]>
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. |
82644107 | 23 | * Copyright (c) 2011, 2017 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. |
34dc7c2f BB |
28 | */ |
29 | ||
34dc7c2f BB |
30 | #include <sys/dmu.h> |
31 | #include <sys/dmu_impl.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/dbuf.h> | |
34 | #include <sys/dnode.h> | |
35 | #include <sys/zfs_context.h> | |
36 | #include <sys/dmu_objset.h> | |
37 | #include <sys/dmu_traverse.h> | |
38 | #include <sys/dsl_dataset.h> | |
39 | #include <sys/dsl_dir.h> | |
40 | #include <sys/dsl_pool.h> | |
41 | #include <sys/dsl_synctask.h> | |
42 | #include <sys/dsl_prop.h> | |
43 | #include <sys/dmu_zfetch.h> | |
44 | #include <sys/zfs_ioctl.h> | |
45 | #include <sys/zap.h> | |
46 | #include <sys/zio_checksum.h> | |
03c6040b | 47 | #include <sys/zio_compress.h> |
428870ff | 48 | #include <sys/sa.h> |
62bdd5eb | 49 | #include <sys/zfeature.h> |
a6255b7f | 50 | #include <sys/abd.h> |
539d33c7 | 51 | #include <sys/trace_dmu.h> |
34dc7c2f BB |
52 | #ifdef _KERNEL |
53 | #include <sys/vmsystm.h> | |
b128c09f | 54 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
55 | #endif |
56 | ||
03c6040b GW |
57 | /* |
58 | * Enable/disable nopwrite feature. | |
59 | */ | |
60 | int zfs_nopwrite_enabled = 1; | |
61 | ||
539d33c7 GM |
62 | /* |
63 | * Tunable to control percentage of dirtied blocks from frees in one TXG. | |
64 | * After this threshold is crossed, additional dirty blocks from frees | |
65 | * wait until the next TXG. | |
66 | * A value of zero will disable this throttle. | |
67 | */ | |
bef78122 | 68 | unsigned long zfs_per_txg_dirty_frees_percent = 30; |
539d33c7 | 69 | |
66aca247 DB |
70 | /* |
71 | * Enable/disable forcing txg sync when dirty in dmu_offset_next. | |
72 | */ | |
73 | int zfs_dmu_offset_next_sync = 0; | |
74 | ||
34dc7c2f | 75 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
b5256303 TC |
76 | { DMU_BSWAP_UINT8, TRUE, FALSE, "unallocated" }, |
77 | { DMU_BSWAP_ZAP, TRUE, FALSE, "object directory" }, | |
78 | { DMU_BSWAP_UINT64, TRUE, FALSE, "object array" }, | |
79 | { DMU_BSWAP_UINT8, TRUE, FALSE, "packed nvlist" }, | |
80 | { DMU_BSWAP_UINT64, TRUE, FALSE, "packed nvlist size" }, | |
81 | { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj" }, | |
82 | { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj header" }, | |
83 | { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA space map header" }, | |
84 | { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA space map" }, | |
85 | { DMU_BSWAP_UINT64, TRUE, TRUE, "ZIL intent log" }, | |
86 | { DMU_BSWAP_DNODE, TRUE, TRUE, "DMU dnode" }, | |
87 | { DMU_BSWAP_OBJSET, TRUE, FALSE, "DMU objset" }, | |
88 | { DMU_BSWAP_UINT64, TRUE, FALSE, "DSL directory" }, | |
89 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DSL directory child map"}, | |
90 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DSL dataset snap map" }, | |
91 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DSL props" }, | |
92 | { DMU_BSWAP_UINT64, TRUE, FALSE, "DSL dataset" }, | |
93 | { DMU_BSWAP_ZNODE, TRUE, FALSE, "ZFS znode" }, | |
94 | { DMU_BSWAP_OLDACL, TRUE, TRUE, "ZFS V0 ACL" }, | |
95 | { DMU_BSWAP_UINT8, FALSE, TRUE, "ZFS plain file" }, | |
96 | { DMU_BSWAP_ZAP, TRUE, TRUE, "ZFS directory" }, | |
97 | { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS master node" }, | |
98 | { DMU_BSWAP_ZAP, TRUE, TRUE, "ZFS delete queue" }, | |
99 | { DMU_BSWAP_UINT8, FALSE, TRUE, "zvol object" }, | |
100 | { DMU_BSWAP_ZAP, TRUE, FALSE, "zvol prop" }, | |
101 | { DMU_BSWAP_UINT8, FALSE, TRUE, "other uint8[]" }, | |
102 | { DMU_BSWAP_UINT64, FALSE, TRUE, "other uint64[]" }, | |
103 | { DMU_BSWAP_ZAP, TRUE, FALSE, "other ZAP" }, | |
104 | { DMU_BSWAP_ZAP, TRUE, FALSE, "persistent error log" }, | |
105 | { DMU_BSWAP_UINT8, TRUE, FALSE, "SPA history" }, | |
106 | { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA history offsets" }, | |
107 | { DMU_BSWAP_ZAP, TRUE, FALSE, "Pool properties" }, | |
108 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DSL permissions" }, | |
109 | { DMU_BSWAP_ACL, TRUE, TRUE, "ZFS ACL" }, | |
110 | { DMU_BSWAP_UINT8, TRUE, TRUE, "ZFS SYSACL" }, | |
111 | { DMU_BSWAP_UINT8, TRUE, TRUE, "FUID table" }, | |
112 | { DMU_BSWAP_UINT64, TRUE, FALSE, "FUID table size" }, | |
113 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DSL dataset next clones"}, | |
114 | { DMU_BSWAP_ZAP, TRUE, FALSE, "scan work queue" }, | |
115 | { DMU_BSWAP_ZAP, TRUE, TRUE, "ZFS user/group used" }, | |
116 | { DMU_BSWAP_ZAP, TRUE, TRUE, "ZFS user/group quota" }, | |
117 | { DMU_BSWAP_ZAP, TRUE, FALSE, "snapshot refcount tags"}, | |
118 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DDT ZAP algorithm" }, | |
119 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DDT statistics" }, | |
120 | { DMU_BSWAP_UINT8, TRUE, TRUE, "System attributes" }, | |
121 | { DMU_BSWAP_ZAP, TRUE, TRUE, "SA master node" }, | |
122 | { DMU_BSWAP_ZAP, TRUE, TRUE, "SA attr registration" }, | |
123 | { DMU_BSWAP_ZAP, TRUE, TRUE, "SA attr layouts" }, | |
124 | { DMU_BSWAP_ZAP, TRUE, FALSE, "scan translations" }, | |
125 | { DMU_BSWAP_UINT8, FALSE, TRUE, "deduplicated block" }, | |
126 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DSL deadlist map" }, | |
127 | { DMU_BSWAP_UINT64, TRUE, FALSE, "DSL deadlist map hdr" }, | |
128 | { DMU_BSWAP_ZAP, TRUE, FALSE, "DSL dir clones" }, | |
129 | { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
130 | }; |
131 | ||
132 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
133 | { byteswap_uint8_array, "uint8" }, | |
134 | { byteswap_uint16_array, "uint16" }, | |
135 | { byteswap_uint32_array, "uint32" }, | |
136 | { byteswap_uint64_array, "uint64" }, | |
137 | { zap_byteswap, "zap" }, | |
138 | { dnode_buf_byteswap, "dnode" }, | |
139 | { dmu_objset_byteswap, "objset" }, | |
140 | { zfs_znode_byteswap, "znode" }, | |
141 | { zfs_oldacl_byteswap, "oldacl" }, | |
142 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
143 | }; |
144 | ||
2bce8049 MA |
145 | int |
146 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, | |
147 | void *tag, dmu_buf_t **dbp) | |
148 | { | |
149 | uint64_t blkid; | |
150 | dmu_buf_impl_t *db; | |
151 | ||
152 | blkid = dbuf_whichblock(dn, 0, offset); | |
153 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
154 | db = dbuf_hold(dn, blkid, tag); | |
155 | rw_exit(&dn->dn_struct_rwlock); | |
156 | ||
157 | if (db == NULL) { | |
158 | *dbp = NULL; | |
159 | return (SET_ERROR(EIO)); | |
160 | } | |
161 | ||
162 | *dbp = &db->db; | |
163 | return (0); | |
164 | } | |
34dc7c2f | 165 | int |
9b67f605 MA |
166 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
167 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
168 | { |
169 | dnode_t *dn; | |
170 | uint64_t blkid; | |
171 | dmu_buf_impl_t *db; | |
172 | int err; | |
428870ff BB |
173 | |
174 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
175 | if (err) |
176 | return (err); | |
fcff0f35 | 177 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
178 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
179 | db = dbuf_hold(dn, blkid, tag); | |
180 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
181 | dnode_rele(dn, FTAG); |
182 | ||
34dc7c2f | 183 | if (db == NULL) { |
9b67f605 MA |
184 | *dbp = NULL; |
185 | return (SET_ERROR(EIO)); | |
186 | } | |
187 | ||
188 | *dbp = &db->db; | |
189 | return (err); | |
190 | } | |
191 | ||
2bce8049 MA |
192 | int |
193 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
194 | void *tag, dmu_buf_t **dbp, int flags) | |
195 | { | |
196 | int err; | |
197 | int db_flags = DB_RF_CANFAIL; | |
198 | ||
199 | if (flags & DMU_READ_NO_PREFETCH) | |
200 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
201 | if (flags & DMU_READ_NO_DECRYPT) |
202 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
203 | |
204 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
205 | if (err == 0) { | |
206 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
207 | err = dbuf_read(db, NULL, db_flags); | |
208 | if (err != 0) { | |
209 | dbuf_rele(db, tag); | |
210 | *dbp = NULL; | |
211 | } | |
212 | } | |
213 | ||
214 | return (err); | |
215 | } | |
216 | ||
9b67f605 MA |
217 | int |
218 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
219 | void *tag, dmu_buf_t **dbp, int flags) | |
220 | { | |
221 | int err; | |
222 | int db_flags = DB_RF_CANFAIL; | |
223 | ||
224 | if (flags & DMU_READ_NO_PREFETCH) | |
225 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
226 | if (flags & DMU_READ_NO_DECRYPT) |
227 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
228 | |
229 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
230 | if (err == 0) { | |
231 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 232 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 233 | if (err != 0) { |
34dc7c2f | 234 | dbuf_rele(db, tag); |
9b67f605 | 235 | *dbp = NULL; |
34dc7c2f BB |
236 | } |
237 | } | |
238 | ||
34dc7c2f BB |
239 | return (err); |
240 | } | |
241 | ||
242 | int | |
243 | dmu_bonus_max(void) | |
244 | { | |
50c957f7 | 245 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
246 | } |
247 | ||
248 | int | |
572e2857 | 249 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 250 | { |
572e2857 BB |
251 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
252 | dnode_t *dn; | |
253 | int error; | |
34dc7c2f | 254 | |
572e2857 BB |
255 | DB_DNODE_ENTER(db); |
256 | dn = DB_DNODE(db); | |
257 | ||
258 | if (dn->dn_bonus != db) { | |
2e528b49 | 259 | error = SET_ERROR(EINVAL); |
572e2857 | 260 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 261 | error = SET_ERROR(EINVAL); |
572e2857 BB |
262 | } else { |
263 | dnode_setbonuslen(dn, newsize, tx); | |
264 | error = 0; | |
265 | } | |
266 | ||
267 | DB_DNODE_EXIT(db); | |
268 | return (error); | |
34dc7c2f BB |
269 | } |
270 | ||
428870ff | 271 | int |
572e2857 | 272 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 273 | { |
572e2857 BB |
274 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
275 | dnode_t *dn; | |
276 | int error; | |
428870ff | 277 | |
572e2857 BB |
278 | DB_DNODE_ENTER(db); |
279 | dn = DB_DNODE(db); | |
428870ff | 280 | |
9ae529ec | 281 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 282 | error = SET_ERROR(EINVAL); |
572e2857 | 283 | } else if (dn->dn_bonus != db) { |
2e528b49 | 284 | error = SET_ERROR(EINVAL); |
572e2857 BB |
285 | } else { |
286 | dnode_setbonus_type(dn, type, tx); | |
287 | error = 0; | |
288 | } | |
428870ff | 289 | |
572e2857 BB |
290 | DB_DNODE_EXIT(db); |
291 | return (error); | |
292 | } | |
293 | ||
294 | dmu_object_type_t | |
295 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
296 | { | |
297 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
298 | dnode_t *dn; | |
299 | dmu_object_type_t type; | |
300 | ||
301 | DB_DNODE_ENTER(db); | |
302 | dn = DB_DNODE(db); | |
303 | type = dn->dn_bonustype; | |
304 | DB_DNODE_EXIT(db); | |
305 | ||
306 | return (type); | |
428870ff BB |
307 | } |
