?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
4185 add new cryptographic checksums to ZFS: SHA-512, Skein, Edon-R
[unleashed.git] / usr / src / uts / common / fs / zfs / dmu.c
blob98b92a1dbe0951a9a24ba7f58d254b1908989f6c
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, 2015 by Delphix. All rights reserved.
24 */
25 /* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */
26 /* Copyright (c) 2013, Joyent, Inc. All rights reserved. */
27 /* Copyright (c) 2014, Nexenta Systems, Inc. All rights reserved. */
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>
52 #endif
53
54 /*
55 * Enable/disable nopwrite feature.
56 */
57 int zfs_nopwrite_enabled = 1;
58
59 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
60 { DMU_BSWAP_UINT8, TRUE, "unallocated" },
61 { DMU_BSWAP_ZAP, TRUE, "object directory" },
62 { DMU_BSWAP_UINT64, TRUE, "object array" },
63 { DMU_BSWAP_UINT8, TRUE, "packed nvlist" },
64 { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" },
65 { DMU_BSWAP_UINT64, TRUE, "bpobj" },
66 { DMU_BSWAP_UINT64, TRUE, "bpobj header" },
67 { DMU_BSWAP_UINT64, TRUE, "SPA space map header" },
68 { DMU_BSWAP_UINT64, TRUE, "SPA space map" },
69 { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" },
70 { DMU_BSWAP_DNODE, TRUE, "DMU dnode" },
71 { DMU_BSWAP_OBJSET, TRUE, "DMU objset" },
72 { DMU_BSWAP_UINT64, TRUE, "DSL directory" },
73 { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"},
74 { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" },
75 { DMU_BSWAP_ZAP, TRUE, "DSL props" },
76 { DMU_BSWAP_UINT64, TRUE, "DSL dataset" },
77 { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" },
78 { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" },
79 { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" },
80 { DMU_BSWAP_ZAP, TRUE, "ZFS directory" },
81 { DMU_BSWAP_ZAP, TRUE, "ZFS master node" },
82 { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" },
83 { DMU_BSWAP_UINT8, FALSE, "zvol object" },
84 { DMU_BSWAP_ZAP, TRUE, "zvol prop" },
85 { DMU_BSWAP_UINT8, FALSE, "other uint8[]" },
86 { DMU_BSWAP_UINT64, FALSE, "other uint64[]" },
87 { DMU_BSWAP_ZAP, TRUE, "other ZAP" },
88 { DMU_BSWAP_ZAP, TRUE, "persistent error log" },
89 { DMU_BSWAP_UINT8, TRUE, "SPA history" },
90 { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" },
91 { DMU_BSWAP_ZAP, TRUE, "Pool properties" },
92 { DMU_BSWAP_ZAP, TRUE, "DSL permissions" },
93 { DMU_BSWAP_ACL, TRUE, "ZFS ACL" },
94 { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" },
95 { DMU_BSWAP_UINT8, TRUE, "FUID table" },
96 { DMU_BSWAP_UINT64, TRUE, "FUID table size" },
97 { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"},
98 { DMU_BSWAP_ZAP, TRUE, "scan work queue" },
99 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" },
100 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" },
101 { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"},
102 { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" },
103 { DMU_BSWAP_ZAP, TRUE, "DDT statistics" },
104 { DMU_BSWAP_UINT8, TRUE, "System attributes" },
105 { DMU_BSWAP_ZAP, TRUE, "SA master node" },
106 { DMU_BSWAP_ZAP, TRUE, "SA attr registration" },
107 { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" },
108 { DMU_BSWAP_ZAP, TRUE, "scan translations" },
109 { DMU_BSWAP_UINT8, FALSE, "deduplicated block" },
110 { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" },
111 { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" },
112 { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" },
113 { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" }
114 };
115
116 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
117 { byteswap_uint8_array, "uint8" },
118 { byteswap_uint16_array, "uint16" },
119 { byteswap_uint32_array, "uint32" },
120 { byteswap_uint64_array, "uint64" },
121 { zap_byteswap, "zap" },
122 { dnode_buf_byteswap, "dnode" },
123 { dmu_objset_byteswap, "objset" },
124 { zfs_znode_byteswap, "znode" },
125 { zfs_oldacl_byteswap, "oldacl" },
126 { zfs_acl_byteswap, "acl" }
127 };
128
129 int
130 dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
131 void *tag, dmu_buf_t **dbp)
132 {
133 dnode_t *dn;
134 uint64_t blkid;
135 dmu_buf_impl_t *db;
136 int err;
137
138 err = dnode_hold(os, object, FTAG, &dn);
139 if (err)
140 return (err);
141 blkid = dbuf_whichblock(dn, 0, offset);
142 rw_enter(&dn->dn_struct_rwlock, RW_READER);
143 db = dbuf_hold(dn, blkid, tag);
144 rw_exit(&dn->dn_struct_rwlock);
145 dnode_rele(dn, FTAG);
146
147 if (db == NULL) {
148 *dbp = NULL;
149 return (SET_ERROR(EIO));
150 }
151
152 *dbp = &db->db;
153 return (err);
154 }
155
156 int
157 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
158 void *tag, dmu_buf_t **dbp, int flags)
159 {
160 int err;
161 int db_flags = DB_RF_CANFAIL;
162
163 if (flags & DMU_READ_NO_PREFETCH)
164 db_flags |= DB_RF_NOPREFETCH;
165
166 err = dmu_buf_hold_noread(os, object, offset, tag, dbp);
167 if (err == 0) {
168 dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
169 err = dbuf_read(db, NULL, db_flags);
170 if (err != 0) {
171 dbuf_rele(db, tag);
172 *dbp = NULL;
173 }
174 }
175
176 return (err);
177 }
178
179 int
180 dmu_bonus_max(void)
181 {
182 return (DN_MAX_BONUSLEN);
183 }
184
185 int
186 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
187 {
188 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
189 dnode_t *dn;
190 int error;
191
192 DB_DNODE_ENTER(db);
193 dn = DB_DNODE(db);
194
195 if (dn->dn_bonus != db) {
196 error = SET_ERROR(EINVAL);
197 } else if (newsize < 0 || newsize > db_fake->db_size) {
198 error = SET_ERROR(EINVAL);
199 } else {
200 dnode_setbonuslen(dn, newsize, tx);
201 error = 0;
202 }
203
204 DB_DNODE_EXIT(db);
205 return (error);
206 }
207
208 int
209 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
210 {
211 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
212 dnode_t *dn;
213 int error;
214
215 DB_DNODE_ENTER(db);
216 dn = DB_DNODE(db);
217
218 if (!DMU_OT_IS_VALID(type)) {
219 error = SET_ERROR(EINVAL);
220 } else if (dn->dn_bonus != db) {
221 error = SET_ERROR(EINVAL);
222 } else {
223 dnode_setbonus_type(dn, type, tx);
224 error = 0;
225 }
226
227 DB_DNODE_EXIT(db);
228 return (error);
229 }
230
231 dmu_object_type_t
232 dmu_get_bonustype(dmu_buf_t *db_fake)
233 {
234 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
235 dnode_t *dn;
236 dmu_object_type_t type;
237
238 DB_DNODE_ENTER(db);
239 dn = DB_DNODE(db);
240 type = dn->dn_bonustype;
241 DB_DNODE_EXIT(db);
242
243 return (type);
244 }
245
246 int
247 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
248 {
249 dnode_t *dn;
250 int error;
251
252 error = dnode_hold(os, object, FTAG, &dn);
253 dbuf_rm_spill(dn, tx);
254 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
255 dnode_rm_spill(dn, tx);
256 rw_exit(&dn->dn_struct_rwlock);
257 dnode_rele(dn, FTAG);
258 return (error);
259 }
260
261 /*
262 * returns ENOENT, EIO, or 0.
