]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
82644107 | 23 | * Copyright (c) 2011, 2017 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
bc77ba73 | 25 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
a08abc1b | 26 | * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved. |
5475aada | 27 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. |
65282ee9 | 28 | * Copyright (c) 2019 Datto Inc. |
34dc7c2f BB |
29 | */ |
30 | ||
34dc7c2f BB |
31 | #include <sys/dmu.h> |
32 | #include <sys/dmu_impl.h> | |
33 | #include <sys/dmu_tx.h> | |
34 | #include <sys/dbuf.h> | |
35 | #include <sys/dnode.h> | |
36 | #include <sys/zfs_context.h> | |
37 | #include <sys/dmu_objset.h> | |
38 | #include <sys/dmu_traverse.h> | |
39 | #include <sys/dsl_dataset.h> | |
40 | #include <sys/dsl_dir.h> | |
41 | #include <sys/dsl_pool.h> | |
42 | #include <sys/dsl_synctask.h> | |
43 | #include <sys/dsl_prop.h> | |
44 | #include <sys/dmu_zfetch.h> | |
45 | #include <sys/zfs_ioctl.h> | |
46 | #include <sys/zap.h> | |
47 | #include <sys/zio_checksum.h> | |
03c6040b | 48 | #include <sys/zio_compress.h> |
428870ff | 49 | #include <sys/sa.h> |
62bdd5eb | 50 | #include <sys/zfeature.h> |
a6255b7f | 51 | #include <sys/abd.h> |
539d33c7 | 52 | #include <sys/trace_dmu.h> |
f763c3d1 | 53 | #include <sys/zfs_rlock.h> |
34dc7c2f BB |
54 | #ifdef _KERNEL |
55 | #include <sys/vmsystm.h> | |
b128c09f | 56 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
57 | #endif |
58 | ||
03c6040b GW |
59 | /* |
60 | * Enable/disable nopwrite feature. | |
61 | */ | |
62 | int zfs_nopwrite_enabled = 1; | |
63 | ||
539d33c7 | 64 | /* |
65282ee9 AP |
65 | * Tunable to control percentage of dirtied L1 blocks from frees allowed into |
66 | * one TXG. After this threshold is crossed, additional dirty blocks from frees | |
67 | * will wait until the next TXG. | |
539d33c7 GM |
68 | * A value of zero will disable this throttle. |
69 | */ | |
65282ee9 | 70 | unsigned long zfs_per_txg_dirty_frees_percent = 5; |
539d33c7 | 71 | |
66aca247 DB |
72 | /* |
73 | * Enable/disable forcing txg sync when dirty in dmu_offset_next. | |
74 | */ | |
75 | int zfs_dmu_offset_next_sync = 0; | |
76 | ||
a1d477c2 MA |
77 | /* |
78 | * This can be used for testing, to ensure that certain actions happen | |
79 | * while in the middle of a remap (which might otherwise complete too | |
65ca2c1e | 80 | * quickly). Used by ztest(8). |
a1d477c2 | 81 | */ |
65ca2c1e | 82 | int zfs_object_remap_one_indirect_delay_ms = 0; |
a1d477c2 | 83 | |
d9b4bf06 MA |
84 | /* |
85 | * Limit the amount we can prefetch with one call to this amount. This | |
86 | * helps to limit the amount of memory that can be used by prefetching. | |
87 | * Larger objects should be prefetched a bit at a time. | |
88 | */ | |
89 | int dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE; | |
90 | ||
34dc7c2f | 91 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
2e5dc449 MA |
92 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" }, |
93 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" }, | |
94 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" }, | |
95 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" }, | |
96 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" }, | |
97 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" }, | |
98 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" }, | |
99 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" }, | |
100 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" }, | |
101 | {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" }, | |
102 | {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" }, | |
103 | {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" }, | |
104 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" }, | |
105 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"}, | |
106 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" }, | |
107 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" }, | |
108 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" }, | |
109 | {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" }, | |
110 | {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" }, | |
111 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" }, | |
112 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" }, | |
113 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" }, | |
114 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" }, | |
115 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" }, | |
116 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" }, | |
117 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" }, | |
118 | {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" }, | |
119 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" }, | |
120 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" }, | |
121 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" }, | |
122 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" }, | |
123 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" }, | |
124 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" }, | |
125 | {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" }, | |
126 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" }, | |
127 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" }, | |
128 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" }, | |
129 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"}, | |
130 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" }, | |
131 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" }, | |
132 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"}, | |
133 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"}, | |
134 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" }, | |
135 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" }, | |
136 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" }, | |
137 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" }, | |
138 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" }, | |
139 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" }, | |
140 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" }, | |
141 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" }, | |
142 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" }, | |
143 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" }, | |
144 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" }, | |
145 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
146 | }; |
147 | ||
148 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
149 | { byteswap_uint8_array, "uint8" }, | |
150 | { byteswap_uint16_array, "uint16" }, | |
151 | { byteswap_uint32_array, "uint32" }, | |
152 | { byteswap_uint64_array, "uint64" }, | |
153 | { zap_byteswap, "zap" }, | |
154 | { dnode_buf_byteswap, "dnode" }, | |
155 | { dmu_objset_byteswap, "objset" }, | |
156 | { zfs_znode_byteswap, "znode" }, | |
157 | { zfs_oldacl_byteswap, "oldacl" }, | |
158 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
159 | }; |
160 | ||
2bce8049 MA |
161 | int |
162 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, | |
163 | void *tag, dmu_buf_t **dbp) | |
164 | { | |
165 | uint64_t blkid; | |
166 | dmu_buf_impl_t *db; | |
167 | ||
168 | blkid = dbuf_whichblock(dn, 0, offset); | |
169 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
170 | db = dbuf_hold(dn, blkid, tag); | |
171 | rw_exit(&dn->dn_struct_rwlock); | |
172 | ||
173 | if (db == NULL) { | |
174 | *dbp = NULL; | |
175 | return (SET_ERROR(EIO)); | |
176 | } | |
177 | ||
178 | *dbp = &db->db; | |
179 | return (0); | |
180 | } | |
34dc7c2f | 181 | int |
9b67f605 MA |
182 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
183 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
184 | { |
185 | dnode_t *dn; | |
186 | uint64_t blkid; | |
187 | dmu_buf_impl_t *db; | |
188 | int err; | |
428870ff BB |
189 | |
190 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
191 | if (err) |
192 | return (err); | |
fcff0f35 | 193 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
194 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
195 | db = dbuf_hold(dn, blkid, tag); | |
196 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
197 | dnode_rele(dn, FTAG); |
198 | ||
34dc7c2f | 199 | if (db == NULL) { |
9b67f605 MA |
200 | *dbp = NULL; |
201 | return (SET_ERROR(EIO)); | |
202 | } | |
203 | ||
204 | *dbp = &db->db; | |
205 | return (err); | |
206 | } | |
207 | ||
2bce8049 MA |
208 | int |
209 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
210 | void *tag, dmu_buf_t **dbp, int flags) | |
211 | { | |
212 | int err; | |
213 | int db_flags = DB_RF_CANFAIL; | |
214 | ||
215 | if (flags & DMU_READ_NO_PREFETCH) | |
216 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
217 | if (flags & DMU_READ_NO_DECRYPT) |
218 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
219 | |
220 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
221 | if (err == 0) { | |
222 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
223 | err = dbuf_read(db, NULL, db_flags); | |
224 | if (err != 0) { | |
225 | dbuf_rele(db, tag); | |
226 | *dbp = NULL; | |
227 | } | |
228 | } | |
229 | ||
230 | return (err); | |
231 | } | |
232 | ||
9b67f605 MA |
233 | int |
234 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
235 | void *tag, dmu_buf_t **dbp, int flags) | |
236 | { | |
237 | int err; | |
238 | int db_flags = DB_RF_CANFAIL; | |
239 | ||
240 | if (flags & DMU_READ_NO_PREFETCH) | |
241 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
242 | if (flags & DMU_READ_NO_DECRYPT) |
243 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
244 | |
245 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
246 | if (err == 0) { | |
247 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 248 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 249 | if (err != 0) { |
34dc7c2f | 250 | dbuf_rele(db, tag); |
9b67f605 | 251 | *dbp = NULL; |
34dc7c2f BB |
252 | } |
253 | } | |
254 | ||
34dc7c2f BB |
255 | return (err); |
256 | } | |
257 | ||
258 | int | |
259 | dmu_bonus_max(void) | |
260 | { | |
50c957f7 | 261 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
262 | } |
263 | ||
264 | int | |
572e2857 | 265 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 266 | { |
572e2857 BB |
267 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
268 | dnode_t *dn; | |
269 | int error; | |
34dc7c2f | 270 | |
572e2857 BB |
271 | DB_DNODE_ENTER(db); |
272 | dn = DB_DNODE(db); | |
273 | ||
274 | if (dn->dn_bonus != db) { | |
2e528b49 | 275 | error = SET_ERROR(EINVAL); |
572e2857 | 276 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 277 | error = SET_ERROR(EINVAL); |
572e2857 BB |
278 | } else { |
279 | dnode_setbonuslen(dn, newsize, tx); | |
280 | error = 0; | |
281 | } | |
282 | ||
283 | DB_DNODE_EXIT(db); | |
284 | return (error); | |
34dc7c2f BB |
285 | } |
286 | ||
428870ff | 287 | int |
572e2857 | 288 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 289 | { |
572e2857 BB |
290 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
291 | dnode_t *dn; | |
292 | int error; | |
428870ff | 293 | |
572e2857 BB |
294 | DB_DNODE_ENTER(db); |
295 | dn = DB_DNODE(db); | |
428870ff | 296 | |
9ae529ec | 297 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 298 | error = SET_ERROR(EINVAL); |
572e2857 | 299 | } else if (dn->dn_bonus != db) { |
2e528b49 | 300 | error = SET_ERROR(EINVAL); |
572e2857 BB |
301 | } else { |
302 | dnode_setbonus_type(dn, type, tx); | |
303 | error = 0; | |
304 | } | |
428870ff | 305 | |
572e2857 BB |
306 | DB_DNODE_EXIT(db); |
307 | return (error); | |
308 | } | |
309 | ||
310 | dmu_object_type_t | |
311 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
312 | { | |
313 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
314 | dnode_t *dn; | |
315 | dmu_object_type_t type; | |
316 | ||
317 | DB_DNODE_ENTER(db); | |
318 | dn = DB_DNODE(db); | |
319 | type = dn->dn_bonustype; | |
320 | DB_DNODE_EXIT(db); | |
321 | ||
322 | return (type); | |
428870ff BB |
323 | } |
324 | ||
325 | int | |
326 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