308 | ||
309 | int | |
310 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
311 | { | |
312 | dnode_t *dn; | |
313 | int error; | |
314 | ||
315 | error = dnode_hold(os, object, FTAG, &dn); | |
316 | dbuf_rm_spill(dn, tx); | |
317 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
318 | dnode_rm_spill(dn, tx); | |
319 | rw_exit(&dn->dn_struct_rwlock); | |
320 | dnode_rele(dn, FTAG); | |
321 | return (error); | |
322 | } | |
323 | ||
34dc7c2f BB |
324 | /* |
325 | * returns ENOENT, EIO, or 0. | |
326 | */ | |
327 | int | |
b5256303 TC |
328 | dmu_bonus_hold_impl(objset_t *os, uint64_t object, void *tag, uint32_t flags, |
329 | dmu_buf_t **dbp) | |
34dc7c2f BB |
330 | { |
331 | dnode_t *dn; | |
332 | dmu_buf_impl_t *db; | |
333 | int error; | |
b5256303 TC |
334 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
335 | ||
336 | if (flags & DMU_READ_NO_PREFETCH) | |
337 | db_flags |= DB_RF_NOPREFETCH; | |
338 | if (flags & DMU_READ_NO_DECRYPT) | |
339 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 340 | |
428870ff | 341 | error = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
342 | if (error) |
343 | return (error); | |
344 | ||
345 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
346 | if (dn->dn_bonus == NULL) { | |
347 | rw_exit(&dn->dn_struct_rwlock); | |
348 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
349 | if (dn->dn_bonus == NULL) | |
350 | dbuf_create_bonus(dn); | |
351 | } | |
352 | db = dn->dn_bonus; | |
34dc7c2f BB |
353 | |
354 | /* as long as the bonus buf is held, the dnode will be held */ | |
572e2857 | 355 | if (refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 356 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 357 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
358 | } |
359 | ||
360 | /* | |
361 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
362 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
363 | * a dnode hold for every dbuf. | |
364 | */ | |
365 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
366 | |
367 | dnode_rele(dn, FTAG); | |
368 | ||
b5256303 TC |
369 | error = dbuf_read(db, NULL, db_flags); |
370 | if (error) { | |
371 | dnode_evict_bonus(dn); | |
372 | dbuf_rele(db, tag); | |
373 | *dbp = NULL; | |
374 | return (error); | |
375 | } | |
34dc7c2f BB |
376 | |
377 | *dbp = &db->db; | |
378 | return (0); | |
379 | } | |
380 | ||
b5256303 TC |
381 | int |
382 | dmu_bonus_hold(objset_t *os, uint64_t obj, void *tag, dmu_buf_t **dbp) | |
383 | { | |
384 | return (dmu_bonus_hold_impl(os, obj, tag, DMU_READ_NO_PREFETCH, dbp)); | |
385 | } | |
386 | ||
428870ff BB |
387 | /* |
388 | * returns ENOENT, EIO, or 0. | |
389 | * | |
390 | * This interface will allocate a blank spill dbuf when a spill blk | |
391 | * doesn't already exist on the dnode. | |
392 | * | |
393 | * if you only want to find an already existing spill db, then | |
394 | * dmu_spill_hold_existing() should be used. | |
395 | */ | |
396 | int | |
397 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
398 | { | |
399 | dmu_buf_impl_t *db = NULL; | |
400 | int err; | |
401 | ||
402 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
403 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
404 | ||
405 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
406 | ||
407 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
408 | rw_exit(&dn->dn_struct_rwlock); | |
409 | ||
b182ac00 | 410 | if (db == NULL) { |
411 | *dbp = NULL; | |
412 | return (SET_ERROR(EIO)); | |
413 | } | |
572e2857 BB |
414 | err = dbuf_read(db, NULL, flags); |
415 | if (err == 0) | |
416 | *dbp = &db->db; | |
b182ac00 | 417 | else { |
572e2857 | 418 | dbuf_rele(db, tag); |
b182ac00 | 419 | *dbp = NULL; |
420 | } | |
428870ff BB |
421 | return (err); |
422 | } | |
423 | ||
424 | int | |
425 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
426 | { | |
572e2857 BB |
427 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
428 | dnode_t *dn; | |
428870ff BB |
429 | int err; |
430 | ||
572e2857 BB |
431 | DB_DNODE_ENTER(db); |
432 | dn = DB_DNODE(db); | |
433 | ||
434 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 435 | err = SET_ERROR(EINVAL); |
572e2857 BB |
436 | } else { |
437 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
438 | ||
439 | if (!dn->dn_have_spill) { | |
2e528b49 | 440 | err = SET_ERROR(ENOENT); |
572e2857 BB |
441 | } else { |
442 | err = dmu_spill_hold_by_dnode(dn, | |
443 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
444 | } | |
428870ff | 445 | |
428870ff | 446 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 447 | } |
572e2857 BB |
448 | |
449 | DB_DNODE_EXIT(db); | |
428870ff BB |
450 | return (err); |
451 | } | |
452 | ||
453 | int | |
454 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
455 | { | |
572e2857 BB |
456 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
457 | dnode_t *dn; | |
458 | int err; | |
459 | ||
460 | DB_DNODE_ENTER(db); | |
461 | dn = DB_DNODE(db); | |
462 | err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); | |
463 | DB_DNODE_EXIT(db); | |
464 | ||
465 | return (err); | |
428870ff BB |
466 | } |
467 | ||
34dc7c2f BB |
468 | /* |
469 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
470 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
471 | * and can induce severe lock contention when writing to several files | |
472 | * whose dnodes are in the same block. | |
473 | */ | |
474 | static int | |
9babb374 | 475 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
7f60329a | 476 | boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
34dc7c2f BB |
477 | { |
478 | dmu_buf_t **dbp; | |
479 | uint64_t blkid, nblks, i; | |
9babb374 | 480 | uint32_t dbuf_flags; |
34dc7c2f BB |
481 | int err; |
482 | zio_t *zio; | |
483 | ||
484 | ASSERT(length <= DMU_MAX_ACCESS); | |
485 | ||
7f60329a MA |
486 | /* |
487 | * Note: We directly notify the prefetch code of this read, so that | |
488 | * we can tell it about the multi-block read. dbuf_read() only knows | |
489 | * about the one block it is accessing. | |
490 | */ | |
491 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
492 | DB_RF_NOPREFETCH; | |
34dc7c2f BB |
493 | |
494 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
495 | if (dn->dn_datablkshift) { | |
496 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
497 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
498 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
499 | } else { |
500 | if (offset + length > dn->dn_datablksz) { | |
501 | zfs_panic_recover("zfs: accessing past end of object " | |
502 | "%llx/%llx (size=%u access=%llu+%llu)", | |
503 | (longlong_t)dn->dn_objset-> | |
504 | os_dsl_dataset->ds_object, | |
505 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
506 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 507 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 508 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
509 | } |
510 | nblks = 1; | |
511 | } | |
79c76d5b | 512 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 513 | |
b128c09f | 514 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
fcff0f35 | 515 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f | 516 | for (i = 0; i < nblks; i++) { |
7f60329a | 517 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f BB |
518 | if (db == NULL) { |
519 | rw_exit(&dn->dn_struct_rwlock); | |
520 | dmu_buf_rele_array(dbp, nblks, tag); | |
521 | zio_nowait(zio); | |
2e528b49 | 522 | return (SET_ERROR(EIO)); |
34dc7c2f | 523 | } |
7f60329a | 524 | |
34dc7c2f | 525 | /* initiate async i/o */ |
7f60329a | 526 | if (read) |
9babb374 | 527 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
528 | dbp[i] = &db->db; |
529 | } | |
7f60329a | 530 | |
755065f3 AM |
531 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
532 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
533 | dmu_zfetch(&dn->dn_zfetch, blkid, nblks, | |
534 | read && DNODE_IS_CACHEABLE(dn)); | |
7f60329a | 535 | } |
34dc7c2f BB |
536 | rw_exit(&dn->dn_struct_rwlock); |
537 | ||
538 | /* wait for async i/o */ | |
539 | err = zio_wait(zio); | |
540 | if (err) { | |
541 | dmu_buf_rele_array(dbp, nblks, tag); | |
542 | return (err); | |
543 | } | |
544 | ||
545 | /* wait for other io to complete */ | |
546 | if (read) { | |
547 | for (i = 0; i < nblks; i++) { | |
548 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
549 | mutex_enter(&db->db_mtx); | |
550 | while (db->db_state == DB_READ || | |
551 | db->db_state == DB_FILL) | |
552 | cv_wait(&db->db_changed, &db->db_mtx); | |
553 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 554 | err = SET_ERROR(EIO); |
34dc7c2f BB |
555 | mutex_exit(&db->db_mtx); |
556 | if (err) { | |
557 | dmu_buf_rele_array(dbp, nblks, tag); | |
558 | return (err); | |
559 | } | |
560 | } | |
561 | } | |
562 | ||
563 | *numbufsp = nblks; | |
564 | *dbpp = dbp; | |
565 | return (0); | |
566 | } | |
567 | ||
568 | static int | |
569 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
570 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
571 | { | |
572 | dnode_t *dn; | |
573 | int err; | |
574 | ||
428870ff | 575 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
576 | if (err) |
577 | return (err); | |
578 | ||
579 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 580 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
581 | |
582 | dnode_rele(dn, FTAG); | |
583 | ||
584 | return (err); | |
585 | } | |
586 | ||
587 | int | |
572e2857 | 588 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
7f60329a MA |
589 | uint64_t length, boolean_t read, void *tag, int *numbufsp, |
590 | dmu_buf_t ***dbpp) | |
34dc7c2f | 591 | { |
572e2857 BB |
592 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
593 | dnode_t *dn; | |
34dc7c2f BB |
594 | int err; |
595 | ||
572e2857 BB |
596 | DB_DNODE_ENTER(db); |
597 | dn = DB_DNODE(db); | |
34dc7c2f | 598 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 599 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 600 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
601 | |
602 | return (err); | |
603 | } | |
604 | ||
605 | void | |
606 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
607 | { | |
608 | int i; | |
609 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
610 | ||
611 | if (numbufs == 0) | |
612 | return; | |
613 | ||
614 | for (i = 0; i < numbufs; i++) { | |
615 | if (dbp[i]) | |
616 | dbuf_rele(dbp[i], tag); | |
617 | } | |
618 | ||
619 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
620 | } | |
621 | ||
e8b96c60 | 622 | /* |
fcff0f35 PD |
623 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
624 | * indirect blocks prefeteched will be those that point to the blocks containing | |
625 | * the data starting at offset, and continuing to offset + len. | |
e8b96c60 | 626 | * |
b5256303 TC |
627 | * Note that if the indirect blocks above the blocks being prefetched are not |
628 | * in cache, they will be asychronously read in. | |
e8b96c60 | 629 | */ |
34dc7c2f | 630 | void |
fcff0f35 PD |
631 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
