]>
Commit | Line | Data |
---|---|---|
70e083d2 TG |
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 | /* | |
22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | * Copyright (c) 2011, 2014 by Delphix. All rights reserved. | |
24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. | |
25 | * Copyright (c) 2014, Nexenta Systems, Inc. All rights reserved. | |
26 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. | |
27 | */ | |
28 | ||
29 | #include <sys/dmu.h> | |
30 | #include <sys/dmu_impl.h> | |
31 | #include <sys/dmu_tx.h> | |
32 | #include <sys/dbuf.h> | |
33 | #include <sys/dnode.h> | |
34 | #include <sys/zfs_context.h> | |
35 | #include <sys/dmu_objset.h> | |
36 | #include <sys/dmu_traverse.h> | |
37 | #include <sys/dsl_dataset.h> | |
38 | #include <sys/dsl_dir.h> | |
39 | #include <sys/dsl_pool.h> | |
40 | #include <sys/dsl_synctask.h> | |
41 | #include <sys/dsl_prop.h> | |
42 | #include <sys/dmu_zfetch.h> | |
43 | #include <sys/zfs_ioctl.h> | |
44 | #include <sys/zap.h> | |
45 | #include <sys/zio_checksum.h> | |
46 | #include <sys/zio_compress.h> | |
47 | #include <sys/sa.h> | |
48 | #include <sys/zfeature.h> | |
49 | #ifdef _KERNEL | |
50 | #include <sys/vmsystm.h> | |
51 | #include <sys/zfs_znode.h> | |
9784fa9e | 52 | #include <linux/kmap_compat.h> |
70e083d2 TG |
53 | #endif |
54 | ||
55 | /* | |
56 | * Enable/disable nopwrite feature. | |
57 | */ | |
58 | int zfs_nopwrite_enabled = 1; | |
59 | ||
60 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { | |
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" } | |
128 | }; | |
129 | ||
130 | int | |
131 | dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, | |
132 | void *tag, dmu_buf_t **dbp) | |
133 | { | |
134 | dnode_t *dn; | |
135 | uint64_t blkid; | |
136 | dmu_buf_impl_t *db; | |
137 | int err; | |
138 | ||
139 | err = dnode_hold(os, object, FTAG, &dn); | |
140 | if (err) | |
141 | return (err); | |
142 | blkid = dbuf_whichblock(dn, offset); | |
143 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
144 | db = dbuf_hold(dn, blkid, tag); | |
145 | rw_exit(&dn->dn_struct_rwlock); | |
146 | dnode_rele(dn, FTAG); | |
147 | ||
148 | if (db == NULL) { | |
149 | *dbp = NULL; | |
150 | return (SET_ERROR(EIO)); | |
151 | } | |
152 | ||
153 | *dbp = &db->db; | |
154 | return (err); | |
155 | } | |
156 | ||
157 | int | |
158 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
159 | void *tag, dmu_buf_t **dbp, int flags) | |
160 | { | |
161 | int err; | |
162 | int db_flags = DB_RF_CANFAIL; | |
163 | ||
164 | if (flags & DMU_READ_NO_PREFETCH) | |
165 | db_flags |= DB_RF_NOPREFETCH; | |
166 | ||
167 | err = dmu_buf_hold_noread(os, object, offset, tag, dbp); | |
168 | if (err == 0) { | |
169 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); | |
170 | err = dbuf_read(db, NULL, db_flags); | |
171 | if (err != 0) { | |
172 | dbuf_rele(db, tag); | |
173 | *dbp = NULL; | |
174 | } | |
175 | } | |
176 | ||
177 | return (err); | |
178 | } | |
179 | ||
180 | int | |
181 | dmu_bonus_max(void) | |
182 | { | |
183 | return (DN_MAX_BONUSLEN); | |
184 | } | |
185 | ||
186 | int | |
187 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) | |
188 | { | |
189 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
190 | dnode_t *dn; | |
191 | int error; | |
192 | ||
193 | DB_DNODE_ENTER(db); | |
194 | dn = DB_DNODE(db); | |
195 | ||
196 | if (dn->dn_bonus != db) { | |
197 | error = SET_ERROR(EINVAL); | |
198 | } else if (newsize < 0 || newsize > db_fake->db_size) { | |
199 | error = SET_ERROR(EINVAL); | |
200 | } else { | |
201 | dnode_setbonuslen(dn, newsize, tx); | |
202 | error = 0; | |
203 | } | |
204 | ||
205 | DB_DNODE_EXIT(db); | |
206 | return (error); | |
207 | } | |
208 | ||
209 | int | |
210 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) | |
211 | { | |
212 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
213 | dnode_t *dn; | |
214 | int error; | |
215 | ||
216 | DB_DNODE_ENTER(db); | |
217 | dn = DB_DNODE(db); | |
218 | ||
219 | if (!DMU_OT_IS_VALID(type)) { | |
220 | error = SET_ERROR(EINVAL); | |
221 | } else if (dn->dn_bonus != db) { | |
222 | error = SET_ERROR(EINVAL); | |
223 | } else { | |
224 | dnode_setbonus_type(dn, type, tx); | |
225 | error = 0; | |
226 | } | |
227 | ||
228 | DB_DNODE_EXIT(db); | |
229 | return (error); | |
230 | } | |
231 | ||
232 | dmu_object_type_t | |
233 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
234 | { | |
235 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
236 | dnode_t *dn; | |
237 | dmu_object_type_t type; | |
238 | ||
239 | DB_DNODE_ENTER(db); | |
240 | dn = DB_DNODE(db); | |
241 | type = dn->dn_bonustype; | |
242 | DB_DNODE_EXIT(db); | |
243 | ||
244 | return (type); | |
245 | } | |
246 | ||
247 | int | |
248 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
249 | { | |
250 | dnode_t *dn; | |
251 | int error; | |
252 | ||
253 | error = dnode_hold(os, object, FTAG, &dn); | |
254 | dbuf_rm_spill(dn, tx); | |
255 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
256 | dnode_rm_spill(dn, tx); | |
257 | rw_exit(&dn->dn_struct_rwlock); | |
258 | dnode_rele(dn, FTAG); | |
259 | return (error); | |
260 | } | |
261 | ||
262 | /* | |
263 | * returns ENOENT, EIO, or 0. | |
264 | */ | |
265 | int | |
266 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) | |
267 | { | |
268 | dnode_t *dn; | |
269 | dmu_buf_impl_t *db; | |
270 | int error; | |
271 | ||
272 | error = dnode_hold(os, object, FTAG, &dn); | |
273 | if (error) | |
274 | return (error); | |
275 | ||
276 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
277 | if (dn->dn_bonus == NULL) { | |
278 | rw_exit(&dn->dn_struct_rwlock); | |
279 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
280 | if (dn->dn_bonus == NULL) | |
281 | dbuf_create_bonus(dn); | |
282 | } | |
283 | db = dn->dn_bonus; | |
284 | ||
285 | /* as long as the bonus buf is held, the dnode will be held */ | |
286 | if (refcount_add(&db->db_holds, tag) == 1) { | |
287 | VERIFY(dnode_add_ref(dn, db)); | |
288 | atomic_inc_32(&dn->dn_dbufs_count); | |
289 | } | |
290 | ||
291 | /* | |
292 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
293 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
294 | * a dnode hold for every dbuf. | |
295 | */ | |
296 | rw_exit(&dn->dn_struct_rwlock); | |
297 | ||
298 | dnode_rele(dn, FTAG); | |
299 | ||
300 | VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); | |
301 | ||
302 | *dbp = &db->db; | |
303 | return (0); | |
304 | } | |
305 | ||
306 | /* | |
307 | * returns ENOENT, EIO, or 0. | |
308 | * | |
309 | * This interface will allocate a blank spill dbuf when a spill blk | |
310 | * doesn't already exist on the dnode. | |
311 | * | |
312 | * if you only want to find an already existing spill db, then | |
313 | * dmu_spill_hold_existing() should be used. | |
314 | */ | |
315 | int | |
316 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
317 | { | |
318 | dmu_buf_impl_t *db = NULL; | |
319 | int err; | |
320 | ||
321 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
322 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
323 | ||
324 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
325 | ||
326 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
327 | rw_exit(&dn->dn_struct_rwlock); | |
328 | ||
329 | ASSERT(db != NULL); | |
330 | err = dbuf_read(db, NULL, flags); | |
331 | if (err == 0) | |
332 | *dbp = &db->db; | |
333 | else | |
334 | dbuf_rele(db, tag); | |
335 | return (err); | |
336 | } | |
337 | ||
338 | int | |
339 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
340 | { | |
341 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; | |
342 | dnode_t *dn; | |
343 | int err; | |
344 | ||
345 | DB_DNODE_ENTER(db); | |
346 | dn = DB_DNODE(db); | |
347 | ||
348 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
349 | err = SET_ERROR(EINVAL); | |
350 | } else { | |
351 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
352 | ||
353 | if (!dn->dn_have_spill) { | |
354 | err = SET_ERROR(ENOENT); | |
355 | } else { | |
356 | err = dmu_spill_hold_by_dnode(dn, | |
357 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
358 | } | |
359 | ||
360 | rw_exit(&dn->dn_struct_rwlock); | |
361 | } | |
362 | ||
363 | DB_DNODE_EXIT(db); | |
364 | return (err); | |
365 | } | |
366 | ||
367 | int | |
368 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
369 | { | |
370 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; | |
371 | dnode_t *dn; | |
372 | int err; | |
373 | ||
374 | DB_DNODE_ENTER(db); | |
375 | dn = DB_DNODE(db); | |
