]>
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. |
34dc7c2f BB |
28 | */ |
29 | ||
34dc7c2f BB |
30 | #include <sys/dmu.h> |
31 | #include <sys/dmu_impl.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/dbuf.h> | |
34 | #include <sys/dnode.h> | |
35 | #include <sys/zfs_context.h> | |
36 | #include <sys/dmu_objset.h> | |
37 | #include <sys/dmu_traverse.h> | |
38 | #include <sys/dsl_dataset.h> | |
39 | #include <sys/dsl_dir.h> | |
40 | #include <sys/dsl_pool.h> | |
41 | #include <sys/dsl_synctask.h> | |
42 | #include <sys/dsl_prop.h> | |
43 | #include <sys/dmu_zfetch.h> | |
44 | #include <sys/zfs_ioctl.h> | |
45 | #include <sys/zap.h> | |
46 | #include <sys/zio_checksum.h> | |
03c6040b | 47 | #include <sys/zio_compress.h> |
428870ff | 48 | #include <sys/sa.h> |
62bdd5eb | 49 | #include <sys/zfeature.h> |
a6255b7f | 50 | #include <sys/abd.h> |
539d33c7 | 51 | #include <sys/trace_dmu.h> |
f763c3d1 | 52 | #include <sys/zfs_rlock.h> |
34dc7c2f BB |
53 | #ifdef _KERNEL |
54 | #include <sys/vmsystm.h> | |
b128c09f | 55 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
56 | #endif |
57 | ||
03c6040b GW |
58 | /* |
59 | * Enable/disable nopwrite feature. | |
60 | */ | |
61 | int zfs_nopwrite_enabled = 1; | |
62 | ||
539d33c7 GM |
63 | /* |
64 | * Tunable to control percentage of dirtied blocks from frees in one TXG. | |
65 | * After this threshold is crossed, additional dirty blocks from frees | |
66 | * wait until the next TXG. | |
67 | * A value of zero will disable this throttle. | |
68 | */ | |
bef78122 | 69 | unsigned long zfs_per_txg_dirty_frees_percent = 30; |
539d33c7 | 70 | |
66aca247 DB |
71 | /* |
72 | * Enable/disable forcing txg sync when dirty in dmu_offset_next. | |
73 | */ | |
74 | int zfs_dmu_offset_next_sync = 0; | |
75 | ||
a1d477c2 MA |
76 | /* |
77 | * This can be used for testing, to ensure that certain actions happen | |
78 | * while in the middle of a remap (which might otherwise complete too | |
79 | * quickly). | |
80 | */ | |
81 | int zfs_object_remap_one_indirect_delay_ticks = 0; | |
82 | ||
34dc7c2f | 83 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
2e5dc449 MA |
84 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" }, |
85 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" }, | |
86 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" }, | |
87 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" }, | |
88 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" }, | |
89 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" }, | |
90 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" }, | |
91 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" }, | |
92 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" }, | |
93 | {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" }, | |
94 | {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" }, | |
95 | {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" }, | |
96 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" }, | |
97 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"}, | |
98 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" }, | |
99 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" }, | |
100 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" }, | |
101 | {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" }, | |
102 | {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" }, | |
103 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" }, | |
104 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" }, | |
105 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" }, | |
106 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" }, | |
107 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" }, | |
108 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" }, | |
109 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" }, | |
110 | {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" }, | |
111 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" }, | |
112 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" }, | |
113 | {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" }, | |
114 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" }, | |
115 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" }, | |
116 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" }, | |
117 | {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" }, | |
118 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" }, | |
119 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" }, | |
120 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" }, | |
121 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"}, | |
122 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" }, | |
123 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" }, | |
124 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"}, | |
125 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"}, | |
126 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" }, | |
127 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" }, | |
128 | {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" }, | |
129 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" }, | |
130 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" }, | |
131 | {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" }, | |
132 | {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" }, | |
133 | {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" }, | |
134 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" }, | |
135 | {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" }, | |
136 | {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" }, | |
137 | {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" } | |
9ae529ec CS |
138 | }; |
139 | ||
140 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
141 | { byteswap_uint8_array, "uint8" }, | |
142 | { byteswap_uint16_array, "uint16" }, | |
143 | { byteswap_uint32_array, "uint32" }, | |
144 | { byteswap_uint64_array, "uint64" }, | |
145 | { zap_byteswap, "zap" }, | |
146 | { dnode_buf_byteswap, "dnode" }, | |
147 | { dmu_objset_byteswap, "objset" }, | |
148 | { zfs_znode_byteswap, "znode" }, | |
149 | { zfs_oldacl_byteswap, "oldacl" }, | |
150 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
151 | }; |
152 | ||
2bce8049 MA |
153 | int |
154 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, | |
155 | void *tag, dmu_buf_t **dbp) | |
156 | { | |
157 | uint64_t blkid; | |
158 | dmu_buf_impl_t *db; | |
159 | ||
160 | blkid = dbuf_whichblock(dn, 0, offset); | |
161 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
162 | db = dbuf_hold(dn, blkid, tag); | |
163 | rw_exit(&dn->dn_struct_rwlock); | |
164 | ||
165 | if (db == NULL) { | |
166 | *dbp = NULL; | |
167 | return (SET_ERROR(EIO)); | |
168 | } | |
169 | ||
170 | *dbp = &db->db; | |
171 | return (0); | |
172 | } | |
34dc7c2f | 173 | int |
9b67f605 MA |
174 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
175 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
176 | { |
177 | dnode_t *dn; | |
178 | uint64_t blkid; | |
179 | dmu_buf_impl_t *db; | |
180 | int err; | |
428870ff BB |
181 | |
182 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
183 | if (err) |
184 | return (err); | |
fcff0f35 | 185 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
186 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
187 | db = dbuf_hold(dn, blkid, tag); | |
188 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
189 | dnode_rele(dn, FTAG); |
190 | ||
34dc7c2f | 191 | if (db == NULL) { |
9b67f605 MA |
192 | *dbp = NULL; |
193 | return (SET_ERROR(EIO)); | |
194 | } | |
195 | ||
196 | *dbp = &db->db; | |
197 | return (err); | |
198 | } | |
199 | ||
2bce8049 MA |
200 | int |
201 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
202 | void *tag, dmu_buf_t **dbp, int flags) | |
203 | { | |
204 | int err; | |
205 | int db_flags = DB_RF_CANFAIL; | |
206 | ||
207 | if (flags & DMU_READ_NO_PREFETCH) | |
208 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
209 | if (flags & DMU_READ_NO_DECRYPT) |
210 | db_flags |= DB_RF_NO_DECRYPT; | |
2bce8049 MA |
211 | |
212 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
213 | if (err == 0) { | |
214 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
215 | err = dbuf_read(db, NULL, db_flags); | |
216 | if (err != 0) { | |
217 | dbuf_rele(db, tag); | |
218 | *dbp = NULL; | |
219 | } | |
220 | } | |
221 | ||
222 | return (err); | |
223 | } | |
224 | ||
9b67f605 MA |
225 | int |
226 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
227 | void *tag, dmu_buf_t **dbp, int flags) | |
228 | { | |
229 | int err; | |
230 | int db_flags = DB_RF_CANFAIL; | |
231 | ||
232 | if (flags & DMU_READ_NO_PREFETCH) | |
233 | db_flags |= DB_RF_NOPREFETCH; | |
b5256303 TC |
234 | if (flags & DMU_READ_NO_DECRYPT) |
235 | db_flags |= DB_RF_NO_DECRYPT; | |
9b67f605 MA |
236 | |
237 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
238 | if (err == 0) { | |
239 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 240 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 241 | if (err != 0) { |
34dc7c2f | 242 | dbuf_rele(db, tag); |
9b67f605 | 243 | *dbp = NULL; |
34dc7c2f BB |
244 | } |
245 | } | |
246 | ||
34dc7c2f BB |
247 | return (err); |
248 | } | |
249 | ||
250 | int | |
251 | dmu_bonus_max(void) | |
252 | { | |
50c957f7 | 253 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
254 | } |
255 | ||
256 | int | |
572e2857 | 257 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 258 | { |
572e2857 BB |
259 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
260 | dnode_t *dn; | |
261 | int error; | |
34dc7c2f | 262 | |
572e2857 BB |
263 | DB_DNODE_ENTER(db); |
264 | dn = DB_DNODE(db); | |
265 | ||
266 | if (dn->dn_bonus != db) { | |
2e528b49 | 267 | error = SET_ERROR(EINVAL); |
572e2857 | 268 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 269 | error = SET_ERROR(EINVAL); |
572e2857 BB |
270 | } else { |
271 | dnode_setbonuslen(dn, newsize, tx); | |
272 | error = 0; | |
273 | } | |
274 | ||
275 | DB_DNODE_EXIT(db); | |
276 | return (error); | |
34dc7c2f BB |
277 | } |
278 | ||
428870ff | 279 | int |
572e2857 | 280 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 281 | { |
572e2857 BB |
282 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
283 | dnode_t *dn; | |
284 | int error; | |
428870ff | 285 | |
572e2857 BB |
286 | DB_DNODE_ENTER(db); |
287 | dn = DB_DNODE(db); | |
428870ff | 288 | |
9ae529ec | 289 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 290 | error = SET_ERROR(EINVAL); |
572e2857 | 291 | } else if (dn->dn_bonus != db) { |
2e528b49 | 292 | error = SET_ERROR(EINVAL); |
572e2857 BB |
293 | } else { |
294 | dnode_setbonus_type(dn, type, tx); | |
295 | error = 0; | |
296 | } | |
428870ff | 297 | |
572e2857 BB |
298 | DB_DNODE_EXIT(db); |
299 | return (error); | |
300 | } | |
301 | ||
302 | dmu_object_type_t | |
303 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
304 | { | |
305 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
306 | dnode_t *dn; | |
307 | dmu_object_type_t type; | |
308 | ||
309 | DB_DNODE_ENTER(db); | |
310 | dn = DB_DNODE(db); | |
311 | type = dn->dn_bonustype; | |
312 | DB_DNODE_EXIT(db); | |
313 | ||
314 | return (type); | |
428870ff BB |
315 | } |
316 | ||
317 | int | |
318 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
319 | { | |
320 | dnode_t *dn; | |
321 | int error; | |
322 | ||
323 | error = dnode_hold(os, object, FTAG, &dn); | |
324 | dbuf_rm_spill(dn, tx); | |
325 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
326 | dnode_rm_spill(dn, tx); | |
327 | rw_exit(&dn->dn_struct_rwlock); | |
