]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
cae5b340 | 21 | |
34dc7c2f | 22 | /* |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
cae5b340 | 24 | * Copyright (c) 2012, 2017 by Delphix. All rights reserved. |
c06d4368 | 25 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
e10b0808 AX |
26 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
27 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. | |
e10b0808 | 28 | * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. |
4e820b5a | 29 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
cae5b340 | 30 | * Copyright 2017 Nexenta Systems, Inc. |
34dc7c2f BB |
31 | */ |
32 | ||
428870ff BB |
33 | /* Portions Copyright 2010 Robert Milkowski */ |
34 | ||
cae5b340 | 35 | #include <sys/zfeature.h> |
34dc7c2f BB |
36 | #include <sys/cred.h> |
37 | #include <sys/zfs_context.h> | |
38 | #include <sys/dmu_objset.h> | |
39 | #include <sys/dsl_dir.h> | |
40 | #include <sys/dsl_dataset.h> | |
41 | #include <sys/dsl_prop.h> | |
42 | #include <sys/dsl_pool.h> | |
43 | #include <sys/dsl_synctask.h> | |
44 | #include <sys/dsl_deleg.h> | |
45 | #include <sys/dnode.h> | |
46 | #include <sys/dbuf.h> | |
47 | #include <sys/zvol.h> | |
48 | #include <sys/dmu_tx.h> | |
34dc7c2f BB |
49 | #include <sys/zap.h> |
50 | #include <sys/zil.h> | |
51 | #include <sys/dmu_impl.h> | |
52 | #include <sys/zfs_ioctl.h> | |
428870ff | 53 | #include <sys/sa.h> |
572e2857 | 54 | #include <sys/zfs_onexit.h> |
a08ee875 | 55 | #include <sys/dsl_destroy.h> |
e10b0808 | 56 | #include <sys/vdev.h> |
cae5b340 AX |
57 | #include <sys/policy.h> |
58 | #include <sys/spa_impl.h> | |
572e2857 BB |
59 | |
60 | /* | |
61 | * Needed to close a window in dnode_move() that allows the objset to be freed | |
62 | * before it can be safely accessed. | |
63 | */ | |
64 | krwlock_t os_lock; | |
65 | ||
e10b0808 | 66 | /* |
cae5b340 | 67 | * Tunable to overwrite the maximum number of threads for the parallelization |
e10b0808 AX |
68 | * of dmu_objset_find_dp, needed to speed up the import of pools with many |
69 | * datasets. | |
70 | * Default is 4 times the number of leaf vdevs. | |
71 | */ | |
72 | int dmu_find_threads = 0; | |
73 | ||
cae5b340 AX |
74 | /* |
75 | * Backfill lower metadnode objects after this many have been freed. | |
76 | * Backfilling negatively impacts object creation rates, so only do it | |
77 | * if there are enough holes to fill. | |
78 | */ | |
79 | int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; | |
80 | ||
e10b0808 AX |
81 | static void dmu_objset_find_dp_cb(void *arg); |
82 | ||
cae5b340 AX |
83 | static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); |
84 | static void dmu_objset_upgrade_stop(objset_t *os); | |
85 | ||
572e2857 BB |
86 | void |
87 | dmu_objset_init(void) | |
88 | { | |
89 | rw_init(&os_lock, NULL, RW_DEFAULT, NULL); | |
90 | } | |
91 | ||
92 | void | |
93 | dmu_objset_fini(void) | |
94 | { | |
95 | rw_destroy(&os_lock); | |
96 | } | |
34dc7c2f BB |
97 | |
98 | spa_t * | |
99 | dmu_objset_spa(objset_t *os) | |
100 | { | |
428870ff | 101 | return (os->os_spa); |
34dc7c2f BB |
102 | } |
103 | ||
104 | zilog_t * | |
105 | dmu_objset_zil(objset_t *os) | |
106 | { | |
428870ff | 107 | return (os->os_zil); |
34dc7c2f BB |
108 | } |
109 | ||
110 | dsl_pool_t * | |
111 | dmu_objset_pool(objset_t *os) | |
112 | { | |
113 | dsl_dataset_t *ds; | |
114 | ||
428870ff | 115 | if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) |
34dc7c2f BB |
116 | return (ds->ds_dir->dd_pool); |
117 | else | |
428870ff | 118 | return (spa_get_dsl(os->os_spa)); |
34dc7c2f BB |
119 | } |
120 | ||
121 | dsl_dataset_t * | |
122 | dmu_objset_ds(objset_t *os) | |
123 | { | |
428870ff | 124 | return (os->os_dsl_dataset); |
34dc7c2f BB |
125 | } |
126 | ||
127 | dmu_objset_type_t | |
128 | dmu_objset_type(objset_t *os) | |
129 | { | |
428870ff | 130 | return (os->os_phys->os_type); |
34dc7c2f BB |
131 | } |
132 | ||
133 | void | |
134 | dmu_objset_name(objset_t *os, char *buf) | |
135 | { | |
428870ff | 136 | dsl_dataset_name(os->os_dsl_dataset, buf); |
34dc7c2f BB |
137 | } |
138 | ||
139 | uint64_t | |
140 | dmu_objset_id(objset_t *os) | |
141 | { | |
428870ff | 142 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
143 | |
144 | return (ds ? ds->ds_object : 0); | |
145 | } | |
146 | ||
cae5b340 AX |
147 | uint64_t |
148 | dmu_objset_dnodesize(objset_t *os) | |
149 | { | |
150 | return (os->os_dnodesize); | |
151 | } | |
152 | ||
ea04106b | 153 | zfs_sync_type_t |
428870ff BB |
154 | dmu_objset_syncprop(objset_t *os) |
155 | { | |
156 | return (os->os_sync); | |
157 | } | |
158 | ||
ea04106b | 159 | zfs_logbias_op_t |
428870ff BB |
160 | dmu_objset_logbias(objset_t *os) |
161 | { | |
162 | return (os->os_logbias); | |
163 | } | |
164 | ||
34dc7c2f BB |
165 | static void |
166 | checksum_changed_cb(void *arg, uint64_t newval) | |
167 | { | |
428870ff | 168 | objset_t *os = arg; |
34dc7c2f BB |
169 | |
170 | /* | |
171 | * Inheritance should have been done by now. | |
172 | */ | |
173 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
174 | ||
428870ff | 175 | os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); |
34dc7c2f BB |
176 | } |
177 | ||
178 | static void | |
179 | compression_changed_cb(void *arg, uint64_t newval) | |
180 | { | |
428870ff | 181 | objset_t *os = arg; |
34dc7c2f BB |
182 | |
183 | /* | |
184 | * Inheritance and range checking should have been done by now. | |
185 | */ | |
186 | ASSERT(newval != ZIO_COMPRESS_INHERIT); | |
187 | ||
e10b0808 AX |
188 | os->os_compress = zio_compress_select(os->os_spa, newval, |
189 | ZIO_COMPRESS_ON); | |
34dc7c2f BB |
190 | } |
191 | ||
192 | static void | |
193 | copies_changed_cb(void *arg, uint64_t newval) | |
194 | { | |
428870ff | 195 | objset_t *os = arg; |
34dc7c2f BB |
196 | |
197 | /* | |
198 | * Inheritance and range checking should have been done by now. | |
199 | */ | |
200 | ASSERT(newval > 0); | |
428870ff | 201 | ASSERT(newval <= spa_max_replication(os->os_spa)); |
34dc7c2f | 202 | |
428870ff BB |
203 | os->os_copies = newval; |
204 | } | |
205 | ||
206 | static void | |
207 | dedup_changed_cb(void *arg, uint64_t newval) | |
208 | { | |
209 | objset_t *os = arg; | |
210 | spa_t *spa = os->os_spa; | |
211 | enum zio_checksum checksum; | |
212 | ||
213 | /* | |
214 | * Inheritance should have been done by now. | |
215 | */ | |
216 | ASSERT(newval != ZIO_CHECKSUM_INHERIT); | |
217 | ||
218 | checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); | |
219 | ||
220 | os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; | |
221 | os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); | |
34dc7c2f BB |
222 | } |
223 | ||
b128c09f BB |
224 | static void |
225 | primary_cache_changed_cb(void *arg, uint64_t newval) | |
226 | { | |
428870ff | 227 | objset_t *os = arg; |
b128c09f BB |
228 | |
229 | /* | |
230 | * Inheritance and range checking should have been done by now. | |
231 | */ | |
232 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
233 | newval == ZFS_CACHE_METADATA); | |
234 | ||
428870ff | 235 | os->os_primary_cache = newval; |
b128c09f BB |
236 | } |
237 | ||
238 | static void | |
239 | secondary_cache_changed_cb(void *arg, uint64_t newval) | |
240 | { | |
428870ff | 241 | objset_t *os = arg; |
b128c09f BB |
242 | |
243 | /* | |
244 | * Inheritance and range checking should have been done by now. | |
245 | */ | |
246 | ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || | |
247 | newval == ZFS_CACHE_METADATA); | |
248 | ||
428870ff BB |
249 | os->os_secondary_cache = newval; |
250 | } | |
251 | ||
252 | static void | |
253 | sync_changed_cb(void *arg, uint64_t newval) | |
254 | { | |
255 | objset_t *os = arg; | |
256 | ||
257 | /* | |
258 | * Inheritance and range checking should have been done by now. | |
259 | */ | |
260 | ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || | |
261 | newval == ZFS_SYNC_DISABLED); | |
262 | ||
263 | os->os_sync = newval; | |
264 | if (os->os_zil) | |
265 | zil_set_sync(os->os_zil, newval); | |
266 | } | |
267 | ||
ea04106b AX |
268 | static void |
269 | redundant_metadata_changed_cb(void *arg, uint64_t newval) | |
270 | { | |
271 | objset_t *os = arg; | |
272 | ||
273 | /* | |
274 | * Inheritance and range checking should have been done by now. | |
275 | */ | |
276 | ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || | |
277 | newval == ZFS_REDUNDANT_METADATA_MOST); | |
278 | ||
279 | os->os_redundant_metadata = newval; | |
280 | } | |
281 | ||
cae5b340 AX |
282 | static void |
283 | dnodesize_changed_cb(void *arg, uint64_t newval) | |
284 | { | |
285 | objset_t *os = arg; | |
286 | ||
287 | switch (newval) { | |
288 | case ZFS_DNSIZE_LEGACY: | |
289 | os->os_dnodesize = DNODE_MIN_SIZE; | |
290 | break; | |
291 | case ZFS_DNSIZE_AUTO: | |
292 | /* | |
293 | * Choose a dnode size that will work well for most | |
294 | * workloads if the user specified "auto". Future code | |
295 | * improvements could dynamically select a dnode size | |
296 | * based on observed workload patterns. | |
297 | */ | |
298 | os->os_dnodesize = DNODE_MIN_SIZE * 2; | |
299 | break; | |
300 | case ZFS_DNSIZE_1K: | |
301 | case ZFS_DNSIZE_2K: | |
302 | case ZFS_DNSIZE_4K: | |
303 | case ZFS_DNSIZE_8K: | |
304 | case ZFS_DNSIZE_16K: | |
305 | os->os_dnodesize = newval; | |
306 | break; | |
307 | } | |
308 | } | |
309 | ||
428870ff BB |
310 | static void |
311 | logbias_changed_cb(void *arg, uint64_t newval) | |
312 | { | |
313 | objset_t *os = arg; | |
314 | ||
315 | ASSERT(newval == ZFS_LOGBIAS_LATENCY || | |
316 | newval == ZFS_LOGBIAS_THROUGHPUT); | |
317 | os->os_logbias = newval; | |
318 | if (os->os_zil) | |
319 | zil_set_logbias(os->os_zil, newval); | |
b128c09f BB |
320 | } |
321 | ||
e10b0808 AX |
322 | static void |
323 | recordsize_changed_cb(void *arg, uint64_t newval) | |
324 | { | |
325 | objset_t *os = arg; | |
326 | ||
327 | os->os_recordsize = newval; | |
328 | } | |
329 | ||
34dc7c2f BB |
330 | void |
331 | dmu_objset_byteswap(void *buf, size_t size) | |
332 | { | |
333 | objset_phys_t *osp = buf; | |
334 | ||
9babb374 | 335 | ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); |
34dc7c2f BB |
336 | dnode_byteswap(&osp->os_meta_dnode); |
337 | byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); | |
338 | osp->os_type = BSWAP_64(osp->os_type); | |
9babb374 BB |
339 | osp->os_flags = BSWAP_64(osp->os_flags); |
340 | if (size == sizeof (objset_phys_t)) { | |
341 | dnode_byteswap(&osp->os_userused_dnode); | |
342 | dnode_byteswap(&osp->os_groupused_dnode); | |
343 | } | |
34dc7c2f BB |
344 | } |
345 | ||
cae5b340 AX |
346 | /* |
347 | * The hash is a CRC-based hash of the objset_t pointer and the object number. | |
348 | */ | |
349 | static uint64_t | |
350 | dnode_hash(const objset_t *os, uint64_t obj) | |
351 | { | |
352 | uintptr_t osv = (uintptr_t)os; | |
353 | uint64_t crc = -1ULL; | |
354 | ||
355 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
356 | /* | |
357 | * The low 6 bits of the pointer don't have much entropy, because | |
358 | * the objset_t is larger than 2^6 bytes long. | |
359 | */ | |
360 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; | |
361 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; | |
362 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; | |
363 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; | |
364 | ||
365 | crc ^= (osv>>14) ^ (obj>>24); | |
366 | ||
367 | return (crc); | |
368 | } | |
369 | ||
370 | unsigned int | |
371 | dnode_multilist_index_func(multilist_t *ml, void *obj) | |
372 | { | |
373 | dnode_t *dn = obj; | |
374 | return (dnode_hash(dn->dn_objset, dn->dn_object) % | |
375 | multilist_get_num_sublists(ml)); | |
376 | } | |
377 | ||
34dc7c2f BB |
378 | int |
379 | dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, | |
428870ff | 380 | objset_t **osp) |
34dc7c2f | 381 | { |