376 | err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); | |
377 | DB_DNODE_EXIT(db); | |
378 | ||
379 | return (err); | |
380 | } | |
381 | ||
382 | /* | |
383 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
384 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
385 | * and can induce severe lock contention when writing to several files | |
386 | * whose dnodes are in the same block. | |
387 | */ | |
388 | static int | |
389 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, | |
390 | int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) | |
391 | { | |
392 | dmu_buf_t **dbp; | |
393 | uint64_t blkid, nblks, i; | |
394 | uint32_t dbuf_flags; | |
395 | int err; | |
396 | zio_t *zio; | |
397 | ||
398 | ASSERT(length <= DMU_MAX_ACCESS); | |
399 | ||
400 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; | |
401 | if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) | |
402 | dbuf_flags |= DB_RF_NOPREFETCH; | |
403 | ||
404 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
405 | if (dn->dn_datablkshift) { | |
406 | int blkshift = dn->dn_datablkshift; | |
407 | nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - | |
408 | P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; | |
409 | } else { | |
410 | if (offset + length > dn->dn_datablksz) { | |
411 | zfs_panic_recover("zfs: accessing past end of object " | |
412 | "%llx/%llx (size=%u access=%llu+%llu)", | |
413 | (longlong_t)dn->dn_objset-> | |
414 | os_dsl_dataset->ds_object, | |
415 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
416 | (longlong_t)offset, (longlong_t)length); | |
417 | rw_exit(&dn->dn_struct_rwlock); | |
418 | return (SET_ERROR(EIO)); | |
419 | } | |
420 | nblks = 1; | |
421 | } | |
422 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); | |
423 | ||
424 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); | |
425 | blkid = dbuf_whichblock(dn, offset); | |
426 | for (i = 0; i < nblks; i++) { | |
427 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); | |
428 | if (db == NULL) { | |
429 | rw_exit(&dn->dn_struct_rwlock); | |
430 | dmu_buf_rele_array(dbp, nblks, tag); | |
431 | zio_nowait(zio); | |
432 | return (SET_ERROR(EIO)); | |
433 | } | |
434 | /* initiate async i/o */ | |
435 | if (read) { | |
436 | (void) dbuf_read(db, zio, dbuf_flags); | |
437 | } | |
438 | dbp[i] = &db->db; | |
439 | } | |
440 | rw_exit(&dn->dn_struct_rwlock); | |
441 | ||
442 | /* wait for async i/o */ | |
443 | err = zio_wait(zio); | |
444 | if (err) { | |
445 | dmu_buf_rele_array(dbp, nblks, tag); | |
446 | return (err); | |
447 | } | |
448 | ||
449 | /* wait for other io to complete */ | |
450 | if (read) { | |
451 | for (i = 0; i < nblks; i++) { | |
452 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
453 | mutex_enter(&db->db_mtx); | |
454 | while (db->db_state == DB_READ || | |
455 | db->db_state == DB_FILL) | |
456 | cv_wait(&db->db_changed, &db->db_mtx); | |
457 | if (db->db_state == DB_UNCACHED) | |
458 | err = SET_ERROR(EIO); | |
459 | mutex_exit(&db->db_mtx); | |
460 | if (err) { | |
461 | dmu_buf_rele_array(dbp, nblks, tag); | |
462 | return (err); | |
463 | } | |
464 | } | |
465 | } | |
466 | ||
467 | *numbufsp = nblks; | |
468 | *dbpp = dbp; | |
469 | return (0); | |
470 | } | |
471 | ||
472 | static int | |
473 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
474 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
475 | { | |
476 | dnode_t *dn; | |
477 | int err; | |
478 | ||
479 | err = dnode_hold(os, object, FTAG, &dn); | |
480 | if (err) | |
481 | return (err); | |
482 | ||
483 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
484 | numbufsp, dbpp, DMU_READ_PREFETCH); | |
485 | ||
486 | dnode_rele(dn, FTAG); | |
487 | ||
488 | return (err); | |
489 | } | |
490 | ||
491 | int | |
492 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, | |
493 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
494 | { | |
495 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
496 | dnode_t *dn; | |
497 | int err; | |
498 | ||
499 | DB_DNODE_ENTER(db); | |
500 | dn = DB_DNODE(db); | |
501 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
502 | numbufsp, dbpp, DMU_READ_PREFETCH); | |
503 | DB_DNODE_EXIT(db); | |
504 | ||
505 | return (err); | |
506 | } | |
507 | ||
508 | void | |
509 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
510 | { | |
511 | int i; | |
512 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
513 | ||
514 | if (numbufs == 0) | |
515 | return; | |
516 | ||
517 | for (i = 0; i < numbufs; i++) { | |
518 | if (dbp[i]) | |
519 | dbuf_rele(dbp[i], tag); | |
520 | } | |
521 | ||
522 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
523 | } | |
524 | ||
525 | /* | |
526 | * Issue prefetch i/os for the given blocks. | |
527 | * | |
528 | * Note: The assumption is that we *know* these blocks will be needed | |
529 | * almost immediately. Therefore, the prefetch i/os will be issued at | |
530 | * ZIO_PRIORITY_SYNC_READ | |
531 | * | |
532 | * Note: indirect blocks and other metadata will be read synchronously, | |
533 | * causing this function to block if they are not already cached. | |
534 | */ | |
535 | void | |
536 | dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) | |
537 | { | |
538 | dnode_t *dn; | |
539 | uint64_t blkid; | |
540 | int nblks, err; | |
541 | ||
542 | if (zfs_prefetch_disable) | |
543 | return; | |
544 | ||
545 | if (len == 0) { /* they're interested in the bonus buffer */ | |
546 | dn = DMU_META_DNODE(os); | |
547 | ||
548 | if (object == 0 || object >= DN_MAX_OBJECT) | |
549 | return; | |
550 | ||
551 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
552 | blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); | |
553 | dbuf_prefetch(dn, blkid, ZIO_PRIORITY_SYNC_READ); | |
554 | rw_exit(&dn->dn_struct_rwlock); | |
555 | return; | |
556 | } | |
557 | ||
558 | /* | |
559 | * XXX - Note, if the dnode for the requested object is not | |
560 | * already cached, we will do a *synchronous* read in the | |
561 | * dnode_hold() call. The same is true for any indirects. | |
562 | */ | |
563 | err = dnode_hold(os, object, FTAG, &dn); | |
564 | if (err != 0) | |
565 | return; | |
566 | ||
567 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
568 | if (dn->dn_datablkshift) { | |
569 | int blkshift = dn->dn_datablkshift; | |
570 | nblks = (P2ROUNDUP(offset + len, 1 << blkshift) - | |
571 | P2ALIGN(offset, 1 << blkshift)) >> blkshift; | |
572 | } else { | |
573 | nblks = (offset < dn->dn_datablksz); | |
574 | } | |
575 | ||
576 | if (nblks != 0) { | |
577 | int i; | |
578 | ||
579 | blkid = dbuf_whichblock(dn, offset); | |
580 | for (i = 0; i < nblks; i++) | |
581 | dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_SYNC_READ); | |
582 | } | |
583 | ||
584 | rw_exit(&dn->dn_struct_rwlock); | |
585 | ||
586 | dnode_rele(dn, FTAG); | |
587 | } | |
588 | ||
589 | /* | |
590 | * Get the next "chunk" of file data to free. We traverse the file from | |
591 | * the end so that the file gets shorter over time (if we crashes in the | |
592 | * middle, this will leave us in a better state). We find allocated file | |
593 | * data by simply searching the allocated level 1 indirects. | |
594 | * | |
595 | * On input, *start should be the first offset that does not need to be | |
596 | * freed (e.g. "offset + length"). On return, *start will be the first | |
597 | * offset that should be freed. | |
598 | */ | |
599 | static int | |
600 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) | |
601 | { | |
602 | uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); | |
603 | /* bytes of data covered by a level-1 indirect block */ | |
604 | uint64_t iblkrange = | |
605 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); | |
606 | uint64_t blks; | |
607 | ||
608 | ASSERT3U(minimum, <=, *start); | |
609 | ||
610 | if (*start - minimum <= iblkrange * maxblks) { | |
611 | *start = minimum; | |
612 | return (0); | |
613 | } | |
614 | ASSERT(ISP2(iblkrange)); | |
615 | ||
616 | for (blks = 0; *start > minimum && blks < maxblks; blks++) { | |
617 | int err; | |
618 | ||
619 | /* | |
620 | * dnode_next_offset(BACKWARDS) will find an allocated L1 | |
621 | * indirect block at or before the input offset. We must | |
622 | * decrement *start so that it is at the end of the region | |
623 | * to search. | |
624 | */ | |
625 | (*start)--; | |
626 | err = dnode_next_offset(dn, | |
627 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); | |
628 | ||
629 | /* if there are no indirect blocks before start, we are done */ | |