328 | dnode_rele(dn, FTAG); | |
329 | return (error); | |
330 | } | |
331 | ||
34dc7c2f BB |
332 | /* |
333 | * returns ENOENT, EIO, or 0. | |
334 | */ | |
335 | int | |
b5256303 TC |
336 | dmu_bonus_hold_impl(objset_t *os, uint64_t object, void *tag, uint32_t flags, |
337 | dmu_buf_t **dbp) | |
34dc7c2f BB |
338 | { |
339 | dnode_t *dn; | |
340 | dmu_buf_impl_t *db; | |
341 | int error; | |
b5256303 TC |
342 | uint32_t db_flags = DB_RF_MUST_SUCCEED; |
343 | ||
344 | if (flags & DMU_READ_NO_PREFETCH) | |
345 | db_flags |= DB_RF_NOPREFETCH; | |
346 | if (flags & DMU_READ_NO_DECRYPT) | |
347 | db_flags |= DB_RF_NO_DECRYPT; | |
34dc7c2f | 348 | |
428870ff | 349 | error = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
350 | if (error) |
351 | return (error); | |
352 | ||
353 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
354 | if (dn->dn_bonus == NULL) { | |
355 | rw_exit(&dn->dn_struct_rwlock); | |
356 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
357 | if (dn->dn_bonus == NULL) | |
358 | dbuf_create_bonus(dn); | |
359 | } | |
360 | db = dn->dn_bonus; | |
34dc7c2f BB |
361 | |
362 | /* as long as the bonus buf is held, the dnode will be held */ | |
c13060e4 | 363 | if (zfs_refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 364 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 365 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
366 | } |
367 | ||
368 | /* | |
369 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
370 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
371 | * a dnode hold for every dbuf. | |
372 | */ | |
373 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
374 | |
375 | dnode_rele(dn, FTAG); | |
376 | ||
b5256303 TC |
377 | error = dbuf_read(db, NULL, db_flags); |
378 | if (error) { | |
379 | dnode_evict_bonus(dn); | |
380 | dbuf_rele(db, tag); | |
381 | *dbp = NULL; | |
382 | return (error); | |
383 | } | |
34dc7c2f BB |
384 | |
385 | *dbp = &db->db; | |
386 | return (0); | |
387 | } | |
388 | ||
b5256303 TC |
389 | int |
390 | dmu_bonus_hold(objset_t *os, uint64_t obj, void *tag, dmu_buf_t **dbp) | |
391 | { | |
392 | return (dmu_bonus_hold_impl(os, obj, tag, DMU_READ_NO_PREFETCH, dbp)); | |
393 | } | |
394 | ||
428870ff BB |
395 | /* |
396 | * returns ENOENT, EIO, or 0. | |
397 | * | |
398 | * This interface will allocate a blank spill dbuf when a spill blk | |
399 | * doesn't already exist on the dnode. | |
400 | * | |
401 | * if you only want to find an already existing spill db, then | |
402 | * dmu_spill_hold_existing() should be used. | |
403 | */ | |
404 | int | |
405 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
406 | { | |
407 | dmu_buf_impl_t *db = NULL; | |
408 | int err; | |
409 | ||
410 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
411 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
412 | ||
413 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
414 | ||
415 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
416 | rw_exit(&dn->dn_struct_rwlock); | |
417 | ||
b182ac00 | 418 | if (db == NULL) { |
419 | *dbp = NULL; | |
420 | return (SET_ERROR(EIO)); | |
421 | } | |
572e2857 BB |
422 | err = dbuf_read(db, NULL, flags); |
423 | if (err == 0) | |
424 | *dbp = &db->db; | |
b182ac00 | 425 | else { |
572e2857 | 426 | dbuf_rele(db, tag); |
b182ac00 | 427 | *dbp = NULL; |
428 | } | |
428870ff BB |
429 | return (err); |
430 | } | |
431 | ||
432 | int | |
433 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
434 | { | |
572e2857 BB |
435 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
436 | dnode_t *dn; | |
428870ff BB |
437 | int err; |
438 | ||
572e2857 BB |
439 | DB_DNODE_ENTER(db); |
440 | dn = DB_DNODE(db); | |
441 | ||
442 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 443 | err = SET_ERROR(EINVAL); |
572e2857 BB |
444 | } else { |
445 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
446 | ||
447 | if (!dn->dn_have_spill) { | |
2e528b49 | 448 | err = SET_ERROR(ENOENT); |
572e2857 BB |
449 | } else { |
450 | err = dmu_spill_hold_by_dnode(dn, | |
451 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
452 | } | |
428870ff | 453 | |
428870ff | 454 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 455 | } |
572e2857 BB |
456 | |
457 | DB_DNODE_EXIT(db); | |
428870ff BB |
458 | return (err); |
459 | } | |
460 | ||
461 | int | |
e7504d7a TC |
462 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag, |
463 | dmu_buf_t **dbp) | |
428870ff | 464 | { |
572e2857 BB |
465 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
466 | dnode_t *dn; | |
467 | int err; | |
e7504d7a TC |
468 | uint32_t db_flags = DB_RF_CANFAIL; |
469 | ||
470 | if (flags & DMU_READ_NO_DECRYPT) | |
471 | db_flags |= DB_RF_NO_DECRYPT; | |
572e2857 BB |
472 | |
473 | DB_DNODE_ENTER(db); | |
474 | dn = DB_DNODE(db); | |
e7504d7a | 475 | err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp); |
572e2857 BB |
476 | DB_DNODE_EXIT(db); |
477 | ||
478 | return (err); | |
428870ff BB |
479 | } |
480 | ||
34dc7c2f BB |
481 | /* |
482 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
483 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
484 | * and can induce severe lock contention when writing to several files | |
485 | * whose dnodes are in the same block. | |
486 | */ | |
487 | static int | |
9babb374 | 488 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
7f60329a | 489 | boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
34dc7c2f BB |
490 | { |
491 | dmu_buf_t **dbp; | |
492 | uint64_t blkid, nblks, i; | |
9babb374 | 493 | uint32_t dbuf_flags; |
34dc7c2f BB |
494 | int err; |
495 | zio_t *zio; | |
496 | ||
497 | ASSERT(length <= DMU_MAX_ACCESS); | |
498 | ||
7f60329a MA |
499 | /* |
500 | * Note: We directly notify the prefetch code of this read, so that | |
501 | * we can tell it about the multi-block read. dbuf_read() only knows | |
502 | * about the one block it is accessing. | |
503 | */ | |
504 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
505 | DB_RF_NOPREFETCH; | |
34dc7c2f BB |
506 | |
507 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
508 | if (dn->dn_datablkshift) { | |
509 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
510 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
511 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
512 | } else { |
513 | if (offset + length > dn->dn_datablksz) { | |
514 | zfs_panic_recover("zfs: accessing past end of object " | |
515 | "%llx/%llx (size=%u access=%llu+%llu)", | |
516 | (longlong_t)dn->dn_objset-> | |
517 | os_dsl_dataset->ds_object, | |
518 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
519 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 520 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 521 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
522 | } |
523 | nblks = 1; | |
524 | } | |
79c76d5b | 525 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 526 | |
b128c09f | 527 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
fcff0f35 | 528 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f | 529 | for (i = 0; i < nblks; i++) { |
7f60329a | 530 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f BB |
531 | if (db == NULL) { |
532 | rw_exit(&dn->dn_struct_rwlock); | |
533 | dmu_buf_rele_array(dbp, nblks, tag); | |
534 | zio_nowait(zio); | |
2e528b49 | 535 | return (SET_ERROR(EIO)); |
34dc7c2f | 536 | } |
7f60329a | 537 | |
34dc7c2f | 538 | /* initiate async i/o */ |
7f60329a | 539 | if (read) |
9babb374 | 540 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
541 | dbp[i] = &db->db; |
542 | } | |
7f60329a | 543 | |
755065f3 AM |
544 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
545 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
546 | dmu_zfetch(&dn->dn_zfetch, blkid, nblks, | |
547 | read && DNODE_IS_CACHEABLE(dn)); | |
7f60329a | 548 | } |
34dc7c2f BB |
549 | rw_exit(&dn->dn_struct_rwlock); |
550 | ||
551 | /* wait for async i/o */ | |
552 | err = zio_wait(zio); | |
553 | if (err) { | |
554 | dmu_buf_rele_array(dbp, nblks, tag); | |
555 | return (err); | |
556 | } | |
557 | ||
558 | /* wait for other io to complete */ | |
559 | if (read) { | |
560 | for (i = 0; i < nblks; i++) { | |
561 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
562 | mutex_enter(&db->db_mtx); | |
563 | while (db->db_state == DB_READ || | |
564 | db->db_state == DB_FILL) | |
565 | cv_wait(&db->db_changed, &db->db_mtx); | |
566 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 567 | err = SET_ERROR(EIO); |
34dc7c2f BB |
568 | mutex_exit(&db->db_mtx); |
569 | if (err) { | |
570 | dmu_buf_rele_array(dbp, nblks, tag); | |
571 | return (err); | |
572 | } | |
573 | } | |
574 | } | |
575 | ||
576 | *numbufsp = nblks; | |
577 | *dbpp = dbp; | |
578 | return (0); | |
579 | } | |
580 | ||
581 | static int | |
582 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
583 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
584 | { | |
585 | dnode_t *dn; | |
586 | int err; | |
587 | ||
428870ff | 588 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
589 | if (err) |
590 | return (err); | |
591 | ||
592 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 593 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
594 | |
595 | dnode_rele(dn, FTAG); | |
596 | ||
597 | return (err); | |
598 | } | |
599 | ||
600 | int | |
572e2857 | 601 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
7f60329a MA |
602 | uint64_t length, boolean_t read, void *tag, int *numbufsp, |
603 | dmu_buf_t ***dbpp) | |
34dc7c2f | 604 | { |
572e2857 BB |
605 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
606 | dnode_t *dn; | |
34dc7c2f BB |
607 | int err; |
608 | ||
572e2857 BB |
609 | DB_DNODE_ENTER(db); |
610 | dn = DB_DNODE(db); | |
34dc7c2f | 611 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 612 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 613 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
614 | |
615 | return (err); | |
616 | } | |
617 | ||
618 | void | |
619 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
620 | { | |
621 | int i; | |
622 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
623 | ||
624 | if (numbufs == 0) | |
625 | return; | |
626 | ||
627 | for (i = 0; i < numbufs; i++) { | |
628 | if (dbp[i]) | |
629 | dbuf_rele(dbp[i], tag); | |
630 | } | |
631 | ||
632 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
633 | } | |
634 | ||
e8b96c60 | 635 | /* |
fcff0f35 PD |
636 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
637 | * indirect blocks prefeteched will be those that point to the blocks containing | |
638 | * the data starting at offset, and continuing to offset + len. | |
e8b96c60 | 639 | * |
b5256303 TC |
640 | * Note that if the indirect blocks above the blocks being prefetched are not |
641 | * in cache, they will be asychronously read in. | |
e8b96c60 | 642 | */ |
34dc7c2f | 643 | void |
fcff0f35 PD |
644 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
645 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
646 | { |
647 | dnode_t *dn; | |
648 | uint64_t blkid; | |
e8b96c60 | 649 | int nblks, err; |
34dc7c2f | 650 | |
34dc7c2f | 651 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 652 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
653 | |
654 | if (object == 0 || object >= DN_MAX_OBJECT) | |
655 | return; | |
656 | ||
657 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
658 | blkid = dbuf_whichblock(dn, level, |
659 | object * sizeof (dnode_phys_t)); | |
660 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
661 | rw_exit(&dn->dn_struct_rwlock); |
662 | return; | |
663 | } | |
664 | ||
665 | /* | |
666 | * XXX - Note, if the dnode for the requested object is not | |