428870ff | 382 | objset_t *os; |
b128c09f | 383 | int i, err; |
34dc7c2f BB |
384 | |
385 | ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); | |
386 | ||
ea04106b | 387 | os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); |
428870ff BB |
388 | os->os_dsl_dataset = ds; |
389 | os->os_spa = spa; | |
390 | os->os_rootbp = bp; | |
391 | if (!BP_IS_HOLE(os->os_rootbp)) { | |
e10b0808 | 392 | arc_flags_t aflags = ARC_FLAG_WAIT; |
ea04106b | 393 | zbookmark_phys_t zb; |
428870ff BB |
394 | SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, |
395 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
396 | ||
397 | if (DMU_OS_IS_L2CACHEABLE(os)) | |
e10b0808 | 398 | aflags |= ARC_FLAG_L2CACHE; |
34dc7c2f | 399 | |
428870ff | 400 | dprintf_bp(os->os_rootbp, "reading %s", ""); |
c06d4368 | 401 | err = arc_read(NULL, spa, os->os_rootbp, |
428870ff | 402 | arc_getbuf_func, &os->os_phys_buf, |
34dc7c2f | 403 | ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); |
a08ee875 | 404 | if (err != 0) { |
428870ff | 405 | kmem_free(os, sizeof (objset_t)); |
b128c09f BB |
406 | /* convert checksum errors into IO errors */ |
407 | if (err == ECKSUM) | |
a08ee875 | 408 | err = SET_ERROR(EIO); |
34dc7c2f BB |
409 | return (err); |
410 | } | |
9babb374 BB |
411 | |
412 | /* Increase the blocksize if we are permitted. */ | |
413 | if (spa_version(spa) >= SPA_VERSION_USERSPACE && | |
428870ff | 414 | arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { |
cae5b340 AX |
415 | arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, |
416 | ARC_BUFC_METADATA, sizeof (objset_phys_t)); | |
9babb374 | 417 | bzero(buf->b_data, sizeof (objset_phys_t)); |
428870ff BB |
418 | bcopy(os->os_phys_buf->b_data, buf->b_data, |
419 | arc_buf_size(os->os_phys_buf)); | |
cae5b340 | 420 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 421 | os->os_phys_buf = buf; |
9babb374 BB |
422 | } |
423 | ||
428870ff BB |
424 | os->os_phys = os->os_phys_buf->b_data; |
425 | os->os_flags = os->os_phys->os_flags; | |
34dc7c2f | 426 | } else { |
9babb374 BB |
427 | int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? |
428 | sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; | |
cae5b340 AX |
429 | os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, |
430 | ARC_BUFC_METADATA, size); | |
428870ff BB |
431 | os->os_phys = os->os_phys_buf->b_data; |
432 | bzero(os->os_phys, size); | |
34dc7c2f BB |
433 | } |
434 | ||
435 | /* | |
436 | * Note: the changed_cb will be called once before the register | |
437 | * func returns, thus changing the checksum/compression from the | |
b128c09f BB |
438 | * default (fletcher2/off). Snapshots don't need to know about |
439 | * checksum/compression/copies. | |
34dc7c2f | 440 | */ |
ea04106b | 441 | if (ds != NULL) { |
cae5b340 AX |
442 | boolean_t needlock = B_FALSE; |
443 | ||
444 | /* | |
445 | * Note: it's valid to open the objset if the dataset is | |
446 | * long-held, in which case the pool_config lock will not | |
447 | * be held. | |
448 | */ | |
449 | if (!dsl_pool_config_held(dmu_objset_pool(os))) { | |
450 | needlock = B_TRUE; | |
451 | dsl_pool_config_enter(dmu_objset_pool(os), FTAG); | |
452 | } | |
a08ee875 LG |
453 | err = dsl_prop_register(ds, |
454 | zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), | |
428870ff | 455 | primary_cache_changed_cb, os); |
a08ee875 LG |
456 | if (err == 0) { |
457 | err = dsl_prop_register(ds, | |
458 | zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), | |
428870ff | 459 | secondary_cache_changed_cb, os); |
a08ee875 | 460 | } |
e10b0808 | 461 | if (!ds->ds_is_snapshot) { |
a08ee875 LG |
462 | if (err == 0) { |
463 | err = dsl_prop_register(ds, | |
464 | zfs_prop_to_name(ZFS_PROP_CHECKSUM), | |
428870ff | 465 | checksum_changed_cb, os); |
a08ee875 LG |
466 | } |
467 | if (err == 0) { | |
468 | err = dsl_prop_register(ds, | |
469 | zfs_prop_to_name(ZFS_PROP_COMPRESSION), | |
428870ff | 470 | compression_changed_cb, os); |
a08ee875 LG |
471 | } |
472 | if (err == 0) { | |
473 | err = dsl_prop_register(ds, | |
474 | zfs_prop_to_name(ZFS_PROP_COPIES), | |
428870ff | 475 | copies_changed_cb, os); |
a08ee875 LG |
476 | } |
477 | if (err == 0) { | |
478 | err = dsl_prop_register(ds, | |
479 | zfs_prop_to_name(ZFS_PROP_DEDUP), | |
428870ff | 480 | dedup_changed_cb, os); |
a08ee875 LG |
481 | } |
482 | if (err == 0) { | |
483 | err = dsl_prop_register(ds, | |
484 | zfs_prop_to_name(ZFS_PROP_LOGBIAS), | |
428870ff | 485 | logbias_changed_cb, os); |
a08ee875 LG |
486 | } |
487 | if (err == 0) { | |
488 | err = dsl_prop_register(ds, | |
489 | zfs_prop_to_name(ZFS_PROP_SYNC), | |
428870ff | 490 | sync_changed_cb, os); |
a08ee875 | 491 | } |
ea04106b AX |
492 | if (err == 0) { |
493 | err = dsl_prop_register(ds, | |
494 | zfs_prop_to_name( | |
495 | ZFS_PROP_REDUNDANT_METADATA), | |
496 | redundant_metadata_changed_cb, os); | |
497 | } | |
e10b0808 AX |
498 | if (err == 0) { |
499 | err = dsl_prop_register(ds, | |
500 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
501 | recordsize_changed_cb, os); | |
502 | } | |
cae5b340 AX |
503 | if (err == 0) { |
504 | err = dsl_prop_register(ds, | |
505 | zfs_prop_to_name(ZFS_PROP_DNODESIZE), | |
506 | dnodesize_changed_cb, os); | |
507 | } | |
b128c09f | 508 | } |
cae5b340 AX |
509 | if (needlock) |
510 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); | |
a08ee875 | 511 | if (err != 0) { |
cae5b340 | 512 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
428870ff | 513 | kmem_free(os, sizeof (objset_t)); |
34dc7c2f BB |
514 | return (err); |
515 | } | |
ea04106b | 516 | } else { |
34dc7c2f | 517 | /* It's the meta-objset. */ |
428870ff | 518 | os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; |
e10b0808 | 519 | os->os_compress = ZIO_COMPRESS_ON; |
428870ff BB |
520 | os->os_copies = spa_max_replication(spa); |
521 | os->os_dedup_checksum = ZIO_CHECKSUM_OFF; | |
ea04106b AX |
522 | os->os_dedup_verify = B_FALSE; |
523 | os->os_logbias = ZFS_LOGBIAS_LATENCY; | |
524 | os->os_sync = ZFS_SYNC_STANDARD; | |
428870ff BB |
525 | os->os_primary_cache = ZFS_CACHE_ALL; |
526 | os->os_secondary_cache = ZFS_CACHE_ALL; | |
cae5b340 | 527 | os->os_dnodesize = DNODE_MIN_SIZE; |
34dc7c2f BB |
528 | } |
529 | ||
e10b0808 | 530 | if (ds == NULL || !ds->ds_is_snapshot) |
572e2857 | 531 | os->os_zil_header = os->os_phys->os_zil_header; |
428870ff | 532 | os->os_zil = zil_alloc(os, &os->os_zil_header); |
34dc7c2f BB |
533 | |
534 | for (i = 0; i < TXG_SIZE; i++) { | |
cae5b340 AX |
535 | os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), |
536 | offsetof(dnode_t, dn_dirty_link[i]), | |
537 | dnode_multilist_index_func); | |
34dc7c2f | 538 | } |
428870ff | 539 | list_create(&os->os_dnodes, sizeof (dnode_t), |
34dc7c2f | 540 | offsetof(dnode_t, dn_link)); |
428870ff | 541 | list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), |
34dc7c2f BB |
542 | offsetof(dmu_buf_impl_t, db_link)); |
543 | ||
e10b0808 AX |
544 | list_link_init(&os->os_evicting_node); |
545 | ||
428870ff | 546 | mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); |
cae5b340 | 547 | mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); |
428870ff BB |
548 | mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); |
549 | mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); | |
cae5b340 AX |
550 | os->os_obj_next_percpu_len = boot_ncpus; |
551 | os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * | |
552 | sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); | |
428870ff | 553 | |
e10b0808 AX |
554 | dnode_special_open(os, &os->os_phys->os_meta_dnode, |
555 | DMU_META_DNODE_OBJECT, &os->os_meta_dnode); | |
428870ff | 556 | if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { |
e10b0808 AX |
557 | dnode_special_open(os, &os->os_phys->os_userused_dnode, |
558 | DMU_USERUSED_OBJECT, &os->os_userused_dnode); | |
559 | dnode_special_open(os, &os->os_phys->os_groupused_dnode, | |
560 | DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); | |
9babb374 | 561 | } |
34dc7c2f | 562 | |
cae5b340 AX |
563 | mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); |
564 | ||
428870ff | 565 | *osp = os; |
34dc7c2f BB |
566 | return (0); |
567 | } | |
568 | ||
428870ff BB |
569 | int |
570 | dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) | |
34dc7c2f | 571 | { |
428870ff | 572 | int err = 0; |
34dc7c2f | 573 | |
cae5b340 AX |
574 | /* |
575 | * We shouldn't be doing anything with dsl_dataset_t's unless the | |
576 | * pool_config lock is held, or the dataset is long-held. | |
577 | */ | |
578 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || | |
579 | dsl_dataset_long_held(ds)); | |
580 | ||
34dc7c2f | 581 | mutex_enter(&ds->ds_opening_lock); |
ea04106b AX |
582 | if (ds->ds_objset == NULL) { |
583 | objset_t *os; | |
cae5b340 | 584 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
34dc7c2f | 585 | err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), |
ea04106b | 586 | ds, dsl_dataset_get_blkptr(ds), &os); |
cae5b340 | 587 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
ea04106b AX |
588 | |
589 | if (err == 0) { | |
590 | mutex_enter(&ds->ds_lock); | |
591 | ASSERT(ds->ds_objset == NULL); | |
592 | ds->ds_objset = os; | |
593 | mutex_exit(&ds->ds_lock); | |
594 | } | |
34dc7c2f | 595 | } |
ea04106b | 596 | *osp = ds->ds_objset; |
34dc7c2f | 597 | mutex_exit(&ds->ds_opening_lock); |
428870ff | 598 | return (err); |
34dc7c2f BB |
599 | } |
600 | ||
a08ee875 LG |
601 | /* |
602 | * Holds the pool while the objset is held. Therefore only one objset | |
603 | * can be held at a time. | |
604 | */ | |
34dc7c2f | 605 | int |
428870ff | 606 | dmu_objset_hold(const char *name, void *tag, objset_t **osp) |
34dc7c2f | 607 | { |
a08ee875 | 608 | dsl_pool_t *dp; |
428870ff | 609 | dsl_dataset_t *ds; |
34dc7c2f BB |
610 | int err; |
611 | ||
a08ee875 LG |
612 | err = dsl_pool_hold(name, tag, &dp); |
613 | if (err != 0) | |
614 | return (err); | |
615 | err = dsl_dataset_hold(dp, name, tag, &ds); | |
616 | if (err != 0) { | |
617 | dsl_pool_rele(dp, tag); | |
428870ff | 618 | return (err); |
a08ee875 | 619 | } |
428870ff BB |
620 | |
621 | err = dmu_objset_from_ds(ds, osp); | |
a08ee875 | 622 | if (err != 0) { |
428870ff | 623 | dsl_dataset_rele(ds, tag); |
a08ee875 LG |
624 | dsl_pool_rele(dp, tag); |
625 | } | |
428870ff | 626 | |
34dc7c2f BB |
627 | return (err); |
628 | } | |
629 | ||
e10b0808 AX |
630 | static int |
631 | dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, | |
632 | boolean_t readonly, void *tag, objset_t **osp) | |
633 | { | |
634 | int err; | |
635 | ||
636 | err = dmu_objset_from_ds(ds, osp); | |
637 | if (err != 0) { | |
638 | dsl_dataset_disown(ds, tag); | |
639 | } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { | |
640 | dsl_dataset_disown(ds, tag); | |
641 | return (SET_ERROR(EINVAL)); | |
642 | } else if (!readonly && dsl_dataset_is_snapshot(ds)) { | |
643 | dsl_dataset_disown(ds, tag); | |
644 | return (SET_ERROR(EROFS)); | |
645 | } | |
646 | return (err); | |
647 | } | |
648 | ||
a08ee875 LG |
649 | /* |
650 | * dsl_pool must not be held when this is called. | |
651 | * Upon successful return, there will be a longhold on the dataset, | |
652 | * and the dsl_pool will not be held. | |
653 | */ | |
34dc7c2f | 654 | int |
428870ff BB |
655 | dmu_objset_own(const char *name, dmu_objset_type_t type, |
656 | boolean_t readonly, void *tag, objset_t **osp) | |
34dc7c2f | 657 | { |
a08ee875 | 658 | dsl_pool_t *dp; |
34dc7c2f BB |
659 | dsl_dataset_t *ds; |
660 | int err; | |
661 | ||
a08ee875 LG |
662 | err = dsl_pool_hold(name, FTAG, &dp); |
663 | if (err != 0) | |
664 | return (err); | |
665 | err = dsl_dataset_own(dp, name, tag, &ds); | |
666 | if (err != 0) { | |
667 | dsl_pool_rele(dp, FTAG); | |
34dc7c2f | 668 | return (err); |
a08ee875 | 669 | } |
e10b0808 | 670 | err = dmu_objset_own_impl(ds, type, readonly, tag, osp); |
