]> 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
cstyle: Resolve C style issues
[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) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 */
26
27 #include <sys/dmu.h>
28 #include <sys/dmu_impl.h>
29 #include <sys/dmu_tx.h>
30 #include <sys/dbuf.h>
31 #include <sys/dnode.h>
32 #include <sys/zfs_context.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/dsl_pool.h>
38 #include <sys/dsl_synctask.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dmu_zfetch.h>
41 #include <sys/zfs_ioctl.h>
42 #include <sys/zap.h>
43 #include <sys/zio_checksum.h>
44 #include <sys/zio_compress.h>
45 #include <sys/sa.h>
46 #ifdef _KERNEL
47 #include <sys/vmsystm.h>
48 #include <sys/zfs_znode.h>
49 #endif
50
51 /*
52 * Enable/disable nopwrite feature.
53 */
54 int zfs_nopwrite_enabled = 1;
55
56 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
57 { DMU_BSWAP_UINT8, TRUE, "unallocated" },
58 { DMU_BSWAP_ZAP, TRUE, "object directory" },
59 { DMU_BSWAP_UINT64, TRUE, "object array" },
60 { DMU_BSWAP_UINT8, TRUE, "packed nvlist" },
61 { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" },
62 { DMU_BSWAP_UINT64, TRUE, "bpobj" },
63 { DMU_BSWAP_UINT64, TRUE, "bpobj header" },
64 { DMU_BSWAP_UINT64, TRUE, "SPA space map header" },
65 { DMU_BSWAP_UINT64, TRUE, "SPA space map" },
66 { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" },
67 { DMU_BSWAP_DNODE, TRUE, "DMU dnode" },
68 { DMU_BSWAP_OBJSET, TRUE, "DMU objset" },
69 { DMU_BSWAP_UINT64, TRUE, "DSL directory" },
70 { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"},
71 { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" },
72 { DMU_BSWAP_ZAP, TRUE, "DSL props" },
73 { DMU_BSWAP_UINT64, TRUE, "DSL dataset" },
74 { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" },
75 { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" },
76 { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" },
77 { DMU_BSWAP_ZAP, TRUE, "ZFS directory" },
78 { DMU_BSWAP_ZAP, TRUE, "ZFS master node" },
79 { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" },
80 { DMU_BSWAP_UINT8, FALSE, "zvol object" },
81 { DMU_BSWAP_ZAP, TRUE, "zvol prop" },
82 { DMU_BSWAP_UINT8, FALSE, "other uint8[]" },
83 { DMU_BSWAP_UINT64, FALSE, "other uint64[]" },
84 { DMU_BSWAP_ZAP, TRUE, "other ZAP" },
85 { DMU_BSWAP_ZAP, TRUE, "persistent error log" },
86 { DMU_BSWAP_UINT8, TRUE, "SPA history" },
87 { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" },
88 { DMU_BSWAP_ZAP, TRUE, "Pool properties" },
89 { DMU_BSWAP_ZAP, TRUE, "DSL permissions" },
90 { DMU_BSWAP_ACL, TRUE, "ZFS ACL" },
91 { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" },
92 { DMU_BSWAP_UINT8, TRUE, "FUID table" },
93 { DMU_BSWAP_UINT64, TRUE, "FUID table size" },
94 { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"},
95 { DMU_BSWAP_ZAP, TRUE, "scan work queue" },
96 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" },
97 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" },
98 { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"},
99 { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" },
100 { DMU_BSWAP_ZAP, TRUE, "DDT statistics" },
101 { DMU_BSWAP_UINT8, TRUE, "System attributes" },
102 { DMU_BSWAP_ZAP, TRUE, "SA master node" },
103 { DMU_BSWAP_ZAP, TRUE, "SA attr registration" },
104 { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" },
105 { DMU_BSWAP_ZAP, TRUE, "scan translations" },
106 { DMU_BSWAP_UINT8, FALSE, "deduplicated block" },
107 { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" },
108 { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" },
109 { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" },
110 { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" }
111 };
112
113 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
114 { byteswap_uint8_array, "uint8" },
115 { byteswap_uint16_array, "uint16" },
116 { byteswap_uint32_array, "uint32" },
117 { byteswap_uint64_array, "uint64" },
118 { zap_byteswap, "zap" },
119 { dnode_buf_byteswap, "dnode" },
120 { dmu_objset_byteswap, "objset" },
121 { zfs_znode_byteswap, "znode" },
122 { zfs_oldacl_byteswap, "oldacl" },
123 { zfs_acl_byteswap, "acl" }
124 };
125
126 int
127 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
128 void *tag, dmu_buf_t **dbp, int flags)
129 {
130 dnode_t *dn;
131 uint64_t blkid;
132 dmu_buf_impl_t *db;
133 int err;
134 int db_flags = DB_RF_CANFAIL;
135
136 if (flags & DMU_READ_NO_PREFETCH)
137 db_flags |= DB_RF_NOPREFETCH;
138
139 err = dnode_hold(os, object, FTAG, &dn);
140 if (err)
141 return (err);
142 blkid = dbuf_whichblock(dn, offset);
143 rw_enter(&dn->dn_struct_rwlock, RW_READER);
144 db = dbuf_hold(dn, blkid, tag);
145 rw_exit(&dn->dn_struct_rwlock);
146 if (db == NULL) {
147 err = SET_ERROR(EIO);
148 } else {
149 err = dbuf_read(db, NULL, db_flags);
150 if (err) {
151 dbuf_rele(db, tag);
152 db = NULL;
153 }
154 }
155
156 dnode_rele(dn, FTAG);
157 *dbp = &db->db; /* NULL db plus first field offset is NULL */
158 return (err);
159 }
160
161 int
162 dmu_bonus_max(void)
163 {
164 return (DN_MAX_BONUSLEN);
165 }
166
167 int
168 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
169 {
170 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
171 dnode_t *dn;
172 int error;
173
174 DB_DNODE_ENTER(db);
175 dn = DB_DNODE(db);
176
177 if (dn->dn_bonus != db) {
178 error = SET_ERROR(EINVAL);
179 } else if (newsize < 0 || newsize > db_fake->db_size) {
180 error = SET_ERROR(EINVAL);
181 } else {
182 dnode_setbonuslen(dn, newsize, tx);
183 error = 0;
184 }
185
186 DB_DNODE_EXIT(db);
187 return (error);
188 }
189
190 int
191 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
192 {
193 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
194 dnode_t *dn;
195 int error;
196
197 DB_DNODE_ENTER(db);
198 dn = DB_DNODE(db);
199
200 if (!DMU_OT_IS_VALID(type)) {
201 error = SET_ERROR(EINVAL);
202 } else if (dn->dn_bonus != db) {
203 error = SET_ERROR(EINVAL);
204 } else {
205 dnode_setbonus_type(dn, type, tx);
206 error = 0;
207 }
208
209 DB_DNODE_EXIT(db);
210 return (error);
211 }
212
213 dmu_object_type_t
214 dmu_get_bonustype(dmu_buf_t *db_fake)
215 {
216 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
217 dnode_t *dn;
218 dmu_object_type_t type;
219
220 DB_DNODE_ENTER(db);
221 dn = DB_DNODE(db);
222 type = dn->dn_bonustype;
223 DB_DNODE_EXIT(db);
224
225 return (type);
226 }
227
228 int
229 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
230 {
231 dnode_t *dn;
232 int error;
233
234 error = dnode_hold(os, object, FTAG, &dn);
235 dbuf_rm_spill(dn, tx);
236 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
237 dnode_rm_spill(dn, tx);
238 rw_exit(&dn->dn_struct_rwlock);
239 dnode_rele(dn, FTAG);
240 return (error);
241 }
242
243 /*
244 * returns ENOENT, EIO, or 0.
