]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
82644107 | 23 | * Copyright (c) 2011, 2017 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
bc77ba73 | 25 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
a08abc1b | 26 | * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved. |
5475aada | 27 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. |
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 | ||
a1d477c2 MA |
77 | /* |
78 | * This can be used for testing, to ensure that certain actions happen | |
79 | * while in the middle of a remap (which might otherwise complete too | |
65ca2c1e | 80 | * quickly). Used by ztest(8). |
a1d477c2 | 81 | */ |
65ca2c1e | 82 | int zfs_object_remap_one_indirect_delay_ms = 0; |
a1d477c2 | 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 | ||
161 | blkid = dbuf_whichblock(dn, 0, offset); | |
162 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
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); | |
fcff0f35 | 186 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
187 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
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, | |
552 | read && DNODE_IS_CACHEABLE(dn)); | |
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 | ||
670 | /* | |
671 | * XXX - Note, if the dnode for the requested object is not | |
672 | * already cached, we will do a *synchronous* read in the | |
673 | * dnode_hold() call. The same is true for any indirects. | |
674 | */ | |
428870ff | 675 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
676 | if (err != 0) |
677 | return; | |
678 | ||
679 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
680 | /* |
681 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
682 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
683 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
684 | * offset) is the first. Then the number we need to prefetch is the | |
685 | * last - first + 1. | |
686 | */ | |
687 | if (level > 0 || dn->dn_datablkshift != 0) { | |
688 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
689 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
690 | } else { |
691 | nblks = (offset < dn->dn_datablksz); | |
692 | } | |
693 | ||
694 | if (nblks != 0) { | |
fcff0f35 | 695 | blkid = dbuf_whichblock(dn, level, offset); |
1c27024e | 696 | for (int i = 0; i < nblks; i++) |
fcff0f35 | 697 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f BB |
698 | } |
699 | ||
700 | rw_exit(&dn->dn_struct_rwlock); | |
701 | ||
702 | dnode_rele(dn, FTAG); | |
703 | } | |
704 | ||
45d1cae3 BB |
705 | /* |
706 | * Get the next "chunk" of file data to free. We traverse the file from | |
707 | * the end so that the file gets shorter over time (if we crashes in the | |
708 | * middle, this will leave us in a better state). We find allocated file | |
709 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
710 | * |
711 | * On input, *start should be the first offset that does not need to be | |
712 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
713 | * offset that should be freed and l1blks is set to the number of level 1 |
714 | * indirect blocks found within the chunk. | |
45d1cae3 | 715 | */ |
b128c09f | 716 | static int |
65282ee9 | 717 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 718 | { |
65282ee9 | 719 | uint64_t blks; |
b663a23d MA |
720 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
721 | /* bytes of data covered by a level-1 indirect block */ | |
45d1cae3 | 722 | uint64_t iblkrange = |
b128c09f BB |
723 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
724 | ||
b663a23d | 725 | ASSERT3U(minimum, <=, *start); |
b128c09f | 726 | |
b663a23d MA |
727 | if (*start - minimum <= iblkrange * maxblks) { |
728 | *start = minimum; | |
65282ee9 AP |
729 | /* |
730 | * Assume full L1 blocks and 128k recordsize to approximate the | |
731 | * expected number of L1 blocks in this chunk | |
732 | */ | |
733 | *l1blks = minimum / (1024 * 128 * 1024); | |
b128c09f BB |
734 | return (0); |
735 | } | |
45d1cae3 | 736 | ASSERT(ISP2(iblkrange)); |
b128c09f | 737 | |
65282ee9 | 738 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
739 | int err; |
740 | ||
b663a23d MA |
741 | /* |
742 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
743 | * indirect block at or before the input offset. We must | |
744 | * decrement *start so that it is at the end of the region | |
745 | * to search. | |
746 | */ | |
747 | (*start)--; | |
b128c09f | 748 | err = dnode_next_offset(dn, |
45d1cae3 | 749 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 750 | |
b663a23d | 751 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 752 | if (err == ESRCH) { |
b663a23d MA |
753 | *start = minimum; |
754 | break; | |
755 | } else if (err != 0) { | |
65282ee9 | 756 | *l1blks = blks; |
b128c09f | 757 | return (err); |
45d1cae3 | 758 | } |
b128c09f | 759 | |
b663a23d | 760 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 761 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 762 | } |
b663a23d MA |
763 | if (*start < minimum) |
764 | *start = minimum; | |
65282ee9 | 765 | *l1blks = blks; |
b128c09f BB |
766 | return (0); |
767 | } | |
768 | ||
a08abc1b GM |
769 | /* |
770 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
771 | * otherwise return false. | |
772 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
773 | */ | |
774 | /*ARGSUSED*/ | |
775 | static boolean_t | |
776 | dmu_objset_zfs_unmounting(objset_t *os) | |
777 | { | |
778 | #ifdef _KERNEL | |
779 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
780 | return (zfs_get_vfs_flag_unmounted(os)); | |
781 | #endif | |
782 | return (B_FALSE); | |
783 | } | |
784 | ||
b128c09f BB |
785 | static int |
786 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 787 | uint64_t length) |
b128c09f | 788 | { |
c97d3069 | 789 | uint64_t object_size; |
b663a23d | 790 | int err; |
539d33c7 GM |
791 | uint64_t dirty_frees_threshold; |
792 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 793 | |
c97d3069 BB |
794 | if (dn == NULL) |
795 | return (SET_ERROR(EINVAL)); | |
796 | ||
797 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 798 | if (offset >= object_size) |
b128c09f | 799 | return (0); |
b128c09f | 800 | |
539d33c7 GM |
801 | if (zfs_per_txg_dirty_frees_percent <= 100) |
802 | dirty_frees_threshold = | |
803 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
804 | else | |
65282ee9 | 805 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 806 | |
b663a23d MA |
807 | if (length == DMU_OBJECT_END || offset + length > object_size) |
808 | length = object_size - offset; | |
809 | ||
810 | while (length != 0) { | |
539d33c7 GM |
811 | uint64_t chunk_end, chunk_begin, chunk_len; |
812 | uint64_t long_free_dirty_all_txgs = 0; | |
65282ee9 | 813 | uint64_t l1blks; |
b663a23d MA |
814 | dmu_tx_t *tx; |
815 | ||
a08abc1b GM |
816 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
817 | return (SET_ERROR(EINTR)); | |
818 | ||
b663a23d MA |
819 | chunk_end = chunk_begin = offset + length; |
820 | ||
821 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 822 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
823 | if (err) |
824 | return (err); | |
b663a23d MA |
825 | ASSERT3U(chunk_begin, >=, offset); |
826 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 827 | |
539d33c7 GM |
828 | chunk_len = chunk_end - chunk_begin; |
829 | ||
830 | mutex_enter(&dp->dp_lock); | |
1c27024e | 831 | for (int t = 0; t < TXG_SIZE; t++) { |
539d33c7 GM |
832 | long_free_dirty_all_txgs += |
833 | dp->dp_long_free_dirty_pertxg[t]; | |
834 | } | |
835 | mutex_exit(&dp->dp_lock); | |
836 | ||
837 | /* | |
838 | * To avoid filling up a TXG with just frees wait for | |
839 | * the next TXG to open before freeing more chunks if | |
840 | * we have reached the threshold of frees | |
841 | */ | |
842 | if (dirty_frees_threshold != 0 && | |
843 | long_free_dirty_all_txgs >= dirty_frees_threshold) { | |
750e1f88 | 844 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); |
1b939560 | 845 | txg_wait_open(dp, 0, B_TRUE); |
539d33c7 GM |
846 | continue; |
847 | } | |
848 | ||
b128c09f | 849 | tx = dmu_tx_create(os); |
539d33c7 | 850 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
851 | |
852 | /* | |
853 | * Mark this transaction as typically resulting in a net | |
854 | * reduction in space used. | |
855 | */ | |
856 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
857 | err = dmu_tx_assign(tx, TXG_WAIT); |
858 | if (err) { | |
859 | dmu_tx_abort(tx); | |
860 | return (err); | |
861 | } | |
539d33c7 | 862 | |
65282ee9 AP |
863 | /* |
864 | * In order to prevent unnecessary write throttling, for each | |
865 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
866 | * in dnode_free_range() below. We compare this number to a | |
867 | * tunable threshold, past which we prevent new L1 dirty freeing | |
868 | * blocks from being added into the open TXG. See | |
869 | * dmu_free_long_range_impl() for details. The threshold | |
870 | * prevents write throttle activation due to dirty freeing L1 | |
871 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
872 | */ | |
539d33c7 GM |
873 | mutex_enter(&dp->dp_lock); |
874 | dp->dp_long_free_dirty_pertxg[dmu_tx_get_txg(tx) & TXG_MASK] += | |