667 | * already cached, we will do a *synchronous* read in the | |
668 | * dnode_hold() call. The same is true for any indirects. | |
669 | */ | |
428870ff | 670 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
671 | if (err != 0) |
672 | return; | |
673 | ||
674 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
675 | /* |
676 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
677 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
678 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
679 | * offset) is the first. Then the number we need to prefetch is the | |
680 | * last - first + 1. | |
681 | */ | |
682 | if (level > 0 || dn->dn_datablkshift != 0) { | |
683 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
684 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
685 | } else { |
686 | nblks = (offset < dn->dn_datablksz); | |
687 | } | |
688 | ||
689 | if (nblks != 0) { | |
fcff0f35 | 690 | blkid = dbuf_whichblock(dn, level, offset); |
1c27024e | 691 | for (int i = 0; i < nblks; i++) |
fcff0f35 | 692 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f BB |
693 | } |
694 | ||
695 | rw_exit(&dn->dn_struct_rwlock); | |
696 | ||
697 | dnode_rele(dn, FTAG); | |
698 | } | |
699 | ||
45d1cae3 BB |
700 | /* |
701 | * Get the next "chunk" of file data to free. We traverse the file from | |
702 | * the end so that the file gets shorter over time (if we crashes in the | |
703 | * middle, this will leave us in a better state). We find allocated file | |
704 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
705 | * |
706 | * On input, *start should be the first offset that does not need to be | |
707 | * freed (e.g. "offset + length"). On return, *start will be the first | |
708 | * offset that should be freed. | |
45d1cae3 | 709 | */ |
b128c09f | 710 | static int |
b663a23d | 711 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) |
b128c09f | 712 | { |
b663a23d MA |
713 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
714 | /* bytes of data covered by a level-1 indirect block */ | |
45d1cae3 | 715 | uint64_t iblkrange = |
b128c09f BB |
716 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
717 | ||
b663a23d | 718 | ASSERT3U(minimum, <=, *start); |
b128c09f | 719 | |
b663a23d MA |
720 | if (*start - minimum <= iblkrange * maxblks) { |
721 | *start = minimum; | |
b128c09f BB |
722 | return (0); |
723 | } | |
45d1cae3 | 724 | ASSERT(ISP2(iblkrange)); |
b128c09f | 725 | |
1c27024e | 726 | for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
727 | int err; |
728 | ||
b663a23d MA |
729 | /* |
730 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
731 | * indirect block at or before the input offset. We must | |
732 | * decrement *start so that it is at the end of the region | |
733 | * to search. | |
734 | */ | |
735 | (*start)--; | |
b128c09f | 736 | err = dnode_next_offset(dn, |
45d1cae3 | 737 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 738 | |
b663a23d | 739 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 740 | if (err == ESRCH) { |
b663a23d MA |
741 | *start = minimum; |
742 | break; | |
743 | } else if (err != 0) { | |
b128c09f | 744 | return (err); |
45d1cae3 | 745 | } |
b128c09f | 746 | |
b663a23d | 747 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 748 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 749 | } |
b663a23d MA |
750 | if (*start < minimum) |
751 | *start = minimum; | |
b128c09f BB |
752 | return (0); |
753 | } | |
754 | ||
a08abc1b GM |
755 | /* |
756 | * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, | |
757 | * otherwise return false. | |
758 | * Used below in dmu_free_long_range_impl() to enable abort when unmounting | |
759 | */ | |
760 | /*ARGSUSED*/ | |
761 | static boolean_t | |
762 | dmu_objset_zfs_unmounting(objset_t *os) | |
763 | { | |
764 | #ifdef _KERNEL | |
765 | if (dmu_objset_type(os) == DMU_OST_ZFS) | |
766 | return (zfs_get_vfs_flag_unmounted(os)); | |
767 | #endif | |
768 | return (B_FALSE); | |
769 | } | |
770 | ||
b128c09f BB |
771 | static int |
772 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
0c03d21a | 773 | uint64_t length) |
b128c09f | 774 | { |
c97d3069 | 775 | uint64_t object_size; |
b663a23d | 776 | int err; |
539d33c7 GM |
777 | uint64_t dirty_frees_threshold; |
778 | dsl_pool_t *dp = dmu_objset_pool(os); | |
b663a23d | 779 | |
c97d3069 BB |
780 | if (dn == NULL) |
781 | return (SET_ERROR(EINVAL)); | |
782 | ||
783 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 784 | if (offset >= object_size) |
b128c09f | 785 | return (0); |
b128c09f | 786 | |
539d33c7 GM |
787 | if (zfs_per_txg_dirty_frees_percent <= 100) |
788 | dirty_frees_threshold = | |
789 | zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; | |
790 | else | |
791 | dirty_frees_threshold = zfs_dirty_data_max / 4; | |
792 | ||
b663a23d MA |
793 | if (length == DMU_OBJECT_END || offset + length > object_size) |
794 | length = object_size - offset; | |
795 | ||
796 | while (length != 0) { | |
539d33c7 GM |
797 | uint64_t chunk_end, chunk_begin, chunk_len; |
798 | uint64_t long_free_dirty_all_txgs = 0; | |
b663a23d MA |
799 | dmu_tx_t *tx; |
800 | ||
a08abc1b GM |
801 | if (dmu_objset_zfs_unmounting(dn->dn_objset)) |
802 | return (SET_ERROR(EINTR)); | |
803 | ||
b663a23d MA |
804 | chunk_end = chunk_begin = offset + length; |
805 | ||
806 | /* move chunk_begin backwards to the beginning of this chunk */ | |
807 | err = get_next_chunk(dn, &chunk_begin, offset); | |
b128c09f BB |
808 | if (err) |
809 | return (err); | |
b663a23d MA |
810 | ASSERT3U(chunk_begin, >=, offset); |
811 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f | 812 | |
539d33c7 GM |
813 | chunk_len = chunk_end - chunk_begin; |
814 | ||
815 | mutex_enter(&dp->dp_lock); | |
1c27024e | 816 | for (int t = 0; t < TXG_SIZE; t++) { |
539d33c7 GM |
817 | long_free_dirty_all_txgs += |
818 | dp->dp_long_free_dirty_pertxg[t]; | |
819 | } | |
820 | mutex_exit(&dp->dp_lock); | |
821 | ||
822 | /* | |
823 | * To avoid filling up a TXG with just frees wait for | |
824 | * the next TXG to open before freeing more chunks if | |
825 | * we have reached the threshold of frees | |
826 | */ | |
827 | if (dirty_frees_threshold != 0 && | |
828 | long_free_dirty_all_txgs >= dirty_frees_threshold) { | |
750e1f88 | 829 | DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay); |
539d33c7 GM |
830 | txg_wait_open(dp, 0); |
831 | continue; | |
832 | } | |
833 | ||
b128c09f | 834 | tx = dmu_tx_create(os); |
539d33c7 | 835 | dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); |
19d55079 MA |
836 | |
837 | /* | |
838 | * Mark this transaction as typically resulting in a net | |
839 | * reduction in space used. | |
840 | */ | |
841 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
842 | err = dmu_tx_assign(tx, TXG_WAIT); |
843 | if (err) { | |
844 | dmu_tx_abort(tx); | |
845 | return (err); | |
846 | } | |
539d33c7 GM |
847 | |
848 | mutex_enter(&dp->dp_lock); | |
849 | dp->dp_long_free_dirty_pertxg[dmu_tx_get_txg(tx) & TXG_MASK] += | |
850 | chunk_len; | |
851 | mutex_exit(&dp->dp_lock); | |
852 | DTRACE_PROBE3(free__long__range, | |
853 | uint64_t, long_free_dirty_all_txgs, uint64_t, chunk_len, | |
854 | uint64_t, dmu_tx_get_txg(tx)); | |
855 | dnode_free_range(dn, chunk_begin, chunk_len, tx); | |
440a3eb9 | 856 | |
b128c09f | 857 | dmu_tx_commit(tx); |
b663a23d | 858 | |
539d33c7 | 859 | length -= chunk_len; |
b128c09f BB |
860 | } |
861 | return (0); | |
862 | } | |
863 | ||
864 | int | |
865 | dmu_free_long_range(objset_t *os, uint64_t object, | |
866 | uint64_t offset, uint64_t length) | |
867 | { | |
868 | dnode_t *dn; | |
869 | int err; | |
870 | ||
428870ff | 871 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
872 | if (err != 0) |
873 | return (err); | |
0c03d21a | 874 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
875 | |
876 | /* | |
877 | * It is important to zero out the maxblkid when freeing the entire | |
878 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
879 | * will take the fast path, and (b) dnode_reallocate() can verify | |
880 | * that the entire file has been freed. | |
881 | */ | |
b0bc7a84 | 882 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
883 | dn->dn_maxblkid = 0; |
884 | ||
b128c09f BB |
885 | dnode_rele(dn, FTAG); |
886 | return (err); | |
887 | } | |
888 | ||
889 | int | |
0c03d21a | 890 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 891 | { |
b128c09f BB |
892 | dmu_tx_t *tx; |
893 | int err; | |
894 | ||
b663a23d | 895 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
896 | if (err != 0) |
897 | return (err); | |
b663a23d MA |
898 | |
899 | tx = dmu_tx_create(os); | |
900 | dmu_tx_hold_bonus(tx, object); | |
901 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 902 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
903 | err = dmu_tx_assign(tx, TXG_WAIT); |
904 | if (err == 0) { | |
35df0bb5 TC |
905 | if (err == 0) |
906 | err = dmu_object_free(os, object, tx); | |
440a3eb9 | 907 | |
b663a23d | 908 | dmu_tx_commit(tx); |
b128c09f | 909 | } else { |
b663a23d | 910 | dmu_tx_abort(tx); |
b128c09f | 911 | } |
b663a23d | 912 | |
b128c09f BB |
913 | return (err); |
914 | } | |
915 | ||
34dc7c2f BB |
916 | int |
917 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
918 | uint64_t size, dmu_tx_t *tx) | |
919 | { | |
920 | dnode_t *dn; | |
428870ff | 921 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
922 | if (err) |
923 | return (err); | |
924 | ASSERT(offset < UINT64_MAX); | |
ee45fbd8 | 925 | ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset); |
34dc7c2f BB |
926 | dnode_free_range(dn, offset, size, tx); |
927 | dnode_rele(dn, FTAG); | |
928 | return (0); | |
929 | } | |
930 | ||
0eef1bde | 931 | static int |
932 | dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, | |
9babb374 | 933 | void *buf, uint32_t flags) |
34dc7c2f | 934 | { |
34dc7c2f | 935 | dmu_buf_t **dbp; |
0eef1bde | 936 | int numbufs, err = 0; |
34dc7c2f BB |
937 | |
938 | /* | |
939 | * Deal with odd block sizes, where there can't be data past the first | |
940 | * block. If we ever do the tail block optimization, we will need to | |
941 | * handle that here as well. | |
942 | */ | |
45d1cae3 | 943 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 944 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
945 | MIN(size, dn->dn_datablksz - offset); |
946 | bzero((char *)buf + newsz, size - newsz); | |
947 | size = newsz; | |
948 | } | |
949 | ||
950 | while (size > 0) { | |
951 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 952 | int i; |
34dc7c2f BB |
953 | |
954 | /* | |
955 | * NB: we could do this block-at-a-time, but it's nice | |
956 | * to be reading in parallel. | |
957 | */ | |
958 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 959 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
960 | if (err) |
961 | break; | |
962 | ||
963 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
964 | uint64_t tocpy; |
965 | int64_t bufoff; | |
34dc7c2f BB |
966 | dmu_buf_t *db = dbp[i]; |
967 | ||
968 | ASSERT(size > 0); | |
969 | ||
970 | bufoff = offset - db->db_offset; | |
c9520ecc | 971 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 972 | |
c9520ecc | 973 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
974 | |
975 | offset += tocpy; | |
976 | size -= tocpy; | |
977 | buf = (char *)buf + tocpy; | |
978 | } | |
979 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
980 | } | |
34dc7c2f BB |
981 | return (err); |
982 | } | |
983 | ||
0eef1bde | 984 | int |
985 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
986 | void *buf, uint32_t flags) | |
34dc7c2f | 987 | { |
