]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
34dc7c2f BB |
10 | * See the License for the specific language governing permissions |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
196bee4c | 23 | * Copyright (c) 2011, 2020 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
bc77ba73 | 25 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
a08abc1b | 26 | * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved. |
5475aada | 27 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. |
65282ee9 | 28 | * Copyright (c) 2019 Datto Inc. |
10b3c7f5 MN |
29 | * Copyright (c) 2019, Klara Inc. |
30 | * Copyright (c) 2019, Allan Jude | |
5405be03 | 31 | * Copyright (c) 2022 Hewlett Packard Enterprise Development LP. |
67a1b037 | 32 | * Copyright (c) 2021, 2022 by Pawel Jakub Dawidek |
34dc7c2f BB |
33 | */ |
34 | ||
34dc7c2f BB |
35 | #include <sys/dmu.h> |
36 | #include <sys/dmu_impl.h> | |
37 | #include <sys/dmu_tx.h> | |
38 | #include <sys/dbuf.h> | |
39 | #include <sys/dnode.h> | |
40 | #include <sys/zfs_context.h> | |
41 | #include <sys/dmu_objset.h> | |
42 | #include <sys/dmu_traverse.h> | |
43 | #include <sys/dsl_dataset.h> | |
44 | #include <sys/dsl_dir.h> | |
45 | #include <sys/dsl_pool.h> | |
46 | #include <sys/dsl_synctask.h> | |
47 | #include <sys/dsl_prop.h> | |
48 | #include <sys/dmu_zfetch.h> | |
49 | #include <sys/zfs_ioctl.h> | |
50 | #include <sys/zap.h> | |
51 | #include <sys/zio_checksum.h> | |
03c6040b | 52 | #include <sys/zio_compress.h> |
428870ff | 53 | #include <sys/sa.h> |
62bdd5eb | 54 | #include <sys/zfeature.h> |
a6255b7f | 55 | #include <sys/abd.h> |
67a1b037 | 56 | #include <sys/brt.h> |
e5d1c27e | 57 | #include <sys/trace_zfs.h> |
64e0fe14 | 58 | #include <sys/zfs_racct.h> |
f763c3d1 | 59 | #include <sys/zfs_rlock.h> |
34dc7c2f BB |
60 | #ifdef _KERNEL |
61 | #include <sys/vmsystm.h> | |
b128c09f | 62 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
63 | #endif |
64 | ||
03c6040b GW |
65 | /* |
66 | * Enable/disable nopwrite feature. | |
67 | */ | |
18168da7 | 68 | static int zfs_nopwrite_enabled = 1; |
03c6040b | 69 | |
539d33c7 | 70 | /* |
65282ee9 AP |
71 | * Tunable to control percentage of dirtied L1 blocks from frees allowed into |
72 | * one TXG. After this threshold is crossed, additional dirty blocks from frees | |
73 | * will wait until the next TXG. | |
539d33c7 GM |
74 | * A value of zero will disable this throttle. |
75 | */ | |
ab8d9c17 | 76 | static uint_t zfs_per_txg_dirty_frees_percent = 30; |
539d33c7 | 77 | |
66aca247 | 78 | /* |
05b3eb6d BB |
79 | * Enable/disable forcing txg sync when dirty checking for holes with lseek(). |
80 | * By default this is enabled to ensure accurate hole reporting, it can result | |
81 | * in a significant performance penalty for lseek(SEEK_HOLE) heavy workloads. | |
82 | * Disabling this option will result in holes never being reported in dirty | |
83 | * files which is always safe. | |
66aca247 | 84 | */ |
18168da7 | 85 | static int zfs_dmu_offset_next_sync = 1; |
66aca247 | 86 | |
d9b4bf06 MA |
87 | /* |
88 | * Limit the amount we can prefetch with one call to this amount. This | |
89 | * helps to limit the amount of memory that can be used by prefetching. | |
90 | * Larger objects should be prefetched a bit at a time. | |
91 | */ | |
34b3d498 AM |
92 | #ifdef _ILP32 |
93 | uint_t dmu_prefetch_max = 8 * 1024 * 1024; | |
94 | #else | |
fdc2d303 | 95 | uint_t dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE; |
34b3d498 | 96 | #endif |
d9b4bf06 | 97 | |
34dc7c2f | 98 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
2e5dc449 MA |
99 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" }, |
100 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" }, | |
101 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" }, | |
102 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" }, | |
103 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" }, | |
104 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" }, | |
105 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" }, | |
106 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" }, | |
107 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" }, | |
108 | {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" }, | |
109 | {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" }, | |
110 | {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" }, | |
111 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" }, | |
112 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"}, | |
113 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" }, | |
114 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" }, | |
115 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" }, | |
116 | {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" }, | |
117 | {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" }, | |
118 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" }, | |
119 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" }, | |
120 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" }, | |
121 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" }, | |
122 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" }, | |
123 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" }, | |
124 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" }, | |
125 | {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" }, | |
126 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" }, | |
127 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" }, | |
128 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" }, | |
129 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" }, | |
130 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" }, | |
131 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" }, | |
132 | {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" }, | |
133 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" }, | |
134 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" }, | |
135 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" }, | |
136 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"}, | |
137 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" }, | |
138 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" }, | |
139 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"}, | |
140 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"}, | |
141 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" }, | |
142 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" }, | |
143 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" }, | |
144 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" }, | |
145 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" }, | |
146 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" }, | |
147 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" }, | |
148 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" }, | |
149 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" }, | |
150 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" }, | |
151 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" }, | |
152 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
153 | }; |
154 | ||
a2d5643f | 155 | dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { |
9ae529ec CS |
156 | { byteswap_uint8_array, "uint8" }, |
157 | { byteswap_uint16_array, "uint16" }, | |
158 | { byteswap_uint32_array, "uint32" }, | |
159 | { byteswap_uint64_array, "uint64" }, | |
160 | { zap_byteswap, "zap" }, | |
161 | { dnode_buf_byteswap, "dnode" }, | |
162 | { dmu_objset_byteswap, "objset" }, | |
163 | { zfs_znode_byteswap, "znode" }, | |
164 | { zfs_oldacl_byteswap, "oldacl" }, | |
165 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
166 | }; |
167 | ||
65c7cc49 | 168 | static int |
2bce8049 | 169 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, |
a926aab9 | 170 | const void *tag, dmu_buf_t **dbp) |
2bce8049 MA |
171 | { |
172 | uint64_t blkid; | |
173 | dmu_buf_impl_t *db; | |
174 | ||
2bce8049 | 175 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 176 | blkid = dbuf_whichblock(dn, 0, offset); |
2bce8049 MA |
177 | db = dbuf_hold(dn, blkid, tag); |
178 | rw_exit(&dn->dn_struct_rwlock); | |
179 | ||
180 | if (db == NULL) { | |
181 | *dbp = NULL; | |
182 | return (SET_ERROR(EIO)); | |
183 | } | |
184 | ||
185 | *dbp = &db->db; | |
186 | return (0); | |
187 | } | |
34dc7c2f | 188 | int |
9b67f605 | 189 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
a926aab9 | 190 | const void *tag, dmu_buf_t **dbp) |
34dc7c2f BB |
191 | { |
192 | dnode_t *dn; | |
193 | uint64_t blkid; | |
194 | dmu_buf_impl_t *db; | |
195 | int err; | |
428870ff BB |
196 | |
197 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
198 | if (err) |
199 | return (err); | |
34dc7c2f | 200 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 201 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
202 | db = dbuf_hold(dn, blkid, tag); |
203 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
204 | dnode_rele(dn, FTAG); |
205 | ||
34dc7c2f | 206 | if (db == NULL) { |
9b67f605 MA |
207 | *dbp = NULL; |
208 | return (SET_ERROR(EIO)); | |
209 | } | |
210 | ||
211 | *dbp = &db->db; | |
212 | return (err); | |
213 | } | |
214 | ||
2bce8049 MA |
215 | int |
216 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
a926aab9 | 217 | const void *tag, dmu_buf_t **dbp, int flags) |
2bce8049 MA |
218 | { |
219 | int err; | |
220 | int db_flags = DB_RF_CANFAIL; | |
221 | ||
222 | if (flags & DMU_READ_NO_PREFETCH) | |
223 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
224 | if (flags & DMU_READ_NO_DECRYPT) |
225 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
226 | |
227 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
228 | if (err == 0) { | |
229 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
230 | err = dbuf_read(db, NULL, db_flags); | |
231 | if (err != 0) { | |
232 | dbuf_rele(db, tag); | |
233 | *dbp = NULL; | |
234 | } | |
235 | } | |
236 | ||
237 | return (err); | |
238 | } | |
239 | ||
9b67f605 MA |
240 | int |
241 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
a926aab9 | 242 | const void *tag, dmu_buf_t **dbp, int flags) |
9b67f605 MA |
243 | { |
244 | int err; | |
245 | int db_flags = DB_RF_CANFAIL; | |
246 | ||
247 | if (flags & DMU_READ_NO_PREFETCH) | |
248 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
249 | if (flags & DMU_READ_NO_DECRYPT) |
250 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
251 | |
252 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
253 | if (err == 0) { | |
254 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 255 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 256 | if (err != 0) { |
34dc7c2f | 257 | dbuf_rele(db, tag); |
9b67f605 | 258 | *dbp = NULL; |
34dc7c2f BB |
259 | } |
260 | } | |
261 | ||
34dc7c2f BB |
262 | return (err); |
263 | } | |
264 | ||
265 | int | |
266 | dmu_bonus_max(void) | |
267 | { | |
50c957f7 | 268 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
269 | } |
270 | ||
271 | int | |
572e2857 | 272 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 273 | { |
572e2857 BB |
274 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
275 | dnode_t *dn; | |
276 | int error; | |
34dc7c2f | 277 | |
572e2857 BB |
278 | DB_DNODE_ENTER(db); |
279 | dn = DB_DNODE(db); | |
280 | ||
281 | if (dn->dn_bonus != db) { | |
2e528b49 | 282 | error = SET_ERROR(EINVAL); |
572e2857 | 283 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 284 | error = SET_ERROR(EINVAL); |
572e2857 BB |
285 | } else { |
286 | dnode_setbonuslen(dn, newsize, tx); | |
287 | error = 0; | |
288 | } | |
289 | ||
290 | DB_DNODE_EXIT(db); | |
291 | return (error); | |
34dc7c2f BB |
292 | } |
293 | ||
428870ff | 294 | int |
572e2857 | 295 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 296 | { |
572e2857 BB |
297 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
298 | dnode_t *dn; | |
299 | int error; | |
428870ff | 300 | |
572e2857 BB |
301 | DB_DNODE_ENTER(db); |
302 | dn = DB_DNODE(db); | |
428870ff | 303 | |
9ae529ec | 304 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 305 | error = SET_ERROR(EINVAL); |
