]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
196bee4c | 23 | * Copyright (c) 2011, 2020 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
bc77ba73 | 25 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
a08abc1b | 26 | * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved. |
5475aada | 27 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. |
65282ee9 | 28 | * Copyright (c) 2019 Datto Inc. |
10b3c7f5 MN |
29 | * Copyright (c) 2019, Klara Inc. |
30 | * Copyright (c) 2019, Allan Jude | |
34dc7c2f BB |
31 | */ |
32 | ||
34dc7c2f BB |
33 | #include <sys/dmu.h> |
34 | #include <sys/dmu_impl.h> | |
35 | #include <sys/dmu_tx.h> | |
36 | #include <sys/dbuf.h> | |
37 | #include <sys/dnode.h> | |
38 | #include <sys/zfs_context.h> | |
39 | #include <sys/dmu_objset.h> | |
40 | #include <sys/dmu_traverse.h> | |
41 | #include <sys/dsl_dataset.h> | |
42 | #include <sys/dsl_dir.h> | |
43 | #include <sys/dsl_pool.h> | |
44 | #include <sys/dsl_synctask.h> | |
45 | #include <sys/dsl_prop.h> | |
46 | #include <sys/dmu_zfetch.h> | |
47 | #include <sys/zfs_ioctl.h> | |
48 | #include <sys/zap.h> | |
49 | #include <sys/zio_checksum.h> | |
03c6040b | 50 | #include <sys/zio_compress.h> |
428870ff | 51 | #include <sys/sa.h> |
62bdd5eb | 52 | #include <sys/zfeature.h> |
a6255b7f | 53 | #include <sys/abd.h> |
e5d1c27e | 54 | #include <sys/trace_zfs.h> |
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 | */ | |
65 | int zfs_nopwrite_enabled = 1; | |
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 | */ | |
65282ee9 | 73 | unsigned long zfs_per_txg_dirty_frees_percent = 5; |
539d33c7 | 74 | |
66aca247 DB |
75 | /* |
76 | * Enable/disable forcing txg sync when dirty in dmu_offset_next. | |
77 | */ | |
78 | int zfs_dmu_offset_next_sync = 0; | |
79 | ||
d9b4bf06 MA |
80 | /* |
81 | * Limit the amount we can prefetch with one call to this amount. This | |
82 | * helps to limit the amount of memory that can be used by prefetching. | |
83 | * Larger objects should be prefetched a bit at a time. | |
84 | */ | |
85 | int dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE; | |
86 | ||
34dc7c2f | 87 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
2e5dc449 MA |
88 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" }, |
89 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" }, | |
90 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" }, | |
91 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" }, | |
92 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" }, | |
93 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" }, | |
94 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" }, | |
95 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" }, | |
96 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" }, | |
97 | {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" }, | |
98 | {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" }, | |
99 | {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" }, | |
100 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" }, | |
101 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"}, | |
102 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" }, | |
103 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" }, | |
104 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" }, | |
105 | {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" }, | |
106 | {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" }, | |
107 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" }, | |
108 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" }, | |
109 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" }, | |
110 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" }, | |
111 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" }, | |
112 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" }, | |
113 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" }, | |
114 | {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" }, | |
115 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" }, | |
116 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" }, | |
117 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" }, | |
118 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" }, | |
119 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" }, | |
120 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" }, | |
121 | {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" }, | |
122 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" }, | |
123 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" }, | |
124 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" }, | |
125 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"}, | |
126 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" }, | |
127 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" }, | |
128 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"}, | |
129 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"}, | |
130 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" }, | |
131 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" }, | |
132 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" }, | |
133 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" }, | |
134 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" }, | |
135 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" }, | |
136 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" }, | |
137 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" }, | |
138 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" }, | |
139 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" }, | |
140 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" }, | |
141 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
142 | }; |
143 | ||
144 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
145 | { byteswap_uint8_array, "uint8" }, | |
146 | { byteswap_uint16_array, "uint16" }, | |
147 | { byteswap_uint32_array, "uint32" }, | |
148 | { byteswap_uint64_array, "uint64" }, | |
149 | { zap_byteswap, "zap" }, | |
150 | { dnode_buf_byteswap, "dnode" }, | |
151 | { dmu_objset_byteswap, "objset" }, | |
152 | { zfs_znode_byteswap, "znode" }, | |
153 | { zfs_oldacl_byteswap, "oldacl" }, | |
154 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
155 | }; |
156 | ||
65c7cc49 | 157 | static int |
2bce8049 MA |
158 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, |
159 | void *tag, dmu_buf_t **dbp) | |
160 | { | |
161 | uint64_t blkid; | |
162 | dmu_buf_impl_t *db; | |
163 | ||
2bce8049 | 164 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 165 | blkid = dbuf_whichblock(dn, 0, offset); |
2bce8049 MA |
166 | db = dbuf_hold(dn, blkid, tag); |
167 | rw_exit(&dn->dn_struct_rwlock); | |
168 | ||
169 | if (db == NULL) { | |
170 | *dbp = NULL; | |
171 | return (SET_ERROR(EIO)); | |
172 | } | |
173 | ||
174 | *dbp = &db->db; | |
175 | return (0); | |
176 | } | |
34dc7c2f | 177 | int |
9b67f605 MA |
178 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
179 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
180 | { |
181 | dnode_t *dn; | |
182 | uint64_t blkid; | |
183 | dmu_buf_impl_t *db; | |
184 | int err; | |
428870ff BB |
185 | |
186 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
187 | if (err) |
188 | return (err); | |
34dc7c2f | 189 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 190 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
191 | db = dbuf_hold(dn, blkid, tag); |
192 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
193 | dnode_rele(dn, FTAG); |
194 | ||
34dc7c2f | 195 | if (db == NULL) { |
9b67f605 MA |
196 | *dbp = NULL; |
197 | return (SET_ERROR(EIO)); | |
198 | } | |
199 | ||
200 | *dbp = &db->db; | |
201 | return (err); | |
202 | } | |
203 | ||
2bce8049 MA |
204 | int |
205 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
206 | void *tag, dmu_buf_t **dbp, int flags) | |
207 | { | |
208 | int err; | |
209 | int db_flags = DB_RF_CANFAIL; | |
210 | ||
211 | if (flags & DMU_READ_NO_PREFETCH) | |
212 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
213 | if (flags & DMU_READ_NO_DECRYPT) |
214 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
215 | |
216 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
217 | if (err == 0) { | |
218 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
219 | err = dbuf_read(db, NULL, db_flags); | |
220 | if (err != 0) { | |
221 | dbuf_rele(db, tag); | |
222 | *dbp = NULL; | |
223 | } | |
224 | } | |
225 | ||
226 | return (err); | |
227 | } | |
228 | ||
9b67f605 MA |
229 | int |
230 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
231 | void *tag, dmu_buf_t **dbp, int flags) | |
232 | { | |
233 | int err; | |
234 | int db_flags = DB_RF_CANFAIL; | |
235 | ||
236 | if (flags & DMU_READ_NO_PREFETCH) | |
237 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
238 | if (flags & DMU_READ_NO_DECRYPT) |
239 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
240 | |
241 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
242 | if (err == 0) { | |
243 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 244 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 245 | if (err != 0) { |
34dc7c2f | 246 | dbuf_rele(db, tag); |
9b67f605 | 247 | *dbp = NULL; |
34dc7c2f BB |
248 | } |
249 | } | |
250 | ||
34dc7c2f BB |
251 | return (err); |
252 | } | |
253 | ||
254 | int | |
255 | dmu_bonus_max(void) | |
256 | { | |
50c957f7 | 257 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
258 | } |
259 | ||
260 | int | |
572e2857 | 261 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 262 | { |
572e2857 BB |
263 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
264 | dnode_t *dn; | |
265 | int error; | |
34dc7c2f | 266 | |
572e2857 BB |
267 | DB_DNODE_ENTER(db); |
268 | dn = DB_DNODE(db); | |
269 | ||
270 | if (dn->dn_bonus != db) { | |
2e528b49 | 271 | error = SET_ERROR(EINVAL); |
572e2857 | 272 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 273 | error = SET_ERROR(EINVAL); |
572e2857 BB |
274 | } else { |
275 | dnode_setbonuslen(dn, newsize, tx); | |
276 | error = 0; | |
277 | } | |
278 | ||
279 | DB_DNODE_EXIT(db); | |
280 | return (error); | |
34dc7c2f BB |
281 | } |
282 | ||
428870ff | 283 | int |
572e2857 | 284 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 285 | { |
572e2857 BB |
286 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
287 | dnode_t *dn; | |
288 | int error; | |
428870ff | 289 | |
572e2857 BB |
290 | DB_DNODE_ENTER(db); |
291 | dn = DB_DNODE(db); | |
428870ff | 292 | |
9ae529ec | 293 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 294 | error = SET_ERROR(EINVAL); |
572e2857 | 295 | } else if (dn->dn_bonus != db) { |
2e528b49 | 296 | error = SET_ERROR(EINVAL); |
