]>
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 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
196bee4c | 23 | * Copyright (c) 2011, 2020 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
bc77ba73 | 25 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
a08abc1b | 26 | * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved. |
5475aada | 27 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. |
65282ee9 | 28 | * Copyright (c) 2019 Datto Inc. |
10b3c7f5 MN |
29 | * Copyright (c) 2019, Klara Inc. |
30 | * Copyright (c) 2019, Allan Jude | |
34dc7c2f BB |
31 | */ |
32 | ||
34dc7c2f BB |
33 | #include <sys/dmu.h> |
34 | #include <sys/dmu_impl.h> | |
35 | #include <sys/dmu_tx.h> | |
36 | #include <sys/dbuf.h> | |
37 | #include <sys/dnode.h> | |
38 | #include <sys/zfs_context.h> | |
39 | #include <sys/dmu_objset.h> | |
40 | #include <sys/dmu_traverse.h> | |
41 | #include <sys/dsl_dataset.h> | |
42 | #include <sys/dsl_dir.h> | |
43 | #include <sys/dsl_pool.h> | |
44 | #include <sys/dsl_synctask.h> | |
45 | #include <sys/dsl_prop.h> | |
46 | #include <sys/dmu_zfetch.h> | |
47 | #include <sys/zfs_ioctl.h> | |
48 | #include <sys/zap.h> | |
49 | #include <sys/zio_checksum.h> | |
03c6040b | 50 | #include <sys/zio_compress.h> |
428870ff | 51 | #include <sys/sa.h> |
62bdd5eb | 52 | #include <sys/zfeature.h> |
a6255b7f | 53 | #include <sys/abd.h> |
e5d1c27e | 54 | #include <sys/trace_zfs.h> |
f763c3d1 | 55 | #include <sys/zfs_rlock.h> |
34dc7c2f BB |
56 | #ifdef _KERNEL |
57 | #include <sys/vmsystm.h> | |
b128c09f | 58 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
59 | #endif |
60 | ||
03c6040b GW |
61 | /* |
62 | * Enable/disable nopwrite feature. | |
63 | */ | |
64 | int zfs_nopwrite_enabled = 1; | |
65 | ||
539d33c7 | 66 | /* |
65282ee9 AP |
67 | * Tunable to control percentage of dirtied L1 blocks from frees allowed into |
68 | * one TXG. After this threshold is crossed, additional dirty blocks from frees | |
69 | * will wait until the next TXG. | |
539d33c7 GM |
70 | * A value of zero will disable this throttle. |
71 | */ | |
65282ee9 | 72 | unsigned long zfs_per_txg_dirty_frees_percent = 5; |
539d33c7 | 73 | |
66aca247 DB |
74 | /* |
75 | * Enable/disable forcing txg sync when dirty in dmu_offset_next. | |
76 | */ | |
77 | int zfs_dmu_offset_next_sync = 0; | |
78 | ||
d9b4bf06 MA |
79 | /* |
80 | * Limit the amount we can prefetch with one call to this amount. This | |
81 | * helps to limit the amount of memory that can be used by prefetching. | |
82 | * Larger objects should be prefetched a bit at a time. | |
83 | */ | |
84 | int dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE; | |
85 | ||
34dc7c2f | 86 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
2e5dc449 MA |
87 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" }, |
88 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" }, | |
89 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" }, | |
90 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" }, | |
91 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" }, | |
92 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" }, | |
93 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" }, | |
94 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" }, | |
95 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" }, | |
96 | {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" }, | |
97 | {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" }, | |
98 | {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" }, | |
99 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" }, | |
100 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"}, | |
101 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" }, | |
102 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" }, | |
103 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" }, | |
104 | {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" }, | |
105 | {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" }, | |
106 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" }, | |
107 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" }, | |
108 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" }, | |
109 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" }, | |
110 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" }, | |
111 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" }, | |
112 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" }, | |
113 | {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" }, | |
114 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" }, | |
115 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" }, | |
116 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" }, | |
117 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" }, | |
118 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" }, | |
119 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" }, | |
120 | {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" }, | |
121 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" }, | |
122 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" }, | |
123 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" }, | |
124 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"}, | |
125 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" }, | |
126 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" }, | |
127 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"}, | |
128 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"}, | |
129 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" }, | |
130 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" }, | |
131 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" }, | |
132 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" }, | |
133 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" }, | |
134 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" }, | |
135 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" }, | |
136 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" }, | |
137 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" }, | |
138 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" }, | |
139 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" }, | |
140 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
141 | }; |
142 | ||
143 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
144 | { byteswap_uint8_array, "uint8" }, | |
145 | { byteswap_uint16_array, "uint16" }, | |
146 | { byteswap_uint32_array, "uint32" }, | |
147 | { byteswap_uint64_array, "uint64" }, | |
148 | { zap_byteswap, "zap" }, | |
149 | { dnode_buf_byteswap, "dnode" }, | |
150 | { dmu_objset_byteswap, "objset" }, | |
151 | { zfs_znode_byteswap, "znode" }, | |
152 | { zfs_oldacl_byteswap, "oldacl" }, | |
153 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
154 | }; |
155 | ||
65c7cc49 | 156 | static int |
2bce8049 MA |
157 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, |
158 | void *tag, dmu_buf_t **dbp) | |
159 | { | |
160 | uint64_t blkid; | |
161 | dmu_buf_impl_t *db; | |
162 | ||
2bce8049 | 163 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 164 | blkid = dbuf_whichblock(dn, 0, offset); |
2bce8049 MA |
165 | db = dbuf_hold(dn, blkid, tag); |
166 | rw_exit(&dn->dn_struct_rwlock); | |
167 | ||
168 | if (db == NULL) { | |
169 | *dbp = NULL; | |
170 | return (SET_ERROR(EIO)); | |
171 | } | |
172 | ||
173 | *dbp = &db->db; | |
174 | return (0); | |
175 | } | |
34dc7c2f | 176 | int |
9b67f605 MA |
177 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
178 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
179 | { |
180 | dnode_t *dn; | |
181 | uint64_t blkid; | |
182 | dmu_buf_impl_t *db; | |
183 | int err; | |
428870ff BB |
184 | |
185 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
186 | if (err) |
187 | return (err); | |
34dc7c2f | 188 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 189 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
190 | db = dbuf_hold(dn, blkid, tag); |
191 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
192 | dnode_rele(dn, FTAG); |
193 | ||
34dc7c2f | 194 | if (db == NULL) { |
9b67f605 MA |
195 | *dbp = NULL; |
196 | return (SET_ERROR(EIO)); | |
197 | } | |
198 | ||
199 | *dbp = &db->db; | |
200 | return (err); | |
201 | } | |
202 | ||
2bce8049 MA |
203 | int |
204 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
205 | void *tag, dmu_buf_t **dbp, int flags) | |
206 | { | |
207 | int err; | |
208 | int db_flags = DB_RF_CANFAIL; | |
209 | ||
210 | if (flags & DMU_READ_NO_PREFETCH) | |
211 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
212 | if (flags & DMU_READ_NO_DECRYPT) |
213 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
214 | |
215 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
216 | if (err == 0) { | |
217 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
218 | err = dbuf_read(db, NULL, db_flags); | |
219 | if (err != 0) { | |
220 | dbuf_rele(db, tag); | |
221 | *dbp = NULL; | |
222 | } | |
223 | } | |
224 | ||
225 | return (err); | |
226 | } | |
227 | ||
9b67f605 MA |
228 | int |
229 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
230 | void *tag, dmu_buf_t **dbp, int flags) | |
231 | { | |
232 | int err; | |
233 | int db_flags = DB_RF_CANFAIL; | |
234 | ||
235 | if (flags & DMU_READ_NO_PREFETCH) | |
236 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
237 | if (flags & DMU_READ_NO_DECRYPT) |
238 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
239 | |
240 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
241 | if (err == 0) { | |
242 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 243 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 244 | if (err != 0) { |
34dc7c2f | 245 | dbuf_rele(db, tag); |
9b67f605 | 246 | *dbp = NULL; |
34dc7c2f BB |
247 | } |
248 | } | |
249 | ||
34dc7c2f BB |
250 | return (err); |
251 | } | |
252 | ||
253 | int | |
254 | dmu_bonus_max(void) | |
255 | { | |
50c957f7 | 256 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
257 | } |
258 | ||
259 | int | |
572e2857 | 260 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 261 | { |
572e2857 BB |
262 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
263 | dnode_t *dn; | |
264 | int error; | |
34dc7c2f | 265 | |
572e2857 BB |
266 | DB_DNODE_ENTER(db); |
267 | dn = DB_DNODE(db); | |
268 | ||
269 | if (dn->dn_bonus != db) { | |
2e528b49 | 270 | error = SET_ERROR(EINVAL); |
572e2857 | 271 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 272 | error = SET_ERROR(EINVAL); |
572e2857 BB |
273 | } else { |
274 | dnode_setbonuslen(dn, newsize, tx); | |
275 | error = 0; | |
276 | } | |
277 | ||
278 | DB_DNODE_EXIT(db); | |
279 | return (error); | |
34dc7c2f BB |
280 | } |
281 | ||
428870ff | 282 | int |
