]> 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
OpenZFS 6513 - partially filled holes lose birth time
[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, 2016 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) 2014, 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 #ifdef _KERNEL
51 #include <sys/vmsystm.h>
52 #include <sys/zfs_znode.h>
53 #endif
54
55 /*
56 * Enable/disable nopwrite feature.
57 */
58 int zfs_nopwrite_enabled = 1;
59
60 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
61 { DMU_BSWAP_UINT8, TRUE, "unallocated" },
62 { DMU_BSWAP_ZAP, TRUE, "object directory" },
63 { DMU_BSWAP_UINT64, TRUE, "object array" },
64 { DMU_BSWAP_UINT8, TRUE, "packed nvlist" },
65 { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" },
66 { DMU_BSWAP_UINT64, TRUE, "bpobj" },
67 { DMU_BSWAP_UINT64, TRUE, "bpobj header" },
68 { DMU_BSWAP_UINT64, TRUE, "SPA space map header" },
69 { DMU_BSWAP_UINT64, TRUE, "SPA space map" },
70 { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" },
71 { DMU_BSWAP_DNODE, TRUE, "DMU dnode" },
72 { DMU_BSWAP_OBJSET, TRUE, "DMU objset" },
73 { DMU_BSWAP_UINT64, TRUE, "DSL directory" },
74 { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"},
75 { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" },
76 { DMU_BSWAP_ZAP, TRUE, "DSL props" },
77 { DMU_BSWAP_UINT64, TRUE, "DSL dataset" },
78 { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" },
79 { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" },
80 { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" },
81 { DMU_BSWAP_ZAP, TRUE, "ZFS directory" },
82 { DMU_BSWAP_ZAP, TRUE, "ZFS master node" },
83 { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" },
84 { DMU_BSWAP_UINT8, FALSE, "zvol object" },
85 { DMU_BSWAP_ZAP, TRUE, "zvol prop" },
86 { DMU_BSWAP_UINT8, FALSE, "other uint8[]" },
87 { DMU_BSWAP_UINT64, FALSE, "other uint64[]" },
88 { DMU_BSWAP_ZAP, TRUE, "other ZAP" },
89 { DMU_BSWAP_ZAP, TRUE, "persistent error log" },
90 { DMU_BSWAP_UINT8, TRUE, "SPA history" },
91 { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" },
92 { DMU_BSWAP_ZAP, TRUE, "Pool properties" },
93 { DMU_BSWAP_ZAP, TRUE, "DSL permissions" },
94 { DMU_BSWAP_ACL, TRUE, "ZFS ACL" },
95 { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" },
96 { DMU_BSWAP_UINT8, TRUE, "FUID table" },
97 { DMU_BSWAP_UINT64, TRUE, "FUID table size" },
98 { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"},
99 { DMU_BSWAP_ZAP, TRUE, "scan work queue" },
100 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" },
101 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" },
102 { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"},
103 { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" },
104 { DMU_BSWAP_ZAP, TRUE, "DDT statistics" },
105 { DMU_BSWAP_UINT8, TRUE, "System attributes" },
106 { DMU_BSWAP_ZAP, TRUE, "SA master node" },
107 { DMU_BSWAP_ZAP, TRUE, "SA attr registration" },
108 { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" },
109 { DMU_BSWAP_ZAP, TRUE, "scan translations" },
110 { DMU_BSWAP_UINT8, FALSE, "deduplicated block" },
111 { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" },
112 { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" },
113 { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" },
114 { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" }
115 };
116
117 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
118 { byteswap_uint8_array, "uint8" },
119 { byteswap_uint16_array, "uint16" },
120 { byteswap_uint32_array, "uint32" },
121 { byteswap_uint64_array, "uint64" },
122 { zap_byteswap, "zap" },
123 { dnode_buf_byteswap, "dnode" },
124 { dmu_objset_byteswap, "objset" },
125 { zfs_znode_byteswap, "znode" },
126 { zfs_oldacl_byteswap, "oldacl" },
127 { zfs_acl_byteswap, "acl" }
128 };
129
130 int
131 dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
132 void *tag, dmu_buf_t **dbp)
133 {
134 dnode_t *dn;
135 uint64_t blkid;
136 dmu_buf_impl_t *db;
137 int err;
138
139 err = dnode_hold(os, object, FTAG, &dn);
140 if (err)
141 return (err);
142 blkid = dbuf_whichblock(dn, 0, offset);
143 rw_enter(&dn->dn_struct_rwlock, RW_READER);
144 db = dbuf_hold(dn, blkid, tag);
145 rw_exit(&dn->dn_struct_rwlock);
146 dnode_rele(dn, FTAG);
147
148 if (db == NULL) {
149 *dbp = NULL;
150 return (SET_ERROR(EIO));
151 }
152
153 *dbp = &db->db;
154 return (err);
155 }
156
157 int
158 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
159 void *tag, dmu_buf_t **dbp, int flags)
160 {
161 int err;
162 int db_flags = DB_RF_CANFAIL;
163
164 if (flags & DMU_READ_NO_PREFETCH)
165 db_flags |= DB_RF_NOPREFETCH;
166
167 err = dmu_buf_hold_noread(os, object, offset, tag, dbp);
168 if (err == 0) {
169 dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
170 err = dbuf_read(db, NULL, db_flags);
171 if (err != 0) {
172 dbuf_rele(db, tag);
173 *dbp = NULL;
174 }
175 }
176
177 return (err);
178 }
179
180 int
181 dmu_bonus_max(void)
182 {
183 return (DN_MAX_BONUSLEN);
184 }
185
186 int
187 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
188 {
189 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
190 dnode_t *dn;
191 int error;
192
193 DB_DNODE_ENTER(db);
194 dn = DB_DNODE(db);
195
196 if (dn->dn_bonus != db) {
197 error = SET_ERROR(EINVAL);
198 } else if (newsize < 0 || newsize > db_fake->db_size) {
199 error = SET_ERROR(EINVAL);
200 } else {
201 dnode_setbonuslen(dn, newsize, tx);
202 error = 0;
203 }
204
205 DB_DNODE_EXIT(db);
206 return (error);
207 }
208
209 int
210 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
211 {
212 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
213 dnode_t *dn;
214 int error;
215
216 DB_DNODE_ENTER(db);
217 dn = DB_DNODE(db);
218
219 if (!DMU_OT_IS_VALID(type)) {
220 error = SET_ERROR(EINVAL);
221 } else if (dn->dn_bonus != db) {
222 error = SET_ERROR(EINVAL);
223 } else {
224 dnode_setbonus_type(dn, type, tx);
225 error = 0;
226 }
227
228 DB_DNODE_EXIT(db);
229 return (error);
230 }
231
232 dmu_object_type_t
233 dmu_get_bonustype(dmu_buf_t *db_fake)
234 {
235 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
236 dnode_t *dn;
237 dmu_object_type_t type;
238
239 DB_DNODE_ENTER(db);
240 dn = DB_DNODE(db);
241 type = dn->dn_bonustype;
242 DB_DNODE_EXIT(db);
243
244 return (type);
245 }
246
247 int
248 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
249 {
250 dnode_t *dn;
251 int error;
252
253 error = dnode_hold(os, object, FTAG, &dn);
254 dbuf_rm_spill(dn, tx);
255 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
256 dnode_rm_spill(dn, tx);
257 rw_exit(&dn->dn_struct_rwlock);
258 dnode_rele(dn, FTAG);
259 return (error);
260 }
261
262 /*
263 * returns ENOENT, EIO, or 0.