a08ee875 | 671 | dsl_pool_rele(dp, FTAG); |
e10b0808 | 672 | |
cae5b340 AX |
673 | if (err == 0 && dmu_objset_userobjspace_upgradable(*osp)) |
674 | dmu_objset_userobjspace_upgrade(*osp); | |
675 | ||
34dc7c2f BB |
676 | return (err); |
677 | } | |
678 | ||
e10b0808 AX |
679 | int |
680 | dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, | |
681 | boolean_t readonly, void *tag, objset_t **osp) | |
682 | { | |
683 | dsl_dataset_t *ds; | |
684 | int err; | |
685 | ||
686 | err = dsl_dataset_own_obj(dp, obj, tag, &ds); | |
687 | if (err != 0) | |
688 | return (err); | |
689 | ||
690 | return (dmu_objset_own_impl(ds, type, readonly, tag, osp)); | |
691 | } | |
692 | ||
34dc7c2f | 693 | void |
428870ff | 694 | dmu_objset_rele(objset_t *os, void *tag) |
34dc7c2f | 695 | { |
a08ee875 | 696 | dsl_pool_t *dp = dmu_objset_pool(os); |
428870ff | 697 | dsl_dataset_rele(os->os_dsl_dataset, tag); |
a08ee875 LG |
698 | dsl_pool_rele(dp, tag); |
699 | } | |
700 | ||
701 | /* | |
702 | * When we are called, os MUST refer to an objset associated with a dataset | |
703 | * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner | |
704 | * == tag. We will then release and reacquire ownership of the dataset while | |
705 | * holding the pool config_rwlock to avoid intervening namespace or ownership | |
706 | * changes may occur. | |
707 | * | |
708 | * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to | |
709 | * release the hold on its dataset and acquire a new one on the dataset of the | |
710 | * same name so that it can be partially torn down and reconstructed. | |
711 | */ | |
712 | void | |
713 | dmu_objset_refresh_ownership(objset_t *os, void *tag) | |
714 | { | |
715 | dsl_pool_t *dp; | |
716 | dsl_dataset_t *ds, *newds; | |
cae5b340 | 717 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
a08ee875 LG |
718 | |
719 | ds = os->os_dsl_dataset; | |
720 | VERIFY3P(ds, !=, NULL); | |
721 | VERIFY3P(ds->ds_owner, ==, tag); | |
722 | VERIFY(dsl_dataset_long_held(ds)); | |
723 | ||
724 | dsl_dataset_name(ds, name); | |
725 | dp = dmu_objset_pool(os); | |
726 | dsl_pool_config_enter(dp, FTAG); | |
727 | dmu_objset_disown(os, tag); | |
728 | VERIFY0(dsl_dataset_own(dp, name, tag, &newds)); | |
729 | VERIFY3P(newds, ==, os->os_dsl_dataset); | |
730 | dsl_pool_config_exit(dp, FTAG); | |
428870ff | 731 | } |
b128c09f | 732 | |
428870ff BB |
733 | void |
734 | dmu_objset_disown(objset_t *os, void *tag) | |
735 | { | |
cae5b340 AX |
736 | /* |
737 | * Stop upgrading thread | |
738 | */ | |
739 | dmu_objset_upgrade_stop(os); | |
428870ff | 740 | dsl_dataset_disown(os->os_dsl_dataset, tag); |
34dc7c2f BB |
741 | } |
742 | ||
a08ee875 | 743 | void |
34dc7c2f BB |
744 | dmu_objset_evict_dbufs(objset_t *os) |
745 | { | |
e10b0808 | 746 | dnode_t *dn_marker; |
34dc7c2f BB |
747 | dnode_t *dn; |
748 | ||
e10b0808 | 749 | dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); |
34dc7c2f | 750 | |
e10b0808 AX |
751 | mutex_enter(&os->os_lock); |
752 | dn = list_head(&os->os_dnodes); | |
753 | while (dn != NULL) { | |
754 | /* | |
755 | * Skip dnodes without holds. We have to do this dance | |
756 | * because dnode_add_ref() only works if there is already a | |
757 | * hold. If the dnode has no holds, then it has no dbufs. | |
758 | */ | |
759 | if (dnode_add_ref(dn, FTAG)) { | |
760 | list_insert_after(&os->os_dnodes, dn, dn_marker); | |
761 | mutex_exit(&os->os_lock); | |
34dc7c2f | 762 | |
e10b0808 AX |
763 | dnode_evict_dbufs(dn); |
764 | dnode_rele(dn, FTAG); | |
34dc7c2f | 765 | |
e10b0808 AX |
766 | mutex_enter(&os->os_lock); |
767 | dn = list_next(&os->os_dnodes, dn_marker); | |
768 | list_remove(&os->os_dnodes, dn_marker); | |
769 | } else { | |
770 | dn = list_next(&os->os_dnodes, dn); | |
771 | } | |
772 | } | |
773 | mutex_exit(&os->os_lock); | |
34dc7c2f | 774 | |
e10b0808 | 775 | kmem_free(dn_marker, sizeof (dnode_t)); |
34dc7c2f | 776 | |
e10b0808 AX |
777 | if (DMU_USERUSED_DNODE(os) != NULL) { |
778 | dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); | |
779 | dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); | |
34dc7c2f | 780 | } |
e10b0808 | 781 | dnode_evict_dbufs(DMU_META_DNODE(os)); |
34dc7c2f BB |
782 | } |
783 | ||
e10b0808 AX |
784 | /* |
785 | * Objset eviction processing is split into into two pieces. | |
786 | * The first marks the objset as evicting, evicts any dbufs that | |
787 | * have a refcount of zero, and then queues up the objset for the | |
788 | * second phase of eviction. Once os->os_dnodes has been cleared by | |
789 | * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. | |
790 | * The second phase closes the special dnodes, dequeues the objset from | |
791 | * the list of those undergoing eviction, and finally frees the objset. | |
792 | * | |
793 | * NOTE: Due to asynchronous eviction processing (invocation of | |
794 | * dnode_buf_pageout()), it is possible for the meta dnode for the | |
795 | * objset to have no holds even though os->os_dnodes is not empty. | |
796 | */ | |
34dc7c2f | 797 | void |
428870ff | 798 | dmu_objset_evict(objset_t *os) |
34dc7c2f | 799 | { |
d6320ddb | 800 | int t; |
34dc7c2f | 801 | |
a08ee875 LG |
802 | dsl_dataset_t *ds = os->os_dsl_dataset; |
803 | ||
d6320ddb | 804 | for (t = 0; t < TXG_SIZE; t++) |
428870ff | 805 | ASSERT(!dmu_objset_is_dirty(os, t)); |
34dc7c2f | 806 | |
cae5b340 AX |
807 | if (ds) |
808 | dsl_prop_unregister_all(ds, os); | |
34dc7c2f | 809 | |
428870ff BB |
810 | if (os->os_sa) |
811 | sa_tear_down(os); | |
812 | ||
a08ee875 | 813 | dmu_objset_evict_dbufs(os); |
34dc7c2f | 814 | |
e10b0808 AX |
815 | mutex_enter(&os->os_lock); |
816 | spa_evicting_os_register(os->os_spa, os); | |
817 | if (list_is_empty(&os->os_dnodes)) { | |
818 | mutex_exit(&os->os_lock); | |
819 | dmu_objset_evict_done(os); | |
820 | } else { | |
821 | mutex_exit(&os->os_lock); | |
822 | } | |
823 | } | |
824 | ||
825 | void | |
826 | dmu_objset_evict_done(objset_t *os) | |
827 | { | |
828 | ASSERT3P(list_head(&os->os_dnodes), ==, NULL); | |
829 | ||
572e2857 BB |
830 | dnode_special_close(&os->os_meta_dnode); |
831 | if (DMU_USERUSED_DNODE(os)) { | |
832 | dnode_special_close(&os->os_userused_dnode); | |
833 | dnode_special_close(&os->os_groupused_dnode); | |
9babb374 | 834 | } |
428870ff BB |
835 | zil_free(os->os_zil); |
836 | ||
cae5b340 | 837 | arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); |
572e2857 BB |
838 | |
839 | /* | |
840 | * This is a barrier to prevent the objset from going away in | |
841 | * dnode_move() until we can safely ensure that the objset is still in | |
842 | * use. We consider the objset valid before the barrier and invalid | |
843 | * after the barrier. | |
844 | */ | |
845 | rw_enter(&os_lock, RW_READER); | |
846 | rw_exit(&os_lock); | |
847 | ||
cae5b340 AX |
848 | kmem_free(os->os_obj_next_percpu, |
849 | os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); | |
850 | ||
428870ff | 851 | mutex_destroy(&os->os_lock); |
cae5b340 | 852 | mutex_destroy(&os->os_userused_lock); |
428870ff BB |
853 | mutex_destroy(&os->os_obj_lock); |
854 | mutex_destroy(&os->os_user_ptr_lock); | |
cae5b340 AX |
855 | mutex_destroy(&os->os_upgrade_lock); |
856 | for (int i = 0; i < TXG_SIZE; i++) { | |
857 | multilist_destroy(os->os_dirty_dnodes[i]); | |
858 | } | |
e10b0808 | 859 | spa_evicting_os_deregister(os->os_spa, os); |
428870ff BB |
860 | kmem_free(os, sizeof (objset_t)); |
861 | } | |
9babb374 | 862 | |
428870ff BB |
863 | timestruc_t |
864 | dmu_objset_snap_cmtime(objset_t *os) | |
865 | { | |
866 | return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); | |
34dc7c2f BB |
867 | } |
868 | ||
869 | /* called from dsl for meta-objset */ | |
428870ff | 870 | objset_t * |
34dc7c2f BB |
871 | dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, |
872 | dmu_objset_type_t type, dmu_tx_t *tx) | |
873 | { | |
428870ff | 874 | objset_t *os; |
34dc7c2f BB |
875 | dnode_t *mdn; |
876 | ||
877 | ASSERT(dmu_tx_is_syncing(tx)); | |
a08ee875 | 878 | |
572e2857 | 879 | if (ds != NULL) |
a08ee875 | 880 | VERIFY0(dmu_objset_from_ds(ds, &os)); |
572e2857 | 881 | else |
a08ee875 | 882 | VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); |
572e2857 BB |
883 | |
884 | mdn = DMU_META_DNODE(os); | |
34dc7c2f | 885 | |
cae5b340 AX |
886 | dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT, |
887 | DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx); | |
34dc7c2f BB |
888 | |
889 | /* | |
890 | * We don't want to have to increase the meta-dnode's nlevels | |
891 | * later, because then we could do it in quescing context while | |
892 | * we are also accessing it in open context. | |
893 | * | |
894 | * This precaution is not necessary for the MOS (ds == NULL), | |
895 | * because the MOS is only updated in syncing context. | |
896 | * This is most fortunate: the MOS is the only objset that | |
897 | * needs to be synced multiple times as spa_sync() iterates | |
898 | * to convergence, so minimizing its dn_nlevels matters. | |
899 | */ | |
900 | if (ds != NULL) { | |
901 | int levels = 1; | |
902 | ||
903 | /* | |
904 | * Determine the number of levels necessary for the meta-dnode | |
cae5b340 AX |
905 | * to contain DN_MAX_OBJECT dnodes. Note that in order to |
906 | * ensure that we do not overflow 64 bits, there has to be | |
907 | * a nlevels that gives us a number of blocks > DN_MAX_OBJECT | |
908 | * but < 2^64. Therefore, | |
909 | * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be | |
910 | * less than (64 - log2(DN_MAX_OBJECT)) (16). | |
34dc7c2f | 911 | */ |
cae5b340 AX |
912 | while ((uint64_t)mdn->dn_nblkptr << |
913 | (mdn->dn_datablkshift - DNODE_SHIFT + | |
34dc7c2f | 914 | (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < |
cae5b340 | 915 | DN_MAX_OBJECT) |
34dc7c2f BB |
916 | levels++; |
917 | ||
918 | mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = | |
919 | mdn->dn_nlevels = levels; | |
920 | } | |
921 | ||
922 | ASSERT(type != DMU_OST_NONE); | |
923 | ASSERT(type != DMU_OST_ANY); | |
924 | ASSERT(type < DMU_OST_NUMTYPES); | |
428870ff BB |
925 | os->os_phys->os_type = type; |
926 | if (dmu_objset_userused_enabled(os)) { | |
927 | os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; | |
cae5b340 AX |
928 | if (dmu_objset_userobjused_enabled(os)) { |
929 | ds->ds_feature_activation_needed[ | |
930 | SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE; | |
931 | os->os_phys->os_flags |= | |
932 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
933 | } | |
428870ff | 934 | os->os_flags = os->os_phys->os_flags; |
9babb374 | 935 | } |
34dc7c2f BB |
936 | |
937 | dsl_dataset_dirty(ds, tx); | |
938 | ||
428870ff | 939 | return (os); |
34dc7c2f BB |
940 | } |
941 | ||
a08ee875 LG |
942 | typedef struct dmu_objset_create_arg { |
943 | const char *doca_name; | |
944 | cred_t *doca_cred; | |
945 | void (*doca_userfunc)(objset_t *os, void *arg, | |
946 | cred_t *cr, dmu_tx_t *tx); | |
947 | void *doca_userarg; | |
948 | dmu_objset_type_t doca_type; | |
949 | uint64_t doca_flags; | |
950 | } dmu_objset_create_arg_t; | |
34dc7c2f BB |
951 | |
952 | /*ARGSUSED*/ | |
953 | static int | |
a08ee875 | 954 | dmu_objset_create_check(void *arg, dmu_tx_t *tx) |
34dc7c2f | 955 | { |
a08ee875 LG |
956 | dmu_objset_create_arg_t *doca = arg; |
957 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
958 | dsl_dir_t *pdd; | |
959 | const char *tail; | |
960 | int error; | |
34dc7c2f | 961 | |
a08ee875 LG |
962 | if (strchr(doca->doca_name, '@') != NULL) |
963 | return (SET_ERROR(EINVAL)); | |
34dc7c2f | 964 | |
cae5b340 AX |
965 | if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) |
966 | return (SET_ERROR(ENAMETOOLONG)); | |
967 | ||
a08ee875 LG |
968 | error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); |
969 | if (error != 0) | |
970 | return (error); | |
971 | if (tail == NULL) { | |
972 | dsl_dir_rele(pdd, FTAG); | |