572e2857 | 283 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 284 | { |
572e2857 BB |
285 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
286 | dnode_t *dn; | |
287 | int error; | |
428870ff | 288 | |
572e2857 BB |
289 | DB_DNODE_ENTER(db); |
290 | dn = DB_DNODE(db); | |
428870ff | 291 | |
9ae529ec | 292 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 293 | error = SET_ERROR(EINVAL); |
572e2857 | 294 | } else if (dn->dn_bonus != db) { |
2e528b49 | 295 | error = SET_ERROR(EINVAL); |
572e2857 BB |
296 | } else { |
297 | dnode_setbonus_type(dn, type, tx); | |
298 | error = 0; | |
299 | } | |
428870ff | 300 | |
572e2857 BB |
301 | DB_DNODE_EXIT(db); |
302 | return (error); | |
303 | } | |
304 | ||
305 | dmu_object_type_t | |
306 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
307 | { | |
308 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
309 | dnode_t *dn; | |
310 | dmu_object_type_t type; | |
311 | ||
312 | DB_DNODE_ENTER(db); | |
313 | dn = DB_DNODE(db); | |
314 | type = dn->dn_bonustype; | |
315 | DB_DNODE_EXIT(db); | |
316 | ||
317 | return (type); | |
428870ff BB |
318 | } |
319 | ||
320 | int | |
321 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
322 | { | |
323 | dnode_t *dn; | |
324 | int error; | |
325 | ||
326 | error = dnode_hold(os, object, FTAG, &dn); | |
327 | dbuf_rm_spill(dn, tx); | |
328 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
329 | dnode_rm_spill(dn, tx); | |
330 | rw_exit(&dn->dn_struct_rwlock); | |
331 | dnode_rele(dn, FTAG); | |
332 | return (error); | |
333 | } | |
334 | ||
34dc7c2f | 335 | /* |
6955b401 BB |
336 | * Lookup and hold the bonus buffer for the provided dnode. If the dnode |
337 | * has not yet been allocated a new bonus dbuf a will be allocated. | |
338 | * Returns ENOENT, EIO, or 0. | |
34dc7c2f | 339 | */ |
6955b401 BB |
340 | int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp, |
341 | uint32_t flags) | |
34dc7c2f | 342 | { |
34dc7c2f BB |
343 | dmu_buf_impl_t *db; |
344 | int error; | |
b5256303 TC |
345 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
346 | ||
347 | if (flags & DMU_READ_NO_PREFETCH) | |
348 | db_flags |= DB_RF_NOPREFETCH; | |
349 | if (flags & DMU_READ_NO_DECRYPT) | |
350 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 351 | |
34dc7c2f BB |
352 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
353 | if (dn->dn_bonus == NULL) { | |
354 | rw_exit(&dn->dn_struct_rwlock); | |
355 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
356 | if (dn->dn_bonus == NULL) | |
357 | dbuf_create_bonus(dn); | |
358 | } | |
359 | db = dn->dn_bonus; | |
34dc7c2f BB |
360 | |
361 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 362 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 363 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 364 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
365 | } |
366 | ||
367 | /* | |
368 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
369 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
370 | * a dnode hold for every dbuf. | |
371 | */ | |
372 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f | 373 | |
b5256303 TC |
374 | error = dbuf_read(db, NULL, db_flags); |
375 | if (error) { | |
376 | dnode_evict_bonus(dn); | |
377 | dbuf_rele(db, tag); | |
378 | *dbp = NULL; | |
379 | return (error); | |
380 | } | |
34dc7c2f BB |
381 | |
382 | *dbp = &db->db; | |
383 | return (0); | |
384 | } | |
385 | ||
b5256303 | 386 | int |
6955b401 | 387 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) |
b5256303 | 388 | { |
6955b401 BB |
389 | dnode_t *dn; |
390 | int error; | |
391 | ||
392 | error = dnode_hold(os, object, FTAG, &dn); | |
393 | if (error) | |
394 | return (error); | |
395 | ||
396 | error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH); | |
397 | dnode_rele(dn, FTAG); | |
398 | ||
399 | return (error); | |
b5256303 TC |
400 | } |
401 | ||
428870ff BB |
402 | /* |
403 | * returns ENOENT, EIO, or 0. | |
404 | * | |
405 | * This interface will allocate a blank spill dbuf when a spill blk | |
406 | * doesn't already exist on the dnode. | |
407 | * | |
408 | * if you only want to find an already existing spill db, then | |
409 | * dmu_spill_hold_existing() should be used. | |
410 | */ | |
411 | int | |
412 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
413 | { | |
414 | dmu_buf_impl_t *db = NULL; | |
415 | int err; | |
416 | ||
417 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
418 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
419 | ||
420 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
421 | ||
422 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
423 | rw_exit(&dn->dn_struct_rwlock); | |
424 | ||
b182ac00 | 425 | if (db == NULL) { |
426 | *dbp = NULL; | |
427 | return (SET_ERROR(EIO)); | |
428 | } | |
572e2857 BB |
429 | err = dbuf_read(db, NULL, flags); |
430 | if (err == 0) | |
431 | *dbp = &db->db; | |
b182ac00 | 432 | else { |
572e2857 | 433 | dbuf_rele(db, tag); |
b182ac00 | 434 | *dbp = NULL; |
435 | } | |
428870ff BB |
436 | return (err); |
437 | } | |
438 | ||
439 | int | |
440 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
441 | { | |
572e2857 BB |
442 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
443 | dnode_t *dn; | |
428870ff BB |
444 | int err; |
445 | ||
572e2857 BB |
446 | DB_DNODE_ENTER(db); |
447 | dn = DB_DNODE(db); | |
448 | ||
449 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 450 | err = SET_ERROR(EINVAL); |
572e2857 BB |
451 | } else { |
452 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
453 | ||
454 | if (!dn->dn_have_spill) { | |
2e528b49 | 455 | err = SET_ERROR(ENOENT); |
572e2857 BB |
456 | } else { |
457 | err = dmu_spill_hold_by_dnode(dn, | |
458 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
459 | } | |
428870ff | 460 | |
428870ff | 461 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 462 | } |
572e2857 BB |
463 | |
464 | DB_DNODE_EXIT(db); | |
428870ff BB |
465 | return (err); |
466 | } | |
467 | ||
468 | int | |
e7504d7a TC |
469 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag, |
470 | dmu_buf_t **dbp) | |
428870ff | 471 | { |
572e2857 BB |
472 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
473 | dnode_t *dn; | |
474 | int err; | |
e7504d7a TC |
475 | uint32_t db_flags = DB_RF_CANFAIL; |
476 | ||
477 | if (flags & DMU_READ_NO_DECRYPT) | |
478 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
479 | |
480 | DB_DNODE_ENTER(db); | |
481 | dn = DB_DNODE(db); | |
e7504d7a | 482 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
483 | DB_DNODE_EXIT(db); |
484 | ||
485 | return (err); | |
428870ff BB |
486 | } |
487 | ||
34dc7c2f BB |
488 | /* |
489 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
490 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
491 | * and can induce severe lock contention when writing to several files | |
492 | * whose dnodes are in the same block. | |
493 | */ | |
af1698f5 | 494 | int |
9babb374 | 495 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
7f60329a | 496 | boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
34dc7c2f BB |
497 | { |
498 | dmu_buf_t **dbp; | |
499 | uint64_t blkid, nblks, i; | |
9babb374 | 500 | uint32_t dbuf_flags; |
34dc7c2f | 501 | int err; |
ec50cd24 | 502 | zio_t *zio = NULL; |
34dc7c2f BB |
503 | |
504 | ASSERT(length <= DMU_MAX_ACCESS); | |
505 | ||
7f60329a MA |
506 | /* |
507 | * Note: We directly notify the prefetch code of this read, so that | |
508 | * we can tell it about the multi-block read. dbuf_read() only knows | |
509 | * about the one block it is accessing. | |
510 | */ | |
511 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
512 | DB_RF_NOPREFETCH; | |
34dc7c2f BB |
513 | |
514 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
515 | if (dn->dn_datablkshift) { | |
516 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
517 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
518 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
519 | } else { |
520 | if (offset + length > dn->dn_datablksz) { | |
521 | zfs_panic_recover("zfs: accessing past end of object " | |
522 | "%llx/%llx (size=%u access=%llu+%llu)", | |
523 | (longlong_t)dn->dn_objset-> | |
524 | os_dsl_dataset->ds_object, | |
525 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
526 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 527 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 528 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
529 | } |
530 | nblks = 1; | |
531 | } | |
79c76d5b | 532 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 533 | |
ec50cd24 FY |
534 | if (read) |
535 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, | |
536 | ZIO_FLAG_CANFAIL); | |
fcff0f35 | 537 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f | 538 | for (i = 0; i < nblks; i++) { |
7f60329a | 539 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f BB |
540 | if (db == NULL) { |
541 | rw_exit(&dn->dn_struct_rwlock); | |
542 | dmu_buf_rele_array(dbp, nblks, tag); | |
ec50cd24 FY |
543 | if (read) |
544 | zio_nowait(zio); | |
2e528b49 | 545 | return (SET_ERROR(EIO)); |
34dc7c2f | 546 | } |
7f60329a | 547 | |
34dc7c2f | 548 | /* initiate async i/o */ |
7f60329a | 549 | if (read) |
9babb374 | 550 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
551 | dbp[i] = &db->db; |
552 | } | |
7f60329a | 553 | |
755065f3 AM |
554 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
555 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
556 | dmu_zfetch(&dn->dn_zfetch, blkid, nblks, | |
f664f1ee | 557 | read && DNODE_IS_CACHEABLE(dn), B_TRUE); |
7f60329a | 558 | } |
34dc7c2f BB |
559 | rw_exit(&dn->dn_struct_rwlock); |
560 | ||
34dc7c2f | 561 | if (read) { |
ec50cd24 FY |
562 | /* wait for async read i/o */ |
563 | err = zio_wait(zio); | |
564 | if (err) { | |
565 | dmu_buf_rele_array(dbp, nblks, tag); | |
566 | return (err); | |
567 | } | |
568 | ||
569 | /* wait for other io to complete */ | |
34dc7c2f BB |
570 | for (i = 0; i < nblks; i++) { |
571 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
572 | mutex_enter(&db->db_mtx); | |
573 | while (db->db_state == DB_READ || | |
574 | db->db_state == DB_FILL) | |
575 | cv_wait(&db->db_changed, &db->db_mtx); | |
576 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 577 | err = SET_ERROR(EIO); |
34dc7c2f BB |
578 | mutex_exit(&db->db_mtx); |
579 | if (err) { | |
580 | dmu_buf_rele_array(dbp, nblks, tag); | |
581 | return (err); | |
582 | } | |
583 | } | |
584 | } | |
585 | ||
586 | *numbufsp = nblks; | |
587 | *dbpp = dbp; | |
588 | return (0); | |
589 | } | |
590 | ||
591 | static int | |
592 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
593 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
594 | { | |
595 | dnode_t *dn; | |
596 | int err; | |
597 | ||