264 */
265 int
266 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
267 {
268 dnode_t *dn;
269 dmu_buf_impl_t *db;
270 int error;
271
272 error = dnode_hold(os, object, FTAG, &dn);
273 if (error)
274 return (error);
275
276 rw_enter(&dn->dn_struct_rwlock, RW_READER);
277 if (dn->dn_bonus == NULL) {
278 rw_exit(&dn->dn_struct_rwlock);
279 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
280 if (dn->dn_bonus == NULL)
281 dbuf_create_bonus(dn);
282 }
283 db = dn->dn_bonus;
284
285 /* as long as the bonus buf is held, the dnode will be held */
286 if (refcount_add(&db->db_holds, tag) == 1) {
287 VERIFY(dnode_add_ref(dn, db));
288 atomic_inc_32(&dn->dn_dbufs_count);
289 }
290
291 /*
292 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
293 * hold and incrementing the dbuf count to ensure that dnode_move() sees
294 * a dnode hold for every dbuf.
295 */
296 rw_exit(&dn->dn_struct_rwlock);
297
298 dnode_rele(dn, FTAG);
299
300 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
301
302 *dbp = &db->db;
303 return (0);
304 }
305
306 /*
307 * returns ENOENT, EIO, or 0.
308 *
309 * This interface will allocate a blank spill dbuf when a spill blk
310 * doesn't already exist on the dnode.
311 *
312 * if you only want to find an already existing spill db, then
313 * dmu_spill_hold_existing() should be used.
314 */
315 int
316 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
317 {
318 dmu_buf_impl_t *db = NULL;
319 int err;
320
321 if ((flags & DB_RF_HAVESTRUCT) == 0)
322 rw_enter(&dn->dn_struct_rwlock, RW_READER);
323
324 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
325
326 if ((flags & DB_RF_HAVESTRUCT) == 0)
327 rw_exit(&dn->dn_struct_rwlock);
328
329 ASSERT(db != NULL);
330 err = dbuf_read(db, NULL, flags);
331 if (err == 0)
332 *dbp = &db->db;
333 else
334 dbuf_rele(db, tag);
335 return (err);
336 }
337
338 int
339 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
340 {
341 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
342 dnode_t *dn;
343 int err;
344
345 DB_DNODE_ENTER(db);
346 dn = DB_DNODE(db);
347
348 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
349 err = SET_ERROR(EINVAL);
350 } else {
351 rw_enter(&dn->dn_struct_rwlock, RW_READER);
352
353 if (!dn->dn_have_spill) {
354 err = SET_ERROR(ENOENT);
355 } else {
356 err = dmu_spill_hold_by_dnode(dn,
357 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
358 }
359
360 rw_exit(&dn->dn_struct_rwlock);
361 }
362
363 DB_DNODE_EXIT(db);
364 return (err);
365 }
366
367 int
368 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
369 {
370 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
371 dnode_t *dn;
372 int err;
373
374 DB_DNODE_ENTER(db);
375 dn = DB_DNODE(db);
376 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
377 DB_DNODE_EXIT(db);
378
379 return (err);
380 }
381
382 /*
383 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
384 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
385 * and can induce severe lock contention when writing to several files
386 * whose dnodes are in the same block.
387 */
388 static int
389 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
390 boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
391 {
392 dmu_buf_t **dbp;
393 uint64_t blkid, nblks, i;
394 uint32_t dbuf_flags;
395 int err;
396 zio_t *zio;
397
398 ASSERT(length <= DMU_MAX_ACCESS);
399
400 /*
401 * Note: We directly notify the prefetch code of this read, so that
402 * we can tell it about the multi-block read. dbuf_read() only knows
403 * about the one block it is accessing.
404 */
405 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT |
406 DB_RF_NOPREFETCH;
407
408 rw_enter(&dn->dn_struct_rwlock, RW_READER);
409 if (dn->dn_datablkshift) {
410 int blkshift = dn->dn_datablkshift;
411 nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) -
412 P2ALIGN(offset, 1ULL << blkshift)) >> blkshift;
413 } else {
414 if (offset + length > dn->dn_datablksz) {
415 zfs_panic_recover("zfs: accessing past end of object "
416 "%llx/%llx (size=%u access=%llu+%llu)",
417 (longlong_t)dn->dn_objset->
418 os_dsl_dataset->ds_object,
419 (longlong_t)dn->dn_object, dn->dn_datablksz,
420 (longlong_t)offset, (longlong_t)length);
421 rw_exit(&dn->dn_struct_rwlock);
422 return (SET_ERROR(EIO));
423 }
424 nblks = 1;
425 }
426 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
427
428 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
429 blkid = dbuf_whichblock(dn, 0, offset);
430 for (i = 0; i < nblks; i++) {
431 dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag);
432 if (db == NULL) {
433 rw_exit(&dn->dn_struct_rwlock);
434 dmu_buf_rele_array(dbp, nblks, tag);
435 zio_nowait(zio);
436 return (SET_ERROR(EIO));
437 }
438
439 /* initiate async i/o */
440 if (read)
441 (void) dbuf_read(db, zio, dbuf_flags);
442 dbp[i] = &db->db;
443 }
444
445 if ((flags & DMU_READ_NO_PREFETCH) == 0 && read &&
446 length <= zfetch_array_rd_sz) {
447 dmu_zfetch(&dn->dn_zfetch, blkid, nblks);
448 }
449 rw_exit(&dn->dn_struct_rwlock);
450
451 /* wait for async i/o */
452 err = zio_wait(zio);
453 if (err) {
454 dmu_buf_rele_array(dbp, nblks, tag);
455 return (err);
456 }
457
458 /* wait for other io to complete */
459 if (read) {
460 for (i = 0; i < nblks; i++) {
461 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
462 mutex_enter(&db->db_mtx);
463 while (db->db_state == DB_READ ||
464 db->db_state == DB_FILL)
465 cv_wait(&db->db_changed, &db->db_mtx);
466 if (db->db_state == DB_UNCACHED)
467 err = SET_ERROR(EIO);
468 mutex_exit(&db->db_mtx);
469 if (err) {
470 dmu_buf_rele_array(dbp, nblks, tag);
471 return (err);
472 }
473 }
474 }
475
476 *numbufsp = nblks;
477 *dbpp = dbp;
478 return (0);
479 }
480
481 static int
482 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
483 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
484 {
485 dnode_t *dn;
486 int err;
487
488 err = dnode_hold(os, object, FTAG, &dn);
489 if (err)
490 return (err);
491
492 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
493 numbufsp, dbpp, DMU_READ_PREFETCH);
494
495 dnode_rele(dn, FTAG);
496
497 return (err);
498 }
499
500 int
501 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
502 uint64_t length, boolean_t read, void *tag, int *numbufsp,
503 dmu_buf_t ***dbpp)
504 {
505 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
506 dnode_t *dn;
507 int err;
508
509 DB_DNODE_ENTER(db);
510 dn = DB_DNODE(db);
511 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
512 numbufsp, dbpp, DMU_READ_PREFETCH);
513 DB_DNODE_EXIT(db);
514
515 return (err);
516 }
517
518 void
519 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
520 {
521 int i;
522 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
523
524 if (numbufs == 0)
525 return;
526
527 for (i = 0; i < numbufs; i++) {
528 if (dbp[i])
529 dbuf_rele(dbp[i], tag);
530 }
531
532 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
533 }
534
535 /*
536 * Issue prefetch i/os for the given blocks. If level is greater than 0, the
537 * indirect blocks prefeteched will be those that point to the blocks containing
538 * the data starting at offset, and continuing to offset + len.
539 *
540 * Note that if the indirect blocks above the blocks being prefetched are not in
541 * cache, they will be asychronously read in.