630 | if (err == ESRCH) { | |
631 | *start = minimum; | |
632 | break; | |
633 | } else if (err != 0) { | |
634 | return (err); | |
635 | } | |
636 | ||
637 | /* set start to the beginning of this L1 indirect */ | |
638 | *start = P2ALIGN(*start, iblkrange); | |
639 | } | |
640 | if (*start < minimum) | |
641 | *start = minimum; | |
642 | return (0); | |
643 | } | |
644 | ||
645 | static int | |
646 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
647 | uint64_t length) | |
648 | { | |
649 | uint64_t object_size; | |
650 | int err; | |
651 | ||
652 | if (dn == NULL) | |
653 | return (SET_ERROR(EINVAL)); | |
654 | ||
655 | object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
656 | if (offset >= object_size) | |
657 | return (0); | |
658 | ||
659 | if (length == DMU_OBJECT_END || offset + length > object_size) | |
660 | length = object_size - offset; | |
661 | ||
662 | while (length != 0) { | |
663 | uint64_t chunk_end, chunk_begin; | |
664 | dmu_tx_t *tx; | |
665 | ||
666 | chunk_end = chunk_begin = offset + length; | |
667 | ||
668 | /* move chunk_begin backwards to the beginning of this chunk */ | |
669 | err = get_next_chunk(dn, &chunk_begin, offset); | |
670 | if (err) | |
671 | return (err); | |
672 | ASSERT3U(chunk_begin, >=, offset); | |
673 | ASSERT3U(chunk_begin, <=, chunk_end); | |
674 | ||
675 | tx = dmu_tx_create(os); | |
676 | dmu_tx_hold_free(tx, dn->dn_object, | |
677 | chunk_begin, chunk_end - chunk_begin); | |
678 | err = dmu_tx_assign(tx, TXG_WAIT); | |
679 | if (err) { | |
680 | dmu_tx_abort(tx); | |
681 | return (err); | |
682 | } | |
683 | dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx); | |
684 | dmu_tx_commit(tx); | |
685 | ||
686 | length -= chunk_end - chunk_begin; | |
687 | } | |
688 | return (0); | |
689 | } | |
690 | ||
691 | int | |
692 | dmu_free_long_range(objset_t *os, uint64_t object, | |
693 | uint64_t offset, uint64_t length) | |
694 | { | |
695 | dnode_t *dn; | |
696 | int err; | |
697 | ||
698 | err = dnode_hold(os, object, FTAG, &dn); | |
699 | if (err != 0) | |
700 | return (err); | |
701 | err = dmu_free_long_range_impl(os, dn, offset, length); | |
702 | ||
703 | /* | |
704 | * It is important to zero out the maxblkid when freeing the entire | |
705 | * file, so that (a) subsequent calls to dmu_free_long_range_impl() | |
706 | * will take the fast path, and (b) dnode_reallocate() can verify | |
707 | * that the entire file has been freed. | |
708 | */ | |
709 | if (err == 0 && offset == 0 && length == DMU_OBJECT_END) | |
710 | dn->dn_maxblkid = 0; | |
711 | ||
712 | dnode_rele(dn, FTAG); | |
713 | return (err); | |
714 | } | |
715 | ||
716 | int | |
717 | dmu_free_long_object(objset_t *os, uint64_t object) | |
718 | { | |
719 | dmu_tx_t *tx; | |
720 | int err; | |
721 | ||
722 | err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); | |
723 | if (err != 0) | |
724 | return (err); | |
725 | ||
726 | tx = dmu_tx_create(os); | |
727 | dmu_tx_hold_bonus(tx, object); | |
728 | dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); | |
729 | err = dmu_tx_assign(tx, TXG_WAIT); | |
730 | if (err == 0) { | |
731 | err = dmu_object_free(os, object, tx); | |
732 | dmu_tx_commit(tx); | |
733 | } else { | |
734 | dmu_tx_abort(tx); | |
735 | } | |
736 | ||
737 | return (err); | |
738 | } | |
739 | ||
740 | int | |
741 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
742 | uint64_t size, dmu_tx_t *tx) | |
743 | { | |
744 | dnode_t *dn; | |
745 | int err = dnode_hold(os, object, FTAG, &dn); | |
746 | if (err) | |
747 | return (err); | |
748 | ASSERT(offset < UINT64_MAX); | |
749 | ASSERT(size == -1ULL || size <= UINT64_MAX - offset); | |
750 | dnode_free_range(dn, offset, size, tx); | |
751 | dnode_rele(dn, FTAG); | |
752 | return (0); | |
753 | } | |
754 | ||
755 | int | |
756 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
757 | void *buf, uint32_t flags) | |
758 | { | |
759 | dnode_t *dn; | |
760 | dmu_buf_t **dbp; | |
761 | int numbufs, err; | |
762 | ||
763 | err = dnode_hold(os, object, FTAG, &dn); | |
764 | if (err) | |
765 | return (err); | |
766 | ||
767 | /* | |
768 | * Deal with odd block sizes, where there can't be data past the first | |
769 | * block. If we ever do the tail block optimization, we will need to | |
770 | * handle that here as well. | |
771 | */ | |
772 | if (dn->dn_maxblkid == 0) { | |
773 | uint64_t newsz = offset > dn->dn_datablksz ? 0 : | |
774 | MIN(size, dn->dn_datablksz - offset); | |
775 | bzero((char *)buf + newsz, size - newsz); | |
776 | size = newsz; | |
777 | } | |
778 | ||
779 | while (size > 0) { | |
780 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
781 | int i; | |
782 | ||
783 | /* | |
784 | * NB: we could do this block-at-a-time, but it's nice | |
785 | * to be reading in parallel. | |
786 | */ | |
787 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
788 | TRUE, FTAG, &numbufs, &dbp, flags); | |
789 | if (err) | |
790 | break; | |
791 | ||
792 | for (i = 0; i < numbufs; i++) { | |
793 | uint64_t tocpy; | |
794 | int64_t bufoff; | |
795 | dmu_buf_t *db = dbp[i]; | |
796 | ||
797 | ASSERT(size > 0); | |
798 | ||
799 | bufoff = offset - db->db_offset; | |
800 | tocpy = MIN(db->db_size - bufoff, size); | |
801 | ||
802 | (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy); | |
803 | ||
804 | offset += tocpy; | |
805 | size -= tocpy; | |
806 | buf = (char *)buf + tocpy; | |
807 | } | |
808 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
809 | } | |
810 | dnode_rele(dn, FTAG); | |
811 | return (err); | |
812 | } | |
813 | ||
814 | void | |
815 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
816 | const void *buf, dmu_tx_t *tx) | |
817 | { | |
818 | dmu_buf_t **dbp; | |
819 | int numbufs, i; | |
820 | ||
821 | if (size == 0) | |
822 | return; | |
823 | ||
824 | VERIFY0(dmu_buf_hold_array(os, object, offset, size, | |
825 | FALSE, FTAG, &numbufs, &dbp)); | |
826 | ||
827 | for (i = 0; i < numbufs; i++) { | |
828 | uint64_t tocpy; | |
829 | int64_t bufoff; | |
830 | dmu_buf_t *db = dbp[i]; | |
831 | ||
832 | ASSERT(size > 0); | |
833 | ||
834 | bufoff = offset - db->db_offset; | |
835 | tocpy = MIN(db->db_size - bufoff, size); | |
836 | ||
837 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
838 | ||
839 | if (tocpy == db->db_size) | |
840 | dmu_buf_will_fill(db, tx); | |
841 | else | |
842 | dmu_buf_will_dirty(db, tx); | |
843 | ||
844 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); | |
845 | ||
846 | if (tocpy == db->db_size) | |
847 | dmu_buf_fill_done(db, tx); | |
848 | ||
849 | offset += tocpy; | |
850 | size -= tocpy; | |
851 | buf = (char *)buf + tocpy; | |
852 | } | |
853 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
854 | } | |
855 | ||
856 | void | |
857 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
858 | dmu_tx_t *tx) | |
859 | { | |
860 | dmu_buf_t **dbp; | |
861 | int numbufs, i; | |
862 | ||
863 | if (size == 0) | |
864 | return; | |
865 | ||
866 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
867 | FALSE, FTAG, &numbufs, &dbp)); | |
868 | ||
869 | for (i = 0; i < numbufs; i++) { | |
870 | dmu_buf_t *db = dbp[i]; | |
871 | ||
872 | dmu_buf_will_not_fill(db, tx); | |
873 | } | |
874 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
875 | } | |
876 | ||
877 | void | |
878 | dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, | |
879 | void *data, uint8_t etype, uint8_t comp, int uncompressed_size, | |
880 | int compressed_size, int byteorder, dmu_tx_t *tx) | |
881 | { | |
882 | dmu_buf_t *db; | |
883 | ||
884 | ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); | |
885 | ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); | |
886 | VERIFY0(dmu_buf_hold_noread(os, object, offset, | |
887 | FTAG, &db)); | |
888 | ||
889 | dmu_buf_write_embedded(db, | |
890 | data, (bp_embedded_type_t)etype, (enum zio_compress)comp, | |
891 | uncompressed_size, compressed_size, byteorder, tx); | |
892 | ||
893 | dmu_buf_rele(db, FTAG); | |
894 | } | |
895 | ||
896 | /* | |
897 | * DMU support for xuio | |
898 | */ | |
899 | kstat_t *xuio_ksp = NULL; | |
900 | ||
901 | typedef struct xuio_stats { | |
902 | /* loaned yet not returned arc_buf */ | |
903 | kstat_named_t xuiostat_onloan_rbuf; | |
904 | kstat_named_t xuiostat_onloan_wbuf; | |
905 | /* whether a copy is made when loaning out a read buffer */ | |
906 | kstat_named_t xuiostat_rbuf_copied; | |
907 | kstat_named_t xuiostat_rbuf_nocopy; | |
908 | /* whether a copy is made when assigning a write buffer */ | |
909 | kstat_named_t xuiostat_wbuf_copied; | |