263 */
264 int
265 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
266 {
267 dnode_t *dn;
268 dmu_buf_impl_t *db;
269 int error;
270
271 error = dnode_hold(os, object, FTAG, &dn);
272 if (error)
273 return (error);
274
275 rw_enter(&dn->dn_struct_rwlock, RW_READER);
276 if (dn->dn_bonus == NULL) {
277 rw_exit(&dn->dn_struct_rwlock);
278 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
279 if (dn->dn_bonus == NULL)
280 dbuf_create_bonus(dn);
281 }
282 db = dn->dn_bonus;
283
284 /* as long as the bonus buf is held, the dnode will be held */
285 if (refcount_add(&db->db_holds, tag) == 1) {
286 VERIFY(dnode_add_ref(dn, db));
287 atomic_inc_32(&dn->dn_dbufs_count);
288 }
289
290 /*
291 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
292 * hold and incrementing the dbuf count to ensure that dnode_move() sees
293 * a dnode hold for every dbuf.
294 */
295 rw_exit(&dn->dn_struct_rwlock);
296
297 dnode_rele(dn, FTAG);
298
299 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
300
301 *dbp = &db->db;
302 return (0);
303 }
304
305 /*
306 * returns ENOENT, EIO, or 0.
307 *
308 * This interface will allocate a blank spill dbuf when a spill blk
309 * doesn't already exist on the dnode.
310 *
311 * if you only want to find an already existing spill db, then
312 * dmu_spill_hold_existing() should be used.
313 */
314 int
315 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
316 {
317 dmu_buf_impl_t *db = NULL;
318 int err;
319
320 if ((flags & DB_RF_HAVESTRUCT) == 0)
321 rw_enter(&dn->dn_struct_rwlock, RW_READER);
322
323 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
324
325 if ((flags & DB_RF_HAVESTRUCT) == 0)
326 rw_exit(&dn->dn_struct_rwlock);
327
328 ASSERT(db != NULL);
329 err = dbuf_read(db, NULL, flags);
330 if (err == 0)
331 *dbp = &db->db;
332 else
333 dbuf_rele(db, tag);
334 return (err);
335 }
336
337 int
338 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
339 {
340 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
341 dnode_t *dn;
342 int err;
343
344 DB_DNODE_ENTER(db);
345 dn = DB_DNODE(db);
346
347 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
348 err = SET_ERROR(EINVAL);
349 } else {
350 rw_enter(&dn->dn_struct_rwlock, RW_READER);
351
352 if (!dn->dn_have_spill) {
353 err = SET_ERROR(ENOENT);
354 } else {
355 err = dmu_spill_hold_by_dnode(dn,
356 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
357 }
358
359 rw_exit(&dn->dn_struct_rwlock);
360 }
361
362 DB_DNODE_EXIT(db);
363 return (err);
364 }
365
366 int
367 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
368 {
369 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
370 dnode_t *dn;
371 int err;
372
373 DB_DNODE_ENTER(db);
374 dn = DB_DNODE(db);
375 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
376 DB_DNODE_EXIT(db);
377
378 return (err);
379 }
380
381 /*
382 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
383 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
384 * and can induce severe lock contention when writing to several files
385 * whose dnodes are in the same block.
386 */
387 static int
388 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
389 boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
390 {
391 dmu_buf_t **dbp;
392 uint64_t blkid, nblks, i;
393 uint32_t dbuf_flags;
394 int err;
395 zio_t *zio;
396
397 ASSERT(length <= DMU_MAX_ACCESS);
398
399 /*
400 * Note: We directly notify the prefetch code of this read, so that
401 * we can tell it about the multi-block read. dbuf_read() only knows
402 * about the one block it is accessing.
403 */
404 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT |
405 DB_RF_NOPREFETCH;
406
407 rw_enter(&dn->dn_struct_rwlock, RW_READER);
408 if (dn->dn_datablkshift) {
409 int blkshift = dn->dn_datablkshift;
410 nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) -
411 P2ALIGN(offset, 1ULL << blkshift)) >> blkshift;
412 } else {
413 if (offset + length > dn->dn_datablksz) {
414 zfs_panic_recover("zfs: accessing past end of object "
415 "%llx/%llx (size=%u access=%llu+%llu)",
416 (longlong_t)dn->dn_objset->
417 os_dsl_dataset->ds_object,
418 (longlong_t)dn->dn_object, dn->dn_datablksz,
419 (longlong_t)offset, (longlong_t)length);
420 rw_exit(&dn->dn_struct_rwlock);
421 return (SET_ERROR(EIO));
422 }
423 nblks = 1;
424 }
425 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
426
427 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
428 blkid = dbuf_whichblock(dn, 0, offset);
429 for (i = 0; i < nblks; i++) {
430 dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag);
431 if (db == NULL) {
432 rw_exit(&dn->dn_struct_rwlock);
433 dmu_buf_rele_array(dbp, nblks, tag);
434 zio_nowait(zio);
435 return (SET_ERROR(EIO));
436 }
437
438 /* initiate async i/o */
439 if (read)
440 (void) dbuf_read(db, zio, dbuf_flags);
441 dbp[i] = &db->db;
442 }
443
444 if ((flags & DMU_READ_NO_PREFETCH) == 0 && read &&
445 length < zfetch_array_rd_sz) {
446 dmu_zfetch(&dn->dn_zfetch, blkid, nblks);
447 }
448 rw_exit(&dn->dn_struct_rwlock);
449
450 /* wait for async i/o */
451 err = zio_wait(zio);
452 if (err) {
453 dmu_buf_rele_array(dbp, nblks, tag);
454 return (err);
455 }
456
457 /* wait for other io to complete */
458 if (read) {
459 for (i = 0; i < nblks; i++) {
460 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
461 mutex_enter(&db->db_mtx);
462 while (db->db_state == DB_READ ||
463 db->db_state == DB_FILL)
464 cv_wait(&db->db_changed, &db->db_mtx);
465 if (db->db_state == DB_UNCACHED)
466 err = SET_ERROR(EIO);
467 mutex_exit(&db->db_mtx);
468 if (err) {
469 dmu_buf_rele_array(dbp, nblks, tag);
470 return (err);
471 }
472 }
473 }
474
475 *numbufsp = nblks;
476 *dbpp = dbp;
477 return (0);
478 }
479
480 static int
481 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
482 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
483 {
484 dnode_t *dn;
485 int err;
486
487 err = dnode_hold(os, object, FTAG, &dn);
488 if (err)
489 return (err);
490
491 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
492 numbufsp, dbpp, DMU_READ_PREFETCH);
493
494 dnode_rele(dn, FTAG);
495
496 return (err);
497 }
498
499 int
500 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
501 uint64_t length, boolean_t read, void *tag, int *numbufsp,
502 dmu_buf_t ***dbpp)
503 {
504 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
505 dnode_t *dn;
506 int err;
507
508 DB_DNODE_ENTER(db);
509 dn = DB_DNODE(db);
510 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
511 numbufsp, dbpp, DMU_READ_PREFETCH);
512 DB_DNODE_EXIT(db);
513
514 return (err);
515 }
516
517 void
518 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
519 {
520 int i;
521 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
522
523 if (numbufs == 0)
524 return;
525
526 for (i = 0; i < numbufs; i++) {
527 if (dbp[i])
528 dbuf_rele(dbp[i], tag);
529 }
530
531 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
532 }
533
534 /*
535 * Issue prefetch i/os for the given blocks. If level is greater than 0, the
536 * indirect blocks prefeteched will be those that point to the blocks containing
537 * the data starting at offset, and continuing to offset + len.
