]>
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. |
5dbd68a3 | 23 | * Copyright (c) 2011, 2014 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
62bdd5eb | 25 | * Copyright (c) 2014, Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
26 | */ |
27 | ||
34dc7c2f BB |
28 | #include <sys/dmu.h> |
29 | #include <sys/dmu_impl.h> | |
30 | #include <sys/dmu_tx.h> | |
31 | #include <sys/dbuf.h> | |
32 | #include <sys/dnode.h> | |
33 | #include <sys/zfs_context.h> | |
34 | #include <sys/dmu_objset.h> | |
35 | #include <sys/dmu_traverse.h> | |
36 | #include <sys/dsl_dataset.h> | |
37 | #include <sys/dsl_dir.h> | |
38 | #include <sys/dsl_pool.h> | |
39 | #include <sys/dsl_synctask.h> | |
40 | #include <sys/dsl_prop.h> | |
41 | #include <sys/dmu_zfetch.h> | |
42 | #include <sys/zfs_ioctl.h> | |
43 | #include <sys/zap.h> | |
44 | #include <sys/zio_checksum.h> | |
03c6040b | 45 | #include <sys/zio_compress.h> |
428870ff | 46 | #include <sys/sa.h> |
62bdd5eb | 47 | #include <sys/zfeature.h> |
34dc7c2f BB |
48 | #ifdef _KERNEL |
49 | #include <sys/vmsystm.h> | |
b128c09f | 50 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
51 | #endif |
52 | ||
03c6040b GW |
53 | /* |
54 | * Enable/disable nopwrite feature. | |
55 | */ | |
56 | int zfs_nopwrite_enabled = 1; | |
57 | ||
34dc7c2f | 58 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { |
9ae529ec CS |
59 | { DMU_BSWAP_UINT8, TRUE, "unallocated" }, |
60 | { DMU_BSWAP_ZAP, TRUE, "object directory" }, | |
61 | { DMU_BSWAP_UINT64, TRUE, "object array" }, | |
62 | { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, | |
63 | { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, | |
64 | { DMU_BSWAP_UINT64, TRUE, "bpobj" }, | |
65 | { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, | |
66 | { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, | |
67 | { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, | |
68 | { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, | |
69 | { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, | |
70 | { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, | |
71 | { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, | |
72 | { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, | |
73 | { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, | |
74 | { DMU_BSWAP_ZAP, TRUE, "DSL props" }, | |
75 | { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, | |
76 | { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, | |
77 | { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, | |
78 | { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, | |
79 | { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, | |
80 | { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, | |
81 | { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, | |
82 | { DMU_BSWAP_UINT8, FALSE, "zvol object" }, | |
83 | { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, | |
84 | { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, | |
85 | { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, | |
86 | { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, | |
87 | { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, | |
88 | { DMU_BSWAP_UINT8, TRUE, "SPA history" }, | |
89 | { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, | |
90 | { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, | |
91 | { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, | |
92 | { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, | |
93 | { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, | |
94 | { DMU_BSWAP_UINT8, TRUE, "FUID table" }, | |
95 | { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, | |
96 | { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, | |
97 | { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, | |
98 | { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, | |
99 | { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, | |
100 | { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, | |
101 | { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, | |
102 | { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, | |
103 | { DMU_BSWAP_UINT8, TRUE, "System attributes" }, | |
104 | { DMU_BSWAP_ZAP, TRUE, "SA master node" }, | |
105 | { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, | |
106 | { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, | |
107 | { DMU_BSWAP_ZAP, TRUE, "scan translations" }, | |
108 | { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, | |
109 | { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, | |
110 | { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, | |
111 | { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, | |
112 | { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } | |
113 | }; | |
114 | ||
115 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
116 | { byteswap_uint8_array, "uint8" }, | |
117 | { byteswap_uint16_array, "uint16" }, | |
118 | { byteswap_uint32_array, "uint32" }, | |
119 | { byteswap_uint64_array, "uint64" }, | |
120 | { zap_byteswap, "zap" }, | |
121 | { dnode_buf_byteswap, "dnode" }, | |
122 | { dmu_objset_byteswap, "objset" }, | |
123 | { zfs_znode_byteswap, "znode" }, | |
124 | { zfs_oldacl_byteswap, "oldacl" }, | |
125 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
126 | }; |
127 | ||
128 | int | |
9b67f605 MA |
129 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, |
130 | void *tag, dmu_buf_t **dbp) | |
34dc7c2f BB |
131 | { |
132 | dnode_t *dn; | |
133 | uint64_t blkid; | |
134 | dmu_buf_impl_t *db; | |
135 | int err; | |
428870ff BB |
136 | |
137 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
138 | if (err) |
139 | return (err); | |
140 | blkid = dbuf_whichblock(dn, offset); | |
141 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
142 | db = dbuf_hold(dn, blkid, tag); | |
143 | rw_exit(&dn->dn_struct_rwlock); | |
9b67f605 MA |
144 | dnode_rele(dn, FTAG); |
145 | ||
34dc7c2f | 146 | if (db == NULL) { |
9b67f605 MA |
147 | *dbp = NULL; |
148 | return (SET_ERROR(EIO)); | |
149 | } | |
150 | ||
151 | *dbp = &db->db; | |
152 | return (err); | |
153 | } | |
154 | ||
155 | int | |
156 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
157 | void *tag, dmu_buf_t **dbp, int flags) | |
158 | { | |
159 | int err; | |
160 | int db_flags = DB_RF_CANFAIL; | |
161 | ||
162 | if (flags & DMU_READ_NO_PREFETCH) | |
163 | db_flags |= DB_RF_NOPREFETCH; | |
164 | ||
165 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
166 | if (err == 0) { | |
167 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
428870ff | 168 | err = dbuf_read(db, NULL, db_flags); |
9b67f605 | 169 | if (err != 0) { |
34dc7c2f | 170 | dbuf_rele(db, tag); |
9b67f605 | 171 | *dbp = NULL; |
34dc7c2f BB |
172 | } |
173 | } | |
174 | ||
34dc7c2f BB |
175 | return (err); |
176 | } | |
177 | ||
178 | int | |
179 | dmu_bonus_max(void) | |
180 | { | |
181 | return (DN_MAX_BONUSLEN); | |
182 | } | |
183 | ||
184 | int | |
572e2857 | 185 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 186 | { |
572e2857 BB |
187 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
188 | dnode_t *dn; | |
189 | int error; | |
34dc7c2f | 190 | |
572e2857 BB |
191 | DB_DNODE_ENTER(db); |
192 | dn = DB_DNODE(db); | |
193 | ||
194 | if (dn->dn_bonus != db) { | |
2e528b49 | 195 | error = SET_ERROR(EINVAL); |
572e2857 | 196 | } else if (newsize < 0 || newsize > db_fake->db_size) { |
2e528b49 | 197 | error = SET_ERROR(EINVAL); |
572e2857 BB |
198 | } else { |
199 | dnode_setbonuslen(dn, newsize, tx); | |
200 | error = 0; | |
201 | } | |
202 | ||
203 | DB_DNODE_EXIT(db); | |
204 | return (error); | |
34dc7c2f BB |
205 | } |
206 | ||
428870ff | 207 | int |
572e2857 | 208 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 209 | { |
572e2857 BB |
210 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
211 | dnode_t *dn; | |
212 | int error; | |
428870ff | 213 | |
572e2857 BB |
214 | DB_DNODE_ENTER(db); |
215 | dn = DB_DNODE(db); | |
428870ff | 216 | |
9ae529ec | 217 | if (!DMU_OT_IS_VALID(type)) { |
2e528b49 | 218 | error = SET_ERROR(EINVAL); |
572e2857 | 219 | } else if (dn->dn_bonus != db) { |
2e528b49 | 220 | error = SET_ERROR(EINVAL); |
572e2857 BB |
221 | } else { |
222 | dnode_setbonus_type(dn, type, tx); | |
223 | error = 0; | |
224 | } | |
428870ff | 225 | |
572e2857 BB |
226 | DB_DNODE_EXIT(db); |
227 | return (error); | |
228 | } | |
229 | ||
230 | dmu_object_type_t | |
231 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
232 | { | |
233 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
234 | dnode_t *dn; | |
235 | dmu_object_type_t type; | |
236 | ||
237 | DB_DNODE_ENTER(db); | |
238 | dn = DB_DNODE(db); | |
239 | type = dn->dn_bonustype; | |
240 | DB_DNODE_EXIT(db); | |
241 | ||
242 | return (type); | |
428870ff BB |
243 | } |
244 | ||
245 | int | |
246 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
247 | { | |
248 | dnode_t *dn; | |
249 | int error; | |
250 | ||
251 | error = dnode_hold(os, object, FTAG, &dn); | |
252 | dbuf_rm_spill(dn, tx); | |
253 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
254 | dnode_rm_spill(dn, tx); | |
255 | rw_exit(&dn->dn_struct_rwlock); | |
256 | dnode_rele(dn, FTAG); | |
257 | return (error); | |
258 | } | |
259 | ||
34dc7c2f BB |
260 | /* |
261 | * returns ENOENT, EIO, or 0. | |
262 | */ | |
263 | int | |
264 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) | |
265 | { | |