65282ee9 | 875 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
876 | mutex_exit(&dp->dp_lock); |
877 | DTRACE_PROBE3(free__long__range, | |
878 | uint64_t, long_free_dirty_all_txgs, uint64_t, chunk_len, | |
879 | uint64_t, dmu_tx_get_txg(tx)); | |
880 | dnode_free_range(dn, chunk_begin, chunk_len, tx); | |
440a3eb9 | 881 | |
b128c09f | 882 | dmu_tx_commit(tx); |
b663a23d | 883 | |
539d33c7 | 884 | length -= chunk_len; |
b128c09f BB |
885 | } |
886 | return (0); | |
887 | } | |
888 | ||
889 | int | |
890 | dmu_free_long_range(objset_t *os, uint64_t object, | |
891 | uint64_t offset, uint64_t length) | |
892 | { | |
893 | dnode_t *dn; | |
894 | int err; | |
895 | ||
428870ff | 896 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
897 | if (err != 0) |
898 | return (err); | |
0c03d21a | 899 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
900 | |
901 | /* | |
902 | * It is important to zero out the maxblkid when freeing the entire | |
903 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
904 | * will take the fast path, and (b) dnode_reallocate() can verify | |
905 | * that the entire file has been freed. | |
906 | */ | |
b0bc7a84 | 907 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
908 | dn->dn_maxblkid = 0; |
909 | ||
b128c09f BB |
910 | dnode_rele(dn, FTAG); |
911 | return (err); | |
912 | } | |
913 | ||
914 | int | |
0c03d21a | 915 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 916 | { |
b128c09f BB |
917 | dmu_tx_t *tx; |
918 | int err; | |
919 | ||
b663a23d | 920 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
921 | if (err != 0) |
922 | return (err); | |
b663a23d MA |
923 | |
924 | tx = dmu_tx_create(os); | |
925 | dmu_tx_hold_bonus(tx, object); | |
926 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 927 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
928 | err = dmu_tx_assign(tx, TXG_WAIT); |
929 | if (err == 0) { | |
35df0bb5 TC |
930 | if (err == 0) |
931 | err = dmu_object_free(os, object, tx); | |
440a3eb9 | 932 | |
b663a23d | 933 | dmu_tx_commit(tx); |
b128c09f | 934 | } else { |
b663a23d | 935 | dmu_tx_abort(tx); |
b128c09f | 936 | } |
b663a23d | 937 | |
b128c09f BB |
938 | return (err); |
939 | } | |
940 | ||
34dc7c2f BB |
941 | int |
942 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
943 | uint64_t size, dmu_tx_t *tx) | |
944 | { | |
945 | dnode_t *dn; | |
428870ff | 946 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
947 | if (err) |
948 | return (err); | |
949 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 950 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
951 | dnode_free_range(dn, offset, size, tx); |
952 | dnode_rele(dn, FTAG); | |
953 | return (0); | |
954 | } | |
955 | ||
0eef1bde | 956 | static int |
957 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 958 | void *buf, uint32_t flags) |
34dc7c2f | 959 | { |
34dc7c2f | 960 | dmu_buf_t **dbp; |
0eef1bde | 961 | int numbufs, err = 0; |
34dc7c2f BB |
962 | |
963 | /* | |
964 | * Deal with odd block sizes, where there can't be data past the first | |
965 | * block. If we ever do the tail block optimization, we will need to | |
966 | * handle that here as well. | |
967 | */ | |
45d1cae3 | 968 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 969 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
970 | MIN(size, dn->dn_datablksz - offset); |
971 | bzero((char *)buf + newsz, size - newsz); | |
972 | size = newsz; | |
973 | } | |
974 | ||
975 | while (size > 0) { | |
976 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 977 | int i; |
34dc7c2f BB |
978 | |
979 | /* | |
980 | * NB: we could do this block-at-a-time, but it's nice | |
981 | * to be reading in parallel. | |
982 | */ | |
983 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 984 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
985 | if (err) |
986 | break; | |
987 | ||
988 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
989 | uint64_t tocpy; |
990 | int64_t bufoff; | |
34dc7c2f BB |
991 | dmu_buf_t *db = dbp[i]; |
992 | ||
993 | ASSERT(size > 0); | |
994 | ||
995 | bufoff = offset - db->db_offset; | |
c9520ecc | 996 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 997 | |
c9520ecc | 998 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
999 | |
1000 | offset += tocpy; | |
1001 | size -= tocpy; | |
1002 | buf = (char *)buf + tocpy; | |
1003 | } | |
1004 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1005 | } | |
34dc7c2f BB |
1006 | return (err); |
1007 | } | |
1008 | ||
0eef1bde | 1009 | int |
1010 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1011 | void *buf, uint32_t flags) | |
34dc7c2f | 1012 | { |
0eef1bde | 1013 | dnode_t *dn; |
1014 | int err; | |
34dc7c2f | 1015 | |
0eef1bde | 1016 | err = dnode_hold(os, object, FTAG, &dn); |
1017 | if (err != 0) | |
1018 | return (err); | |
34dc7c2f | 1019 | |
0eef1bde | 1020 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1021 | dnode_rele(dn, FTAG); | |
1022 | return (err); | |
1023 | } | |
1024 | ||
1025 | int | |
1026 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1027 | uint32_t flags) | |
1028 | { | |
1029 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1030 | } | |
1031 | ||
1032 | static void | |
1033 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1034 | const void *buf, dmu_tx_t *tx) | |
1035 | { | |
1036 | int i; | |
34dc7c2f BB |
1037 | |
1038 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1039 | uint64_t tocpy; |
1040 | int64_t bufoff; | |
34dc7c2f BB |
1041 | dmu_buf_t *db = dbp[i]; |
1042 | ||
1043 | ASSERT(size > 0); | |
1044 | ||
1045 | bufoff = offset - db->db_offset; | |
c9520ecc | 1046 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1047 | |
1048 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1049 | ||
1050 | if (tocpy == db->db_size) | |
1051 | dmu_buf_will_fill(db, tx); | |
1052 | else | |
1053 | dmu_buf_will_dirty(db, tx); | |
1054 | ||
60101509 | 1055 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1056 | |
1057 | if (tocpy == db->db_size) | |
1058 | dmu_buf_fill_done(db, tx); | |
1059 | ||
1060 | offset += tocpy; | |
1061 | size -= tocpy; | |
1062 | buf = (char *)buf + tocpy; | |
1063 | } | |
0eef1bde | 1064 | } |
1065 | ||
1066 | void | |
1067 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1068 | const void *buf, dmu_tx_t *tx) | |
1069 | { | |
1070 | dmu_buf_t **dbp; | |
1071 | int numbufs; | |
1072 | ||
1073 | if (size == 0) | |
1074 | return; | |
1075 | ||
1076 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1077 | FALSE, FTAG, &numbufs, &dbp)); | |
1078 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1079 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1080 | } | |
1081 | ||
1082 | void | |
1083 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1084 | const void *buf, dmu_tx_t *tx) | |
1085 | { | |
1086 | dmu_buf_t **dbp; | |
1087 | int numbufs; | |
1088 | ||
1089 | if (size == 0) | |
1090 | return; | |
1091 | ||
1092 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1093 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1094 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1095 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1096 | } | |
1097 | ||
a1d477c2 MA |
1098 | static int |
1099 | dmu_object_remap_one_indirect(objset_t *os, dnode_t *dn, | |
1100 | uint64_t last_removal_txg, uint64_t offset) | |
1101 | { | |
1102 | uint64_t l1blkid = dbuf_whichblock(dn, 1, offset); | |
65ca2c1e | 1103 | dnode_t *dn_tx; |
a1d477c2 MA |
1104 | int err = 0; |
1105 | ||
1106 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
1107 | dmu_buf_impl_t *dbuf = dbuf_hold_level(dn, 1, l1blkid, FTAG); | |
1108 | ASSERT3P(dbuf, !=, NULL); | |
1109 | ||
1110 | /* | |
1111 | * If the block hasn't been written yet, this default will ensure | |
1112 | * we don't try to remap it. | |
1113 | */ | |
1114 | uint64_t birth = UINT64_MAX; | |
1115 | ASSERT3U(last_removal_txg, !=, UINT64_MAX); | |
1116 | if (dbuf->db_blkptr != NULL) | |
1117 | birth = dbuf->db_blkptr->blk_birth; | |
1118 | rw_exit(&dn->dn_struct_rwlock); | |
1119 | ||
1120 | /* | |
1121 | * If this L1 was already written after the last removal, then we've | |
65ca2c1e BB |
1122 | * already tried to remap it. An additional hold is taken after the |
1123 | * dmu_tx_assign() to handle the case where the dnode is freed while | |
1124 | * waiting for the next open txg. | |
a1d477c2 MA |
1125 | */ |
1126 | if (birth <= last_removal_txg && | |
1127 | dbuf_read(dbuf, NULL, DB_RF_MUST_SUCCEED) == 0 && | |
1128 | dbuf_can_remap(dbuf)) { | |
1129 | dmu_tx_t *tx = dmu_tx_create(os); | |
1130 | dmu_tx_hold_remap_l1indirect(tx, dn->dn_object); | |
1131 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1132 | if (err == 0) { | |
65ca2c1e BB |
1133 | err = dnode_hold(os, dn->dn_object, FTAG, &dn_tx); |
1134 | if (err == 0) { | |
1135 | (void) dbuf_dirty(dbuf, tx); | |
1136 | dnode_rele(dn_tx, FTAG); | |
1137 | } | |
a1d477c2 MA |
1138 | dmu_tx_commit(tx); |
1139 | } else { | |
1140 | dmu_tx_abort(tx); | |
1141 | } | |
1142 | } | |
1143 | ||
1144 | dbuf_rele(dbuf, FTAG); | |
1145 | ||
65ca2c1e | 1146 | delay(MSEC_TO_TICK(zfs_object_remap_one_indirect_delay_ms)); |
a1d477c2 MA |
1147 | |