538 *
539 * Note that if the indirect blocks above the blocks being prefetched are not in
540 * cache, they will be asychronously read in.
541 */
542 void
543 dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
544 uint64_t len, zio_priority_t pri)
545 {
546 dnode_t *dn;
547 uint64_t blkid;
548 int nblks, err;
549
550 if (len == 0) { /* they're interested in the bonus buffer */
551 dn = DMU_META_DNODE(os);
552
553 if (object == 0 || object >= DN_MAX_OBJECT)
554 return;
555
556 rw_enter(&dn->dn_struct_rwlock, RW_READER);
557 blkid = dbuf_whichblock(dn, level,
558 object * sizeof (dnode_phys_t));
559 dbuf_prefetch(dn, level, blkid, pri, 0);
560 rw_exit(&dn->dn_struct_rwlock);
561 return;
562 }
563
564 /*
565 * XXX - Note, if the dnode for the requested object is not
566 * already cached, we will do a *synchronous* read in the
567 * dnode_hold() call. The same is true for any indirects.
568 */
569 err = dnode_hold(os, object, FTAG, &dn);
570 if (err != 0)
571 return;
572
573 rw_enter(&dn->dn_struct_rwlock, RW_READER);
574 /*
575 * offset + len - 1 is the last byte we want to prefetch for, and offset
576 * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the
577 * last block we want to prefetch, and dbuf_whichblock(dn, level,
578 * offset) is the first. Then the number we need to prefetch is the
579 * last - first + 1.
580 */
581 if (level > 0 || dn->dn_datablkshift != 0) {
582 nblks = dbuf_whichblock(dn, level, offset + len - 1) -
583 dbuf_whichblock(dn, level, offset) + 1;
584 } else {
585 nblks = (offset < dn->dn_datablksz);
586 }
587
588 if (nblks != 0) {
589 blkid = dbuf_whichblock(dn, level, offset);
590 for (int i = 0; i < nblks; i++)
591 dbuf_prefetch(dn, level, blkid + i, pri, 0);
592 }
593
594 rw_exit(&dn->dn_struct_rwlock);
595
596 dnode_rele(dn, FTAG);
597 }
598
599 /*
600 * Get the next "chunk" of file data to free. We traverse the file from
601 * the end so that the file gets shorter over time (if we crashes in the
602 * middle, this will leave us in a better state). We find allocated file
603 * data by simply searching the allocated level 1 indirects.
604 *
605 * On input, *start should be the first offset that does not need to be
606 * freed (e.g. "offset + length"). On return, *start will be the first
607 * offset that should be freed.
608 */
609 static int
610 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum)
611 {
612 uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
613 /* bytes of data covered by a level-1 indirect block */
614 uint64_t iblkrange =
615 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
616
617 ASSERT3U(minimum, <=, *start);
618
619 if (*start - minimum <= iblkrange * maxblks) {
620 *start = minimum;
621 return (0);
622 }
623 ASSERT(ISP2(iblkrange));
624
625 for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) {
626 int err;
627
628 /*
629 * dnode_next_offset(BACKWARDS) will find an allocated L1
630 * indirect block at or before the input offset. We must
631 * decrement *start so that it is at the end of the region
632 * to search.
633 */
634 (*start)--;
635 err = dnode_next_offset(dn,
636 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
637
638 /* if there are no indirect blocks before start, we are done */
639 if (err == ESRCH) {
640 *start = minimum;
641 break;
642 } else if (err != 0) {
643 return (err);
644 }
645
646 /* set start to the beginning of this L1 indirect */
647 *start = P2ALIGN(*start, iblkrange);
648 }
649 if (*start < minimum)
650 *start = minimum;
651 return (0);
652 }
653
654 static int
655 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
656 uint64_t length)
657 {
658 uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
659 int err;
660
661 if (offset >= object_size)
662 return (0);
663
664 if (length == DMU_OBJECT_END || offset + length > object_size)
665 length = object_size - offset;
666
667 while (length != 0) {
668 uint64_t chunk_end, chunk_begin;
669
670 chunk_end = chunk_begin = offset + length;
671
672 /* move chunk_begin backwards to the beginning of this chunk */
673 err = get_next_chunk(dn, &chunk_begin, offset);
674 if (err)
675 return (err);
676 ASSERT3U(chunk_begin, >=, offset);
677 ASSERT3U(chunk_begin, <=, chunk_end);
678
679 dmu_tx_t *tx = dmu_tx_create(os);
680 dmu_tx_hold_free(tx, dn->dn_object,
681 chunk_begin, chunk_end - chunk_begin);
682
683 /*
684 * Mark this transaction as typically resulting in a net
685 * reduction in space used.
686 */
687 dmu_tx_mark_netfree(tx);
688 err = dmu_tx_assign(tx, TXG_WAIT);
689 if (err) {
690 dmu_tx_abort(tx);
691 return (err);
692 }
693 dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx);
694 dmu_tx_commit(tx);
695
696 length -= chunk_end - chunk_begin;
697 }
698 return (0);
699 }
700
701 int
702 dmu_free_long_range(objset_t *os, uint64_t object,
703 uint64_t offset, uint64_t length)
704 {
705 dnode_t *dn;
706 int err;
707
708 err = dnode_hold(os, object, FTAG, &dn);
709 if (err != 0)
710 return (err);
711 err = dmu_free_long_range_impl(os, dn, offset, length);
712
713 /*
714 * It is important to zero out the maxblkid when freeing the entire
715 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
716 * will take the fast path, and (b) dnode_reallocate() can verify
717 * that the entire file has been freed.
718 */
719 if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
720 dn->dn_maxblkid = 0;
721
722 dnode_rele(dn, FTAG);
723 return (err);
724 }
725
726 int
727 dmu_free_long_object(objset_t *os, uint64_t object)
728 {
729 dmu_tx_t *tx;
730 int err;
731
732 err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
733 if (err != 0)
734 return (err);
735
736 tx = dmu_tx_create(os);
737 dmu_tx_hold_bonus(tx, object);
738 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
739 dmu_tx_mark_netfree(tx);
740 err = dmu_tx_assign(tx, TXG_WAIT);
741 if (err == 0) {
742 err = dmu_object_free(os, object, tx);
743 dmu_tx_commit(tx);
744 } else {
745 dmu_tx_abort(tx);
746 }
747
748 return (err);
749 }
750
751 int
752 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
753 uint64_t size, dmu_tx_t *tx)
754 {
755 dnode_t *dn;
756 int err = dnode_hold(os, object, FTAG, &dn);
757 if (err)
758 return (err);
759 ASSERT(offset < UINT64_MAX);
760 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
761 dnode_free_range(dn, offset, size, tx);
762 dnode_rele(dn, FTAG);
763 return (0);
764 }
765
766 int
767 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
768 void *buf, uint32_t flags)
769 {
770 dnode_t *dn;
771 dmu_buf_t **dbp;
772 int numbufs, err;
773
774 err = dnode_hold(os, object, FTAG, &dn);
775 if (err)
776 return (err);
777
778 /*
779 * Deal with odd block sizes, where there can't be data past the first
780 * block. If we ever do the tail block optimization, we will need to
781 * handle that here as well.