428870ff | 598 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
599 | if (err) |
600 | return (err); | |
601 | ||
602 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 603 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
604 | |
605 | dnode_rele(dn, FTAG); | |
606 | ||
607 | return (err); | |
608 | } | |
609 | ||
610 | int | |
572e2857 | 611 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
7f60329a MA |
612 | uint64_t length, boolean_t read, void *tag, int *numbufsp, |
613 | dmu_buf_t ***dbpp) | |
34dc7c2f | 614 | { |
572e2857 BB |
615 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
616 | dnode_t *dn; | |
34dc7c2f BB |
617 | int err; |
618 | ||
572e2857 BB |
619 | DB_DNODE_ENTER(db); |
620 | dn = DB_DNODE(db); | |
34dc7c2f | 621 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 622 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 623 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
624 | |
625 | return (err); | |
626 | } | |
627 | ||
628 | void | |
629 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
630 | { | |
631 | int i; | |
632 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
633 | ||
634 | if (numbufs == 0) | |
635 | return; | |
636 | ||
637 | for (i = 0; i < numbufs; i++) { | |
638 | if (dbp[i]) | |
639 | dbuf_rele(dbp[i], tag); | |
640 | } | |
641 | ||
642 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
643 | } | |
644 | ||
e8b96c60 | 645 | /* |
fcff0f35 | 646 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
e1cfd73f | 647 | * indirect blocks prefetched will be those that point to the blocks containing |
fcff0f35 | 648 | * the data starting at offset, and continuing to offset + len. |
e8b96c60 | 649 | * |
b5256303 | 650 | * Note that if the indirect blocks above the blocks being prefetched are not |
e1cfd73f | 651 | * in cache, they will be asynchronously read in. |
e8b96c60 | 652 | */ |
34dc7c2f | 653 | void |
fcff0f35 PD |
654 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
655 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
656 | { |
657 | dnode_t *dn; | |
658 | uint64_t blkid; | |
e8b96c60 | 659 | int nblks, err; |
34dc7c2f | 660 | |
34dc7c2f | 661 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 662 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
663 | |
664 | if (object == 0 || object >= DN_MAX_OBJECT) | |
665 | return; | |
666 | ||
667 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
668 | blkid = dbuf_whichblock(dn, level, |
669 | object * sizeof (dnode_phys_t)); | |
670 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
671 | rw_exit(&dn->dn_struct_rwlock); |
672 | return; | |
673 | } | |
674 | ||
d9b4bf06 MA |
675 | /* |
676 | * See comment before the definition of dmu_prefetch_max. | |
677 | */ | |
678 | len = MIN(len, dmu_prefetch_max); | |
679 | ||
34dc7c2f BB |
680 | /* |
681 | * XXX - Note, if the dnode for the requested object is not | |
682 | * already cached, we will do a *synchronous* read in the | |
683 | * dnode_hold() call. The same is true for any indirects. | |
684 | */ | |
428870ff | 685 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
686 | if (err != 0) |
687 | return; | |
688 | ||
fcff0f35 PD |
689 | /* |
690 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
691 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
692 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
693 | * offset) is the first. Then the number we need to prefetch is the | |
694 | * last - first + 1. | |
695 | */ | |
f664f1ee | 696 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
fcff0f35 PD |
697 | if (level > 0 || dn->dn_datablkshift != 0) { |
698 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
699 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
700 | } else { |
701 | nblks = (offset < dn->dn_datablksz); | |
702 | } | |
703 | ||
704 | if (nblks != 0) { | |
fcff0f35 | 705 | blkid = dbuf_whichblock(dn, level, offset); |
1c27024e | 706 | for (int i = 0; i < nblks; i++) |
fcff0f35 | 707 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f | 708 | } |
34dc7c2f BB |
709 | rw_exit(&dn->dn_struct_rwlock); |
710 | ||
711 | dnode_rele(dn, FTAG); | |
712 | } | |
713 | ||
45d1cae3 BB |
714 | /* |
715 | * Get the next "chunk" of file data to free. We traverse the file from | |
716 | * the end so that the file gets shorter over time (if we crashes in the | |
717 | * middle, this will leave us in a better state). We find allocated file | |
718 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
719 | * |
720 | * On input, *start should be the first offset that does not need to be | |
721 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
722 | * offset that should be freed and l1blks is set to the number of level 1 |
723 | * indirect blocks found within the chunk. | |
45d1cae3 | 724 | */ |
b128c09f | 725 | static int |
65282ee9 | 726 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 727 | { |
65282ee9 | 728 | uint64_t blks; |
b663a23d MA |
729 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
730 | /* bytes of data covered by a level-1 indirect block */ | |
ec4afd27 OM |
731 | uint64_t iblkrange = (uint64_t)dn->dn_datablksz * |
732 | EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
b128c09f | 733 | |
b663a23d | 734 | ASSERT3U(minimum, <=, *start); |
b128c09f | 735 | |
f4c594da TC |
736 | /* |
737 | * Check if we can free the entire range assuming that all of the | |
738 | * L1 blocks in this range have data. If we can, we use this | |
739 | * worst case value as an estimate so we can avoid having to look | |
740 | * at the object's actual data. | |
741 | */ | |
742 | uint64_t total_l1blks = | |
743 | (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / | |
744 | iblkrange; | |
745 | if (total_l1blks <= maxblks) { | |
746 | *l1blks = total_l1blks; | |
b663a23d | 747 | *start = minimum; |
b128c09f BB |
748 | return (0); |
749 | } | |
45d1cae3 | 750 | ASSERT(ISP2(iblkrange)); |
b128c09f | 751 | |
65282ee9 | 752 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
753 | int err; |
754 | ||
b663a23d MA |
755 | /* |
756 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
757 | * indirect block at or before the input offset. We must | |
758 | * decrement *start so that it is at the end of the region | |
759 | * to search. | |
760 | */ | |
761 | (*start)--; | |
f4c594da | 762 | |
b128c09f | 763 | err = dnode_next_offset(dn, |
45d1cae3 | 764 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 765 | |
b663a23d | 766 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 767 | if (err == ESRCH) { |
b663a23d MA |
768 | *start = minimum; |
769 | break; | |
770 | } else if (err != 0) { | |
65282ee9 | 771 | *l1blks = blks; |
b128c09f | 772 | return (err); |
45d1cae3 | 773 | } |
b128c09f | 774 | |
b663a23d | 775 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 776 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 777 | } |
b663a23d MA |
778 | if (*start < minimum) |
779 | *start = minimum; | |
65282ee9 | 780 | *l1blks = blks; |
f4c594da | 781 | |
b128c09f BB |
782 | return (0); |
783 | } | |
784 | ||
a08abc1b GM |
785 | /* |
786 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
787 | * otherwise return false. | |
788 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
789 | */ | |
790 | /*ARGSUSED*/ | |
791 | static boolean_t | |
792 | dmu_objset_zfs_unmounting(objset_t *os) | |
793 | { | |
794 | #ifdef _KERNEL | |
795 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
796 | return (zfs_get_vfs_flag_unmounted(os)); | |
797 | #endif | |
798 | return (B_FALSE); | |
799 | } | |
800 | ||
b128c09f BB |
801 | static int |
802 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 803 | uint64_t length) |
b128c09f | 804 | { |
c97d3069 | 805 | uint64_t object_size; |
b663a23d | 806 | int err; |
539d33c7 GM |
807 | uint64_t dirty_frees_threshold; |
808 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 809 | |
c97d3069 BB |
810 | if (dn == NULL) |
811 | return (SET_ERROR(EINVAL)); | |
812 | ||
813 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 814 | if (offset >= object_size) |
b128c09f | 815 | return (0); |
b128c09f | 816 | |
539d33c7 GM |
817 | if (zfs_per_txg_dirty_frees_percent <= 100) |
818 | dirty_frees_threshold = | |
819 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
820 | else | |
65282ee9 | 821 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 822 | |
b663a23d MA |
823 | if (length == DMU_OBJECT_END || offset + length > object_size) |
824 | length = object_size - offset; | |
825 | ||
826 | while (length != 0) { | |
539d33c7 | 827 | uint64_t chunk_end, chunk_begin, chunk_len; |
65282ee9 | 828 | uint64_t l1blks; |
b663a23d MA |
829 | dmu_tx_t *tx; |
830 | ||
a08abc1b GM |
831 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
832 | return (SET_ERROR(EINTR)); | |
833 | ||
b663a23d MA |
834 | chunk_end = chunk_begin = offset + length; |
835 | ||
836 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 837 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
838 | if (err) |
839 | return (err); | |
b663a23d MA |
840 | ASSERT3U(chunk_begin, >=, offset); |
841 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 842 | |
539d33c7 GM |
843 | chunk_len = chunk_end - chunk_begin; |
844 | ||
b128c09f | 845 | tx = dmu_tx_create(os); |
539d33c7 | 846 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
847 | |
848 | /* | |
849 | * Mark this transaction as typically resulting in a net | |
850 | * reduction in space used. | |
851 | */ | |
852 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
853 | err = dmu_tx_assign(tx, TXG_WAIT); |
854 | if (err) { | |
855 | dmu_tx_abort(tx); | |
856 | return (err); | |
857 | } | |
539d33c7 | 858 | |
f4c594da TC |
859 | uint64_t txg = dmu_tx_get_txg(tx); |
860 | ||
861 | mutex_enter(&dp->dp_lock); | |
862 | uint64_t long_free_dirty = | |
863 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; | |
864 | mutex_exit(&dp->dp_lock); | |
865 | ||
866 | /* | |
867 | * To avoid filling up a TXG with just frees, wait for | |
868 | * the next TXG to open before freeing more chunks if | |
869 | * we have reached the threshold of frees. | |
870 | */ | |
871 | if (dirty_frees_threshold != 0 && | |
872 | long_free_dirty >= dirty_frees_threshold) { | |
873 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); | |
874 | dmu_tx_commit(tx); | |
875 | txg_wait_open(dp, 0, B_TRUE); | |
876 | continue; | |
877 | } | |
878 | ||
65282ee9 AP |
879 | /* |
880 | * In order to prevent unnecessary write throttling, for each | |
881 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
882 | * in dnode_free_range() below. We compare this number to a | |
883 | * tunable threshold, past which we prevent new L1 dirty freeing | |
884 | * blocks from being added into the open TXG. See | |
885 | * dmu_free_long_range_impl() for details. The threshold | |
