]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
34dc7c2f BB |
10 | * See the License for the specific language governing permissions |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
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> |
64e0fe14 | 55 | #include <sys/zfs_racct.h> |
f763c3d1 | 56 | #include <sys/zfs_rlock.h> |
34dc7c2f BB |
57 | #ifdef _KERNEL |
58 | #include <sys/vmsystm.h> | |
b128c09f | 59 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
60 | #endif |
61 | ||
03c6040b GW |
62 | /* |
63 | * Enable/disable nopwrite feature. | |
64 | */ | |
18168da7 | 65 | static int zfs_nopwrite_enabled = 1; |
03c6040b | 66 | |
539d33c7 | 67 | /* |
65282ee9 AP |
68 | * Tunable to control percentage of dirtied L1 blocks from frees allowed into |
69 | * one TXG. After this threshold is crossed, additional dirty blocks from frees | |
70 | * will wait until the next TXG. | |
539d33c7 GM |
71 | * A value of zero will disable this throttle. |
72 | */ | |
18168da7 | 73 | static unsigned long zfs_per_txg_dirty_frees_percent = 5; |
539d33c7 | 74 | |
66aca247 | 75 | /* |
05b3eb6d BB |
76 | * Enable/disable forcing txg sync when dirty checking for holes with lseek(). |
77 | * By default this is enabled to ensure accurate hole reporting, it can result | |
78 | * in a significant performance penalty for lseek(SEEK_HOLE) heavy workloads. | |
79 | * Disabling this option will result in holes never being reported in dirty | |
80 | * files which is always safe. | |
66aca247 | 81 | */ |
18168da7 | 82 | static int zfs_dmu_offset_next_sync = 1; |
66aca247 | 83 | |
d9b4bf06 MA |
84 | /* |
85 | * Limit the amount we can prefetch with one call to this amount. This | |
86 | * helps to limit the amount of memory that can be used by prefetching. | |
87 | * Larger objects should be prefetched a bit at a time. | |
88 | */ | |
fdc2d303 | 89 | uint_t dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE; |
d9b4bf06 | 90 | |
34dc7c2f | 91 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
2e5dc449 MA |
92 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" }, |
93 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" }, | |
94 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" }, | |
95 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" }, | |
96 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" }, | |
97 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" }, | |
98 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" }, | |
99 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" }, | |
100 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" }, | |
101 | {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" }, | |
102 | {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" }, | |
103 | {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" }, | |
104 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" }, | |
105 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"}, | |
106 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" }, | |
107 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" }, | |
108 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" }, | |
109 | {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" }, | |
110 | {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" }, | |
111 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" }, | |
112 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" }, | |
113 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" }, | |
114 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" }, | |
115 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" }, | |
116 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" }, | |
117 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" }, | |
118 | {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" }, | |
119 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" }, | |
120 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" }, | |
121 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" }, | |
122 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" }, | |
123 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" }, | |
124 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" }, | |
125 | {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" }, | |
126 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" }, | |
127 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" }, | |
128 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" }, | |
129 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"}, | |
130 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" }, | |
131 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" }, | |
132 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"}, | |
133 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"}, | |
134 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" }, | |
135 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" }, | |
136 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" }, | |
137 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" }, | |
138 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" }, | |
139 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" }, | |
140 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" }, | |
141 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" }, | |
142 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" }, | |
143 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" }, | |
144 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" }, | |
145 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
146 | }; |
147 | ||
148 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
149 | { byteswap_uint8_array, "uint8" }, | |
150 | { byteswap_uint16_array, "uint16" }, | |
151 | { byteswap_uint32_array, "uint32" }, | |
152 | { byteswap_uint64_array, "uint64" }, | |
153 | { zap_byteswap, "zap" }, | |
154 | { dnode_buf_byteswap, "dnode" }, | |
155 | { dmu_objset_byteswap, "objset" }, | |
156 | { zfs_znode_byteswap, "znode" }, | |
157 | { zfs_oldacl_byteswap, "oldacl" }, | |
158 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
159 | }; |
160 | ||
65c7cc49 | 161 | static int |
2bce8049 | 162 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, |
a926aab9 | 163 | const void *tag, dmu_buf_t **dbp) |
2bce8049 MA |
164 | { |
165 | uint64_t blkid; | |
166 | dmu_buf_impl_t *db; | |
167 | ||
2bce8049 | 168 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 169 | blkid = dbuf_whichblock(dn, 0, offset); |
2bce8049 MA |
170 | db = dbuf_hold(dn, blkid, tag); |
171 | rw_exit(&dn->dn_struct_rwlock); | |
172 | ||
173 | if (db == NULL) { | |
174 | *dbp = NULL; | |
175 | return (SET_ERROR(EIO)); | |
176 | } | |
177 | ||
178 | *dbp = &db->db; | |
179 | return (0); | |
180 | } | |
34dc7c2f | 181 | int |
9b67f605 | 182 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
a926aab9 | 183 | const void *tag, dmu_buf_t **dbp) |
34dc7c2f BB |
184 | { |
185 | dnode_t *dn; | |
186 | uint64_t blkid; | |
187 | dmu_buf_impl_t *db; | |
188 | int err; | |
428870ff BB |
189 | |
190 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
191 | if (err) |
192 | return (err); | |
34dc7c2f | 193 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 194 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
195 | db = dbuf_hold(dn, blkid, tag); |
196 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
197 | dnode_rele(dn, FTAG); |
198 | ||
34dc7c2f | 199 | if (db == NULL) { |
9b67f605 MA |
200 | *dbp = NULL; |
201 | return (SET_ERROR(EIO)); | |
202 | } | |
203 | ||
204 | *dbp = &db->db; | |
205 | return (err); | |
206 | } | |
207 | ||
2bce8049 MA |
208 | int |
209 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
a926aab9 | 210 | const void *tag, dmu_buf_t **dbp, int flags) |
2bce8049 MA |
211 | { |
212 | int err; | |
213 | int db_flags = DB_RF_CANFAIL; | |
214 | ||
215 | if (flags & DMU_READ_NO_PREFETCH) | |
216 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
217 | if (flags & DMU_READ_NO_DECRYPT) |
218 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
219 | |
220 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
221 | if (err == 0) { | |
222 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
223 | err = dbuf_read(db, NULL, db_flags); | |
224 | if (err != 0) { | |
225 | dbuf_rele(db, tag); | |
226 | *dbp = NULL; | |
227 | } | |
228 | } | |
229 | ||
230 | return (err); | |
231 | } | |
232 | ||
9b67f605 MA |
233 | int |
234 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
a926aab9 | 235 | const void *tag, dmu_buf_t **dbp, int flags) |
9b67f605 MA |
236 | { |
237 | int err; | |
238 | int db_flags = DB_RF_CANFAIL; | |
239 | ||
240 | if (flags & DMU_READ_NO_PREFETCH) | |
241 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
242 | if (flags & DMU_READ_NO_DECRYPT) |
243 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
244 | |
245 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
246 | if (err == 0) { | |
247 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 248 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 249 | if (err != 0) { |
34dc7c2f | 250 | dbuf_rele(db, tag); |
9b67f605 | 251 | *dbp = NULL; |
34dc7c2f BB |
252 | } |
253 | } | |
254 | ||
34dc7c2f BB |
255 | return (err); |
256 | } | |
257 | ||
258 | int | |
259 | dmu_bonus_max(void) | |
260 | { | |
50c957f7 | 261 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
262 | } |
263 | ||
264 | int | |
572e2857 | 265 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 266 | { |
572e2857 BB |
267 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
268 | dnode_t *dn; | |
269 | int error; | |
34dc7c2f | 270 | |
572e2857 BB |
271 | DB_DNODE_ENTER(db); |
272 | dn = DB_DNODE(db); | |
273 | ||
274 | if (dn->dn_bonus != db) { | |
2e528b49 | 275 | error = SET_ERROR(EINVAL); |
572e2857 | 276 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 277 | error = SET_ERROR(EINVAL); |
572e2857 BB |
278 | } else { |
279 | dnode_setbonuslen(dn, newsize, tx); | |
280 | error = 0; | |
281 | } | |
282 | ||
283 | DB_DNODE_EXIT(db); | |
284 | return (error); | |
34dc7c2f BB |
285 | } |
286 | ||
428870ff | 287 | int |
572e2857 | 288 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 289 | { |
572e2857 BB |
290 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
291 | dnode_t *dn; | |
292 | int error; | |
428870ff | 293 | |
572e2857 BB |
294 | DB_DNODE_ENTER(db); |
295 | dn = DB_DNODE(db); | |
428870ff | 296 | |
9ae529ec | 297 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 298 | error = SET_ERROR(EINVAL); |
572e2857 | 299 | } else if (dn->dn_bonus != db) { |
2e528b49 | 300 | error = SET_ERROR(EINVAL); |
572e2857 BB |
301 | } else { |
302 | dnode_setbonus_type(dn, type, tx); | |
303 | error = 0; | |
304 | } | |
428870ff | 305 | |
572e2857 BB |
306 | DB_DNODE_EXIT(db); |
307 | return (error); | |
308 | } | |
309 | ||
310 | dmu_object_type_t | |
311 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