782 */
783 if (dn->dn_maxblkid == 0) {
784 int newsz = offset > dn->dn_datablksz ? 0 :
785 MIN(size, dn->dn_datablksz - offset);
786 bzero((char *)buf + newsz, size - newsz);
787 size = newsz;
788 }
789
790 while (size > 0) {
791 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
792 int i;
793
794 /*
795 * NB: we could do this block-at-a-time, but it's nice
796 * to be reading in parallel.
797 */
798 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
799 TRUE, FTAG, &numbufs, &dbp, flags);
800 if (err)
801 break;
802
803 for (i = 0; i < numbufs; i++) {
804 int tocpy;
805 int bufoff;
806 dmu_buf_t *db = dbp[i];
807
808 ASSERT(size > 0);
809
810 bufoff = offset - db->db_offset;
811 tocpy = (int)MIN(db->db_size - bufoff, size);
812
813 bcopy((char *)db->db_data + bufoff, buf, tocpy);
814
815 offset += tocpy;
816 size -= tocpy;
817 buf = (char *)buf + tocpy;
818 }
819 dmu_buf_rele_array(dbp, numbufs, FTAG);
820 }
821 dnode_rele(dn, FTAG);
822 return (err);
823 }
824
825 void
826 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
827 const void *buf, dmu_tx_t *tx)
828 {
829 dmu_buf_t **dbp;
830 int numbufs, i;
831
832 if (size == 0)
833 return;
834
835 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
836 FALSE, FTAG, &numbufs, &dbp));
837
838 for (i = 0; i < numbufs; i++) {
839 int tocpy;
840 int bufoff;
841 dmu_buf_t *db = dbp[i];
842
843 ASSERT(size > 0);
844
845 bufoff = offset - db->db_offset;
846 tocpy = (int)MIN(db->db_size - bufoff, size);
847
848 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
849
850 if (tocpy == db->db_size)
851 dmu_buf_will_fill(db, tx);
852 else
853 dmu_buf_will_dirty(db, tx);
854
855 bcopy(buf, (char *)db->db_data + bufoff, tocpy);
856
857 if (tocpy == db->db_size)
858 dmu_buf_fill_done(db, tx);
859
860 offset += tocpy;
861 size -= tocpy;
862 buf = (char *)buf + tocpy;
863 }
864 dmu_buf_rele_array(dbp, numbufs, FTAG);
865 }
866
867 void
868 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
869 dmu_tx_t *tx)
870 {
871 dmu_buf_t **dbp;
872 int numbufs, i;
873
874 if (size == 0)
875 return;
876
877 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
878 FALSE, FTAG, &numbufs, &dbp));
879
880 for (i = 0; i < numbufs; i++) {
881 dmu_buf_t *db = dbp[i];
882
883 dmu_buf_will_not_fill(db, tx);
884 }
885 dmu_buf_rele_array(dbp, numbufs, FTAG);
886 }
887
888 void
889 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
890 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
891 int compressed_size, int byteorder, dmu_tx_t *tx)
892 {
893 dmu_buf_t *db;
894
895 ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
896 ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
897 VERIFY0(dmu_buf_hold_noread(os, object, offset,
898 FTAG, &db));
899
900 dmu_buf_write_embedded(db,
901 data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
902 uncompressed_size, compressed_size, byteorder, tx);
903
904 dmu_buf_rele(db, FTAG);
905 }
906
907 /*
908 * DMU support for xuio
909 */
910 kstat_t *xuio_ksp = NULL;
911
912 int
913 dmu_xuio_init(xuio_t *xuio, int nblk)
914 {
915 dmu_xuio_t *priv;
916 uio_t *uio = &xuio->xu_uio;
917
918 uio->uio_iovcnt = nblk;
919 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
920
921 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
922 priv->cnt = nblk;
923 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
924 priv->iovp = uio->uio_iov;
925 XUIO_XUZC_PRIV(xuio) = priv;
926
927 if (XUIO_XUZC_RW(xuio) == UIO_READ)
928 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
929 else
930 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
931
932 return (0);
933 }
934
935 void
936 dmu_xuio_fini(xuio_t *xuio)
937 {
938 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
939 int nblk = priv->cnt;
940
941 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
942 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
943 kmem_free(priv, sizeof (dmu_xuio_t));
944
945 if (XUIO_XUZC_RW(xuio) == UIO_READ)
946 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
947 else
948 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
949 }
950
951 /*
952 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
953 * and increase priv->next by 1.
954 */
955 int
956 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
957 {
958 struct iovec *iov;
959 uio_t *uio = &xuio->xu_uio;
960 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
961 int i = priv->next++;
962
963 ASSERT(i < priv->cnt);
964 ASSERT(off + n <= arc_buf_size(abuf));
965 iov = uio->uio_iov + i;
966 iov->iov_base = (char *)abuf->b_data + off;
967 iov->iov_len = n;
968 priv->bufs[i] = abuf;
969 return (0);
970 }
971
972 int
973 dmu_xuio_cnt(xuio_t *xuio)
974 {
975 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
976 return (priv->cnt);
977 }
978
979 arc_buf_t *
980 dmu_xuio_arcbuf(xuio_t *xuio, int i)
981 {
982 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
983
984 ASSERT(i < priv->cnt);
985 return (priv->bufs[i]);
986 }
987
988 void
989 dmu_xuio_clear(xuio_t *xuio, int i)
990 {
991 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
992
993 ASSERT(i < priv->cnt);
994 priv->bufs[i] = NULL;
995 }
996
997 static void
998 xuio_stat_init(void)
999 {
1000 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
1001 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
1002 KSTAT_FLAG_VIRTUAL);
1003 if (xuio_ksp != NULL) {
1004 xuio_ksp->ks_data = &xuio_stats;
1005 kstat_install(xuio_ksp);
1006 }
1007 }
1008
1009 static void
1010 xuio_stat_fini(void)
1011 {
1012 if (xuio_ksp != NULL) {
1013 kstat_delete(xuio_ksp);
1014 xuio_ksp = NULL;
1015 }
1016 }
1017
1018 void
1019 xuio_stat_wbuf_copied()
1020 {
1021 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1022 }
1023
1024 void
1025 xuio_stat_wbuf_nocopy()
1026 {
1027 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1028 }
1029
1030 #ifdef _KERNEL
1031 static int
1032 dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
1033 {
1034 dmu_buf_t **dbp;
1035 int numbufs, i, err;
1036 xuio_t *xuio = NULL;
1037
1038 /*
1039 * NB: we could do this block-at-a-time, but it's nice
1040 * to be reading in parallel.
1041 */
1042 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1043 TRUE, FTAG, &numbufs, &dbp, 0);
1044 if (err)
1045 return (err);
1046
1047 if (uio->uio_extflg == UIO_XUIO)
1048 xuio = (xuio_t *)uio;
1049
1050 for (i = 0; i < numbufs; i++) {
1051 int tocpy;
1052 int bufoff;
1053 dmu_buf_t *db = dbp[i];
1054
1055 ASSERT(size > 0);
1056
1057 bufoff = uio->uio_loffset - db->db_offset;
1058 tocpy = (int)MIN(db->db_size - bufoff, size);
1059
1060 if (xuio) {
1061 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1062 arc_buf_t *dbuf_abuf = dbi->db_buf;
1063 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1064 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1065 if (!err) {
1066 uio->uio_resid -= tocpy;
1067 uio->uio_loffset += tocpy;
1068 }
1069
1070 if (abuf == dbuf_abuf)
1071 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1072 else
1073 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1074 } else {
1075 err = uiomove((char *)db->db_data + bufoff, tocpy,
1076 UIO_READ, uio);
1077 }
1078 if (err)
1079 break;
1080
1081 size -= tocpy;
1082 }
1083 dmu_buf_rele_array(dbp, numbufs, FTAG);
1084
1085 return (err);
1086 }
1087
1088 /*
1089 * Read 'size' bytes into the uio buffer.