632 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
633 | { |
634 | dnode_t *dn; | |
635 | uint64_t blkid; | |
e8b96c60 | 636 | int nblks, err; |
34dc7c2f | 637 | |
34dc7c2f | 638 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 639 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
640 | |
641 | if (object == 0 || object >= DN_MAX_OBJECT) | |
642 | return; | |
643 | ||
644 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
645 | blkid = dbuf_whichblock(dn, level, |
646 | object * sizeof (dnode_phys_t)); | |
647 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
648 | rw_exit(&dn->dn_struct_rwlock); |
649 | return; | |
650 | } | |
651 | ||
652 | /* | |
653 | * XXX - Note, if the dnode for the requested object is not | |
654 | * already cached, we will do a *synchronous* read in the | |
655 | * dnode_hold() call. The same is true for any indirects. | |
656 | */ | |
428870ff | 657 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
658 | if (err != 0) |
659 | return; | |
660 | ||
661 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
662 | /* |
663 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
664 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
665 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
666 | * offset) is the first. Then the number we need to prefetch is the | |
667 | * last - first + 1. | |
668 | */ | |
669 | if (level > 0 || dn->dn_datablkshift != 0) { | |
670 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
671 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
672 | } else { |
673 | nblks = (offset < dn->dn_datablksz); | |
674 | } | |
675 | ||
676 | if (nblks != 0) { | |
e8b96c60 MA |
677 | int i; |
678 | ||
fcff0f35 | 679 | blkid = dbuf_whichblock(dn, level, offset); |
34dc7c2f | 680 | for (i = 0; i < nblks; i++) |
fcff0f35 | 681 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f BB |
682 | } |
683 | ||
684 | rw_exit(&dn->dn_struct_rwlock); | |
685 | ||
686 | dnode_rele(dn, FTAG); | |
687 | } | |
688 | ||
45d1cae3 BB |
689 | /* |
690 | * Get the next "chunk" of file data to free. We traverse the file from | |
691 | * the end so that the file gets shorter over time (if we crashes in the | |
692 | * middle, this will leave us in a better state). We find allocated file | |
693 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
694 | * |
695 | * On input, *start should be the first offset that does not need to be | |
696 | * freed (e.g. "offset + length"). On return, *start will be the first | |
697 | * offset that should be freed. | |
45d1cae3 | 698 | */ |
b128c09f | 699 | static int |
b663a23d | 700 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) |
b128c09f | 701 | { |
b663a23d MA |
702 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
703 | /* bytes of data covered by a level-1 indirect block */ | |
45d1cae3 | 704 | uint64_t iblkrange = |
b128c09f | 705 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
b663a23d | 706 | uint64_t blks; |
b128c09f | 707 | |
b663a23d | 708 | ASSERT3U(minimum, <=, *start); |
b128c09f | 709 | |
b663a23d MA |
710 | if (*start - minimum <= iblkrange * maxblks) { |
711 | *start = minimum; | |
b128c09f BB |
712 | return (0); |
713 | } | |
45d1cae3 | 714 | ASSERT(ISP2(iblkrange)); |
b128c09f | 715 | |
b663a23d | 716 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
717 | int err; |
718 | ||
b663a23d MA |
719 | /* |
720 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
721 | * indirect block at or before the input offset. We must | |
722 | * decrement *start so that it is at the end of the region | |
723 | * to search. | |
724 | */ | |
725 | (*start)--; | |
b128c09f | 726 | err = dnode_next_offset(dn, |
45d1cae3 | 727 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 728 | |
b663a23d | 729 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 730 | if (err == ESRCH) { |
b663a23d MA |
731 | *start = minimum; |
732 | break; | |
733 | } else if (err != 0) { | |
b128c09f | 734 | return (err); |
45d1cae3 | 735 | } |
b128c09f | 736 | |
b663a23d | 737 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 738 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 739 | } |
b663a23d MA |
740 | if (*start < minimum) |
741 | *start = minimum; | |
b128c09f BB |
742 | return (0); |
743 | } | |
744 | ||
a08abc1b GM |
745 | /* |
746 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
747 | * otherwise return false. | |
748 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
749 | */ | |
750 | /*ARGSUSED*/ | |
751 | static boolean_t | |
752 | dmu_objset_zfs_unmounting(objset_t *os) | |
753 | { | |
754 | #ifdef _KERNEL | |
755 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
756 | return (zfs_get_vfs_flag_unmounted(os)); | |
757 | #endif | |
758 | return (B_FALSE); | |
759 | } | |
760 | ||
b128c09f BB |
761 | static int |
762 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
b663a23d | 763 | uint64_t length) |
b128c09f | 764 | { |
c97d3069 | 765 | uint64_t object_size; |
b663a23d | 766 | int err; |
539d33c7 GM |
767 | uint64_t dirty_frees_threshold; |
768 | dsl_pool_t *dp = dmu_objset_pool(os); | |
769 | int t; | |
b663a23d | 770 | |
c97d3069 BB |
771 | if (dn == NULL) |
772 | return (SET_ERROR(EINVAL)); | |
773 | ||
774 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 775 | if (offset >= object_size) |
b128c09f | 776 | return (0); |
b128c09f | 777 | |
539d33c7 GM |
778 | if (zfs_per_txg_dirty_frees_percent <= 100) |
779 | dirty_frees_threshold = | |
780 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
781 | else | |
782 | dirty_frees_threshold = zfs_dirty_data_max / 4; | |
783 | ||
b663a23d MA |
784 | if (length == DMU_OBJECT_END || offset + length > object_size) |
785 | length = object_size - offset; | |
786 | ||
787 | while (length != 0) { | |
539d33c7 GM |
788 | uint64_t chunk_end, chunk_begin, chunk_len; |
789 | uint64_t long_free_dirty_all_txgs = 0; | |
b663a23d MA |
790 | dmu_tx_t *tx; |
791 | ||
a08abc1b GM |
792 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
793 | return (SET_ERROR(EINTR)); | |
794 | ||
b663a23d MA |
795 | chunk_end = chunk_begin = offset + length; |
796 | ||
797 | /* move chunk_begin backwards to the beginning of this chunk */ | |
798 | err = get_next_chunk(dn, &chunk_begin, offset); | |
b128c09f BB |
799 | if (err) |
800 | return (err); | |
b663a23d MA |
801 | ASSERT3U(chunk_begin, >=, offset); |
802 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 803 | |
539d33c7 GM |
804 | chunk_len = chunk_end - chunk_begin; |
805 | ||
806 | mutex_enter(&dp->dp_lock); | |
807 | for (t = 0; t < TXG_SIZE; t++) { | |
808 | long_free_dirty_all_txgs += | |
809 | dp->dp_long_free_dirty_pertxg[t]; | |
810 | } | |
811 | mutex_exit(&dp->dp_lock); | |
812 | ||
813 | /* | |
814 | * To avoid filling up a TXG with just frees wait for | |
815 | * the next TXG to open before freeing more chunks if | |
816 | * we have reached the threshold of frees | |
817 | */ | |
818 | if (dirty_frees_threshold != 0 && | |
819 | long_free_dirty_all_txgs >= dirty_frees_threshold) { | |
820 | txg_wait_open(dp, 0); | |
821 | continue; | |
822 | } | |
823 | ||
b128c09f | 824 | tx = dmu_tx_create(os); |
539d33c7 | 825 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
826 | |
827 | /* | |
828 | * Mark this transaction as typically resulting in a net | |
829 | * reduction in space used. | |
830 | */ | |
831 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
832 | err = dmu_tx_assign(tx, TXG_WAIT); |
833 | if (err) { | |
834 | dmu_tx_abort(tx); | |
835 | return (err); | |
836 | } | |
539d33c7 GM |
837 | |
838 | mutex_enter(&dp->dp_lock); | |
839 | dp->dp_long_free_dirty_pertxg[dmu_tx_get_txg(tx) & TXG_MASK] += | |
840 | chunk_len; | |
841 | mutex_exit(&dp->dp_lock); | |
842 | DTRACE_PROBE3(free__long__range, | |
843 | uint64_t, long_free_dirty_all_txgs, uint64_t, chunk_len, | |
844 | uint64_t, dmu_tx_get_txg(tx)); | |
845 | dnode_free_range(dn, chunk_begin, chunk_len, tx); | |
b128c09f | 846 | dmu_tx_commit(tx); |
b663a23d | 847 | |
539d33c7 | 848 | length -= chunk_len; |
b128c09f BB |
849 | } |
850 | return (0); | |
851 | } | |
852 | ||
853 | int | |
854 | dmu_free_long_range(objset_t *os, uint64_t object, | |
855 | uint64_t offset, uint64_t length) | |
856 | { | |
857 | dnode_t *dn; | |
858 | int err; | |
859 | ||
428870ff | 860 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
861 | if (err != 0) |
862 | return (err); | |
b663a23d | 863 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
864 | |
865 | /* | |
866 | * It is important to zero out the maxblkid when freeing the entire | |
867 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
868 | * will take the fast path, and (b) dnode_reallocate() can verify | |
869 | * that the entire file has been freed. | |
870 | */ | |
b0bc7a84 | 871 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
872 | dn->dn_maxblkid = 0; |
873 | ||
b128c09f BB |
874 | dnode_rele(dn, FTAG); |
875 | return (err); | |
876 | } | |
877 | ||
878 | int | |
b663a23d | 879 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 880 | { |
b128c09f BB |
881 | dmu_tx_t *tx; |
882 | int err; | |
883 | ||
b663a23d | 884 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
885 | if (err != 0) |
886 | return (err); | |
b663a23d MA |
887 | |
888 | tx = dmu_tx_create(os); | |
889 | dmu_tx_hold_bonus(tx, object); | |
890 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 891 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
892 | err = dmu_tx_assign(tx, TXG_WAIT); |
893 | if (err == 0) { | |
894 | err = dmu_object_free(os, object, tx); | |
895 | dmu_tx_commit(tx); | |
b128c09f | 896 | } else { |
b663a23d | 897 | dmu_tx_abort(tx); |
b128c09f | 898 | } |
b663a23d | 899 | |
b128c09f BB |
900 | return (err); |
901 | } | |
902 | ||
34dc7c2f BB |
903 | int |
904 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
905 | uint64_t size, dmu_tx_t *tx) | |
906 | { | |
907 | dnode_t *dn; | |
428870ff | 908 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
909 | if (err) |
910 | return (err); | |
911 | ASSERT(offset < UINT64_MAX); | |
912 | ASSERT(size == -1ULL || size <= UINT64_MAX - offset); | |
913 | dnode_free_range(dn, offset, size, tx); | |
914 | dnode_rele(dn, FTAG); | |
915 | return (0); | |
916 | } | |
917 | ||
0eef1bde | 918 | static int |
919 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 920 | void *buf, uint32_t flags) |
34dc7c2f | 921 | { |
34dc7c2f | 922 | dmu_buf_t **dbp; |
0eef1bde | 923 | int numbufs, err = 0; |
34dc7c2f BB |
924 | |
925 | /* | |
926 | * Deal with odd block sizes, where there can't be data past the first | |
927 | * block. If we ever do the tail block optimization, we will need to | |
928 | * handle that here as well. | |
929 | */ | |
45d1cae3 | 930 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 931 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
932 | MIN(size, dn->dn_datablksz - offset); |
933 | bzero((char *)buf + newsz, size - newsz); | |
934 | size = newsz; | |
935 | } | |
936 | ||