910 | kstat_named_t xuiostat_wbuf_nocopy; | |
911 | } xuio_stats_t; | |
912 | ||
913 | static xuio_stats_t xuio_stats = { | |
914 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
915 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
916 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
917 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
918 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
919 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
920 | }; | |
921 | ||
922 | #define XUIOSTAT_INCR(stat, val) \ | |
923 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
924 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
925 | ||
926 | int | |
927 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
928 | { | |
929 | dmu_xuio_t *priv; | |
930 | uio_t *uio = &xuio->xu_uio; | |
931 | ||
932 | uio->uio_iovcnt = nblk; | |
933 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); | |
934 | ||
935 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); | |
936 | priv->cnt = nblk; | |
937 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); | |
938 | priv->iovp = (iovec_t *)uio->uio_iov; | |
939 | XUIO_XUZC_PRIV(xuio) = priv; | |
940 | ||
941 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
942 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
943 | else | |
944 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
945 | ||
946 | return (0); | |
947 | } | |
948 | ||
949 | void | |
950 | dmu_xuio_fini(xuio_t *xuio) | |
951 | { | |
952 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
953 | int nblk = priv->cnt; | |
954 | ||
955 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
956 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
957 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
958 | ||
959 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
960 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
961 | else | |
962 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
963 | } | |
964 | ||
965 | /* | |
966 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
967 | * and increase priv->next by 1. | |
968 | */ | |
969 | int | |
970 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
971 | { | |
972 | struct iovec *iov; | |
973 | uio_t *uio = &xuio->xu_uio; | |
974 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
975 | int i = priv->next++; | |
976 | ||
977 | ASSERT(i < priv->cnt); | |
978 | ASSERT(off + n <= arc_buf_size(abuf)); | |
979 | iov = (iovec_t *)uio->uio_iov + i; | |
980 | iov->iov_base = (char *)abuf->b_data + off; | |
981 | iov->iov_len = n; | |
982 | priv->bufs[i] = abuf; | |
983 | return (0); | |
984 | } | |
985 | ||
986 | int | |
987 | dmu_xuio_cnt(xuio_t *xuio) | |
988 | { | |
989 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
990 | return (priv->cnt); | |
991 | } | |
992 | ||
993 | arc_buf_t * | |
994 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
995 | { | |
996 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
997 | ||
998 | ASSERT(i < priv->cnt); | |
999 | return (priv->bufs[i]); | |
1000 | } | |
1001 | ||
1002 | void | |
1003 | dmu_xuio_clear(xuio_t *xuio, int i) | |
1004 | { | |
1005 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
1006 | ||
1007 | ASSERT(i < priv->cnt); | |
1008 | priv->bufs[i] = NULL; | |
1009 | } | |
1010 | ||
1011 | static void | |
1012 | xuio_stat_init(void) | |
1013 | { | |
1014 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
1015 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
1016 | KSTAT_FLAG_VIRTUAL); | |
1017 | if (xuio_ksp != NULL) { | |
1018 | xuio_ksp->ks_data = &xuio_stats; | |
1019 | kstat_install(xuio_ksp); | |
1020 | } | |
1021 | } | |
1022 | ||
1023 | static void | |
1024 | xuio_stat_fini(void) | |
1025 | { | |
1026 | if (xuio_ksp != NULL) { | |
1027 | kstat_delete(xuio_ksp); | |
1028 | xuio_ksp = NULL; | |
1029 | } | |
1030 | } | |
1031 | ||
1032 | void | |
1033 | xuio_stat_wbuf_copied() | |
1034 | { | |
1035 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1036 | } | |
1037 | ||
1038 | void | |
1039 | xuio_stat_wbuf_nocopy() | |
1040 | { | |
1041 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
1042 | } | |
1043 | ||
1044 | #ifdef _KERNEL | |
1045 | ||
1046 | /* | |
1047 | * Copy up to size bytes between arg_buf and req based on the data direction | |
1048 | * described by the req. If an entire req's data cannot be transfered in one | |
1049 | * pass, you should pass in @req_offset to indicate where to continue. The | |
1050 | * return value is the number of bytes successfully copied to arg_buf. | |
1051 | */ | |
1052 | static int | |
1053 | dmu_bio_copy(void *arg_buf, int size, struct bio *bio, size_t bio_offset) | |
1054 | { | |
1055 | struct bio_vec bv, *bvp = &bv; | |
1056 | bvec_iterator_t iter; | |
1057 | char *bv_buf; | |
1058 | int tocpy, bv_len, bv_offset; | |
1059 | int offset = 0; | |
9784fa9e | 1060 | void *paddr; |
70e083d2 TG |
1061 | |
1062 | bio_for_each_segment4(bv, bvp, bio, iter) { | |
1063 | ||
1064 | /* | |
1065 | * Fully consumed the passed arg_buf. We use goto here because | |
1066 | * rq_for_each_segment is a double loop | |
1067 | */ | |
1068 | ASSERT3S(offset, <=, size); | |
1069 | if (size == offset) | |
1070 | goto out; | |
1071 | ||
1072 | /* Skip already copied bvp */ | |
1073 | if (bio_offset >= bvp->bv_len) { | |
1074 | bio_offset -= bvp->bv_len; | |
1075 | continue; | |
1076 | } | |
1077 | ||
1078 | bv_len = bvp->bv_len - bio_offset; | |
1079 | bv_offset = bvp->bv_offset + bio_offset; | |
1080 | bio_offset = 0; | |
1081 | ||
1082 | tocpy = MIN(bv_len, size - offset); | |
1083 | ASSERT3S(tocpy, >=, 0); | |
1084 | ||
9784fa9e CIK |
1085 | paddr = zfs_kmap_atomic(bvp->bv_page, KM_USER0); |
1086 | bv_buf = paddr + bv_offset; | |
1087 | ASSERT3P(paddr, !=, NULL); | |
70e083d2 TG |
1088 | |
1089 | if (bio_data_dir(bio) == WRITE) | |
1090 | memcpy(arg_buf + offset, bv_buf, tocpy); | |
1091 | else | |
1092 | memcpy(bv_buf, arg_buf + offset, tocpy); | |
9784fa9e | 1093 | zfs_kunmap_atomic(paddr, KM_USER0); |
70e083d2 TG |
1094 | offset += tocpy; |
1095 | } | |
1096 | out: | |
1097 | return (offset); | |
1098 | } | |
1099 | ||
1100 | int | |
1101 | dmu_read_bio(objset_t *os, uint64_t object, struct bio *bio) | |
1102 | { | |
1103 | uint64_t offset = BIO_BI_SECTOR(bio) << 9; | |
1104 | uint64_t size = BIO_BI_SIZE(bio); | |
1105 | dmu_buf_t **dbp; | |
1106 | int numbufs, i, err; | |
1107 | size_t bio_offset; | |
1108 | ||
1109 | /* | |
1110 | * NB: we could do this block-at-a-time, but it's nice | |
1111 | * to be reading in parallel. | |
1112 | */ | |
1113 | err = dmu_buf_hold_array(os, object, offset, size, TRUE, FTAG, | |
1114 | &numbufs, &dbp); | |
1115 | if (err) | |
1116 | return (err); | |
1117 | ||
1118 | bio_offset = 0; | |
1119 | for (i = 0; i < numbufs; i++) { | |
1120 | uint64_t tocpy; | |
1121 | int64_t bufoff; | |
1122 | int didcpy; | |
1123 | dmu_buf_t *db = dbp[i]; | |
1124 | ||
1125 | bufoff = offset - db->db_offset; | |
1126 | ASSERT3S(bufoff, >=, 0); | |
1127 | ||
1128 | tocpy = MIN(db->db_size - bufoff, size); | |
1129 | if (tocpy == 0) | |
1130 | break; | |
1131 | ||
1132 | didcpy = dmu_bio_copy(db->db_data + bufoff, tocpy, bio, | |
1133 | bio_offset); | |
1134 | ||
1135 | if (didcpy < tocpy) | |
1136 | err = EIO; | |
1137 | ||
1138 | if (err) | |
1139 | break; | |
1140 | ||
1141 | size -= tocpy; | |
1142 | offset += didcpy; | |
1143 | bio_offset += didcpy; | |
1144 | err = 0; | |
1145 | } | |
1146 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1147 | ||
1148 | return (err); | |
1149 | } | |
1150 | ||
1151 | int | |
1152 | dmu_write_bio(objset_t *os, uint64_t object, struct bio *bio, dmu_tx_t *tx) | |
1153 | { | |
1154 | uint64_t offset = BIO_BI_SECTOR(bio) << 9; | |
1155 | uint64_t size = BIO_BI_SIZE(bio); | |
1156 | dmu_buf_t **dbp; | |
1157 | int numbufs, i, err; | |
1158 | size_t bio_offset; | |
1159 | ||
1160 | if (size == 0) | |
1161 | return (0); | |
1162 | ||
1163 | err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1164 | &numbufs, &dbp); | |
1165 | if (err) | |
1166 | return (err); | |
1167 | ||
1168 | bio_offset = 0; | |
1169 | for (i = 0; i < numbufs; i++) { | |
1170 | uint64_t tocpy; | |
1171 | int64_t bufoff; | |
1172 | int didcpy; | |
1173 | dmu_buf_t *db = dbp[i]; | |
1174 | ||
1175 | bufoff = offset - db->db_offset; | |
1176 | ASSERT3S(bufoff, >=, 0); | |
1177 | ||
1178 | tocpy = MIN(db->db_size - bufoff, size); | |
1179 | if (tocpy == 0) | |
1180 | break; | |
1181 | ||
1182 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1183 | ||
1184 | if (tocpy == db->db_size) | |