1148 | return (err); | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * Remap all blockpointers in the object, if possible, so that they reference | |
1153 | * only concrete vdevs. | |
1154 | * | |
1155 | * To do this, iterate over the L0 blockpointers and remap any that reference | |
1156 | * an indirect vdev. Note that we only examine L0 blockpointers; since we | |
1157 | * cannot guarantee that we can remap all blockpointer anyways (due to split | |
1158 | * blocks), we do not want to make the code unnecessarily complicated to | |
1159 | * catch the unlikely case that there is an L1 block on an indirect vdev that | |
1160 | * contains no indirect blockpointers. | |
1161 | */ | |
1162 | int | |
1163 | dmu_object_remap_indirects(objset_t *os, uint64_t object, | |
1164 | uint64_t last_removal_txg) | |
1165 | { | |
1166 | uint64_t offset, l1span; | |
1167 | int err; | |
65ca2c1e | 1168 | dnode_t *dn, *dn_tx; |
a1d477c2 MA |
1169 | |
1170 | err = dnode_hold(os, object, FTAG, &dn); | |
1171 | if (err != 0) { | |
1172 | return (err); | |
1173 | } | |
1174 | ||
1175 | if (dn->dn_nlevels <= 1) { | |
1176 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
1177 | err = SET_ERROR(EINTR); | |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * If the dnode has no indirect blocks, we cannot dirty them. | |
1182 | * We still want to remap the blkptr(s) in the dnode if | |
65ca2c1e BB |
1183 | * appropriate, so mark it as dirty. An additional hold is |
1184 | * taken after the dmu_tx_assign() to handle the case where | |
1185 | * the dnode is freed while waiting for the next open txg. | |
a1d477c2 MA |
1186 | */ |
1187 | if (err == 0 && dnode_needs_remap(dn)) { | |
1188 | dmu_tx_t *tx = dmu_tx_create(os); | |
65ca2c1e BB |
1189 | dmu_tx_hold_bonus(tx, object); |
1190 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1191 | if (err == 0) { | |
1192 | err = dnode_hold(os, object, FTAG, &dn_tx); | |
1193 | if (err == 0) { | |
1194 | dnode_setdirty(dn_tx, tx); | |
1195 | dnode_rele(dn_tx, FTAG); | |
1196 | } | |
a1d477c2 MA |
1197 | dmu_tx_commit(tx); |
1198 | } else { | |
1199 | dmu_tx_abort(tx); | |
1200 | } | |
1201 | } | |
1202 | ||
1203 | dnode_rele(dn, FTAG); | |
1204 | return (err); | |
1205 | } | |
1206 | ||
1207 | offset = 0; | |
1208 | l1span = 1ULL << (dn->dn_indblkshift - SPA_BLKPTRSHIFT + | |
1209 | dn->dn_datablkshift); | |
1210 | /* | |
1211 | * Find the next L1 indirect that is not a hole. | |
1212 | */ | |
1213 | while (dnode_next_offset(dn, 0, &offset, 2, 1, 0) == 0) { | |
1214 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
1215 | err = SET_ERROR(EINTR); | |
1216 | break; | |
1217 | } | |
1218 | if ((err = dmu_object_remap_one_indirect(os, dn, | |
1219 | last_removal_txg, offset)) != 0) { | |
1220 | break; | |
1221 | } | |
1222 | offset += l1span; | |
1223 | } | |
1224 | ||
1225 | dnode_rele(dn, FTAG); | |
1226 | return (err); | |
1227 | } | |
1228 | ||
b128c09f BB |
1229 | void |
1230 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1231 | dmu_tx_t *tx) | |
1232 | { | |
1233 | dmu_buf_t **dbp; | |
1234 | int numbufs, i; | |
1235 | ||
1236 | if (size == 0) | |
1237 | return; | |
1238 | ||
1239 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1240 | FALSE, FTAG, &numbufs, &dbp)); | |
1241 | ||
1242 | for (i = 0; i < numbufs; i++) { | |
1243 | dmu_buf_t *db = dbp[i]; | |
1244 | ||
1245 | dmu_buf_will_not_fill(db, tx); | |
1246 | } | |
1247 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1248 | } | |
1249 | ||
9b67f605 MA |
1250 | void |
1251 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1252 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1253 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1254 | { | |
1255 | dmu_buf_t *db; | |
1256 | ||
1257 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1258 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1259 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1260 | FTAG, &db)); | |
1261 | ||
1262 | dmu_buf_write_embedded(db, | |
1263 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1264 | uncompressed_size, compressed_size, byteorder, tx); | |
1265 | ||
1266 | dmu_buf_rele(db, FTAG); | |
1267 | } | |
1268 | ||
428870ff BB |
1269 | /* |
1270 | * DMU support for xuio | |
1271 | */ | |
1272 | kstat_t *xuio_ksp = NULL; | |
1273 | ||
59e6e7ca BB |
1274 | typedef struct xuio_stats { |
1275 | /* loaned yet not returned arc_buf */ | |
1276 | kstat_named_t xuiostat_onloan_rbuf; | |
1277 | kstat_named_t xuiostat_onloan_wbuf; | |
1278 | /* whether a copy is made when loaning out a read buffer */ | |
1279 | kstat_named_t xuiostat_rbuf_copied; | |
1280 | kstat_named_t xuiostat_rbuf_nocopy; | |
1281 | /* whether a copy is made when assigning a write buffer */ | |
1282 | kstat_named_t xuiostat_wbuf_copied; | |
1283 | kstat_named_t xuiostat_wbuf_nocopy; | |
1284 | } xuio_stats_t; | |
1285 | ||
1286 | static xuio_stats_t xuio_stats = { | |
1287 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
1288 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
1289 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
1290 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
1291 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
1292 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
1293 | }; | |
1294 | ||
d1d7e268 MK |
1295 | #define XUIOSTAT_INCR(stat, val) \ |
1296 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
1297 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
59e6e7ca | 1298 | |
5a6765cf | 1299 | #ifdef HAVE_UIO_ZEROCOPY |
428870ff BB |
1300 | int |
1301 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
1302 | { | |
1303 | dmu_xuio_t *priv; | |
1304 | uio_t *uio = &xuio->xu_uio; | |
1305 | ||
1306 | uio->uio_iovcnt = nblk; | |
79c76d5b | 1307 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
428870ff | 1308 | |
79c76d5b | 1309 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
428870ff | 1310 | priv->cnt = nblk; |
79c76d5b | 1311 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
5475aada | 1312 | priv->iovp = (iovec_t *)uio->uio_iov; |
428870ff BB |
1313 | XUIO_XUZC_PRIV(xuio) = priv; |
1314 | ||
1315 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1316 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
1317 | else | |
1318 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
1319 | ||
1320 | return (0); | |
1321 | } | |
1322 | ||
1323 | void | |
1324 | dmu_xuio_fini(xuio_t *xuio) | |
1325 | { | |
1326 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1327 | int nblk = priv->cnt; | |
1328 | ||
1329 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
1330 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
1331 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
1332 | ||
1333 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1334 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
1335 | else | |
1336 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
1337 | } | |
1338 | ||
1339 | /* | |
1340 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
1341 | * and increase priv->next by 1. | |
1342 | */ | |
1343 | int | |
1344 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
1345 | { | |
1346 | struct iovec *iov; | |
1347 | uio_t *uio = &xuio->xu_uio; | |
1348 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1349 | int i = priv->next++; | |
1350 | ||
1351 | ASSERT(i < priv->cnt); | |
2aa34383 | 1352 | ASSERT(off + n <= arc_buf_lsize(abuf)); |
5475aada | 1353 | iov = (iovec_t *)uio->uio_iov + i; |
428870ff BB |
1354 | iov->iov_base = (char *)abuf->b_data + off; |
1355 | iov->iov_len = n; | |
1356 | priv->bufs[i] = abuf; | |
1357 | return (0); | |
1358 | } | |
1359 | ||
1360 | int | |
1361 | dmu_xuio_cnt(xuio_t *xuio) | |
1362 | { | |
1363 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1364 | return (priv->cnt); | |
1365 | } | |
1366 | ||
1367 | arc_buf_t * | |
1368 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
1369 | { | |
1370 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1371 | ||
1372 | ASSERT(i < priv->cnt); | |
1373 | return (priv->bufs[i]); | |
1374 | } | |
1375 | ||
1376 | void | |
1377 | dmu_xuio_clear(xuio_t *xuio, int i) | |
1378 | { | |
1379 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1380 | ||
1381 | ASSERT(i < priv->cnt); | |
1382 | priv->bufs[i] = NULL; | |
1383 | } | |
5a6765cf | 1384 | #endif /* HAVE_UIO_ZEROCOPY */ |
428870ff BB |
1385 | |
1386 | static void | |
1387 | xuio_stat_init(void) | |
1388 | { | |
1389 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1390 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1391 | KSTAT_FLAG_VIRTUAL); | |
1392 | if (xuio_ksp != NULL) { | |
1393 | xuio_ksp->ks_data = &xuio_stats; | |
1394 | kstat_install(xuio_ksp); | |
1395 | } | |
1396 | } | |
1397 | ||
1398 | static void | |
1399 | xuio_stat_fini(void) | |
1400 | { | |
1401 | if (xuio_ksp != NULL) { | |
1402 | kstat_delete(xuio_ksp); | |
1403 | xuio_ksp = NULL; | |
1404 | } | |
1405 | } | |
1406 | ||
1407 | void | |
5043684a | 1408 | xuio_stat_wbuf_copied(void) |
428870ff BB |
1409 | { |
1410 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1411 | } | |
1412 | ||
1413 | void | |
5043684a | 1414 | xuio_stat_wbuf_nocopy(void) |
428870ff BB |
1415 | { |
1416 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1417 | } | |
1418 | ||