542 */
543 void
544 dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
545 uint64_t len, zio_priority_t pri)
546 {
547 dnode_t *dn;
548 uint64_t blkid;
549 int nblks, err;
550
551 if (len == 0) { /* they're interested in the bonus buffer */
552 dn = DMU_META_DNODE(os);
553
554 if (object == 0 || object >= DN_MAX_OBJECT)
555 return;
556
557 rw_enter(&dn->dn_struct_rwlock, RW_READER);
558 blkid = dbuf_whichblock(dn, level,
559 object * sizeof (dnode_phys_t));
560 dbuf_prefetch(dn, level, blkid, pri, 0);
561 rw_exit(&dn->dn_struct_rwlock);
562 return;
563 }
564
565 /*
566 * XXX - Note, if the dnode for the requested object is not
567 * already cached, we will do a *synchronous* read in the
568 * dnode_hold() call. The same is true for any indirects.
569 */
570 err = dnode_hold(os, object, FTAG, &dn);
571 if (err != 0)
572 return;
573
574 rw_enter(&dn->dn_struct_rwlock, RW_READER);
575 /*
576 * offset + len - 1 is the last byte we want to prefetch for, and offset
577 * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the
578 * last block we want to prefetch, and dbuf_whichblock(dn, level,
579 * offset) is the first. Then the number we need to prefetch is the
580 * last - first + 1.
581 */
582 if (level > 0 || dn->dn_datablkshift != 0) {
583 nblks = dbuf_whichblock(dn, level, offset + len - 1) -
584 dbuf_whichblock(dn, level, offset) + 1;
585 } else {
586 nblks = (offset < dn->dn_datablksz);
587 }
588
589 if (nblks != 0) {
590 int i;
591
592 blkid = dbuf_whichblock(dn, level, offset);
593 for (i = 0; i < nblks; i++)
594 dbuf_prefetch(dn, level, blkid + i, pri, 0);
595 }
596
597 rw_exit(&dn->dn_struct_rwlock);
598
599 dnode_rele(dn, FTAG);
600 }
601
602 /*
603 * Get the next "chunk" of file data to free. We traverse the file from
604 * the end so that the file gets shorter over time (if we crashes in the
605 * middle, this will leave us in a better state). We find allocated file
606 * data by simply searching the allocated level 1 indirects.
607 *
608 * On input, *start should be the first offset that does not need to be
609 * freed (e.g. "offset + length"). On return, *start will be the first
610 * offset that should be freed.
611 */
612 static int
613 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum)
614 {
615 uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
616 /* bytes of data covered by a level-1 indirect block */
617 uint64_t iblkrange =
618 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
619 uint64_t blks;
620
621 ASSERT3U(minimum, <=, *start);
622
623 if (*start - minimum <= iblkrange * maxblks) {
624 *start = minimum;
625 return (0);
626 }
627 ASSERT(ISP2(iblkrange));
628
629 for (blks = 0; *start > minimum && blks < maxblks; blks++) {
630 int err;
631
632 /*
633 * dnode_next_offset(BACKWARDS) will find an allocated L1
634 * indirect block at or before the input offset. We must
635 * decrement *start so that it is at the end of the region
636 * to search.
637 */
638 (*start)--;
639 err = dnode_next_offset(dn,
640 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
641
642 /* if there are no indirect blocks before start, we are done */
643 if (err == ESRCH) {
644 *start = minimum;
645 break;
646 } else if (err != 0) {
647 return (err);
648 }
649
650 /* set start to the beginning of this L1 indirect */
651 *start = P2ALIGN(*start, iblkrange);
652 }
653 if (*start < minimum)
654 *start = minimum;
655 return (0);
656 }
657
658 static int
659 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
660 uint64_t length)
661 {
662 uint64_t object_size;
663 int err;
664
665 if (dn == NULL)
666 return (SET_ERROR(EINVAL));
667
668 object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
669 if (offset >= object_size)
670 return (0);
671
672 if (length == DMU_OBJECT_END || offset + length > object_size)
673 length = object_size - offset;
674
675 while (length != 0) {
676 uint64_t chunk_end, chunk_begin;
677 dmu_tx_t *tx;
678
679 chunk_end = chunk_begin = offset + length;
680
681 /* move chunk_begin backwards to the beginning of this chunk */
682 err = get_next_chunk(dn, &chunk_begin, offset);
683 if (err)
684 return (err);
685 ASSERT3U(chunk_begin, >=, offset);
686 ASSERT3U(chunk_begin, <=, chunk_end);
687
688 tx = dmu_tx_create(os);
689 dmu_tx_hold_free(tx, dn->dn_object,
690 chunk_begin, chunk_end - chunk_begin);
691
692 /*
693 * Mark this transaction as typically resulting in a net
694 * reduction in space used.
695 */
696 dmu_tx_mark_netfree(tx);
697 err = dmu_tx_assign(tx, TXG_WAIT);
698 if (err) {
699 dmu_tx_abort(tx);
700 return (err);
701 }
702 dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx);
703 dmu_tx_commit(tx);
704
705 length -= chunk_end - chunk_begin;
706 }
707 return (0);
708 }
709
710 int
711 dmu_free_long_range(objset_t *os, uint64_t object,
712 uint64_t offset, uint64_t length)
713 {
714 dnode_t *dn;
715 int err;
716
717 err = dnode_hold(os, object, FTAG, &dn);
718 if (err != 0)
719 return (err);
720 err = dmu_free_long_range_impl(os, dn, offset, length);
721
722 /*
723 * It is important to zero out the maxblkid when freeing the entire
724 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
725 * will take the fast path, and (b) dnode_reallocate() can verify
726 * that the entire file has been freed.
727 */
728 if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
729 dn->dn_maxblkid = 0;
730
731 dnode_rele(dn, FTAG);
732 return (err);
733 }
734
735 int
736 dmu_free_long_object(objset_t *os, uint64_t object)
737 {
738 dmu_tx_t *tx;
739 int err;
740
741 err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
742 if (err != 0)
743 return (err);
744
745 tx = dmu_tx_create(os);
746 dmu_tx_hold_bonus(tx, object);
747 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
748 dmu_tx_mark_netfree(tx);
749 err = dmu_tx_assign(tx, TXG_WAIT);
750 if (err == 0) {
751 err = dmu_object_free(os, object, tx);
752 dmu_tx_commit(tx);
753 } else {
754 dmu_tx_abort(tx);
755 }
756
757 return (err);
758 }
759
760 int
761 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
762 uint64_t size, dmu_tx_t *tx)
763 {
764 dnode_t *dn;
765 int err = dnode_hold(os, object, FTAG, &dn);
766 if (err)
767 return (err);
768 ASSERT(offset < UINT64_MAX);
769 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
770 dnode_free_range(dn, offset, size, tx);
771 dnode_rele(dn, FTAG);
772 return (0);
773 }
774
775 int
776 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
777 void *buf, uint32_t flags)
778 {
779 dnode_t *dn;
780 dmu_buf_t **dbp;
781 int numbufs, err;
782
783 err = dnode_hold(os, object, FTAG, &dn);
784 if (err)
785 return (err);
786
787 /*
788 * Deal with odd block sizes, where there can't be data past the first
789 * block. If we ever do the tail block optimization, we will need to
790 * handle that here as well.
791 */
792 if (dn->dn_maxblkid == 0) {
793 uint64_t newsz = offset > dn->dn_datablksz ? 0 :
794 MIN(size, dn->dn_datablksz - offset);
795 bzero((char *)buf + newsz, size - newsz);
796 size = newsz;
797 }
798
799 while (size > 0) {
800 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
801 int i;
802
803 /*
804 * NB: we could do this block-at-a-time, but it's nice
805 * to be reading in parallel.
