]>
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. |
30af21b0 | 23 | * Copyright (c) 2011, 2018 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
bc77ba73 | 25 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
a08abc1b | 26 | * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved. |
5475aada | 27 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. |
65282ee9 | 28 | * Copyright (c) 2019 Datto Inc. |
34dc7c2f BB |
29 | */ |
30 | ||
34dc7c2f BB |
31 | #include <sys/dmu.h> |
32 | #include <sys/dmu_impl.h> | |
33 | #include <sys/dmu_tx.h> | |
34 | #include <sys/dbuf.h> | |
35 | #include <sys/dnode.h> | |
36 | #include <sys/zfs_context.h> | |
37 | #include <sys/dmu_objset.h> | |
38 | #include <sys/dmu_traverse.h> | |
39 | #include <sys/dsl_dataset.h> | |
40 | #include <sys/dsl_dir.h> | |
41 | #include <sys/dsl_pool.h> | |
42 | #include <sys/dsl_synctask.h> | |
43 | #include <sys/dsl_prop.h> | |
44 | #include <sys/dmu_zfetch.h> | |
45 | #include <sys/zfs_ioctl.h> | |
46 | #include <sys/zap.h> | |
47 | #include <sys/zio_checksum.h> | |
03c6040b | 48 | #include <sys/zio_compress.h> |
428870ff | 49 | #include <sys/sa.h> |
62bdd5eb | 50 | #include <sys/zfeature.h> |
a6255b7f | 51 | #include <sys/abd.h> |
539d33c7 | 52 | #include <sys/trace_dmu.h> |
f763c3d1 | 53 | #include <sys/zfs_rlock.h> |
34dc7c2f BB |
54 | #ifdef _KERNEL |
55 | #include <sys/vmsystm.h> | |
b128c09f | 56 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
57 | #endif |
58 | ||
03c6040b GW |
59 | /* |
60 | * Enable/disable nopwrite feature. | |
61 | */ | |
62 | int zfs_nopwrite_enabled = 1; | |
63 | ||
539d33c7 | 64 | /* |
65282ee9 AP |
65 | * Tunable to control percentage of dirtied L1 blocks from frees allowed into |
66 | * one TXG. After this threshold is crossed, additional dirty blocks from frees | |
67 | * will wait until the next TXG. | |
539d33c7 GM |
68 | * A value of zero will disable this throttle. |
69 | */ | |
65282ee9 | 70 | unsigned long zfs_per_txg_dirty_frees_percent = 5; |
539d33c7 | 71 | |
66aca247 DB |
72 | /* |
73 | * Enable/disable forcing txg sync when dirty in dmu_offset_next. | |
74 | */ | |
75 | int zfs_dmu_offset_next_sync = 0; | |
76 | ||
d9b4bf06 MA |
77 | /* |
78 | * Limit the amount we can prefetch with one call to this amount. This | |
79 | * helps to limit the amount of memory that can be used by prefetching. | |
80 | * Larger objects should be prefetched a bit at a time. | |
81 | */ | |
82 | int dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE; | |
83 | ||
34dc7c2f | 84 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
2e5dc449 MA |
85 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" }, |
86 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" }, | |
87 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" }, | |
88 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" }, | |
89 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" }, | |
90 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" }, | |
91 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" }, | |
92 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" }, | |
93 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" }, | |
94 | {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" }, | |
95 | {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" }, | |
96 | {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" }, | |
97 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" }, | |
98 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"}, | |
99 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" }, | |
100 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" }, | |
101 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" }, | |
102 | {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" }, | |
103 | {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" }, | |
104 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" }, | |
105 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" }, | |
106 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" }, | |
107 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" }, | |
108 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" }, | |
109 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" }, | |
110 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" }, | |
111 | {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" }, | |
112 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" }, | |
113 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" }, | |
114 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" }, | |
115 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" }, | |
116 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" }, | |
117 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" }, | |
118 | {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" }, | |
119 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" }, | |
120 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" }, | |
121 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" }, | |
122 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"}, | |
123 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" }, | |
124 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" }, | |
125 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"}, | |
126 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"}, | |
127 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" }, | |
128 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" }, | |
129 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" }, | |
130 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" }, | |
131 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" }, | |
132 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" }, | |
133 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" }, | |
134 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" }, | |
135 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" }, | |
136 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" }, | |
137 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" }, | |
138 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
139 | }; |
140 | ||
141 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
142 | { byteswap_uint8_array, "uint8" }, | |
143 | { byteswap_uint16_array, "uint16" }, | |
144 | { byteswap_uint32_array, "uint32" }, | |
145 | { byteswap_uint64_array, "uint64" }, | |
146 | { zap_byteswap, "zap" }, | |
147 | { dnode_buf_byteswap, "dnode" }, | |
148 | { dmu_objset_byteswap, "objset" }, | |
149 | { zfs_znode_byteswap, "znode" }, | |
150 | { zfs_oldacl_byteswap, "oldacl" }, | |
151 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
152 | }; |
153 | ||
2bce8049 MA |
154 | int |
155 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, | |
156 | void *tag, dmu_buf_t **dbp) | |
157 | { | |
158 | uint64_t blkid; | |
159 | dmu_buf_impl_t *db; | |
160 | ||
2bce8049 | 161 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 162 | blkid = dbuf_whichblock(dn, 0, offset); |
2bce8049 MA |
163 | db = dbuf_hold(dn, blkid, tag); |
164 | rw_exit(&dn->dn_struct_rwlock); | |
165 | ||
166 | if (db == NULL) { | |
167 | *dbp = NULL; | |
168 | return (SET_ERROR(EIO)); | |
169 | } | |
170 | ||
171 | *dbp = &db->db; | |
172 | return (0); | |
173 | } | |
34dc7c2f | 174 | int |
9b67f605 MA |
175 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
176 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
177 | { |
178 | dnode_t *dn; | |
179 | uint64_t blkid; | |
180 | dmu_buf_impl_t *db; | |
181 | int err; | |
428870ff BB |
182 | |
183 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
184 | if (err) |
185 | return (err); | |
34dc7c2f | 186 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 187 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
188 | db = dbuf_hold(dn, blkid, tag); |
189 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
190 | dnode_rele(dn, FTAG); |
191 | ||
34dc7c2f | 192 | if (db == NULL) { |
9b67f605 MA |
193 | *dbp = NULL; |
194 | return (SET_ERROR(EIO)); | |
195 | } | |
196 | ||
197 | *dbp = &db->db; | |
198 | return (err); | |
199 | } | |
200 | ||
2bce8049 MA |
201 | int |
202 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
203 | void *tag, dmu_buf_t **dbp, int flags) | |
204 | { | |
205 | int err; | |
206 | int db_flags = DB_RF_CANFAIL; | |
207 | ||
208 | if (flags & DMU_READ_NO_PREFETCH) | |
209 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
210 | if (flags & DMU_READ_NO_DECRYPT) |
211 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
212 | |
213 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
214 | if (err == 0) { | |
215 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
216 | err = dbuf_read(db, NULL, db_flags); | |
217 | if (err != 0) { | |
218 | dbuf_rele(db, tag); | |
219 | *dbp = NULL; | |
220 | } | |
221 | } | |
222 | ||
223 | return (err); | |
224 | } | |
225 | ||
9b67f605 MA |
226 | int |
227 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
228 | void *tag, dmu_buf_t **dbp, int flags) | |
229 | { | |
230 | int err; | |
231 | int db_flags = DB_RF_CANFAIL; | |
232 | ||
233 | if (flags & DMU_READ_NO_PREFETCH) | |
234 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
235 | if (flags & DMU_READ_NO_DECRYPT) |
236 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
237 | |
238 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
239 | if (err == 0) { | |
240 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 241 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 242 | if (err != 0) { |
34dc7c2f | 243 | dbuf_rele(db, tag); |
9b67f605 | 244 | *dbp = NULL; |
34dc7c2f BB |
245 | } |
246 | } | |
247 | ||
34dc7c2f BB |
248 | return (err); |
249 | } | |
250 | ||
251 | int | |
252 | dmu_bonus_max(void) | |
253 | { | |
50c957f7 | 254 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
255 | } |
256 | ||
257 | int | |
572e2857 | 258 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 259 | { |
572e2857 BB |
260 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
261 | dnode_t *dn; | |
262 | int error; | |
34dc7c2f | 263 | |
572e2857 BB |
264 | DB_DNODE_ENTER(db); |
265 | dn = DB_DNODE(db); | |
266 | ||
267 | if (dn->dn_bonus != db) { | |
2e528b49 | 268 | error = SET_ERROR(EINVAL); |
572e2857 | 269 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 270 | error = SET_ERROR(EINVAL); |
572e2857 BB |
271 | } else { |
272 | dnode_setbonuslen(dn, newsize, tx); | |
273 | error = 0; | |
274 | } | |
275 | ||
276 | DB_DNODE_EXIT(db); | |
277 | return (error); | |
34dc7c2f BB |
278 | } |
279 | ||
428870ff | 280 | int |
572e2857 | 281 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 282 | { |
572e2857 BB |
283 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
284 | dnode_t *dn; | |
285 | int error; | |
428870ff | 286 | |
572e2857 BB |
287 | DB_DNODE_ENTER(db); |
288 | dn = DB_DNODE(db); | |
428870ff | 289 | |
9ae529ec | 290 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 291 | error = SET_ERROR(EINVAL); |
572e2857 | 292 | } else if (dn->dn_bonus != db) { |
2e528b49 | 293 | error = SET_ERROR(EINVAL); |
572e2857 BB |
294 | } else { |
295 | dnode_setbonus_type(dn, type, tx); | |
296 | error = 0; | |
297 | } | |
428870ff | 298 | |
572e2857 BB |
299 | DB_DNODE_EXIT(db); |
300 | return (error); | |
301 | } | |
302 | ||
303 | dmu_object_type_t | |
304 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
305 | { | |
306 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
307 | dnode_t *dn; | |