245 */
246 int
247 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
248 {
249 dnode_t *dn;
250 dmu_buf_impl_t *db;
251 int error;
252
253 error = dnode_hold(os, object, FTAG, &dn);
254 if (error)
255 return (error);
256
257 rw_enter(&dn->dn_struct_rwlock, RW_READER);
258 if (dn->dn_bonus == NULL) {
259 rw_exit(&dn->dn_struct_rwlock);
260 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
261 if (dn->dn_bonus == NULL)
262 dbuf_create_bonus(dn);
263 }
264 db = dn->dn_bonus;
265
266 /* as long as the bonus buf is held, the dnode will be held */
267 if (refcount_add(&db->db_holds, tag) == 1) {
268 VERIFY(dnode_add_ref(dn, db));
269 (void) atomic_inc_32_nv(&dn->dn_dbufs_count);
270 }
271
272 /*
273 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
274 * hold and incrementing the dbuf count to ensure that dnode_move() sees
275 * a dnode hold for every dbuf.
276 */
277 rw_exit(&dn->dn_struct_rwlock);
278
279 dnode_rele(dn, FTAG);
280
281 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
282
283 *dbp = &db->db;
284 return (0);
285 }
286
287 /*
288 * returns ENOENT, EIO, or 0.
289 *
290 * This interface will allocate a blank spill dbuf when a spill blk
291 * doesn't already exist on the dnode.
292 *
293 * if you only want to find an already existing spill db, then
294 * dmu_spill_hold_existing() should be used.
295 */
296 int
297 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
298 {
299 dmu_buf_impl_t *db = NULL;
300 int err;
301
302 if ((flags & DB_RF_HAVESTRUCT) == 0)
303 rw_enter(&dn->dn_struct_rwlock, RW_READER);
304
305 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
306
307 if ((flags & DB_RF_HAVESTRUCT) == 0)
308 rw_exit(&dn->dn_struct_rwlock);
309
310 ASSERT(db != NULL);
311 err = dbuf_read(db, NULL, flags);
312 if (err == 0)
313 *dbp = &db->db;
314 else
315 dbuf_rele(db, tag);
316 return (err);
317 }
318
319 int
320 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
321 {
322 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
323 dnode_t *dn;
324 int err;
325
326 DB_DNODE_ENTER(db);
327 dn = DB_DNODE(db);
328
329 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
330 err = SET_ERROR(EINVAL);
331 } else {
332 rw_enter(&dn->dn_struct_rwlock, RW_READER);
333
334 if (!dn->dn_have_spill) {
335 err = SET_ERROR(ENOENT);
336 } else {
337 err = dmu_spill_hold_by_dnode(dn,
338 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
339 }
340
341 rw_exit(&dn->dn_struct_rwlock);
342 }
343
344 DB_DNODE_EXIT(db);
345 return (err);
346 }
347
348 int
349 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
350 {
351 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
352 dnode_t *dn;
353 int err;
354
355 DB_DNODE_ENTER(db);
356 dn = DB_DNODE(db);
357 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
358 DB_DNODE_EXIT(db);
359
360 return (err);
361 }
362
363 /*
364 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
365 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
366 * and can induce severe lock contention when writing to several files
367 * whose dnodes are in the same block.
368 */
369 static int
370 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
371 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
372 {
373 dmu_buf_t **dbp;
374 uint64_t blkid, nblks, i;
375 uint32_t dbuf_flags;
376 int err;
377 zio_t *zio;
378
379 ASSERT(length <= DMU_MAX_ACCESS);
380
381 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
382 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
383 dbuf_flags |= DB_RF_NOPREFETCH;
384
385 rw_enter(&dn->dn_struct_rwlock, RW_READER);
386 if (dn->dn_datablkshift) {
387 int blkshift = dn->dn_datablkshift;
388 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
389 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
390 } else {
391 if (offset + length > dn->dn_datablksz) {
392 zfs_panic_recover("zfs: accessing past end of object "
393 "%llx/%llx (size=%u access=%llu+%llu)",
394 (longlong_t)dn->dn_objset->
395 os_dsl_dataset->ds_object,
396 (longlong_t)dn->dn_object, dn->dn_datablksz,
397 (longlong_t)offset, (longlong_t)length);
398 rw_exit(&dn->dn_struct_rwlock);
399 return (SET_ERROR(EIO));
400 }
401 nblks = 1;
402 }
403 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks,
404 KM_PUSHPAGE | KM_NODEBUG);
405
406 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
407 blkid = dbuf_whichblock(dn, offset);
408 for (i = 0; i < nblks; i++) {
409 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
410 if (db == NULL) {
411 rw_exit(&dn->dn_struct_rwlock);
412 dmu_buf_rele_array(dbp, nblks, tag);
413 zio_nowait(zio);
414 return (SET_ERROR(EIO));
415 }
416 /* initiate async i/o */
417 if (read) {
418 (void) dbuf_read(db, zio, dbuf_flags);
419 }
420 dbp[i] = &db->db;
421 }
422 rw_exit(&dn->dn_struct_rwlock);
423
424 /* wait for async i/o */
425 err = zio_wait(zio);
426 if (err) {
427 dmu_buf_rele_array(dbp, nblks, tag);
428 return (err);
429 }
430
431 /* wait for other io to complete */
432 if (read) {
433 for (i = 0; i < nblks; i++) {
434 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
435 mutex_enter(&db->db_mtx);
436 while (db->db_state == DB_READ ||
437 db->db_state == DB_FILL)
438 cv_wait(&db->db_changed, &db->db_mtx);
439 if (db->db_state == DB_UNCACHED)
440 err = SET_ERROR(EIO);
441 mutex_exit(&db->db_mtx);
442 if (err) {
443 dmu_buf_rele_array(dbp, nblks, tag);
444 return (err);
445 }
446 }
447 }
448
449 *numbufsp = nblks;
450 *dbpp = dbp;
451 return (0);
452 }
453
454 static int
455 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
456 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
457 {
458 dnode_t *dn;
459 int err;
460
461 err = dnode_hold(os, object, FTAG, &dn);
462 if (err)
463 return (err);
464
465 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
466 numbufsp, dbpp, DMU_READ_PREFETCH);
467
468 dnode_rele(dn, FTAG);
469
470 return (err);
471 }
472
473 int
474 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
475 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
476 {
477 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
478 dnode_t *dn;
479 int err;
480
481 DB_DNODE_ENTER(db);
482 dn = DB_DNODE(db);
483 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
484 numbufsp, dbpp, DMU_READ_PREFETCH);
485 DB_DNODE_EXIT(db);
486
487 return (err);
488 }
489
490 void
491 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
492 {
493 int i;
494 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
495
496 if (numbufs == 0)
497 return;
498
499 for (i = 0; i < numbufs; i++) {
500 if (dbp[i])
501 dbuf_rele(dbp[i], tag);
502 }
503
504 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
505 }
506
507 /*
508 * Issue prefetch i/os for the given blocks.
509 *
510 * Note: The assumption is that we *know* these blocks will be needed
511 * almost immediately. Therefore, the prefetch i/os will be issued at
512 * ZIO_PRIORITY_SYNC_READ
513 *
514 * Note: indirect blocks and other metadata will be read synchronously,
515 * causing this function to block if they are not already cached.
516 */
517 void
518 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
519 {
520 dnode_t *dn;
521 uint64_t blkid;
522 int nblks, err;
523
524 if (zfs_prefetch_disable)
525 return;
526
527 if (len == 0) { /* they're interested in the bonus buffer */
528 dn = DMU_META_DNODE(os);
529
530 if (object == 0 || object >= DN_MAX_OBJECT)
531 return;
532
533 rw_enter(&dn->dn_struct_rwlock, RW_READER);
534 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
535 dbuf_prefetch(dn, blkid, ZIO_PRIORITY_SYNC_READ);
536 rw_exit(&dn->dn_struct_rwlock);
537 return;
538 }
539
540 /*
541 * XXX - Note, if the dnode for the requested object is not
542 * already cached, we will do a *synchronous* read in the
543 * dnode_hold() call. The same is true for any indirects.
