]>
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. |
bc77ba73 | 23 | * Copyright (c) 2011, 2016 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. |
62bdd5eb | 26 | * Copyright (c) 2014, 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> |
34dc7c2f BB |
50 | #ifdef _KERNEL |
51 | #include <sys/vmsystm.h> | |
b128c09f | 52 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
53 | #endif |
54 | ||
03c6040b GW |
55 | /* |
56 | * Enable/disable nopwrite feature. | |
57 | */ | |
58 | int zfs_nopwrite_enabled = 1; | |
59 | ||
34dc7c2f | 60 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
9ae529ec CS |
61 | { DMU_BSWAP_UINT8, TRUE, "unallocated" }, |
62 | { DMU_BSWAP_ZAP, TRUE, "object directory" }, | |
63 | { DMU_BSWAP_UINT64, TRUE, "object array" }, | |
64 | { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, | |
65 | { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, | |
66 | { DMU_BSWAP_UINT64, TRUE, "bpobj" }, | |
67 | { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, | |
68 | { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, | |
69 | { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, | |
70 | { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, | |
71 | { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, | |
72 | { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, | |
73 | { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, | |
74 | { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, | |
75 | { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, | |
76 | { DMU_BSWAP_ZAP, TRUE, "DSL props" }, | |
77 | { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, | |
78 | { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, | |
79 | { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, | |
80 | { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, | |
81 | { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, | |
82 | { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, | |
83 | { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, | |
84 | { DMU_BSWAP_UINT8, FALSE, "zvol object" }, | |
85 | { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, | |
86 | { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, | |
87 | { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, | |
88 | { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, | |
89 | { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, | |
90 | { DMU_BSWAP_UINT8, TRUE, "SPA history" }, | |
91 | { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, | |
92 | { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, | |
93 | { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, | |
94 | { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, | |
95 | { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, | |
96 | { DMU_BSWAP_UINT8, TRUE, "FUID table" }, | |
97 | { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, | |
98 | { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, | |
99 | { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, | |
100 | { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, | |
101 | { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, | |
102 | { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, | |
103 | { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, | |
104 | { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, | |
105 | { DMU_BSWAP_UINT8, TRUE, "System attributes" }, | |
106 | { DMU_BSWAP_ZAP, TRUE, "SA master node" }, | |
107 | { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, | |
108 | { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, | |
109 | { DMU_BSWAP_ZAP, TRUE, "scan translations" }, | |
110 | { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, | |
111 | { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, | |
112 | { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, | |
113 | { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, | |
114 | { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } | |
115 | }; | |
116 | ||
117 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
118 | { byteswap_uint8_array, "uint8" }, | |
119 | { byteswap_uint16_array, "uint16" }, | |
120 | { byteswap_uint32_array, "uint32" }, | |
121 | { byteswap_uint64_array, "uint64" }, | |
122 | { zap_byteswap, "zap" }, | |
123 | { dnode_buf_byteswap, "dnode" }, | |
124 | { dmu_objset_byteswap, "objset" }, | |
125 | { zfs_znode_byteswap, "znode" }, | |
126 | { zfs_oldacl_byteswap, "oldacl" }, | |
127 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
128 | }; |
129 | ||
2bce8049 MA |
130 | int |
131 | dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, | |
132 | void *tag, dmu_buf_t **dbp) | |
133 | { | |
134 | uint64_t blkid; | |
135 | dmu_buf_impl_t *db; | |
136 | ||
137 | blkid = dbuf_whichblock(dn, 0, offset); | |
138 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
139 | db = dbuf_hold(dn, blkid, tag); | |
140 | rw_exit(&dn->dn_struct_rwlock); | |
141 | ||
142 | if (db == NULL) { | |
143 | *dbp = NULL; | |
144 | return (SET_ERROR(EIO)); | |
145 | } | |
146 | ||
147 | *dbp = &db->db; | |
148 | return (0); | |
149 | } | |
34dc7c2f | 150 | int |
9b67f605 MA |
151 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
152 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
153 | { |
154 | dnode_t *dn; | |
155 | uint64_t blkid; | |
156 | dmu_buf_impl_t *db; | |
157 | int err; | |
428870ff BB |
158 | |
159 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
160 | if (err) |
161 | return (err); | |
fcff0f35 | 162 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f BB |
163 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
164 | db = dbuf_hold(dn, blkid, tag); | |
165 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
166 | dnode_rele(dn, FTAG); |
167 | ||
34dc7c2f | 168 | if (db == NULL) { |
9b67f605 MA |
169 | *dbp = NULL; |
170 | return (SET_ERROR(EIO)); | |
171 | } | |
172 | ||
173 | *dbp = &db->db; | |
174 | return (err); | |
175 | } | |
176 | ||
2bce8049 MA |
177 | int |
178 | dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, | |
179 | void *tag, dmu_buf_t **dbp, int flags) | |
180 | { | |
181 | int err; | |
182 | int db_flags = DB_RF_CANFAIL; | |
183 | ||
184 | if (flags & DMU_READ_NO_PREFETCH) | |
185 | db_flags |= DB_RF_NOPREFETCH; | |
186 | ||
187 | err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); | |
188 | if (err == 0) { | |
189 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
190 | err = dbuf_read(db, NULL, db_flags); | |
191 | if (err != 0) { | |
192 | dbuf_rele(db, tag); | |
193 | *dbp = NULL; | |
194 | } | |
195 | } | |
196 | ||
197 | return (err); | |
198 | } | |
199 | ||
9b67f605 MA |
200 | int |
201 | dmu_buf_hold(objset_t *os, uint64_t object, 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; | |
209 | ||
210 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
211 | if (err == 0) { | |
212 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 213 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 214 | if (err != 0) { |
34dc7c2f | 215 | dbuf_rele(db, tag); |
9b67f605 | 216 | *dbp = NULL; |
34dc7c2f BB |
217 | } |
218 | } | |
219 | ||
34dc7c2f BB |
220 | return (err); |
221 | } | |
222 | ||
223 | int | |
224 | dmu_bonus_max(void) | |
225 | { | |
50c957f7 | 226 | return (DN_OLD_MAX_BONUSLEN); |
34dc7c2f BB |
227 | } |
228 | ||
229 | int | |
572e2857 | 230 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 231 | { |
572e2857 BB |
232 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
233 | dnode_t *dn; | |
234 | int error; | |
34dc7c2f | 235 | |
572e2857 BB |
236 | DB_DNODE_ENTER(db); |
237 | dn = DB_DNODE(db); | |
238 | ||
239 | if (dn->dn_bonus != db) { | |
2e528b49 | 240 | error = SET_ERROR(EINVAL); |
572e2857 | 241 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 242 | error = SET_ERROR(EINVAL); |
572e2857 BB |
243 | } else { |
244 | dnode_setbonuslen(dn, newsize, tx); | |
245 | error = 0; | |
246 | } | |
247 | ||
248 | DB_DNODE_EXIT(db); | |
249 | return (error); | |
34dc7c2f BB |
250 | } |
251 | ||
428870ff | 252 | int |
572e2857 | 253 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 254 | { |
572e2857 BB |
255 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
256 | dnode_t *dn; | |
257 | int error; | |
428870ff | 258 | |
572e2857 BB |
259 | DB_DNODE_ENTER(db); |
260 | dn = DB_DNODE(db); | |
428870ff | 261 | |
9ae529ec | 262 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 263 | error = SET_ERROR(EINVAL); |
572e2857 | 264 | } else if (dn->dn_bonus != db) { |
2e528b49 | 265 | error = SET_ERROR(EINVAL); |
572e2857 BB |
266 | } else { |
267 | dnode_setbonus_type(dn, type, tx); | |
268 | error = 0; | |
269 | } | |
428870ff | 270 | |
572e2857 BB |
271 | DB_DNODE_EXIT(db); |
272 | return (error); | |
273 | } | |
274 | ||
275 | dmu_object_type_t | |
276 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
277 | { | |
278 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
279 | dnode_t *dn; | |
280 | dmu_object_type_t type; | |
281 | ||
282 | DB_DNODE_ENTER(db); | |
283 | dn = DB_DNODE(db); | |
284 | type = dn->dn_bonustype; | |
285 | DB_DNODE_EXIT(db); | |
286 | ||
287 | return (type); | |
428870ff BB |
288 | } |
289 | ||
290 | int | |
291 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
292 | { | |
293 | dnode_t *dn; | |
294 | int error; | |