1090 * From object zdb->db_object.
1091 * Starting at offset uio->uio_loffset.
1092 *
1093 * If the caller already has a dbuf in the target object
1094 * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1095 * because we don't have to find the dnode_t for the object.
1096 */
1097 int
1098 dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size)
1099 {
1100 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1101 dnode_t *dn;
1102 int err;
1103
1104 if (size == 0)
1105 return (0);
1106
1107 DB_DNODE_ENTER(db);
1108 dn = DB_DNODE(db);
1109 err = dmu_read_uio_dnode(dn, uio, size);
1110 DB_DNODE_EXIT(db);
1111
1112 return (err);
1113 }
1114
1115 /*
1116 * Read 'size' bytes into the uio buffer.
1117 * From the specified object
1118 * Starting at offset uio->uio_loffset.
1119 */
1120 int
1121 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1122 {
1123 dnode_t *dn;
1124 int err;
1125
1126 if (size == 0)
1127 return (0);
1128
1129 err = dnode_hold(os, object, FTAG, &dn);
1130 if (err)
1131 return (err);
1132
1133 err = dmu_read_uio_dnode(dn, uio, size);
1134
1135 dnode_rele(dn, FTAG);
1136
1137 return (err);
1138 }
1139
1140 static int
1141 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1142 {
1143 dmu_buf_t **dbp;
1144 int numbufs;
1145 int err = 0;
1146 int i;
1147
1148 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1149 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1150 if (err)
1151 return (err);
1152
1153 for (i = 0; i < numbufs; i++) {
1154 int tocpy;
1155 int bufoff;
1156 dmu_buf_t *db = dbp[i];
1157
1158 ASSERT(size > 0);
1159
1160 bufoff = uio->uio_loffset - db->db_offset;
1161 tocpy = (int)MIN(db->db_size - bufoff, size);
1162
1163 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1164
1165 if (tocpy == db->db_size)
1166 dmu_buf_will_fill(db, tx);
1167 else
1168 dmu_buf_will_dirty(db, tx);
1169
1170 /*
1171 * XXX uiomove could block forever (eg. nfs-backed
1172 * pages). There needs to be a uiolockdown() function
1173 * to lock the pages in memory, so that uiomove won't
1174 * block.
1175 */
1176 err = uiomove((char *)db->db_data + bufoff, tocpy,
1177 UIO_WRITE, uio);
1178
1179 if (tocpy == db->db_size)
1180 dmu_buf_fill_done(db, tx);
1181
1182 if (err)
1183 break;
1184
1185 size -= tocpy;
1186 }
1187
1188 dmu_buf_rele_array(dbp, numbufs, FTAG);
1189 return (err);
1190 }
1191
1192 /*
1193 * Write 'size' bytes from the uio buffer.
1194 * To object zdb->db_object.
1195 * Starting at offset uio->uio_loffset.
1196 *
1197 * If the caller already has a dbuf in the target object
1198 * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1199 * because we don't have to find the dnode_t for the object.
1200 */
1201 int
1202 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1203 dmu_tx_t *tx)
1204 {
1205 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1206 dnode_t *dn;
1207 int err;
1208
1209 if (size == 0)
1210 return (0);
1211
1212 DB_DNODE_ENTER(db);
1213 dn = DB_DNODE(db);
1214 err = dmu_write_uio_dnode(dn, uio, size, tx);
1215 DB_DNODE_EXIT(db);
1216
1217 return (err);
1218 }
1219
1220 /*
1221 * Write 'size' bytes from the uio buffer.
1222 * To the specified object.
1223 * Starting at offset uio->uio_loffset.
1224 */
1225 int
1226 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1227 dmu_tx_t *tx)
1228 {
1229 dnode_t *dn;
1230 int err;
1231
1232 if (size == 0)
1233 return (0);
1234
1235 err = dnode_hold(os, object, FTAG, &dn);
1236 if (err)
1237 return (err);
1238
1239 err = dmu_write_uio_dnode(dn, uio, size, tx);
1240
1241 dnode_rele(dn, FTAG);
1242
1243 return (err);
1244 }
1245
1246 int
1247 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1248 page_t *pp, dmu_tx_t *tx)
1249 {
1250 dmu_buf_t **dbp;
1251 int numbufs, i;
1252 int err;
1253
1254 if (size == 0)
1255 return (0);
1256
1257 err = dmu_buf_hold_array(os, object, offset, size,
1258 FALSE, FTAG, &numbufs, &dbp);
1259 if (err)
1260 return (err);
1261
1262 for (i = 0; i < numbufs; i++) {
1263 int tocpy, copied, thiscpy;
1264 int bufoff;
1265 dmu_buf_t *db = dbp[i];
1266 caddr_t va;
1267
1268 ASSERT(size > 0);
1269 ASSERT3U(db->db_size, >=, PAGESIZE);
1270
1271 bufoff = offset - db->db_offset;
1272 tocpy = (int)MIN(db->db_size - bufoff, size);
1273
1274 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1275
1276 if (tocpy == db->db_size)
1277 dmu_buf_will_fill(db, tx);
1278 else
1279 dmu_buf_will_dirty(db, tx);
1280
1281 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1282 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
1283 thiscpy = MIN(PAGESIZE, tocpy - copied);
1284 va = zfs_map_page(pp, S_READ);
1285 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1286 zfs_unmap_page(pp, va);
1287 pp = pp->p_next;
1288 bufoff += PAGESIZE;
1289 }
1290
1291 if (tocpy == db->db_size)
1292 dmu_buf_fill_done(db, tx);
1293
1294 offset += tocpy;
1295 size -= tocpy;
1296 }
1297 dmu_buf_rele_array(dbp, numbufs, FTAG);
1298 return (err);
1299 }
1300 #endif
1301
1302 /*
1303 * Allocate a loaned anonymous arc buffer.
1304 */
1305 arc_buf_t *
1306 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1307 {
1308 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1309
1310 return (arc_loan_buf(db->db_objset->os_spa, size));
1311 }
1312
1313 /*
1314 * Free a loaned arc buffer.
1315 */
1316 void
1317 dmu_return_arcbuf(arc_buf_t *buf)
1318 {
1319 arc_return_buf(buf, FTAG);
1320 VERIFY(arc_buf_remove_ref(buf, FTAG));
1321 }
1322
1323 /*
1324 * When possible directly assign passed loaned arc buffer to a dbuf.
1325 * If this is not possible copy the contents of passed arc buf via
1326 * dmu_write().
1327 */
1328 void
1329 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1330 dmu_tx_t *tx)
1331 {
1332 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1333 dnode_t *dn;
1334 dmu_buf_impl_t *db;
1335 uint32_t blksz = (uint32_t)arc_buf_size(buf);
1336 uint64_t blkid;
1337
1338 DB_DNODE_ENTER(dbuf);
1339 dn = DB_DNODE(dbuf);
1340 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1341 blkid = dbuf_whichblock(dn, 0, offset);
1342 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1343 rw_exit(&dn->dn_struct_rwlock);
1344 DB_DNODE_EXIT(dbuf);
1345
1346 /*
1347 * We can only assign if the offset is aligned, the arc buf is the
1348 * same size as the dbuf, and the dbuf is not metadata. It
1349 * can't be metadata because the loaned arc buf comes from the
1350 * user-data kmem arena.