308 | dmu_object_type_t type; | |
309 | ||
310 | DB_DNODE_ENTER(db); | |
311 | dn = DB_DNODE(db); | |
312 | type = dn->dn_bonustype; | |
313 | DB_DNODE_EXIT(db); | |
314 | ||
315 | return (type); | |
428870ff BB |
316 | } |
317 | ||
318 | int | |
319 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
320 | { | |
321 | dnode_t *dn; | |
322 | int error; | |
323 | ||
324 | error = dnode_hold(os, object, FTAG, &dn); | |
325 | dbuf_rm_spill(dn, tx); | |
326 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
327 | dnode_rm_spill(dn, tx); | |
328 | rw_exit(&dn->dn_struct_rwlock); | |
329 | dnode_rele(dn, FTAG); | |
330 | return (error); | |
331 | } | |
332 | ||
34dc7c2f | 333 | /* |
6955b401 BB |
334 | * Lookup and hold the bonus buffer for the provided dnode. If the dnode |
335 | * has not yet been allocated a new bonus dbuf a will be allocated. | |
336 | * Returns ENOENT, EIO, or 0. | |
34dc7c2f | 337 | */ |
6955b401 BB |
338 | int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp, |
339 | uint32_t flags) | |
34dc7c2f | 340 | { |
34dc7c2f BB |
341 | dmu_buf_impl_t *db; |
342 | int error; | |
b5256303 TC |
343 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
344 | ||
345 | if (flags & DMU_READ_NO_PREFETCH) | |
346 | db_flags |= DB_RF_NOPREFETCH; | |
347 | if (flags & DMU_READ_NO_DECRYPT) | |
348 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 349 | |
34dc7c2f BB |
350 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
351 | if (dn->dn_bonus == NULL) { | |
352 | rw_exit(&dn->dn_struct_rwlock); | |
353 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
354 | if (dn->dn_bonus == NULL) | |
355 | dbuf_create_bonus(dn); | |
356 | } | |
357 | db = dn->dn_bonus; | |
34dc7c2f BB |
358 | |
359 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 360 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 361 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 362 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
363 | } |
364 | ||
365 | /* | |
366 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
367 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
368 | * a dnode hold for every dbuf. | |
369 | */ | |
370 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f | 371 | |
b5256303 TC |
372 | error = dbuf_read(db, NULL, db_flags); |
373 | if (error) { | |
374 | dnode_evict_bonus(dn); | |
375 | dbuf_rele(db, tag); | |
376 | *dbp = NULL; | |
377 | return (error); | |
378 | } | |
34dc7c2f BB |
379 | |
380 | *dbp = &db->db; | |
381 | return (0); | |
382 | } | |
383 | ||
b5256303 | 384 | int |
6955b401 | 385 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) |
b5256303 | 386 | { |
6955b401 BB |
387 | dnode_t *dn; |
388 | int error; | |
389 | ||
390 | error = dnode_hold(os, object, FTAG, &dn); | |
391 | if (error) | |
392 | return (error); | |
393 | ||
394 | error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH); | |
395 | dnode_rele(dn, FTAG); | |
396 | ||
397 | return (error); | |
b5256303 TC |
398 | } |
399 | ||
428870ff BB |
400 | /* |
401 | * returns ENOENT, EIO, or 0. | |
402 | * | |
403 | * This interface will allocate a blank spill dbuf when a spill blk | |
404 | * doesn't already exist on the dnode. | |
405 | * | |
406 | * if you only want to find an already existing spill db, then | |
407 | * dmu_spill_hold_existing() should be used. | |
408 | */ | |
409 | int | |
410 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
411 | { | |
412 | dmu_buf_impl_t *db = NULL; | |
413 | int err; | |
414 | ||
415 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
416 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
417 | ||
418 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
419 | ||
420 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
421 | rw_exit(&dn->dn_struct_rwlock); | |
422 | ||
b182ac00 | 423 | if (db == NULL) { |
424 | *dbp = NULL; | |
425 | return (SET_ERROR(EIO)); | |
426 | } | |
572e2857 BB |
427 | err = dbuf_read(db, NULL, flags); |
428 | if (err == 0) | |
429 | *dbp = &db->db; | |
b182ac00 | 430 | else { |
572e2857 | 431 | dbuf_rele(db, tag); |
b182ac00 | 432 | *dbp = NULL; |
433 | } | |
428870ff BB |
434 | return (err); |
435 | } | |
436 | ||
437 | int | |
438 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
439 | { | |
572e2857 BB |
440 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
441 | dnode_t *dn; | |
428870ff BB |
442 | int err; |
443 | ||
572e2857 BB |
444 | DB_DNODE_ENTER(db); |
445 | dn = DB_DNODE(db); | |
446 | ||
447 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 448 | err = SET_ERROR(EINVAL); |
572e2857 BB |
449 | } else { |
450 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
451 | ||
452 | if (!dn->dn_have_spill) { | |
2e528b49 | 453 | err = SET_ERROR(ENOENT); |
572e2857 BB |
454 | } else { |
455 | err = dmu_spill_hold_by_dnode(dn, | |
456 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
457 | } | |
428870ff | 458 | |
428870ff | 459 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 460 | } |
572e2857 BB |
461 | |
462 | DB_DNODE_EXIT(db); | |
428870ff BB |
463 | return (err); |
464 | } | |
465 | ||
466 | int | |
e7504d7a TC |
467 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag, |
468 | dmu_buf_t **dbp) | |
428870ff | 469 | { |
572e2857 BB |
470 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
471 | dnode_t *dn; | |
472 | int err; | |
e7504d7a TC |
473 | uint32_t db_flags = DB_RF_CANFAIL; |
474 | ||
475 | if (flags & DMU_READ_NO_DECRYPT) | |
476 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
477 | |
478 | DB_DNODE_ENTER(db); | |
479 | dn = DB_DNODE(db); | |
e7504d7a | 480 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
481 | DB_DNODE_EXIT(db); |
482 | ||
483 | return (err); | |
428870ff BB |
484 | } |
485 | ||
34dc7c2f BB |
486 | /* |
487 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
488 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
489 | * and can induce severe lock contention when writing to several files | |
490 | * whose dnodes are in the same block. | |
491 | */ | |
492 | static int | |
9babb374 | 493 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
7f60329a | 494 | boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
34dc7c2f BB |
495 | { |
496 | dmu_buf_t **dbp; | |
497 | uint64_t blkid, nblks, i; | |
9babb374 | 498 | uint32_t dbuf_flags; |
34dc7c2f BB |
499 | int err; |
500 | zio_t *zio; | |
501 | ||
502 | ASSERT(length <= DMU_MAX_ACCESS); | |
503 | ||
7f60329a MA |
504 | /* |
505 | * Note: We directly notify the prefetch code of this read, so that | |
506 | * we can tell it about the multi-block read. dbuf_read() only knows | |
507 | * about the one block it is accessing. | |
508 | */ | |
509 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
510 | DB_RF_NOPREFETCH; | |
34dc7c2f BB |
511 | |
512 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
513 | if (dn->dn_datablkshift) { | |
514 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
515 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
516 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
517 | } else { |
518 | if (offset + length > dn->dn_datablksz) { | |
519 | zfs_panic_recover("zfs: accessing past end of object " | |
520 | "%llx/%llx (size=%u access=%llu+%llu)", | |
521 | (longlong_t)dn->dn_objset-> | |
522 | os_dsl_dataset->ds_object, | |
523 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
524 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 525 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 526 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
527 | } |
528 | nblks = 1; | |
529 | } | |
79c76d5b | 530 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 531 | |
b128c09f | 532 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
fcff0f35 | 533 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f | 534 | for (i = 0; i < nblks; i++) { |
7f60329a | 535 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f BB |
536 | if (db == NULL) { |
537 | rw_exit(&dn->dn_struct_rwlock); | |
538 | dmu_buf_rele_array(dbp, nblks, tag); | |
539 | zio_nowait(zio); | |
2e528b49 | 540 | return (SET_ERROR(EIO)); |
34dc7c2f | 541 | } |
7f60329a | 542 | |
34dc7c2f | 543 | /* initiate async i/o */ |
7f60329a | 544 | if (read) |
9babb374 | 545 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
546 | dbp[i] = &db->db; |
547 | } | |
7f60329a | 548 | |
755065f3 AM |
549 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
550 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
551 | dmu_zfetch(&dn->dn_zfetch, blkid, nblks, | |
f664f1ee | 552 | read && DNODE_IS_CACHEABLE(dn), B_TRUE); |
7f60329a | 553 | } |
34dc7c2f BB |
554 | rw_exit(&dn->dn_struct_rwlock); |
555 | ||
556 | /* wait for async i/o */ | |
557 | err = zio_wait(zio); | |
558 | if (err) { | |
559 | dmu_buf_rele_array(dbp, nblks, tag); | |
560 | return (err); | |
561 | } | |
562 | ||
563 | /* wait for other io to complete */ | |
564 | if (read) { | |
565 | for (i = 0; i < nblks; i++) { | |
566 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
567 | mutex_enter(&db->db_mtx); | |
568 | while (db->db_state == DB_READ || | |
569 | db->db_state == DB_FILL) | |
570 | cv_wait(&db->db_changed, &db->db_mtx); | |
571 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 572 | err = SET_ERROR(EIO); |
34dc7c2f BB |
573 | mutex_exit(&db->db_mtx); |
574 | if (err) { | |
575 | dmu_buf_rele_array(dbp, nblks, tag); | |
576 | return (err); | |
577 | } | |
578 | } | |
579 | } | |
580 | ||
581 | *numbufsp = nblks; | |
582 | *dbpp = dbp; | |
583 | return (0); | |
584 | } | |
585 | ||
586 | static int | |
587 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
588 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
589 | { | |
590 | dnode_t *dn; | |
591 | int err; | |
592 | ||
428870ff | 593 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
594 | if (err) |
595 | return (err); | |
596 | ||
597 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 598 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
599 | |
600 | dnode_rele(dn, FTAG); | |
601 | ||
602 | return (err); | |
603 | } | |
604 | ||
605 | int | |
572e2857 | 606 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
7f60329a MA |
607 | uint64_t length, boolean_t read, void *tag, int *numbufsp, |
608 | dmu_buf_t ***dbpp) | |
34dc7c2f | 609 | { |
572e2857 BB |
610 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
611 | dnode_t *dn; | |
34dc7c2f BB |
612 | int err; |
613 | ||
572e2857 BB |
614 | DB_DNODE_ENTER(db); |
615 | dn = DB_DNODE(db); | |
34dc7c2f | 616 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 617 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 618 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
619 | |
620 | return (err); | |
621 | } | |
622 | ||
623 | void | |
624 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
625 | { | |
626 | int i; | |
627 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
628 | ||
629 | if (numbufs == 0) | |
630 | return; | |
631 | ||
632 | for (i = 0; i < numbufs; i++) { | |
633 | if (dbp[i]) | |
634 | dbuf_rele(dbp[i], tag); | |
635 | } | |
636 | ||
637 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
638 | } | |
639 | ||
e8b96c60 | 640 | /* |
fcff0f35 PD |
641 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
642 | * indirect blocks prefeteched will be those that point to the blocks containing | |
643 | * the data starting at offset, and continuing to offset + len. | |
e8b96c60 | 644 | * |
b5256303 TC |
645 | * Note that if the indirect blocks above the blocks being prefetched are not |