1185 | dmu_buf_will_fill(db, tx); | |
1186 | else | |
1187 | dmu_buf_will_dirty(db, tx); | |
1188 | ||
1189 | didcpy = dmu_bio_copy(db->db_data + bufoff, tocpy, bio, | |
1190 | bio_offset); | |
1191 | ||
1192 | if (tocpy == db->db_size) | |
1193 | dmu_buf_fill_done(db, tx); | |
1194 | ||
1195 | if (didcpy < tocpy) | |
1196 | err = EIO; | |
1197 | ||
1198 | if (err) | |
1199 | break; | |
1200 | ||
1201 | size -= tocpy; | |
1202 | offset += didcpy; | |
1203 | bio_offset += didcpy; | |
1204 | err = 0; | |
1205 | } | |
1206 | ||
1207 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1208 | return (err); | |
1209 | } | |
1210 | ||
1211 | static int | |
1212 | dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) | |
1213 | { | |
1214 | dmu_buf_t **dbp; | |
1215 | int numbufs, i, err; | |
1216 | xuio_t *xuio = NULL; | |
1217 | ||
1218 | /* | |
1219 | * NB: we could do this block-at-a-time, but it's nice | |
1220 | * to be reading in parallel. | |
1221 | */ | |
1222 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1223 | TRUE, FTAG, &numbufs, &dbp, 0); | |
1224 | if (err) | |
1225 | return (err); | |
1226 | ||
1227 | for (i = 0; i < numbufs; i++) { | |
1228 | uint64_t tocpy; | |
1229 | int64_t bufoff; | |
1230 | dmu_buf_t *db = dbp[i]; | |
1231 | ||
1232 | ASSERT(size > 0); | |
1233 | ||
1234 | bufoff = uio->uio_loffset - db->db_offset; | |
1235 | tocpy = MIN(db->db_size - bufoff, size); | |
1236 | ||
1237 | if (xuio) { | |
1238 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1239 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1240 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1241 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1242 | if (!err) { | |
1243 | uio->uio_resid -= tocpy; | |
1244 | uio->uio_loffset += tocpy; | |
1245 | } | |
1246 | ||
1247 | if (abuf == dbuf_abuf) | |
1248 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1249 | else | |
1250 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
1251 | } else { | |
1252 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1253 | UIO_READ, uio); | |
1254 | } | |
1255 | if (err) | |
1256 | break; | |
1257 | ||
1258 | size -= tocpy; | |
1259 | } | |
1260 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1261 | ||
1262 | return (err); | |
1263 | } | |
1264 | ||
1265 | /* | |
1266 | * Read 'size' bytes into the uio buffer. | |
1267 | * From object zdb->db_object. | |
1268 | * Starting at offset uio->uio_loffset. | |
1269 | * | |
1270 | * If the caller already has a dbuf in the target object | |
1271 | * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), | |
1272 | * because we don't have to find the dnode_t for the object. | |
1273 | */ | |
1274 | int | |
1275 | dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) | |
1276 | { | |
1277 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1278 | dnode_t *dn; | |
1279 | int err; | |
1280 | ||
1281 | if (size == 0) | |
1282 | return (0); | |
1283 | ||
1284 | DB_DNODE_ENTER(db); | |
1285 | dn = DB_DNODE(db); | |
1286 | err = dmu_read_uio_dnode(dn, uio, size); | |
1287 | DB_DNODE_EXIT(db); | |
1288 | ||
1289 | return (err); | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Read 'size' bytes into the uio buffer. | |
1294 | * From the specified object | |
1295 | * Starting at offset uio->uio_loffset. | |
1296 | */ | |
1297 | int | |
1298 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1299 | { | |
1300 | dnode_t *dn; | |
1301 | int err; | |
1302 | ||
1303 | if (size == 0) | |
1304 | return (0); | |
1305 | ||
1306 | err = dnode_hold(os, object, FTAG, &dn); | |
1307 | if (err) | |
1308 | return (err); | |
1309 | ||
1310 | err = dmu_read_uio_dnode(dn, uio, size); | |
1311 | ||
1312 | dnode_rele(dn, FTAG); | |
1313 | ||
1314 | return (err); | |
1315 | } | |
1316 | ||
1317 | static int | |
1318 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) | |
1319 | { | |
1320 | dmu_buf_t **dbp; | |
1321 | int numbufs; | |
1322 | int err = 0; | |
1323 | int i; | |
1324 | ||
1325 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1326 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1327 | if (err) | |
1328 | return (err); | |
1329 | ||
1330 | for (i = 0; i < numbufs; i++) { | |
1331 | uint64_t tocpy; | |
1332 | int64_t bufoff; | |
1333 | dmu_buf_t *db = dbp[i]; | |
1334 | ||
1335 | ASSERT(size > 0); | |
1336 | ||
1337 | bufoff = uio->uio_loffset - db->db_offset; | |
1338 | tocpy = MIN(db->db_size - bufoff, size); | |
1339 | ||
1340 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1341 | ||
1342 | if (tocpy == db->db_size) | |
1343 | dmu_buf_will_fill(db, tx); | |
1344 | else | |
1345 | dmu_buf_will_dirty(db, tx); | |
1346 | ||
1347 | /* | |
1348 | * XXX uiomove could block forever (eg.nfs-backed | |
1349 | * pages). There needs to be a uiolockdown() function | |
1350 | * to lock the pages in memory, so that uiomove won't | |
1351 | * block. | |
1352 | */ | |
1353 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1354 | UIO_WRITE, uio); | |
1355 | ||
1356 | if (tocpy == db->db_size) | |
1357 | dmu_buf_fill_done(db, tx); | |
1358 | ||
1359 | if (err) | |
1360 | break; | |
1361 | ||
1362 | size -= tocpy; | |
1363 | } | |
1364 | ||
1365 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1366 | return (err); | |
1367 | } | |
1368 | ||
1369 | /* | |
1370 | * Write 'size' bytes from the uio buffer. | |
1371 | * To object zdb->db_object. | |
1372 | * Starting at offset uio->uio_loffset. | |
1373 | * | |
1374 | * If the caller already has a dbuf in the target object | |
1375 | * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), | |
1376 | * because we don't have to find the dnode_t for the object. | |
1377 | */ | |
1378 | int | |
1379 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1380 | dmu_tx_t *tx) | |
1381 | { | |
1382 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; | |
1383 | dnode_t *dn; | |
1384 | int err; | |
1385 | ||
1386 | if (size == 0) | |
1387 | return (0); | |
1388 | ||
1389 | DB_DNODE_ENTER(db); | |
1390 | dn = DB_DNODE(db); | |
1391 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1392 | DB_DNODE_EXIT(db); | |
1393 | ||
1394 | return (err); | |
1395 | } | |
1396 | ||
1397 | /* | |
1398 | * Write 'size' bytes from the uio buffer. | |
1399 | * To the specified object. | |
1400 | * Starting at offset uio->uio_loffset. | |
1401 | */ | |
1402 | int | |
1403 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1404 | dmu_tx_t *tx) | |
1405 | { | |
1406 | dnode_t *dn; | |
1407 | int err; | |
1408 | ||
1409 | if (size == 0) | |
1410 | return (0); | |
1411 | ||
1412 | err = dnode_hold(os, object, FTAG, &dn); | |
1413 | if (err) | |
1414 | return (err); | |
1415 | ||
1416 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1417 | ||
1418 | dnode_rele(dn, FTAG); | |
1419 | ||
1420 | return (err); | |
1421 | } | |
1422 | #endif /* _KERNEL */ | |
1423 | ||
1424 | /* | |
1425 | * Allocate a loaned anonymous arc buffer. | |
1426 | */ | |
1427 | arc_buf_t * | |
1428 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1429 | { | |
1430 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; | |
1431 | ||
1432 | return (arc_loan_buf(db->db_objset->os_spa, size)); | |
1433 | } | |
1434 | ||
1435 | /* | |
1436 | * Free a loaned arc buffer. | |
1437 | */ | |
1438 | void | |
1439 | dmu_return_arcbuf(arc_buf_t *buf) | |
1440 | { | |
1441 | arc_return_buf(buf, FTAG); | |
1442 | VERIFY(arc_buf_remove_ref(buf, FTAG)); | |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1447 | * If this is not possible copy the contents of passed arc buf via | |
1448 | * dmu_write(). | |
1449 | */ | |
1450 | void | |
1451 | dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, | |
1452 | dmu_tx_t *tx) | |
1453 | { | |
1454 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; | |
1455 | dnode_t *dn; | |
1456 | dmu_buf_impl_t *db; | |
1457 | uint32_t blksz = (uint32_t)arc_buf_size(buf); | |
1458 | uint64_t blkid; | |
1459 | ||
1460 | DB_DNODE_ENTER(dbuf); | |
1461 | dn = DB_DNODE(dbuf); | |
1462 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
1463 | blkid = dbuf_whichblock(dn, offset); | |
1464 | VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); | |
1465 | rw_exit(&dn->dn_struct_rwlock); | |
1466 | DB_DNODE_EXIT(dbuf); | |
1467 | ||
1468 | /* | |
1469 | * We can only assign if the offset is aligned, the arc buf is the | |
1470 | * same size as the dbuf, and the dbuf is not metadata. It | |
1471 | * can't be metadata because the loaned arc buf comes from the | |
1472 | * user-data kmem area. | |
1473 | */ | |