34dc7c2f | 1419 | #ifdef _KERNEL |
5228cf01 | 1420 | int |
804e0504 | 1421 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) |
872e8d26 BB |
1422 | { |
1423 | dmu_buf_t **dbp; | |
1424 | int numbufs, i, err; | |
5a6765cf | 1425 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 | 1426 | xuio_t *xuio = NULL; |
5a6765cf | 1427 | #endif |
872e8d26 BB |
1428 | |
1429 | /* | |
1430 | * NB: we could do this block-at-a-time, but it's nice | |
1431 | * to be reading in parallel. | |
1432 | */ | |
804e0504 MA |
1433 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, |
1434 | TRUE, FTAG, &numbufs, &dbp, 0); | |
872e8d26 BB |
1435 | if (err) |
1436 | return (err); | |
1437 | ||
1438 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1439 | uint64_t tocpy; |
1440 | int64_t bufoff; | |
872e8d26 BB |
1441 | dmu_buf_t *db = dbp[i]; |
1442 | ||
1443 | ASSERT(size > 0); | |
1444 | ||
1445 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1446 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1447 | |
5a6765cf | 1448 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 BB |
1449 | if (xuio) { |
1450 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1451 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1452 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1453 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1454 | if (!err) { | |
1455 | uio->uio_resid -= tocpy; | |
1456 | uio->uio_loffset += tocpy; | |
1457 | } | |
1458 | ||
1459 | if (abuf == dbuf_abuf) | |
1460 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1461 | else | |
1462 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
5a6765cf | 1463 | } else |
1464 | #endif | |
872e8d26 BB |
1465 | err = uiomove((char *)db->db_data + bufoff, tocpy, |
1466 | UIO_READ, uio); | |
872e8d26 BB |
1467 | if (err) |
1468 | break; | |
1469 | ||
1470 | size -= tocpy; | |
1471 | } | |
1472 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1473 | ||
1474 | return (err); | |
1475 | } | |
1476 | ||
804e0504 MA |
1477 | /* |
1478 | * Read 'size' bytes into the uio buffer. | |
1479 | * From object zdb->db_object. | |
1480 | * Starting at offset uio->uio_loffset. | |
1481 | * | |
1482 | * If the caller already has a dbuf in the target object | |
1483 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1484 | * because we don't have to find the dnode_t for the object. | |
1485 | */ | |
1486 | int | |
1487 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1488 | { | |
1489 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1490 | dnode_t *dn; | |
1491 | int err; | |
1492 | ||
1493 | if (size == 0) | |
1494 | return (0); | |
1495 | ||
1496 | DB_DNODE_ENTER(db); | |
1497 | dn = DB_DNODE(db); | |
1498 | err = dmu_read_uio_dnode(dn, uio, size); | |
1499 | DB_DNODE_EXIT(db); | |
1500 | ||
1501 | return (err); | |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * Read 'size' bytes into the uio buffer. | |
1506 | * From the specified object | |
1507 | * Starting at offset uio->uio_loffset. | |
1508 | */ | |
1509 | int | |
1510 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
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_read_uio_dnode(dn, uio, size); | |
1523 | ||
1524 | dnode_rele(dn, FTAG); | |
1525 | ||
1526 | return (err); | |
1527 | } | |
1528 | ||
5228cf01 | 1529 | int |
872e8d26 BB |
1530 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) |
1531 | { | |
1532 | dmu_buf_t **dbp; | |
1533 | int numbufs; | |
1534 | int err = 0; | |
1535 | int i; | |
1536 | ||
1537 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1538 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1539 | if (err) | |
1540 | return (err); | |
1541 | ||
1542 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1543 | uint64_t tocpy; |
1544 | int64_t bufoff; | |
872e8d26 BB |
1545 | dmu_buf_t *db = dbp[i]; |
1546 | ||
1547 | ASSERT(size > 0); | |
1548 | ||
1549 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1550 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1551 | |
1552 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1553 | ||
1554 | if (tocpy == db->db_size) | |
1555 | dmu_buf_will_fill(db, tx); | |
1556 | else | |
1557 | dmu_buf_will_dirty(db, tx); | |
1558 | ||
1559 | /* | |
1560 | * XXX uiomove could block forever (eg.nfs-backed | |
1561 | * pages). There needs to be a uiolockdown() function | |
1562 | * to lock the pages in memory, so that uiomove won't | |
1563 | * block. | |
1564 | */ | |
1565 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1566 | UIO_WRITE, uio); | |
1567 | ||
1568 | if (tocpy == db->db_size) | |
1569 | dmu_buf_fill_done(db, tx); | |
1570 | ||
1571 | if (err) | |
1572 | break; | |
1573 | ||
1574 | size -= tocpy; | |
1575 | } | |
1576 | ||
1577 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1578 | return (err); | |
1579 | } | |
1580 | ||
804e0504 MA |
1581 | /* |
1582 | * Write 'size' bytes from the uio buffer. | |
1583 | * To object zdb->db_object. | |
1584 | * Starting at offset uio->uio_loffset. | |
1585 | * | |
1586 | * If the caller already has a dbuf in the target object | |
1587 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1588 | * because we don't have to find the dnode_t for the object. | |
1589 | */ | |
428870ff BB |
1590 | int |
1591 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1592 | dmu_tx_t *tx) | |
1593 | { | |
572e2857 BB |
1594 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1595 | dnode_t *dn; | |
1596 | int err; | |
1597 | ||
428870ff BB |
1598 | if (size == 0) |
1599 | return (0); | |
1600 | ||
572e2857 BB |
1601 | DB_DNODE_ENTER(db); |
1602 | dn = DB_DNODE(db); | |
1603 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1604 | DB_DNODE_EXIT(db); | |
1605 | ||
1606 | return (err); | |
428870ff BB |
1607 | } |
1608 | ||
804e0504 MA |
1609 | /* |
1610 | * Write 'size' bytes from the uio buffer. | |
1611 | * To the specified object. | |
1612 | * Starting at offset uio->uio_loffset. | |
1613 | */ | |
428870ff BB |
1614 | int |
1615 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1616 | dmu_tx_t *tx) | |
1617 | { | |
1618 | dnode_t *dn; | |
1619 | int err; | |
1620 | ||
1621 | if (size == 0) | |
1622 | return (0); | |
1623 | ||
1624 | err = dnode_hold(os, object, FTAG, &dn); | |
1625 | if (err) | |
1626 | return (err); | |
1627 | ||
1628 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1629 | ||
1630 | dnode_rele(dn, FTAG); | |
1631 | ||
1632 | return (err); | |
1633 | } | |
872e8d26 | 1634 | #endif /* _KERNEL */ |
34dc7c2f | 1635 | |
9babb374 BB |
1636 | /* |
1637 | * Allocate a loaned anonymous arc buffer. | |
1638 | */ | |
1639 | arc_buf_t * | |
1640 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1641 | { | |
572e2857 | 1642 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1643 | |
2aa34383 | 1644 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1645 | } |
1646 | ||
1647 | /* | |
1648 | * Free a loaned arc buffer. | |
1649 | */ | |
1650 | void | |
1651 | dmu_return_arcbuf(arc_buf_t *buf) | |
1652 | { | |
1653 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1654 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1655 | } |
1656 | ||
b5256303 TC |
1657 | void |
1658 | dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset, | |
1659 | dmu_buf_t *handle, dmu_tx_t *tx) | |
1660 | { | |
1661 | dmu_buf_t *dst_handle; | |
1662 | dmu_buf_impl_t *dstdb; | |
1663 | dmu_buf_impl_t *srcdb = (dmu_buf_impl_t *)handle; | |
149ce888 | 1664 | dmu_object_type_t type; |
b5256303 TC |
1665 | arc_buf_t *abuf; |
1666 | uint64_t datalen; | |
1667 | boolean_t byteorder; | |
1668 | uint8_t salt[ZIO_DATA_SALT_LEN]; | |
1669 | uint8_t iv[ZIO_DATA_IV_LEN]; | |
1670 | uint8_t mac[ZIO_DATA_MAC_LEN]; | |
1671 | ||
1672 | ASSERT3P(srcdb->db_buf, !=, NULL); | |
1673 | ||
1674 | /* hold the db that we want to write to */ | |
1675 | VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &dst_handle, | |
1676 | DMU_READ_NO_DECRYPT)); | |
1677 | dstdb = (dmu_buf_impl_t *)dst_handle; | |
1678 | datalen = arc_buf_size(srcdb->db_buf); | |
1679 | ||
149ce888 TC |
1680 | DB_DNODE_ENTER(dstdb); |
1681 | type = DB_DNODE(dstdb)->dn_type; | |
1682 | DB_DNODE_EXIT(dstdb); | |
1683 | ||
b5256303 TC |
1684 | /* allocated an arc buffer that matches the type of srcdb->db_buf */ |
1685 | if (arc_is_encrypted(srcdb->db_buf)) { | |
1686 | arc_get_raw_params(srcdb->db_buf, &byteorder, salt, iv, mac); | |
1687 | abuf = arc_loan_raw_buf(os->os_spa, dmu_objset_id(os), | |
149ce888 | 1688 | byteorder, salt, iv, mac, type, |
b5256303 TC |
1689 | datalen, arc_buf_lsize(srcdb->db_buf), |
1690 | arc_get_compression(srcdb->db_buf)); | |
1691 | } else { | |
1692 | /* we won't get a compressed db back from dmu_buf_hold() */ | |
1693 | ASSERT3U(arc_get_compression(srcdb->db_buf), | |
1694 | ==, ZIO_COMPRESS_OFF); | |
1695 | abuf = arc_loan_buf(os->os_spa, | |
149ce888 | 1696 | DMU_OT_IS_METADATA(type), datalen); |
b5256303 TC |
1697 | } |
1698 | ||
1699 | ASSERT3U(datalen, ==, arc_buf_size(abuf)); | |
1700 | ||
1701 | /* copy the data to the new buffer and assign it to the dstdb */ | |
1702 | bcopy(srcdb->db_buf->b_data, abuf->b_data, datalen); | |
1703 | dbuf_assign_arcbuf(dstdb, abuf, tx); | |
1704 | dmu_buf_rele(dst_handle, FTAG); | |
1705 | } | |
1706 | ||
9babb374 BB |
1707 | /* |