937 | while (size > 0) { | |
938 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 939 | int i; |
34dc7c2f BB |
940 | |
941 | /* | |
942 | * NB: we could do this block-at-a-time, but it's nice | |
943 | * to be reading in parallel. | |
944 | */ | |
945 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 946 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
947 | if (err) |
948 | break; | |
949 | ||
950 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
951 | uint64_t tocpy; |
952 | int64_t bufoff; | |
34dc7c2f BB |
953 | dmu_buf_t *db = dbp[i]; |
954 | ||
955 | ASSERT(size > 0); | |
956 | ||
957 | bufoff = offset - db->db_offset; | |
c9520ecc | 958 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 959 | |
c9520ecc | 960 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
961 | |
962 | offset += tocpy; | |
963 | size -= tocpy; | |
964 | buf = (char *)buf + tocpy; | |
965 | } | |
966 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
967 | } | |
34dc7c2f BB |
968 | return (err); |
969 | } | |
970 | ||
0eef1bde | 971 | int |
972 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
973 | void *buf, uint32_t flags) | |
34dc7c2f | 974 | { |
0eef1bde | 975 | dnode_t *dn; |
976 | int err; | |
34dc7c2f | 977 | |
0eef1bde | 978 | err = dnode_hold(os, object, FTAG, &dn); |
979 | if (err != 0) | |
980 | return (err); | |
34dc7c2f | 981 | |
0eef1bde | 982 | err = dmu_read_impl(dn, offset, size, buf, flags); |
983 | dnode_rele(dn, FTAG); | |
984 | return (err); | |
985 | } | |
986 | ||
987 | int | |
988 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
989 | uint32_t flags) | |
990 | { | |
991 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
992 | } | |
993 | ||
994 | static void | |
995 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
996 | const void *buf, dmu_tx_t *tx) | |
997 | { | |
998 | int i; | |
34dc7c2f BB |
999 | |
1000 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1001 | uint64_t tocpy; |
1002 | int64_t bufoff; | |
34dc7c2f BB |
1003 | dmu_buf_t *db = dbp[i]; |
1004 | ||
1005 | ASSERT(size > 0); | |
1006 | ||
1007 | bufoff = offset - db->db_offset; | |
c9520ecc | 1008 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1009 | |
1010 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1011 | ||
1012 | if (tocpy == db->db_size) | |
1013 | dmu_buf_will_fill(db, tx); | |
1014 | else | |
1015 | dmu_buf_will_dirty(db, tx); | |
1016 | ||
60101509 | 1017 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1018 | |
1019 | if (tocpy == db->db_size) | |
1020 | dmu_buf_fill_done(db, tx); | |
1021 | ||
1022 | offset += tocpy; | |
1023 | size -= tocpy; | |
1024 | buf = (char *)buf + tocpy; | |
1025 | } | |
0eef1bde | 1026 | } |
1027 | ||
1028 | void | |
1029 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1030 | const void *buf, dmu_tx_t *tx) | |
1031 | { | |
1032 | dmu_buf_t **dbp; | |
1033 | int numbufs; | |
1034 | ||
1035 | if (size == 0) | |
1036 | return; | |
1037 | ||
1038 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1039 | FALSE, FTAG, &numbufs, &dbp)); | |
1040 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1041 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1042 | } | |
1043 | ||
1044 | void | |
1045 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1046 | const void *buf, dmu_tx_t *tx) | |
1047 | { | |
1048 | dmu_buf_t **dbp; | |
1049 | int numbufs; | |
1050 | ||
1051 | if (size == 0) | |
1052 | return; | |
1053 | ||
1054 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1055 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1056 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1057 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1058 | } | |
1059 | ||
b128c09f BB |
1060 | void |
1061 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1062 | dmu_tx_t *tx) | |
1063 | { | |
1064 | dmu_buf_t **dbp; | |
1065 | int numbufs, i; | |
1066 | ||
1067 | if (size == 0) | |
1068 | return; | |
1069 | ||
1070 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1071 | FALSE, FTAG, &numbufs, &dbp)); | |
1072 | ||
1073 | for (i = 0; i < numbufs; i++) { | |
1074 | dmu_buf_t *db = dbp[i]; | |
1075 | ||
1076 | dmu_buf_will_not_fill(db, tx); | |
1077 | } | |
1078 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1079 | } | |
1080 | ||
9b67f605 MA |
1081 | void |
1082 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1083 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1084 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1085 | { | |
1086 | dmu_buf_t *db; | |
1087 | ||
1088 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1089 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1090 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1091 | FTAG, &db)); | |
1092 | ||
1093 | dmu_buf_write_embedded(db, | |
1094 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1095 | uncompressed_size, compressed_size, byteorder, tx); | |
1096 | ||
1097 | dmu_buf_rele(db, FTAG); | |
1098 | } | |
1099 | ||
428870ff BB |
1100 | /* |
1101 | * DMU support for xuio | |
1102 | */ | |
1103 | kstat_t *xuio_ksp = NULL; | |
1104 | ||
59e6e7ca BB |
1105 | typedef struct xuio_stats { |
1106 | /* loaned yet not returned arc_buf */ | |
1107 | kstat_named_t xuiostat_onloan_rbuf; | |
1108 | kstat_named_t xuiostat_onloan_wbuf; | |
1109 | /* whether a copy is made when loaning out a read buffer */ | |
1110 | kstat_named_t xuiostat_rbuf_copied; | |
1111 | kstat_named_t xuiostat_rbuf_nocopy; | |
1112 | /* whether a copy is made when assigning a write buffer */ | |
1113 | kstat_named_t xuiostat_wbuf_copied; | |
1114 | kstat_named_t xuiostat_wbuf_nocopy; | |
1115 | } xuio_stats_t; | |
1116 | ||
1117 | static xuio_stats_t xuio_stats = { | |
1118 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
1119 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
1120 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
1121 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
1122 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
1123 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
1124 | }; | |
1125 | ||
d1d7e268 MK |
1126 | #define XUIOSTAT_INCR(stat, val) \ |
1127 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
1128 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
59e6e7ca | 1129 | |
5a6765cf | 1130 | #ifdef HAVE_UIO_ZEROCOPY |
428870ff BB |
1131 | int |
1132 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
1133 | { | |
1134 | dmu_xuio_t *priv; | |
1135 | uio_t *uio = &xuio->xu_uio; | |
1136 | ||
1137 | uio->uio_iovcnt = nblk; | |
79c76d5b | 1138 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
428870ff | 1139 | |
79c76d5b | 1140 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
428870ff | 1141 | priv->cnt = nblk; |
79c76d5b | 1142 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
5475aada | 1143 | priv->iovp = (iovec_t *)uio->uio_iov; |
428870ff BB |
1144 | XUIO_XUZC_PRIV(xuio) = priv; |
1145 | ||
1146 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1147 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
1148 | else | |
1149 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
1150 | ||
1151 | return (0); | |
1152 | } | |
1153 | ||
1154 | void | |
1155 | dmu_xuio_fini(xuio_t *xuio) | |
1156 | { | |
1157 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1158 | int nblk = priv->cnt; | |
1159 | ||
1160 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
1161 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
1162 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
1163 | ||
1164 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1165 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
1166 | else | |
1167 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
1168 | } | |
1169 | ||
1170 | /* | |
1171 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
1172 | * and increase priv->next by 1. | |
1173 | */ | |
1174 | int | |
1175 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
1176 | { | |
1177 | struct iovec *iov; | |
1178 | uio_t *uio = &xuio->xu_uio; | |
1179 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1180 | int i = priv->next++; | |
1181 | ||
1182 | ASSERT(i < priv->cnt); | |
2aa34383 | 1183 | ASSERT(off + n <= arc_buf_lsize(abuf)); |
5475aada | 1184 | iov = (iovec_t *)uio->uio_iov + i; |
428870ff BB |
1185 | iov->iov_base = (char *)abuf->b_data + off; |
1186 | iov->iov_len = n; | |
1187 | priv->bufs[i] = abuf; | |
1188 | return (0); | |
1189 | } | |
1190 | ||
1191 | int | |
1192 | dmu_xuio_cnt(xuio_t *xuio) | |
1193 | { | |
1194 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1195 | return (priv->cnt); | |
1196 | } | |
1197 | ||
1198 | arc_buf_t * | |
1199 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
1200 | { | |
1201 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1202 | ||
1203 | ASSERT(i < priv->cnt); | |
1204 | return (priv->bufs[i]); | |
1205 | } | |
1206 | ||
1207 | void | |
1208 | dmu_xuio_clear(xuio_t *xuio, int i) | |
1209 | { | |
1210 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1211 | ||
1212 | ASSERT(i < priv->cnt); | |
1213 | priv->bufs[i] = NULL; | |
1214 | } | |
5a6765cf | 1215 | #endif /* HAVE_UIO_ZEROCOPY */ |
428870ff BB |
1216 | |
1217 | static void | |
1218 | xuio_stat_init(void) | |
1219 | { | |
1220 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1221 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1222 | KSTAT_FLAG_VIRTUAL); | |
1223 | if (xuio_ksp != NULL) { | |
1224 | xuio_ksp->ks_data = &xuio_stats; | |
1225 | kstat_install(xuio_ksp); | |
1226 | } | |
1227 | } | |
1228 | ||
1229 | static void | |
1230 | xuio_stat_fini(void) | |
1231 | { | |
1232 | if (xuio_ksp != NULL) { | |
1233 | kstat_delete(xuio_ksp); | |
1234 | xuio_ksp = NULL; | |
1235 | } | |
1236 | } | |
1237 | ||
1238 | void | |
5043684a | 1239 | xuio_stat_wbuf_copied(void) |
428870ff BB |
1240 | { |
1241 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1242 | } | |
1243 | ||
1244 | void | |
5043684a | 1245 | xuio_stat_wbuf_nocopy(void) |
428870ff BB |
1246 | { |
1247 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1248 | } | |
1249 | ||
34dc7c2f | 1250 | #ifdef _KERNEL |
5228cf01 | 1251 | int |
804e0504 | 1252 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) |
872e8d26 BB |
1253 | { |
1254 | dmu_buf_t **dbp; | |
1255 | int numbufs, i, err; | |
5a6765cf | 1256 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 | 1257 | xuio_t *xuio = NULL; |
5a6765cf | 1258 | #endif |
872e8d26 BB |
1259 | |
1260 | /* | |
1261 | * NB: we could do this block-at-a-time, but it's nice | |
1262 | * to be reading in parallel. | |
1263 | */ | |
804e0504 MA |
1264 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, |
1265 | TRUE, FTAG, &numbufs, &dbp, 0); | |
872e8d26 BB |
1266 | if (err) |
1267 | return (err); | |
1268 | ||
1269 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1270 | uint64_t tocpy; |
1271 | int64_t bufoff; | |
872e8d26 BB |
1272 | dmu_buf_t *db = dbp[i]; |
1273 | ||
1274 | ASSERT(size > 0); | |
1275 | ||
1276 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1277 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1278 | |