646 | * in cache, they will be asychronously read in. | |
e8b96c60 | 647 | */ |
34dc7c2f | 648 | void |
fcff0f35 PD |
649 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
650 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
651 | { |
652 | dnode_t *dn; | |
653 | uint64_t blkid; | |
e8b96c60 | 654 | int nblks, err; |
34dc7c2f | 655 | |
34dc7c2f | 656 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 657 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
658 | |
659 | if (object == 0 || object >= DN_MAX_OBJECT) | |
660 | return; | |
661 | ||
662 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
663 | blkid = dbuf_whichblock(dn, level, |
664 | object * sizeof (dnode_phys_t)); | |
665 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
666 | rw_exit(&dn->dn_struct_rwlock); |
667 | return; | |
668 | } | |
669 | ||
d9b4bf06 MA |
670 | /* |
671 | * See comment before the definition of dmu_prefetch_max. | |
672 | */ | |
673 | len = MIN(len, dmu_prefetch_max); | |
674 | ||
34dc7c2f BB |
675 | /* |
676 | * XXX - Note, if the dnode for the requested object is not | |
677 | * already cached, we will do a *synchronous* read in the | |
678 | * dnode_hold() call. The same is true for any indirects. | |
679 | */ | |
428870ff | 680 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
681 | if (err != 0) |
682 | return; | |
683 | ||
fcff0f35 PD |
684 | /* |
685 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
686 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
687 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
688 | * offset) is the first. Then the number we need to prefetch is the | |
689 | * last - first + 1. | |
690 | */ | |
f664f1ee | 691 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
fcff0f35 PD |
692 | if (level > 0 || dn->dn_datablkshift != 0) { |
693 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
694 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
695 | } else { |
696 | nblks = (offset < dn->dn_datablksz); | |
697 | } | |
698 | ||
699 | if (nblks != 0) { | |
fcff0f35 | 700 | blkid = dbuf_whichblock(dn, level, offset); |
1c27024e | 701 | for (int i = 0; i < nblks; i++) |
fcff0f35 | 702 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f | 703 | } |
34dc7c2f BB |
704 | rw_exit(&dn->dn_struct_rwlock); |
705 | ||
706 | dnode_rele(dn, FTAG); | |
707 | } | |
708 | ||
45d1cae3 BB |
709 | /* |
710 | * Get the next "chunk" of file data to free. We traverse the file from | |
711 | * the end so that the file gets shorter over time (if we crashes in the | |
712 | * middle, this will leave us in a better state). We find allocated file | |
713 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
714 | * |
715 | * On input, *start should be the first offset that does not need to be | |
716 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
717 | * offset that should be freed and l1blks is set to the number of level 1 |
718 | * indirect blocks found within the chunk. | |
45d1cae3 | 719 | */ |
b128c09f | 720 | static int |
65282ee9 | 721 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 722 | { |
65282ee9 | 723 | uint64_t blks; |
b663a23d MA |
724 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
725 | /* bytes of data covered by a level-1 indirect block */ | |
ec4afd27 OM |
726 | uint64_t iblkrange = (uint64_t)dn->dn_datablksz * |
727 | EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
b128c09f | 728 | |
b663a23d | 729 | ASSERT3U(minimum, <=, *start); |
b128c09f | 730 | |
f4c594da TC |
731 | /* |
732 | * Check if we can free the entire range assuming that all of the | |
733 | * L1 blocks in this range have data. If we can, we use this | |
734 | * worst case value as an estimate so we can avoid having to look | |
735 | * at the object's actual data. | |
736 | */ | |
737 | uint64_t total_l1blks = | |
738 | (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / | |
739 | iblkrange; | |
740 | if (total_l1blks <= maxblks) { | |
741 | *l1blks = total_l1blks; | |
b663a23d | 742 | *start = minimum; |
b128c09f BB |
743 | return (0); |
744 | } | |
45d1cae3 | 745 | ASSERT(ISP2(iblkrange)); |
b128c09f | 746 | |
65282ee9 | 747 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
748 | int err; |
749 | ||
b663a23d MA |
750 | /* |
751 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
752 | * indirect block at or before the input offset. We must | |
753 | * decrement *start so that it is at the end of the region | |
754 | * to search. | |
755 | */ | |
756 | (*start)--; | |
f4c594da | 757 | |
b128c09f | 758 | err = dnode_next_offset(dn, |
45d1cae3 | 759 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 760 | |
b663a23d | 761 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 762 | if (err == ESRCH) { |
b663a23d MA |
763 | *start = minimum; |
764 | break; | |
765 | } else if (err != 0) { | |
65282ee9 | 766 | *l1blks = blks; |
b128c09f | 767 | return (err); |
45d1cae3 | 768 | } |
b128c09f | 769 | |
b663a23d | 770 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 771 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 772 | } |
b663a23d MA |
773 | if (*start < minimum) |
774 | *start = minimum; | |
65282ee9 | 775 | *l1blks = blks; |
f4c594da | 776 | |
b128c09f BB |
777 | return (0); |
778 | } | |
779 | ||
a08abc1b GM |
780 | /* |
781 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
782 | * otherwise return false. | |
783 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
784 | */ | |
785 | /*ARGSUSED*/ | |
786 | static boolean_t | |
787 | dmu_objset_zfs_unmounting(objset_t *os) | |
788 | { | |
789 | #ifdef _KERNEL | |
790 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
791 | return (zfs_get_vfs_flag_unmounted(os)); | |
792 | #endif | |
793 | return (B_FALSE); | |
794 | } | |
795 | ||
b128c09f BB |
796 | static int |
797 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 798 | uint64_t length) |
b128c09f | 799 | { |
c97d3069 | 800 | uint64_t object_size; |
b663a23d | 801 | int err; |
539d33c7 GM |
802 | uint64_t dirty_frees_threshold; |
803 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 804 | |
c97d3069 BB |
805 | if (dn == NULL) |
806 | return (SET_ERROR(EINVAL)); | |
807 | ||
808 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 809 | if (offset >= object_size) |
b128c09f | 810 | return (0); |
b128c09f | 811 | |
539d33c7 GM |
812 | if (zfs_per_txg_dirty_frees_percent <= 100) |
813 | dirty_frees_threshold = | |
814 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
815 | else | |
65282ee9 | 816 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 817 | |
b663a23d MA |
818 | if (length == DMU_OBJECT_END || offset + length > object_size) |
819 | length = object_size - offset; | |
820 | ||
821 | while (length != 0) { | |
539d33c7 | 822 | uint64_t chunk_end, chunk_begin, chunk_len; |
65282ee9 | 823 | uint64_t l1blks; |
b663a23d MA |
824 | dmu_tx_t *tx; |
825 | ||
a08abc1b GM |
826 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
827 | return (SET_ERROR(EINTR)); | |
828 | ||
b663a23d MA |
829 | chunk_end = chunk_begin = offset + length; |
830 | ||
831 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 832 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
833 | if (err) |
834 | return (err); | |
b663a23d MA |
835 | ASSERT3U(chunk_begin, >=, offset); |
836 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 837 | |
539d33c7 GM |
838 | chunk_len = chunk_end - chunk_begin; |
839 | ||
b128c09f | 840 | tx = dmu_tx_create(os); |
539d33c7 | 841 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
842 | |
843 | /* | |
844 | * Mark this transaction as typically resulting in a net | |
845 | * reduction in space used. | |
846 | */ | |
847 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
848 | err = dmu_tx_assign(tx, TXG_WAIT); |
849 | if (err) { | |
850 | dmu_tx_abort(tx); | |
851 | return (err); | |
852 | } | |
539d33c7 | 853 | |
f4c594da TC |
854 | uint64_t txg = dmu_tx_get_txg(tx); |
855 | ||
856 | mutex_enter(&dp->dp_lock); | |
857 | uint64_t long_free_dirty = | |
858 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; | |
859 | mutex_exit(&dp->dp_lock); | |
860 | ||
861 | /* | |
862 | * To avoid filling up a TXG with just frees, wait for | |
863 | * the next TXG to open before freeing more chunks if | |
864 | * we have reached the threshold of frees. | |
865 | */ | |
866 | if (dirty_frees_threshold != 0 && | |
867 | long_free_dirty >= dirty_frees_threshold) { | |
868 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); | |
869 | dmu_tx_commit(tx); | |
870 | txg_wait_open(dp, 0, B_TRUE); | |
871 | continue; | |
872 | } | |
873 | ||
65282ee9 AP |
874 | /* |
875 | * In order to prevent unnecessary write throttling, for each | |
876 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
877 | * in dnode_free_range() below. We compare this number to a | |
878 | * tunable threshold, past which we prevent new L1 dirty freeing | |
879 | * blocks from being added into the open TXG. See | |
880 | * dmu_free_long_range_impl() for details. The threshold | |
881 | * prevents write throttle activation due to dirty freeing L1 | |
882 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
883 | */ | |
539d33c7 | 884 | mutex_enter(&dp->dp_lock); |
f4c594da | 885 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += |
65282ee9 | 886 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
887 | mutex_exit(&dp->dp_lock); |
888 | DTRACE_PROBE3(free__long__range, | |
f4c594da TC |
889 | uint64_t, long_free_dirty, uint64_t, chunk_len, |
890 | uint64_t, txg); | |
539d33c7 | 891 | dnode_free_range(dn, chunk_begin, chunk_len, tx); |
440a3eb9 | 892 | |
b128c09f | 893 | dmu_tx_commit(tx); |
b663a23d | 894 | |
539d33c7 | 895 | length -= chunk_len; |
b128c09f BB |
896 | } |
897 | return (0); | |
898 | } | |
899 | ||
900 | int | |
901 | dmu_free_long_range(objset_t *os, uint64_t object, | |
902 | uint64_t offset, uint64_t length) | |
903 | { | |
904 | dnode_t *dn; | |
905 | int err; | |
906 | ||
428870ff | 907 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
908 | if (err != 0) |
909 | return (err); | |
0c03d21a | 910 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
911 | |
912 | /* | |
913 | * It is important to zero out the maxblkid when freeing the entire | |
914 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
915 | * will take the fast path, and (b) dnode_reallocate() can verify | |
916 | * that the entire file has been freed. | |
917 | */ | |
b0bc7a84 | 918 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
919 | dn->dn_maxblkid = 0; |
920 | ||
b128c09f BB |
921 | dnode_rele(dn, FTAG); |
922 | return (err); | |
923 | } | |
924 | ||
925 | int | |
0c03d21a | 926 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 927 | { |
b128c09f BB |
928 | dmu_tx_t *tx; |
929 | int err; | |
930 | ||
b663a23d | 931 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
932 | if (err != 0) |
933 | return (err); | |
b663a23d MA |
934 | |
935 | tx = dmu_tx_create(os); | |
936 | dmu_tx_hold_bonus(tx, object); | |
937 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 938 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
939 | err = dmu_tx_assign(tx, TXG_WAIT); |
940 | if (err == 0) { | |
35df0bb5 TC |
941 | if (err == 0) |
942 | err = dmu_object_free(os, object, tx); | |
440a3eb9 | 943 | |
b663a23d | 944 | dmu_tx_commit(tx); |
b128c09f | 945 | } else { |
b663a23d | 946 | dmu_tx_abort(tx); |
b128c09f | 947 | } |
b663a23d | 948 | |
b128c09f BB |
949 | return (err); |
950 | } | |
951 | ||