295 | ||
296 | error = dnode_hold(os, object, FTAG, &dn); | |
297 | dbuf_rm_spill(dn, tx); | |
298 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
299 | dnode_rm_spill(dn, tx); | |
300 | rw_exit(&dn->dn_struct_rwlock); | |
301 | dnode_rele(dn, FTAG); | |
302 | return (error); | |
303 | } | |
304 | ||
34dc7c2f BB |
305 | /* |
306 | * returns ENOENT, EIO, or 0. | |
307 | */ | |
308 | int | |
309 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) | |
310 | { | |
311 | dnode_t *dn; | |
312 | dmu_buf_impl_t *db; | |
313 | int error; | |
314 | ||
428870ff | 315 | error = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
316 | if (error) |
317 | return (error); | |
318 | ||
319 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
320 | if (dn->dn_bonus == NULL) { | |
321 | rw_exit(&dn->dn_struct_rwlock); | |
322 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
323 | if (dn->dn_bonus == NULL) | |
324 | dbuf_create_bonus(dn); | |
325 | } | |
326 | db = dn->dn_bonus; | |
34dc7c2f BB |
327 | |
328 | /* as long as the bonus buf is held, the dnode will be held */ | |
572e2857 | 329 | if (refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 330 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 331 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
332 | } |
333 | ||
334 | /* | |
335 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
336 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
337 | * a dnode hold for every dbuf. | |
338 | */ | |
339 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
340 | |
341 | dnode_rele(dn, FTAG); | |
342 | ||
428870ff | 343 | VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); |
34dc7c2f BB |
344 | |
345 | *dbp = &db->db; | |
346 | return (0); | |
347 | } | |
348 | ||
428870ff BB |
349 | /* |
350 | * returns ENOENT, EIO, or 0. | |
351 | * | |
352 | * This interface will allocate a blank spill dbuf when a spill blk | |
353 | * doesn't already exist on the dnode. | |
354 | * | |
355 | * if you only want to find an already existing spill db, then | |
356 | * dmu_spill_hold_existing() should be used. | |
357 | */ | |
358 | int | |
359 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
360 | { | |
361 | dmu_buf_impl_t *db = NULL; | |
362 | int err; | |
363 | ||
364 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
365 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
366 | ||
367 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
368 | ||
369 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
370 | rw_exit(&dn->dn_struct_rwlock); | |
371 | ||
372 | ASSERT(db != NULL); | |
572e2857 BB |
373 | err = dbuf_read(db, NULL, flags); |
374 | if (err == 0) | |
375 | *dbp = &db->db; | |
376 | else | |
377 | dbuf_rele(db, tag); | |
428870ff BB |
378 | return (err); |
379 | } | |
380 | ||
381 | int | |
382 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
383 | { | |
572e2857 BB |
384 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
385 | dnode_t *dn; | |
428870ff BB |
386 | int err; |
387 | ||
572e2857 BB |
388 | DB_DNODE_ENTER(db); |
389 | dn = DB_DNODE(db); | |
390 | ||
391 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 392 | err = SET_ERROR(EINVAL); |
572e2857 BB |
393 | } else { |
394 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
395 | ||
396 | if (!dn->dn_have_spill) { | |
2e528b49 | 397 | err = SET_ERROR(ENOENT); |
572e2857 BB |
398 | } else { |
399 | err = dmu_spill_hold_by_dnode(dn, | |
400 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
401 | } | |
428870ff | 402 | |
428870ff | 403 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 404 | } |
572e2857 BB |
405 | |
406 | DB_DNODE_EXIT(db); | |
428870ff BB |
407 | return (err); |
408 | } | |
409 | ||
410 | int | |
411 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
412 | { | |
572e2857 BB |
413 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
414 | dnode_t *dn; | |
415 | int err; | |
416 | ||
417 | DB_DNODE_ENTER(db); | |
418 | dn = DB_DNODE(db); | |
419 | err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); | |
420 | DB_DNODE_EXIT(db); | |
421 | ||
422 | return (err); | |
428870ff BB |
423 | } |
424 | ||
34dc7c2f BB |
425 | /* |
426 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
427 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
428 | * and can induce severe lock contention when writing to several files | |
429 | * whose dnodes are in the same block. | |
430 | */ | |
431 | static int | |
9babb374 | 432 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
7f60329a | 433 | boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
34dc7c2f BB |
434 | { |
435 | dmu_buf_t **dbp; | |
436 | uint64_t blkid, nblks, i; | |
9babb374 | 437 | uint32_t dbuf_flags; |
34dc7c2f BB |
438 | int err; |
439 | zio_t *zio; | |
440 | ||
441 | ASSERT(length <= DMU_MAX_ACCESS); | |
442 | ||
7f60329a MA |
443 | /* |
444 | * Note: We directly notify the prefetch code of this read, so that | |
445 | * we can tell it about the multi-block read. dbuf_read() only knows | |
446 | * about the one block it is accessing. | |
447 | */ | |
448 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | | |
449 | DB_RF_NOPREFETCH; | |
34dc7c2f BB |
450 | |
451 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
452 | if (dn->dn_datablkshift) { | |
453 | int blkshift = dn->dn_datablkshift; | |
7f60329a MA |
454 | nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - |
455 | P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; | |
34dc7c2f BB |
456 | } else { |
457 | if (offset + length > dn->dn_datablksz) { | |
458 | zfs_panic_recover("zfs: accessing past end of object " | |
459 | "%llx/%llx (size=%u access=%llu+%llu)", | |
460 | (longlong_t)dn->dn_objset-> | |
461 | os_dsl_dataset->ds_object, | |
462 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
463 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 464 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 465 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
466 | } |
467 | nblks = 1; | |
468 | } | |
79c76d5b | 469 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 470 | |
b128c09f | 471 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
fcff0f35 | 472 | blkid = dbuf_whichblock(dn, 0, offset); |
34dc7c2f | 473 | for (i = 0; i < nblks; i++) { |
7f60329a | 474 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); |
34dc7c2f BB |
475 | if (db == NULL) { |
476 | rw_exit(&dn->dn_struct_rwlock); | |
477 | dmu_buf_rele_array(dbp, nblks, tag); | |
478 | zio_nowait(zio); | |
2e528b49 | 479 | return (SET_ERROR(EIO)); |
34dc7c2f | 480 | } |
7f60329a | 481 | |
34dc7c2f | 482 | /* initiate async i/o */ |
7f60329a | 483 | if (read) |
9babb374 | 484 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
485 | dbp[i] = &db->db; |
486 | } | |
7f60329a | 487 | |
755065f3 AM |
488 | if ((flags & DMU_READ_NO_PREFETCH) == 0 && |
489 | DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { | |
490 | dmu_zfetch(&dn->dn_zfetch, blkid, nblks, | |
491 | read && DNODE_IS_CACHEABLE(dn)); | |
7f60329a | 492 | } |
34dc7c2f BB |
493 | rw_exit(&dn->dn_struct_rwlock); |
494 | ||
495 | /* wait for async i/o */ | |
496 | err = zio_wait(zio); | |
497 | if (err) { | |
498 | dmu_buf_rele_array(dbp, nblks, tag); | |
499 | return (err); | |
500 | } | |
501 | ||
502 | /* wait for other io to complete */ | |
503 | if (read) { | |
504 | for (i = 0; i < nblks; i++) { | |
505 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
506 | mutex_enter(&db->db_mtx); | |
507 | while (db->db_state == DB_READ || | |
508 | db->db_state == DB_FILL) | |
509 | cv_wait(&db->db_changed, &db->db_mtx); | |
510 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 511 | err = SET_ERROR(EIO); |
34dc7c2f BB |
512 | mutex_exit(&db->db_mtx); |
513 | if (err) { | |
514 | dmu_buf_rele_array(dbp, nblks, tag); | |
515 | return (err); | |
516 | } | |
517 | } | |
518 | } | |
519 | ||
520 | *numbufsp = nblks; | |
521 | *dbpp = dbp; | |
522 | return (0); | |
523 | } | |
524 | ||
525 | static int | |
526 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
527 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
528 | { | |
529 | dnode_t *dn; | |
530 | int err; | |
531 | ||
428870ff | 532 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
533 | if (err) |
534 | return (err); | |
535 | ||
536 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 537 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
538 | |
539 | dnode_rele(dn, FTAG); | |
540 | ||
541 | return (err); | |
542 | } | |
543 | ||
544 | int | |
572e2857 | 545 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
7f60329a MA |
546 | uint64_t length, boolean_t read, void *tag, int *numbufsp, |
547 | dmu_buf_t ***dbpp) | |
34dc7c2f | 548 | { |
572e2857 BB |
549 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
550 | dnode_t *dn; | |
34dc7c2f BB |
551 | int err; |
552 | ||
572e2857 BB |
553 | DB_DNODE_ENTER(db); |
554 | dn = DB_DNODE(db); | |
34dc7c2f | 555 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 556 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 557 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
558 | |
559 | return (err); | |
560 | } | |
561 | ||
562 | void | |
563 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
564 | { | |
565 | int i; | |
566 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