5a6765cf | 1279 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 BB |
1280 | if (xuio) { |
1281 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1282 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1283 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1284 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1285 | if (!err) { | |
1286 | uio->uio_resid -= tocpy; | |
1287 | uio->uio_loffset += tocpy; | |
1288 | } | |
1289 | ||
1290 | if (abuf == dbuf_abuf) | |
1291 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1292 | else | |
1293 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
5a6765cf | 1294 | } else |
1295 | #endif | |
872e8d26 BB |
1296 | err = uiomove((char *)db->db_data + bufoff, tocpy, |
1297 | UIO_READ, uio); | |
872e8d26 BB |
1298 | if (err) |
1299 | break; | |
1300 | ||
1301 | size -= tocpy; | |
1302 | } | |
1303 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1304 | ||
1305 | return (err); | |
1306 | } | |
1307 | ||
804e0504 MA |
1308 | /* |
1309 | * Read 'size' bytes into the uio buffer. | |
1310 | * From object zdb->db_object. | |
1311 | * Starting at offset uio->uio_loffset. | |
1312 | * | |
1313 | * If the caller already has a dbuf in the target object | |
1314 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1315 | * because we don't have to find the dnode_t for the object. | |
1316 | */ | |
1317 | int | |
1318 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1319 | { | |
1320 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1321 | dnode_t *dn; | |
1322 | int err; | |
1323 | ||
1324 | if (size == 0) | |
1325 | return (0); | |
1326 | ||
1327 | DB_DNODE_ENTER(db); | |
1328 | dn = DB_DNODE(db); | |
1329 | err = dmu_read_uio_dnode(dn, uio, size); | |
1330 | DB_DNODE_EXIT(db); | |
1331 | ||
1332 | return (err); | |
1333 | } | |
1334 | ||
1335 | /* | |
1336 | * Read 'size' bytes into the uio buffer. | |
1337 | * From the specified object | |
1338 | * Starting at offset uio->uio_loffset. | |
1339 | */ | |
1340 | int | |
1341 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1342 | { | |
1343 | dnode_t *dn; | |
1344 | int err; | |
1345 | ||
1346 | if (size == 0) | |
1347 | return (0); | |
1348 | ||
1349 | err = dnode_hold(os, object, FTAG, &dn); | |
1350 | if (err) | |
1351 | return (err); | |
1352 | ||
1353 | err = dmu_read_uio_dnode(dn, uio, size); | |
1354 | ||
1355 | dnode_rele(dn, FTAG); | |
1356 | ||
1357 | return (err); | |
1358 | } | |
1359 | ||
5228cf01 | 1360 | int |
872e8d26 BB |
1361 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) |
1362 | { | |
1363 | dmu_buf_t **dbp; | |
1364 | int numbufs; | |
1365 | int err = 0; | |
1366 | int i; | |
1367 | ||
1368 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1369 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1370 | if (err) | |
1371 | return (err); | |
1372 | ||
1373 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1374 | uint64_t tocpy; |
1375 | int64_t bufoff; | |
872e8d26 BB |
1376 | dmu_buf_t *db = dbp[i]; |
1377 | ||
1378 | ASSERT(size > 0); | |
1379 | ||
1380 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1381 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1382 | |
1383 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1384 | ||
1385 | if (tocpy == db->db_size) | |
1386 | dmu_buf_will_fill(db, tx); | |
1387 | else | |
1388 | dmu_buf_will_dirty(db, tx); | |
1389 | ||
1390 | /* | |
1391 | * XXX uiomove could block forever (eg.nfs-backed | |
1392 | * pages). There needs to be a uiolockdown() function | |
1393 | * to lock the pages in memory, so that uiomove won't | |
1394 | * block. | |
1395 | */ | |
1396 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1397 | UIO_WRITE, uio); | |
1398 | ||
1399 | if (tocpy == db->db_size) | |
1400 | dmu_buf_fill_done(db, tx); | |
1401 | ||
1402 | if (err) | |
1403 | break; | |
1404 | ||
1405 | size -= tocpy; | |
1406 | } | |
1407 | ||
1408 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1409 | return (err); | |
1410 | } | |
1411 | ||
804e0504 MA |
1412 | /* |
1413 | * Write 'size' bytes from the uio buffer. | |
1414 | * To object zdb->db_object. | |
1415 | * Starting at offset uio->uio_loffset. | |
1416 | * | |
1417 | * If the caller already has a dbuf in the target object | |
1418 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1419 | * because we don't have to find the dnode_t for the object. | |
1420 | */ | |
428870ff BB |
1421 | int |
1422 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1423 | dmu_tx_t *tx) | |
1424 | { | |
572e2857 BB |
1425 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1426 | dnode_t *dn; | |
1427 | int err; | |
1428 | ||
428870ff BB |
1429 | if (size == 0) |
1430 | return (0); | |
1431 | ||
572e2857 BB |
1432 | DB_DNODE_ENTER(db); |
1433 | dn = DB_DNODE(db); | |
1434 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1435 | DB_DNODE_EXIT(db); | |
1436 | ||
1437 | return (err); | |
428870ff BB |
1438 | } |
1439 | ||
804e0504 MA |
1440 | /* |
1441 | * Write 'size' bytes from the uio buffer. | |
1442 | * To the specified object. | |
1443 | * Starting at offset uio->uio_loffset. | |
1444 | */ | |
428870ff BB |
1445 | int |
1446 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1447 | dmu_tx_t *tx) | |
1448 | { | |
1449 | dnode_t *dn; | |
1450 | int err; | |
1451 | ||
1452 | if (size == 0) | |
1453 | return (0); | |
1454 | ||
1455 | err = dnode_hold(os, object, FTAG, &dn); | |
1456 | if (err) | |
1457 | return (err); | |
1458 | ||
1459 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1460 | ||
1461 | dnode_rele(dn, FTAG); | |
1462 | ||
1463 | return (err); | |
1464 | } | |
872e8d26 | 1465 | #endif /* _KERNEL */ |
34dc7c2f | 1466 | |
9babb374 BB |
1467 | /* |
1468 | * Allocate a loaned anonymous arc buffer. | |
1469 | */ | |
1470 | arc_buf_t * | |
1471 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1472 | { | |
572e2857 | 1473 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1474 | |
2aa34383 | 1475 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1476 | } |
1477 | ||
1478 | /* | |
1479 | * Free a loaned arc buffer. | |
1480 | */ | |
1481 | void | |
1482 | dmu_return_arcbuf(arc_buf_t *buf) | |
1483 | { | |
1484 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1485 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1486 | } |
1487 | ||
b5256303 TC |
1488 | void |
1489 | dmu_assign_arcbuf_impl(dmu_buf_t *handle, arc_buf_t *buf, dmu_tx_t *tx) | |
1490 | { | |
1491 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; | |
1492 | dbuf_assign_arcbuf(db, buf, tx); | |
1493 | } | |
1494 | ||
1495 | void | |
1496 | dmu_convert_to_raw(dmu_buf_t *handle, boolean_t byteorder, const uint8_t *salt, | |
1497 | const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx) | |
1498 | { | |
1499 | dmu_object_type_t type; | |
1500 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; | |
1501 | uint64_t dsobj = dmu_objset_id(db->db_objset); | |
1502 | ||
1503 | ASSERT3P(db->db_buf, !=, NULL); | |
1504 | ASSERT3U(dsobj, !=, 0); | |
1505 | ||
1506 | dmu_buf_will_change_crypt_params(handle, tx); | |
1507 | ||
1508 | DB_DNODE_ENTER(db); | |
1509 | type = DB_DNODE(db)->dn_type; | |
1510 | DB_DNODE_EXIT(db); | |
1511 | ||
1512 | /* | |
1513 | * This technically violates the assumption the dmu code makes | |
1514 | * that dnode blocks are only released in syncing context. | |
1515 | */ | |
1516 | (void) arc_release(db->db_buf, db); | |
1517 | arc_convert_to_raw(db->db_buf, dsobj, byteorder, type, salt, iv, mac); | |
1518 | } | |
1519 | ||
1520 | void | |
1521 | dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset, | |
1522 | dmu_buf_t *handle, dmu_tx_t *tx) | |
1523 | { | |
1524 | dmu_buf_t *dst_handle; | |
1525 | dmu_buf_impl_t *dstdb; | |
1526 | dmu_buf_impl_t *srcdb = (dmu_buf_impl_t *)handle; | |
1527 | arc_buf_t *abuf; | |
1528 | uint64_t datalen; | |
1529 | boolean_t byteorder; | |
1530 | uint8_t salt[ZIO_DATA_SALT_LEN]; | |
1531 | uint8_t iv[ZIO_DATA_IV_LEN]; | |
1532 | uint8_t mac[ZIO_DATA_MAC_LEN]; | |
1533 | ||
1534 | ASSERT3P(srcdb->db_buf, !=, NULL); | |
1535 | ||
1536 | /* hold the db that we want to write to */ | |
1537 | VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &dst_handle, | |
1538 | DMU_READ_NO_DECRYPT)); | |
1539 | dstdb = (dmu_buf_impl_t *)dst_handle; | |
1540 | datalen = arc_buf_size(srcdb->db_buf); | |
1541 | ||
1542 | /* allocated an arc buffer that matches the type of srcdb->db_buf */ | |
1543 | if (arc_is_encrypted(srcdb->db_buf)) { | |
1544 | arc_get_raw_params(srcdb->db_buf, &byteorder, salt, iv, mac); | |
1545 | abuf = arc_loan_raw_buf(os->os_spa, dmu_objset_id(os), | |
1546 | byteorder, salt, iv, mac, DB_DNODE(dstdb)->dn_type, | |
1547 | datalen, arc_buf_lsize(srcdb->db_buf), | |
1548 | arc_get_compression(srcdb->db_buf)); | |
1549 | } else { | |
1550 | /* we won't get a compressed db back from dmu_buf_hold() */ | |
1551 | ASSERT3U(arc_get_compression(srcdb->db_buf), | |
1552 | ==, ZIO_COMPRESS_OFF); | |
1553 | abuf = arc_loan_buf(os->os_spa, | |
1554 | DMU_OT_IS_METADATA(DB_DNODE(dstdb)->dn_type), datalen); | |
1555 | } | |
1556 | ||
1557 | ASSERT3U(datalen, ==, arc_buf_size(abuf)); | |
1558 | ||
1559 | /* copy the data to the new buffer and assign it to the dstdb */ | |
1560 | bcopy(srcdb->db_buf->b_data, abuf->b_data, datalen); | |
1561 | dbuf_assign_arcbuf(dstdb, abuf, tx); | |
1562 | dmu_buf_rele(dst_handle, FTAG); | |
1563 | } | |
1564 | ||
9babb374 BB |
1565 | /* |
1566 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1567 | * If this is not possible copy the contents of passed arc buf via | |
1568 | * dmu_write(). | |
1569 | */ | |
1570 | void | |
1571 | dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, | |
1572 | dmu_tx_t *tx) | |
1573 | { | |
572e2857 BB |
1574 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1575 | dnode_t *dn; | |
9babb374 | 1576 | dmu_buf_impl_t *db; |
2aa34383 | 1577 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1578 | uint64_t blkid; |
1579 | ||
572e2857 BB |
1580 | DB_DNODE_ENTER(dbuf); |
1581 | dn = DB_DNODE(dbuf); | |
9babb374 | 1582 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
fcff0f35 | 1583 | blkid = dbuf_whichblock(dn, 0, offset); |
9babb374 BB |
1584 | VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); |
1585 | rw_exit(&dn->dn_struct_rwlock); | |
572e2857 | 1586 | DB_DNODE_EXIT(dbuf); |
9babb374 | 1587 | |
88904bb3 MA |
1588 | /* |
1589 | * We can only assign if the offset is aligned, the arc buf is the | |
2aa34383 | 1590 | * same size as the dbuf, and the dbuf is not metadata. |
88904bb3 | 1591 | */ |
2aa34383 | 1592 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
9babb374 BB |
1593 | dbuf_assign_arcbuf(db, buf, tx); |
1594 | dbuf_rele(db, FTAG); | |
1595 | } else { | |
572e2857 BB |
1596 | objset_t *os; |
1597 | uint64_t object; | |
1598 | ||
2aa34383 DK |
1599 | /* compressed bufs must always be assignable to their dbuf */ |
1600 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1601 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1602 | |