327 | { | |
328 | dnode_t *dn; | |
329 | int error; | |
330 | ||
331 | error = dnode_hold(os, object, FTAG, &dn); | |
332 | dbuf_rm_spill(dn, tx); | |
333 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
334 | dnode_rm_spill(dn, tx); | |
335 | rw_exit(&dn->dn_struct_rwlock); | |
336 | dnode_rele(dn, FTAG); | |
337 | return (error); | |
338 | } | |
339 | ||
34dc7c2f | 340 | /* |
6955b401 BB |
341 | * Lookup and hold the bonus buffer for the provided dnode. If the dnode |
342 | * has not yet been allocated a new bonus dbuf a will be allocated. | |
343 | * Returns ENOENT, EIO, or 0. | |
34dc7c2f | 344 | */ |
6955b401 BB |
345 | int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp, |
346 | uint32_t flags) | |
34dc7c2f | 347 | { |
34dc7c2f BB |
348 | dmu_buf_impl_t *db; |
349 | int error; | |
b5256303 TC |
350 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
351 | ||
352 | if (flags & DMU_READ_NO_PREFETCH) | |
353 | db_flags |= DB_RF_NOPREFETCH; | |
354 | if (flags & DMU_READ_NO_DECRYPT) | |
355 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 356 | |
34dc7c2f BB |
357 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
358 | if (dn->dn_bonus == NULL) { | |
359 | rw_exit(&dn->dn_struct_rwlock); | |
360 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
361 | if (dn->dn_bonus == NULL) | |
362 | dbuf_create_bonus(dn); | |
363 | } | |
364 | db = dn->dn_bonus; | |
34dc7c2f BB |
365 | |
366 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 367 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 368 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 369 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
370 | } |
371 | ||
372 | /* | |
373 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
374 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
375 | * a dnode hold for every dbuf. | |
376 | */ | |
377 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f | 378 | |
b5256303 TC |
379 | error = dbuf_read(db, NULL, db_flags); |
380 | if (error) { | |
381 | dnode_evict_bonus(dn); | |
382 | dbuf_rele(db, tag); | |
383 | *dbp = NULL; | |
384 | return (error); | |
385 | } | |
34dc7c2f BB |
386 | |
387 | *dbp = &db->db; | |
388 | return (0); | |
389 | } | |
390 | ||
b5256303 | 391 | int |
6955b401 | 392 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) |
b5256303 | 393 | { |
6955b401 BB |
394 | dnode_t *dn; |
395 | int error; | |
396 | ||
397 | error = dnode_hold(os, object, FTAG, &dn); | |
398 | if (error) | |
399 | return (error); | |
400 | ||
401 | error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH); | |
402 | dnode_rele(dn, FTAG); | |
403 | ||
404 | return (error); | |
b5256303 TC |
405 | } |
406 | ||
428870ff BB |
407 | /* |
408 | * returns ENOENT, EIO, or 0. | |
409 | * | |
410 | * This interface will allocate a blank spill dbuf when a spill blk | |
411 | * doesn't already exist on the dnode. | |
412 | * | |
413 | * if you only want to find an already existing spill db, then | |
414 | * dmu_spill_hold_existing() should be used. | |
415 | */ | |
416 | int | |
417 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
418 | { | |
419 | dmu_buf_impl_t *db = NULL; | |
420 | int err; | |
421 | ||
422 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
423 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
424 | ||
425 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
426 | ||
427 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
428 | rw_exit(&dn->dn_struct_rwlock); | |
429 | ||
b182ac00 | 430 | if (db == NULL) { |
431 | *dbp = NULL; | |
432 | return (SET_ERROR(EIO)); | |
433 | } | |
572e2857 BB |
434 | err = dbuf_read(db, NULL, flags); |
435 | if (err == 0) | |
436 | *dbp = &db->db; | |
b182ac00 | 437 | else { |
572e2857 | 438 | dbuf_rele(db, tag); |
b182ac00 | 439 | *dbp = NULL; |
440 | } | |
428870ff BB |
441 | return (err); |
442 | } | |
443 | ||
444 | int | |
445 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
446 | { | |
572e2857 BB |
447 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
448 | dnode_t *dn; | |
428870ff BB |
449 | int err; |
450 | ||
572e2857 BB |
451 | DB_DNODE_ENTER(db); |
452 | dn = DB_DNODE(db); | |
453 | ||
454 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 455 | err = SET_ERROR(EINVAL); |
572e2857 BB |
456 | } else { |
457 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
458 | ||
459 | if (!dn->dn_have_spill) { | |
2e528b49 | 460 | err = SET_ERROR(ENOENT); |
572e2857 BB |
461 | } else { |
462 | err = dmu_spill_hold_by_dnode(dn, | |
463 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
464 | } | |
428870ff | 465 | |
428870ff | 466 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 467 | } |
572e2857 BB |
468 | |
469 | DB_DNODE_EXIT(db); | |
428870ff BB |
470 | return (err); |
471 | } | |
472 | ||
473 | int | |
e7504d7a TC |
474 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag, |
475 | dmu_buf_t **dbp) | |
428870ff | 476 | { |
572e2857 BB |
477 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
478 | dnode_t *dn; | |
479 | int err; | |
e7504d7a TC |
480 | uint32_t db_flags = DB_RF_CANFAIL; |
481 | ||
482 | if (flags & DMU_READ_NO_DECRYPT) | |
483 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
484 | |
485 | DB_DNODE_ENTER(db); | |
486 | dn = DB_DNODE(db); | |
e7504d7a | 487 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
488 | DB_DNODE_EXIT(db); |
489 | ||
490 | return (err); | |
428870ff BB |
491 | } |
492 | ||
34dc7c2f BB |
493 | /* |
494 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
495 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
496 | * and can induce severe lock contention when writing to several files | |
497 | * whose dnodes are in the same block. | |
498 | */ | |
499 | static int | |
9babb374 | 500 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
7f60329a | 501 | boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
34dc7c2f BB |
502 | { |
503 | dmu_buf_t **dbp; | |
504 | uint64_t blkid, nblks, i; | |
9babb374 | 505 | uint32_t dbuf_flags; |
34dc7c2f BB |
506 | int err; |
507 | zio_t *zio; | |
508 | ||
509 | ASSERT(length <= DMU_MAX_ACCESS); | |
510 | ||
7f60329a MA |
511 | /* |
512 | * Note: We directly notify the prefetch code of this read, so that | |
513 | * we can tell it about the multi-block read. dbuf_read() only knows | |
514 | * about the one block it is accessing. | |
515 | */ | |
516 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
517 | DB_RF_NOPREFETCH; | |
34dc7c2f BB |
518 | |
519 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
520 | if (dn->dn_datablkshift) { | |
521 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
522 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
523 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
524 | } else { |
525 | if (offset + length > dn->dn_datablksz) { | |
526 | zfs_panic_recover("zfs: accessing past end of object " | |
527 | "%llx/%llx (size=%u access=%llu+%llu)", | |
528 | (longlong_t)dn->dn_objset-> | |
529 | os_dsl_dataset->ds_object, | |
530 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
531 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 532 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 533 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
534 | } |
535 | nblks = 1; | |
536 | } | |
79c76d5b | 537 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 538 | |
b128c09f | 539 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
fcff0f35 | 540 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f | 541 | for (i = 0; i < nblks; i++) { |
7f60329a | 542 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f BB |
543 | if (db == NULL) { |
544 | rw_exit(&dn->dn_struct_rwlock); | |
545 | dmu_buf_rele_array(dbp, nblks, tag); | |
546 | zio_nowait(zio); | |
2e528b49 | 547 | return (SET_ERROR(EIO)); |
34dc7c2f | 548 | } |
7f60329a | 549 | |
34dc7c2f | 550 | /* initiate async i/o */ |
7f60329a | 551 | if (read) |
9babb374 | 552 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
553 | dbp[i] = &db->db; |
554 | } | |
7f60329a | 555 | |
755065f3 AM |
556 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
557 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
558 | dmu_zfetch(&dn->dn_zfetch, blkid, nblks, | |
559 | read && DNODE_IS_CACHEABLE(dn)); | |
7f60329a | 560 | } |
34dc7c2f BB |
561 | rw_exit(&dn->dn_struct_rwlock); |
562 | ||
563 | /* wait for async i/o */ | |
564 | err = zio_wait(zio); | |
565 | if (err) { | |
566 | dmu_buf_rele_array(dbp, nblks, tag); | |
567 | return (err); | |
568 | } | |
569 | ||
570 | /* wait for other io to complete */ | |
571 | if (read) { | |
572 | for (i = 0; i < nblks; i++) { | |
573 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
574 | mutex_enter(&db->db_mtx); | |
575 | while (db->db_state == DB_READ || | |
576 | db->db_state == DB_FILL) | |
577 | cv_wait(&db->db_changed, &db->db_mtx); | |
578 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 579 | err = SET_ERROR(EIO); |
34dc7c2f BB |
580 | mutex_exit(&db->db_mtx); |
581 | if (err) { | |
582 | dmu_buf_rele_array(dbp, nblks, tag); | |
583 | return (err); | |
584 | } | |
585 | } | |
586 | } | |
587 | ||
588 | *numbufsp = nblks; | |
589 | *dbpp = dbp; | |
590 | return (0); | |
591 | } | |
592 | ||
593 | static int | |
594 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
595 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
596 | { | |
597 | dnode_t *dn; | |
598 | int err; | |
599 | ||
428870ff | 600 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
601 | if (err) |
602 | return (err); | |
603 | ||
604 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 605 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
606 | |
607 | dnode_rele(dn, FTAG); | |
608 | ||
609 | return (err); | |
610 | } | |
611 | ||
612 | int | |
572e2857 | 613 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
7f60329a MA |
614 | uint64_t length, boolean_t read, void *tag, int *numbufsp, |
615 | dmu_buf_t ***dbpp) | |
34dc7c2f | 616 | { |
572e2857 BB |
617 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
618 | dnode_t *dn; | |
34dc7c2f BB |
619 | int err; |
620 | ||
572e2857 BB |
621 | DB_DNODE_ENTER(db); |
622 | dn = DB_DNODE(db); | |
34dc7c2f | 623 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 624 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 625 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
626 | |
627 | return (err); | |
628 | } | |
629 | ||
630 | void | |
631 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
632 | { | |
633 | int i; | |
634 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
635 | ||
636 | if (numbufs == 0) | |
637 | return; | |
638 | ||
639 | for (i = 0; i < numbufs; i++) { | |
640 | if (dbp[i]) | |
641 | dbuf_rele(dbp[i], tag); | |
642 | } | |
643 | ||
644 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
645 | } | |
646 | ||
e8b96c60 | 647 | /* |
fcff0f35 PD |
648 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
649 | * indirect blocks prefeteched will be those that point to the blocks containing | |
650 | * the data starting at offset, and continuing to offset + len. | |
e8b96c60 | 651 | * |
b5256303 TC |
652 | * Note that if the indirect blocks above the blocks being prefetched are not |
653 | * in cache, they will be asychronously read in. | |
e8b96c60 | 654 | */ |
34dc7c2f | 655 | void |
fcff0f35 PD |
656 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
657 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
658 | { |
659 | dnode_t *dn; | |
660 | uint64_t blkid; | |
e8b96c60 | 661 | int nblks, err; |
34dc7c2f | 662 | |
34dc7c2f | 663 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 664 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
665 | |
666 | if (object == 0 || object >= DN_MAX_OBJECT) | |
667 | return; | |
668 | ||
669 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
670 | blkid = dbuf_whichblock(dn, level, |
671 | object * sizeof (dnode_phys_t)); | |
672 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
673 | rw_exit(&dn->dn_struct_rwlock); |
674 | return; | |
675 | } | |
676 | ||
d9b4bf06 MA |
677 | /* |
678 | * See comment before the definition of dmu_prefetch_max. | |
