]> 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
Illumos 5162 - zfs recv should use loaned arc buffer to avoid copy
[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, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2014, Nexenta Systems, Inc. All rights reserved.
26 */
27
28 #include <sys/dmu.h>
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dbuf.h>
32 #include <sys/dnode.h>
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_pool.h>
39 #include <sys/dsl_synctask.h>
40 #include <sys/dsl_prop.h>
41 #include <sys/dmu_zfetch.h>
42 #include <sys/zfs_ioctl.h>
43 #include <sys/zap.h>
44 #include <sys/zio_checksum.h>
45 #include <sys/zio_compress.h>
46 #include <sys/sa.h>
47 #include <sys/zfeature.h>
48 #ifdef _KERNEL
49 #include <sys/vmsystm.h>
50 #include <sys/zfs_znode.h>
51 #endif
52
53 /*
54 * Enable/disable nopwrite feature.
55 */
56 int zfs_nopwrite_enabled = 1;
57
58 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
59 { DMU_BSWAP_UINT8, TRUE, "unallocated" },
60 { DMU_BSWAP_ZAP, TRUE, "object directory" },
61 { DMU_BSWAP_UINT64, TRUE, "object array" },
62 { DMU_BSWAP_UINT8, TRUE, "packed nvlist" },
63 { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" },
64 { DMU_BSWAP_UINT64, TRUE, "bpobj" },
65 { DMU_BSWAP_UINT64, TRUE, "bpobj header" },
66 { DMU_BSWAP_UINT64, TRUE, "SPA space map header" },
67 { DMU_BSWAP_UINT64, TRUE, "SPA space map" },
68 { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" },
69 { DMU_BSWAP_DNODE, TRUE, "DMU dnode" },
70 { DMU_BSWAP_OBJSET, TRUE, "DMU objset" },
71 { DMU_BSWAP_UINT64, TRUE, "DSL directory" },
72 { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"},
73 { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" },
74 { DMU_BSWAP_ZAP, TRUE, "DSL props" },
75 { DMU_BSWAP_UINT64, TRUE, "DSL dataset" },
76 { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" },
77 { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" },
78 { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" },
79 { DMU_BSWAP_ZAP, TRUE, "ZFS directory" },
80 { DMU_BSWAP_ZAP, TRUE, "ZFS master node" },
81 { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" },
82 { DMU_BSWAP_UINT8, FALSE, "zvol object" },
83 { DMU_BSWAP_ZAP, TRUE, "zvol prop" },
84 { DMU_BSWAP_UINT8, FALSE, "other uint8[]" },
85 { DMU_BSWAP_UINT64, FALSE, "other uint64[]" },
86 { DMU_BSWAP_ZAP, TRUE, "other ZAP" },
87 { DMU_BSWAP_ZAP, TRUE, "persistent error log" },
88 { DMU_BSWAP_UINT8, TRUE, "SPA history" },
89 { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" },
90 { DMU_BSWAP_ZAP, TRUE, "Pool properties" },
91 { DMU_BSWAP_ZAP, TRUE, "DSL permissions" },
92 { DMU_BSWAP_ACL, TRUE, "ZFS ACL" },
93 { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" },
94 { DMU_BSWAP_UINT8, TRUE, "FUID table" },
95 { DMU_BSWAP_UINT64, TRUE, "FUID table size" },
96 { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"},
97 { DMU_BSWAP_ZAP, TRUE, "scan work queue" },
98 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" },
99 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" },
100 { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"},
101 { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" },
102 { DMU_BSWAP_ZAP, TRUE, "DDT statistics" },
103 { DMU_BSWAP_UINT8, TRUE, "System attributes" },
104 { DMU_BSWAP_ZAP, TRUE, "SA master node" },
105 { DMU_BSWAP_ZAP, TRUE, "SA attr registration" },
106 { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" },
107 { DMU_BSWAP_ZAP, TRUE, "scan translations" },
108 { DMU_BSWAP_UINT8, FALSE, "deduplicated block" },
109 { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" },
110 { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" },
111 { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" },
112 { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" }
113 };
114
115 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
116 { byteswap_uint8_array, "uint8" },
117 { byteswap_uint16_array, "uint16" },
118 { byteswap_uint32_array, "uint32" },
119 { byteswap_uint64_array, "uint64" },
120 { zap_byteswap, "zap" },
121 { dnode_buf_byteswap, "dnode" },
122 { dmu_objset_byteswap, "objset" },
123 { zfs_znode_byteswap, "znode" },
124 { zfs_oldacl_byteswap, "oldacl" },
125 { zfs_acl_byteswap, "acl" }
126 };
127
128 int
129 dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
130 void *tag, dmu_buf_t **dbp)
131 {
132 dnode_t *dn;
133 uint64_t blkid;
134 dmu_buf_impl_t *db;
135 int err;
136
137 err = dnode_hold(os, object, FTAG, &dn);
138 if (err)
139 return (err);
140 blkid = dbuf_whichblock(dn, offset);
141 rw_enter(&dn->dn_struct_rwlock, RW_READER);
142 db = dbuf_hold(dn, blkid, tag);
143 rw_exit(&dn->dn_struct_rwlock);
144 dnode_rele(dn, FTAG);
145
146 if (db == NULL) {
147 *dbp = NULL;
148 return (SET_ERROR(EIO));
149 }
150
151 *dbp = &db->db;
152 return (err);
153 }
154
155 int
156 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
157 void *tag, dmu_buf_t **dbp, int flags)
158 {
159 int err;
160 int db_flags = DB_RF_CANFAIL;
161
162 if (flags & DMU_READ_NO_PREFETCH)
163 db_flags |= DB_RF_NOPREFETCH;
164