1708 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1709 | * If this is not possible copy the contents of passed arc buf via | |
1710 | * dmu_write(). | |
1711 | */ | |
305781da | 1712 | int |
440a3eb9 | 1713 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1714 | dmu_tx_t *tx) |
1715 | { | |
9babb374 | 1716 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1717 | objset_t *os = dn->dn_objset; |
1718 | uint64_t object = dn->dn_object; | |
2aa34383 | 1719 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1720 | uint64_t blkid; |
1721 | ||
1722 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1723 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1724 | db = dbuf_hold(dn, blkid, FTAG); |
1725 | if (db == NULL) | |
1726 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1727 | rw_exit(&dn->dn_struct_rwlock); |
1728 | ||
88904bb3 MA |
1729 | /* |
1730 | * We can only assign if the offset is aligned, the arc buf is the | |
2aa34383 | 1731 | * same size as the dbuf, and the dbuf is not metadata. |
88904bb3 | 1732 | */ |
2aa34383 | 1733 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
9babb374 BB |
1734 | dbuf_assign_arcbuf(db, buf, tx); |
1735 | dbuf_rele(db, FTAG); | |
1736 | } else { | |
2aa34383 DK |
1737 | /* compressed bufs must always be assignable to their dbuf */ |
1738 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1739 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1740 | |
9babb374 | 1741 | dbuf_rele(db, FTAG); |
572e2857 | 1742 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1743 | dmu_return_arcbuf(buf); |
428870ff | 1744 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 | 1745 | } |
305781da TC |
1746 | |
1747 | return (0); | |
9babb374 BB |
1748 | } |
1749 | ||
305781da | 1750 | int |
440a3eb9 TC |
1751 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1752 | dmu_tx_t *tx) | |
1753 | { | |
305781da | 1754 | int err; |
440a3eb9 TC |
1755 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1756 | ||
1757 | DB_DNODE_ENTER(dbuf); | |
305781da | 1758 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1759 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1760 | |
1761 | return (err); | |
440a3eb9 TC |
1762 | } |
1763 | ||
34dc7c2f | 1764 | typedef struct { |
428870ff BB |
1765 | dbuf_dirty_record_t *dsa_dr; |
1766 | dmu_sync_cb_t *dsa_done; | |
1767 | zgd_t *dsa_zgd; | |
1768 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1769 | } dmu_sync_arg_t; |
1770 | ||
b128c09f BB |
1771 | /* ARGSUSED */ |
1772 | static void | |
1773 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1774 | { | |
428870ff BB |
1775 | dmu_sync_arg_t *dsa = varg; |
1776 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1777 | blkptr_t *bp = zio->io_bp; |
1778 | ||
428870ff BB |
1779 | if (zio->io_error == 0) { |
1780 | if (BP_IS_HOLE(bp)) { | |
1781 | /* | |
1782 | * A block of zeros may compress to a hole, but the | |
1783 | * block size still needs to be known for replay. | |
1784 | */ | |
1785 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1786 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1787 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1788 | BP_SET_FILL(bp, 1); |
428870ff | 1789 | } |
b128c09f BB |
1790 | } |
1791 | } | |
1792 | ||
428870ff BB |
1793 | static void |
1794 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1795 | { | |
1796 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1797 | } | |
1798 | ||
34dc7c2f BB |
1799 | /* ARGSUSED */ |
1800 | static void | |
1801 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1802 | { | |
428870ff BB |
1803 | dmu_sync_arg_t *dsa = varg; |
1804 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1805 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1806 | zgd_t *zgd = dsa->dsa_zgd; |
1807 | ||
1808 | /* | |
1809 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1810 | * the writes for the lwb have completed. | |
1811 | */ | |
1812 | if (zio->io_error == 0) { | |
1813 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1814 | } | |
34dc7c2f | 1815 | |
34dc7c2f BB |
1816 | mutex_enter(&db->db_mtx); |
1817 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1818 | if (zio->io_error == 0) { |
03c6040b GW |
1819 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1820 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1821 | blkptr_t *bp = zio->io_bp; |
1822 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1823 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1824 | |
1825 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1826 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1827 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1828 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1829 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1830 | } |
428870ff BB |
1831 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1832 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1833 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1834 | |
1835 | /* | |
1836 | * Old style holes are filled with all zeros, whereas | |
1837 | * new-style holes maintain their lsize, type, level, | |
1838 | * and birth time (see zio_write_compress). While we | |
1839 | * need to reset the BP_SET_LSIZE() call that happened | |
1840 | * in dmu_sync_ready for old style holes, we do *not* | |
1841 | * want to wipe out the information contained in new | |
1842 | * style holes. Thus, only zero out the block pointer if | |
1843 | * it's an old style hole. | |
1844 | */ | |
1845 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1846 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1847 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1848 | } else { | |
1849 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1850 | } | |
34dc7c2f BB |
1851 | cv_broadcast(&db->db_changed); |
1852 | mutex_exit(&db->db_mtx); | |
1853 | ||
428870ff | 1854 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1855 | |
428870ff BB |
1856 | kmem_free(dsa, sizeof (*dsa)); |
1857 | } | |
1858 | ||
1859 | static void | |
1860 | dmu_sync_late_arrival_done(zio_t *zio) | |
1861 | { | |
1862 | blkptr_t *bp = zio->io_bp; | |
1863 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1864 | zgd_t *zgd = dsa->dsa_zgd; |
1865 | ||
1866 | if (zio->io_error == 0) { | |
1867 | /* | |
1868 | * Record the vdev(s) backing this blkptr so they can be | |
1869 | * flushed after the writes for the lwb have completed. | |
1870 | */ | |
1871 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1872 | ||
1873 | if (!BP_IS_HOLE(bp)) { | |
1874 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); | |
1875 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); | |
1876 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1877 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1878 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1879 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1880 | } | |
428870ff BB |
1881 | } |
1882 | ||
1883 | dmu_tx_commit(dsa->dsa_tx); | |
1884 | ||
1885 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1886 | ||
a6255b7f | 1887 | abd_put(zio->io_abd); |
428870ff BB |
1888 | kmem_free(dsa, sizeof (*dsa)); |
1889 | } | |
1890 | ||
1891 | static int | |
1892 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1893 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1894 | { |
1895 | dmu_sync_arg_t *dsa; | |
1896 | dmu_tx_t *tx; | |
1897 | ||
1898 | tx = dmu_tx_create(os); | |
1899 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1900 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1901 | dmu_tx_abort(tx); | |
2e528b49 MA |
1902 | /* Make zl_get_data do txg_waited_synced() */ |
1903 | return (SET_ERROR(EIO)); | |
428870ff BB |
1904 | } |
1905 | ||
1ce23dca PS |
1906 | /* |
1907 | * In order to prevent the zgd's lwb from being free'd prior to | |
1908 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1909 | * the lwb's "max txg" takes this tx's txg into account. | |
1910 | */ | |
1911 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1912 | ||
79c76d5b | 1913 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1914 | dsa->dsa_dr = NULL; |
1915 | dsa->dsa_done = done; | |
1916 | dsa->dsa_zgd = zgd; | |
1917 | dsa->dsa_tx = tx; | |
1918 | ||
02dc43bc MA |
1919 | /* |
1920 | * Since we are currently syncing this txg, it's nontrivial to | |
1921 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1922 | * | |
1923 | * When syncing, the db_blkptr is initially the BP of the previous | |
1924 | * txg. We can not nopwrite against it because it will be changed | |
1925 | * (this is similar to the non-late-arrival case where the dbuf is | |
1926 | * dirty in a future txg). | |
1927 | * | |
1928 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1929 | * We can not nopwrite against it because although the BP will not | |
1930 | * (typically) be changed, the data has not yet been persisted to this | |
1931 | * location. | |
1932 | * | |
1933 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1934 | * possible to always nopwrite, because the data that was written in | |
1935 | * this txg is the same data that we are trying to write. However we | |
1936 | * would need to check that this dbuf is not dirty in any future | |