973 | return (SET_ERROR(EEXIST)); | |
34dc7c2f | 974 | } |
e10b0808 AX |
975 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, |
976 | doca->doca_cred); | |
a08ee875 | 977 | dsl_dir_rele(pdd, FTAG); |
34dc7c2f | 978 | |
e10b0808 | 979 | return (error); |
34dc7c2f BB |
980 | } |
981 | ||
982 | static void | |
a08ee875 | 983 | dmu_objset_create_sync(void *arg, dmu_tx_t *tx) |
34dc7c2f | 984 | { |
a08ee875 LG |
985 | dmu_objset_create_arg_t *doca = arg; |
986 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
987 | dsl_dir_t *pdd; | |
988 | const char *tail; | |
989 | dsl_dataset_t *ds; | |
572e2857 | 990 | uint64_t obj; |
a08ee875 LG |
991 | blkptr_t *bp; |
992 | objset_t *os; | |
34dc7c2f | 993 | |
a08ee875 | 994 | VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); |
34dc7c2f | 995 | |
a08ee875 LG |
996 | obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, |
997 | doca->doca_cred, tx); | |
428870ff | 998 | |
a08ee875 | 999 | VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); |
cae5b340 | 1000 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
a08ee875 LG |
1001 | bp = dsl_dataset_get_blkptr(ds); |
1002 | os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, | |
1003 | ds, bp, doca->doca_type, tx); | |
cae5b340 | 1004 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
34dc7c2f | 1005 | |
a08ee875 LG |
1006 | if (doca->doca_userfunc != NULL) { |
1007 | doca->doca_userfunc(os, doca->doca_userarg, | |
1008 | doca->doca_cred, tx); | |
34dc7c2f BB |
1009 | } |
1010 | ||
a08ee875 | 1011 | spa_history_log_internal_ds(ds, "create", tx, ""); |
4e820b5a AX |
1012 | zvol_create_minors(dp->dp_spa, doca->doca_name, B_TRUE); |
1013 | ||
a08ee875 LG |
1014 | dsl_dataset_rele(ds, FTAG); |
1015 | dsl_dir_rele(pdd, FTAG); | |
34dc7c2f BB |
1016 | } |
1017 | ||
1018 | int | |
428870ff | 1019 | dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, |
34dc7c2f BB |
1020 | void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) |
1021 | { | |
a08ee875 | 1022 | dmu_objset_create_arg_t doca; |
34dc7c2f | 1023 | |
a08ee875 LG |
1024 | doca.doca_name = name; |
1025 | doca.doca_cred = CRED(); | |
1026 | doca.doca_flags = flags; | |
1027 | doca.doca_userfunc = func; | |
1028 | doca.doca_userarg = arg; | |
1029 | doca.doca_type = type; | |
34dc7c2f | 1030 | |
a08ee875 | 1031 | return (dsl_sync_task(name, |
e10b0808 AX |
1032 | dmu_objset_create_check, dmu_objset_create_sync, &doca, |
1033 | 5, ZFS_SPACE_CHECK_NORMAL)); | |
34dc7c2f BB |
1034 | } |
1035 | ||
a08ee875 LG |
1036 | typedef struct dmu_objset_clone_arg { |
1037 | const char *doca_clone; | |
1038 | const char *doca_origin; | |
1039 | cred_t *doca_cred; | |
1040 | } dmu_objset_clone_arg_t; | |
1041 | ||
1042 | /*ARGSUSED*/ | |
1043 | static int | |
1044 | dmu_objset_clone_check(void *arg, dmu_tx_t *tx) | |
34dc7c2f | 1045 | { |
a08ee875 | 1046 | dmu_objset_clone_arg_t *doca = arg; |
428870ff BB |
1047 | dsl_dir_t *pdd; |
1048 | const char *tail; | |
a08ee875 LG |
1049 | int error; |
1050 | dsl_dataset_t *origin; | |
1051 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
34dc7c2f | 1052 | |
a08ee875 LG |
1053 | if (strchr(doca->doca_clone, '@') != NULL) |
1054 | return (SET_ERROR(EINVAL)); | |
1055 | ||
cae5b340 AX |
1056 | if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) |
1057 | return (SET_ERROR(ENAMETOOLONG)); | |
1058 | ||
a08ee875 LG |
1059 | error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); |
1060 | if (error != 0) | |
1061 | return (error); | |
428870ff | 1062 | if (tail == NULL) { |
a08ee875 LG |
1063 | dsl_dir_rele(pdd, FTAG); |
1064 | return (SET_ERROR(EEXIST)); | |
34dc7c2f | 1065 | } |
e10b0808 AX |
1066 | |
1067 | error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, | |
1068 | doca->doca_cred); | |
1069 | if (error != 0) { | |
a08ee875 | 1070 | dsl_dir_rele(pdd, FTAG); |
e10b0808 | 1071 | return (SET_ERROR(EDQUOT)); |
428870ff | 1072 | } |
a08ee875 | 1073 | dsl_dir_rele(pdd, FTAG); |
428870ff | 1074 | |
a08ee875 LG |
1075 | error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); |
1076 | if (error != 0) | |
572e2857 BB |
1077 | return (error); |
1078 | ||
a08ee875 | 1079 | /* You can only clone snapshots, not the head datasets. */ |
e10b0808 | 1080 | if (!origin->ds_is_snapshot) { |
a08ee875 LG |
1081 | dsl_dataset_rele(origin, FTAG); |
1082 | return (SET_ERROR(EINVAL)); | |
572e2857 | 1083 | } |
a08ee875 LG |
1084 | dsl_dataset_rele(origin, FTAG); |
1085 | ||
1086 | return (0); | |
9babb374 BB |
1087 | } |
1088 | ||
1089 | static void | |
a08ee875 | 1090 | dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) |
9babb374 | 1091 | { |
a08ee875 LG |
1092 | dmu_objset_clone_arg_t *doca = arg; |
1093 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1094 | dsl_dir_t *pdd; | |
1095 | const char *tail; | |
1096 | dsl_dataset_t *origin, *ds; | |
1097 | uint64_t obj; | |
cae5b340 | 1098 | char namebuf[ZFS_MAX_DATASET_NAME_LEN]; |
572e2857 | 1099 | |
a08ee875 LG |
1100 | VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); |
1101 | VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); | |
572e2857 | 1102 | |
a08ee875 LG |
1103 | obj = dsl_dataset_create_sync(pdd, tail, origin, 0, |
1104 | doca->doca_cred, tx); | |
572e2857 | 1105 | |
a08ee875 LG |
1106 | VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); |
1107 | dsl_dataset_name(origin, namebuf); | |
1108 | spa_history_log_internal_ds(ds, "clone", tx, | |
1109 | "origin=%s (%llu)", namebuf, origin->ds_object); | |
4e820b5a | 1110 | zvol_create_minors(dp->dp_spa, doca->doca_clone, B_TRUE); |
a08ee875 LG |
1111 | dsl_dataset_rele(ds, FTAG); |
1112 | dsl_dataset_rele(origin, FTAG); | |
1113 | dsl_dir_rele(pdd, FTAG); | |
9babb374 | 1114 | } |
34dc7c2f | 1115 | |
a08ee875 LG |
1116 | int |
1117 | dmu_objset_clone(const char *clone, const char *origin) | |
34dc7c2f | 1118 | { |
a08ee875 | 1119 | dmu_objset_clone_arg_t doca; |
34dc7c2f | 1120 | |
a08ee875 LG |
1121 | doca.doca_clone = clone; |
1122 | doca.doca_origin = origin; | |
1123 | doca.doca_cred = CRED(); | |
34dc7c2f | 1124 | |
a08ee875 | 1125 | return (dsl_sync_task(clone, |
e10b0808 AX |
1126 | dmu_objset_clone_check, dmu_objset_clone_sync, &doca, |
1127 | 5, ZFS_SPACE_CHECK_NORMAL)); | |
34dc7c2f BB |
1128 | } |
1129 | ||
1130 | int | |
a08ee875 | 1131 | dmu_objset_snapshot_one(const char *fsname, const char *snapname) |
34dc7c2f | 1132 | { |
34dc7c2f | 1133 | int err; |
a08ee875 LG |
1134 | char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); |
1135 | nvlist_t *snaps = fnvlist_alloc(); | |
34dc7c2f | 1136 | |
a08ee875 LG |
1137 | fnvlist_add_boolean(snaps, longsnap); |
1138 | strfree(longsnap); | |
1139 | err = dsl_dataset_snapshot(snaps, NULL, NULL); | |
1140 | fnvlist_free(snaps); | |
34dc7c2f BB |
1141 | return (err); |
1142 | } | |
1143 | ||
1144 | static void | |
cae5b340 AX |
1145 | dmu_objset_upgrade_task_cb(void *data) |
1146 | { | |
1147 | objset_t *os = data; | |
1148 | ||
1149 | mutex_enter(&os->os_upgrade_lock); | |
1150 | os->os_upgrade_status = EINTR; | |
1151 | if (!os->os_upgrade_exit) { | |
1152 | mutex_exit(&os->os_upgrade_lock); | |
1153 | ||
1154 | os->os_upgrade_status = os->os_upgrade_cb(os); | |
1155 | mutex_enter(&os->os_upgrade_lock); | |
1156 | } | |
1157 | os->os_upgrade_exit = B_TRUE; | |
1158 | os->os_upgrade_id = 0; | |
1159 | mutex_exit(&os->os_upgrade_lock); | |
1160 | } | |
1161 | ||
1162 | static void | |
1163 | dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) | |
1164 | { | |
1165 | if (os->os_upgrade_id != 0) | |
1166 | return; | |
1167 | ||
1168 | mutex_enter(&os->os_upgrade_lock); | |
1169 | if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { | |
1170 | os->os_upgrade_exit = B_FALSE; | |
1171 | os->os_upgrade_cb = cb; | |
1172 | os->os_upgrade_id = taskq_dispatch( | |
1173 | os->os_spa->spa_upgrade_taskq, | |
1174 | dmu_objset_upgrade_task_cb, os, TQ_SLEEP); | |
1175 | if (os->os_upgrade_id == TASKQID_INVALID) | |
1176 | os->os_upgrade_status = ENOMEM; | |
1177 | } | |
1178 | mutex_exit(&os->os_upgrade_lock); | |
1179 | } | |
1180 | ||
1181 | static void | |
1182 | dmu_objset_upgrade_stop(objset_t *os) | |
1183 | { | |
1184 | mutex_enter(&os->os_upgrade_lock); | |
1185 | os->os_upgrade_exit = B_TRUE; | |
1186 | if (os->os_upgrade_id != 0) { | |
1187 | taskqid_t id = os->os_upgrade_id; | |
1188 | ||
1189 | os->os_upgrade_id = 0; | |
1190 | mutex_exit(&os->os_upgrade_lock); | |
1191 | ||
1192 | taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id); | |
1193 | } else { | |
1194 | mutex_exit(&os->os_upgrade_lock); | |
1195 | } | |
1196 | } | |
1197 | ||
1198 | static void | |
1199 | dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) | |
34dc7c2f BB |
1200 | { |
1201 | dnode_t *dn; | |
1202 | ||
cae5b340 | 1203 | while ((dn = multilist_sublist_head(list)) != NULL) { |
34dc7c2f BB |
1204 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); |
1205 | ASSERT(dn->dn_dbuf->db_data_pending); | |
1206 | /* | |
9babb374 BB |
1207 | * Initialize dn_zio outside dnode_sync() because the |
1208 | * meta-dnode needs to set it ouside dnode_sync(). | |
34dc7c2f BB |
1209 | */ |
1210 | dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; | |
1211 | ASSERT(dn->dn_zio); | |
1212 | ||
1213 | ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); | |
cae5b340 | 1214 | multilist_sublist_remove(list, dn); |
9babb374 | 1215 | |
cae5b340 AX |
1216 | multilist_t *newlist = dn->dn_objset->os_synced_dnodes; |
1217 | if (newlist != NULL) { | |
9babb374 | 1218 | (void) dnode_add_ref(dn, newlist); |
cae5b340 | 1219 | multilist_insert(newlist, dn); |
9babb374 BB |
1220 | } |
1221 | ||
34dc7c2f BB |
1222 | dnode_sync(dn, tx); |
1223 | } | |
1224 | } | |
1225 | ||
1226 | /* ARGSUSED */ | |
1227 | static void | |
428870ff | 1228 | dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) |
34dc7c2f | 1229 | { |
d6320ddb BB |
1230 | int i; |
1231 | ||
b128c09f | 1232 | blkptr_t *bp = zio->io_bp; |
428870ff | 1233 | objset_t *os = arg; |
34dc7c2f | 1234 | dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; |
34dc7c2f | 1235 | |
ea04106b | 1236 | ASSERT(!BP_IS_EMBEDDED(bp)); |
a08ee875 LG |
1237 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); |
1238 | ASSERT0(BP_GET_LEVEL(bp)); | |
34dc7c2f BB |
1239 | |
1240 | /* | |
9babb374 BB |
1241 | * Update rootbp fill count: it should be the number of objects |
1242 | * allocated in the object set (not counting the "special" | |
1243 | * objects that are stored in the objset_phys_t -- the meta | |
1244 | * dnode and user/group accounting objects). | |
34dc7c2f | 1245 | */ |
9babb374 | 1246 | bp->blk_fill = 0; |
d6320ddb | 1247 | for (i = 0; i < dnp->dn_nblkptr; i++) |
ea04106b | 1248 | bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]); |
cae5b340 AX |
1249 | if (os->os_dsl_dataset != NULL) |
1250 | rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); | |
1251 | *os->os_rootbp = *bp; | |
1252 | if (os->os_dsl_dataset != NULL) | |
1253 | rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); | |
428870ff BB |
1254 | } |
1255 | ||
1256 | /* ARGSUSED */ | |
1257 | static void | |
1258 | dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) | |
1259 | { | |
1260 | blkptr_t *bp = zio->io_bp; | |
1261 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1262 | objset_t *os = arg; | |
34dc7c2f | 1263 | |
b128c09f | 1264 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
428870ff | 1265 | ASSERT(BP_EQUAL(bp, bp_orig)); |
b128c09f | 1266 | } else { |
428870ff BB |
1267 | dsl_dataset_t *ds = os->os_dsl_dataset; |
1268 | dmu_tx_t *tx = os->os_synctx; | |
1269 | ||
1270 | (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); | |
1271 | dsl_dataset_block_born(ds, bp, tx); | |
34dc7c2f | 1272 | } |
cae5b340 | 1273 | kmem_free(bp, sizeof (*bp)); |
34dc7c2f BB |
1274 | } |
1275 | ||
cae5b340 AX |
1276 | typedef struct sync_dnodes_arg { |