572e2857 BB |
297 | } else { |
298 | dnode_setbonus_type(dn, type, tx); | |
299 | error = 0; | |
300 | } | |
428870ff | 301 | |
572e2857 BB |
302 | DB_DNODE_EXIT(db); |
303 | return (error); | |
304 | } | |
305 | ||
306 | dmu_object_type_t | |
307 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
308 | { | |
309 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
310 | dnode_t *dn; | |
311 | dmu_object_type_t type; | |
312 | ||
313 | DB_DNODE_ENTER(db); | |
314 | dn = DB_DNODE(db); | |
315 | type = dn->dn_bonustype; | |
316 | DB_DNODE_EXIT(db); | |
317 | ||
318 | return (type); | |
428870ff BB |
319 | } |
320 | ||
321 | int | |
322 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
323 | { | |
324 | dnode_t *dn; | |
325 | int error; | |
326 | ||
327 | error = dnode_hold(os, object, FTAG, &dn); | |
328 | dbuf_rm_spill(dn, tx); | |
329 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
330 | dnode_rm_spill(dn, tx); | |
331 | rw_exit(&dn->dn_struct_rwlock); | |
332 | dnode_rele(dn, FTAG); | |
333 | return (error); | |
334 | } | |
335 | ||
34dc7c2f | 336 | /* |
6955b401 BB |
337 | * Lookup and hold the bonus buffer for the provided dnode. If the dnode |
338 | * has not yet been allocated a new bonus dbuf a will be allocated. | |
339 | * Returns ENOENT, EIO, or 0. | |
34dc7c2f | 340 | */ |
6955b401 BB |
341 | int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp, |
342 | uint32_t flags) | |
34dc7c2f | 343 | { |
34dc7c2f BB |
344 | dmu_buf_impl_t *db; |
345 | int error; | |
b5256303 TC |
346 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
347 | ||
348 | if (flags & DMU_READ_NO_PREFETCH) | |
349 | db_flags |= DB_RF_NOPREFETCH; | |
350 | if (flags & DMU_READ_NO_DECRYPT) | |
351 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 352 | |
34dc7c2f BB |
353 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
354 | if (dn->dn_bonus == NULL) { | |
355 | rw_exit(&dn->dn_struct_rwlock); | |
356 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
357 | if (dn->dn_bonus == NULL) | |
358 | dbuf_create_bonus(dn); | |
359 | } | |
360 | db = dn->dn_bonus; | |
34dc7c2f BB |
361 | |
362 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 363 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 364 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 365 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
366 | } |
367 | ||
368 | /* | |
369 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
370 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
371 | * a dnode hold for every dbuf. | |
372 | */ | |
373 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f | 374 | |
b5256303 TC |
375 | error = dbuf_read(db, NULL, db_flags); |
376 | if (error) { | |
377 | dnode_evict_bonus(dn); | |
378 | dbuf_rele(db, tag); | |
379 | *dbp = NULL; | |
380 | return (error); | |
381 | } | |
34dc7c2f BB |
382 | |
383 | *dbp = &db->db; | |
384 | return (0); | |
385 | } | |
386 | ||
b5256303 | 387 | int |
6955b401 | 388 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) |
b5256303 | 389 | { |
6955b401 BB |
390 | dnode_t *dn; |
391 | int error; | |
392 | ||
393 | error = dnode_hold(os, object, FTAG, &dn); | |
394 | if (error) | |
395 | return (error); | |
396 | ||
397 | error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH); | |
398 | dnode_rele(dn, FTAG); | |
399 | ||
400 | return (error); | |
b5256303 TC |
401 | } |
402 | ||
428870ff BB |
403 | /* |
404 | * returns ENOENT, EIO, or 0. | |
405 | * | |
406 | * This interface will allocate a blank spill dbuf when a spill blk | |
407 | * doesn't already exist on the dnode. | |
408 | * | |
409 | * if you only want to find an already existing spill db, then | |
410 | * dmu_spill_hold_existing() should be used. | |
411 | */ | |
412 | int | |
413 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
414 | { | |
415 | dmu_buf_impl_t *db = NULL; | |
416 | int err; | |
417 | ||
418 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
419 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
420 | ||
421 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
422 | ||
423 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
424 | rw_exit(&dn->dn_struct_rwlock); | |
425 | ||
b182ac00 | 426 | if (db == NULL) { |
427 | *dbp = NULL; | |
428 | return (SET_ERROR(EIO)); | |
429 | } | |
572e2857 BB |
430 | err = dbuf_read(db, NULL, flags); |
431 | if (err == 0) | |
432 | *dbp = &db->db; | |
b182ac00 | 433 | else { |
572e2857 | 434 | dbuf_rele(db, tag); |
b182ac00 | 435 | *dbp = NULL; |
436 | } | |
428870ff BB |
437 | return (err); |
438 | } | |
439 | ||
440 | int | |
441 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
442 | { | |
572e2857 BB |
443 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
444 | dnode_t *dn; | |
428870ff BB |
445 | int err; |
446 | ||
572e2857 BB |
447 | DB_DNODE_ENTER(db); |
448 | dn = DB_DNODE(db); | |
449 | ||
450 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 451 | err = SET_ERROR(EINVAL); |
572e2857 BB |
452 | } else { |
453 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
454 | ||
455 | if (!dn->dn_have_spill) { | |
2e528b49 | 456 | err = SET_ERROR(ENOENT); |
572e2857 BB |
457 | } else { |
458 | err = dmu_spill_hold_by_dnode(dn, | |
459 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
460 | } | |
428870ff | 461 | |
428870ff | 462 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 463 | } |
572e2857 BB |
464 | |
465 | DB_DNODE_EXIT(db); | |
428870ff BB |
466 | return (err); |
467 | } | |
468 | ||
469 | int | |
e7504d7a TC |
470 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag, |
471 | dmu_buf_t **dbp) | |
428870ff | 472 | { |
572e2857 BB |
473 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
474 | dnode_t *dn; | |
475 | int err; | |
e7504d7a TC |
476 | uint32_t db_flags = DB_RF_CANFAIL; |
477 | ||
478 | if (flags & DMU_READ_NO_DECRYPT) | |
479 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
480 | |
481 | DB_DNODE_ENTER(db); | |
482 | dn = DB_DNODE(db); | |
e7504d7a | 483 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
484 | DB_DNODE_EXIT(db); |
485 | ||
486 | return (err); | |
428870ff BB |
487 | } |
488 | ||
34dc7c2f BB |
489 | /* |
490 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
491 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
492 | * and can induce severe lock contention when writing to several files | |
493 | * whose dnodes are in the same block. | |
494 | */ | |
af1698f5 | 495 | int |
9babb374 | 496 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
7f60329a | 497 | boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
34dc7c2f BB |
498 | { |
499 | dmu_buf_t **dbp; | |
500 | uint64_t blkid, nblks, i; | |
9babb374 | 501 | uint32_t dbuf_flags; |
34dc7c2f | 502 | int err; |
ec50cd24 | 503 | zio_t *zio = NULL; |
34dc7c2f BB |
504 | |
505 | ASSERT(length <= DMU_MAX_ACCESS); | |
506 | ||
7f60329a MA |
507 | /* |
508 | * Note: We directly notify the prefetch code of this read, so that | |
509 | * we can tell it about the multi-block read. dbuf_read() only knows | |
510 | * about the one block it is accessing. | |
511 | */ | |
512 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
513 | DB_RF_NOPREFETCH; | |
34dc7c2f BB |
514 | |
515 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
516 | if (dn->dn_datablkshift) { | |
517 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
518 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
519 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
520 | } else { |
521 | if (offset + length > dn->dn_datablksz) { | |
522 | zfs_panic_recover("zfs: accessing past end of object " | |
523 | "%llx/%llx (size=%u access=%llu+%llu)", | |
524 | (longlong_t)dn->dn_objset-> | |
525 | os_dsl_dataset->ds_object, | |
526 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
527 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 528 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 529 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
530 | } |
531 | nblks = 1; | |
532 | } | |
79c76d5b | 533 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 534 | |
ec50cd24 FY |
535 | if (read) |
536 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, | |
537 | ZIO_FLAG_CANFAIL); | |
fcff0f35 | 538 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f | 539 | for (i = 0; i < nblks; i++) { |
7f60329a | 540 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f BB |
541 | if (db == NULL) { |
542 | rw_exit(&dn->dn_struct_rwlock); | |
543 | dmu_buf_rele_array(dbp, nblks, tag); | |
ec50cd24 FY |
544 | if (read) |
545 | zio_nowait(zio); | |
2e528b49 | 546 | return (SET_ERROR(EIO)); |
34dc7c2f | 547 | } |
7f60329a | 548 | |
34dc7c2f | 549 | /* initiate async i/o */ |
7f60329a | 550 | if (read) |
9babb374 | 551 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
552 | dbp[i] = &db->db; |
553 | } | |
7f60329a | 554 | |
64e0fe14 RM |
555 | if (!read) |
556 | zfs_racct_write(length, nblks); | |
557 | ||
755065f3 AM |
558 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
559 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
560 | dmu_zfetch(&dn->dn_zfetch, blkid, nblks, | |
f664f1ee | 561 | read && DNODE_IS_CACHEABLE(dn), B_TRUE); |
7f60329a | 562 | } |
34dc7c2f BB |
563 | rw_exit(&dn->dn_struct_rwlock); |
564 | ||
34dc7c2f | 565 | if (read) { |
ec50cd24 FY |
566 | /* wait for async read i/o */ |
567 | err = zio_wait(zio); | |
568 | if (err) { | |
569 | dmu_buf_rele_array(dbp, nblks, tag); | |
570 | return (err); | |
571 | } | |
572 | ||
573 | /* wait for other io to complete */ | |
34dc7c2f BB |
574 | for (i = 0; i < nblks; i++) { |
575 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
576 | mutex_enter(&db->db_mtx); | |
577 | while (db->db_state == DB_READ || | |
578 | db->db_state == DB_FILL) | |
579 | cv_wait(&db->db_changed, &db->db_mtx); | |
580 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 581 | err = SET_ERROR(EIO); |
34dc7c2f BB |
582 | mutex_exit(&db->db_mtx); |
583 | if (err) { | |
584 | dmu_buf_rele_array(dbp, nblks, tag); | |
585 | return (err); | |
586 | } | |
587 | } | |
588 | } | |
589 | ||
590 | *numbufsp = nblks; | |
591 | *dbpp = dbp; | |
592 | return (0); | |
593 | } | |
594 | ||
595 | static int | |
596 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
597 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
598 | { | |
599 | dnode_t *dn; | |
600 | int err; | |
601 | ||
428870ff | 602 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
603 | if (err) |
604 | return (err); | |
605 | ||
606 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 607 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
608 | |
609 | dnode_rele(dn, FTAG); | |
610 | ||
611 | return (err); | |
612 | } | |