1937 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1938 | * don't nopwrite in this case. | |
1939 | */ | |
1940 | zp->zp_nopwrite = B_FALSE; | |
1941 | ||
a6255b7f DQ |
1942 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1943 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1944 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1945 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1946 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1947 | |
1948 | return (0); | |
34dc7c2f BB |
1949 | } |
1950 | ||
1951 | /* | |
1952 | * Intent log support: sync the block associated with db to disk. | |
1953 | * N.B. and XXX: the caller is responsible for making sure that the | |
1954 | * data isn't changing while dmu_sync() is writing it. | |
1955 | * | |
1956 | * Return values: | |
1957 | * | |
03c6040b | 1958 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1959 | * The caller should not log the write. |
1960 | * | |
1961 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1962 | * The caller should not log the write. | |
1963 | * | |
1964 | * EALREADY: this block is already in the process of being synced. | |
1965 | * The caller should track its progress (somehow). | |
1966 | * | |
428870ff BB |
1967 | * EIO: could not do the I/O. |
1968 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1969 | * |
428870ff BB |
1970 | * 0: the I/O has been initiated. |
1971 | * The caller should log this blkptr in the done callback. | |
1972 | * It is possible that the I/O will fail, in which case | |
1973 | * the error will be reported to the done callback and | |
1974 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1975 | */ |
1976 | int | |
428870ff | 1977 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1978 | { |
428870ff BB |
1979 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1980 | objset_t *os = db->db_objset; | |
1981 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1982 | dbuf_dirty_record_t *dr; |
428870ff | 1983 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1984 | zbookmark_phys_t zb; |
428870ff | 1985 | zio_prop_t zp; |
572e2857 | 1986 | dnode_t *dn; |
34dc7c2f | 1987 | |
428870ff | 1988 | ASSERT(pio != NULL); |
34dc7c2f BB |
1989 | ASSERT(txg != 0); |
1990 | ||
428870ff BB |
1991 | SET_BOOKMARK(&zb, ds->ds_object, |
1992 | db->db.db_object, db->db_level, db->db_blkid); | |
1993 | ||
572e2857 BB |
1994 | DB_DNODE_ENTER(db); |
1995 | dn = DB_DNODE(db); | |
82644107 | 1996 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1997 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1998 | |
1999 | /* | |
428870ff | 2000 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 2001 | */ |
428870ff BB |
2002 | if (txg > spa_freeze_txg(os->os_spa)) |
2003 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
2004 | |
2005 | /* | |
428870ff BB |
2006 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
2007 | * and us. If we determine that this txg is not yet syncing, | |
2008 | * but it begins to sync a moment later, that's OK because the | |
2009 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 2010 | */ |
428870ff BB |
2011 | mutex_enter(&db->db_mtx); |
2012 | ||
2013 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 2014 | /* |
428870ff | 2015 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 2016 | */ |
428870ff | 2017 | mutex_exit(&db->db_mtx); |
2e528b49 | 2018 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
2019 | } |
2020 | ||
428870ff BB |
2021 | if (txg <= spa_syncing_txg(os->os_spa)) { |
2022 | /* | |
2023 | * This txg is currently syncing, so we can't mess with | |
2024 | * the dirty record anymore; just write a new log block. | |
2025 | */ | |
2026 | mutex_exit(&db->db_mtx); | |
2027 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
2028 | } |
2029 | ||
2030 | dr = db->db_last_dirty; | |
428870ff | 2031 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 2032 | dr = dr->dr_next; |
428870ff BB |
2033 | |
2034 | if (dr == NULL) { | |
34dc7c2f | 2035 | /* |
428870ff | 2036 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
2037 | * There's no need to log writes to freed blocks, so we're done. |
2038 | */ | |
2039 | mutex_exit(&db->db_mtx); | |
2e528b49 | 2040 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2041 | } |
2042 | ||
03c6040b GW |
2043 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); |
2044 | ||
02dc43bc MA |
2045 | if (db->db_blkptr != NULL) { |
2046 | /* | |
2047 | * We need to fill in zgd_bp with the current blkptr so that | |
2048 | * the nopwrite code can check if we're writing the same | |
2049 | * data that's already on disk. We can only nopwrite if we | |
2050 | * are sure that after making the copy, db_blkptr will not | |
2051 | * change until our i/o completes. We ensure this by | |
2052 | * holding the db_mtx, and only allowing nopwrite if the | |
2053 | * block is not already dirty (see below). This is verified | |
2054 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
2055 | * not changed. | |
2056 | */ | |
2057 | *zgd->zgd_bp = *db->db_blkptr; | |
2058 | } | |
2059 | ||
03c6040b | 2060 | /* |
f3c517d8 MA |
2061 | * Assume the on-disk data is X, the current syncing data (in |
2062 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
2063 | * in dmu_sync). | |
2064 | * | |
2065 | * We usually want to perform a nopwrite if X and Z are the | |
2066 | * same. However, if Y is different (i.e. the BP is going to | |
2067 | * change before this write takes effect), then a nopwrite will | |
2068 | * be incorrect - we would override with X, which could have | |
2069 | * been freed when Y was written. | |
2070 | * | |
2071 | * (Note that this is not a concern when we are nop-writing from | |
2072 | * syncing context, because X and Y must be identical, because | |
2073 | * all previous txgs have been synced.) | |
2074 | * | |
2075 | * Therefore, we disable nopwrite if the current BP could change | |
2076 | * before this TXG. There are two ways it could change: by | |
2077 | * being dirty (dr_next is non-NULL), or by being freed | |
2078 | * (dnode_block_freed()). This behavior is verified by | |
2079 | * zio_done(), which VERIFYs that the override BP is identical | |
2080 | * to the on-disk BP. | |
03c6040b | 2081 | */ |
f3c517d8 MA |
2082 | DB_DNODE_ENTER(db); |
2083 | dn = DB_DNODE(db); | |
2084 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
03c6040b | 2085 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 2086 | DB_DNODE_EXIT(db); |
03c6040b | 2087 | |
34dc7c2f | 2088 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
2089 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
2090 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 2091 | /* |
428870ff BB |
2092 | * We have already issued a sync write for this buffer, |
2093 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
2094 | * have been dirtied since, or we would have cleared the state. |
2095 | */ | |
34dc7c2f | 2096 | mutex_exit(&db->db_mtx); |
2e528b49 | 2097 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
2098 | } |
2099 | ||
428870ff | 2100 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 2101 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 2102 | mutex_exit(&db->db_mtx); |
34dc7c2f | 2103 | |
79c76d5b | 2104 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
2105 | dsa->dsa_dr = dr; |
2106 | dsa->dsa_done = done; | |
2107 | dsa->dsa_zgd = zgd; | |
2108 | dsa->dsa_tx = NULL; | |
b128c09f | 2109 | |
428870ff | 2110 | zio_nowait(arc_write(pio, os->os_spa, txg, |
02dc43bc | 2111 | zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 2112 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 2113 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 2114 | |
428870ff | 2115 | return (0); |
34dc7c2f BB |
2116 | } |
2117 | ||
b5256303 TC |
2118 | int |
2119 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
2120 | { | |
2121 | dnode_t *dn; | |
2122 | int err; | |
2123 | ||
2124 | err = dnode_hold(os, object, FTAG, &dn); | |
2125 | if (err) | |
2126 | return (err); | |
2127 | err = dnode_set_nlevels(dn, nlevels, tx); | |
2128 | dnode_rele(dn, FTAG); | |
2129 | return (err); | |
2130 | } | |
2131 | ||
34dc7c2f BB |
2132 | int |
2133 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 2134 | dmu_tx_t *tx) |
34dc7c2f BB |
2135 | { |
2136 | dnode_t *dn; | |
2137 | int err; | |
2138 | ||
428870ff | 2139 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2140 | if (err) |
2141 | return (err); | |
2142 | err = dnode_set_blksz(dn, size, ibs, tx); | |
2143 | dnode_rele(dn, FTAG); | |
2144 | return (err); | |
2145 | } | |
2146 | ||
ae76f45c TC |
2147 | int |
2148 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
2149 | dmu_tx_t *tx) | |
2150 | { | |
2151 | dnode_t *dn; | |
2152 | int err; | |
2153 | ||
2154 | err = dnode_hold(os, object, FTAG, &dn); | |
2155 | if (err) | |
2156 | return (err); | |
2157 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 2158 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
2159 | rw_exit(&dn->dn_struct_rwlock); |
2160 | dnode_rele(dn, FTAG); | |
2161 | return (0); | |
2162 | } | |
2163 | ||
34dc7c2f BB |
2164 | void |
2165 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 2166 | dmu_tx_t *tx) |
34dc7c2f BB |
2167 | { |
2168 | dnode_t *dn; | |
2169 | ||