1277 | multilist_t *sda_list; | |
1278 | int sda_sublist_idx; | |
1279 | multilist_t *sda_newlist; | |
1280 | dmu_tx_t *sda_tx; | |
1281 | } sync_dnodes_arg_t; | |
1282 | ||
1283 | static void | |
1284 | sync_dnodes_task(void *arg) | |
1285 | { | |
1286 | sync_dnodes_arg_t *sda = arg; | |
1287 | ||
1288 | multilist_sublist_t *ms = | |
1289 | multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); | |
1290 | ||
1291 | dmu_objset_sync_dnodes(ms, sda->sda_tx); | |
1292 | ||
1293 | multilist_sublist_unlock(ms); | |
1294 | ||
1295 | kmem_free(sda, sizeof (*sda)); | |
1296 | } | |
1297 | ||
1298 | ||
34dc7c2f BB |
1299 | /* called from dsl */ |
1300 | void | |
428870ff | 1301 | dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) |
34dc7c2f BB |
1302 | { |
1303 | int txgoff; | |
ea04106b | 1304 | zbookmark_phys_t zb; |
428870ff | 1305 | zio_prop_t zp; |
34dc7c2f BB |
1306 | zio_t *zio; |
1307 | list_t *list; | |
1308 | dbuf_dirty_record_t *dr; | |
cae5b340 AX |
1309 | blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); |
1310 | *blkptr_copy = *os->os_rootbp; | |
34dc7c2f BB |
1311 | |
1312 | dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); | |
1313 | ||
1314 | ASSERT(dmu_tx_is_syncing(tx)); | |
1315 | /* XXX the write_done callback should really give us the tx... */ | |
1316 | os->os_synctx = tx; | |
1317 | ||
1318 | if (os->os_dsl_dataset == NULL) { | |
1319 | /* | |
1320 | * This is the MOS. If we have upgraded, | |
1321 | * spa_max_replication() could change, so reset | |
1322 | * os_copies here. | |
1323 | */ | |
1324 | os->os_copies = spa_max_replication(os->os_spa); | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * Create the root block IO | |
1329 | */ | |
428870ff BB |
1330 | SET_BOOKMARK(&zb, os->os_dsl_dataset ? |
1331 | os->os_dsl_dataset->ds_object : DMU_META_OBJSET, | |
1332 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
c06d4368 | 1333 | arc_release(os->os_phys_buf, &os->os_phys_buf); |
b128c09f | 1334 | |
428870ff | 1335 | dmu_write_policy(os, NULL, 0, 0, &zp); |
9babb374 | 1336 | |
428870ff | 1337 | zio = arc_write(pio, os->os_spa, tx->tx_txg, |
cae5b340 AX |
1338 | blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), |
1339 | &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, | |
1340 | os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); | |
34dc7c2f BB |
1341 | |
1342 | /* | |
9babb374 | 1343 | * Sync special dnodes - the parent IO for the sync is the root block |
34dc7c2f | 1344 | */ |
572e2857 BB |
1345 | DMU_META_DNODE(os)->dn_zio = zio; |
1346 | dnode_sync(DMU_META_DNODE(os), tx); | |
34dc7c2f | 1347 | |
9babb374 BB |
1348 | os->os_phys->os_flags = os->os_flags; |
1349 | ||
572e2857 BB |
1350 | if (DMU_USERUSED_DNODE(os) && |
1351 | DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { | |
1352 | DMU_USERUSED_DNODE(os)->dn_zio = zio; | |
1353 | dnode_sync(DMU_USERUSED_DNODE(os), tx); | |
1354 | DMU_GROUPUSED_DNODE(os)->dn_zio = zio; | |
1355 | dnode_sync(DMU_GROUPUSED_DNODE(os), tx); | |
9babb374 BB |
1356 | } |
1357 | ||
34dc7c2f BB |
1358 | txgoff = tx->tx_txg & TXG_MASK; |
1359 | ||
9babb374 | 1360 | if (dmu_objset_userused_enabled(os)) { |
9babb374 BB |
1361 | /* |
1362 | * We must create the list here because it uses the | |
cae5b340 AX |
1363 | * dn_dirty_link[] of this txg. But it may already |
1364 | * exist because we call dsl_dataset_sync() twice per txg. | |
9babb374 | 1365 | */ |
cae5b340 AX |
1366 | if (os->os_synced_dnodes == NULL) { |
1367 | os->os_synced_dnodes = | |
1368 | multilist_create(sizeof (dnode_t), | |
1369 | offsetof(dnode_t, dn_dirty_link[txgoff]), | |
1370 | dnode_multilist_index_func); | |
1371 | } else { | |
1372 | ASSERT3U(os->os_synced_dnodes->ml_offset, ==, | |
1373 | offsetof(dnode_t, dn_dirty_link[txgoff])); | |
1374 | } | |
9babb374 BB |
1375 | } |
1376 | ||
cae5b340 AX |
1377 | for (int i = 0; |
1378 | i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { | |
1379 | sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); | |
1380 | sda->sda_list = os->os_dirty_dnodes[txgoff]; | |
1381 | sda->sda_sublist_idx = i; | |
1382 | sda->sda_tx = tx; | |
1383 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1384 | sync_dnodes_task, sda, 0); | |
1385 | /* callback frees sda */ | |
1386 | } | |
1387 | taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); | |
34dc7c2f | 1388 | |
572e2857 | 1389 | list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; |
cae5b340 | 1390 | while ((dr = list_head(list)) != NULL) { |
a08ee875 | 1391 | ASSERT0(dr->dr_dbuf->db_level); |
34dc7c2f BB |
1392 | list_remove(list, dr); |
1393 | if (dr->dr_zio) | |
1394 | zio_nowait(dr->dr_zio); | |
1395 | } | |
cae5b340 AX |
1396 | |
1397 | /* Enable dnode backfill if enough objects have been freed. */ | |
1398 | if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { | |
1399 | os->os_rescan_dnodes = B_TRUE; | |
1400 | os->os_freed_dnodes = 0; | |
1401 | } | |
1402 | ||
34dc7c2f BB |
1403 | /* |
1404 | * Free intent log blocks up to this tx. | |
1405 | */ | |
1406 | zil_sync(os->os_zil, tx); | |
b128c09f | 1407 | os->os_phys->os_zil_header = os->os_zil_header; |
34dc7c2f BB |
1408 | zio_nowait(zio); |
1409 | } | |
1410 | ||
428870ff BB |
1411 | boolean_t |
1412 | dmu_objset_is_dirty(objset_t *os, uint64_t txg) | |
1413 | { | |
cae5b340 | 1414 | return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); |
428870ff BB |
1415 | } |
1416 | ||
572e2857 | 1417 | static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; |
9babb374 BB |
1418 | |
1419 | void | |
1420 | dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) | |
1421 | { | |
1422 | used_cbs[ost] = cb; | |
1423 | } | |
1424 | ||
1425 | boolean_t | |
428870ff | 1426 | dmu_objset_userused_enabled(objset_t *os) |
9babb374 BB |
1427 | { |
1428 | return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && | |
572e2857 BB |
1429 | used_cbs[os->os_phys->os_type] != NULL && |
1430 | DMU_USERUSED_DNODE(os) != NULL); | |
9babb374 BB |
1431 | } |
1432 | ||
cae5b340 AX |
1433 | boolean_t |
1434 | dmu_objset_userobjused_enabled(objset_t *os) | |
1435 | { | |
1436 | return (dmu_objset_userused_enabled(os) && | |
1437 | spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); | |
1438 | } | |
1439 | ||
1440 | typedef struct userquota_node { | |
1441 | /* must be in the first filed, see userquota_update_cache() */ | |
1442 | char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1443 | int64_t uqn_delta; | |
1444 | avl_node_t uqn_node; | |
1445 | } userquota_node_t; | |
1446 | ||
1447 | typedef struct userquota_cache { | |
1448 | avl_tree_t uqc_user_deltas; | |
1449 | avl_tree_t uqc_group_deltas; | |
1450 | } userquota_cache_t; | |
1451 | ||
1452 | static int | |
1453 | userquota_compare(const void *l, const void *r) | |
1454 | { | |
1455 | const userquota_node_t *luqn = l; | |
1456 | const userquota_node_t *ruqn = r; | |
1457 | int rv; | |
1458 | ||
1459 | /* | |
1460 | * NB: can only access uqn_id because userquota_update_cache() doesn't | |
1461 | * pass in an entire userquota_node_t. | |
1462 | */ | |
1463 | rv = strcmp(luqn->uqn_id, ruqn->uqn_id); | |
1464 | ||
1465 | return (AVL_ISIGN(rv)); | |
1466 | } | |
1467 | ||
428870ff | 1468 | static void |
cae5b340 AX |
1469 | do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) |
1470 | { | |
1471 | void *cookie; | |
1472 | userquota_node_t *uqn; | |
1473 | ||
1474 | ASSERT(dmu_tx_is_syncing(tx)); | |
1475 | ||
1476 | cookie = NULL; | |
1477 | while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, | |
1478 | &cookie)) != NULL) { | |
1479 | /* | |
1480 | * os_userused_lock protects against concurrent calls to | |
1481 | * zap_increment_int(). It's needed because zap_increment_int() | |
1482 | * is not thread-safe (i.e. not atomic). | |
1483 | */ | |
1484 | mutex_enter(&os->os_userused_lock); | |
1485 | VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, | |
1486 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1487 | mutex_exit(&os->os_userused_lock); | |
1488 | kmem_free(uqn, sizeof (*uqn)); | |
1489 | } | |
1490 | avl_destroy(&cache->uqc_user_deltas); | |
1491 | ||
1492 | cookie = NULL; | |
1493 | while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, | |
1494 | &cookie)) != NULL) { | |
1495 | mutex_enter(&os->os_userused_lock); | |
1496 | VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, | |
1497 | uqn->uqn_id, uqn->uqn_delta, tx)); | |
1498 | mutex_exit(&os->os_userused_lock); | |
1499 | kmem_free(uqn, sizeof (*uqn)); | |
1500 | } | |
1501 | avl_destroy(&cache->uqc_group_deltas); | |
1502 | } | |
1503 | ||
1504 | static void | |
1505 | userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) | |
1506 | { | |
1507 | userquota_node_t *uqn; | |
1508 | avl_index_t idx; | |
1509 | ||
1510 | ASSERT(strlen(id) < sizeof (uqn->uqn_id)); | |
1511 | /* | |
1512 | * Use id directly for searching because uqn_id is the first field of | |
1513 | * userquota_node_t and fields after uqn_id won't be accessed in | |
1514 | * avl_find(). | |
1515 | */ | |
1516 | uqn = avl_find(avl, (const void *)id, &idx); | |
1517 | if (uqn == NULL) { | |
1518 | uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); | |
1519 | strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); | |
1520 | avl_insert(avl, uqn, idx); | |
1521 | } | |
1522 | uqn->uqn_delta += delta; | |
1523 | } | |
1524 | ||
1525 | static void | |
1526 | do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags, | |
1527 | uint64_t user, uint64_t group, boolean_t subtract) | |
428870ff BB |
1528 | { |
1529 | if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { | |
cae5b340 AX |
1530 | int64_t delta = DNODE_MIN_SIZE + used; |
1531 | char name[20]; | |
1532 | ||
428870ff BB |
1533 | if (subtract) |
1534 | delta = -delta; | |
cae5b340 AX |
1535 | |
1536 | (void) sprintf(name, "%llx", (longlong_t)user); | |
1537 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); | |
1538 | ||
1539 | (void) sprintf(name, "%llx", (longlong_t)group); | |
1540 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); | |
428870ff BB |
1541 | } |
1542 | } | |
1543 | ||
cae5b340 AX |
1544 | static void |
1545 | do_userobjquota_update(userquota_cache_t *cache, uint64_t flags, | |
1546 | uint64_t user, uint64_t group, boolean_t subtract) | |
9babb374 | 1547 | { |
cae5b340 AX |
1548 | if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { |
1549 | char name[20 + DMU_OBJACCT_PREFIX_LEN]; | |
1550 | int delta = subtract ? -1 : 1; | |
1551 | ||
1552 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1553 | (longlong_t)user); | |
1554 | userquota_update_cache(&cache->uqc_user_deltas, name, delta); | |
1555 | ||
1556 | (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", | |
1557 | (longlong_t)group); | |
1558 | userquota_update_cache(&cache->uqc_group_deltas, name, delta); | |
1559 | } | |
1560 | } | |
1561 | ||
1562 | typedef struct userquota_updates_arg { | |
1563 | objset_t *uua_os; | |
1564 | int uua_sublist_idx; | |
1565 | dmu_tx_t *uua_tx; | |
1566 | } userquota_updates_arg_t; | |
1567 | ||
1568 | static void | |
1569 | userquota_updates_task(void *arg) | |
1570 | { | |
1571 | userquota_updates_arg_t *uua = arg; | |
1572 | objset_t *os = uua->uua_os; | |
1573 | dmu_tx_t *tx = uua->uua_tx; | |
9babb374 | 1574 | dnode_t *dn; |
cae5b340 | 1575 | userquota_cache_t cache = { { 0 } }; |
9babb374 | 1576 | |
cae5b340 AX |
1577 | multilist_sublist_t *list = |
1578 | multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); | |
9babb374 | 1579 | |
cae5b340 AX |
1580 | ASSERT(multilist_sublist_head(list) == NULL || |
1581 | dmu_objset_userused_enabled(os)); | |
1582 | avl_create(&cache.uqc_user_deltas, userquota_compare, | |
1583 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
1584 | avl_create(&cache.uqc_group_deltas, userquota_compare, | |
1585 | sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); | |