613 | ||
614 | int | |
572e2857 | 615 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
7f60329a MA |
616 | uint64_t length, boolean_t read, void *tag, int *numbufsp, |
617 | dmu_buf_t ***dbpp) | |
34dc7c2f | 618 | { |
572e2857 BB |
619 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
620 | dnode_t *dn; | |
34dc7c2f BB |
621 | int err; |
622 | ||
572e2857 BB |
623 | DB_DNODE_ENTER(db); |
624 | dn = DB_DNODE(db); | |
34dc7c2f | 625 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 626 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 627 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
628 | |
629 | return (err); | |
630 | } | |
631 | ||
632 | void | |
633 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
634 | { | |
635 | int i; | |
636 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
637 | ||
638 | if (numbufs == 0) | |
639 | return; | |
640 | ||
641 | for (i = 0; i < numbufs; i++) { | |
642 | if (dbp[i]) | |
643 | dbuf_rele(dbp[i], tag); | |
644 | } | |
645 | ||
646 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
647 | } | |
648 | ||
e8b96c60 | 649 | /* |
fcff0f35 | 650 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
e1cfd73f | 651 | * indirect blocks prefetched will be those that point to the blocks containing |
fcff0f35 | 652 | * the data starting at offset, and continuing to offset + len. |
e8b96c60 | 653 | * |
b5256303 | 654 | * Note that if the indirect blocks above the blocks being prefetched are not |
e1cfd73f | 655 | * in cache, they will be asynchronously read in. |
e8b96c60 | 656 | */ |
34dc7c2f | 657 | void |
fcff0f35 PD |
658 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
659 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
660 | { |
661 | dnode_t *dn; | |
662 | uint64_t blkid; | |
e8b96c60 | 663 | int nblks, err; |
34dc7c2f | 664 | |
34dc7c2f | 665 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 666 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
667 | |
668 | if (object == 0 || object >= DN_MAX_OBJECT) | |
669 | return; | |
670 | ||
671 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
672 | blkid = dbuf_whichblock(dn, level, |
673 | object * sizeof (dnode_phys_t)); | |
674 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
675 | rw_exit(&dn->dn_struct_rwlock); |
676 | return; | |
677 | } | |
678 | ||
d9b4bf06 MA |
679 | /* |
680 | * See comment before the definition of dmu_prefetch_max. | |
681 | */ | |
682 | len = MIN(len, dmu_prefetch_max); | |
683 | ||
34dc7c2f BB |
684 | /* |
685 | * XXX - Note, if the dnode for the requested object is not | |
686 | * already cached, we will do a *synchronous* read in the | |
687 | * dnode_hold() call. The same is true for any indirects. | |
688 | */ | |
428870ff | 689 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
690 | if (err != 0) |
691 | return; | |
692 | ||
fcff0f35 PD |
693 | /* |
694 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
695 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
696 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
697 | * offset) is the first. Then the number we need to prefetch is the | |
698 | * last - first + 1. | |
699 | */ | |
f664f1ee | 700 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
fcff0f35 PD |
701 | if (level > 0 || dn->dn_datablkshift != 0) { |
702 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
703 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
704 | } else { |
705 | nblks = (offset < dn->dn_datablksz); | |
706 | } | |
707 | ||
708 | if (nblks != 0) { | |
fcff0f35 | 709 | blkid = dbuf_whichblock(dn, level, offset); |
1c27024e | 710 | for (int i = 0; i < nblks; i++) |
fcff0f35 | 711 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f | 712 | } |
34dc7c2f BB |
713 | rw_exit(&dn->dn_struct_rwlock); |
714 | ||
715 | dnode_rele(dn, FTAG); | |
716 | } | |
717 | ||
45d1cae3 BB |
718 | /* |
719 | * Get the next "chunk" of file data to free. We traverse the file from | |
720 | * the end so that the file gets shorter over time (if we crashes in the | |
721 | * middle, this will leave us in a better state). We find allocated file | |
722 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
723 | * |
724 | * On input, *start should be the first offset that does not need to be | |
725 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
726 | * offset that should be freed and l1blks is set to the number of level 1 |
727 | * indirect blocks found within the chunk. | |
45d1cae3 | 728 | */ |
b128c09f | 729 | static int |
65282ee9 | 730 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 731 | { |
65282ee9 | 732 | uint64_t blks; |
b663a23d MA |
733 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
734 | /* bytes of data covered by a level-1 indirect block */ | |
ec4afd27 OM |
735 | uint64_t iblkrange = (uint64_t)dn->dn_datablksz * |
736 | EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
b128c09f | 737 | |
b663a23d | 738 | ASSERT3U(minimum, <=, *start); |
b128c09f | 739 | |
f4c594da TC |
740 | /* |
741 | * Check if we can free the entire range assuming that all of the | |
742 | * L1 blocks in this range have data. If we can, we use this | |
743 | * worst case value as an estimate so we can avoid having to look | |
744 | * at the object's actual data. | |
745 | */ | |
746 | uint64_t total_l1blks = | |
747 | (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / | |
748 | iblkrange; | |
749 | if (total_l1blks <= maxblks) { | |
750 | *l1blks = total_l1blks; | |
b663a23d | 751 | *start = minimum; |
b128c09f BB |
752 | return (0); |
753 | } | |
45d1cae3 | 754 | ASSERT(ISP2(iblkrange)); |
b128c09f | 755 | |
65282ee9 | 756 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
757 | int err; |
758 | ||
b663a23d MA |
759 | /* |
760 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
761 | * indirect block at or before the input offset. We must | |
762 | * decrement *start so that it is at the end of the region | |
763 | * to search. | |
764 | */ | |
765 | (*start)--; | |
f4c594da | 766 | |
b128c09f | 767 | err = dnode_next_offset(dn, |
45d1cae3 | 768 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 769 | |
b663a23d | 770 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 771 | if (err == ESRCH) { |
b663a23d MA |
772 | *start = minimum; |
773 | break; | |
774 | } else if (err != 0) { | |
65282ee9 | 775 | *l1blks = blks; |
b128c09f | 776 | return (err); |
45d1cae3 | 777 | } |
b128c09f | 778 | |
b663a23d | 779 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 780 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 781 | } |
b663a23d MA |
782 | if (*start < minimum) |
783 | *start = minimum; | |
65282ee9 | 784 | *l1blks = blks; |
f4c594da | 785 | |
b128c09f BB |
786 | return (0); |
787 | } | |
788 | ||
a08abc1b GM |
789 | /* |
790 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
791 | * otherwise return false. | |
792 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
793 | */ | |
794 | /*ARGSUSED*/ | |
795 | static boolean_t | |
796 | dmu_objset_zfs_unmounting(objset_t *os) | |
797 | { | |
798 | #ifdef _KERNEL | |
799 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
800 | return (zfs_get_vfs_flag_unmounted(os)); | |
801 | #endif | |
802 | return (B_FALSE); | |
803 | } | |
804 | ||
b128c09f BB |
805 | static int |
806 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 807 | uint64_t length) |
b128c09f | 808 | { |
c97d3069 | 809 | uint64_t object_size; |
b663a23d | 810 | int err; |
539d33c7 GM |
811 | uint64_t dirty_frees_threshold; |
812 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 813 | |
c97d3069 BB |
814 | if (dn == NULL) |
815 | return (SET_ERROR(EINVAL)); | |
816 | ||
817 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 818 | if (offset >= object_size) |
b128c09f | 819 | return (0); |
b128c09f | 820 | |
539d33c7 GM |
821 | if (zfs_per_txg_dirty_frees_percent <= 100) |
822 | dirty_frees_threshold = | |
823 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
824 | else | |
65282ee9 | 825 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 826 | |
b663a23d MA |
827 | if (length == DMU_OBJECT_END || offset + length > object_size) |
828 | length = object_size - offset; | |
829 | ||
830 | while (length != 0) { | |
539d33c7 | 831 | uint64_t chunk_end, chunk_begin, chunk_len; |
65282ee9 | 832 | uint64_t l1blks; |
b663a23d MA |
833 | dmu_tx_t *tx; |
834 | ||
a08abc1b GM |
835 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
836 | return (SET_ERROR(EINTR)); | |
837 | ||
b663a23d MA |
838 | chunk_end = chunk_begin = offset + length; |
839 | ||
840 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 841 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
842 | if (err) |
843 | return (err); | |
b663a23d MA |
844 | ASSERT3U(chunk_begin, >=, offset); |
845 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 846 | |
539d33c7 GM |
847 | chunk_len = chunk_end - chunk_begin; |
848 | ||
b128c09f | 849 | tx = dmu_tx_create(os); |
539d33c7 | 850 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
851 | |
852 | /* | |
853 | * Mark this transaction as typically resulting in a net | |
854 | * reduction in space used. | |
855 | */ | |
856 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
857 | err = dmu_tx_assign(tx, TXG_WAIT); |
858 | if (err) { | |
859 | dmu_tx_abort(tx); | |
860 | return (err); | |
861 | } | |
539d33c7 | 862 | |
f4c594da TC |
863 | uint64_t txg = dmu_tx_get_txg(tx); |
864 | ||
865 | mutex_enter(&dp->dp_lock); | |
866 | uint64_t long_free_dirty = | |
867 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; | |
868 | mutex_exit(&dp->dp_lock); | |
869 | ||
870 | /* | |
871 | * To avoid filling up a TXG with just frees, wait for | |
872 | * the next TXG to open before freeing more chunks if | |
873 | * we have reached the threshold of frees. | |
874 | */ | |
875 | if (dirty_frees_threshold != 0 && | |
876 | long_free_dirty >= dirty_frees_threshold) { | |
877 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); | |
878 | dmu_tx_commit(tx); | |
879 | txg_wait_open(dp, 0, B_TRUE); | |
880 | continue; | |
881 | } | |
882 | ||
65282ee9 AP |
883 | /* |
884 | * In order to prevent unnecessary write throttling, for each | |
885 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
886 | * in dnode_free_range() below. We compare this number to a | |
887 | * tunable threshold, past which we prevent new L1 dirty freeing | |
888 | * blocks from being added into the open TXG. See | |
889 | * dmu_free_long_range_impl() for details. The threshold | |
890 | * prevents write throttle activation due to dirty freeing L1 | |
891 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
892 | */ | |
539d33c7 | 893 | mutex_enter(&dp->dp_lock); |