9b67f605 MA |
2170 | /* |
2171 | * Send streams include each object's checksum function. This | |
2172 | * check ensures that the receiving system can understand the | |
2173 | * checksum function transmitted. | |
2174 | */ | |
2175 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
2176 | ||
2177 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
2178 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
2179 | dn->dn_checksum = checksum; |
2180 | dnode_setdirty(dn, tx); | |
2181 | dnode_rele(dn, FTAG); | |
2182 | } | |
2183 | ||
2184 | void | |
2185 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 2186 | dmu_tx_t *tx) |
34dc7c2f BB |
2187 | { |
2188 | dnode_t *dn; | |
2189 | ||
9b67f605 MA |
2190 | /* |
2191 | * Send streams include each object's compression function. This | |
2192 | * check ensures that the receiving system can understand the | |
2193 | * compression function transmitted. | |
2194 | */ | |
2195 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
2196 | ||
2197 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
2198 | dn->dn_compress = compress; |
2199 | dnode_setdirty(dn, tx); | |
2200 | dnode_rele(dn, FTAG); | |
2201 | } | |
2202 | ||
faf0f58c MA |
2203 | /* |
2204 | * When the "redundant_metadata" property is set to "most", only indirect | |
2205 | * blocks of this level and higher will have an additional ditto block. | |
2206 | */ | |
2207 | int zfs_redundant_metadata_most_ditto_level = 2; | |
2208 | ||
428870ff | 2209 | void |
82644107 | 2210 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
2211 | { |
2212 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 2213 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 2214 | (wp & WP_SPILL)); |
428870ff BB |
2215 | enum zio_checksum checksum = os->os_checksum; |
2216 | enum zio_compress compress = os->os_compress; | |
2217 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
03c6040b GW |
2218 | boolean_t dedup = B_FALSE; |
2219 | boolean_t nopwrite = B_FALSE; | |
428870ff | 2220 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 2221 | boolean_t encrypt = B_FALSE; |
428870ff | 2222 | int copies = os->os_copies; |
a7004725 | 2223 | |
428870ff | 2224 | /* |
03c6040b GW |
2225 | * We maintain different write policies for each of the following |
2226 | * types of data: | |
2227 | * 1. metadata | |
2228 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
2229 | * 3. all other level 0 blocks | |
428870ff BB |
2230 | */ |
2231 | if (ismd) { | |
b1d21733 TC |
2232 | /* |
2233 | * XXX -- we should design a compression algorithm | |
2234 | * that specializes in arrays of bps. | |
2235 | */ | |
2236 | compress = zio_compress_select(os->os_spa, | |
2237 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 2238 | |
428870ff BB |
2239 | /* |
2240 | * Metadata always gets checksummed. If the data | |
2241 | * checksum is multi-bit correctable, and it's not a | |
2242 | * ZBT-style checksum, then it's suitable for metadata | |
2243 | * as well. Otherwise, the metadata checksum defaults | |
2244 | * to fletcher4. | |
2245 | */ | |
3c67d83a TH |
2246 | if (!(zio_checksum_table[checksum].ci_flags & |
2247 | ZCHECKSUM_FLAG_METADATA) || | |
2248 | (zio_checksum_table[checksum].ci_flags & | |
2249 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 2250 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
2251 | |
2252 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
2253 | (os->os_redundant_metadata == | |
2254 | ZFS_REDUNDANT_METADATA_MOST && | |
2255 | (level >= zfs_redundant_metadata_most_ditto_level || | |
2256 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
2257 | copies++; | |
03c6040b GW |
2258 | } else if (wp & WP_NOFILL) { |
2259 | ASSERT(level == 0); | |
428870ff | 2260 | |
428870ff | 2261 | /* |
03c6040b GW |
2262 | * If we're writing preallocated blocks, we aren't actually |
2263 | * writing them so don't set any policy properties. These | |
2264 | * blocks are currently only used by an external subsystem | |
2265 | * outside of zfs (i.e. dump) and not written by the zio | |
2266 | * pipeline. | |
428870ff | 2267 | */ |
03c6040b GW |
2268 | compress = ZIO_COMPRESS_OFF; |
2269 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2270 | } else { |
99197f03 JG |
2271 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2272 | compress); | |
428870ff | 2273 | |
03c6040b GW |
2274 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2275 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2276 | dedup_checksum; | |
428870ff | 2277 | |
03c6040b GW |
2278 | /* |
2279 | * Determine dedup setting. If we are in dmu_sync(), | |
2280 | * we won't actually dedup now because that's all | |
2281 | * done in syncing context; but we do want to use the | |
2282 | * dedup checkum. If the checksum is not strong | |
2283 | * enough to ensure unique signatures, force | |
2284 | * dedup_verify. | |
2285 | */ | |
2286 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2287 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2288 | if (!(zio_checksum_table[checksum].ci_flags & |
2289 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2290 | dedup_verify = B_TRUE; |
2291 | } | |
428870ff | 2292 | |
03c6040b | 2293 | /* |
3c67d83a TH |
2294 | * Enable nopwrite if we have secure enough checksum |
2295 | * algorithm (see comment in zio_nop_write) and | |
2296 | * compression is enabled. We don't enable nopwrite if | |
2297 | * dedup is enabled as the two features are mutually | |
2298 | * exclusive. | |
03c6040b | 2299 | */ |
3c67d83a TH |
2300 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2301 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2302 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2303 | } |
2304 | ||
b5256303 TC |
2305 | /* |
2306 | * All objects in an encrypted objset are protected from modification | |
2307 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2308 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2309 | * not subject to nopwrite since writing the same data will still | |
2310 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2311 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2312 | * object types. | |
2313 | */ | |
2314 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2315 | encrypt = B_TRUE; | |
2316 | ||
2317 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2318 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2319 | nopwrite = B_FALSE; | |
2320 | } else { | |
2321 | dedup = B_FALSE; | |
2322 | } | |
2323 | ||
ae76f45c TC |
2324 | if (level <= 0 && |
2325 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2326 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2327 | } |
b5256303 | 2328 | } |
2aa34383 | 2329 | |
b5256303 TC |
2330 | zp->zp_compress = compress; |
2331 | zp->zp_checksum = checksum; | |
428870ff BB |
2332 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2333 | zp->zp_level = level; | |
faf0f58c | 2334 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2335 | zp->zp_dedup = dedup; |
2336 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2337 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2338 | zp->zp_encrypt = encrypt; |
2339 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
2340 | bzero(zp->zp_salt, ZIO_DATA_SALT_LEN); | |
2341 | bzero(zp->zp_iv, ZIO_DATA_IV_LEN); | |
2342 | bzero(zp->zp_mac, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2343 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2344 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2345 | |
2346 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2347 | } |
2348 | ||
66aca247 DB |
2349 | /* |
2350 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2351 | * in order to determine the location of holes. In order to accurately | |
2352 | * report holes all dirty data must be synced to disk. This causes extremely | |
2353 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2354 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2355 | * report the dnode as having no holes which is always a safe thing to do. | |
2356 | */ | |
34dc7c2f BB |
2357 | int |
2358 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2359 | { | |
2360 | dnode_t *dn; | |
2361 | int i, err; | |
66aca247 | 2362 | boolean_t clean = B_TRUE; |
34dc7c2f | 2363 | |
428870ff | 2364 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2365 | if (err) |
2366 | return (err); | |
66aca247 | 2367 | |
34dc7c2f | 2368 | /* |
ec4f9b8f BB |
2369 | * Check if there are dirty data blocks or frees which have not been |
2370 | * synced. Dirty spill and bonus blocks which are external to the | |
2371 | * object can ignored when reporting holes. | |
34dc7c2f | 2372 | */ |
ec4f9b8f | 2373 | mutex_enter(&dn->dn_mtx); |
454365bb | 2374 | for (i = 0; i < TXG_SIZE; i++) { |
edc1e713 | 2375 | if (multilist_link_active(&dn->dn_dirty_link[i])) { |
ec4f9b8f BB |
2376 | |
2377 | if (dn->dn_free_ranges[i] != NULL) { | |
2378 | clean = B_FALSE; | |
2379 | break; | |
2380 | } | |
2381 | ||
2382 | list_t *list = &dn->dn_dirty_records[i]; | |
2383 | dbuf_dirty_record_t *dr; | |
2384 | ||
2385 | for (dr = list_head(list); dr != NULL; | |
2386 | dr = list_next(list, dr)) { | |
2387 | dmu_buf_impl_t *db = dr->dr_dbuf; | |
2388 | ||
2389 | if (db->db_blkid == DMU_SPILL_BLKID || | |