0eef1bde | 988 | dnode_t *dn; |
989 | int err; | |
34dc7c2f | 990 | |
0eef1bde | 991 | err = dnode_hold(os, object, FTAG, &dn); |
992 | if (err != 0) | |
993 | return (err); | |
34dc7c2f | 994 | |
0eef1bde | 995 | err = dmu_read_impl(dn, offset, size, buf, flags); |
996 | dnode_rele(dn, FTAG); | |
997 | return (err); | |
998 | } | |
999 | ||
1000 | int | |
1001 | dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, | |
1002 | uint32_t flags) | |
1003 | { | |
1004 | return (dmu_read_impl(dn, offset, size, buf, flags)); | |
1005 | } | |
1006 | ||
1007 | static void | |
1008 | dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, | |
1009 | const void *buf, dmu_tx_t *tx) | |
1010 | { | |
1011 | int i; | |
34dc7c2f BB |
1012 | |
1013 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1014 | uint64_t tocpy; |
1015 | int64_t bufoff; | |
34dc7c2f BB |
1016 | dmu_buf_t *db = dbp[i]; |
1017 | ||
1018 | ASSERT(size > 0); | |
1019 | ||
1020 | bufoff = offset - db->db_offset; | |
c9520ecc | 1021 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
1022 | |
1023 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1024 | ||
1025 | if (tocpy == db->db_size) | |
1026 | dmu_buf_will_fill(db, tx); | |
1027 | else | |
1028 | dmu_buf_will_dirty(db, tx); | |
1029 | ||
60101509 | 1030 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
1031 | |
1032 | if (tocpy == db->db_size) | |
1033 | dmu_buf_fill_done(db, tx); | |
1034 | ||
1035 | offset += tocpy; | |
1036 | size -= tocpy; | |
1037 | buf = (char *)buf + tocpy; | |
1038 | } | |
0eef1bde | 1039 | } |
1040 | ||
1041 | void | |
1042 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1043 | const void *buf, dmu_tx_t *tx) | |
1044 | { | |
1045 | dmu_buf_t **dbp; | |
1046 | int numbufs; | |
1047 | ||
1048 | if (size == 0) | |
1049 | return; | |
1050 | ||
1051 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
1052 | FALSE, FTAG, &numbufs, &dbp)); | |
1053 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
1054 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1055 | } | |
1056 | ||
1057 | void | |
1058 | dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, | |
1059 | const void *buf, dmu_tx_t *tx) | |
1060 | { | |
1061 | dmu_buf_t **dbp; | |
1062 | int numbufs; | |
1063 | ||
1064 | if (size == 0) | |
1065 | return; | |
1066 | ||
1067 | VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, | |
1068 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); | |
1069 | dmu_write_impl(dbp, numbufs, offset, size, buf, tx); | |
34dc7c2f BB |
1070 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1071 | } | |
1072 | ||
a1d477c2 MA |
1073 | static int |
1074 | dmu_object_remap_one_indirect(objset_t *os, dnode_t *dn, | |
1075 | uint64_t last_removal_txg, uint64_t offset) | |
1076 | { | |
1077 | uint64_t l1blkid = dbuf_whichblock(dn, 1, offset); | |
1078 | int err = 0; | |
1079 | ||
1080 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
1081 | dmu_buf_impl_t *dbuf = dbuf_hold_level(dn, 1, l1blkid, FTAG); | |
1082 | ASSERT3P(dbuf, !=, NULL); | |
1083 | ||
1084 | /* | |
1085 | * If the block hasn't been written yet, this default will ensure | |
1086 | * we don't try to remap it. | |
1087 | */ | |
1088 | uint64_t birth = UINT64_MAX; | |
1089 | ASSERT3U(last_removal_txg, !=, UINT64_MAX); | |
1090 | if (dbuf->db_blkptr != NULL) | |
1091 | birth = dbuf->db_blkptr->blk_birth; | |
1092 | rw_exit(&dn->dn_struct_rwlock); | |
1093 | ||
1094 | /* | |
1095 | * If this L1 was already written after the last removal, then we've | |
1096 | * already tried to remap it. | |
1097 | */ | |
1098 | if (birth <= last_removal_txg && | |
1099 | dbuf_read(dbuf, NULL, DB_RF_MUST_SUCCEED) == 0 && | |
1100 | dbuf_can_remap(dbuf)) { | |
1101 | dmu_tx_t *tx = dmu_tx_create(os); | |
1102 | dmu_tx_hold_remap_l1indirect(tx, dn->dn_object); | |
1103 | err = dmu_tx_assign(tx, TXG_WAIT); | |
1104 | if (err == 0) { | |
1105 | (void) dbuf_dirty(dbuf, tx); | |
1106 | dmu_tx_commit(tx); | |
1107 | } else { | |
1108 | dmu_tx_abort(tx); | |
1109 | } | |
1110 | } | |
1111 | ||
1112 | dbuf_rele(dbuf, FTAG); | |
1113 | ||
1114 | delay(zfs_object_remap_one_indirect_delay_ticks); | |
1115 | ||
1116 | return (err); | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * Remap all blockpointers in the object, if possible, so that they reference | |
1121 | * only concrete vdevs. | |
1122 | * | |
1123 | * To do this, iterate over the L0 blockpointers and remap any that reference | |
1124 | * an indirect vdev. Note that we only examine L0 blockpointers; since we | |
1125 | * cannot guarantee that we can remap all blockpointer anyways (due to split | |
1126 | * blocks), we do not want to make the code unnecessarily complicated to | |
1127 | * catch the unlikely case that there is an L1 block on an indirect vdev that | |
1128 | * contains no indirect blockpointers. | |
1129 | */ | |
1130 | int | |
1131 | dmu_object_remap_indirects(objset_t *os, uint64_t object, | |
1132 | uint64_t last_removal_txg) | |
1133 | { | |
1134 | uint64_t offset, l1span; | |
1135 | int err; | |
1136 | dnode_t *dn; | |
1137 | ||
1138 | err = dnode_hold(os, object, FTAG, &dn); | |
1139 | if (err != 0) { | |
1140 | return (err); | |
1141 | } | |
1142 | ||
1143 | if (dn->dn_nlevels <= 1) { | |
1144 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
1145 | err = SET_ERROR(EINTR); | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * If the dnode has no indirect blocks, we cannot dirty them. | |
1150 | * We still want to remap the blkptr(s) in the dnode if | |
1151 | * appropriate, so mark it as dirty. | |
1152 | */ | |
1153 | if (err == 0 && dnode_needs_remap(dn)) { | |
1154 | dmu_tx_t *tx = dmu_tx_create(os); | |
1155 | dmu_tx_hold_bonus(tx, dn->dn_object); | |
1156 | if ((err = dmu_tx_assign(tx, TXG_WAIT)) == 0) { | |
1157 | dnode_setdirty(dn, tx); | |
1158 | dmu_tx_commit(tx); | |
1159 | } else { | |
1160 | dmu_tx_abort(tx); | |
1161 | } | |
1162 | } | |
1163 | ||
1164 | dnode_rele(dn, FTAG); | |
1165 | return (err); | |
1166 | } | |
1167 | ||
1168 | offset = 0; | |
1169 | l1span = 1ULL << (dn->dn_indblkshift - SPA_BLKPTRSHIFT + | |
1170 | dn->dn_datablkshift); | |
1171 | /* | |
1172 | * Find the next L1 indirect that is not a hole. | |
1173 | */ | |
1174 | while (dnode_next_offset(dn, 0, &offset, 2, 1, 0) == 0) { | |
1175 | if (issig(JUSTLOOKING) && issig(FORREAL)) { | |
1176 | err = SET_ERROR(EINTR); | |
1177 | break; | |
1178 | } | |
1179 | if ((err = dmu_object_remap_one_indirect(os, dn, | |
1180 | last_removal_txg, offset)) != 0) { | |
1181 | break; | |
1182 | } | |
1183 | offset += l1span; | |
1184 | } | |
1185 | ||
1186 | dnode_rele(dn, FTAG); | |
1187 | return (err); | |
1188 | } | |
1189 | ||
b128c09f BB |
1190 | void |
1191 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
1192 | dmu_tx_t *tx) | |
1193 | { | |
1194 | dmu_buf_t **dbp; | |
1195 | int numbufs, i; | |
1196 | ||
1197 | if (size == 0) | |
1198 | return; | |
1199 | ||
1200 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
1201 | FALSE, FTAG, &numbufs, &dbp)); | |
1202 | ||
1203 | for (i = 0; i < numbufs; i++) { | |
1204 | dmu_buf_t *db = dbp[i]; | |
1205 | ||
1206 | dmu_buf_will_not_fill(db, tx); | |
1207 | } | |
1208 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1209 | } | |
1210 | ||
9b67f605 MA |
1211 | void |
1212 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
1213 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
1214 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
1215 | { | |
1216 | dmu_buf_t *db; | |
1217 | ||
1218 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
1219 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
1220 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
1221 | FTAG, &db)); | |
1222 | ||
1223 | dmu_buf_write_embedded(db, | |
1224 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
1225 | uncompressed_size, compressed_size, byteorder, tx); | |
1226 | ||
1227 | dmu_buf_rele(db, FTAG); | |
1228 | } | |
1229 | ||
428870ff BB |
1230 | /* |
1231 | * DMU support for xuio | |
1232 | */ | |
1233 | kstat_t *xuio_ksp = NULL; | |
1234 | ||
59e6e7ca BB |
1235 | typedef struct xuio_stats { |
1236 | /* loaned yet not returned arc_buf */ | |
1237 | kstat_named_t xuiostat_onloan_rbuf; | |
1238 | kstat_named_t xuiostat_onloan_wbuf; | |
1239 | /* whether a copy is made when loaning out a read buffer */ | |
1240 | kstat_named_t xuiostat_rbuf_copied; | |
1241 | kstat_named_t xuiostat_rbuf_nocopy; | |
1242 | /* whether a copy is made when assigning a write buffer */ | |
1243 | kstat_named_t xuiostat_wbuf_copied; | |
1244 | kstat_named_t xuiostat_wbuf_nocopy; | |
1245 | } xuio_stats_t; | |
1246 | ||
1247 | static xuio_stats_t xuio_stats = { | |
1248 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
1249 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
1250 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
1251 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
1252 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
1253 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
1254 | }; | |
1255 | ||
d1d7e268 MK |
1256 | #define XUIOSTAT_INCR(stat, val) \ |
1257 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
1258 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
59e6e7ca | 1259 | |
5a6765cf | 1260 | #ifdef HAVE_UIO_ZEROCOPY |
428870ff BB |
1261 | int |
1262 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
1263 | { | |
1264 | dmu_xuio_t *priv; | |
1265 | uio_t *uio = &xuio->xu_uio; | |
1266 | ||
1267 | uio->uio_iovcnt = nblk; | |
79c76d5b | 1268 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
428870ff | 1269 | |
79c76d5b | 1270 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
428870ff | 1271 | priv->cnt = nblk; |
79c76d5b | 1272 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
5475aada | 1273 | priv->iovp = (iovec_t *)uio->uio_iov; |
428870ff BB |
1274 | XUIO_XUZC_PRIV(xuio) = priv; |
1275 | ||
1276 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1277 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
1278 | else | |
1279 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
1280 | ||
1281 | return (0); | |
1282 | } | |
1283 | ||
1284 | void | |
1285 | dmu_xuio_fini(xuio_t *xuio) | |
1286 | { | |
1287 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1288 | int nblk = priv->cnt; | |
1289 | ||
1290 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
1291 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
1292 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
1293 | ||
1294 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1295 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
1296 | else | |
1297 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
1302 | * and increase priv->next by 1. | |
1303 | */ | |
1304 | int | |
1305 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
1306 | { | |
1307 | struct iovec *iov; | |
1308 | uio_t *uio = &xuio->xu_uio; | |
1309 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1310 | int i = priv->next++; | |
1311 | ||
1312 | ASSERT(i < priv->cnt); | |
2aa34383 | 1313 | ASSERT(off + n <= arc_buf_lsize(abuf)); |
5475aada | 1314 | iov = (iovec_t *)uio->uio_iov + i; |
428870ff BB |
1315 | iov->iov_base = (char *)abuf->b_data + off; |
1316 | iov->iov_len = n; | |
1317 | priv->bufs[i] = abuf; | |
1318 | return (0); | |
1319 | } | |
1320 | ||
1321 | int | |
1322 | dmu_xuio_cnt(xuio_t *xuio) | |
1323 | { | |
1324 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1325 | return (priv->cnt); | |
1326 | } | |
1327 | ||
1328 | arc_buf_t * | |
1329 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
1330 | { | |
1331 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1332 | ||