34dc7c2f BB |
952 | int |
953 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
954 | uint64_t size, dmu_tx_t *tx) | |
955 | { | |
956 | dnode_t *dn; | |
428870ff | 957 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
958 | if (err) |
959 | return (err); | |
960 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 961 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
962 | dnode_free_range(dn, offset, size, tx); |
963 | dnode_rele(dn, FTAG); | |
964 | return (0); | |
965 | } | |
966 | ||
0eef1bde | 967 | static int |
968 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 969 | void *buf, uint32_t flags) |
34dc7c2f | 970 | { |
34dc7c2f | 971 | dmu_buf_t **dbp; |
0eef1bde | 972 | int numbufs, err = 0; |
34dc7c2f BB |
973 | |
974 | /* | |
975 | * Deal with odd block sizes, where there can't be data past the first | |
976 | * block. If we ever do the tail block optimization, we will need to | |
977 | * handle that here as well. | |
978 | */ | |
45d1cae3 | 979 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 980 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
981 | MIN(size, dn->dn_datablksz - offset); |
982 | bzero((char *)buf + newsz, size - newsz); | |
983 | size = newsz; | |
984 | } | |
985 | ||
986 | while (size > 0) { | |
987 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 988 | int i; |
34dc7c2f BB |
989 | |
990 | /* | |
991 | * NB: we could do this block-at-a-time, but it's nice | |
992 | * to be reading in parallel. | |
993 | */ | |
994 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 995 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
996 | if (err) |
997 | break; | |
998 | ||
999 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1000 | uint64_t tocpy; |
1001 | int64_t bufoff; | |
34dc7c2f BB |
1002 | dmu_buf_t *db = dbp[i]; |
1003 | ||
1004 | ASSERT(size > 0); | |
1005 | ||
1006 | bufoff = offset - db->db_offset; | |
c9520ecc | 1007 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 1008 | |
c9520ecc | 1009 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
1010 | |
1011 | offset += tocpy; | |
1012 | size -= tocpy; | |
1013 | buf = (char *)buf + tocpy; | |
1014 | } | |
1015 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1016 | } | |
34dc7c2f BB |
1017 | return (err); |
1018 | } | |
1019 | ||
0eef1bde | 1020 | int |
1021 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1022 | void *buf, uint32_t flags) | |
34dc7c2f | 1023 | { |
0eef1bde | 1024 | dnode_t *dn; |
1025 | int err; | |
34dc7c2f | 1026 | |
0eef1bde | 1027 | err = dnode_hold(os, object, FTAG, &dn); |
1028 | if (err != 0) | |
1029 | return (err); | |
34dc7c2f | 1030 | |
0eef1bde | 1031 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1032 | dnode_rele(dn, FTAG); | |
1033 | return (err); | |
1034 | } | |
1035 | ||
1036 | int | |
1037 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1038 | uint32_t flags) | |
1039 | { | |
1040 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1041 | } | |
1042 | ||
1043 | static void | |
1044 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1045 | const void *buf, dmu_tx_t *tx) | |
1046 | { | |
1047 | int i; | |
34dc7c2f BB |
1048 | |
1049 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1050 | uint64_t tocpy; |
1051 | int64_t bufoff; | |
34dc7c2f BB |
1052 | dmu_buf_t *db = dbp[i]; |
1053 | ||
1054 | ASSERT(size > 0); | |
1055 | ||
1056 | bufoff = offset - db->db_offset; | |
c9520ecc | 1057 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1058 | |
1059 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1060 | ||
1061 | if (tocpy == db->db_size) | |
1062 | dmu_buf_will_fill(db, tx); | |
1063 | else | |
1064 | dmu_buf_will_dirty(db, tx); | |
1065 | ||
60101509 | 1066 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1067 | |
1068 | if (tocpy == db->db_size) | |
1069 | dmu_buf_fill_done(db, tx); | |
1070 | ||
1071 | offset += tocpy; | |
1072 | size -= tocpy; | |
1073 | buf = (char *)buf + tocpy; | |
1074 | } | |
0eef1bde | 1075 | } |
1076 | ||
1077 | void | |
1078 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1079 | const void *buf, dmu_tx_t *tx) | |
1080 | { | |
1081 | dmu_buf_t **dbp; | |
1082 | int numbufs; | |
1083 | ||
1084 | if (size == 0) | |
1085 | return; | |
1086 | ||
1087 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1088 | FALSE, FTAG, &numbufs, &dbp)); | |
1089 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1090 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1091 | } | |
1092 | ||
1093 | void | |
1094 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1095 | const void *buf, dmu_tx_t *tx) | |
1096 | { | |
1097 | dmu_buf_t **dbp; | |
1098 | int numbufs; | |
1099 | ||
1100 | if (size == 0) | |
1101 | return; | |
1102 | ||
1103 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1104 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1105 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1106 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1107 | } | |
1108 | ||
b128c09f BB |
1109 | void |
1110 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1111 | dmu_tx_t *tx) | |
1112 | { | |
1113 | dmu_buf_t **dbp; | |
1114 | int numbufs, i; | |
1115 | ||
1116 | if (size == 0) | |
1117 | return; | |
1118 | ||
1119 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1120 | FALSE, FTAG, &numbufs, &dbp)); | |
1121 | ||
1122 | for (i = 0; i < numbufs; i++) { | |
1123 | dmu_buf_t *db = dbp[i]; | |
1124 | ||
1125 | dmu_buf_will_not_fill(db, tx); | |
1126 | } | |
1127 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1128 | } | |
1129 | ||
9b67f605 MA |
1130 | void |
1131 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1132 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1133 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1134 | { | |
1135 | dmu_buf_t *db; | |
1136 | ||
1137 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1138 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1139 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1140 | FTAG, &db)); | |
1141 | ||
1142 | dmu_buf_write_embedded(db, | |
1143 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1144 | uncompressed_size, compressed_size, byteorder, tx); | |
1145 | ||
1146 | dmu_buf_rele(db, FTAG); | |
1147 | } | |
1148 | ||
30af21b0 PD |
1149 | void |
1150 | dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1151 | dmu_tx_t *tx) | |
1152 | { | |
1153 | int numbufs, i; | |
1154 | dmu_buf_t **dbp; | |
1155 | ||
1156 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1157 | &numbufs, &dbp)); | |
1158 | for (i = 0; i < numbufs; i++) | |
1159 | dmu_buf_redact(dbp[i], tx); | |
1160 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1161 | } | |
1162 | ||
428870ff BB |
1163 | /* |
1164 | * DMU support for xuio | |
1165 | */ | |
1166 | kstat_t *xuio_ksp = NULL; | |
1167 | ||
59e6e7ca BB |
1168 | typedef struct xuio_stats { |
1169 | /* loaned yet not returned arc_buf */ | |
1170 | kstat_named_t xuiostat_onloan_rbuf; | |
1171 | kstat_named_t xuiostat_onloan_wbuf; | |
1172 | /* whether a copy is made when loaning out a read buffer */ | |
1173 | kstat_named_t xuiostat_rbuf_copied; | |
1174 | kstat_named_t xuiostat_rbuf_nocopy; | |
1175 | /* whether a copy is made when assigning a write buffer */ | |
1176 | kstat_named_t xuiostat_wbuf_copied; | |
1177 | kstat_named_t xuiostat_wbuf_nocopy; | |
1178 | } xuio_stats_t; | |
1179 | ||
1180 | static xuio_stats_t xuio_stats = { | |
1181 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
1182 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
1183 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
1184 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
1185 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
1186 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
1187 | }; | |
1188 | ||
d1d7e268 MK |
1189 | #define XUIOSTAT_INCR(stat, val) \ |
1190 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
1191 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
59e6e7ca | 1192 | |
5a6765cf | 1193 | #ifdef HAVE_UIO_ZEROCOPY |
428870ff BB |
1194 | int |
1195 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
1196 | { | |
1197 | dmu_xuio_t *priv; | |
1198 | uio_t *uio = &xuio->xu_uio; | |
1199 | ||
1200 | uio->uio_iovcnt = nblk; | |
79c76d5b | 1201 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
428870ff | 1202 | |
79c76d5b | 1203 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
428870ff | 1204 | priv->cnt = nblk; |
79c76d5b | 1205 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
5475aada | 1206 | priv->iovp = (iovec_t *)uio->uio_iov; |
428870ff BB |
1207 | XUIO_XUZC_PRIV(xuio) = priv; |
1208 | ||
1209 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1210 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
1211 | else | |
1212 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
1213 | ||
1214 | return (0); | |
1215 | } | |
1216 | ||
1217 | void | |
1218 | dmu_xuio_fini(xuio_t *xuio) | |
1219 | { | |
1220 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1221 | int nblk = priv->cnt; | |
1222 | ||
1223 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
1224 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
1225 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
1226 | ||
1227 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1228 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
1229 | else | |
1230 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
1235 | * and increase priv->next by 1. | |
1236 | */ | |
1237 | int | |
1238 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
1239 | { | |
1240 | struct iovec *iov; | |
1241 | uio_t *uio = &xuio->xu_uio; | |
1242 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1243 | int i = priv->next++; | |
1244 | ||
1245 | ASSERT(i < priv->cnt); | |
2aa34383 | 1246 | ASSERT(off + n <= arc_buf_lsize(abuf)); |
5475aada | 1247 | iov = (iovec_t *)uio->uio_iov + i; |
428870ff BB |
1248 | iov->iov_base = (char *)abuf->b_data + off; |
1249 | iov->iov_len = n; | |
1250 | priv->bufs[i] = abuf; | |
1251 | return (0); | |
1252 | } | |
1253 | ||
1254 | int | |
1255 | dmu_xuio_cnt(xuio_t *xuio) | |
1256 | { | |
1257 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1258 | return (priv->cnt); | |
1259 | } | |
1260 | ||
1261 | arc_buf_t * | |
1262 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
1263 | { | |
1264 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1265 | ||
1266 | ASSERT(i < priv->cnt); | |
1267 | return (priv->bufs[i]); | |
1268 | } | |
1269 | ||
1270 | void | |
1271 | dmu_xuio_clear(xuio_t *xuio, int i) | |
1272 | { | |
1273 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1274 | ||
1275 | ASSERT(i < priv->cnt); | |
1276 | priv->bufs[i] = NULL; | |
1277 | } | |
5a6765cf | 1278 | #endif /* HAVE_UIO_ZEROCOPY */ |
428870ff BB |
1279 | |
1280 | static void | |
1281 | xuio_stat_init(void) | |
1282 | { | |
1283 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1284 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1285 | KSTAT_FLAG_VIRTUAL); | |
1286 | if (xuio_ksp != NULL) { | |
1287 | xuio_ksp->ks_data = &xuio_stats; | |
1288 | kstat_install(xuio_ksp); | |
1289 | } | |
1290 | } | |
1291 | ||
1292 | static void | |
1293 | xuio_stat_fini(void) | |
1294 | { | |
1295 | if (xuio_ksp != NULL) { | |
1296 | kstat_delete(xuio_ksp); | |
1297 | xuio_ksp = NULL; | |
1298 | } | |
1299 | } | |
1300 | ||
1301 | void | |
5043684a | 1302 | xuio_stat_wbuf_copied(void) |
428870ff BB |
1303 | { |
1304 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1305 | } | |