1474 | if (offset == db->db.db_offset && blksz == db->db.db_size && | |
1475 | DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) { | |
1476 | dbuf_assign_arcbuf(db, buf, tx); | |
1477 | dbuf_rele(db, FTAG); | |
1478 | } else { | |
1479 | objset_t *os; | |
1480 | uint64_t object; | |
1481 | ||
1482 | DB_DNODE_ENTER(dbuf); | |
1483 | dn = DB_DNODE(dbuf); | |
1484 | os = dn->dn_objset; | |
1485 | object = dn->dn_object; | |
1486 | DB_DNODE_EXIT(dbuf); | |
1487 | ||
1488 | dbuf_rele(db, FTAG); | |
1489 | dmu_write(os, object, offset, blksz, buf->b_data, tx); | |
1490 | dmu_return_arcbuf(buf); | |
1491 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
1492 | } | |
1493 | } | |
1494 | ||
1495 | typedef struct { | |
1496 | dbuf_dirty_record_t *dsa_dr; | |
1497 | dmu_sync_cb_t *dsa_done; | |
1498 | zgd_t *dsa_zgd; | |
1499 | dmu_tx_t *dsa_tx; | |
1500 | } dmu_sync_arg_t; | |
1501 | ||
1502 | /* ARGSUSED */ | |
1503 | static void | |
1504 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1505 | { | |
1506 | dmu_sync_arg_t *dsa = varg; | |
1507 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
1508 | blkptr_t *bp = zio->io_bp; | |
1509 | ||
1510 | if (zio->io_error == 0) { | |
1511 | if (BP_IS_HOLE(bp)) { | |
1512 | /* | |
1513 | * A block of zeros may compress to a hole, but the | |
1514 | * block size still needs to be known for replay. | |
1515 | */ | |
1516 | BP_SET_LSIZE(bp, db->db_size); | |
1517 | } else if (!BP_IS_EMBEDDED(bp)) { | |
1518 | ASSERT(BP_GET_LEVEL(bp) == 0); | |
1519 | bp->blk_fill = 1; | |
1520 | } | |
1521 | } | |
1522 | } | |
1523 | ||
1524 | static void | |
1525 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1526 | { | |
1527 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1528 | } | |
1529 | ||
1530 | /* ARGSUSED */ | |
1531 | static void | |
1532 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1533 | { | |
1534 | dmu_sync_arg_t *dsa = varg; | |
1535 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
1536 | dmu_buf_impl_t *db = dr->dr_dbuf; | |
1537 | ||
1538 | mutex_enter(&db->db_mtx); | |
1539 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
1540 | if (zio->io_error == 0) { | |
1541 | dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); | |
1542 | if (dr->dt.dl.dr_nopwrite) { | |
1543 | ASSERTV(blkptr_t *bp = zio->io_bp); | |
1544 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); | |
1545 | ASSERTV(uint8_t chksum = BP_GET_CHECKSUM(bp_orig)); | |
1546 | ||
1547 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
1548 | ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); | |
1549 | ASSERT(zio_checksum_table[chksum].ci_dedup); | |
1550 | } | |
1551 | dr->dt.dl.dr_overridden_by = *zio->io_bp; | |
1552 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1553 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
1554 | ||
1555 | /* | |
1556 | * Old style holes are filled with all zeros, whereas | |
1557 | * new-style holes maintain their lsize, type, level, | |
1558 | * and birth time (see zio_write_compress). While we | |
1559 | * need to reset the BP_SET_LSIZE() call that happened | |
1560 | * in dmu_sync_ready for old style holes, we do *not* | |
1561 | * want to wipe out the information contained in new | |
1562 | * style holes. Thus, only zero out the block pointer if | |
1563 | * it's an old style hole. | |
1564 | */ | |
1565 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && | |
1566 | dr->dt.dl.dr_overridden_by.blk_birth == 0) | |
1567 | BP_ZERO(&dr->dt.dl.dr_overridden_by); | |
1568 | } else { | |
1569 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1570 | } | |
1571 | cv_broadcast(&db->db_changed); | |
1572 | mutex_exit(&db->db_mtx); | |
1573 | ||
1574 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1575 | ||
1576 | kmem_free(dsa, sizeof (*dsa)); | |
1577 | } | |
1578 | ||
1579 | static void | |
1580 | dmu_sync_late_arrival_done(zio_t *zio) | |
1581 | { | |
1582 | blkptr_t *bp = zio->io_bp; | |
1583 | dmu_sync_arg_t *dsa = zio->io_private; | |
1584 | ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig); | |
1585 | ||
1586 | if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { | |
1587 | /* | |
1588 | * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) | |
1589 | * then there is nothing to do here. Otherwise, free the | |
1590 | * newly allocated block in this txg. | |
1591 | */ | |
1592 | if (zio->io_flags & ZIO_FLAG_NOPWRITE) { | |
1593 | ASSERT(BP_EQUAL(bp, bp_orig)); | |
1594 | } else { | |
1595 | ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); | |
1596 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1597 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1598 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1599 | } | |
1600 | } | |
1601 | ||
1602 | dmu_tx_commit(dsa->dsa_tx); | |
1603 | ||
1604 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1605 | ||
1606 | kmem_free(dsa, sizeof (*dsa)); | |
1607 | } | |
1608 | ||
1609 | static int | |
1610 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
1611 | zio_prop_t *zp, zbookmark_phys_t *zb) | |
1612 | { | |
1613 | dmu_sync_arg_t *dsa; | |
1614 | dmu_tx_t *tx; | |
1615 | ||
1616 | tx = dmu_tx_create(os); | |
1617 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1618 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1619 | dmu_tx_abort(tx); | |
1620 | /* Make zl_get_data do txg_waited_synced() */ | |
1621 | return (SET_ERROR(EIO)); | |
1622 | } | |
1623 | ||
1624 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); | |
1625 | dsa->dsa_dr = NULL; | |
1626 | dsa->dsa_done = done; | |
1627 | dsa->dsa_zgd = zgd; | |
1628 | dsa->dsa_tx = tx; | |
1629 | ||
1630 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, | |
1631 | zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, | |
1632 | dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa, | |
1633 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL|ZIO_FLAG_FASTWRITE, zb)); | |
1634 | ||
1635 | return (0); | |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * Intent log support: sync the block associated with db to disk. | |
1640 | * N.B. and XXX: the caller is responsible for making sure that the | |
1641 | * data isn't changing while dmu_sync() is writing it. | |
1642 | * | |
1643 | * Return values: | |
1644 | * | |
1645 | * EEXIST: this txg has already been synced, so there's nothing to do. | |
1646 | * The caller should not log the write. | |
1647 | * | |
1648 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1649 | * The caller should not log the write. | |
1650 | * | |
1651 | * EALREADY: this block is already in the process of being synced. | |
1652 | * The caller should track its progress (somehow). | |
1653 | * | |
1654 | * EIO: could not do the I/O. | |
1655 | * The caller should do a txg_wait_synced(). | |
1656 | * | |
1657 | * 0: the I/O has been initiated. | |
1658 | * The caller should log this blkptr in the done callback. | |
1659 | * It is possible that the I/O will fail, in which case | |
1660 | * the error will be reported to the done callback and | |
1661 | * propagated to pio from zio_done(). | |
1662 | */ | |
1663 | int | |
1664 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) | |
1665 | { | |
1666 | blkptr_t *bp = zgd->zgd_bp; | |
1667 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; | |
1668 | objset_t *os = db->db_objset; | |
1669 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
1670 | dbuf_dirty_record_t *dr; | |
1671 | dmu_sync_arg_t *dsa; | |
1672 | zbookmark_phys_t zb; | |
1673 | zio_prop_t zp; | |
1674 | dnode_t *dn; | |
1675 | ||
1676 | ASSERT(pio != NULL); | |
1677 | ASSERT(txg != 0); | |
1678 | ||
1679 | SET_BOOKMARK(&zb, ds->ds_object, | |
1680 | db->db.db_object, db->db_level, db->db_blkid); | |
1681 | ||
1682 | DB_DNODE_ENTER(db); | |
1683 | dn = DB_DNODE(db); | |
1684 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); | |
1685 | DB_DNODE_EXIT(db); | |
1686 | ||
1687 | /* | |
1688 | * If we're frozen (running ziltest), we always need to generate a bp. | |
1689 | */ | |
1690 | if (txg > spa_freeze_txg(os->os_spa)) | |
1691 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
1692 | ||
1693 | /* | |
1694 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() | |
1695 | * and us. If we determine that this txg is not yet syncing, | |
1696 | * but it begins to sync a moment later, that's OK because the | |
1697 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
1698 | */ | |
1699 | mutex_enter(&db->db_mtx); | |
1700 | ||
1701 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