f4c594da | 894 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += |
65282ee9 | 895 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
896 | mutex_exit(&dp->dp_lock); |
897 | DTRACE_PROBE3(free__long__range, | |
f4c594da TC |
898 | uint64_t, long_free_dirty, uint64_t, chunk_len, |
899 | uint64_t, txg); | |
539d33c7 | 900 | dnode_free_range(dn, chunk_begin, chunk_len, tx); |
440a3eb9 | 901 | |
b128c09f | 902 | dmu_tx_commit(tx); |
b663a23d | 903 | |
539d33c7 | 904 | length -= chunk_len; |
b128c09f BB |
905 | } |
906 | return (0); | |
907 | } | |
908 | ||
909 | int | |
910 | dmu_free_long_range(objset_t *os, uint64_t object, | |
911 | uint64_t offset, uint64_t length) | |
912 | { | |
913 | dnode_t *dn; | |
914 | int err; | |
915 | ||
428870ff | 916 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
917 | if (err != 0) |
918 | return (err); | |
0c03d21a | 919 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
920 | |
921 | /* | |
922 | * It is important to zero out the maxblkid when freeing the entire | |
923 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
924 | * will take the fast path, and (b) dnode_reallocate() can verify | |
925 | * that the entire file has been freed. | |
926 | */ | |
b0bc7a84 | 927 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
928 | dn->dn_maxblkid = 0; |
929 | ||
b128c09f BB |
930 | dnode_rele(dn, FTAG); |
931 | return (err); | |
932 | } | |
933 | ||
934 | int | |
0c03d21a | 935 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 936 | { |
b128c09f BB |
937 | dmu_tx_t *tx; |
938 | int err; | |
939 | ||
b663a23d | 940 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
941 | if (err != 0) |
942 | return (err); | |
b663a23d MA |
943 | |
944 | tx = dmu_tx_create(os); | |
945 | dmu_tx_hold_bonus(tx, object); | |
946 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 947 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
948 | err = dmu_tx_assign(tx, TXG_WAIT); |
949 | if (err == 0) { | |
35df0bb5 TC |
950 | if (err == 0) |
951 | err = dmu_object_free(os, object, tx); | |
440a3eb9 | 952 | |
b663a23d | 953 | dmu_tx_commit(tx); |
b128c09f | 954 | } else { |
b663a23d | 955 | dmu_tx_abort(tx); |
b128c09f | 956 | } |
b663a23d | 957 | |
b128c09f BB |
958 | return (err); |
959 | } | |
960 | ||
34dc7c2f BB |
961 | int |
962 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
963 | uint64_t size, dmu_tx_t *tx) | |
964 | { | |
965 | dnode_t *dn; | |
428870ff | 966 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
967 | if (err) |
968 | return (err); | |
969 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 970 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
971 | dnode_free_range(dn, offset, size, tx); |
972 | dnode_rele(dn, FTAG); | |
973 | return (0); | |
974 | } | |
975 | ||
0eef1bde | 976 | static int |
977 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 978 | void *buf, uint32_t flags) |
34dc7c2f | 979 | { |
34dc7c2f | 980 | dmu_buf_t **dbp; |
0eef1bde | 981 | int numbufs, err = 0; |
34dc7c2f BB |
982 | |
983 | /* | |
984 | * Deal with odd block sizes, where there can't be data past the first | |
985 | * block. If we ever do the tail block optimization, we will need to | |
986 | * handle that here as well. | |
987 | */ | |
45d1cae3 | 988 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 989 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
990 | MIN(size, dn->dn_datablksz - offset); |
991 | bzero((char *)buf + newsz, size - newsz); | |
992 | size = newsz; | |
993 | } | |
994 | ||
995 | while (size > 0) { | |
996 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 997 | int i; |
34dc7c2f BB |
998 | |
999 | /* | |
1000 | * NB: we could do this block-at-a-time, but it's nice | |
1001 | * to be reading in parallel. | |
1002 | */ | |
1003 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 1004 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
1005 | if (err) |
1006 | break; | |
1007 | ||
1008 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1009 | uint64_t tocpy; |
1010 | int64_t bufoff; | |
34dc7c2f BB |
1011 | dmu_buf_t *db = dbp[i]; |
1012 | ||
1013 | ASSERT(size > 0); | |
1014 | ||
1015 | bufoff = offset - db->db_offset; | |
c9520ecc | 1016 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 1017 | |
c9520ecc | 1018 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
1019 | |
1020 | offset += tocpy; | |
1021 | size -= tocpy; | |
1022 | buf = (char *)buf + tocpy; | |
1023 | } | |
1024 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1025 | } | |
34dc7c2f BB |
1026 | return (err); |
1027 | } | |
1028 | ||
0eef1bde | 1029 | int |
1030 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1031 | void *buf, uint32_t flags) | |
34dc7c2f | 1032 | { |
0eef1bde | 1033 | dnode_t *dn; |
1034 | int err; | |
34dc7c2f | 1035 | |
0eef1bde | 1036 | err = dnode_hold(os, object, FTAG, &dn); |
1037 | if (err != 0) | |
1038 | return (err); | |
34dc7c2f | 1039 | |
0eef1bde | 1040 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1041 | dnode_rele(dn, FTAG); | |
1042 | return (err); | |
1043 | } | |
1044 | ||
1045 | int | |
1046 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1047 | uint32_t flags) | |
1048 | { | |
1049 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1050 | } | |
1051 | ||
1052 | static void | |
1053 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1054 | const void *buf, dmu_tx_t *tx) | |
1055 | { | |
1056 | int i; | |
34dc7c2f BB |
1057 | |
1058 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1059 | uint64_t tocpy; |
1060 | int64_t bufoff; | |
34dc7c2f BB |
1061 | dmu_buf_t *db = dbp[i]; |
1062 | ||
1063 | ASSERT(size > 0); | |
1064 | ||
1065 | bufoff = offset - db->db_offset; | |
c9520ecc | 1066 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1067 | |
1068 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1069 | ||
1070 | if (tocpy == db->db_size) | |
1071 | dmu_buf_will_fill(db, tx); | |
1072 | else | |
1073 | dmu_buf_will_dirty(db, tx); | |
1074 | ||
60101509 | 1075 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1076 | |
1077 | if (tocpy == db->db_size) | |
1078 | dmu_buf_fill_done(db, tx); | |
1079 | ||
1080 | offset += tocpy; | |
1081 | size -= tocpy; | |
1082 | buf = (char *)buf + tocpy; | |
1083 | } | |
0eef1bde | 1084 | } |
1085 | ||
1086 | void | |
1087 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1088 | const void *buf, dmu_tx_t *tx) | |
1089 | { | |
1090 | dmu_buf_t **dbp; | |
1091 | int numbufs; | |
1092 | ||
1093 | if (size == 0) | |
1094 | return; | |
1095 | ||
1096 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1097 | FALSE, FTAG, &numbufs, &dbp)); | |
1098 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1099 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1100 | } | |
1101 | ||
0f8ff49e SD |
1102 | /* |
1103 | * Note: Lustre is an external consumer of this interface. | |
1104 | */ | |
0eef1bde | 1105 | void |
1106 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1107 | const void *buf, dmu_tx_t *tx) | |
1108 | { | |
1109 | dmu_buf_t **dbp; | |
1110 | int numbufs; | |
1111 | ||
1112 | if (size == 0) | |
1113 | return; | |
1114 | ||
1115 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1116 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1117 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1118 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1119 | } | |
1120 | ||
b128c09f BB |
1121 | void |
1122 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1123 | dmu_tx_t *tx) | |
1124 | { | |
1125 | dmu_buf_t **dbp; | |
1126 | int numbufs, i; | |
1127 | ||
1128 | if (size == 0) | |
1129 | return; | |
1130 | ||
1131 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1132 | FALSE, FTAG, &numbufs, &dbp)); | |
1133 | ||
1134 | for (i = 0; i < numbufs; i++) { | |
1135 | dmu_buf_t *db = dbp[i]; | |
1136 | ||
1137 | dmu_buf_will_not_fill(db, tx); | |
1138 | } | |
1139 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1140 | } | |
1141 | ||
9b67f605 MA |
1142 | void |
1143 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1144 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1145 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1146 | { | |
1147 | dmu_buf_t *db; | |
1148 | ||
1149 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1150 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1151 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1152 | FTAG, &db)); | |
1153 | ||
1154 | dmu_buf_write_embedded(db, | |
1155 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1156 | uncompressed_size, compressed_size, byteorder, tx); | |
1157 | ||
1158 | dmu_buf_rele(db, FTAG); | |
1159 | } | |
1160 | ||
30af21b0 PD |
1161 | void |
1162 | dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1163 | dmu_tx_t *tx) | |
1164 | { | |
1165 | int numbufs, i; | |
1166 | dmu_buf_t **dbp; | |
1167 | ||
1168 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1169 | &numbufs, &dbp)); | |
1170 | for (i = 0; i < numbufs; i++) | |
1171 | dmu_buf_redact(dbp[i], tx); | |
1172 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1173 | } | |
1174 | ||
34dc7c2f | 1175 | #ifdef _KERNEL |
5228cf01 | 1176 | int |
d0cd9a5c | 1177 | dmu_read_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size) |
872e8d26 BB |
1178 | { |
1179 | dmu_buf_t **dbp; | |
1180 | int numbufs, i, err; | |
872e8d26 BB |
1181 | |
1182 | /* | |
1183 | * NB: we could do this block-at-a-time, but it's nice | |
1184 | * to be reading in parallel. | |
1185 | */ | |
d0cd9a5c | 1186 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
804e0504 | 1187 | TRUE, FTAG, &numbufs, &dbp, 0); |
872e8d26 BB |
1188 | if (err) |
1189 | return (err); | |
1190 | ||
1191 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1192 | uint64_t tocpy; |
1193 | int64_t bufoff; | |
872e8d26 BB |
1194 | dmu_buf_t *db = dbp[i]; |
1195 | ||
1196 | ASSERT(size > 0); | |
1197 | ||
d0cd9a5c | 1198 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1199 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1200 | |
d0cd9a5c BA |
1201 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, tocpy, |
1202 | UIO_READ, uio); | |
1203 | ||
872e8d26 BB |
1204 | if (err) |
1205 | break; | |
1206 | ||
1207 | size -= tocpy; | |
1208 | } | |
1209 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1210 | ||
1211 | return (err); | |
1212 | } | |
1213 | ||
804e0504 MA |
1214 | /* |
1215 | * Read 'size' bytes into the uio buffer. | |
1216 | * From object zdb->db_object. | |
d0cd9a5c | 1217 | * Starting at zfs_uio_offset(uio). |
804e0504 MA |
1218 | * |
1219 | * If the caller already has a dbuf in the target object | |
1220 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1221 | * because we don't have to find the dnode_t for the object. | |