544 */
545 err = dnode_hold(os, object, FTAG, &dn);
546 if (err != 0)
547 return;
548
549 rw_enter(&dn->dn_struct_rwlock, RW_READER);
550 if (dn->dn_datablkshift) {
551 int blkshift = dn->dn_datablkshift;
552 nblks = (P2ROUNDUP(offset + len, 1 << blkshift) -
553 P2ALIGN(offset, 1 << blkshift)) >> blkshift;
554 } else {
555 nblks = (offset < dn->dn_datablksz);
556 }
557
558 if (nblks != 0) {
559 int i;
560
561 blkid = dbuf_whichblock(dn, offset);
562 for (i = 0; i < nblks; i++)
563 dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_SYNC_READ);
564 }
565
566 rw_exit(&dn->dn_struct_rwlock);
567
568 dnode_rele(dn, FTAG);
569 }
570
571 /*
572 * Get the next "chunk" of file data to free. We traverse the file from
573 * the end so that the file gets shorter over time (if we crashes in the
574 * middle, this will leave us in a better state). We find allocated file
575 * data by simply searching the allocated level 1 indirects.
576 *
577 * On input, *start should be the first offset that does not need to be
578 * freed (e.g. "offset + length"). On return, *start will be the first
579 * offset that should be freed.
580 */
581 static int
582 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum)
583 {
584 uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
585 /* bytes of data covered by a level-1 indirect block */
586 uint64_t iblkrange =
587 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
588 uint64_t blks;
589
590 ASSERT3U(minimum, <=, *start);
591
592 if (*start - minimum <= iblkrange * maxblks) {
593 *start = minimum;
594 return (0);
595 }
596 ASSERT(ISP2(iblkrange));
597
598 for (blks = 0; *start > minimum && blks < maxblks; blks++) {
599 int err;
600
601 /*
602 * dnode_next_offset(BACKWARDS) will find an allocated L1
603 * indirect block at or before the input offset. We must
604 * decrement *start so that it is at the end of the region
605 * to search.
606 */
607 (*start)--;
608 err = dnode_next_offset(dn,
609 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
610
611 /* if there are no indirect blocks before start, we are done */
612 if (err == ESRCH) {
613 *start = minimum;
614 break;
615 } else if (err != 0) {
616 return (err);
617 }
618
619 /* set start to the beginning of this L1 indirect */
620 *start = P2ALIGN(*start, iblkrange);
621 }
622 if (*start < minimum)
623 *start = minimum;
624 return (0);
625 }
626
627 static int
628 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
629 uint64_t length)
630 {
631 uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
632 int err;
633
634 if (offset >= object_size)
635 return (0);
636
637 if (length == DMU_OBJECT_END || offset + length > object_size)
638 length = object_size - offset;
639
640 while (length != 0) {
641 uint64_t chunk_end, chunk_begin;
642 dmu_tx_t *tx;
643
644 chunk_end = chunk_begin = offset + length;
645
646 /* move chunk_begin backwards to the beginning of this chunk */
647 err = get_next_chunk(dn, &chunk_begin, offset);
648 if (err)
649 return (err);
650 ASSERT3U(chunk_begin, >=, offset);
651 ASSERT3U(chunk_begin, <=, chunk_end);
652
653 tx = dmu_tx_create(os);
654 dmu_tx_hold_free(tx, dn->dn_object,
655 chunk_begin, chunk_end - chunk_begin);
656 err = dmu_tx_assign(tx, TXG_WAIT);
657 if (err) {
658 dmu_tx_abort(tx);
659 return (err);
660 }
661 dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx);
662 dmu_tx_commit(tx);
663
664 length -= chunk_end - chunk_begin;
665 }
666 return (0);
667 }
668
669 int
670 dmu_free_long_range(objset_t *os, uint64_t object,
671 uint64_t offset, uint64_t length)
672 {
673 dnode_t *dn;
674 int err;
675
676 err = dnode_hold(os, object, FTAG, &dn);
677 if (err != 0)
678 return (err);
679 err = dmu_free_long_range_impl(os, dn, offset, length);
680
681 /*
682 * It is important to zero out the maxblkid when freeing the entire
683 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
684 * will take the fast path, and (b) dnode_reallocate() can verify
685 * that the entire file has been freed.
686 */
687 if (offset == 0 && length == DMU_OBJECT_END)
688 dn->dn_maxblkid = 0;
689
690 dnode_rele(dn, FTAG);
691 return (err);
692 }
693
694 int
695 dmu_free_long_object(objset_t *os, uint64_t object)
696 {
697 dmu_tx_t *tx;
698 int err;
699
700 err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
701 if (err != 0)
702 return (err);
703
704 tx = dmu_tx_create(os);
705 dmu_tx_hold_bonus(tx, object);
706 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
707 err = dmu_tx_assign(tx, TXG_WAIT);
708 if (err == 0) {
709 err = dmu_object_free(os, object, tx);
710 dmu_tx_commit(tx);
711 } else {
712 dmu_tx_abort(tx);
713 }
714
715 return (err);
716 }
717
718 int
719 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
720 uint64_t size, dmu_tx_t *tx)
721 {
722 dnode_t *dn;
723 int err = dnode_hold(os, object, FTAG, &dn);
724 if (err)
725 return (err);
726 ASSERT(offset < UINT64_MAX);
727 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
728 dnode_free_range(dn, offset, size, tx);
729 dnode_rele(dn, FTAG);
730 return (0);
731 }
732
733 int
734 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
735 void *buf, uint32_t flags)
736 {
737 dnode_t *dn;
738 dmu_buf_t **dbp;
739 int numbufs, err;
740
741 err = dnode_hold(os, object, FTAG, &dn);
742 if (err)
743 return (err);
744
745 /*
746 * Deal with odd block sizes, where there can't be data past the first
747 * block. If we ever do the tail block optimization, we will need to
748 * handle that here as well.
749 */
750 if (dn->dn_maxblkid == 0) {
751 int newsz = offset > dn->dn_datablksz ? 0 :
752 MIN(size, dn->dn_datablksz - offset);
753 bzero((char *)buf + newsz, size - newsz);
754 size = newsz;
755 }
756
757 while (size > 0) {
758 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
759 int i;
760
761 /*
762 * NB: we could do this block-at-a-time, but it's nice
763 * to be reading in parallel.