1586 | ||
1587 | while ((dn = multilist_sublist_head(list)) != NULL) { | |
572e2857 | 1588 | int flags; |
9babb374 | 1589 | ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); |
9babb374 BB |
1590 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || |
1591 | dn->dn_phys->dn_flags & | |
1592 | DNODE_FLAG_USERUSED_ACCOUNTED); | |
1593 | ||
572e2857 BB |
1594 | flags = dn->dn_id_flags; |
1595 | ASSERT(flags); | |
1596 | if (flags & DN_ID_OLD_EXIST) { | |
cae5b340 AX |
1597 | do_userquota_update(&cache, |
1598 | dn->dn_oldused, dn->dn_oldflags, | |
1599 | dn->dn_olduid, dn->dn_oldgid, B_TRUE); | |
1600 | do_userobjquota_update(&cache, dn->dn_oldflags, | |
1601 | dn->dn_olduid, dn->dn_oldgid, B_TRUE); | |
428870ff | 1602 | } |
572e2857 | 1603 | if (flags & DN_ID_NEW_EXIST) { |
cae5b340 AX |
1604 | do_userquota_update(&cache, |
1605 | DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, | |
1606 | dn->dn_newuid, dn->dn_newgid, B_FALSE); | |
1607 | do_userobjquota_update(&cache, dn->dn_phys->dn_flags, | |
1608 | dn->dn_newuid, dn->dn_newgid, B_FALSE); | |
428870ff BB |
1609 | } |
1610 | ||
572e2857 | 1611 | mutex_enter(&dn->dn_mtx); |
428870ff BB |
1612 | dn->dn_oldused = 0; |
1613 | dn->dn_oldflags = 0; | |
1614 | if (dn->dn_id_flags & DN_ID_NEW_EXIST) { | |
1615 | dn->dn_olduid = dn->dn_newuid; | |
1616 | dn->dn_oldgid = dn->dn_newgid; | |
1617 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
1618 | if (dn->dn_bonuslen == 0) | |
1619 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
1620 | else | |
1621 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
1622 | } | |
1623 | dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); | |
9babb374 BB |
1624 | mutex_exit(&dn->dn_mtx); |
1625 | ||
cae5b340 AX |
1626 | multilist_sublist_remove(list, dn); |
1627 | dnode_rele(dn, os->os_synced_dnodes); | |
1628 | } | |
1629 | do_userquota_cacheflush(os, &cache, tx); | |
1630 | multilist_sublist_unlock(list); | |
1631 | kmem_free(uua, sizeof (*uua)); | |
1632 | } | |
1633 | ||
1634 | void | |
1635 | dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) | |
1636 | { | |
1637 | if (!dmu_objset_userused_enabled(os)) | |
1638 | return; | |
1639 | ||
1640 | /* Allocate the user/groupused objects if necessary. */ | |
1641 | if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { | |
1642 | VERIFY0(zap_create_claim(os, | |
1643 | DMU_USERUSED_OBJECT, | |
1644 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
1645 | VERIFY0(zap_create_claim(os, | |
1646 | DMU_GROUPUSED_OBJECT, | |
1647 | DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); | |
1648 | } | |
1649 | ||
1650 | for (int i = 0; | |
1651 | i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { | |
1652 | userquota_updates_arg_t *uua = | |
1653 | kmem_alloc(sizeof (*uua), KM_SLEEP); | |
1654 | uua->uua_os = os; | |
1655 | uua->uua_sublist_idx = i; | |
1656 | uua->uua_tx = tx; | |
1657 | /* note: caller does taskq_wait() */ | |
1658 | (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, | |
1659 | userquota_updates_task, uua, 0); | |
1660 | /* callback frees uua */ | |
9babb374 BB |
1661 | } |
1662 | } | |
1663 | ||
428870ff BB |
1664 | /* |
1665 | * Returns a pointer to data to find uid/gid from | |
1666 | * | |
1667 | * If a dirty record for transaction group that is syncing can't | |
1668 | * be found then NULL is returned. In the NULL case it is assumed | |
1669 | * the uid/gid aren't changing. | |
1670 | */ | |
1671 | static void * | |
1672 | dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) | |
1673 | { | |
1674 | dbuf_dirty_record_t *dr, **drp; | |
1675 | void *data; | |
1676 | ||
1677 | if (db->db_dirtycnt == 0) | |
1678 | return (db->db.db_data); /* Nothing is changing */ | |
1679 | ||
1680 | for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) | |
1681 | if (dr->dr_txg == tx->tx_txg) | |
1682 | break; | |
1683 | ||
572e2857 | 1684 | if (dr == NULL) { |
428870ff | 1685 | data = NULL; |
572e2857 BB |
1686 | } else { |
1687 | dnode_t *dn; | |
1688 | ||
1689 | DB_DNODE_ENTER(dr->dr_dbuf); | |
1690 | dn = DB_DNODE(dr->dr_dbuf); | |
1691 | ||
1692 | if (dn->dn_bonuslen == 0 && | |
1693 | dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) | |
1694 | data = dr->dt.dl.dr_data->b_data; | |
1695 | else | |
1696 | data = dr->dt.dl.dr_data; | |
1697 | ||
1698 | DB_DNODE_EXIT(dr->dr_dbuf); | |
1699 | } | |
1700 | ||
428870ff BB |
1701 | return (data); |
1702 | } | |
1703 | ||
1704 | void | |
1705 | dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) | |
1706 | { | |
1707 | objset_t *os = dn->dn_objset; | |
1708 | void *data = NULL; | |
1709 | dmu_buf_impl_t *db = NULL; | |
a08ee875 LG |
1710 | uint64_t *user = NULL; |
1711 | uint64_t *group = NULL; | |
428870ff BB |
1712 | int flags = dn->dn_id_flags; |
1713 | int error; | |
1714 | boolean_t have_spill = B_FALSE; | |
1715 | ||
1716 | if (!dmu_objset_userused_enabled(dn->dn_objset)) | |
1717 | return; | |
1718 | ||
1719 | if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| | |
1720 | DN_ID_CHKED_SPILL))) | |
1721 | return; | |
1722 | ||
1723 | if (before && dn->dn_bonuslen != 0) | |
1724 | data = DN_BONUS(dn->dn_phys); | |
1725 | else if (!before && dn->dn_bonuslen != 0) { | |
1726 | if (dn->dn_bonus) { | |
1727 | db = dn->dn_bonus; | |
1728 | mutex_enter(&db->db_mtx); | |
1729 | data = dmu_objset_userquota_find_data(db, tx); | |
1730 | } else { | |
1731 | data = DN_BONUS(dn->dn_phys); | |
1732 | } | |
1733 | } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { | |
1734 | int rf = 0; | |
1735 | ||
1736 | if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) | |
1737 | rf |= DB_RF_HAVESTRUCT; | |
572e2857 BB |
1738 | error = dmu_spill_hold_by_dnode(dn, |
1739 | rf | DB_RF_MUST_SUCCEED, | |
428870ff BB |
1740 | FTAG, (dmu_buf_t **)&db); |
1741 | ASSERT(error == 0); | |
1742 | mutex_enter(&db->db_mtx); | |
1743 | data = (before) ? db->db.db_data : | |
1744 | dmu_objset_userquota_find_data(db, tx); | |
1745 | have_spill = B_TRUE; | |
1746 | } else { | |
1747 | mutex_enter(&dn->dn_mtx); | |
1748 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
1749 | mutex_exit(&dn->dn_mtx); | |
1750 | return; | |
1751 | } | |
1752 | ||
1753 | if (before) { | |
1754 | ASSERT(data); | |
1755 | user = &dn->dn_olduid; | |
1756 | group = &dn->dn_oldgid; | |
1757 | } else if (data) { | |
1758 | user = &dn->dn_newuid; | |
1759 | group = &dn->dn_newgid; | |
1760 | } | |
1761 | ||
1762 | /* | |
1763 | * Must always call the callback in case the object | |
1764 | * type has changed and that type isn't an object type to track | |
1765 | */ | |
1766 | error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, | |
1767 | user, group); | |
1768 | ||
1769 | /* | |
1770 | * Preserve existing uid/gid when the callback can't determine | |
1771 | * what the new uid/gid are and the callback returned EEXIST. | |
1772 | * The EEXIST error tells us to just use the existing uid/gid. | |
1773 | * If we don't know what the old values are then just assign | |
1774 | * them to 0, since that is a new file being created. | |
1775 | */ | |
1776 | if (!before && data == NULL && error == EEXIST) { | |
1777 | if (flags & DN_ID_OLD_EXIST) { | |
1778 | dn->dn_newuid = dn->dn_olduid; | |
1779 | dn->dn_newgid = dn->dn_oldgid; | |
1780 | } else { | |
1781 | dn->dn_newuid = 0; | |
1782 | dn->dn_newgid = 0; | |
1783 | } | |
1784 | error = 0; | |
1785 | } | |
1786 | ||
1787 | if (db) | |
1788 | mutex_exit(&db->db_mtx); | |
1789 | ||
1790 | mutex_enter(&dn->dn_mtx); | |
1791 | if (error == 0 && before) | |
1792 | dn->dn_id_flags |= DN_ID_OLD_EXIST; | |
1793 | if (error == 0 && !before) | |
1794 | dn->dn_id_flags |= DN_ID_NEW_EXIST; | |
1795 | ||
1796 | if (have_spill) { | |
1797 | dn->dn_id_flags |= DN_ID_CHKED_SPILL; | |
1798 | } else { | |
1799 | dn->dn_id_flags |= DN_ID_CHKED_BONUS; | |
1800 | } | |
1801 | mutex_exit(&dn->dn_mtx); | |
1802 | if (have_spill) | |
1803 | dmu_buf_rele((dmu_buf_t *)db, FTAG); | |
1804 | } | |
1805 | ||
9babb374 BB |
1806 | boolean_t |
1807 | dmu_objset_userspace_present(objset_t *os) | |
1808 | { | |
428870ff | 1809 | return (os->os_phys->os_flags & |
9babb374 BB |
1810 | OBJSET_FLAG_USERACCOUNTING_COMPLETE); |
1811 | } | |
1812 | ||
cae5b340 AX |
1813 | boolean_t |
1814 | dmu_objset_userobjspace_present(objset_t *os) | |
1815 | { | |
1816 | return (os->os_phys->os_flags & | |
1817 | OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); | |
1818 | } | |
1819 | ||
1820 | static int | |
1821 | dmu_objset_space_upgrade(objset_t *os) | |
9babb374 BB |
1822 | { |
1823 | uint64_t obj; | |
1824 | int err = 0; | |
1825 | ||
9babb374 BB |
1826 | /* |
1827 | * We simply need to mark every object dirty, so that it will be | |
1828 | * synced out and now accounted. If this is called | |
1829 | * concurrently, or if we already did some work before crashing, | |
1830 | * that's fine, since we track each object's accounted state | |
1831 | * independently. | |
1832 | */ | |
1833 | ||
1834 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { | |
45d1cae3 | 1835 | dmu_tx_t *tx; |
9babb374 BB |
1836 | dmu_buf_t *db; |
1837 | int objerr; | |
1838 | ||
cae5b340 AX |
1839 | mutex_enter(&os->os_upgrade_lock); |
1840 | if (os->os_upgrade_exit) | |
1841 | err = SET_ERROR(EINTR); | |
1842 | mutex_exit(&os->os_upgrade_lock); | |
1843 | if (err != 0) | |
1844 | return (err); | |
1845 | ||
9babb374 | 1846 | if (issig(JUSTLOOKING) && issig(FORREAL)) |
a08ee875 | 1847 | return (SET_ERROR(EINTR)); |
9babb374 BB |
1848 | |
1849 | objerr = dmu_bonus_hold(os, obj, FTAG, &db); | |
a08ee875 | 1850 | if (objerr != 0) |
9babb374 | 1851 | continue; |
45d1cae3 | 1852 | tx = dmu_tx_create(os); |
9babb374 BB |
1853 | dmu_tx_hold_bonus(tx, obj); |
1854 | objerr = dmu_tx_assign(tx, TXG_WAIT); | |
a08ee875 | 1855 | if (objerr != 0) { |
9babb374 BB |
1856 | dmu_tx_abort(tx); |
1857 | continue; | |
1858 | } | |
1859 | dmu_buf_will_dirty(db, tx); | |
1860 | dmu_buf_rele(db, FTAG); | |
1861 | dmu_tx_commit(tx); | |
1862 | } | |
cae5b340 AX |
1863 | return (0); |
1864 | } | |
1865 | ||
1866 | int | |
1867 | dmu_objset_userspace_upgrade(objset_t *os) | |
1868 | { | |
1869 | int err = 0; | |
1870 | ||
1871 | if (dmu_objset_userspace_present(os)) | |
1872 | return (0); | |
1873 | if (dmu_objset_is_snapshot(os)) | |
1874 | return (SET_ERROR(EINVAL)); | |
1875 | if (!dmu_objset_userused_enabled(os)) | |
1876 | return (SET_ERROR(ENOTSUP)); | |
1877 | ||
1878 | err = dmu_objset_space_upgrade(os); | |
1879 | if (err) | |
1880 | return (err); | |
9babb374 | 1881 | |
428870ff | 1882 | os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; |
9babb374 BB |
1883 | txg_wait_synced(dmu_objset_pool(os), 0); |
1884 | return (0); | |
1885 | } | |
1886 | ||
cae5b340 AX |
1887 | static int |
1888 | dmu_objset_userobjspace_upgrade_cb(objset_t *os) | |
1889 | { | |
1890 | int err = 0; | |
1891 | ||
1892 | if (dmu_objset_userobjspace_present(os)) | |
1893 | return (0); | |
1894 | if (dmu_objset_is_snapshot(os)) | |
1895 | return (SET_ERROR(EINVAL)); | |
1896 | if (!dmu_objset_userobjused_enabled(os)) | |
1897 | return (SET_ERROR(ENOTSUP)); | |
1898 | ||
1899 | dmu_objset_ds(os)->ds_feature_activation_needed[ | |
1900 | SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE; | |
1901 | ||
1902 | err = dmu_objset_space_upgrade(os); | |
1903 | if (err) | |
1904 | return (err); | |
1905 | ||
1906 | os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; | |
1907 | txg_wait_synced(dmu_objset_pool(os), 0); | |
1908 | return (0); | |
1909 | } | |
1910 | ||
1911 | void | |
1912 | dmu_objset_userobjspace_upgrade(objset_t *os) | |
1913 | { | |
1914 | dmu_objset_upgrade(os, dmu_objset_userobjspace_upgrade_cb); | |
1915 | } | |
1916 | ||
1917 | boolean_t | |
1918 | dmu_objset_userobjspace_upgradable(objset_t *os) | |
1919 | { | |
1920 | return (dmu_objset_type(os) == DMU_OST_ZFS && | |