1306 | ||
1307 | void | |
5043684a | 1308 | xuio_stat_wbuf_nocopy(void) |
428870ff BB |
1309 | { |
1310 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1311 | } | |
1312 | ||
34dc7c2f | 1313 | #ifdef _KERNEL |
5228cf01 | 1314 | int |
804e0504 | 1315 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) |
872e8d26 BB |
1316 | { |
1317 | dmu_buf_t **dbp; | |
1318 | int numbufs, i, err; | |
5a6765cf | 1319 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 | 1320 | xuio_t *xuio = NULL; |
5a6765cf | 1321 | #endif |
872e8d26 BB |
1322 | |
1323 | /* | |
1324 | * NB: we could do this block-at-a-time, but it's nice | |
1325 | * to be reading in parallel. | |
1326 | */ | |
804e0504 MA |
1327 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, |
1328 | TRUE, FTAG, &numbufs, &dbp, 0); | |
872e8d26 BB |
1329 | if (err) |
1330 | return (err); | |
1331 | ||
1332 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1333 | uint64_t tocpy; |
1334 | int64_t bufoff; | |
872e8d26 BB |
1335 | dmu_buf_t *db = dbp[i]; |
1336 | ||
1337 | ASSERT(size > 0); | |
1338 | ||
1339 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1340 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1341 | |
5a6765cf | 1342 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 BB |
1343 | if (xuio) { |
1344 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1345 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1346 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1347 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1348 | if (!err) { | |
1349 | uio->uio_resid -= tocpy; | |
1350 | uio->uio_loffset += tocpy; | |
1351 | } | |
1352 | ||
1353 | if (abuf == dbuf_abuf) | |
1354 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1355 | else | |
1356 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
5a6765cf | 1357 | } else |
1358 | #endif | |
872e8d26 BB |
1359 | err = uiomove((char *)db->db_data + bufoff, tocpy, |
1360 | UIO_READ, uio); | |
872e8d26 BB |
1361 | if (err) |
1362 | break; | |
1363 | ||
1364 | size -= tocpy; | |
1365 | } | |
1366 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1367 | ||
1368 | return (err); | |
1369 | } | |
1370 | ||
804e0504 MA |
1371 | /* |
1372 | * Read 'size' bytes into the uio buffer. | |
1373 | * From object zdb->db_object. | |
1374 | * Starting at offset uio->uio_loffset. | |
1375 | * | |
1376 | * If the caller already has a dbuf in the target object | |
1377 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1378 | * because we don't have to find the dnode_t for the object. | |
1379 | */ | |
1380 | int | |
1381 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1382 | { | |
1383 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1384 | dnode_t *dn; | |
1385 | int err; | |
1386 | ||
1387 | if (size == 0) | |
1388 | return (0); | |
1389 | ||
1390 | DB_DNODE_ENTER(db); | |
1391 | dn = DB_DNODE(db); | |
1392 | err = dmu_read_uio_dnode(dn, uio, size); | |
1393 | DB_DNODE_EXIT(db); | |
1394 | ||
1395 | return (err); | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * Read 'size' bytes into the uio buffer. | |
1400 | * From the specified object | |
1401 | * Starting at offset uio->uio_loffset. | |
1402 | */ | |
1403 | int | |
1404 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1405 | { | |
1406 | dnode_t *dn; | |
1407 | int err; | |
1408 | ||
1409 | if (size == 0) | |
1410 | return (0); | |
1411 | ||
1412 | err = dnode_hold(os, object, FTAG, &dn); | |
1413 | if (err) | |
1414 | return (err); | |
1415 | ||
1416 | err = dmu_read_uio_dnode(dn, uio, size); | |
1417 | ||
1418 | dnode_rele(dn, FTAG); | |
1419 | ||
1420 | return (err); | |
1421 | } | |
1422 | ||
5228cf01 | 1423 | int |
872e8d26 BB |
1424 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) |
1425 | { | |
1426 | dmu_buf_t **dbp; | |
1427 | int numbufs; | |
1428 | int err = 0; | |
1429 | int i; | |
1430 | ||
1431 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1432 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1433 | if (err) | |
1434 | return (err); | |
1435 | ||
1436 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1437 | uint64_t tocpy; |
1438 | int64_t bufoff; | |
872e8d26 BB |
1439 | dmu_buf_t *db = dbp[i]; |
1440 | ||
1441 | ASSERT(size > 0); | |
1442 | ||
1443 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1444 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1445 | |
1446 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1447 | ||
1448 | if (tocpy == db->db_size) | |
1449 | dmu_buf_will_fill(db, tx); | |
1450 | else | |
1451 | dmu_buf_will_dirty(db, tx); | |
1452 | ||
1453 | /* | |
1454 | * XXX uiomove could block forever (eg.nfs-backed | |
1455 | * pages). There needs to be a uiolockdown() function | |
1456 | * to lock the pages in memory, so that uiomove won't | |
1457 | * block. | |
1458 | */ | |
1459 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1460 | UIO_WRITE, uio); | |
1461 | ||
1462 | if (tocpy == db->db_size) | |
1463 | dmu_buf_fill_done(db, tx); | |
1464 | ||
1465 | if (err) | |
1466 | break; | |
1467 | ||
1468 | size -= tocpy; | |
1469 | } | |
1470 | ||
1471 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1472 | return (err); | |
1473 | } | |
1474 | ||
804e0504 MA |
1475 | /* |
1476 | * Write 'size' bytes from the uio buffer. | |
1477 | * To object zdb->db_object. | |
1478 | * Starting at offset uio->uio_loffset. | |
1479 | * | |
1480 | * If the caller already has a dbuf in the target object | |
1481 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1482 | * because we don't have to find the dnode_t for the object. | |
1483 | */ | |
428870ff BB |
1484 | int |
1485 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1486 | dmu_tx_t *tx) | |
1487 | { | |
572e2857 BB |
1488 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1489 | dnode_t *dn; | |
1490 | int err; | |
1491 | ||
428870ff BB |
1492 | if (size == 0) |
1493 | return (0); | |
1494 | ||
572e2857 BB |
1495 | DB_DNODE_ENTER(db); |
1496 | dn = DB_DNODE(db); | |
1497 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1498 | DB_DNODE_EXIT(db); | |
1499 | ||
1500 | return (err); | |
428870ff BB |
1501 | } |
1502 | ||
804e0504 MA |
1503 | /* |
1504 | * Write 'size' bytes from the uio buffer. | |
1505 | * To the specified object. | |
1506 | * Starting at offset uio->uio_loffset. | |
1507 | */ | |
428870ff BB |
1508 | int |
1509 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1510 | dmu_tx_t *tx) | |
1511 | { | |
1512 | dnode_t *dn; | |
1513 | int err; | |
1514 | ||
1515 | if (size == 0) | |
1516 | return (0); | |
1517 | ||
1518 | err = dnode_hold(os, object, FTAG, &dn); | |
1519 | if (err) | |
1520 | return (err); | |
1521 | ||
1522 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1523 | ||
1524 | dnode_rele(dn, FTAG); | |
1525 | ||
1526 | return (err); | |
1527 | } | |
872e8d26 | 1528 | #endif /* _KERNEL */ |
34dc7c2f | 1529 | |
9babb374 BB |
1530 | /* |
1531 | * Allocate a loaned anonymous arc buffer. | |
1532 | */ | |
1533 | arc_buf_t * | |
1534 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1535 | { | |
572e2857 | 1536 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1537 | |
2aa34383 | 1538 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1539 | } |
1540 | ||
1541 | /* | |
1542 | * Free a loaned arc buffer. | |
1543 | */ | |
1544 | void | |
1545 | dmu_return_arcbuf(arc_buf_t *buf) | |
1546 | { | |
1547 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1548 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1549 | } |
1550 | ||
b5256303 TC |
1551 | void |
1552 | dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset, | |
1553 | dmu_buf_t *handle, dmu_tx_t *tx) | |
1554 | { | |
1555 | dmu_buf_t *dst_handle; | |
1556 | dmu_buf_impl_t *dstdb; | |
1557 | dmu_buf_impl_t *srcdb = (dmu_buf_impl_t *)handle; | |
149ce888 | 1558 | dmu_object_type_t type; |
b5256303 TC |
1559 | arc_buf_t *abuf; |
1560 | uint64_t datalen; | |
1561 | boolean_t byteorder; | |
1562 | uint8_t salt[ZIO_DATA_SALT_LEN]; | |
1563 | uint8_t iv[ZIO_DATA_IV_LEN]; | |
1564 | uint8_t mac[ZIO_DATA_MAC_LEN]; | |
1565 | ||
1566 | ASSERT3P(srcdb->db_buf, !=, NULL); | |
1567 | ||
1568 | /* hold the db that we want to write to */ | |
1569 | VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &dst_handle, | |
1570 | DMU_READ_NO_DECRYPT)); | |
1571 | dstdb = (dmu_buf_impl_t *)dst_handle; | |
1572 | datalen = arc_buf_size(srcdb->db_buf); | |
1573 | ||
149ce888 TC |
1574 | DB_DNODE_ENTER(dstdb); |
1575 | type = DB_DNODE(dstdb)->dn_type; | |
1576 | DB_DNODE_EXIT(dstdb); | |
1577 | ||
b5256303 TC |
1578 | /* allocated an arc buffer that matches the type of srcdb->db_buf */ |
1579 | if (arc_is_encrypted(srcdb->db_buf)) { | |
1580 | arc_get_raw_params(srcdb->db_buf, &byteorder, salt, iv, mac); | |
1581 | abuf = arc_loan_raw_buf(os->os_spa, dmu_objset_id(os), | |
149ce888 | 1582 | byteorder, salt, iv, mac, type, |
b5256303 TC |
1583 | datalen, arc_buf_lsize(srcdb->db_buf), |
1584 | arc_get_compression(srcdb->db_buf)); | |
1585 | } else { | |
1586 | /* we won't get a compressed db back from dmu_buf_hold() */ | |
1587 | ASSERT3U(arc_get_compression(srcdb->db_buf), | |
1588 | ==, ZIO_COMPRESS_OFF); | |
1589 | abuf = arc_loan_buf(os->os_spa, | |
149ce888 | 1590 | DMU_OT_IS_METADATA(type), datalen); |
b5256303 TC |
1591 | } |
1592 | ||
1593 | ASSERT3U(datalen, ==, arc_buf_size(abuf)); | |
1594 | ||
1595 | /* copy the data to the new buffer and assign it to the dstdb */ | |
1596 | bcopy(srcdb->db_buf->b_data, abuf->b_data, datalen); | |
1597 | dbuf_assign_arcbuf(dstdb, abuf, tx); | |
1598 | dmu_buf_rele(dst_handle, FTAG); | |
1599 | } | |
1600 | ||
9babb374 BB |
1601 | /* |
1602 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1603 | * If this is not possible copy the contents of passed arc buf via | |
1604 | * dmu_write(). | |
1605 | */ | |
305781da | 1606 | int |
440a3eb9 | 1607 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1608 | dmu_tx_t *tx) |
1609 | { | |
9babb374 | 1610 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1611 | objset_t *os = dn->dn_objset; |
1612 | uint64_t object = dn->dn_object; | |
2aa34383 | 1613 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1614 | uint64_t blkid; |
1615 | ||
1616 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1617 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1618 | db = dbuf_hold(dn, blkid, FTAG); |
1619 | if (db == NULL) | |
1620 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1621 | rw_exit(&dn->dn_struct_rwlock); |
1622 | ||
88904bb3 MA |
1623 | /* |
1624 | * We can only assign if the offset is aligned, the arc buf is the | |
2aa34383 | 1625 | * same size as the dbuf, and the dbuf is not metadata. |
88904bb3 | 1626 | */ |
2aa34383 | 1627 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
9babb374 BB |
1628 | dbuf_assign_arcbuf(db, buf, tx); |
1629 | dbuf_rele(db, FTAG); | |
1630 | } else { | |
2aa34383 DK |
1631 | /* compressed bufs must always be assignable to their dbuf */ |
1632 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1633 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1634 | |
9babb374 | 1635 | dbuf_rele(db, FTAG); |
572e2857 | 1636 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1637 | dmu_return_arcbuf(buf); |
428870ff | 1638 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 | 1639 | } |
305781da TC |
1640 | |
1641 | return (0); | |
9babb374 BB |
1642 | } |
1643 | ||
305781da | 1644 | int |
440a3eb9 TC |
1645 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1646 | dmu_tx_t *tx) | |