1333 | ASSERT(i < priv->cnt); | |
1334 | return (priv->bufs[i]); | |
1335 | } | |
1336 | ||
1337 | void | |
1338 | dmu_xuio_clear(xuio_t *xuio, int i) | |
1339 | { | |
1340 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1341 | ||
1342 | ASSERT(i < priv->cnt); | |
1343 | priv->bufs[i] = NULL; | |
1344 | } | |
5a6765cf | 1345 | #endif /* HAVE_UIO_ZEROCOPY */ |
428870ff BB |
1346 | |
1347 | static void | |
1348 | xuio_stat_init(void) | |
1349 | { | |
1350 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1351 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1352 | KSTAT_FLAG_VIRTUAL); | |
1353 | if (xuio_ksp != NULL) { | |
1354 | xuio_ksp->ks_data = &xuio_stats; | |
1355 | kstat_install(xuio_ksp); | |
1356 | } | |
1357 | } | |
1358 | ||
1359 | static void | |
1360 | xuio_stat_fini(void) | |
1361 | { | |
1362 | if (xuio_ksp != NULL) { | |
1363 | kstat_delete(xuio_ksp); | |
1364 | xuio_ksp = NULL; | |
1365 | } | |
1366 | } | |
1367 | ||
1368 | void | |
5043684a | 1369 | xuio_stat_wbuf_copied(void) |
428870ff BB |
1370 | { |
1371 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1372 | } | |
1373 | ||
1374 | void | |
5043684a | 1375 | xuio_stat_wbuf_nocopy(void) |
428870ff BB |
1376 | { |
1377 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1378 | } | |
1379 | ||
34dc7c2f | 1380 | #ifdef _KERNEL |
5228cf01 | 1381 | int |
804e0504 | 1382 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) |
872e8d26 BB |
1383 | { |
1384 | dmu_buf_t **dbp; | |
1385 | int numbufs, i, err; | |
5a6765cf | 1386 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 | 1387 | xuio_t *xuio = NULL; |
5a6765cf | 1388 | #endif |
872e8d26 BB |
1389 | |
1390 | /* | |
1391 | * NB: we could do this block-at-a-time, but it's nice | |
1392 | * to be reading in parallel. | |
1393 | */ | |
804e0504 MA |
1394 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, |
1395 | TRUE, FTAG, &numbufs, &dbp, 0); | |
872e8d26 BB |
1396 | if (err) |
1397 | return (err); | |
1398 | ||
1399 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1400 | uint64_t tocpy; |
1401 | int64_t bufoff; | |
872e8d26 BB |
1402 | dmu_buf_t *db = dbp[i]; |
1403 | ||
1404 | ASSERT(size > 0); | |
1405 | ||
1406 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1407 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 | 1408 | |
5a6765cf | 1409 | #ifdef HAVE_UIO_ZEROCOPY |
872e8d26 BB |
1410 | if (xuio) { |
1411 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1412 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1413 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1414 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1415 | if (!err) { | |
1416 | uio->uio_resid -= tocpy; | |
1417 | uio->uio_loffset += tocpy; | |
1418 | } | |
1419 | ||
1420 | if (abuf == dbuf_abuf) | |
1421 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1422 | else | |
1423 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
5a6765cf | 1424 | } else |
1425 | #endif | |
872e8d26 BB |
1426 | err = uiomove((char *)db->db_data + bufoff, tocpy, |
1427 | UIO_READ, uio); | |
872e8d26 BB |
1428 | if (err) |
1429 | break; | |
1430 | ||
1431 | size -= tocpy; | |
1432 | } | |
1433 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1434 | ||
1435 | return (err); | |
1436 | } | |
1437 | ||
804e0504 MA |
1438 | /* |
1439 | * Read 'size' bytes into the uio buffer. | |
1440 | * From object zdb->db_object. | |
1441 | * Starting at offset uio->uio_loffset. | |
1442 | * | |
1443 | * If the caller already has a dbuf in the target object | |
1444 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1445 | * because we don't have to find the dnode_t for the object. | |
1446 | */ | |
1447 | int | |
1448 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1449 | { | |
1450 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1451 | dnode_t *dn; | |
1452 | int err; | |
1453 | ||
1454 | if (size == 0) | |
1455 | return (0); | |
1456 | ||
1457 | DB_DNODE_ENTER(db); | |
1458 | dn = DB_DNODE(db); | |
1459 | err = dmu_read_uio_dnode(dn, uio, size); | |
1460 | DB_DNODE_EXIT(db); | |
1461 | ||
1462 | return (err); | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Read 'size' bytes into the uio buffer. | |
1467 | * From the specified object | |
1468 | * Starting at offset uio->uio_loffset. | |
1469 | */ | |
1470 | int | |
1471 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1472 | { | |
1473 | dnode_t *dn; | |
1474 | int err; | |
1475 | ||
1476 | if (size == 0) | |
1477 | return (0); | |
1478 | ||
1479 | err = dnode_hold(os, object, FTAG, &dn); | |
1480 | if (err) | |
1481 | return (err); | |
1482 | ||
1483 | err = dmu_read_uio_dnode(dn, uio, size); | |
1484 | ||
1485 | dnode_rele(dn, FTAG); | |
1486 | ||
1487 | return (err); | |
1488 | } | |
1489 | ||
5228cf01 | 1490 | int |
872e8d26 BB |
1491 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) |
1492 | { | |
1493 | dmu_buf_t **dbp; | |
1494 | int numbufs; | |
1495 | int err = 0; | |
1496 | int i; | |
1497 | ||
1498 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1499 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1500 | if (err) | |
1501 | return (err); | |
1502 | ||
1503 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1504 | uint64_t tocpy; |
1505 | int64_t bufoff; | |
872e8d26 BB |
1506 | dmu_buf_t *db = dbp[i]; |
1507 | ||
1508 | ASSERT(size > 0); | |
1509 | ||
1510 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1511 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1512 | |
1513 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1514 | ||
1515 | if (tocpy == db->db_size) | |
1516 | dmu_buf_will_fill(db, tx); | |
1517 | else | |
1518 | dmu_buf_will_dirty(db, tx); | |
1519 | ||
1520 | /* | |
1521 | * XXX uiomove could block forever (eg.nfs-backed | |
1522 | * pages). There needs to be a uiolockdown() function | |
1523 | * to lock the pages in memory, so that uiomove won't | |
1524 | * block. | |
1525 | */ | |
1526 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1527 | UIO_WRITE, uio); | |
1528 | ||
1529 | if (tocpy == db->db_size) | |
1530 | dmu_buf_fill_done(db, tx); | |
1531 | ||
1532 | if (err) | |
1533 | break; | |
1534 | ||
1535 | size -= tocpy; | |
1536 | } | |
1537 | ||
1538 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1539 | return (err); | |
1540 | } | |
1541 | ||
804e0504 MA |
1542 | /* |
1543 | * Write 'size' bytes from the uio buffer. | |
1544 | * To object zdb->db_object. | |
1545 | * Starting at offset uio->uio_loffset. | |
1546 | * | |
1547 | * If the caller already has a dbuf in the target object | |
1548 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1549 | * because we don't have to find the dnode_t for the object. | |
1550 | */ | |
428870ff BB |
1551 | int |
1552 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1553 | dmu_tx_t *tx) | |
1554 | { | |
572e2857 BB |
1555 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1556 | dnode_t *dn; | |
1557 | int err; | |
1558 | ||
428870ff BB |
1559 | if (size == 0) |
1560 | return (0); | |
1561 | ||
572e2857 BB |
1562 | DB_DNODE_ENTER(db); |
1563 | dn = DB_DNODE(db); | |
1564 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1565 | DB_DNODE_EXIT(db); | |
1566 | ||
1567 | return (err); | |
428870ff BB |
1568 | } |
1569 | ||
804e0504 MA |
1570 | /* |
1571 | * Write 'size' bytes from the uio buffer. | |
1572 | * To the specified object. | |
1573 | * Starting at offset uio->uio_loffset. | |
1574 | */ | |
428870ff BB |
1575 | int |
1576 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1577 | dmu_tx_t *tx) | |
1578 | { | |
1579 | dnode_t *dn; | |
1580 | int err; | |
1581 | ||
1582 | if (size == 0) | |
1583 | return (0); | |
1584 | ||
1585 | err = dnode_hold(os, object, FTAG, &dn); | |
1586 | if (err) | |
1587 | return (err); | |
1588 | ||
1589 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1590 | ||
1591 | dnode_rele(dn, FTAG); | |
1592 | ||
1593 | return (err); | |
1594 | } | |
872e8d26 | 1595 | #endif /* _KERNEL */ |
34dc7c2f | 1596 | |
9babb374 BB |
1597 | /* |
1598 | * Allocate a loaned anonymous arc buffer. | |
1599 | */ | |
1600 | arc_buf_t * | |
1601 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1602 | { | |
572e2857 | 1603 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1604 | |
2aa34383 | 1605 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1606 | } |
1607 | ||
1608 | /* | |
1609 | * Free a loaned arc buffer. | |
1610 | */ | |
1611 | void | |
1612 | dmu_return_arcbuf(arc_buf_t *buf) | |
1613 | { | |
1614 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1615 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1616 | } |
1617 | ||
b5256303 TC |
1618 | void |
1619 | dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset, | |
1620 | dmu_buf_t *handle, dmu_tx_t *tx) | |
1621 | { | |
1622 | dmu_buf_t *dst_handle; | |
1623 | dmu_buf_impl_t *dstdb; | |
1624 | dmu_buf_impl_t *srcdb = (dmu_buf_impl_t *)handle; | |
149ce888 | 1625 | dmu_object_type_t type; |
b5256303 TC |
1626 | arc_buf_t *abuf; |
1627 | uint64_t datalen; | |
1628 | boolean_t byteorder; | |
1629 | uint8_t salt[ZIO_DATA_SALT_LEN]; | |
1630 | uint8_t iv[ZIO_DATA_IV_LEN]; | |
1631 | uint8_t mac[ZIO_DATA_MAC_LEN]; | |
1632 | ||
1633 | ASSERT3P(srcdb->db_buf, !=, NULL); | |
1634 | ||
1635 | /* hold the db that we want to write to */ | |
1636 | VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &dst_handle, | |
1637 | DMU_READ_NO_DECRYPT)); | |
1638 | dstdb = (dmu_buf_impl_t *)dst_handle; | |
1639 | datalen = arc_buf_size(srcdb->db_buf); | |
1640 | ||
149ce888 TC |
1641 | DB_DNODE_ENTER(dstdb); |
1642 | type = DB_DNODE(dstdb)->dn_type; | |
1643 | DB_DNODE_EXIT(dstdb); | |
1644 | ||
b5256303 TC |
1645 | /* allocated an arc buffer that matches the type of srcdb->db_buf */ |
1646 | if (arc_is_encrypted(srcdb->db_buf)) { | |
1647 | arc_get_raw_params(srcdb->db_buf, &byteorder, salt, iv, mac); | |
1648 | abuf = arc_loan_raw_buf(os->os_spa, dmu_objset_id(os), | |
149ce888 | 1649 | byteorder, salt, iv, mac, type, |
b5256303 TC |
1650 | datalen, arc_buf_lsize(srcdb->db_buf), |
1651 | arc_get_compression(srcdb->db_buf)); | |
1652 | } else { | |
1653 | /* we won't get a compressed db back from dmu_buf_hold() */ | |
1654 | ASSERT3U(arc_get_compression(srcdb->db_buf), | |
1655 | ==, ZIO_COMPRESS_OFF); | |
1656 | abuf = arc_loan_buf(os->os_spa, | |
149ce888 | 1657 | DMU_OT_IS_METADATA(type), datalen); |
b5256303 TC |
1658 | } |
1659 | ||
1660 | ASSERT3U(datalen, ==, arc_buf_size(abuf)); | |
1661 | ||
1662 | /* copy the data to the new buffer and assign it to the dstdb */ | |
1663 | bcopy(srcdb->db_buf->b_data, abuf->b_data, datalen); | |
1664 | dbuf_assign_arcbuf(dstdb, abuf, tx); | |
1665 | dmu_buf_rele(dst_handle, FTAG); | |
1666 | } | |
1667 | ||
9babb374 BB |
1668 | /* |
1669 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1670 | * If this is not possible copy the contents of passed arc buf via | |
1671 | * dmu_write(). | |
1672 | */ | |
1673 | void | |
440a3eb9 | 1674 | dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, |
9babb374 BB |
1675 | dmu_tx_t *tx) |
1676 | { | |
9babb374 | 1677 | dmu_buf_impl_t *db; |
440a3eb9 TC |
1678 | objset_t *os = dn->dn_objset; |
1679 | uint64_t object = dn->dn_object; | |
2aa34383 | 1680 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1681 | uint64_t blkid; |
1682 | ||
1683 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 | 1684 | blkid = dbuf_whichblock(dn, 0, offset); |
9babb374 BB |
1685 | VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); |
1686 | rw_exit(&dn->dn_struct_rwlock); | |
1687 | ||
88904bb3 MA |
1688 | /* |
1689 | * We can only assign if the offset is aligned, the arc buf is the | |