312 | { | |
313 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
314 | dnode_t *dn; | |
315 | dmu_object_type_t type; | |
316 | ||
317 | DB_DNODE_ENTER(db); | |
318 | dn = DB_DNODE(db); | |
319 | type = dn->dn_bonustype; | |
320 | DB_DNODE_EXIT(db); | |
321 | ||
322 | return (type); | |
428870ff BB |
323 | } |
324 | ||
325 | int | |
326 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
327 | { | |
328 | dnode_t *dn; | |
329 | int error; | |
330 | ||
331 | error = dnode_hold(os, object, FTAG, &dn); | |
332 | dbuf_rm_spill(dn, tx); | |
333 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
334 | dnode_rm_spill(dn, tx); | |
335 | rw_exit(&dn->dn_struct_rwlock); | |
336 | dnode_rele(dn, FTAG); | |
337 | return (error); | |
338 | } | |
339 | ||
34dc7c2f | 340 | /* |
6955b401 BB |
341 | * Lookup and hold the bonus buffer for the provided dnode. If the dnode |
342 | * has not yet been allocated a new bonus dbuf a will be allocated. | |
343 | * Returns ENOENT, EIO, or 0. | |
34dc7c2f | 344 | */ |
a926aab9 | 345 | int dmu_bonus_hold_by_dnode(dnode_t *dn, const void *tag, dmu_buf_t **dbp, |
6955b401 | 346 | uint32_t flags) |
34dc7c2f | 347 | { |
34dc7c2f BB |
348 | dmu_buf_impl_t *db; |
349 | int error; | |
b5256303 TC |
350 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
351 | ||
352 | if (flags & DMU_READ_NO_PREFETCH) | |
353 | db_flags |= DB_RF_NOPREFETCH; | |
354 | if (flags & DMU_READ_NO_DECRYPT) | |
355 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 356 | |
34dc7c2f BB |
357 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
358 | if (dn->dn_bonus == NULL) { | |
359 | rw_exit(&dn->dn_struct_rwlock); | |
360 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
361 | if (dn->dn_bonus == NULL) | |
362 | dbuf_create_bonus(dn); | |
363 | } | |
364 | db = dn->dn_bonus; | |
34dc7c2f BB |
365 | |
366 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 367 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 368 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 369 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
370 | } |
371 | ||
372 | /* | |
373 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
374 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
375 | * a dnode hold for every dbuf. | |
376 | */ | |
377 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f | 378 | |
b5256303 TC |
379 | error = dbuf_read(db, NULL, db_flags); |
380 | if (error) { | |
381 | dnode_evict_bonus(dn); | |
382 | dbuf_rele(db, tag); | |
383 | *dbp = NULL; | |
384 | return (error); | |
385 | } | |
34dc7c2f BB |
386 | |
387 | *dbp = &db->db; | |
388 | return (0); | |
389 | } | |
390 | ||
b5256303 | 391 | int |
a926aab9 | 392 | dmu_bonus_hold(objset_t *os, uint64_t object, const void *tag, dmu_buf_t **dbp) |
b5256303 | 393 | { |
6955b401 BB |
394 | dnode_t *dn; |
395 | int error; | |
396 | ||
397 | error = dnode_hold(os, object, FTAG, &dn); | |
398 | if (error) | |
399 | return (error); | |
400 | ||
401 | error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH); | |
402 | dnode_rele(dn, FTAG); | |
403 | ||
404 | return (error); | |
b5256303 TC |
405 | } |
406 | ||
428870ff BB |
407 | /* |
408 | * returns ENOENT, EIO, or 0. | |
409 | * | |
410 | * This interface will allocate a blank spill dbuf when a spill blk | |
411 | * doesn't already exist on the dnode. | |
412 | * | |
413 | * if you only want to find an already existing spill db, then | |
414 | * dmu_spill_hold_existing() should be used. | |
415 | */ | |
416 | int | |
a926aab9 AZ |
417 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, const void *tag, |
418 | dmu_buf_t **dbp) | |
428870ff BB |
419 | { |
420 | dmu_buf_impl_t *db = NULL; | |
421 | int err; | |
422 | ||
423 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
424 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
425 | ||
426 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
427 | ||
428 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
429 | rw_exit(&dn->dn_struct_rwlock); | |
430 | ||
b182ac00 | 431 | if (db == NULL) { |
432 | *dbp = NULL; | |
433 | return (SET_ERROR(EIO)); | |
434 | } | |
572e2857 BB |
435 | err = dbuf_read(db, NULL, flags); |
436 | if (err == 0) | |
437 | *dbp = &db->db; | |
b182ac00 | 438 | else { |
572e2857 | 439 | dbuf_rele(db, tag); |
b182ac00 | 440 | *dbp = NULL; |
441 | } | |
428870ff BB |
442 | return (err); |
443 | } | |
444 | ||
445 | int | |
a926aab9 | 446 | dmu_spill_hold_existing(dmu_buf_t *bonus, const void *tag, dmu_buf_t **dbp) |
428870ff | 447 | { |
572e2857 BB |
448 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
449 | dnode_t *dn; | |
428870ff BB |
450 | int err; |
451 | ||
572e2857 BB |
452 | DB_DNODE_ENTER(db); |
453 | dn = DB_DNODE(db); | |
454 | ||
455 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 456 | err = SET_ERROR(EINVAL); |
572e2857 BB |
457 | } else { |
458 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
459 | ||
460 | if (!dn->dn_have_spill) { | |
2e528b49 | 461 | err = SET_ERROR(ENOENT); |
572e2857 BB |
462 | } else { |
463 | err = dmu_spill_hold_by_dnode(dn, | |
464 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
465 | } | |
428870ff | 466 | |
428870ff | 467 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 468 | } |
572e2857 BB |
469 | |
470 | DB_DNODE_EXIT(db); | |
428870ff BB |
471 | return (err); |
472 | } | |
473 | ||
474 | int | |
a926aab9 | 475 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, const void *tag, |
e7504d7a | 476 | dmu_buf_t **dbp) |
428870ff | 477 | { |
572e2857 BB |
478 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
479 | dnode_t *dn; | |
480 | int err; | |
e7504d7a TC |
481 | uint32_t db_flags = DB_RF_CANFAIL; |
482 | ||
483 | if (flags & DMU_READ_NO_DECRYPT) | |
484 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
485 | |
486 | DB_DNODE_ENTER(db); | |
487 | dn = DB_DNODE(db); | |
e7504d7a | 488 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
489 | DB_DNODE_EXIT(db); |
490 | ||
491 | return (err); | |
428870ff BB |
492 | } |
493 | ||
34dc7c2f BB |
494 | /* |
495 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
496 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
497 | * and can induce severe lock contention when writing to several files | |
498 | * whose dnodes are in the same block. | |
499 | */ | |
af1698f5 | 500 | int |
9babb374 | 501 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
a926aab9 AZ |
502 | boolean_t read, const void *tag, int *numbufsp, dmu_buf_t ***dbpp, |
503 | uint32_t flags) | |
34dc7c2f BB |
504 | { |
505 | dmu_buf_t **dbp; | |
891568c9 | 506 | zstream_t *zs = NULL; |
34dc7c2f | 507 | uint64_t blkid, nblks, i; |
9babb374 | 508 | uint32_t dbuf_flags; |
34dc7c2f | 509 | int err; |
ec50cd24 | 510 | zio_t *zio = NULL; |
891568c9 | 511 | boolean_t missed = B_FALSE; |
34dc7c2f BB |
512 | |
513 | ASSERT(length <= DMU_MAX_ACCESS); | |
514 | ||
7f60329a MA |
515 | /* |
516 | * Note: We directly notify the prefetch code of this read, so that | |
517 | * we can tell it about the multi-block read. dbuf_read() only knows | |
518 | * about the one block it is accessing. | |
519 | */ | |
520 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
521 | DB_RF_NOPREFETCH; | |
34dc7c2f | 522 | |
e8cf3a4f AP |
523 | if ((flags & DMU_READ_NO_DECRYPT) != 0) |
524 | dbuf_flags |= DB_RF_NO_DECRYPT; | |
525 | ||
34dc7c2f BB |
526 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
527 | if (dn->dn_datablkshift) { | |
528 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
529 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
530 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
531 | } else { |
532 | if (offset + length > dn->dn_datablksz) { | |
533 | zfs_panic_recover("zfs: accessing past end of object " | |
534 | "%llx/%llx (size=%u access=%llu+%llu)", | |
535 | (longlong_t)dn->dn_objset-> | |
536 | os_dsl_dataset->ds_object, | |
537 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
538 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 539 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 540 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
541 | } |
542 | nblks = 1; | |
543 | } | |
79c76d5b | 544 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 545 | |
ec50cd24 FY |
546 | if (read) |
547 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, | |
548 | ZIO_FLAG_CANFAIL); | |
fcff0f35 | 549 | blkid = dbuf_whichblock(dn, 0, offset); |
891568c9 AM |
550 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
551 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
552 | /* | |
553 | * Prepare the zfetch before initiating the demand reads, so | |
554 | * that if multiple threads block on same indirect block, we | |
555 | * base predictions on the original less racy request order. | |
556 | */ | |
557 | zs = dmu_zfetch_prepare(&dn->dn_zfetch, blkid, nblks, | |
558 | read && DNODE_IS_CACHEABLE(dn), B_TRUE); | |
559 | } | |
34dc7c2f | 560 | for (i = 0; i < nblks; i++) { |
7f60329a | 561 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f | 562 | if (db == NULL) { |
891568c9 AM |
563 | if (zs) |
564 | dmu_zfetch_run(zs, missed, B_TRUE); | |
34dc7c2f BB |
565 | rw_exit(&dn->dn_struct_rwlock); |
566 | dmu_buf_rele_array(dbp, nblks, tag); | |
ec50cd24 FY |
567 | if (read) |
568 | zio_nowait(zio); | |
2e528b49 | 569 | return (SET_ERROR(EIO)); |
34dc7c2f | 570 | } |
7f60329a | 571 | |
891568c9 AM |
572 | /* |
573 | * Initiate async demand data read. | |
574 | * We check the db_state after calling dbuf_read() because | |
575 | * (1) dbuf_read() may change the state to CACHED due to a | |
576 | * hit in the ARC, and (2) on a cache miss, a child will | |
577 | * have been added to "zio" but not yet completed, so the | |
578 | * state will not yet be CACHED. | |
579 | */ | |
580 | if (read) { | |
9babb374 | 581 | (void) dbuf_read(db, zio, dbuf_flags); |
891568c9 AM |
582 | if (db->db_state != DB_CACHED) |
583 | missed = B_TRUE; | |
584 | } | |
34dc7c2f BB |
585 | dbp[i] = &db->db; |
586 | } | |
7f60329a | 587 | |
64e0fe14 RM |
588 | if (!read) |
589 | zfs_racct_write(length, nblks); | |
590 | ||
891568c9 AM |
591 | if (zs) |
592 | dmu_zfetch_run(zs, missed, B_TRUE); | |
34dc7c2f BB |
593 | rw_exit(&dn->dn_struct_rwlock); |
594 | ||
34dc7c2f | 595 | if (read) { |
ec50cd24 FY |
596 | /* wait for async read i/o */ |
597 | err = zio_wait(zio); | |
598 | if (err) { | |
599 | dmu_buf_rele_array(dbp, nblks, tag); | |
600 | return (err); | |
601 | } | |
602 | ||
603 | /* wait for other io to complete */ | |
34dc7c2f BB |
604 | for (i = 0; i < nblks; i++) { |
605 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
606 | mutex_enter(&db->db_mtx); | |
607 | while (db->db_state == DB_READ || | |
608 | db->db_state == DB_FILL) | |
609 | cv_wait(&db->db_changed, &db->db_mtx); | |
610 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 611 | err = SET_ERROR(EIO); |
34dc7c2f BB |
612 | mutex_exit(&db->db_mtx); |