572e2857 BB |
1603 | DB_DNODE_ENTER(dbuf); |
1604 | dn = DB_DNODE(dbuf); | |
1605 | os = dn->dn_objset; | |
1606 | object = dn->dn_object; | |
1607 | DB_DNODE_EXIT(dbuf); | |
1608 | ||
9babb374 | 1609 | dbuf_rele(db, FTAG); |
572e2857 | 1610 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1611 | dmu_return_arcbuf(buf); |
428870ff | 1612 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 BB |
1613 | } |
1614 | } | |
1615 | ||
34dc7c2f | 1616 | typedef struct { |
428870ff BB |
1617 | dbuf_dirty_record_t *dsa_dr; |
1618 | dmu_sync_cb_t *dsa_done; | |
1619 | zgd_t *dsa_zgd; | |
1620 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1621 | } dmu_sync_arg_t; |
1622 | ||
b128c09f BB |
1623 | /* ARGSUSED */ |
1624 | static void | |
1625 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1626 | { | |
428870ff BB |
1627 | dmu_sync_arg_t *dsa = varg; |
1628 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1629 | blkptr_t *bp = zio->io_bp; |
1630 | ||
428870ff BB |
1631 | if (zio->io_error == 0) { |
1632 | if (BP_IS_HOLE(bp)) { | |
1633 | /* | |
1634 | * A block of zeros may compress to a hole, but the | |
1635 | * block size still needs to be known for replay. | |
1636 | */ | |
1637 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1638 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1639 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1640 | BP_SET_FILL(bp, 1); |
428870ff | 1641 | } |
b128c09f BB |
1642 | } |
1643 | } | |
1644 | ||
428870ff BB |
1645 | static void |
1646 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1647 | { | |
1648 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1649 | } | |
1650 | ||
34dc7c2f BB |
1651 | /* ARGSUSED */ |
1652 | static void | |
1653 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1654 | { | |
428870ff BB |
1655 | dmu_sync_arg_t *dsa = varg; |
1656 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1657 | dmu_buf_impl_t *db = dr->dr_dbuf; |
34dc7c2f | 1658 | |
34dc7c2f BB |
1659 | mutex_enter(&db->db_mtx); |
1660 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1661 | if (zio->io_error == 0) { |
03c6040b GW |
1662 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1663 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1664 | blkptr_t *bp = zio->io_bp; |
1665 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1666 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1667 | |
1668 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1669 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1670 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1671 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1672 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1673 | } |
428870ff BB |
1674 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1675 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1676 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1677 | |
1678 | /* | |
1679 | * Old style holes are filled with all zeros, whereas | |
1680 | * new-style holes maintain their lsize, type, level, | |
1681 | * and birth time (see zio_write_compress). While we | |
1682 | * need to reset the BP_SET_LSIZE() call that happened | |
1683 | * in dmu_sync_ready for old style holes, we do *not* | |
1684 | * want to wipe out the information contained in new | |
1685 | * style holes. Thus, only zero out the block pointer if | |
1686 | * it's an old style hole. | |
1687 | */ | |
1688 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1689 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1690 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1691 | } else { | |
1692 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1693 | } | |
34dc7c2f BB |
1694 | cv_broadcast(&db->db_changed); |
1695 | mutex_exit(&db->db_mtx); | |
1696 | ||
428870ff | 1697 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1698 | |
428870ff BB |
1699 | kmem_free(dsa, sizeof (*dsa)); |
1700 | } | |
1701 | ||
1702 | static void | |
1703 | dmu_sync_late_arrival_done(zio_t *zio) | |
1704 | { | |
1705 | blkptr_t *bp = zio->io_bp; | |
1706 | dmu_sync_arg_t *dsa = zio->io_private; | |
03c6040b | 1707 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); |
428870ff BB |
1708 | |
1709 | if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { | |
02dc43bc MA |
1710 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); |
1711 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1712 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1713 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1714 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
428870ff BB |
1715 | } |
1716 | ||
1717 | dmu_tx_commit(dsa->dsa_tx); | |
1718 | ||
1719 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1720 | ||
a6255b7f | 1721 | abd_put(zio->io_abd); |
428870ff BB |
1722 | kmem_free(dsa, sizeof (*dsa)); |
1723 | } | |
1724 | ||
1725 | static int | |
1726 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1727 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1728 | { |
1729 | dmu_sync_arg_t *dsa; | |
1730 | dmu_tx_t *tx; | |
1731 | ||
1732 | tx = dmu_tx_create(os); | |
1733 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1734 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1735 | dmu_tx_abort(tx); | |
2e528b49 MA |
1736 | /* Make zl_get_data do txg_waited_synced() */ |
1737 | return (SET_ERROR(EIO)); | |
428870ff BB |
1738 | } |
1739 | ||
79c76d5b | 1740 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1741 | dsa->dsa_dr = NULL; |
1742 | dsa->dsa_done = done; | |
1743 | dsa->dsa_zgd = zgd; | |
1744 | dsa->dsa_tx = tx; | |
1745 | ||
02dc43bc MA |
1746 | /* |
1747 | * Since we are currently syncing this txg, it's nontrivial to | |
1748 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1749 | * | |
1750 | * When syncing, the db_blkptr is initially the BP of the previous | |
1751 | * txg. We can not nopwrite against it because it will be changed | |
1752 | * (this is similar to the non-late-arrival case where the dbuf is | |
1753 | * dirty in a future txg). | |
1754 | * | |
1755 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1756 | * We can not nopwrite against it because although the BP will not | |
1757 | * (typically) be changed, the data has not yet been persisted to this | |
1758 | * location. | |
1759 | * | |
1760 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1761 | * possible to always nopwrite, because the data that was written in | |
1762 | * this txg is the same data that we are trying to write. However we | |
1763 | * would need to check that this dbuf is not dirty in any future | |
1764 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1765 | * don't nopwrite in this case. | |
1766 | */ | |
1767 | zp->zp_nopwrite = B_FALSE; | |
1768 | ||
a6255b7f DQ |
1769 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1770 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1771 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1772 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1773 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1774 | |
1775 | return (0); | |
34dc7c2f BB |
1776 | } |
1777 | ||
1778 | /* | |
1779 | * Intent log support: sync the block associated with db to disk. | |
1780 | * N.B. and XXX: the caller is responsible for making sure that the | |
1781 | * data isn't changing while dmu_sync() is writing it. | |
1782 | * | |
1783 | * Return values: | |
1784 | * | |
03c6040b | 1785 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1786 | * The caller should not log the write. |
1787 | * | |
1788 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1789 | * The caller should not log the write. | |
1790 | * | |
1791 | * EALREADY: this block is already in the process of being synced. | |
1792 | * The caller should track its progress (somehow). | |
1793 | * | |
428870ff BB |
1794 | * EIO: could not do the I/O. |
1795 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1796 | * |
428870ff BB |
1797 | * 0: the I/O has been initiated. |
1798 | * The caller should log this blkptr in the done callback. | |
1799 | * It is possible that the I/O will fail, in which case | |
1800 | * the error will be reported to the done callback and | |
1801 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1802 | */ |
1803 | int | |
428870ff | 1804 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1805 | { |
428870ff BB |
1806 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1807 | objset_t *os = db->db_objset; | |
1808 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1809 | dbuf_dirty_record_t *dr; |
428870ff | 1810 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1811 | zbookmark_phys_t zb; |
428870ff | 1812 | zio_prop_t zp; |
572e2857 | 1813 | dnode_t *dn; |
34dc7c2f | 1814 | |
428870ff | 1815 | ASSERT(pio != NULL); |
34dc7c2f BB |
1816 | ASSERT(txg != 0); |
1817 | ||
428870ff BB |
1818 | SET_BOOKMARK(&zb, ds->ds_object, |
1819 | db->db.db_object, db->db_level, db->db_blkid); | |
1820 | ||
572e2857 BB |
1821 | DB_DNODE_ENTER(db); |
1822 | dn = DB_DNODE(db); | |
82644107 | 1823 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1824 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1825 | |
1826 | /* | |
428870ff | 1827 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1828 | */ |
428870ff BB |
1829 | if (txg > spa_freeze_txg(os->os_spa)) |
1830 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1831 | |
1832 | /* | |
428870ff BB |
1833 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1834 | * and us. If we determine that this txg is not yet syncing, | |
1835 | * but it begins to sync a moment later, that's OK because the | |
1836 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1837 | */ |
428870ff BB |
1838 | mutex_enter(&db->db_mtx); |
1839 | ||
1840 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1841 | /* |
428870ff | 1842 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1843 | */ |
428870ff | 1844 | mutex_exit(&db->db_mtx); |
2e528b49 | 1845 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1846 | } |
1847 | ||
428870ff BB |
1848 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1849 | /* | |
1850 | * This txg is currently syncing, so we can't mess with | |
1851 | * the dirty record anymore; just write a new log block. | |
1852 | */ | |
1853 | mutex_exit(&db->db_mtx); | |
1854 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1855 | } |
1856 | ||
1857 | dr = db->db_last_dirty; | |
428870ff | 1858 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 1859 | dr = dr->dr_next; |
428870ff BB |
1860 | |
1861 | if (dr == NULL) { | |
34dc7c2f | 1862 | /* |
428870ff | 1863 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1864 | * There's no need to log writes to freed blocks, so we're done. |
1865 | */ | |
1866 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1867 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1868 | } |
1869 | ||
03c6040b GW |
1870 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); |
1871 | ||
02dc43bc MA |