679 | */ | |
680 | len = MIN(len, dmu_prefetch_max); | |
681 | ||
34dc7c2f BB |
682 | /* |
683 | * XXX - Note, if the dnode for the requested object is not | |
684 | * already cached, we will do a *synchronous* read in the | |
685 | * dnode_hold() call. The same is true for any indirects. | |
686 | */ | |
428870ff | 687 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
688 | if (err != 0) |
689 | return; | |
690 | ||
691 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
692 | /* |
693 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
694 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
695 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
696 | * offset) is the first. Then the number we need to prefetch is the | |
697 | * last - first + 1. | |
698 | */ | |
699 | if (level > 0 || dn->dn_datablkshift != 0) { | |
700 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
701 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
702 | } else { |
703 | nblks = (offset < dn->dn_datablksz); | |
704 | } | |
705 | ||
706 | if (nblks != 0) { | |
fcff0f35 | 707 | blkid = dbuf_whichblock(dn, level, offset); |
1c27024e | 708 | for (int i = 0; i < nblks; i++) |
fcff0f35 | 709 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f BB |
710 | } |
711 | ||
712 | rw_exit(&dn->dn_struct_rwlock); | |
713 | ||
714 | dnode_rele(dn, FTAG); | |
715 | } | |
716 | ||
45d1cae3 BB |
717 | /* |
718 | * Get the next "chunk" of file data to free. We traverse the file from | |
719 | * the end so that the file gets shorter over time (if we crashes in the | |
720 | * middle, this will leave us in a better state). We find allocated file | |
721 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
722 | * |
723 | * On input, *start should be the first offset that does not need to be | |
724 | * freed (e.g. "offset + length"). On return, *start will be the first | |
65282ee9 AP |
725 | * offset that should be freed and l1blks is set to the number of level 1 |
726 | * indirect blocks found within the chunk. | |
45d1cae3 | 727 | */ |
b128c09f | 728 | static int |
65282ee9 | 729 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) |
b128c09f | 730 | { |
65282ee9 | 731 | uint64_t blks; |
b663a23d MA |
732 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
733 | /* bytes of data covered by a level-1 indirect block */ | |
ec4afd27 OM |
734 | uint64_t iblkrange = (uint64_t)dn->dn_datablksz * |
735 | EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
b128c09f | 736 | |
b663a23d | 737 | ASSERT3U(minimum, <=, *start); |
b128c09f | 738 | |
f4c594da TC |
739 | /* |
740 | * Check if we can free the entire range assuming that all of the | |
741 | * L1 blocks in this range have data. If we can, we use this | |
742 | * worst case value as an estimate so we can avoid having to look | |
743 | * at the object's actual data. | |
744 | */ | |
745 | uint64_t total_l1blks = | |
746 | (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / | |
747 | iblkrange; | |
748 | if (total_l1blks <= maxblks) { | |
749 | *l1blks = total_l1blks; | |
b663a23d | 750 | *start = minimum; |
b128c09f BB |
751 | return (0); |
752 | } | |
45d1cae3 | 753 | ASSERT(ISP2(iblkrange)); |
b128c09f | 754 | |
65282ee9 | 755 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
756 | int err; |
757 | ||
b663a23d MA |
758 | /* |
759 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
760 | * indirect block at or before the input offset. We must | |
761 | * decrement *start so that it is at the end of the region | |
762 | * to search. | |
763 | */ | |
764 | (*start)--; | |
f4c594da | 765 | |
b128c09f | 766 | err = dnode_next_offset(dn, |
45d1cae3 | 767 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 768 | |
b663a23d | 769 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 770 | if (err == ESRCH) { |
b663a23d MA |
771 | *start = minimum; |
772 | break; | |
773 | } else if (err != 0) { | |
65282ee9 | 774 | *l1blks = blks; |
b128c09f | 775 | return (err); |
45d1cae3 | 776 | } |
b128c09f | 777 | |
b663a23d | 778 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 779 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 780 | } |
b663a23d MA |
781 | if (*start < minimum) |
782 | *start = minimum; | |
65282ee9 | 783 | *l1blks = blks; |
f4c594da | 784 | |
b128c09f BB |
785 | return (0); |
786 | } | |
787 | ||
a08abc1b GM |
788 | /* |
789 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
790 | * otherwise return false. | |
791 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
792 | */ | |
793 | /*ARGSUSED*/ | |
794 | static boolean_t | |
795 | dmu_objset_zfs_unmounting(objset_t *os) | |
796 | { | |
797 | #ifdef _KERNEL | |
798 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
799 | return (zfs_get_vfs_flag_unmounted(os)); | |
800 | #endif | |
801 | return (B_FALSE); | |
802 | } | |
803 | ||
b128c09f BB |
804 | static int |
805 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 806 | uint64_t length) |
b128c09f | 807 | { |
c97d3069 | 808 | uint64_t object_size; |
b663a23d | 809 | int err; |
539d33c7 GM |
810 | uint64_t dirty_frees_threshold; |
811 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 812 | |
c97d3069 BB |
813 | if (dn == NULL) |
814 | return (SET_ERROR(EINVAL)); | |
815 | ||
816 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 817 | if (offset >= object_size) |
b128c09f | 818 | return (0); |
b128c09f | 819 | |
539d33c7 GM |
820 | if (zfs_per_txg_dirty_frees_percent <= 100) |
821 | dirty_frees_threshold = | |
822 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
823 | else | |
65282ee9 | 824 | dirty_frees_threshold = zfs_dirty_data_max / 20; |
539d33c7 | 825 | |
b663a23d MA |
826 | if (length == DMU_OBJECT_END || offset + length > object_size) |
827 | length = object_size - offset; | |
828 | ||
829 | while (length != 0) { | |
539d33c7 | 830 | uint64_t chunk_end, chunk_begin, chunk_len; |
65282ee9 | 831 | uint64_t l1blks; |
b663a23d MA |
832 | dmu_tx_t *tx; |
833 | ||
a08abc1b GM |
834 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
835 | return (SET_ERROR(EINTR)); | |
836 | ||
b663a23d MA |
837 | chunk_end = chunk_begin = offset + length; |
838 | ||
839 | /* move chunk_begin backwards to the beginning of this chunk */ | |
65282ee9 | 840 | err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); |
b128c09f BB |
841 | if (err) |
842 | return (err); | |
b663a23d MA |
843 | ASSERT3U(chunk_begin, >=, offset); |
844 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 845 | |
539d33c7 GM |
846 | chunk_len = chunk_end - chunk_begin; |
847 | ||
b128c09f | 848 | tx = dmu_tx_create(os); |
539d33c7 | 849 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
850 | |
851 | /* | |
852 | * Mark this transaction as typically resulting in a net | |
853 | * reduction in space used. | |
854 | */ | |
855 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
856 | err = dmu_tx_assign(tx, TXG_WAIT); |
857 | if (err) { | |
858 | dmu_tx_abort(tx); | |
859 | return (err); | |
860 | } | |
539d33c7 | 861 | |
f4c594da TC |
862 | uint64_t txg = dmu_tx_get_txg(tx); |
863 | ||
864 | mutex_enter(&dp->dp_lock); | |
865 | uint64_t long_free_dirty = | |
866 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; | |
867 | mutex_exit(&dp->dp_lock); | |
868 | ||
869 | /* | |
870 | * To avoid filling up a TXG with just frees, wait for | |
871 | * the next TXG to open before freeing more chunks if | |
872 | * we have reached the threshold of frees. | |
873 | */ | |
874 | if (dirty_frees_threshold != 0 && | |
875 | long_free_dirty >= dirty_frees_threshold) { | |
876 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); | |
877 | dmu_tx_commit(tx); | |
878 | txg_wait_open(dp, 0, B_TRUE); | |
879 | continue; | |
880 | } | |
881 | ||
65282ee9 AP |
882 | /* |
883 | * In order to prevent unnecessary write throttling, for each | |
884 | * TXG, we track the cumulative size of L1 blocks being dirtied | |
885 | * in dnode_free_range() below. We compare this number to a | |
886 | * tunable threshold, past which we prevent new L1 dirty freeing | |
887 | * blocks from being added into the open TXG. See | |
888 | * dmu_free_long_range_impl() for details. The threshold | |
889 | * prevents write throttle activation due to dirty freeing L1 | |
890 | * blocks taking up a large percentage of zfs_dirty_data_max. | |
891 | */ | |
539d33c7 | 892 | mutex_enter(&dp->dp_lock); |
f4c594da | 893 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += |
65282ee9 | 894 | l1blks << dn->dn_indblkshift; |
539d33c7 GM |
895 | mutex_exit(&dp->dp_lock); |
896 | DTRACE_PROBE3(free__long__range, | |
f4c594da TC |
897 | uint64_t, long_free_dirty, uint64_t, chunk_len, |
898 | uint64_t, txg); | |
539d33c7 | 899 | dnode_free_range(dn, chunk_begin, chunk_len, tx); |
440a3eb9 | 900 | |
b128c09f | 901 | dmu_tx_commit(tx); |
b663a23d | 902 | |
539d33c7 | 903 | length -= chunk_len; |
b128c09f BB |
904 | } |
905 | return (0); | |
906 | } | |
907 | ||
908 | int | |
909 | dmu_free_long_range(objset_t *os, uint64_t object, | |
910 | uint64_t offset, uint64_t length) | |
911 | { | |
912 | dnode_t *dn; | |
913 | int err; | |
914 | ||
428870ff | 915 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
916 | if (err != 0) |
917 | return (err); | |
0c03d21a | 918 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
919 | |
920 | /* | |
921 | * It is important to zero out the maxblkid when freeing the entire | |
922 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
923 | * will take the fast path, and (b) dnode_reallocate() can verify | |
924 | * that the entire file has been freed. | |
925 | */ | |
b0bc7a84 | 926 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
927 | dn->dn_maxblkid = 0; |
928 | ||
b128c09f BB |
929 | dnode_rele(dn, FTAG); |
930 | return (err); | |
931 | } | |
932 | ||
933 | int | |
0c03d21a | 934 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 935 | { |
b128c09f BB |
936 | dmu_tx_t *tx; |
937 | int err; | |
938 | ||
b663a23d | 939 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
940 | if (err != 0) |
941 | return (err); | |
b663a23d MA |
942 | |
943 | tx = dmu_tx_create(os); | |
944 | dmu_tx_hold_bonus(tx, object); | |
945 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 946 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
947 | err = dmu_tx_assign(tx, TXG_WAIT); |
948 | if (err == 0) { | |
35df0bb5 TC |
949 | if (err == 0) |
950 | err = dmu_object_free(os, object, tx); | |
440a3eb9 | 951 | |
b663a23d | 952 | dmu_tx_commit(tx); |
b128c09f | 953 | } else { |
b663a23d | 954 | dmu_tx_abort(tx); |
b128c09f | 955 | } |
b663a23d | 956 | |
b128c09f BB |
957 | return (err); |
958 | } | |
959 | ||
34dc7c2f BB |
960 | int |
961 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
962 | uint64_t size, dmu_tx_t *tx) | |
963 | { | |
964 | dnode_t *dn; | |
428870ff | 965 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
966 | if (err) |
967 | return (err); | |
968 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 969 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
970 | dnode_free_range(dn, offset, size, tx); |
971 | dnode_rele(dn, FTAG); | |
972 | return (0); | |
973 | } | |
974 | ||
0eef1bde | 975 | static int |
976 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 977 | void *buf, uint32_t flags) |
34dc7c2f | 978 | { |
34dc7c2f | 979 | dmu_buf_t **dbp; |
0eef1bde | 980 | int numbufs, err = 0; |
34dc7c2f BB |
981 | |
982 | /* | |
983 | * Deal with odd block sizes, where there can't be data past the first | |
984 | * block. If we ever do the tail block optimization, we will need to | |
985 | * handle that here as well. | |
986 | */ | |
45d1cae3 | 987 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 988 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
989 | MIN(size, dn->dn_datablksz - offset); |
990 | bzero((char *)buf + newsz, size - newsz); | |
991 | size = newsz; | |
992 | } | |
993 | ||
994 | while (size > 0) { | |
995 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 996 | int i; |
34dc7c2f BB |
997 | |
998 | /* | |
999 | * NB: we could do this block-at-a-time, but it's nice | |
1000 | * to be reading in parallel. | |