1222 | */ | |
1223 | int | |
d0cd9a5c | 1224 | dmu_read_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1225 | { |
1226 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1227 | dnode_t *dn; | |
1228 | int err; | |
1229 | ||
1230 | if (size == 0) | |
1231 | return (0); | |
1232 | ||
1233 | DB_DNODE_ENTER(db); | |
1234 | dn = DB_DNODE(db); | |
1235 | err = dmu_read_uio_dnode(dn, uio, size); | |
1236 | DB_DNODE_EXIT(db); | |
1237 | ||
1238 | return (err); | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * Read 'size' bytes into the uio buffer. | |
1243 | * From the specified object | |
d0cd9a5c | 1244 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1245 | */ |
1246 | int | |
d0cd9a5c | 1247 | dmu_read_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1248 | { |
1249 | dnode_t *dn; | |
1250 | int err; | |
1251 | ||
1252 | if (size == 0) | |
1253 | return (0); | |
1254 | ||
1255 | err = dnode_hold(os, object, FTAG, &dn); | |
1256 | if (err) | |
1257 | return (err); | |
1258 | ||
1259 | err = dmu_read_uio_dnode(dn, uio, size); | |
1260 | ||
1261 | dnode_rele(dn, FTAG); | |
1262 | ||
1263 | return (err); | |
1264 | } | |
1265 | ||
5228cf01 | 1266 | int |
d0cd9a5c | 1267 | dmu_write_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size, dmu_tx_t *tx) |
872e8d26 BB |
1268 | { |
1269 | dmu_buf_t **dbp; | |
1270 | int numbufs; | |
1271 | int err = 0; | |
1272 | int i; | |
1273 | ||
d0cd9a5c | 1274 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
872e8d26 BB |
1275 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); |
1276 | if (err) | |
1277 | return (err); | |
1278 | ||
1279 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1280 | uint64_t tocpy; |
1281 | int64_t bufoff; | |
872e8d26 BB |
1282 | dmu_buf_t *db = dbp[i]; |
1283 | ||
1284 | ASSERT(size > 0); | |
1285 | ||
d0cd9a5c | 1286 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1287 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1288 | |
1289 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1290 | ||
1291 | if (tocpy == db->db_size) | |
1292 | dmu_buf_will_fill(db, tx); | |
1293 | else | |
1294 | dmu_buf_will_dirty(db, tx); | |
1295 | ||
1296 | /* | |
d0cd9a5c | 1297 | * XXX zfs_uiomove could block forever (eg.nfs-backed |
872e8d26 | 1298 | * pages). There needs to be a uiolockdown() function |
d0cd9a5c | 1299 | * to lock the pages in memory, so that zfs_uiomove won't |
872e8d26 BB |
1300 | * block. |
1301 | */ | |
d0cd9a5c BA |
1302 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, |
1303 | tocpy, UIO_WRITE, uio); | |
1304 | ||
872e8d26 BB |
1305 | if (tocpy == db->db_size) |
1306 | dmu_buf_fill_done(db, tx); | |
1307 | ||
1308 | if (err) | |
1309 | break; | |
1310 | ||
1311 | size -= tocpy; | |
1312 | } | |
1313 | ||
1314 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1315 | return (err); | |
1316 | } | |
1317 | ||
804e0504 MA |
1318 | /* |
1319 | * Write 'size' bytes from the uio buffer. | |
1320 | * To object zdb->db_object. | |
d0cd9a5c | 1321 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1322 | * |
1323 | * If the caller already has a dbuf in the target object | |
1324 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1325 | * because we don't have to find the dnode_t for the object. | |
1326 | */ | |
428870ff | 1327 | int |
d0cd9a5c | 1328 | dmu_write_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1329 | dmu_tx_t *tx) |
1330 | { | |
572e2857 BB |
1331 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1332 | dnode_t *dn; | |
1333 | int err; | |
1334 | ||
428870ff BB |
1335 | if (size == 0) |
1336 | return (0); | |
1337 | ||
572e2857 BB |
1338 | DB_DNODE_ENTER(db); |
1339 | dn = DB_DNODE(db); | |
1340 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1341 | DB_DNODE_EXIT(db); | |
1342 | ||
1343 | return (err); | |
428870ff BB |
1344 | } |
1345 | ||
804e0504 MA |
1346 | /* |
1347 | * Write 'size' bytes from the uio buffer. | |
1348 | * To the specified object. | |
d0cd9a5c | 1349 | * Starting at offset zfs_uio_offset(uio). |
804e0504 | 1350 | */ |
428870ff | 1351 | int |
d0cd9a5c | 1352 | dmu_write_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1353 | dmu_tx_t *tx) |
1354 | { | |
1355 | dnode_t *dn; | |
1356 | int err; | |
1357 | ||
1358 | if (size == 0) | |
1359 | return (0); | |
1360 | ||
1361 | err = dnode_hold(os, object, FTAG, &dn); | |
1362 | if (err) | |
1363 | return (err); | |
1364 | ||
1365 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1366 | ||
1367 | dnode_rele(dn, FTAG); | |
1368 | ||
1369 | return (err); | |
1370 | } | |
872e8d26 | 1371 | #endif /* _KERNEL */ |
34dc7c2f | 1372 | |
9babb374 BB |
1373 | /* |
1374 | * Allocate a loaned anonymous arc buffer. | |
1375 | */ | |
1376 | arc_buf_t * | |
1377 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1378 | { | |
572e2857 | 1379 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1380 | |
2aa34383 | 1381 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1382 | } |
1383 | ||
1384 | /* | |
1385 | * Free a loaned arc buffer. | |
1386 | */ | |
1387 | void | |
1388 | dmu_return_arcbuf(arc_buf_t *buf) | |
1389 | { | |
1390 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1391 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1392 | } |
1393 | ||
ba67d821 MA |
1394 | /* |
1395 | * A "lightweight" write is faster than a regular write (e.g. | |
1396 | * dmu_write_by_dnode() or dmu_assign_arcbuf_by_dnode()), because it avoids the | |
1397 | * CPU cost of creating a dmu_buf_impl_t and arc_buf_[hdr_]_t. However, the | |
1398 | * data can not be read or overwritten until the transaction's txg has been | |
1399 | * synced. This makes it appropriate for workloads that are known to be | |
1400 | * (temporarily) write-only, like "zfs receive". | |
1401 | * | |
1402 | * A single block is written, starting at the specified offset in bytes. If | |
1403 | * the call is successful, it returns 0 and the provided abd has been | |
1404 | * consumed (the caller should not free it). | |
1405 | */ | |
1406 | int | |
1407 | dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, | |
1408 | const zio_prop_t *zp, enum zio_flag flags, dmu_tx_t *tx) | |
1409 | { | |
1410 | dbuf_dirty_record_t *dr = | |
1411 | dbuf_dirty_lightweight(dn, dbuf_whichblock(dn, 0, offset), tx); | |
1412 | if (dr == NULL) | |
1413 | return (SET_ERROR(EIO)); | |
1414 | dr->dt.dll.dr_abd = abd; | |
1415 | dr->dt.dll.dr_props = *zp; | |
1416 | dr->dt.dll.dr_flags = flags; | |
1417 | return (0); | |
1418 | } | |
1419 | ||
9babb374 BB |
1420 | /* |
1421 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1422 | * If this is not possible copy the contents of passed arc buf via | |
1423 | * dmu_write(). | |
1424 | */ | |
305781da | 1425 | int |
440a3eb9 | 1426 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1427 | dmu_tx_t *tx) |
1428 | { | |
9babb374 | 1429 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1430 | objset_t *os = dn->dn_objset; |
1431 | uint64_t object = dn->dn_object; | |
2aa34383 | 1432 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1433 | uint64_t blkid; |
1434 | ||
1435 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1436 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1437 | db = dbuf_hold(dn, blkid, FTAG); |
1438 | if (db == NULL) | |
1439 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1440 | rw_exit(&dn->dn_struct_rwlock); |
1441 | ||
88904bb3 | 1442 | /* |
ba67d821 MA |
1443 | * We can only assign if the offset is aligned and the arc buf is the |
1444 | * same size as the dbuf. | |
88904bb3 | 1445 | */ |
2aa34383 | 1446 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
64e0fe14 | 1447 | zfs_racct_write(blksz, 1); |
9babb374 BB |
1448 | dbuf_assign_arcbuf(db, buf, tx); |
1449 | dbuf_rele(db, FTAG); | |
1450 | } else { | |
2aa34383 DK |
1451 | /* compressed bufs must always be assignable to their dbuf */ |
1452 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1453 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1454 | |
9babb374 | 1455 | dbuf_rele(db, FTAG); |
572e2857 | 1456 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 BB |
1457 | dmu_return_arcbuf(buf); |
1458 | } | |
305781da TC |
1459 | |
1460 | return (0); | |
9babb374 BB |
1461 | } |
1462 | ||
305781da | 1463 | int |
440a3eb9 TC |
1464 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1465 | dmu_tx_t *tx) | |
1466 | { | |
305781da | 1467 | int err; |
440a3eb9 TC |
1468 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1469 | ||
1470 | DB_DNODE_ENTER(dbuf); | |
305781da | 1471 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1472 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1473 | |
1474 | return (err); | |
440a3eb9 TC |
1475 | } |
1476 | ||
34dc7c2f | 1477 | typedef struct { |
428870ff BB |
1478 | dbuf_dirty_record_t *dsa_dr; |
1479 | dmu_sync_cb_t *dsa_done; | |
1480 | zgd_t *dsa_zgd; | |
1481 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1482 | } dmu_sync_arg_t; |
1483 | ||
b128c09f BB |
1484 | /* ARGSUSED */ |
1485 | static void | |
1486 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1487 | { | |
428870ff BB |
1488 | dmu_sync_arg_t *dsa = varg; |
1489 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1490 | blkptr_t *bp = zio->io_bp; |
1491 | ||
428870ff BB |
1492 | if (zio->io_error == 0) { |
1493 | if (BP_IS_HOLE(bp)) { | |
1494 | /* | |
1495 | * A block of zeros may compress to a hole, but the | |
1496 | * block size still needs to be known for replay. | |
1497 | */ | |
1498 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1499 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1500 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1501 | BP_SET_FILL(bp, 1); |
428870ff | 1502 | } |
b128c09f BB |
1503 | } |
1504 | } | |
1505 | ||
428870ff BB |
1506 | static void |
1507 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1508 | { | |
1509 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1510 | } | |
1511 | ||
34dc7c2f BB |
1512 | /* ARGSUSED */ |
1513 | static void | |
1514 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1515 | { | |
428870ff BB |
1516 | dmu_sync_arg_t *dsa = varg; |
1517 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1518 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1519 | zgd_t *zgd = dsa->dsa_zgd; |
1520 | ||
1521 | /* | |
1522 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1523 | * the writes for the lwb have completed. | |
1524 | */ | |
1525 | if (zio->io_error == 0) { | |
1526 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1527 | } | |
34dc7c2f | 1528 | |
34dc7c2f BB |
1529 | mutex_enter(&db->db_mtx); |
1530 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1531 | if (zio->io_error == 0) { |