165 err = dmu_buf_hold_noread(os, object, offset, tag, dbp);
166 if (err == 0) {
167 dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
168 err = dbuf_read(db, NULL, db_flags);
169 if (err != 0) {
170 dbuf_rele(db, tag);
171 *dbp = NULL;
172 }
173 }
174
175 return (err);
176 }
177
178 int
179 dmu_bonus_max(void)
180 {
181 return (DN_MAX_BONUSLEN);
182 }
183
184 int
185 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
186 {
187 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
188 dnode_t *dn;
189 int error;
190
191 DB_DNODE_ENTER(db);
192 dn = DB_DNODE(db);
193
194 if (dn->dn_bonus != db) {
195 error = SET_ERROR(EINVAL);
196 } else if (newsize < 0 || newsize > db_fake->db_size) {
197 error = SET_ERROR(EINVAL);
198 } else {
199 dnode_setbonuslen(dn, newsize, tx);
200 error = 0;
201 }
202
203 DB_DNODE_EXIT(db);
204 return (error);
205 }
206
207 int
208 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
209 {
210 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
211 dnode_t *dn;
212 int error;
213
214 DB_DNODE_ENTER(db);
215 dn = DB_DNODE(db);
216
217 if (!DMU_OT_IS_VALID(type)) {
218 error = SET_ERROR(EINVAL);
219 } else if (dn->dn_bonus != db) {
220 error = SET_ERROR(EINVAL);
221 } else {
222 dnode_setbonus_type(dn, type, tx);
223 error = 0;
224 }
225
226 DB_DNODE_EXIT(db);
227 return (error);
228 }
229
230 dmu_object_type_t
231 dmu_get_bonustype(dmu_buf_t *db_fake)
232 {
233 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
234 dnode_t *dn;
235 dmu_object_type_t type;
236
237 DB_DNODE_ENTER(db);
238 dn = DB_DNODE(db);
239 type = dn->dn_bonustype;
240 DB_DNODE_EXIT(db);
241
242 return (type);
243 }
244
245 int
246 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
247 {
248 dnode_t *dn;
249 int error;
250
251 error = dnode_hold(os, object, FTAG, &dn);
252 dbuf_rm_spill(dn, tx);
253 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
254 dnode_rm_spill(dn, tx);
255 rw_exit(&dn->dn_struct_rwlock);
256 dnode_rele(dn, FTAG);
257 return (error);
258 }
259
260 /*
261 * returns ENOENT, EIO, or 0.
262 */
263 int
264 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
265 {
266 dnode_t *dn;
267 dmu_buf_impl_t *db;
268 int error;
269
270 error = dnode_hold(os, object, FTAG, &dn);
271 if (error)
272 return (error);
273
274 rw_enter(&dn->dn_struct_rwlock, RW_READER);
275 if (dn->dn_bonus == NULL) {
276 rw_exit(&dn->dn_struct_rwlock);
277 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
278 if (dn->dn_bonus == NULL)
279 dbuf_create_bonus(dn);
280 }
281 db = dn->dn_bonus;
282
283 /* as long as the bonus buf is held, the dnode will be held */
284 if (refcount_add(&db->db_holds, tag) == 1) {
285 VERIFY(dnode_add_ref(dn, db));
286 (void) atomic_inc_32_nv(&dn->dn_dbufs_count);
287 }
288
289 /*
290 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
291 * hold and incrementing the dbuf count to ensure that dnode_move() sees
292 * a dnode hold for every dbuf.
293 */
294 rw_exit(&dn->dn_struct_rwlock);
295
296 dnode_rele(dn, FTAG);
297
298 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
299
300 *dbp = &db->db;
301 return (0);
302 }
303
304 /*
305 * returns ENOENT, EIO, or 0.
306 *
307 * This interface will allocate a blank spill dbuf when a spill blk
308 * doesn't already exist on the dnode.
309 *
310 * if you only want to find an already existing spill db, then
311 * dmu_spill_hold_existing() should be used.
312 */
313 int
314 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
315 {
316 dmu_buf_impl_t *db = NULL;
317 int err;
318
319 if ((flags & DB_RF_HAVESTRUCT) == 0)
320 rw_enter(&dn->dn_struct_rwlock, RW_READER);
321
322 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
323
324 if ((flags & DB_RF_HAVESTRUCT) == 0)
325 rw_exit(&dn->dn_struct_rwlock);
326
327 ASSERT(db != NULL);
328 err = dbuf_read(db, NULL, flags);
329 if (err == 0)
330 *dbp = &db->db;
331 else
332 dbuf_rele(db, tag);
333 return (err);
334 }
335
336 int
337 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
338 {
339 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
340 dnode_t *dn;
341 int err;
342
343 DB_DNODE_ENTER(db);
344 dn = DB_DNODE(db);
345
346 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
347 err = SET_ERROR(EINVAL);
348 } else {
349 rw_enter(&dn->dn_struct_rwlock, RW_READER);
350
351 if (!dn->dn_have_spill) {
352 err = SET_ERROR(ENOENT);
353 } else {
354 err = dmu_spill_hold_by_dnode(dn,
355 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
356 }
357
358 rw_exit(&dn->dn_struct_rwlock);
359 }
360
361 DB_DNODE_EXIT(db);
362 return (err);
363 }
364
365 int
366 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
367 {
368 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
369 dnode_t *dn;
370 int err;
371
372 DB_DNODE_ENTER(db);
373 dn = DB_DNODE(db);
374 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
375 DB_DNODE_EXIT(db);
376
377 return (err);
378 }
379
380 /*
381 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
382 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
383 * and can induce severe lock contention when writing to several files
384 * whose dnodes are in the same block.