1921 | !dmu_objset_is_snapshot(os) && | |
1922 | dmu_objset_userobjused_enabled(os) && | |
1923 | !dmu_objset_userobjspace_present(os)); | |
1924 | } | |
1925 | ||
34dc7c2f BB |
1926 | void |
1927 | dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, | |
1928 | uint64_t *usedobjsp, uint64_t *availobjsp) | |
1929 | { | |
428870ff | 1930 | dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, |
34dc7c2f BB |
1931 | usedobjsp, availobjsp); |
1932 | } | |
1933 | ||
1934 | uint64_t | |
1935 | dmu_objset_fsid_guid(objset_t *os) | |
1936 | { | |
428870ff | 1937 | return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); |
34dc7c2f BB |
1938 | } |
1939 | ||
1940 | void | |
1941 | dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) | |
1942 | { | |
428870ff BB |
1943 | stat->dds_type = os->os_phys->os_type; |
1944 | if (os->os_dsl_dataset) | |
1945 | dsl_dataset_fast_stat(os->os_dsl_dataset, stat); | |
34dc7c2f BB |
1946 | } |
1947 | ||
1948 | void | |
1949 | dmu_objset_stats(objset_t *os, nvlist_t *nv) | |
1950 | { | |
428870ff BB |
1951 | ASSERT(os->os_dsl_dataset || |
1952 | os->os_phys->os_type == DMU_OST_META); | |
34dc7c2f | 1953 | |
428870ff BB |
1954 | if (os->os_dsl_dataset != NULL) |
1955 | dsl_dataset_stats(os->os_dsl_dataset, nv); | |
34dc7c2f BB |
1956 | |
1957 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, | |
428870ff | 1958 | os->os_phys->os_type); |
9babb374 BB |
1959 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, |
1960 | dmu_objset_userspace_present(os)); | |
34dc7c2f BB |
1961 | } |
1962 | ||
1963 | int | |
1964 | dmu_objset_is_snapshot(objset_t *os) | |
1965 | { | |
428870ff | 1966 | if (os->os_dsl_dataset != NULL) |
e10b0808 | 1967 | return (os->os_dsl_dataset->ds_is_snapshot); |
34dc7c2f BB |
1968 | else |
1969 | return (B_FALSE); | |
1970 | } | |
1971 | ||
1972 | int | |
1973 | dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, | |
1974 | boolean_t *conflict) | |
1975 | { | |
428870ff | 1976 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
1977 | uint64_t ignored; |
1978 | ||
e10b0808 | 1979 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
a08ee875 | 1980 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1981 | |
1982 | return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, | |
e10b0808 | 1983 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, |
cae5b340 | 1984 | MT_NORMALIZE, real, maxlen, conflict)); |
34dc7c2f BB |
1985 | } |
1986 | ||
1987 | int | |
1988 | dmu_snapshot_list_next(objset_t *os, int namelen, char *name, | |
1989 | uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) | |
1990 | { | |
428870ff | 1991 | dsl_dataset_t *ds = os->os_dsl_dataset; |
34dc7c2f BB |
1992 | zap_cursor_t cursor; |
1993 | zap_attribute_t attr; | |
1994 | ||
a08ee875 LG |
1995 | ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); |
1996 | ||
e10b0808 | 1997 | if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) |
a08ee875 | 1998 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1999 | |
2000 | zap_cursor_init_serialized(&cursor, | |
2001 | ds->ds_dir->dd_pool->dp_meta_objset, | |
e10b0808 | 2002 | dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); |
34dc7c2f BB |
2003 | |
2004 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2005 | zap_cursor_fini(&cursor); | |
a08ee875 | 2006 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2007 | } |
2008 | ||
2009 | if (strlen(attr.za_name) + 1 > namelen) { | |
2010 | zap_cursor_fini(&cursor); | |
a08ee875 | 2011 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2012 | } |
2013 | ||
2014 | (void) strcpy(name, attr.za_name); | |
2015 | if (idp) | |
2016 | *idp = attr.za_first_integer; | |
2017 | if (case_conflict) | |
2018 | *case_conflict = attr.za_normalization_conflict; | |
2019 | zap_cursor_advance(&cursor); | |
2020 | *offp = zap_cursor_serialize(&cursor); | |
2021 | zap_cursor_fini(&cursor); | |
2022 | ||
2023 | return (0); | |
2024 | } | |
2025 | ||
ebe7e575 | 2026 | int |
6772fb67 | 2027 | dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) |
ebe7e575 | 2028 | { |
a08ee875 | 2029 | return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); |
ebe7e575 BB |
2030 | } |
2031 | ||
34dc7c2f BB |
2032 | int |
2033 | dmu_dir_list_next(objset_t *os, int namelen, char *name, | |
2034 | uint64_t *idp, uint64_t *offp) | |
2035 | { | |
428870ff | 2036 | dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; |
34dc7c2f BB |
2037 | zap_cursor_t cursor; |
2038 | zap_attribute_t attr; | |
2039 | ||
2040 | /* there is no next dir on a snapshot! */ | |
428870ff | 2041 | if (os->os_dsl_dataset->ds_object != |
e10b0808 | 2042 | dsl_dir_phys(dd)->dd_head_dataset_obj) |
a08ee875 | 2043 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2044 | |
2045 | zap_cursor_init_serialized(&cursor, | |
2046 | dd->dd_pool->dp_meta_objset, | |
e10b0808 | 2047 | dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); |
34dc7c2f BB |
2048 | |
2049 | if (zap_cursor_retrieve(&cursor, &attr) != 0) { | |
2050 | zap_cursor_fini(&cursor); | |
a08ee875 | 2051 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2052 | } |
2053 | ||
2054 | if (strlen(attr.za_name) + 1 > namelen) { | |
2055 | zap_cursor_fini(&cursor); | |
a08ee875 | 2056 | return (SET_ERROR(ENAMETOOLONG)); |
34dc7c2f BB |
2057 | } |
2058 | ||
2059 | (void) strcpy(name, attr.za_name); | |
2060 | if (idp) | |
2061 | *idp = attr.za_first_integer; | |
2062 | zap_cursor_advance(&cursor); | |
2063 | *offp = zap_cursor_serialize(&cursor); | |
2064 | zap_cursor_fini(&cursor); | |
2065 | ||
2066 | return (0); | |
2067 | } | |
2068 | ||
e10b0808 AX |
2069 | typedef struct dmu_objset_find_ctx { |
2070 | taskq_t *dc_tq; | |
2071 | dsl_pool_t *dc_dp; | |
2072 | uint64_t dc_ddobj; | |
cae5b340 | 2073 | char *dc_ddname; /* last component of ddobj's name */ |
e10b0808 AX |
2074 | int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); |
2075 | void *dc_arg; | |
2076 | int dc_flags; | |
2077 | kmutex_t *dc_error_lock; | |
2078 | int *dc_error; | |
2079 | } dmu_objset_find_ctx_t; | |
2080 | ||
2081 | static void | |
2082 | dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) | |
b128c09f | 2083 | { |
e10b0808 AX |
2084 | dsl_pool_t *dp = dcp->dc_dp; |
2085 | dmu_objset_find_ctx_t *child_dcp; | |
a08ee875 LG |
2086 | dsl_dir_t *dd; |
2087 | dsl_dataset_t *ds; | |
2088 | zap_cursor_t zc; | |
2089 | zap_attribute_t *attr; | |
2090 | uint64_t thisobj; | |
e10b0808 | 2091 | int err = 0; |
a08ee875 | 2092 | |
e10b0808 AX |
2093 | /* don't process if there already was an error */ |
2094 | if (*dcp->dc_error != 0) | |
2095 | goto out; | |
a08ee875 | 2096 | |
cae5b340 AX |
2097 | /* |
2098 | * Note: passing the name (dc_ddname) here is optional, but it | |
2099 | * improves performance because we don't need to call | |
2100 | * zap_value_search() to determine the name. | |
2101 | */ | |
2102 | err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); | |
a08ee875 | 2103 | if (err != 0) |
e10b0808 | 2104 | goto out; |
a08ee875 LG |
2105 | |
2106 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ | |
2107 | if (dd->dd_myname[0] == '$') { | |
2108 | dsl_dir_rele(dd, FTAG); | |
e10b0808 | 2109 | goto out; |
a08ee875 LG |
2110 | } |
2111 | ||
e10b0808 | 2112 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
ea04106b | 2113 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
a08ee875 LG |
2114 | |
2115 | /* | |
2116 | * Iterate over all children. | |
2117 | */ | |
e10b0808 | 2118 | if (dcp->dc_flags & DS_FIND_CHILDREN) { |
a08ee875 | 2119 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
e10b0808 | 2120 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
a08ee875 LG |
2121 | zap_cursor_retrieve(&zc, attr) == 0; |
2122 | (void) zap_cursor_advance(&zc)) { | |
2123 | ASSERT3U(attr->za_integer_length, ==, | |
2124 | sizeof (uint64_t)); | |
2125 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2126 | ||
cae5b340 AX |
2127 | child_dcp = |
2128 | kmem_alloc(sizeof (*child_dcp), KM_SLEEP); | |
e10b0808 AX |
2129 | *child_dcp = *dcp; |
2130 | child_dcp->dc_ddobj = attr->za_first_integer; | |
cae5b340 | 2131 | child_dcp->dc_ddname = spa_strdup(attr->za_name); |
e10b0808 AX |
2132 | if (dcp->dc_tq != NULL) |
2133 | (void) taskq_dispatch(dcp->dc_tq, | |
2134 | dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); | |
2135 | else | |
2136 | dmu_objset_find_dp_impl(child_dcp); | |
a08ee875 LG |
2137 | } |
2138 | zap_cursor_fini(&zc); | |
a08ee875 LG |
2139 | } |
2140 | ||
2141 | /* | |
2142 | * Iterate over all snapshots. | |
2143 | */ | |
e10b0808 | 2144 | if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { |
a08ee875 LG |
2145 | dsl_dataset_t *ds; |
2146 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
2147 | ||
2148 | if (err == 0) { | |
e10b0808 AX |
2149 | uint64_t snapobj; |
2150 | ||
2151 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
a08ee875 LG |
2152 | dsl_dataset_rele(ds, FTAG); |
2153 | ||
2154 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2155 | zap_cursor_retrieve(&zc, attr) == 0; | |
2156 | (void) zap_cursor_advance(&zc)) { | |
2157 | ASSERT3U(attr->za_integer_length, ==, | |
2158 | sizeof (uint64_t)); | |
2159 | ASSERT3U(attr->za_num_integers, ==, 1); | |
2160 | ||
2161 | err = dsl_dataset_hold_obj(dp, | |
2162 | attr->za_first_integer, FTAG, &ds); | |
2163 | if (err != 0) | |
2164 | break; | |
e10b0808 | 2165 | err = dcp->dc_func(dp, ds, dcp->dc_arg); |
a08ee875 LG |
2166 | dsl_dataset_rele(ds, FTAG); |
2167 | if (err != 0) | |
2168 | break; | |
2169 | } | |
2170 | zap_cursor_fini(&zc); | |
2171 | } | |
2172 | } | |
2173 | ||
a08ee875 LG |
2174 | kmem_free(attr, sizeof (zap_attribute_t)); |
2175 | ||
cae5b340 AX |
2176 | if (err != 0) { |
2177 | dsl_dir_rele(dd, FTAG); | |
e10b0808 | 2178 | goto out; |
cae5b340 | 2179 | } |
a08ee875 LG |
2180 | |
2181 | /* | |
2182 | * Apply to self. | |
2183 | */ | |
2184 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); | |
cae5b340 AX |
2185 | |
2186 | /* | |
2187 | * Note: we hold the dir while calling dsl_dataset_hold_obj() so | |
2188 | * that the dir will remain cached, and we won't have to re-instantiate | |
2189 | * it (which could be expensive due to finding its name via | |
2190 | * zap_value_search()). | |
2191 | */ | |
2192 | dsl_dir_rele(dd, FTAG); | |
a08ee875 | 2193 | if (err != 0) |
e10b0808 AX |
2194 | goto out; |
2195 | err = dcp->dc_func(dp, ds, dcp->dc_arg); | |
a08ee875 | 2196 | dsl_dataset_rele(ds, FTAG); |
e10b0808 AX |
2197 | |
2198 | out: | |
2199 | if (err != 0) { | |
2200 | mutex_enter(dcp->dc_error_lock); | |
2201 | /* only keep first error */ | |
2202 | if (*dcp->dc_error == 0) | |
2203 | *dcp->dc_error = err; | |
2204 | mutex_exit(dcp->dc_error_lock); | |
2205 | } | |
2206 | ||
cae5b340 AX |
2207 | if (dcp->dc_ddname != NULL) |
2208 | spa_strfree(dcp->dc_ddname); | |
e10b0808 AX |
2209 | kmem_free(dcp, sizeof (*dcp)); |
2210 | } | |
2211 | ||
2212 | static void | |
2213 | dmu_objset_find_dp_cb(void *arg) | |
2214 | { | |
2215 | dmu_objset_find_ctx_t *dcp = arg; | |
2216 | dsl_pool_t *dp = dcp->dc_dp; | |
2217 | ||
2218 | /* | |
2219 | * We need to get a pool_config_lock here, as there are several | |
2220 | * asssert(pool_config_held) down the stack. Getting a lock via | |
2221 | * dsl_pool_config_enter is risky, as it might be stalled by a | |
2222 | * pending writer. This would deadlock, as the write lock can | |
2223 | * only be granted when our parent thread gives up the lock. | |
2224 | * The _prio interface gives us priority over a pending writer. | |
2225 | */ | |
2226 | dsl_pool_config_enter_prio(dp, FTAG); | |
2227 | ||
2228 | dmu_objset_find_dp_impl(dcp); | |
2229 | ||
2230 | dsl_pool_config_exit(dp, FTAG); | |
2231 | } | |
2232 | ||
2233 | /* | |
2234 | * Find objsets under and including ddobj, call func(ds) on each. | |
2235 | * The order for the enumeration is completely undefined. | |