572e2857 | 306 | } else if (dn->dn_bonus != db) { |
2e528b49 | 307 | error = SET_ERROR(EINVAL); |
572e2857 BB |
308 | } else { |
309 | dnode_setbonus_type(dn, type, tx); | |
310 | error = 0; | |
311 | } | |
428870ff | 312 | |
572e2857 BB |
313 | DB_DNODE_EXIT(db); |
314 | return (error); | |
315 | } | |
316 | ||
317 | dmu_object_type_t | |
318 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
319 | { | |
320 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
321 | dnode_t *dn; | |
322 | dmu_object_type_t type; | |
323 | ||
324 | DB_DNODE_ENTER(db); | |
325 | dn = DB_DNODE(db); | |
326 | type = dn->dn_bonustype; | |
327 | DB_DNODE_EXIT(db); | |
328 | ||
329 | return (type); | |
428870ff BB |
330 | } |
331 | ||
332 | int | |
333 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
334 | { | |
335 | dnode_t *dn; | |
336 | int error; | |
337 | ||
338 | error = dnode_hold(os, object, FTAG, &dn); | |
339 | dbuf_rm_spill(dn, tx); | |
340 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
341 | dnode_rm_spill(dn, tx); | |
342 | rw_exit(&dn->dn_struct_rwlock); | |
343 | dnode_rele(dn, FTAG); | |
344 | return (error); | |
345 | } | |
346 | ||
34dc7c2f | 347 | /* |
6955b401 BB |
348 | * Lookup and hold the bonus buffer for the provided dnode. If the dnode |
349 | * has not yet been allocated a new bonus dbuf a will be allocated. | |
350 | * Returns ENOENT, EIO, or 0. | |
34dc7c2f | 351 | */ |
a926aab9 | 352 | int dmu_bonus_hold_by_dnode(dnode_t *dn, const void *tag, dmu_buf_t **dbp, |
6955b401 | 353 | uint32_t flags) |
34dc7c2f | 354 | { |
34dc7c2f BB |
355 | dmu_buf_impl_t *db; |
356 | int error; | |
b5256303 TC |
357 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
358 | ||
359 | if (flags & DMU_READ_NO_PREFETCH) | |
360 | db_flags |= DB_RF_NOPREFETCH; | |
361 | if (flags & DMU_READ_NO_DECRYPT) | |
362 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 363 | |
34dc7c2f BB |
364 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
365 | if (dn->dn_bonus == NULL) { | |
c9e39da9 RY |
366 | if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { |
367 | rw_exit(&dn->dn_struct_rwlock); | |
368 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369 | } | |
34dc7c2f BB |
370 | if (dn->dn_bonus == NULL) |
371 | dbuf_create_bonus(dn); | |
372 | } | |
373 | db = dn->dn_bonus; | |
34dc7c2f BB |
374 | |
375 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 376 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 377 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 378 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
379 | } |
380 | ||
381 | /* | |
382 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
383 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
384 | * a dnode hold for every dbuf. | |
385 | */ | |
386 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f | 387 | |
b5256303 TC |
388 | error = dbuf_read(db, NULL, db_flags); |
389 | if (error) { | |
390 | dnode_evict_bonus(dn); | |
391 | dbuf_rele(db, tag); | |
392 | *dbp = NULL; | |
393 | return (error); | |
394 | } | |
34dc7c2f BB |
395 | |
396 | *dbp = &db->db; | |
397 | return (0); | |
398 | } | |
399 | ||
b5256303 | 400 | int |
a926aab9 | 401 | dmu_bonus_hold(objset_t *os, uint64_t object, const void *tag, dmu_buf_t **dbp) |
b5256303 | 402 | { |
6955b401 BB |
403 | dnode_t *dn; |
404 | int error; | |
405 | ||
406 | error = dnode_hold(os, object, FTAG, &dn); | |
407 | if (error) | |
408 | return (error); | |
409 | ||
410 | error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH); | |
411 | dnode_rele(dn, FTAG); | |
412 | ||
413 | return (error); | |
b5256303 TC |
414 | } |
415 | ||
428870ff BB |
416 | /* |
417 | * returns ENOENT, EIO, or 0. | |
418 | * | |
419 | * This interface will allocate a blank spill dbuf when a spill blk | |
420 | * doesn't already exist on the dnode. | |
421 | * | |
422 | * if you only want to find an already existing spill db, then | |
423 | * dmu_spill_hold_existing() should be used. | |
424 | */ | |
425 | int | |
a926aab9 AZ |
426 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, const void *tag, |
427 | dmu_buf_t **dbp) | |
428870ff BB |
428 | { |
429 | dmu_buf_impl_t *db = NULL; | |
430 | int err; | |
431 | ||
432 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
433 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
434 | ||
435 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
436 | ||
437 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
438 | rw_exit(&dn->dn_struct_rwlock); | |
439 | ||
b182ac00 | 440 | if (db == NULL) { |
441 | *dbp = NULL; | |
442 | return (SET_ERROR(EIO)); | |
443 | } | |
572e2857 BB |
444 | err = dbuf_read(db, NULL, flags); |
445 | if (err == 0) | |
446 | *dbp = &db->db; | |
b182ac00 | 447 | else { |
572e2857 | 448 | dbuf_rele(db, tag); |
b182ac00 | 449 | *dbp = NULL; |
450 | } | |
428870ff BB |
451 | return (err); |
452 | } | |
453 | ||
454 | int | |
a926aab9 | 455 | dmu_spill_hold_existing(dmu_buf_t *bonus, const void *tag, dmu_buf_t **dbp) |
428870ff | 456 | { |
572e2857 BB |
457 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
458 | dnode_t *dn; | |
428870ff BB |
459 | int err; |
460 | ||
572e2857 BB |
461 | DB_DNODE_ENTER(db); |
462 | dn = DB_DNODE(db); | |
463 | ||
464 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 465 | err = SET_ERROR(EINVAL); |
572e2857 BB |
466 | } else { |
467 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
468 | ||
469 | if (!dn->dn_have_spill) { | |
2e528b49 | 470 | err = SET_ERROR(ENOENT); |
572e2857 BB |
471 | } else { |
472 | err = dmu_spill_hold_by_dnode(dn, | |
473 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
474 | } | |
428870ff | 475 | |
428870ff | 476 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 477 | } |
572e2857 BB |
478 | |
479 | DB_DNODE_EXIT(db); | |
428870ff BB |
480 | return (err); |
481 | } | |
482 | ||
483 | int | |
a926aab9 | 484 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, const void *tag, |
e7504d7a | 485 | dmu_buf_t **dbp) |
428870ff | 486 | { |
572e2857 BB |
487 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
488 | dnode_t *dn; | |
489 | int err; | |
e7504d7a TC |
490 | uint32_t db_flags = DB_RF_CANFAIL; |
491 | ||
492 | if (flags & DMU_READ_NO_DECRYPT) | |
493 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
494 | |
495 | DB_DNODE_ENTER(db); | |
496 | dn = DB_DNODE(db); | |
e7504d7a | 497 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
498 | DB_DNODE_EXIT(db); |
499 | ||
500 | return (err); | |
428870ff BB |
501 | } |
502 | ||
34dc7c2f BB |
503 | /* |
504 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
505 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
506 | * and can induce severe lock contention when writing to several files | |
507 | * whose dnodes are in the same block. | |
508 | */ | |
af1698f5 | 509 | int |
9babb374 | 510 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
a926aab9 AZ |
511 | boolean_t read, const void *tag, int *numbufsp, dmu_buf_t ***dbpp, |
512 | uint32_t flags) | |
34dc7c2f BB |
513 | { |
514 | dmu_buf_t **dbp; | |
891568c9 | 515 | zstream_t *zs = NULL; |
34dc7c2f | 516 | uint64_t blkid, nblks, i; |
9babb374 | 517 | uint32_t dbuf_flags; |
34dc7c2f | 518 | int err; |
ec50cd24 | 519 | zio_t *zio = NULL; |
891568c9 | 520 | boolean_t missed = B_FALSE; |
34dc7c2f | 521 | |
67a1b037 | 522 | ASSERT(!read || length <= DMU_MAX_ACCESS); |
34dc7c2f | 523 | |
7f60329a MA |
524 | /* |
525 | * Note: We directly notify the prefetch code of this read, so that | |
526 | * we can tell it about the multi-block read. dbuf_read() only knows | |
527 | * about the one block it is accessing. | |
528 | */ | |
529 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
530 | DB_RF_NOPREFETCH; | |
34dc7c2f | 531 | |
e8cf3a4f AP |
532 | if ((flags & DMU_READ_NO_DECRYPT) != 0) |
533 | dbuf_flags |= DB_RF_NO_DECRYPT; | |
534 | ||
34dc7c2f BB |
535 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
536 | if (dn->dn_datablkshift) { | |
537 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
538 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
539 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
540 | } else { |
541 | if (offset + length > dn->dn_datablksz) { | |
542 | zfs_panic_recover("zfs: accessing past end of object " | |
543 | "%llx/%llx (size=%u access=%llu+%llu)", | |
544 | (longlong_t)dn->dn_objset-> | |
545 | os_dsl_dataset->ds_object, | |
546 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
547 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 548 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 549 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
550 | } |
551 | nblks = 1; | |
552 | } | |
79c76d5b | 553 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 554 | |
ec50cd24 FY |
555 | if (read) |
556 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, | |
557 | ZIO_FLAG_CANFAIL); | |
fcff0f35 | 558 | blkid = dbuf_whichblock(dn, 0, offset); |
34b3d498 | 559 | if ((flags & DMU_READ_NO_PREFETCH) == 0) { |
891568c9 AM |
560 | /* |
561 | * Prepare the zfetch before initiating the demand reads, so | |
562 | * that if multiple threads block on same indirect block, we | |
563 | * base predictions on the original less racy request order. | |
564 | */ | |
ed2f7ba0 AM |
565 | zs = dmu_zfetch_prepare(&dn->dn_zfetch, blkid, nblks, read, |
566 | B_TRUE); | |
891568c9 | 567 | } |
34dc7c2f | 568 | for (i = 0; i < nblks; i++) { |
7f60329a | 569 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f | 570 | if (db == NULL) { |
891568c9 AM |
571 | if (zs) |
572 | dmu_zfetch_run(zs, missed, B_TRUE); | |
34dc7c2f BB |
573 | rw_exit(&dn->dn_struct_rwlock); |
574 | dmu_buf_rele_array(dbp, nblks, tag); | |
ec50cd24 FY |
575 | if (read) |
576 | zio_nowait(zio); | |
2e528b49 | 577 | return (SET_ERROR(EIO)); |
34dc7c2f | 578 | } |
7f60329a | 579 | |
891568c9 AM |
580 | /* |
581 | * Initiate async demand data read. | |
582 | * We check the db_state after calling dbuf_read() because | |
583 | * (1) dbuf_read() may change the state to CACHED due to a | |
584 | * hit in the ARC, and (2) on a cache miss, a child will | |
585 | * have been added to "zio" but not yet completed, so the | |
586 | * state will not yet be CACHED. | |
587 | */ | |
588 | if (read) { | |
ed2f7ba0 AM |
589 | if (i == nblks - 1 && blkid + i < dn->dn_maxblkid && |
590 | offset + length < db->db.db_offset + | |
591 | db->db.db_size) { | |
592 | if (offset <= db->db.db_offset) | |
593 | dbuf_flags |= DB_RF_PARTIAL_FIRST; | |
594 | else | |
595 | dbuf_flags |= DB_RF_PARTIAL_MORE; | |
596 | } | |
9babb374 | 597 | (void) dbuf_read(db, zio, dbuf_flags); |
891568c9 AM |
598 | if (db->db_state != DB_CACHED) |
599 | missed = B_TRUE; | |
600 | } | |
34dc7c2f BB |
601 | dbp[i] = &db->db; |
602 | } | |
7f60329a | 603 | |
64e0fe14 RM |
604 | if (!read) |
605 | zfs_racct_write(length, nblks); | |
606 | ||
891568c9 AM |
607 | if (zs) |
608 | dmu_zfetch_run(zs, missed, B_TRUE); | |
34dc7c2f BB |
609 | rw_exit(&dn->dn_struct_rwlock); |
610 | ||
34dc7c2f | 611 | if (read) { |
ec50cd24 FY |
612 | /* wait for async read i/o */ |
613 | err = zio_wait(zio); | |
614 | if (err) { | |
615 | dmu_buf_rele_array(dbp, nblks, tag); | |
616 | return (err); | |
617 | } | |
618 | ||
619 | /* wait for other io to complete */ | |
34dc7c2f BB |
620 | for (i = 0; i < nblks; i++) { |
621 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
622 | mutex_enter(&db->db_mtx); | |
623 | while (db->db_state == DB_READ || | |
624 | db->db_state == DB_FILL) | |
625 | cv_wait(&db->db_changed, &db->db_mtx); | |
626 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 627 | err = SET_ERROR(EIO); |
34dc7c2f BB |
628 | mutex_exit(&db->db_mtx); |
629 | if (err) { | |
630 | dmu_buf_rele_array(dbp, nblks, tag); | |
631 | return (err); | |
632 | } | |
633 | } | |
634 | } | |
635 | ||
636 | *numbufsp = nblks; | |
637 | *dbpp = dbp; | |
638 | return (0); | |
639 | } | |
640 | ||
afbc6179 | 641 | int |
34dc7c2f | 642 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, |
a926aab9 AZ |
643 | uint64_t length, int read, const void *tag, int *numbufsp, |
644 | dmu_buf_t ***dbpp) | |
34dc7c2f BB |
645 | { |
646 | dnode_t *dn; | |
647 | int err; | |