1351 */
1352 if (offset == db->db.db_offset && blksz == db->db.db_size &&
1353 DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) {
1354 dbuf_assign_arcbuf(db, buf, tx);
1355 dbuf_rele(db, FTAG);
1356 } else {
1357 objset_t *os;
1358 uint64_t object;
1359
1360 DB_DNODE_ENTER(dbuf);
1361 dn = DB_DNODE(dbuf);
1362 os = dn->dn_objset;
1363 object = dn->dn_object;
1364 DB_DNODE_EXIT(dbuf);
1365
1366 dbuf_rele(db, FTAG);
1367 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1368 dmu_return_arcbuf(buf);
1369 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1370 }
1371 }
1372
1373 typedef struct {
1374 dbuf_dirty_record_t *dsa_dr;
1375 dmu_sync_cb_t *dsa_done;
1376 zgd_t *dsa_zgd;
1377 dmu_tx_t *dsa_tx;
1378 } dmu_sync_arg_t;
1379
1380 /* ARGSUSED */
1381 static void
1382 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1383 {
1384 dmu_sync_arg_t *dsa = varg;
1385 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1386 blkptr_t *bp = zio->io_bp;
1387
1388 if (zio->io_error == 0) {
1389 if (BP_IS_HOLE(bp)) {
1390 /*
1391 * A block of zeros may compress to a hole, but the
1392 * block size still needs to be known for replay.
1393 */
1394 BP_SET_LSIZE(bp, db->db_size);
1395 } else if (!BP_IS_EMBEDDED(bp)) {
1396 ASSERT(BP_GET_LEVEL(bp) == 0);
1397 bp->blk_fill = 1;
1398 }
1399 }
1400 }
1401
1402 static void
1403 dmu_sync_late_arrival_ready(zio_t *zio)
1404 {
1405 dmu_sync_ready(zio, NULL, zio->io_private);
1406 }
1407
1408 /* ARGSUSED */
1409 static void
1410 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1411 {
1412 dmu_sync_arg_t *dsa = varg;
1413 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1414 dmu_buf_impl_t *db = dr->dr_dbuf;
1415
1416 mutex_enter(&db->db_mtx);
1417 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1418 if (zio->io_error == 0) {
1419 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1420 if (dr->dt.dl.dr_nopwrite) {
1421 blkptr_t *bp = zio->io_bp;
1422 blkptr_t *bp_orig = &zio->io_bp_orig;
1423 uint8_t chksum = BP_GET_CHECKSUM(bp_orig);
1424
1425 ASSERT(BP_EQUAL(bp, bp_orig));
1426 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1427 ASSERT(zio_checksum_table[chksum].ci_flags &
1428 ZCHECKSUM_FLAG_NOPWRITE);
1429 }
1430 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1431 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1432 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1433
1434 /*
1435 * Old style holes are filled with all zeros, whereas
1436 * new-style holes maintain their lsize, type, level,
1437 * and birth time (see zio_write_compress). While we
1438 * need to reset the BP_SET_LSIZE() call that happened
1439 * in dmu_sync_ready for old style holes, we do *not*
1440 * want to wipe out the information contained in new
1441 * style holes. Thus, only zero out the block pointer if
1442 * it's an old style hole.
1443 */
1444 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) &&
1445 dr->dt.dl.dr_overridden_by.blk_birth == 0)
1446 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1447 } else {
1448 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1449 }
1450 cv_broadcast(&db->db_changed);
1451 mutex_exit(&db->db_mtx);
1452
1453 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1454
1455 kmem_free(dsa, sizeof (*dsa));
1456 }
1457
1458 static void
1459 dmu_sync_late_arrival_done(zio_t *zio)
1460 {
1461 blkptr_t *bp = zio->io_bp;
1462 dmu_sync_arg_t *dsa = zio->io_private;
1463 blkptr_t *bp_orig = &zio->io_bp_orig;
1464
1465 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1466 /*
1467 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1468 * then there is nothing to do here. Otherwise, free the
1469 * newly allocated block in this txg.
1470 */
1471 if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
1472 ASSERT(BP_EQUAL(bp, bp_orig));
1473 } else {
1474 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1475 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1476 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1477 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1478 }
1479 }
1480
1481 dmu_tx_commit(dsa->dsa_tx);
1482
1483 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1484
1485 kmem_free(dsa, sizeof (*dsa));
1486 }
1487
1488 static int
1489 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1490 zio_prop_t *zp, zbookmark_phys_t *zb)
1491 {
1492 dmu_sync_arg_t *dsa;
1493 dmu_tx_t *tx;
1494
1495 tx = dmu_tx_create(os);
1496 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1497 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1498 dmu_tx_abort(tx);
1499 /* Make zl_get_data do txg_waited_synced() */
1500 return (SET_ERROR(EIO));
1501 }
1502
1503 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1504 dsa->dsa_dr = NULL;
1505 dsa->dsa_done = done;
1506 dsa->dsa_zgd = zgd;
1507 dsa->dsa_tx = tx;
1508
1509 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1510 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
1511 dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa,
1512 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
1513
1514 return (0);
1515 }
1516
1517 /*
1518 * Intent log support: sync the block associated with db to disk.
1519 * N.B. and XXX: the caller is responsible for making sure that the
1520 * data isn't changing while dmu_sync() is writing it.
1521 *
1522 * Return values:
1523 *
1524 * EEXIST: this txg has already been synced, so there's nothing to do.
1525 * The caller should not log the write.
1526 *
1527 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1528 * The caller should not log the write.
1529 *
1530 * EALREADY: this block is already in the process of being synced.
1531 * The caller should track its progress (somehow).
1532 *
1533 * EIO: could not do the I/O.
1534 * The caller should do a txg_wait_synced().
1535 *
1536 * 0: the I/O has been initiated.
1537 * The caller should log this blkptr in the done callback.
1538 * It is possible that the I/O will fail, in which case
1539 * the error will be reported to the done callback and
1540 * propagated to pio from zio_done().
1541 */
1542 int
1543 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1544 {
1545 blkptr_t *bp = zgd->zgd_bp;
1546 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1547 objset_t *os = db->db_objset;
1548 dsl_dataset_t *ds = os->os_dsl_dataset;
1549 dbuf_dirty_record_t *dr;
1550 dmu_sync_arg_t *dsa;
1551 zbookmark_phys_t zb;
1552 zio_prop_t zp;
1553 dnode_t *dn;
1554
1555 ASSERT(pio != NULL);
1556 ASSERT(txg != 0);
1557
1558 SET_BOOKMARK(&zb, ds->ds_object,
1559 db->db.db_object, db->db_level, db->db_blkid);
1560
1561 DB_DNODE_ENTER(db);
1562 dn = DB_DNODE(db);
1563 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1564 DB_DNODE_EXIT(db);
1565
1566 /*
1567 * If we're frozen (running ziltest), we always need to generate a bp.