266 | dnode_t *dn; | |
267 | dmu_buf_impl_t *db; | |
268 | int error; | |
269 | ||
428870ff | 270 | error = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
271 | if (error) |
272 | return (error); | |
273 | ||
274 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
275 | if (dn->dn_bonus == NULL) { | |
276 | rw_exit(&dn->dn_struct_rwlock); | |
277 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
278 | if (dn->dn_bonus == NULL) | |
279 | dbuf_create_bonus(dn); | |
280 | } | |
281 | db = dn->dn_bonus; | |
34dc7c2f BB |
282 | |
283 | /* as long as the bonus buf is held, the dnode will be held */ | |
572e2857 | 284 | if (refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 285 | VERIFY(dnode_add_ref(dn, db)); |
73ad4a9f | 286 | atomic_inc_32(&dn->dn_dbufs_count); |
572e2857 BB |
287 | } |
288 | ||
289 | /* | |
290 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
291 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
292 | * a dnode hold for every dbuf. | |
293 | */ | |
294 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
295 | |
296 | dnode_rele(dn, FTAG); | |
297 | ||
428870ff | 298 | VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); |
34dc7c2f BB |
299 | |
300 | *dbp = &db->db; | |
301 | return (0); | |
302 | } | |
303 | ||
428870ff BB |
304 | /* |
305 | * returns ENOENT, EIO, or 0. | |
306 | * | |
307 | * This interface will allocate a blank spill dbuf when a spill blk | |
308 | * doesn't already exist on the dnode. | |
309 | * | |
310 | * if you only want to find an already existing spill db, then | |
311 | * dmu_spill_hold_existing() should be used. | |
312 | */ | |
313 | int | |
314 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
315 | { | |
316 | dmu_buf_impl_t *db = NULL; | |
317 | int err; | |
318 | ||
319 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
320 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
321 | ||
322 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
323 | ||
324 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
325 | rw_exit(&dn->dn_struct_rwlock); | |
326 | ||
327 | ASSERT(db != NULL); | |
572e2857 BB |
328 | err = dbuf_read(db, NULL, flags); |
329 | if (err == 0) | |
330 | *dbp = &db->db; | |
331 | else | |
332 | dbuf_rele(db, tag); | |
428870ff BB |
333 | return (err); |
334 | } | |
335 | ||
336 | int | |
337 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
338 | { | |
572e2857 BB |
339 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
340 | dnode_t *dn; | |
428870ff BB |
341 | int err; |
342 | ||
572e2857 BB |
343 | DB_DNODE_ENTER(db); |
344 | dn = DB_DNODE(db); | |
345 | ||
346 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
2e528b49 | 347 | err = SET_ERROR(EINVAL); |
572e2857 BB |
348 | } else { |
349 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
350 | ||
351 | if (!dn->dn_have_spill) { | |
2e528b49 | 352 | err = SET_ERROR(ENOENT); |
572e2857 BB |
353 | } else { |
354 | err = dmu_spill_hold_by_dnode(dn, | |
355 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
356 | } | |
428870ff | 357 | |
428870ff | 358 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 359 | } |
572e2857 BB |
360 | |
361 | DB_DNODE_EXIT(db); | |
428870ff BB |
362 | return (err); |
363 | } | |
364 | ||
365 | int | |
366 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
367 | { | |
572e2857 BB |
368 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
369 | dnode_t *dn; | |
370 | int err; | |
371 | ||
372 | DB_DNODE_ENTER(db); | |
373 | dn = DB_DNODE(db); | |
374 | err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); | |
375 | DB_DNODE_EXIT(db); | |
376 | ||
377 | return (err); | |
428870ff BB |
378 | } |
379 | ||
34dc7c2f BB |
380 | /* |
381 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
382 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
383 | * and can induce severe lock contention when writing to several files | |
384 | * whose dnodes are in the same block. | |
385 | */ | |
386 | static int | |
9babb374 BB |
387 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
388 | int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) | |
34dc7c2f BB |
389 | { |
390 | dmu_buf_t **dbp; | |
391 | uint64_t blkid, nblks, i; | |
9babb374 | 392 | uint32_t dbuf_flags; |
34dc7c2f BB |
393 | int err; |
394 | zio_t *zio; | |
395 | ||
396 | ASSERT(length <= DMU_MAX_ACCESS); | |
397 | ||
45d1cae3 | 398 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; |
9babb374 BB |
399 | if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) |
400 | dbuf_flags |= DB_RF_NOPREFETCH; | |
34dc7c2f BB |
401 | |
402 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
403 | if (dn->dn_datablkshift) { | |
404 | int blkshift = dn->dn_datablkshift; | |
405 | nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - | |
406 | P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; | |
407 | } else { | |
408 | if (offset + length > dn->dn_datablksz) { | |
409 | zfs_panic_recover("zfs: accessing past end of object " | |
410 | "%llx/%llx (size=%u access=%llu+%llu)", | |
411 | (longlong_t)dn->dn_objset-> | |
412 | os_dsl_dataset->ds_object, | |
413 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
414 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 415 | rw_exit(&dn->dn_struct_rwlock); |
2e528b49 | 416 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
417 | } |
418 | nblks = 1; | |
419 | } | |
79c76d5b | 420 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
34dc7c2f | 421 | |
b128c09f | 422 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
34dc7c2f BB |
423 | blkid = dbuf_whichblock(dn, offset); |
424 | for (i = 0; i < nblks; i++) { | |
425 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); | |
426 | if (db == NULL) { | |
427 | rw_exit(&dn->dn_struct_rwlock); | |
428 | dmu_buf_rele_array(dbp, nblks, tag); | |
429 | zio_nowait(zio); | |
2e528b49 | 430 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
431 | } |
432 | /* initiate async i/o */ | |
433 | if (read) { | |
9babb374 | 434 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
435 | } |
436 | dbp[i] = &db->db; | |
437 | } | |
438 | rw_exit(&dn->dn_struct_rwlock); | |
439 | ||
440 | /* wait for async i/o */ | |
441 | err = zio_wait(zio); | |
442 | if (err) { | |
443 | dmu_buf_rele_array(dbp, nblks, tag); | |
444 | return (err); | |
445 | } | |
446 | ||
447 | /* wait for other io to complete */ | |
448 | if (read) { | |
449 | for (i = 0; i < nblks; i++) { | |
450 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
451 | mutex_enter(&db->db_mtx); | |
452 | while (db->db_state == DB_READ || | |
453 | db->db_state == DB_FILL) | |
454 | cv_wait(&db->db_changed, &db->db_mtx); | |
455 | if (db->db_state == DB_UNCACHED) | |
2e528b49 | 456 | err = SET_ERROR(EIO); |
34dc7c2f BB |
457 | mutex_exit(&db->db_mtx); |
458 | if (err) { | |
459 | dmu_buf_rele_array(dbp, nblks, tag); | |
460 | return (err); | |
461 | } | |
462 | } | |
463 | } | |
464 | ||
465 | *numbufsp = nblks; | |
466 | *dbpp = dbp; | |
467 | return (0); | |
468 | } | |
469 | ||
470 | static int | |
471 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
472 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
473 | { | |
474 | dnode_t *dn; | |
475 | int err; | |
476 | ||
428870ff | 477 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
478 | if (err) |
479 | return (err); | |
480 | ||
481 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 482 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
483 | |
484 | dnode_rele(dn, FTAG); | |
485 | ||
486 | return (err); | |
487 | } | |
488 | ||
489 | int | |
572e2857 | 490 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
34dc7c2f BB |
491 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) |
492 | { | |
572e2857 BB |
493 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
494 | dnode_t *dn; | |
34dc7c2f BB |
495 | int err; |
496 | ||
572e2857 BB |
497 | DB_DNODE_ENTER(db); |
498 | dn = DB_DNODE(db); | |
34dc7c2f | 499 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 500 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 501 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
502 | |
503 | return (err); | |
504 | } | |
505 | ||
506 | void | |
507 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
508 | { | |
509 | int i; | |
510 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
511 | ||
512 | if (numbufs == 0) | |
513 | return; | |
514 | ||
515 | for (i = 0; i < numbufs; i++) { | |
516 | if (dbp[i]) | |
517 | dbuf_rele(dbp[i], tag); | |
518 | } | |
519 | ||
520 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
521 | } | |
522 | ||
e8b96c60 MA |
523 | /* |
524 | * Issue prefetch i/os for the given blocks. | |
525 | * | |
526 | * Note: The assumption is that we *know* these blocks will be needed | |
527 | * almost immediately. Therefore, the prefetch i/os will be issued at | |
528 | * ZIO_PRIORITY_SYNC_READ | |
529 | * | |
530 | * Note: indirect blocks and other metadata will be read synchronously, | |
531 | * causing this function to block if they are not already cached. | |
532 | */ | |
34dc7c2f BB |
533 | void |
534 | dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) | |
535 | { | |
536 | dnode_t *dn; | |
537 | uint64_t blkid; | |
e8b96c60 | 538 | int nblks, err; |
34dc7c2f BB |
539 | |
540 | if (zfs_prefetch_disable) | |
541 | return; | |
542 | ||
543 | if (len == 0) { /* they're interested in the bonus buffer */ | |