648 | ||
428870ff | 649 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
650 | if (err) |
651 | return (err); | |
652 | ||
653 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 654 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
655 | |
656 | dnode_rele(dn, FTAG); | |
657 | ||
658 | return (err); | |
659 | } | |
660 | ||
661 | int | |
572e2857 | 662 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
a926aab9 | 663 | uint64_t length, boolean_t read, const void *tag, int *numbufsp, |
7f60329a | 664 | dmu_buf_t ***dbpp) |
34dc7c2f | 665 | { |
572e2857 BB |
666 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
667 | dnode_t *dn; | |
34dc7c2f BB |
668 | int err; |
669 | ||
572e2857 BB |
670 | DB_DNODE_ENTER(db); |
671 | dn = DB_DNODE(db); | |
34dc7c2f | 672 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 673 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 674 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
675 | |
676 | return (err); | |
677 | } | |
678 | ||
679 | void | |
a926aab9 | 680 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, const void *tag) |
34dc7c2f BB |
681 | { |
682 | int i; | |
683 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
684 | ||
685 | if (numbufs == 0) | |
686 | return; | |
687 | ||
688 | for (i = 0; i < numbufs; i++) { | |
689 | if (dbp[i]) | |
690 | dbuf_rele(dbp[i], tag); | |
691 | } | |
692 | ||
693 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
694 | } | |
695 | ||
e8b96c60 | 696 | /* |
6c94e649 | 697 | * Issue prefetch I/Os for the given blocks. If level is greater than 0, the |
e1cfd73f | 698 | * indirect blocks prefetched will be those that point to the blocks containing |
6c94e649 AM |
699 | * the data starting at offset, and continuing to offset + len. If the range |
700 | * it too long, prefetch the first dmu_prefetch_max bytes as requested, while | |
701 | * for the rest only a higher level, also fitting within dmu_prefetch_max. It | |
702 | * should primarily help random reads, since for long sequential reads there is | |
703 | * a speculative prefetcher. | |
e8b96c60 | 704 | * |
b5256303 | 705 | * Note that if the indirect blocks above the blocks being prefetched are not |
6c94e649 AM |
706 | * in cache, they will be asynchronously read in. Dnode read by dnode_hold() |
707 | * is currently synchronous. | |
e8b96c60 | 708 | */ |
34dc7c2f | 709 | void |
fcff0f35 PD |
710 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
711 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
712 | { |
713 | dnode_t *dn; | |
6c94e649 AM |
714 | int64_t level2 = level; |
715 | uint64_t start, end, start2, end2; | |
34dc7c2f | 716 | |
6c94e649 AM |
717 | if (dmu_prefetch_max == 0 || len == 0) { |
718 | dmu_prefetch_dnode(os, object, pri); | |
34dc7c2f BB |
719 | return; |
720 | } | |
721 | ||
6c94e649 | 722 | if (dnode_hold(os, object, FTAG, &dn) != 0) |
34dc7c2f BB |
723 | return; |
724 | ||
fcff0f35 | 725 | /* |
6c94e649 AM |
726 | * Depending on len we may do two prefetches: blocks [start, end) at |
727 | * level, and following blocks [start2, end2) at higher level2. | |
fcff0f35 | 728 | */ |
f664f1ee | 729 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
6c94e649 AM |
730 | if (dn->dn_datablkshift != 0) { |
731 | /* | |
732 | * The object has multiple blocks. Calculate the full range | |
733 | * of blocks [start, end2) and then split it into two parts, | |
734 | * so that the first [start, end) fits into dmu_prefetch_max. | |
735 | */ | |
736 | start = dbuf_whichblock(dn, level, offset); | |
737 | end2 = dbuf_whichblock(dn, level, offset + len - 1) + 1; | |
738 | uint8_t ibs = dn->dn_indblkshift; | |
739 | uint8_t bs = (level == 0) ? dn->dn_datablkshift : ibs; | |
740 | uint_t limit = P2ROUNDUP(dmu_prefetch_max, 1 << bs) >> bs; | |
741 | start2 = end = MIN(end2, start + limit); | |
742 | ||
743 | /* | |
744 | * Find level2 where [start2, end2) fits into dmu_prefetch_max. | |
745 | */ | |
746 | uint8_t ibps = ibs - SPA_BLKPTRSHIFT; | |
747 | limit = P2ROUNDUP(dmu_prefetch_max, 1 << ibs) >> ibs; | |
748 | do { | |
749 | level2++; | |
750 | start2 = P2ROUNDUP(start2, 1 << ibps) >> ibps; | |
751 | end2 = P2ROUNDUP(end2, 1 << ibps) >> ibps; | |
752 | } while (end2 - start2 > limit); | |
34dc7c2f | 753 | } else { |
6c94e649 AM |
754 | /* There is only one block. Prefetch it or nothing. */ |
755 | start = start2 = end2 = 0; | |
756 | end = start + (level == 0 && offset < dn->dn_datablksz); | |
34dc7c2f BB |
757 | } |
758 | ||
6c94e649 AM |
759 | for (uint64_t i = start; i < end; i++) |
760 | dbuf_prefetch(dn, level, i, pri, 0); | |
761 | for (uint64_t i = start2; i < end2; i++) | |
762 | dbuf_prefetch(dn, level2, i, pri, 0); | |
34dc7c2f BB |
763 | rw_exit(&dn->dn_struct_rwlock); |
764 | ||
765 | dnode_rele(dn, FTAG); | |
766 | } | |
767 | ||
6c94e649 AM |
768 | /* |
769 | * Issue prefetch I/Os for the given object's dnode. | |
770 | */ | |
771 | void | |
772 | dmu_prefetch_dnode(objset_t *os, uint64_t object, zio_priority_t pri) | |
773 | { | |
774 | if (object == 0 || object >= DN_MAX_OBJECT) | |
775 | return; | |
776 | ||
777 | dnode_t *dn = DMU_META_DNODE(os); | |
778 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
779 | uint64_t blkid = dbuf_whichblock(dn, 0, object * sizeof (dnode_phys_t)); | |
780 | dbuf_prefetch(dn, 0, blkid, pri, 0); | |
781 | rw_exit(&dn->dn_struct_rwlock); | |
782 | } | |
783 | ||
45d1cae3 BB |
784 | /* |
785 | * Get the next "chunk" of file data to free. We traverse the file from | |
786 | * the end so that the file gets shorter over time (if we crashes in the | |
787 | * middle, this will leave us in a better state). We find allocated file | |
788 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
789 | * |
790 | * On input, *start should be the first offset that does not need to be | |
791 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
792 | * offset that should be freed and l1blks is set to the number of level 1 |
793 | * indirect blocks found within the chunk. | |
45d1cae3 | 794 | */ |
b128c09f | 795 | static int |
65282ee9 | 796 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 797 | { |
65282ee9 | 798 | uint64_t blks; |
b663a23d MA |
799 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
800 | /* bytes of data covered by a level-1 indirect block */ | |
ec4afd27 OM |
801 | uint64_t iblkrange = (uint64_t)dn->dn_datablksz * |
802 | EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
b128c09f | 803 | |
b663a23d | 804 | ASSERT3U(minimum, <=, *start); |
b128c09f | 805 | |
f4c594da TC |
806 | /* |
807 | * Check if we can free the entire range assuming that all of the | |
808 | * L1 blocks in this range have data. If we can, we use this | |
809 | * worst case value as an estimate so we can avoid having to look | |
810 | * at the object's actual data. | |
811 | */ | |
812 | uint64_t total_l1blks = | |
813 | (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / | |
814 | iblkrange; | |
815 | if (total_l1blks <= maxblks) { | |
816 | *l1blks = total_l1blks; | |
b663a23d | 817 | *start = minimum; |
b128c09f BB |
818 | return (0); |
819 | } | |
45d1cae3 | 820 | ASSERT(ISP2(iblkrange)); |
b128c09f | 821 | |
65282ee9 | 822 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
823 | int err; |
824 | ||
b663a23d MA |
825 | /* |
826 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
827 | * indirect block at or before the input offset. We must | |
828 | * decrement *start so that it is at the end of the region | |
829 | * to search. | |
830 | */ | |
831 | (*start)--; | |
f4c594da | 832 | |
b128c09f | 833 | err = dnode_next_offset(dn, |
45d1cae3 | 834 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 835 | |
b663a23d | 836 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 837 | if (err == ESRCH) { |
b663a23d MA |
838 | *start = minimum; |
839 | break; | |
840 | } else if (err != 0) { | |
65282ee9 | 841 | *l1blks = blks; |
b128c09f | 842 | return (err); |
45d1cae3 | 843 | } |
b128c09f | 844 | |
b663a23d | 845 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 846 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 847 | } |
b663a23d MA |
848 | if (*start < minimum) |
849 | *start = minimum; | |
65282ee9 | 850 | *l1blks = blks; |
f4c594da | 851 | |
b128c09f BB |
852 | return (0); |
853 | } | |
854 | ||
a08abc1b GM |
855 | /* |
856 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
857 | * otherwise return false. | |
858 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
859 | */ | |
a08abc1b GM |
860 | static boolean_t |
861 | dmu_objset_zfs_unmounting(objset_t *os) | |
862 | { | |
863 | #ifdef _KERNEL | |
864 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
865 | return (zfs_get_vfs_flag_unmounted(os)); | |
14e4e3cb AZ |
866 | #else |
867 | (void) os; | |
a08abc1b GM |
868 | #endif |
869 | return (B_FALSE); | |
870 | } | |
871 | ||
b128c09f BB |
872 | static int |
873 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 874 | uint64_t length) |
b128c09f | 875 | { |
c97d3069 | 876 | uint64_t object_size; |
b663a23d | 877 | int err; |
539d33c7 GM |
878 | uint64_t dirty_frees_threshold; |
879 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 880 | |
c97d3069 BB |
881 | if (dn == NULL) |
882 | return (SET_ERROR(EINVAL)); | |
883 | ||
884 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 885 | if (offset >= object_size) |
b128c09f | 886 | return (0); |
b128c09f | 887 | |
539d33c7 GM |
888 | if (zfs_per_txg_dirty_frees_percent <= 100) |
889 | dirty_frees_threshold = | |
890 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
891 | else | |
65282ee9 | 892 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 893 | |
b663a23d MA |
894 | if (length == DMU_OBJECT_END || offset + length > object_size) |
895 | length = object_size - offset; | |
896 | ||
897 | while (length != 0) { | |
539d33c7 | 898 | uint64_t chunk_end, chunk_begin, chunk_len; |
65282ee9 | 899 | uint64_t l1blks; |
b663a23d MA |
900 | dmu_tx_t *tx; |
901 | ||
a08abc1b GM |
902 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
903 | return (SET_ERROR(EINTR)); | |
904 | ||
b663a23d MA |
905 | chunk_end = chunk_begin = offset + length; |
906 | ||
907 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 908 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
909 | if (err) |
910 | return (err); | |
b663a23d MA |
911 | ASSERT3U(chunk_begin, >=, offset); |
912 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 913 | |
539d33c7 GM |
914 | chunk_len = chunk_end - chunk_begin; |
915 | ||
b128c09f | 916 | tx = dmu_tx_create(os); |
539d33c7 | 917 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
918 | |
919 | /* | |
920 | * Mark this transaction as typically resulting in a net | |
921 | * reduction in space used. | |
922 | */ | |
923 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
924 | err = dmu_tx_assign(tx, TXG_WAIT); |
925 | if (err) { | |
926 | dmu_tx_abort(tx); | |
927 | return (err); | |
928 | } | |
539d33c7 | 929 | |
f4c594da TC |
930 | uint64_t txg = dmu_tx_get_txg(tx); |
931 | ||
932 | mutex_enter(&dp->dp_lock); | |
933 | uint64_t long_free_dirty = | |
934 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; | |
935 | mutex_exit(&dp->dp_lock); | |
936 | ||
937 | /* | |
938 | * To avoid filling up a TXG with just frees, wait for | |
939 | * the next TXG to open before freeing more chunks if | |
940 | * we have reached the threshold of frees. | |
941 | */ | |
942 | if (dirty_frees_threshold != 0 && | |
943 | long_free_dirty >= dirty_frees_threshold) { | |
944 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); | |
945 | dmu_tx_commit(tx); | |
946 | txg_wait_open(dp, 0, B_TRUE); | |
947 | continue; | |
948 | } | |
949 | ||
65282ee9 AP |
950 | /* |
951 | * In order to prevent unnecessary write throttling, for each | |
952 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
953 | * in dnode_free_range() below. We compare this number to a | |
954 | * tunable threshold, past which we prevent new L1 dirty freeing | |
955 | * blocks from being added into the open TXG. See | |
956 | * dmu_free_long_range_impl() for details. The threshold | |
957 | * prevents write throttle activation due to dirty freeing L1 | |