2aa34383 | 1690 | * same size as the dbuf, and the dbuf is not metadata. |
88904bb3 | 1691 | */ |
2aa34383 | 1692 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
9babb374 BB |
1693 | dbuf_assign_arcbuf(db, buf, tx); |
1694 | dbuf_rele(db, FTAG); | |
1695 | } else { | |
2aa34383 DK |
1696 | /* compressed bufs must always be assignable to their dbuf */ |
1697 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1698 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1699 | |
9babb374 | 1700 | dbuf_rele(db, FTAG); |
572e2857 | 1701 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1702 | dmu_return_arcbuf(buf); |
428870ff | 1703 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 BB |
1704 | } |
1705 | } | |
1706 | ||
440a3eb9 TC |
1707 | void |
1708 | dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, | |
1709 | dmu_tx_t *tx) | |
1710 | { | |
1711 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; | |
1712 | ||
1713 | DB_DNODE_ENTER(dbuf); | |
1714 | dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx); | |
1715 | DB_DNODE_EXIT(dbuf); | |
1716 | } | |
1717 | ||
34dc7c2f | 1718 | typedef struct { |
428870ff BB |
1719 | dbuf_dirty_record_t *dsa_dr; |
1720 | dmu_sync_cb_t *dsa_done; | |
1721 | zgd_t *dsa_zgd; | |
1722 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1723 | } dmu_sync_arg_t; |
1724 | ||
b128c09f BB |
1725 | /* ARGSUSED */ |
1726 | static void | |
1727 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1728 | { | |
428870ff BB |
1729 | dmu_sync_arg_t *dsa = varg; |
1730 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1731 | blkptr_t *bp = zio->io_bp; |
1732 | ||
428870ff BB |
1733 | if (zio->io_error == 0) { |
1734 | if (BP_IS_HOLE(bp)) { | |
1735 | /* | |
1736 | * A block of zeros may compress to a hole, but the | |
1737 | * block size still needs to be known for replay. | |
1738 | */ | |
1739 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1740 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff | 1741 | ASSERT(BP_GET_LEVEL(bp) == 0); |
b5256303 | 1742 | BP_SET_FILL(bp, 1); |
428870ff | 1743 | } |
b128c09f BB |
1744 | } |
1745 | } | |
1746 | ||
428870ff BB |
1747 | static void |
1748 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1749 | { | |
1750 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1751 | } | |
1752 | ||
34dc7c2f BB |
1753 | /* ARGSUSED */ |
1754 | static void | |
1755 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1756 | { | |
428870ff BB |
1757 | dmu_sync_arg_t *dsa = varg; |
1758 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1759 | dmu_buf_impl_t *db = dr->dr_dbuf; |
34dc7c2f | 1760 | |
34dc7c2f BB |
1761 | mutex_enter(&db->db_mtx); |
1762 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1763 | if (zio->io_error == 0) { |
03c6040b GW |
1764 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1765 | if (dr->dt.dl.dr_nopwrite) { | |
02dc43bc MA |
1766 | blkptr_t *bp = zio->io_bp; |
1767 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1768 | uint8_t chksum = BP_GET_CHECKSUM(bp_orig); | |
03c6040b GW |
1769 | |
1770 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
02dc43bc | 1771 | VERIFY(BP_EQUAL(bp, db->db_blkptr)); |
03c6040b | 1772 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); |
02dc43bc | 1773 | VERIFY(zio_checksum_table[chksum].ci_flags & |
3c67d83a | 1774 | ZCHECKSUM_FLAG_NOPWRITE); |
03c6040b | 1775 | } |
428870ff BB |
1776 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1777 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1778 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1779 | |
1780 | /* | |
1781 | * Old style holes are filled with all zeros, whereas | |
1782 | * new-style holes maintain their lsize, type, level, | |
1783 | * and birth time (see zio_write_compress). While we | |
1784 | * need to reset the BP_SET_LSIZE() call that happened | |
1785 | * in dmu_sync_ready for old style holes, we do *not* | |
1786 | * want to wipe out the information contained in new | |
1787 | * style holes. Thus, only zero out the block pointer if | |
1788 | * it's an old style hole. | |
1789 | */ | |
1790 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1791 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1792 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1793 | } else { | |
1794 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1795 | } | |
34dc7c2f BB |
1796 | cv_broadcast(&db->db_changed); |
1797 | mutex_exit(&db->db_mtx); | |
1798 | ||
428870ff | 1799 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1800 | |
428870ff BB |
1801 | kmem_free(dsa, sizeof (*dsa)); |
1802 | } | |
1803 | ||
1804 | static void | |
1805 | dmu_sync_late_arrival_done(zio_t *zio) | |
1806 | { | |
1807 | blkptr_t *bp = zio->io_bp; | |
1808 | dmu_sync_arg_t *dsa = zio->io_private; | |
03c6040b | 1809 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); |
428870ff BB |
1810 | |
1811 | if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { | |
02dc43bc MA |
1812 | ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); |
1813 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1814 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1815 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1816 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
428870ff BB |
1817 | } |
1818 | ||
1819 | dmu_tx_commit(dsa->dsa_tx); | |
1820 | ||
1821 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1822 | ||
a6255b7f | 1823 | abd_put(zio->io_abd); |
428870ff BB |
1824 | kmem_free(dsa, sizeof (*dsa)); |
1825 | } | |
1826 | ||
1827 | static int | |
1828 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1829 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1830 | { |
1831 | dmu_sync_arg_t *dsa; | |
1832 | dmu_tx_t *tx; | |
1833 | ||
1834 | tx = dmu_tx_create(os); | |
1835 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1836 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1837 | dmu_tx_abort(tx); | |
2e528b49 MA |
1838 | /* Make zl_get_data do txg_waited_synced() */ |
1839 | return (SET_ERROR(EIO)); | |
428870ff BB |
1840 | } |
1841 | ||
1ce23dca PS |
1842 | /* |
1843 | * In order to prevent the zgd's lwb from being free'd prior to | |
1844 | * dmu_sync_late_arrival_done() being called, we have to ensure | |
1845 | * the lwb's "max txg" takes this tx's txg into account. | |
1846 | */ | |
1847 | zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); | |
1848 | ||
79c76d5b | 1849 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1850 | dsa->dsa_dr = NULL; |
1851 | dsa->dsa_done = done; | |
1852 | dsa->dsa_zgd = zgd; | |
1853 | dsa->dsa_tx = tx; | |
1854 | ||
02dc43bc MA |
1855 | /* |
1856 | * Since we are currently syncing this txg, it's nontrivial to | |
1857 | * determine what BP to nopwrite against, so we disable nopwrite. | |
1858 | * | |
1859 | * When syncing, the db_blkptr is initially the BP of the previous | |
1860 | * txg. We can not nopwrite against it because it will be changed | |
1861 | * (this is similar to the non-late-arrival case where the dbuf is | |
1862 | * dirty in a future txg). | |
1863 | * | |
1864 | * Then dbuf_write_ready() sets bp_blkptr to the location we will write. | |
1865 | * We can not nopwrite against it because although the BP will not | |
1866 | * (typically) be changed, the data has not yet been persisted to this | |
1867 | * location. | |
1868 | * | |
1869 | * Finally, when dbuf_write_done() is called, it is theoretically | |
1870 | * possible to always nopwrite, because the data that was written in | |
1871 | * this txg is the same data that we are trying to write. However we | |
1872 | * would need to check that this dbuf is not dirty in any future | |
1873 | * txg's (as we do in the normal dmu_sync() path). For simplicity, we | |
1874 | * don't nopwrite in this case. | |
1875 | */ | |
1876 | zp->zp_nopwrite = B_FALSE; | |
1877 | ||
a6255b7f DQ |
1878 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
1879 | abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), | |
1880 | zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, | |
1881 | dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, | |
1882 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1883 | |
1884 | return (0); | |
34dc7c2f BB |
1885 | } |
1886 | ||
1887 | /* | |
1888 | * Intent log support: sync the block associated with db to disk. | |
1889 | * N.B. and XXX: the caller is responsible for making sure that the | |
1890 | * data isn't changing while dmu_sync() is writing it. | |
1891 | * | |
1892 | * Return values: | |
1893 | * | |
03c6040b | 1894 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1895 | * The caller should not log the write. |
1896 | * | |
1897 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1898 | * The caller should not log the write. | |
1899 | * | |
1900 | * EALREADY: this block is already in the process of being synced. | |
1901 | * The caller should track its progress (somehow). | |
1902 | * | |
428870ff BB |
1903 | * EIO: could not do the I/O. |
1904 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1905 | * |
428870ff BB |
1906 | * 0: the I/O has been initiated. |
1907 | * The caller should log this blkptr in the done callback. | |
1908 | * It is possible that the I/O will fail, in which case | |
1909 | * the error will be reported to the done callback and | |
1910 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1911 | */ |
1912 | int | |
428870ff | 1913 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1914 | { |
428870ff BB |
1915 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
1916 | objset_t *os = db->db_objset; | |
1917 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1918 | dbuf_dirty_record_t *dr; |
428870ff | 1919 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1920 | zbookmark_phys_t zb; |
428870ff | 1921 | zio_prop_t zp; |
572e2857 | 1922 | dnode_t *dn; |
34dc7c2f | 1923 | |
428870ff | 1924 | ASSERT(pio != NULL); |
34dc7c2f BB |
1925 | ASSERT(txg != 0); |
1926 | ||
428870ff BB |
1927 | SET_BOOKMARK(&zb, ds->ds_object, |
1928 | db->db.db_object, db->db_level, db->db_blkid); | |
1929 | ||
572e2857 BB |
1930 | DB_DNODE_ENTER(db); |
1931 | dn = DB_DNODE(db); | |
82644107 | 1932 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); |
572e2857 | 1933 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1934 | |
1935 | /* | |
428870ff | 1936 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1937 | */ |
428870ff BB |
1938 | if (txg > spa_freeze_txg(os->os_spa)) |
1939 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1940 | |
1941 | /* | |
428870ff BB |
1942 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1943 | * and us. If we determine that this txg is not yet syncing, | |
1944 | * but it begins to sync a moment later, that's OK because the | |
1945 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1946 | */ |
428870ff BB |
1947 | mutex_enter(&db->db_mtx); |
1948 | ||
1949 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1950 | /* |
428870ff | 1951 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1952 | */ |
428870ff | 1953 | mutex_exit(&db->db_mtx); |
2e528b49 | 1954 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1955 | } |
1956 | ||
428870ff BB |
1957 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1958 | /* | |
1959 | * This txg is currently syncing, so we can't mess with | |
1960 | * the dirty record anymore; just write a new log block. | |
1961 | */ | |
1962 | mutex_exit(&db->db_mtx); | |
1963 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1964 | } |
1965 | ||
1966 | dr = db->db_last_dirty; | |
428870ff | 1967 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 1968 | dr = dr->dr_next; |