1872 | if (db->db_blkptr != NULL) { |
1873 | /* | |
1874 | * We need to fill in zgd_bp with the current blkptr so that | |
1875 | * the nopwrite code can check if we're writing the same | |
1876 | * data that's already on disk. We can only nopwrite if we | |
1877 | * are sure that after making the copy, db_blkptr will not | |
1878 | * change until our i/o completes. We ensure this by | |
1879 | * holding the db_mtx, and only allowing nopwrite if the | |
1880 | * block is not already dirty (see below). This is verified | |
1881 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1882 | * not changed. | |
1883 | */ | |
1884 | *zgd->zgd_bp = *db->db_blkptr; | |
1885 | } | |
1886 | ||
03c6040b | 1887 | /* |
f3c517d8 MA |
1888 | * Assume the on-disk data is X, the current syncing data (in |
1889 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1890 | * in dmu_sync). | |
1891 | * | |
1892 | * We usually want to perform a nopwrite if X and Z are the | |
1893 | * same. However, if Y is different (i.e. the BP is going to | |
1894 | * change before this write takes effect), then a nopwrite will | |
1895 | * be incorrect - we would override with X, which could have | |
1896 | * been freed when Y was written. | |
1897 | * | |
1898 | * (Note that this is not a concern when we are nop-writing from | |
1899 | * syncing context, because X and Y must be identical, because | |
1900 | * all previous txgs have been synced.) | |
1901 | * | |
1902 | * Therefore, we disable nopwrite if the current BP could change | |
1903 | * before this TXG. There are two ways it could change: by | |
1904 | * being dirty (dr_next is non-NULL), or by being freed | |
1905 | * (dnode_block_freed()). This behavior is verified by | |
1906 | * zio_done(), which VERIFYs that the override BP is identical | |
1907 | * to the on-disk BP. | |
03c6040b | 1908 | */ |
f3c517d8 MA |
1909 | DB_DNODE_ENTER(db); |
1910 | dn = DB_DNODE(db); | |
1911 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
03c6040b | 1912 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1913 | DB_DNODE_EXIT(db); |
03c6040b | 1914 | |
34dc7c2f | 1915 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1916 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1917 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1918 | /* |
428870ff BB |
1919 | * We have already issued a sync write for this buffer, |
1920 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1921 | * have been dirtied since, or we would have cleared the state. |
1922 | */ | |
34dc7c2f | 1923 | mutex_exit(&db->db_mtx); |
2e528b49 | 1924 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1925 | } |
1926 | ||
428870ff | 1927 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1928 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1929 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1930 | |
79c76d5b | 1931 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1932 | dsa->dsa_dr = dr; |
1933 | dsa->dsa_done = done; | |
1934 | dsa->dsa_zgd = zgd; | |
1935 | dsa->dsa_tx = NULL; | |
b128c09f | 1936 | |
428870ff | 1937 | zio_nowait(arc_write(pio, os->os_spa, txg, |
02dc43bc | 1938 | zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 1939 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 1940 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 1941 | |
428870ff | 1942 | return (0); |
34dc7c2f BB |
1943 | } |
1944 | ||
b5256303 TC |
1945 | int |
1946 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
1947 | { | |
1948 | dnode_t *dn; | |
1949 | int err; | |
1950 | ||
1951 | err = dnode_hold(os, object, FTAG, &dn); | |
1952 | if (err) | |
1953 | return (err); | |
1954 | err = dnode_set_nlevels(dn, nlevels, tx); | |
1955 | dnode_rele(dn, FTAG); | |
1956 | return (err); | |
1957 | } | |
1958 | ||
34dc7c2f BB |
1959 | int |
1960 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 1961 | dmu_tx_t *tx) |
34dc7c2f BB |
1962 | { |
1963 | dnode_t *dn; | |
1964 | int err; | |
1965 | ||
428870ff | 1966 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1967 | if (err) |
1968 | return (err); | |
1969 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1970 | dnode_rele(dn, FTAG); | |
1971 | return (err); | |
1972 | } | |
1973 | ||
1974 | void | |
1975 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 1976 | dmu_tx_t *tx) |
34dc7c2f BB |
1977 | { |
1978 | dnode_t *dn; | |
1979 | ||
9b67f605 MA |
1980 | /* |
1981 | * Send streams include each object's checksum function. This | |
1982 | * check ensures that the receiving system can understand the | |
1983 | * checksum function transmitted. | |
1984 | */ | |
1985 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1986 | ||
1987 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1988 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1989 | dn->dn_checksum = checksum; |
1990 | dnode_setdirty(dn, tx); | |
1991 | dnode_rele(dn, FTAG); | |
1992 | } | |
1993 | ||
1994 | void | |
1995 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 1996 | dmu_tx_t *tx) |
34dc7c2f BB |
1997 | { |
1998 | dnode_t *dn; | |
1999 | ||
9b67f605 MA |
2000 | /* |
2001 | * Send streams include each object's compression function. This | |
2002 | * check ensures that the receiving system can understand the | |
2003 | * compression function transmitted. | |
2004 | */ | |
2005 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
2006 | ||
2007 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
2008 | dn->dn_compress = compress; |
2009 | dnode_setdirty(dn, tx); | |
2010 | dnode_rele(dn, FTAG); | |
2011 | } | |
2012 | ||
428870ff BB |
2013 | int zfs_mdcomp_disable = 0; |
2014 | ||
faf0f58c MA |
2015 | /* |
2016 | * When the "redundant_metadata" property is set to "most", only indirect | |
2017 | * blocks of this level and higher will have an additional ditto block. | |
2018 | */ | |
2019 | int zfs_redundant_metadata_most_ditto_level = 2; | |
2020 | ||
428870ff | 2021 | void |
82644107 | 2022 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
2023 | { |
2024 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 2025 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 2026 | (wp & WP_SPILL)); |
428870ff BB |
2027 | enum zio_checksum checksum = os->os_checksum; |
2028 | enum zio_compress compress = os->os_compress; | |
2029 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
03c6040b GW |
2030 | boolean_t dedup = B_FALSE; |
2031 | boolean_t nopwrite = B_FALSE; | |
428870ff | 2032 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 2033 | boolean_t encrypt = B_FALSE; |
428870ff | 2034 | int copies = os->os_copies; |
a7004725 | 2035 | |
428870ff | 2036 | /* |
03c6040b GW |
2037 | * We maintain different write policies for each of the following |
2038 | * types of data: | |
2039 | * 1. metadata | |
2040 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
2041 | * 3. all other level 0 blocks | |
428870ff BB |
2042 | */ |
2043 | if (ismd) { | |
62bdd5eb DL |
2044 | if (zfs_mdcomp_disable) { |
2045 | compress = ZIO_COMPRESS_EMPTY; | |
62bdd5eb | 2046 | } else { |
99197f03 JG |
2047 | /* |
2048 | * XXX -- we should design a compression algorithm | |
2049 | * that specializes in arrays of bps. | |
2050 | */ | |
2051 | compress = zio_compress_select(os->os_spa, | |
2052 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
62bdd5eb | 2053 | } |
03c6040b | 2054 | |
428870ff BB |
2055 | /* |
2056 | * Metadata always gets checksummed. If the data | |
2057 | * checksum is multi-bit correctable, and it's not a | |
2058 | * ZBT-style checksum, then it's suitable for metadata | |
2059 | * as well. Otherwise, the metadata checksum defaults | |
2060 | * to fletcher4. | |
2061 | */ | |
3c67d83a TH |
2062 | if (!(zio_checksum_table[checksum].ci_flags & |
2063 | ZCHECKSUM_FLAG_METADATA) || | |
2064 | (zio_checksum_table[checksum].ci_flags & | |
2065 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 2066 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
2067 | |
2068 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
2069 | (os->os_redundant_metadata == | |
2070 | ZFS_REDUNDANT_METADATA_MOST && | |
2071 | (level >= zfs_redundant_metadata_most_ditto_level || | |
2072 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
2073 | copies++; | |
03c6040b GW |
2074 | } else if (wp & WP_NOFILL) { |
2075 | ASSERT(level == 0); | |
428870ff | 2076 | |
428870ff | 2077 | /* |
03c6040b GW |
2078 | * If we're writing preallocated blocks, we aren't actually |
2079 | * writing them so don't set any policy properties. These | |
2080 | * blocks are currently only used by an external subsystem | |
2081 | * outside of zfs (i.e. dump) and not written by the zio | |
2082 | * pipeline. | |
428870ff | 2083 | */ |
03c6040b GW |
2084 | compress = ZIO_COMPRESS_OFF; |
2085 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2086 | } else { |
99197f03 JG |
2087 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2088 | compress); | |
428870ff | 2089 | |
03c6040b GW |
2090 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2091 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2092 | dedup_checksum; | |
428870ff | 2093 | |
03c6040b GW |
2094 | /* |
2095 | * Determine dedup setting. If we are in dmu_sync(), | |
2096 | * we won't actually dedup now because that's all | |
2097 | * done in syncing context; but we do want to use the | |
2098 | * dedup checkum. If the checksum is not strong | |
2099 | * enough to ensure unique signatures, force | |
2100 | * dedup_verify. | |
2101 | */ | |
2102 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2103 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2104 | if (!(zio_checksum_table[checksum].ci_flags & |
2105 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2106 | dedup_verify = B_TRUE; |
2107 | } | |
428870ff | 2108 | |
03c6040b | 2109 | /* |
3c67d83a TH |
2110 | * Enable nopwrite if we have secure enough checksum |
2111 | * algorithm (see comment in zio_nop_write) and | |
2112 | * compression is enabled. We don't enable nopwrite if | |
2113 | * dedup is enabled as the two features are mutually | |
2114 | * exclusive. | |
03c6040b | 2115 | */ |
3c67d83a TH |
2116 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2117 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2118 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2119 | } |
2120 | ||
b5256303 TC |
2121 | /* |
2122 | * All objects in an encrypted objset are protected from modification | |
2123 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2124 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2125 | * not subject to nopwrite since writing the same data will still | |
2126 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2127 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2128 | * object types. | |
2129 | */ | |
2130 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2131 | encrypt = B_TRUE; | |
2132 | ||
2133 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2134 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2135 | nopwrite = B_FALSE; | |
2136 | } else { | |
