]>
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 | ||
bdb7df42 | 168 | 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 | } | |
bdb7df42 | 188 | |
34dc7c2f | 189 | int |
9b67f605 | 190 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
a926aab9 | 191 | const void *tag, dmu_buf_t **dbp) |
34dc7c2f BB |
192 | { |
193 | dnode_t *dn; | |
194 | uint64_t blkid; | |
195 | dmu_buf_impl_t *db; | |
196 | int err; | |
428870ff BB |
197 | |
198 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
199 | if (err) |
200 | return (err); | |
34dc7c2f | 201 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
f664f1ee | 202 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
203 | db = dbuf_hold(dn, blkid, tag); |
204 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
205 | dnode_rele(dn, FTAG); |
206 | ||
34dc7c2f | 207 | if (db == NULL) { |
9b67f605 MA |
208 | *dbp = NULL; |
209 | return (SET_ERROR(EIO)); | |
210 | } | |
211 | ||
212 | *dbp = &db->db; | |
213 | return (err); | |
214 | } | |
215 | ||
2bce8049 MA |
216 | int |
217 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
a926aab9 | 218 | const void *tag, dmu_buf_t **dbp, int flags) |
2bce8049 MA |
219 | { |
220 | int err; | |
221 | int db_flags = DB_RF_CANFAIL; | |
222 | ||
223 | if (flags & DMU_READ_NO_PREFETCH) | |
224 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
225 | if (flags & DMU_READ_NO_DECRYPT) |
226 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
227 | |
228 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
229 | if (err == 0) { | |
230 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
231 | err = dbuf_read(db, NULL, db_flags); | |
232 | if (err != 0) { | |
233 | dbuf_rele(db, tag); | |
234 | *dbp = NULL; | |
235 | } | |
236 | } | |
237 | ||
238 | return (err); | |
239 | } | |
240 | ||
9b67f605 MA |
241 | int |
242 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
a926aab9 | 243 | const void *tag, dmu_buf_t **dbp, int flags) |
9b67f605 MA |
244 | { |
245 | int err; | |
246 | int db_flags = DB_RF_CANFAIL; | |
247 | ||
248 | if (flags & DMU_READ_NO_PREFETCH) | |
249 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
250 | if (flags & DMU_READ_NO_DECRYPT) |
251 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
252 | |
253 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
254 | if (err == 0) { | |
255 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 256 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 257 | if (err != 0) { |
34dc7c2f | 258 | dbuf_rele(db, tag); |
9b67f605 | 259 | *dbp = NULL; |
34dc7c2f BB |
260 | } |
261 | } | |
262 | ||
34dc7c2f BB |
263 | return (err); |
264 | } | |
265 | ||
266 | int | |
267 | dmu_bonus_max(void) | |
268 | { | |
50c957f7 | 269 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
270 | } |
271 | ||
272 | int | |
572e2857 | 273 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 274 | { |
572e2857 BB |
275 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
276 | dnode_t *dn; | |
277 | int error; | |
34dc7c2f | 278 | |
572e2857 BB |
279 | DB_DNODE_ENTER(db); |
280 | dn = DB_DNODE(db); | |
281 | ||
282 | if (dn->dn_bonus != db) { | |
2e528b49 | 283 | error = SET_ERROR(EINVAL); |
572e2857 | 284 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 285 | error = SET_ERROR(EINVAL); |
572e2857 BB |
286 | } else { |
287 | dnode_setbonuslen(dn, newsize, tx); | |
288 | error = 0; | |
289 | } | |
290 | ||
291 | DB_DNODE_EXIT(db); | |
292 | return (error); | |
34dc7c2f BB |
293 | } |
294 | ||
428870ff | 295 | int |
572e2857 | 296 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 297 | { |
572e2857 BB |
298 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
299 | dnode_t *dn; | |
300 | int error; | |
428870ff | 301 | |
572e2857 BB |
302 | DB_DNODE_ENTER(db); |
303 | dn = DB_DNODE(db); | |
428870ff | 304 | |
9ae529ec | 305 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 306 | error = SET_ERROR(EINVAL); |
572e2857 | 307 | } else if (dn->dn_bonus != db) { |
2e528b49 | 308 | error = SET_ERROR(EINVAL); |
572e2857 BB |
309 | } else { |
310 | dnode_setbonus_type(dn, type, tx); | |
311 | error = 0; | |
312 | } | |
428870ff | 313 | |
572e2857 BB |
314 | DB_DNODE_EXIT(db); |
315 | return (error); | |
316 | } | |
317 | ||
318 | dmu_object_type_t | |
319 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
320 | { | |
321 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
322 | dnode_t *dn; | |
323 | dmu_object_type_t type; | |
324 | ||
325 | DB_DNODE_ENTER(db); | |
326 | dn = DB_DNODE(db); | |
327 | type = dn->dn_bonustype; | |
328 | DB_DNODE_EXIT(db); | |
329 | ||
330 | return (type); | |
428870ff BB |
331 | } |
332 | ||
333 | int | |
334 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
335 | { | |
336 | dnode_t *dn; | |
337 | int error; | |
338 | ||
339 | error = dnode_hold(os, object, FTAG, &dn); | |
340 | dbuf_rm_spill(dn, tx); | |
341 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
342 | dnode_rm_spill(dn, tx); | |
343 | rw_exit(&dn->dn_struct_rwlock); | |
344 | dnode_rele(dn, FTAG); | |
345 | return (error); | |
346 | } | |
347 | ||
34dc7c2f | 348 | /* |
6955b401 BB |
349 | * Lookup and hold the bonus buffer for the provided dnode. If the dnode |
350 | * has not yet been allocated a new bonus dbuf a will be allocated. | |
351 | * Returns ENOENT, EIO, or 0. | |
34dc7c2f | 352 | */ |
a926aab9 | 353 | int dmu_bonus_hold_by_dnode(dnode_t *dn, const void *tag, dmu_buf_t **dbp, |
6955b401 | 354 | uint32_t flags) |
34dc7c2f | 355 | { |
34dc7c2f BB |
356 | dmu_buf_impl_t *db; |
357 | int error; | |
b5256303 TC |
358 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
359 | ||
360 | if (flags & DMU_READ_NO_PREFETCH) | |
361 | db_flags |= DB_RF_NOPREFETCH; | |
362 | if (flags & DMU_READ_NO_DECRYPT) | |
363 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 364 | |
34dc7c2f BB |
365 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
366 | if (dn->dn_bonus == NULL) { | |
c9e39da9 RY |
367 | if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { |
368 | rw_exit(&dn->dn_struct_rwlock); | |
369 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
370 | } | |
34dc7c2f BB |
371 | if (dn->dn_bonus == NULL) |
372 | dbuf_create_bonus(dn); | |
373 | } | |
374 | db = dn->dn_bonus; | |
34dc7c2f BB |
375 | |
376 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 377 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 378 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 379 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
380 | } |
381 | ||
382 | /* | |
383 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
384 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
385 | * a dnode hold for every dbuf. | |
386 | */ | |
387 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f | 388 | |
b5256303 TC |
389 | error = dbuf_read(db, NULL, db_flags); |
390 | if (error) { | |
391 | dnode_evict_bonus(dn); | |
392 | dbuf_rele(db, tag); | |
393 | *dbp = NULL; | |
394 | return (error); | |
395 | } | |
34dc7c2f BB |
396 | |
397 | *dbp = &db->db; | |
398 | return (0); | |
399 | } | |
400 | ||
b5256303 | 401 | int |
a926aab9 | 402 | dmu_bonus_hold(objset_t *os, uint64_t object, const void *tag, dmu_buf_t **dbp) |
b5256303 | 403 | { |
6955b401 BB |
404 | dnode_t *dn; |
405 | int error; | |
406 | ||
407 | error = dnode_hold(os, object, FTAG, &dn); | |
408 | if (error) | |
409 | return (error); | |
410 | ||
411 | error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH); | |
412 | dnode_rele(dn, FTAG); | |
413 | ||
414 | return (error); | |
b5256303 TC |
415 | } |
416 | ||
428870ff BB |
417 | /* |
418 | * returns ENOENT, EIO, or 0. | |
419 | * | |
420 | * This interface will allocate a blank spill dbuf when a spill blk | |
421 | * doesn't already exist on the dnode. | |
422 | * | |
423 | * if you only want to find an already existing spill db, then | |
424 | * dmu_spill_hold_existing() should be used. | |
425 | */ | |
426 | int | |
a926aab9 AZ |
427 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, const void *tag, |
428 | dmu_buf_t **dbp) | |
428870ff BB |
429 | { |
430 | dmu_buf_impl_t *db = NULL; | |
431 | int err; | |
432 | ||
433 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
434 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
435 | ||
436 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
437 | ||
438 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
439 | rw_exit(&dn->dn_struct_rwlock); | |
440 | ||
b182ac00 | 441 | if (db == NULL) { |
442 | *dbp = NULL; | |
443 | return (SET_ERROR(EIO)); | |
444 | } | |
572e2857 BB |
445 | err = dbuf_read(db, NULL, flags); |
446 | if (err == 0) | |
447 | *dbp = &db->db; | |
b182ac00 | 448 | else { |
572e2857 | 449 | dbuf_rele(db, tag); |
b182ac00 | 450 | *dbp = NULL; |
451 | } | |
428870ff BB |
452 | return (err); |
453 | } | |
454 | ||
455 | int | |
a926aab9 | 456 | dmu_spill_hold_existing(dmu_buf_t *bonus, const void *tag, dmu_buf_t **dbp) |
428870ff | 457 | { |
572e2857 BB |
458 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
459 | dnode_t *dn; | |
428870ff BB |
460 | int err; |
461 | ||
572e2857 BB |
462 | DB_DNODE_ENTER(db); |
463 | dn = DB_DNODE(db); | |
464 | ||
465 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 466 | err = SET_ERROR(EINVAL); |
572e2857 BB |
467 | } else { |
468 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
469 | ||
470 | if (!dn->dn_have_spill) { | |
2e528b49 | 471 | err = SET_ERROR(ENOENT); |
572e2857 BB |
472 | } else { |
473 | err = dmu_spill_hold_by_dnode(dn, | |
474 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
475 | } | |
428870ff | 476 | |
428870ff | 477 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 478 | } |
572e2857 BB |
479 | |
480 | DB_DNODE_EXIT(db); | |
428870ff BB |
481 | return (err); |
482 | } | |
483 | ||
484 | int | |
a926aab9 | 485 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, const void *tag, |
e7504d7a | 486 | dmu_buf_t **dbp) |
428870ff | 487 | { |
572e2857 BB |
488 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
489 | dnode_t *dn; | |
490 | int err; | |
e7504d7a TC |
491 | uint32_t db_flags = DB_RF_CANFAIL; |
492 | ||
493 | if (flags & DMU_READ_NO_DECRYPT) | |
494 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
495 | |
496 | DB_DNODE_ENTER(db); | |
497 | dn = DB_DNODE(db); | |
e7504d7a | 498 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
499 | DB_DNODE_EXIT(db); |
500 | ||
501 | return (err); | |
428870ff BB |
502 | } |
503 | ||
34dc7c2f BB |
504 | /* |
505 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
506 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
507 | * and can induce severe lock contention when writing to several files | |
508 | * whose dnodes are in the same block. | |
509 | */ | |
af1698f5 | 510 | int |
9babb374 | 511 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
a926aab9 AZ |
512 | boolean_t read, const void *tag, int *numbufsp, dmu_buf_t ***dbpp, |
513 | uint32_t flags) | |
34dc7c2f BB |
514 | { |
515 | dmu_buf_t **dbp; | |
891568c9 | 516 | zstream_t *zs = NULL; |
34dc7c2f | 517 | uint64_t blkid, nblks, i; |
9babb374 | 518 | uint32_t dbuf_flags; |
34dc7c2f | 519 | int err; |
ec50cd24 | 520 | zio_t *zio = NULL; |
891568c9 | 521 | boolean_t missed = B_FALSE; |