572e2857 | 544 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
545 | |
546 | if (object == 0 || object >= DN_MAX_OBJECT) | |
547 | return; | |
548 | ||
549 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
550 | blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); | |
e8b96c60 | 551 | dbuf_prefetch(dn, blkid, ZIO_PRIORITY_SYNC_READ); |
34dc7c2f BB |
552 | rw_exit(&dn->dn_struct_rwlock); |
553 | return; | |
554 | } | |
555 | ||
556 | /* | |
557 | * XXX - Note, if the dnode for the requested object is not | |
558 | * already cached, we will do a *synchronous* read in the | |
559 | * dnode_hold() call. The same is true for any indirects. | |
560 | */ | |
428870ff | 561 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
562 | if (err != 0) |
563 | return; | |
564 | ||
565 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
566 | if (dn->dn_datablkshift) { | |
567 | int blkshift = dn->dn_datablkshift; | |
e8b96c60 MA |
568 | nblks = (P2ROUNDUP(offset + len, 1 << blkshift) - |
569 | P2ALIGN(offset, 1 << blkshift)) >> blkshift; | |
34dc7c2f BB |
570 | } else { |
571 | nblks = (offset < dn->dn_datablksz); | |
572 | } | |
573 | ||
574 | if (nblks != 0) { | |
e8b96c60 MA |
575 | int i; |
576 | ||
34dc7c2f BB |
577 | blkid = dbuf_whichblock(dn, offset); |
578 | for (i = 0; i < nblks; i++) | |
e8b96c60 | 579 | dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_SYNC_READ); |
34dc7c2f BB |
580 | } |
581 | ||
582 | rw_exit(&dn->dn_struct_rwlock); | |
583 | ||
584 | dnode_rele(dn, FTAG); | |
585 | } | |
586 | ||
45d1cae3 BB |
587 | /* |
588 | * Get the next "chunk" of file data to free. We traverse the file from | |
589 | * the end so that the file gets shorter over time (if we crashes in the | |
590 | * middle, this will leave us in a better state). We find allocated file | |
591 | * data by simply searching the allocated level 1 indirects. | |
b663a23d MA |
592 | * |
593 | * On input, *start should be the first offset that does not need to be | |
594 | * freed (e.g. "offset + length"). On return, *start will be the first | |
595 | * offset that should be freed. | |
45d1cae3 | 596 | */ |
b128c09f | 597 | static int |
b663a23d | 598 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) |
b128c09f | 599 | { |
b663a23d MA |
600 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); |
601 | /* bytes of data covered by a level-1 indirect block */ | |
45d1cae3 | 602 | uint64_t iblkrange = |
b128c09f | 603 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
b663a23d | 604 | uint64_t blks; |
b128c09f | 605 | |
b663a23d | 606 | ASSERT3U(minimum, <=, *start); |
b128c09f | 607 | |
b663a23d MA |
608 | if (*start - minimum <= iblkrange * maxblks) { |
609 | *start = minimum; | |
b128c09f BB |
610 | return (0); |
611 | } | |
45d1cae3 | 612 | ASSERT(ISP2(iblkrange)); |
b128c09f | 613 | |
b663a23d | 614 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { |
b128c09f BB |
615 | int err; |
616 | ||
b663a23d MA |
617 | /* |
618 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
619 | * indirect block at or before the input offset. We must | |
620 | * decrement *start so that it is at the end of the region | |
621 | * to search. | |
622 | */ | |
623 | (*start)--; | |
b128c09f | 624 | err = dnode_next_offset(dn, |
45d1cae3 | 625 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 626 | |
b663a23d | 627 | /* if there are no indirect blocks before start, we are done */ |
45d1cae3 | 628 | if (err == ESRCH) { |
b663a23d MA |
629 | *start = minimum; |
630 | break; | |
631 | } else if (err != 0) { | |
b128c09f | 632 | return (err); |
45d1cae3 | 633 | } |
b128c09f | 634 | |
b663a23d | 635 | /* set start to the beginning of this L1 indirect */ |
45d1cae3 | 636 | *start = P2ALIGN(*start, iblkrange); |
b128c09f | 637 | } |
b663a23d MA |
638 | if (*start < minimum) |
639 | *start = minimum; | |
b128c09f BB |
640 | return (0); |
641 | } | |
642 | ||
643 | static int | |
644 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
b663a23d | 645 | uint64_t length) |
b128c09f | 646 | { |
b663a23d MA |
647 | uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
648 | int err; | |
649 | ||
650 | if (offset >= object_size) | |
b128c09f | 651 | return (0); |
b128c09f | 652 | |
b663a23d MA |
653 | if (length == DMU_OBJECT_END || offset + length > object_size) |
654 | length = object_size - offset; | |
655 | ||
656 | while (length != 0) { | |
657 | uint64_t chunk_end, chunk_begin; | |
658 | dmu_tx_t *tx; | |
659 | ||
660 | chunk_end = chunk_begin = offset + length; | |
661 | ||
662 | /* move chunk_begin backwards to the beginning of this chunk */ | |
663 | err = get_next_chunk(dn, &chunk_begin, offset); | |
b128c09f BB |
664 | if (err) |
665 | return (err); | |
b663a23d MA |
666 | ASSERT3U(chunk_begin, >=, offset); |
667 | ASSERT3U(chunk_begin, <=, chunk_end); | |
b128c09f BB |
668 | |
669 | tx = dmu_tx_create(os); | |
b663a23d MA |
670 | dmu_tx_hold_free(tx, dn->dn_object, |
671 | chunk_begin, chunk_end - chunk_begin); | |
b128c09f BB |
672 | err = dmu_tx_assign(tx, TXG_WAIT); |
673 | if (err) { | |
674 | dmu_tx_abort(tx); | |
675 | return (err); | |
676 | } | |
b663a23d | 677 | dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx); |
b128c09f | 678 | dmu_tx_commit(tx); |
b663a23d MA |
679 | |
680 | length -= chunk_end - chunk_begin; | |
b128c09f BB |
681 | } |
682 | return (0); | |
683 | } | |
684 | ||
685 | int | |
686 | dmu_free_long_range(objset_t *os, uint64_t object, | |
687 | uint64_t offset, uint64_t length) | |
688 | { | |
689 | dnode_t *dn; | |
690 | int err; | |
691 | ||
428870ff | 692 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
693 | if (err != 0) |
694 | return (err); | |
b663a23d | 695 | err = dmu_free_long_range_impl(os, dn, offset, length); |
92bc214c MA |
696 | |
697 | /* | |
698 | * It is important to zero out the maxblkid when freeing the entire | |
699 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
700 | * will take the fast path, and (b) dnode_reallocate() can verify | |
701 | * that the entire file has been freed. | |
702 | */ | |
b0bc7a84 | 703 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) |
92bc214c MA |
704 | dn->dn_maxblkid = 0; |
705 | ||
b128c09f BB |
706 | dnode_rele(dn, FTAG); |
707 | return (err); | |
708 | } | |
709 | ||
710 | int | |
b663a23d | 711 | dmu_free_long_object(objset_t *os, uint64_t object) |
b128c09f | 712 | { |
b128c09f BB |
713 | dmu_tx_t *tx; |
714 | int err; | |
715 | ||
b663a23d | 716 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); |
b128c09f BB |
717 | if (err != 0) |
718 | return (err); | |
b663a23d MA |
719 | |
720 | tx = dmu_tx_create(os); | |
721 | dmu_tx_hold_bonus(tx, object); | |
722 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
723 | err = dmu_tx_assign(tx, TXG_WAIT); | |
724 | if (err == 0) { | |
725 | err = dmu_object_free(os, object, tx); | |
726 | dmu_tx_commit(tx); | |
b128c09f | 727 | } else { |
b663a23d | 728 | dmu_tx_abort(tx); |
b128c09f | 729 | } |
b663a23d | 730 | |
b128c09f BB |
731 | return (err); |
732 | } | |
733 | ||
34dc7c2f BB |
734 | int |
735 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
736 | uint64_t size, dmu_tx_t *tx) | |
737 | { | |
738 | dnode_t *dn; | |
428870ff | 739 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
740 | if (err) |
741 | return (err); | |
742 | ASSERT(offset < UINT64_MAX); | |
743 | ASSERT(size == -1ULL || size <= UINT64_MAX - offset); | |
744 | dnode_free_range(dn, offset, size, tx); | |
745 | dnode_rele(dn, FTAG); | |
746 | return (0); | |
747 | } | |
748 | ||
749 | int | |
750 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
9babb374 | 751 | void *buf, uint32_t flags) |
34dc7c2f BB |
752 | { |
753 | dnode_t *dn; | |
754 | dmu_buf_t **dbp; | |
45d1cae3 | 755 | int numbufs, err; |
34dc7c2f | 756 | |
428870ff | 757 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
758 | if (err) |
759 | return (err); | |
760 | ||
761 | /* | |
762 | * Deal with odd block sizes, where there can't be data past the first | |
763 | * block. If we ever do the tail block optimization, we will need to | |
764 | * handle that here as well. | |
765 | */ | |
45d1cae3 | 766 | if (dn->dn_maxblkid == 0) { |
c9520ecc | 767 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : |
34dc7c2f BB |
768 | MIN(size, dn->dn_datablksz - offset); |
769 | bzero((char *)buf + newsz, size - newsz); | |
770 | size = newsz; | |
771 | } | |
772 | ||
773 | while (size > 0) { | |
774 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 775 | int i; |
34dc7c2f BB |
776 | |
777 | /* | |
778 | * NB: we could do this block-at-a-time, but it's nice | |
779 | * to be reading in parallel. | |
780 | */ | |
781 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 782 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
783 | if (err) |
784 | break; | |
785 | ||
786 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
787 | uint64_t tocpy; |
788 | int64_t bufoff; | |
34dc7c2f BB |
789 | dmu_buf_t *db = dbp[i]; |
790 | ||
791 | ASSERT(size > 0); | |
792 | ||
793 | bufoff = offset - db->db_offset; | |
c9520ecc | 794 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f | 795 | |
c9520ecc | 796 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); |
34dc7c2f BB |
797 | |
798 | offset += tocpy; | |
799 | size -= tocpy; | |
800 | buf = (char *)buf + tocpy; | |
801 | } | |
802 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
803 | } | |