1647 | { | |
305781da | 1648 | int err; |
440a3eb9 TC |
1649 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1650 | ||
1651 | DB_DNODE_ENTER(dbuf); | |
305781da | 1652 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1653 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1654 | |
1655 | return (err); | |
440a3eb9 TC |
1656 | } |
1657 | ||
34dc7c2f | 1658 | typedef struct { |
428870ff BB |
1659 | dbuf_dirty_record_t *dsa_dr; |
1660 | dmu_sync_cb_t *dsa_done; | |
1661 | zgd_t *dsa_zgd; | |
1662 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1663 | } dmu_sync_arg_t; |
1664 | ||
b128c09f BB |
1665 | /* ARGSUSED */ |
1666 | static void | |
1667 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1668 | { | |
428870ff BB |
1669 | dmu_sync_arg_t *dsa = varg; |
1670 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1671 | blkptr_t *bp = zio->io_bp; |
1672 | ||
428870ff BB |
1673 | if (zio->io_error == 0) { |
1674 | if (BP_IS_HOLE(bp)) { | |
1675 | /* | |
1676 | * A block of zeros may compress to a hole, but the | |
1677 | * block size still needs to be known for replay. | |
1678 | */ | |
1679 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1680 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1681 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1682 | BP_SET_FILL(bp, 1); |
428870ff | 1683 | } |
b128c09f BB |
1684 | } |
1685 | } | |
1686 | ||
428870ff BB |
1687 | static void |
1688 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1689 | { | |
1690 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1691 | } | |
1692 | ||
34dc7c2f BB |
1693 | /* ARGSUSED */ |
1694 | static void | |
1695 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1696 | { | |
428870ff BB |
1697 | dmu_sync_arg_t *dsa = varg; |
1698 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1699 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1700 | zgd_t *zgd = dsa->dsa_zgd; |
1701 | ||
1702 | /* | |
1703 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1704 | * the writes for the lwb have completed. | |
1705 | */ | |
1706 | if (zio->io_error == 0) { | |
1707 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1708 | } | |
34dc7c2f | 1709 | |
34dc7c2f BB |
1710 | mutex_enter(&db->db_mtx); |
1711 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1712 | if (zio->io_error == 0) { |
03c6040b GW |
1713 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1714 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1715 | blkptr_t *bp = zio->io_bp; |
1716 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1717 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1718 | |
1719 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1720 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1721 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1722 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1723 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1724 | } |
428870ff BB |
1725 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1726 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1727 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1728 | |
1729 | /* | |
1730 | * Old style holes are filled with all zeros, whereas | |
1731 | * new-style holes maintain their lsize, type, level, | |
1732 | * and birth time (see zio_write_compress). While we | |
1733 | * need to reset the BP_SET_LSIZE() call that happened | |
1734 | * in dmu_sync_ready for old style holes, we do *not* | |
1735 | * want to wipe out the information contained in new | |
1736 | * style holes. Thus, only zero out the block pointer if | |
1737 | * it's an old style hole. | |
1738 | */ | |
1739 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1740 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1741 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1742 | } else { | |
1743 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1744 | } | |
34dc7c2f BB |
1745 | cv_broadcast(&db->db_changed); |
1746 | mutex_exit(&db->db_mtx); | |
1747 | ||
428870ff | 1748 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1749 | |
428870ff BB |
1750 | kmem_free(dsa, sizeof (*dsa)); |
1751 | } | |
1752 | ||
1753 | static void | |
1754 | dmu_sync_late_arrival_done(zio_t *zio) | |
1755 | { | |
1756 | blkptr_t *bp = zio->io_bp; | |
1757 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1758 | zgd_t *zgd = dsa->dsa_zgd; |
1759 | ||
1760 | if (zio->io_error == 0) { | |
1761 | /* | |
1762 | * Record the vdev(s) backing this blkptr so they can be | |
1763 | * flushed after the writes for the lwb have completed. | |
1764 | */ | |
1765 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1766 | ||
1767 | if (!BP_IS_HOLE(bp)) { | |
1768 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); | |
1769 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); | |
1770 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1771 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1772 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1773 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1774 | } | |
428870ff BB |
1775 | } |
1776 | ||
1777 | dmu_tx_commit(dsa->dsa_tx); | |
1778 | ||
1779 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1780 | ||
a6255b7f | 1781 | abd_put(zio->io_abd); |
428870ff BB |
1782 | kmem_free(dsa, sizeof (*dsa)); |
1783 | } | |
1784 | ||
1785 | static int | |
1786 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1787 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1788 | { |
1789 | dmu_sync_arg_t *dsa; | |
1790 | dmu_tx_t *tx; | |
1791 | ||
1792 | tx = dmu_tx_create(os); | |
1793 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1794 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1795 | dmu_tx_abort(tx); | |
2e528b49 MA |
1796 | /* Make zl_get_data do txg_waited_synced() */ |
1797 | return (SET_ERROR(EIO)); | |
428870ff BB |
1798 | } |
1799 | ||
1ce23dca PS |
1800 | /* |
1801 | * In order to prevent the zgd's lwb from being free'd prior to | |
1802 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1803 | * the lwb's "max txg" takes this tx's txg into account. | |
1804 | */ | |
1805 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1806 | ||
79c76d5b | 1807 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1808 | dsa->dsa_dr = NULL; |
1809 | dsa->dsa_done = done; | |
1810 | dsa->dsa_zgd = zgd; | |
1811 | dsa->dsa_tx = tx; | |
1812 | ||
02dc43bc MA |
1813 | /* |
1814 | * Since we are currently syncing this txg, it's nontrivial to | |
1815 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1816 | * | |
1817 | * When syncing, the db_blkptr is initially the BP of the previous | |
1818 | * txg. We can not nopwrite against it because it will be changed | |
1819 | * (this is similar to the non-late-arrival case where the dbuf is | |
1820 | * dirty in a future txg). | |
1821 | * | |
1822 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1823 | * We can not nopwrite against it because although the BP will not | |
1824 | * (typically) be changed, the data has not yet been persisted to this | |
1825 | * location. | |
1826 | * | |
1827 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1828 | * possible to always nopwrite, because the data that was written in | |
1829 | * this txg is the same data that we are trying to write. However we | |
1830 | * would need to check that this dbuf is not dirty in any future | |
1831 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1832 | * don't nopwrite in this case. | |
1833 | */ | |
1834 | zp->zp_nopwrite = B_FALSE; | |
1835 | ||
a6255b7f DQ |
1836 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1837 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1838 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1839 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1840 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1841 | |
1842 | return (0); | |
34dc7c2f BB |
1843 | } |
1844 | ||
1845 | /* | |
1846 | * Intent log support: sync the block associated with db to disk. | |
1847 | * N.B. and XXX: the caller is responsible for making sure that the | |
1848 | * data isn't changing while dmu_sync() is writing it. | |
1849 | * | |
1850 | * Return values: | |
1851 | * | |
03c6040b | 1852 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1853 | * The caller should not log the write. |
1854 | * | |
1855 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1856 | * The caller should not log the write. | |
1857 | * | |
1858 | * EALREADY: this block is already in the process of being synced. | |
1859 | * The caller should track its progress (somehow). | |
1860 | * | |
428870ff BB |
1861 | * EIO: could not do the I/O. |
1862 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1863 | * |
428870ff BB |
1864 | * 0: the I/O has been initiated. |
1865 | * The caller should log this blkptr in the done callback. | |
1866 | * It is possible that the I/O will fail, in which case | |
1867 | * the error will be reported to the done callback and | |
1868 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1869 | */ |
1870 | int | |
428870ff | 1871 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1872 | { |
428870ff BB |
1873 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1874 | objset_t *os = db->db_objset; | |
1875 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1876 | dbuf_dirty_record_t *dr; |
428870ff | 1877 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1878 | zbookmark_phys_t zb; |
428870ff | 1879 | zio_prop_t zp; |
572e2857 | 1880 | dnode_t *dn; |
34dc7c2f | 1881 | |
428870ff | 1882 | ASSERT(pio != NULL); |
34dc7c2f BB |
1883 | ASSERT(txg != 0); |
1884 | ||
428870ff BB |
1885 | SET_BOOKMARK(&zb, ds->ds_object, |
1886 | db->db.db_object, db->db_level, db->db_blkid); | |
1887 | ||
572e2857 BB |
1888 | DB_DNODE_ENTER(db); |
1889 | dn = DB_DNODE(db); | |
82644107 | 1890 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1891 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1892 | |
1893 | /* | |
428870ff | 1894 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1895 | */ |
428870ff BB |
1896 | if (txg > spa_freeze_txg(os->os_spa)) |
1897 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1898 | |
1899 | /* | |
428870ff BB |
1900 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1901 | * and us. If we determine that this txg is not yet syncing, | |
1902 | * but it begins to sync a moment later, that's OK because the | |
1903 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1904 | */ |
428870ff BB |
1905 | mutex_enter(&db->db_mtx); |
1906 | ||
1907 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1908 | /* |
428870ff | 1909 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1910 | */ |
428870ff | 1911 | mutex_exit(&db->db_mtx); |
2e528b49 | 1912 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1913 | } |
1914 | ||
428870ff BB |
1915 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1916 | /* | |
1917 | * This txg is currently syncing, so we can't mess with | |
1918 | * the dirty record anymore; just write a new log block. | |
1919 | */ | |
1920 | mutex_exit(&db->db_mtx); | |
1921 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1922 | } |
1923 | ||
1924 | dr = db->db_last_dirty; | |
428870ff | 1925 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 1926 | dr = dr->dr_next; |
428870ff BB |
1927 | |
1928 | if (dr == NULL) { | |
34dc7c2f | 1929 | /* |
428870ff | 1930 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1931 | * There's no need to log writes to freed blocks, so we're done. |