34dc7c2f | 522 | |
67a1b037 | 523 | ASSERT(!read || length <= DMU_MAX_ACCESS); |
34dc7c2f | 524 | |
7f60329a MA |
525 | /* |
526 | * Note: We directly notify the prefetch code of this read, so that | |
527 | * we can tell it about the multi-block read. dbuf_read() only knows | |
528 | * about the one block it is accessing. | |
529 | */ | |
530 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
531 | DB_RF_NOPREFETCH; | |
34dc7c2f | 532 | |
e8cf3a4f AP |
533 | if ((flags & DMU_READ_NO_DECRYPT) != 0) |
534 | dbuf_flags |= DB_RF_NO_DECRYPT; | |
535 | ||
34dc7c2f BB |
536 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
537 | if (dn->dn_datablkshift) { | |
538 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
539 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
540 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
541 | } else { |
542 | if (offset + length > dn->dn_datablksz) { | |
543 | zfs_panic_recover("zfs: accessing past end of object " | |
544 | "%llx/%llx (size=%u access=%llu+%llu)", | |
545 | (longlong_t)dn->dn_objset-> | |
546 | os_dsl_dataset->ds_object, | |
547 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
548 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 549 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 550 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
551 | } |
552 | nblks = 1; | |
553 | } | |
79c76d5b | 554 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 555 | |
ec50cd24 FY |
556 | if (read) |
557 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, | |
558 | ZIO_FLAG_CANFAIL); | |
fcff0f35 | 559 | blkid = dbuf_whichblock(dn, 0, offset); |
34b3d498 | 560 | if ((flags & DMU_READ_NO_PREFETCH) == 0) { |
891568c9 AM |
561 | /* |
562 | * Prepare the zfetch before initiating the demand reads, so | |
563 | * that if multiple threads block on same indirect block, we | |
564 | * base predictions on the original less racy request order. | |
565 | */ | |
ed2f7ba0 AM |
566 | zs = dmu_zfetch_prepare(&dn->dn_zfetch, blkid, nblks, read, |
567 | B_TRUE); | |
891568c9 | 568 | } |
34dc7c2f | 569 | for (i = 0; i < nblks; i++) { |
7f60329a | 570 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f | 571 | if (db == NULL) { |
891568c9 AM |
572 | if (zs) |
573 | dmu_zfetch_run(zs, missed, B_TRUE); | |
34dc7c2f BB |
574 | rw_exit(&dn->dn_struct_rwlock); |
575 | dmu_buf_rele_array(dbp, nblks, tag); | |
ec50cd24 FY |
576 | if (read) |
577 | zio_nowait(zio); | |
2e528b49 | 578 | return (SET_ERROR(EIO)); |
34dc7c2f | 579 | } |
7f60329a | 580 | |
891568c9 AM |
581 | /* |
582 | * Initiate async demand data read. | |
583 | * We check the db_state after calling dbuf_read() because | |
584 | * (1) dbuf_read() may change the state to CACHED due to a | |
585 | * hit in the ARC, and (2) on a cache miss, a child will | |
586 | * have been added to "zio" but not yet completed, so the | |
587 | * state will not yet be CACHED. | |
588 | */ | |
589 | if (read) { | |
ed2f7ba0 AM |
590 | if (i == nblks - 1 && blkid + i < dn->dn_maxblkid && |
591 | offset + length < db->db.db_offset + | |
592 | db->db.db_size) { | |
593 | if (offset <= db->db.db_offset) | |
594 | dbuf_flags |= DB_RF_PARTIAL_FIRST; | |
595 | else | |
596 | dbuf_flags |= DB_RF_PARTIAL_MORE; | |
597 | } | |
9babb374 | 598 | (void) dbuf_read(db, zio, dbuf_flags); |
891568c9 AM |
599 | if (db->db_state != DB_CACHED) |
600 | missed = B_TRUE; | |
601 | } | |
34dc7c2f BB |
602 | dbp[i] = &db->db; |
603 | } | |
7f60329a | 604 | |
64e0fe14 RM |
605 | if (!read) |
606 | zfs_racct_write(length, nblks); | |
607 | ||
891568c9 AM |
608 | if (zs) |
609 | dmu_zfetch_run(zs, missed, B_TRUE); | |
34dc7c2f BB |
610 | rw_exit(&dn->dn_struct_rwlock); |
611 | ||
34dc7c2f | 612 | if (read) { |
ec50cd24 FY |
613 | /* wait for async read i/o */ |
614 | err = zio_wait(zio); | |
615 | if (err) { | |
616 | dmu_buf_rele_array(dbp, nblks, tag); | |
617 | return (err); | |
618 | } | |
619 | ||
620 | /* wait for other io to complete */ | |
34dc7c2f BB |
621 | for (i = 0; i < nblks; i++) { |
622 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
623 | mutex_enter(&db->db_mtx); | |
624 | while (db->db_state == DB_READ || | |
625 | db->db_state == DB_FILL) | |
626 | cv_wait(&db->db_changed, &db->db_mtx); | |
627 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 628 | err = SET_ERROR(EIO); |
34dc7c2f BB |
629 | mutex_exit(&db->db_mtx); |
630 | if (err) { | |
631 | dmu_buf_rele_array(dbp, nblks, tag); | |
632 | return (err); | |
633 | } | |
634 | } | |
635 | } | |
636 | ||
637 | *numbufsp = nblks; | |
638 | *dbpp = dbp; | |
639 | return (0); | |
640 | } | |
641 | ||
afbc6179 | 642 | int |
34dc7c2f | 643 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, |
a926aab9 AZ |
644 | uint64_t length, int read, const void *tag, int *numbufsp, |
645 | dmu_buf_t ***dbpp) | |
34dc7c2f BB |
646 | { |
647 | dnode_t *dn; | |
648 | int err; | |
649 | ||
428870ff | 650 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
651 | if (err) |
652 | return (err); | |
653 | ||
654 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 655 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
656 | |
657 | dnode_rele(dn, FTAG); | |
658 | ||
659 | return (err); | |
660 | } | |
661 | ||
662 | int | |
572e2857 | 663 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
a926aab9 | 664 | uint64_t length, boolean_t read, const void *tag, int *numbufsp, |
7f60329a | 665 | dmu_buf_t ***dbpp) |
34dc7c2f | 666 | { |
572e2857 BB |
667 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
668 | dnode_t *dn; | |
34dc7c2f BB |
669 | int err; |
670 | ||
572e2857 BB |
671 | DB_DNODE_ENTER(db); |
672 | dn = DB_DNODE(db); | |
34dc7c2f | 673 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 674 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 675 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
676 | |
677 | return (err); | |
678 | } | |
679 | ||
680 | void | |
a926aab9 | 681 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, const void *tag) |
34dc7c2f BB |
682 | { |
683 | int i; | |
684 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
685 | ||
686 | if (numbufs == 0) | |
687 | return; | |
688 | ||
689 | for (i = 0; i < numbufs; i++) { | |
690 | if (dbp[i]) | |
691 | dbuf_rele(dbp[i], tag); | |
692 | } | |
693 | ||
694 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
695 | } | |
696 | ||
e8b96c60 | 697 | /* |
6c94e649 | 698 | * Issue prefetch I/Os for the given blocks. If level is greater than 0, the |
e1cfd73f | 699 | * indirect blocks prefetched will be those that point to the blocks containing |
6c94e649 AM |
700 | * the data starting at offset, and continuing to offset + len. If the range |
701 | * it too long, prefetch the first dmu_prefetch_max bytes as requested, while | |
702 | * for the rest only a higher level, also fitting within dmu_prefetch_max. It | |
703 | * should primarily help random reads, since for long sequential reads there is | |
704 | * a speculative prefetcher. | |
e8b96c60 | 705 | * |
b5256303 | 706 | * Note that if the indirect blocks above the blocks being prefetched are not |
6c94e649 AM |
707 | * in cache, they will be asynchronously read in. Dnode read by dnode_hold() |
708 | * is currently synchronous. | |
e8b96c60 | 709 | */ |
34dc7c2f | 710 | void |
fcff0f35 PD |
711 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
712 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
713 | { |
714 | dnode_t *dn; | |
6c94e649 AM |
715 | int64_t level2 = level; |
716 | uint64_t start, end, start2, end2; | |
34dc7c2f | 717 | |
6c94e649 AM |
718 | if (dmu_prefetch_max == 0 || len == 0) { |
719 | dmu_prefetch_dnode(os, object, pri); | |
34dc7c2f BB |
720 | return; |
721 | } | |
722 | ||
6c94e649 | 723 | if (dnode_hold(os, object, FTAG, &dn) != 0) |
34dc7c2f BB |
724 | return; |
725 | ||
fcff0f35 | 726 | /* |
6c94e649 AM |
727 | * Depending on len we may do two prefetches: blocks [start, end) at |
728 | * level, and following blocks [start2, end2) at higher level2. | |
fcff0f35 | 729 | */ |
f664f1ee | 730 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
6c94e649 AM |
731 | if (dn->dn_datablkshift != 0) { |
732 | /* | |
733 | * The object has multiple blocks. Calculate the full range | |
734 | * of blocks [start, end2) and then split it into two parts, | |
735 | * so that the first [start, end) fits into dmu_prefetch_max. | |
736 | */ | |
737 | start = dbuf_whichblock(dn, level, offset); | |
738 | end2 = dbuf_whichblock(dn, level, offset + len - 1) + 1; | |
739 | uint8_t ibs = dn->dn_indblkshift; | |
740 | uint8_t bs = (level == 0) ? dn->dn_datablkshift : ibs; | |
741 | uint_t limit = P2ROUNDUP(dmu_prefetch_max, 1 << bs) >> bs; | |
742 | start2 = end = MIN(end2, start + limit); | |
743 | ||
744 | /* | |
745 | * Find level2 where [start2, end2) fits into dmu_prefetch_max. | |
746 | */ | |
747 | uint8_t ibps = ibs - SPA_BLKPTRSHIFT; | |
748 | limit = P2ROUNDUP(dmu_prefetch_max, 1 << ibs) >> ibs; | |
749 | do { | |
750 | level2++; | |
751 | start2 = P2ROUNDUP(start2, 1 << ibps) >> ibps; | |
752 | end2 = P2ROUNDUP(end2, 1 << ibps) >> ibps; | |
753 | } while (end2 - start2 > limit); | |
34dc7c2f | 754 | } else { |
6c94e649 AM |
755 | /* There is only one block. Prefetch it or nothing. */ |
756 | start = start2 = end2 = 0; | |
757 | end = start + (level == 0 && offset < dn->dn_datablksz); | |
34dc7c2f BB |
758 | } |
759 | ||
6c94e649 AM |
760 | for (uint64_t i = start; i < end; i++) |
761 | dbuf_prefetch(dn, level, i, pri, 0); | |
762 | for (uint64_t i = start2; i < end2; i++) | |
763 | dbuf_prefetch(dn, level2, i, pri, 0); | |
34dc7c2f BB |
764 | rw_exit(&dn->dn_struct_rwlock); |
765 | ||
766 | dnode_rele(dn, FTAG); | |
767 | } | |
768 | ||
6c94e649 AM |
769 | /* |
770 | * Issue prefetch I/Os for the given object's dnode. | |
771 | */ | |
772 | void | |
773 | dmu_prefetch_dnode(objset_t *os, uint64_t object, zio_priority_t pri) | |
774 | { | |
775 | if (object == 0 || object >= DN_MAX_OBJECT) | |
776 | return; | |
777 | ||
778 | dnode_t *dn = DMU_META_DNODE(os); | |
779 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
780 | uint64_t blkid = dbuf_whichblock(dn, 0, object * sizeof (dnode_phys_t)); | |
781 | dbuf_prefetch(dn, 0, blkid, pri, 0); | |
782 | rw_exit(&dn->dn_struct_rwlock); | |
783 | } | |
784 | ||
45d1cae3 BB |
785 | /* |
786 | * Get the next "chunk" of file data to free. We traverse the file from | |
787 | * the end so that the file gets shorter over time (if we crashes in the | |
788 | * middle, this will leave us in a better state). We find allocated file | |
789 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
790 | * |
791 | * On input, *start should be the first offset that does not need to be | |
792 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
793 | * offset that should be freed and l1blks is set to the number of level 1 |
794 | * indirect blocks found within the chunk. | |
45d1cae3 | 795 | */ |
b128c09f | 796 | static int |
65282ee9 | 797 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 798 | { |
65282ee9 | 799 | uint64_t blks; |
b663a23d MA |
800 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
801 | /* bytes of data covered by a level-1 indirect block */ | |
ec4afd27 OM |
802 | uint64_t iblkrange = (uint64_t)dn->dn_datablksz * |