806 */
807 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
808 TRUE, FTAG, &numbufs, &dbp, flags);
809 if (err)
810 break;
811
812 for (i = 0; i < numbufs; i++) {
813 uint64_t tocpy;
814 int64_t bufoff;
815 dmu_buf_t *db = dbp[i];
816
817 ASSERT(size > 0);
818
819 bufoff = offset - db->db_offset;
820 tocpy = MIN(db->db_size - bufoff, size);
821
822 (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy);
823
824 offset += tocpy;
825 size -= tocpy;
826 buf = (char *)buf + tocpy;
827 }
828 dmu_buf_rele_array(dbp, numbufs, FTAG);
829 }
830 dnode_rele(dn, FTAG);
831 return (err);
832 }
833
834 void
835 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
836 const void *buf, dmu_tx_t *tx)
837 {
838 dmu_buf_t **dbp;
839 int numbufs, i;
840
841 if (size == 0)
842 return;
843
844 VERIFY0(dmu_buf_hold_array(os, object, offset, size,
845 FALSE, FTAG, &numbufs, &dbp));
846
847 for (i = 0; i < numbufs; i++) {
848 uint64_t tocpy;
849 int64_t bufoff;
850 dmu_buf_t *db = dbp[i];
851
852 ASSERT(size > 0);
853
854 bufoff = offset - db->db_offset;
855 tocpy = MIN(db->db_size - bufoff, size);
856
857 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
858
859 if (tocpy == db->db_size)
860 dmu_buf_will_fill(db, tx);
861 else
862 dmu_buf_will_dirty(db, tx);
863
864 (void) memcpy((char *)db->db_data + bufoff, buf, tocpy);
865
866 if (tocpy == db->db_size)
867 dmu_buf_fill_done(db, tx);
868
869 offset += tocpy;
870 size -= tocpy;
871 buf = (char *)buf + tocpy;
872 }
873 dmu_buf_rele_array(dbp, numbufs, FTAG);
874 }
875
876 void
877 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
878 dmu_tx_t *tx)
879 {
880 dmu_buf_t **dbp;
881 int numbufs, i;
882
883 if (size == 0)
884 return;
885
886 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
887 FALSE, FTAG, &numbufs, &dbp));
888
889 for (i = 0; i < numbufs; i++) {
890 dmu_buf_t *db = dbp[i];
891
892 dmu_buf_will_not_fill(db, tx);
893 }
894 dmu_buf_rele_array(dbp, numbufs, FTAG);
895 }
896
897 void
898 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
899 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
900 int compressed_size, int byteorder, dmu_tx_t *tx)
901 {
902 dmu_buf_t *db;
903
904 ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
905 ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
906 VERIFY0(dmu_buf_hold_noread(os, object, offset,
907 FTAG, &db));
908
909 dmu_buf_write_embedded(db,
910 data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
911 uncompressed_size, compressed_size, byteorder, tx);
912
913 dmu_buf_rele(db, FTAG);
914 }
915
916 /*
917 * DMU support for xuio
918 */
919 kstat_t *xuio_ksp = NULL;
920
921 typedef struct xuio_stats {
922 /* loaned yet not returned arc_buf */
923 kstat_named_t xuiostat_onloan_rbuf;
924 kstat_named_t xuiostat_onloan_wbuf;
925 /* whether a copy is made when loaning out a read buffer */
926 kstat_named_t xuiostat_rbuf_copied;
927 kstat_named_t xuiostat_rbuf_nocopy;
928 /* whether a copy is made when assigning a write buffer */
929 kstat_named_t xuiostat_wbuf_copied;
930 kstat_named_t xuiostat_wbuf_nocopy;
931 } xuio_stats_t;
932
933 static xuio_stats_t xuio_stats = {
934 { "onloan_read_buf", KSTAT_DATA_UINT64 },
935 { "onloan_write_buf", KSTAT_DATA_UINT64 },
936 { "read_buf_copied", KSTAT_DATA_UINT64 },
937 { "read_buf_nocopy", KSTAT_DATA_UINT64 },
938 { "write_buf_copied", KSTAT_DATA_UINT64 },
939 { "write_buf_nocopy", KSTAT_DATA_UINT64 }
940 };
941
942 #define XUIOSTAT_INCR(stat, val) \
943 atomic_add_64(&xuio_stats.stat.value.ui64, (val))
944 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
945
946 int
947 dmu_xuio_init(xuio_t *xuio, int nblk)
948 {
949 dmu_xuio_t *priv;
950 uio_t *uio = &xuio->xu_uio;
951
952 uio->uio_iovcnt = nblk;
953 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
954
955 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
956 priv->cnt = nblk;
957 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
958 priv->iovp = (iovec_t *)uio->uio_iov;
959 XUIO_XUZC_PRIV(xuio) = priv;
960
961 if (XUIO_XUZC_RW(xuio) == UIO_READ)
962 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
963 else
964 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
965
966 return (0);
967 }
968
969 void
970 dmu_xuio_fini(xuio_t *xuio)
971 {
972 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
973 int nblk = priv->cnt;
974
975 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
976 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
977 kmem_free(priv, sizeof (dmu_xuio_t));
978
979 if (XUIO_XUZC_RW(xuio) == UIO_READ)
980 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
981 else
982 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
983 }
984
985 /*
986 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
987 * and increase priv->next by 1.
988 */
989 int
990 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
991 {
992 struct iovec *iov;
993 uio_t *uio = &xuio->xu_uio;
994 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
995 int i = priv->next++;
996
997 ASSERT(i < priv->cnt);
998 ASSERT(off + n <= arc_buf_size(abuf));
999 iov = (iovec_t *)uio->uio_iov + i;
1000 iov->iov_base = (char *)abuf->b_data + off;
1001 iov->iov_len = n;
1002 priv->bufs[i] = abuf;
1003 return (0);
1004 }
1005
1006 int
1007 dmu_xuio_cnt(xuio_t *xuio)
1008 {
1009 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1010 return (priv->cnt);
1011 }
1012
1013 arc_buf_t *
1014 dmu_xuio_arcbuf(xuio_t *xuio, int i)
1015 {
1016 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1017
1018 ASSERT(i < priv->cnt);
1019 return (priv->bufs[i]);
1020 }
1021
1022 void
1023 dmu_xuio_clear(xuio_t *xuio, int i)
1024 {
1025 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1026
1027 ASSERT(i < priv->cnt);
1028 priv->bufs[i] = NULL;
1029 }
1030
1031 static void
1032 xuio_stat_init(void)
1033 {
1034 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
1035 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
1036 KSTAT_FLAG_VIRTUAL);
1037 if (xuio_ksp != NULL) {
1038 xuio_ksp->ks_data = &xuio_stats;
1039 kstat_install(xuio_ksp);
1040 }
1041 }
1042
1043 static void
1044 xuio_stat_fini(void)
1045 {
1046 if (xuio_ksp != NULL) {
1047 kstat_delete(xuio_ksp);
1048 xuio_ksp = NULL;
1049 }
1050 }
1051
1052 void
1053 xuio_stat_wbuf_copied()
1054 {
1055 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1056 }
1057
1058 void
1059 xuio_stat_wbuf_nocopy()
1060 {
1061 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1062 }
1063
1064 #ifdef _KERNEL
1065 static int
1066 dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
1067 {
1068 dmu_buf_t **dbp;
1069 int numbufs, i, err;
1070 xuio_t *xuio = NULL;
1071
1072 /*
1073 * NB: we could do this block-at-a-time, but it's nice
1074 * to be reading in parallel.