958 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
959 | */ | |
539d33c7 | 960 | mutex_enter(&dp->dp_lock); |
f4c594da | 961 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += |
65282ee9 | 962 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
963 | mutex_exit(&dp->dp_lock); |
964 | DTRACE_PROBE3(free__long__range, | |
f4c594da TC |
965 | uint64_t, long_free_dirty, uint64_t, chunk_len, |
966 | uint64_t, txg); | |
539d33c7 | 967 | dnode_free_range(dn, chunk_begin, chunk_len, tx); |
440a3eb9 | 968 | |
b128c09f | 969 | dmu_tx_commit(tx); |
b663a23d | 970 | |
539d33c7 | 971 | length -= chunk_len; |
b128c09f BB |
972 | } |
973 | return (0); | |
974 | } | |
975 | ||
976 | int | |
977 | dmu_free_long_range(objset_t *os, uint64_t object, | |
978 | uint64_t offset, uint64_t length) | |
979 | { | |
980 | dnode_t *dn; | |
981 | int err; | |
982 | ||
428870ff | 983 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
984 | if (err != 0) |
985 | return (err); | |
0c03d21a | 986 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
987 | |
988 | /* | |
989 | * It is important to zero out the maxblkid when freeing the entire | |
990 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
991 | * will take the fast path, and (b) dnode_reallocate() can verify | |
992 | * that the entire file has been freed. | |
993 | */ | |
b0bc7a84 | 994 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
995 | dn->dn_maxblkid = 0; |
996 | ||
b128c09f BB |
997 | dnode_rele(dn, FTAG); |
998 | return (err); | |
999 | } | |
1000 | ||
1001 | int | |
0c03d21a | 1002 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 1003 | { |
b128c09f BB |
1004 | dmu_tx_t *tx; |
1005 | int err; | |
1006 | ||
b663a23d | 1007 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
1008 | if (err != 0) |
1009 | return (err); | |
b663a23d MA |
1010 | |
1011 | tx = dmu_tx_create(os); | |
1012 | dmu_tx_hold_bonus(tx, object); | |
1013 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 1014 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
1015 | err = dmu_tx_assign(tx, TXG_WAIT); |
1016 | if (err == 0) { | |
60b618a9 | 1017 | err = dmu_object_free(os, object, tx); |
b663a23d | 1018 | dmu_tx_commit(tx); |
b128c09f | 1019 | } else { |
b663a23d | 1020 | dmu_tx_abort(tx); |
b128c09f | 1021 | } |
b663a23d | 1022 | |
b128c09f BB |
1023 | return (err); |
1024 | } | |
1025 | ||
34dc7c2f BB |
1026 | int |
1027 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
1028 | uint64_t size, dmu_tx_t *tx) | |
1029 | { | |
1030 | dnode_t *dn; | |
428870ff | 1031 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1032 | if (err) |
1033 | return (err); | |
1034 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 1035 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
1036 | dnode_free_range(dn, offset, size, tx); |
1037 | dnode_rele(dn, FTAG); | |
1038 | return (0); | |
1039 | } | |
1040 | ||
0eef1bde | 1041 | static int |
1042 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 1043 | void *buf, uint32_t flags) |
34dc7c2f | 1044 | { |
34dc7c2f | 1045 | dmu_buf_t **dbp; |
0eef1bde | 1046 | int numbufs, err = 0; |
34dc7c2f BB |
1047 | |
1048 | /* | |
1049 | * Deal with odd block sizes, where there can't be data past the first | |
1050 | * block. If we ever do the tail block optimization, we will need to | |
1051 | * handle that here as well. | |
1052 | */ | |
45d1cae3 | 1053 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 1054 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f | 1055 | MIN(size, dn->dn_datablksz - offset); |
861166b0 | 1056 | memset((char *)buf + newsz, 0, size - newsz); |
34dc7c2f BB |
1057 | size = newsz; |
1058 | } | |
1059 | ||
1060 | while (size > 0) { | |
1061 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 1062 | int i; |
34dc7c2f BB |
1063 | |
1064 | /* | |
1065 | * NB: we could do this block-at-a-time, but it's nice | |
1066 | * to be reading in parallel. | |
1067 | */ | |
1068 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 1069 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
1070 | if (err) |
1071 | break; | |
1072 | ||
1073 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1074 | uint64_t tocpy; |
1075 | int64_t bufoff; | |
34dc7c2f BB |
1076 | dmu_buf_t *db = dbp[i]; |
1077 | ||
1078 | ASSERT(size > 0); | |
1079 | ||
1080 | bufoff = offset - db->db_offset; | |
c9520ecc | 1081 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 1082 | |
c9520ecc | 1083 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
1084 | |
1085 | offset += tocpy; | |
1086 | size -= tocpy; | |
1087 | buf = (char *)buf + tocpy; | |
1088 | } | |
1089 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1090 | } | |
34dc7c2f BB |
1091 | return (err); |
1092 | } | |
1093 | ||
0eef1bde | 1094 | int |
1095 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1096 | void *buf, uint32_t flags) | |
34dc7c2f | 1097 | { |
0eef1bde | 1098 | dnode_t *dn; |
1099 | int err; | |
34dc7c2f | 1100 | |
0eef1bde | 1101 | err = dnode_hold(os, object, FTAG, &dn); |
1102 | if (err != 0) | |
1103 | return (err); | |
34dc7c2f | 1104 | |
0eef1bde | 1105 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1106 | dnode_rele(dn, FTAG); | |
1107 | return (err); | |
1108 | } | |
1109 | ||
1110 | int | |
1111 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1112 | uint32_t flags) | |
1113 | { | |
1114 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1115 | } | |
1116 | ||
1117 | static void | |
1118 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1119 | const void *buf, dmu_tx_t *tx) | |
1120 | { | |
1121 | int i; | |
34dc7c2f BB |
1122 | |
1123 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1124 | uint64_t tocpy; |
1125 | int64_t bufoff; | |
34dc7c2f BB |
1126 | dmu_buf_t *db = dbp[i]; |
1127 | ||
1128 | ASSERT(size > 0); | |
1129 | ||
1130 | bufoff = offset - db->db_offset; | |
c9520ecc | 1131 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1132 | |
1133 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1134 | ||
1135 | if (tocpy == db->db_size) | |
1136 | dmu_buf_will_fill(db, tx); | |
1137 | else | |
1138 | dmu_buf_will_dirty(db, tx); | |
1139 | ||
60101509 | 1140 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1141 | |
1142 | if (tocpy == db->db_size) | |
1143 | dmu_buf_fill_done(db, tx); | |
1144 | ||
1145 | offset += tocpy; | |
1146 | size -= tocpy; | |
1147 | buf = (char *)buf + tocpy; | |
1148 | } | |
0eef1bde | 1149 | } |
1150 | ||
1151 | void | |
1152 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1153 | const void *buf, dmu_tx_t *tx) | |
1154 | { | |
1155 | dmu_buf_t **dbp; | |
1156 | int numbufs; | |
1157 | ||
1158 | if (size == 0) | |
1159 | return; | |
1160 | ||
1161 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1162 | FALSE, FTAG, &numbufs, &dbp)); | |
1163 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1164 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1165 | } | |
1166 | ||
0f8ff49e SD |
1167 | /* |
1168 | * Note: Lustre is an external consumer of this interface. | |
1169 | */ | |
0eef1bde | 1170 | void |
1171 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1172 | const void *buf, dmu_tx_t *tx) | |
1173 | { | |
1174 | dmu_buf_t **dbp; | |
1175 | int numbufs; | |
1176 | ||
1177 | if (size == 0) | |
1178 | return; | |
1179 | ||
1180 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1181 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1182 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1183 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1184 | } | |
1185 | ||
b128c09f BB |
1186 | void |
1187 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1188 | dmu_tx_t *tx) | |
1189 | { | |
1190 | dmu_buf_t **dbp; | |
1191 | int numbufs, i; | |
1192 | ||
1193 | if (size == 0) | |
1194 | return; | |
1195 | ||
1196 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1197 | FALSE, FTAG, &numbufs, &dbp)); | |
1198 | ||
1199 | for (i = 0; i < numbufs; i++) { | |
1200 | dmu_buf_t *db = dbp[i]; | |
1201 | ||
1202 | dmu_buf_will_not_fill(db, tx); | |
1203 | } | |
1204 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1205 | } | |
1206 | ||
9b67f605 MA |
1207 | void |
1208 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1209 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1210 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1211 | { | |
1212 | dmu_buf_t *db; | |
1213 | ||
1214 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1215 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1216 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1217 | FTAG, &db)); | |
1218 | ||
1219 | dmu_buf_write_embedded(db, | |
1220 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1221 | uncompressed_size, compressed_size, byteorder, tx); | |
1222 | ||
1223 | dmu_buf_rele(db, FTAG); | |
1224 | } | |
1225 | ||
30af21b0 PD |
1226 | void |
1227 | dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1228 | dmu_tx_t *tx) | |
1229 | { | |
1230 | int numbufs, i; | |
1231 | dmu_buf_t **dbp; | |
1232 | ||
1233 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1234 | &numbufs, &dbp)); | |
1235 | for (i = 0; i < numbufs; i++) | |
1236 | dmu_buf_redact(dbp[i], tx); | |
1237 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1238 | } | |
1239 | ||
34dc7c2f | 1240 | #ifdef _KERNEL |
5228cf01 | 1241 | int |
d0cd9a5c | 1242 | dmu_read_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size) |
872e8d26 BB |
1243 | { |
1244 | dmu_buf_t **dbp; | |
1245 | int numbufs, i, err; | |
872e8d26 BB |
1246 | |
1247 | /* | |
1248 | * NB: we could do this block-at-a-time, but it's nice | |
1249 | * to be reading in parallel. | |
1250 | */ | |
d0cd9a5c | 1251 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
804e0504 | 1252 | TRUE, FTAG, &numbufs, &dbp, 0); |
872e8d26 BB |
1253 | if (err) |
1254 | return (err); | |
1255 | ||
1256 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1257 | uint64_t tocpy; |
1258 | int64_t bufoff; | |
872e8d26 BB |
1259 | dmu_buf_t *db = dbp[i]; |
1260 | ||
1261 | ASSERT(size > 0); | |
1262 | ||
d0cd9a5c | 1263 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1264 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1265 | |
d0cd9a5c BA |
1266 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, tocpy, |
1267 | UIO_READ, uio); | |
1268 | ||
872e8d26 BB |
1269 | if (err) |
1270 | break; | |
1271 | ||
1272 | size -= tocpy; | |
1273 | } | |
1274 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1275 | ||
1276 | return (err); | |
1277 | } | |
1278 | ||
804e0504 MA |
1279 | /* |
1280 | * Read 'size' bytes into the uio buffer. | |
1281 | * From object zdb->db_object. | |
d0cd9a5c | 1282 | * Starting at zfs_uio_offset(uio). |
804e0504 MA |
1283 | * |
1284 | * If the caller already has a dbuf in the target object | |
1285 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1286 | * because we don't have to find the dnode_t for the object. | |
1287 | */ | |
1288 | int | |
d0cd9a5c | 1289 | dmu_read_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1290 | { |
1291 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1292 | dnode_t *dn; | |
1293 | int err; | |
1294 | ||
1295 | if (size == 0) | |
1296 | return (0); | |
1297 | ||
1298 | DB_DNODE_ENTER(db); | |
1299 | dn = DB_DNODE(db); | |
1300 | err = dmu_read_uio_dnode(dn, uio, size); | |
1301 | DB_DNODE_EXIT(db); | |
1302 | ||
1303 | return (err); | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * Read 'size' bytes into the uio buffer. | |
1308 | * From the specified object | |
d0cd9a5c | 1309 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1310 | */ |
1311 | int | |
d0cd9a5c | 1312 | dmu_read_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1313 | { |
1314 | dnode_t *dn; | |
1315 | int err; | |
1316 | ||
1317 | if (size == 0) | |
1318 | return (0); | |
1319 | ||
1320 | err = dnode_hold(os, object, FTAG, &dn); | |
1321 | if (err) | |
1322 | return (err); | |
1323 | ||
1324 | err = dmu_read_uio_dnode(dn, uio, size); | |
1325 | ||
1326 | dnode_rele(dn, FTAG); | |
1327 | ||
1328 | return (err); | |
1329 | } | |
1330 | ||
5228cf01 | 1331 | int |
d0cd9a5c | 1332 | dmu_write_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size, dmu_tx_t *tx) |