1001 | */ | |
1002 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 1003 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
1004 | if (err) |
1005 | break; | |
1006 | ||
1007 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1008 | uint64_t tocpy; |
1009 | int64_t bufoff; | |
34dc7c2f BB |
1010 | dmu_buf_t *db = dbp[i]; |
1011 | ||
1012 | ASSERT(size > 0); | |
1013 | ||
1014 | bufoff = offset - db->db_offset; | |
c9520ecc | 1015 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 1016 | |
c9520ecc | 1017 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
1018 | |
1019 | offset += tocpy; | |
1020 | size -= tocpy; | |
1021 | buf = (char *)buf + tocpy; | |
1022 | } | |
1023 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1024 | } | |
34dc7c2f BB |
1025 | return (err); |
1026 | } | |
1027 | ||
0eef1bde | 1028 | int |
1029 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1030 | void *buf, uint32_t flags) | |
34dc7c2f | 1031 | { |
0eef1bde | 1032 | dnode_t *dn; |
1033 | int err; | |
34dc7c2f | 1034 | |
0eef1bde | 1035 | err = dnode_hold(os, object, FTAG, &dn); |
1036 | if (err != 0) | |
1037 | return (err); | |
34dc7c2f | 1038 | |
0eef1bde | 1039 | err = dmu_read_impl(dn, offset, size, buf, flags); |
1040 | dnode_rele(dn, FTAG); | |
1041 | return (err); | |
1042 | } | |
1043 | ||
1044 | int | |
1045 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1046 | uint32_t flags) | |
1047 | { | |
1048 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1049 | } | |
1050 | ||
1051 | static void | |
1052 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1053 | const void *buf, dmu_tx_t *tx) | |
1054 | { | |
1055 | int i; | |
34dc7c2f BB |
1056 | |
1057 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1058 | uint64_t tocpy; |
1059 | int64_t bufoff; | |
34dc7c2f BB |
1060 | dmu_buf_t *db = dbp[i]; |
1061 | ||
1062 | ASSERT(size > 0); | |
1063 | ||
1064 | bufoff = offset - db->db_offset; | |
c9520ecc | 1065 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1066 | |
1067 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1068 | ||
1069 | if (tocpy == db->db_size) | |
1070 | dmu_buf_will_fill(db, tx); | |
1071 | else | |
1072 | dmu_buf_will_dirty(db, tx); | |
1073 | ||
60101509 | 1074 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1075 | |
1076 | if (tocpy == db->db_size) | |
1077 | dmu_buf_fill_done(db, tx); | |
1078 | ||
1079 | offset += tocpy; | |
1080 | size -= tocpy; | |
1081 | buf = (char *)buf + tocpy; | |
1082 | } | |
0eef1bde | 1083 | } |
1084 | ||
1085 | void | |
1086 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1087 | const void *buf, dmu_tx_t *tx) | |
1088 | { | |
1089 | dmu_buf_t **dbp; | |
1090 | int numbufs; | |
1091 | ||
1092 | if (size == 0) | |
1093 | return; | |
1094 | ||
1095 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1096 | FALSE, FTAG, &numbufs, &dbp)); | |
1097 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1098 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1099 | } | |
1100 | ||
1101 | void | |
1102 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1103 | const void *buf, dmu_tx_t *tx) | |
1104 | { | |
1105 | dmu_buf_t **dbp; | |
1106 | int numbufs; | |
1107 | ||
1108 | if (size == 0) | |
1109 | return; | |
1110 | ||
1111 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1112 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1113 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1114 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1115 | } | |
1116 | ||
a1d477c2 MA |
1117 | static int |
1118 | dmu_object_remap_one_indirect(objset_t *os, dnode_t *dn, | |
1119 | uint64_t last_removal_txg, uint64_t offset) | |
1120 | { | |
1121 | uint64_t l1blkid = dbuf_whichblock(dn, 1, offset); | |
65ca2c1e | 1122 | dnode_t *dn_tx; |
a1d477c2 MA |
1123 | int err = 0; |
1124 | ||
1125 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
1126 | dmu_buf_impl_t *dbuf = dbuf_hold_level(dn, 1, l1blkid, FTAG); | |
1127 | ASSERT3P(dbuf, !=, NULL); | |
1128 | ||
1129 | /* | |
1130 | * If the block hasn't been written yet, this default will ensure | |
1131 | * we don't try to remap it. | |
1132 | */ | |
1133 | uint64_t birth = UINT64_MAX; | |
1134 | ASSERT3U(last_removal_txg, !=, UINT64_MAX); | |
1135 | if (dbuf->db_blkptr != NULL) | |
1136 | birth = dbuf->db_blkptr->blk_birth; | |
1137 | rw_exit(&dn->dn_struct_rwlock); | |
1138 | ||
1139 | /* | |
1140 | * If this L1 was already written after the last removal, then we've | |
65ca2c1e BB |
1141 | * already tried to remap it. An additional hold is taken after the |
1142 | * dmu_tx_assign() to handle the case where the dnode is freed while | |
1143 | * waiting for the next open txg. | |
a1d477c2 MA |
1144 | */ |
1145 | if (birth <= last_removal_txg && | |
1146 | dbuf_read(dbuf, NULL, DB_RF_MUST_SUCCEED) == 0 && | |
1147 | dbuf_can_remap(dbuf)) { | |
1148 | dmu_tx_t *tx = dmu_tx_create(os); | |
1149 | dmu_tx_hold_remap_l1indirect(tx, dn->dn_object); | |
1150 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1151 | if (err == 0) { | |
65ca2c1e BB |
1152 | err = dnode_hold(os, dn->dn_object, FTAG, &dn_tx); |
1153 | if (err == 0) { | |
1154 | (void) dbuf_dirty(dbuf, tx); | |
1155 | dnode_rele(dn_tx, FTAG); | |
1156 | } | |
a1d477c2 MA |
1157 | dmu_tx_commit(tx); |
1158 | } else { | |
1159 | dmu_tx_abort(tx); | |
1160 | } | |
1161 | } | |
1162 | ||
1163 | dbuf_rele(dbuf, FTAG); | |
1164 | ||
65ca2c1e | 1165 | delay(MSEC_TO_TICK(zfs_object_remap_one_indirect_delay_ms)); |
a1d477c2 MA |
1166 | |
1167 | return (err); | |
1168 | } | |
1169 | ||
1170 | /* | |
1171 | * Remap all blockpointers in the object, if possible, so that they reference | |
1172 | * only concrete vdevs. | |
1173 | * | |
1174 | * To do this, iterate over the L0 blockpointers and remap any that reference | |
1175 | * an indirect vdev. Note that we only examine L0 blockpointers; since we | |
1176 | * cannot guarantee that we can remap all blockpointer anyways (due to split | |
1177 | * blocks), we do not want to make the code unnecessarily complicated to | |
1178 | * catch the unlikely case that there is an L1 block on an indirect vdev that | |
1179 | * contains no indirect blockpointers. | |
1180 | */ | |
1181 | int | |
1182 | dmu_object_remap_indirects(objset_t *os, uint64_t object, | |
1183 | uint64_t last_removal_txg) | |
1184 | { | |
1185 | uint64_t offset, l1span; | |
1186 | int err; | |
65ca2c1e | 1187 | dnode_t *dn, *dn_tx; |
a1d477c2 MA |
1188 | |
1189 | err = dnode_hold(os, object, FTAG, &dn); | |
1190 | if (err != 0) { | |
1191 | return (err); | |
1192 | } | |
1193 | ||
1194 | if (dn->dn_nlevels <= 1) { | |
1195 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
1196 | err = SET_ERROR(EINTR); | |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * If the dnode has no indirect blocks, we cannot dirty them. | |
1201 | * We still want to remap the blkptr(s) in the dnode if | |
65ca2c1e BB |
1202 | * appropriate, so mark it as dirty. An additional hold is |
1203 | * taken after the dmu_tx_assign() to handle the case where | |
1204 | * the dnode is freed while waiting for the next open txg. | |
a1d477c2 MA |
1205 | */ |
1206 | if (err == 0 && dnode_needs_remap(dn)) { | |
1207 | dmu_tx_t *tx = dmu_tx_create(os); | |
65ca2c1e BB |
1208 | dmu_tx_hold_bonus(tx, object); |
1209 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1210 | if (err == 0) { | |
1211 | err = dnode_hold(os, object, FTAG, &dn_tx); | |
1212 | if (err == 0) { | |
1213 | dnode_setdirty(dn_tx, tx); | |
1214 | dnode_rele(dn_tx, FTAG); | |
1215 | } | |
a1d477c2 MA |
1216 | dmu_tx_commit(tx); |
1217 | } else { | |
1218 | dmu_tx_abort(tx); | |
1219 | } | |
1220 | } | |
1221 | ||
1222 | dnode_rele(dn, FTAG); | |
1223 | return (err); | |
1224 | } | |
1225 | ||
1226 | offset = 0; | |
1227 | l1span = 1ULL << (dn->dn_indblkshift - SPA_BLKPTRSHIFT + | |
1228 | dn->dn_datablkshift); | |
1229 | /* | |
1230 | * Find the next L1 indirect that is not a hole. | |
1231 | */ | |
1232 | while (dnode_next_offset(dn, 0, &offset, 2, 1, 0) == 0) { | |
1233 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
1234 | err = SET_ERROR(EINTR); | |
1235 | break; | |
1236 | } | |
1237 | if ((err = dmu_object_remap_one_indirect(os, dn, | |
1238 | last_removal_txg, offset)) != 0) { | |
1239 | break; | |
1240 | } | |
1241 | offset += l1span; | |
1242 | } | |
1243 | ||
1244 | dnode_rele(dn, FTAG); | |
1245 | return (err); | |
1246 | } | |
1247 | ||
b128c09f BB |
1248 | void |
1249 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1250 | dmu_tx_t *tx) | |
1251 | { | |
1252 | dmu_buf_t **dbp; | |
1253 | int numbufs, i; | |
1254 | ||
1255 | if (size == 0) | |
1256 | return; | |
1257 | ||
1258 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1259 | FALSE, FTAG, &numbufs, &dbp)); | |
1260 | ||
1261 | for (i = 0; i < numbufs; i++) { | |
1262 | dmu_buf_t *db = dbp[i]; | |
1263 | ||
1264 | dmu_buf_will_not_fill(db, tx); | |
1265 | } | |
1266 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1267 | } | |
1268 | ||
9b67f605 MA |
1269 | void |
1270 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1271 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1272 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1273 | { | |
1274 | dmu_buf_t *db; | |
1275 | ||
1276 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1277 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1278 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1279 | FTAG, &db)); | |
1280 | ||
1281 | dmu_buf_write_embedded(db, | |
1282 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1283 | uncompressed_size, compressed_size, byteorder, tx); | |
1284 | ||
1285 | dmu_buf_rele(db, FTAG); | |
1286 | } | |
1287 | ||
428870ff BB |
1288 | /* |
1289 | * DMU support for xuio | |
1290 | */ | |
1291 | kstat_t *xuio_ksp = NULL; | |
1292 | ||
59e6e7ca BB |
1293 | typedef struct xuio_stats { |
1294 | /* loaned yet not returned arc_buf */ | |
1295 | kstat_named_t xuiostat_onloan_rbuf; | |
1296 | kstat_named_t xuiostat_onloan_wbuf; | |
1297 | /* whether a copy is made when loaning out a read buffer */ | |
1298 | kstat_named_t xuiostat_rbuf_copied; | |
1299 | kstat_named_t xuiostat_rbuf_nocopy; | |
1300 | /* whether a copy is made when assigning a write buffer */ | |
1301 | kstat_named_t xuiostat_wbuf_copied; | |
1302 | kstat_named_t xuiostat_wbuf_nocopy; | |
1303 | } xuio_stats_t; | |
1304 | ||
1305 | static xuio_stats_t xuio_stats = { | |
1306 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
1307 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
1308 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
1309 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
1310 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
1311 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
1312 | }; | |
1313 | ||
d1d7e268 MK |
1314 | #define XUIOSTAT_INCR(stat, val) \ |
1315 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
1316 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
59e6e7ca | 1317 | |
5a6765cf | 1318 | #ifdef HAVE_UIO_ZEROCOPY |
428870ff BB |
1319 | int |
1320 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
1321 | { | |
1322 | dmu_xuio_t *priv; | |
1323 | uio_t *uio = &xuio->xu_uio; | |
1324 | ||
1325 | uio->uio_iovcnt = nblk; | |
79c76d5b | 1326 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
428870ff | 1327 | |
79c76d5b | 1328 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
428870ff | 1329 | priv->cnt = nblk; |
79c76d5b | 1330 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
5475aada | 1331 | priv->iovp = (iovec_t *)uio->uio_iov; |
428870ff BB |
1332 | XUIO_XUZC_PRIV(xuio) = priv; |
1333 | ||
1334 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1335 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
1336 | else | |
1337 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
1338 | ||
1339 | return (0); | |
1340 | } | |
1341 | ||
1342 | void | |
1343 | dmu_xuio_fini(xuio_t *xuio) | |
1344 | { | |
1345 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1346 | int nblk = priv->cnt; | |
1347 | ||
1348 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