764 */
765 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
766 TRUE, FTAG, &numbufs, &dbp, flags);
767 if (err)
768 break;
769
770 for (i = 0; i < numbufs; i++) {
771 int tocpy;
772 int bufoff;
773 dmu_buf_t *db = dbp[i];
774
775 ASSERT(size > 0);
776
777 bufoff = offset - db->db_offset;
778 tocpy = (int)MIN(db->db_size - bufoff, size);
779
780 bcopy((char *)db->db_data + bufoff, buf, tocpy);
781
782 offset += tocpy;
783 size -= tocpy;
784 buf = (char *)buf + tocpy;
785 }
786 dmu_buf_rele_array(dbp, numbufs, FTAG);
787 }
788 dnode_rele(dn, FTAG);
789 return (err);
790 }
791
792 void
793 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
794 const void *buf, dmu_tx_t *tx)
795 {
796 dmu_buf_t **dbp;
797 int numbufs, i;
798
799 if (size == 0)
800 return;
801
802 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
803 FALSE, FTAG, &numbufs, &dbp));
804
805 for (i = 0; i < numbufs; i++) {
806 int tocpy;
807 int bufoff;
808 dmu_buf_t *db = dbp[i];
809
810 ASSERT(size > 0);
811
812 bufoff = offset - db->db_offset;
813 tocpy = (int)MIN(db->db_size - bufoff, size);
814
815 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
816
817 if (tocpy == db->db_size)
818 dmu_buf_will_fill(db, tx);
819 else
820 dmu_buf_will_dirty(db, tx);
821
822 (void) memcpy((char *)db->db_data + bufoff, buf, tocpy);
823
824 if (tocpy == db->db_size)
825 dmu_buf_fill_done(db, tx);
826
827 offset += tocpy;
828 size -= tocpy;
829 buf = (char *)buf + tocpy;
830 }
831 dmu_buf_rele_array(dbp, numbufs, FTAG);
832 }
833
834 void
835 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
836 dmu_tx_t *tx)
837 {
838 dmu_buf_t **dbp;
839 int numbufs, i;
840
841 if (size == 0)
842 return;
843
844 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
845 FALSE, FTAG, &numbufs, &dbp));
846
847 for (i = 0; i < numbufs; i++) {
848 dmu_buf_t *db = dbp[i];
849
850 dmu_buf_will_not_fill(db, tx);
851 }
852 dmu_buf_rele_array(dbp, numbufs, FTAG);
853 }
854
855 /*
856 * DMU support for xuio
857 */
858 kstat_t *xuio_ksp = NULL;
859
860 typedef struct xuio_stats {
861 /* loaned yet not returned arc_buf */
862 kstat_named_t xuiostat_onloan_rbuf;
863 kstat_named_t xuiostat_onloan_wbuf;
864 /* whether a copy is made when loaning out a read buffer */
865 kstat_named_t xuiostat_rbuf_copied;
866 kstat_named_t xuiostat_rbuf_nocopy;
867 /* whether a copy is made when assigning a write buffer */
868 kstat_named_t xuiostat_wbuf_copied;
869 kstat_named_t xuiostat_wbuf_nocopy;
870 } xuio_stats_t;
871
872 static xuio_stats_t xuio_stats = {
873 { "onloan_read_buf", KSTAT_DATA_UINT64 },
874 { "onloan_write_buf", KSTAT_DATA_UINT64 },
875 { "read_buf_copied", KSTAT_DATA_UINT64 },
876 { "read_buf_nocopy", KSTAT_DATA_UINT64 },
877 { "write_buf_copied", KSTAT_DATA_UINT64 },
878 { "write_buf_nocopy", KSTAT_DATA_UINT64 }
879 };
880
881 #define XUIOSTAT_INCR(stat, val) \
882 atomic_add_64(&xuio_stats.stat.value.ui64, (val))
883 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
884
885 int
886 dmu_xuio_init(xuio_t *xuio, int nblk)
887 {
888 dmu_xuio_t *priv;
889 uio_t *uio = &xuio->xu_uio;
890
891 uio->uio_iovcnt = nblk;
892 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_PUSHPAGE);
893
894 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_PUSHPAGE);
895 priv->cnt = nblk;
896 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_PUSHPAGE);
897 priv->iovp = uio->uio_iov;
898 XUIO_XUZC_PRIV(xuio) = priv;
899
900 if (XUIO_XUZC_RW(xuio) == UIO_READ)
901 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
902 else
903 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
904
905 return (0);
906 }
907
908 void
909 dmu_xuio_fini(xuio_t *xuio)
910 {
911 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
912 int nblk = priv->cnt;
913
914 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
915 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
916 kmem_free(priv, sizeof (dmu_xuio_t));
917
918 if (XUIO_XUZC_RW(xuio) == UIO_READ)
919 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
920 else
921 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
922 }
923
924 /*
925 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
926 * and increase priv->next by 1.
927 */
928 int
929 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
930 {
931 struct iovec *iov;
932 uio_t *uio = &xuio->xu_uio;
933 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
934 int i = priv->next++;
935
936 ASSERT(i < priv->cnt);
937 ASSERT(off + n <= arc_buf_size(abuf));
938 iov = uio->uio_iov + i;
939 iov->iov_base = (char *)abuf->b_data + off;
940 iov->iov_len = n;
941 priv->bufs[i] = abuf;
942 return (0);
943 }
944
945 int
946 dmu_xuio_cnt(xuio_t *xuio)
947 {
948 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
949 return (priv->cnt);
950 }
951
952 arc_buf_t *
953 dmu_xuio_arcbuf(xuio_t *xuio, int i)
954 {
955 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
956
957 ASSERT(i < priv->cnt);
958 return (priv->bufs[i]);
959 }
960
961 void
962 dmu_xuio_clear(xuio_t *xuio, int i)
963 {
964 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
965
966 ASSERT(i < priv->cnt);
967 priv->bufs[i] = NULL;
968 }
969
970 static void
971 xuio_stat_init(void)
972 {
973 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
974 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
975 KSTAT_FLAG_VIRTUAL);
976 if (xuio_ksp != NULL) {
977 xuio_ksp->ks_data = &xuio_stats;
978 kstat_install(xuio_ksp);
979 }
980 }
981
982 static void
983 xuio_stat_fini(void)
984 {
985 if (xuio_ksp != NULL) {
986 kstat_delete(xuio_ksp);
987 xuio_ksp = NULL;
988 }
989 }
990
991 void
992 xuio_stat_wbuf_copied()
993 {
994 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
995 }
996
997 void
998 xuio_stat_wbuf_nocopy()
999 {
1000 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1001 }
1002
1003 #ifdef _KERNEL
1004
1005 /*
1006 * Copy up to size bytes between arg_buf and req based on the data direction
1007 * described by the req. If an entire req's data cannot be transfered the
1008 * req's is updated such that it's current index and bv offsets correctly
1009 * reference any residual data which could not be copied. The return value
1010 * is the number of bytes successfully copied to arg_buf.
1011 */
1012 static int
1013 dmu_req_copy(void *arg_buf, int size, int *offset, struct request *req)
1014 {
1015 struct bio_vec *bv;
1016 struct req_iterator iter;
1017 char *bv_buf;
1018 int tocpy;
1019
1020 *offset = 0;
1021 rq_for_each_segment(bv, req, iter) {
1022
1023 /* Fully consumed the passed arg_buf */
1024 ASSERT3S(*offset, <=, size);
1025 if (size == *offset)
1026 break;
1027
1028 /* Skip fully consumed bv's */
1029 if (bv->bv_len == 0)
1030 continue;
1031
1032 tocpy = MIN(bv->bv_len, size - *offset);
1033 ASSERT3S(tocpy, >=, 0);
1034
1035 bv_buf = page_address(bv->bv_page) + bv->bv_offset;
1036 ASSERT3P(bv_buf, !=, NULL);
1037
1038 if (rq_data_dir(req) == WRITE)
1039 memcpy(arg_buf + *offset, bv_buf, tocpy);
1040 else
1041 memcpy(bv_buf, arg_buf + *offset, tocpy);
1042
1043 *offset += tocpy;
1044 bv->bv_offset += tocpy;
1045 bv->bv_len -= tocpy;
1046 }
1047
1048 return (0);
1049 }
1050
1051 static void
1052 dmu_bio_put(struct bio *bio)
1053 {
1054 struct bio *bio_next;
1055
1056 while (bio) {
1057 bio_next = bio->bi_next;
1058 bio_put(bio);
1059 bio = bio_next;
1060 }
1061 }
1062
1063 static int
1064 dmu_bio_clone(struct bio *bio, struct bio **bio_copy)
1065 {
1066 struct bio *bio_root = NULL;
1067 struct bio *bio_last = NULL;
1068 struct bio *bio_new;
1069
1070 if (bio == NULL)
1071 return (EINVAL);
1072
1073 while (bio) {
1074 bio_new = bio_clone(bio, GFP_NOIO);
1075 if (bio_new == NULL) {
1076 dmu_bio_put(bio_root);
1077 return (ENOMEM);
1078 }
1079
1080 if (bio_last) {
1081 bio_last->bi_next = bio_new;
1082 bio_last = bio_new;
1083 } else {
1084 bio_root = bio_new;
1085 bio_last = bio_new;
1086 }
1087
1088 bio = bio->bi_next;
1089 }
1090
1091 *bio_copy = bio_root;
1092
1093 return (0);
1094 }
1095
1096 int
1097 dmu_read_req(objset_t *os, uint64_t object, struct request *req)
1098 {
1099 uint64_t size = blk_rq_bytes(req);
1100 uint64_t offset = blk_rq_pos(req) << 9;
1101 struct bio *bio_saved = req->bio;
1102 dmu_buf_t **dbp;
1103 int numbufs, i, err;
1104
1105 /*
1106 * NB: we could do this block-at-a-time, but it's nice
1107 * to be reading in parallel.