567 | ||
568 | if (numbufs == 0) | |
569 | return; | |
570 | ||
571 | for (i = 0; i < numbufs; i++) { | |
572 | if (dbp[i]) | |
573 | dbuf_rele(dbp[i], tag); | |
574 | } | |
575 | ||
576 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
577 | } | |
578 | ||
e8b96c60 | 579 | /* |
fcff0f35 PD |
580 | * Issue prefetch i/os for the given blocks. If level is greater than 0, the |
581 | * indirect blocks prefeteched will be those that point to the blocks containing | |
582 | * the data starting at offset, and continuing to offset + len. | |
e8b96c60 | 583 | * |
fcff0f35 PD |
584 | * Note that if the indirect blocks above the blocks being prefetched are not in |
585 | * cache, they will be asychronously read in. | |
e8b96c60 | 586 | */ |
34dc7c2f | 587 | void |
fcff0f35 PD |
588 | dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, |
589 | uint64_t len, zio_priority_t pri) | |
34dc7c2f BB |
590 | { |
591 | dnode_t *dn; | |
592 | uint64_t blkid; | |
e8b96c60 | 593 | int nblks, err; |
34dc7c2f | 594 | |
34dc7c2f | 595 | if (len == 0) { /* they're interested in the bonus buffer */ |
572e2857 | 596 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
597 | |
598 | if (object == 0 || object >= DN_MAX_OBJECT) | |
599 | return; | |
600 | ||
601 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
602 | blkid = dbuf_whichblock(dn, level, |
603 | object * sizeof (dnode_phys_t)); | |
604 | dbuf_prefetch(dn, level, blkid, pri, 0); | |
34dc7c2f BB |
605 | rw_exit(&dn->dn_struct_rwlock); |
606 | return; | |
607 | } | |
608 | ||
609 | /* | |
610 | * XXX - Note, if the dnode for the requested object is not | |
611 | * already cached, we will do a *synchronous* read in the | |
612 | * dnode_hold() call. The same is true for any indirects. | |
613 | */ | |
428870ff | 614 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
615 | if (err != 0) |
616 | return; | |
617 | ||
618 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
fcff0f35 PD |
619 | /* |
620 | * offset + len - 1 is the last byte we want to prefetch for, and offset | |
621 | * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the | |
622 | * last block we want to prefetch, and dbuf_whichblock(dn, level, | |
623 | * offset) is the first. Then the number we need to prefetch is the | |
624 | * last - first + 1. | |
625 | */ | |
626 | if (level > 0 || dn->dn_datablkshift != 0) { | |
627 | nblks = dbuf_whichblock(dn, level, offset + len - 1) - | |
628 | dbuf_whichblock(dn, level, offset) + 1; | |
34dc7c2f BB |
629 | } else { |
630 | nblks = (offset < dn->dn_datablksz); | |
631 | } | |
632 | ||
633 | if (nblks != 0) { | |
e8b96c60 MA |
634 | int i; |
635 | ||
fcff0f35 | 636 | blkid = dbuf_whichblock(dn, level, offset); |
34dc7c2f | 637 | for (i = 0; i < nblks; i++) |
fcff0f35 | 638 | dbuf_prefetch(dn, level, blkid + i, pri, 0); |
34dc7c2f BB |
639 | } |
640 | ||
641 | rw_exit(&dn->dn_struct_rwlock); | |
642 | ||
643 | dnode_rele(dn, FTAG); | |
644 | } | |
645 | ||
45d1cae3 BB |
646 | /* |
647 | * Get the next "chunk" of file data to free. We traverse the file from | |
648 | * the end so that the file gets shorter over time (if we crashes in the | |
649 | * middle, this will leave us in a better state). We find allocated file | |
650 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
651 | * |
652 | * On input, *start should be the first offset that does not need to be | |
653 | * freed (e.g. "offset + length"). On return, *start will be the first | |
654 | * offset that should be freed. | |
45d1cae3 | 655 | */ |
b128c09f | 656 | static int |
b663a23d | 657 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) |
b128c09f | 658 | { |
b663a23d MA |
659 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
660 | /* bytes of data covered by a level-1 indirect block */ | |
45d1cae3 | 661 | uint64_t iblkrange = |
b128c09f | 662 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
b663a23d | 663 | uint64_t blks; |
b128c09f | 664 | |
b663a23d | 665 | ASSERT3U(minimum, <=, *start); |
b128c09f | 666 | |
b663a23d MA |
667 | if (*start - minimum <= iblkrange * maxblks) { |
668 | *start = minimum; | |
b128c09f BB |
669 | return (0); |
670 | } | |
45d1cae3 | 671 | ASSERT(ISP2(iblkrange)); |
b128c09f | 672 | |
b663a23d | 673 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
674 | int err; |
675 | ||
b663a23d MA |
676 | /* |
677 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
678 | * indirect block at or before the input offset. We must | |
679 | * decrement *start so that it is at the end of the region | |
680 | * to search. | |
681 | */ | |
682 | (*start)--; | |
b128c09f | 683 | err = dnode_next_offset(dn, |
45d1cae3 | 684 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 685 | |
b663a23d | 686 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 687 | if (err == ESRCH) { |
b663a23d MA |
688 | *start = minimum; |
689 | break; | |
690 | } else if (err != 0) { | |
b128c09f | 691 | return (err); |
45d1cae3 | 692 | } |
b128c09f | 693 | |
b663a23d | 694 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 695 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 696 | } |
b663a23d MA |
697 | if (*start < minimum) |
698 | *start = minimum; | |
b128c09f BB |
699 | return (0); |
700 | } | |
701 | ||
702 | static int | |
703 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
b663a23d | 704 | uint64_t length) |
b128c09f | 705 | { |
c97d3069 | 706 | uint64_t object_size; |
b663a23d MA |
707 | int err; |
708 | ||
c97d3069 BB |
709 | if (dn == NULL) |
710 | return (SET_ERROR(EINVAL)); | |
711 | ||
712 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
b663a23d | 713 | if (offset >= object_size) |
b128c09f | 714 | return (0); |
b128c09f | 715 | |
b663a23d MA |
716 | if (length == DMU_OBJECT_END || offset + length > object_size) |
717 | length = object_size - offset; | |
718 | ||
719 | while (length != 0) { | |
720 | uint64_t chunk_end, chunk_begin; | |
721 | dmu_tx_t *tx; | |
722 | ||
723 | chunk_end = chunk_begin = offset + length; | |
724 | ||
725 | /* move chunk_begin backwards to the beginning of this chunk */ | |
726 | err = get_next_chunk(dn, &chunk_begin, offset); | |
b128c09f BB |
727 | if (err) |
728 | return (err); | |
b663a23d MA |
729 | ASSERT3U(chunk_begin, >=, offset); |
730 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f BB |
731 | |
732 | tx = dmu_tx_create(os); | |
b663a23d MA |
733 | dmu_tx_hold_free(tx, dn->dn_object, |
734 | chunk_begin, chunk_end - chunk_begin); | |
19d55079 MA |
735 | |
736 | /* | |
737 | * Mark this transaction as typically resulting in a net | |
738 | * reduction in space used. | |
739 | */ | |
740 | dmu_tx_mark_netfree(tx); | |
b128c09f BB |
741 | err = dmu_tx_assign(tx, TXG_WAIT); |
742 | if (err) { | |
743 | dmu_tx_abort(tx); | |
744 | return (err); | |
745 | } | |
b663a23d | 746 | dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx); |
b128c09f | 747 | dmu_tx_commit(tx); |
b663a23d MA |
748 | |
749 | length -= chunk_end - chunk_begin; | |
b128c09f BB |
750 | } |
751 | return (0); | |
752 | } | |
753 | ||
754 | int | |
755 | dmu_free_long_range(objset_t *os, uint64_t object, | |
756 | uint64_t offset, uint64_t length) | |
757 | { | |
758 | dnode_t *dn; | |
759 | int err; | |
760 | ||
428870ff | 761 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
762 | if (err != 0) |
763 | return (err); | |
b663a23d | 764 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
765 | |
766 | /* | |
767 | * It is important to zero out the maxblkid when freeing the entire | |
768 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
769 | * will take the fast path, and (b) dnode_reallocate() can verify | |
770 | * that the entire file has been freed. | |
771 | */ | |
b0bc7a84 | 772 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
773 | dn->dn_maxblkid = 0; |
774 | ||
b128c09f BB |
775 | dnode_rele(dn, FTAG); |
776 | return (err); | |
777 | } | |
778 | ||
779 | int | |
b663a23d | 780 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 781 | { |
b128c09f BB |
782 | dmu_tx_t *tx; |
783 | int err; | |
784 | ||
b663a23d | 785 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
786 | if (err != 0) |
787 | return (err); | |
b663a23d MA |
788 | |
789 | tx = dmu_tx_create(os); | |
790 | dmu_tx_hold_bonus(tx, object); | |
791 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
19d55079 | 792 | dmu_tx_mark_netfree(tx); |
b663a23d MA |
793 | err = dmu_tx_assign(tx, TXG_WAIT); |
794 | if (err == 0) { | |
795 | err = dmu_object_free(os, object, tx); | |
796 | dmu_tx_commit(tx); | |
b128c09f | 797 | } else { |
b663a23d | 798 | dmu_tx_abort(tx); |
b128c09f | 799 | } |
b663a23d | 800 | |
b128c09f BB |
801 | return (err); |
802 | } | |
803 | ||
34dc7c2f BB |
804 | int |
805 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
806 | uint64_t size, dmu_tx_t *tx) | |
807 | { | |
808 | dnode_t *dn; | |
428870ff | 809 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
810 | if (err) |
811 | return (err); | |
812 | ASSERT(offset < UINT64_MAX); | |
813 | ASSERT(size == -1ULL || size <= UINT64_MAX - offset); | |
814 | dnode_free_range(dn, offset, size, tx); | |
815 | dnode_rele(dn, FTAG); | |
816 | return (0); | |
817 | } | |
818 | ||
819 | int | |
820 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
9babb374 | 821 | void *buf, uint32_t flags) |