1568 */
1569 if (txg > spa_freeze_txg(os->os_spa))
1570 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1571
1572 /*
1573 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1574 * and us. If we determine that this txg is not yet syncing,
1575 * but it begins to sync a moment later, that's OK because the
1576 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1577 */
1578 mutex_enter(&db->db_mtx);
1579
1580 if (txg <= spa_last_synced_txg(os->os_spa)) {
1581 /*
1582 * This txg has already synced. There's nothing to do.
1583 */
1584 mutex_exit(&db->db_mtx);
1585 return (SET_ERROR(EEXIST));
1586 }
1587
1588 if (txg <= spa_syncing_txg(os->os_spa)) {
1589 /*
1590 * This txg is currently syncing, so we can't mess with
1591 * the dirty record anymore; just write a new log block.
1592 */
1593 mutex_exit(&db->db_mtx);
1594 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1595 }
1596
1597 dr = db->db_last_dirty;
1598 while (dr && dr->dr_txg != txg)
1599 dr = dr->dr_next;
1600
1601 if (dr == NULL) {
1602 /*
1603 * There's no dr for this dbuf, so it must have been freed.
1604 * There's no need to log writes to freed blocks, so we're done.
1605 */
1606 mutex_exit(&db->db_mtx);
1607 return (SET_ERROR(ENOENT));
1608 }
1609
1610 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
1611
1612 /*
1613 * Assume the on-disk data is X, the current syncing data (in
1614 * txg - 1) is Y, and the current in-memory data is Z (currently
1615 * in dmu_sync).
1616 *
1617 * We usually want to perform a nopwrite if X and Z are the
1618 * same. However, if Y is different (i.e. the BP is going to
1619 * change before this write takes effect), then a nopwrite will
1620 * be incorrect - we would override with X, which could have
1621 * been freed when Y was written.
1622 *
1623 * (Note that this is not a concern when we are nop-writing from
1624 * syncing context, because X and Y must be identical, because
1625 * all previous txgs have been synced.)
1626 *
1627 * Therefore, we disable nopwrite if the current BP could change
1628 * before this TXG. There are two ways it could change: by
1629 * being dirty (dr_next is non-NULL), or by being freed
1630 * (dnode_block_freed()). This behavior is verified by
1631 * zio_done(), which VERIFYs that the override BP is identical
1632 * to the on-disk BP.
1633 */
1634 DB_DNODE_ENTER(db);
1635 dn = DB_DNODE(db);
1636 if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid))
1637 zp.zp_nopwrite = B_FALSE;
1638 DB_DNODE_EXIT(db);
1639
1640 ASSERT(dr->dr_txg == txg);
1641 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1642 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1643 /*
1644 * We have already issued a sync write for this buffer,
1645 * or this buffer has already been synced. It could not
1646 * have been dirtied since, or we would have cleared the state.
1647 */
1648 mutex_exit(&db->db_mtx);
1649 return (SET_ERROR(EALREADY));
1650 }
1651
1652 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1653 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1654 mutex_exit(&db->db_mtx);
1655
1656 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1657 dsa->dsa_dr = dr;
1658 dsa->dsa_done = done;
1659 dsa->dsa_zgd = zgd;
1660 dsa->dsa_tx = NULL;
1661
1662 zio_nowait(arc_write(pio, os->os_spa, txg,
1663 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
1664 DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready,
1665 NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE,
1666 ZIO_FLAG_CANFAIL, &zb));
1667
1668 return (0);
1669 }
1670
1671 int
1672 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1673 dmu_tx_t *tx)
1674 {
1675 dnode_t *dn;
1676 int err;
1677
1678 err = dnode_hold(os, object, FTAG, &dn);
1679 if (err)
1680 return (err);
1681 err = dnode_set_blksz(dn, size, ibs, tx);
1682 dnode_rele(dn, FTAG);
1683 return (err);
1684 }
1685
1686 void
1687 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1688 dmu_tx_t *tx)
1689 {
1690 dnode_t *dn;
1691
1692 /*
1693 * Send streams include each object's checksum function. This
1694 * check ensures that the receiving system can understand the
1695 * checksum function transmitted.
1696 */
1697 ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
1698
1699 VERIFY0(dnode_hold(os, object, FTAG, &dn));
1700 ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
1701 dn->dn_checksum = checksum;
1702 dnode_setdirty(dn, tx);
1703 dnode_rele(dn, FTAG);
1704 }
1705
1706 void
1707 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1708 dmu_tx_t *tx)
1709 {
1710 dnode_t *dn;
1711
1712 /*
1713 * Send streams include each object's compression function. This
1714 * check ensures that the receiving system can understand the
1715 * compression function transmitted.
1716 */
1717 ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
1718
1719 VERIFY0(dnode_hold(os, object, FTAG, &dn));
1720 dn->dn_compress = compress;
1721 dnode_setdirty(dn, tx);
1722 dnode_rele(dn, FTAG);
1723 }
1724
1725 int zfs_mdcomp_disable = 0;
1726
1727 /*
1728 * When the "redundant_metadata" property is set to "most", only indirect
1729 * blocks of this level and higher will have an additional ditto block.
1730 */
1731 int zfs_redundant_metadata_most_ditto_level = 2;
1732
1733 void
1734 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1735 {
1736 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1737 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1738 (wp & WP_SPILL));
1739 enum zio_checksum checksum = os->os_checksum;
1740 enum zio_compress compress = os->os_compress;
1741 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1742 boolean_t dedup = B_FALSE;
1743 boolean_t nopwrite = B_FALSE;
1744 boolean_t dedup_verify = os->os_dedup_verify;
1745 int copies = os->os_copies;
1746
1747 /*
1748 * We maintain different write policies for each of the following
1749 * types of data:
1750 * 1. metadata
1751 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
1752 * 3. all other level 0 blocks
1753 */
1754 if (ismd) {
1755 if (zfs_mdcomp_disable) {
1756 compress = ZIO_COMPRESS_EMPTY;
1757 } else {
1758 /*
1759 * XXX -- we should design a compression algorithm
1760 * that specializes in arrays of bps.
1761 */
1762 compress = zio_compress_select(os->os_spa,
1763 ZIO_COMPRESS_ON, ZIO_COMPRESS_ON);
1764 }
1765
1766 /*
1767 * Metadata always gets checksummed. If the data
1768 * checksum is multi-bit correctable, and it's not a
1769 * ZBT-style checksum, then it's suitable for metadata
1770 * as well. Otherwise, the metadata checksum defaults
1771 * to fletcher4.
1772 */
1773 if (!(zio_checksum_table[checksum].ci_flags &
1774 ZCHECKSUM_FLAG_METADATA) ||
1775 (zio_checksum_table[checksum].ci_flags &
1776 ZCHECKSUM_FLAG_EMBEDDED))
1777 checksum = ZIO_CHECKSUM_FLETCHER_4;
1778
1779 if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
1780 (os->os_redundant_metadata ==
1781 ZFS_REDUNDANT_METADATA_MOST &&
1782 (level >= zfs_redundant_metadata_most_ditto_level ||
1783 DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))))
1784 copies++;
1785 } else if (wp & WP_NOFILL) {
1786 ASSERT(level == 0);
1787
1788 /*
1789 * If we're writing preallocated blocks, we aren't actually
1790 * writing them so don't set any policy properties. These
1791 * blocks are currently only used by an external subsystem
1792 * outside of zfs (i.e. dump) and not written by the zio
1793 * pipeline.