1349 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
1350 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
1351 | ||
1352 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1353 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
1354 | else | |
1355 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
1360 | * and increase priv->next by 1. | |
1361 | */ | |
1362 | int | |
1363 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
1364 | { | |
1365 | struct iovec *iov; | |
1366 | uio_t *uio = &xuio->xu_uio; | |
1367 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1368 | int i = priv->next++; | |
1369 | ||
1370 | ASSERT(i < priv->cnt); | |
2aa34383 | 1371 | ASSERT(off + n <= arc_buf_lsize(abuf)); |
5475aada | 1372 | iov = (iovec_t *)uio->uio_iov + i; |
428870ff BB |
1373 | iov->iov_base = (char *)abuf->b_data + off; |
1374 | iov->iov_len = n; | |
1375 | priv->bufs[i] = abuf; | |
1376 | return (0); | |
1377 | } | |
1378 | ||
1379 | int | |
1380 | dmu_xuio_cnt(xuio_t *xuio) | |
1381 | { | |
1382 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1383 | return (priv->cnt); | |
1384 | } | |
1385 | ||
1386 | arc_buf_t * | |
1387 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
1388 | { | |
1389 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1390 | ||
1391 | ASSERT(i < priv->cnt); | |
1392 | return (priv->bufs[i]); | |
1393 | } | |
1394 | ||
1395 | void | |
1396 | dmu_xuio_clear(xuio_t *xuio, int i) | |
1397 | { | |
1398 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1399 | ||
1400 | ASSERT(i < priv->cnt); | |
1401 | priv->bufs[i] = NULL; | |
1402 | } | |
5a6765cf | 1403 | #endif /* HAVE_UIO_ZEROCOPY */ |
428870ff BB |
1404 | |
1405 | static void | |
1406 | xuio_stat_init(void) | |
1407 | { | |
1408 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1409 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1410 | KSTAT_FLAG_VIRTUAL); | |
1411 | if (xuio_ksp != NULL) { | |
1412 | xuio_ksp->ks_data = &xuio_stats; | |
1413 | kstat_install(xuio_ksp); | |
1414 | } | |
1415 | } | |
1416 | ||
1417 | static void | |
1418 | xuio_stat_fini(void) | |
1419 | { | |
1420 | if (xuio_ksp != NULL) { | |
1421 | kstat_delete(xuio_ksp); | |
1422 | xuio_ksp = NULL; | |
1423 | } | |
1424 | } | |
1425 | ||
1426 | void | |
5043684a | 1427 | xuio_stat_wbuf_copied(void) |
428870ff BB |
1428 | { |
1429 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1430 | } | |
1431 | ||
1432 | void | |
5043684a | 1433 | xuio_stat_wbuf_nocopy(void) |
428870ff BB |
1434 | { |
1435 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1436 | } | |
1437 | ||
34dc7c2f | 1438 | #ifdef _KERNEL |
5228cf01 | 1439 | int |
804e0504 | 1440 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) |
872e8d26 BB |
1441 | { |
1442 | dmu_buf_t **dbp; | |
1443 | int numbufs, i, err; | |
5a6765cf | 1444 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 | 1445 | xuio_t *xuio = NULL; |
5a6765cf | 1446 | #endif |
872e8d26 BB |
1447 | |
1448 | /* | |
1449 | * NB: we could do this block-at-a-time, but it's nice | |
1450 | * to be reading in parallel. | |
1451 | */ | |
804e0504 MA |
1452 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, |
1453 | TRUE, FTAG, &numbufs, &dbp, 0); | |
872e8d26 BB |
1454 | if (err) |
1455 | return (err); | |
1456 | ||
1457 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1458 | uint64_t tocpy; |
1459 | int64_t bufoff; | |
872e8d26 BB |
1460 | dmu_buf_t *db = dbp[i]; |
1461 | ||
1462 | ASSERT(size > 0); | |
1463 | ||
1464 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1465 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1466 | |
5a6765cf | 1467 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 BB |
1468 | if (xuio) { |
1469 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1470 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1471 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1472 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1473 | if (!err) { | |
1474 | uio->uio_resid -= tocpy; | |
1475 | uio->uio_loffset += tocpy; | |
1476 | } | |
1477 | ||
1478 | if (abuf == dbuf_abuf) | |
1479 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1480 | else | |
1481 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
5a6765cf | 1482 | } else |
1483 | #endif | |
872e8d26 BB |
1484 | err = uiomove((char *)db->db_data + bufoff, tocpy, |
1485 | UIO_READ, uio); | |
872e8d26 BB |
1486 | if (err) |
1487 | break; | |
1488 | ||
1489 | size -= tocpy; | |
1490 | } | |
1491 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1492 | ||
1493 | return (err); | |
1494 | } | |
1495 | ||
804e0504 MA |
1496 | /* |
1497 | * Read 'size' bytes into the uio buffer. | |
1498 | * From object zdb->db_object. | |
1499 | * Starting at offset uio->uio_loffset. | |
1500 | * | |
1501 | * If the caller already has a dbuf in the target object | |
1502 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1503 | * because we don't have to find the dnode_t for the object. | |
1504 | */ | |
1505 | int | |
1506 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1507 | { | |
1508 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1509 | dnode_t *dn; | |
1510 | int err; | |
1511 | ||
1512 | if (size == 0) | |
1513 | return (0); | |
1514 | ||
1515 | DB_DNODE_ENTER(db); | |
1516 | dn = DB_DNODE(db); | |
1517 | err = dmu_read_uio_dnode(dn, uio, size); | |
1518 | DB_DNODE_EXIT(db); | |
1519 | ||
1520 | return (err); | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Read 'size' bytes into the uio buffer. | |
1525 | * From the specified object | |
1526 | * Starting at offset uio->uio_loffset. | |
1527 | */ | |
1528 | int | |
1529 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1530 | { | |
1531 | dnode_t *dn; | |
1532 | int err; | |
1533 | ||
1534 | if (size == 0) | |
1535 | return (0); | |
1536 | ||
1537 | err = dnode_hold(os, object, FTAG, &dn); | |
1538 | if (err) | |
1539 | return (err); | |
1540 | ||
1541 | err = dmu_read_uio_dnode(dn, uio, size); | |
1542 | ||
1543 | dnode_rele(dn, FTAG); | |
1544 | ||
1545 | return (err); | |
1546 | } | |
1547 | ||
5228cf01 | 1548 | int |
872e8d26 BB |
1549 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) |
1550 | { | |
1551 | dmu_buf_t **dbp; | |
1552 | int numbufs; | |
1553 | int err = 0; | |
1554 | int i; | |
1555 | ||
1556 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1557 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1558 | if (err) | |
1559 | return (err); | |
1560 | ||
1561 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1562 | uint64_t tocpy; |
1563 | int64_t bufoff; | |
872e8d26 BB |
1564 | dmu_buf_t *db = dbp[i]; |
1565 | ||
1566 | ASSERT(size > 0); | |
1567 | ||
1568 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1569 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1570 | |
1571 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1572 | ||
1573 | if (tocpy == db->db_size) | |
1574 | dmu_buf_will_fill(db, tx); | |
1575 | else | |
1576 | dmu_buf_will_dirty(db, tx); | |
1577 | ||
1578 | /* | |
1579 | * XXX uiomove could block forever (eg.nfs-backed | |
1580 | * pages). There needs to be a uiolockdown() function | |
1581 | * to lock the pages in memory, so that uiomove won't | |
1582 | * block. | |
1583 | */ | |
1584 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1585 | UIO_WRITE, uio); | |
1586 | ||
1587 | if (tocpy == db->db_size) | |
1588 | dmu_buf_fill_done(db, tx); | |
1589 | ||
1590 | if (err) | |
1591 | break; | |
1592 | ||
1593 | size -= tocpy; | |
1594 | } | |
1595 | ||
1596 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1597 | return (err); | |
1598 | } | |
1599 | ||
804e0504 MA |
1600 | /* |
1601 | * Write 'size' bytes from the uio buffer. | |
1602 | * To object zdb->db_object. | |
1603 | * Starting at offset uio->uio_loffset. | |
1604 | * | |
1605 | * If the caller already has a dbuf in the target object | |
1606 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1607 | * because we don't have to find the dnode_t for the object. | |
1608 | */ | |
428870ff BB |
1609 | int |
1610 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1611 | dmu_tx_t *tx) | |
1612 | { | |
572e2857 BB |
1613 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1614 | dnode_t *dn; | |
1615 | int err; | |
1616 | ||
428870ff BB |
1617 | if (size == 0) |
1618 | return (0); | |
1619 | ||
572e2857 BB |
1620 | DB_DNODE_ENTER(db); |
1621 | dn = DB_DNODE(db); | |
1622 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1623 | DB_DNODE_EXIT(db); | |
1624 | ||
1625 | return (err); | |
428870ff BB |
1626 | } |
1627 | ||
804e0504 MA |
1628 | /* |
1629 | * Write 'size' bytes from the uio buffer. | |
1630 | * To the specified object. | |
1631 | * Starting at offset uio->uio_loffset. | |
1632 | */ | |
428870ff BB |
1633 | int |
1634 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1635 | dmu_tx_t *tx) | |
1636 | { | |
1637 | dnode_t *dn; | |
1638 | int err; | |
1639 | ||
1640 | if (size == 0) | |
1641 | return (0); | |
1642 | ||
1643 | err = dnode_hold(os, object, FTAG, &dn); | |
1644 | if (err) | |
1645 | return (err); | |
1646 | ||
1647 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1648 | ||
1649 | dnode_rele(dn, FTAG); | |
1650 | ||
1651 | return (err); | |
1652 | } | |
872e8d26 | 1653 | #endif /* _KERNEL */ |
34dc7c2f | 1654 | |
9babb374 BB |
1655 | /* |
1656 | * Allocate a loaned anonymous arc buffer. | |
1657 | */ | |
1658 | arc_buf_t * | |
1659 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1660 | { | |
572e2857 | 1661 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1662 | |
2aa34383 | 1663 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1664 | } |
1665 | ||
1666 | /* | |
1667 | * Free a loaned arc buffer. | |
1668 | */ | |
1669 | void | |
1670 | dmu_return_arcbuf(arc_buf_t *buf) | |
1671 | { | |
1672 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1673 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1674 | } |
1675 | ||
b5256303 TC |
1676 | void |
1677 | dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset, | |
1678 | dmu_buf_t *handle, dmu_tx_t *tx) | |
1679 | { | |
1680 | dmu_buf_t *dst_handle; | |
1681 | dmu_buf_impl_t *dstdb; | |
1682 | dmu_buf_impl_t *srcdb = (dmu_buf_impl_t *)handle; | |
149ce888 | 1683 | dmu_object_type_t type; |
b5256303 TC |
1684 | arc_buf_t *abuf; |
1685 | uint64_t datalen; | |
1686 | boolean_t byteorder; | |
1687 | uint8_t salt[ZIO_DATA_SALT_LEN]; | |
1688 | uint8_t iv[ZIO_DATA_IV_LEN]; | |
1689 | uint8_t mac[ZIO_DATA_MAC_LEN]; | |
1690 | ||
1691 | ASSERT3P(srcdb->db_buf, !=, NULL); | |
1692 | ||
1693 | /* hold the db that we want to write to */ | |
1694 | VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &dst_handle, | |
1695 | DMU_READ_NO_DECRYPT)); | |
1696 | dstdb = (dmu_buf_impl_t *)dst_handle; | |
1697 | datalen = arc_buf_size(srcdb->db_buf); | |
1698 | ||
149ce888 TC |
1699 | DB_DNODE_ENTER(dstdb); |
1700 | type = DB_DNODE(dstdb)->dn_type; | |
1701 | DB_DNODE_EXIT(dstdb); | |
1702 | ||
b5256303 TC |
1703 | /* allocated an arc buffer that matches the type of srcdb->db_buf */ |
1704 | if (arc_is_encrypted(srcdb->db_buf)) { | |
1705 | arc_get_raw_params(srcdb->db_buf, &byteorder, salt, iv, mac); | |
1706 | abuf = arc_loan_raw_buf(os->os_spa, dmu_objset_id(os), | |
149ce888 | 1707 | byteorder, salt, iv, mac, type, |
b5256303 TC |
1708 | datalen, arc_buf_lsize(srcdb->db_buf), |
1709 | arc_get_compression(srcdb->db_buf)); | |
1710 | } else { | |
1711 | /* we won't get a compressed db back from dmu_buf_hold() */ | |
1712 | ASSERT3U(arc_get_compression(srcdb->db_buf), | |
1713 | ==, ZIO_COMPRESS_OFF); | |
1714 | abuf = arc_loan_buf(os->os_spa, | |
149ce888 | 1715 | DMU_OT_IS_METADATA(type), datalen); |
b5256303 TC |
1716 | } |
1717 | ||
1718 | ASSERT3U(datalen, ==, arc_buf_size(abuf)); | |
1719 | ||
1720 | /* copy the data to the new buffer and assign it to the dstdb */ | |
1721 | bcopy(srcdb->db_buf->b_data, abuf->b_data, datalen); | |
1722 | dbuf_assign_arcbuf(dstdb, abuf, tx); | |
1723 | dmu_buf_rele(dst_handle, FTAG); | |
1724 | } | |
1725 | ||
9babb374 BB |
1726 | /* |