872e8d26 BB |
1333 | { |
1334 | dmu_buf_t **dbp; | |
1335 | int numbufs; | |
1336 | int err = 0; | |
1337 | int i; | |
1338 | ||
d0cd9a5c | 1339 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
872e8d26 BB |
1340 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); |
1341 | if (err) | |
1342 | return (err); | |
1343 | ||
1344 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1345 | uint64_t tocpy; |
1346 | int64_t bufoff; | |
872e8d26 BB |
1347 | dmu_buf_t *db = dbp[i]; |
1348 | ||
1349 | ASSERT(size > 0); | |
1350 | ||
d0cd9a5c | 1351 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1352 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1353 | |
1354 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1355 | ||
1356 | if (tocpy == db->db_size) | |
1357 | dmu_buf_will_fill(db, tx); | |
1358 | else | |
1359 | dmu_buf_will_dirty(db, tx); | |
1360 | ||
1361 | /* | |
d0cd9a5c | 1362 | * XXX zfs_uiomove could block forever (eg.nfs-backed |
872e8d26 | 1363 | * pages). There needs to be a uiolockdown() function |
d0cd9a5c | 1364 | * to lock the pages in memory, so that zfs_uiomove won't |
872e8d26 BB |
1365 | * block. |
1366 | */ | |
d0cd9a5c BA |
1367 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, |
1368 | tocpy, UIO_WRITE, uio); | |
1369 | ||
872e8d26 BB |
1370 | if (tocpy == db->db_size) |
1371 | dmu_buf_fill_done(db, tx); | |
1372 | ||
1373 | if (err) | |
1374 | break; | |
1375 | ||
1376 | size -= tocpy; | |
1377 | } | |
1378 | ||
1379 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1380 | return (err); | |
1381 | } | |
1382 | ||
804e0504 MA |
1383 | /* |
1384 | * Write 'size' bytes from the uio buffer. | |
1385 | * To object zdb->db_object. | |
d0cd9a5c | 1386 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1387 | * |
1388 | * If the caller already has a dbuf in the target object | |
1389 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1390 | * because we don't have to find the dnode_t for the object. | |
1391 | */ | |
428870ff | 1392 | int |
d0cd9a5c | 1393 | dmu_write_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1394 | dmu_tx_t *tx) |
1395 | { | |
572e2857 BB |
1396 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1397 | dnode_t *dn; | |
1398 | int err; | |
1399 | ||
428870ff BB |
1400 | if (size == 0) |
1401 | return (0); | |
1402 | ||
572e2857 BB |
1403 | DB_DNODE_ENTER(db); |
1404 | dn = DB_DNODE(db); | |
1405 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1406 | DB_DNODE_EXIT(db); | |
1407 | ||
1408 | return (err); | |
428870ff BB |
1409 | } |
1410 | ||
804e0504 MA |
1411 | /* |
1412 | * Write 'size' bytes from the uio buffer. | |
1413 | * To the specified object. | |
d0cd9a5c | 1414 | * Starting at offset zfs_uio_offset(uio). |
804e0504 | 1415 | */ |
428870ff | 1416 | int |
d0cd9a5c | 1417 | dmu_write_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1418 | dmu_tx_t *tx) |
1419 | { | |
1420 | dnode_t *dn; | |
1421 | int err; | |
1422 | ||
1423 | if (size == 0) | |
1424 | return (0); | |
1425 | ||
1426 | err = dnode_hold(os, object, FTAG, &dn); | |
1427 | if (err) | |
1428 | return (err); | |
1429 | ||
1430 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1431 | ||
1432 | dnode_rele(dn, FTAG); | |
1433 | ||
1434 | return (err); | |
1435 | } | |
872e8d26 | 1436 | #endif /* _KERNEL */ |
34dc7c2f | 1437 | |
9babb374 BB |
1438 | /* |
1439 | * Allocate a loaned anonymous arc buffer. | |
1440 | */ | |
1441 | arc_buf_t * | |
1442 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1443 | { | |
572e2857 | 1444 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1445 | |
2aa34383 | 1446 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1447 | } |
1448 | ||
1449 | /* | |
1450 | * Free a loaned arc buffer. | |
1451 | */ | |
1452 | void | |
1453 | dmu_return_arcbuf(arc_buf_t *buf) | |
1454 | { | |
1455 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1456 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1457 | } |
1458 | ||
ba67d821 MA |
1459 | /* |
1460 | * A "lightweight" write is faster than a regular write (e.g. | |
1461 | * dmu_write_by_dnode() or dmu_assign_arcbuf_by_dnode()), because it avoids the | |
1462 | * CPU cost of creating a dmu_buf_impl_t and arc_buf_[hdr_]_t. However, the | |
1463 | * data can not be read or overwritten until the transaction's txg has been | |
1464 | * synced. This makes it appropriate for workloads that are known to be | |
1465 | * (temporarily) write-only, like "zfs receive". | |
1466 | * | |
1467 | * A single block is written, starting at the specified offset in bytes. If | |
1468 | * the call is successful, it returns 0 and the provided abd has been | |
1469 | * consumed (the caller should not free it). | |
1470 | */ | |
1471 | int | |
1472 | dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, | |
4938d01d | 1473 | const zio_prop_t *zp, zio_flag_t flags, dmu_tx_t *tx) |
ba67d821 MA |
1474 | { |
1475 | dbuf_dirty_record_t *dr = | |
1476 | dbuf_dirty_lightweight(dn, dbuf_whichblock(dn, 0, offset), tx); | |
1477 | if (dr == NULL) | |
1478 | return (SET_ERROR(EIO)); | |
1479 | dr->dt.dll.dr_abd = abd; | |
1480 | dr->dt.dll.dr_props = *zp; | |
1481 | dr->dt.dll.dr_flags = flags; | |
1482 | return (0); | |
1483 | } | |
1484 | ||
9babb374 BB |
1485 | /* |
1486 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1487 | * If this is not possible copy the contents of passed arc buf via | |
1488 | * dmu_write(). | |
1489 | */ | |
305781da | 1490 | int |
440a3eb9 | 1491 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1492 | dmu_tx_t *tx) |
1493 | { | |
9babb374 | 1494 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1495 | objset_t *os = dn->dn_objset; |
1496 | uint64_t object = dn->dn_object; | |
2aa34383 | 1497 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1498 | uint64_t blkid; |
1499 | ||
1500 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1501 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1502 | db = dbuf_hold(dn, blkid, FTAG); |
1503 | if (db == NULL) | |
1504 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1505 | rw_exit(&dn->dn_struct_rwlock); |
1506 | ||
88904bb3 | 1507 | /* |
ba67d821 MA |
1508 | * We can only assign if the offset is aligned and the arc buf is the |
1509 | * same size as the dbuf. | |
88904bb3 | 1510 | */ |
2aa34383 | 1511 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
64e0fe14 | 1512 | zfs_racct_write(blksz, 1); |
9babb374 BB |
1513 | dbuf_assign_arcbuf(db, buf, tx); |
1514 | dbuf_rele(db, FTAG); | |
1515 | } else { | |
2aa34383 DK |
1516 | /* compressed bufs must always be assignable to their dbuf */ |
1517 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1518 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1519 | |
9babb374 | 1520 | dbuf_rele(db, FTAG); |
572e2857 | 1521 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 BB |
1522 | dmu_return_arcbuf(buf); |
1523 | } | |
305781da TC |
1524 | |
1525 | return (0); | |
9babb374 BB |
1526 | } |
1527 | ||
305781da | 1528 | int |
440a3eb9 TC |
1529 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1530 | dmu_tx_t *tx) | |
1531 | { | |
305781da | 1532 | int err; |
440a3eb9 TC |
1533 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1534 | ||
1535 | DB_DNODE_ENTER(dbuf); | |
305781da | 1536 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1537 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1538 | |
1539 | return (err); | |
440a3eb9 TC |
1540 | } |
1541 | ||
34dc7c2f | 1542 | typedef struct { |
428870ff BB |
1543 | dbuf_dirty_record_t *dsa_dr; |
1544 | dmu_sync_cb_t *dsa_done; | |
1545 | zgd_t *dsa_zgd; | |
1546 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1547 | } dmu_sync_arg_t; |
1548 | ||
b128c09f BB |
1549 | static void |
1550 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1551 | { | |
14e4e3cb | 1552 | (void) buf; |
428870ff BB |
1553 | dmu_sync_arg_t *dsa = varg; |
1554 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1555 | blkptr_t *bp = zio->io_bp; |
1556 | ||
428870ff BB |
1557 | if (zio->io_error == 0) { |
1558 | if (BP_IS_HOLE(bp)) { | |
1559 | /* | |
1560 | * A block of zeros may compress to a hole, but the | |
1561 | * block size still needs to be known for replay. | |
1562 | */ | |
1563 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1564 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1565 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1566 | BP_SET_FILL(bp, 1); |
428870ff | 1567 | } |
b128c09f BB |
1568 | } |
1569 | } | |
1570 | ||
428870ff BB |
1571 | static void |
1572 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1573 | { | |
1574 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1575 | } | |
1576 | ||
34dc7c2f BB |
1577 | static void |
1578 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1579 | { | |
14e4e3cb | 1580 | (void) buf; |
428870ff BB |
1581 | dmu_sync_arg_t *dsa = varg; |
1582 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1583 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1584 | zgd_t *zgd = dsa->dsa_zgd; |
1585 | ||
1586 | /* | |
1587 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1588 | * the writes for the lwb have completed. | |
1589 | */ | |
1590 | if (zio->io_error == 0) { | |
1591 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1592 | } | |
34dc7c2f | 1593 | |
34dc7c2f BB |
1594 | mutex_enter(&db->db_mtx); |
1595 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1596 | if (zio->io_error == 0) { |
03c6040b GW |
1597 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1598 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1599 | blkptr_t *bp = zio->io_bp; |
1600 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1601 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1602 | |
1603 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1604 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1605 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1606 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1607 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1608 | } |
428870ff BB |
1609 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1610 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1611 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1612 | |
1613 | /* | |
1614 | * Old style holes are filled with all zeros, whereas | |
1615 | * new-style holes maintain their lsize, type, level, | |
1616 | * and birth time (see zio_write_compress). While we | |
1617 | * need to reset the BP_SET_LSIZE() call that happened | |
1618 | * in dmu_sync_ready for old style holes, we do *not* | |
1619 | * want to wipe out the information contained in new | |
1620 | * style holes. Thus, only zero out the block pointer if | |
1621 | * it's an old style hole. | |
1622 | */ | |
1623 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1624 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1625 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1626 | } else { | |
1627 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1628 | } | |
34dc7c2f BB |
1629 | cv_broadcast(&db->db_changed); |
1630 | mutex_exit(&db->db_mtx); | |
1631 | ||
428870ff | 1632 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1633 | |
428870ff BB |
1634 | kmem_free(dsa, sizeof (*dsa)); |
1635 | } | |
1636 | ||
1637 | static void | |
1638 | dmu_sync_late_arrival_done(zio_t *zio) | |
1639 | { | |
1640 | blkptr_t *bp = zio->io_bp; | |
1641 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1642 | zgd_t *zgd = dsa->dsa_zgd; |
1643 | ||
1644 | if (zio->io_error == 0) { | |
1645 | /* | |
1646 | * Record the vdev(s) backing this blkptr so they can be | |
1647 | * flushed after the writes for the lwb have completed. | |
1648 | */ | |
1649 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1650 | ||
1651 | if (!BP_IS_HOLE(bp)) { | |
2a8ba608 | 1652 | blkptr_t *bp_orig __maybe_unused = &zio->io_bp_orig; |
900d09b2 PS |
1653 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); |
1654 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1655 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1656 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1657 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1658 | } | |