803 | EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
b128c09f | 804 | |
b663a23d | 805 | ASSERT3U(minimum, <=, *start); |
b128c09f | 806 | |
f4c594da TC |
807 | /* |
808 | * Check if we can free the entire range assuming that all of the | |
809 | * L1 blocks in this range have data. If we can, we use this | |
810 | * worst case value as an estimate so we can avoid having to look | |
811 | * at the object's actual data. | |
812 | */ | |
813 | uint64_t total_l1blks = | |
814 | (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / | |
815 | iblkrange; | |
816 | if (total_l1blks <= maxblks) { | |
817 | *l1blks = total_l1blks; | |
b663a23d | 818 | *start = minimum; |
b128c09f BB |
819 | return (0); |
820 | } | |
45d1cae3 | 821 | ASSERT(ISP2(iblkrange)); |
b128c09f | 822 | |
65282ee9 | 823 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
824 | int err; |
825 | ||
b663a23d MA |
826 | /* |
827 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
828 | * indirect block at or before the input offset. We must | |
829 | * decrement *start so that it is at the end of the region | |
830 | * to search. | |
831 | */ | |
832 | (*start)--; | |
f4c594da | 833 | |
b128c09f | 834 | err = dnode_next_offset(dn, |
45d1cae3 | 835 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 836 | |
b663a23d | 837 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 838 | if (err == ESRCH) { |
b663a23d MA |
839 | *start = minimum; |
840 | break; | |
841 | } else if (err != 0) { | |
65282ee9 | 842 | *l1blks = blks; |
b128c09f | 843 | return (err); |
45d1cae3 | 844 | } |
b128c09f | 845 | |
b663a23d | 846 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 847 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 848 | } |
b663a23d MA |
849 | if (*start < minimum) |
850 | *start = minimum; | |
65282ee9 | 851 | *l1blks = blks; |
f4c594da | 852 | |
b128c09f BB |
853 | return (0); |
854 | } | |
855 | ||
a08abc1b GM |
856 | /* |
857 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
858 | * otherwise return false. | |
859 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
860 | */ | |
a08abc1b GM |
861 | static boolean_t |
862 | dmu_objset_zfs_unmounting(objset_t *os) | |
863 | { | |
864 | #ifdef _KERNEL | |
865 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
866 | return (zfs_get_vfs_flag_unmounted(os)); | |
14e4e3cb AZ |
867 | #else |
868 | (void) os; | |
a08abc1b GM |
869 | #endif |
870 | return (B_FALSE); | |
871 | } | |
872 | ||
b128c09f BB |
873 | static int |
874 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 875 | uint64_t length) |
b128c09f | 876 | { |
c97d3069 | 877 | uint64_t object_size; |
b663a23d | 878 | int err; |
539d33c7 GM |
879 | uint64_t dirty_frees_threshold; |
880 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 881 | |
c97d3069 BB |
882 | if (dn == NULL) |
883 | return (SET_ERROR(EINVAL)); | |
884 | ||
885 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 886 | if (offset >= object_size) |
b128c09f | 887 | return (0); |
b128c09f | 888 | |
539d33c7 GM |
889 | if (zfs_per_txg_dirty_frees_percent <= 100) |
890 | dirty_frees_threshold = | |
891 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
892 | else | |
65282ee9 | 893 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 894 | |
b663a23d MA |
895 | if (length == DMU_OBJECT_END || offset + length > object_size) |
896 | length = object_size - offset; | |
897 | ||
898 | while (length != 0) { | |
539d33c7 | 899 | uint64_t chunk_end, chunk_begin, chunk_len; |
65282ee9 | 900 | uint64_t l1blks; |
b663a23d MA |
901 | dmu_tx_t *tx; |
902 | ||
a08abc1b GM |
903 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
904 | return (SET_ERROR(EINTR)); | |
905 | ||
b663a23d MA |
906 | chunk_end = chunk_begin = offset + length; |
907 | ||
908 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 909 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
910 | if (err) |
911 | return (err); | |
b663a23d MA |
912 | ASSERT3U(chunk_begin, >=, offset); |
913 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 914 | |
539d33c7 GM |
915 | chunk_len = chunk_end - chunk_begin; |
916 | ||
b128c09f | 917 | tx = dmu_tx_create(os); |
539d33c7 | 918 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
919 | |
920 | /* | |
921 | * Mark this transaction as typically resulting in a net | |
922 | * reduction in space used. | |
923 | */ | |
924 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
925 | err = dmu_tx_assign(tx, TXG_WAIT); |
926 | if (err) { | |
927 | dmu_tx_abort(tx); | |
928 | return (err); | |
929 | } | |
539d33c7 | 930 | |
f4c594da TC |
931 | uint64_t txg = dmu_tx_get_txg(tx); |
932 | ||
933 | mutex_enter(&dp->dp_lock); | |
934 | uint64_t long_free_dirty = | |
935 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; | |
936 | mutex_exit(&dp->dp_lock); | |
937 | ||
938 | /* | |
939 | * To avoid filling up a TXG with just frees, wait for | |
940 | * the next TXG to open before freeing more chunks if | |
941 | * we have reached the threshold of frees. | |
942 | */ | |
943 | if (dirty_frees_threshold != 0 && | |
944 | long_free_dirty >= dirty_frees_threshold) { | |
945 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); | |
946 | dmu_tx_commit(tx); | |
947 | txg_wait_open(dp, 0, B_TRUE); | |
948 | continue; | |
949 | } | |
950 | ||
65282ee9 AP |
951 | /* |
952 | * In order to prevent unnecessary write throttling, for each | |
953 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
954 | * in dnode_free_range() below. We compare this number to a | |
955 | * tunable threshold, past which we prevent new L1 dirty freeing | |
956 | * blocks from being added into the open TXG. See | |
957 | * dmu_free_long_range_impl() for details. The threshold | |
958 | * prevents write throttle activation due to dirty freeing L1 | |
959 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
960 | */ | |
539d33c7 | 961 | mutex_enter(&dp->dp_lock); |
f4c594da | 962 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += |
65282ee9 | 963 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
964 | mutex_exit(&dp->dp_lock); |
965 | DTRACE_PROBE3(free__long__range, | |
f4c594da TC |
966 | uint64_t, long_free_dirty, uint64_t, chunk_len, |
967 | uint64_t, txg); | |
539d33c7 | 968 | dnode_free_range(dn, chunk_begin, chunk_len, tx); |
440a3eb9 | 969 | |
b128c09f | 970 | dmu_tx_commit(tx); |
b663a23d | 971 | |
539d33c7 | 972 | length -= chunk_len; |
b128c09f BB |
973 | } |
974 | return (0); | |
975 | } | |
976 | ||
977 | int | |
978 | dmu_free_long_range(objset_t *os, uint64_t object, | |
979 | uint64_t offset, uint64_t length) | |
980 | { | |
981 | dnode_t *dn; | |
982 | int err; | |
983 | ||
428870ff | 984 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
985 | if (err != 0) |
986 | return (err); | |
0c03d21a | 987 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
988 | |
989 | /* | |
990 | * It is important to zero out the maxblkid when freeing the entire | |
991 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
992 | * will take the fast path, and (b) dnode_reallocate() can verify | |
993 | * that the entire file has been freed. | |
994 | */ | |
b0bc7a84 | 995 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
996 | dn->dn_maxblkid = 0; |
997 | ||
b128c09f BB |
998 | dnode_rele(dn, FTAG); |
999 | return (err); | |
1000 | } | |
1001 | ||
1002 | int | |
0c03d21a | 1003 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 1004 | { |
b128c09f BB |
1005 | dmu_tx_t *tx; |
1006 | int err; | |
1007 | ||
b663a23d | 1008 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
1009 | if (err != 0) |
1010 | return (err); | |
b663a23d MA |
1011 | |
1012 | tx = dmu_tx_create(os); | |
1013 | dmu_tx_hold_bonus(tx, object); | |
1014 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 1015 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
1016 | err = dmu_tx_assign(tx, TXG_WAIT); |
1017 | if (err == 0) { | |
60b618a9 | 1018 | err = dmu_object_free(os, object, tx); |
b663a23d | 1019 | dmu_tx_commit(tx); |
b128c09f | 1020 | } else { |
b663a23d | 1021 | dmu_tx_abort(tx); |
b128c09f | 1022 | } |
b663a23d | 1023 | |
b128c09f BB |
1024 | return (err); |
1025 | } | |
1026 | ||
34dc7c2f BB |
1027 | int |
1028 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
1029 | uint64_t size, dmu_tx_t *tx) | |
1030 | { | |
1031 | dnode_t *dn; | |
428870ff | 1032 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1033 | if (err) |
1034 | return (err); | |
1035 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 1036 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
1037 | dnode_free_range(dn, offset, size, tx); |
1038 | dnode_rele(dn, FTAG); | |
1039 | return (0); | |
1040 | } | |
1041 | ||
0eef1bde | 1042 | static int |
1043 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 1044 | void *buf, uint32_t flags) |
34dc7c2f | 1045 | { |
34dc7c2f | 1046 | dmu_buf_t **dbp; |
0eef1bde | 1047 | int numbufs, err = 0; |
34dc7c2f BB |
1048 | |
1049 | /* | |
1050 | * Deal with odd block sizes, where there can't be data past the first | |
1051 | * block. If we ever do the tail block optimization, we will need to | |
1052 | * handle that here as well. | |
1053 | */ | |
45d1cae3 | 1054 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 1055 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f | 1056 | MIN(size, dn->dn_datablksz - offset); |
861166b0 | 1057 | memset((char *)buf + newsz, 0, size - newsz); |
34dc7c2f BB |
1058 | size = newsz; |
1059 | } | |
1060 | ||
1061 | while (size > 0) { | |
1062 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 1063 | int i; |
34dc7c2f BB |
1064 | |
1065 | /* | |
1066 | * NB: we could do this block-at-a-time, but it's nice | |
1067 | * to be reading in parallel. | |
1068 | */ | |
1069 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 1070 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
1071 | if (err) |
1072 | break; | |
1073 | ||
1074 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1075 | uint64_t tocpy; |
1076 | int64_t bufoff; | |
34dc7c2f BB |
1077 | dmu_buf_t *db = dbp[i]; |
1078 | ||
1079 | ASSERT(size > 0); | |
1080 | ||
1081 | bufoff = offset - db->db_offset; | |
c9520ecc | 1082 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 1083 | |
c9520ecc | 1084 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
1085 | |
1086 | offset += tocpy; | |
1087 | size -= tocpy; | |
1088 | buf = (char *)buf + tocpy; | |
1089 | } | |
1090 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1091 | } | |
34dc7c2f BB |
1092 | return (err); |
1093 | } | |
1094 | ||
0eef1bde | 1095 | int |
1096 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1097 | void *buf, uint32_t flags) | |
34dc7c2f | 1098 | { |
0eef1bde | 1099 | dnode_t *dn; |
1100 | int err; | |
34dc7c2f | 1101 | |
0eef1bde | 1102 | err = dnode_hold(os, object, FTAG, &dn); |
1103 | if (err != 0) | |
1104 | return (err); | |
34dc7c2f | 1105 | |
0eef1bde | 1106 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1107 | dnode_rele(dn, FTAG); | |
1108 | return (err); | |
1109 | } | |
1110 | ||
1111 | int | |
1112 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1113 | uint32_t flags) | |
1114 | { | |
1115 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1116 | } | |
1117 | ||
1118 | static void | |