385 */
386 static int
387 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
388 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
389 {
390 dmu_buf_t **dbp;
391 uint64_t blkid, nblks, i;
392 uint32_t dbuf_flags;
393 int err;
394 zio_t *zio;
395
396 ASSERT(length <= DMU_MAX_ACCESS);
397
398 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
399 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
400 dbuf_flags |= DB_RF_NOPREFETCH;
401
402 rw_enter(&dn->dn_struct_rwlock, RW_READER);
403 if (dn->dn_datablkshift) {
404 int blkshift = dn->dn_datablkshift;
405 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
406 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
407 } else {
408 if (offset + length > dn->dn_datablksz) {
409 zfs_panic_recover("zfs: accessing past end of object "
410 "%llx/%llx (size=%u access=%llu+%llu)",
411 (longlong_t)dn->dn_objset->
412 os_dsl_dataset->ds_object,
413 (longlong_t)dn->dn_object, dn->dn_datablksz,
414 (longlong_t)offset, (longlong_t)length);
415 rw_exit(&dn->dn_struct_rwlock);
416 return (SET_ERROR(EIO));
417 }
418 nblks = 1;
419 }
420 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks,
421 KM_PUSHPAGE | KM_NODEBUG);
422
423 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
424 blkid = dbuf_whichblock(dn, offset);
425 for (i = 0; i < nblks; i++) {
426 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
427 if (db == NULL) {
428 rw_exit(&dn->dn_struct_rwlock);
429 dmu_buf_rele_array(dbp, nblks, tag);
430 zio_nowait(zio);
431 return (SET_ERROR(EIO));
432 }
433 /* initiate async i/o */
434 if (read) {
435 (void) dbuf_read(db, zio, dbuf_flags);
436 }
437 dbp[i] = &db->db;
438 }
439 rw_exit(&dn->dn_struct_rwlock);
440
441 /* wait for async i/o */
442 err = zio_wait(zio);
443 if (err) {
444 dmu_buf_rele_array(dbp, nblks, tag);
445 return (err);
446 }
447
448 /* wait for other io to complete */
449 if (read) {
450 for (i = 0; i < nblks; i++) {
451 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
452 mutex_enter(&db->db_mtx);
453 while (db->db_state == DB_READ ||
454 db->db_state == DB_FILL)
455 cv_wait(&db->db_changed, &db->db_mtx);
456 if (db->db_state == DB_UNCACHED)
457 err = SET_ERROR(EIO);
458 mutex_exit(&db->db_mtx);
459 if (err) {
460 dmu_buf_rele_array(dbp, nblks, tag);
461 return (err);
462 }
463 }
464 }
465
466 *numbufsp = nblks;
467 *dbpp = dbp;
468 return (0);
469 }
470
471 static int
472 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
473 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
474 {
475 dnode_t *dn;
476 int err;
477
478 err = dnode_hold(os, object, FTAG, &dn);
479 if (err)
480 return (err);
481
482 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
483 numbufsp, dbpp, DMU_READ_PREFETCH);
484
485 dnode_rele(dn, FTAG);
486
487 return (err);
488 }
489
490 int
491 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
492 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
493 {
494 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
495 dnode_t *dn;
496 int err;
497
498 DB_DNODE_ENTER(db);
499 dn = DB_DNODE(db);
500 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
501 numbufsp, dbpp, DMU_READ_PREFETCH);
502 DB_DNODE_EXIT(db);
503
504 return (err);
505 }
506
507 void
508 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
509 {
510 int i;
511 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
512
513 if (numbufs == 0)
514 return;
515
516 for (i = 0; i < numbufs; i++) {
517 if (dbp[i])
518 dbuf_rele(dbp[i], tag);
519 }
520
521 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
522 }
523
524 /*
525 * Issue prefetch i/os for the given blocks.
526 *
527 * Note: The assumption is that we *know* these blocks will be needed
528 * almost immediately. Therefore, the prefetch i/os will be issued at
529 * ZIO_PRIORITY_SYNC_READ
530 *
531 * Note: indirect blocks and other metadata will be read synchronously,
532 * causing this function to block if they are not already cached.
533 */
534 void
535 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
536 {
537 dnode_t *dn;
538 uint64_t blkid;
539 int nblks, err;
540
541 if (zfs_prefetch_disable)
542 return;
543
544 if (len == 0) { /* they're interested in the bonus buffer */
545 dn = DMU_META_DNODE(os);
546
547 if (object == 0 || object >= DN_MAX_OBJECT)
548 return;
549
550 rw_enter(&dn->dn_struct_rwlock, RW_READER);
551 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
552 dbuf_prefetch(dn, blkid, ZIO_PRIORITY_SYNC_READ);
553 rw_exit(&dn->dn_struct_rwlock);
554 return;
555 }
556
557 /*
558 * XXX - Note, if the dnode for the requested object is not
559 * already cached, we will do a *synchronous* read in the
560 * dnode_hold() call. The same is true for any indirects.
561 */
562 err = dnode_hold(os, object, FTAG, &dn);
563 if (err != 0)
564 return;
565
566 rw_enter(&dn->dn_struct_rwlock, RW_READER);
567 if (dn->dn_datablkshift) {
568 int blkshift = dn->dn_datablkshift;
569 nblks = (P2ROUNDUP(offset + len, 1 << blkshift) -
570 P2ALIGN(offset, 1 << blkshift)) >> blkshift;
571 } else {
572 nblks = (offset < dn->dn_datablksz);
573 }
574
575 if (nblks != 0) {
576 int i;
577
578 blkid = dbuf_whichblock(dn, offset);
579 for (i = 0; i < nblks; i++)
580 dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_SYNC_READ);
581 }
582
583 rw_exit(&dn->dn_struct_rwlock);
584
585 dnode_rele(dn, FTAG);
586 }
587
588 /*
589 * Get the next "chunk" of file data to free. We traverse the file from
590 * the end so that the file gets shorter over time (if we crashes in the
591 * middle, this will leave us in a better state). We find allocated file
592 * data by simply searching the allocated level 1 indirects.