804 | dnode_rele(dn, FTAG); | |
805 | return (err); | |
806 | } | |
807 | ||
808 | void | |
809 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
810 | const void *buf, dmu_tx_t *tx) | |
811 | { | |
812 | dmu_buf_t **dbp; | |
813 | int numbufs, i; | |
814 | ||
815 | if (size == 0) | |
816 | return; | |
817 | ||
e6659763 | 818 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, |
34dc7c2f BB |
819 | FALSE, FTAG, &numbufs, &dbp)); |
820 | ||
821 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
822 | uint64_t tocpy; |
823 | int64_t bufoff; | |
34dc7c2f BB |
824 | dmu_buf_t *db = dbp[i]; |
825 | ||
826 | ASSERT(size > 0); | |
827 | ||
828 | bufoff = offset - db->db_offset; | |
c9520ecc | 829 | tocpy = MIN(db->db_size - bufoff, size); |
34dc7c2f BB |
830 | |
831 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
832 | ||
833 | if (tocpy == db->db_size) | |
834 | dmu_buf_will_fill(db, tx); | |
835 | else | |
836 | dmu_buf_will_dirty(db, tx); | |
837 | ||
60101509 | 838 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
839 | |
840 | if (tocpy == db->db_size) | |
841 | dmu_buf_fill_done(db, tx); | |
842 | ||
843 | offset += tocpy; | |
844 | size -= tocpy; | |
845 | buf = (char *)buf + tocpy; | |
846 | } | |
847 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
848 | } | |
849 | ||
b128c09f BB |
850 | void |
851 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
852 | dmu_tx_t *tx) | |
853 | { | |
854 | dmu_buf_t **dbp; | |
855 | int numbufs, i; | |
856 | ||
857 | if (size == 0) | |
858 | return; | |
859 | ||
860 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
861 | FALSE, FTAG, &numbufs, &dbp)); | |
862 | ||
863 | for (i = 0; i < numbufs; i++) { | |
864 | dmu_buf_t *db = dbp[i]; | |
865 | ||
866 | dmu_buf_will_not_fill(db, tx); | |
867 | } | |
868 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
869 | } | |
870 | ||
9b67f605 MA |
871 | void |
872 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
873 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
874 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
875 | { | |
876 | dmu_buf_t *db; | |
877 | ||
878 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
879 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
880 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
881 | FTAG, &db)); | |
882 | ||
883 | dmu_buf_write_embedded(db, | |
884 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
885 | uncompressed_size, compressed_size, byteorder, tx); | |
886 | ||
887 | dmu_buf_rele(db, FTAG); | |
888 | } | |
889 | ||
428870ff BB |
890 | /* |
891 | * DMU support for xuio | |
892 | */ | |
893 | kstat_t *xuio_ksp = NULL; | |
894 | ||
59e6e7ca BB |
895 | typedef struct xuio_stats { |
896 | /* loaned yet not returned arc_buf */ | |
897 | kstat_named_t xuiostat_onloan_rbuf; | |
898 | kstat_named_t xuiostat_onloan_wbuf; | |
899 | /* whether a copy is made when loaning out a read buffer */ | |
900 | kstat_named_t xuiostat_rbuf_copied; | |
901 | kstat_named_t xuiostat_rbuf_nocopy; | |
902 | /* whether a copy is made when assigning a write buffer */ | |
903 | kstat_named_t xuiostat_wbuf_copied; | |
904 | kstat_named_t xuiostat_wbuf_nocopy; | |
905 | } xuio_stats_t; | |
906 | ||
907 | static xuio_stats_t xuio_stats = { | |
908 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
909 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
910 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
911 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
912 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
913 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
914 | }; | |
915 | ||
d1d7e268 MK |
916 | #define XUIOSTAT_INCR(stat, val) \ |
917 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
918 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
59e6e7ca | 919 | |
428870ff BB |
920 | int |
921 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
922 | { | |
923 | dmu_xuio_t *priv; | |
924 | uio_t *uio = &xuio->xu_uio; | |
925 | ||
926 | uio->uio_iovcnt = nblk; | |
79c76d5b | 927 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
428870ff | 928 | |
79c76d5b | 929 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
428870ff | 930 | priv->cnt = nblk; |
79c76d5b | 931 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
428870ff BB |
932 | priv->iovp = uio->uio_iov; |
933 | XUIO_XUZC_PRIV(xuio) = priv; | |
934 | ||
935 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
936 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
937 | else | |
938 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
939 | ||
940 | return (0); | |
941 | } | |
942 | ||
943 | void | |
944 | dmu_xuio_fini(xuio_t *xuio) | |
945 | { | |
946 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
947 | int nblk = priv->cnt; | |
948 | ||
949 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
950 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
951 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
952 | ||
953 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
954 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
955 | else | |
956 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
957 | } | |
958 | ||
959 | /* | |
960 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
961 | * and increase priv->next by 1. | |
962 | */ | |
963 | int | |
964 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
965 | { | |
966 | struct iovec *iov; | |
967 | uio_t *uio = &xuio->xu_uio; | |
968 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
969 | int i = priv->next++; | |
970 | ||
971 | ASSERT(i < priv->cnt); | |
972 | ASSERT(off + n <= arc_buf_size(abuf)); | |
973 | iov = uio->uio_iov + i; | |
974 | iov->iov_base = (char *)abuf->b_data + off; | |
975 | iov->iov_len = n; | |
976 | priv->bufs[i] = abuf; | |
977 | return (0); | |
978 | } | |
979 | ||
980 | int | |
981 | dmu_xuio_cnt(xuio_t *xuio) | |
982 | { | |
983 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
984 | return (priv->cnt); | |
985 | } | |
986 | ||
987 | arc_buf_t * | |
988 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
989 | { | |
990 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
991 | ||
992 | ASSERT(i < priv->cnt); | |
993 | return (priv->bufs[i]); | |
994 | } | |
995 | ||
996 | void | |
997 | dmu_xuio_clear(xuio_t *xuio, int i) | |
998 | { | |
999 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1000 | ||
1001 | ASSERT(i < priv->cnt); | |
1002 | priv->bufs[i] = NULL; | |
1003 | } | |
1004 | ||
1005 | static void | |
1006 | xuio_stat_init(void) | |
1007 | { | |
1008 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1009 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1010 | KSTAT_FLAG_VIRTUAL); | |
1011 | if (xuio_ksp != NULL) { | |
1012 | xuio_ksp->ks_data = &xuio_stats; | |
1013 | kstat_install(xuio_ksp); | |
1014 | } | |
1015 | } | |
1016 | ||
1017 | static void | |
1018 | xuio_stat_fini(void) | |
1019 | { | |
1020 | if (xuio_ksp != NULL) { | |
1021 | kstat_delete(xuio_ksp); | |
1022 | xuio_ksp = NULL; | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | void | |
1027 | xuio_stat_wbuf_copied() | |
1028 | { | |
1029 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1030 | } | |
1031 | ||
1032 | void | |
1033 | xuio_stat_wbuf_nocopy() | |
1034 | { | |
1035 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1036 | } | |
1037 | ||
34dc7c2f | 1038 | #ifdef _KERNEL |
60101509 BB |
1039 | |
1040 | /* | |
1041 | * Copy up to size bytes between arg_buf and req based on the data direction | |
215b4634 CC |
1042 | * described by the req. If an entire req's data cannot be transfered in one |
1043 | * pass, you should pass in @req_offset to indicate where to continue. The | |
1044 | * return value is the number of bytes successfully copied to arg_buf. | |
60101509 BB |
1045 | */ |
1046 | static int | |
215b4634 | 1047 | dmu_req_copy(void *arg_buf, int size, struct request *req, size_t req_offset) |
60101509 | 1048 | { |
b761912b | 1049 | struct bio_vec bv, *bvp; |
60101509 BB |
1050 | struct req_iterator iter; |
1051 | char *bv_buf; | |
215b4634 CC |
1052 | int tocpy, bv_len, bv_offset; |
1053 | int offset = 0; | |
60101509 | 1054 | |
b761912b | 1055 | rq_for_each_segment4(bv, bvp, req, iter) { |
215b4634 CC |
1056 | /* |
1057 | * Fully consumed the passed arg_buf. We use goto here because | |
1058 | * rq_for_each_segment is a double loop | |
1059 | */ | |
1060 | ASSERT3S(offset, <=, size); | |
1061 | if (size == offset) | |
1062 | goto out; | |
60101509 | 1063 | |
215b4634 | 1064 | /* Skip already copied bv */ |
b761912b CC |
1065 | if (req_offset >= bv.bv_len) { |
1066 | req_offset -= bv.bv_len; | |
60101509 | 1067 | continue; |
215b4634 CC |
1068 | } |
1069 | ||
b761912b CC |
1070 | bv_len = bv.bv_len - req_offset; |
1071 | bv_offset = bv.bv_offset + req_offset; | |
215b4634 | 1072 | req_offset = 0; |
60101509 | 1073 | |
215b4634 | 1074 | tocpy = MIN(bv_len, size - offset); |
60101509 BB |
1075 | ASSERT3S(tocpy, >=, 0); |
1076 | ||
b761912b | 1077 | bv_buf = page_address(bv.bv_page) + bv_offset; |
60101509 BB |
1078 | ASSERT3P(bv_buf, !=, NULL); |
1079 | ||
1080 | if (rq_data_dir(req) == WRITE) | |
215b4634 | 1081 | memcpy(arg_buf + offset, bv_buf, tocpy); |
60101509 | 1082 | else |
215b4634 | 1083 | memcpy(bv_buf, arg_buf + offset, tocpy); |
60101509 | 1084 | |
215b4634 | 1085 | offset += tocpy; |
60101509 | 1086 | } |
215b4634 CC |
1087 | out: |