1119 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1120 | const void *buf, dmu_tx_t *tx) | |
1121 | { | |
1122 | int i; | |
34dc7c2f BB |
1123 | |
1124 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1125 | uint64_t tocpy; |
1126 | int64_t bufoff; | |
34dc7c2f BB |
1127 | dmu_buf_t *db = dbp[i]; |
1128 | ||
1129 | ASSERT(size > 0); | |
1130 | ||
1131 | bufoff = offset - db->db_offset; | |
c9520ecc | 1132 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1133 | |
1134 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1135 | ||
1136 | if (tocpy == db->db_size) | |
1137 | dmu_buf_will_fill(db, tx); | |
1138 | else | |
1139 | dmu_buf_will_dirty(db, tx); | |
1140 | ||
60101509 | 1141 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1142 | |
1143 | if (tocpy == db->db_size) | |
1144 | dmu_buf_fill_done(db, tx); | |
1145 | ||
1146 | offset += tocpy; | |
1147 | size -= tocpy; | |
1148 | buf = (char *)buf + tocpy; | |
1149 | } | |
0eef1bde | 1150 | } |
1151 | ||
1152 | void | |
1153 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1154 | const void *buf, dmu_tx_t *tx) | |
1155 | { | |
1156 | dmu_buf_t **dbp; | |
1157 | int numbufs; | |
1158 | ||
1159 | if (size == 0) | |
1160 | return; | |
1161 | ||
1162 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1163 | FALSE, FTAG, &numbufs, &dbp)); | |
1164 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1165 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1166 | } | |
1167 | ||
0f8ff49e SD |
1168 | /* |
1169 | * Note: Lustre is an external consumer of this interface. | |
1170 | */ | |
0eef1bde | 1171 | void |
1172 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1173 | const void *buf, dmu_tx_t *tx) | |
1174 | { | |
1175 | dmu_buf_t **dbp; | |
1176 | int numbufs; | |
1177 | ||
1178 | if (size == 0) | |
1179 | return; | |
1180 | ||
1181 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1182 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1183 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1184 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1185 | } | |
1186 | ||
b128c09f BB |
1187 | void |
1188 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1189 | dmu_tx_t *tx) | |
1190 | { | |
1191 | dmu_buf_t **dbp; | |
1192 | int numbufs, i; | |
1193 | ||
1194 | if (size == 0) | |
1195 | return; | |
1196 | ||
1197 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1198 | FALSE, FTAG, &numbufs, &dbp)); | |
1199 | ||
1200 | for (i = 0; i < numbufs; i++) { | |
1201 | dmu_buf_t *db = dbp[i]; | |
1202 | ||
1203 | dmu_buf_will_not_fill(db, tx); | |
1204 | } | |
1205 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1206 | } | |
1207 | ||
9b67f605 MA |
1208 | void |
1209 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1210 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1211 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1212 | { | |
1213 | dmu_buf_t *db; | |
1214 | ||
1215 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1216 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1217 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1218 | FTAG, &db)); | |
1219 | ||
1220 | dmu_buf_write_embedded(db, | |
1221 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1222 | uncompressed_size, compressed_size, byteorder, tx); | |
1223 | ||
1224 | dmu_buf_rele(db, FTAG); | |
1225 | } | |
1226 | ||
30af21b0 PD |
1227 | void |
1228 | dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1229 | dmu_tx_t *tx) | |
1230 | { | |
1231 | int numbufs, i; | |
1232 | dmu_buf_t **dbp; | |
1233 | ||
1234 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1235 | &numbufs, &dbp)); | |
1236 | for (i = 0; i < numbufs; i++) | |
1237 | dmu_buf_redact(dbp[i], tx); | |
1238 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1239 | } | |
1240 | ||
34dc7c2f | 1241 | #ifdef _KERNEL |
5228cf01 | 1242 | int |
d0cd9a5c | 1243 | dmu_read_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size) |
872e8d26 BB |
1244 | { |
1245 | dmu_buf_t **dbp; | |
1246 | int numbufs, i, err; | |
872e8d26 BB |
1247 | |
1248 | /* | |
1249 | * NB: we could do this block-at-a-time, but it's nice | |
1250 | * to be reading in parallel. | |
1251 | */ | |
d0cd9a5c | 1252 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
804e0504 | 1253 | TRUE, FTAG, &numbufs, &dbp, 0); |
872e8d26 BB |
1254 | if (err) |
1255 | return (err); | |
1256 | ||
1257 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1258 | uint64_t tocpy; |
1259 | int64_t bufoff; | |
872e8d26 BB |
1260 | dmu_buf_t *db = dbp[i]; |
1261 | ||
1262 | ASSERT(size > 0); | |
1263 | ||
d0cd9a5c | 1264 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1265 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1266 | |
d0cd9a5c BA |
1267 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, tocpy, |
1268 | UIO_READ, uio); | |
1269 | ||
872e8d26 BB |
1270 | if (err) |
1271 | break; | |
1272 | ||
1273 | size -= tocpy; | |
1274 | } | |
1275 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1276 | ||
1277 | return (err); | |
1278 | } | |
1279 | ||
804e0504 MA |
1280 | /* |
1281 | * Read 'size' bytes into the uio buffer. | |
1282 | * From object zdb->db_object. | |
d0cd9a5c | 1283 | * Starting at zfs_uio_offset(uio). |
804e0504 MA |
1284 | * |
1285 | * If the caller already has a dbuf in the target object | |
1286 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1287 | * because we don't have to find the dnode_t for the object. | |
1288 | */ | |
1289 | int | |
d0cd9a5c | 1290 | dmu_read_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1291 | { |
1292 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1293 | dnode_t *dn; | |
1294 | int err; | |
1295 | ||
1296 | if (size == 0) | |
1297 | return (0); | |
1298 | ||
1299 | DB_DNODE_ENTER(db); | |
1300 | dn = DB_DNODE(db); | |
1301 | err = dmu_read_uio_dnode(dn, uio, size); | |
1302 | DB_DNODE_EXIT(db); | |
1303 | ||
1304 | return (err); | |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Read 'size' bytes into the uio buffer. | |
1309 | * From the specified object | |
d0cd9a5c | 1310 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1311 | */ |
1312 | int | |
d0cd9a5c | 1313 | dmu_read_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size) |
804e0504 MA |
1314 | { |
1315 | dnode_t *dn; | |
1316 | int err; | |
1317 | ||
1318 | if (size == 0) | |
1319 | return (0); | |
1320 | ||
1321 | err = dnode_hold(os, object, FTAG, &dn); | |
1322 | if (err) | |
1323 | return (err); | |
1324 | ||
1325 | err = dmu_read_uio_dnode(dn, uio, size); | |
1326 | ||
1327 | dnode_rele(dn, FTAG); | |
1328 | ||
1329 | return (err); | |
1330 | } | |
1331 | ||
5228cf01 | 1332 | int |
d0cd9a5c | 1333 | dmu_write_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size, dmu_tx_t *tx) |
872e8d26 BB |
1334 | { |
1335 | dmu_buf_t **dbp; | |
1336 | int numbufs; | |
1337 | int err = 0; | |
1338 | int i; | |
1339 | ||
d0cd9a5c | 1340 | err = dmu_buf_hold_array_by_dnode(dn, zfs_uio_offset(uio), size, |
872e8d26 BB |
1341 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); |
1342 | if (err) | |
1343 | return (err); | |
1344 | ||
1345 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1346 | uint64_t tocpy; |
1347 | int64_t bufoff; | |
872e8d26 BB |
1348 | dmu_buf_t *db = dbp[i]; |
1349 | ||
1350 | ASSERT(size > 0); | |
1351 | ||
d0cd9a5c | 1352 | bufoff = zfs_uio_offset(uio) - db->db_offset; |
c9520ecc | 1353 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1354 | |
1355 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1356 | ||
1357 | if (tocpy == db->db_size) | |
1358 | dmu_buf_will_fill(db, tx); | |
1359 | else | |
1360 | dmu_buf_will_dirty(db, tx); | |
1361 | ||
1362 | /* | |
d0cd9a5c | 1363 | * XXX zfs_uiomove could block forever (eg.nfs-backed |
872e8d26 | 1364 | * pages). There needs to be a uiolockdown() function |
d0cd9a5c | 1365 | * to lock the pages in memory, so that zfs_uiomove won't |
872e8d26 BB |
1366 | * block. |
1367 | */ | |
d0cd9a5c BA |
1368 | err = zfs_uio_fault_move((char *)db->db_data + bufoff, |
1369 | tocpy, UIO_WRITE, uio); | |
1370 | ||
872e8d26 BB |
1371 | if (tocpy == db->db_size) |
1372 | dmu_buf_fill_done(db, tx); | |
1373 | ||
1374 | if (err) | |
1375 | break; | |
1376 | ||
1377 | size -= tocpy; | |
1378 | } | |
1379 | ||
1380 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1381 | return (err); | |
1382 | } | |
1383 | ||
804e0504 MA |
1384 | /* |
1385 | * Write 'size' bytes from the uio buffer. | |
1386 | * To object zdb->db_object. | |
d0cd9a5c | 1387 | * Starting at offset zfs_uio_offset(uio). |
804e0504 MA |
1388 | * |
1389 | * If the caller already has a dbuf in the target object | |
1390 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1391 | * because we don't have to find the dnode_t for the object. | |
1392 | */ | |
428870ff | 1393 | int |
d0cd9a5c | 1394 | dmu_write_uio_dbuf(dmu_buf_t *zdb, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1395 | dmu_tx_t *tx) |
1396 | { | |
572e2857 BB |
1397 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1398 | dnode_t *dn; | |
1399 | int err; | |
1400 | ||
428870ff BB |
1401 | if (size == 0) |
1402 | return (0); | |
1403 | ||
572e2857 BB |
1404 | DB_DNODE_ENTER(db); |
1405 | dn = DB_DNODE(db); | |
1406 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1407 | DB_DNODE_EXIT(db); | |
1408 | ||
1409 | return (err); | |
428870ff BB |
1410 | } |
1411 | ||
804e0504 MA |
1412 | /* |
1413 | * Write 'size' bytes from the uio buffer. | |
1414 | * To the specified object. | |
d0cd9a5c | 1415 | * Starting at offset zfs_uio_offset(uio). |
804e0504 | 1416 | */ |
428870ff | 1417 | int |
d0cd9a5c | 1418 | dmu_write_uio(objset_t *os, uint64_t object, zfs_uio_t *uio, uint64_t size, |
428870ff BB |
1419 | dmu_tx_t *tx) |
1420 | { | |
1421 | dnode_t *dn; | |
1422 | int err; | |
1423 | ||
1424 | if (size == 0) | |
1425 | return (0); | |
1426 | ||
1427 | err = dnode_hold(os, object, FTAG, &dn); | |
1428 | if (err) | |
1429 | return (err); | |
1430 | ||
1431 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1432 | ||
1433 | dnode_rele(dn, FTAG); | |
1434 | ||
1435 | return (err); | |
1436 | } | |
872e8d26 | 1437 | #endif /* _KERNEL */ |
34dc7c2f | 1438 | |
9babb374 BB |
1439 | /* |
1440 | * Allocate a loaned anonymous arc buffer. | |
1441 | */ | |
1442 | arc_buf_t * | |
1443 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1444 | { | |
572e2857 | 1445 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1446 | |
2aa34383 | 1447 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1448 | } |
1449 | ||
1450 | /* | |
1451 | * Free a loaned arc buffer. | |
1452 | */ | |
1453 | void | |
1454 | dmu_return_arcbuf(arc_buf_t *buf) | |
1455 | { | |
1456 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1457 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1458 | } |
1459 | ||
ba67d821 MA |
1460 | /* |
1461 | * A "lightweight" write is faster than a regular write (e.g. | |
1462 | * dmu_write_by_dnode() or dmu_assign_arcbuf_by_dnode()), because it avoids the | |
1463 | * CPU cost of creating a dmu_buf_impl_t and arc_buf_[hdr_]_t. However, the | |
1464 | * data can not be read or overwritten until the transaction's txg has been | |
1465 | * synced. This makes it appropriate for workloads that are known to be | |