1108 */
1109 err = dmu_buf_hold_array(os, object, offset, size, TRUE, FTAG,
1110 &numbufs, &dbp);
1111 if (err)
1112 return (err);
1113
1114 /*
1115 * Clone the bio list so the bv->bv_offset and bv->bv_len members
1116 * can be safely modified. The original bio list is relinked in to
1117 * the request when the function exits. This is required because
1118 * some file systems blindly assume that these values will remain
1119 * constant between bio_submit() and the IO completion callback.
1120 */
1121 err = dmu_bio_clone(bio_saved, &req->bio);
1122 if (err)
1123 goto error;
1124
1125 for (i = 0; i < numbufs; i++) {
1126 int tocpy, didcpy, bufoff;
1127 dmu_buf_t *db = dbp[i];
1128
1129 bufoff = offset - db->db_offset;
1130 ASSERT3S(bufoff, >=, 0);
1131
1132 tocpy = (int)MIN(db->db_size - bufoff, size);
1133 if (tocpy == 0)
1134 break;
1135
1136 err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req);
1137
1138 if (didcpy < tocpy)
1139 err = EIO;
1140
1141 if (err)
1142 break;
1143
1144 size -= tocpy;
1145 offset += didcpy;
1146 err = 0;
1147 }
1148
1149 dmu_bio_put(req->bio);
1150 req->bio = bio_saved;
1151 error:
1152 dmu_buf_rele_array(dbp, numbufs, FTAG);
1153
1154 return (err);
1155 }
1156
1157 int
1158 dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx)
1159 {
1160 uint64_t size = blk_rq_bytes(req);
1161 uint64_t offset = blk_rq_pos(req) << 9;
1162 struct bio *bio_saved = req->bio;
1163 dmu_buf_t **dbp;
1164 int numbufs;
1165 int err = 0;
1166 int i;
1167
1168 if (size == 0)
1169 return (0);
1170
1171 err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG,
1172 &numbufs, &dbp);
1173 if (err)
1174 return (err);
1175
1176 /*
1177 * Clone the bio list so the bv->bv_offset and bv->bv_len members
1178 * can be safely modified. The original bio list is relinked in to
1179 * the request when the function exits. This is required because
1180 * some file systems blindly assume that these values will remain
1181 * constant between bio_submit() and the IO completion callback.
1182 */
1183 err = dmu_bio_clone(bio_saved, &req->bio);
1184 if (err)
1185 goto error;
1186
1187 for (i = 0; i < numbufs; i++) {
1188 int tocpy, didcpy, bufoff;
1189 dmu_buf_t *db = dbp[i];
1190
1191 bufoff = offset - db->db_offset;
1192 ASSERT3S(bufoff, >=, 0);
1193
1194 tocpy = (int)MIN(db->db_size - bufoff, size);
1195 if (tocpy == 0)
1196 break;
1197
1198 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1199
1200 if (tocpy == db->db_size)
1201 dmu_buf_will_fill(db, tx);
1202 else
1203 dmu_buf_will_dirty(db, tx);
1204
1205 err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req);
1206
1207 if (tocpy == db->db_size)
1208 dmu_buf_fill_done(db, tx);
1209
1210 if (didcpy < tocpy)
1211 err = EIO;
1212
1213 if (err)
1214 break;
1215
1216 size -= tocpy;
1217 offset += didcpy;
1218 err = 0;
1219 }
1220
1221 dmu_bio_put(req->bio);
1222 req->bio = bio_saved;
1223 error:
1224 dmu_buf_rele_array(dbp, numbufs, FTAG);
1225
1226 return (err);
1227 }
1228
1229 int
1230 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1231 {
1232 dmu_buf_t **dbp;
1233 int numbufs, i, err;
1234 xuio_t *xuio = NULL;
1235
1236 /*
1237 * NB: we could do this block-at-a-time, but it's nice
1238 * to be reading in parallel.
1239 */
1240 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
1241 &numbufs, &dbp);
1242 if (err)
1243 return (err);
1244
1245 for (i = 0; i < numbufs; i++) {
1246 int tocpy;
1247 int bufoff;
1248 dmu_buf_t *db = dbp[i];
1249
1250 ASSERT(size > 0);
1251
1252 bufoff = uio->uio_loffset - db->db_offset;
1253 tocpy = (int)MIN(db->db_size - bufoff, size);
1254
1255 if (xuio) {
1256 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1257 arc_buf_t *dbuf_abuf = dbi->db_buf;
1258 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1259 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1260 if (!err) {
1261 uio->uio_resid -= tocpy;
1262 uio->uio_loffset += tocpy;
1263 }
1264
1265 if (abuf == dbuf_abuf)
1266 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1267 else
1268 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1269 } else {
1270 err = uiomove((char *)db->db_data + bufoff, tocpy,
1271 UIO_READ, uio);
1272 }
1273 if (err)
1274 break;
1275
1276 size -= tocpy;
1277 }
1278 dmu_buf_rele_array(dbp, numbufs, FTAG);
1279
1280 return (err);
1281 }
1282
1283 static int
1284 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1285 {
1286 dmu_buf_t **dbp;
1287 int numbufs;
1288 int err = 0;
1289 int i;
1290
1291 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1292 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1293 if (err)
1294 return (err);
1295
1296 for (i = 0; i < numbufs; i++) {
1297 int tocpy;
1298 int bufoff;
1299 dmu_buf_t *db = dbp[i];
1300
1301 ASSERT(size > 0);
1302
1303 bufoff = uio->uio_loffset - db->db_offset;
1304 tocpy = (int)MIN(db->db_size - bufoff, size);
1305
1306 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1307
1308 if (tocpy == db->db_size)
1309 dmu_buf_will_fill(db, tx);
1310 else
1311 dmu_buf_will_dirty(db, tx);
1312
1313 /*
1314 * XXX uiomove could block forever (eg.nfs-backed
1315 * pages). There needs to be a uiolockdown() function
1316 * to lock the pages in memory, so that uiomove won't
1317 * block.