428870ff BB |
1659 | } |
1660 | ||
1661 | dmu_tx_commit(dsa->dsa_tx); | |
1662 | ||
1663 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1664 | ||
e2af2acc | 1665 | abd_free(zio->io_abd); |
428870ff BB |
1666 | kmem_free(dsa, sizeof (*dsa)); |
1667 | } | |
1668 | ||
1669 | static int | |
1670 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1671 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1672 | { |
1673 | dmu_sync_arg_t *dsa; | |
1674 | dmu_tx_t *tx; | |
1675 | ||
1676 | tx = dmu_tx_create(os); | |
1677 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
704c80f0 AM |
1678 | /* |
1679 | * This transaction does not produce any dirty data or log blocks, so | |
1680 | * it should not be throttled. All other cases wait for TXG sync, by | |
1681 | * which time the log block we are writing will be obsolete, so we can | |
1682 | * skip waiting and just return error here instead. | |
1683 | */ | |
1684 | if (dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE) != 0) { | |
428870ff | 1685 | dmu_tx_abort(tx); |
2e528b49 MA |
1686 | /* Make zl_get_data do txg_waited_synced() */ |
1687 | return (SET_ERROR(EIO)); | |
428870ff BB |
1688 | } |
1689 | ||
1ce23dca PS |
1690 | /* |
1691 | * In order to prevent the zgd's lwb from being free'd prior to | |
1692 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1693 | * the lwb's "max txg" takes this tx's txg into account. | |
1694 | */ | |
1695 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1696 | ||
79c76d5b | 1697 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1698 | dsa->dsa_dr = NULL; |
1699 | dsa->dsa_done = done; | |
1700 | dsa->dsa_zgd = zgd; | |
1701 | dsa->dsa_tx = tx; | |
1702 | ||
02dc43bc MA |
1703 | /* |
1704 | * Since we are currently syncing this txg, it's nontrivial to | |
1705 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1706 | * | |
1707 | * When syncing, the db_blkptr is initially the BP of the previous | |
1708 | * txg. We can not nopwrite against it because it will be changed | |
1709 | * (this is similar to the non-late-arrival case where the dbuf is | |
1710 | * dirty in a future txg). | |
1711 | * | |
1712 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1713 | * We can not nopwrite against it because although the BP will not | |
1714 | * (typically) be changed, the data has not yet been persisted to this | |
1715 | * location. | |
1716 | * | |
1717 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1718 | * possible to always nopwrite, because the data that was written in | |
1719 | * this txg is the same data that we are trying to write. However we | |
1720 | * would need to check that this dbuf is not dirty in any future | |
1721 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1722 | * don't nopwrite in this case. | |
1723 | */ | |
1724 | zp->zp_nopwrite = B_FALSE; | |
1725 | ||
a6255b7f DQ |
1726 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1727 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1728 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
ccec7fbe | 1729 | dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, |
a6255b7f | 1730 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); |
428870ff BB |
1731 | |
1732 | return (0); | |
34dc7c2f BB |
1733 | } |
1734 | ||
1735 | /* | |
1736 | * Intent log support: sync the block associated with db to disk. | |
1737 | * N.B. and XXX: the caller is responsible for making sure that the | |
1738 | * data isn't changing while dmu_sync() is writing it. | |
1739 | * | |
1740 | * Return values: | |
1741 | * | |
03c6040b | 1742 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1743 | * The caller should not log the write. |
1744 | * | |
1745 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1746 | * The caller should not log the write. | |
1747 | * | |
1748 | * EALREADY: this block is already in the process of being synced. | |
1749 | * The caller should track its progress (somehow). | |
1750 | * | |
428870ff BB |
1751 | * EIO: could not do the I/O. |
1752 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1753 | * |
428870ff BB |
1754 | * 0: the I/O has been initiated. |
1755 | * The caller should log this blkptr in the done callback. | |
1756 | * It is possible that the I/O will fail, in which case | |
1757 | * the error will be reported to the done callback and | |
1758 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1759 | */ |
1760 | int | |
428870ff | 1761 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1762 | { |
428870ff BB |
1763 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1764 | objset_t *os = db->db_objset; | |
1765 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
cccbed9f | 1766 | dbuf_dirty_record_t *dr, *dr_next; |
428870ff | 1767 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1768 | zbookmark_phys_t zb; |
428870ff | 1769 | zio_prop_t zp; |
572e2857 | 1770 | dnode_t *dn; |
34dc7c2f | 1771 | |
428870ff | 1772 | ASSERT(pio != NULL); |
34dc7c2f BB |
1773 | ASSERT(txg != 0); |
1774 | ||
428870ff BB |
1775 | SET_BOOKMARK(&zb, ds->ds_object, |
1776 | db->db.db_object, db->db_level, db->db_blkid); | |
1777 | ||
572e2857 BB |
1778 | DB_DNODE_ENTER(db); |
1779 | dn = DB_DNODE(db); | |
82644107 | 1780 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1781 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1782 | |
1783 | /* | |
428870ff | 1784 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1785 | */ |
428870ff BB |
1786 | if (txg > spa_freeze_txg(os->os_spa)) |
1787 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1788 | |
1789 | /* | |
428870ff BB |
1790 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1791 | * and us. If we determine that this txg is not yet syncing, | |
1792 | * but it begins to sync a moment later, that's OK because the | |
1793 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1794 | */ |
428870ff BB |
1795 | mutex_enter(&db->db_mtx); |
1796 | ||
1797 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1798 | /* |
428870ff | 1799 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1800 | */ |
428870ff | 1801 | mutex_exit(&db->db_mtx); |
2e528b49 | 1802 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1803 | } |
1804 | ||
428870ff BB |
1805 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1806 | /* | |
1807 | * This txg is currently syncing, so we can't mess with | |
1808 | * the dirty record anymore; just write a new log block. | |
1809 | */ | |
1810 | mutex_exit(&db->db_mtx); | |
1811 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1812 | } |
1813 | ||
cccbed9f | 1814 | dr = dbuf_find_dirty_eq(db, txg); |
428870ff BB |
1815 | |
1816 | if (dr == NULL) { | |
34dc7c2f | 1817 | /* |
428870ff | 1818 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1819 | * There's no need to log writes to freed blocks, so we're done. |
1820 | */ | |
1821 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1822 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1823 | } |
1824 | ||
cccbed9f MM |
1825 | dr_next = list_next(&db->db_dirty_records, dr); |
1826 | ASSERT(dr_next == NULL || dr_next->dr_txg < txg); | |
03c6040b | 1827 | |
02dc43bc MA |
1828 | if (db->db_blkptr != NULL) { |
1829 | /* | |
1830 | * We need to fill in zgd_bp with the current blkptr so that | |
1831 | * the nopwrite code can check if we're writing the same | |
1832 | * data that's already on disk. We can only nopwrite if we | |
1833 | * are sure that after making the copy, db_blkptr will not | |
1834 | * change until our i/o completes. We ensure this by | |
1835 | * holding the db_mtx, and only allowing nopwrite if the | |
1836 | * block is not already dirty (see below). This is verified | |
1837 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1838 | * not changed. | |
1839 | */ | |
1840 | *zgd->zgd_bp = *db->db_blkptr; | |
1841 | } | |
1842 | ||
03c6040b | 1843 | /* |
f3c517d8 MA |
1844 | * Assume the on-disk data is X, the current syncing data (in |
1845 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1846 | * in dmu_sync). | |
1847 | * | |
1848 | * We usually want to perform a nopwrite if X and Z are the | |
1849 | * same. However, if Y is different (i.e. the BP is going to | |
1850 | * change before this write takes effect), then a nopwrite will | |
1851 | * be incorrect - we would override with X, which could have | |
1852 | * been freed when Y was written. | |
1853 | * | |
1854 | * (Note that this is not a concern when we are nop-writing from | |
1855 | * syncing context, because X and Y must be identical, because | |
1856 | * all previous txgs have been synced.) | |
1857 | * | |
1858 | * Therefore, we disable nopwrite if the current BP could change | |
1859 | * before this TXG. There are two ways it could change: by | |
1860 | * being dirty (dr_next is non-NULL), or by being freed | |
1861 | * (dnode_block_freed()). This behavior is verified by | |
1862 | * zio_done(), which VERIFYs that the override BP is identical | |
1863 | * to the on-disk BP. | |
03c6040b | 1864 | */ |
f3c517d8 MA |
1865 | DB_DNODE_ENTER(db); |
1866 | dn = DB_DNODE(db); | |
cccbed9f | 1867 | if (dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) |
03c6040b | 1868 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1869 | DB_DNODE_EXIT(db); |
03c6040b | 1870 | |
34dc7c2f | 1871 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1872 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1873 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1874 | /* |
428870ff BB |
1875 | * We have already issued a sync write for this buffer, |
1876 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1877 | * have been dirtied since, or we would have cleared the state. |
1878 | */ | |
34dc7c2f | 1879 | mutex_exit(&db->db_mtx); |
2e528b49 | 1880 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1881 | } |
1882 | ||
428870ff | 1883 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1884 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1885 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1886 | |
79c76d5b | 1887 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1888 | dsa->dsa_dr = dr; |
1889 | dsa->dsa_done = done; | |
1890 | dsa->dsa_zgd = zgd; | |
1891 | dsa->dsa_tx = NULL; | |
b128c09f | 1892 | |
ed2f7ba0 AM |
1893 | zio_nowait(arc_write(pio, os->os_spa, txg, zgd->zgd_bp, |
1894 | dr->dt.dl.dr_data, !DBUF_IS_CACHEABLE(db), dbuf_is_l2cacheable(db), | |
ccec7fbe | 1895 | &zp, dmu_sync_ready, NULL, dmu_sync_done, dsa, |
bc77ba73 | 1896 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 1897 | |
428870ff | 1898 | return (0); |
34dc7c2f BB |
1899 | } |
1900 | ||
b5256303 TC |
1901 | int |
1902 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
1903 | { | |
1904 | dnode_t *dn; | |
1905 | int err; | |
1906 | ||
1907 | err = dnode_hold(os, object, FTAG, &dn); | |
1908 | if (err) | |
1909 | return (err); | |
1910 | err = dnode_set_nlevels(dn, nlevels, tx); | |
1911 | dnode_rele(dn, FTAG); | |
1912 | return (err); | |
1913 | } | |
1914 | ||
34dc7c2f BB |
1915 | int |
1916 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 1917 | dmu_tx_t *tx) |
34dc7c2f BB |
1918 | { |
1919 | dnode_t *dn; | |
1920 | int err; | |
1921 | ||
428870ff | 1922 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1923 | if (err) |
1924 | return (err); | |
1925 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1926 | dnode_rele(dn, FTAG); | |
1927 | return (err); | |
1928 | } | |
1929 | ||
ae76f45c TC |
1930 | int |
1931 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
1932 | dmu_tx_t *tx) | |
1933 | { | |
1934 | dnode_t *dn; | |
1935 | int err; | |
1936 | ||
1937 | err = dnode_hold(os, object, FTAG, &dn); | |
1938 | if (err) | |
1939 | return (err); | |
1940 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 1941 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
1942 | rw_exit(&dn->dn_struct_rwlock); |
1943 | dnode_rele(dn, FTAG); | |
1944 | return (0); | |
1945 | } | |
1946 | ||
34dc7c2f BB |
1947 | void |
1948 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 1949 | dmu_tx_t *tx) |
34dc7c2f BB |
1950 | { |
1951 | dnode_t *dn; | |
1952 | ||
9b67f605 MA |
1953 | /* |