428870ff BB |
1969 | |
1970 | if (dr == NULL) { | |
34dc7c2f | 1971 | /* |
428870ff | 1972 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1973 | * There's no need to log writes to freed blocks, so we're done. |
1974 | */ | |
1975 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1976 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1977 | } |
1978 | ||
03c6040b GW |
1979 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); |
1980 | ||
02dc43bc MA |
1981 | if (db->db_blkptr != NULL) { |
1982 | /* | |
1983 | * We need to fill in zgd_bp with the current blkptr so that | |
1984 | * the nopwrite code can check if we're writing the same | |
1985 | * data that's already on disk. We can only nopwrite if we | |
1986 | * are sure that after making the copy, db_blkptr will not | |
1987 | * change until our i/o completes. We ensure this by | |
1988 | * holding the db_mtx, and only allowing nopwrite if the | |
1989 | * block is not already dirty (see below). This is verified | |
1990 | * by dmu_sync_done(), which VERIFYs that the db_blkptr has | |
1991 | * not changed. | |
1992 | */ | |
1993 | *zgd->zgd_bp = *db->db_blkptr; | |
1994 | } | |
1995 | ||
03c6040b | 1996 | /* |
f3c517d8 MA |
1997 | * Assume the on-disk data is X, the current syncing data (in |
1998 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1999 | * in dmu_sync). | |
2000 | * | |
2001 | * We usually want to perform a nopwrite if X and Z are the | |
2002 | * same. However, if Y is different (i.e. the BP is going to | |
2003 | * change before this write takes effect), then a nopwrite will | |
2004 | * be incorrect - we would override with X, which could have | |
2005 | * been freed when Y was written. | |
2006 | * | |
2007 | * (Note that this is not a concern when we are nop-writing from | |
2008 | * syncing context, because X and Y must be identical, because | |
2009 | * all previous txgs have been synced.) | |
2010 | * | |
2011 | * Therefore, we disable nopwrite if the current BP could change | |
2012 | * before this TXG. There are two ways it could change: by | |
2013 | * being dirty (dr_next is non-NULL), or by being freed | |
2014 | * (dnode_block_freed()). This behavior is verified by | |
2015 | * zio_done(), which VERIFYs that the override BP is identical | |
2016 | * to the on-disk BP. | |
03c6040b | 2017 | */ |
f3c517d8 MA |
2018 | DB_DNODE_ENTER(db); |
2019 | dn = DB_DNODE(db); | |
2020 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
03c6040b | 2021 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 2022 | DB_DNODE_EXIT(db); |
03c6040b | 2023 | |
34dc7c2f | 2024 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
2025 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
2026 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 2027 | /* |
428870ff BB |
2028 | * We have already issued a sync write for this buffer, |
2029 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
2030 | * have been dirtied since, or we would have cleared the state. |
2031 | */ | |
34dc7c2f | 2032 | mutex_exit(&db->db_mtx); |
2e528b49 | 2033 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
2034 | } |
2035 | ||
428870ff | 2036 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 2037 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 2038 | mutex_exit(&db->db_mtx); |
34dc7c2f | 2039 | |
79c76d5b | 2040 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
2041 | dsa->dsa_dr = dr; |
2042 | dsa->dsa_done = done; | |
2043 | dsa->dsa_zgd = zgd; | |
2044 | dsa->dsa_tx = NULL; | |
b128c09f | 2045 | |
428870ff | 2046 | zio_nowait(arc_write(pio, os->os_spa, txg, |
02dc43bc | 2047 | zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 2048 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 2049 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 2050 | |
428870ff | 2051 | return (0); |
34dc7c2f BB |
2052 | } |
2053 | ||
b5256303 TC |
2054 | int |
2055 | dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx) | |
2056 | { | |
2057 | dnode_t *dn; | |
2058 | int err; | |
2059 | ||
2060 | err = dnode_hold(os, object, FTAG, &dn); | |
2061 | if (err) | |
2062 | return (err); | |
2063 | err = dnode_set_nlevels(dn, nlevels, tx); | |
2064 | dnode_rele(dn, FTAG); | |
2065 | return (err); | |
2066 | } | |
2067 | ||
34dc7c2f BB |
2068 | int |
2069 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
4ea3f864 | 2070 | dmu_tx_t *tx) |
34dc7c2f BB |
2071 | { |
2072 | dnode_t *dn; | |
2073 | int err; | |
2074 | ||
428870ff | 2075 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2076 | if (err) |
2077 | return (err); | |
2078 | err = dnode_set_blksz(dn, size, ibs, tx); | |
2079 | dnode_rele(dn, FTAG); | |
2080 | return (err); | |
2081 | } | |
2082 | ||
ae76f45c TC |
2083 | int |
2084 | dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid, | |
2085 | dmu_tx_t *tx) | |
2086 | { | |
2087 | dnode_t *dn; | |
2088 | int err; | |
2089 | ||
2090 | err = dnode_hold(os, object, FTAG, &dn); | |
2091 | if (err) | |
2092 | return (err); | |
2093 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
2094 | dnode_new_blkid(dn, maxblkid, tx, B_FALSE); | |
2095 | rw_exit(&dn->dn_struct_rwlock); | |
2096 | dnode_rele(dn, FTAG); | |
2097 | return (0); | |
2098 | } | |
2099 | ||
34dc7c2f BB |
2100 | void |
2101 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
4ea3f864 | 2102 | dmu_tx_t *tx) |
34dc7c2f BB |
2103 | { |
2104 | dnode_t *dn; | |
2105 | ||
9b67f605 MA |
2106 | /* |
2107 | * Send streams include each object's checksum function. This | |
2108 | * check ensures that the receiving system can understand the | |
2109 | * checksum function transmitted. | |
2110 | */ | |
2111 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
2112 | ||
2113 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
2114 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
2115 | dn->dn_checksum = checksum; |
2116 | dnode_setdirty(dn, tx); | |
2117 | dnode_rele(dn, FTAG); | |
2118 | } | |
2119 | ||
2120 | void | |
2121 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
4ea3f864 | 2122 | dmu_tx_t *tx) |
34dc7c2f BB |
2123 | { |
2124 | dnode_t *dn; | |
2125 | ||
9b67f605 MA |
2126 | /* |
2127 | * Send streams include each object's compression function. This | |
2128 | * check ensures that the receiving system can understand the | |
2129 | * compression function transmitted. | |
2130 | */ | |
2131 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
2132 | ||
2133 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
2134 | dn->dn_compress = compress; |
2135 | dnode_setdirty(dn, tx); | |
2136 | dnode_rele(dn, FTAG); | |
2137 | } | |
2138 | ||
faf0f58c MA |
2139 | /* |
2140 | * When the "redundant_metadata" property is set to "most", only indirect | |
2141 | * blocks of this level and higher will have an additional ditto block. | |
2142 | */ | |
2143 | int zfs_redundant_metadata_most_ditto_level = 2; | |
2144 | ||
428870ff | 2145 | void |
82644107 | 2146 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
428870ff BB |
2147 | { |
2148 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 2149 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 2150 | (wp & WP_SPILL)); |
428870ff BB |
2151 | enum zio_checksum checksum = os->os_checksum; |
2152 | enum zio_compress compress = os->os_compress; | |
2153 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
03c6040b GW |
2154 | boolean_t dedup = B_FALSE; |
2155 | boolean_t nopwrite = B_FALSE; | |
428870ff | 2156 | boolean_t dedup_verify = os->os_dedup_verify; |
b5256303 | 2157 | boolean_t encrypt = B_FALSE; |
428870ff | 2158 | int copies = os->os_copies; |
a7004725 | 2159 | |
428870ff | 2160 | /* |
03c6040b GW |
2161 | * We maintain different write policies for each of the following |
2162 | * types of data: | |
2163 | * 1. metadata | |
2164 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
2165 | * 3. all other level 0 blocks | |
428870ff BB |
2166 | */ |
2167 | if (ismd) { | |
b1d21733 TC |
2168 | /* |
2169 | * XXX -- we should design a compression algorithm | |
2170 | * that specializes in arrays of bps. | |
2171 | */ | |
2172 | compress = zio_compress_select(os->os_spa, | |
2173 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
03c6040b | 2174 | |
428870ff BB |
2175 | /* |
2176 | * Metadata always gets checksummed. If the data | |
2177 | * checksum is multi-bit correctable, and it's not a | |
2178 | * ZBT-style checksum, then it's suitable for metadata | |
2179 | * as well. Otherwise, the metadata checksum defaults | |
2180 | * to fletcher4. | |
2181 | */ | |
3c67d83a TH |
2182 | if (!(zio_checksum_table[checksum].ci_flags & |
2183 | ZCHECKSUM_FLAG_METADATA) || | |
2184 | (zio_checksum_table[checksum].ci_flags & | |
2185 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 2186 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
2187 | |
2188 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
2189 | (os->os_redundant_metadata == | |
2190 | ZFS_REDUNDANT_METADATA_MOST && | |
2191 | (level >= zfs_redundant_metadata_most_ditto_level || | |
2192 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
2193 | copies++; | |
03c6040b GW |
2194 | } else if (wp & WP_NOFILL) { |
2195 | ASSERT(level == 0); | |
428870ff | 2196 | |
428870ff | 2197 | /* |
03c6040b GW |
2198 | * If we're writing preallocated blocks, we aren't actually |
2199 | * writing them so don't set any policy properties. These | |
2200 | * blocks are currently only used by an external subsystem | |
2201 | * outside of zfs (i.e. dump) and not written by the zio | |
2202 | * pipeline. | |
428870ff | 2203 | */ |
03c6040b GW |
2204 | compress = ZIO_COMPRESS_OFF; |
2205 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 2206 | } else { |
99197f03 JG |
2207 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
2208 | compress); | |
428870ff | 2209 | |
03c6040b GW |
2210 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
2211 | zio_checksum_select(dn->dn_checksum, checksum) : | |
2212 | dedup_checksum; | |
428870ff | 2213 | |
03c6040b GW |
2214 | /* |
2215 | * Determine dedup setting. If we are in dmu_sync(), | |
2216 | * we won't actually dedup now because that's all | |
2217 | * done in syncing context; but we do want to use the | |
2218 | * dedup checkum. If the checksum is not strong | |
2219 | * enough to ensure unique signatures, force | |
2220 | * dedup_verify. | |
2221 | */ | |
2222 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
2223 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
2224 | if (!(zio_checksum_table[checksum].ci_flags & |
2225 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
2226 | dedup_verify = B_TRUE; |
2227 | } | |
428870ff | 2228 | |
03c6040b | 2229 | /* |
3c67d83a TH |
2230 | * Enable nopwrite if we have secure enough checksum |
2231 | * algorithm (see comment in zio_nop_write) and | |
2232 | * compression is enabled. We don't enable nopwrite if | |
2233 | * dedup is enabled as the two features are mutually | |
2234 | * exclusive. | |
03c6040b | 2235 | */ |
3c67d83a TH |
2236 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
2237 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 2238 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
2239 | } |
2240 | ||
b5256303 TC |
2241 | /* |
2242 | * All objects in an encrypted objset are protected from modification | |
2243 | * via a MAC. Encrypted objects store their IV and salt in the last DVA | |
2244 | * in the bp, so we cannot use all copies. Encrypted objects are also | |
2245 | * not subject to nopwrite since writing the same data will still | |
2246 | * result in a new ciphertext. Only encrypted blocks can be dedup'd | |