1318 */
1319 err = uiomove((char *)db->db_data + bufoff, tocpy,
1320 UIO_WRITE, uio);
1321
1322 if (tocpy == db->db_size)
1323 dmu_buf_fill_done(db, tx);
1324
1325 if (err)
1326 break;
1327
1328 size -= tocpy;
1329 }
1330
1331 dmu_buf_rele_array(dbp, numbufs, FTAG);
1332 return (err);
1333 }
1334
1335 int
1336 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1337 dmu_tx_t *tx)
1338 {
1339 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1340 dnode_t *dn;
1341 int err;
1342
1343 if (size == 0)
1344 return (0);
1345
1346 DB_DNODE_ENTER(db);
1347 dn = DB_DNODE(db);
1348 err = dmu_write_uio_dnode(dn, uio, size, tx);
1349 DB_DNODE_EXIT(db);
1350
1351 return (err);
1352 }
1353
1354 int
1355 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1356 dmu_tx_t *tx)
1357 {
1358 dnode_t *dn;
1359 int err;
1360
1361 if (size == 0)
1362 return (0);
1363
1364 err = dnode_hold(os, object, FTAG, &dn);
1365 if (err)
1366 return (err);
1367
1368 err = dmu_write_uio_dnode(dn, uio, size, tx);
1369
1370 dnode_rele(dn, FTAG);
1371
1372 return (err);
1373 }
1374 #endif /* _KERNEL */
1375
1376 /*
1377 * Allocate a loaned anonymous arc buffer.
1378 */
1379 arc_buf_t *
1380 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1381 {
1382 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1383 spa_t *spa;
1384
1385 DB_GET_SPA(&spa, db);
1386 return (arc_loan_buf(spa, size));
1387 }
1388
1389 /*
1390 * Free a loaned arc buffer.
1391 */
1392 void
1393 dmu_return_arcbuf(arc_buf_t *buf)
1394 {
1395 arc_return_buf(buf, FTAG);
1396 VERIFY(arc_buf_remove_ref(buf, FTAG));
1397 }
1398
1399 /*
1400 * When possible directly assign passed loaned arc buffer to a dbuf.
1401 * If this is not possible copy the contents of passed arc buf via
1402 * dmu_write().
1403 */
1404 void
1405 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1406 dmu_tx_t *tx)
1407 {
1408 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1409 dnode_t *dn;
1410 dmu_buf_impl_t *db;
1411 uint32_t blksz = (uint32_t)arc_buf_size(buf);
1412 uint64_t blkid;
1413
1414 DB_DNODE_ENTER(dbuf);
1415 dn = DB_DNODE(dbuf);
1416 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1417 blkid = dbuf_whichblock(dn, offset);
1418 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1419 rw_exit(&dn->dn_struct_rwlock);
1420 DB_DNODE_EXIT(dbuf);
1421
1422 if (offset == db->db.db_offset && blksz == db->db.db_size) {
1423 dbuf_assign_arcbuf(db, buf, tx);
1424 dbuf_rele(db, FTAG);
1425 } else {
1426 objset_t *os;
1427 uint64_t object;
1428
1429 DB_DNODE_ENTER(dbuf);
1430 dn = DB_DNODE(dbuf);
1431 os = dn->dn_objset;
1432 object = dn->dn_object;
1433 DB_DNODE_EXIT(dbuf);
1434
1435 dbuf_rele(db, FTAG);
1436 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1437 dmu_return_arcbuf(buf);
1438 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1439 }
1440 }
1441
1442 typedef struct {
1443 dbuf_dirty_record_t *dsa_dr;
1444 dmu_sync_cb_t *dsa_done;
1445 zgd_t *dsa_zgd;
1446 dmu_tx_t *dsa_tx;
1447 } dmu_sync_arg_t;
1448
1449 /* ARGSUSED */
1450 static void
1451 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1452 {
1453 dmu_sync_arg_t *dsa = varg;
1454 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1455 blkptr_t *bp = zio->io_bp;
1456
1457 if (zio->io_error == 0) {
1458 if (BP_IS_HOLE(bp)) {
1459 /*
1460 * A block of zeros may compress to a hole, but the
1461 * block size still needs to be known for replay.
1462 */
1463 BP_SET_LSIZE(bp, db->db_size);
1464 } else {
1465 ASSERT(BP_GET_LEVEL(bp) == 0);
1466 bp->blk_fill = 1;
1467 }
1468 }
1469 }
1470
1471 static void
1472 dmu_sync_late_arrival_ready(zio_t *zio)
1473 {
1474 dmu_sync_ready(zio, NULL, zio->io_private);
1475 }
1476
1477 /* ARGSUSED */
1478 static void
1479 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1480 {
1481 dmu_sync_arg_t *dsa = varg;
1482 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1483 dmu_buf_impl_t *db = dr->dr_dbuf;
1484
1485 mutex_enter(&db->db_mtx);
1486 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1487 if (zio->io_error == 0) {
1488 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1489 if (dr->dt.dl.dr_nopwrite) {
1490 ASSERTV(blkptr_t *bp = zio->io_bp);
1491 ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
1492 ASSERTV(uint8_t chksum = BP_GET_CHECKSUM(bp_orig));
1493
1494 ASSERT(BP_EQUAL(bp, bp_orig));
1495 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1496 ASSERT(zio_checksum_table[chksum].ci_dedup);
1497 }
1498 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1499 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1500 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1501 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
1502 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1503 } else {
1504 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1505 }
1506 cv_broadcast(&db->db_changed);
1507 mutex_exit(&db->db_mtx);
1508
1509 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1510
1511 kmem_free(dsa, sizeof (*dsa));
1512 }
1513
1514 static void
1515 dmu_sync_late_arrival_done(zio_t *zio)
1516 {
1517 blkptr_t *bp = zio->io_bp;
1518 dmu_sync_arg_t *dsa = zio->io_private;
1519 ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
1520
1521 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1522 /*
1523 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1524 * then there is nothing to do here. Otherwise, free the
1525 * newly allocated block in this txg.
1526 */
1527 if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
1528 ASSERT(BP_EQUAL(bp, bp_orig));
1529 } else {
1530 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1531 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1532 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1533 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1534 }
1535 }
1536
1537 dmu_tx_commit(dsa->dsa_tx);
1538
1539 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1540
1541 kmem_free(dsa, sizeof (*dsa));
1542 }
1543
1544 static int
1545 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1546 zio_prop_t *zp, zbookmark_t *zb)
1547 {
1548 dmu_sync_arg_t *dsa;
1549 dmu_tx_t *tx;
1550
1551 tx = dmu_tx_create(os);
1552 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1553 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1554 dmu_tx_abort(tx);
1555 /* Make zl_get_data do txg_waited_synced() */
1556 return (SET_ERROR(EIO));
1557 }
1558
1559 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE);
1560 dsa->dsa_dr = NULL;
1561 dsa->dsa_done = done;
1562 dsa->dsa_zgd = zgd;
1563 dsa->dsa_tx = tx;
1564
1565 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1566 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
1567 dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa,
1568 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL|ZIO_FLAG_FASTWRITE, zb));
1569
1570 return (0);
1571 }
1572
1573 /*
1574 * Intent log support: sync the block associated with db to disk.
1575 * N.B. and XXX: the caller is responsible for making sure that the
1576 * data isn't changing while dmu_sync() is writing it.
1577 *
1578 * Return values:
1579 *
1580 * EEXIST: this txg has already been synced, so there's nothing to do.
1581 * The caller should not log the write.
1582 *
1583 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1584 * The caller should not log the write.
1585 *
1586 * EALREADY: this block is already in the process of being synced.
1587 * The caller should track its progress (somehow).
1588 *
1589 * EIO: could not do the I/O.
1590 * The caller should do a txg_wait_synced().
1591 *
1592 * 0: the I/O has been initiated.
1593 * The caller should log this blkptr in the done callback.
1594 * It is possible that the I/O will fail, in which case
1595 * the error will be reported to the done callback and
1596 * propagated to pio from zio_done().