1466 | * (temporarily) write-only, like "zfs receive". | |
1467 | * | |
1468 | * A single block is written, starting at the specified offset in bytes. If | |
1469 | * the call is successful, it returns 0 and the provided abd has been | |
1470 | * consumed (the caller should not free it). | |
1471 | */ | |
1472 | int | |
1473 | dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, | |
4938d01d | 1474 | const zio_prop_t *zp, zio_flag_t flags, dmu_tx_t *tx) |
ba67d821 MA |
1475 | { |
1476 | dbuf_dirty_record_t *dr = | |
1477 | dbuf_dirty_lightweight(dn, dbuf_whichblock(dn, 0, offset), tx); | |
1478 | if (dr == NULL) | |
1479 | return (SET_ERROR(EIO)); | |
1480 | dr->dt.dll.dr_abd = abd; | |
1481 | dr->dt.dll.dr_props = *zp; | |
1482 | dr->dt.dll.dr_flags = flags; | |
1483 | return (0); | |
1484 | } | |
1485 | ||
9babb374 BB |
1486 | /* |
1487 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1488 | * If this is not possible copy the contents of passed arc buf via | |
1489 | * dmu_write(). | |
1490 | */ | |
305781da | 1491 | int |
440a3eb9 | 1492 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1493 | dmu_tx_t *tx) |
1494 | { | |
9babb374 | 1495 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1496 | objset_t *os = dn->dn_objset; |
1497 | uint64_t object = dn->dn_object; | |
2aa34383 | 1498 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1499 | uint64_t blkid; |
1500 | ||
1501 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1502 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1503 | db = dbuf_hold(dn, blkid, FTAG); |
1504 | if (db == NULL) | |
1505 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1506 | rw_exit(&dn->dn_struct_rwlock); |
1507 | ||
88904bb3 | 1508 | /* |
ba67d821 MA |
1509 | * We can only assign if the offset is aligned and the arc buf is the |
1510 | * same size as the dbuf. | |
88904bb3 | 1511 | */ |
2aa34383 | 1512 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
64e0fe14 | 1513 | zfs_racct_write(blksz, 1); |
9babb374 BB |
1514 | dbuf_assign_arcbuf(db, buf, tx); |
1515 | dbuf_rele(db, FTAG); | |
1516 | } else { | |
2aa34383 DK |
1517 | /* compressed bufs must always be assignable to their dbuf */ |
1518 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1519 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1520 | |
9babb374 | 1521 | dbuf_rele(db, FTAG); |
572e2857 | 1522 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 BB |
1523 | dmu_return_arcbuf(buf); |
1524 | } | |
305781da TC |
1525 | |
1526 | return (0); | |
9babb374 BB |
1527 | } |
1528 | ||
305781da | 1529 | int |
440a3eb9 TC |
1530 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1531 | dmu_tx_t *tx) | |
1532 | { | |
305781da | 1533 | int err; |
440a3eb9 TC |
1534 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1535 | ||
1536 | DB_DNODE_ENTER(dbuf); | |
305781da | 1537 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1538 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1539 | |
1540 | return (err); | |
440a3eb9 TC |
1541 | } |
1542 | ||
34dc7c2f | 1543 | typedef struct { |
428870ff BB |
1544 | dbuf_dirty_record_t *dsa_dr; |
1545 | dmu_sync_cb_t *dsa_done; | |
1546 | zgd_t *dsa_zgd; | |
1547 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1548 | } dmu_sync_arg_t; |
1549 | ||
b128c09f BB |
1550 | static void |
1551 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1552 | { | |
14e4e3cb | 1553 | (void) buf; |
428870ff BB |
1554 | dmu_sync_arg_t *dsa = varg; |
1555 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1556 | blkptr_t *bp = zio->io_bp; |
1557 | ||
428870ff BB |
1558 | if (zio->io_error == 0) { |
1559 | if (BP_IS_HOLE(bp)) { | |
1560 | /* | |
1561 | * A block of zeros may compress to a hole, but the | |
1562 | * block size still needs to be known for replay. | |
1563 | */ | |
1564 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1565 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1566 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1567 | BP_SET_FILL(bp, 1); |
428870ff | 1568 | } |
b128c09f BB |
1569 | } |
1570 | } | |
1571 | ||
428870ff BB |
1572 | static void |
1573 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1574 | { | |
1575 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1576 | } | |
1577 | ||
34dc7c2f BB |
1578 | static void |
1579 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1580 | { | |
14e4e3cb | 1581 | (void) buf; |
428870ff BB |
1582 | dmu_sync_arg_t *dsa = varg; |
1583 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1584 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1585 | zgd_t *zgd = dsa->dsa_zgd; |
1586 | ||
1587 | /* | |
1588 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1589 | * the writes for the lwb have completed. | |
1590 | */ | |
1591 | if (zio->io_error == 0) { | |
1592 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1593 | } | |
34dc7c2f | 1594 | |
34dc7c2f BB |
1595 | mutex_enter(&db->db_mtx); |
1596 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1597 | if (zio->io_error == 0) { |
03c6040b GW |
1598 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1599 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1600 | blkptr_t *bp = zio->io_bp; |
1601 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1602 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1603 | |
1604 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1605 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1606 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1607 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1608 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1609 | } |
428870ff BB |
1610 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1611 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1612 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1613 | |
1614 | /* | |
1615 | * Old style holes are filled with all zeros, whereas | |
1616 | * new-style holes maintain their lsize, type, level, | |
1617 | * and birth time (see zio_write_compress). While we | |
1618 | * need to reset the BP_SET_LSIZE() call that happened | |
1619 | * in dmu_sync_ready for old style holes, we do *not* | |
1620 | * want to wipe out the information contained in new | |
1621 | * style holes. Thus, only zero out the block pointer if | |
1622 | * it's an old style hole. | |
1623 | */ | |
1624 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1625 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1626 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1627 | } else { | |
1628 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1629 | } | |
34dc7c2f BB |
1630 | cv_broadcast(&db->db_changed); |
1631 | mutex_exit(&db->db_mtx); | |
1632 | ||
428870ff | 1633 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1634 | |
428870ff BB |
1635 | kmem_free(dsa, sizeof (*dsa)); |
1636 | } | |
1637 | ||
1638 | static void | |
1639 | dmu_sync_late_arrival_done(zio_t *zio) | |
1640 | { | |
1641 | blkptr_t *bp = zio->io_bp; | |
1642 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1643 | zgd_t *zgd = dsa->dsa_zgd; |
1644 | ||
1645 | if (zio->io_error == 0) { | |
1646 | /* | |
1647 | * Record the vdev(s) backing this blkptr so they can be | |
1648 | * flushed after the writes for the lwb have completed. | |
1649 | */ | |
1650 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1651 | ||
1652 | if (!BP_IS_HOLE(bp)) { | |
2a8ba608 | 1653 | blkptr_t *bp_orig __maybe_unused = &zio->io_bp_orig; |
900d09b2 PS |
1654 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); |
1655 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1656 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1657 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1658 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1659 | } | |
428870ff BB |
1660 | } |
1661 | ||
1662 | dmu_tx_commit(dsa->dsa_tx); | |
1663 | ||
1664 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1665 | ||
e2af2acc | 1666 | abd_free(zio->io_abd); |
428870ff BB |
1667 | kmem_free(dsa, sizeof (*dsa)); |
1668 | } | |
1669 | ||
1670 | static int | |
1671 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1672 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1673 | { |
1674 | dmu_sync_arg_t *dsa; | |
1675 | dmu_tx_t *tx; | |
bdb7df42 AM |
1676 | int error; |
1677 | ||
1678 | error = dbuf_read((dmu_buf_impl_t *)zgd->zgd_db, NULL, | |
1679 | DB_RF_CANFAIL | DB_RF_NOPREFETCH); | |
1680 | if (error != 0) | |
1681 | return (error); | |
428870ff BB |
1682 | |
1683 | tx = dmu_tx_create(os); | |
1684 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
704c80f0 AM |
1685 | /* |
1686 | * This transaction does not produce any dirty data or log blocks, so | |
1687 | * it should not be throttled. All other cases wait for TXG sync, by | |
1688 | * which time the log block we are writing will be obsolete, so we can | |
1689 | * skip waiting and just return error here instead. | |
1690 | */ | |
1691 | if (dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE) != 0) { | |
428870ff | 1692 | dmu_tx_abort(tx); |
2e528b49 MA |
1693 | /* Make zl_get_data do txg_waited_synced() */ |
1694 | return (SET_ERROR(EIO)); | |
428870ff BB |
1695 | } |
1696 | ||
1ce23dca PS |
1697 | /* |
1698 | * In order to prevent the zgd's lwb from being free'd prior to | |
1699 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1700 | * the lwb's "max txg" takes this tx's txg into account. | |
1701 | */ | |
1702 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1703 | ||
79c76d5b | 1704 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1705 | dsa->dsa_dr = NULL; |
1706 | dsa->dsa_done = done; | |
1707 | dsa->dsa_zgd = zgd; | |
1708 | dsa->dsa_tx = tx; | |
1709 | ||
02dc43bc MA |
1710 | /* |
1711 | * Since we are currently syncing this txg, it's nontrivial to | |
1712 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1713 | * | |
1714 | * When syncing, the db_blkptr is initially the BP of the previous | |
1715 | * txg. We can not nopwrite against it because it will be changed | |
1716 | * (this is similar to the non-late-arrival case where the dbuf is | |
1717 | * dirty in a future txg). | |
1718 | * | |
1719 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1720 | * We can not nopwrite against it because although the BP will not | |
1721 | * (typically) be changed, the data has not yet been persisted to this | |
1722 | * location. | |
1723 | * | |
1724 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1725 | * possible to always nopwrite, because the data that was written in | |
1726 | * this txg is the same data that we are trying to write. However we | |
1727 | * would need to check that this dbuf is not dirty in any future | |
1728 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1729 | * don't nopwrite in this case. | |
1730 | */ | |
1731 | zp->zp_nopwrite = B_FALSE; | |
1732 | ||
a6255b7f DQ |
1733 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1734 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1735 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
ccec7fbe | 1736 | dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, |
a6255b7f | 1737 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); |
428870ff BB |
1738 | |
1739 | return (0); | |
34dc7c2f BB |
1740 | } |