593 *
594 * On input, *start should be the first offset that does not need to be
595 * freed (e.g. "offset + length"). On return, *start will be the first
596 * offset that should be freed.
597 */
598 static int
599 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum)
600 {
601 uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
602 /* bytes of data covered by a level-1 indirect block */
603 uint64_t iblkrange =
604 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
605 uint64_t blks;
606
607 ASSERT3U(minimum, <=, *start);
608
609 if (*start - minimum <= iblkrange * maxblks) {
610 *start = minimum;
611 return (0);
612 }
613 ASSERT(ISP2(iblkrange));
614
615 for (blks = 0; *start > minimum && blks < maxblks; blks++) {
616 int err;
617
618 /*
619 * dnode_next_offset(BACKWARDS) will find an allocated L1
620 * indirect block at or before the input offset. We must
621 * decrement *start so that it is at the end of the region
622 * to search.
623 */
624 (*start)--;
625 err = dnode_next_offset(dn,
626 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
627
628 /* if there are no indirect blocks before start, we are done */
629 if (err == ESRCH) {
630 *start = minimum;
631 break;
632 } else if (err != 0) {
633 return (err);
634 }
635
636 /* set start to the beginning of this L1 indirect */
637 *start = P2ALIGN(*start, iblkrange);
638 }
639 if (*start < minimum)
640 *start = minimum;
641 return (0);
642 }
643
644 static int
645 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
646 uint64_t length)
647 {
648 uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
649 int err;
650
651 if (offset >= object_size)
652 return (0);
653
654 if (length == DMU_OBJECT_END || offset + length > object_size)
655 length = object_size - offset;
656
657 while (length != 0) {
658 uint64_t chunk_end, chunk_begin;
659 dmu_tx_t *tx;
660
661 chunk_end = chunk_begin = offset + length;
662
663 /* move chunk_begin backwards to the beginning of this chunk */
664 err = get_next_chunk(dn, &chunk_begin, offset);
665 if (err)
666 return (err);
667 ASSERT3U(chunk_begin, >=, offset);
668 ASSERT3U(chunk_begin, <=, chunk_end);
669
670 tx = dmu_tx_create(os);
671 dmu_tx_hold_free(tx, dn->dn_object,
672 chunk_begin, chunk_end - chunk_begin);
673 err = dmu_tx_assign(tx, TXG_WAIT);
674 if (err) {
675 dmu_tx_abort(tx);
676 return (err);
677 }
678 dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx);
679 dmu_tx_commit(tx);
680
681 length -= chunk_end - chunk_begin;
682 }
683 return (0);
684 }
685
686 int
687 dmu_free_long_range(objset_t *os, uint64_t object,
688 uint64_t offset, uint64_t length)
689 {
690 dnode_t *dn;
691 int err;
692
693 err = dnode_hold(os, object, FTAG, &dn);
694 if (err != 0)
695 return (err);
696 err = dmu_free_long_range_impl(os, dn, offset, length);
697
698 /*
699 * It is important to zero out the maxblkid when freeing the entire
700 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
701 * will take the fast path, and (b) dnode_reallocate() can verify
702 * that the entire file has been freed.
703 */
704 if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
705 dn->dn_maxblkid = 0;
706
707 dnode_rele(dn, FTAG);
708 return (err);
709 }
710
711 int
712 dmu_free_long_object(objset_t *os, uint64_t object)
713 {
714 dmu_tx_t *tx;
715 int err;
716
717 err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
718 if (err != 0)
719 return (err);
720
721 tx = dmu_tx_create(os);
722 dmu_tx_hold_bonus(tx, object);
723 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
724 err = dmu_tx_assign(tx, TXG_WAIT);
725 if (err == 0) {
726 err = dmu_object_free(os, object, tx);
727 dmu_tx_commit(tx);
728 } else {
729 dmu_tx_abort(tx);
730 }
731
732 return (err);
733 }
734
735 int
736 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
737 uint64_t size, dmu_tx_t *tx)
738 {
739 dnode_t *dn;
740 int err = dnode_hold(os, object, FTAG, &dn);
741 if (err)
742 return (err);
743 ASSERT(offset < UINT64_MAX);
744 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
745 dnode_free_range(dn, offset, size, tx);
746 dnode_rele(dn, FTAG);
747 return (0);
748 }
749
750 int
751 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
752 void *buf, uint32_t flags)
753 {
754 dnode_t *dn;
755 dmu_buf_t **dbp;
756 int numbufs, err;
757
758 err = dnode_hold(os, object, FTAG, &dn);
759 if (err)
760 return (err);
761
762 /*
763 * Deal with odd block sizes, where there can't be data past the first
764 * block. If we ever do the tail block optimization, we will need to
765 * handle that here as well.
766 */
767 if (dn->dn_maxblkid == 0) {
768 int newsz = offset > dn->dn_datablksz ? 0 :
769 MIN(size, dn->dn_datablksz - offset);
770 bzero((char *)buf + newsz, size - newsz);
771 size = newsz;
772 }
773
774 while (size > 0) {
775 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
776 int i;
777
778 /*
779 * NB: we could do this block-at-a-time, but it's nice
780 * to be reading in parallel.