1088 | return (offset); | |
60101509 BB |
1089 | } |
1090 | ||
34dc7c2f | 1091 | int |
60101509 | 1092 | dmu_read_req(objset_t *os, uint64_t object, struct request *req) |
34dc7c2f | 1093 | { |
60101509 BB |
1094 | uint64_t size = blk_rq_bytes(req); |
1095 | uint64_t offset = blk_rq_pos(req) << 9; | |
34dc7c2f BB |
1096 | dmu_buf_t **dbp; |
1097 | int numbufs, i, err; | |
215b4634 | 1098 | size_t req_offset; |
34dc7c2f BB |
1099 | |
1100 | /* | |
1101 | * NB: we could do this block-at-a-time, but it's nice | |
1102 | * to be reading in parallel. | |
1103 | */ | |
60101509 | 1104 | err = dmu_buf_hold_array(os, object, offset, size, TRUE, FTAG, |
d1d7e268 | 1105 | &numbufs, &dbp); |
34dc7c2f BB |
1106 | if (err) |
1107 | return (err); | |
1108 | ||
215b4634 | 1109 | req_offset = 0; |
34dc7c2f | 1110 | for (i = 0; i < numbufs; i++) { |
c9520ecc JZ |
1111 | uint64_t tocpy; |
1112 | int64_t bufoff; | |
1113 | int didcpy; | |
34dc7c2f BB |
1114 | dmu_buf_t *db = dbp[i]; |
1115 | ||
60101509 BB |
1116 | bufoff = offset - db->db_offset; |
1117 | ASSERT3S(bufoff, >=, 0); | |
34dc7c2f | 1118 | |
c9520ecc | 1119 | tocpy = MIN(db->db_size - bufoff, size); |
60101509 BB |
1120 | if (tocpy == 0) |
1121 | break; | |
34dc7c2f | 1122 | |
215b4634 CC |
1123 | didcpy = dmu_req_copy(db->db_data + bufoff, tocpy, req, |
1124 | req_offset); | |
60101509 BB |
1125 | |
1126 | if (didcpy < tocpy) | |
1127 | err = EIO; | |
428870ff | 1128 | |
34dc7c2f BB |
1129 | if (err) |
1130 | break; | |
1131 | ||
1132 | size -= tocpy; | |
60101509 | 1133 | offset += didcpy; |
215b4634 | 1134 | req_offset += didcpy; |
60101509 | 1135 | err = 0; |
34dc7c2f BB |
1136 | } |
1137 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1138 | ||
1139 | return (err); | |
1140 | } | |
1141 | ||
60101509 BB |
1142 | int |
1143 | dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx) | |
34dc7c2f | 1144 | { |
60101509 BB |
1145 | uint64_t size = blk_rq_bytes(req); |
1146 | uint64_t offset = blk_rq_pos(req) << 9; | |
34dc7c2f | 1147 | dmu_buf_t **dbp; |
215b4634 CC |
1148 | int numbufs, i, err; |
1149 | size_t req_offset; | |
34dc7c2f | 1150 | |
60101509 BB |
1151 | if (size == 0) |
1152 | return (0); | |
1153 | ||
1154 | err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
d1d7e268 | 1155 | &numbufs, &dbp); |
34dc7c2f BB |
1156 | if (err) |
1157 | return (err); | |
1158 | ||
215b4634 | 1159 | req_offset = 0; |
34dc7c2f | 1160 | for (i = 0; i < numbufs; i++) { |
c9520ecc JZ |
1161 | uint64_t tocpy; |
1162 | int64_t bufoff; | |
1163 | int didcpy; | |
34dc7c2f BB |
1164 | dmu_buf_t *db = dbp[i]; |
1165 | ||
60101509 BB |
1166 | bufoff = offset - db->db_offset; |
1167 | ASSERT3S(bufoff, >=, 0); | |
34dc7c2f | 1168 | |
c9520ecc | 1169 | tocpy = MIN(db->db_size - bufoff, size); |
60101509 BB |
1170 | if (tocpy == 0) |
1171 | break; | |
34dc7c2f BB |
1172 | |
1173 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1174 | ||
1175 | if (tocpy == db->db_size) | |
1176 | dmu_buf_will_fill(db, tx); | |
1177 | else | |
1178 | dmu_buf_will_dirty(db, tx); | |
1179 | ||
215b4634 CC |
1180 | didcpy = dmu_req_copy(db->db_data + bufoff, tocpy, req, |
1181 | req_offset); | |
34dc7c2f BB |
1182 | |
1183 | if (tocpy == db->db_size) | |
1184 | dmu_buf_fill_done(db, tx); | |
1185 | ||
60101509 BB |
1186 | if (didcpy < tocpy) |
1187 | err = EIO; | |
1188 | ||
34dc7c2f BB |
1189 | if (err) |
1190 | break; | |
1191 | ||
1192 | size -= tocpy; | |
60101509 | 1193 | offset += didcpy; |
215b4634 | 1194 | req_offset += didcpy; |
60101509 | 1195 | err = 0; |
34dc7c2f | 1196 | } |
428870ff | 1197 | |
34dc7c2f BB |
1198 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1199 | return (err); | |
1200 | } | |
1201 | ||
872e8d26 BB |
1202 | int |
1203 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1204 | { | |
1205 | dmu_buf_t **dbp; | |
1206 | int numbufs, i, err; | |
1207 | xuio_t *xuio = NULL; | |
1208 | ||
1209 | /* | |
1210 | * NB: we could do this block-at-a-time, but it's nice | |
1211 | * to be reading in parallel. | |
1212 | */ | |
1213 | err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, | |
1214 | &numbufs, &dbp); | |
1215 | if (err) | |
1216 | return (err); | |
1217 | ||
1218 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1219 | uint64_t tocpy; |
1220 | int64_t bufoff; | |
872e8d26 BB |
1221 | dmu_buf_t *db = dbp[i]; |
1222 | ||
1223 | ASSERT(size > 0); | |
1224 | ||
1225 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1226 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1227 | |
1228 | if (xuio) { | |
1229 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1230 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1231 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1232 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1233 | if (!err) { | |
1234 | uio->uio_resid -= tocpy; | |
1235 | uio->uio_loffset += tocpy; | |
1236 | } | |
1237 | ||
1238 | if (abuf == dbuf_abuf) | |
1239 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1240 | else | |
1241 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
1242 | } else { | |
1243 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1244 | UIO_READ, uio); | |
1245 | } | |
1246 | if (err) | |
1247 | break; | |
1248 | ||
1249 | size -= tocpy; | |
1250 | } | |
1251 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1252 | ||
1253 | return (err); | |
1254 | } | |
1255 | ||
1256 | static int | |
1257 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) | |
1258 | { | |
1259 | dmu_buf_t **dbp; | |
1260 | int numbufs; | |
1261 | int err = 0; | |
1262 | int i; | |
1263 | ||
1264 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1265 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1266 | if (err) | |
1267 | return (err); | |
1268 | ||
1269 | for (i = 0; i < numbufs; i++) { | |
c9520ecc JZ |
1270 | uint64_t tocpy; |
1271 | int64_t bufoff; | |
872e8d26 BB |
1272 | dmu_buf_t *db = dbp[i]; |
1273 | ||
1274 | ASSERT(size > 0); | |
1275 | ||
1276 | bufoff = uio->uio_loffset - db->db_offset; | |
c9520ecc | 1277 | tocpy = MIN(db->db_size - bufoff, size); |
872e8d26 BB |
1278 | |
1279 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1280 | ||
1281 | if (tocpy == db->db_size) | |
1282 | dmu_buf_will_fill(db, tx); | |
1283 | else | |
1284 | dmu_buf_will_dirty(db, tx); | |
1285 | ||
1286 | /* | |
1287 | * XXX uiomove could block forever (eg.nfs-backed | |
1288 | * pages). There needs to be a uiolockdown() function | |
1289 | * to lock the pages in memory, so that uiomove won't | |
1290 | * block. | |
1291 | */ | |
1292 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1293 | UIO_WRITE, uio); | |
1294 | ||
1295 | if (tocpy == db->db_size) | |
1296 | dmu_buf_fill_done(db, tx); | |
1297 | ||
1298 | if (err) | |
1299 | break; | |
1300 | ||
1301 | size -= tocpy; | |
1302 | } | |
1303 | ||
1304 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1305 | return (err); | |
1306 | } | |
1307 | ||
428870ff BB |
1308 | int |
1309 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1310 | dmu_tx_t *tx) | |
1311 | { | |
572e2857 BB |
1312 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1313 | dnode_t *dn; | |
1314 | int err; | |
1315 | ||
428870ff BB |
1316 | if (size == 0) |
1317 | return (0); | |
1318 | ||
572e2857 BB |
1319 | DB_DNODE_ENTER(db); |
1320 | dn = DB_DNODE(db); | |
1321 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1322 | DB_DNODE_EXIT(db); | |
1323 | ||
1324 | return (err); | |
428870ff BB |
1325 | } |
1326 | ||
1327 | int | |
1328 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1329 | dmu_tx_t *tx) | |
1330 | { | |
1331 | dnode_t *dn; | |
1332 | int err; | |
1333 | ||
1334 | if (size == 0) | |
1335 | return (0); | |
1336 | ||
1337 | err = dnode_hold(os, object, FTAG, &dn); | |
1338 | if (err) | |
1339 | return (err); | |
1340 | ||
1341 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1342 | ||
1343 | dnode_rele(dn, FTAG); | |
1344 | ||
1345 | return (err); | |
1346 | } | |
872e8d26 | 1347 | #endif /* _KERNEL */ |
34dc7c2f | 1348 | |
9babb374 BB |
1349 | /* |
1350 | * Allocate a loaned anonymous arc buffer. | |
1351 | */ | |
1352 | arc_buf_t * | |
1353 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1354 | { | |
572e2857 | 1355 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
9babb374 | 1356 | |
b0bc7a84 | 1357 | return (arc_loan_buf(db->db_objset->os_spa, size)); |
9babb374 BB |
1358 | } |
1359 | ||
1360 | /* | |
1361 | * Free a loaned arc buffer. | |
1362 | */ | |
1363 | void | |
1364 | dmu_return_arcbuf(arc_buf_t *buf) | |
1365 | { | |
1366 | arc_return_buf(buf, FTAG); | |
13fe0198 | 1367 | VERIFY(arc_buf_remove_ref(buf, FTAG)); |
9babb374 BB |
1368 | } |
1369 | ||
1370 | /* | |
1371 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1372 | * If this is not possible copy the contents of passed arc buf via | |
1373 | * dmu_write(). | |
1374 | */ | |
1375 | void | |
1376 | dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, | |
1377 | dmu_tx_t *tx) | |
1378 | { | |
572e2857 BB |
1379 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1380 | dnode_t *dn; | |
9babb374 BB |
1381 | dmu_buf_impl_t *db; |
1382 | uint32_t blksz = (uint32_t)arc_buf_size(buf); | |
1383 | uint64_t blkid; | |
1384 | ||
572e2857 BB |
1385 | DB_DNODE_ENTER(dbuf); |
1386 | dn = DB_DNODE(dbuf); | |
9babb374 BB |
1387 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
1388 | blkid = dbuf_whichblock(dn, offset); | |