34dc7c2f BB |
822 | { |
823 | dnode_t *dn; | |
824 | dmu_buf_t **dbp; | |
45d1cae3 | 825 | int numbufs, err; |
34dc7c2f | 826 | |
428870ff | 827 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
828 | if (err) |
829 | return (err); | |
830 | ||
831 | /* | |
832 | * Deal with odd block sizes, where there can't be data past the first | |
833 | * block. If we ever do the tail block optimization, we will need to | |
834 | * handle that here as well. | |
835 | */ | |
45d1cae3 | 836 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 837 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
838 | MIN(size, dn->dn_datablksz - offset); |
839 | bzero((char *)buf + newsz, size - newsz); | |
840 | size = newsz; | |
841 | } | |
842 | ||
843 | while (size > 0) { | |
844 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 845 | int i; |
34dc7c2f BB |
846 | |
847 | /* | |
848 | * NB: we could do this block-at-a-time, but it's nice | |
849 | * to be reading in parallel. | |
850 | */ | |
851 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 852 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
853 | if (err) |
854 | break; | |
855 | ||
856 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
857 | uint64_t tocpy; |
858 | int64_t bufoff; | |
34dc7c2f BB |
859 | dmu_buf_t *db = dbp[i]; |
860 | ||
861 | ASSERT(size > 0); | |
862 | ||
863 | bufoff = offset - db->db_offset; | |
c9520ecc | 864 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 865 | |
c9520ecc | 866 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
867 | |
868 | offset += tocpy; | |
869 | size -= tocpy; | |
870 | buf = (char *)buf + tocpy; | |
871 | } | |
872 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
873 | } | |
874 | dnode_rele(dn, FTAG); | |
875 | return (err); | |
876 | } | |
877 | ||
878 | void | |
879 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
880 | const void *buf, dmu_tx_t *tx) | |
881 | { | |
882 | dmu_buf_t **dbp; | |
883 | int numbufs, i; | |
884 | ||
885 | if (size == 0) | |
886 | return; | |
887 | ||
e6659763 | 888 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, |
34dc7c2f BB |
889 | FALSE, FTAG, &numbufs, &dbp)); |
890 | ||
891 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
892 | uint64_t tocpy; |
893 | int64_t bufoff; | |
34dc7c2f BB |
894 | dmu_buf_t *db = dbp[i]; |
895 | ||
896 | ASSERT(size > 0); | |
897 | ||
898 | bufoff = offset - db->db_offset; | |
c9520ecc | 899 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
900 | |
901 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
902 | ||
903 | if (tocpy == db->db_size) | |
904 | dmu_buf_will_fill(db, tx); | |
905 | else | |
906 | dmu_buf_will_dirty(db, tx); | |
907 | ||
60101509 | 908 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
909 | |
910 | if (tocpy == db->db_size) | |
911 | dmu_buf_fill_done(db, tx); | |
912 | ||
913 | offset += tocpy; | |
914 | size -= tocpy; | |
915 | buf = (char *)buf + tocpy; | |
916 | } | |
917 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
918 | } | |
919 | ||
b128c09f BB |
920 | void |
921 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
922 | dmu_tx_t *tx) | |
923 | { | |
924 | dmu_buf_t **dbp; | |
925 | int numbufs, i; | |
926 | ||
927 | if (size == 0) | |
928 | return; | |
929 | ||
930 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
931 | FALSE, FTAG, &numbufs, &dbp)); | |
932 | ||
933 | for (i = 0; i < numbufs; i++) { | |
934 | dmu_buf_t *db = dbp[i]; | |
935 | ||
936 | dmu_buf_will_not_fill(db, tx); | |
937 | } | |
938 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
939 | } | |
940 | ||
9b67f605 MA |
941 | void |
942 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
943 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
944 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
945 | { | |
946 | dmu_buf_t *db; | |
947 | ||
948 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
949 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
950 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
951 | FTAG, &db)); | |
952 | ||
953 | dmu_buf_write_embedded(db, | |
954 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
955 | uncompressed_size, compressed_size, byteorder, tx); | |
956 | ||
957 | dmu_buf_rele(db, FTAG); | |
958 | } | |
959 | ||
428870ff BB |
960 | /* |
961 | * DMU support for xuio | |
962 | */ | |
963 | kstat_t *xuio_ksp = NULL; | |
964 | ||
59e6e7ca BB |
965 | typedef struct xuio_stats { |
966 | /* loaned yet not returned arc_buf */ | |
967 | kstat_named_t xuiostat_onloan_rbuf; | |
968 | kstat_named_t xuiostat_onloan_wbuf; | |
969 | /* whether a copy is made when loaning out a read buffer */ | |
970 | kstat_named_t xuiostat_rbuf_copied; | |
971 | kstat_named_t xuiostat_rbuf_nocopy; | |
972 | /* whether a copy is made when assigning a write buffer */ | |
973 | kstat_named_t xuiostat_wbuf_copied; | |
974 | kstat_named_t xuiostat_wbuf_nocopy; | |
975 | } xuio_stats_t; | |
976 | ||
977 | static xuio_stats_t xuio_stats = { | |
978 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
979 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
980 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
981 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
982 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
983 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
984 | }; | |
985 | ||
d1d7e268 MK |
986 | #define XUIOSTAT_INCR(stat, val) \ |
987 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
988 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
59e6e7ca | 989 | |
428870ff BB |
990 | int |
991 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
992 | { | |
993 | dmu_xuio_t *priv; | |
994 | uio_t *uio = &xuio->xu_uio; | |
995 | ||
996 | uio->uio_iovcnt = nblk; | |
79c76d5b | 997 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
428870ff | 998 | |
79c76d5b | 999 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
428870ff | 1000 | priv->cnt = nblk; |
79c76d5b | 1001 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
5475aada | 1002 | priv->iovp = (iovec_t *)uio->uio_iov; |
428870ff BB |
1003 | XUIO_XUZC_PRIV(xuio) = priv; |
1004 | ||
1005 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1006 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
1007 | else | |
1008 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
1009 | ||
1010 | return (0); | |
1011 | } | |
1012 | ||
1013 | void | |
1014 | dmu_xuio_fini(xuio_t *xuio) | |
1015 | { | |
1016 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1017 | int nblk = priv->cnt; | |
1018 | ||
1019 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
1020 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
1021 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
1022 | ||
1023 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
1024 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
1025 | else | |
1026 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
1031 | * and increase priv->next by 1. | |
1032 | */ | |
1033 | int | |
1034 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
1035 | { | |
1036 | struct iovec *iov; | |
1037 | uio_t *uio = &xuio->xu_uio; | |
1038 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1039 | int i = priv->next++; | |
1040 | ||
1041 | ASSERT(i < priv->cnt); | |
2aa34383 | 1042 | ASSERT(off + n <= arc_buf_lsize(abuf)); |
5475aada | 1043 | iov = (iovec_t *)uio->uio_iov + i; |
428870ff BB |
1044 | iov->iov_base = (char *)abuf->b_data + off; |
1045 | iov->iov_len = n; | |
1046 | priv->bufs[i] = abuf; | |
1047 | return (0); | |
1048 | } | |
1049 | ||
1050 | int | |
1051 | dmu_xuio_cnt(xuio_t *xuio) | |
1052 | { | |
1053 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1054 | return (priv->cnt); | |
1055 | } | |
1056 | ||
1057 | arc_buf_t * | |
1058 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
1059 | { | |
1060 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1061 | ||
1062 | ASSERT(i < priv->cnt); | |
1063 | return (priv->bufs[i]); | |
1064 | } | |
1065 | ||
1066 | void | |
1067 | dmu_xuio_clear(xuio_t *xuio, int i) | |
1068 | { | |
1069 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1070 | ||
1071 | ASSERT(i < priv->cnt); | |
1072 | priv->bufs[i] = NULL; | |
1073 | } | |
1074 | ||
1075 | static void | |
1076 | xuio_stat_init(void) | |
1077 | { | |
1078 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1079 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1080 | KSTAT_FLAG_VIRTUAL); | |
1081 | if (xuio_ksp != NULL) { | |
1082 | xuio_ksp->ks_data = &xuio_stats; | |
1083 | kstat_install(xuio_ksp); | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | static void | |
1088 | xuio_stat_fini(void) | |
1089 | { | |
1090 | if (xuio_ksp != NULL) { | |
1091 | kstat_delete(xuio_ksp); | |
1092 | xuio_ksp = NULL; | |
1093 | } | |
1094 | } | |
1095 | ||
1096 | void | |
1097 | xuio_stat_wbuf_copied() | |
1098 | { | |
1099 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1100 | } | |
1101 | ||
1102 | void | |
1103 | xuio_stat_wbuf_nocopy() | |
1104 | { | |
1105 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1106 | } | |
1107 | ||
34dc7c2f | 1108 | #ifdef _KERNEL |
804e0504 MA |
1109 | static int |
1110 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) | |
872e8d26 BB |
1111 | { |
1112 | dmu_buf_t **dbp; | |
1113 | int numbufs, i, err; | |