781 */
782 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
783 TRUE, FTAG, &numbufs, &dbp, flags);
784 if (err)
785 break;
786
787 for (i = 0; i < numbufs; i++) {
788 int tocpy;
789 int bufoff;
790 dmu_buf_t *db = dbp[i];
791
792 ASSERT(size > 0);
793
794 bufoff = offset - db->db_offset;
795 tocpy = (int)MIN(db->db_size - bufoff, size);
796
797 bcopy((char *)db->db_data + bufoff, buf, tocpy);
798
799 offset += tocpy;
800 size -= tocpy;
801 buf = (char *)buf + tocpy;
802 }
803 dmu_buf_rele_array(dbp, numbufs, FTAG);
804 }
805 dnode_rele(dn, FTAG);
806 return (err);
807 }
808
809 void
810 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
811 const void *buf, dmu_tx_t *tx)
812 {
813 dmu_buf_t **dbp;
814 int numbufs, i;
815
816 if (size == 0)
817 return;
818
819 VERIFY0(dmu_buf_hold_array(os, object, offset, size,
820 FALSE, FTAG, &numbufs, &dbp));
821
822 for (i = 0; i < numbufs; i++) {
823 int tocpy;
824 int bufoff;
825 dmu_buf_t *db = dbp[i];
826
827 ASSERT(size > 0);
828
829 bufoff = offset - db->db_offset;
830 tocpy = (int)MIN(db->db_size - bufoff, size);
831
832 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
833
834 if (tocpy == db->db_size)
835 dmu_buf_will_fill(db, tx);
836 else
837 dmu_buf_will_dirty(db, tx);
838
839 (void) memcpy((char *)db->db_data + bufoff, buf, tocpy);
840
841 if (tocpy == db->db_size)
842 dmu_buf_fill_done(db, tx);
843
844 offset += tocpy;
845 size -= tocpy;
846 buf = (char *)buf + tocpy;
847 }
848 dmu_buf_rele_array(dbp, numbufs, FTAG);
849 }
850
851 void
852 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
853 dmu_tx_t *tx)
854 {
855 dmu_buf_t **dbp;
856 int numbufs, i;
857
858 if (size == 0)
859 return;
860
861 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
862 FALSE, FTAG, &numbufs, &dbp));
863
864 for (i = 0; i < numbufs; i++) {
865 dmu_buf_t *db = dbp[i];
866
867 dmu_buf_will_not_fill(db, tx);
868 }
869 dmu_buf_rele_array(dbp, numbufs, FTAG);
870 }
871
872 void
873 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
874 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
875 int compressed_size, int byteorder, dmu_tx_t *tx)
876 {
877 dmu_buf_t *db;
878
879 ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
880 ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
881 VERIFY0(dmu_buf_hold_noread(os, object, offset,
882 FTAG, &db));
883
884 dmu_buf_write_embedded(db,
885 data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
886 uncompressed_size, compressed_size, byteorder, tx);
887
888 dmu_buf_rele(db, FTAG);
889 }
890
891 /*
892 * DMU support for xuio
893 */
894 kstat_t *xuio_ksp = NULL;
895
896 typedef struct xuio_stats {
897 /* loaned yet not returned arc_buf */
898 kstat_named_t xuiostat_onloan_rbuf;
899 kstat_named_t xuiostat_onloan_wbuf;
900 /* whether a copy is made when loaning out a read buffer */
901 kstat_named_t xuiostat_rbuf_copied;
902 kstat_named_t xuiostat_rbuf_nocopy;
903 /* whether a copy is made when assigning a write buffer */
904 kstat_named_t xuiostat_wbuf_copied;
905 kstat_named_t xuiostat_wbuf_nocopy;
906 } xuio_stats_t;
907
908 static xuio_stats_t xuio_stats = {
909 { "onloan_read_buf", KSTAT_DATA_UINT64 },
910 { "onloan_write_buf", KSTAT_DATA_UINT64 },
911 { "read_buf_copied", KSTAT_DATA_UINT64 },
912 { "read_buf_nocopy", KSTAT_DATA_UINT64 },
913 { "write_buf_copied", KSTAT_DATA_UINT64 },
914 { "write_buf_nocopy", KSTAT_DATA_UINT64 }
915 };
916
917 #define XUIOSTAT_INCR(stat, val) \
918 atomic_add_64(&xuio_stats.stat.value.ui64, (val))
919 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
920
921 int
922 dmu_xuio_init(xuio_t *xuio, int nblk)
923 {
924 dmu_xuio_t *priv;
925 uio_t *uio = &xuio->xu_uio;
926
927 uio->uio_iovcnt = nblk;
928 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_PUSHPAGE);
929
930 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_PUSHPAGE);
931 priv->cnt = nblk;
932 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_PUSHPAGE);
933 priv->iovp = uio->uio_iov;
934 XUIO_XUZC_PRIV(xuio) = priv;
935
936 if (XUIO_XUZC_RW(xuio) == UIO_READ)
937 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
938 else
939 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
940
941 return (0);
942 }
943
944 void
945 dmu_xuio_fini(xuio_t *xuio)
946 {
947 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
948 int nblk = priv->cnt;
949
950 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
951 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
952 kmem_free(priv, sizeof (dmu_xuio_t));
953
954 if (XUIO_XUZC_RW(xuio) == UIO_READ)
955 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
956 else
957 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
958 }
959
960 /*
961 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
962 * and increase priv->next by 1.