03c6040b GW |
1532 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1533 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1534 | blkptr_t *bp = zio->io_bp; |
1535 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1536 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1537 | |
1538 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1539 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1540 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1541 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1542 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1543 | } |
428870ff BB |
1544 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1545 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1546 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1547 | |
1548 | /* | |
1549 | * Old style holes are filled with all zeros, whereas | |
1550 | * new-style holes maintain their lsize, type, level, | |
1551 | * and birth time (see zio_write_compress). While we | |
1552 | * need to reset the BP_SET_LSIZE() call that happened | |
1553 | * in dmu_sync_ready for old style holes, we do *not* | |
1554 | * want to wipe out the information contained in new | |
1555 | * style holes. Thus, only zero out the block pointer if | |
1556 | * it's an old style hole. | |
1557 | */ | |
1558 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1559 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1560 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1561 | } else { | |
1562 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1563 | } | |
34dc7c2f BB |
1564 | cv_broadcast(&db->db_changed); |
1565 | mutex_exit(&db->db_mtx); | |
1566 | ||
428870ff | 1567 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1568 | |
428870ff BB |
1569 | kmem_free(dsa, sizeof (*dsa)); |
1570 | } | |
1571 | ||
1572 | static void | |
1573 | dmu_sync_late_arrival_done(zio_t *zio) | |
1574 | { | |
1575 | blkptr_t *bp = zio->io_bp; | |
1576 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1577 | zgd_t *zgd = dsa->dsa_zgd; |
1578 | ||
1579 | if (zio->io_error == 0) { | |
1580 | /* | |
1581 | * Record the vdev(s) backing this blkptr so they can be | |
1582 | * flushed after the writes for the lwb have completed. | |
1583 | */ | |
1584 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1585 | ||
1586 | if (!BP_IS_HOLE(bp)) { | |
2a8ba608 | 1587 | blkptr_t *bp_orig __maybe_unused = &zio->io_bp_orig; |
900d09b2 PS |
1588 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); |
1589 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1590 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1591 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1592 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1593 | } | |
428870ff BB |
1594 | } |
1595 | ||
1596 | dmu_tx_commit(dsa->dsa_tx); | |
1597 | ||
1598 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1599 | ||
e2af2acc | 1600 | abd_free(zio->io_abd); |
428870ff BB |
1601 | kmem_free(dsa, sizeof (*dsa)); |
1602 | } | |
1603 | ||
1604 | static int | |
1605 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1606 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1607 | { |
1608 | dmu_sync_arg_t *dsa; | |
1609 | dmu_tx_t *tx; | |
1610 | ||
1611 | tx = dmu_tx_create(os); | |
1612 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1613 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1614 | dmu_tx_abort(tx); | |
2e528b49 MA |
1615 | /* Make zl_get_data do txg_waited_synced() */ |
1616 | return (SET_ERROR(EIO)); | |
428870ff BB |
1617 | } |
1618 | ||
1ce23dca PS |
1619 | /* |
1620 | * In order to prevent the zgd's lwb from being free'd prior to | |
1621 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1622 | * the lwb's "max txg" takes this tx's txg into account. | |
1623 | */ | |
1624 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1625 | ||
79c76d5b | 1626 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1627 | dsa->dsa_dr = NULL; |
1628 | dsa->dsa_done = done; | |
1629 | dsa->dsa_zgd = zgd; | |
1630 | dsa->dsa_tx = tx; | |
1631 | ||
02dc43bc MA |
1632 | /* |
1633 | * Since we are currently syncing this txg, it's nontrivial to | |
1634 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1635 | * | |
1636 | * When syncing, the db_blkptr is initially the BP of the previous | |
1637 | * txg. We can not nopwrite against it because it will be changed | |
1638 | * (this is similar to the non-late-arrival case where the dbuf is | |
1639 | * dirty in a future txg). | |
1640 | * | |
1641 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1642 | * We can not nopwrite against it because although the BP will not | |
1643 | * (typically) be changed, the data has not yet been persisted to this | |
1644 | * location. | |
1645 | * | |
1646 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1647 | * possible to always nopwrite, because the data that was written in | |
1648 | * this txg is the same data that we are trying to write. However we | |
1649 | * would need to check that this dbuf is not dirty in any future | |
1650 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1651 | * don't nopwrite in this case. | |
1652 | */ | |
1653 | zp->zp_nopwrite = B_FALSE; | |
1654 | ||
a6255b7f DQ |
1655 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1656 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1657 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1658 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1659 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1660 | |
1661 | return (0); | |
34dc7c2f BB |
1662 | } |
1663 | ||
1664 | /* | |
1665 | * Intent log support: sync the block associated with db to disk. | |
1666 | * N.B. and XXX: the caller is responsible for making sure that the | |
1667 | * data isn't changing while dmu_sync() is writing it. | |
1668 | * | |
1669 | * Return values: | |
1670 | * | |
03c6040b | 1671 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1672 | * The caller should not log the write. |
1673 | * | |
1674 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1675 | * The caller should not log the write. | |
1676 | * | |
1677 | * EALREADY: this block is already in the process of being synced. | |
1678 | * The caller should track its progress (somehow). | |
1679 | * | |
428870ff BB |
1680 | * EIO: could not do the I/O. |
1681 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1682 | * |
428870ff BB |
1683 | * 0: the I/O has been initiated. |
1684 | * The caller should log this blkptr in the done callback. | |
1685 | * It is possible that the I/O will fail, in which case | |
1686 | * the error will be reported to the done callback and | |
1687 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1688 | */ |
1689 | int | |
428870ff | 1690 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1691 | { |
428870ff BB |
1692 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1693 | objset_t *os = db->db_objset; | |
1694 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
cccbed9f | 1695 | dbuf_dirty_record_t *dr, *dr_next; |
428870ff | 1696 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1697 | zbookmark_phys_t zb; |
428870ff | 1698 | zio_prop_t zp; |
572e2857 | 1699 | dnode_t *dn; |
34dc7c2f | 1700 | |
428870ff | 1701 | ASSERT(pio != NULL); |
34dc7c2f BB |
1702 | ASSERT(txg != 0); |
1703 | ||
428870ff BB |
1704 | SET_BOOKMARK(&zb, ds->ds_object, |
1705 | db->db.db_object, db->db_level, db->db_blkid); | |
1706 | ||
572e2857 BB |
1707 | DB_DNODE_ENTER(db); |
1708 | dn = DB_DNODE(db); | |
82644107 | 1709 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1710 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1711 | |
1712 | /* | |
428870ff | 1713 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1714 | */ |
428870ff BB |
1715 | if (txg > spa_freeze_txg(os->os_spa)) |
1716 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1717 | |
1718 | /* | |
428870ff BB |
1719 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1720 | * and us. If we determine that this txg is not yet syncing, | |
1721 | * but it begins to sync a moment later, that's OK because the | |
1722 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1723 | */ |
428870ff BB |
1724 | mutex_enter(&db->db_mtx); |
1725 | ||
1726 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1727 | /* |
428870ff | 1728 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1729 | */ |
428870ff | 1730 | mutex_exit(&db->db_mtx); |
2e528b49 | 1731 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1732 | } |
1733 | ||
428870ff BB |
1734 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1735 | /* | |
1736 | * This txg is currently syncing, so we can't mess with | |
1737 | * the dirty record anymore; just write a new log block. | |
1738 | */ | |
1739 | mutex_exit(&db->db_mtx); | |
1740 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1741 | } |
1742 | ||
cccbed9f | 1743 | dr = dbuf_find_dirty_eq(db, txg); |
428870ff BB |
1744 | |
1745 | if (dr == NULL) { | |
34dc7c2f | 1746 | /* |
428870ff | 1747 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1748 | * There's no need to log writes to freed blocks, so we're done. |
1749 | */ | |
1750 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1751 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1752 | } |
1753 | ||
cccbed9f MM |
1754 | dr_next = list_next(&db->db_dirty_records, dr); |
1755 | ASSERT(dr_next == NULL || dr_next->dr_txg < txg); | |
03c6040b | 1756 | |
02dc43bc MA |
1757 | if (db->db_blkptr != NULL) { |
1758 | /* | |
1759 | * We need to fill in zgd_bp with the current blkptr so that | |
1760 | * the nopwrite code can check if we're writing the same | |
1761 | * data that's already on disk. We can only nopwrite if we | |
1762 | * are sure that after making the copy, db_blkptr will not | |
1763 | * change until our i/o completes. We ensure this by | |
1764 | * holding the db_mtx, and only allowing nopwrite if the | |
1765 | * block is not already dirty (see below). This is verified | |
1766 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1767 | * not changed. | |
1768 | */ | |
1769 | *zgd->zgd_bp = *db->db_blkptr; | |
1770 | } | |
1771 | ||
03c6040b | 1772 | /* |
f3c517d8 MA |
1773 | * Assume the on-disk data is X, the current syncing data (in |
1774 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1775 | * in dmu_sync). | |
1776 | * | |
1777 | * We usually want to perform a nopwrite if X and Z are the | |
1778 | * same. However, if Y is different (i.e. the BP is going to | |
1779 | * change before this write takes effect), then a nopwrite will | |
1780 | * be incorrect - we would override with X, which could have | |
1781 | * been freed when Y was written. | |
1782 | * | |