1727 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1728 | * If this is not possible copy the contents of passed arc buf via | |
1729 | * dmu_write(). | |
1730 | */ | |
305781da | 1731 | int |
440a3eb9 | 1732 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1733 | dmu_tx_t *tx) |
1734 | { | |
9babb374 | 1735 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1736 | objset_t *os = dn->dn_objset; |
1737 | uint64_t object = dn->dn_object; | |
2aa34383 | 1738 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1739 | uint64_t blkid; |
1740 | ||
1741 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1742 | blkid = dbuf_whichblock(dn, 0, offset); |
305781da TC |
1743 | db = dbuf_hold(dn, blkid, FTAG); |
1744 | if (db == NULL) | |
1745 | return (SET_ERROR(EIO)); | |
9babb374 BB |
1746 | rw_exit(&dn->dn_struct_rwlock); |
1747 | ||
88904bb3 MA |
1748 | /* |
1749 | * We can only assign if the offset is aligned, the arc buf is the | |
2aa34383 | 1750 | * same size as the dbuf, and the dbuf is not metadata. |
88904bb3 | 1751 | */ |
2aa34383 | 1752 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
9babb374 BB |
1753 | dbuf_assign_arcbuf(db, buf, tx); |
1754 | dbuf_rele(db, FTAG); | |
1755 | } else { | |
2aa34383 DK |
1756 | /* compressed bufs must always be assignable to their dbuf */ |
1757 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1758 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1759 | |
9babb374 | 1760 | dbuf_rele(db, FTAG); |
572e2857 | 1761 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1762 | dmu_return_arcbuf(buf); |
428870ff | 1763 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 | 1764 | } |
305781da TC |
1765 | |
1766 | return (0); | |
9babb374 BB |
1767 | } |
1768 | ||
305781da | 1769 | int |
440a3eb9 TC |
1770 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
1771 | dmu_tx_t *tx) | |
1772 | { | |
305781da | 1773 | int err; |
440a3eb9 TC |
1774 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1775 | ||
1776 | DB_DNODE_ENTER(dbuf); | |
305781da | 1777 | err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); |
440a3eb9 | 1778 | DB_DNODE_EXIT(dbuf); |
305781da TC |
1779 | |
1780 | return (err); | |
440a3eb9 TC |
1781 | } |
1782 | ||
34dc7c2f | 1783 | typedef struct { |
428870ff BB |
1784 | dbuf_dirty_record_t *dsa_dr; |
1785 | dmu_sync_cb_t *dsa_done; | |
1786 | zgd_t *dsa_zgd; | |
1787 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1788 | } dmu_sync_arg_t; |
1789 | ||
b128c09f BB |
1790 | /* ARGSUSED */ |
1791 | static void | |
1792 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1793 | { | |
428870ff BB |
1794 | dmu_sync_arg_t *dsa = varg; |
1795 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1796 | blkptr_t *bp = zio->io_bp; |
1797 | ||
428870ff BB |
1798 | if (zio->io_error == 0) { |
1799 | if (BP_IS_HOLE(bp)) { | |
1800 | /* | |
1801 | * A block of zeros may compress to a hole, but the | |
1802 | * block size still needs to be known for replay. | |
1803 | */ | |
1804 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1805 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1806 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1807 | BP_SET_FILL(bp, 1); |
428870ff | 1808 | } |
b128c09f BB |
1809 | } |
1810 | } | |
1811 | ||
428870ff BB |
1812 | static void |
1813 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1814 | { | |
1815 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1816 | } | |
1817 | ||
34dc7c2f BB |
1818 | /* ARGSUSED */ |
1819 | static void | |
1820 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1821 | { | |
428870ff BB |
1822 | dmu_sync_arg_t *dsa = varg; |
1823 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1824 | dmu_buf_impl_t *db = dr->dr_dbuf; |
900d09b2 PS |
1825 | zgd_t *zgd = dsa->dsa_zgd; |
1826 | ||
1827 | /* | |
1828 | * Record the vdev(s) backing this blkptr so they can be flushed after | |
1829 | * the writes for the lwb have completed. | |
1830 | */ | |
1831 | if (zio->io_error == 0) { | |
1832 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1833 | } | |
34dc7c2f | 1834 | |
34dc7c2f BB |
1835 | mutex_enter(&db->db_mtx); |
1836 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1837 | if (zio->io_error == 0) { |
03c6040b GW |
1838 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1839 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1840 | blkptr_t *bp = zio->io_bp; |
1841 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1842 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1843 | |
1844 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1845 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1846 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1847 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1848 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1849 | } |
428870ff BB |
1850 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1851 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1852 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1853 | |
1854 | /* | |
1855 | * Old style holes are filled with all zeros, whereas | |
1856 | * new-style holes maintain their lsize, type, level, | |
1857 | * and birth time (see zio_write_compress). While we | |
1858 | * need to reset the BP_SET_LSIZE() call that happened | |
1859 | * in dmu_sync_ready for old style holes, we do *not* | |
1860 | * want to wipe out the information contained in new | |
1861 | * style holes. Thus, only zero out the block pointer if | |
1862 | * it's an old style hole. | |
1863 | */ | |
1864 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1865 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1866 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1867 | } else { | |
1868 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1869 | } | |
34dc7c2f BB |
1870 | cv_broadcast(&db->db_changed); |
1871 | mutex_exit(&db->db_mtx); | |
1872 | ||
428870ff | 1873 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1874 | |
428870ff BB |
1875 | kmem_free(dsa, sizeof (*dsa)); |
1876 | } | |
1877 | ||
1878 | static void | |
1879 | dmu_sync_late_arrival_done(zio_t *zio) | |
1880 | { | |
1881 | blkptr_t *bp = zio->io_bp; | |
1882 | dmu_sync_arg_t *dsa = zio->io_private; | |
900d09b2 PS |
1883 | zgd_t *zgd = dsa->dsa_zgd; |
1884 | ||
1885 | if (zio->io_error == 0) { | |
1886 | /* | |
1887 | * Record the vdev(s) backing this blkptr so they can be | |
1888 | * flushed after the writes for the lwb have completed. | |
1889 | */ | |
1890 | zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); | |
1891 | ||
1892 | if (!BP_IS_HOLE(bp)) { | |
1893 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); | |
1894 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); | |
1895 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1896 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1897 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1898 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1899 | } | |
428870ff BB |
1900 | } |
1901 | ||
1902 | dmu_tx_commit(dsa->dsa_tx); | |
1903 | ||
1904 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1905 | ||
a6255b7f | 1906 | abd_put(zio->io_abd); |
428870ff BB |
1907 | kmem_free(dsa, sizeof (*dsa)); |
1908 | } | |
1909 | ||
1910 | static int | |
1911 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1912 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1913 | { |
1914 | dmu_sync_arg_t *dsa; | |
1915 | dmu_tx_t *tx; | |
1916 | ||
1917 | tx = dmu_tx_create(os); | |
1918 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1919 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1920 | dmu_tx_abort(tx); | |
2e528b49 MA |
1921 | /* Make zl_get_data do txg_waited_synced() */ |
1922 | return (SET_ERROR(EIO)); | |
428870ff BB |
1923 | } |
1924 | ||
1ce23dca PS |
1925 | /* |
1926 | * In order to prevent the zgd's lwb from being free'd prior to | |
1927 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1928 | * the lwb's "max txg" takes this tx's txg into account. | |
1929 | */ | |
1930 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1931 | ||
79c76d5b | 1932 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1933 | dsa->dsa_dr = NULL; |
1934 | dsa->dsa_done = done; | |
1935 | dsa->dsa_zgd = zgd; | |
1936 | dsa->dsa_tx = tx; | |
1937 | ||
02dc43bc MA |
1938 | /* |
1939 | * Since we are currently syncing this txg, it's nontrivial to | |
1940 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1941 | * | |
1942 | * When syncing, the db_blkptr is initially the BP of the previous | |
1943 | * txg. We can not nopwrite against it because it will be changed | |
1944 | * (this is similar to the non-late-arrival case where the dbuf is | |
1945 | * dirty in a future txg). | |
1946 | * | |
1947 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1948 | * We can not nopwrite against it because although the BP will not | |
1949 | * (typically) be changed, the data has not yet been persisted to this | |
1950 | * location. | |
1951 | * | |
1952 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1953 | * possible to always nopwrite, because the data that was written in | |
1954 | * this txg is the same data that we are trying to write. However we | |
1955 | * would need to check that this dbuf is not dirty in any future | |
1956 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1957 | * don't nopwrite in this case. | |
1958 | */ | |
1959 | zp->zp_nopwrite = B_FALSE; | |
1960 | ||
a6255b7f DQ |
1961 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1962 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1963 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1964 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1965 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1966 | |
1967 | return (0); | |
34dc7c2f BB |
1968 | } |
1969 | ||
1970 | /* | |
1971 | * Intent log support: sync the block associated with db to disk. | |
1972 | * N.B. and XXX: the caller is responsible for making sure that the | |
1973 | * data isn't changing while dmu_sync() is writing it. | |
1974 | * | |
1975 | * Return values: | |
1976 | * | |
03c6040b | 1977 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1978 | * The caller should not log the write. |
1979 | * | |
1980 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1981 | * The caller should not log the write. | |
1982 | * | |
1983 | * EALREADY: this block is already in the process of being synced. | |
1984 | * The caller should track its progress (somehow). | |
1985 | * | |
428870ff BB |
1986 | * EIO: could not do the I/O. |
1987 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1988 | * |
428870ff BB |
1989 | * 0: the I/O has been initiated. |
1990 | * The caller should log this blkptr in the done callback. | |
1991 | * It is possible that the I/O will fail, in which case | |
1992 | * the error will be reported to the done callback and | |
1993 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1994 | */ |
1995 | int | |
428870ff | 1996 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1997 | { |
428870ff BB |
1998 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1999 | objset_t *os = db->db_objset; | |
2000 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 2001 | dbuf_dirty_record_t *dr; |
428870ff | 2002 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 2003 | zbookmark_phys_t zb; |
428870ff | 2004 | zio_prop_t zp; |
572e2857 | 2005 | dnode_t *dn; |
34dc7c2f | 2006 | |
428870ff | 2007 | ASSERT(pio != NULL); |
34dc7c2f BB |
2008 | ASSERT(txg != 0); |
2009 | ||
428870ff BB |
2010 | SET_BOOKMARK(&zb, ds->ds_object, |
2011 | db->db.db_object, db->db_level, db->db_blkid); | |
2012 | ||
572e2857 BB |
2013 | DB_DNODE_ENTER(db); |
2014 | dn = DB_DNODE(db); | |
82644107 | 2015 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 2016 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2017 | |
2018 | /* | |
428870ff | 2019 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 2020 | */ |
428870ff BB |
2021 | if (txg > spa_freeze_txg(os->os_spa)) |
2022 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
2023 | |
2024 | /* | |
428870ff BB |
2025 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