1702 | /* | |
1703 | * This txg has already synced. There's nothing to do. | |
1704 | */ | |
1705 | mutex_exit(&db->db_mtx); | |
1706 | return (SET_ERROR(EEXIST)); | |
1707 | } | |
1708 | ||
1709 | if (txg <= spa_syncing_txg(os->os_spa)) { | |
1710 | /* | |
1711 | * This txg is currently syncing, so we can't mess with | |
1712 | * the dirty record anymore; just write a new log block. | |
1713 | */ | |
1714 | mutex_exit(&db->db_mtx); | |
1715 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
1716 | } | |
1717 | ||
1718 | dr = db->db_last_dirty; | |
1719 | while (dr && dr->dr_txg != txg) | |
1720 | dr = dr->dr_next; | |
1721 | ||
1722 | if (dr == NULL) { | |
1723 | /* | |
1724 | * There's no dr for this dbuf, so it must have been freed. | |
1725 | * There's no need to log writes to freed blocks, so we're done. | |
1726 | */ | |
1727 | mutex_exit(&db->db_mtx); | |
1728 | return (SET_ERROR(ENOENT)); | |
1729 | } | |
1730 | ||
1731 | ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); | |
1732 | ||
1733 | /* | |
1734 | * Assume the on-disk data is X, the current syncing data (in | |
1735 | * txg - 1) is Y, and the current in-memory data is Z (currently | |
1736 | * in dmu_sync). | |
1737 | * | |
1738 | * We usually want to perform a nopwrite if X and Z are the | |
1739 | * same. However, if Y is different (i.e. the BP is going to | |
1740 | * change before this write takes effect), then a nopwrite will | |
1741 | * be incorrect - we would override with X, which could have | |
1742 | * been freed when Y was written. | |
1743 | * | |
1744 | * (Note that this is not a concern when we are nop-writing from | |
1745 | * syncing context, because X and Y must be identical, because | |
1746 | * all previous txgs have been synced.) | |
1747 | * | |
1748 | * Therefore, we disable nopwrite if the current BP could change | |
1749 | * before this TXG. There are two ways it could change: by | |
1750 | * being dirty (dr_next is non-NULL), or by being freed | |
1751 | * (dnode_block_freed()). This behavior is verified by | |
1752 | * zio_done(), which VERIFYs that the override BP is identical | |
1753 | * to the on-disk BP. | |
1754 | */ | |
1755 | DB_DNODE_ENTER(db); | |
1756 | dn = DB_DNODE(db); | |
1757 | if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) | |
1758 | zp.zp_nopwrite = B_FALSE; | |
1759 | DB_DNODE_EXIT(db); | |
1760 | ||
1761 | ASSERT(dr->dr_txg == txg); | |
1762 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || | |
1763 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
1764 | /* | |
1765 | * We have already issued a sync write for this buffer, | |
1766 | * or this buffer has already been synced. It could not | |
1767 | * have been dirtied since, or we would have cleared the state. | |
1768 | */ | |
1769 | mutex_exit(&db->db_mtx); | |
1770 | return (SET_ERROR(EALREADY)); | |
1771 | } | |
1772 | ||
1773 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); | |
1774 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; | |
1775 | mutex_exit(&db->db_mtx); | |
1776 | ||
1777 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); | |
1778 | dsa->dsa_dr = dr; | |
1779 | dsa->dsa_done = done; | |
1780 | dsa->dsa_zgd = zgd; | |
1781 | dsa->dsa_tx = NULL; | |
1782 | ||
1783 | zio_nowait(arc_write(pio, os->os_spa, txg, | |
1784 | bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), | |
1785 | DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, | |
1786 | NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE, | |
1787 | ZIO_FLAG_CANFAIL, &zb)); | |
1788 | ||
1789 | return (0); | |
1790 | } | |
1791 | ||
1792 | int | |
1793 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
1794 | dmu_tx_t *tx) | |
1795 | { | |
1796 | dnode_t *dn; | |
1797 | int err; | |
1798 | ||
1799 | err = dnode_hold(os, object, FTAG, &dn); | |
1800 | if (err) | |
1801 | return (err); | |
1802 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1803 | dnode_rele(dn, FTAG); | |
1804 | return (err); | |
1805 | } | |
1806 | ||
1807 | void | |
1808 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
1809 | dmu_tx_t *tx) | |
1810 | { | |
1811 | dnode_t *dn; | |
1812 | ||
1813 | /* | |
1814 | * Send streams include each object's checksum function. This | |
1815 | * check ensures that the receiving system can understand the | |
1816 | * checksum function transmitted. | |
1817 | */ | |
1818 | ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); | |
1819 | ||
1820 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1821 | ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); | |
1822 | dn->dn_checksum = checksum; | |
1823 | dnode_setdirty(dn, tx); | |
1824 | dnode_rele(dn, FTAG); | |
1825 | } | |
1826 | ||
1827 | void | |
1828 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
1829 | dmu_tx_t *tx) | |
1830 | { | |
1831 | dnode_t *dn; | |
1832 | ||
1833 | /* | |
1834 | * Send streams include each object's compression function. This | |
1835 | * check ensures that the receiving system can understand the | |
1836 | * compression function transmitted. | |
1837 | */ | |
1838 | ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); | |
1839 | ||
1840 | VERIFY0(dnode_hold(os, object, FTAG, &dn)); | |
1841 | dn->dn_compress = compress; | |
1842 | dnode_setdirty(dn, tx); | |
1843 | dnode_rele(dn, FTAG); | |
1844 | } | |
1845 | ||
1846 | int zfs_mdcomp_disable = 0; | |
1847 | ||
1848 | /* | |
1849 | * When the "redundant_metadata" property is set to "most", only indirect | |
1850 | * blocks of this level and higher will have an additional ditto block. | |
1851 | */ | |
1852 | int zfs_redundant_metadata_most_ditto_level = 2; | |
1853 | ||
1854 | void | |
1855 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) | |
1856 | { | |
1857 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
1858 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || | |
1859 | (wp & WP_SPILL)); | |
1860 | enum zio_checksum checksum = os->os_checksum; | |
1861 | enum zio_compress compress = os->os_compress; | |
1862 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
1863 | boolean_t dedup = B_FALSE; | |
1864 | boolean_t nopwrite = B_FALSE; | |
1865 | boolean_t dedup_verify = os->os_dedup_verify; | |
1866 | int copies = os->os_copies; | |
1867 | ||
1868 | /* | |
1869 | * We maintain different write policies for each of the following | |
1870 | * types of data: | |
1871 | * 1. metadata | |
1872 | * 2. preallocated blocks (i.e. level-0 blocks of a dump device) | |
1873 | * 3. all other level 0 blocks | |
1874 | */ | |
1875 | if (ismd) { | |
1876 | if (zfs_mdcomp_disable) { | |
1877 | compress = ZIO_COMPRESS_EMPTY; | |
1878 | } else { | |
1879 | /* | |
1880 | * XXX -- we should design a compression algorithm | |
1881 | * that specializes in arrays of bps. | |
1882 | */ | |
1883 | compress = zio_compress_select(os->os_spa, | |
1884 | ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * Metadata always gets checksummed. If the data | |
1889 | * checksum is multi-bit correctable, and it's not a | |
1890 | * ZBT-style checksum, then it's suitable for metadata | |
1891 | * as well. Otherwise, the metadata checksum defaults | |
1892 | * to fletcher4. | |
1893 | */ | |
1894 | if (zio_checksum_table[checksum].ci_correctable < 1 || | |
1895 | zio_checksum_table[checksum].ci_eck) | |
1896 | checksum = ZIO_CHECKSUM_FLETCHER_4; | |
1897 | ||
1898 | if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || | |
1899 | (os->os_redundant_metadata == | |
1900 | ZFS_REDUNDANT_METADATA_MOST && | |
1901 | (level >= zfs_redundant_metadata_most_ditto_level || | |
1902 | DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) | |
1903 | copies++; | |
1904 | } else if (wp & WP_NOFILL) { | |
1905 | ASSERT(level == 0); | |
1906 | ||
1907 | /* | |
1908 | * If we're writing preallocated blocks, we aren't actually | |
1909 | * writing them so don't set any policy properties. These | |
1910 | * blocks are currently only used by an external subsystem | |
1911 | * outside of zfs (i.e. dump) and not written by the zio | |
1912 | * pipeline. | |
1913 | */ | |
1914 | compress = ZIO_COMPRESS_OFF; | |
1915 | checksum = ZIO_CHECKSUM_OFF; | |
1916 | } else { | |
1917 | compress = zio_compress_select(os->os_spa, dn->dn_compress, | |
1918 | compress); | |
1919 | ||
1920 | checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? | |
1921 | zio_checksum_select(dn->dn_checksum, checksum) : | |
1922 | dedup_checksum; | |
1923 | ||
1924 | /* | |
1925 | * Determine dedup setting. If we are in dmu_sync(), | |
1926 | * we won't actually dedup now because that's all | |