963 */
964 int
965 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
966 {
967 struct iovec *iov;
968 uio_t *uio = &xuio->xu_uio;
969 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
970 int i = priv->next++;
971
972 ASSERT(i < priv->cnt);
973 ASSERT(off + n <= arc_buf_size(abuf));
974 iov = uio->uio_iov + i;
975 iov->iov_base = (char *)abuf->b_data + off;
976 iov->iov_len = n;
977 priv->bufs[i] = abuf;
978 return (0);
979 }
980
981 int
982 dmu_xuio_cnt(xuio_t *xuio)
983 {
984 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
985 return (priv->cnt);
986 }
987
988 arc_buf_t *
989 dmu_xuio_arcbuf(xuio_t *xuio, int i)
990 {
991 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
992
993 ASSERT(i < priv->cnt);
994 return (priv->bufs[i]);
995 }
996
997 void
998 dmu_xuio_clear(xuio_t *xuio, int i)
999 {
1000 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1001
1002 ASSERT(i < priv->cnt);
1003 priv->bufs[i] = NULL;
1004 }
1005
1006 static void
1007 xuio_stat_init(void)
1008 {
1009 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
1010 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
1011 KSTAT_FLAG_VIRTUAL);
1012 if (xuio_ksp != NULL) {
1013 xuio_ksp->ks_data = &xuio_stats;
1014 kstat_install(xuio_ksp);
1015 }
1016 }
1017
1018 static void
1019 xuio_stat_fini(void)
1020 {
1021 if (xuio_ksp != NULL) {
1022 kstat_delete(xuio_ksp);
1023 xuio_ksp = NULL;
1024 }
1025 }
1026
1027 void
1028 xuio_stat_wbuf_copied()
1029 {
1030 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1031 }
1032
1033 void
1034 xuio_stat_wbuf_nocopy()
1035 {
1036 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1037 }
1038
1039 #ifdef _KERNEL
1040
1041 /*
1042 * Copy up to size bytes between arg_buf and req based on the data direction
1043 * described by the req. If an entire req's data cannot be transfered in one
1044 * pass, you should pass in @req_offset to indicate where to continue. The
1045 * return value is the number of bytes successfully copied to arg_buf.
1046 */
1047 static int
1048 dmu_req_copy(void *arg_buf, int size, struct request *req, size_t req_offset)
1049 {
1050 struct bio_vec bv, *bvp;
1051 struct req_iterator iter;
1052 char *bv_buf;
1053 int tocpy, bv_len, bv_offset;
1054 int offset = 0;
1055
1056 rq_for_each_segment4(bv, bvp, req, iter) {
1057 /*
1058 * Fully consumed the passed arg_buf. We use goto here because
1059 * rq_for_each_segment is a double loop
1060 */
1061 ASSERT3S(offset, <=, size);
1062 if (size == offset)
1063 goto out;
1064
1065 /* Skip already copied bv */
1066 if (req_offset >= bv.bv_len) {
1067 req_offset -= bv.bv_len;
1068 continue;
1069 }
1070
1071 bv_len = bv.bv_len - req_offset;
1072 bv_offset = bv.bv_offset + req_offset;
1073 req_offset = 0;
1074
1075 tocpy = MIN(bv_len, size - offset);
1076 ASSERT3S(tocpy, >=, 0);
1077
1078 bv_buf = page_address(bv.bv_page) + bv_offset;
1079 ASSERT3P(bv_buf, !=, NULL);
1080
1081 if (rq_data_dir(req) == WRITE)
1082 memcpy(arg_buf + offset, bv_buf, tocpy);
1083 else
1084 memcpy(bv_buf, arg_buf + offset, tocpy);
1085
1086 offset += tocpy;
1087 }
1088 out:
1089 return (offset);
1090 }
1091
1092 int
1093 dmu_read_req(objset_t *os, uint64_t object, struct request *req)
1094 {
1095 uint64_t size = blk_rq_bytes(req);
1096 uint64_t offset = blk_rq_pos(req) << 9;
1097 dmu_buf_t **dbp;
1098 int numbufs, i, err;
1099 size_t req_offset;
1100
1101 /*
1102 * NB: we could do this block-at-a-time, but it's nice
1103 * to be reading in parallel.
1104 */
1105 err = dmu_buf_hold_array(os, object, offset, size, TRUE, FTAG,
1106 &numbufs, &dbp);
1107 if (err)
1108 return (err);
1109
1110 req_offset = 0;
1111 for (i = 0; i < numbufs; i++) {
1112 int tocpy, didcpy, bufoff;
1113 dmu_buf_t *db = dbp[i];
1114
1115 bufoff = offset - db->db_offset;
1116 ASSERT3S(bufoff, >=, 0);
1117
1118 tocpy = (int)MIN(db->db_size - bufoff, size);
1119 if (tocpy == 0)
1120 break;
1121
1122 didcpy = dmu_req_copy(db->db_data + bufoff, tocpy, req,
1123 req_offset);
1124
1125 if (didcpy < tocpy)
1126 err = EIO;
1127
1128 if (err)
1129 break;
1130
1131 size -= tocpy;
1132 offset += didcpy;
1133 req_offset += didcpy;
1134 err = 0;
1135 }
1136 dmu_buf_rele_array(dbp, numbufs, FTAG);
1137
1138 return (err);
1139 }
1140
1141 int
1142 dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx)
1143 {
1144 uint64_t size = blk_rq_bytes(req);
1145 uint64_t offset = blk_rq_pos(req) << 9;
1146 dmu_buf_t **dbp;
1147 int numbufs, i, err;
1148 size_t req_offset;
1149
1150 if (size == 0)
1151 return (0);
1152
1153 err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG,
1154 &numbufs, &dbp);
1155 if (err)
1156 return (err);
1157
1158 req_offset = 0;
1159 for (i = 0; i < numbufs; i++) {
1160 int tocpy, didcpy, bufoff;
1161 dmu_buf_t *db = dbp[i];
1162
1163 bufoff = offset - db->db_offset;
1164 ASSERT3S(bufoff, >=, 0);
1165
1166 tocpy = (int)MIN(db->db_size - bufoff, size);
1167 if (tocpy == 0)
1168 break;
1169
1170 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1171
1172 if (tocpy == db->db_size)
1173 dmu_buf_will_fill(db, tx);
1174 else
1175 dmu_buf_will_dirty(db, tx);
1176
1177 didcpy = dmu_req_copy(db->db_data + bufoff, tocpy, req,
1178 req_offset);
1179
1180 if (tocpy == db->db_size)
1181 dmu_buf_fill_done(db, tx);
1182
1183 if (didcpy < tocpy)
1184 err = EIO;
1185
1186 if (err)
1187 break;
1188
1189 size -= tocpy;
1190 offset += didcpy;
1191 req_offset += didcpy;
1192 err = 0;
1193 }
1194
1195 dmu_buf_rele_array(dbp, numbufs, FTAG);
1196 return (err);
1197 }
1198
1199 int
1200 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1201 {
1202 dmu_buf_t **dbp;
1203 int numbufs, i, err;
1204 xuio_t *xuio = NULL;
1205
1206 /*
1207 * NB: we could do this block-at-a-time, but it's nice
1208 * to be reading in parallel.