1741 | ||
1742 | /* | |
1743 | * Intent log support: sync the block associated with db to disk. | |
1744 | * N.B. and XXX: the caller is responsible for making sure that the | |
1745 | * data isn't changing while dmu_sync() is writing it. | |
1746 | * | |
1747 | * Return values: | |
1748 | * | |
03c6040b | 1749 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1750 | * The caller should not log the write. |
1751 | * | |
1752 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1753 | * The caller should not log the write. | |
1754 | * | |
1755 | * EALREADY: this block is already in the process of being synced. | |
1756 | * The caller should track its progress (somehow). | |
1757 | * | |
428870ff BB |
1758 | * EIO: could not do the I/O. |
1759 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1760 | * |
428870ff BB |
1761 | * 0: the I/O has been initiated. |
1762 | * The caller should log this blkptr in the done callback. | |
1763 | * It is possible that the I/O will fail, in which case | |
1764 | * the error will be reported to the done callback and | |
1765 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1766 | */ |
1767 | int | |
428870ff | 1768 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1769 | { |
428870ff BB |
1770 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1771 | objset_t *os = db->db_objset; | |
1772 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
cccbed9f | 1773 | dbuf_dirty_record_t *dr, *dr_next; |
428870ff | 1774 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1775 | zbookmark_phys_t zb; |
428870ff | 1776 | zio_prop_t zp; |
572e2857 | 1777 | dnode_t *dn; |
34dc7c2f | 1778 | |
428870ff | 1779 | ASSERT(pio != NULL); |
34dc7c2f BB |
1780 | ASSERT(txg != 0); |
1781 | ||
428870ff BB |
1782 | SET_BOOKMARK(&zb, ds->ds_object, |
1783 | db->db.db_object, db->db_level, db->db_blkid); | |
1784 | ||
572e2857 BB |
1785 | DB_DNODE_ENTER(db); |
1786 | dn = DB_DNODE(db); | |
82644107 | 1787 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1788 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1789 | |
1790 | /* | |
428870ff | 1791 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1792 | */ |
428870ff BB |
1793 | if (txg > spa_freeze_txg(os->os_spa)) |
1794 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1795 | |
1796 | /* | |
428870ff BB |
1797 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1798 | * and us. If we determine that this txg is not yet syncing, | |
1799 | * but it begins to sync a moment later, that's OK because the | |
1800 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1801 | */ |
428870ff BB |
1802 | mutex_enter(&db->db_mtx); |
1803 | ||
1804 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1805 | /* |
428870ff | 1806 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1807 | */ |
428870ff | 1808 | mutex_exit(&db->db_mtx); |
2e528b49 | 1809 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1810 | } |
1811 | ||
428870ff BB |
1812 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1813 | /* | |
1814 | * This txg is currently syncing, so we can't mess with | |
1815 | * the dirty record anymore; just write a new log block. | |
1816 | */ | |
1817 | mutex_exit(&db->db_mtx); | |
1818 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1819 | } |
1820 | ||
cccbed9f | 1821 | dr = dbuf_find_dirty_eq(db, txg); |
428870ff BB |
1822 | |
1823 | if (dr == NULL) { | |
34dc7c2f | 1824 | /* |
428870ff | 1825 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1826 | * There's no need to log writes to freed blocks, so we're done. |
1827 | */ | |
1828 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1829 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1830 | } |
1831 | ||
cccbed9f MM |
1832 | dr_next = list_next(&db->db_dirty_records, dr); |
1833 | ASSERT(dr_next == NULL || dr_next->dr_txg < txg); | |
03c6040b | 1834 | |
02dc43bc MA |
1835 | if (db->db_blkptr != NULL) { |
1836 | /* | |
1837 | * We need to fill in zgd_bp with the current blkptr so that | |
1838 | * the nopwrite code can check if we're writing the same | |
1839 | * data that's already on disk. We can only nopwrite if we | |
1840 | * are sure that after making the copy, db_blkptr will not | |
1841 | * change until our i/o completes. We ensure this by | |
1842 | * holding the db_mtx, and only allowing nopwrite if the | |
1843 | * block is not already dirty (see below). This is verified | |
1844 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1845 | * not changed. | |
1846 | */ | |
1847 | *zgd->zgd_bp = *db->db_blkptr; | |
1848 | } | |
1849 | ||
03c6040b | 1850 | /* |
f3c517d8 MA |
1851 | * Assume the on-disk data is X, the current syncing data (in |
1852 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1853 | * in dmu_sync). | |
1854 | * | |
1855 | * We usually want to perform a nopwrite if X and Z are the | |
1856 | * same. However, if Y is different (i.e. the BP is going to | |
1857 | * change before this write takes effect), then a nopwrite will | |
1858 | * be incorrect - we would override with X, which could have | |
1859 | * been freed when Y was written. | |
1860 | * | |
1861 | * (Note that this is not a concern when we are nop-writing from | |
1862 | * syncing context, because X and Y must be identical, because | |
1863 | * all previous txgs have been synced.) | |
1864 | * | |
1865 | * Therefore, we disable nopwrite if the current BP could change | |
1866 | * before this TXG. There are two ways it could change: by | |
1867 | * being dirty (dr_next is non-NULL), or by being freed | |
1868 | * (dnode_block_freed()). This behavior is verified by | |
1869 | * zio_done(), which VERIFYs that the override BP is identical | |
1870 | * to the on-disk BP. | |
03c6040b | 1871 | */ |
f3c517d8 MA |
1872 | DB_DNODE_ENTER(db); |
1873 | dn = DB_DNODE(db); | |
cccbed9f | 1874 | if (dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) |
03c6040b | 1875 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1876 | DB_DNODE_EXIT(db); |
03c6040b | 1877 | |
34dc7c2f | 1878 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1879 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1880 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1881 | /* |
428870ff BB |
1882 | * We have already issued a sync write for this buffer, |
1883 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1884 | * have been dirtied since, or we would have cleared the state. |
1885 | */ | |
34dc7c2f | 1886 | mutex_exit(&db->db_mtx); |
2e528b49 | 1887 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1888 | } |
1889 | ||
428870ff | 1890 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1891 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1892 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1893 | |
79c76d5b | 1894 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1895 | dsa->dsa_dr = dr; |
1896 | dsa->dsa_done = done; | |
1897 | dsa->dsa_zgd = zgd; | |
1898 | dsa->dsa_tx = NULL; | |
b128c09f | 1899 | |
ed2f7ba0 AM |
1900 | zio_nowait(arc_write(pio, os->os_spa, txg, zgd->zgd_bp, |
1901 | dr->dt.dl.dr_data, !DBUF_IS_CACHEABLE(db), dbuf_is_l2cacheable(db), | |
ccec7fbe | 1902 | &zp, dmu_sync_ready, NULL, dmu_sync_done, dsa, |
bc77ba73 | 1903 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 1904 | |
428870ff | 1905 | return (0); |
34dc7c2f BB |
1906 | } |
1907 | ||
b5256303 TC |
1908 | int |
1909 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
1910 | { | |
1911 | dnode_t *dn; | |
1912 | int err; | |
1913 | ||
1914 | err = dnode_hold(os, object, FTAG, &dn); | |
1915 | if (err) | |
1916 | return (err); | |
1917 | err = dnode_set_nlevels(dn, nlevels, tx); | |
1918 | dnode_rele(dn, FTAG); | |
1919 | return (err); | |
1920 | } | |
1921 | ||
34dc7c2f BB |
1922 | int |
1923 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 1924 | dmu_tx_t *tx) |
34dc7c2f BB |
1925 | { |
1926 | dnode_t *dn; | |
1927 | int err; | |
1928 | ||
428870ff | 1929 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1930 | if (err) |
1931 | return (err); | |
1932 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1933 | dnode_rele(dn, FTAG); | |
1934 | return (err); | |
1935 | } | |
1936 | ||
ae76f45c TC |
1937 | int |
1938 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
1939 | dmu_tx_t *tx) | |
1940 | { | |
1941 | dnode_t *dn; | |
1942 | int err; | |
1943 | ||
1944 | err = dnode_hold(os, object, FTAG, &dn); | |
1945 | if (err) | |
1946 | return (err); | |
1947 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 1948 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
1949 | rw_exit(&dn->dn_struct_rwlock); |
1950 | dnode_rele(dn, FTAG); | |
1951 | return (0); | |
1952 | } | |
1953 | ||
34dc7c2f BB |
1954 | void |
1955 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 1956 | dmu_tx_t *tx) |
34dc7c2f BB |
1957 | { |
1958 | dnode_t *dn; | |
1959 | ||
9b67f605 MA |
1960 | /* |
1961 | * Send streams include each object's checksum function. This | |
1962 | * check ensures that the receiving system can understand the | |
1963 | * checksum function transmitted. | |
1964 | */ | |
1965 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1966 | ||
1967 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1968 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1969 | dn->dn_checksum = checksum; |
1970 | dnode_setdirty(dn, tx); | |
1971 | dnode_rele(dn, FTAG); | |
1972 | } | |
1973 | ||
1974 | void | |
1975 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 1976 | dmu_tx_t *tx) |
34dc7c2f BB |
1977 | { |
1978 | dnode_t *dn; | |
1979 | ||
9b67f605 MA |
1980 | /* |
1981 | * Send streams include each object's compression function. This | |
1982 | * check ensures that the receiving system can understand the | |
1983 | * compression function transmitted. | |
1984 | */ | |
1985 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1986 | ||
1987 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
1988 | dn->dn_compress = compress; |
1989 | dnode_setdirty(dn, tx); | |
1990 | dnode_rele(dn, FTAG); | |
1991 | } | |
1992 | ||
faf0f58c MA |
1993 | /* |
1994 | * When the "redundant_metadata" property is set to "most", only indirect | |
1995 | * blocks of this level and higher will have an additional ditto block. | |
1996 | */ | |
18168da7 | 1997 | static const int zfs_redundant_metadata_most_ditto_level = 2; |
faf0f58c | 1998 | |
428870ff | 1999 | void |
82644107 | 2000 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
2001 | { |
2002 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 2003 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 2004 | (wp & WP_SPILL)); |
428870ff BB |
2005 | enum zio_checksum checksum = os->os_checksum; |
2006 | enum zio_compress compress = os->os_compress; | |
10b3c7f5 | 2007 | uint8_t complevel = os->os_complevel; |
428870ff | 2008 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; |
03c6040b GW |
2009 | boolean_t dedup = B_FALSE; |
2010 | boolean_t nopwrite = B_FALSE; | |
428870ff | 2011 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 2012 | boolean_t encrypt = B_FALSE; |
428870ff | 2013 | int copies = os->os_copies; |
a7004725 | 2014 | |
428870ff | 2015 | /* |
03c6040b GW |
2016 | * We maintain different write policies for each of the following |
2017 | * types of data: | |