2026 | * and us. If we determine that this txg is not yet syncing, | |
2027 | * but it begins to sync a moment later, that's OK because the | |
2028 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 2029 | */ |
428870ff BB |
2030 | mutex_enter(&db->db_mtx); |
2031 | ||
2032 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 2033 | /* |
428870ff | 2034 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 2035 | */ |
428870ff | 2036 | mutex_exit(&db->db_mtx); |
2e528b49 | 2037 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
2038 | } |
2039 | ||
428870ff BB |
2040 | if (txg <= spa_syncing_txg(os->os_spa)) { |
2041 | /* | |
2042 | * This txg is currently syncing, so we can't mess with | |
2043 | * the dirty record anymore; just write a new log block. | |
2044 | */ | |
2045 | mutex_exit(&db->db_mtx); | |
2046 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
2047 | } |
2048 | ||
2049 | dr = db->db_last_dirty; | |
428870ff | 2050 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 2051 | dr = dr->dr_next; |
428870ff BB |
2052 | |
2053 | if (dr == NULL) { | |
34dc7c2f | 2054 | /* |
428870ff | 2055 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
2056 | * There's no need to log writes to freed blocks, so we're done. |
2057 | */ | |
2058 | mutex_exit(&db->db_mtx); | |
2e528b49 | 2059 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
2060 | } |
2061 | ||
03c6040b GW |
2062 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); |
2063 | ||
02dc43bc MA |
2064 | if (db->db_blkptr != NULL) { |
2065 | /* | |
2066 | * We need to fill in zgd_bp with the current blkptr so that | |
2067 | * the nopwrite code can check if we're writing the same | |
2068 | * data that's already on disk. We can only nopwrite if we | |
2069 | * are sure that after making the copy, db_blkptr will not | |
2070 | * change until our i/o completes. We ensure this by | |
2071 | * holding the db_mtx, and only allowing nopwrite if the | |
2072 | * block is not already dirty (see below). This is verified | |
2073 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
2074 | * not changed. | |
2075 | */ | |
2076 | *zgd->zgd_bp = *db->db_blkptr; | |
2077 | } | |
2078 | ||
03c6040b | 2079 | /* |
f3c517d8 MA |
2080 | * Assume the on-disk data is X, the current syncing data (in |
2081 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
2082 | * in dmu_sync). | |
2083 | * | |
2084 | * We usually want to perform a nopwrite if X and Z are the | |
2085 | * same. However, if Y is different (i.e. the BP is going to | |
2086 | * change before this write takes effect), then a nopwrite will | |
2087 | * be incorrect - we would override with X, which could have | |
2088 | * been freed when Y was written. | |
2089 | * | |
2090 | * (Note that this is not a concern when we are nop-writing from | |
2091 | * syncing context, because X and Y must be identical, because | |
2092 | * all previous txgs have been synced.) | |
2093 | * | |
2094 | * Therefore, we disable nopwrite if the current BP could change | |
2095 | * before this TXG. There are two ways it could change: by | |
2096 | * being dirty (dr_next is non-NULL), or by being freed | |
2097 | * (dnode_block_freed()). This behavior is verified by | |
2098 | * zio_done(), which VERIFYs that the override BP is identical | |
2099 | * to the on-disk BP. | |
03c6040b | 2100 | */ |
f3c517d8 MA |
2101 | DB_DNODE_ENTER(db); |
2102 | dn = DB_DNODE(db); | |
2103 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
03c6040b | 2104 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 2105 | DB_DNODE_EXIT(db); |
03c6040b | 2106 | |
34dc7c2f | 2107 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
2108 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
2109 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 2110 | /* |
428870ff BB |
2111 | * We have already issued a sync write for this buffer, |
2112 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
2113 | * have been dirtied since, or we would have cleared the state. |
2114 | */ | |
34dc7c2f | 2115 | mutex_exit(&db->db_mtx); |
2e528b49 | 2116 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
2117 | } |
2118 | ||
428870ff | 2119 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 2120 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 2121 | mutex_exit(&db->db_mtx); |
34dc7c2f | 2122 | |
79c76d5b | 2123 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
2124 | dsa->dsa_dr = dr; |
2125 | dsa->dsa_done = done; | |
2126 | dsa->dsa_zgd = zgd; | |
2127 | dsa->dsa_tx = NULL; | |
b128c09f | 2128 | |
428870ff | 2129 | zio_nowait(arc_write(pio, os->os_spa, txg, |
02dc43bc | 2130 | zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 2131 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 2132 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 2133 | |
428870ff | 2134 | return (0); |
34dc7c2f BB |
2135 | } |
2136 | ||
b5256303 TC |
2137 | int |
2138 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
2139 | { | |
2140 | dnode_t *dn; | |
2141 | int err; | |
2142 | ||
2143 | err = dnode_hold(os, object, FTAG, &dn); | |
2144 | if (err) | |
2145 | return (err); | |
2146 | err = dnode_set_nlevels(dn, nlevels, tx); | |
2147 | dnode_rele(dn, FTAG); | |
2148 | return (err); | |
2149 | } | |
2150 | ||
34dc7c2f BB |
2151 | int |
2152 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 2153 | dmu_tx_t *tx) |
34dc7c2f BB |
2154 | { |
2155 | dnode_t *dn; | |
2156 | int err; | |
2157 | ||
428870ff | 2158 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2159 | if (err) |
2160 | return (err); | |
2161 | err = dnode_set_blksz(dn, size, ibs, tx); | |
2162 | dnode_rele(dn, FTAG); | |
2163 | return (err); | |
2164 | } | |
2165 | ||
ae76f45c TC |
2166 | int |
2167 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
2168 | dmu_tx_t *tx) | |
2169 | { | |
2170 | dnode_t *dn; | |
2171 | int err; | |
2172 | ||
2173 | err = dnode_hold(os, object, FTAG, &dn); | |
2174 | if (err) | |
2175 | return (err); | |
2176 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
369aa501 | 2177 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE); |
ae76f45c TC |
2178 | rw_exit(&dn->dn_struct_rwlock); |
2179 | dnode_rele(dn, FTAG); | |
2180 | return (0); | |
2181 | } | |
2182 | ||
34dc7c2f BB |
2183 | void |
2184 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 2185 | dmu_tx_t *tx) |
34dc7c2f BB |
2186 | { |
2187 | dnode_t *dn; | |
2188 | ||
9b67f605 MA |
2189 | /* |
2190 | * Send streams include each object's checksum function. This | |
2191 | * check ensures that the receiving system can understand the | |
2192 | * checksum function transmitted. | |
2193 | */ | |
2194 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
2195 | ||
2196 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
2197 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
2198 | dn->dn_checksum = checksum; |
2199 | dnode_setdirty(dn, tx); | |
2200 | dnode_rele(dn, FTAG); | |
2201 | } | |
2202 | ||
2203 | void | |
2204 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 2205 | dmu_tx_t *tx) |
34dc7c2f BB |
2206 | { |
2207 | dnode_t *dn; | |
2208 | ||
9b67f605 MA |
2209 | /* |
2210 | * Send streams include each object's compression function. This | |
2211 | * check ensures that the receiving system can understand the | |
2212 | * compression function transmitted. | |
2213 | */ | |
2214 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
2215 | ||
2216 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
2217 | dn->dn_compress = compress; |
2218 | dnode_setdirty(dn, tx); | |
2219 | dnode_rele(dn, FTAG); | |
2220 | } | |
2221 | ||
faf0f58c MA |
2222 | /* |
2223 | * When the "redundant_metadata" property is set to "most", only indirect | |
2224 | * blocks of this level and higher will have an additional ditto block. | |
2225 | */ | |
2226 | int zfs_redundant_metadata_most_ditto_level = 2; | |
2227 | ||
428870ff | 2228 | void |
82644107 | 2229 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
2230 | { |
2231 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 2232 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 2233 | (wp & WP_SPILL)); |
428870ff BB |
2234 | enum zio_checksum checksum = os->os_checksum; |
2235 | enum zio_compress compress = os->os_compress; | |
2236 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
03c6040b GW |
2237 | boolean_t dedup = B_FALSE; |
2238 | boolean_t nopwrite = B_FALSE; | |
428870ff | 2239 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 2240 | boolean_t encrypt = B_FALSE; |
428870ff | 2241 | int copies = os->os_copies; |
a7004725 | 2242 | |
428870ff | 2243 | /* |
03c6040b GW |
2244 | * We maintain different write policies for each of the following |
2245 | * types of data: | |
2246 | * 1. metadata | |
2247 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
2248 | * 3. all other level 0 blocks | |
428870ff BB |
2249 | */ |
2250 | if (ismd) { | |
b1d21733 TC |
2251 | /* |
2252 | * XXX -- we should design a compression algorithm | |
2253 | * that specializes in arrays of bps. | |
2254 | */ | |
2255 | compress = zio_compress_select(os->os_spa, | |
2256 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 2257 | |
428870ff BB |
2258 | /* |
2259 | * Metadata always gets checksummed. If the data | |
2260 | * checksum is multi-bit correctable, and it's not a | |
2261 | * ZBT-style checksum, then it's suitable for metadata | |
2262 | * as well. Otherwise, the metadata checksum defaults | |
2263 | * to fletcher4. | |
2264 | */ | |
3c67d83a TH |
2265 | if (!(zio_checksum_table[checksum].ci_flags & |
2266 | ZCHECKSUM_FLAG_METADATA) || | |
2267 | (zio_checksum_table[checksum].ci_flags & | |
2268 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 2269 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
2270 | |
2271 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
2272 | (os->os_redundant_metadata == | |
2273 | ZFS_REDUNDANT_METADATA_MOST && | |
2274 | (level >= zfs_redundant_metadata_most_ditto_level || | |
2275 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
2276 | copies++; | |
03c6040b GW |
2277 | } else if (wp & WP_NOFILL) { |
2278 | ASSERT(level == 0); | |
428870ff | 2279 | |
428870ff | 2280 | /* |
03c6040b GW |
2281 | * If we're writing preallocated blocks, we aren't actually |
2282 | * writing them so don't set any policy properties. These | |
2283 | * blocks are currently only used by an external subsystem | |
2284 | * outside of zfs (i.e. dump) and not written by the zio | |
2285 | * pipeline. | |
428870ff | 2286 | */ |
03c6040b GW |
2287 | compress = ZIO_COMPRESS_OFF; |
2288 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2289 | } else { |
99197f03 JG |
2290 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2291 | compress); | |
428870ff | 2292 | |
03c6040b GW |
2293 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2294 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2295 | dedup_checksum; | |
428870ff | 2296 | |
03c6040b GW |
2297 | /* |
2298 | * Determine dedup setting. If we are in dmu_sync(), | |
2299 | * we won't actually dedup now because that's all | |
2300 | * done in syncing context; but we do want to use the | |
2301 | * dedup checkum. If the checksum is not strong | |
2302 | * enough to ensure unique signatures, force | |
2303 | * dedup_verify. | |
2304 | */ | |
2305 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2306 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2307 | if (!(zio_checksum_table[checksum].ci_flags & |
2308 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2309 | dedup_verify = B_TRUE; |
2310 | } | |
428870ff | 2311 | |
03c6040b | 2312 | /* |
3c67d83a TH |
2313 | * Enable nopwrite if we have secure enough checksum |
2314 | * algorithm (see comment in zio_nop_write) and | |
2315 | * compression is enabled. We don't enable nopwrite if | |
2316 | * dedup is enabled as the two features are mutually | |
2317 | * exclusive. | |
03c6040b | 2318 | */ |
3c67d83a TH |
2319 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2320 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2321 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2322 | } |