1389 | VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); | |
1390 | rw_exit(&dn->dn_struct_rwlock); | |
572e2857 | 1391 | DB_DNODE_EXIT(dbuf); |
9babb374 | 1392 | |
88904bb3 MA |
1393 | /* |
1394 | * We can only assign if the offset is aligned, the arc buf is the | |
1395 | * same size as the dbuf, and the dbuf is not metadata. It | |
1396 | * can't be metadata because the loaned arc buf comes from the | |
1397 | * user-data kmem area. | |
1398 | */ | |
1399 | if (offset == db->db.db_offset && blksz == db->db.db_size && | |
1400 | DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) { | |
9babb374 BB |
1401 | dbuf_assign_arcbuf(db, buf, tx); |
1402 | dbuf_rele(db, FTAG); | |
1403 | } else { | |
572e2857 BB |
1404 | objset_t *os; |
1405 | uint64_t object; | |
1406 | ||
1407 | DB_DNODE_ENTER(dbuf); | |
1408 | dn = DB_DNODE(dbuf); | |
1409 | os = dn->dn_objset; | |
1410 | object = dn->dn_object; | |
1411 | DB_DNODE_EXIT(dbuf); | |
1412 | ||
9babb374 | 1413 | dbuf_rele(db, FTAG); |
572e2857 | 1414 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1415 | dmu_return_arcbuf(buf); |
428870ff | 1416 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 BB |
1417 | } |
1418 | } | |
1419 | ||
34dc7c2f | 1420 | typedef struct { |
428870ff BB |
1421 | dbuf_dirty_record_t *dsa_dr; |
1422 | dmu_sync_cb_t *dsa_done; | |
1423 | zgd_t *dsa_zgd; | |
1424 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1425 | } dmu_sync_arg_t; |
1426 | ||
b128c09f BB |
1427 | /* ARGSUSED */ |
1428 | static void | |
1429 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1430 | { | |
428870ff BB |
1431 | dmu_sync_arg_t *dsa = varg; |
1432 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1433 | blkptr_t *bp = zio->io_bp; |
1434 | ||
428870ff BB |
1435 | if (zio->io_error == 0) { |
1436 | if (BP_IS_HOLE(bp)) { | |
1437 | /* | |
1438 | * A block of zeros may compress to a hole, but the | |
1439 | * block size still needs to be known for replay. | |
1440 | */ | |
1441 | BP_SET_LSIZE(bp, db->db_size); | |
9b67f605 | 1442 | } else if (!BP_IS_EMBEDDED(bp)) { |
428870ff BB |
1443 | ASSERT(BP_GET_LEVEL(bp) == 0); |
1444 | bp->blk_fill = 1; | |
1445 | } | |
b128c09f BB |
1446 | } |
1447 | } | |
1448 | ||
428870ff BB |
1449 | static void |
1450 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1451 | { | |
1452 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1453 | } | |
1454 | ||
34dc7c2f BB |
1455 | /* ARGSUSED */ |
1456 | static void | |
1457 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1458 | { | |
428870ff BB |
1459 | dmu_sync_arg_t *dsa = varg; |
1460 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1461 | dmu_buf_impl_t *db = dr->dr_dbuf; |
34dc7c2f | 1462 | |
34dc7c2f BB |
1463 | mutex_enter(&db->db_mtx); |
1464 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff | 1465 | if (zio->io_error == 0) { |
03c6040b GW |
1466 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); |
1467 | if (dr->dt.dl.dr_nopwrite) { | |
1468 | ASSERTV(blkptr_t *bp = zio->io_bp); | |
1469 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); | |
1470 | ASSERTV(uint8_t chksum = BP_GET_CHECKSUM(bp_orig)); | |
1471 | ||
1472 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
1473 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); | |
1474 | ASSERT(zio_checksum_table[chksum].ci_dedup); | |
1475 | } | |
428870ff BB |
1476 | dr->dt.dl.dr_overridden_by = *zio->io_bp; |
1477 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1478 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
a4069eef PS |
1479 | |
1480 | /* | |
1481 | * Old style holes are filled with all zeros, whereas | |
1482 | * new-style holes maintain their lsize, type, level, | |
1483 | * and birth time (see zio_write_compress). While we | |
1484 | * need to reset the BP_SET_LSIZE() call that happened | |
1485 | * in dmu_sync_ready for old style holes, we do *not* | |
1486 | * want to wipe out the information contained in new | |
1487 | * style holes. Thus, only zero out the block pointer if | |
1488 | * it's an old style hole. | |
1489 | */ | |
1490 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1491 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
428870ff BB |
1492 | BP_ZERO(&dr->dt.dl.dr_overridden_by); |
1493 | } else { | |
1494 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1495 | } | |
34dc7c2f BB |
1496 | cv_broadcast(&db->db_changed); |
1497 | mutex_exit(&db->db_mtx); | |
1498 | ||
428870ff | 1499 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1500 | |
428870ff BB |
1501 | kmem_free(dsa, sizeof (*dsa)); |
1502 | } | |
1503 | ||
1504 | static void | |
1505 | dmu_sync_late_arrival_done(zio_t *zio) | |
1506 | { | |
1507 | blkptr_t *bp = zio->io_bp; | |
1508 | dmu_sync_arg_t *dsa = zio->io_private; | |
03c6040b | 1509 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); |
428870ff BB |
1510 | |
1511 | if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { | |
03c6040b GW |
1512 | /* |
1513 | * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) | |
1514 | * then there is nothing to do here. Otherwise, free the | |
1515 | * newly allocated block in this txg. | |
1516 | */ | |
1517 | if (zio->io_flags & ZIO_FLAG_NOPWRITE) { | |
1518 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
1519 | } else { | |
1520 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1521 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1522 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1523 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1524 | } | |
428870ff BB |
1525 | } |
1526 | ||
1527 | dmu_tx_commit(dsa->dsa_tx); | |
1528 | ||
1529 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1530 | ||
1531 | kmem_free(dsa, sizeof (*dsa)); | |
1532 | } | |
1533 | ||
1534 | static int | |
1535 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
5dbd68a3 | 1536 | zio_prop_t *zp, zbookmark_phys_t *zb) |
428870ff BB |
1537 | { |
1538 | dmu_sync_arg_t *dsa; | |
1539 | dmu_tx_t *tx; | |
1540 | ||
1541 | tx = dmu_tx_create(os); | |
1542 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1543 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1544 | dmu_tx_abort(tx); | |
2e528b49 MA |
1545 | /* Make zl_get_data do txg_waited_synced() */ |
1546 | return (SET_ERROR(EIO)); | |
428870ff BB |
1547 | } |
1548 | ||
79c76d5b | 1549 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1550 | dsa->dsa_dr = NULL; |
1551 | dsa->dsa_done = done; | |
1552 | dsa->dsa_zgd = zgd; | |
1553 | dsa->dsa_tx = tx; | |
1554 | ||
1555 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, | |
1556 | zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, | |
e8b96c60 | 1557 | dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa, |
d1d7e268 | 1558 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL|ZIO_FLAG_FASTWRITE, zb)); |
428870ff BB |
1559 | |
1560 | return (0); | |
34dc7c2f BB |
1561 | } |
1562 | ||
1563 | /* | |
1564 | * Intent log support: sync the block associated with db to disk. | |
1565 | * N.B. and XXX: the caller is responsible for making sure that the | |
1566 | * data isn't changing while dmu_sync() is writing it. | |
1567 | * | |
1568 | * Return values: | |
1569 | * | |
03c6040b | 1570 | * EEXIST: this txg has already been synced, so there's nothing to do. |
34dc7c2f BB |
1571 | * The caller should not log the write. |
1572 | * | |
1573 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1574 | * The caller should not log the write. | |
1575 | * | |
1576 | * EALREADY: this block is already in the process of being synced. | |
1577 | * The caller should track its progress (somehow). | |
1578 | * | |
428870ff BB |
1579 | * EIO: could not do the I/O. |
1580 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1581 | * |
428870ff BB |
1582 | * 0: the I/O has been initiated. |
1583 | * The caller should log this blkptr in the done callback. | |
1584 | * It is possible that the I/O will fail, in which case | |
1585 | * the error will be reported to the done callback and | |
1586 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1587 | */ |
1588 | int | |
428870ff | 1589 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1590 | { |
428870ff BB |
1591 | blkptr_t *bp = zgd->zgd_bp; |
1592 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; | |
1593 | objset_t *os = db->db_objset; | |
1594 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1595 | dbuf_dirty_record_t *dr; |
428870ff | 1596 | dmu_sync_arg_t *dsa; |
5dbd68a3 | 1597 | zbookmark_phys_t zb; |
428870ff | 1598 | zio_prop_t zp; |
572e2857 | 1599 | dnode_t *dn; |
34dc7c2f | 1600 | |
428870ff | 1601 | ASSERT(pio != NULL); |
34dc7c2f BB |
1602 | ASSERT(txg != 0); |
1603 | ||
428870ff BB |
1604 | SET_BOOKMARK(&zb, ds->ds_object, |
1605 | db->db.db_object, db->db_level, db->db_blkid); | |
1606 | ||
572e2857 BB |
1607 | DB_DNODE_ENTER(db); |
1608 | dn = DB_DNODE(db); | |
1609 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); | |
1610 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
1611 | |
1612 | /* | |
428870ff | 1613 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1614 | */ |
428870ff BB |
1615 | if (txg > spa_freeze_txg(os->os_spa)) |
1616 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1617 | |
1618 | /* | |
428870ff BB |
1619 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1620 | * and us. If we determine that this txg is not yet syncing, | |