1927 | * done in syncing context; but we do want to use the | |
1928 | * dedup checkum. If the checksum is not strong | |
1929 | * enough to ensure unique signatures, force | |
1930 | * dedup_verify. | |
1931 | */ | |
1932 | if (dedup_checksum != ZIO_CHECKSUM_OFF) { | |
1933 | dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; | |
1934 | if (!zio_checksum_table[checksum].ci_dedup) | |
1935 | dedup_verify = B_TRUE; | |
1936 | } | |
1937 | ||
1938 | /* | |
1939 | * Enable nopwrite if we have a cryptographically secure | |
1940 | * checksum that has no known collisions (i.e. SHA-256) | |
1941 | * and compression is enabled. We don't enable nopwrite if | |
1942 | * dedup is enabled as the two features are mutually exclusive. | |
1943 | */ | |
1944 | nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup && | |
1945 | compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); | |
1946 | } | |
1947 | ||
1948 | zp->zp_checksum = checksum; | |
1949 | zp->zp_compress = compress; | |
1950 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; | |
1951 | zp->zp_level = level; | |
1952 | zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); | |
1953 | zp->zp_dedup = dedup; | |
1954 | zp->zp_dedup_verify = dedup && dedup_verify; | |
1955 | zp->zp_nopwrite = nopwrite; | |
1956 | } | |
1957 | ||
1958 | int | |
1959 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
1960 | { | |
1961 | dnode_t *dn; | |
1962 | int i, err; | |
1963 | ||
1964 | err = dnode_hold(os, object, FTAG, &dn); | |
1965 | if (err) | |
1966 | return (err); | |
1967 | /* | |
1968 | * Sync any current changes before | |
1969 | * we go trundling through the block pointers. | |
1970 | */ | |
1971 | for (i = 0; i < TXG_SIZE; i++) { | |
1972 | if (list_link_active(&dn->dn_dirty_link[i])) | |
1973 | break; | |
1974 | } | |
1975 | if (i != TXG_SIZE) { | |
1976 | dnode_rele(dn, FTAG); | |
1977 | txg_wait_synced(dmu_objset_pool(os), 0); | |
1978 | err = dnode_hold(os, object, FTAG, &dn); | |
1979 | if (err) | |
1980 | return (err); | |
1981 | } | |
1982 | ||
1983 | err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); | |
1984 | dnode_rele(dn, FTAG); | |
1985 | ||
1986 | return (err); | |
1987 | } | |
1988 | ||
1989 | void | |
1990 | __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
1991 | { | |
1992 | dnode_phys_t *dnp = dn->dn_phys; | |
1993 | int i; | |
1994 | ||
1995 | doi->doi_data_block_size = dn->dn_datablksz; | |
1996 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
1997 | 1ULL << dn->dn_indblkshift : 0; | |
1998 | doi->doi_type = dn->dn_type; | |
1999 | doi->doi_bonus_type = dn->dn_bonustype; | |
2000 | doi->doi_bonus_size = dn->dn_bonuslen; | |
2001 | doi->doi_indirection = dn->dn_nlevels; | |
2002 | doi->doi_checksum = dn->dn_checksum; | |
2003 | doi->doi_compress = dn->dn_compress; | |
2004 | doi->doi_nblkptr = dn->dn_nblkptr; | |
2005 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; | |
2006 | doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; | |
2007 | doi->doi_fill_count = 0; | |
2008 | for (i = 0; i < dnp->dn_nblkptr; i++) | |
2009 | doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); | |
2010 | } | |
2011 | ||
2012 | void | |
2013 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
2014 | { | |
2015 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
2016 | mutex_enter(&dn->dn_mtx); | |
2017 | ||
2018 | __dmu_object_info_from_dnode(dn, doi); | |
2019 | ||
2020 | mutex_exit(&dn->dn_mtx); | |
2021 | rw_exit(&dn->dn_struct_rwlock); | |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * Get information on a DMU object. | |
2026 | * If doi is NULL, just indicates whether the object exists. | |
2027 | */ | |
2028 | int | |
2029 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
2030 | { | |
2031 | dnode_t *dn; | |
2032 | int err = dnode_hold(os, object, FTAG, &dn); | |
2033 | ||
2034 | if (err) | |
2035 | return (err); | |
2036 | ||
2037 | if (doi != NULL) | |
2038 | dmu_object_info_from_dnode(dn, doi); | |
2039 | ||
2040 | dnode_rele(dn, FTAG); | |
2041 | return (0); | |
2042 | } | |
2043 | ||
2044 | /* | |
2045 | * As above, but faster; can be used when you have a held dbuf in hand. | |
2046 | */ | |
2047 | void | |
2048 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) | |
2049 | { | |
2050 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2051 | ||
2052 | DB_DNODE_ENTER(db); | |
2053 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
2054 | DB_DNODE_EXIT(db); | |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * Faster still when you only care about the size. | |
2059 | * This is specifically optimized for zfs_getattr(). | |
2060 | */ | |
2061 | void | |
2062 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, | |
2063 | u_longlong_t *nblk512) | |
2064 | { | |
2065 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
2066 | dnode_t *dn; | |
2067 | ||
2068 | DB_DNODE_ENTER(db); | |
2069 | dn = DB_DNODE(db); | |
2070 | ||
2071 | *blksize = dn->dn_datablksz; | |
2072 | /* add 1 for dnode space */ | |
2073 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> | |
2074 | SPA_MINBLOCKSHIFT) + 1; | |
2075 | DB_DNODE_EXIT(db); | |
2076 | } | |
2077 | ||
2078 | void | |
2079 | byteswap_uint64_array(void *vbuf, size_t size) | |
2080 | { | |
2081 | uint64_t *buf = vbuf; | |
2082 | size_t count = size >> 3; | |
2083 | int i; | |
2084 | ||
2085 | ASSERT((size & 7) == 0); | |
2086 | ||
2087 | for (i = 0; i < count; i++) | |
2088 | buf[i] = BSWAP_64(buf[i]); | |
2089 | } | |
2090 | ||
2091 | void | |
2092 | byteswap_uint32_array(void *vbuf, size_t size) | |
2093 | { | |
2094 | uint32_t *buf = vbuf; | |
2095 | size_t count = size >> 2; | |
2096 | int i; | |
2097 | ||
2098 | ASSERT((size & 3) == 0); | |
2099 | ||
2100 | for (i = 0; i < count; i++) | |
2101 | buf[i] = BSWAP_32(buf[i]); | |
2102 | } | |
2103 | ||
2104 | void | |
2105 | byteswap_uint16_array(void *vbuf, size_t size) | |
2106 | { | |
2107 | uint16_t *buf = vbuf; | |
2108 | size_t count = size >> 1; | |
2109 | int i; | |
2110 | ||
2111 | ASSERT((size & 1) == 0); | |
2112 | ||
2113 | for (i = 0; i < count; i++) | |
2114 | buf[i] = BSWAP_16(buf[i]); | |
2115 | } | |
2116 | ||
2117 | /* ARGSUSED */ | |
2118 | void | |
2119 | byteswap_uint8_array(void *vbuf, size_t size) | |
2120 | { | |
2121 | } | |
2122 | ||
2123 | void | |
2124 | dmu_init(void) | |
2125 | { | |
2126 | zfs_dbgmsg_init(); | |
2127 | sa_cache_init(); | |
2128 | xuio_stat_init(); | |
2129 | dmu_objset_init(); | |
2130 | dnode_init(); | |
2131 | dbuf_init(); | |
2132 | zfetch_init(); | |
2133 | dmu_tx_init(); | |
2134 | l2arc_init(); | |
2135 | arc_init(); | |
2136 | } | |
2137 | ||
2138 | void | |
2139 | dmu_fini(void) | |
2140 | { | |
2141 | arc_fini(); /* arc depends on l2arc, so arc must go first */ | |
2142 | l2arc_fini(); | |
2143 | dmu_tx_fini(); | |
2144 | zfetch_fini(); | |
2145 | dbuf_fini(); | |
2146 | dnode_fini(); | |
2147 | dmu_objset_fini(); | |
2148 | xuio_stat_fini(); | |
2149 | sa_cache_fini(); | |
2150 | zfs_dbgmsg_fini(); | |
2151 | } | |
2152 | ||
2153 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
2154 | EXPORT_SYMBOL(dmu_bonus_hold); | |
2155 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); | |
2156 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
2157 | EXPORT_SYMBOL(dmu_prefetch); | |
2158 | EXPORT_SYMBOL(dmu_free_range); | |
2159 | EXPORT_SYMBOL(dmu_free_long_range); | |
2160 | EXPORT_SYMBOL(dmu_free_long_object); | |
2161 | EXPORT_SYMBOL(dmu_read); | |
2162 | EXPORT_SYMBOL(dmu_write); | |
2163 | EXPORT_SYMBOL(dmu_prealloc); | |
2164 | EXPORT_SYMBOL(dmu_object_info); | |
2165 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
2166 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
2167 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
2168 | EXPORT_SYMBOL(dmu_object_set_blocksize); | |
2169 | EXPORT_SYMBOL(dmu_object_set_checksum); | |
2170 | EXPORT_SYMBOL(dmu_object_set_compress); | |
2171 | EXPORT_SYMBOL(dmu_write_policy); | |
2172 | EXPORT_SYMBOL(dmu_sync); | |
2173 | EXPORT_SYMBOL(dmu_request_arcbuf); | |
2174 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
2175 | EXPORT_SYMBOL(dmu_assign_arcbuf); | |
2176 | EXPORT_SYMBOL(dmu_buf_hold); | |
2177 | EXPORT_SYMBOL(dmu_ot); | |
2178 | ||
2179 | module_param(zfs_mdcomp_disable, int, 0644); | |
2180 | MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression"); | |
2181 | ||
2182 | module_param(zfs_nopwrite_enabled, int, 0644); | |
2183 | MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes"); | |
2184 | ||
2185 | #endif |