1209 */
1210 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
1211 &numbufs, &dbp);
1212 if (err)
1213 return (err);
1214
1215 for (i = 0; i < numbufs; i++) {
1216 int tocpy;
1217 int bufoff;
1218 dmu_buf_t *db = dbp[i];
1219
1220 ASSERT(size > 0);
1221
1222 bufoff = uio->uio_loffset - db->db_offset;
1223 tocpy = (int)MIN(db->db_size - bufoff, size);
1224
1225 if (xuio) {
1226 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1227 arc_buf_t *dbuf_abuf = dbi->db_buf;
1228 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1229 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1230 if (!err) {
1231 uio->uio_resid -= tocpy;
1232 uio->uio_loffset += tocpy;
1233 }
1234
1235 if (abuf == dbuf_abuf)
1236 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1237 else
1238 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1239 } else {
1240 err = uiomove((char *)db->db_data + bufoff, tocpy,
1241 UIO_READ, uio);
1242 }
1243 if (err)
1244 break;
1245
1246 size -= tocpy;
1247 }
1248 dmu_buf_rele_array(dbp, numbufs, FTAG);
1249
1250 return (err);
1251 }
1252
1253 static int
1254 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1255 {
1256 dmu_buf_t **dbp;
1257 int numbufs;
1258 int err = 0;
1259 int i;
1260
1261 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1262 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1263 if (err)
1264 return (err);
1265
1266 for (i = 0; i < numbufs; i++) {
1267 int tocpy;
1268 int bufoff;
1269 dmu_buf_t *db = dbp[i];
1270
1271 ASSERT(size > 0);
1272
1273 bufoff = uio->uio_loffset - db->db_offset;
1274 tocpy = (int)MIN(db->db_size - bufoff, size);
1275
1276 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1277
1278 if (tocpy == db->db_size)
1279 dmu_buf_will_fill(db, tx);
1280 else
1281 dmu_buf_will_dirty(db, tx);
1282
1283 /*
1284 * XXX uiomove could block forever (eg.nfs-backed
1285 * pages). There needs to be a uiolockdown() function
1286 * to lock the pages in memory, so that uiomove won't
1287 * block.
1288 */
1289 err = uiomove((char *)db->db_data + bufoff, tocpy,
1290 UIO_WRITE, uio);
1291
1292 if (tocpy == db->db_size)
1293 dmu_buf_fill_done(db, tx);
1294
1295 if (err)
1296 break;
1297
1298 size -= tocpy;
1299 }
1300
1301 dmu_buf_rele_array(dbp, numbufs, FTAG);
1302 return (err);
1303 }
1304
1305 int
1306 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1307 dmu_tx_t *tx)
1308 {
1309 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1310 dnode_t *dn;
1311 int err;
1312
1313 if (size == 0)
1314 return (0);
1315
1316 DB_DNODE_ENTER(db);
1317 dn = DB_DNODE(db);
1318 err = dmu_write_uio_dnode(dn, uio, size, tx);
1319 DB_DNODE_EXIT(db);
1320
1321 return (err);
1322 }
1323
1324 int
1325 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1326 dmu_tx_t *tx)
1327 {
1328 dnode_t *dn;
1329 int err;
1330
1331 if (size == 0)
1332 return (0);
1333
1334 err = dnode_hold(os, object, FTAG, &dn);
1335 if (err)
1336 return (err);
1337
1338 err = dmu_write_uio_dnode(dn, uio, size, tx);
1339
1340 dnode_rele(dn, FTAG);
1341
1342 return (err);
1343 }
1344 #endif /* _KERNEL */
1345
1346 /*
1347 * Allocate a loaned anonymous arc buffer.
1348 */
1349 arc_buf_t *
1350 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1351 {
1352 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1353
1354 return (arc_loan_buf(db->db_objset->os_spa, size));
1355 }
1356
1357 /*
1358 * Free a loaned arc buffer.
1359 */
1360 void
1361 dmu_return_arcbuf(arc_buf_t *buf)
1362 {
1363 arc_return_buf(buf, FTAG);
1364 VERIFY(arc_buf_remove_ref(buf, FTAG));
1365 }
1366
1367 /*
1368 * When possible directly assign passed loaned arc buffer to a dbuf.
1369 * If this is not possible copy the contents of passed arc buf via
1370 * dmu_write().
1371 */
1372 void
1373 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1374 dmu_tx_t *tx)
1375 {
1376 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1377 dnode_t *dn;
1378 dmu_buf_impl_t *db;
1379 uint32_t blksz = (uint32_t)arc_buf_size(buf);
1380 uint64_t blkid;
1381
1382 DB_DNODE_ENTER(dbuf);
1383 dn = DB_DNODE(dbuf);
1384 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1385 blkid = dbuf_whichblock(dn, offset);
1386 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1387 rw_exit(&dn->dn_struct_rwlock);
1388 DB_DNODE_EXIT(dbuf);
1389
1390 /*
1391 * We can only assign if the offset is aligned, the arc buf is the
1392 * same size as the dbuf, and the dbuf is not metadata. It
1393 * can't be metadata because the loaned arc buf comes from the
1394 * user-data kmem area.