1114 | xuio_t *xuio = NULL; | |
1115 | ||
1116 | /* | |
1117 | * NB: we could do this block-at-a-time, but it's nice | |
1118 | * to be reading in parallel. | |
1119 | */ | |
804e0504 MA |
1120 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, |
1121 | TRUE, FTAG, &numbufs, &dbp, 0); | |
872e8d26 BB |
1122 | if (err) |
1123 | return (err); | |
1124 | ||
1125 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1126 | uint64_t tocpy; |
1127 | int64_t bufoff; | |
872e8d26 BB |
1128 | dmu_buf_t *db = dbp[i]; |
1129 | ||
1130 | ASSERT(size > 0); | |
1131 | ||
1132 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1133 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1134 | |
1135 | if (xuio) { | |
1136 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1137 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1138 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1139 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1140 | if (!err) { | |
1141 | uio->uio_resid -= tocpy; | |
1142 | uio->uio_loffset += tocpy; | |
1143 | } | |
1144 | ||
1145 | if (abuf == dbuf_abuf) | |
1146 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1147 | else | |
1148 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
1149 | } else { | |
1150 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1151 | UIO_READ, uio); | |
1152 | } | |
1153 | if (err) | |
1154 | break; | |
1155 | ||
1156 | size -= tocpy; | |
1157 | } | |
1158 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1159 | ||
1160 | return (err); | |
1161 | } | |
1162 | ||
804e0504 MA |
1163 | /* |
1164 | * Read 'size' bytes into the uio buffer. | |
1165 | * From object zdb->db_object. | |
1166 | * Starting at offset uio->uio_loffset. | |
1167 | * | |
1168 | * If the caller already has a dbuf in the target object | |
1169 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1170 | * because we don't have to find the dnode_t for the object. | |
1171 | */ | |
1172 | int | |
1173 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1174 | { | |
1175 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1176 | dnode_t *dn; | |
1177 | int err; | |
1178 | ||
1179 | if (size == 0) | |
1180 | return (0); | |
1181 | ||
1182 | DB_DNODE_ENTER(db); | |
1183 | dn = DB_DNODE(db); | |
1184 | err = dmu_read_uio_dnode(dn, uio, size); | |
1185 | DB_DNODE_EXIT(db); | |
1186 | ||
1187 | return (err); | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * Read 'size' bytes into the uio buffer. | |
1192 | * From the specified object | |
1193 | * Starting at offset uio->uio_loffset. | |
1194 | */ | |
1195 | int | |
1196 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1197 | { | |
1198 | dnode_t *dn; | |
1199 | int err; | |
1200 | ||
1201 | if (size == 0) | |
1202 | return (0); | |
1203 | ||
1204 | err = dnode_hold(os, object, FTAG, &dn); | |
1205 | if (err) | |
1206 | return (err); | |
1207 | ||
1208 | err = dmu_read_uio_dnode(dn, uio, size); | |
1209 | ||
1210 | dnode_rele(dn, FTAG); | |
1211 | ||
1212 | return (err); | |
1213 | } | |
1214 | ||
872e8d26 BB |
1215 | static int |
1216 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) | |
1217 | { | |
1218 | dmu_buf_t **dbp; | |
1219 | int numbufs; | |
1220 | int err = 0; | |
1221 | int i; | |
1222 | ||
1223 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1224 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1225 | if (err) | |
1226 | return (err); | |
1227 | ||
1228 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1229 | uint64_t tocpy; |
1230 | int64_t bufoff; | |
872e8d26 BB |
1231 | dmu_buf_t *db = dbp[i]; |
1232 | ||
1233 | ASSERT(size > 0); | |
1234 | ||
1235 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1236 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1237 | |
1238 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1239 | ||
1240 | if (tocpy == db->db_size) | |
1241 | dmu_buf_will_fill(db, tx); | |
1242 | else | |
1243 | dmu_buf_will_dirty(db, tx); | |
1244 | ||
1245 | /* | |
1246 | * XXX uiomove could block forever (eg.nfs-backed | |
1247 | * pages). There needs to be a uiolockdown() function | |
1248 | * to lock the pages in memory, so that uiomove won't | |
1249 | * block. | |
1250 | */ | |
1251 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1252 | UIO_WRITE, uio); | |
1253 | ||
1254 | if (tocpy == db->db_size) | |
1255 | dmu_buf_fill_done(db, tx); | |
1256 | ||
1257 | if (err) | |
1258 | break; | |
1259 | ||
1260 | size -= tocpy; | |
1261 | } | |
1262 | ||
1263 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1264 | return (err); | |
1265 | } | |
1266 | ||
804e0504 MA |
1267 | /* |
1268 | * Write 'size' bytes from the uio buffer. | |
1269 | * To object zdb->db_object. | |
1270 | * Starting at offset uio->uio_loffset. | |
1271 | * | |
1272 | * If the caller already has a dbuf in the target object | |
1273 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1274 | * because we don't have to find the dnode_t for the object. | |
1275 | */ | |
428870ff BB |
1276 | int |
1277 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1278 | dmu_tx_t *tx) | |
1279 | { | |
572e2857 BB |
1280 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1281 | dnode_t *dn; | |
1282 | int err; | |
1283 | ||
428870ff BB |
1284 | if (size == 0) |
1285 | return (0); | |
1286 | ||
572e2857 BB |
1287 | DB_DNODE_ENTER(db); |
1288 | dn = DB_DNODE(db); | |
1289 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1290 | DB_DNODE_EXIT(db); | |
1291 | ||
1292 | return (err); | |
428870ff BB |
1293 | } |
1294 | ||
804e0504 MA |
1295 | /* |
1296 | * Write 'size' bytes from the uio buffer. | |
1297 | * To the specified object. | |
1298 | * Starting at offset uio->uio_loffset. | |
1299 | */ | |
428870ff BB |
1300 | int |
1301 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1302 | dmu_tx_t *tx) | |
1303 | { | |
1304 | dnode_t *dn; | |
1305 | int err; | |
1306 | ||
1307 | if (size == 0) | |
1308 | return (0); | |
1309 | ||
1310 | err = dnode_hold(os, object, FTAG, &dn); | |
1311 | if (err) | |
1312 | return (err); | |
1313 | ||
1314 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1315 | ||
1316 | dnode_rele(dn, FTAG); | |
1317 | ||
1318 | return (err); | |
1319 | } | |
872e8d26 | 1320 | #endif /* _KERNEL */ |
34dc7c2f | 1321 | |
9babb374 BB |
1322 | /* |
1323 | * Allocate a loaned anonymous arc buffer. | |
1324 | */ | |
1325 | arc_buf_t * | |
1326 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1327 | { | |
572e2857 | 1328 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1329 | |
2aa34383 | 1330 | return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); |
9babb374 BB |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * Free a loaned arc buffer. | |
1335 | */ | |
1336 | void | |
1337 | dmu_return_arcbuf(arc_buf_t *buf) | |
1338 | { | |
1339 | arc_return_buf(buf, FTAG); | |
d3c2ae1c | 1340 | arc_buf_destroy(buf, FTAG); |
9babb374 BB |
1341 | } |
1342 | ||
1343 | /* | |
1344 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1345 | * If this is not possible copy the contents of passed arc buf via | |
1346 | * dmu_write(). | |
1347 | */ | |
1348 | void | |
1349 | dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, | |
1350 | dmu_tx_t *tx) | |
1351 | { | |
572e2857 BB |
1352 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1353 | dnode_t *dn; | |
9babb374 | 1354 | dmu_buf_impl_t *db; |
2aa34383 | 1355 | uint32_t blksz = (uint32_t)arc_buf_lsize(buf); |
9babb374 BB |
1356 | uint64_t blkid; |
1357 | ||
572e2857 BB |
1358 | DB_DNODE_ENTER(dbuf); |
1359 | dn = DB_DNODE(dbuf); | |
9babb374 | 1360 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
fcff0f35 | 1361 | blkid = dbuf_whichblock(dn, 0, offset); |
9babb374 BB |
1362 | VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); |
1363 | rw_exit(&dn->dn_struct_rwlock); | |
572e2857 | 1364 | DB_DNODE_EXIT(dbuf); |
9babb374 | 1365 | |
88904bb3 MA |
1366 | /* |
1367 | * We can only assign if the offset is aligned, the arc buf is the | |
2aa34383 | 1368 | * same size as the dbuf, and the dbuf is not metadata. |
88904bb3 | 1369 | */ |
2aa34383 | 1370 | if (offset == db->db.db_offset && blksz == db->db.db_size) { |
9babb374 BB |
1371 | dbuf_assign_arcbuf(db, buf, tx); |
1372 | dbuf_rele(db, FTAG); | |
1373 | } else { | |
572e2857 BB |
1374 | objset_t *os; |
1375 | uint64_t object; | |
1376 | ||
2aa34383 DK |
1377 | /* compressed bufs must always be assignable to their dbuf */ |
1378 | ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); | |
524b4217 | 1379 | ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); |
2aa34383 | 1380 | |
572e2857 BB |
1381 | DB_DNODE_ENTER(dbuf); |
1382 | dn = DB_DNODE(dbuf); | |
1383 | os = dn->dn_objset; | |
1384 | object = dn->dn_object; | |
1385 | DB_DNODE_EXIT(dbuf); | |
1386 | ||
9babb374 | 1387 | dbuf_rele(db, FTAG); |
572e2857 | 1388 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1389 | dmu_return_arcbuf(buf); |
428870ff | 1390 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 BB |
1391 | } |
1392 | } | |
1393 | ||
34dc7c2f | 1394 | typedef struct { |
428870ff BB |
1395 | dbuf_dirty_record_t *dsa_dr; |
1396 | dmu_sync_cb_t *dsa_done; | |
1397 | zgd_t *dsa_zgd; | |
1398 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1399 | } dmu_sync_arg_t; |
1400 | ||
b128c09f BB |
1401 | /* ARGSUSED */ |
1402 | static void | |