1075 */
1076 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1077 TRUE, FTAG, &numbufs, &dbp, 0);
1078 if (err)
1079 return (err);
1080
1081 for (i = 0; i < numbufs; i++) {
1082 uint64_t tocpy;
1083 int64_t bufoff;
1084 dmu_buf_t *db = dbp[i];
1085
1086 ASSERT(size > 0);
1087
1088 bufoff = uio->uio_loffset - db->db_offset;
1089 tocpy = MIN(db->db_size - bufoff, size);
1090
1091 if (xuio) {
1092 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1093 arc_buf_t *dbuf_abuf = dbi->db_buf;
1094 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1095 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1096 if (!err) {
1097 uio->uio_resid -= tocpy;
1098 uio->uio_loffset += tocpy;
1099 }
1100
1101 if (abuf == dbuf_abuf)
1102 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1103 else
1104 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1105 } else {
1106 err = uiomove((char *)db->db_data + bufoff, tocpy,
1107 UIO_READ, uio);
1108 }
1109 if (err)
1110 break;
1111
1112 size -= tocpy;
1113 }
1114 dmu_buf_rele_array(dbp, numbufs, FTAG);
1115
1116 return (err);
1117 }
1118
1119 /*
1120 * Read 'size' bytes into the uio buffer.
1121 * From object zdb->db_object.
1122 * Starting at offset uio->uio_loffset.
1123 *
1124 * If the caller already has a dbuf in the target object
1125 * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1126 * because we don't have to find the dnode_t for the object.
1127 */
1128 int
1129 dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size)
1130 {
1131 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1132 dnode_t *dn;
1133 int err;
1134
1135 if (size == 0)
1136 return (0);
1137
1138 DB_DNODE_ENTER(db);
1139 dn = DB_DNODE(db);
1140 err = dmu_read_uio_dnode(dn, uio, size);
1141 DB_DNODE_EXIT(db);
1142
1143 return (err);
1144 }
1145
1146 /*
1147 * Read 'size' bytes into the uio buffer.
1148 * From the specified object
1149 * Starting at offset uio->uio_loffset.
1150 */
1151 int
1152 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1153 {
1154 dnode_t *dn;
1155 int err;
1156
1157 if (size == 0)
1158 return (0);
1159
1160 err = dnode_hold(os, object, FTAG, &dn);
1161 if (err)
1162 return (err);
1163
1164 err = dmu_read_uio_dnode(dn, uio, size);
1165
1166 dnode_rele(dn, FTAG);
1167
1168 return (err);
1169 }
1170
1171 static int
1172 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1173 {
1174 dmu_buf_t **dbp;
1175 int numbufs;
1176 int err = 0;
1177 int i;
1178
1179 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1180 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1181 if (err)
1182 return (err);
1183
1184 for (i = 0; i < numbufs; i++) {
1185 uint64_t tocpy;
1186 int64_t bufoff;
1187 dmu_buf_t *db = dbp[i];
1188
1189 ASSERT(size > 0);
1190
1191 bufoff = uio->uio_loffset - db->db_offset;
1192 tocpy = MIN(db->db_size - bufoff, size);
1193
1194 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1195
1196 if (tocpy == db->db_size)
1197 dmu_buf_will_fill(db, tx);
1198 else
1199 dmu_buf_will_dirty(db, tx);
1200
1201 /*
1202 * XXX uiomove could block forever (eg.nfs-backed
1203 * pages). There needs to be a uiolockdown() function
1204 * to lock the pages in memory, so that uiomove won't
1205 * block.
1206 */
1207 err = uiomove((char *)db->db_data + bufoff, tocpy,
1208 UIO_WRITE, uio);
1209
1210 if (tocpy == db->db_size)
1211 dmu_buf_fill_done(db, tx);
1212
1213 if (err)
1214 break;
1215
1216 size -= tocpy;
1217 }
1218
1219 dmu_buf_rele_array(dbp, numbufs, FTAG);
1220 return (err);
1221 }
1222
1223 /*
1224 * Write 'size' bytes from the uio buffer.
1225 * To object zdb->db_object.
1226 * Starting at offset uio->uio_loffset.
1227 *
1228 * If the caller already has a dbuf in the target object
1229 * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1230 * because we don't have to find the dnode_t for the object.
1231 */
1232 int
1233 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1234 dmu_tx_t *tx)
1235 {
1236 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1237 dnode_t *dn;
1238 int err;
1239
1240 if (size == 0)
1241 return (0);
1242
1243 DB_DNODE_ENTER(db);
1244 dn = DB_DNODE(db);
1245 err = dmu_write_uio_dnode(dn, uio, size, tx);
1246 DB_DNODE_EXIT(db);
1247
1248 return (err);
1249 }
1250
1251 /*
1252 * Write 'size' bytes from the uio buffer.
1253 * To the specified object.
1254 * Starting at offset uio->uio_loffset.
1255 */
1256 int
1257 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1258 dmu_tx_t *tx)
1259 {
1260 dnode_t *dn;
1261 int err;
1262
1263 if (size == 0)
1264 return (0);
1265
1266 err = dnode_hold(os, object, FTAG, &dn);
1267 if (err)
1268 return (err);
1269
1270 err = dmu_write_uio_dnode(dn, uio, size, tx);
1271
1272 dnode_rele(dn, FTAG);
1273
1274 return (err);
1275 }
1276 #endif /* _KERNEL */
1277
1278 /*
1279 * Allocate a loaned anonymous arc buffer.
1280 */
1281 arc_buf_t *
1282 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1283 {
1284 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1285
1286 return (arc_loan_buf(db->db_objset->os_spa, size));
1287 }
1288
1289 /*
1290 * Free a loaned arc buffer.
1291 */
1292 void
1293 dmu_return_arcbuf(arc_buf_t *buf)
1294 {
1295 arc_return_buf(buf, FTAG);
1296 VERIFY(arc_buf_remove_ref(buf, FTAG));
1297 }
1298
1299 /*
1300 * When possible directly assign passed loaned arc buffer to a dbuf.
1301 * If this is not possible copy the contents of passed arc buf via
1302 * dmu_write().
1303 */
1304 void
1305 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1306 dmu_tx_t *tx)
1307 {
1308 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1309 dnode_t *dn;
1310 dmu_buf_impl_t *db;
1311 uint32_t blksz = (uint32_t)arc_buf_size(buf);
1312 uint64_t blkid;
1313
1314 DB_DNODE_ENTER(dbuf);
1315 dn = DB_DNODE(dbuf);
1316 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1317 blkid = dbuf_whichblock(dn, 0, offset);
1318 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1319 rw_exit(&dn->dn_struct_rwlock);
1320 DB_DNODE_EXIT(dbuf);
1321
1322 /*
1323 * We can only assign if the offset is aligned, the arc buf is the
1324 * same size as the dbuf, and the dbuf is not metadata. It
1325 * can't be metadata because the loaned arc buf comes from the
1326 * user-data kmem area.