1621 | * but it begins to sync a moment later, that's OK because the | |
1622 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1623 | */ |
428870ff BB |
1624 | mutex_enter(&db->db_mtx); |
1625 | ||
1626 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1627 | /* |
428870ff | 1628 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1629 | */ |
428870ff | 1630 | mutex_exit(&db->db_mtx); |
2e528b49 | 1631 | return (SET_ERROR(EEXIST)); |
34dc7c2f BB |
1632 | } |
1633 | ||
428870ff BB |
1634 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1635 | /* | |
1636 | * This txg is currently syncing, so we can't mess with | |
1637 | * the dirty record anymore; just write a new log block. | |
1638 | */ | |
1639 | mutex_exit(&db->db_mtx); | |
1640 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1641 | } |
1642 | ||
1643 | dr = db->db_last_dirty; | |
428870ff | 1644 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 1645 | dr = dr->dr_next; |
428870ff BB |
1646 | |
1647 | if (dr == NULL) { | |
34dc7c2f | 1648 | /* |
428870ff | 1649 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1650 | * There's no need to log writes to freed blocks, so we're done. |
1651 | */ | |
1652 | mutex_exit(&db->db_mtx); | |
2e528b49 | 1653 | return (SET_ERROR(ENOENT)); |
34dc7c2f BB |
1654 | } |
1655 | ||
03c6040b GW |
1656 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); |
1657 | ||
1658 | /* | |
f3c517d8 MA |
1659 | * Assume the on-disk data is X, the current syncing data (in |
1660 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1661 | * in dmu_sync). | |
1662 | * | |
1663 | * We usually want to perform a nopwrite if X and Z are the | |
1664 | * same. However, if Y is different (i.e. the BP is going to | |
1665 | * change before this write takes effect), then a nopwrite will | |
1666 | * be incorrect - we would override with X, which could have | |
1667 | * been freed when Y was written. | |
1668 | * | |
1669 | * (Note that this is not a concern when we are nop-writing from | |
1670 | * syncing context, because X and Y must be identical, because | |
1671 | * all previous txgs have been synced.) | |
1672 | * | |
1673 | * Therefore, we disable nopwrite if the current BP could change | |
1674 | * before this TXG. There are two ways it could change: by | |
1675 | * being dirty (dr_next is non-NULL), or by being freed | |
1676 | * (dnode_block_freed()). This behavior is verified by | |
1677 | * zio_done(), which VERIFYs that the override BP is identical | |
1678 | * to the on-disk BP. | |
03c6040b | 1679 | */ |
f3c517d8 MA |
1680 | DB_DNODE_ENTER(db); |
1681 | dn = DB_DNODE(db); | |
1682 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
03c6040b | 1683 | zp.zp_nopwrite = B_FALSE; |
f3c517d8 | 1684 | DB_DNODE_EXIT(db); |
03c6040b | 1685 | |
34dc7c2f | 1686 | ASSERT(dr->dr_txg == txg); |
428870ff BB |
1687 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1688 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1689 | /* |
428870ff BB |
1690 | * We have already issued a sync write for this buffer, |
1691 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1692 | * have been dirtied since, or we would have cleared the state. |
1693 | */ | |
34dc7c2f | 1694 | mutex_exit(&db->db_mtx); |
2e528b49 | 1695 | return (SET_ERROR(EALREADY)); |
34dc7c2f BB |
1696 | } |
1697 | ||
428870ff | 1698 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1699 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1700 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1701 | |
79c76d5b | 1702 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
428870ff BB |
1703 | dsa->dsa_dr = dr; |
1704 | dsa->dsa_done = done; | |
1705 | dsa->dsa_zgd = zgd; | |
1706 | dsa->dsa_tx = NULL; | |
b128c09f | 1707 | |
428870ff | 1708 | zio_nowait(arc_write(pio, os->os_spa, txg, |
3a17a7a9 | 1709 | bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
e8b96c60 MA |
1710 | DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, |
1711 | NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE, | |
1712 | ZIO_FLAG_CANFAIL, &zb)); | |
b128c09f | 1713 | |
428870ff | 1714 | return (0); |
34dc7c2f BB |
1715 | } |
1716 | ||
1717 | int | |
1718 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
1719 | dmu_tx_t *tx) | |
1720 | { | |
1721 | dnode_t *dn; | |
1722 | int err; | |
1723 | ||
428870ff | 1724 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1725 | if (err) |
1726 | return (err); | |
1727 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1728 | dnode_rele(dn, FTAG); | |
1729 | return (err); | |
1730 | } | |
1731 | ||
1732 | void | |
1733 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
1734 | dmu_tx_t *tx) | |
1735 | { | |
1736 | dnode_t *dn; | |
1737 | ||
9b67f605 MA |
1738 | /* |
1739 | * Send streams include each object's checksum function. This | |
1740 | * check ensures that the receiving system can understand the | |
1741 | * checksum function transmitted. | |
1742 | */ | |
1743 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1744 | ||
1745 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1746 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
34dc7c2f BB |
1747 | dn->dn_checksum = checksum; |
1748 | dnode_setdirty(dn, tx); | |
1749 | dnode_rele(dn, FTAG); | |
1750 | } | |
1751 | ||
1752 | void | |
1753 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
1754 | dmu_tx_t *tx) | |
1755 | { | |
1756 | dnode_t *dn; | |
1757 | ||
9b67f605 MA |
1758 | /* |
1759 | * Send streams include each object's compression function. This | |
1760 | * check ensures that the receiving system can understand the | |
1761 | * compression function transmitted. | |
1762 | */ | |
1763 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1764 | ||
1765 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
34dc7c2f BB |
1766 | dn->dn_compress = compress; |
1767 | dnode_setdirty(dn, tx); | |
1768 | dnode_rele(dn, FTAG); | |
1769 | } | |
1770 | ||
428870ff BB |
1771 | int zfs_mdcomp_disable = 0; |
1772 | ||
faf0f58c MA |
1773 | /* |
1774 | * When the "redundant_metadata" property is set to "most", only indirect | |
1775 | * blocks of this level and higher will have an additional ditto block. | |
1776 | */ | |
1777 | int zfs_redundant_metadata_most_ditto_level = 2; | |
1778 | ||
428870ff BB |
1779 | void |
1780 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) | |
1781 | { | |
1782 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 1783 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 1784 | (wp & WP_SPILL)); |
428870ff BB |
1785 | enum zio_checksum checksum = os->os_checksum; |
1786 | enum zio_compress compress = os->os_compress; | |
1787 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
03c6040b GW |
1788 | boolean_t dedup = B_FALSE; |
1789 | boolean_t nopwrite = B_FALSE; | |
428870ff BB |
1790 | boolean_t dedup_verify = os->os_dedup_verify; |
1791 | int copies = os->os_copies; | |
1792 | ||
1793 | /* | |
03c6040b GW |
1794 | * We maintain different write policies for each of the following |
1795 | * types of data: | |
1796 | * 1. metadata | |
1797 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
1798 | * 3. all other level 0 blocks | |
428870ff BB |
1799 | */ |
1800 | if (ismd) { | |
03c6040b GW |
1801 | /* |
1802 | * XXX -- we should design a compression algorithm | |
1803 | * that specializes in arrays of bps. | |
1804 | */ | |
62bdd5eb DL |
1805 | boolean_t lz4_ac = spa_feature_is_active(os->os_spa, |
1806 | SPA_FEATURE_LZ4_COMPRESS); | |
1807 | ||
1808 | if (zfs_mdcomp_disable) { | |
1809 | compress = ZIO_COMPRESS_EMPTY; | |
1810 | } else if (lz4_ac) { | |
1811 | compress = ZIO_COMPRESS_LZ4; | |
1812 | } else { | |
1813 | compress = ZIO_COMPRESS_LZJB; | |
1814 | } | |
03c6040b | 1815 | |
428870ff BB |
1816 | /* |
1817 | * Metadata always gets checksummed. If the data | |
1818 | * checksum is multi-bit correctable, and it's not a | |
1819 | * ZBT-style checksum, then it's suitable for metadata | |
1820 | * as well. Otherwise, the metadata checksum defaults | |
1821 | * to fletcher4. | |
1822 | */ | |
1823 | if (zio_checksum_table[checksum].ci_correctable < 1 || | |
1824 | zio_checksum_table[checksum].ci_eck) | |
1825 | checksum = ZIO_CHECKSUM_FLETCHER_4; | |
faf0f58c MA |
1826 | |
1827 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
1828 | (os->os_redundant_metadata == | |
1829 | ZFS_REDUNDANT_METADATA_MOST && | |
1830 | (level >= zfs_redundant_metadata_most_ditto_level || | |
1831 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
1832 | copies++; | |
03c6040b GW |
1833 | } else if (wp & WP_NOFILL) { |
1834 | ASSERT(level == 0); | |
428870ff | 1835 | |
428870ff | 1836 | /* |
03c6040b GW |
1837 | * If we're writing preallocated blocks, we aren't actually |
1838 | * writing them so don't set any policy properties. These | |
1839 | * blocks are currently only used by an external subsystem | |
1840 | * outside of zfs (i.e. dump) and not written by the zio | |
1841 | * pipeline. | |
428870ff | 1842 | */ |
03c6040b GW |
1843 | compress = ZIO_COMPRESS_OFF; |
1844 | checksum = ZIO_CHECKSUM_OFF; | |
428870ff BB |
1845 | } else { |
1846 | compress = zio_compress_select(dn->dn_compress, compress); | |
428870ff | 1847 | |
03c6040b GW |
1848 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? |
1849 | zio_checksum_select(dn->dn_checksum, checksum) : | |
1850 | dedup_checksum; | |
428870ff | 1851 | |
03c6040b GW |
1852 | /* |