613 | if (err) { | |
614 | dmu_buf_rele_array(dbp, nblks, tag); | |
615 | return (err); | |
616 | } | |
617 | } | |
618 | } | |
619 | ||
620 | *numbufsp = nblks; | |
621 | *dbpp = dbp; | |
622 | return (0); | |
623 | } | |
624 | ||
afbc6179 | 625 | int |
34dc7c2f | 626 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, |
a926aab9 AZ |
627 | uint64_t length, int read, const void *tag, int *numbufsp, |
628 | dmu_buf_t ***dbpp) | |
34dc7c2f BB |
629 | { |
630 | dnode_t *dn; | |
631 | int err; | |
632 | ||
428870ff | 633 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
634 | if (err) |
635 | return (err); | |
636 | ||
637 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 638 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
639 | |
640 | dnode_rele(dn, FTAG); | |
641 | ||
642 | return (err); | |
643 | } | |
644 | ||
645 | int | |
572e2857 | 646 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
a926aab9 | 647 | uint64_t length, boolean_t read, const void *tag, int *numbufsp, |
7f60329a | 648 | dmu_buf_t ***dbpp) |
34dc7c2f | 649 | { |
572e2857 BB |
650 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
651 | dnode_t *dn; | |
34dc7c2f BB |
652 | int err; |
653 | ||
572e2857 BB |
654 | DB_DNODE_ENTER(db); |
655 | dn = DB_DNODE(db); | |
34dc7c2f | 656 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 657 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 658 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
659 | |
660 | return (err); | |
661 | } | |
662 | ||
663 | void | |
a926aab9 | 664 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, const void *tag) |
34dc7c2f BB |
665 | { |
666 | int i; | |
667 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
668 | ||
669 | if (numbufs == 0) | |
670 | return; | |
671 | ||
672 | for (i = 0; i < numbufs; i++) { | |
673 | if (dbp[i]) | |
674 | dbuf_rele(dbp[i], tag); | |
675 | } | |
676 | ||
677 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
678 | } | |
679 | ||
e8b96c60 | 680 | /* |
fcff0f35 | 681 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
e1cfd73f | 682 | * indirect blocks prefetched will be those that point to the blocks containing |
fcff0f35 | 683 | * the data starting at offset, and continuing to offset + len. |
e8b96c60 | 684 | * |
b5256303 | 685 | * Note that if the indirect blocks above the blocks being prefetched are not |
e1cfd73f | 686 | * in cache, they will be asynchronously read in. |
e8b96c60 | 687 | */ |
34dc7c2f | 688 | void |
fcff0f35 PD |
689 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
690 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
691 | { |
692 | dnode_t *dn; | |
693 | uint64_t blkid; | |
e8b96c60 | 694 | int nblks, err; |
34dc7c2f | 695 | |
34dc7c2f | 696 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 697 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
698 | |
699 | if (object == 0 || object >= DN_MAX_OBJECT) | |
700 | return; | |
701 | ||
702 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
703 | blkid = dbuf_whichblock(dn, level, |
704 | object * sizeof (dnode_phys_t)); | |
705 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
706 | rw_exit(&dn->dn_struct_rwlock); |
707 | return; | |
708 | } | |
709 | ||
d9b4bf06 MA |
710 | /* |
711 | * See comment before the definition of dmu_prefetch_max. | |
712 | */ | |
713 | len = MIN(len, dmu_prefetch_max); | |
714 | ||
34dc7c2f BB |
715 | /* |
716 | * XXX - Note, if the dnode for the requested object is not | |
717 | * already cached, we will do a *synchronous* read in the | |
718 | * dnode_hold() call. The same is true for any indirects. | |
719 | */ | |
428870ff | 720 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
721 | if (err != 0) |
722 | return; | |
723 | ||
fcff0f35 PD |
724 | /* |
725 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
726 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
727 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
728 | * offset) is the first. Then the number we need to prefetch is the | |
729 | * last - first + 1. | |
730 | */ | |
f664f1ee | 731 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
fcff0f35 PD |
732 | if (level > 0 || dn->dn_datablkshift != 0) { |
733 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
734 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
735 | } else { |
736 | nblks = (offset < dn->dn_datablksz); | |
737 | } | |
738 | ||
739 | if (nblks != 0) { | |
fcff0f35 | 740 | blkid = dbuf_whichblock(dn, level, offset); |
1c27024e | 741 | for (int i = 0; i < nblks; i++) |
fcff0f35 | 742 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f | 743 | } |
34dc7c2f BB |
744 | rw_exit(&dn->dn_struct_rwlock); |
745 | ||
746 | dnode_rele(dn, FTAG); | |
747 | } | |
748 | ||
45d1cae3 BB |
749 | /* |
750 | * Get the next "chunk" of file data to free. We traverse the file from | |
751 | * the end so that the file gets shorter over time (if we crashes in the | |
752 | * middle, this will leave us in a better state). We find allocated file | |
753 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
754 | * |
755 | * On input, *start should be the first offset that does not need to be | |
756 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
757 | * offset that should be freed and l1blks is set to the number of level 1 |
758 | * indirect blocks found within the chunk. | |
45d1cae3 | 759 | */ |
b128c09f | 760 | static int |
65282ee9 | 761 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 762 | { |
65282ee9 | 763 | uint64_t blks; |
b663a23d MA |
764 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
765 | /* bytes of data covered by a level-1 indirect block */ | |
ec4afd27 OM |
766 | uint64_t iblkrange = (uint64_t)dn->dn_datablksz * |
767 | EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
b128c09f | 768 | |
b663a23d | 769 | ASSERT3U(minimum, <=, *start); |
b128c09f | 770 | |
f4c594da TC |
771 | /* |
772 | * Check if we can free the entire range assuming that all of the | |
773 | * L1 blocks in this range have data. If we can, we use this | |
774 | * worst case value as an estimate so we can avoid having to look | |
775 | * at the object's actual data. | |
776 | */ | |
777 | uint64_t total_l1blks = | |
778 | (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / | |
779 | iblkrange; | |
780 | if (total_l1blks <= maxblks) { | |
781 | *l1blks = total_l1blks; | |
b663a23d | 782 | *start = minimum; |
b128c09f BB |
783 | return (0); |
784 | } | |
45d1cae3 | 785 | ASSERT(ISP2(iblkrange)); |
b128c09f | 786 | |
65282ee9 | 787 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
788 | int err; |
789 | ||
b663a23d MA |
790 | /* |
791 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
792 | * indirect block at or before the input offset. We must | |
793 | * decrement *start so that it is at the end of the region | |
794 | * to search. | |
795 | */ | |
796 | (*start)--; | |
f4c594da | 797 | |
b128c09f | 798 | err = dnode_next_offset(dn, |
45d1cae3 | 799 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 800 | |
b663a23d | 801 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 802 | if (err == ESRCH) { |
b663a23d MA |
803 | *start = minimum; |
804 | break; | |
805 | } else if (err != 0) { | |
65282ee9 | 806 | *l1blks = blks; |
b128c09f | 807 | return (err); |
45d1cae3 | 808 | } |
b128c09f | 809 | |
b663a23d | 810 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 811 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 812 | } |
b663a23d MA |
813 | if (*start < minimum) |
814 | *start = minimum; | |
65282ee9 | 815 | *l1blks = blks; |
f4c594da | 816 | |
b128c09f BB |
817 | return (0); |
818 | } | |
819 | ||
a08abc1b GM |
820 | /* |
821 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
822 | * otherwise return false. | |
823 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
824 | */ | |
a08abc1b GM |
825 | static boolean_t |
826 | dmu_objset_zfs_unmounting(objset_t *os) | |
827 | { | |
828 | #ifdef _KERNEL | |
829 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
830 | return (zfs_get_vfs_flag_unmounted(os)); | |
14e4e3cb AZ |
831 | #else |
832 | (void) os; | |
a08abc1b GM |
833 | #endif |
834 | return (B_FALSE); | |
835 | } | |
836 | ||
b128c09f BB |
837 | static int |
838 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 839 | uint64_t length) |
b128c09f | 840 | { |
c97d3069 | 841 | uint64_t object_size; |
b663a23d | 842 | int err; |
539d33c7 GM |
843 | uint64_t dirty_frees_threshold; |
844 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 845 | |
c97d3069 BB |
846 | if (dn == NULL) |
847 | return (SET_ERROR(EINVAL)); | |
848 | ||
849 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 850 | if (offset >= object_size) |
b128c09f | 851 | return (0); |
b128c09f | 852 | |
539d33c7 GM |
853 | if (zfs_per_txg_dirty_frees_percent <= 100) |
854 | dirty_frees_threshold = | |
855 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
856 | else | |
65282ee9 | 857 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 858 | |
b663a23d MA |
859 | if (length == DMU_OBJECT_END || offset + length > object_size) |
860 | length = object_size - offset; | |
861 | ||
862 | while (length != 0) { | |
539d33c7 | 863 | uint64_t chunk_end, chunk_begin, chunk_len; |
65282ee9 | 864 | uint64_t l1blks; |
b663a23d MA |
865 | dmu_tx_t *tx; |
866 | ||
a08abc1b GM |
867 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
868 | return (SET_ERROR(EINTR)); | |
869 | ||
b663a23d MA |
870 | chunk_end = chunk_begin = offset + length; |
871 | ||
872 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 873 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
874 | if (err) |
875 | return (err); | |
b663a23d MA |
876 | ASSERT3U(chunk_begin, >=, offset); |
877 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 878 | |
539d33c7 GM |
879 | chunk_len = chunk_end - chunk_begin; |
880 | ||
b128c09f | 881 | tx = dmu_tx_create(os); |
539d33c7 | 882 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
883 | |
884 | /* | |
885 | * Mark this transaction as typically resulting in a net | |
886 | * reduction in space used. | |
887 | */ | |
888 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
889 | err = dmu_tx_assign(tx, TXG_WAIT); |
890 | if (err) { | |
891 | dmu_tx_abort(tx); | |
892 | return (err); | |
893 | } | |
539d33c7 | 894 | |
f4c594da TC |
895 | uint64_t txg = dmu_tx_get_txg(tx); |
896 | ||
897 | mutex_enter(&dp->dp_lock); | |
898 | uint64_t long_free_dirty = | |
899 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; | |
900 | mutex_exit(&dp->dp_lock); | |
901 | ||
902 | /* | |
903 | * To avoid filling up a TXG with just frees, wait for | |
904 | * the next TXG to open before freeing more chunks if | |
905 | * we have reached the threshold of frees. | |
906 | */ | |
907 | if (dirty_frees_threshold != 0 && | |
908 | long_free_dirty >= dirty_frees_threshold) { | |
909 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); | |
910 | dmu_tx_commit(tx); | |