1783 | * (Note that this is not a concern when we are nop-writing from | |
1784 | * syncing context, because X and Y must be identical, because | |
1785 | * all previous txgs have been synced.) | |
1786 | * | |
1787 | * Therefore, we disable nopwrite if the current BP could change | |
1788 | * before this TXG. There are two ways it could change: by | |
1789 | * being dirty (dr_next is non-NULL), or by being freed | |
1790 | * (dnode_block_freed()). This behavior is verified by | |
1791 | * zio_done(), which VERIFYs that the override BP is identical | |
1792 | * to the on-disk BP. | |
03c6040b | 1793 | */ |
f3c517d8 MA |
1794 | DB_DNODE_ENTER(db); |
1795 | dn = DB_DNODE(db); | |
cccbed9f | 1796 | if (dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) |
03c6040b | 1797 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1798 | DB_DNODE_EXIT(db); |
03c6040b | 1799 | |
34dc7c2f | 1800 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1801 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1802 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1803 | /* |
428870ff BB |
1804 | * We have already issued a sync write for this buffer, |
1805 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1806 | * have been dirtied since, or we would have cleared the state. |
1807 | */ | |
34dc7c2f | 1808 | mutex_exit(&db->db_mtx); |
2e528b49 | 1809 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1810 | } |
1811 | ||
428870ff | 1812 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1813 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1814 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1815 | |
79c76d5b | 1816 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1817 | dsa->dsa_dr = dr; |
1818 | dsa->dsa_done = done; | |
1819 | dsa->dsa_zgd = zgd; | |
1820 | dsa->dsa_tx = NULL; | |
b128c09f | 1821 | |
428870ff | 1822 | zio_nowait(arc_write(pio, os->os_spa, txg, |
02dc43bc | 1823 | zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 1824 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 1825 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 1826 | |
428870ff | 1827 | return (0); |
34dc7c2f BB |
1828 | } |
1829 | ||
b5256303 TC |
1830 | int |
1831 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
1832 | { | |
1833 | dnode_t *dn; | |
1834 | int err; | |
1835 | ||
1836 | err = dnode_hold(os, object, FTAG, &dn); | |
1837 | if (err) | |
1838 | return (err); | |
1839 | err = dnode_set_nlevels(dn, nlevels, tx); | |
1840 | dnode_rele(dn, FTAG); | |
1841 | return (err); | |
1842 | } | |
1843 | ||
34dc7c2f BB |
1844 | int |
1845 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 1846 | dmu_tx_t *tx) |
34dc7c2f BB |
1847 | { |
1848 | dnode_t *dn; | |
1849 | int err; | |
1850 | ||
428870ff | 1851 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1852 | if (err) |
1853 | return (err); | |
1854 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1855 | dnode_rele(dn, FTAG); | |
1856 | return (err); | |
1857 | } | |
1858 | ||
ae76f45c TC |
1859 | int |
1860 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
1861 | 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 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 1870 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
1871 | rw_exit(&dn->dn_struct_rwlock); |
1872 | dnode_rele(dn, FTAG); | |
1873 | return (0); | |
1874 | } | |
1875 | ||
34dc7c2f BB |
1876 | void |
1877 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 1878 | dmu_tx_t *tx) |
34dc7c2f BB |
1879 | { |
1880 | dnode_t *dn; | |
1881 | ||
9b67f605 MA |
1882 | /* |
1883 | * Send streams include each object's checksum function. This | |
1884 | * check ensures that the receiving system can understand the | |
1885 | * checksum function transmitted. | |
1886 | */ | |
1887 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1888 | ||
1889 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1890 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1891 | dn->dn_checksum = checksum; |
1892 | dnode_setdirty(dn, tx); | |
1893 | dnode_rele(dn, FTAG); | |
1894 | } | |
1895 | ||
1896 | void | |
1897 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 1898 | dmu_tx_t *tx) |
34dc7c2f BB |
1899 | { |
1900 | dnode_t *dn; | |
1901 | ||
9b67f605 MA |
1902 | /* |
1903 | * Send streams include each object's compression function. This | |
1904 | * check ensures that the receiving system can understand the | |
1905 | * compression function transmitted. | |
1906 | */ | |
1907 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1908 | ||
1909 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
1910 | dn->dn_compress = compress; |
1911 | dnode_setdirty(dn, tx); | |
1912 | dnode_rele(dn, FTAG); | |
1913 | } | |
1914 | ||
faf0f58c MA |
1915 | /* |
1916 | * When the "redundant_metadata" property is set to "most", only indirect | |
1917 | * blocks of this level and higher will have an additional ditto block. | |
1918 | */ | |
1919 | int zfs_redundant_metadata_most_ditto_level = 2; | |
1920 | ||
428870ff | 1921 | void |
82644107 | 1922 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
1923 | { |
1924 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 1925 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 1926 | (wp & WP_SPILL)); |
428870ff BB |
1927 | enum zio_checksum checksum = os->os_checksum; |
1928 | enum zio_compress compress = os->os_compress; | |
10b3c7f5 | 1929 | uint8_t complevel = os->os_complevel; |
428870ff | 1930 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; |
03c6040b GW |
1931 | boolean_t dedup = B_FALSE; |
1932 | boolean_t nopwrite = B_FALSE; | |
428870ff | 1933 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 1934 | boolean_t encrypt = B_FALSE; |
428870ff | 1935 | int copies = os->os_copies; |
a7004725 | 1936 | |
428870ff | 1937 | /* |
03c6040b GW |
1938 | * We maintain different write policies for each of the following |
1939 | * types of data: | |
1940 | * 1. metadata | |
1941 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
1942 | * 3. all other level 0 blocks | |
428870ff BB |
1943 | */ |
1944 | if (ismd) { | |
b1d21733 TC |
1945 | /* |
1946 | * XXX -- we should design a compression algorithm | |
1947 | * that specializes in arrays of bps. | |
1948 | */ | |
1949 | compress = zio_compress_select(os->os_spa, | |
1950 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 1951 | |
428870ff BB |
1952 | /* |
1953 | * Metadata always gets checksummed. If the data | |
1954 | * checksum is multi-bit correctable, and it's not a | |
1955 | * ZBT-style checksum, then it's suitable for metadata | |
1956 | * as well. Otherwise, the metadata checksum defaults | |
1957 | * to fletcher4. | |
1958 | */ | |
3c67d83a TH |
1959 | if (!(zio_checksum_table[checksum].ci_flags & |
1960 | ZCHECKSUM_FLAG_METADATA) || | |
1961 | (zio_checksum_table[checksum].ci_flags & | |
1962 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 1963 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
1964 | |
1965 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
1966 | (os->os_redundant_metadata == | |
1967 | ZFS_REDUNDANT_METADATA_MOST && | |
1968 | (level >= zfs_redundant_metadata_most_ditto_level || | |
1969 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
1970 | copies++; | |
03c6040b GW |
1971 | } else if (wp & WP_NOFILL) { |
1972 | ASSERT(level == 0); | |
428870ff | 1973 | |
428870ff | 1974 | /* |
03c6040b GW |
1975 | * If we're writing preallocated blocks, we aren't actually |
1976 | * writing them so don't set any policy properties. These | |
1977 | * blocks are currently only used by an external subsystem | |
1978 | * outside of zfs (i.e. dump) and not written by the zio | |
1979 | * pipeline. | |
428870ff | 1980 | */ |
03c6040b GW |
1981 | compress = ZIO_COMPRESS_OFF; |
1982 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 1983 | } else { |
99197f03 JG |
1984 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
1985 | compress); | |
10b3c7f5 MN |
1986 | complevel = zio_complevel_select(os->os_spa, compress, |
1987 | complevel, complevel); | |
428870ff | 1988 | |
03c6040b GW |
1989 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
1990 | zio_checksum_select(dn->dn_checksum, checksum) : | |
1991 | dedup_checksum; | |
428870ff | 1992 | |
03c6040b GW |
1993 | /* |
1994 | * Determine dedup setting. If we are in dmu_sync(), | |
1995 | * we won't actually dedup now because that's all | |
1996 | * done in syncing context; but we do want to use the | |
e1cfd73f | 1997 | * dedup checksum. If the checksum is not strong |
03c6040b GW |
1998 | * enough to ensure unique signatures, force |
1999 | * dedup_verify. | |
2000 | */ | |
2001 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2002 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2003 | if (!(zio_checksum_table[checksum].ci_flags & |
2004 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2005 | dedup_verify = B_TRUE; |
2006 | } | |
428870ff | 2007 | |
03c6040b | 2008 | /* |
3c67d83a TH |
2009 | * Enable nopwrite if we have secure enough checksum |
2010 | * algorithm (see comment in zio_nop_write) and | |
2011 | * compression is enabled. We don't enable nopwrite if | |
2012 | * dedup is enabled as the two features are mutually | |
2013 | * exclusive. | |
03c6040b | 2014 | */ |
3c67d83a TH |
2015 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2016 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2017 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2018 | } |
2019 | ||
b5256303 TC |
2020 | /* |
2021 | * All objects in an encrypted objset are protected from modification | |
2022 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2023 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2024 | * not subject to nopwrite since writing the same data will still | |
2025 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2026 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2027 | * object types. | |
2028 | */ | |
2029 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2030 | encrypt = B_TRUE; | |
2031 | ||
2032 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2033 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2034 | nopwrite = B_FALSE; | |
2035 | } else { | |
2036 | dedup = B_FALSE; | |
2037 | } | |
2038 | ||
ae76f45c TC |
2039 | if (level <= 0 && |
2040 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2041 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2042 | } |
b5256303 | 2043 | } |
2aa34383 | 2044 | |
b5256303 | 2045 | zp->zp_compress = compress; |
10b3c7f5 | 2046 | zp->zp_complevel = complevel; |
b5256303 | 2047 | zp->zp_checksum = checksum; |
428870ff BB |
2048 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2049 | zp->zp_level = level; | |