1853 | * Determine dedup setting. If we are in dmu_sync(), | |
1854 | * we won't actually dedup now because that's all | |
1855 | * done in syncing context; but we do want to use the | |
1856 | * dedup checkum. If the checksum is not strong | |
1857 | * enough to ensure unique signatures, force | |
1858 | * dedup_verify. | |
1859 | */ | |
1860 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
1861 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
1862 | if (!zio_checksum_table[checksum].ci_dedup) | |
1863 | dedup_verify = B_TRUE; | |
1864 | } | |
428870ff | 1865 | |
03c6040b GW |
1866 | /* |
1867 | * Enable nopwrite if we have a cryptographically secure | |
1868 | * checksum that has no known collisions (i.e. SHA-256) | |
1869 | * and compression is enabled. We don't enable nopwrite if | |
1870 | * dedup is enabled as the two features are mutually exclusive. | |
1871 | */ | |
1872 | nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup && | |
1873 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); | |
428870ff BB |
1874 | } |
1875 | ||
1876 | zp->zp_checksum = checksum; | |
1877 | zp->zp_compress = compress; | |
1878 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; | |
1879 | zp->zp_level = level; | |
faf0f58c | 1880 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); |
428870ff BB |
1881 | zp->zp_dedup = dedup; |
1882 | zp->zp_dedup_verify = dedup && dedup_verify; | |
03c6040b | 1883 | zp->zp_nopwrite = nopwrite; |
428870ff BB |
1884 | } |
1885 | ||
34dc7c2f BB |
1886 | int |
1887 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
1888 | { | |
1889 | dnode_t *dn; | |
1890 | int i, err; | |
1891 | ||
428870ff | 1892 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1893 | if (err) |
1894 | return (err); | |
1895 | /* | |
1896 | * Sync any current changes before | |
1897 | * we go trundling through the block pointers. | |
1898 | */ | |
1899 | for (i = 0; i < TXG_SIZE; i++) { | |
1900 | if (list_link_active(&dn->dn_dirty_link[i])) | |
1901 | break; | |
1902 | } | |
1903 | if (i != TXG_SIZE) { | |
1904 | dnode_rele(dn, FTAG); | |
1905 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 1906 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1907 | if (err) |
1908 | return (err); | |
1909 | } | |
1910 | ||
b128c09f | 1911 | err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); |
34dc7c2f BB |
1912 | dnode_rele(dn, FTAG); |
1913 | ||
1914 | return (err); | |
1915 | } | |
1916 | ||
1917 | void | |
e0b0ca98 | 1918 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) |
34dc7c2f | 1919 | { |
e0b0ca98 | 1920 | dnode_phys_t *dnp = dn->dn_phys; |
d6320ddb | 1921 | int i; |
428870ff | 1922 | |
34dc7c2f BB |
1923 | doi->doi_data_block_size = dn->dn_datablksz; |
1924 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
1925 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
1926 | doi->doi_type = dn->dn_type; |
1927 | doi->doi_bonus_type = dn->dn_bonustype; | |
1928 | doi->doi_bonus_size = dn->dn_bonuslen; | |
34dc7c2f BB |
1929 | doi->doi_indirection = dn->dn_nlevels; |
1930 | doi->doi_checksum = dn->dn_checksum; | |
1931 | doi->doi_compress = dn->dn_compress; | |
6c59307a | 1932 | doi->doi_nblkptr = dn->dn_nblkptr; |
428870ff | 1933 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
d1fada1e | 1934 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; |
428870ff | 1935 | doi->doi_fill_count = 0; |
d6320ddb | 1936 | for (i = 0; i < dnp->dn_nblkptr; i++) |
9b67f605 | 1937 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); |
e0b0ca98 BB |
1938 | } |
1939 | ||
1940 | void | |
1941 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
1942 | { | |
1943 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
1944 | mutex_enter(&dn->dn_mtx); | |
1945 | ||
1946 | __dmu_object_info_from_dnode(dn, doi); | |
34dc7c2f BB |
1947 | |
1948 | mutex_exit(&dn->dn_mtx); | |
1949 | rw_exit(&dn->dn_struct_rwlock); | |
1950 | } | |
1951 | ||
1952 | /* | |
1953 | * Get information on a DMU object. | |
1954 | * If doi is NULL, just indicates whether the object exists. | |
1955 | */ | |
1956 | int | |
1957 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
1958 | { | |
1959 | dnode_t *dn; | |
428870ff | 1960 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1961 | |
1962 | if (err) | |
1963 | return (err); | |
1964 | ||
1965 | if (doi != NULL) | |
1966 | dmu_object_info_from_dnode(dn, doi); | |
1967 | ||
1968 | dnode_rele(dn, FTAG); | |
1969 | return (0); | |
1970 | } | |
1971 | ||
1972 | /* | |
1973 | * As above, but faster; can be used when you have a held dbuf in hand. | |
1974 | */ | |
1975 | void | |
572e2857 | 1976 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 1977 | { |
572e2857 BB |
1978 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1979 | ||
1980 | DB_DNODE_ENTER(db); | |
1981 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
1982 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
1983 | } |
1984 | ||
1985 | /* | |
1986 | * Faster still when you only care about the size. | |
1987 | * This is specifically optimized for zfs_getattr(). | |
1988 | */ | |
1989 | void | |
572e2857 BB |
1990 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
1991 | u_longlong_t *nblk512) | |
34dc7c2f | 1992 | { |
572e2857 BB |
1993 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1994 | dnode_t *dn; | |
1995 | ||
1996 | DB_DNODE_ENTER(db); | |
1997 | dn = DB_DNODE(db); | |
34dc7c2f BB |
1998 | |
1999 | *blksize = dn->dn_datablksz; | |
2000 | /* add 1 for dnode space */ | |
2001 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> | |
2002 | SPA_MINBLOCKSHIFT) + 1; | |
572e2857 | 2003 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
2004 | } |
2005 | ||
2006 | void | |
2007 | byteswap_uint64_array(void *vbuf, size_t size) | |
2008 | { | |
2009 | uint64_t *buf = vbuf; | |
2010 | size_t count = size >> 3; | |
2011 | int i; | |
2012 | ||
2013 | ASSERT((size & 7) == 0); | |
2014 | ||
2015 | for (i = 0; i < count; i++) | |
2016 | buf[i] = BSWAP_64(buf[i]); | |
2017 | } | |
2018 | ||
2019 | void | |
2020 | byteswap_uint32_array(void *vbuf, size_t size) | |
2021 | { | |
2022 | uint32_t *buf = vbuf; | |
2023 | size_t count = size >> 2; | |
2024 | int i; | |
2025 | ||
2026 | ASSERT((size & 3) == 0); | |
2027 | ||
2028 | for (i = 0; i < count; i++) | |
2029 | buf[i] = BSWAP_32(buf[i]); | |
2030 | } | |
2031 | ||
2032 | void | |
2033 | byteswap_uint16_array(void *vbuf, size_t size) | |
2034 | { | |
2035 | uint16_t *buf = vbuf; | |
2036 | size_t count = size >> 1; | |
2037 | int i; | |
2038 | ||
2039 | ASSERT((size & 1) == 0); | |
2040 | ||
2041 | for (i = 0; i < count; i++) | |
2042 | buf[i] = BSWAP_16(buf[i]); | |
2043 | } | |
2044 | ||
2045 | /* ARGSUSED */ | |
2046 | void | |
2047 | byteswap_uint8_array(void *vbuf, size_t size) | |
2048 | { | |
2049 | } | |
2050 | ||
2051 | void | |
2052 | dmu_init(void) | |
2053 | { | |
428870ff | 2054 | zfs_dbgmsg_init(); |
572e2857 BB |
2055 | sa_cache_init(); |
2056 | xuio_stat_init(); | |
2057 | dmu_objset_init(); | |
34dc7c2f | 2058 | dnode_init(); |
572e2857 | 2059 | dbuf_init(); |
428870ff | 2060 | zfetch_init(); |
570827e1 | 2061 | dmu_tx_init(); |
34dc7c2f | 2062 | l2arc_init(); |
29809a6c | 2063 | arc_init(); |
34dc7c2f BB |
2064 | } |
2065 | ||
2066 | void | |
2067 | dmu_fini(void) | |
2068 | { | |
e49f1e20 | 2069 | arc_fini(); /* arc depends on l2arc, so arc must go first */ |
29809a6c | 2070 | l2arc_fini(); |
570827e1 | 2071 | dmu_tx_fini(); |
428870ff | 2072 | zfetch_fini(); |
34dc7c2f | 2073 | dbuf_fini(); |
572e2857 BB |
2074 | dnode_fini(); |
2075 | dmu_objset_fini(); | |
428870ff BB |
2076 | xuio_stat_fini(); |
2077 | sa_cache_fini(); | |
2078 | zfs_dbgmsg_fini(); | |
34dc7c2f | 2079 | } |
c28b2279 BB |
2080 | |
2081 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
2082 | EXPORT_SYMBOL(dmu_bonus_hold); | |
a473d90c AZ |
2083 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
2084 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 2085 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 2086 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 | 2087 | EXPORT_SYMBOL(dmu_free_long_range); |
b663a23d | 2088 | EXPORT_SYMBOL(dmu_free_long_object); |
c28b2279 BB |
2089 | EXPORT_SYMBOL(dmu_read); |
2090 | EXPORT_SYMBOL(dmu_write); | |
57b650b8 | 2091 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
2092 | EXPORT_SYMBOL(dmu_object_info); |
2093 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2094 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2095 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
c28b2279 BB |
2096 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
2097 | EXPORT_SYMBOL(dmu_object_set_checksum); | |
2098 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
2099 | EXPORT_SYMBOL(dmu_write_policy); |
2100 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
2101 | EXPORT_SYMBOL(dmu_request_arcbuf); |
2102 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
2103 | EXPORT_SYMBOL(dmu_assign_arcbuf); | |
2104 | EXPORT_SYMBOL(dmu_buf_hold); | |
c28b2279 | 2105 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 BB |
2106 | |
2107 | module_param(zfs_mdcomp_disable, int, 0644); | |
2108 | MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression"); | |
03c6040b GW |
2109 | |
2110 | module_param(zfs_nopwrite_enabled, int, 0644); | |
2111 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2112 | ||
c28b2279 | 2113 | #endif |