2018 | * 1. metadata | |
2019 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
2020 | * 3. all other level 0 blocks | |
428870ff BB |
2021 | */ |
2022 | if (ismd) { | |
b1d21733 TC |
2023 | /* |
2024 | * XXX -- we should design a compression algorithm | |
2025 | * that specializes in arrays of bps. | |
2026 | */ | |
2027 | compress = zio_compress_select(os->os_spa, | |
2028 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 2029 | |
428870ff BB |
2030 | /* |
2031 | * Metadata always gets checksummed. If the data | |
2032 | * checksum is multi-bit correctable, and it's not a | |
2033 | * ZBT-style checksum, then it's suitable for metadata | |
2034 | * as well. Otherwise, the metadata checksum defaults | |
2035 | * to fletcher4. | |
2036 | */ | |
3c67d83a TH |
2037 | if (!(zio_checksum_table[checksum].ci_flags & |
2038 | ZCHECKSUM_FLAG_METADATA) || | |
2039 | (zio_checksum_table[checksum].ci_flags & | |
2040 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 2041 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c | 2042 | |
5405be03 A |
2043 | switch (os->os_redundant_metadata) { |
2044 | case ZFS_REDUNDANT_METADATA_ALL: | |
faf0f58c | 2045 | copies++; |
5405be03 A |
2046 | break; |
2047 | case ZFS_REDUNDANT_METADATA_MOST: | |
2048 | if (level >= zfs_redundant_metadata_most_ditto_level || | |
2049 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)) | |
2050 | copies++; | |
2051 | break; | |
2052 | case ZFS_REDUNDANT_METADATA_SOME: | |
2053 | if (DMU_OT_IS_CRITICAL(type)) | |
2054 | copies++; | |
2055 | break; | |
2056 | case ZFS_REDUNDANT_METADATA_NONE: | |
2057 | break; | |
2058 | } | |
03c6040b GW |
2059 | } else if (wp & WP_NOFILL) { |
2060 | ASSERT(level == 0); | |
428870ff | 2061 | |
428870ff | 2062 | /* |
03c6040b GW |
2063 | * If we're writing preallocated blocks, we aren't actually |
2064 | * writing them so don't set any policy properties. These | |
2065 | * blocks are currently only used by an external subsystem | |
2066 | * outside of zfs (i.e. dump) and not written by the zio | |
2067 | * pipeline. | |
428870ff | 2068 | */ |
03c6040b GW |
2069 | compress = ZIO_COMPRESS_OFF; |
2070 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2071 | } else { |
99197f03 JG |
2072 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2073 | compress); | |
10b3c7f5 MN |
2074 | complevel = zio_complevel_select(os->os_spa, compress, |
2075 | complevel, complevel); | |
428870ff | 2076 | |
03c6040b GW |
2077 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2078 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2079 | dedup_checksum; | |
428870ff | 2080 | |
03c6040b GW |
2081 | /* |
2082 | * Determine dedup setting. If we are in dmu_sync(), | |
2083 | * we won't actually dedup now because that's all | |
2084 | * done in syncing context; but we do want to use the | |
e1cfd73f | 2085 | * dedup checksum. If the checksum is not strong |
03c6040b GW |
2086 | * enough to ensure unique signatures, force |
2087 | * dedup_verify. | |
2088 | */ | |
2089 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2090 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2091 | if (!(zio_checksum_table[checksum].ci_flags & |
2092 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2093 | dedup_verify = B_TRUE; |
2094 | } | |
428870ff | 2095 | |
03c6040b | 2096 | /* |
3c67d83a TH |
2097 | * Enable nopwrite if we have secure enough checksum |
2098 | * algorithm (see comment in zio_nop_write) and | |
2099 | * compression is enabled. We don't enable nopwrite if | |
2100 | * dedup is enabled as the two features are mutually | |
2101 | * exclusive. | |
03c6040b | 2102 | */ |
3c67d83a TH |
2103 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2104 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2105 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2106 | } |
2107 | ||
b5256303 TC |
2108 | /* |
2109 | * All objects in an encrypted objset are protected from modification | |
2110 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2111 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2112 | * not subject to nopwrite since writing the same data will still | |
2113 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2114 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2115 | * object types. | |
2116 | */ | |
2117 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2118 | encrypt = B_TRUE; | |
2119 | ||
2120 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2121 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2122 | nopwrite = B_FALSE; | |
2123 | } else { | |
2124 | dedup = B_FALSE; | |
2125 | } | |
2126 | ||
ae76f45c TC |
2127 | if (level <= 0 && |
2128 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2129 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2130 | } |
b5256303 | 2131 | } |
2aa34383 | 2132 | |
b5256303 | 2133 | zp->zp_compress = compress; |
10b3c7f5 | 2134 | zp->zp_complevel = complevel; |
b5256303 | 2135 | zp->zp_checksum = checksum; |
428870ff BB |
2136 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2137 | zp->zp_level = level; | |
faf0f58c | 2138 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2139 | zp->zp_dedup = dedup; |
2140 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2141 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2142 | zp->zp_encrypt = encrypt; |
2143 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
861166b0 AZ |
2144 | memset(zp->zp_salt, 0, ZIO_DATA_SALT_LEN); |
2145 | memset(zp->zp_iv, 0, ZIO_DATA_IV_LEN); | |
2146 | memset(zp->zp_mac, 0, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2147 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2148 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2149 | |
2150 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2151 | } |
2152 | ||
66aca247 | 2153 | /* |
64bfa6ba BB |
2154 | * Reports the location of data and holes in an object. In order to |
2155 | * accurately report holes all dirty data must be synced to disk. This | |
2156 | * causes extremely poor performance when seeking for holes in a dirty file. | |
2157 | * As a compromise, only provide hole data when the dnode is clean. When | |
2158 | * a dnode is dirty report the dnode as having no holes by returning EBUSY | |
2159 | * which is always safe to do. | |
66aca247 | 2160 | */ |
34dc7c2f BB |
2161 | int |
2162 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2163 | { | |
2164 | dnode_t *dn; | |
64bfa6ba | 2165 | int restarted = 0, err; |
34dc7c2f | 2166 | |
de198f2d | 2167 | restart: |
428870ff | 2168 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2169 | if (err) |
2170 | return (err); | |
66aca247 | 2171 | |
de198f2d | 2172 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
66aca247 | 2173 | |
de198f2d BB |
2174 | if (dnode_is_dirty(dn)) { |
2175 | /* | |
2176 | * If the zfs_dmu_offset_next_sync module option is enabled | |
64bfa6ba BB |
2177 | * then hole reporting has been requested. Dirty dnodes |
2178 | * must be synced to disk to accurately report holes. | |
de198f2d | 2179 | * |
64bfa6ba BB |
2180 | * Provided a RL_READER rangelock spanning 0-UINT64_MAX is |
2181 | * held by the caller only a single restart will be required. | |
2182 | * We tolerate callers which do not hold the rangelock by | |
2183 | * returning EBUSY and not reporting holes after one restart. | |
de198f2d BB |
2184 | */ |
2185 | if (zfs_dmu_offset_next_sync) { | |
2186 | rw_exit(&dn->dn_struct_rwlock); | |
2187 | dnode_rele(dn, FTAG); | |
64bfa6ba BB |
2188 | |
2189 | if (restarted) | |
2190 | return (SET_ERROR(EBUSY)); | |
2191 | ||
de198f2d | 2192 | txg_wait_synced(dmu_objset_pool(os), 0); |
64bfa6ba | 2193 | restarted = 1; |
de198f2d BB |
2194 | goto restart; |
2195 | } | |
34dc7c2f | 2196 | |
66aca247 | 2197 | err = SET_ERROR(EBUSY); |
de198f2d BB |
2198 | } else { |
2199 | err = dnode_next_offset(dn, DNODE_FIND_HAVELOCK | | |
2200 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2201 | } | |
66aca247 | 2202 | |
de198f2d | 2203 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
2204 | dnode_rele(dn, FTAG); |
2205 | ||
2206 | return (err); | |
67a1b037 PJD |
2207 | } |
2208 | ||
2209 | int | |
2210 | dmu_read_l0_bps(objset_t *os, uint64_t object, uint64_t offset, uint64_t length, | |
d0d91f18 | 2211 | blkptr_t *bps, size_t *nbpsp) |
67a1b037 PJD |
2212 | { |
2213 | dmu_buf_t **dbp, *dbuf; | |
2214 | dmu_buf_impl_t *db; | |
2215 | blkptr_t *bp; | |
2216 | int error, numbufs; | |
2217 | ||
2218 | error = dmu_buf_hold_array(os, object, offset, length, FALSE, FTAG, | |
2219 | &numbufs, &dbp); | |
2220 | if (error != 0) { | |
2221 | if (error == ESRCH) { | |
2222 | error = SET_ERROR(ENXIO); | |
2223 | } | |
2224 | return (error); | |
2225 | } | |
2226 | ||
2227 | ASSERT3U(numbufs, <=, *nbpsp); | |
2228 | ||
2229 | for (int i = 0; i < numbufs; i++) { | |
2230 | dbuf = dbp[i]; | |
2231 | db = (dmu_buf_impl_t *)dbuf; | |
ce0e1cc4 PJD |
2232 | |
2233 | mutex_enter(&db->db_mtx); | |
67a1b037 | 2234 | |
67a1b037 PJD |
2235 | if (!list_is_empty(&db->db_dirty_records)) { |
2236 | dbuf_dirty_record_t *dr; | |
2237 | ||
2238 | dr = list_head(&db->db_dirty_records); | |
2239 | if (dr->dt.dl.dr_brtwrite) { | |
2240 | /* | |
2241 | * This is very special case where we clone a | |
2242 | * block and in the same transaction group we | |
2243 | * read its BP (most likely to clone the clone). | |
2244 | */ | |
2245 | bp = &dr->dt.dl.dr_overridden_by; | |
2246 | } else { | |
2247 | /* | |
2248 | * The block was modified in the same | |
2249 | * transaction group. | |
2250 | */ | |
ce0e1cc4 | 2251 | mutex_exit(&db->db_mtx); |
67a1b037 PJD |
2252 | error = SET_ERROR(EAGAIN); |
2253 | goto out; | |
2254 | } | |
ce0e1cc4 PJD |
2255 | } else { |
2256 | bp = db->db_blkptr; | |
67a1b037 | 2257 | } |
ce0e1cc4 PJD |
2258 | |
2259 | mutex_exit(&db->db_mtx); | |
2260 | ||
67a1b037 PJD |
2261 | if (bp == NULL) { |
2262 | /* | |
2263 | * The block was created in this transaction group, | |
2264 | * so it has no BP yet. | |
2265 | */ | |
2266 | error = SET_ERROR(EAGAIN); | |
2267 | goto out; | |
2268 | } | |
67a1b037 PJD |
2269 | /* |
2270 | * Make sure we clone only data blocks. | |
2271 | */ | |
2272 | if (BP_IS_METADATA(bp) && !BP_IS_HOLE(bp)) { | |
2273 | error = SET_ERROR(EINVAL); | |
2274 | goto out; | |
2275 | } | |
2276 | ||
55b764e0 AM |
2277 | /* |
2278 | * If the block was allocated in transaction group that is not | |
2279 | * yet synced, we could clone it, but we couldn't write this | |
2280 | * operation into ZIL, or it may be impossible to replay, since | |
2281 | * the block may appear not yet allocated at that point. | |
2282 | */ | |
2283 | if (BP_PHYSICAL_BIRTH(bp) > spa_freeze_txg(os->os_spa)) { | |
2284 | error = SET_ERROR(EINVAL); | |
2285 | goto out; | |
2286 | } | |
2287 | if (BP_PHYSICAL_BIRTH(bp) > spa_last_synced_txg(os->os_spa)) { | |
2288 | error = SET_ERROR(EAGAIN); | |
2289 | goto out; | |
2290 | } | |
2291 | ||
67a1b037 PJD |
2292 | bps[i] = *bp; |
2293 | } | |
2294 | ||
2295 | *nbpsp = numbufs; | |
2296 | out: | |
2297 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
2298 | ||
2299 | return (error); | |
2300 | } | |
2301 | ||
bd8c6bd6 | 2302 | int |
67a1b037 | 2303 | dmu_brt_clone(objset_t *os, uint64_t object, uint64_t offset, uint64_t length, |