911 | txg_wait_open(dp, 0, B_TRUE); | |
912 | continue; | |
913 | } | |
914 | ||
65282ee9 AP |
915 | /* |
916 | * In order to prevent unnecessary write throttling, for each | |
917 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
918 | * in dnode_free_range() below. We compare this number to a | |
919 | * tunable threshold, past which we prevent new L1 dirty freeing | |
920 | * blocks from being added into the open TXG. See | |
921 | * dmu_free_long_range_impl() for details. The threshold | |
922 | * prevents write throttle activation due to dirty freeing L1 | |
923 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
924 | */ | |
539d33c7 | 925 | mutex_enter(&dp->dp_lock); |
f4c594da | 926 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += |
65282ee9 | 927 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
928 | mutex_exit(&dp->dp_lock); |
929 | DTRACE_PROBE3(free__long__range, | |
f4c594da TC |
930 | uint64_t, long_free_dirty, uint64_t, chunk_len, |
931 | uint64_t, txg); | |
539d33c7 | 932 | dnode_free_range(dn, chunk_begin, chunk_len, tx); |
440a3eb9 | 933 | |
b128c09f | 934 | dmu_tx_commit(tx); |
b663a23d | 935 | |
539d33c7 | 936 | length -= chunk_len; |
b128c09f BB |
937 | } |
938 | return (0); | |
939 | } | |
940 | ||
941 | int | |
942 | dmu_free_long_range(objset_t *os, uint64_t object, | |
943 | uint64_t offset, uint64_t length) | |
944 | { | |
945 | dnode_t *dn; | |
946 | int err; | |
947 | ||
428870ff | 948 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
949 | if (err != 0) |
950 | return (err); | |
0c03d21a | 951 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
952 | |
953 | /* | |
954 | * It is important to zero out the maxblkid when freeing the entire | |
955 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
956 | * will take the fast path, and (b) dnode_reallocate() can verify | |
957 | * that the entire file has been freed. | |
958 | */ | |
b0bc7a84 | 959 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
960 | dn->dn_maxblkid = 0; |
961 | ||
b128c09f BB |
962 | dnode_rele(dn, FTAG); |
963 | return (err); | |
964 | } | |
965 | ||
966 | int | |
0c03d21a | 967 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 968 | { |
b128c09f BB |
969 | dmu_tx_t *tx; |
970 | int err; | |
971 | ||
b663a23d | 972 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
973 | if (err != 0) |
974 | return (err); | |
b663a23d MA |
975 | |
976 | tx = dmu_tx_create(os); | |
977 | dmu_tx_hold_bonus(tx, object); | |
978 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 979 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
980 | err = dmu_tx_assign(tx, TXG_WAIT); |
981 | if (err == 0) { | |
60b618a9 | 982 | err = dmu_object_free(os, object, tx); |
b663a23d | 983 | dmu_tx_commit(tx); |
b128c09f | 984 | } else { |
b663a23d | 985 | dmu_tx_abort(tx); |
b128c09f | 986 | } |
b663a23d | 987 | |
b128c09f BB |
988 | return (err); |
989 | } | |
990 | ||
34dc7c2f BB |
991 | int |
992 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
993 | uint64_t size, dmu_tx_t *tx) | |
994 | { | |
995 | dnode_t *dn; | |
428870ff | 996 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
997 | if (err) |
998 | return (err); | |
999 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 1000 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
1001 | dnode_free_range(dn, offset, size, tx); |
1002 | dnode_rele(dn, FTAG); | |
1003 | return (0); | |
1004 | } | |
1005 | ||
0eef1bde | 1006 | static int |
1007 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 1008 | void *buf, uint32_t flags) |
34dc7c2f | 1009 | { |
34dc7c2f | 1010 | dmu_buf_t **dbp; |
0eef1bde | 1011 | int numbufs, err = 0; |
34dc7c2f BB |
1012 | |
1013 | /* | |
1014 | * Deal with odd block sizes, where there can't be data past the first | |
1015 | * block. If we ever do the tail block optimization, we will need to | |
1016 | * handle that here as well. | |
1017 | */ | |
45d1cae3 | 1018 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 1019 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f | 1020 | MIN(size, dn->dn_datablksz - offset); |
861166b0 | 1021 | memset((char *)buf + newsz, 0, size - newsz); |
34dc7c2f BB |
1022 | size = newsz; |
1023 | } | |
1024 | ||
1025 | while (size > 0) { | |
1026 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 1027 | int i; |
34dc7c2f BB |
1028 | |
1029 | /* | |
1030 | * NB: we could do this block-at-a-time, but it's nice | |
1031 | * to be reading in parallel. | |
1032 | */ | |
1033 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 1034 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
1035 | if (err) |
1036 | break; | |
1037 | ||
1038 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1039 | uint64_t tocpy; |
1040 | int64_t bufoff; | |
34dc7c2f BB |
1041 | dmu_buf_t *db = dbp[i]; |
1042 | ||
1043 | ASSERT(size > 0); | |
1044 | ||
1045 | bufoff = offset - db->db_offset; | |
c9520ecc | 1046 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 1047 | |
c9520ecc | 1048 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
1049 | |
1050 | offset += tocpy; | |
1051 | size -= tocpy; | |
1052 | buf = (char *)buf + tocpy; | |
1053 | } | |
1054 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1055 | } | |
34dc7c2f BB |
1056 | return (err); |
1057 | } | |
1058 | ||
0eef1bde | 1059 | int |
1060 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1061 | void *buf, uint32_t flags) | |
34dc7c2f | 1062 | { |
0eef1bde | 1063 | dnode_t *dn; |
1064 | int err; | |
34dc7c2f | 1065 | |
0eef1bde | 1066 | err = dnode_hold(os, object, FTAG, &dn); |
1067 | if (err != 0) | |
1068 | return (err); | |
34dc7c2f | 1069 | |
0eef1bde | 1070 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1071 | dnode_rele(dn, FTAG); | |
1072 | return (err); | |
1073 | } | |
1074 | ||
1075 | int | |
1076 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1077 | uint32_t flags) | |
1078 | { | |
1079 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1080 | } | |
1081 | ||
1082 | static void | |
1083 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1084 | const void *buf, dmu_tx_t *tx) | |
1085 | { | |
1086 | int i; | |
34dc7c2f BB |
1087 | |
1088 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1089 | uint64_t tocpy; |
1090 | int64_t bufoff; | |
34dc7c2f BB |
1091 | dmu_buf_t *db = dbp[i]; |
1092 | ||
1093 | ASSERT(size > 0); | |
1094 | ||
1095 | bufoff = offset - db->db_offset; | |
c9520ecc | 1096 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1097 | |
1098 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1099 | ||
1100 | if (tocpy == db->db_size) | |
1101 | dmu_buf_will_fill(db, tx); | |
1102 | else | |
1103 | dmu_buf_will_dirty(db, tx); | |
1104 | ||
60101509 | 1105 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1106 | |
1107 | if (tocpy == db->db_size) | |
1108 | dmu_buf_fill_done(db, tx); | |
1109 | ||
1110 | offset += tocpy; | |
1111 | size -= tocpy; | |
1112 | buf = (char *)buf + tocpy; | |
1113 | } | |
0eef1bde | 1114 | } |
1115 | ||
1116 | void | |
1117 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1118 | const void *buf, dmu_tx_t *tx) | |
1119 | { | |
1120 | dmu_buf_t **dbp; | |
1121 | int numbufs; | |
1122 | ||
1123 | if (size == 0) | |
1124 | return; | |
1125 | ||
1126 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1127 | FALSE, FTAG, &numbufs, &dbp)); | |
1128 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1129 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1130 | } | |
1131 | ||
0f8ff49e SD |
1132 | /* |
1133 | * Note: Lustre is an external consumer of this interface. | |
1134 | */ | |
0eef1bde | 1135 | void |
1136 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1137 | const void *buf, dmu_tx_t *tx) | |
1138 | { | |
1139 | dmu_buf_t **dbp; | |
1140 | int numbufs; | |
1141 | ||
1142 | if (size == 0) | |
1143 | return; | |
1144 | ||
1145 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1146 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1147 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1148 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1149 | } | |
1150 | ||
b128c09f BB |
1151 | void |
1152 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1153 | dmu_tx_t *tx) | |
1154 | { | |
1155 | dmu_buf_t **dbp; | |
1156 | int numbufs, i; | |
1157 | ||
1158 | if (size == 0) | |
1159 | return; | |
1160 | ||
1161 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1162 | FALSE, FTAG, &numbufs, &dbp)); | |
1163 | ||
1164 | for (i = 0; i < numbufs; i++) { | |
1165 | dmu_buf_t *db = dbp[i]; | |
1166 | ||
1167 | dmu_buf_will_not_fill(db, tx); | |
1168 | } | |
1169 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1170 | } | |
1171 | ||
9b67f605 MA |
1172 | void |
1173 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1174 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1175 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1176 | { | |
1177 | dmu_buf_t *db; | |
1178 | ||
1179 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1180 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1181 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1182 | FTAG, &db)); | |
1183 | ||
1184 | dmu_buf_write_embedded(db, | |
1185 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1186 | uncompressed_size, compressed_size, byteorder, tx); | |
1187 | ||
1188 | dmu_buf_rele(db, FTAG); | |
1189 | } | |
1190 | ||
30af21b0 PD |
1191 | void |
1192 | dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1193 | dmu_tx_t *tx) | |
1194 | { | |
1195 | int numbufs, i; | |
1196 | dmu_buf_t **dbp; | |
1197 | ||
1198 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1199 | &numbufs, &dbp)); | |
1200 | for (i = 0; i < numbufs; i++) | |
1201 | dmu_buf_redact(dbp[i], tx); | |
1202 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1203 | } | |
1204 | ||
34dc7c2f | 1205 | #ifdef _KERNEL |
5228cf01 | 1206 | int |
d0cd9a5c | 1207 | dmu_read_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size) |
872e8d26 BB |
1208 | { |
1209 | dmu_buf_t **dbp; | |
1210 | int numbufs, i, err; | |
872e8d26 BB |
1211 | |
1212 | /* | |
1213 | * NB: we could do this block-at-a-time, but it's nice | |
1214 | * to be reading in parallel. | |
1215 | */ | |
d0cd9a5c | 1216 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
804e0504 | 1217 | TRUE, FTAG, &numbufs, &dbp, 0); |
872e8d26 BB |
1218 | if (err) |
1219 | return (err); | |
1220 | ||
1221 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1222 | uint64_t tocpy; |
1223 | int64_t bufoff; | |
872e8d26 BB |
1224 | dmu_buf_t *db = dbp[i]; |
1225 | ||
1226 | ASSERT(size > 0); | |
1227 | ||
d0cd9a5c | 1228 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1229 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1230 | |
d0cd9a5c BA |
1231 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, tocpy, |
1232 | UIO_READ, uio); | |
1233 | ||
872e8d26 BB |
1234 | if (err) |
1235 | break; | |
1236 | ||
1237 | size -= tocpy; | |
1238 | } | |