2236 | * func is called with dsl_pool_config held. | |
2237 | */ | |
2238 | int | |
2239 | dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, | |
2240 | int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) | |
2241 | { | |
2242 | int error = 0; | |
2243 | taskq_t *tq = NULL; | |
2244 | int ntasks; | |
2245 | dmu_objset_find_ctx_t *dcp; | |
2246 | kmutex_t err_lock; | |
2247 | ||
2248 | mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); | |
2249 | dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); | |
2250 | dcp->dc_tq = NULL; | |
2251 | dcp->dc_dp = dp; | |
2252 | dcp->dc_ddobj = ddobj; | |
cae5b340 | 2253 | dcp->dc_ddname = NULL; |
e10b0808 AX |
2254 | dcp->dc_func = func; |
2255 | dcp->dc_arg = arg; | |
2256 | dcp->dc_flags = flags; | |
2257 | dcp->dc_error_lock = &err_lock; | |
2258 | dcp->dc_error = &error; | |
2259 | ||
2260 | if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { | |
2261 | /* | |
2262 | * In case a write lock is held we can't make use of | |
2263 | * parallelism, as down the stack of the worker threads | |
2264 | * the lock is asserted via dsl_pool_config_held. | |
2265 | * In case of a read lock this is solved by getting a read | |
2266 | * lock in each worker thread, which isn't possible in case | |
2267 | * of a writer lock. So we fall back to the synchronous path | |
2268 | * here. | |
2269 | * In the future it might be possible to get some magic into | |
2270 | * dsl_pool_config_held in a way that it returns true for | |
2271 | * the worker threads so that a single lock held from this | |
2272 | * thread suffices. For now, stay single threaded. | |
2273 | */ | |
2274 | dmu_objset_find_dp_impl(dcp); | |
cae5b340 | 2275 | mutex_destroy(&err_lock); |
e10b0808 AX |
2276 | |
2277 | return (error); | |
2278 | } | |
2279 | ||
2280 | ntasks = dmu_find_threads; | |
2281 | if (ntasks == 0) | |
2282 | ntasks = vdev_count_leaves(dp->dp_spa) * 4; | |
2283 | tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, | |
2284 | INT_MAX, 0); | |
2285 | if (tq == NULL) { | |
2286 | kmem_free(dcp, sizeof (*dcp)); | |
cae5b340 AX |
2287 | mutex_destroy(&err_lock); |
2288 | ||
e10b0808 AX |
2289 | return (SET_ERROR(ENOMEM)); |
2290 | } | |
2291 | dcp->dc_tq = tq; | |
2292 | ||
2293 | /* dcp will be freed by task */ | |
2294 | (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); | |
2295 | ||
2296 | /* | |
2297 | * PORTING: this code relies on the property of taskq_wait to wait | |
2298 | * until no more tasks are queued and no more tasks are active. As | |
2299 | * we always queue new tasks from within other tasks, task_wait | |
2300 | * reliably waits for the full recursion to finish, even though we | |
2301 | * enqueue new tasks after taskq_wait has been called. | |
2302 | * On platforms other than illumos, taskq_wait may not have this | |
2303 | * property. | |
2304 | */ | |
2305 | taskq_wait(tq); | |
2306 | taskq_destroy(tq); | |
2307 | mutex_destroy(&err_lock); | |
2308 | ||
2309 | return (error); | |
b128c09f BB |
2310 | } |
2311 | ||
2312 | /* | |
a08ee875 LG |
2313 | * Find all objsets under name, and for each, call 'func(child_name, arg)'. |
2314 | * The dp_config_rwlock must not be held when this is called, and it | |
2315 | * will not be held when the callback is called. | |
2316 | * Therefore this function should only be used when the pool is not changing | |
2317 | * (e.g. in syncing context), or the callback can deal with the possible races. | |
b128c09f | 2318 | */ |
a08ee875 LG |
2319 | static int |
2320 | dmu_objset_find_impl(spa_t *spa, const char *name, | |
2321 | int func(const char *, void *), void *arg, int flags) | |
34dc7c2f BB |
2322 | { |
2323 | dsl_dir_t *dd; | |
a08ee875 | 2324 | dsl_pool_t *dp = spa_get_dsl(spa); |
b128c09f | 2325 | dsl_dataset_t *ds; |
34dc7c2f BB |
2326 | zap_cursor_t zc; |
2327 | zap_attribute_t *attr; | |
2328 | char *child; | |
b128c09f BB |
2329 | uint64_t thisobj; |
2330 | int err; | |
34dc7c2f | 2331 | |
a08ee875 LG |
2332 | dsl_pool_config_enter(dp, FTAG); |
2333 | ||
2334 | err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); | |
2335 | if (err != 0) { | |
2336 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f | 2337 | return (err); |
a08ee875 | 2338 | } |
34dc7c2f | 2339 | |
b128c09f BB |
2340 | /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ |
2341 | if (dd->dd_myname[0] == '$') { | |
a08ee875 LG |
2342 | dsl_dir_rele(dd, FTAG); |
2343 | dsl_pool_config_exit(dp, FTAG); | |
b128c09f BB |
2344 | return (0); |
2345 | } | |
2346 | ||
e10b0808 | 2347 | thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; |
ea04106b | 2348 | attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); |
34dc7c2f BB |
2349 | |
2350 | /* | |
2351 | * Iterate over all children. | |
2352 | */ | |
2353 | if (flags & DS_FIND_CHILDREN) { | |
b128c09f | 2354 | for (zap_cursor_init(&zc, dp->dp_meta_objset, |
e10b0808 | 2355 | dsl_dir_phys(dd)->dd_child_dir_zapobj); |
34dc7c2f BB |
2356 | zap_cursor_retrieve(&zc, attr) == 0; |
2357 | (void) zap_cursor_advance(&zc)) { | |
a08ee875 LG |
2358 | ASSERT3U(attr->za_integer_length, ==, |
2359 | sizeof (uint64_t)); | |
2360 | ASSERT3U(attr->za_num_integers, ==, 1); | |
34dc7c2f | 2361 | |
428870ff | 2362 | child = kmem_asprintf("%s/%s", name, attr->za_name); |
a08ee875 LG |
2363 | dsl_pool_config_exit(dp, FTAG); |
2364 | err = dmu_objset_find_impl(spa, child, | |
2365 | func, arg, flags); | |
2366 | dsl_pool_config_enter(dp, FTAG); | |
428870ff | 2367 | strfree(child); |
a08ee875 | 2368 | if (err != 0) |
34dc7c2f BB |
2369 | break; |
2370 | } | |
2371 | zap_cursor_fini(&zc); | |
2372 | ||
a08ee875 LG |
2373 | if (err != 0) { |
2374 | dsl_dir_rele(dd, FTAG); | |
2375 | dsl_pool_config_exit(dp, FTAG); | |
34dc7c2f BB |
2376 | kmem_free(attr, sizeof (zap_attribute_t)); |
2377 | return (err); | |
2378 | } | |
2379 | } | |
2380 | ||
2381 | /* | |
2382 | * Iterate over all snapshots. | |
2383 | */ | |
b128c09f | 2384 | if (flags & DS_FIND_SNAPSHOTS) { |
b128c09f | 2385 | err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); |
b128c09f BB |
2386 | |
2387 | if (err == 0) { | |
e10b0808 AX |
2388 | uint64_t snapobj; |
2389 | ||
2390 | snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; | |
b128c09f BB |
2391 | dsl_dataset_rele(ds, FTAG); |
2392 | ||
2393 | for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); | |
2394 | zap_cursor_retrieve(&zc, attr) == 0; | |
2395 | (void) zap_cursor_advance(&zc)) { | |
a08ee875 | 2396 | ASSERT3U(attr->za_integer_length, ==, |
b128c09f | 2397 | sizeof (uint64_t)); |
a08ee875 | 2398 | ASSERT3U(attr->za_num_integers, ==, 1); |
b128c09f | 2399 | |
428870ff BB |
2400 | child = kmem_asprintf("%s@%s", |
2401 | name, attr->za_name); | |
a08ee875 LG |
2402 | dsl_pool_config_exit(dp, FTAG); |
2403 | err = func(child, arg); | |
2404 | dsl_pool_config_enter(dp, FTAG); | |
428870ff | 2405 | strfree(child); |
a08ee875 | 2406 | if (err != 0) |
b128c09f BB |
2407 | break; |
2408 | } | |
2409 | zap_cursor_fini(&zc); | |
34dc7c2f | 2410 | } |
34dc7c2f BB |
2411 | } |
2412 | ||
a08ee875 | 2413 | dsl_dir_rele(dd, FTAG); |
34dc7c2f | 2414 | kmem_free(attr, sizeof (zap_attribute_t)); |
a08ee875 | 2415 | dsl_pool_config_exit(dp, FTAG); |
34dc7c2f | 2416 | |
a08ee875 | 2417 | if (err != 0) |
34dc7c2f BB |
2418 | return (err); |
2419 | ||
a08ee875 LG |
2420 | /* Apply to self. */ |
2421 | return (func(name, arg)); | |
34dc7c2f BB |
2422 | } |
2423 | ||
a08ee875 LG |
2424 | /* |
2425 | * See comment above dmu_objset_find_impl(). | |
2426 | */ | |
d164b209 | 2427 | int |
a08ee875 LG |
2428 | dmu_objset_find(char *name, int func(const char *, void *), void *arg, |
2429 | int flags) | |
d164b209 | 2430 | { |
a08ee875 LG |
2431 | spa_t *spa; |
2432 | int error; | |
d164b209 | 2433 | |
a08ee875 LG |
2434 | error = spa_open(name, &spa, FTAG); |
2435 | if (error != 0) | |
2436 | return (error); | |
2437 | error = dmu_objset_find_impl(spa, name, func, arg, flags); | |
2438 | spa_close(spa, FTAG); | |
2439 | return (error); | |
d164b209 BB |
2440 | } |
2441 | ||
34dc7c2f BB |
2442 | void |
2443 | dmu_objset_set_user(objset_t *os, void *user_ptr) | |
2444 | { | |
428870ff BB |
2445 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2446 | os->os_user_ptr = user_ptr; | |
34dc7c2f BB |
2447 | } |
2448 | ||
2449 | void * | |
2450 | dmu_objset_get_user(objset_t *os) | |
2451 | { | |
428870ff BB |
2452 | ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); |
2453 | return (os->os_user_ptr); | |
34dc7c2f | 2454 | } |
c28b2279 | 2455 | |
a08ee875 LG |
2456 | /* |
2457 | * Determine name of filesystem, given name of snapshot. | |
cae5b340 | 2458 | * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes |
a08ee875 LG |
2459 | */ |
2460 | int | |
2461 | dmu_fsname(const char *snapname, char *buf) | |
2462 | { | |
2463 | char *atp = strchr(snapname, '@'); | |
2464 | if (atp == NULL) | |
2465 | return (SET_ERROR(EINVAL)); | |
cae5b340 | 2466 | if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) |
a08ee875 LG |
2467 | return (SET_ERROR(ENAMETOOLONG)); |
2468 | (void) strlcpy(buf, snapname, atp - snapname + 1); | |
2469 | return (0); | |
2470 | } | |
2471 | ||
cae5b340 AX |
2472 | /* |
2473 | * Call when we think we're going to write/free space in open context to track | |
2474 | * the amount of dirty data in the open txg, which is also the amount | |
2475 | * of memory that can not be evicted until this txg syncs. | |
2476 | */ | |
2477 | void | |
2478 | dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) | |
2479 | { | |
2480 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
2481 | int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); | |
2482 | ||
2483 | if (ds != NULL) { | |
2484 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
2485 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
2486 | } | |
2487 | } | |
2488 | ||
c28b2279 | 2489 | #if defined(_KERNEL) && defined(HAVE_SPL) |
f0fd83be | 2490 | EXPORT_SYMBOL(dmu_objset_zil); |
c28b2279 | 2491 | EXPORT_SYMBOL(dmu_objset_pool); |
f0fd83be BB |
2492 | EXPORT_SYMBOL(dmu_objset_ds); |
2493 | EXPORT_SYMBOL(dmu_objset_type); | |
c28b2279 BB |
2494 | EXPORT_SYMBOL(dmu_objset_name); |
2495 | EXPORT_SYMBOL(dmu_objset_hold); | |
2496 | EXPORT_SYMBOL(dmu_objset_own); | |
2497 | EXPORT_SYMBOL(dmu_objset_rele); | |
2498 | EXPORT_SYMBOL(dmu_objset_disown); | |
2499 | EXPORT_SYMBOL(dmu_objset_from_ds); | |
2500 | EXPORT_SYMBOL(dmu_objset_create); | |
2501 | EXPORT_SYMBOL(dmu_objset_clone); | |
c28b2279 BB |
2502 | EXPORT_SYMBOL(dmu_objset_stats); |
2503 | EXPORT_SYMBOL(dmu_objset_fast_stat); | |
54a179e7 | 2504 | EXPORT_SYMBOL(dmu_objset_spa); |
c28b2279 BB |
2505 | EXPORT_SYMBOL(dmu_objset_space); |
2506 | EXPORT_SYMBOL(dmu_objset_fsid_guid); | |
2507 | EXPORT_SYMBOL(dmu_objset_find); | |
c28b2279 BB |
2508 | EXPORT_SYMBOL(dmu_objset_byteswap); |
2509 | EXPORT_SYMBOL(dmu_objset_evict_dbufs); | |
2510 | EXPORT_SYMBOL(dmu_objset_snap_cmtime); | |
cae5b340 | 2511 | EXPORT_SYMBOL(dmu_objset_dnodesize); |
c28b2279 BB |
2512 | |
2513 | EXPORT_SYMBOL(dmu_objset_sync); | |
2514 | EXPORT_SYMBOL(dmu_objset_is_dirty); | |
2515 | EXPORT_SYMBOL(dmu_objset_create_impl); | |
2516 | EXPORT_SYMBOL(dmu_objset_open_impl); | |
2517 | EXPORT_SYMBOL(dmu_objset_evict); | |
2518 | EXPORT_SYMBOL(dmu_objset_register_type); | |
2519 | EXPORT_SYMBOL(dmu_objset_do_userquota_updates); | |
2520 | EXPORT_SYMBOL(dmu_objset_userquota_get_ids); | |
2521 | EXPORT_SYMBOL(dmu_objset_userused_enabled); | |
2522 | EXPORT_SYMBOL(dmu_objset_userspace_upgrade); | |
2523 | EXPORT_SYMBOL(dmu_objset_userspace_present); | |
cae5b340 AX |
2524 | EXPORT_SYMBOL(dmu_objset_userobjused_enabled); |
2525 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgrade); | |
2526 | EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); | |
2527 | EXPORT_SYMBOL(dmu_objset_userobjspace_present); | |
c28b2279 | 2528 | #endif |