2247 | * to avoid ambiguity in the dedup code since the DDT does not store | |
2248 | * object types. | |
2249 | */ | |
2250 | if (os->os_encrypted && (wp & WP_NOFILL) == 0) { | |
2251 | encrypt = B_TRUE; | |
2252 | ||
2253 | if (DMU_OT_IS_ENCRYPTED(type)) { | |
2254 | copies = MIN(copies, SPA_DVAS_PER_BP - 1); | |
2255 | nopwrite = B_FALSE; | |
2256 | } else { | |
2257 | dedup = B_FALSE; | |
2258 | } | |
2259 | ||
ae76f45c TC |
2260 | if (level <= 0 && |
2261 | (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) { | |
b5256303 | 2262 | compress = ZIO_COMPRESS_EMPTY; |
ae76f45c | 2263 | } |
b5256303 | 2264 | } |
2aa34383 | 2265 | |
b5256303 TC |
2266 | zp->zp_compress = compress; |
2267 | zp->zp_checksum = checksum; | |
428870ff BB |
2268 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
2269 | zp->zp_level = level; | |
faf0f58c | 2270 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
2271 | zp->zp_dedup = dedup; |
2272 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 2273 | zp->zp_nopwrite = nopwrite; |
b5256303 TC |
2274 | zp->zp_encrypt = encrypt; |
2275 | zp->zp_byteorder = ZFS_HOST_BYTEORDER; | |
2276 | bzero(zp->zp_salt, ZIO_DATA_SALT_LEN); | |
2277 | bzero(zp->zp_iv, ZIO_DATA_IV_LEN); | |
2278 | bzero(zp->zp_mac, ZIO_DATA_MAC_LEN); | |
cc99f275 DB |
2279 | zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? |
2280 | os->os_zpl_special_smallblock : 0; | |
b5256303 TC |
2281 | |
2282 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
428870ff BB |
2283 | } |
2284 | ||
66aca247 DB |
2285 | /* |
2286 | * This function is only called from zfs_holey_common() for zpl_llseek() | |
2287 | * in order to determine the location of holes. In order to accurately | |
2288 | * report holes all dirty data must be synced to disk. This causes extremely | |
2289 | * poor performance when seeking for holes in a dirty file. As a compromise, | |
2290 | * only provide hole data when the dnode is clean. When a dnode is dirty | |
2291 | * report the dnode as having no holes which is always a safe thing to do. | |
2292 | */ | |
34dc7c2f BB |
2293 | int |
2294 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
2295 | { | |
2296 | dnode_t *dn; | |
2297 | int i, err; | |
66aca247 | 2298 | boolean_t clean = B_TRUE; |
34dc7c2f | 2299 | |
428870ff | 2300 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2301 | if (err) |
2302 | return (err); | |
66aca247 | 2303 | |
34dc7c2f | 2304 | /* |
66aca247 | 2305 | * Check if dnode is dirty |
34dc7c2f | 2306 | */ |
454365bb | 2307 | for (i = 0; i < TXG_SIZE; i++) { |
edc1e713 | 2308 | if (multilist_link_active(&dn->dn_dirty_link[i])) { |
454365bb BB |
2309 | clean = B_FALSE; |
2310 | break; | |
66aca247 | 2311 | } |
34dc7c2f | 2312 | } |
66aca247 DB |
2313 | |
2314 | /* | |
2315 | * If compatibility option is on, sync any current changes before | |
2316 | * we go trundling through the block pointers. | |
2317 | */ | |
2318 | if (!clean && zfs_dmu_offset_next_sync) { | |
2319 | clean = B_TRUE; | |
34dc7c2f BB |
2320 | dnode_rele(dn, FTAG); |
2321 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 2322 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2323 | if (err) |
2324 | return (err); | |
2325 | } | |
2326 | ||
66aca247 DB |
2327 | if (clean) |
2328 | err = dnode_next_offset(dn, | |
2329 | (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
2330 | else | |
2331 | err = SET_ERROR(EBUSY); | |
2332 | ||
34dc7c2f BB |
2333 | dnode_rele(dn, FTAG); |
2334 | ||
2335 | return (err); | |
2336 | } | |
2337 | ||
2338 | void | |
e0b0ca98 | 2339 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 2340 | { |
e0b0ca98 | 2341 | dnode_phys_t *dnp = dn->dn_phys; |
428870ff | 2342 | |
34dc7c2f BB |
2343 | doi->doi_data_block_size = dn->dn_datablksz; |
2344 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
2345 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
2346 | doi->doi_type = dn->dn_type; |
2347 | doi->doi_bonus_type = dn->dn_bonustype; | |
2348 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 2349 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
2350 | doi->doi_indirection = dn->dn_nlevels; |
2351 | doi->doi_checksum = dn->dn_checksum; | |
2352 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 2353 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 2354 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 2355 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 2356 | doi->doi_fill_count = 0; |
1c27024e | 2357 | for (int i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 2358 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
2359 | } |
2360 | ||
2361 | void | |
2362 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2363 | { | |
2364 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2365 | mutex_enter(&dn->dn_mtx); | |
2366 | ||
2367 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
2368 | |
2369 | mutex_exit(&dn->dn_mtx); | |
2370 | rw_exit(&dn->dn_struct_rwlock); | |
2371 | } | |
2372 | ||
2373 | /* | |
2374 | * Get information on a DMU object. | |
2375 | * If doi is NULL, just indicates whether the object exists. | |
2376 | */ | |
2377 | int | |
2378 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2379 | { | |
2380 | dnode_t *dn; | |
428870ff | 2381 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
2382 | |
2383 | if (err) | |
2384 | return (err); | |
2385 | ||
2386 | if (doi != NULL) | |
2387 | dmu_object_info_from_dnode(dn, doi); | |
2388 | ||
2389 | dnode_rele(dn, FTAG); | |
2390 | return (0); | |
2391 | } | |
2392 | ||
2393 | /* | |
2394 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2395 | */ | |
2396 | void | |
572e2857 | 2397 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 2398 | { |
572e2857 BB |
2399 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2400 | ||
2401 | DB_DNODE_ENTER(db); | |
2402 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2403 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
2404 | } |
2405 | ||
2406 | /* | |
2407 | * Faster still when you only care about the size. | |
2408 | * This is specifically optimized for zfs_getattr(). | |
2409 | */ | |
2410 | void | |
572e2857 BB |
2411 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
2412 | u_longlong_t *nblk512) | |
34dc7c2f | 2413 | { |
572e2857 BB |
2414 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
2415 | dnode_t *dn; | |
2416 | ||
2417 | DB_DNODE_ENTER(db); | |
2418 | dn = DB_DNODE(db); | |
34dc7c2f BB |
2419 | |
2420 | *blksize = dn->dn_datablksz; | |
50c957f7 | 2421 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 2422 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
2423 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
2424 | DB_DNODE_EXIT(db); | |
2425 | } | |
2426 | ||
2427 | void | |
2428 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
2429 | { | |
2430 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2431 | dnode_t *dn; | |
2432 | ||
2433 | DB_DNODE_ENTER(db); | |
2434 | dn = DB_DNODE(db); | |
2435 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2436 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2437 | } |
2438 | ||
2439 | void | |
2440 | byteswap_uint64_array(void *vbuf, size_t size) | |
2441 | { | |
2442 | uint64_t *buf = vbuf; | |
2443 | size_t count = size >> 3; | |
2444 | int i; | |
2445 | ||
2446 | ASSERT((size & 7) == 0); | |
2447 | ||
2448 | for (i = 0; i < count; i++) | |
2449 | buf[i] = BSWAP_64(buf[i]); | |
2450 | } | |
2451 | ||
2452 | void | |
2453 | byteswap_uint32_array(void *vbuf, size_t size) | |
2454 | { | |
2455 | uint32_t *buf = vbuf; | |
2456 | size_t count = size >> 2; | |
2457 | int i; | |
2458 | ||
2459 | ASSERT((size & 3) == 0); | |
2460 | ||
2461 | for (i = 0; i < count; i++) | |
2462 | buf[i] = BSWAP_32(buf[i]); | |
2463 | } | |
2464 | ||
2465 | void | |
2466 | byteswap_uint16_array(void *vbuf, size_t size) | |
2467 | { | |
2468 | uint16_t *buf = vbuf; | |
2469 | size_t count = size >> 1; | |
2470 | int i; | |
2471 | ||
2472 | ASSERT((size & 1) == 0); | |
2473 | ||
2474 | for (i = 0; i < count; i++) | |
2475 | buf[i] = BSWAP_16(buf[i]); | |
2476 | } | |
2477 | ||
2478 | /* ARGSUSED */ | |
2479 | void | |
2480 | byteswap_uint8_array(void *vbuf, size_t size) | |
2481 | { | |
2482 | } | |
2483 | ||
2484 | void | |
2485 | dmu_init(void) | |
2486 | { | |
a6255b7f | 2487 | abd_init(); |
428870ff | 2488 | zfs_dbgmsg_init(); |
572e2857 BB |
2489 | sa_cache_init(); |
2490 | xuio_stat_init(); | |
2491 | dmu_objset_init(); | |
34dc7c2f | 2492 | dnode_init(); |
428870ff | 2493 | zfetch_init(); |
570827e1 | 2494 | dmu_tx_init(); |
34dc7c2f | 2495 | l2arc_init(); |
29809a6c | 2496 | arc_init(); |
d3c2ae1c | 2497 | dbuf_init(); |
34dc7c2f BB |
2498 | } |
2499 | ||
2500 | void | |
2501 | dmu_fini(void) | |
2502 | { | |
e49f1e20 | 2503 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2504 | l2arc_fini(); |
570827e1 | 2505 | dmu_tx_fini(); |
428870ff | 2506 | zfetch_fini(); |
34dc7c2f | 2507 | dbuf_fini(); |
572e2857 BB |
2508 | dnode_fini(); |
2509 | dmu_objset_fini(); | |
428870ff BB |
2510 | xuio_stat_fini(); |
2511 | sa_cache_fini(); | |
2512 | zfs_dbgmsg_fini(); | |
a6255b7f | 2513 | abd_fini(); |
34dc7c2f | 2514 | } |
c28b2279 | 2515 | |
93ce2b4c | 2516 | #if defined(_KERNEL) |
c28b2279 | 2517 | EXPORT_SYMBOL(dmu_bonus_hold); |
a473d90c AZ |
2518 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2519 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2520 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2521 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2522 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2523 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 | 2524 | EXPORT_SYMBOL(dmu_read); |
0eef1bde | 2525 | EXPORT_SYMBOL(dmu_read_by_dnode); |
c28b2279 | 2526 | EXPORT_SYMBOL(dmu_write); |
0eef1bde | 2527 | EXPORT_SYMBOL(dmu_write_by_dnode); |
57b650b8 | 2528 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2529 | EXPORT_SYMBOL(dmu_object_info); |
2530 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2531 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2532 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2533 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
b5256303 | 2534 | EXPORT_SYMBOL(dmu_object_set_nlevels); |
c28b2279 | 2535 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
ae76f45c | 2536 | EXPORT_SYMBOL(dmu_object_set_maxblkid); |
c28b2279 BB |
2537 | EXPORT_SYMBOL(dmu_object_set_checksum); |
2538 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
2539 | EXPORT_SYMBOL(dmu_write_policy); |
2540 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2541 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2542 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
440a3eb9 TC |
2543 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode); |
2544 | EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf); | |
b10c77f7 | 2545 | EXPORT_SYMBOL(dmu_buf_hold); |
c28b2279 | 2546 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 | 2547 | |
bef78122 | 2548 | /* BEGIN CSTYLED */ |
03c6040b GW |
2549 | module_param(zfs_nopwrite_enabled, int, 0644); |
2550 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2551 | ||
bef78122 DQ |
2552 | module_param(zfs_per_txg_dirty_frees_percent, ulong, 0644); |
2553 | MODULE_PARM_DESC(zfs_per_txg_dirty_frees_percent, | |
2554 | "percentage of dirtied blocks from frees in one TXG"); | |
66aca247 DB |
2555 | |
2556 | module_param(zfs_dmu_offset_next_sync, int, 0644); | |
2557 | MODULE_PARM_DESC(zfs_dmu_offset_next_sync, | |
2558 | "Enable forcing txg sync to find holes"); | |
2559 | ||
bef78122 | 2560 | /* END CSTYLED */ |
66aca247 | 2561 | |
c28b2279 | 2562 | #endif |