1239 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1240 | ||
1241 | return (err); | |
1242 | } | |
1243 | ||
804e0504 MA |
1244 | /* |
1245 | * Read 'size' bytes into the uio buffer. | |
1246 | * From object zdb->db_object. | |
d0cd9a5c | 1247 | * Starting at zfs_uio_offset(uio). |
804e0504 MA |
1248 | * |
1249 | * If the caller already has a dbuf in the target object | |
1250 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1251 | * because we don't have to find the dnode_t for the object. | |
1252 | */ | |
1253 | int | |
d0cd9a5c | 1254 | dmu_read_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1255 | { |
1256 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1257 | dnode_t *dn; | |
1258 | int err; | |
1259 | ||
1260 | if (size == 0) | |
1261 | return (0); | |
1262 | ||
1263 | DB_DNODE_ENTER(db); | |
1264 | dn = DB_DNODE(db); | |
1265 | err = dmu_read_uio_dnode(dn, uio, size); | |
1266 | DB_DNODE_EXIT(db); | |
1267 | ||
1268 | return (err); | |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * Read 'size' bytes into the uio buffer. | |
1273 | * From the specified object | |
d0cd9a5c | 1274 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1275 | */ |
1276 | int | |
d0cd9a5c | 1277 | dmu_read_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1278 | { |
1279 | dnode_t *dn; | |
1280 | int err; | |
1281 | ||
1282 | if (size == 0) | |
1283 | return (0); | |
1284 | ||
1285 | err = dnode_hold(os, object, FTAG, &dn); | |
1286 | if (err) | |
1287 | return (err); | |
1288 | ||
1289 | err = dmu_read_uio_dnode(dn, uio, size); | |
1290 | ||
1291 | dnode_rele(dn, FTAG); | |
1292 | ||
1293 | return (err); | |
1294 | } | |
1295 | ||
5228cf01 | 1296 | int |
d0cd9a5c | 1297 | dmu_write_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size, dmu_tx_t *tx) |
872e8d26 BB |
1298 | { |
1299 | dmu_buf_t **dbp; | |
1300 | int numbufs; | |
1301 | int err = 0; | |
1302 | int i; | |
1303 | ||
d0cd9a5c | 1304 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
872e8d26 BB |
1305 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); |
1306 | if (err) | |
1307 | return (err); | |
1308 | ||
1309 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1310 | uint64_t tocpy; |
1311 | int64_t bufoff; | |
872e8d26 BB |
1312 | dmu_buf_t *db = dbp[i]; |
1313 | ||
1314 | ASSERT(size > 0); | |
1315 | ||
d0cd9a5c | 1316 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1317 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1318 | |
1319 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1320 | ||
1321 | if (tocpy == db->db_size) | |
1322 | dmu_buf_will_fill(db, tx); | |
1323 | else | |
1324 | dmu_buf_will_dirty(db, tx); | |
1325 | ||
1326 | /* | |
d0cd9a5c | 1327 | * XXX zfs_uiomove could block forever (eg.nfs-backed |
872e8d26 | 1328 | * pages). There needs to be a uiolockdown() function |
d0cd9a5c | 1329 | * to lock the pages in memory, so that zfs_uiomove won't |
872e8d26 BB |
1330 | * block. |
1331 | */ | |
d0cd9a5c BA |
1332 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, |
1333 | tocpy, UIO_WRITE, uio); | |
1334 | ||
872e8d26 BB |
1335 | if (tocpy == db->db_size) |
1336 | dmu_buf_fill_done(db, tx); | |
1337 | ||
1338 | if (err) | |
1339 | break; | |
1340 | ||
1341 | size -= tocpy; | |
1342 | } | |
1343 | ||
1344 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1345 | return (err); | |
1346 | } | |
1347 | ||
804e0504 MA |
1348 | /* |
1349 | * Write 'size' bytes from the uio buffer. | |
1350 | * To object zdb->db_object. | |
d0cd9a5c | 1351 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1352 | * |
1353 | * If the caller already has a dbuf in the target object | |
1354 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1355 | * because we don't have to find the dnode_t for the object. | |
1356 | */ | |
428870ff | 1357 | int |
d0cd9a5c | 1358 | dmu_write_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1359 | dmu_tx_t *tx) |
1360 | { | |
572e2857 BB |
1361 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1362 | dnode_t *dn; | |
1363 | int err; | |
1364 | ||
428870ff BB |
1365 | if (size == 0) |
1366 | return (0); | |
1367 | ||
572e2857 BB |
1368 | DB_DNODE_ENTER(db); |
1369 | dn = DB_DNODE(db); | |
1370 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1371 | DB_DNODE_EXIT(db); | |
1372 | ||
1373 | return (err); | |
428870ff BB |
1374 | } |
1375 | ||
804e0504 MA |
1376 | /* |
1377 | * Write 'size' bytes from the uio buffer. | |
1378 | * To the specified object. | |
d0cd9a5c | 1379 | * Starting at offset zfs_uio_offset(uio). |
804e0504 | 1380 | */ |
428870ff | 1381 | int |
d0cd9a5c | 1382 | dmu_write_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1383 | dmu_tx_t *tx) |
1384 | { | |
1385 | dnode_t *dn; | |
1386 | int err; | |
1387 | ||
1388 | if (size == 0) | |
1389 | return (0); | |
1390 | ||
1391 | err = dnode_hold(os, object, FTAG, &dn); | |
1392 | if (err) | |
1393 | return (err); | |
1394 | ||
1395 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1396 | ||
1397 | dnode_rele(dn, FTAG); | |
1398 | ||
1399 | return (err); | |
1400 | } | |
872e8d26 | 1401 | #endif /* _KERNEL */ |
34dc7c2f | 1402 | |
9babb374 BB |
1403 | /* |
1404 | * Allocate a loaned anonymous arc buffer. | |
1405 | */ | |
1406 | arc_buf_t * | |
1407 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1408 | { | |
572e2857 | 1409 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1410 | |
2aa34383 | 1411 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1412 | } |
1413 | ||
1414 | /* | |
1415 | * Free a loaned arc buffer. | |
1416 | */ | |
1417 | void | |
1418 | dmu_return_arcbuf(arc_buf_t *buf) | |
1419 | { | |
1420 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1421 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1422 | } |
1423 | ||
ba67d821 MA |
1424 | /* |
1425 | * A "lightweight" write is faster than a regular write (e.g. | |
1426 | * dmu_write_by_dnode() or dmu_assign_arcbuf_by_dnode()), because it avoids the | |
1427 | * CPU cost of creating a dmu_buf_impl_t and arc_buf_[hdr_]_t. However, the | |
1428 | * data can not be read or overwritten until the transaction's txg has been | |
1429 | * synced. This makes it appropriate for workloads that are known to be | |
1430 | * (temporarily) write-only, like "zfs receive". | |
1431 | * | |
1432 | * A single block is written, starting at the specified offset in bytes. If | |
1433 | * the call is successful, it returns 0 and the provided abd has been | |
1434 | * consumed (the caller should not free it). | |
1435 | */ | |
1436 | int | |
1437 | dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, | |
1438 | const zio_prop_t *zp, enum zio_flag flags, dmu_tx_t *tx) | |
1439 | { | |
1440 | dbuf_dirty_record_t *dr = | |
1441 | dbuf_dirty_lightweight(dn, dbuf_whichblock(dn, 0, offset), tx); | |
1442 | if (dr == NULL) | |
1443 | return (SET_ERROR(EIO)); | |
1444 | dr->dt.dll.dr_abd = abd; | |
1445 | dr->dt.dll.dr_props = *zp; | |
1446 | dr->dt.dll.dr_flags = flags; | |
1447 | return (0); | |
1448 | } | |
1449 | ||
9babb374 BB |
1450 | /* |
1451 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1452 | * If this is not possible copy the contents of passed arc buf via | |
1453 | * dmu_write(). | |
1454 | */ | |
305781da | 1455 | int |
440a3eb9 | 1456 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1457 | dmu_tx_t *tx) |
1458 | { | |
9babb374 | 1459 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1460 | objset_t *os = dn->dn_objset; |
1461 | uint64_t object = dn->dn_object; | |
2aa34383 | 1462 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1463 | uint64_t blkid; |
1464 | ||
1465 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1466 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1467 | db = dbuf_hold(dn, blkid, FTAG); |
1468 | if (db == NULL) | |
1469 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1470 | rw_exit(&dn->dn_struct_rwlock); |
1471 | ||
88904bb3 | 1472 | /* |
ba67d821 MA |
1473 | * We can only assign if the offset is aligned and the arc buf is the |
1474 | * same size as the dbuf. | |
88904bb3 | 1475 | */ |
2aa34383 | 1476 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
64e0fe14 | 1477 | zfs_racct_write(blksz, 1); |
9babb374 BB |
1478 | dbuf_assign_arcbuf(db, buf, tx); |
1479 | dbuf_rele(db, FTAG); | |
1480 | } else { | |
2aa34383 DK |
1481 | /* compressed bufs must always be assignable to their dbuf */ |
1482 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1483 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1484 | |
9babb374 | 1485 | dbuf_rele(db, FTAG); |
572e2857 | 1486 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 BB |
1487 | dmu_return_arcbuf(buf); |
1488 | } | |
305781da TC |
1489 | |
1490 | return (0); | |
9babb374 BB |
1491 | } |
1492 | ||
305781da | 1493 | int |
440a3eb9 TC |
1494 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1495 | dmu_tx_t *tx) | |
1496 | { | |
305781da | 1497 | int err; |
440a3eb9 TC |
1498 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1499 | ||
1500 | DB_DNODE_ENTER(dbuf); | |
305781da | 1501 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1502 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1503 | |
1504 | return (err); | |
440a3eb9 TC |
1505 | } |
1506 | ||
34dc7c2f | 1507 | typedef struct { |
428870ff BB |
1508 | dbuf_dirty_record_t *dsa_dr; |
1509 | dmu_sync_cb_t *dsa_done; | |
1510 | zgd_t *dsa_zgd; | |
1511 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1512 | } dmu_sync_arg_t; |
1513 | ||
b128c09f BB |
1514 | static void |
1515 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1516 | { | |
14e4e3cb | 1517 | (void) buf; |
428870ff BB |
1518 | dmu_sync_arg_t *dsa = varg; |
1519 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1520 | blkptr_t *bp = zio->io_bp; |
1521 | ||
428870ff BB |
1522 | if (zio->io_error == 0) { |
1523 | if (BP_IS_HOLE(bp)) { | |
1524 | /* | |
1525 | * A block of zeros may compress to a hole, but the | |
1526 | * block size still needs to be known for replay. | |
1527 | */ | |
1528 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1529 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1530 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1531 | BP_SET_FILL(bp, 1); |
428870ff | 1532 | } |
b128c09f BB |
1533 | } |
1534 | } | |
1535 | ||
428870ff BB |
1536 | static void |
1537 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1538 | { | |
1539 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1540 | } | |
1541 | ||
34dc7c2f BB |
1542 | static void |
1543 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1544 | { | |
14e4e3cb | 1545 | (void) buf; |
428870ff BB |
1546 | dmu_sync_arg_t *dsa = varg; |
1547 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1548 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1549 | zgd_t *zgd = dsa->dsa_zgd; |
1550 | ||
1551 | /* | |
1552 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1553 | * the writes for the lwb have completed. | |
1554 | */ | |