1403 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1404 | { | |
428870ff BB |
1405 | dmu_sync_arg_t *dsa = varg; |
1406 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1407 | blkptr_t *bp = zio->io_bp; |
1408 | ||
428870ff BB |
1409 | if (zio->io_error == 0) { |
1410 | if (BP_IS_HOLE(bp)) { | |
1411 | /* | |
1412 | * A block of zeros may compress to a hole, but the | |
1413 | * block size still needs to be known for replay. | |
1414 | */ | |
1415 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1416 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff BB |
1417 | ASSERT(BP_GET_LEVEL(bp) == 0); |
1418 | bp->blk_fill = 1; | |
1419 | } | |
b128c09f BB |
1420 | } |
1421 | } | |
1422 | ||
428870ff BB |
1423 | static void |
1424 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1425 | { | |
1426 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1427 | } | |
1428 | ||
34dc7c2f BB |
1429 | /* ARGSUSED */ |
1430 | static void | |
1431 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1432 | { | |
428870ff BB |
1433 | dmu_sync_arg_t *dsa = varg; |
1434 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1435 | dmu_buf_impl_t *db = dr->dr_dbuf; |
34dc7c2f | 1436 | |
34dc7c2f BB |
1437 | mutex_enter(&db->db_mtx); |
1438 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1439 | if (zio->io_error == 0) { |
03c6040b GW |
1440 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1441 | if (dr->dt.dl.dr_nopwrite) { | |
1442 | ASSERTV(blkptr_t *bp = zio->io_bp); | |
1443 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); | |
1444 | ASSERTV(uint8_t chksum = BP_GET_CHECKSUM(bp_orig)); | |
1445 | ||
1446 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
1447 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); | |
3c67d83a TH |
1448 | ASSERT(zio_checksum_table[chksum].ci_flags & |
1449 | ZCHECKSUM_FLAG_NOPWRITE); | |
03c6040b | 1450 | } |
428870ff BB |
1451 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1452 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1453 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1454 | |
1455 | /* | |
1456 | * Old style holes are filled with all zeros, whereas | |
1457 | * new-style holes maintain their lsize, type, level, | |
1458 | * and birth time (see zio_write_compress). While we | |
1459 | * need to reset the BP_SET_LSIZE() call that happened | |
1460 | * in dmu_sync_ready for old style holes, we do *not* | |
1461 | * want to wipe out the information contained in new | |
1462 | * style holes. Thus, only zero out the block pointer if | |
1463 | * it's an old style hole. | |
1464 | */ | |
1465 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1466 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1467 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1468 | } else { | |
1469 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1470 | } | |
34dc7c2f BB |
1471 | cv_broadcast(&db->db_changed); |
1472 | mutex_exit(&db->db_mtx); | |
1473 | ||
428870ff | 1474 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1475 | |
428870ff BB |
1476 | kmem_free(dsa, sizeof (*dsa)); |
1477 | } | |
1478 | ||
1479 | static void | |
1480 | dmu_sync_late_arrival_done(zio_t *zio) | |
1481 | { | |
1482 | blkptr_t *bp = zio->io_bp; | |
1483 | dmu_sync_arg_t *dsa = zio->io_private; | |
03c6040b | 1484 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); |
428870ff BB |
1485 | |
1486 | if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { | |
03c6040b GW |
1487 | /* |
1488 | * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) | |
1489 | * then there is nothing to do here. Otherwise, free the | |
1490 | * newly allocated block in this txg. | |
1491 | */ | |
1492 | if (zio->io_flags & ZIO_FLAG_NOPWRITE) { | |
1493 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
1494 | } else { | |
1495 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1496 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1497 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1498 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1499 | } | |
428870ff BB |
1500 | } |
1501 | ||
1502 | dmu_tx_commit(dsa->dsa_tx); | |
1503 | ||
1504 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1505 | ||
1506 | kmem_free(dsa, sizeof (*dsa)); | |
1507 | } | |
1508 | ||
1509 | static int | |
1510 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1511 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1512 | { |
1513 | dmu_sync_arg_t *dsa; | |
1514 | dmu_tx_t *tx; | |
1515 | ||
1516 | tx = dmu_tx_create(os); | |
1517 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1518 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1519 | dmu_tx_abort(tx); | |
2e528b49 MA |
1520 | /* Make zl_get_data do txg_waited_synced() */ |
1521 | return (SET_ERROR(EIO)); | |
428870ff BB |
1522 | } |
1523 | ||
79c76d5b | 1524 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1525 | dsa->dsa_dr = NULL; |
1526 | dsa->dsa_done = done; | |
1527 | dsa->dsa_zgd = zgd; | |
1528 | dsa->dsa_tx = tx; | |
1529 | ||
bc77ba73 PD |
1530 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), |
1531 | zgd->zgd_bp, zgd->zgd_db->db_data, zgd->zgd_db->db_size, | |
2aa34383 | 1532 | zgd->zgd_db->db_size, zp, dmu_sync_late_arrival_ready, NULL, |
bc77ba73 PD |
1533 | NULL, dmu_sync_late_arrival_done, dsa, ZIO_PRIORITY_SYNC_WRITE, |
1534 | ZIO_FLAG_CANFAIL, zb)); | |
428870ff BB |
1535 | |
1536 | return (0); | |
34dc7c2f BB |
1537 | } |
1538 | ||
1539 | /* | |
1540 | * Intent log support: sync the block associated with db to disk. | |
1541 | * N.B. and XXX: the caller is responsible for making sure that the | |
1542 | * data isn't changing while dmu_sync() is writing it. | |
1543 | * | |
1544 | * Return values: | |
1545 | * | |
03c6040b | 1546 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1547 | * The caller should not log the write. |
1548 | * | |
1549 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1550 | * The caller should not log the write. | |
1551 | * | |
1552 | * EALREADY: this block is already in the process of being synced. | |
1553 | * The caller should track its progress (somehow). | |
1554 | * | |
428870ff BB |
1555 | * EIO: could not do the I/O. |
1556 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1557 | * |
428870ff BB |
1558 | * 0: the I/O has been initiated. |
1559 | * The caller should log this blkptr in the done callback. | |
1560 | * It is possible that the I/O will fail, in which case | |
1561 | * the error will be reported to the done callback and | |
1562 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1563 | */ |
1564 | int | |
428870ff | 1565 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1566 | { |
428870ff BB |
1567 | blkptr_t *bp = zgd->zgd_bp; |
1568 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; | |
1569 | objset_t *os = db->db_objset; | |
1570 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1571 | dbuf_dirty_record_t *dr; |
428870ff | 1572 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1573 | zbookmark_phys_t zb; |
428870ff | 1574 | zio_prop_t zp; |
572e2857 | 1575 | dnode_t *dn; |
34dc7c2f | 1576 | |
428870ff | 1577 | ASSERT(pio != NULL); |
34dc7c2f BB |
1578 | ASSERT(txg != 0); |
1579 | ||
428870ff BB |
1580 | SET_BOOKMARK(&zb, ds->ds_object, |
1581 | db->db.db_object, db->db_level, db->db_blkid); | |
1582 | ||
572e2857 BB |
1583 | DB_DNODE_ENTER(db); |
1584 | dn = DB_DNODE(db); | |
2aa34383 DK |
1585 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, |
1586 | ZIO_COMPRESS_INHERIT, &zp); | |
572e2857 | 1587 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1588 | |
1589 | /* | |
428870ff | 1590 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1591 | */ |
428870ff BB |
1592 | if (txg > spa_freeze_txg(os->os_spa)) |
1593 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1594 | |
1595 | /* | |
428870ff BB |
1596 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1597 | * and us. If we determine that this txg is not yet syncing, | |
1598 | * but it begins to sync a moment later, that's OK because the | |
1599 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1600 | */ |
428870ff BB |
1601 | mutex_enter(&db->db_mtx); |
1602 | ||
1603 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1604 | /* |
428870ff | 1605 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1606 | */ |
428870ff | 1607 | mutex_exit(&db->db_mtx); |
2e528b49 | 1608 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1609 | } |
1610 | ||
428870ff BB |
1611 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1612 | /* | |
1613 | * This txg is currently syncing, so we can't mess with | |
1614 | * the dirty record anymore; just write a new log block. | |
1615 | */ | |
1616 | mutex_exit(&db->db_mtx); | |
1617 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1618 | } |
1619 | ||
1620 | dr = db->db_last_dirty; | |
428870ff | 1621 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 1622 | dr = dr->dr_next; |
428870ff BB |
1623 | |
1624 | if (dr == NULL) { | |
34dc7c2f | 1625 | /* |
428870ff | 1626 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1627 | * There's no need to log writes to freed blocks, so we're done. |
1628 | */ | |
1629 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1630 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1631 | } |
1632 | ||
