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