1555 | if (zio->io_error == 0) { | |
1556 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1557 | } | |
34dc7c2f | 1558 | |
34dc7c2f BB |
1559 | mutex_enter(&db->db_mtx); |
1560 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1561 | if (zio->io_error == 0) { |
03c6040b GW |
1562 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1563 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1564 | blkptr_t *bp = zio->io_bp; |
1565 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1566 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1567 | |
1568 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1569 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1570 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1571 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1572 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1573 | } |
428870ff BB |
1574 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1575 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1576 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1577 | |
1578 | /* | |
1579 | * Old style holes are filled with all zeros, whereas | |
1580 | * new-style holes maintain their lsize, type, level, | |
1581 | * and birth time (see zio_write_compress). While we | |
1582 | * need to reset the BP_SET_LSIZE() call that happened | |
1583 | * in dmu_sync_ready for old style holes, we do *not* | |
1584 | * want to wipe out the information contained in new | |
1585 | * style holes. Thus, only zero out the block pointer if | |
1586 | * it's an old style hole. | |
1587 | */ | |
1588 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1589 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1590 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1591 | } else { | |
1592 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1593 | } | |
34dc7c2f BB |
1594 | cv_broadcast(&db->db_changed); |
1595 | mutex_exit(&db->db_mtx); | |
1596 | ||
428870ff | 1597 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1598 | |
428870ff BB |
1599 | kmem_free(dsa, sizeof (*dsa)); |
1600 | } | |
1601 | ||
1602 | static void | |
1603 | dmu_sync_late_arrival_done(zio_t *zio) | |
1604 | { | |
1605 | blkptr_t *bp = zio->io_bp; | |
1606 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1607 | zgd_t *zgd = dsa->dsa_zgd; |
1608 | ||
1609 | if (zio->io_error == 0) { | |
1610 | /* | |
1611 | * Record the vdev(s) backing this blkptr so they can be | |
1612 | * flushed after the writes for the lwb have completed. | |
1613 | */ | |
1614 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1615 | ||
1616 | if (!BP_IS_HOLE(bp)) { | |
2a8ba608 | 1617 | blkptr_t *bp_orig __maybe_unused = &zio->io_bp_orig; |
900d09b2 PS |
1618 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); |
1619 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1620 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1621 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1622 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1623 | } | |
428870ff BB |
1624 | } |
1625 | ||
1626 | dmu_tx_commit(dsa->dsa_tx); | |
1627 | ||
1628 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1629 | ||
e2af2acc | 1630 | abd_free(zio->io_abd); |
428870ff BB |
1631 | kmem_free(dsa, sizeof (*dsa)); |
1632 | } | |
1633 | ||
1634 | static int | |
1635 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1636 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1637 | { |
1638 | dmu_sync_arg_t *dsa; | |
1639 | dmu_tx_t *tx; | |
1640 | ||
1641 | tx = dmu_tx_create(os); | |
1642 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1643 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1644 | dmu_tx_abort(tx); | |
2e528b49 MA |
1645 | /* Make zl_get_data do txg_waited_synced() */ |
1646 | return (SET_ERROR(EIO)); | |
428870ff BB |
1647 | } |
1648 | ||
1ce23dca PS |
1649 | /* |
1650 | * In order to prevent the zgd's lwb from being free'd prior to | |
1651 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1652 | * the lwb's "max txg" takes this tx's txg into account. | |
1653 | */ | |
1654 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1655 | ||
79c76d5b | 1656 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1657 | dsa->dsa_dr = NULL; |
1658 | dsa->dsa_done = done; | |
1659 | dsa->dsa_zgd = zgd; | |
1660 | dsa->dsa_tx = tx; | |
1661 | ||
02dc43bc MA |
1662 | /* |
1663 | * Since we are currently syncing this txg, it's nontrivial to | |
1664 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1665 | * | |
1666 | * When syncing, the db_blkptr is initially the BP of the previous | |
1667 | * txg. We can not nopwrite against it because it will be changed | |
1668 | * (this is similar to the non-late-arrival case where the dbuf is | |
1669 | * dirty in a future txg). | |
1670 | * | |
1671 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1672 | * We can not nopwrite against it because although the BP will not | |
1673 | * (typically) be changed, the data has not yet been persisted to this | |
1674 | * location. | |
1675 | * | |
1676 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1677 | * possible to always nopwrite, because the data that was written in | |
1678 | * this txg is the same data that we are trying to write. However we | |
1679 | * would need to check that this dbuf is not dirty in any future | |
1680 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1681 | * don't nopwrite in this case. | |
1682 | */ | |
1683 | zp->zp_nopwrite = B_FALSE; | |
1684 | ||
a6255b7f DQ |
1685 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1686 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1687 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1688 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1689 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1690 | |
1691 | return (0); | |
34dc7c2f BB |
1692 | } |
1693 | ||
1694 | /* | |
1695 | * Intent log support: sync the block associated with db to disk. | |
1696 | * N.B. and XXX: the caller is responsible for making sure that the | |
1697 | * data isn't changing while dmu_sync() is writing it. | |
1698 | * | |
1699 | * Return values: | |
1700 | * | |
03c6040b | 1701 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1702 | * The caller should not log the write. |
1703 | * | |
1704 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1705 | * The caller should not log the write. | |
1706 | * | |
1707 | * EALREADY: this block is already in the process of being synced. | |
1708 | * The caller should track its progress (somehow). | |
1709 | * | |
428870ff BB |
1710 | * EIO: could not do the I/O. |
1711 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1712 | * |
428870ff BB |
1713 | * 0: the I/O has been initiated. |
1714 | * The caller should log this blkptr in the done callback. | |
1715 | * It is possible that the I/O will fail, in which case | |
1716 | * the error will be reported to the done callback and | |
1717 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1718 | */ |
1719 | int | |
428870ff | 1720 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1721 | { |
428870ff BB |
1722 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1723 | objset_t *os = db->db_objset; | |
1724 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
cccbed9f | 1725 | dbuf_dirty_record_t *dr, *dr_next; |
428870ff | 1726 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1727 | zbookmark_phys_t zb; |
428870ff | 1728 | zio_prop_t zp; |
572e2857 | 1729 | dnode_t *dn; |
34dc7c2f | 1730 | |
428870ff | 1731 | ASSERT(pio != NULL); |
34dc7c2f BB |
1732 | ASSERT(txg != 0); |
1733 | ||
428870ff BB |
1734 | SET_BOOKMARK(&zb, ds->ds_object, |
1735 | db->db.db_object, db->db_level, db->db_blkid); | |
1736 | ||
572e2857 BB |
1737 | DB_DNODE_ENTER(db); |
1738 | dn = DB_DNODE(db); | |
82644107 | 1739 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1740 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1741 | |
1742 | /* | |
428870ff | 1743 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1744 | */ |
428870ff BB |
1745 | if (txg > spa_freeze_txg(os->os_spa)) |
1746 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1747 | |
1748 | /* | |
428870ff BB |
1749 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1750 | * and us. If we determine that this txg is not yet syncing, | |
1751 | * but it begins to sync a moment later, that's OK because the | |
1752 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1753 | */ |
428870ff BB |
1754 | mutex_enter(&db->db_mtx); |
1755 | ||
1756 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1757 | /* |
428870ff | 1758 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1759 | */ |
428870ff | 1760 | mutex_exit(&db->db_mtx); |
2e528b49 | 1761 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1762 | } |
1763 | ||
428870ff BB |
1764 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1765 | /* | |
1766 | * This txg is currently syncing, so we can't mess with | |
1767 | * the dirty record anymore; just write a new log block. | |
1768 | */ | |
1769 | mutex_exit(&db->db_mtx); | |
1770 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1771 | } |
1772 | ||
cccbed9f | 1773 | dr = dbuf_find_dirty_eq(db, txg); |
428870ff BB |
1774 | |
1775 | if (dr == NULL) { | |
34dc7c2f | 1776 | /* |
428870ff | 1777 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1778 | * There's no need to log writes to freed blocks, so we're done. |
1779 | */ | |
1780 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1781 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1782 | } |
1783 | ||
cccbed9f MM |
1784 | dr_next = list_next(&db->db_dirty_records, dr); |
1785 | ASSERT(dr_next == NULL || dr_next->dr_txg < txg); | |
03c6040b | 1786 | |
02dc43bc MA |
1787 | if (db->db_blkptr != NULL) { |
1788 | /* | |
1789 | * We need to fill in zgd_bp with the current blkptr so that | |
1790 | * the nopwrite code can check if we're writing the same | |
1791 | * data that's already on disk. We can only nopwrite if we | |
1792 | * are sure that after making the copy, db_blkptr will not | |
1793 | * change until our i/o completes. We ensure this by | |
1794 | * holding the db_mtx, and only allowing nopwrite if the | |
1795 | * block is not already dirty (see below). This is verified | |
1796 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1797 | * not changed. | |
1798 | */ | |
1799 | *zgd->zgd_bp = *db->db_blkptr; | |
1800 | } | |
1801 | ||
03c6040b | 1802 | /* |
f3c517d8 MA |
1803 | * Assume the on-disk data is X, the current syncing data (in |
1804 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1805 | * in dmu_sync). | |
1806 | * | |
1807 | * We usually want to perform a nopwrite if X and Z are the | |
1808 | * same. However, if Y is different (i.e. the BP is going to | |
1809 | * change before this write takes effect), then a nopwrite will | |
1810 | * be incorrect - we would override with X, which could have | |