1954 | * Send streams include each object's checksum function. This | |
1955 | * check ensures that the receiving system can understand the | |
1956 | * checksum function transmitted. | |
1957 | */ | |
1958 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1959 | ||
1960 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1961 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1962 | dn->dn_checksum = checksum; |
1963 | dnode_setdirty(dn, tx); | |
1964 | dnode_rele(dn, FTAG); | |
1965 | } | |
1966 | ||
1967 | void | |
1968 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 1969 | dmu_tx_t *tx) |
34dc7c2f BB |
1970 | { |
1971 | dnode_t *dn; | |
1972 | ||
9b67f605 MA |
1973 | /* |
1974 | * Send streams include each object's compression function. This | |
1975 | * check ensures that the receiving system can understand the | |
1976 | * compression function transmitted. | |
1977 | */ | |
1978 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1979 | ||
1980 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
1981 | dn->dn_compress = compress; |
1982 | dnode_setdirty(dn, tx); | |
1983 | dnode_rele(dn, FTAG); | |
1984 | } | |
1985 | ||
faf0f58c MA |
1986 | /* |
1987 | * When the "redundant_metadata" property is set to "most", only indirect | |
1988 | * blocks of this level and higher will have an additional ditto block. | |
1989 | */ | |
18168da7 | 1990 | static const int zfs_redundant_metadata_most_ditto_level = 2; |
faf0f58c | 1991 | |
428870ff | 1992 | void |
82644107 | 1993 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
1994 | { |
1995 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 1996 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 1997 | (wp & WP_SPILL)); |
428870ff BB |
1998 | enum zio_checksum checksum = os->os_checksum; |
1999 | enum zio_compress compress = os->os_compress; | |
10b3c7f5 | 2000 | uint8_t complevel = os->os_complevel; |
428870ff | 2001 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; |
03c6040b GW |
2002 | boolean_t dedup = B_FALSE; |
2003 | boolean_t nopwrite = B_FALSE; | |
428870ff | 2004 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 2005 | boolean_t encrypt = B_FALSE; |
428870ff | 2006 | int copies = os->os_copies; |
a7004725 | 2007 | |
428870ff | 2008 | /* |
03c6040b GW |
2009 | * We maintain different write policies for each of the following |
2010 | * types of data: | |
2011 | * 1. metadata | |
2012 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
2013 | * 3. all other level 0 blocks | |
428870ff BB |
2014 | */ |
2015 | if (ismd) { | |
b1d21733 TC |
2016 | /* |
2017 | * XXX -- we should design a compression algorithm | |
2018 | * that specializes in arrays of bps. | |
2019 | */ | |
2020 | compress = zio_compress_select(os->os_spa, | |
2021 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 2022 | |
428870ff BB |
2023 | /* |
2024 | * Metadata always gets checksummed. If the data | |
2025 | * checksum is multi-bit correctable, and it's not a | |
2026 | * ZBT-style checksum, then it's suitable for metadata | |
2027 | * as well. Otherwise, the metadata checksum defaults | |
2028 | * to fletcher4. | |
2029 | */ | |
3c67d83a TH |
2030 | if (!(zio_checksum_table[checksum].ci_flags & |
2031 | ZCHECKSUM_FLAG_METADATA) || | |
2032 | (zio_checksum_table[checksum].ci_flags & | |
2033 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 2034 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c | 2035 | |
5405be03 A |
2036 | switch (os->os_redundant_metadata) { |
2037 | case ZFS_REDUNDANT_METADATA_ALL: | |
faf0f58c | 2038 | copies++; |
5405be03 A |
2039 | break; |
2040 | case ZFS_REDUNDANT_METADATA_MOST: | |
2041 | if (level >= zfs_redundant_metadata_most_ditto_level || | |
2042 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)) | |
2043 | copies++; | |
2044 | break; | |
2045 | case ZFS_REDUNDANT_METADATA_SOME: | |
2046 | if (DMU_OT_IS_CRITICAL(type)) | |
2047 | copies++; | |
2048 | break; | |
2049 | case ZFS_REDUNDANT_METADATA_NONE: | |
2050 | break; | |
2051 | } | |
03c6040b GW |
2052 | } else if (wp & WP_NOFILL) { |
2053 | ASSERT(level == 0); | |
428870ff | 2054 | |
428870ff | 2055 | /* |
03c6040b GW |
2056 | * If we're writing preallocated blocks, we aren't actually |
2057 | * writing them so don't set any policy properties. These | |
2058 | * blocks are currently only used by an external subsystem | |
2059 | * outside of zfs (i.e. dump) and not written by the zio | |
2060 | * pipeline. | |
428870ff | 2061 | */ |
03c6040b GW |
2062 | compress = ZIO_COMPRESS_OFF; |
2063 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2064 | } else { |
99197f03 JG |
2065 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2066 | compress); | |
10b3c7f5 MN |
2067 | complevel = zio_complevel_select(os->os_spa, compress, |
2068 | complevel, complevel); | |
428870ff | 2069 | |
03c6040b GW |
2070 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2071 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2072 | dedup_checksum; | |
428870ff | 2073 | |
03c6040b GW |
2074 | /* |
2075 | * Determine dedup setting. If we are in dmu_sync(), | |
2076 | * we won't actually dedup now because that's all | |
2077 | * done in syncing context; but we do want to use the | |
e1cfd73f | 2078 | * dedup checksum. If the checksum is not strong |
03c6040b GW |
2079 | * enough to ensure unique signatures, force |
2080 | * dedup_verify. | |
2081 | */ | |
2082 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2083 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2084 | if (!(zio_checksum_table[checksum].ci_flags & |
2085 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2086 | dedup_verify = B_TRUE; |
2087 | } | |
428870ff | 2088 | |
03c6040b | 2089 | /* |
3c67d83a TH |
2090 | * Enable nopwrite if we have secure enough checksum |
2091 | * algorithm (see comment in zio_nop_write) and | |
2092 | * compression is enabled. We don't enable nopwrite if | |
2093 | * dedup is enabled as the two features are mutually | |
2094 | * exclusive. | |
03c6040b | 2095 | */ |
3c67d83a TH |
2096 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2097 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2098 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2099 | } |
2100 | ||
b5256303 TC |
2101 | /* |
2102 | * All objects in an encrypted objset are protected from modification | |
2103 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2104 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2105 | * not subject to nopwrite since writing the same data will still | |
2106 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2107 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2108 | * object types. | |
2109 | */ | |
2110 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2111 | encrypt = B_TRUE; | |
2112 | ||
2113 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2114 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2115 | nopwrite = B_FALSE; | |
2116 | } else { | |
2117 | dedup = B_FALSE; | |
2118 | } | |
2119 | ||
ae76f45c TC |
2120 | if (level <= 0 && |
2121 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2122 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2123 | } |
b5256303 | 2124 | } |
2aa34383 | 2125 | |
b5256303 | 2126 | zp->zp_compress = compress; |
10b3c7f5 | 2127 | zp->zp_complevel = complevel; |
b5256303 | 2128 | zp->zp_checksum = checksum; |
428870ff BB |
2129 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2130 | zp->zp_level = level; | |
faf0f58c | 2131 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2132 | zp->zp_dedup = dedup; |
2133 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2134 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2135 | zp->zp_encrypt = encrypt; |
2136 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
861166b0 AZ |
2137 | memset(zp->zp_salt, 0, ZIO_DATA_SALT_LEN); |
2138 | memset(zp->zp_iv, 0, ZIO_DATA_IV_LEN); | |
2139 | memset(zp->zp_mac, 0, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2140 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2141 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2142 | |
2143 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2144 | } |
2145 | ||
66aca247 | 2146 | /* |
64bfa6ba BB |
2147 | * Reports the location of data and holes in an object. In order to |
2148 | * accurately report holes all dirty data must be synced to disk. This | |
2149 | * causes extremely poor performance when seeking for holes in a dirty file. | |
2150 | * As a compromise, only provide hole data when the dnode is clean. When | |
2151 | * a dnode is dirty report the dnode as having no holes by returning EBUSY | |
2152 | * which is always safe to do. | |
66aca247 | 2153 | */ |
34dc7c2f BB |
2154 | int |
2155 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2156 | { | |
2157 | dnode_t *dn; | |
64bfa6ba | 2158 | int restarted = 0, err; |
34dc7c2f | 2159 | |
de198f2d | 2160 | restart: |
428870ff | 2161 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2162 | if (err) |
2163 | return (err); | |
66aca247 | 2164 | |
de198f2d | 2165 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
66aca247 | 2166 | |
de198f2d BB |
2167 | if (dnode_is_dirty(dn)) { |
2168 | /* | |
2169 | * If the zfs_dmu_offset_next_sync module option is enabled | |
64bfa6ba BB |
2170 | * then hole reporting has been requested. Dirty dnodes |
2171 | * must be synced to disk to accurately report holes. | |
de198f2d | 2172 | * |
64bfa6ba BB |
2173 | * Provided a RL_READER rangelock spanning 0-UINT64_MAX is |
2174 | * held by the caller only a single restart will be required. | |
2175 | * We tolerate callers which do not hold the rangelock by | |
2176 | * returning EBUSY and not reporting holes after one restart. | |
de198f2d BB |
2177 | */ |
2178 | if (zfs_dmu_offset_next_sync) { | |
2179 | rw_exit(&dn->dn_struct_rwlock); | |
2180 | dnode_rele(dn, FTAG); | |
64bfa6ba BB |
2181 | |
2182 | if (restarted) | |
2183 | return (SET_ERROR(EBUSY)); | |
2184 | ||
de198f2d | 2185 | txg_wait_synced(dmu_objset_pool(os), 0); |
64bfa6ba | 2186 | restarted = 1; |
de198f2d BB |
2187 | goto restart; |
2188 | } | |
34dc7c2f | 2189 | |
66aca247 | 2190 | err = SET_ERROR(EBUSY); |
de198f2d BB |
2191 | } else { |
2192 | err = dnode_next_offset(dn, DNODE_FIND_HAVELOCK | | |
2193 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2194 | } | |
66aca247 | 2195 | |
de198f2d | 2196 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
2197 | dnode_rele(dn, FTAG); |
2198 | ||
2199 | return (err); | |
67a1b037 PJD |
2200 | } |
2201 | ||
2202 | int | |
2203 | dmu_read_l0_bps(objset_t *os, uint64_t object, uint64_t offset, uint64_t length, | |
d0d91f18 | 2204 | blkptr_t *bps, size_t *nbpsp) |
67a1b037 PJD |
2205 | { |
2206 | dmu_buf_t **dbp, *dbuf; | |
2207 | dmu_buf_impl_t *db; | |
2208 | blkptr_t *bp; | |
2209 | int error, numbufs; | |
2210 | ||
2211 | error = dmu_buf_hold_array(os, object, offset, length, FALSE, FTAG, | |
2212 | &numbufs, &dbp); | |
2213 | if (error != 0) { | |
2214 | if (error == ESRCH) { | |
2215 | error = SET_ERROR(ENXIO); | |
2216 | } | |
2217 | return (error); | |
2218 | } | |
2219 | ||
2220 | ASSERT3U(numbufs, <=, *nbpsp); | |
2221 | ||
2222 | for (int i = 0; i < numbufs; i++) { | |
2223 | dbuf = dbp[i]; | |
2224 | db = (dmu_buf_impl_t *)dbuf; | |
ce0e1cc4 PJD |
2225 | |
2226 | mutex_enter(&db->db_mtx); | |
67a1b037 | 2227 | |
67a1b037 PJD |
2228 | if (!list_is_empty(&db->db_dirty_records)) { |
2229 | dbuf_dirty_record_t *dr; | |
2230 | ||
2231 | dr = list_head(&db->db_dirty_records); | |
2232 | if (dr->dt.dl.dr_brtwrite) { | |
2233 | /* | |
2234 | * This is very special case where we clone a | |
2235 | * block and in the same transaction group we | |
2236 | * read its BP (most likely to clone the clone). | |
2237 | */ | |
2238 | bp = &dr->dt.dl.dr_overridden_by; | |
2239 | } else { | |
2240 | /* | |
2241 | * The block was modified in the same | |
2242 | * transaction group. | |
2243 | */ | |
ce0e1cc4 | 2244 | mutex_exit(&db->db_mtx); |
67a1b037 PJD |
2245 | error = SET_ERROR(EAGAIN); |
2246 | goto out; | |
2247 | } | |
ce0e1cc4 PJD |
2248 | } else { |
2249 | bp = db->db_blkptr; | |
67a1b037 | 2250 | } |
ce0e1cc4 PJD |
2251 | |
2252 | mutex_exit(&db->db_mtx); | |
2253 | ||