2390 | db->db_blkid == DMU_BONUS_BLKID) | |
2391 | continue; | |
2392 | ||
2393 | clean = B_FALSE; | |
2394 | break; | |
2395 | } | |
66aca247 | 2396 | } |
ec4f9b8f BB |
2397 | |
2398 | if (clean == B_FALSE) | |
2399 | break; | |
34dc7c2f | 2400 | } |
ec4f9b8f | 2401 | mutex_exit(&dn->dn_mtx); |
66aca247 DB |
2402 | |
2403 | /* | |
2404 | * If compatibility option is on, sync any current changes before | |
2405 | * we go trundling through the block pointers. | |
2406 | */ | |
2407 | if (!clean && zfs_dmu_offset_next_sync) { | |
2408 | clean = B_TRUE; | |
34dc7c2f BB |
2409 | dnode_rele(dn, FTAG); |
2410 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 2411 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2412 | if (err) |
2413 | return (err); | |
2414 | } | |
2415 | ||
66aca247 DB |
2416 | if (clean) |
2417 | err = dnode_next_offset(dn, | |
2418 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2419 | else | |
2420 | err = SET_ERROR(EBUSY); | |
2421 | ||
34dc7c2f BB |
2422 | dnode_rele(dn, FTAG); |
2423 | ||
2424 | return (err); | |
2425 | } | |
2426 | ||
2427 | void | |
e0b0ca98 | 2428 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2429 | { |
e0b0ca98 | 2430 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2431 | |
34dc7c2f BB |
2432 | doi->doi_data_block_size = dn->dn_datablksz; |
2433 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2434 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2435 | doi->doi_type = dn->dn_type; |
2436 | doi->doi_bonus_type = dn->dn_bonustype; | |
2437 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2438 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2439 | doi->doi_indirection = dn->dn_nlevels; |
2440 | doi->doi_checksum = dn->dn_checksum; | |
2441 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2442 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2443 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2444 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2445 | doi->doi_fill_count = 0; |
1c27024e | 2446 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2447 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2448 | } |
2449 | ||
2450 | void | |
2451 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2452 | { | |
2453 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2454 | mutex_enter(&dn->dn_mtx); | |
2455 | ||
2456 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2457 | |
2458 | mutex_exit(&dn->dn_mtx); | |
2459 | rw_exit(&dn->dn_struct_rwlock); | |
2460 | } | |
2461 | ||
2462 | /* | |
2463 | * Get information on a DMU object. | |
2464 | * If doi is NULL, just indicates whether the object exists. | |
2465 | */ | |
2466 | int | |
2467 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2468 | { | |
2469 | dnode_t *dn; | |
428870ff | 2470 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2471 | |
2472 | if (err) | |
2473 | return (err); | |
2474 | ||
2475 | if (doi != NULL) | |
2476 | dmu_object_info_from_dnode(dn, doi); | |
2477 | ||
2478 | dnode_rele(dn, FTAG); | |
2479 | return (0); | |
2480 | } | |
2481 | ||
2482 | /* | |
2483 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2484 | */ | |
2485 | void | |
572e2857 | 2486 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2487 | { |
572e2857 BB |
2488 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2489 | ||
2490 | DB_DNODE_ENTER(db); | |
2491 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2492 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2493 | } |
2494 | ||
2495 | /* | |
2496 | * Faster still when you only care about the size. | |
2497 | * This is specifically optimized for zfs_getattr(). | |
2498 | */ | |
2499 | void | |
572e2857 BB |
2500 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2501 | u_longlong_t *nblk512) | |
34dc7c2f | 2502 | { |
572e2857 BB |
2503 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2504 | dnode_t *dn; | |
2505 | ||
2506 | DB_DNODE_ENTER(db); | |
2507 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2508 | |
2509 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2510 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2511 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2512 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2513 | DB_DNODE_EXIT(db); | |
2514 | } | |
2515 | ||
2516 | void | |
2517 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2518 | { | |
2519 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2520 | dnode_t *dn; | |
2521 | ||
2522 | DB_DNODE_ENTER(db); | |
2523 | dn = DB_DNODE(db); | |
2524 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2525 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2526 | } |
2527 | ||
2528 | void | |
2529 | byteswap_uint64_array(void *vbuf, size_t size) | |
2530 | { | |
2531 | uint64_t *buf = vbuf; | |
2532 | size_t count = size >> 3; | |
2533 | int i; | |
2534 | ||
2535 | ASSERT((size & 7) == 0); | |
2536 | ||
2537 | for (i = 0; i < count; i++) | |
2538 | buf[i] = BSWAP_64(buf[i]); | |
2539 | } | |
2540 | ||
2541 | void | |
2542 | byteswap_uint32_array(void *vbuf, size_t size) | |
2543 | { | |
2544 | uint32_t *buf = vbuf; | |
2545 | size_t count = size >> 2; | |
2546 | int i; | |
2547 | ||
2548 | ASSERT((size & 3) == 0); | |
2549 | ||
2550 | for (i = 0; i < count; i++) | |
2551 | buf[i] = BSWAP_32(buf[i]); | |
2552 | } | |
2553 | ||
2554 | void | |
2555 | byteswap_uint16_array(void *vbuf, size_t size) | |
2556 | { | |
2557 | uint16_t *buf = vbuf; | |
2558 | size_t count = size >> 1; | |
2559 | int i; | |
2560 | ||
2561 | ASSERT((size & 1) == 0); | |
2562 | ||
2563 | for (i = 0; i < count; i++) | |
2564 | buf[i] = BSWAP_16(buf[i]); | |
2565 | } | |
2566 | ||
2567 | /* ARGSUSED */ | |
2568 | void | |
2569 | byteswap_uint8_array(void *vbuf, size_t size) | |
2570 | { | |
2571 | } | |
2572 | ||
2573 | void | |
2574 | dmu_init(void) | |
2575 | { | |
a6255b7f | 2576 | abd_init(); |
428870ff | 2577 | zfs_dbgmsg_init(); |
572e2857 BB |
2578 | sa_cache_init(); |
2579 | xuio_stat_init(); | |
2580 | dmu_objset_init(); | |
34dc7c2f | 2581 | dnode_init(); |
428870ff | 2582 | zfetch_init(); |
570827e1 | 2583 | dmu_tx_init(); |
34dc7c2f | 2584 | l2arc_init(); |
29809a6c | 2585 | arc_init(); |
d3c2ae1c | 2586 | dbuf_init(); |
34dc7c2f BB |
2587 | } |
2588 | ||
2589 | void | |
2590 | dmu_fini(void) | |
2591 | { | |
e49f1e20 | 2592 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2593 | l2arc_fini(); |
570827e1 | 2594 | dmu_tx_fini(); |
428870ff | 2595 | zfetch_fini(); |
34dc7c2f | 2596 | dbuf_fini(); |
572e2857 BB |
2597 | dnode_fini(); |
2598 | dmu_objset_fini(); | |
428870ff BB |
2599 | xuio_stat_fini(); |
2600 | sa_cache_fini(); | |
2601 | zfs_dbgmsg_fini(); | |
a6255b7f | 2602 | abd_fini(); |
34dc7c2f | 2603 | } |
c28b2279 | 2604 | |
93ce2b4c | 2605 | #if defined(_KERNEL) |
c28b2279 | 2606 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2607 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2608 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2609 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2610 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2611 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2612 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2613 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2614 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2615 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2616 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2617 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2618 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2619 | EXPORT_SYMBOL(dmu_object_info); |
2620 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2621 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2622 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2623 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2624 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2625 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2626 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2627 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2628 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
2629 | EXPORT_SYMBOL(dmu_write_policy); |
2630 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2631 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2632 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2633 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2634 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2635 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2636 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2637 | |
bef78122 | 2638 | /* BEGIN CSTYLED */ |
03c6040b GW |
2639 | module_param(zfs_nopwrite_enabled, int, 0644); |
2640 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2641 | ||
bef78122 DQ |
2642 | module_param(zfs_per_txg_dirty_frees_percent, ulong, 0644); |
2643 | MODULE_PARM_DESC(zfs_per_txg_dirty_frees_percent, | |
2644 | "percentage of dirtied blocks from frees in one TXG"); | |
66aca247 DB |
2645 | |
2646 | module_param(zfs_dmu_offset_next_sync, int, 0644); | |
2647 | MODULE_PARM_DESC(zfs_dmu_offset_next_sync, | |
2648 | "Enable forcing txg sync to find holes"); | |
2649 | ||
bef78122 | 2650 | /* END CSTYLED */ |
66aca247 | 2651 | |
c28b2279 | 2652 | #endif |