886 | * prevents write throttle activation due to dirty freeing L1 | |
887 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
888 | */ | |
539d33c7 | 889 | mutex_enter(&dp->dp_lock); |
f4c594da | 890 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += |
65282ee9 | 891 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
892 | mutex_exit(&dp->dp_lock); |
893 | DTRACE_PROBE3(free__long__range, | |
f4c594da TC |
894 | uint64_t, long_free_dirty, uint64_t, chunk_len, |
895 | uint64_t, txg); | |
539d33c7 | 896 | dnode_free_range(dn, chunk_begin, chunk_len, tx); |
440a3eb9 | 897 | |
b128c09f | 898 | dmu_tx_commit(tx); |
b663a23d | 899 | |
539d33c7 | 900 | length -= chunk_len; |
b128c09f BB |
901 | } |
902 | return (0); | |
903 | } | |
904 | ||
905 | int | |
906 | dmu_free_long_range(objset_t *os, uint64_t object, | |
907 | uint64_t offset, uint64_t length) | |
908 | { | |
909 | dnode_t *dn; | |
910 | int err; | |
911 | ||
428870ff | 912 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
913 | if (err != 0) |
914 | return (err); | |
0c03d21a | 915 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
916 | |
917 | /* | |
918 | * It is important to zero out the maxblkid when freeing the entire | |
919 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
920 | * will take the fast path, and (b) dnode_reallocate() can verify | |
921 | * that the entire file has been freed. | |
922 | */ | |
b0bc7a84 | 923 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
924 | dn->dn_maxblkid = 0; |
925 | ||
b128c09f BB |
926 | dnode_rele(dn, FTAG); |
927 | return (err); | |
928 | } | |
929 | ||
930 | int | |
0c03d21a | 931 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 932 | { |
b128c09f BB |
933 | dmu_tx_t *tx; |
934 | int err; | |
935 | ||
b663a23d | 936 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
937 | if (err != 0) |
938 | return (err); | |
b663a23d MA |
939 | |
940 | tx = dmu_tx_create(os); | |
941 | dmu_tx_hold_bonus(tx, object); | |
942 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 943 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
944 | err = dmu_tx_assign(tx, TXG_WAIT); |
945 | if (err == 0) { | |
35df0bb5 TC |
946 | if (err == 0) |
947 | err = dmu_object_free(os, object, tx); | |
440a3eb9 | 948 | |
b663a23d | 949 | dmu_tx_commit(tx); |
b128c09f | 950 | } else { |
b663a23d | 951 | dmu_tx_abort(tx); |
b128c09f | 952 | } |
b663a23d | 953 | |
b128c09f BB |
954 | return (err); |
955 | } | |
956 | ||
34dc7c2f BB |
957 | int |
958 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
959 | uint64_t size, dmu_tx_t *tx) | |
960 | { | |
961 | dnode_t *dn; | |
428870ff | 962 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
963 | if (err) |
964 | return (err); | |
965 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 966 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
967 | dnode_free_range(dn, offset, size, tx); |
968 | dnode_rele(dn, FTAG); | |
969 | return (0); | |
970 | } | |
971 | ||
0eef1bde | 972 | static int |
973 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 974 | void *buf, uint32_t flags) |
34dc7c2f | 975 | { |
34dc7c2f | 976 | dmu_buf_t **dbp; |
0eef1bde | 977 | int numbufs, err = 0; |
34dc7c2f BB |
978 | |
979 | /* | |
980 | * Deal with odd block sizes, where there can't be data past the first | |
981 | * block. If we ever do the tail block optimization, we will need to | |
982 | * handle that here as well. | |
983 | */ | |
45d1cae3 | 984 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 985 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
986 | MIN(size, dn->dn_datablksz - offset); |
987 | bzero((char *)buf + newsz, size - newsz); | |
988 | size = newsz; | |
989 | } | |
990 | ||
991 | while (size > 0) { | |
992 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 993 | int i; |
34dc7c2f BB |
994 | |
995 | /* | |
996 | * NB: we could do this block-at-a-time, but it's nice | |
997 | * to be reading in parallel. | |
998 | */ | |
999 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 1000 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
1001 | if (err) |
1002 | break; | |
1003 | ||
1004 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1005 | uint64_t tocpy; |
1006 | int64_t bufoff; | |
34dc7c2f BB |
1007 | dmu_buf_t *db = dbp[i]; |
1008 | ||
1009 | ASSERT(size > 0); | |
1010 | ||
1011 | bufoff = offset - db->db_offset; | |
c9520ecc | 1012 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 1013 | |
c9520ecc | 1014 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
1015 | |
1016 | offset += tocpy; | |
1017 | size -= tocpy; | |
1018 | buf = (char *)buf + tocpy; | |
1019 | } | |
1020 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1021 | } | |
34dc7c2f BB |
1022 | return (err); |
1023 | } | |
1024 | ||
0eef1bde | 1025 | int |
1026 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1027 | void *buf, uint32_t flags) | |
34dc7c2f | 1028 | { |
0eef1bde | 1029 | dnode_t *dn; |
1030 | int err; | |
34dc7c2f | 1031 | |
0eef1bde | 1032 | err = dnode_hold(os, object, FTAG, &dn); |
1033 | if (err != 0) | |
1034 | return (err); | |
34dc7c2f | 1035 | |
0eef1bde | 1036 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1037 | dnode_rele(dn, FTAG); | |
1038 | return (err); | |
1039 | } | |
1040 | ||
1041 | int | |
1042 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1043 | uint32_t flags) | |
1044 | { | |
1045 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1046 | } | |
1047 | ||
1048 | static void | |
1049 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1050 | const void *buf, dmu_tx_t *tx) | |
1051 | { | |
1052 | int i; | |
34dc7c2f BB |
1053 | |
1054 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1055 | uint64_t tocpy; |
1056 | int64_t bufoff; | |
34dc7c2f BB |
1057 | dmu_buf_t *db = dbp[i]; |
1058 | ||
1059 | ASSERT(size > 0); | |
1060 | ||
1061 | bufoff = offset - db->db_offset; | |
c9520ecc | 1062 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1063 | |
1064 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1065 | ||
1066 | if (tocpy == db->db_size) | |
1067 | dmu_buf_will_fill(db, tx); | |
1068 | else | |
1069 | dmu_buf_will_dirty(db, tx); | |
1070 | ||
60101509 | 1071 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1072 | |
1073 | if (tocpy == db->db_size) | |
1074 | dmu_buf_fill_done(db, tx); | |
1075 | ||
1076 | offset += tocpy; | |
1077 | size -= tocpy; | |
1078 | buf = (char *)buf + tocpy; | |
1079 | } | |
0eef1bde | 1080 | } |
1081 | ||
1082 | void | |
1083 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1084 | const void *buf, dmu_tx_t *tx) | |
1085 | { | |
1086 | dmu_buf_t **dbp; | |
1087 | int numbufs; | |
1088 | ||
1089 | if (size == 0) | |
1090 | return; | |
1091 | ||
1092 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1093 | FALSE, FTAG, &numbufs, &dbp)); | |
1094 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1095 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1096 | } | |
1097 | ||
0f8ff49e SD |
1098 | /* |
1099 | * Note: Lustre is an external consumer of this interface. | |
1100 | */ | |
0eef1bde | 1101 | void |
1102 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1103 | const void *buf, dmu_tx_t *tx) | |
1104 | { | |
1105 | dmu_buf_t **dbp; | |
1106 | int numbufs; | |
1107 | ||
1108 | if (size == 0) | |
1109 | return; | |
1110 | ||
1111 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1112 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1113 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1114 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1115 | } | |
1116 | ||
b128c09f BB |
1117 | void |
1118 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1119 | dmu_tx_t *tx) | |
1120 | { | |
1121 | dmu_buf_t **dbp; | |
1122 | int numbufs, i; | |
1123 | ||
1124 | if (size == 0) | |
1125 | return; | |
1126 | ||
1127 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1128 | FALSE, FTAG, &numbufs, &dbp)); | |
1129 | ||
1130 | for (i = 0; i < numbufs; i++) { | |
1131 | dmu_buf_t *db = dbp[i]; | |
1132 | ||
1133 | dmu_buf_will_not_fill(db, tx); | |
1134 | } | |
1135 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1136 | } | |
1137 | ||
9b67f605 MA |
1138 | void |
1139 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1140 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1141 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1142 | { | |
1143 | dmu_buf_t *db; | |
1144 | ||
1145 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1146 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1147 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1148 | FTAG, &db)); | |
1149 | ||
1150 | dmu_buf_write_embedded(db, | |
1151 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1152 | uncompressed_size, compressed_size, byteorder, tx); | |
1153 | ||
1154 | dmu_buf_rele(db, FTAG); | |
1155 | } | |
1156 | ||
30af21b0 PD |
1157 | void |
1158 | dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1159 | dmu_tx_t *tx) | |
1160 | { | |
1161 | int numbufs, i; | |
1162 | dmu_buf_t **dbp; | |
1163 | ||
1164 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1165 | &numbufs, &dbp)); | |
1166 | for (i = 0; i < numbufs; i++) | |
1167 | dmu_buf_redact(dbp[i], tx); | |
1168 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1169 | } | |
1170 | ||
34dc7c2f | 1171 | #ifdef _KERNEL |
5228cf01 | 1172 | int |
804e0504 | 1173 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) |
872e8d26 BB |
1174 | { |
1175 | dmu_buf_t **dbp; | |
1176 | int numbufs, i, err; | |
872e8d26 BB |
1177 | |
1178 | /* | |
1179 | * NB: we could do this block-at-a-time, but it's nice | |
1180 | * to be reading in parallel. | |
1181 | */ | |
883a40ff | 1182 | err = dmu_buf_hold_array_by_dnode(dn, uio_offset(uio), size, |
804e0504 | 1183 | TRUE, FTAG, &numbufs, &dbp, 0); |
872e8d26 BB |
1184 | if (err) |
1185 | return (err); | |
1186 | ||
1187 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1188 | uint64_t tocpy; |
1189 | int64_t bufoff; | |
872e8d26 BB |
1190 | dmu_buf_t *db = dbp[i]; |
1191 | ||
1192 | ASSERT(size > 0); | |
1193 | ||
883a40ff | 1194 | bufoff = uio_offset(uio) - db->db_offset; |
c9520ecc | 1195 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1196 | |
01c4f2bf MM |
1197 | #ifdef __FreeBSD__ |
1198 | err = vn_io_fault_uiomove((char *)db->db_data + bufoff, | |
1199 | tocpy, uio); | |
1200 | #else | |
872e8d26 BB |
1201 | err = uiomove((char *)db->db_data + bufoff, tocpy, |
1202 | UIO_READ, uio); | |
01c4f2bf | 1203 | #endif |
872e8d26 BB |
1204 | if (err) |
1205 | break; | |
1206 | ||
1207 | size -= tocpy; | |
1208 | } | |
1209 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1210 | ||
1211 | return (err); | |
1212 | } | |
1213 | ||
804e0504 MA |
1214 | /* |
1215 | * Read 'size' bytes into the uio buffer. | |