1597 */
1598 int
1599 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1600 {
1601 blkptr_t *bp = zgd->zgd_bp;
1602 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1603 objset_t *os = db->db_objset;
1604 dsl_dataset_t *ds = os->os_dsl_dataset;
1605 dbuf_dirty_record_t *dr;
1606 dmu_sync_arg_t *dsa;
1607 zbookmark_t zb;
1608 zio_prop_t zp;
1609 dnode_t *dn;
1610
1611 ASSERT(pio != NULL);
1612 ASSERT(txg != 0);
1613
1614 SET_BOOKMARK(&zb, ds->ds_object,
1615 db->db.db_object, db->db_level, db->db_blkid);
1616
1617 DB_DNODE_ENTER(db);
1618 dn = DB_DNODE(db);
1619 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1620 DB_DNODE_EXIT(db);
1621
1622 /*
1623 * If we're frozen (running ziltest), we always need to generate a bp.
1624 */
1625 if (txg > spa_freeze_txg(os->os_spa))
1626 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1627
1628 /*
1629 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1630 * and us. If we determine that this txg is not yet syncing,
1631 * but it begins to sync a moment later, that's OK because the
1632 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1633 */
1634 mutex_enter(&db->db_mtx);
1635
1636 if (txg <= spa_last_synced_txg(os->os_spa)) {
1637 /*
1638 * This txg has already synced. There's nothing to do.
1639 */
1640 mutex_exit(&db->db_mtx);
1641 return (SET_ERROR(EEXIST));
1642 }
1643
1644 if (txg <= spa_syncing_txg(os->os_spa)) {
1645 /*
1646 * This txg is currently syncing, so we can't mess with
1647 * the dirty record anymore; just write a new log block.
1648 */
1649 mutex_exit(&db->db_mtx);
1650 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1651 }
1652
1653 dr = db->db_last_dirty;
1654 while (dr && dr->dr_txg != txg)
1655 dr = dr->dr_next;
1656
1657 if (dr == NULL) {
1658 /*
1659 * There's no dr for this dbuf, so it must have been freed.
1660 * There's no need to log writes to freed blocks, so we're done.
1661 */
1662 mutex_exit(&db->db_mtx);
1663 return (SET_ERROR(ENOENT));
1664 }
1665
1666 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
1667
1668 /*
1669 * Assume the on-disk data is X, the current syncing data is Y,
1670 * and the current in-memory data is Z (currently in dmu_sync).
1671 * X and Z are identical but Y is has been modified. Normally,
1672 * when X and Z are the same we will perform a nopwrite but if Y
1673 * is different we must disable nopwrite since the resulting write
1674 * of Y to disk can free the block containing X. If we allowed a
1675 * nopwrite to occur the block pointing to Z would reference a freed
1676 * block. Since this is a rare case we simplify this by disabling
1677 * nopwrite if the current dmu_sync-ing dbuf has been modified in
1678 * a previous transaction.
1679 */
1680 if (dr->dr_next)
1681 zp.zp_nopwrite = B_FALSE;
1682
1683 ASSERT(dr->dr_txg == txg);
1684 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1685 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1686 /*
1687 * We have already issued a sync write for this buffer,
1688 * or this buffer has already been synced. It could not
1689 * have been dirtied since, or we would have cleared the state.
1690 */
1691 mutex_exit(&db->db_mtx);
1692 return (SET_ERROR(EALREADY));
1693 }
1694
1695 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1696 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1697 mutex_exit(&db->db_mtx);
1698
1699 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE);
1700 dsa->dsa_dr = dr;
1701 dsa->dsa_done = done;
1702 dsa->dsa_zgd = zgd;
1703 dsa->dsa_tx = NULL;
1704
1705 zio_nowait(arc_write(pio, os->os_spa, txg,
1706 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
1707 DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready,
1708 NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE,
1709 ZIO_FLAG_CANFAIL, &zb));
1710
1711 return (0);
1712 }
1713
1714 int
1715 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1716 dmu_tx_t *tx)
1717 {
1718 dnode_t *dn;
1719 int err;
1720
1721 err = dnode_hold(os, object, FTAG, &dn);
1722 if (err)
1723 return (err);
1724 err = dnode_set_blksz(dn, size, ibs, tx);
1725 dnode_rele(dn, FTAG);
1726 return (err);
1727 }
1728
1729 void
1730 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1731 dmu_tx_t *tx)
1732 {
1733 dnode_t *dn;
1734
1735 /* XXX assumes dnode_hold will not get an i/o error */
1736 (void) dnode_hold(os, object, FTAG, &dn);
1737 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
1738 dn->dn_checksum = checksum;
1739 dnode_setdirty(dn, tx);
1740 dnode_rele(dn, FTAG);
1741 }
1742
1743 void
1744 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1745 dmu_tx_t *tx)
1746 {
1747 dnode_t *dn;
1748
1749 /* XXX assumes dnode_hold will not get an i/o error */
1750 (void) dnode_hold(os, object, FTAG, &dn);
1751 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
1752 dn->dn_compress = compress;
1753 dnode_setdirty(dn, tx);
1754 dnode_rele(dn, FTAG);
1755 }
1756
1757 int zfs_mdcomp_disable = 0;
1758
1759 void
1760 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1761 {
1762 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1763 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1764 (wp & WP_SPILL));
1765 enum zio_checksum checksum = os->os_checksum;
1766 enum zio_compress compress = os->os_compress;
1767 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1768 boolean_t dedup = B_FALSE;
1769 boolean_t nopwrite = B_FALSE;
1770 boolean_t dedup_verify = os->os_dedup_verify;
1771 int copies = os->os_copies;
1772
1773 /*
1774 * We maintain different write policies for each of the following
1775 * types of data:
1776 * 1. metadata
1777 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
1778 * 3. all other level 0 blocks
1779 */
1780 if (ismd) {
1781 /*
1782 * XXX -- we should design a compression algorithm
1783 * that specializes in arrays of bps.
1784 */
1785 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
1786 ZIO_COMPRESS_LZJB;
1787
1788 /*
1789 * Metadata always gets checksummed. If the data
1790 * checksum is multi-bit correctable, and it's not a
1791 * ZBT-style checksum, then it's suitable for metadata
1792 * as well. Otherwise, the metadata checksum defaults
1793 * to fletcher4.
1794 */
1795 if (zio_checksum_table[checksum].ci_correctable < 1 ||
1796 zio_checksum_table[checksum].ci_eck)
1797 checksum = ZIO_CHECKSUM_FLETCHER_4;
1798 } else if (wp & WP_NOFILL) {
1799 ASSERT(level == 0);
1800
1801 /*
1802 * If we're writing preallocated blocks, we aren't actually
1803 * writing them so don't set any policy properties. These
1804 * blocks are currently only used by an external subsystem
1805 * outside of zfs (i.e. dump) and not written by the zio
1806 * pipeline.
1807 */
1808 compress = ZIO_COMPRESS_OFF;
1809 checksum = ZIO_CHECKSUM_OFF;
1810 } else {
1811 compress = zio_compress_select(dn->dn_compress, compress);
1812
1813 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
1814 zio_checksum_select(dn->dn_checksum, checksum) :
1815 dedup_checksum;
1816
1817 /*
1818 * Determine dedup setting. If we are in dmu_sync(),
1819 * we won't actually dedup now because that's all
1820 * done in syncing context; but we do want to use the
1821 * dedup checkum. If the checksum is not strong
1822 * enough to ensure unique signatures, force
1823 * dedup_verify.
1824 */
1825 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
1826 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
1827 if (!zio_checksum_table[checksum].ci_dedup)
1828 dedup_verify = B_TRUE;
1829 }
1830
1831 /*
1832 * Enable nopwrite if we have a cryptographically secure
1833 * checksum that has no known collisions (i.e. SHA-256)
1834 * and compression is enabled. We don't enable nopwrite if
1835 * dedup is enabled as the two features are mutually exclusive.