1811 | * been freed when Y was written. | |
1812 | * | |
1813 | * (Note that this is not a concern when we are nop-writing from | |
1814 | * syncing context, because X and Y must be identical, because | |
1815 | * all previous txgs have been synced.) | |
1816 | * | |
1817 | * Therefore, we disable nopwrite if the current BP could change | |
1818 | * before this TXG. There are two ways it could change: by | |
1819 | * being dirty (dr_next is non-NULL), or by being freed | |
1820 | * (dnode_block_freed()). This behavior is verified by | |
1821 | * zio_done(), which VERIFYs that the override BP is identical | |
1822 | * to the on-disk BP. | |
03c6040b | 1823 | */ |
f3c517d8 MA |
1824 | DB_DNODE_ENTER(db); |
1825 | dn = DB_DNODE(db); | |
cccbed9f | 1826 | if (dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) |
03c6040b | 1827 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1828 | DB_DNODE_EXIT(db); |
03c6040b | 1829 | |
34dc7c2f | 1830 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1831 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1832 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1833 | /* |
428870ff BB |
1834 | * We have already issued a sync write for this buffer, |
1835 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1836 | * have been dirtied since, or we would have cleared the state. |
1837 | */ | |
34dc7c2f | 1838 | mutex_exit(&db->db_mtx); |
2e528b49 | 1839 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1840 | } |
1841 | ||
428870ff | 1842 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1843 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1844 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1845 | |
79c76d5b | 1846 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1847 | dsa->dsa_dr = dr; |
1848 | dsa->dsa_done = done; | |
1849 | dsa->dsa_zgd = zgd; | |
1850 | dsa->dsa_tx = NULL; | |
b128c09f | 1851 | |
428870ff | 1852 | zio_nowait(arc_write(pio, os->os_spa, txg, |
c9d62d13 | 1853 | zgd->zgd_bp, dr->dt.dl.dr_data, dbuf_is_l2cacheable(db), |
d3c2ae1c | 1854 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 1855 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 1856 | |
428870ff | 1857 | return (0); |
34dc7c2f BB |
1858 | } |
1859 | ||
b5256303 TC |
1860 | int |
1861 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
1862 | { | |
1863 | dnode_t *dn; | |
1864 | int err; | |
1865 | ||
1866 | err = dnode_hold(os, object, FTAG, &dn); | |
1867 | if (err) | |
1868 | return (err); | |
1869 | err = dnode_set_nlevels(dn, nlevels, tx); | |
1870 | dnode_rele(dn, FTAG); | |
1871 | return (err); | |
1872 | } | |
1873 | ||
34dc7c2f BB |
1874 | int |
1875 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 1876 | dmu_tx_t *tx) |
34dc7c2f BB |
1877 | { |
1878 | dnode_t *dn; | |
1879 | int err; | |
1880 | ||
428870ff | 1881 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1882 | if (err) |
1883 | return (err); | |
1884 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1885 | dnode_rele(dn, FTAG); | |
1886 | return (err); | |
1887 | } | |
1888 | ||
ae76f45c TC |
1889 | int |
1890 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
1891 | dmu_tx_t *tx) | |
1892 | { | |
1893 | dnode_t *dn; | |
1894 | int err; | |
1895 | ||
1896 | err = dnode_hold(os, object, FTAG, &dn); | |
1897 | if (err) | |
1898 | return (err); | |
1899 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 1900 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
1901 | rw_exit(&dn->dn_struct_rwlock); |
1902 | dnode_rele(dn, FTAG); | |
1903 | return (0); | |
1904 | } | |
1905 | ||
34dc7c2f BB |
1906 | void |
1907 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 1908 | dmu_tx_t *tx) |
34dc7c2f BB |
1909 | { |
1910 | dnode_t *dn; | |
1911 | ||
9b67f605 MA |
1912 | /* |
1913 | * Send streams include each object's checksum function. This | |
1914 | * check ensures that the receiving system can understand the | |
1915 | * checksum function transmitted. | |
1916 | */ | |
1917 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1918 | ||
1919 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1920 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1921 | dn->dn_checksum = checksum; |
1922 | dnode_setdirty(dn, tx); | |
1923 | dnode_rele(dn, FTAG); | |
1924 | } | |
1925 | ||
1926 | void | |
1927 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 1928 | dmu_tx_t *tx) |
34dc7c2f BB |
1929 | { |
1930 | dnode_t *dn; | |
1931 | ||
9b67f605 MA |
1932 | /* |
1933 | * Send streams include each object's compression function. This | |
1934 | * check ensures that the receiving system can understand the | |
1935 | * compression function transmitted. | |
1936 | */ | |
1937 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1938 | ||
1939 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
1940 | dn->dn_compress = compress; |
1941 | dnode_setdirty(dn, tx); | |
1942 | dnode_rele(dn, FTAG); | |
1943 | } | |
1944 | ||
faf0f58c MA |
1945 | /* |
1946 | * When the "redundant_metadata" property is set to "most", only indirect | |
1947 | * blocks of this level and higher will have an additional ditto block. | |
1948 | */ | |
18168da7 | 1949 | static const int zfs_redundant_metadata_most_ditto_level = 2; |
faf0f58c | 1950 | |
428870ff | 1951 | void |
82644107 | 1952 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
1953 | { |
1954 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 1955 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 1956 | (wp & WP_SPILL)); |
428870ff BB |
1957 | enum zio_checksum checksum = os->os_checksum; |
1958 | enum zio_compress compress = os->os_compress; | |
10b3c7f5 | 1959 | uint8_t complevel = os->os_complevel; |
428870ff | 1960 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; |
03c6040b GW |
1961 | boolean_t dedup = B_FALSE; |
1962 | boolean_t nopwrite = B_FALSE; | |
428870ff | 1963 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 1964 | boolean_t encrypt = B_FALSE; |
428870ff | 1965 | int copies = os->os_copies; |
a7004725 | 1966 | |
428870ff | 1967 | /* |
03c6040b GW |
1968 | * We maintain different write policies for each of the following |
1969 | * types of data: | |
1970 | * 1. metadata | |
1971 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
1972 | * 3. all other level 0 blocks | |
428870ff BB |
1973 | */ |
1974 | if (ismd) { | |
b1d21733 TC |
1975 | /* |
1976 | * XXX -- we should design a compression algorithm | |
1977 | * that specializes in arrays of bps. | |
1978 | */ | |
1979 | compress = zio_compress_select(os->os_spa, | |
1980 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 1981 | |
428870ff BB |
1982 | /* |
1983 | * Metadata always gets checksummed. If the data | |
1984 | * checksum is multi-bit correctable, and it's not a | |
1985 | * ZBT-style checksum, then it's suitable for metadata | |
1986 | * as well. Otherwise, the metadata checksum defaults | |
1987 | * to fletcher4. | |
1988 | */ | |
3c67d83a TH |
1989 | if (!(zio_checksum_table[checksum].ci_flags & |
1990 | ZCHECKSUM_FLAG_METADATA) || | |
1991 | (zio_checksum_table[checksum].ci_flags & | |
1992 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 1993 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
1994 | |
1995 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
1996 | (os->os_redundant_metadata == | |
1997 | ZFS_REDUNDANT_METADATA_MOST && | |
1998 | (level >= zfs_redundant_metadata_most_ditto_level || | |
1999 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
2000 | copies++; | |
03c6040b GW |
2001 | } else if (wp & WP_NOFILL) { |
2002 | ASSERT(level == 0); | |
428870ff | 2003 | |
428870ff | 2004 | /* |
03c6040b GW |
2005 | * If we're writing preallocated blocks, we aren't actually |
2006 | * writing them so don't set any policy properties. These | |
2007 | * blocks are currently only used by an external subsystem | |
2008 | * outside of zfs (i.e. dump) and not written by the zio | |
2009 | * pipeline. | |
428870ff | 2010 | */ |
03c6040b GW |
2011 | compress = ZIO_COMPRESS_OFF; |
2012 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2013 | } else { |
99197f03 JG |
2014 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2015 | compress); | |
10b3c7f5 MN |
2016 | complevel = zio_complevel_select(os->os_spa, compress, |
2017 | complevel, complevel); | |
428870ff | 2018 | |
03c6040b GW |
2019 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2020 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2021 | dedup_checksum; | |
428870ff | 2022 | |
03c6040b GW |
2023 | /* |
2024 | * Determine dedup setting. If we are in dmu_sync(), | |
2025 | * we won't actually dedup now because that's all | |
2026 | * done in syncing context; but we do want to use the | |
e1cfd73f | 2027 | * dedup checksum. If the checksum is not strong |
03c6040b GW |
2028 | * enough to ensure unique signatures, force |
2029 | * dedup_verify. | |
2030 | */ | |
2031 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2032 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2033 | if (!(zio_checksum_table[checksum].ci_flags & |
2034 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2035 | dedup_verify = B_TRUE; |
2036 | } | |
428870ff | 2037 | |
03c6040b | 2038 | /* |
3c67d83a TH |
2039 | * Enable nopwrite if we have secure enough checksum |
2040 | * algorithm (see comment in zio_nop_write) and | |
2041 | * compression is enabled. We don't enable nopwrite if | |
2042 | * dedup is enabled as the two features are mutually | |
2043 | * exclusive. | |
03c6040b | 2044 | */ |
3c67d83a TH |
2045 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2046 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2047 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2048 | } |
2049 | ||
b5256303 TC |
2050 | /* |
2051 | * All objects in an encrypted objset are protected from modification | |
2052 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2053 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2054 | * not subject to nopwrite since writing the same data will still | |
2055 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2056 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2057 | * object types. | |
2058 | */ | |
2059 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2060 | encrypt = B_TRUE; | |
2061 | ||
2062 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2063 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2064 | nopwrite = B_FALSE; | |
2065 | } else { | |
2066 | dedup = B_FALSE; | |
2067 | } | |
2068 | ||
ae76f45c TC |
2069 | if (level <= 0 && |
2070 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2071 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2072 | } |
b5256303 | 2073 | } |
2aa34383 | 2074 | |
b5256303 | 2075 | zp->zp_compress = compress; |
10b3c7f5 | 2076 | zp->zp_complevel = complevel; |
b5256303 | 2077 | zp->zp_checksum = checksum; |
428870ff BB |
2078 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2079 | zp->zp_level = level; | |