1327 */
1328 if (offset == db->db.db_offset && blksz == db->db.db_size &&
1329 DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) {
1330 dbuf_assign_arcbuf(db, buf, tx);
1331 dbuf_rele(db, FTAG);
1332 } else {
1333 objset_t *os;
1334 uint64_t object;
1335
1336 DB_DNODE_ENTER(dbuf);
1337 dn = DB_DNODE(dbuf);
1338 os = dn->dn_objset;
1339 object = dn->dn_object;
1340 DB_DNODE_EXIT(dbuf);
1341
1342 dbuf_rele(db, FTAG);
1343 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1344 dmu_return_arcbuf(buf);
1345 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1346 }
1347 }
1348
1349 typedef struct {
1350 dbuf_dirty_record_t *dsa_dr;
1351 dmu_sync_cb_t *dsa_done;
1352 zgd_t *dsa_zgd;
1353 dmu_tx_t *dsa_tx;
1354 } dmu_sync_arg_t;
1355
1356 /* ARGSUSED */
1357 static void
1358 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1359 {
1360 dmu_sync_arg_t *dsa = varg;
1361 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1362 blkptr_t *bp = zio->io_bp;
1363
1364 if (zio->io_error == 0) {
1365 if (BP_IS_HOLE(bp)) {
1366 /*
1367 * A block of zeros may compress to a hole, but the
1368 * block size still needs to be known for replay.
1369 */
1370 BP_SET_LSIZE(bp, db->db_size);
1371 } else if (!BP_IS_EMBEDDED(bp)) {
1372 ASSERT(BP_GET_LEVEL(bp) == 0);
1373 bp->blk_fill = 1;
1374 }
1375 }
1376 }
1377
1378 static void
1379 dmu_sync_late_arrival_ready(zio_t *zio)
1380 {
1381 dmu_sync_ready(zio, NULL, zio->io_private);
1382 }
1383
1384 /* ARGSUSED */
1385 static void
1386 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1387 {
1388 dmu_sync_arg_t *dsa = varg;
1389 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1390 dmu_buf_impl_t *db = dr->dr_dbuf;
1391
1392 mutex_enter(&db->db_mtx);
1393 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1394 if (zio->io_error == 0) {
1395 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1396 if (dr->dt.dl.dr_nopwrite) {
1397 ASSERTV(blkptr_t *bp = zio->io_bp);
1398 ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
1399 ASSERTV(uint8_t chksum = BP_GET_CHECKSUM(bp_orig));
1400
1401 ASSERT(BP_EQUAL(bp, bp_orig));
1402 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1403 ASSERT(zio_checksum_table[chksum].ci_dedup);
1404 }
1405 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1406 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1407 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1408
1409 /*
1410 * Old style holes are filled with all zeros, whereas
1411 * new-style holes maintain their lsize, type, level,
1412 * and birth time (see zio_write_compress). While we
1413 * need to reset the BP_SET_LSIZE() call that happened
1414 * in dmu_sync_ready for old style holes, we do *not*
1415 * want to wipe out the information contained in new
1416 * style holes. Thus, only zero out the block pointer if
1417 * it's an old style hole.
1418 */
1419 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) &&
1420 dr->dt.dl.dr_overridden_by.blk_birth == 0)
1421 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1422 } else {
1423 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1424 }
1425 cv_broadcast(&db->db_changed);
1426 mutex_exit(&db->db_mtx);
1427
1428 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1429
1430 kmem_free(dsa, sizeof (*dsa));
1431 }
1432
1433 static void
1434 dmu_sync_late_arrival_done(zio_t *zio)
1435 {
1436 blkptr_t *bp = zio->io_bp;
1437 dmu_sync_arg_t *dsa = zio->io_private;
1438 ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
1439
1440 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1441 /*
1442 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1443 * then there is nothing to do here. Otherwise, free the
1444 * newly allocated block in this txg.
1445 */
1446 if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
1447 ASSERT(BP_EQUAL(bp, bp_orig));
1448 } else {
1449 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1450 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1451 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1452 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1453 }
1454 }
1455
1456 dmu_tx_commit(dsa->dsa_tx);
1457
1458 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1459
1460 kmem_free(dsa, sizeof (*dsa));
1461 }
1462
1463 static int
1464 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1465 zio_prop_t *zp, zbookmark_phys_t *zb)
1466 {
1467 dmu_sync_arg_t *dsa;
1468 dmu_tx_t *tx;
1469
1470 tx = dmu_tx_create(os);
1471 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1472 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1473 dmu_tx_abort(tx);
1474 /* Make zl_get_data do txg_waited_synced() */
1475 return (SET_ERROR(EIO));
1476 }
1477
1478 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1479 dsa->dsa_dr = NULL;
1480 dsa->dsa_done = done;
1481 dsa->dsa_zgd = zgd;
1482 dsa->dsa_tx = tx;
1483
1484 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx),
1485 zgd->zgd_bp, zgd->zgd_db->db_data, zgd->zgd_db->db_size,
1486 zp, dmu_sync_late_arrival_ready, NULL,
1487 NULL, dmu_sync_late_arrival_done, dsa, ZIO_PRIORITY_SYNC_WRITE,
1488 ZIO_FLAG_CANFAIL, zb));
1489
1490 return (0);
1491 }
1492
1493 /*
1494 * Intent log support: sync the block associated with db to disk.
1495 * N.B. and XXX: the caller is responsible for making sure that the
1496 * data isn't changing while dmu_sync() is writing it.
1497 *
1498 * Return values:
1499 *
1500 * EEXIST: this txg has already been synced, so there's nothing to do.
1501 * The caller should not log the write.
1502 *
1503 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1504 * The caller should not log the write.
1505 *
1506 * EALREADY: this block is already in the process of being synced.
1507 * The caller should track its progress (somehow).
1508 *
1509 * EIO: could not do the I/O.
1510 * The caller should do a txg_wait_synced().
1511 *
1512 * 0: the I/O has been initiated.
1513 * The caller should log this blkptr in the done callback.
1514 * It is possible that the I/O will fail, in which case
1515 * the error will be reported to the done callback and
1516 * propagated to pio from zio_done().
1517 */
1518 int
1519 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1520 {
1521 blkptr_t *bp = zgd->zgd_bp;
1522 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1523 objset_t *os = db->db_objset;
1524 dsl_dataset_t *ds = os->os_dsl_dataset;
1525 dbuf_dirty_record_t *dr;
1526 dmu_sync_arg_t *dsa;
1527 zbookmark_phys_t zb;
1528 zio_prop_t zp;
1529 dnode_t *dn;
1530
1531 ASSERT(pio != NULL);
1532 ASSERT(txg != 0);
1533
1534 SET_BOOKMARK(&zb, ds->ds_object,
1535 db->db.db_object, db->db_level, db->db_blkid);
1536
1537 DB_DNODE_ENTER(db);
1538 dn = DB_DNODE(db);
1539 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1540 DB_DNODE_EXIT(db);
1541
1542 /*
1543 * If we're frozen (running ziltest), we always need to generate a bp.
1544 */
1545 if (txg > spa_freeze_txg(os->os_spa))
1546 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1547
1548 /*
1549 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1550 * and us. If we determine that this txg is not yet syncing,
1551 * but it begins to sync a moment later, that's OK because the
1552 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1553 */
1554 mutex_enter(&db->db_mtx);
1555
1556 if (txg <= spa_last_synced_txg(os->os_spa)) {
1557 /*
1558 * This txg has already synced. There's nothing to do.
1559 */
1560 mutex_exit(&db->db_mtx);
1561 return (SET_ERROR(EEXIST));
1562 }
1563
1564 if (txg <= spa_syncing_txg(os->os_spa)) {
1565 /*
1566 * This txg is currently syncing, so we can't mess with
1567 * the dirty record anymore; just write a new log block.
1568 */
1569 mutex_exit(&db->db_mtx);
1570 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1571 }
1572
1573 dr = db->db_last_dirty;
1574 while (dr && dr->dr_txg != txg)
1575 dr = dr->dr_next;
1576
1577 if (dr == NULL) {
1578 /*
1579 * There's no dr for this dbuf, so it must have been freed.