67a1b037 PJD |
2254 | if (bp == NULL) { |
2255 | /* | |
2256 | * The block was created in this transaction group, | |
2257 | * so it has no BP yet. | |
2258 | */ | |
2259 | error = SET_ERROR(EAGAIN); | |
2260 | goto out; | |
2261 | } | |
67a1b037 PJD |
2262 | /* |
2263 | * Make sure we clone only data blocks. | |
2264 | */ | |
2265 | if (BP_IS_METADATA(bp) && !BP_IS_HOLE(bp)) { | |
2266 | error = SET_ERROR(EINVAL); | |
2267 | goto out; | |
2268 | } | |
2269 | ||
2270 | bps[i] = *bp; | |
2271 | } | |
2272 | ||
2273 | *nbpsp = numbufs; | |
2274 | out: | |
2275 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
2276 | ||
2277 | return (error); | |
2278 | } | |
2279 | ||
bd8c6bd6 | 2280 | int |
67a1b037 PJD |
2281 | dmu_brt_clone(objset_t *os, uint64_t object, uint64_t offset, uint64_t length, |
2282 | dmu_tx_t *tx, const blkptr_t *bps, size_t nbps, boolean_t replay) | |
2283 | { | |
2284 | spa_t *spa; | |
2285 | dmu_buf_t **dbp, *dbuf; | |
2286 | dmu_buf_impl_t *db; | |
2287 | struct dirty_leaf *dl; | |
2288 | dbuf_dirty_record_t *dr; | |
2289 | const blkptr_t *bp; | |
bd8c6bd6 | 2290 | int error = 0, i, numbufs; |
67a1b037 PJD |
2291 | |
2292 | spa = os->os_spa; | |
2293 | ||
2294 | VERIFY0(dmu_buf_hold_array(os, object, offset, length, FALSE, FTAG, | |
2295 | &numbufs, &dbp)); | |
2296 | ASSERT3U(nbps, ==, numbufs); | |
2297 | ||
bd8c6bd6 | 2298 | /* |
fbbe5e96 | 2299 | * Before we start cloning make sure that the dbufs sizes match new BPs |
bd8c6bd6 PJD |
2300 | * sizes. If they don't, that's a no-go, as we are not able to shrink |
2301 | * dbufs. | |
2302 | */ | |
2303 | for (i = 0; i < numbufs; i++) { | |
2304 | dbuf = dbp[i]; | |
2305 | db = (dmu_buf_impl_t *)dbuf; | |
2306 | bp = &bps[i]; | |
2307 | ||
2308 | ASSERT0(db->db_level); | |
2309 | ASSERT(db->db_blkid != DMU_BONUS_BLKID); | |
e6107668 | 2310 | ASSERT(db->db_blkid != DMU_SPILL_BLKID); |
bd8c6bd6 PJD |
2311 | |
2312 | if (!BP_IS_HOLE(bp) && BP_GET_LSIZE(bp) != dbuf->db_size) { | |
2313 | error = SET_ERROR(EXDEV); | |
2314 | goto out; | |
2315 | } | |
2316 | } | |
2317 | ||
2318 | for (i = 0; i < numbufs; i++) { | |
67a1b037 PJD |
2319 | dbuf = dbp[i]; |
2320 | db = (dmu_buf_impl_t *)dbuf; | |
2321 | bp = &bps[i]; | |
2322 | ||
2323 | ASSERT0(db->db_level); | |
2324 | ASSERT(db->db_blkid != DMU_BONUS_BLKID); | |
e6107668 | 2325 | ASSERT(db->db_blkid != DMU_SPILL_BLKID); |
67a1b037 PJD |
2326 | ASSERT(BP_IS_HOLE(bp) || dbuf->db_size == BP_GET_LSIZE(bp)); |
2327 | ||
555ef90c | 2328 | dmu_buf_will_clone(dbuf, tx); |
ce0e1cc4 PJD |
2329 | |
2330 | mutex_enter(&db->db_mtx); | |
2331 | ||
67a1b037 | 2332 | dr = list_head(&db->db_dirty_records); |
ce0e1cc4 | 2333 | VERIFY(dr != NULL); |
67a1b037 PJD |
2334 | ASSERT3U(dr->dr_txg, ==, tx->tx_txg); |
2335 | dl = &dr->dt.dl; | |
2336 | dl->dr_overridden_by = *bp; | |
2337 | dl->dr_brtwrite = B_TRUE; | |
67a1b037 PJD |
2338 | dl->dr_override_state = DR_OVERRIDDEN; |
2339 | if (BP_IS_HOLE(bp)) { | |
2340 | dl->dr_overridden_by.blk_birth = 0; | |
2341 | dl->dr_overridden_by.blk_phys_birth = 0; | |
2342 | } else { | |
2343 | dl->dr_overridden_by.blk_birth = dr->dr_txg; | |
c71fe716 PJD |
2344 | if (!BP_IS_EMBEDDED(bp)) { |
2345 | dl->dr_overridden_by.blk_phys_birth = | |
2346 | BP_PHYSICAL_BIRTH(bp); | |
2347 | } | |
67a1b037 PJD |
2348 | } |
2349 | ||
ce0e1cc4 PJD |
2350 | mutex_exit(&db->db_mtx); |
2351 | ||
67a1b037 PJD |
2352 | /* |
2353 | * When data in embedded into BP there is no need to create | |
2354 | * BRT entry as there is no data block. Just copy the BP as | |
2355 | * it contains the data. | |
2356 | * Also, when replaying ZIL we don't want to bump references | |
2357 | * in the BRT as it was already done during ZIL claim. | |
2358 | */ | |
2359 | if (!replay && !BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) { | |
2360 | brt_pending_add(spa, bp, tx); | |
2361 | } | |
2362 | } | |
bd8c6bd6 | 2363 | out: |
67a1b037 | 2364 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
bd8c6bd6 PJD |
2365 | |
2366 | return (error); | |
34dc7c2f BB |
2367 | } |
2368 | ||
2369 | void | |
e0b0ca98 | 2370 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2371 | { |
e0b0ca98 | 2372 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2373 | |
34dc7c2f BB |
2374 | doi->doi_data_block_size = dn->dn_datablksz; |
2375 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2376 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2377 | doi->doi_type = dn->dn_type; |
2378 | doi->doi_bonus_type = dn->dn_bonustype; | |
2379 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2380 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2381 | doi->doi_indirection = dn->dn_nlevels; |
2382 | doi->doi_checksum = dn->dn_checksum; | |
2383 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2384 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2385 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2386 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2387 | doi->doi_fill_count = 0; |
1c27024e | 2388 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2389 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2390 | } |
2391 | ||
2392 | void | |
2393 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2394 | { | |
2395 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2396 | mutex_enter(&dn->dn_mtx); | |
2397 | ||
2398 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2399 | |
2400 | mutex_exit(&dn->dn_mtx); | |
2401 | rw_exit(&dn->dn_struct_rwlock); | |
2402 | } | |
2403 | ||
2404 | /* | |
2405 | * Get information on a DMU object. | |
2406 | * If doi is NULL, just indicates whether the object exists. | |
2407 | */ | |
2408 | int | |
2409 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2410 | { | |
2411 | dnode_t *dn; | |
428870ff | 2412 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2413 | |
2414 | if (err) | |
2415 | return (err); | |
2416 | ||
2417 | if (doi != NULL) | |
2418 | dmu_object_info_from_dnode(dn, doi); | |
2419 | ||
2420 | dnode_rele(dn, FTAG); | |
2421 | return (0); | |
2422 | } | |
2423 | ||
2424 | /* | |
2425 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2426 | */ | |
2427 | void | |
572e2857 | 2428 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2429 | { |
572e2857 BB |
2430 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2431 | ||
2432 | DB_DNODE_ENTER(db); | |
2433 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2434 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2435 | } |
2436 | ||
2437 | /* | |
2438 | * Faster still when you only care about the size. | |
34dc7c2f BB |
2439 | */ |
2440 | void | |
572e2857 BB |
2441 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2442 | u_longlong_t *nblk512) | |
34dc7c2f | 2443 | { |
572e2857 BB |
2444 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2445 | dnode_t *dn; | |
2446 | ||
2447 | DB_DNODE_ENTER(db); | |
2448 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2449 | |
2450 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2451 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2452 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2453 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2454 | DB_DNODE_EXIT(db); | |
2455 | } | |
2456 | ||
2457 | void | |
2458 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2459 | { | |
2460 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2461 | dnode_t *dn; | |
2462 | ||
2463 | DB_DNODE_ENTER(db); | |
2464 | dn = DB_DNODE(db); | |
2465 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2466 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2467 | } |
2468 | ||
2469 | void | |
2470 | byteswap_uint64_array(void *vbuf, size_t size) | |
2471 | { | |
2472 | uint64_t *buf = vbuf; | |
2473 | size_t count = size >> 3; | |
2474 | int i; | |
2475 | ||
2476 | ASSERT((size & 7) == 0); | |
2477 | ||
2478 | for (i = 0; i < count; i++) | |
2479 | buf[i] = BSWAP_64(buf[i]); | |
2480 | } | |
2481 | ||
2482 | void | |
2483 | byteswap_uint32_array(void *vbuf, size_t size) | |
2484 | { | |
2485 | uint32_t *buf = vbuf; | |
2486 | size_t count = size >> 2; | |
2487 | int i; | |
2488 | ||
2489 | ASSERT((size & 3) == 0); | |
2490 | ||
2491 | for (i = 0; i < count; i++) | |
2492 | buf[i] = BSWAP_32(buf[i]); | |
2493 | } | |
2494 | ||
2495 | void | |
2496 | byteswap_uint16_array(void *vbuf, size_t size) | |
2497 | { | |
2498 | uint16_t *buf = vbuf; | |
2499 | size_t count = size >> 1; | |
2500 | int i; | |
2501 | ||
2502 | ASSERT((size & 1) == 0); | |
2503 | ||
2504 | for (i = 0; i < count; i++) | |
2505 | buf[i] = BSWAP_16(buf[i]); | |
2506 | } | |
2507 | ||
34dc7c2f BB |
2508 | void |
2509 | byteswap_uint8_array(void *vbuf, size_t size) | |
2510 | { | |
14e4e3cb | 2511 | (void) vbuf, (void) size; |
34dc7c2f BB |
2512 | } |
2513 | ||
2514 | void | |
2515 | dmu_init(void) | |
2516 | { | |
a6255b7f | 2517 | abd_init(); |
428870ff | 2518 | zfs_dbgmsg_init(); |
572e2857 | 2519 | sa_cache_init(); |
572e2857 | 2520 | dmu_objset_init(); |
34dc7c2f | 2521 | dnode_init(); |
428870ff | 2522 | zfetch_init(); |
570827e1 | 2523 | dmu_tx_init(); |
34dc7c2f | 2524 | l2arc_init(); |
29809a6c | 2525 | arc_init(); |
d3c2ae1c | 2526 | dbuf_init(); |
34dc7c2f BB |
2527 | } |
2528 | ||
2529 | void | |
2530 | dmu_fini(void) | |
2531 | { | |
e49f1e20 | 2532 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2533 | l2arc_fini(); |
570827e1 | 2534 | dmu_tx_fini(); |
428870ff | 2535 | zfetch_fini(); |
34dc7c2f | 2536 | dbuf_fini(); |
572e2857 BB |
2537 | dnode_fini(); |
2538 | dmu_objset_fini(); | |
428870ff BB |
2539 | sa_cache_fini(); |
2540 | zfs_dbgmsg_fini(); | |
a6255b7f | 2541 | abd_fini(); |
34dc7c2f | 2542 | } |
c28b2279 | 2543 | |
c28b2279 | 2544 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2545 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2546 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2547 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2548 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2549 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2550 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2551 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2552 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2553 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2554 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2555 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2556 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2557 | EXPORT_SYMBOL(dmu_object_info); |
2558 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2559 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2560 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2561 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2562 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2563 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2564 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2565 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2566 | EXPORT_SYMBOL(dmu_object_set_compress); | |
94dac3e8 | 2567 | EXPORT_SYMBOL(dmu_offset_next); |
57b650b8 BB |
2568 | EXPORT_SYMBOL(dmu_write_policy); |
2569 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2570 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2571 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2572 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2573 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2574 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2575 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2576 | |
03fdcb9a MM |
2577 | ZFS_MODULE_PARAM(zfs, zfs_, nopwrite_enabled, INT, ZMOD_RW, |
2578 | "Enable NOP writes"); | |
03c6040b | 2579 | |
ab8d9c17 | 2580 | ZFS_MODULE_PARAM(zfs, zfs_, per_txg_dirty_frees_percent, UINT, ZMOD_RW, |
03fdcb9a | 2581 | "Percentage of dirtied blocks from frees in one TXG"); |
66aca247 | 2582 | |
03fdcb9a | 2583 | ZFS_MODULE_PARAM(zfs, zfs_, dmu_offset_next_sync, INT, ZMOD_RW, |
66aca247 DB |
2584 | "Enable forcing txg sync to find holes"); |
2585 | ||
7ada752a | 2586 | /* CSTYLED */ |
fdc2d303 | 2587 | ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, UINT, ZMOD_RW, |
d9b4bf06 | 2588 | "Limit one prefetch call to this size"); |