faf0f58c | 2080 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2081 | zp->zp_dedup = dedup; |
2082 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2083 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2084 | zp->zp_encrypt = encrypt; |
2085 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
861166b0 AZ |
2086 | memset(zp->zp_salt, 0, ZIO_DATA_SALT_LEN); |
2087 | memset(zp->zp_iv, 0, ZIO_DATA_IV_LEN); | |
2088 | memset(zp->zp_mac, 0, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2089 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2090 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2091 | |
2092 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2093 | } |
2094 | ||
66aca247 DB |
2095 | /* |
2096 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2097 | * in order to determine the location of holes. In order to accurately | |
2098 | * report holes all dirty data must be synced to disk. This causes extremely | |
2099 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2100 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2101 | * report the dnode as having no holes which is always a safe thing to do. | |
2102 | */ | |
34dc7c2f BB |
2103 | int |
2104 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2105 | { | |
2106 | dnode_t *dn; | |
de198f2d | 2107 | int err; |
34dc7c2f | 2108 | |
de198f2d | 2109 | restart: |
428870ff | 2110 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2111 | if (err) |
2112 | return (err); | |
66aca247 | 2113 | |
de198f2d | 2114 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
66aca247 | 2115 | |
de198f2d BB |
2116 | if (dnode_is_dirty(dn)) { |
2117 | /* | |
2118 | * If the zfs_dmu_offset_next_sync module option is enabled | |
2119 | * then strict hole reporting has been requested. Dirty | |
2120 | * dnodes must be synced to disk to accurately report all | |
05b3eb6d BB |
2121 | * holes. When disabled dirty dnodes are reported to not |
2122 | * have any holes which is always safe. | |
de198f2d BB |
2123 | * |
2124 | * When called by zfs_holey_common() the zp->z_rangelock | |
2125 | * is held to prevent zfs_write() and mmap writeback from | |
2126 | * re-dirtying the dnode after txg_wait_synced(). | |
2127 | */ | |
2128 | if (zfs_dmu_offset_next_sync) { | |
2129 | rw_exit(&dn->dn_struct_rwlock); | |
2130 | dnode_rele(dn, FTAG); | |
2131 | txg_wait_synced(dmu_objset_pool(os), 0); | |
2132 | goto restart; | |
2133 | } | |
34dc7c2f | 2134 | |
66aca247 | 2135 | err = SET_ERROR(EBUSY); |
de198f2d BB |
2136 | } else { |
2137 | err = dnode_next_offset(dn, DNODE_FIND_HAVELOCK | | |
2138 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2139 | } | |
66aca247 | 2140 | |
de198f2d | 2141 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
2142 | dnode_rele(dn, FTAG); |
2143 | ||
2144 | return (err); | |
2145 | } | |
2146 | ||
2147 | void | |
e0b0ca98 | 2148 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2149 | { |
e0b0ca98 | 2150 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2151 | |
34dc7c2f BB |
2152 | doi->doi_data_block_size = dn->dn_datablksz; |
2153 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2154 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2155 | doi->doi_type = dn->dn_type; |
2156 | doi->doi_bonus_type = dn->dn_bonustype; | |
2157 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2158 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2159 | doi->doi_indirection = dn->dn_nlevels; |
2160 | doi->doi_checksum = dn->dn_checksum; | |
2161 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2162 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2163 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2164 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2165 | doi->doi_fill_count = 0; |
1c27024e | 2166 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2167 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2168 | } |
2169 | ||
2170 | void | |
2171 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2172 | { | |
2173 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2174 | mutex_enter(&dn->dn_mtx); | |
2175 | ||
2176 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2177 | |
2178 | mutex_exit(&dn->dn_mtx); | |
2179 | rw_exit(&dn->dn_struct_rwlock); | |
2180 | } | |
2181 | ||
2182 | /* | |
2183 | * Get information on a DMU object. | |
2184 | * If doi is NULL, just indicates whether the object exists. | |
2185 | */ | |
2186 | int | |
2187 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2188 | { | |
2189 | dnode_t *dn; | |
428870ff | 2190 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2191 | |
2192 | if (err) | |
2193 | return (err); | |
2194 | ||
2195 | if (doi != NULL) | |
2196 | dmu_object_info_from_dnode(dn, doi); | |
2197 | ||
2198 | dnode_rele(dn, FTAG); | |
2199 | return (0); | |
2200 | } | |
2201 | ||
2202 | /* | |
2203 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2204 | */ | |
2205 | void | |
572e2857 | 2206 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2207 | { |
572e2857 BB |
2208 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2209 | ||
2210 | DB_DNODE_ENTER(db); | |
2211 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2212 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2213 | } |
2214 | ||
2215 | /* | |
2216 | * Faster still when you only care about the size. | |
34dc7c2f BB |
2217 | */ |
2218 | void | |
572e2857 BB |
2219 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2220 | u_longlong_t *nblk512) | |
34dc7c2f | 2221 | { |
572e2857 BB |
2222 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2223 | dnode_t *dn; | |
2224 | ||
2225 | DB_DNODE_ENTER(db); | |
2226 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2227 | |
2228 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2229 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2230 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2231 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2232 | DB_DNODE_EXIT(db); | |
2233 | } | |
2234 | ||
2235 | void | |
2236 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2237 | { | |
2238 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2239 | dnode_t *dn; | |
2240 | ||
2241 | DB_DNODE_ENTER(db); | |
2242 | dn = DB_DNODE(db); | |
2243 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2244 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2245 | } |
2246 | ||
2247 | void | |
2248 | byteswap_uint64_array(void *vbuf, size_t size) | |
2249 | { | |
2250 | uint64_t *buf = vbuf; | |
2251 | size_t count = size >> 3; | |
2252 | int i; | |
2253 | ||
2254 | ASSERT((size & 7) == 0); | |
2255 | ||
2256 | for (i = 0; i < count; i++) | |
2257 | buf[i] = BSWAP_64(buf[i]); | |
2258 | } | |
2259 | ||
2260 | void | |
2261 | byteswap_uint32_array(void *vbuf, size_t size) | |
2262 | { | |
2263 | uint32_t *buf = vbuf; | |
2264 | size_t count = size >> 2; | |
2265 | int i; | |
2266 | ||
2267 | ASSERT((size & 3) == 0); | |
2268 | ||
2269 | for (i = 0; i < count; i++) | |
2270 | buf[i] = BSWAP_32(buf[i]); | |
2271 | } | |
2272 | ||
2273 | void | |
2274 | byteswap_uint16_array(void *vbuf, size_t size) | |
2275 | { | |
2276 | uint16_t *buf = vbuf; | |
2277 | size_t count = size >> 1; | |
2278 | int i; | |
2279 | ||
2280 | ASSERT((size & 1) == 0); | |
2281 | ||
2282 | for (i = 0; i < count; i++) | |
2283 | buf[i] = BSWAP_16(buf[i]); | |
2284 | } | |
2285 | ||
34dc7c2f BB |
2286 | void |
2287 | byteswap_uint8_array(void *vbuf, size_t size) | |
2288 | { | |
14e4e3cb | 2289 | (void) vbuf, (void) size; |
34dc7c2f BB |
2290 | } |
2291 | ||
2292 | void | |
2293 | dmu_init(void) | |
2294 | { | |
a6255b7f | 2295 | abd_init(); |
428870ff | 2296 | zfs_dbgmsg_init(); |
572e2857 | 2297 | sa_cache_init(); |
572e2857 | 2298 | dmu_objset_init(); |
34dc7c2f | 2299 | dnode_init(); |
428870ff | 2300 | zfetch_init(); |
570827e1 | 2301 | dmu_tx_init(); |
34dc7c2f | 2302 | l2arc_init(); |
29809a6c | 2303 | arc_init(); |
d3c2ae1c | 2304 | dbuf_init(); |
34dc7c2f BB |
2305 | } |
2306 | ||
2307 | void | |
2308 | dmu_fini(void) | |
2309 | { | |
e49f1e20 | 2310 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2311 | l2arc_fini(); |
570827e1 | 2312 | dmu_tx_fini(); |
428870ff | 2313 | zfetch_fini(); |
34dc7c2f | 2314 | dbuf_fini(); |
572e2857 BB |
2315 | dnode_fini(); |
2316 | dmu_objset_fini(); | |
428870ff BB |
2317 | sa_cache_fini(); |
2318 | zfs_dbgmsg_fini(); | |
a6255b7f | 2319 | abd_fini(); |
34dc7c2f | 2320 | } |
c28b2279 | 2321 | |
c28b2279 | 2322 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2323 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2324 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2325 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2326 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2327 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2328 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2329 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2330 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2331 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2332 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2333 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2334 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2335 | EXPORT_SYMBOL(dmu_object_info); |
2336 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2337 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2338 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2339 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2340 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2341 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2342 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2343 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2344 | EXPORT_SYMBOL(dmu_object_set_compress); | |
94dac3e8 | 2345 | EXPORT_SYMBOL(dmu_offset_next); |
57b650b8 BB |
2346 | EXPORT_SYMBOL(dmu_write_policy); |
2347 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2348 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2349 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2350 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2351 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2352 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2353 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2354 | |
03fdcb9a MM |
2355 | ZFS_MODULE_PARAM(zfs, zfs_, nopwrite_enabled, INT, ZMOD_RW, |
2356 | "Enable NOP writes"); | |
03c6040b | 2357 | |
03fdcb9a MM |
2358 | ZFS_MODULE_PARAM(zfs, zfs_, per_txg_dirty_frees_percent, ULONG, ZMOD_RW, |
2359 | "Percentage of dirtied blocks from frees in one TXG"); | |
66aca247 | 2360 | |
03fdcb9a | 2361 | ZFS_MODULE_PARAM(zfs, zfs_, dmu_offset_next_sync, INT, ZMOD_RW, |
66aca247 DB |
2362 | "Enable forcing txg sync to find holes"); |
2363 | ||
7ada752a | 2364 | /* CSTYLED */ |
fdc2d303 | 2365 | ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, UINT, ZMOD_RW, |
d9b4bf06 | 2366 | "Limit one prefetch call to this size"); |