsl_pool_config_enter(dp, FTAG); | |
2928 | strfree(child); | |
2929 | if (err != 0) | |
2930 | break; | |
2931 | } | |
2932 | zap_cursor_fini(&zc); | |
2933 | } | |
2934 | } | |
2935 | ||
2936 | dsl_dir_rele(dd, FTAG); | |
2937 | kmem_free(attr, sizeof (zap_attribute_t)); | |
2938 | dsl_pool_config_exit(dp, FTAG); | |
2939 | ||
2940 | if (err != 0) | |
2941 | return (err); | |
2942 | ||
2943 | /* Apply to self. */ | |
2944 | return (func(name, arg)); | |
2945 | } | |
2946 | ||
2947 | /* | |
2948 | * See comment above dmu_objset_find_impl(). | |
2949 | */ | |
2950 | int | |
2951 | dmu_objset_find(char *name, int func(const char *, void *), void *arg, | |
2952 | int flags) | |
2953 | { | |
2954 | spa_t *spa; | |
2955 | int error; | |
2956 | ||
2957 | error = spa_open(name, &spa, FTAG); | |
2958 | if (error != 0) | |
2959 | return (error); | |
2960 | error = dmu_objset_find_impl(spa, name, func, arg, flags); | |
2961 | spa_close(spa, FTAG); | |
2962 | return (error); | |
2963 | } | |
2964 | ||
2965 | boolean_t | |
2966 | dmu_objset_incompatible_encryption_version(objset_t *os) | |
2967 | { | |
2968 | return (dsl_dir_incompatible_encryption_version( | |
2969 | os->os_dsl_dataset->ds_dir)); | |
2970 | } | |
2971 | ||
2972 | void | |
2973 | dmu_objset_set_user(objset_t *os, void *user_ptr) | |
2974 | { | |
2975 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); | |
2976 | os->os_user_ptr = user_ptr; | |
2977 | } | |
2978 | ||
2979 | void * | |
2980 | dmu_objset_get_user(objset_t *os) | |
2981 | { | |
2982 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); | |
2983 | return (os->os_user_ptr); | |
2984 | } | |
2985 | ||
2986 | /* | |
2987 | * Determine name of filesystem, given name of snapshot. | |
2988 | * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes | |
2989 | */ | |
2990 | int | |
2991 | dmu_fsname(const char *snapname, char *buf) | |
2992 | { | |
2993 | char *atp = strchr(snapname, '@'); | |
2994 | if (atp == NULL) | |
2995 | return (SET_ERROR(EINVAL)); | |
2996 | if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) | |
2997 | return (SET_ERROR(ENAMETOOLONG)); | |
2998 | (void) strlcpy(buf, snapname, atp - snapname + 1); | |
2999 | return (0); | |
3000 | } | |
3001 | ||
3002 | /* | |
3003 | * Call when we think we're going to write/free space in open context to track | |
3004 | * the amount of dirty data in the open txg, which is also the amount | |
3005 | * of memory that can not be evicted until this txg syncs. | |
3006 | */ | |
3007 | void | |
3008 | dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) | |
3009 | { | |
3010 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
3011 | int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); | |
3012 | ||
3013 | if (ds != NULL) { | |
3014 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
3015 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
3016 | } | |
3017 | } | |
3018 | ||
3019 | #if defined(_KERNEL) | |
3020 | EXPORT_SYMBOL(dmu_objset_zil); | |
3021 | EXPORT_SYMBOL(dmu_objset_pool); | |
3022 | EXPORT_SYMBOL(dmu_objset_ds); | |
3023 | EXPORT_SYMBOL(dmu_objset_type); | |
3024 | EXPORT_SYMBOL(dmu_objset_name); | |
3025 | EXPORT_SYMBOL(dmu_objset_hold); | |
3026 | EXPORT_SYMBOL(dmu_objset_hold_flags); | |
3027 | EXPORT_SYMBOL(dmu_objset_own); | |
3028 | EXPORT_SYMBOL(dmu_objset_rele); | |
3029 | EXPORT_SYMBOL(dmu_objset_rele_flags); | |
3030 | EXPORT_SYMBOL(dmu_objset_disown); | |
3031 | EXPORT_SYMBOL(dmu_objset_from_ds); | |
3032 | EXPORT_SYMBOL(dmu_objset_create); | |
3033 | EXPORT_SYMBOL(dmu_objset_clone); | |
3034 | EXPORT_SYMBOL(dmu_objset_stats); | |
3035 | EXPORT_SYMBOL(dmu_objset_fast_stat); | |
3036 | EXPORT_SYMBOL(dmu_objset_spa); | |
3037 | EXPORT_SYMBOL(dmu_objset_space); | |
3038 | EXPORT_SYMBOL(dmu_objset_fsid_guid); | |
3039 | EXPORT_SYMBOL(dmu_objset_find); | |
3040 | EXPORT_SYMBOL(dmu_objset_byteswap); | |
3041 | EXPORT_SYMBOL(dmu_objset_evict_dbufs); | |
3042 | EXPORT_SYMBOL(dmu_objset_snap_cmtime); | |
3043 | EXPORT_SYMBOL(dmu_objset_dnodesize); | |
3044 | ||
3045 | EXPORT_SYMBOL(dmu_objset_sync); | |
3046 | EXPORT_SYMBOL(dmu_objset_is_dirty); | |
3047 | EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); | |
3048 | EXPORT_SYMBOL(dmu_objset_create_impl); | |
3049 | EXPORT_SYMBOL(dmu_objset_open_impl); | |
3050 | EXPORT_SYMBOL(dmu_objset_evict); | |
3051 | EXPORT_SYMBOL(dmu_objset_register_type); | |
3052 | EXPORT_SYMBOL(dmu_objset_do_userquota_updates); | |
3053 | EXPORT_SYMBOL(dmu_objset_userquota_get_ids); | |
3054 | EXPORT_SYMBOL(dmu_objset_userused_enabled); | |
3055 | EXPORT_SYMBOL(dmu_objset_userspace_upgrade); | |
3056 | EXPORT_SYMBOL(dmu_objset_userspace_present); | |
3057 | EXPORT_SYMBOL(dmu_objset_userobjused_enabled); | |
3058 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); | |
3059 | EXPORT_SYMBOL(dmu_objset_userobjspace_present); | |
3060 | EXPORT_SYMBOL(dmu_objset_projectquota_enabled); | |
3061 | EXPORT_SYMBOL(dmu_objset_projectquota_present); | |
3062 | EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); | |
3063 | EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); | |
3064 | #endif |