1836 */
1837 nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup &&
1838 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
1839 }
1840
1841 zp->zp_checksum = checksum;
1842 zp->zp_compress = compress;
1843 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
1844 zp->zp_level = level;
1845 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa));
1846 zp->zp_dedup = dedup;
1847 zp->zp_dedup_verify = dedup && dedup_verify;
1848 zp->zp_nopwrite = nopwrite;
1849 }
1850
1851 int
1852 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1853 {
1854 dnode_t *dn;
1855 int i, err;
1856
1857 err = dnode_hold(os, object, FTAG, &dn);
1858 if (err)
1859 return (err);
1860 /*
1861 * Sync any current changes before
1862 * we go trundling through the block pointers.
1863 */
1864 for (i = 0; i < TXG_SIZE; i++) {
1865 if (list_link_active(&dn->dn_dirty_link[i]))
1866 break;
1867 }
1868 if (i != TXG_SIZE) {
1869 dnode_rele(dn, FTAG);
1870 txg_wait_synced(dmu_objset_pool(os), 0);
1871 err = dnode_hold(os, object, FTAG, &dn);
1872 if (err)
1873 return (err);
1874 }
1875
1876 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1877 dnode_rele(dn, FTAG);
1878
1879 return (err);
1880 }
1881
1882 void
1883 __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1884 {
1885 dnode_phys_t *dnp = dn->dn_phys;
1886 int i;
1887
1888 doi->doi_data_block_size = dn->dn_datablksz;
1889 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1890 1ULL << dn->dn_indblkshift : 0;
1891 doi->doi_type = dn->dn_type;
1892 doi->doi_bonus_type = dn->dn_bonustype;
1893 doi->doi_bonus_size = dn->dn_bonuslen;
1894 doi->doi_indirection = dn->dn_nlevels;
1895 doi->doi_checksum = dn->dn_checksum;
1896 doi->doi_compress = dn->dn_compress;
1897 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1898 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
1899 doi->doi_fill_count = 0;
1900 for (i = 0; i < dnp->dn_nblkptr; i++)
1901 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
1902 }
1903
1904 void
1905 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1906 {
1907 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1908 mutex_enter(&dn->dn_mtx);
1909
1910 __dmu_object_info_from_dnode(dn, doi);
1911
1912 mutex_exit(&dn->dn_mtx);
1913 rw_exit(&dn->dn_struct_rwlock);
1914 }
1915
1916 /*
1917 * Get information on a DMU object.
1918 * If doi is NULL, just indicates whether the object exists.
1919 */
1920 int
1921 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1922 {
1923 dnode_t *dn;
1924 int err = dnode_hold(os, object, FTAG, &dn);
1925
1926 if (err)
1927 return (err);
1928
1929 if (doi != NULL)
1930 dmu_object_info_from_dnode(dn, doi);
1931
1932 dnode_rele(dn, FTAG);
1933 return (0);
1934 }
1935
1936 /*
1937 * As above, but faster; can be used when you have a held dbuf in hand.
1938 */
1939 void
1940 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1941 {
1942 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1943
1944 DB_DNODE_ENTER(db);
1945 dmu_object_info_from_dnode(DB_DNODE(db), doi);
1946 DB_DNODE_EXIT(db);
1947 }
1948
1949 /*
1950 * Faster still when you only care about the size.
1951 * This is specifically optimized for zfs_getattr().
1952 */
1953 void
1954 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1955 u_longlong_t *nblk512)
1956 {
1957 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1958 dnode_t *dn;
1959
1960 DB_DNODE_ENTER(db);
1961 dn = DB_DNODE(db);
1962
1963 *blksize = dn->dn_datablksz;
1964 /* add 1 for dnode space */
1965 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1966 SPA_MINBLOCKSHIFT) + 1;
1967 DB_DNODE_EXIT(db);
1968 }
1969
1970 void
1971 byteswap_uint64_array(void *vbuf, size_t size)
1972 {
1973 uint64_t *buf = vbuf;
1974 size_t count = size >> 3;
1975 int i;
1976
1977 ASSERT((size & 7) == 0);
1978
1979 for (i = 0; i < count; i++)
1980 buf[i] = BSWAP_64(buf[i]);
1981 }
1982
1983 void
1984 byteswap_uint32_array(void *vbuf, size_t size)
1985 {
1986 uint32_t *buf = vbuf;
1987 size_t count = size >> 2;
1988 int i;
1989
1990 ASSERT((size & 3) == 0);
1991
1992 for (i = 0; i < count; i++)
1993 buf[i] = BSWAP_32(buf[i]);
1994 }
1995
1996 void
1997 byteswap_uint16_array(void *vbuf, size_t size)
1998 {
1999 uint16_t *buf = vbuf;
2000 size_t count = size >> 1;
2001 int i;
2002
2003 ASSERT((size & 1) == 0);
2004
2005 for (i = 0; i < count; i++)
2006 buf[i] = BSWAP_16(buf[i]);
2007 }
2008
2009 /* ARGSUSED */
2010 void
2011 byteswap_uint8_array(void *vbuf, size_t size)
2012 {
2013 }
2014
2015 void
2016 dmu_init(void)
2017 {
2018 zfs_dbgmsg_init();
2019 sa_cache_init();
2020 xuio_stat_init();
2021 dmu_objset_init();
2022 dnode_init();
2023 dbuf_init();
2024 zfetch_init();
2025 dmu_tx_init();
2026 l2arc_init();
2027 arc_init();
2028 }
2029
2030 void
2031 dmu_fini(void)
2032 {
2033 arc_fini(); /* arc depends on l2arc, so arc must go first */
2034 l2arc_fini();
2035 dmu_tx_fini();
2036 zfetch_fini();
2037 dbuf_fini();
2038 dnode_fini();
2039 dmu_objset_fini();
2040 xuio_stat_fini();
2041 sa_cache_fini();
2042 zfs_dbgmsg_fini();
2043 }
2044
2045 #if defined(_KERNEL) && defined(HAVE_SPL)
2046 EXPORT_SYMBOL(dmu_bonus_hold);
2047 EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus);
2048 EXPORT_SYMBOL(dmu_buf_rele_array);
2049 EXPORT_SYMBOL(dmu_prefetch);
2050 EXPORT_SYMBOL(dmu_free_range);
2051 EXPORT_SYMBOL(dmu_free_long_range);
2052 EXPORT_SYMBOL(dmu_free_long_object);
2053 EXPORT_SYMBOL(dmu_read);
2054 EXPORT_SYMBOL(dmu_write);
2055 EXPORT_SYMBOL(dmu_prealloc);
2056 EXPORT_SYMBOL(dmu_object_info);
2057 EXPORT_SYMBOL(dmu_object_info_from_dnode);
2058 EXPORT_SYMBOL(dmu_object_info_from_db);
2059 EXPORT_SYMBOL(dmu_object_size_from_db);
2060 EXPORT_SYMBOL(dmu_object_set_blocksize);
2061 EXPORT_SYMBOL(dmu_object_set_checksum);
2062 EXPORT_SYMBOL(dmu_object_set_compress);
2063 EXPORT_SYMBOL(dmu_write_policy);
2064 EXPORT_SYMBOL(dmu_sync);
2065 EXPORT_SYMBOL(dmu_request_arcbuf);
2066 EXPORT_SYMBOL(dmu_return_arcbuf);
2067 EXPORT_SYMBOL(dmu_assign_arcbuf);
2068 EXPORT_SYMBOL(dmu_buf_hold);
2069 EXPORT_SYMBOL(dmu_ot);
2070
2071 module_param(zfs_mdcomp_disable, int, 0644);
2072 MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression");
2073
2074 module_param(zfs_nopwrite_enabled, int, 0644);
2075 MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes");
2076
2077 #endif
ZFS On Linux mirror for PVE