1216 | * From object zdb->db_object. | |
1217 | * Starting at offset uio->uio_loffset. | |
1218 | * | |
1219 | * If the caller already has a dbuf in the target object | |
1220 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1221 | * because we don't have to find the dnode_t for the object. | |
1222 | */ | |
1223 | int | |
1224 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1225 | { | |
1226 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1227 | dnode_t *dn; | |
1228 | int err; | |
1229 | ||
1230 | if (size == 0) | |
1231 | return (0); | |
1232 | ||
1233 | DB_DNODE_ENTER(db); | |
1234 | dn = DB_DNODE(db); | |
1235 | err = dmu_read_uio_dnode(dn, uio, size); | |
1236 | DB_DNODE_EXIT(db); | |
1237 | ||
1238 | return (err); | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * Read 'size' bytes into the uio buffer. | |
1243 | * From the specified object | |
1244 | * Starting at offset uio->uio_loffset. | |
1245 | */ | |
1246 | int | |
1247 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1248 | { | |
1249 | dnode_t *dn; | |
1250 | int err; | |
1251 | ||
1252 | if (size == 0) | |
1253 | return (0); | |
1254 | ||
1255 | err = dnode_hold(os, object, FTAG, &dn); | |
1256 | if (err) | |
1257 | return (err); | |
1258 | ||
1259 | err = dmu_read_uio_dnode(dn, uio, size); | |
1260 | ||
1261 | dnode_rele(dn, FTAG); | |
1262 | ||
1263 | return (err); | |
1264 | } | |
1265 | ||
5228cf01 | 1266 | int |
872e8d26 BB |
1267 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) |
1268 | { | |
1269 | dmu_buf_t **dbp; | |
1270 | int numbufs; | |
1271 | int err = 0; | |
1272 | int i; | |
1273 | ||
883a40ff | 1274 | err = dmu_buf_hold_array_by_dnode(dn, uio_offset(uio), size, |
872e8d26 BB |
1275 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); |
1276 | if (err) | |
1277 | return (err); | |
1278 | ||
1279 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1280 | uint64_t tocpy; |
1281 | int64_t bufoff; | |
872e8d26 BB |
1282 | dmu_buf_t *db = dbp[i]; |
1283 | ||
1284 | ASSERT(size > 0); | |
1285 | ||
883a40ff | 1286 | bufoff = uio_offset(uio) - db->db_offset; |
c9520ecc | 1287 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1288 | |
1289 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1290 | ||
1291 | if (tocpy == db->db_size) | |
1292 | dmu_buf_will_fill(db, tx); | |
1293 | else | |
1294 | dmu_buf_will_dirty(db, tx); | |
1295 | ||
1296 | /* | |
1297 | * XXX uiomove could block forever (eg.nfs-backed | |
1298 | * pages). There needs to be a uiolockdown() function | |
1299 | * to lock the pages in memory, so that uiomove won't | |
1300 | * block. | |
1301 | */ | |
01c4f2bf MM |
1302 | #ifdef __FreeBSD__ |
1303 | err = vn_io_fault_uiomove((char *)db->db_data + bufoff, | |
1304 | tocpy, uio); | |
1305 | #else | |
872e8d26 BB |
1306 | err = uiomove((char *)db->db_data + bufoff, tocpy, |
1307 | UIO_WRITE, uio); | |
01c4f2bf | 1308 | #endif |
872e8d26 BB |
1309 | if (tocpy == db->db_size) |
1310 | dmu_buf_fill_done(db, tx); | |
1311 | ||
1312 | if (err) | |
1313 | break; | |
1314 | ||
1315 | size -= tocpy; | |
1316 | } | |
1317 | ||
1318 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1319 | return (err); | |
1320 | } | |
1321 | ||
804e0504 MA |
1322 | /* |
1323 | * Write 'size' bytes from the uio buffer. | |
1324 | * To object zdb->db_object. | |
1325 | * Starting at offset uio->uio_loffset. | |
1326 | * | |
1327 | * If the caller already has a dbuf in the target object | |
1328 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1329 | * because we don't have to find the dnode_t for the object. | |
1330 | */ | |
428870ff BB |
1331 | int |
1332 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1333 | dmu_tx_t *tx) | |
1334 | { | |
572e2857 BB |
1335 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1336 | dnode_t *dn; | |
1337 | int err; | |
1338 | ||
428870ff BB |
1339 | if (size == 0) |
1340 | return (0); | |
1341 | ||
572e2857 BB |
1342 | DB_DNODE_ENTER(db); |
1343 | dn = DB_DNODE(db); | |
1344 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1345 | DB_DNODE_EXIT(db); | |
1346 | ||
1347 | return (err); | |
428870ff BB |
1348 | } |
1349 | ||
804e0504 MA |
1350 | /* |
1351 | * Write 'size' bytes from the uio buffer. | |
1352 | * To the specified object. | |
1353 | * Starting at offset uio->uio_loffset. | |
1354 | */ | |
428870ff BB |
1355 | int |
1356 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1357 | dmu_tx_t *tx) | |
1358 | { | |
1359 | dnode_t *dn; | |
1360 | int err; | |
1361 | ||
1362 | if (size == 0) | |
1363 | return (0); | |
1364 | ||
1365 | err = dnode_hold(os, object, FTAG, &dn); | |
1366 | if (err) | |
1367 | return (err); | |
1368 | ||
1369 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1370 | ||
1371 | dnode_rele(dn, FTAG); | |
1372 | ||
1373 | return (err); | |
1374 | } | |
872e8d26 | 1375 | #endif /* _KERNEL */ |
34dc7c2f | 1376 | |
9babb374 BB |
1377 | /* |
1378 | * Allocate a loaned anonymous arc buffer. | |
1379 | */ | |
1380 | arc_buf_t * | |
1381 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1382 | { | |
572e2857 | 1383 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1384 | |
2aa34383 | 1385 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1386 | } |
1387 | ||
1388 | /* | |
1389 | * Free a loaned arc buffer. | |
1390 | */ | |
1391 | void | |
1392 | dmu_return_arcbuf(arc_buf_t *buf) | |
1393 | { | |
1394 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1395 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1396 | } |
1397 | ||
ba67d821 MA |
1398 | /* |
1399 | * A "lightweight" write is faster than a regular write (e.g. | |
1400 | * dmu_write_by_dnode() or dmu_assign_arcbuf_by_dnode()), because it avoids the | |
1401 | * CPU cost of creating a dmu_buf_impl_t and arc_buf_[hdr_]_t. However, the | |
1402 | * data can not be read or overwritten until the transaction's txg has been | |
1403 | * synced. This makes it appropriate for workloads that are known to be | |
1404 | * (temporarily) write-only, like "zfs receive". | |
1405 | * | |
1406 | * A single block is written, starting at the specified offset in bytes. If | |
1407 | * the call is successful, it returns 0 and the provided abd has been | |
1408 | * consumed (the caller should not free it). | |
1409 | */ | |
1410 | int | |
1411 | dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, | |
1412 | const zio_prop_t *zp, enum zio_flag flags, dmu_tx_t *tx) | |
1413 | { | |
1414 | dbuf_dirty_record_t *dr = | |
1415 | dbuf_dirty_lightweight(dn, dbuf_whichblock(dn, 0, offset), tx); | |
1416 | if (dr == NULL) | |
1417 | return (SET_ERROR(EIO)); | |
1418 | dr->dt.dll.dr_abd = abd; | |
1419 | dr->dt.dll.dr_props = *zp; | |
1420 | dr->dt.dll.dr_flags = flags; | |
1421 | return (0); | |
1422 | } | |
1423 | ||
9babb374 BB |
1424 | /* |
1425 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1426 | * If this is not possible copy the contents of passed arc buf via | |
1427 | * dmu_write(). | |
1428 | */ | |
305781da | 1429 | int |
440a3eb9 | 1430 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1431 | dmu_tx_t *tx) |
1432 | { | |
9babb374 | 1433 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1434 | objset_t *os = dn->dn_objset; |
1435 | uint64_t object = dn->dn_object; | |
2aa34383 | 1436 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1437 | uint64_t blkid; |
1438 | ||
1439 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1440 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1441 | db = dbuf_hold(dn, blkid, FTAG); |
1442 | if (db == NULL) | |
1443 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1444 | rw_exit(&dn->dn_struct_rwlock); |
1445 | ||
88904bb3 | 1446 | /* |
ba67d821 MA |
1447 | * We can only assign if the offset is aligned and the arc buf is the |
1448 | * same size as the dbuf. | |
88904bb3 | 1449 | */ |
2aa34383 | 1450 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
9babb374 BB |
1451 | dbuf_assign_arcbuf(db, buf, tx); |
1452 | dbuf_rele(db, FTAG); | |
1453 | } else { | |
2aa34383 DK |
1454 | /* compressed bufs must always be assignable to their dbuf */ |
1455 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1456 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1457 | |
9babb374 | 1458 | dbuf_rele(db, FTAG); |
572e2857 | 1459 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 BB |
1460 | dmu_return_arcbuf(buf); |
1461 | } | |
305781da TC |
1462 | |
1463 | return (0); | |
9babb374 BB |
1464 | } |
1465 | ||
305781da | 1466 | int |
440a3eb9 TC |
1467 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1468 | dmu_tx_t *tx) | |
1469 | { | |
305781da | 1470 | int err; |
440a3eb9 TC |
1471 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1472 | ||
1473 | DB_DNODE_ENTER(dbuf); | |
305781da | 1474 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1475 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1476 | |
1477 | return (err); | |
440a3eb9 TC |
1478 | } |
1479 | ||
34dc7c2f | 1480 | typedef struct { |
428870ff BB |
1481 | dbuf_dirty_record_t *dsa_dr; |
1482 | dmu_sync_cb_t *dsa_done; | |
1483 | zgd_t *dsa_zgd; | |
1484 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1485 | } dmu_sync_arg_t; |
1486 | ||
b128c09f BB |
1487 | /* ARGSUSED */ |
1488 | static void | |
1489 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1490 | { | |
428870ff BB |
1491 | dmu_sync_arg_t *dsa = varg; |
1492 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1493 | blkptr_t *bp = zio->io_bp; |
1494 | ||
428870ff BB |
1495 | if (zio->io_error == 0) { |
1496 | if (BP_IS_HOLE(bp)) { | |
1497 | /* | |
1498 | * A block of zeros may compress to a hole, but the | |
1499 | * block size still needs to be known for replay. | |
1500 | */ | |
1501 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1502 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1503 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1504 | BP_SET_FILL(bp, 1); |
428870ff | 1505 | } |
b128c09f BB |
1506 | } |
1507 | } | |
1508 | ||
428870ff BB |
1509 | static void |
1510 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1511 | { | |
1512 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1513 | } | |
1514 | ||
34dc7c2f BB |
1515 | /* ARGSUSED */ |
1516 | static void | |
1517 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1518 | { | |
428870ff BB |
1519 | dmu_sync_arg_t *dsa = varg; |
1520 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1521 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1522 | zgd_t *zgd = dsa->dsa_zgd; |
1523 | ||
1524 | /* | |
1525 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1526 | * the writes for the lwb have completed. | |
1527 | */ | |
1528 | if (zio->io_error == 0) { | |
1529 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1530 | } | |
34dc7c2f | 1531 | |
34dc7c2f BB |