2137 | dedup = B_FALSE; | |
2138 | } | |
2139 | ||
2140 | if (type == DMU_OT_DNODE || type == DMU_OT_OBJSET) | |
2141 | compress = ZIO_COMPRESS_EMPTY; | |
2142 | } | |
2aa34383 | 2143 | |
b5256303 TC |
2144 | zp->zp_compress = compress; |
2145 | zp->zp_checksum = checksum; | |
428870ff BB |
2146 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2147 | zp->zp_level = level; | |
faf0f58c | 2148 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2149 | zp->zp_dedup = dedup; |
2150 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2151 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2152 | zp->zp_encrypt = encrypt; |
2153 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
2154 | bzero(zp->zp_salt, ZIO_DATA_SALT_LEN); | |
2155 | bzero(zp->zp_iv, ZIO_DATA_IV_LEN); | |
2156 | bzero(zp->zp_mac, ZIO_DATA_MAC_LEN); | |
2157 | ||
2158 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2159 | } |
2160 | ||
66aca247 DB |
2161 | /* |
2162 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2163 | * in order to determine the location of holes. In order to accurately | |
2164 | * report holes all dirty data must be synced to disk. This causes extremely | |
2165 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2166 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2167 | * report the dnode as having no holes which is always a safe thing to do. | |
2168 | */ | |
34dc7c2f BB |
2169 | int |
2170 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2171 | { | |
2172 | dnode_t *dn; | |
2173 | int i, err; | |
66aca247 | 2174 | boolean_t clean = B_TRUE; |
34dc7c2f | 2175 | |
428870ff | 2176 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2177 | if (err) |
2178 | return (err); | |
66aca247 | 2179 | |
34dc7c2f | 2180 | /* |
66aca247 | 2181 | * Check if dnode is dirty |
34dc7c2f | 2182 | */ |
66aca247 DB |
2183 | if (dn->dn_dirtyctx != DN_UNDIRTIED) { |
2184 | for (i = 0; i < TXG_SIZE; i++) { | |
2185 | if (!list_is_empty(&dn->dn_dirty_records[i])) { | |
2186 | clean = B_FALSE; | |
2187 | break; | |
2188 | } | |
2189 | } | |
34dc7c2f | 2190 | } |
66aca247 DB |
2191 | |
2192 | /* | |
2193 | * If compatibility option is on, sync any current changes before | |
2194 | * we go trundling through the block pointers. | |
2195 | */ | |
2196 | if (!clean && zfs_dmu_offset_next_sync) { | |
2197 | clean = B_TRUE; | |
34dc7c2f BB |
2198 | dnode_rele(dn, FTAG); |
2199 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 2200 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2201 | if (err) |
2202 | return (err); | |
2203 | } | |
2204 | ||
66aca247 DB |
2205 | if (clean) |
2206 | err = dnode_next_offset(dn, | |
2207 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2208 | else | |
2209 | err = SET_ERROR(EBUSY); | |
2210 | ||
34dc7c2f BB |
2211 | dnode_rele(dn, FTAG); |
2212 | ||
2213 | return (err); | |
2214 | } | |
2215 | ||
2216 | void | |
e0b0ca98 | 2217 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2218 | { |
e0b0ca98 | 2219 | dnode_phys_t *dnp = dn->dn_phys; |
d6320ddb | 2220 | int i; |
428870ff | 2221 | |
34dc7c2f BB |
2222 | doi->doi_data_block_size = dn->dn_datablksz; |
2223 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2224 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2225 | doi->doi_type = dn->dn_type; |
2226 | doi->doi_bonus_type = dn->dn_bonustype; | |
2227 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2228 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2229 | doi->doi_indirection = dn->dn_nlevels; |
2230 | doi->doi_checksum = dn->dn_checksum; | |
2231 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2232 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2233 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2234 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2235 | doi->doi_fill_count = 0; |
d6320ddb | 2236 | for (i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2237 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2238 | } |
2239 | ||
2240 | void | |
2241 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2242 | { | |
2243 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2244 | mutex_enter(&dn->dn_mtx); | |
2245 | ||
2246 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2247 | |
2248 | mutex_exit(&dn->dn_mtx); | |
2249 | rw_exit(&dn->dn_struct_rwlock); | |
2250 | } | |
2251 | ||
2252 | /* | |
2253 | * Get information on a DMU object. | |
2254 | * If doi is NULL, just indicates whether the object exists. | |
2255 | */ | |
2256 | int | |
2257 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2258 | { | |
2259 | dnode_t *dn; | |
428870ff | 2260 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2261 | |
2262 | if (err) | |
2263 | return (err); | |
2264 | ||
2265 | if (doi != NULL) | |
2266 | dmu_object_info_from_dnode(dn, doi); | |
2267 | ||
2268 | dnode_rele(dn, FTAG); | |
2269 | return (0); | |
2270 | } | |
2271 | ||
2272 | /* | |
2273 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2274 | */ | |
2275 | void | |
572e2857 | 2276 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2277 | { |
572e2857 BB |
2278 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2279 | ||
2280 | DB_DNODE_ENTER(db); | |
2281 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2282 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2283 | } |
2284 | ||
2285 | /* | |
2286 | * Faster still when you only care about the size. | |
2287 | * This is specifically optimized for zfs_getattr(). | |
2288 | */ | |
2289 | void | |
572e2857 BB |
2290 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2291 | u_longlong_t *nblk512) | |
34dc7c2f | 2292 | { |
572e2857 BB |
2293 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2294 | dnode_t *dn; | |
2295 | ||
2296 | DB_DNODE_ENTER(db); | |
2297 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2298 | |
2299 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2300 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2301 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2302 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2303 | DB_DNODE_EXIT(db); | |
2304 | } | |
2305 | ||
2306 | void | |
2307 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2308 | { | |
2309 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2310 | dnode_t *dn; | |
2311 | ||
2312 | DB_DNODE_ENTER(db); | |
2313 | dn = DB_DNODE(db); | |
2314 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2315 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2316 | } |
2317 | ||
2318 | void | |
2319 | byteswap_uint64_array(void *vbuf, size_t size) | |
2320 | { | |
2321 | uint64_t *buf = vbuf; | |
2322 | size_t count = size >> 3; | |
2323 | int i; | |
2324 | ||
2325 | ASSERT((size & 7) == 0); | |
2326 | ||
2327 | for (i = 0; i < count; i++) | |
2328 | buf[i] = BSWAP_64(buf[i]); | |
2329 | } | |
2330 | ||
2331 | void | |
2332 | byteswap_uint32_array(void *vbuf, size_t size) | |
2333 | { | |
2334 | uint32_t *buf = vbuf; | |
2335 | size_t count = size >> 2; | |
2336 | int i; | |
2337 | ||
2338 | ASSERT((size & 3) == 0); | |
2339 | ||
2340 | for (i = 0; i < count; i++) | |
2341 | buf[i] = BSWAP_32(buf[i]); | |
2342 | } | |
2343 | ||
2344 | void | |
2345 | byteswap_uint16_array(void *vbuf, size_t size) | |
2346 | { | |
2347 | uint16_t *buf = vbuf; | |
2348 | size_t count = size >> 1; | |
2349 | int i; | |
2350 | ||
2351 | ASSERT((size & 1) == 0); | |
2352 | ||
2353 | for (i = 0; i < count; i++) | |
2354 | buf[i] = BSWAP_16(buf[i]); | |
2355 | } | |
2356 | ||
2357 | /* ARGSUSED */ | |
2358 | void | |
2359 | byteswap_uint8_array(void *vbuf, size_t size) | |
2360 | { | |
2361 | } | |
2362 | ||
2363 | void | |
2364 | dmu_init(void) | |
2365 | { | |
a6255b7f | 2366 | abd_init(); |
428870ff | 2367 | zfs_dbgmsg_init(); |
572e2857 BB |
2368 | sa_cache_init(); |
2369 | xuio_stat_init(); | |
2370 | dmu_objset_init(); | |
34dc7c2f | 2371 | dnode_init(); |
428870ff | 2372 | zfetch_init(); |
570827e1 | 2373 | dmu_tx_init(); |
34dc7c2f | 2374 | l2arc_init(); |
29809a6c | 2375 | arc_init(); |
d3c2ae1c | 2376 | dbuf_init(); |
34dc7c2f BB |
2377 | } |
2378 | ||
2379 | void | |
2380 | dmu_fini(void) | |
2381 | { | |
e49f1e20 | 2382 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2383 | l2arc_fini(); |
570827e1 | 2384 | dmu_tx_fini(); |
428870ff | 2385 | zfetch_fini(); |
34dc7c2f | 2386 | dbuf_fini(); |
572e2857 BB |
2387 | dnode_fini(); |
2388 | dmu_objset_fini(); | |
428870ff BB |
2389 | xuio_stat_fini(); |
2390 | sa_cache_fini(); | |
2391 | zfs_dbgmsg_fini(); | |
a6255b7f | 2392 | abd_fini(); |
34dc7c2f | 2393 | } |
c28b2279 BB |
2394 | |
2395 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
2396 | EXPORT_SYMBOL(dmu_bonus_hold); | |
a473d90c AZ |
2397 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2398 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2399 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2400 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2401 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2402 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2403 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2404 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2405 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2406 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2407 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2408 | EXPORT_SYMBOL(dmu_object_info); |
2409 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2410 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2411 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2412 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2413 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 BB |
2414 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
2415 | EXPORT_SYMBOL(dmu_object_set_checksum); | |
2416 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
2417 | EXPORT_SYMBOL(dmu_write_policy); |
2418 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2419 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2420 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
2421 | EXPORT_SYMBOL(dmu_assign_arcbuf); | |
2422 | EXPORT_SYMBOL(dmu_buf_hold); | |
c28b2279 | 2423 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2424 | |
bef78122 | 2425 | /* BEGIN CSTYLED */ |
afec56b4 BB |
2426 | module_param(zfs_mdcomp_disable, int, 0644); |
2427 | MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression"); | |
03c6040b GW |
2428 | |
2429 | module_param(zfs_nopwrite_enabled, int, 0644); | |
2430 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2431 | ||
bef78122 DQ |
2432 | module_param(zfs_per_txg_dirty_frees_percent, ulong, 0644); |
2433 | MODULE_PARM_DESC(zfs_per_txg_dirty_frees_percent, | |
2434 | "percentage of dirtied blocks from frees in one TXG"); | |
66aca247 DB |
2435 | |
2436 | module_param(zfs_dmu_offset_next_sync, int, 0644); | |
2437 | MODULE_PARM_DESC(zfs_dmu_offset_next_sync, | |
2438 | "Enable forcing txg sync to find holes"); | |
2439 | ||
bef78122 | 2440 | /* END CSTYLED */ |
66aca247 | 2441 | |
c28b2279 | 2442 | #endif |