03c6040b GW |
1633 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); |
1634 | ||
1635 | /* | |
f3c517d8 MA |
1636 | * Assume the on-disk data is X, the current syncing data (in |
1637 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1638 | * in dmu_sync). | |
1639 | * | |
1640 | * We usually want to perform a nopwrite if X and Z are the | |
1641 | * same. However, if Y is different (i.e. the BP is going to | |
1642 | * change before this write takes effect), then a nopwrite will | |
1643 | * be incorrect - we would override with X, which could have | |
1644 | * been freed when Y was written. | |
1645 | * | |
1646 | * (Note that this is not a concern when we are nop-writing from | |
1647 | * syncing context, because X and Y must be identical, because | |
1648 | * all previous txgs have been synced.) | |
1649 | * | |
1650 | * Therefore, we disable nopwrite if the current BP could change | |
1651 | * before this TXG. There are two ways it could change: by | |
1652 | * being dirty (dr_next is non-NULL), or by being freed | |
1653 | * (dnode_block_freed()). This behavior is verified by | |
1654 | * zio_done(), which VERIFYs that the override BP is identical | |
1655 | * to the on-disk BP. | |
03c6040b | 1656 | */ |
f3c517d8 MA |
1657 | DB_DNODE_ENTER(db); |
1658 | dn = DB_DNODE(db); | |
1659 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
03c6040b | 1660 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1661 | DB_DNODE_EXIT(db); |
03c6040b | 1662 | |
34dc7c2f | 1663 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1664 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1665 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1666 | /* |
428870ff BB |
1667 | * We have already issued a sync write for this buffer, |
1668 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1669 | * have been dirtied since, or we would have cleared the state. |
1670 | */ | |
34dc7c2f | 1671 | mutex_exit(&db->db_mtx); |
2e528b49 | 1672 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1673 | } |
1674 | ||
428870ff | 1675 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1676 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1677 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1678 | |
79c76d5b | 1679 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1680 | dsa->dsa_dr = dr; |
1681 | dsa->dsa_done = done; | |
1682 | dsa->dsa_zgd = zgd; | |
1683 | dsa->dsa_tx = NULL; | |
b128c09f | 1684 | |
428870ff | 1685 | zio_nowait(arc_write(pio, os->os_spa, txg, |
3a17a7a9 | 1686 | bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
d3c2ae1c | 1687 | &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, |
bc77ba73 | 1688 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
b128c09f | 1689 | |
428870ff | 1690 | return (0); |
34dc7c2f BB |
1691 | } |
1692 | ||
1693 | int | |
1694 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
1695 | dmu_tx_t *tx) | |
1696 | { | |
1697 | dnode_t *dn; | |
1698 | int err; | |
1699 | ||
428870ff | 1700 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1701 | if (err) |
1702 | return (err); | |
1703 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1704 | dnode_rele(dn, FTAG); | |
1705 | return (err); | |
1706 | } | |
1707 | ||
1708 | void | |
1709 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
1710 | dmu_tx_t *tx) | |
1711 | { | |
1712 | dnode_t *dn; | |
1713 | ||
9b67f605 MA |
1714 | /* |
1715 | * Send streams include each object's checksum function. This | |
1716 | * check ensures that the receiving system can understand the | |
1717 | * checksum function transmitted. | |
1718 | */ | |
1719 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1720 | ||
1721 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1722 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1723 | dn->dn_checksum = checksum; |
1724 | dnode_setdirty(dn, tx); | |
1725 | dnode_rele(dn, FTAG); | |
1726 | } | |
1727 | ||
1728 | void | |
1729 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
1730 | dmu_tx_t *tx) | |
1731 | { | |
1732 | dnode_t *dn; | |
1733 | ||
9b67f605 MA |
1734 | /* |
1735 | * Send streams include each object's compression function. This | |
1736 | * check ensures that the receiving system can understand the | |
1737 | * compression function transmitted. | |
1738 | */ | |
1739 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1740 | ||
1741 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
1742 | dn->dn_compress = compress; |
1743 | dnode_setdirty(dn, tx); | |
1744 | dnode_rele(dn, FTAG); | |
1745 | } | |
1746 | ||
428870ff BB |
1747 | int zfs_mdcomp_disable = 0; |
1748 | ||
faf0f58c MA |
1749 | /* |
1750 | * When the "redundant_metadata" property is set to "most", only indirect | |
1751 | * blocks of this level and higher will have an additional ditto block. | |
1752 | */ | |
1753 | int zfs_redundant_metadata_most_ditto_level = 2; | |
1754 | ||
428870ff | 1755 | void |
2aa34383 DK |
1756 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, |
1757 | enum zio_compress override_compress, zio_prop_t *zp) | |
428870ff BB |
1758 | { |
1759 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 1760 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 1761 | (wp & WP_SPILL)); |
428870ff BB |
1762 | enum zio_checksum checksum = os->os_checksum; |
1763 | enum zio_compress compress = os->os_compress; | |
1764 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
03c6040b GW |
1765 | boolean_t dedup = B_FALSE; |
1766 | boolean_t nopwrite = B_FALSE; | |
428870ff BB |
1767 | boolean_t dedup_verify = os->os_dedup_verify; |
1768 | int copies = os->os_copies; | |
1769 | ||
1770 | /* | |
03c6040b GW |
1771 | * We maintain different write policies for each of the following |
1772 | * types of data: | |
1773 | * 1. metadata | |
1774 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
1775 | * 3. all other level 0 blocks | |
428870ff BB |
1776 | */ |
1777 | if (ismd) { | |
62bdd5eb DL |
1778 | if (zfs_mdcomp_disable) { |
1779 | compress = ZIO_COMPRESS_EMPTY; | |
62bdd5eb | 1780 | } else { |
99197f03 JG |
1781 | /* |
1782 | * XXX -- we should design a compression algorithm | |
1783 | * that specializes in arrays of bps. | |
1784 | */ | |
1785 | compress = zio_compress_select(os->os_spa, | |
1786 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
62bdd5eb | 1787 | } |
03c6040b | 1788 | |
428870ff BB |
1789 | /* |
1790 | * Metadata always gets checksummed. If the data | |
1791 | * checksum is multi-bit correctable, and it's not a | |
1792 | * ZBT-style checksum, then it's suitable for metadata | |
1793 | * as well. Otherwise, the metadata checksum defaults | |
1794 | * to fletcher4. | |
1795 | */ | |
3c67d83a TH |
1796 | if (!(zio_checksum_table[checksum].ci_flags & |
1797 | ZCHECKSUM_FLAG_METADATA) || | |
1798 | (zio_checksum_table[checksum].ci_flags & | |
1799 | ZCHECKSUM_FLAG_EMBEDDED)) | |
428870ff | 1800 | checksum = ZIO_CHECKSUM_FLETCHER_4; |
faf0f58c MA |
1801 | |
1802 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
1803 | (os->os_redundant_metadata == | |
1804 | ZFS_REDUNDANT_METADATA_MOST && | |
1805 | (level >= zfs_redundant_metadata_most_ditto_level || | |
1806 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
1807 | copies++; | |
03c6040b GW |
1808 | } else if (wp & WP_NOFILL) { |
1809 | ASSERT(level == 0); | |
428870ff | 1810 | |
428870ff | 1811 | /* |
03c6040b GW |
1812 | * If we're writing preallocated blocks, we aren't actually |
1813 | * writing them so don't set any policy properties. These | |
1814 | * blocks are currently only used by an external subsystem | |
1815 | * outside of zfs (i.e. dump) and not written by the zio | |
1816 | * pipeline. | |
428870ff | 1817 | */ |
03c6040b GW |
1818 | compress = ZIO_COMPRESS_OFF; |
1819 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff | 1820 | } else { |
99197f03 JG |
1821 | compress = zio_compress_select(os->os_spa, dn->dn_compress, |
1822 | compress); | |
428870ff | 1823 | |
03c6040b GW |
1824 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
1825 | zio_checksum_select(dn->dn_checksum, checksum) : | |
1826 | dedup_checksum; | |
428870ff | 1827 | |
03c6040b GW |
1828 | /* |
1829 | * Determine dedup setting. If we are in dmu_sync(), | |
1830 | * we won't actually dedup now because that's all | |
1831 | * done in syncing context; but we do want to use the | |
1832 | * dedup checkum. If the checksum is not strong | |
1833 | * enough to ensure unique signatures, force | |
1834 | * dedup_verify. | |
1835 | */ | |
1836 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
1837 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
3c67d83a TH |
1838 | if (!(zio_checksum_table[checksum].ci_flags & |
1839 | ZCHECKSUM_FLAG_DEDUP)) | |
03c6040b GW |
1840 | dedup_verify = B_TRUE; |
1841 | } | |
428870ff | 1842 | |
03c6040b | 1843 | /* |
3c67d83a TH |
1844 | * Enable nopwrite if we have secure enough checksum |
1845 | * algorithm (see comment in zio_nop_write) and | |
1846 | * compression is enabled. We don't enable nopwrite if | |
1847 | * dedup is enabled as the two features are mutually | |
1848 | * exclusive. | |
03c6040b | 1849 | */ |
3c67d83a TH |
1850 | nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & |
1851 | ZCHECKSUM_FLAG_NOPWRITE) && | |
03c6040b | 1852 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); |