1532 | mutex_enter(&db->db_mtx); |
1533 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1534 | if (zio->io_error == 0) { |
03c6040b GW |
1535 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1536 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1537 | blkptr_t *bp = zio->io_bp; |
1538 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1539 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1540 | |
1541 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1542 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1543 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1544 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1545 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1546 | } |
428870ff BB |
1547 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1548 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1549 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1550 | |
1551 | /* | |
1552 | * Old style holes are filled with all zeros, whereas | |
1553 | * new-style holes maintain their lsize, type, level, | |
1554 | * and birth time (see zio_write_compress). While we | |
1555 | * need to reset the BP_SET_LSIZE() call that happened | |
1556 | * in dmu_sync_ready for old style holes, we do *not* | |
1557 | * want to wipe out the information contained in new | |
1558 | * style holes. Thus, only zero out the block pointer if | |
1559 | * it's an old style hole. | |
1560 | */ | |
1561 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1562 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1563 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1564 | } else { | |
1565 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1566 | } | |
34dc7c2f BB |
1567 | cv_broadcast(&db->db_changed); |
1568 | mutex_exit(&db->db_mtx); | |
1569 | ||
428870ff | 1570 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1571 | |
428870ff BB |
1572 | kmem_free(dsa, sizeof (*dsa)); |
1573 | } | |
1574 | ||
1575 | static void | |
1576 | dmu_sync_late_arrival_done(zio_t *zio) | |
1577 | { | |
1578 | blkptr_t *bp = zio->io_bp; | |
1579 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1580 | zgd_t *zgd = dsa->dsa_zgd; |
1581 | ||
1582 | if (zio->io_error == 0) { | |
1583 | /* | |
1584 | * Record the vdev(s) backing this blkptr so they can be | |
1585 | * flushed after the writes for the lwb have completed. | |
1586 | */ | |
1587 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1588 | ||
1589 | if (!BP_IS_HOLE(bp)) { | |
2a8ba608 | 1590 | blkptr_t *bp_orig __maybe_unused = &zio->io_bp_orig; |
900d09b2 PS |
1591 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); |
1592 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1593 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1594 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1595 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1596 | } | |
428870ff BB |
1597 | } |
1598 | ||
1599 | dmu_tx_commit(dsa->dsa_tx); | |
1600 | ||
1601 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1602 | ||
a6255b7f | 1603 | abd_put(zio->io_abd); |
428870ff BB |
1604 | kmem_free(dsa, sizeof (*dsa)); |
1605 | } | |
1606 | ||
1607 | static int | |
1608 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1609 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1610 | { |
1611 | dmu_sync_arg_t *dsa; | |
1612 | dmu_tx_t *tx; | |
1613 | ||
1614 | tx = dmu_tx_create(os); | |
1615 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1616 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1617 | dmu_tx_abort(tx); | |
2e528b49 MA |
1618 | /* Make zl_get_data do txg_waited_synced() */ |
1619 | return (SET_ERROR(EIO)); | |
428870ff BB |
1620 | } |
1621 | ||
1ce23dca PS |
1622 | /* |
1623 | * In order to prevent the zgd's lwb from being free'd prior to | |
1624 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1625 | * the lwb's "max txg" takes this tx's txg into account. | |
1626 | */ | |
1627 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1628 | ||
79c76d5b | 1629 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1630 | dsa->dsa_dr = NULL; |
1631 | dsa->dsa_done = done; | |
1632 | dsa->dsa_zgd = zgd; | |
1633 | dsa->dsa_tx = tx; | |
1634 | ||
02dc43bc MA |
1635 | /* |
1636 | * Since we are currently syncing this txg, it's nontrivial to | |
1637 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1638 | * | |
1639 | * When syncing, the db_blkptr is initially the BP of the previous | |
1640 | * txg. We can not nopwrite against it because it will be changed | |
1641 | * (this is similar to the non-late-arrival case where the dbuf is | |
1642 | * dirty in a future txg). | |
1643 | * | |
1644 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1645 | * We can not nopwrite against it because although the BP will not | |
1646 | * (typically) be changed, the data has not yet been persisted to this | |
1647 | * location. | |
1648 | * | |
1649 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1650 | * possible to always nopwrite, because the data that was written in | |
1651 | * this txg is the same data that we are trying to write. However we | |
1652 | * would need to check that this dbuf is not dirty in any future | |
1653 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1654 | * don't nopwrite in this case. | |
1655 | */ | |
1656 | zp->zp_nopwrite = B_FALSE; | |
1657 | ||
a6255b7f DQ |
1658 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1659 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1660 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1661 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1662 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1663 | |
1664 | return (0); | |
34dc7c2f BB |
1665 | } |
1666 | ||
1667 | /* | |
1668 | * Intent log support: sync the block associated with db to disk. | |
1669 | * N.B. and XXX: the caller is responsible for making sure that the | |
1670 | * data isn't changing while dmu_sync() is writing it. | |
1671 | * | |
1672 | * Return values: | |
1673 | * | |
03c6040b | 1674 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1675 | * The caller should not log the write. |
1676 | * | |
1677 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1678 | * The caller should not log the write. | |
1679 | * | |
1680 | * EALREADY: this block is already in the process of being synced. | |
1681 | * The caller should track its progress (somehow). | |
1682 | * | |
428870ff BB |
1683 | * EIO: could not do the I/O. |
1684 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1685 | * |
428870ff BB |
1686 | * 0: the I/O has been initiated. |
1687 | * The caller should log this blkptr in the done callback. | |
1688 | * It is possible that the I/O will fail, in which case | |
1689 | * the error will be reported to the done callback and | |
1690 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1691 | */ |
1692 | int | |
428870ff | 1693 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1694 | { |
428870ff BB |
1695 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1696 | objset_t *os = db->db_objset; | |
1697 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
cccbed9f | 1698 | dbuf_dirty_record_t *dr, *dr_next; |
428870ff | 1699 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1700 | zbookmark_phys_t zb; |
428870ff | 1701 | zio_prop_t zp; |
572e2857 | 1702 | dnode_t *dn; |
34dc7c2f | 1703 | |
428870ff | 1704 | ASSERT(pio != NULL); |
34dc7c2f BB |
1705 | ASSERT(txg != 0); |
1706 | ||
428870ff BB |
1707 | SET_BOOKMARK(&zb, ds->ds_object, |
1708 | db->db.db_object, db->db_level, db->db_blkid); | |
1709 | ||
572e2857 BB |
1710 | DB_DNODE_ENTER(db); |
1711 | dn = DB_DNODE(db); | |
82644107 | 1712 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1713 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1714 | |
1715 | /* | |
428870ff | 1716 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1717 | */ |
428870ff BB |
1718 | if (txg > spa_freeze_txg(os->os_spa)) |
1719 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1720 | |
1721 | /* | |
428870ff BB |
1722 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1723 | * and us. If we determine that this txg is not yet syncing, | |
1724 | * but it begins to sync a moment later, that's OK because the | |
1725 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1726 | */ |
428870ff BB |
1727 | mutex_enter(&db->db_mtx); |
1728 | ||
1729 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1730 | /* |
428870ff | 1731 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1732 | */ |
428870ff | 1733 | mutex_exit(&db->db_mtx); |
2e528b49 | 1734 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1735 | } |
1736 | ||
428870ff BB |
1737 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1738 | /* | |
1739 | * This txg is currently syncing, so we can't mess with | |
1740 | * the dirty record anymore; just write a new log block. | |
1741 | */ | |
1742 | mutex_exit(&db->db_mtx); | |
1743 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1744 | } |
1745 | ||
cccbed9f | 1746 | dr = dbuf_find_dirty_eq(db, txg); |
428870ff BB |
1747 | |
1748 | if (dr == NULL) { | |
34dc7c2f | 1749 | /* |
428870ff | 1750 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1751 | * There's no need to log writes to freed blocks, so we're done. |
1752 | */ | |
1753 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1754 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1755 | } |
1756 | ||
cccbed9f MM |
1757 | dr_next = list_next(&db->db_dirty_records, dr); |
1758 | ASSERT(dr_next == NULL || dr_next->dr_txg < txg); | |
03c6040b | 1759 | |
02dc43bc MA |
1760 | if (db->db_blkptr != NULL) { |
1761 | /* | |
1762 | * We need to fill in zgd_bp with the current blkptr so that | |
1763 | * the nopwrite code can check if we're writing the same | |
1764 | * data that's already on disk. We can only nopwrite if we | |
1765 | * are sure that after making the copy, db_blkptr will not | |
1766 | * change until our i/o completes. We ensure this by | |
1767 | * holding the db_mtx, and only allowing nopwrite if the | |
1768 | * block is not already dirty (see below). This is verified | |
1769 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1770 | * not changed. | |
1771 | */ | |
1772 | *zgd->zgd_bp = *db->db_blkptr; | |
1773 | } | |
1774 | ||
03c6040b | 1775 | /* |
f3c517d8 MA |
1776 | * Assume the on-disk data is X, the current syncing data (in |
1777 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1778 | * in dmu_sync). | |
1779 | * | |
1780 | * We usually want to perform a nopwrite if X and Z are the | |
1781 | * same. However, if Y is different (i.e. the BP is going to | |
1782 | * change before this write takes effect), then a nopwrite will | |
1783 | * be incorrect - we would override with X, which could have | |
1784 | * been freed when Y was written. | |