a03ebd9b | 2304 | dmu_tx_t *tx, const blkptr_t *bps, size_t nbps) |
67a1b037 PJD |
2305 | { |
2306 | spa_t *spa; | |
2307 | dmu_buf_t **dbp, *dbuf; | |
2308 | dmu_buf_impl_t *db; | |
2309 | struct dirty_leaf *dl; | |
2310 | dbuf_dirty_record_t *dr; | |
2311 | const blkptr_t *bp; | |
bd8c6bd6 | 2312 | int error = 0, i, numbufs; |
67a1b037 PJD |
2313 | |
2314 | spa = os->os_spa; | |
2315 | ||
2316 | VERIFY0(dmu_buf_hold_array(os, object, offset, length, FALSE, FTAG, | |
2317 | &numbufs, &dbp)); | |
2318 | ASSERT3U(nbps, ==, numbufs); | |
2319 | ||
bd8c6bd6 | 2320 | /* |
fbbe5e96 | 2321 | * Before we start cloning make sure that the dbufs sizes match new BPs |
bd8c6bd6 PJD |
2322 | * sizes. If they don't, that's a no-go, as we are not able to shrink |
2323 | * dbufs. | |
2324 | */ | |
2325 | for (i = 0; i < numbufs; i++) { | |
2326 | dbuf = dbp[i]; | |
2327 | db = (dmu_buf_impl_t *)dbuf; | |
2328 | bp = &bps[i]; | |
2329 | ||
2330 | ASSERT0(db->db_level); | |
2331 | ASSERT(db->db_blkid != DMU_BONUS_BLKID); | |
e6107668 | 2332 | ASSERT(db->db_blkid != DMU_SPILL_BLKID); |
bd8c6bd6 PJD |
2333 | |
2334 | if (!BP_IS_HOLE(bp) && BP_GET_LSIZE(bp) != dbuf->db_size) { | |
2335 | error = SET_ERROR(EXDEV); | |
2336 | goto out; | |
2337 | } | |
2338 | } | |
2339 | ||
2340 | for (i = 0; i < numbufs; i++) { | |
67a1b037 PJD |
2341 | dbuf = dbp[i]; |
2342 | db = (dmu_buf_impl_t *)dbuf; | |
2343 | bp = &bps[i]; | |
2344 | ||
2345 | ASSERT0(db->db_level); | |
2346 | ASSERT(db->db_blkid != DMU_BONUS_BLKID); | |
e6107668 | 2347 | ASSERT(db->db_blkid != DMU_SPILL_BLKID); |
67a1b037 PJD |
2348 | ASSERT(BP_IS_HOLE(bp) || dbuf->db_size == BP_GET_LSIZE(bp)); |
2349 | ||
555ef90c | 2350 | dmu_buf_will_clone(dbuf, tx); |
ce0e1cc4 PJD |
2351 | |
2352 | mutex_enter(&db->db_mtx); | |
2353 | ||
67a1b037 | 2354 | dr = list_head(&db->db_dirty_records); |
ce0e1cc4 | 2355 | VERIFY(dr != NULL); |
67a1b037 PJD |
2356 | ASSERT3U(dr->dr_txg, ==, tx->tx_txg); |
2357 | dl = &dr->dt.dl; | |
2358 | dl->dr_overridden_by = *bp; | |
2359 | dl->dr_brtwrite = B_TRUE; | |
67a1b037 PJD |
2360 | dl->dr_override_state = DR_OVERRIDDEN; |
2361 | if (BP_IS_HOLE(bp)) { | |
2362 | dl->dr_overridden_by.blk_birth = 0; | |
2363 | dl->dr_overridden_by.blk_phys_birth = 0; | |
2364 | } else { | |
2365 | dl->dr_overridden_by.blk_birth = dr->dr_txg; | |
c71fe716 PJD |
2366 | if (!BP_IS_EMBEDDED(bp)) { |
2367 | dl->dr_overridden_by.blk_phys_birth = | |
2368 | BP_PHYSICAL_BIRTH(bp); | |
2369 | } | |
67a1b037 PJD |
2370 | } |
2371 | ||
ce0e1cc4 PJD |
2372 | mutex_exit(&db->db_mtx); |
2373 | ||
67a1b037 PJD |
2374 | /* |
2375 | * When data in embedded into BP there is no need to create | |
2376 | * BRT entry as there is no data block. Just copy the BP as | |
2377 | * it contains the data. | |
67a1b037 | 2378 | */ |
a03ebd9b | 2379 | if (!BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) { |
67a1b037 PJD |
2380 | brt_pending_add(spa, bp, tx); |
2381 | } | |
2382 | } | |
bd8c6bd6 | 2383 | out: |
67a1b037 | 2384 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
bd8c6bd6 PJD |
2385 | |
2386 | return (error); | |
34dc7c2f BB |
2387 | } |
2388 | ||
2389 | void | |
e0b0ca98 | 2390 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2391 | { |
e0b0ca98 | 2392 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2393 | |
34dc7c2f BB |
2394 | doi->doi_data_block_size = dn->dn_datablksz; |
2395 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2396 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2397 | doi->doi_type = dn->dn_type; |
2398 | doi->doi_bonus_type = dn->dn_bonustype; | |
2399 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2400 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2401 | doi->doi_indirection = dn->dn_nlevels; |
2402 | doi->doi_checksum = dn->dn_checksum; | |
2403 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2404 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2405 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2406 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2407 | doi->doi_fill_count = 0; |
1c27024e | 2408 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2409 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2410 | } |
2411 | ||
2412 | void | |
2413 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2414 | { | |
2415 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2416 | mutex_enter(&dn->dn_mtx); | |
2417 | ||
2418 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2419 | |
2420 | mutex_exit(&dn->dn_mtx); | |
2421 | rw_exit(&dn->dn_struct_rwlock); | |
2422 | } | |
2423 | ||
2424 | /* | |
2425 | * Get information on a DMU object. | |
2426 | * If doi is NULL, just indicates whether the object exists. | |
2427 | */ | |
2428 | int | |
2429 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2430 | { | |
2431 | dnode_t *dn; | |
428870ff | 2432 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2433 | |
2434 | if (err) | |
2435 | return (err); | |
2436 | ||
2437 | if (doi != NULL) | |
2438 | dmu_object_info_from_dnode(dn, doi); | |
2439 | ||
2440 | dnode_rele(dn, FTAG); | |
2441 | return (0); | |
2442 | } | |
2443 | ||
2444 | /* | |
2445 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2446 | */ | |
2447 | void | |
572e2857 | 2448 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2449 | { |
572e2857 BB |
2450 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2451 | ||
2452 | DB_DNODE_ENTER(db); | |
2453 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2454 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2455 | } |
2456 | ||
2457 | /* | |
2458 | * Faster still when you only care about the size. | |
34dc7c2f BB |
2459 | */ |
2460 | void | |
572e2857 BB |
2461 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2462 | u_longlong_t *nblk512) | |
34dc7c2f | 2463 | { |
572e2857 BB |
2464 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2465 | dnode_t *dn; | |
2466 | ||
2467 | DB_DNODE_ENTER(db); | |
2468 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2469 | |
2470 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2471 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2472 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2473 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2474 | DB_DNODE_EXIT(db); | |
2475 | } | |
2476 | ||
2477 | void | |
2478 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2479 | { | |
2480 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2481 | dnode_t *dn; | |
2482 | ||
2483 | DB_DNODE_ENTER(db); | |
2484 | dn = DB_DNODE(db); | |
2485 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2486 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2487 | } |
2488 | ||
2489 | void | |
2490 | byteswap_uint64_array(void *vbuf, size_t size) | |
2491 | { | |
2492 | uint64_t *buf = vbuf; | |
2493 | size_t count = size >> 3; | |
2494 | int i; | |
2495 | ||
2496 | ASSERT((size & 7) == 0); | |
2497 | ||
2498 | for (i = 0; i < count; i++) | |
2499 | buf[i] = BSWAP_64(buf[i]); | |
2500 | } | |
2501 | ||
2502 | void | |
2503 | byteswap_uint32_array(void *vbuf, size_t size) | |
2504 | { | |
2505 | uint32_t *buf = vbuf; | |
2506 | size_t count = size >> 2; | |
2507 | int i; | |
2508 | ||
2509 | ASSERT((size & 3) == 0); | |
2510 | ||
2511 | for (i = 0; i < count; i++) | |
2512 | buf[i] = BSWAP_32(buf[i]); | |
2513 | } | |
2514 | ||
2515 | void | |
2516 | byteswap_uint16_array(void *vbuf, size_t size) | |
2517 | { | |
2518 | uint16_t *buf = vbuf; | |
2519 | size_t count = size >> 1; | |
2520 | int i; | |
2521 | ||
2522 | ASSERT((size & 1) == 0); | |
2523 | ||
2524 | for (i = 0; i < count; i++) | |
2525 | buf[i] = BSWAP_16(buf[i]); | |
2526 | } | |
2527 | ||
34dc7c2f BB |
2528 | void |
2529 | byteswap_uint8_array(void *vbuf, size_t size) | |
2530 | { | |
14e4e3cb | 2531 | (void) vbuf, (void) size; |
34dc7c2f BB |
2532 | } |
2533 | ||
2534 | void | |
2535 | dmu_init(void) | |
2536 | { | |
a6255b7f | 2537 | abd_init(); |
428870ff | 2538 | zfs_dbgmsg_init(); |
572e2857 | 2539 | sa_cache_init(); |
572e2857 | 2540 | dmu_objset_init(); |
34dc7c2f | 2541 | dnode_init(); |
428870ff | 2542 | zfetch_init(); |
570827e1 | 2543 | dmu_tx_init(); |
34dc7c2f | 2544 | l2arc_init(); |
29809a6c | 2545 | arc_init(); |
d3c2ae1c | 2546 | dbuf_init(); |
34dc7c2f BB |
2547 | } |
2548 | ||
2549 | void | |
2550 | dmu_fini(void) | |
2551 | { | |
e49f1e20 | 2552 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2553 | l2arc_fini(); |
570827e1 | 2554 | dmu_tx_fini(); |
428870ff | 2555 | zfetch_fini(); |
34dc7c2f | 2556 | dbuf_fini(); |
572e2857 BB |
2557 | dnode_fini(); |
2558 | dmu_objset_fini(); | |
428870ff BB |
2559 | sa_cache_fini(); |
2560 | zfs_dbgmsg_fini(); | |
a6255b7f | 2561 | abd_fini(); |
34dc7c2f | 2562 | } |
c28b2279 | 2563 | |
c28b2279 | 2564 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2565 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2566 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2567 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2568 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2569 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2570 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2571 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2572 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2573 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2574 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2575 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2576 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2577 | EXPORT_SYMBOL(dmu_object_info); |
2578 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2579 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2580 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2581 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2582 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2583 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2584 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2585 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2586 | EXPORT_SYMBOL(dmu_object_set_compress); | |
94dac3e8 | 2587 | EXPORT_SYMBOL(dmu_offset_next); |
57b650b8 BB |
2588 | EXPORT_SYMBOL(dmu_write_policy); |
2589 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2590 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2591 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2592 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2593 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2594 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2595 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2596 | |
03fdcb9a MM |
2597 | ZFS_MODULE_PARAM(zfs, zfs_, nopwrite_enabled, INT, ZMOD_RW, |
2598 | "Enable NOP writes"); | |
03c6040b | 2599 | |
ab8d9c17 | 2600 | ZFS_MODULE_PARAM(zfs, zfs_, per_txg_dirty_frees_percent, UINT, ZMOD_RW, |
03fdcb9a | 2601 | "Percentage of dirtied blocks from frees in one TXG"); |
66aca247 | 2602 | |
03fdcb9a | 2603 | ZFS_MODULE_PARAM(zfs, zfs_, dmu_offset_next_sync, INT, ZMOD_RW, |
66aca247 DB |
2604 | "Enable forcing txg sync to find holes"); |
2605 | ||
7ada752a | 2606 | /* CSTYLED */ |
fdc2d303 | 2607 | ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, UINT, ZMOD_RW, |
d9b4bf06 | 2608 | "Limit one prefetch call to this size"); |