faf0f58c | 2050 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2051 | zp->zp_dedup = dedup; |
2052 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2053 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2054 | zp->zp_encrypt = encrypt; |
2055 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
2056 | bzero(zp->zp_salt, ZIO_DATA_SALT_LEN); | |
2057 | bzero(zp->zp_iv, ZIO_DATA_IV_LEN); | |
2058 | bzero(zp->zp_mac, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2059 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2060 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2061 | |
2062 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2063 | } |
2064 | ||
66aca247 DB |
2065 | /* |
2066 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2067 | * in order to determine the location of holes. In order to accurately | |
2068 | * report holes all dirty data must be synced to disk. This causes extremely | |
2069 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2070 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2071 | * report the dnode as having no holes which is always a safe thing to do. | |
2072 | */ | |
34dc7c2f BB |
2073 | int |
2074 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2075 | { | |
2076 | dnode_t *dn; | |
2077 | int i, err; | |
66aca247 | 2078 | boolean_t clean = B_TRUE; |
34dc7c2f | 2079 | |
428870ff | 2080 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2081 | if (err) |
2082 | return (err); | |
66aca247 | 2083 | |
34dc7c2f | 2084 | /* |
2531ce37 | 2085 | * Check if dnode is dirty |
34dc7c2f | 2086 | */ |
454365bb | 2087 | for (i = 0; i < TXG_SIZE; i++) { |
edc1e713 | 2088 | if (multilist_link_active(&dn->dn_dirty_link[i])) { |
2531ce37 | 2089 | clean = B_FALSE; |
ec4f9b8f | 2090 | break; |
2531ce37 | 2091 | } |
34dc7c2f | 2092 | } |
66aca247 DB |
2093 | |
2094 | /* | |
2095 | * If compatibility option is on, sync any current changes before | |
2096 | * we go trundling through the block pointers. | |
2097 | */ | |
2098 | if (!clean && zfs_dmu_offset_next_sync) { | |
2099 | clean = B_TRUE; | |
34dc7c2f BB |
2100 | dnode_rele(dn, FTAG); |
2101 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 2102 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2103 | if (err) |
2104 | return (err); | |
2105 | } | |
2106 | ||
66aca247 DB |
2107 | if (clean) |
2108 | err = dnode_next_offset(dn, | |
2109 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2110 | else | |
2111 | err = SET_ERROR(EBUSY); | |
2112 | ||
34dc7c2f BB |
2113 | dnode_rele(dn, FTAG); |
2114 | ||
2115 | return (err); | |
2116 | } | |
2117 | ||
2118 | void | |
e0b0ca98 | 2119 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2120 | { |
e0b0ca98 | 2121 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2122 | |
34dc7c2f BB |
2123 | doi->doi_data_block_size = dn->dn_datablksz; |
2124 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2125 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2126 | doi->doi_type = dn->dn_type; |
2127 | doi->doi_bonus_type = dn->dn_bonustype; | |
2128 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2129 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2130 | doi->doi_indirection = dn->dn_nlevels; |
2131 | doi->doi_checksum = dn->dn_checksum; | |
2132 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2133 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2134 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2135 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2136 | doi->doi_fill_count = 0; |
1c27024e | 2137 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2138 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2139 | } |
2140 | ||
2141 | void | |
2142 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2143 | { | |
2144 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2145 | mutex_enter(&dn->dn_mtx); | |
2146 | ||
2147 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2148 | |
2149 | mutex_exit(&dn->dn_mtx); | |
2150 | rw_exit(&dn->dn_struct_rwlock); | |
2151 | } | |
2152 | ||
2153 | /* | |
2154 | * Get information on a DMU object. | |
2155 | * If doi is NULL, just indicates whether the object exists. | |
2156 | */ | |
2157 | int | |
2158 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2159 | { | |
2160 | dnode_t *dn; | |
428870ff | 2161 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2162 | |
2163 | if (err) | |
2164 | return (err); | |
2165 | ||
2166 | if (doi != NULL) | |
2167 | dmu_object_info_from_dnode(dn, doi); | |
2168 | ||
2169 | dnode_rele(dn, FTAG); | |
2170 | return (0); | |
2171 | } | |
2172 | ||
2173 | /* | |
2174 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2175 | */ | |
2176 | void | |
572e2857 | 2177 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2178 | { |
572e2857 BB |
2179 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2180 | ||
2181 | DB_DNODE_ENTER(db); | |
2182 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2183 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2184 | } |
2185 | ||
2186 | /* | |
2187 | * Faster still when you only care about the size. | |
34dc7c2f BB |
2188 | */ |
2189 | void | |
572e2857 BB |
2190 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2191 | u_longlong_t *nblk512) | |
34dc7c2f | 2192 | { |
572e2857 BB |
2193 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2194 | dnode_t *dn; | |
2195 | ||
2196 | DB_DNODE_ENTER(db); | |
2197 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2198 | |
2199 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2200 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2201 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2202 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2203 | DB_DNODE_EXIT(db); | |
2204 | } | |
2205 | ||
2206 | void | |
2207 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2208 | { | |
2209 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2210 | dnode_t *dn; | |
2211 | ||
2212 | DB_DNODE_ENTER(db); | |
2213 | dn = DB_DNODE(db); | |
2214 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2215 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2216 | } |
2217 | ||
2218 | void | |
2219 | byteswap_uint64_array(void *vbuf, size_t size) | |
2220 | { | |
2221 | uint64_t *buf = vbuf; | |
2222 | size_t count = size >> 3; | |
2223 | int i; | |
2224 | ||
2225 | ASSERT((size & 7) == 0); | |
2226 | ||
2227 | for (i = 0; i < count; i++) | |
2228 | buf[i] = BSWAP_64(buf[i]); | |
2229 | } | |
2230 | ||
2231 | void | |
2232 | byteswap_uint32_array(void *vbuf, size_t size) | |
2233 | { | |
2234 | uint32_t *buf = vbuf; | |
2235 | size_t count = size >> 2; | |
2236 | int i; | |
2237 | ||
2238 | ASSERT((size & 3) == 0); | |
2239 | ||
2240 | for (i = 0; i < count; i++) | |
2241 | buf[i] = BSWAP_32(buf[i]); | |
2242 | } | |
2243 | ||
2244 | void | |
2245 | byteswap_uint16_array(void *vbuf, size_t size) | |
2246 | { | |
2247 | uint16_t *buf = vbuf; | |
2248 | size_t count = size >> 1; | |
2249 | int i; | |
2250 | ||
2251 | ASSERT((size & 1) == 0); | |
2252 | ||
2253 | for (i = 0; i < count; i++) | |
2254 | buf[i] = BSWAP_16(buf[i]); | |
2255 | } | |
2256 | ||
2257 | /* ARGSUSED */ | |
2258 | void | |
2259 | byteswap_uint8_array(void *vbuf, size_t size) | |
2260 | { | |
2261 | } | |
2262 | ||
2263 | void | |
2264 | dmu_init(void) | |
2265 | { | |
a6255b7f | 2266 | abd_init(); |
428870ff | 2267 | zfs_dbgmsg_init(); |
572e2857 | 2268 | sa_cache_init(); |
572e2857 | 2269 | dmu_objset_init(); |
34dc7c2f | 2270 | dnode_init(); |
428870ff | 2271 | zfetch_init(); |
570827e1 | 2272 | dmu_tx_init(); |
34dc7c2f | 2273 | l2arc_init(); |
29809a6c | 2274 | arc_init(); |
d3c2ae1c | 2275 | dbuf_init(); |
34dc7c2f BB |
2276 | } |
2277 | ||
2278 | void | |
2279 | dmu_fini(void) | |
2280 | { | |
e49f1e20 | 2281 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2282 | l2arc_fini(); |
570827e1 | 2283 | dmu_tx_fini(); |
428870ff | 2284 | zfetch_fini(); |
34dc7c2f | 2285 | dbuf_fini(); |
572e2857 BB |
2286 | dnode_fini(); |
2287 | dmu_objset_fini(); | |
428870ff BB |
2288 | sa_cache_fini(); |
2289 | zfs_dbgmsg_fini(); | |
a6255b7f | 2290 | abd_fini(); |
34dc7c2f | 2291 | } |
c28b2279 | 2292 | |
c28b2279 | 2293 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2294 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2295 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2296 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2297 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2298 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2299 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2300 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2301 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2302 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2303 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2304 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2305 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2306 | EXPORT_SYMBOL(dmu_object_info); |
2307 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2308 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2309 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2310 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2311 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2312 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2313 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2314 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2315 | EXPORT_SYMBOL(dmu_object_set_compress); | |
94dac3e8 | 2316 | EXPORT_SYMBOL(dmu_offset_next); |
57b650b8 BB |
2317 | EXPORT_SYMBOL(dmu_write_policy); |
2318 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2319 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2320 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2321 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2322 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2323 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2324 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2325 | |
bef78122 | 2326 | /* BEGIN CSTYLED */ |
03fdcb9a MM |
2327 | ZFS_MODULE_PARAM(zfs, zfs_, nopwrite_enabled, INT, ZMOD_RW, |
2328 | "Enable NOP writes"); | |
03c6040b | 2329 | |
03fdcb9a MM |
2330 | ZFS_MODULE_PARAM(zfs, zfs_, per_txg_dirty_frees_percent, ULONG, ZMOD_RW, |
2331 | "Percentage of dirtied blocks from frees in one TXG"); | |
66aca247 | 2332 | |
03fdcb9a | 2333 | ZFS_MODULE_PARAM(zfs, zfs_, dmu_offset_next_sync, INT, ZMOD_RW, |
66aca247 DB |
2334 | "Enable forcing txg sync to find holes"); |
2335 | ||
03fdcb9a | 2336 | ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, INT, ZMOD_RW, |
d9b4bf06 | 2337 | "Limit one prefetch call to this size"); |
bef78122 | 2338 | /* END CSTYLED */ |