1785 | * | |
1786 | * (Note that this is not a concern when we are nop-writing from | |
1787 | * syncing context, because X and Y must be identical, because | |
1788 | * all previous txgs have been synced.) | |
1789 | * | |
1790 | * Therefore, we disable nopwrite if the current BP could change | |
1791 | * before this TXG. There are two ways it could change: by | |
1792 | * being dirty (dr_next is non-NULL), or by being freed | |
1793 | * (dnode_block_freed()). This behavior is verified by | |
1794 | * zio_done(), which VERIFYs that the override BP is identical | |
1795 | * to the on-disk BP. | |
03c6040b | 1796 | */ |
f3c517d8 MA |
1797 | DB_DNODE_ENTER(db); |
1798 | dn = DB_DNODE(db); | |
cccbed9f | 1799 | if (dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) |
03c6040b | 1800 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1801 | DB_DNODE_EXIT(db); |
03c6040b | 1802 | |
34dc7c2f | 1803 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1804 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1805 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1806 | /* |
428870ff BB |
1807 | * We have already issued a sync write for this buffer, |
1808 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1809 | * have been dirtied since, or we would have cleared the state. |
1810 | */ | |
34dc7c2f | 1811 | mutex_exit(&db->db_mtx); |
2e528b49 | 1812 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1813 | } |
1814 | ||
428870ff | 1815 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1816 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1817 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1818 | |
79c76d5b | 1819 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1820 | dsa->dsa_dr = dr; |
1821 | dsa->dsa_done = done; | |
1822 | dsa->dsa_zgd = zgd; | |
1823 | dsa->dsa_tx = NULL; | |
b128c09f | 1824 | |
428870ff | 1825 | zio_nowait(arc_write(pio, os->os_spa, txg, |
02dc43bc | 1826 | zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 1827 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 1828 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 1829 | |
428870ff | 1830 | return (0); |
34dc7c2f BB |
1831 | } |
1832 | ||
b5256303 TC |
1833 | int |
1834 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
1835 | { | |
1836 | dnode_t *dn; | |
1837 | int err; | |
1838 | ||
1839 | err = dnode_hold(os, object, FTAG, &dn); | |
1840 | if (err) | |
1841 | return (err); | |
1842 | err = dnode_set_nlevels(dn, nlevels, tx); | |
1843 | dnode_rele(dn, FTAG); | |
1844 | return (err); | |
1845 | } | |
1846 | ||
34dc7c2f BB |
1847 | int |
1848 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 1849 | dmu_tx_t *tx) |
34dc7c2f BB |
1850 | { |
1851 | dnode_t *dn; | |
1852 | int err; | |
1853 | ||
428870ff | 1854 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1855 | if (err) |
1856 | return (err); | |
1857 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1858 | dnode_rele(dn, FTAG); | |
1859 | return (err); | |
1860 | } | |
1861 | ||
ae76f45c TC |
1862 | int |
1863 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
1864 | dmu_tx_t *tx) | |
1865 | { | |
1866 | dnode_t *dn; | |
1867 | int err; | |
1868 | ||
1869 | err = dnode_hold(os, object, FTAG, &dn); | |
1870 | if (err) | |
1871 | return (err); | |
1872 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 1873 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
1874 | rw_exit(&dn->dn_struct_rwlock); |
1875 | dnode_rele(dn, FTAG); | |
1876 | return (0); | |
1877 | } | |
1878 | ||
34dc7c2f BB |
1879 | void |
1880 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 1881 | dmu_tx_t *tx) |
34dc7c2f BB |
1882 | { |
1883 | dnode_t *dn; | |
1884 | ||
9b67f605 MA |
1885 | /* |
1886 | * Send streams include each object's checksum function. This | |
1887 | * check ensures that the receiving system can understand the | |
1888 | * checksum function transmitted. | |
1889 | */ | |
1890 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1891 | ||
1892 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1893 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1894 | dn->dn_checksum = checksum; |
1895 | dnode_setdirty(dn, tx); | |
1896 | dnode_rele(dn, FTAG); | |
1897 | } | |
1898 | ||
1899 | void | |
1900 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 1901 | dmu_tx_t *tx) |
34dc7c2f BB |
1902 | { |
1903 | dnode_t *dn; | |
1904 | ||
9b67f605 MA |
1905 | /* |
1906 | * Send streams include each object's compression function. This | |
1907 | * check ensures that the receiving system can understand the | |
1908 | * compression function transmitted. | |
1909 | */ | |
1910 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1911 | ||
1912 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
1913 | dn->dn_compress = compress; |
1914 | dnode_setdirty(dn, tx); | |
1915 | dnode_rele(dn, FTAG); | |
1916 | } | |
1917 | ||
faf0f58c MA |
1918 | /* |
1919 | * When the "redundant_metadata" property is set to "most", only indirect | |
1920 | * blocks of this level and higher will have an additional ditto block. | |
1921 | */ | |
1922 | int zfs_redundant_metadata_most_ditto_level = 2; | |
1923 | ||
428870ff | 1924 | void |
82644107 | 1925 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
1926 | { |
1927 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 1928 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 1929 | (wp & WP_SPILL)); |
428870ff BB |
1930 | enum zio_checksum checksum = os->os_checksum; |
1931 | enum zio_compress compress = os->os_compress; | |
10b3c7f5 | 1932 | uint8_t complevel = os->os_complevel; |
428870ff | 1933 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; |
03c6040b GW |
1934 | boolean_t dedup = B_FALSE; |
1935 | boolean_t nopwrite = B_FALSE; | |
428870ff | 1936 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 1937 | boolean_t encrypt = B_FALSE; |
428870ff | 1938 | int copies = os->os_copies; |
a7004725 | 1939 | |
428870ff | 1940 | /* |
03c6040b GW |
1941 | * We maintain different write policies for each of the following |
1942 | * types of data: | |
1943 | * 1. metadata | |
1944 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
1945 | * 3. all other level 0 blocks | |
428870ff BB |
1946 | */ |
1947 | if (ismd) { | |
b1d21733 TC |
1948 | /* |
1949 | * XXX -- we should design a compression algorithm | |
1950 | * that specializes in arrays of bps. | |
1951 | */ | |
1952 | compress = zio_compress_select(os->os_spa, | |
1953 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 1954 | |
428870ff BB |
1955 | /* |
1956 | * Metadata always gets checksummed. If the data | |
1957 | * checksum is multi-bit correctable, and it's not a | |
1958 | * ZBT-style checksum, then it's suitable for metadata | |
1959 | * as well. Otherwise, the metadata checksum defaults | |
1960 | * to fletcher4. | |
1961 | */ | |
3c67d83a TH |
1962 | if (!(zio_checksum_table[checksum].ci_flags & |
1963 | ZCHECKSUM_FLAG_METADATA) || | |
1964 | (zio_checksum_table[checksum].ci_flags & | |
1965 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 1966 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
1967 | |
1968 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
1969 | (os->os_redundant_metadata == | |
1970 | ZFS_REDUNDANT_METADATA_MOST && | |
1971 | (level >= zfs_redundant_metadata_most_ditto_level || | |
1972 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
1973 | copies++; | |
03c6040b GW |
1974 | } else if (wp & WP_NOFILL) { |
1975 | ASSERT(level == 0); | |
428870ff | 1976 | |
428870ff | 1977 | /* |
03c6040b GW |
1978 | * If we're writing preallocated blocks, we aren't actually |
1979 | * writing them so don't set any policy properties. These | |
1980 | * blocks are currently only used by an external subsystem | |
1981 | * outside of zfs (i.e. dump) and not written by the zio | |
1982 | * pipeline. | |
428870ff | 1983 | */ |
03c6040b GW |
1984 | compress = ZIO_COMPRESS_OFF; |
1985 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 1986 | } else { |
99197f03 JG |
1987 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
1988 | compress); | |
10b3c7f5 MN |
1989 | complevel = zio_complevel_select(os->os_spa, compress, |
1990 | complevel, complevel); | |
428870ff | 1991 | |
03c6040b GW |
1992 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
1993 | zio_checksum_select(dn->dn_checksum, checksum) : | |
1994 | dedup_checksum; | |
428870ff | 1995 | |
03c6040b GW |
1996 | /* |
1997 | * Determine dedup setting. If we are in dmu_sync(), | |
1998 | * we won't actually dedup now because that's all | |
1999 | * done in syncing context; but we do want to use the | |
e1cfd73f | 2000 | * dedup checksum. If the checksum is not strong |
03c6040b GW |
2001 | * enough to ensure unique signatures, force |
2002 | * dedup_verify. | |
2003 | */ | |
2004 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2005 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2006 | if (!(zio_checksum_table[checksum].ci_flags & |
2007 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2008 | dedup_verify = B_TRUE; |
2009 | } | |
428870ff | 2010 | |
03c6040b | 2011 | /* |
3c67d83a TH |
2012 | * Enable nopwrite if we have secure enough checksum |
2013 | * algorithm (see comment in zio_nop_write) and | |
2014 | * compression is enabled. We don't enable nopwrite if | |
2015 | * dedup is enabled as the two features are mutually | |
2016 | * exclusive. | |
03c6040b | 2017 | */ |
3c67d83a TH |
2018 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2019 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2020 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2021 | } |
2022 | ||
b5256303 TC |
2023 | /* |
2024 | * All objects in an encrypted objset are protected from modification | |
2025 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2026 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2027 | * not subject to nopwrite since writing the same data will still | |
2028 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2029 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2030 | * object types. | |
2031 | */ | |
2032 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2033 | encrypt = B_TRUE; | |
2034 | ||
2035 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2036 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2037 | nopwrite = B_FALSE; | |
2038 | } else { | |
2039 | dedup = B_FALSE; | |
2040 | } | |
2041 | ||
ae76f45c TC |
2042 | if (level <= 0 && |
2043 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2044 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2045 | } |
b5256303 | 2046 | } |
2aa34383 | 2047 | |
b5256303 | 2048 | zp->zp_compress = compress; |
10b3c7f5 | 2049 | zp->zp_complevel = complevel; |
b5256303 | 2050 | zp->zp_checksum = checksum; |
428870ff BB |