2323 | ||
b5256303 TC |
2324 | /* |
2325 | * All objects in an encrypted objset are protected from modification | |
2326 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2327 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2328 | * not subject to nopwrite since writing the same data will still | |
2329 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2330 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2331 | * object types. | |
2332 | */ | |
2333 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2334 | encrypt = B_TRUE; | |
2335 | ||
2336 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2337 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2338 | nopwrite = B_FALSE; | |
2339 | } else { | |
2340 | dedup = B_FALSE; | |
2341 | } | |
2342 | ||
ae76f45c TC |
2343 | if (level <= 0 && |
2344 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2345 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2346 | } |
b5256303 | 2347 | } |
2aa34383 | 2348 | |
b5256303 TC |
2349 | zp->zp_compress = compress; |
2350 | zp->zp_checksum = checksum; | |
428870ff BB |
2351 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2352 | zp->zp_level = level; | |
faf0f58c | 2353 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2354 | zp->zp_dedup = dedup; |
2355 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2356 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2357 | zp->zp_encrypt = encrypt; |
2358 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
2359 | bzero(zp->zp_salt, ZIO_DATA_SALT_LEN); | |
2360 | bzero(zp->zp_iv, ZIO_DATA_IV_LEN); | |
2361 | bzero(zp->zp_mac, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2362 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2363 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2364 | |
2365 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2366 | } |
2367 | ||
66aca247 DB |
2368 | /* |
2369 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2370 | * in order to determine the location of holes. In order to accurately | |
2371 | * report holes all dirty data must be synced to disk. This causes extremely | |
2372 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2373 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2374 | * report the dnode as having no holes which is always a safe thing to do. | |
2375 | */ | |
34dc7c2f BB |
2376 | int |
2377 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2378 | { | |
2379 | dnode_t *dn; | |
2380 | int i, err; | |
66aca247 | 2381 | boolean_t clean = B_TRUE; |
34dc7c2f | 2382 | |
428870ff | 2383 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2384 | if (err) |
2385 | return (err); | |
66aca247 | 2386 | |
34dc7c2f | 2387 | /* |
2531ce37 | 2388 | * Check if dnode is dirty |
34dc7c2f | 2389 | */ |
454365bb | 2390 | for (i = 0; i < TXG_SIZE; i++) { |
edc1e713 | 2391 | if (multilist_link_active(&dn->dn_dirty_link[i])) { |
2531ce37 | 2392 | clean = B_FALSE; |
ec4f9b8f | 2393 | break; |
2531ce37 | 2394 | } |
34dc7c2f | 2395 | } |
66aca247 DB |
2396 | |
2397 | /* | |
2398 | * If compatibility option is on, sync any current changes before | |
2399 | * we go trundling through the block pointers. | |
2400 | */ | |
2401 | if (!clean && zfs_dmu_offset_next_sync) { | |
2402 | clean = B_TRUE; | |
34dc7c2f BB |
2403 | dnode_rele(dn, FTAG); |
2404 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 2405 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2406 | if (err) |
2407 | return (err); | |
2408 | } | |
2409 | ||
66aca247 DB |
2410 | if (clean) |
2411 | err = dnode_next_offset(dn, | |
2412 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2413 | else | |
2414 | err = SET_ERROR(EBUSY); | |
2415 | ||
34dc7c2f BB |
2416 | dnode_rele(dn, FTAG); |
2417 | ||
2418 | return (err); | |
2419 | } | |
2420 | ||
2421 | void | |
e0b0ca98 | 2422 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2423 | { |
e0b0ca98 | 2424 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2425 | |
34dc7c2f BB |
2426 | doi->doi_data_block_size = dn->dn_datablksz; |
2427 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2428 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2429 | doi->doi_type = dn->dn_type; |
2430 | doi->doi_bonus_type = dn->dn_bonustype; | |
2431 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2432 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2433 | doi->doi_indirection = dn->dn_nlevels; |
2434 | doi->doi_checksum = dn->dn_checksum; | |
2435 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2436 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2437 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2438 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2439 | doi->doi_fill_count = 0; |
1c27024e | 2440 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2441 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2442 | } |
2443 | ||
2444 | void | |
2445 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2446 | { | |
2447 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2448 | mutex_enter(&dn->dn_mtx); | |
2449 | ||
2450 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2451 | |
2452 | mutex_exit(&dn->dn_mtx); | |
2453 | rw_exit(&dn->dn_struct_rwlock); | |
2454 | } | |
2455 | ||
2456 | /* | |
2457 | * Get information on a DMU object. | |
2458 | * If doi is NULL, just indicates whether the object exists. | |
2459 | */ | |
2460 | int | |
2461 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2462 | { | |
2463 | dnode_t *dn; | |
428870ff | 2464 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2465 | |
2466 | if (err) | |
2467 | return (err); | |
2468 | ||
2469 | if (doi != NULL) | |
2470 | dmu_object_info_from_dnode(dn, doi); | |
2471 | ||
2472 | dnode_rele(dn, FTAG); | |
2473 | return (0); | |
2474 | } | |
2475 | ||
2476 | /* | |
2477 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2478 | */ | |
2479 | void | |
572e2857 | 2480 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2481 | { |
572e2857 BB |
2482 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2483 | ||
2484 | DB_DNODE_ENTER(db); | |
2485 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2486 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2487 | } |
2488 | ||
2489 | /* | |
2490 | * Faster still when you only care about the size. | |
2491 | * This is specifically optimized for zfs_getattr(). | |
2492 | */ | |
2493 | void | |
572e2857 BB |
2494 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2495 | u_longlong_t *nblk512) | |
34dc7c2f | 2496 | { |
572e2857 BB |
2497 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2498 | dnode_t *dn; | |
2499 | ||
2500 | DB_DNODE_ENTER(db); | |
2501 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2502 | |
2503 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2504 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2505 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2506 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2507 | DB_DNODE_EXIT(db); | |
2508 | } | |
2509 | ||
2510 | void | |
2511 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2512 | { | |
2513 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2514 | dnode_t *dn; | |
2515 | ||
2516 | DB_DNODE_ENTER(db); | |
2517 | dn = DB_DNODE(db); | |
2518 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2519 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2520 | } |
2521 | ||
2522 | void | |
2523 | byteswap_uint64_array(void *vbuf, size_t size) | |
2524 | { | |
2525 | uint64_t *buf = vbuf; | |
2526 | size_t count = size >> 3; | |
2527 | int i; | |
2528 | ||
2529 | ASSERT((size & 7) == 0); | |
2530 | ||
2531 | for (i = 0; i < count; i++) | |
2532 | buf[i] = BSWAP_64(buf[i]); | |
2533 | } | |
2534 | ||
2535 | void | |
2536 | byteswap_uint32_array(void *vbuf, size_t size) | |
2537 | { | |
2538 | uint32_t *buf = vbuf; | |
2539 | size_t count = size >> 2; | |
2540 | int i; | |
2541 | ||
2542 | ASSERT((size & 3) == 0); | |
2543 | ||
2544 | for (i = 0; i < count; i++) | |
2545 | buf[i] = BSWAP_32(buf[i]); | |
2546 | } | |
2547 | ||
2548 | void | |
2549 | byteswap_uint16_array(void *vbuf, size_t size) | |
2550 | { | |
2551 | uint16_t *buf = vbuf; | |
2552 | size_t count = size >> 1; | |
2553 | int i; | |
2554 | ||
2555 | ASSERT((size & 1) == 0); | |
2556 | ||
2557 | for (i = 0; i < count; i++) | |
2558 | buf[i] = BSWAP_16(buf[i]); | |
2559 | } | |
2560 | ||
2561 | /* ARGSUSED */ | |
2562 | void | |
2563 | byteswap_uint8_array(void *vbuf, size_t size) | |
2564 | { | |
2565 | } | |
2566 | ||
2567 | void | |
2568 | dmu_init(void) | |
2569 | { | |
a6255b7f | 2570 | abd_init(); |
428870ff | 2571 | zfs_dbgmsg_init(); |
572e2857 BB |
2572 | sa_cache_init(); |
2573 | xuio_stat_init(); | |
2574 | dmu_objset_init(); | |
34dc7c2f | 2575 | dnode_init(); |
428870ff | 2576 | zfetch_init(); |
570827e1 | 2577 | dmu_tx_init(); |
34dc7c2f | 2578 | l2arc_init(); |
29809a6c | 2579 | arc_init(); |
d3c2ae1c | 2580 | dbuf_init(); |
34dc7c2f BB |
2581 | } |
2582 | ||
2583 | void | |
2584 | dmu_fini(void) | |
2585 | { | |
e49f1e20 | 2586 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2587 | l2arc_fini(); |
570827e1 | 2588 | dmu_tx_fini(); |
428870ff | 2589 | zfetch_fini(); |
34dc7c2f | 2590 | dbuf_fini(); |
572e2857 BB |
2591 | dnode_fini(); |
2592 | dmu_objset_fini(); | |
428870ff BB |
2593 | xuio_stat_fini(); |
2594 | sa_cache_fini(); | |
2595 | zfs_dbgmsg_fini(); | |
a6255b7f | 2596 | abd_fini(); |
34dc7c2f | 2597 | } |
c28b2279 | 2598 | |
93ce2b4c | 2599 | #if defined(_KERNEL) |
c28b2279 | 2600 | EXPORT_SYMBOL(dmu_bonus_hold); |
6955b401 | 2601 | EXPORT_SYMBOL(dmu_bonus_hold_by_dnode); |
a473d90c AZ |
2602 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2603 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2604 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2605 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2606 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2607 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2608 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2609 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2610 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2611 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2612 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2613 | EXPORT_SYMBOL(dmu_object_info); |
2614 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2615 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2616 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2617 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2618 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2619 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2620 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2621 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2622 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
2623 | EXPORT_SYMBOL(dmu_write_policy); |
2624 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2625 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2626 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2627 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2628 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2629 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2630 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2631 | |
bef78122 | 2632 | /* BEGIN CSTYLED */ |
03c6040b GW |
2633 | module_param(zfs_nopwrite_enabled, int, 0644); |
2634 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2635 | ||
bef78122 DQ |
2636 | module_param(zfs_per_txg_dirty_frees_percent, ulong, 0644); |
2637 | MODULE_PARM_DESC(zfs_per_txg_dirty_frees_percent, | |
2638 | "percentage of dirtied blocks from frees in one TXG"); | |
66aca247 DB |
2639 | |
2640 | module_param(zfs_dmu_offset_next_sync, int, 0644); | |
2641 | MODULE_PARM_DESC(zfs_dmu_offset_next_sync, | |
2642 | "Enable forcing txg sync to find holes"); | |
2643 | ||
d9b4bf06 MA |
2644 | module_param(dmu_prefetch_max, int, 0644); |
2645 | MODULE_PARM_DESC(dmu_prefetch_max, | |
2646 | "Limit one prefetch call to this size"); | |
2647 | ||
bef78122 | 2648 | /* END CSTYLED */ |
66aca247 | 2649 | |
c28b2279 | 2650 | #endif |