1794 */
1795 compress = ZIO_COMPRESS_OFF;
1796 checksum = ZIO_CHECKSUM_NOPARITY;
1797 } else {
1798 compress = zio_compress_select(os->os_spa, dn->dn_compress,
1799 compress);
1800
1801 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
1802 zio_checksum_select(dn->dn_checksum, checksum) :
1803 dedup_checksum;
1804
1805 /*
1806 * Determine dedup setting. If we are in dmu_sync(),
1807 * we won't actually dedup now because that's all
1808 * done in syncing context; but we do want to use the
1809 * dedup checkum. If the checksum is not strong
1810 * enough to ensure unique signatures, force
1811 * dedup_verify.
1812 */
1813 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
1814 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
1815 if (!(zio_checksum_table[checksum].ci_flags &
1816 ZCHECKSUM_FLAG_DEDUP))
1817 dedup_verify = B_TRUE;
1818 }
1819
1820 /*
1821 * Enable nopwrite if we have secure enough checksum
1822 * algorithm (see comment in zio_nop_write) and
1823 * compression is enabled. We don't enable nopwrite if
1824 * dedup is enabled as the two features are mutually
1825 * exclusive.
1826 */
1827 nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags &
1828 ZCHECKSUM_FLAG_NOPWRITE) &&
1829 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
1830 }
1831
1832 zp->zp_checksum = checksum;
1833 zp->zp_compress = compress;
1834 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
1835 zp->zp_level = level;
1836 zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
1837 zp->zp_dedup = dedup;
1838 zp->zp_dedup_verify = dedup && dedup_verify;
1839 zp->zp_nopwrite = nopwrite;
1840 }
1841
1842 int
1843 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1844 {
1845 dnode_t *dn;
1846 int err;
1847
1848 /*
1849 * Sync any current changes before
1850 * we go trundling through the block pointers.
1851 */
1852 err = dmu_object_wait_synced(os, object);
1853 if (err) {
1854 return (err);
1855 }
1856
1857 err = dnode_hold(os, object, FTAG, &dn);
1858 if (err) {
1859 return (err);
1860 }
1861
1862 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1863 dnode_rele(dn, FTAG);
1864
1865 return (err);
1866 }
1867
1868 /*
1869 * Given the ZFS object, if it contains any dirty nodes
1870 * this function flushes all dirty blocks to disk. This
1871 * ensures the DMU object info is updated. A more efficient
1872 * future version might just find the TXG with the maximum
1873 * ID and wait for that to be synced.
1874 */
1875 int
1876 dmu_object_wait_synced(objset_t *os, uint64_t object) {
1877 dnode_t *dn;
1878 int error, i;
1879
1880 error = dnode_hold(os, object, FTAG, &dn);
1881 if (error) {
1882 return (error);
1883 }
1884
1885 for (i = 0; i < TXG_SIZE; i++) {
1886 if (list_link_active(&dn->dn_dirty_link[i])) {
1887 break;
1888 }
1889 }
1890 dnode_rele(dn, FTAG);
1891 if (i != TXG_SIZE) {
1892 txg_wait_synced(dmu_objset_pool(os), 0);
1893 }
1894
1895 return (0);
1896 }
1897
1898 void
1899 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1900 {
1901 dnode_phys_t *dnp;
1902
1903 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1904 mutex_enter(&dn->dn_mtx);
1905
1906 dnp = dn->dn_phys;
1907
1908 doi->doi_data_block_size = dn->dn_datablksz;
1909 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1910 1ULL << dn->dn_indblkshift : 0;
1911 doi->doi_type = dn->dn_type;
1912 doi->doi_bonus_type = dn->dn_bonustype;
1913 doi->doi_bonus_size = dn->dn_bonuslen;
1914 doi->doi_indirection = dn->dn_nlevels;
1915 doi->doi_checksum = dn->dn_checksum;
1916 doi->doi_compress = dn->dn_compress;
1917 doi->doi_nblkptr = dn->dn_nblkptr;
1918 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1919 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
1920 doi->doi_fill_count = 0;
1921 for (int i = 0; i < dnp->dn_nblkptr; i++)
1922 doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
1923
1924 mutex_exit(&dn->dn_mtx);
1925 rw_exit(&dn->dn_struct_rwlock);
1926 }
1927
1928 /*
1929 * Get information on a DMU object.
1930 * If doi is NULL, just indicates whether the object exists.
1931 */
1932 int
1933 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1934 {
1935 dnode_t *dn;
1936 int err = dnode_hold(os, object, FTAG, &dn);
1937
1938 if (err)
1939 return (err);
1940
1941 if (doi != NULL)
1942 dmu_object_info_from_dnode(dn, doi);
1943
1944 dnode_rele(dn, FTAG);
1945 return (0);
1946 }
1947
1948 /*
1949 * As above, but faster; can be used when you have a held dbuf in hand.
1950 */
1951 void
1952 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1953 {
1954 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1955
1956 DB_DNODE_ENTER(db);
1957 dmu_object_info_from_dnode(DB_DNODE(db), doi);
1958 DB_DNODE_EXIT(db);
1959 }
1960
1961 /*
1962 * Faster still when you only care about the size.
1963 * This is specifically optimized for zfs_getattr().
1964 */
1965 void
1966 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1967 u_longlong_t *nblk512)
1968 {
1969 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1970 dnode_t *dn;
1971
1972 DB_DNODE_ENTER(db);
1973 dn = DB_DNODE(db);
1974
1975 *blksize = dn->dn_datablksz;
1976 /* add 1 for dnode space */
1977 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1978 SPA_MINBLOCKSHIFT) + 1;
1979 DB_DNODE_EXIT(db);
1980 }
1981
1982 void
1983 byteswap_uint64_array(void *vbuf, size_t size)
1984 {
1985 uint64_t *buf = vbuf;
1986 size_t count = size >> 3;
1987 int i;
1988
1989 ASSERT((size & 7) == 0);
1990
1991 for (i = 0; i < count; i++)
1992 buf[i] = BSWAP_64(buf[i]);
1993 }
1994
1995 void
1996 byteswap_uint32_array(void *vbuf, size_t size)
1997 {
1998 uint32_t *buf = vbuf;
1999 size_t count = size >> 2;
2000 int i;
2001
2002 ASSERT((size & 3) == 0);
2003
2004 for (i = 0; i < count; i++)
2005 buf[i] = BSWAP_32(buf[i]);
2006 }
2007
2008 void
2009 byteswap_uint16_array(void *vbuf, size_t size)
2010 {
2011 uint16_t *buf = vbuf;
2012 size_t count = size >> 1;
2013 int i;
2014
2015 ASSERT((size & 1) == 0);
2016
2017 for (i = 0; i < count; i++)
2018 buf[i] = BSWAP_16(buf[i]);
2019 }
2020
2021 /* ARGSUSED */
2022 void
2023 byteswap_uint8_array(void *vbuf, size_t size)
2024 {
2025 }
2026
2027 void
2028 dmu_init(void)
2029 {
2030 zfs_dbgmsg_init();
2031 sa_cache_init();
2032 xuio_stat_init();
2033 dmu_objset_init();
2034 dnode_init();
2035 dbuf_init();
2036 zfetch_init();
2037 l2arc_init();
2038 arc_init();
2039 }
2040
2041 void
2042 dmu_fini(void)
2043 {
2044 arc_fini(); /* arc depends on l2arc, so arc must go first */
2045 l2arc_fini();
2046 zfetch_fini();
2047 dbuf_fini();
2048 dnode_fini();
2049 dmu_objset_fini();
2050 xuio_stat_fini();
2051 sa_cache_fini();
2052 zfs_dbgmsg_fini();
2053 }
Unleashed OS