1580 * There's no need to log writes to freed blocks, so we're done.
1581 */
1582 mutex_exit(&db->db_mtx);
1583 return (SET_ERROR(ENOENT));
1584 }
1585
1586 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
1587
1588 /*
1589 * Assume the on-disk data is X, the current syncing data (in
1590 * txg - 1) is Y, and the current in-memory data is Z (currently
1591 * in dmu_sync).
1592 *
1593 * We usually want to perform a nopwrite if X and Z are the
1594 * same. However, if Y is different (i.e. the BP is going to
1595 * change before this write takes effect), then a nopwrite will
1596 * be incorrect - we would override with X, which could have
1597 * been freed when Y was written.
1598 *
1599 * (Note that this is not a concern when we are nop-writing from
1600 * syncing context, because X and Y must be identical, because
1601 * all previous txgs have been synced.)
1602 *
1603 * Therefore, we disable nopwrite if the current BP could change
1604 * before this TXG. There are two ways it could change: by
1605 * being dirty (dr_next is non-NULL), or by being freed
1606 * (dnode_block_freed()). This behavior is verified by
1607 * zio_done(), which VERIFYs that the override BP is identical
1608 * to the on-disk BP.
1609 */
1610 DB_DNODE_ENTER(db);
1611 dn = DB_DNODE(db);
1612 if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid))
1613 zp.zp_nopwrite = B_FALSE;
1614 DB_DNODE_EXIT(db);
1615
1616 ASSERT(dr->dr_txg == txg);
1617 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1618 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1619 /*
1620 * We have already issued a sync write for this buffer,
1621 * or this buffer has already been synced. It could not
1622 * have been dirtied since, or we would have cleared the state.
1623 */
1624 mutex_exit(&db->db_mtx);
1625 return (SET_ERROR(EALREADY));
1626 }
1627
1628 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1629 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1630 mutex_exit(&db->db_mtx);
1631
1632 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1633 dsa->dsa_dr = dr;
1634 dsa->dsa_done = done;
1635 dsa->dsa_zgd = zgd;
1636 dsa->dsa_tx = NULL;
1637
1638 zio_nowait(arc_write(pio, os->os_spa, txg,
1639 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
1640 DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready,
1641 NULL, NULL, dmu_sync_done, dsa,
1642 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
1643
1644 return (0);
1645 }
1646
1647 int
1648 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1649 dmu_tx_t *tx)
1650 {
1651 dnode_t *dn;
1652 int err;
1653
1654 err = dnode_hold(os, object, FTAG, &dn);
1655 if (err)
1656 return (err);
1657 err = dnode_set_blksz(dn, size, ibs, tx);
1658 dnode_rele(dn, FTAG);
1659 return (err);
1660 }
1661
1662 void
1663 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1664 dmu_tx_t *tx)
1665 {
1666 dnode_t *dn;
1667
1668 /*
1669 * Send streams include each object's checksum function. This
1670 * check ensures that the receiving system can understand the
1671 * checksum function transmitted.
1672 */
1673 ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
1674
1675 VERIFY0(dnode_hold(os, object, FTAG, &dn));
1676 ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
1677 dn->dn_checksum = checksum;
1678 dnode_setdirty(dn, tx);
1679 dnode_rele(dn, FTAG);
1680 }
1681
1682 void
1683 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1684 dmu_tx_t *tx)
1685 {
1686 dnode_t *dn;
1687
1688 /*
1689 * Send streams include each object's compression function. This
1690 * check ensures that the receiving system can understand the
1691 * compression function transmitted.
1692 */
1693 ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
1694
1695 VERIFY0(dnode_hold(os, object, FTAG, &dn));
1696 dn->dn_compress = compress;
1697 dnode_setdirty(dn, tx);
1698 dnode_rele(dn, FTAG);
1699 }
1700
1701 int zfs_mdcomp_disable = 0;
1702
1703 /*
1704 * When the "redundant_metadata" property is set to "most", only indirect
1705 * blocks of this level and higher will have an additional ditto block.
1706 */
1707 int zfs_redundant_metadata_most_ditto_level = 2;
1708
1709 void
1710 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1711 {
1712 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1713 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1714 (wp & WP_SPILL));
1715 enum zio_checksum checksum = os->os_checksum;
1716 enum zio_compress compress = os->os_compress;
1717 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1718 boolean_t dedup = B_FALSE;
1719 boolean_t nopwrite = B_FALSE;
1720 boolean_t dedup_verify = os->os_dedup_verify;
1721 int copies = os->os_copies;
1722
1723 /*
1724 * We maintain different write policies for each of the following
1725 * types of data:
1726 * 1. metadata
1727 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
1728 * 3. all other level 0 blocks
1729 */
1730 if (ismd) {
1731 if (zfs_mdcomp_disable) {
1732 compress = ZIO_COMPRESS_EMPTY;
1733 } else {
1734 /*
1735 * XXX -- we should design a compression algorithm
1736 * that specializes in arrays of bps.
1737 */
1738 compress = zio_compress_select(os->os_spa,
1739 ZIO_COMPRESS_ON, ZIO_COMPRESS_ON);
1740 }
1741
1742 /*
1743 * Metadata always gets checksummed. If the data
1744 * checksum is multi-bit correctable, and it's not a
1745 * ZBT-style checksum, then it's suitable for metadata
1746 * as well. Otherwise, the metadata checksum defaults
1747 * to fletcher4.
1748 */
1749 if (zio_checksum_table[checksum].ci_correctable < 1 ||
1750 zio_checksum_table[checksum].ci_eck)
1751 checksum = ZIO_CHECKSUM_FLETCHER_4;
1752
1753 if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
1754 (os->os_redundant_metadata ==
1755 ZFS_REDUNDANT_METADATA_MOST &&
1756 (level >= zfs_redundant_metadata_most_ditto_level ||
1757 DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))))
1758 copies++;
1759 } else if (wp & WP_NOFILL) {
1760 ASSERT(level == 0);
1761
1762 /*
1763 * If we're writing preallocated blocks, we aren't actually
1764 * writing them so don't set any policy properties. These
1765 * blocks are currently only used by an external subsystem
1766 * outside of zfs (i.e. dump) and not written by the zio
1767 * pipeline.
1768 */
1769 compress = ZIO_COMPRESS_OFF;
1770 checksum = ZIO_CHECKSUM_OFF;
1771 } else {
1772 compress = zio_compress_select(os->os_spa, dn->dn_compress,
1773 compress);
1774
1775 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
1776 zio_checksum_select(dn->dn_checksum, checksum) :
1777 dedup_checksum;
1778
1779 /*
1780 * Determine dedup setting. If we are in dmu_sync(),
1781 * we won't actually dedup now because that's all
1782 * done in syncing context; but we do want to use the
1783 * dedup checkum. If the checksum is not strong
1784 * enough to ensure unique signatures, force
1785 * dedup_verify.
1786 */
1787 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
1788 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
1789 if (!zio_checksum_table[checksum].ci_dedup)
1790 dedup_verify = B_TRUE;
1791 }
1792
1793 /*
1794 * Enable nopwrite if we have a cryptographically secure
1795 * checksum that has no known collisions (i.e. SHA-256)
1796 * and compression is enabled. We don't enable nopwrite if
1797 * dedup is enabled as the two features are mutually exclusive.