2051 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2052 | zp->zp_level = level; | |
faf0f58c | 2053 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2054 | zp->zp_dedup = dedup; |
2055 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2056 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2057 | zp->zp_encrypt = encrypt; |
2058 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
2059 | bzero(zp->zp_salt, ZIO_DATA_SALT_LEN); | |
2060 | bzero(zp->zp_iv, ZIO_DATA_IV_LEN); | |
2061 | bzero(zp->zp_mac, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2062 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2063 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2064 | |
2065 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2066 | } |
2067 | ||
66aca247 DB |
2068 | /* |
2069 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2070 | * in order to determine the location of holes. In order to accurately | |
2071 | * report holes all dirty data must be synced to disk. This causes extremely | |
2072 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2073 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2074 | * report the dnode as having no holes which is always a safe thing to do. | |
2075 | */ | |
34dc7c2f BB |
2076 | int |
2077 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2078 | { | |
2079 | dnode_t *dn; | |
2080 | int i, err; | |
66aca247 | 2081 | boolean_t clean = B_TRUE; |
34dc7c2f | 2082 | |
428870ff | 2083 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2084 | if (err) |
2085 | return (err); | |
66aca247 | 2086 | |
34dc7c2f | 2087 | /* |
2531ce37 | 2088 | * Check if dnode is dirty |
34dc7c2f | 2089 | */ |
454365bb | 2090 | for (i = 0; i < TXG_SIZE; i++) { |
edc1e713 | 2091 | if (multilist_link_active(&dn->dn_dirty_link[i])) { |
2531ce37 | 2092 | clean = B_FALSE; |
ec4f9b8f | 2093 | break; |
2531ce37 | 2094 | } |
34dc7c2f | 2095 | } |
66aca247 DB |
2096 | |
2097 | /* | |
2098 | * If compatibility option is on, sync any current changes before | |
2099 | * we go trundling through the block pointers. | |
2100 | */ | |
2101 | if (!clean && zfs_dmu_offset_next_sync) { | |
2102 | clean = B_TRUE; | |
34dc7c2f BB |
2103 | dnode_rele(dn, FTAG); |
2104 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 2105 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2106 | if (err) |
2107 | return (err); | |
2108 | } | |
2109 | ||
66aca247 DB |
2110 | if (clean) |
2111 | err = dnode_next_offset(dn, | |
2112 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2113 | else | |
2114 | err = SET_ERROR(EBUSY); | |
2115 | ||
34dc7c2f BB |
2116 | dnode_rele(dn, FTAG); |
2117 | ||
2118 | return (err); | |
2119 | } | |
2120 | ||
2121 | void | |
e0b0ca98 | 2122 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2123 | { |
e0b0ca98 | 2124 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2125 | |
34dc7c2f BB |
2126 | doi->doi_data_block_size = dn->dn_datablksz; |
2127 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2128 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2129 | doi->doi_type = dn->dn_type; |
2130 | doi->doi_bonus_type = dn->dn_bonustype; | |
2131 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2132 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2133 | doi->doi_indirection = dn->dn_nlevels; |
2134 | doi->doi_checksum = dn->dn_checksum; | |
2135 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2136 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2137 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2138 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2139 | doi->doi_fill_count = 0; |
1c27024e | 2140 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2141 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2142 | } |
2143 | ||
2144 | void | |
2145 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2146 | { | |
2147 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2148 | mutex_enter(&dn->dn_mtx); | |
2149 | ||
2150 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2151 | |
2152 | mutex_exit(&dn->dn_mtx); | |
2153 | rw_exit(&dn->dn_struct_rwlock); | |
2154 | } | |
2155 | ||
2156 | /* | |
2157 | * Get information on a DMU object. | |
2158 | * If doi is NULL, just indicates whether the object exists. | |
2159 | */ | |
2160 | int | |
2161 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2162 | { | |
2163 | dnode_t *dn; | |
428870ff | 2164 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2165 | |
2166 | if (err) | |
2167 | return (err); | |
2168 | ||
2169 | if (doi != NULL) | |
2170 | dmu_object_info_from_dnode(dn, doi); | |
2171 | ||
2172 | dnode_rele(dn, FTAG); | |
2173 | return (0); | |
2174 | } | |
2175 | ||
2176 | /* | |
2177 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2178 | */ | |
2179 | void | |
572e2857 | 2180 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2181 | { |
572e2857 BB |
2182 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2183 | ||
2184 | DB_DNODE_ENTER(db); | |
2185 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2186 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2187 | } |
2188 | ||
2189 | /* | |
2190 | * Faster still when you only care about the size. | |
34dc7c2f BB |
2191 | */ |
2192 | void | |
572e2857 BB |
2193 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2194 | u_longlong_t *nblk512) | |
34dc7c2f | 2195 | { |
572e2857 BB |
2196 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2197 | dnode_t *dn; | |
2198 | ||
2199 | DB_DNODE_ENTER(db); | |
2200 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2201 | |
2202 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2203 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2204 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2205 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2206 | DB_DNODE_EXIT(db); | |
2207 | } | |
2208 | ||
2209 | void | |
2210 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2211 | { | |
2212 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2213 | dnode_t *dn; | |
2214 | ||
2215 | DB_DNODE_ENTER(db); | |
2216 | dn = DB_DNODE(db); | |
2217 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2218 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2219 | } |
2220 | ||
2221 | void | |
2222 | byteswap_uint64_array(void *vbuf, size_t size) | |
2223 | { | |
2224 | uint64_t *buf = vbuf; | |
2225 | size_t count = size >> 3; | |
2226 | int i; | |
2227 | ||
2228 | ASSERT((size & 7) == 0); | |
2229 | ||
2230 | for (i = 0; i < count; i++) | |
2231 | buf[i] = BSWAP_64(buf[i]); | |
2232 | } | |
2233 | ||
2234 | void | |
2235 | byteswap_uint32_array(void *vbuf, size_t size) | |
2236 | { | |
2237 | uint32_t *buf = vbuf; | |
2238 | size_t count = size >> 2; | |
2239 | int i; | |
2240 | ||
2241 | ASSERT((size & 3) == 0); | |
2242 | ||
2243 | for (i = 0; i < count; i++) | |
2244 | buf[i] = BSWAP_32(buf[i]); | |
2245 | } | |
2246 | ||
2247 | void | |
2248 | byteswap_uint16_array(void *vbuf, size_t size) | |
2249 | { | |
2250 | uint16_t *buf = vbuf; | |
2251 | size_t count = size >> 1; | |
2252 | int i; | |
2253 | ||
2254 | ASSERT((size & 1) == 0); | |
2255 | ||
2256 | for (i = 0; i < count; i++) | |
2257 | buf[i] = BSWAP_16(buf[i]); | |
2258 | } | |
2259 | ||
2260 | /* ARGSUSED */ | |
2261 | void | |
2262 | byteswap_uint8_array(void *vbuf, size_t size) | |
2263 | { | |
2264 | } | |
2265 | ||
2266 | void | |
2267 | dmu_init(void) | |
2268 | { | |
a6255b7f | 2269 | abd_init(); |
428870ff | 2270 | zfs_dbgmsg_init(); |
572e2857 | 2271 | sa_cache_init(); |
572e2857 | 2272 | dmu_objset_init(); |
34dc7c2f | 2273 | dnode_init(); |
428870ff | 2274 | zfetch_init(); |
570827e1 | 2275 | dmu_tx_init(); |
34dc7c2f | 2276 | l2arc_init(); |
29809a6c | 2277 | arc_init(); |
d3c2ae1c | 2278 | dbuf_init(); |
34dc7c2f BB |
2279 | } |
2280 | ||
2281 | void | |
2282 | dmu_fini(void) | |
2283 | { | |
e49f1e20 | 2284 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2285 | l2arc_fini(); |
570827e1 | 2286 | dmu_tx_fini(); |
428870ff | 2287 | zfetch_fini(); |
34dc7c2f | 2288 | dbuf_fini(); |
572e2857 BB |
2289 | dnode_fini(); |
2290 | dmu_objset_fini(); | |
428870ff BB |
2291 | sa_cache_fini(); |
2292 | zfs_dbgmsg_fini(); | |
a6255b7f | 2293 | abd_fini(); |
34dc7c2f | 2294 | } |
c28b2279 | 2295 | |
c28b2279 | 2296 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2297 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2298 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2299 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2300 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2301 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2302 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2303 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2304 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2305 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2306 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2307 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2308 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2309 | EXPORT_SYMBOL(dmu_object_info); |
2310 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2311 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2312 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2313 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2314 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2315 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2316 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2317 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2318 | EXPORT_SYMBOL(dmu_object_set_compress); | |
94dac3e8 | 2319 | EXPORT_SYMBOL(dmu_offset_next); |
57b650b8 BB |
2320 | EXPORT_SYMBOL(dmu_write_policy); |
2321 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2322 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2323 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2324 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2325 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2326 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2327 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2328 | |
bef78122 | 2329 | /* BEGIN CSTYLED */ |
03fdcb9a MM |
2330 | ZFS_MODULE_PARAM(zfs, zfs_, nopwrite_enabled, INT, ZMOD_RW, |
2331 | "Enable NOP writes"); | |
03c6040b | 2332 | |
03fdcb9a MM |
2333 | ZFS_MODULE_PARAM(zfs, zfs_, per_txg_dirty_frees_percent, ULONG, ZMOD_RW, |
2334 | "Percentage of dirtied blocks from frees in one TXG"); | |
66aca247 | 2335 | |
03fdcb9a | 2336 | ZFS_MODULE_PARAM(zfs, zfs_, dmu_offset_next_sync, INT, ZMOD_RW, |
66aca247 DB |
2337 | "Enable forcing txg sync to find holes"); |
2338 | ||
03fdcb9a | 2339 | ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, INT, ZMOD_RW, |
d9b4bf06 | 2340 | "Limit one prefetch call to this size"); |
bef78122 | 2341 | /* END CSTYLED */ |