428870ff BB |
1853 | } |
1854 | ||
1855 | zp->zp_checksum = checksum; | |
2aa34383 DK |
1856 | |
1857 | /* | |
1858 | * If we're writing a pre-compressed buffer, the compression type we use | |
1859 | * must match the data. If it hasn't been compressed yet, then we should | |
1860 | * use the value dictated by the policies above. | |
1861 | */ | |
1862 | zp->zp_compress = override_compress != ZIO_COMPRESS_INHERIT | |
1863 | ? override_compress : compress; | |
1864 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); | |
1865 | ||
428870ff BB |
1866 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; |
1867 | zp->zp_level = level; | |
faf0f58c | 1868 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
1869 | zp->zp_dedup = dedup; |
1870 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 1871 | zp->zp_nopwrite = nopwrite; |
428870ff BB |
1872 | } |
1873 | ||
34dc7c2f BB |
1874 | int |
1875 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
1876 | { | |
1877 | dnode_t *dn; | |
1878 | int i, err; | |
1879 | ||
428870ff | 1880 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1881 | if (err) |
1882 | return (err); | |
1883 | /* | |
1884 | * Sync any current changes before | |
1885 | * we go trundling through the block pointers. | |
1886 | */ | |
1887 | for (i = 0; i < TXG_SIZE; i++) { | |
1888 | if (list_link_active(&dn->dn_dirty_link[i])) | |
1889 | break; | |
1890 | } | |
1891 | if (i != TXG_SIZE) { | |
1892 | dnode_rele(dn, FTAG); | |
1893 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 1894 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1895 | if (err) |
1896 | return (err); | |
1897 | } | |
1898 | ||
b128c09f | 1899 | err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); |
34dc7c2f BB |
1900 | dnode_rele(dn, FTAG); |
1901 | ||
1902 | return (err); | |
1903 | } | |
1904 | ||
1905 | void | |
e0b0ca98 | 1906 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 1907 | { |
e0b0ca98 | 1908 | dnode_phys_t *dnp = dn->dn_phys; |
d6320ddb | 1909 | int i; |
428870ff | 1910 | |
34dc7c2f BB |
1911 | doi->doi_data_block_size = dn->dn_datablksz; |
1912 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
1913 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
1914 | doi->doi_type = dn->dn_type; |
1915 | doi->doi_bonus_type = dn->dn_bonustype; | |
1916 | doi->doi_bonus_size = dn->dn_bonuslen; | |
50c957f7 | 1917 | doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; |
34dc7c2f BB |
1918 | doi->doi_indirection = dn->dn_nlevels; |
1919 | doi->doi_checksum = dn->dn_checksum; | |
1920 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 1921 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 1922 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 1923 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 1924 | doi->doi_fill_count = 0; |
d6320ddb | 1925 | for (i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 1926 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
1927 | } |
1928 | ||
1929 | void | |
1930 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
1931 | { | |
1932 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
1933 | mutex_enter(&dn->dn_mtx); | |
1934 | ||
1935 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
1936 | |
1937 | mutex_exit(&dn->dn_mtx); | |
1938 | rw_exit(&dn->dn_struct_rwlock); | |
1939 | } | |
1940 | ||
1941 | /* | |
1942 | * Get information on a DMU object. | |
1943 | * If doi is NULL, just indicates whether the object exists. | |
1944 | */ | |
1945 | int | |
1946 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
1947 | { | |
1948 | dnode_t *dn; | |
428870ff | 1949 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1950 | |
1951 | if (err) | |
1952 | return (err); | |
1953 | ||
1954 | if (doi != NULL) | |
1955 | dmu_object_info_from_dnode(dn, doi); | |
1956 | ||
1957 | dnode_rele(dn, FTAG); | |
1958 | return (0); | |
1959 | } | |
1960 | ||
1961 | /* | |
1962 | * As above, but faster; can be used when you have a held dbuf in hand. | |
1963 | */ | |
1964 | void | |
572e2857 | 1965 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 1966 | { |
572e2857 BB |
1967 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1968 | ||
1969 | DB_DNODE_ENTER(db); | |
1970 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
1971 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
1972 | } |
1973 | ||
1974 | /* | |
1975 | * Faster still when you only care about the size. | |
1976 | * This is specifically optimized for zfs_getattr(). | |
1977 | */ | |
1978 | void | |
572e2857 BB |
1979 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
1980 | u_longlong_t *nblk512) | |
34dc7c2f | 1981 | { |
572e2857 BB |
1982 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1983 | dnode_t *dn; | |
1984 | ||
1985 | DB_DNODE_ENTER(db); | |
1986 | dn = DB_DNODE(db); | |
34dc7c2f BB |
1987 | |
1988 | *blksize = dn->dn_datablksz; | |
50c957f7 | 1989 | /* add in number of slots used for the dnode itself */ |
34dc7c2f | 1990 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> |
50c957f7 NB |
1991 | SPA_MINBLOCKSHIFT) + dn->dn_num_slots; |
1992 | DB_DNODE_EXIT(db); | |
1993 | } | |
1994 | ||
1995 | void | |
1996 | dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) | |
1997 | { | |
1998 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
1999 | dnode_t *dn; | |
2000 | ||
2001 | DB_DNODE_ENTER(db); | |
2002 | dn = DB_DNODE(db); | |
2003 | *dnsize = dn->dn_num_slots << DNODE_SHIFT; | |
572e2857 | 2004 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2005 | } |
2006 | ||
2007 | void | |
2008 | byteswap_uint64_array(void *vbuf, size_t size) | |
2009 | { | |
2010 | uint64_t *buf = vbuf; | |
2011 | size_t count = size >> 3; | |
2012 | int i; | |
2013 | ||
2014 | ASSERT((size & 7) == 0); | |
2015 | ||
2016 | for (i = 0; i < count; i++) | |
2017 | buf[i] = BSWAP_64(buf[i]); | |
2018 | } | |
2019 | ||
2020 | void | |
2021 | byteswap_uint32_array(void *vbuf, size_t size) | |
2022 | { | |
2023 | uint32_t *buf = vbuf; | |
2024 | size_t count = size >> 2; | |
2025 | int i; | |
2026 | ||
2027 | ASSERT((size & 3) == 0); | |
2028 | ||
2029 | for (i = 0; i < count; i++) | |
2030 | buf[i] = BSWAP_32(buf[i]); | |
2031 | } | |
2032 | ||
2033 | void | |
2034 | byteswap_uint16_array(void *vbuf, size_t size) | |
2035 | { | |
2036 | uint16_t *buf = vbuf; | |
2037 | size_t count = size >> 1; | |
2038 | int i; | |
2039 | ||
2040 | ASSERT((size & 1) == 0); | |
2041 | ||
2042 | for (i = 0; i < count; i++) | |
2043 | buf[i] = BSWAP_16(buf[i]); | |
2044 | } | |
2045 | ||
2046 | /* ARGSUSED */ | |
2047 | void | |
2048 | byteswap_uint8_array(void *vbuf, size_t size) | |
2049 | { | |
2050 | } | |
2051 | ||
2052 | void | |
2053 | dmu_init(void) | |
2054 | { | |
428870ff | 2055 | zfs_dbgmsg_init(); |
572e2857 BB |
2056 | sa_cache_init(); |
2057 | xuio_stat_init(); | |
2058 | dmu_objset_init(); | |
34dc7c2f | 2059 | dnode_init(); |
428870ff | 2060 | zfetch_init(); |
570827e1 | 2061 | dmu_tx_init(); |
34dc7c2f | 2062 | l2arc_init(); |
29809a6c | 2063 | arc_init(); |
d3c2ae1c | 2064 | dbuf_init(); |
34dc7c2f BB |
2065 | } |
2066 | ||
2067 | void | |
2068 | dmu_fini(void) | |
2069 | { | |
e49f1e20 | 2070 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2071 | l2arc_fini(); |
570827e1 | 2072 | dmu_tx_fini(); |
428870ff | 2073 | zfetch_fini(); |
34dc7c2f | 2074 | dbuf_fini(); |
572e2857 BB |
2075 | dnode_fini(); |
2076 | dmu_objset_fini(); | |
428870ff BB |
2077 | xuio_stat_fini(); |
2078 | sa_cache_fini(); | |
2079 | zfs_dbgmsg_fini(); | |
34dc7c2f | 2080 | } |
c28b2279 BB |
2081 | |
2082 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
2083 | EXPORT_SYMBOL(dmu_bonus_hold); | |
a473d90c AZ |
2084 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2085 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2086 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2087 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2088 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2089 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 BB |
2090 | EXPORT_SYMBOL(dmu_read); |
2091 | EXPORT_SYMBOL(dmu_write); | |
57b650b8 | 2092 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2093 | EXPORT_SYMBOL(dmu_object_info); |
2094 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2095 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2096 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
50c957f7 | 2097 | EXPORT_SYMBOL(dmu_object_dnsize_from_db); |
c28b2279 BB |
2098 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
2099 | EXPORT_SYMBOL(dmu_object_set_checksum); | |
2100 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
2101 | EXPORT_SYMBOL(dmu_write_policy); |
2102 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2103 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2104 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
2105 | EXPORT_SYMBOL(dmu_assign_arcbuf); | |
2106 | EXPORT_SYMBOL(dmu_buf_hold); | |
c28b2279 | 2107 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 BB |
2108 | |
2109 | module_param(zfs_mdcomp_disable, int, 0644); | |
2110 | MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression"); | |
03c6040b GW |
2111 | |
2112 | module_param(zfs_nopwrite_enabled, int, 0644); | |
2113 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2114 | ||
c28b2279 | 2115 | #endif |