1798 */
1799 nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup &&
1800 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
1801 }
1802
1803 zp->zp_checksum = checksum;
1804 zp->zp_compress = compress;
1805 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
1806 zp->zp_level = level;
1807 zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
1808 zp->zp_dedup = dedup;
1809 zp->zp_dedup_verify = dedup && dedup_verify;
1810 zp->zp_nopwrite = nopwrite;
1811 }
1812
1813 int
1814 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1815 {
1816 dnode_t *dn;
1817 int i, err;
1818
1819 err = dnode_hold(os, object, FTAG, &dn);
1820 if (err)
1821 return (err);
1822 /*
1823 * Sync any current changes before
1824 * we go trundling through the block pointers.
1825 */
1826 for (i = 0; i < TXG_SIZE; i++) {
1827 if (list_link_active(&dn->dn_dirty_link[i]))
1828 break;
1829 }
1830 if (i != TXG_SIZE) {
1831 dnode_rele(dn, FTAG);
1832 txg_wait_synced(dmu_objset_pool(os), 0);
1833 err = dnode_hold(os, object, FTAG, &dn);
1834 if (err)
1835 return (err);
1836 }
1837
1838 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1839 dnode_rele(dn, FTAG);
1840
1841 return (err);
1842 }
1843
1844 void
1845 __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1846 {
1847 dnode_phys_t *dnp = dn->dn_phys;
1848 int i;
1849
1850 doi->doi_data_block_size = dn->dn_datablksz;
1851 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1852 1ULL << dn->dn_indblkshift : 0;
1853 doi->doi_type = dn->dn_type;
1854 doi->doi_bonus_type = dn->dn_bonustype;
1855 doi->doi_bonus_size = dn->dn_bonuslen;
1856 doi->doi_indirection = dn->dn_nlevels;
1857 doi->doi_checksum = dn->dn_checksum;
1858 doi->doi_compress = dn->dn_compress;
1859 doi->doi_nblkptr = dn->dn_nblkptr;
1860 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1861 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
1862 doi->doi_fill_count = 0;
1863 for (i = 0; i < dnp->dn_nblkptr; i++)
1864 doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
1865 }
1866
1867 void
1868 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1869 {
1870 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1871 mutex_enter(&dn->dn_mtx);
1872
1873 __dmu_object_info_from_dnode(dn, doi);
1874
1875 mutex_exit(&dn->dn_mtx);
1876 rw_exit(&dn->dn_struct_rwlock);
1877 }
1878
1879 /*
1880 * Get information on a DMU object.
1881 * If doi is NULL, just indicates whether the object exists.
1882 */
1883 int
1884 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1885 {
1886 dnode_t *dn;
1887 int err = dnode_hold(os, object, FTAG, &dn);
1888
1889 if (err)
1890 return (err);
1891
1892 if (doi != NULL)
1893 dmu_object_info_from_dnode(dn, doi);
1894
1895 dnode_rele(dn, FTAG);
1896 return (0);
1897 }
1898
1899 /*
1900 * As above, but faster; can be used when you have a held dbuf in hand.
1901 */
1902 void
1903 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1904 {
1905 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1906
1907 DB_DNODE_ENTER(db);
1908 dmu_object_info_from_dnode(DB_DNODE(db), doi);
1909 DB_DNODE_EXIT(db);
1910 }
1911
1912 /*
1913 * Faster still when you only care about the size.
1914 * This is specifically optimized for zfs_getattr().
1915 */
1916 void
1917 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1918 u_longlong_t *nblk512)
1919 {
1920 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1921 dnode_t *dn;
1922
1923 DB_DNODE_ENTER(db);
1924 dn = DB_DNODE(db);
1925
1926 *blksize = dn->dn_datablksz;
1927 /* add 1 for dnode space */
1928 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1929 SPA_MINBLOCKSHIFT) + 1;
1930 DB_DNODE_EXIT(db);
1931 }
1932
1933 void
1934 byteswap_uint64_array(void *vbuf, size_t size)
1935 {
1936 uint64_t *buf = vbuf;
1937 size_t count = size >> 3;
1938 int i;
1939
1940 ASSERT((size & 7) == 0);
1941
1942 for (i = 0; i < count; i++)
1943 buf[i] = BSWAP_64(buf[i]);
1944 }
1945
1946 void
1947 byteswap_uint32_array(void *vbuf, size_t size)
1948 {
1949 uint32_t *buf = vbuf;
1950 size_t count = size >> 2;
1951 int i;
1952
1953 ASSERT((size & 3) == 0);
1954
1955 for (i = 0; i < count; i++)
1956 buf[i] = BSWAP_32(buf[i]);
1957 }
1958
1959 void
1960 byteswap_uint16_array(void *vbuf, size_t size)
1961 {
1962 uint16_t *buf = vbuf;
1963 size_t count = size >> 1;
1964 int i;
1965
1966 ASSERT((size & 1) == 0);
1967
1968 for (i = 0; i < count; i++)
1969 buf[i] = BSWAP_16(buf[i]);
1970 }
1971
1972 /* ARGSUSED */
1973 void
1974 byteswap_uint8_array(void *vbuf, size_t size)
1975 {
1976 }
1977
1978 void
1979 dmu_init(void)
1980 {
1981 zfs_dbgmsg_init();
1982 sa_cache_init();
1983 xuio_stat_init();
1984 dmu_objset_init();
1985 dnode_init();
1986 dbuf_init();
1987 zfetch_init();
1988 dmu_tx_init();
1989 l2arc_init();
1990 arc_init();
1991 }
1992
1993 void
1994 dmu_fini(void)
1995 {
1996 arc_fini(); /* arc depends on l2arc, so arc must go first */
1997 l2arc_fini();
1998 dmu_tx_fini();
1999 zfetch_fini();
2000 dbuf_fini();
2001 dnode_fini();
2002 dmu_objset_fini();
2003 xuio_stat_fini();
2004 sa_cache_fini();
2005 zfs_dbgmsg_fini();
2006 }
2007
2008 #if defined(_KERNEL) && defined(HAVE_SPL)
2009 EXPORT_SYMBOL(dmu_bonus_hold);
2010 EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus);
2011 EXPORT_SYMBOL(dmu_buf_rele_array);
2012 EXPORT_SYMBOL(dmu_prefetch);
2013 EXPORT_SYMBOL(dmu_free_range);
2014 EXPORT_SYMBOL(dmu_free_long_range);
2015 EXPORT_SYMBOL(dmu_free_long_object);
2016 EXPORT_SYMBOL(dmu_read);
2017 EXPORT_SYMBOL(dmu_write);
2018 EXPORT_SYMBOL(dmu_prealloc);
2019 EXPORT_SYMBOL(dmu_object_info);
2020 EXPORT_SYMBOL(dmu_object_info_from_dnode);
2021 EXPORT_SYMBOL(dmu_object_info_from_db);
2022 EXPORT_SYMBOL(dmu_object_size_from_db);
2023 EXPORT_SYMBOL(dmu_object_set_blocksize);
2024 EXPORT_SYMBOL(dmu_object_set_checksum);
2025 EXPORT_SYMBOL(dmu_object_set_compress);
2026 EXPORT_SYMBOL(dmu_write_policy);
2027 EXPORT_SYMBOL(dmu_sync);
2028 EXPORT_SYMBOL(dmu_request_arcbuf);
2029 EXPORT_SYMBOL(dmu_return_arcbuf);
2030 EXPORT_SYMBOL(dmu_assign_arcbuf);
2031 EXPORT_SYMBOL(dmu_buf_hold);
2032 EXPORT_SYMBOL(dmu_ot);
2033
2034 module_param(zfs_mdcomp_disable, int, 0644);
2035 MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression");
2036
2037 module_param(zfs_nopwrite_enabled, int, 0644);
2038 MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes");
2039
2040 #endif
ZFS On Linux mirror for PVE