1932 | */ | |
1933 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1934 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1935 | } |
1936 | ||
03c6040b GW |
1937 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); |
1938 | ||
02dc43bc MA |
1939 | if (db->db_blkptr != NULL) { |
1940 | /* | |
1941 | * We need to fill in zgd_bp with the current blkptr so that | |
1942 | * the nopwrite code can check if we're writing the same | |
1943 | * data that's already on disk. We can only nopwrite if we | |
1944 | * are sure that after making the copy, db_blkptr will not | |
1945 | * change until our i/o completes. We ensure this by | |
1946 | * holding the db_mtx, and only allowing nopwrite if the | |
1947 | * block is not already dirty (see below). This is verified | |
1948 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1949 | * not changed. | |
1950 | */ | |
1951 | *zgd->zgd_bp = *db->db_blkptr; | |
1952 | } | |
1953 | ||
03c6040b | 1954 | /* |
f3c517d8 MA |
1955 | * Assume the on-disk data is X, the current syncing data (in |
1956 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1957 | * in dmu_sync). | |
1958 | * | |
1959 | * We usually want to perform a nopwrite if X and Z are the | |
1960 | * same. However, if Y is different (i.e. the BP is going to | |
1961 | * change before this write takes effect), then a nopwrite will | |
1962 | * be incorrect - we would override with X, which could have | |
1963 | * been freed when Y was written. | |
1964 | * | |
1965 | * (Note that this is not a concern when we are nop-writing from | |
1966 | * syncing context, because X and Y must be identical, because | |
1967 | * all previous txgs have been synced.) | |
1968 | * | |
1969 | * Therefore, we disable nopwrite if the current BP could change | |
1970 | * before this TXG. There are two ways it could change: by | |
1971 | * being dirty (dr_next is non-NULL), or by being freed | |
1972 | * (dnode_block_freed()). This behavior is verified by | |
1973 | * zio_done(), which VERIFYs that the override BP is identical | |
1974 | * to the on-disk BP. | |
03c6040b | 1975 | */ |
f3c517d8 MA |
1976 | DB_DNODE_ENTER(db); |
1977 | dn = DB_DNODE(db); | |
1978 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
03c6040b | 1979 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1980 | DB_DNODE_EXIT(db); |
03c6040b | 1981 | |
34dc7c2f | 1982 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1983 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1984 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1985 | /* |
428870ff BB |
1986 | * We have already issued a sync write for this buffer, |
1987 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1988 | * have been dirtied since, or we would have cleared the state. |
1989 | */ | |
34dc7c2f | 1990 | mutex_exit(&db->db_mtx); |
2e528b49 | 1991 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1992 | } |
1993 | ||
428870ff | 1994 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1995 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1996 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1997 | |
79c76d5b | 1998 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1999 | dsa->dsa_dr = dr; |
2000 | dsa->dsa_done = done; | |
2001 | dsa->dsa_zgd = zgd; | |
2002 | dsa->dsa_tx = NULL; | |
b128c09f | 2003 | |
428870ff | 2004 | zio_nowait(arc_write(pio, os->os_spa, txg, |
02dc43bc | 2005 | zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 2006 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 2007 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 2008 | |
428870ff | 2009 | return (0); |
34dc7c2f BB |
2010 | } |
2011 | ||
b5256303 TC |
2012 | int |
2013 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
2014 | { | |
2015 | dnode_t *dn; | |
2016 | int err; | |
2017 | ||
2018 | err = dnode_hold(os, object, FTAG, &dn); | |
2019 | if (err) | |
2020 | return (err); | |
2021 | err = dnode_set_nlevels(dn, nlevels, tx); | |
2022 | dnode_rele(dn, FTAG); | |
2023 | return (err); | |
2024 | } | |
2025 | ||
34dc7c2f BB |
2026 | int |
2027 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 2028 | dmu_tx_t *tx) |
34dc7c2f BB |
2029 | { |
2030 | dnode_t *dn; | |
2031 | int err; | |
2032 | ||
428870ff | 2033 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2034 | if (err) |
2035 | return (err); | |
2036 | err = dnode_set_blksz(dn, size, ibs, tx); | |
2037 | dnode_rele(dn, FTAG); | |
2038 | return (err); | |
2039 | } | |
2040 | ||
ae76f45c TC |
2041 | int |
2042 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
2043 | dmu_tx_t *tx) | |
2044 | { | |
2045 | dnode_t *dn; | |
2046 | int err; | |
2047 | ||
2048 | err = dnode_hold(os, object, FTAG, &dn); | |
2049 | if (err) | |
2050 | return (err); | |
2051 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 2052 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
2053 | rw_exit(&dn->dn_struct_rwlock); |
2054 | dnode_rele(dn, FTAG); | |
2055 | return (0); | |
2056 | } | |
2057 | ||
34dc7c2f BB |
2058 | void |
2059 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 2060 | dmu_tx_t *tx) |
34dc7c2f BB |
2061 | { |
2062 | dnode_t *dn; | |
2063 | ||
9b67f605 MA |
2064 | /* |
2065 | * Send streams include each object's checksum function. This | |
2066 | * check ensures that the receiving system can understand the | |
2067 | * checksum function transmitted. | |
2068 | */ | |
2069 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
2070 | ||
2071 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
2072 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
2073 | dn->dn_checksum = checksum; |
2074 | dnode_setdirty(dn, tx); | |
2075 | dnode_rele(dn, FTAG); | |
2076 | } | |
2077 | ||
2078 | void | |
2079 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 2080 | dmu_tx_t *tx) |
34dc7c2f BB |
2081 | { |
2082 | dnode_t *dn; | |
2083 | ||
9b67f605 MA |
2084 | /* |
2085 | * Send streams include each object's compression function. This | |
2086 | * check ensures that the receiving system can understand the | |
2087 | * compression function transmitted. | |
2088 | */ | |
2089 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
2090 | ||
2091 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
2092 | dn->dn_compress = compress; |
2093 | dnode_setdirty(dn, tx); | |
2094 | dnode_rele(dn, FTAG); | |
2095 | } | |
2096 | ||
faf0f58c MA |
2097 | /* |
2098 | * When the "redundant_metadata" property is set to "most", only indirect | |
2099 | * blocks of this level and higher will have an additional ditto block. | |
2100 | */ | |
2101 | int zfs_redundant_metadata_most_ditto_level = 2; | |
2102 | ||
428870ff | 2103 | void |
82644107 | 2104 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
2105 | { |
2106 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 2107 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 2108 | (wp & WP_SPILL)); |
428870ff BB |
2109 | enum zio_checksum checksum = os->os_checksum; |
2110 | enum zio_compress compress = os->os_compress; | |
2111 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
03c6040b GW |
2112 | boolean_t dedup = B_FALSE; |
2113 | boolean_t nopwrite = B_FALSE; | |
428870ff | 2114 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 2115 | boolean_t encrypt = B_FALSE; |
428870ff | 2116 | int copies = os->os_copies; |
a7004725 | 2117 | |
428870ff | 2118 | /* |
03c6040b GW |
2119 | * We maintain different write policies for each of the following |
2120 | * types of data: | |
2121 | * 1. metadata | |
2122 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
2123 | * 3. all other level 0 blocks | |
428870ff BB |
2124 | */ |
2125 | if (ismd) { | |
b1d21733 TC |
2126 | /* |
2127 | * XXX -- we should design a compression algorithm | |
2128 | * that specializes in arrays of bps. | |
2129 | */ | |
2130 | compress = zio_compress_select(os->os_spa, | |
2131 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 2132 | |
428870ff BB |
2133 | /* |
2134 | * Metadata always gets checksummed. If the data | |
2135 | * checksum is multi-bit correctable, and it's not a | |
2136 | * ZBT-style checksum, then it's suitable for metadata | |
2137 | * as well. Otherwise, the metadata checksum defaults | |
2138 | * to fletcher4. | |
2139 | */ | |
3c67d83a TH |
2140 | if (!(zio_checksum_table[checksum].ci_flags & |
2141 | ZCHECKSUM_FLAG_METADATA) || | |
2142 | (zio_checksum_table[checksum].ci_flags & | |
2143 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 2144 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
2145 | |
2146 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
2147 | (os->os_redundant_metadata == | |
2148 | ZFS_REDUNDANT_METADATA_MOST && | |
2149 | (level >= zfs_redundant_metadata_most_ditto_level || | |
2150 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
2151 | copies++; | |
03c6040b GW |
2152 | } else if (wp & WP_NOFILL) { |
2153 | ASSERT(level == 0); | |
428870ff | 2154 | |
428870ff | 2155 | /* |
03c6040b GW |
2156 | * If we're writing preallocated blocks, we aren't actually |
2157 | * writing them so don't set any policy properties. These | |
2158 | * blocks are currently only used by an external subsystem | |
2159 | * outside of zfs (i.e. dump) and not written by the zio | |
2160 | * pipeline. | |
428870ff | 2161 | */ |
03c6040b GW |
2162 | compress = ZIO_COMPRESS_OFF; |
2163 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2164 | } else { |
99197f03 JG |
2165 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2166 | compress); | |
428870ff | 2167 | |
03c6040b GW |
2168 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2169 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2170 | dedup_checksum; | |
428870ff | 2171 | |
03c6040b GW |
2172 | /* |
2173 | * Determine dedup setting. If we are in dmu_sync(), | |
2174 | * we won't actually dedup now because that's all | |
2175 | * done in syncing context; but we do want to use the | |
2176 | * dedup checkum. If the checksum is not strong | |
2177 | * enough to ensure unique signatures, force | |
2178 | * dedup_verify. | |
2179 | */ | |
2180 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2181 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2182 | if (!(zio_checksum_table[checksum].ci_flags & |
2183 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2184 | dedup_verify = B_TRUE; |
2185 | } | |
428870ff | 2186 | |
03c6040b | 2187 | /* |
3c67d83a TH |
2188 | * Enable nopwrite if we have secure enough checksum |
2189 | * algorithm (see comment in zio_nop_write) and | |
2190 | * compression is enabled. We don't enable nopwrite if | |
2191 | * dedup is enabled as the two features are mutually | |
2192 | * exclusive. | |
03c6040b | 2193 | */ |
3c67d83a TH |
2194 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2195 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2196 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2197 | } |
2198 | ||
b5256303 TC |
2199 | /* |
2200 | * All objects in an encrypted objset are protected from modification | |
2201 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2202 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2203 | * not subject to nopwrite since writing the same data will still | |
2204 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2205 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2206 | * object types. | |
2207 | */ | |
2208 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2209 | encrypt = B_TRUE; | |
2210 | ||
2211 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2212 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2213 | nopwrite = B_FALSE; | |
2214 | } else { | |
2215 | dedup = B_FALSE; | |
2216 | } | |
2217 | ||
ae76f45c TC |
2218 | if (level <= 0 && |
2219 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2220 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2221 | } |
b5256303 | 2222 | } |
2aa34383 | 2223 | |
b5256303 TC |
2224 | zp->zp_compress = compress; |
2225 | zp->zp_checksum = checksum; | |
428870ff BB |
2226 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2227 | zp->zp_level = level; | |
faf0f58c | 2228 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2229 | zp->zp_dedup = dedup; |
2230 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2231 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2232 | zp->zp_encrypt = encrypt; |
2233 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
2234 | bzero(zp->zp_salt, ZIO_DATA_SALT_LEN); | |
2235 | bzero(zp->zp_iv, ZIO_DATA_IV_LEN); | |
2236 | bzero(zp->zp_mac, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2237 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2238 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2239 | |
2240 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2241 | } |
2242 | ||
66aca247 DB |
2243 | /* |
2244 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2245 | * in order to determine the location of holes. In order to accurately | |
2246 | * report holes all dirty data must be synced to disk. This causes extremely | |
2247 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2248 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2249 | * report the dnode as having no holes which is always a safe thing to do. | |
2250 | */ | |
34dc7c2f BB |
2251 | int |
2252 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2253 | { | |
2254 | dnode_t *dn; | |
2255 | int i, err; | |
66aca247 | 2256 | boolean_t clean = B_TRUE; |
34dc7c2f | 2257 | |
428870ff | 2258 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2259 | if (err) |
2260 | return (err); | |
66aca247 | 2261 | |
34dc7c2f | 2262 | /* |
2531ce37 | 2263 | * Check if dnode is dirty |
34dc7c2f | 2264 | */ |
454365bb | 2265 | for (i = 0; i < TXG_SIZE; i++) { |
edc1e713 | 2266 | if (multilist_link_active(&dn->dn_dirty_link[i])) { |
2531ce37 | 2267 | clean = B_FALSE; |
ec4f9b8f | 2268 | break; |
2531ce37 | 2269 | } |
34dc7c2f | 2270 | } |
66aca247 DB |
2271 | |
2272 | /* | |
2273 | * If compatibility option is on, sync any current changes before | |
2274 | * we go trundling through the block pointers. | |
2275 | */ | |
2276 | if (!clean && zfs_dmu_offset_next_sync) { | |
2277 | clean = B_TRUE; | |
34dc7c2f BB |
2278 | dnode_rele(dn, FTAG); |
2279 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 2280 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2281 | if (err) |
2282 | return (err); | |
2283 | } | |
2284 | ||
66aca247 DB |
2285 | if (clean) |
2286 | err = dnode_next_offset(dn, | |
2287 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2288 | else | |
2289 | err = SET_ERROR(EBUSY); | |
2290 | ||
34dc7c2f BB |
2291 | dnode_rele(dn, FTAG); |
2292 | ||
2293 | return (err); | |
2294 | } | |
2295 | ||
2296 | void | |
e0b0ca98 | 2297 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2298 | { |
e0b0ca98 | 2299 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2300 | |
34dc7c2f BB |
2301 | doi->doi_data_block_size = dn->dn_datablksz; |
2302 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2303 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2304 | doi->doi_type = dn->dn_type; |
2305 | doi->doi_bonus_type = dn->dn_bonustype; | |
2306 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2307 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2308 | doi->doi_indirection = dn->dn_nlevels; |
2309 | doi->doi_checksum = dn->dn_checksum; | |
2310 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2311 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2312 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2313 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2314 | doi->doi_fill_count = 0; |
1c27024e | 2315 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2316 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2317 | } |
2318 | ||
2319 | void | |
2320 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2321 | { | |
2322 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2323 | mutex_enter(&dn->dn_mtx); | |
2324 | ||
2325 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2326 | |
2327 | mutex_exit(&dn->dn_mtx); | |
2328 | rw_exit(&dn->dn_struct_rwlock); | |
2329 | } | |
2330 | ||
2331 | /* | |
2332 | * Get information on a DMU object. | |
2333 | * If doi is NULL, just indicates whether the object exists. | |
2334 | */ | |
2335 | int | |
2336 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2337 | { | |
2338 | dnode_t *dn; | |
428870ff | 2339 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2340 | |
2341 | if (err) | |
2342 | return (err); | |
2343 | ||
2344 | if (doi != NULL) | |
2345 | dmu_object_info_from_dnode(dn, doi); | |
2346 | ||
2347 | dnode_rele(dn, FTAG); | |
2348 | return (0); | |
2349 | } | |
2350 | ||
2351 | /* | |
2352 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2353 | */ | |
2354 | void | |
572e2857 | 2355 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2356 | { |
572e2857 BB |
2357 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2358 | ||
2359 | DB_DNODE_ENTER(db); | |
2360 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2361 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2362 | } |
2363 | ||
2364 | /* | |
2365 | * Faster still when you only care about the size. | |
2366 | * This is specifically optimized for zfs_getattr(). | |
2367 | */ | |
2368 | void | |
572e2857 BB |
2369 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2370 | u_longlong_t *nblk512) | |
34dc7c2f | 2371 | { |
572e2857 BB |
2372 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2373 | dnode_t *dn; | |
2374 | ||
2375 | DB_DNODE_ENTER(db); | |
2376 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2377 | |
2378 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2379 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2380 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2381 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2382 | DB_DNODE_EXIT(db); | |
2383 | } | |
2384 | ||
2385 | void | |
2386 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2387 | { | |
2388 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2389 | dnode_t *dn; | |
2390 | ||
2391 | DB_DNODE_ENTER(db); | |
2392 | dn = DB_DNODE(db); | |
2393 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2394 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2395 | } |
2396 | ||
2397 | void | |
2398 | byteswap_uint64_array(void *vbuf, size_t size) | |
2399 | { | |
2400 | uint64_t *buf = vbuf; | |
2401 | size_t count = size >> 3; | |
2402 | int i; | |
2403 | ||
2404 | ASSERT((size & 7) == 0); | |
2405 | ||
2406 | for (i = 0; i < count; i++) | |
2407 | buf[i] = BSWAP_64(buf[i]); | |
2408 | } | |
2409 | ||
2410 | void | |
2411 | byteswap_uint32_array(void *vbuf, size_t size) | |
2412 | { | |
2413 | uint32_t *buf = vbuf; | |
2414 | size_t count = size >> 2; | |
2415 | int i; | |
2416 | ||
2417 | ASSERT((size & 3) == 0); | |
2418 | ||
2419 | for (i = 0; i < count; i++) | |
2420 | buf[i] = BSWAP_32(buf[i]); | |
2421 | } | |
2422 | ||
2423 | void | |
2424 | byteswap_uint16_array(void *vbuf, size_t size) | |
2425 | { | |
2426 | uint16_t *buf = vbuf; | |
2427 | size_t count = size >> 1; | |
2428 | int i; | |
2429 | ||
2430 | ASSERT((size & 1) == 0); | |
2431 | ||
2432 | for (i = 0; i < count; i++) | |
2433 | buf[i] = BSWAP_16(buf[i]); | |
2434 | } | |
2435 | ||
2436 | /* ARGSUSED */ | |
2437 | void | |
2438 | byteswap_uint8_array(void *vbuf, size_t size) | |
2439 | { | |
2440 | } | |
2441 | ||
2442 | void | |
2443 | dmu_init(void) | |
2444 | { | |
a6255b7f | 2445 | abd_init(); |
428870ff | 2446 | zfs_dbgmsg_init(); |
572e2857 BB |
2447 | sa_cache_init(); |
2448 | xuio_stat_init(); | |
2449 | dmu_objset_init(); | |
34dc7c2f | 2450 | dnode_init(); |
428870ff | 2451 | zfetch_init(); |
570827e1 | 2452 | dmu_tx_init(); |
34dc7c2f | 2453 | l2arc_init(); |
29809a6c | 2454 | arc_init(); |
d3c2ae1c | 2455 | dbuf_init(); |
34dc7c2f BB |
2456 | } |
2457 | ||
2458 | void | |
2459 | dmu_fini(void) | |
2460 | { | |
e49f1e20 | 2461 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2462 | l2arc_fini(); |
570827e1 | 2463 | dmu_tx_fini(); |
428870ff | 2464 | zfetch_fini(); |
34dc7c2f | 2465 | dbuf_fini(); |
572e2857 BB |
2466 | dnode_fini(); |
2467 | dmu_objset_fini(); | |
428870ff BB |
2468 | xuio_stat_fini(); |
2469 | sa_cache_fini(); | |
2470 | zfs_dbgmsg_fini(); | |
a6255b7f | 2471 | abd_fini(); |
34dc7c2f | 2472 | } |
c28b2279 | 2473 | |
93ce2b4c | 2474 | #if defined(_KERNEL) |
c28b2279 | 2475 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2476 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2477 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2478 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2479 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2480 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2481 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2482 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2483 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2484 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2485 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2486 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2487 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2488 | EXPORT_SYMBOL(dmu_object_info); |
2489 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2490 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2491 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2492 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2493 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2494 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2495 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2496 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2497 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
2498 | EXPORT_SYMBOL(dmu_write_policy); |
2499 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2500 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2501 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2502 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2503 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2504 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2505 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2506 | |
bef78122 | 2507 | /* BEGIN CSTYLED */ |
03c6040b GW |
2508 | module_param(zfs_nopwrite_enabled, int, 0644); |
2509 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2510 | ||
bef78122 DQ |
2511 | module_param(zfs_per_txg_dirty_frees_percent, ulong, 0644); |
2512 | MODULE_PARM_DESC(zfs_per_txg_dirty_frees_percent, | |
2513 | "percentage of dirtied blocks from frees in one TXG"); | |
66aca247 DB |
2514 | |
2515 | module_param(zfs_dmu_offset_next_sync, int, 0644); | |
2516 | MODULE_PARM_DESC(zfs_dmu_offset_next_sync, | |
2517 | "Enable forcing txg sync to find holes"); | |
2518 | ||
d9b4bf06 MA |
2519 | module_param(dmu_prefetch_max, int, 0644); |
2520 | MODULE_PARM_DESC(dmu_prefetch_max, | |
2521 | "Limit one prefetch call to this size"); | |
2522 | ||
bef78122 | 2523 | /* END CSTYLED */ |
66aca247 | 2524 | |
c28b2279 | 2525 | #endif |