1395 */
1396 if (offset == db->db.db_offset && blksz == db->db.db_size &&
1397 DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) {
1398 dbuf_assign_arcbuf(db, buf, tx);
1399 dbuf_rele(db, FTAG);
1400 } else {
1401 objset_t *os;
1402 uint64_t object;
1403
1404 DB_DNODE_ENTER(dbuf);
1405 dn = DB_DNODE(dbuf);
1406 os = dn->dn_objset;
1407 object = dn->dn_object;
1408 DB_DNODE_EXIT(dbuf);
1409
1410 dbuf_rele(db, FTAG);
1411 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1412 dmu_return_arcbuf(buf);
1413 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1414 }
1415 }
1416
1417 typedef struct {
1418 dbuf_dirty_record_t *dsa_dr;
1419 dmu_sync_cb_t *dsa_done;
1420 zgd_t *dsa_zgd;
1421 dmu_tx_t *dsa_tx;
1422 } dmu_sync_arg_t;
1423
1424 /* ARGSUSED */
1425 static void
1426 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1427 {
1428 dmu_sync_arg_t *dsa = varg;
1429 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1430 blkptr_t *bp = zio->io_bp;
1431
1432 if (zio->io_error == 0) {
1433 if (BP_IS_HOLE(bp)) {
1434 /*
1435 * A block of zeros may compress to a hole, but the
1436 * block size still needs to be known for replay.
1437 */
1438 BP_SET_LSIZE(bp, db->db_size);
1439 } else if (!BP_IS_EMBEDDED(bp)) {
1440 ASSERT(BP_GET_LEVEL(bp) == 0);
1441 bp->blk_fill = 1;
1442 }
1443 }
1444 }
1445
1446 static void
1447 dmu_sync_late_arrival_ready(zio_t *zio)
1448 {
1449 dmu_sync_ready(zio, NULL, zio->io_private);
1450 }
1451
1452 /* ARGSUSED */
1453 static void
1454 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1455 {
1456 dmu_sync_arg_t *dsa = varg;
1457 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1458 dmu_buf_impl_t *db = dr->dr_dbuf;
1459
1460 mutex_enter(&db->db_mtx);
1461 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1462 if (zio->io_error == 0) {
1463 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1464 if (dr->dt.dl.dr_nopwrite) {
1465 ASSERTV(blkptr_t *bp = zio->io_bp);
1466 ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
1467 ASSERTV(uint8_t chksum = BP_GET_CHECKSUM(bp_orig));
1468
1469 ASSERT(BP_EQUAL(bp, bp_orig));
1470 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1471 ASSERT(zio_checksum_table[chksum].ci_dedup);
1472 }
1473 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1474 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1475 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1476 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
1477 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1478 } else {
1479 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1480 }
1481 cv_broadcast(&db->db_changed);
1482 mutex_exit(&db->db_mtx);
1483
1484 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1485
1486 kmem_free(dsa, sizeof (*dsa));
1487 }
1488
1489 static void
1490 dmu_sync_late_arrival_done(zio_t *zio)
1491 {
1492 blkptr_t *bp = zio->io_bp;
1493 dmu_sync_arg_t *dsa = zio->io_private;
1494 ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
1495
1496 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1497 /*
1498 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1499 * then there is nothing to do here. Otherwise, free the
1500 * newly allocated block in this txg.
1501 */
1502 if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
1503 ASSERT(BP_EQUAL(bp, bp_orig));
1504 } else {
1505 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1506 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1507 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1508 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1509 }
1510 }
1511
1512 dmu_tx_commit(dsa->dsa_tx);
1513
1514 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1515
1516 kmem_free(dsa, sizeof (*dsa));
1517 }
1518
1519 static int
1520 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1521 zio_prop_t *zp, zbookmark_phys_t *zb)
1522 {
1523 dmu_sync_arg_t *dsa;
1524 dmu_tx_t *tx;
1525
1526 tx = dmu_tx_create(os);
1527 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1528 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1529 dmu_tx_abort(tx);
1530 /* Make zl_get_data do txg_waited_synced() */
1531 return (SET_ERROR(EIO));
1532 }
1533
1534 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE);
1535 dsa->dsa_dr = NULL;
1536 dsa->dsa_done = done;
1537 dsa->dsa_zgd = zgd;
1538 dsa->dsa_tx = tx;
1539
1540 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1541 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
1542 dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa,
1543 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL|ZIO_FLAG_FASTWRITE, zb));
1544
1545 return (0);
1546 }
1547
1548 /*
1549 * Intent log support: sync the block associated with db to disk.
1550 * N.B. and XXX: the caller is responsible for making sure that the
1551 * data isn't changing while dmu_sync() is writing it.
1552 *
1553 * Return values:
1554 *
1555 * EEXIST: this txg has already been synced, so there's nothing to do.
1556 * The caller should not log the write.
1557 *
1558 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1559 * The caller should not log the write.
1560 *
1561 * EALREADY: this block is already in the process of being synced.
1562 * The caller should track its progress (somehow).
1563 *
1564 * EIO: could not do the I/O.
1565 * The caller should do a txg_wait_synced().
1566 *
1567 * 0: the I/O has been initiated.
1568 * The caller should log this blkptr in the done callback.
1569 * It is possible that the I/O will fail, in which case
1570 * the error will be reported to the done callback and
1571 * propagated to pio from zio_done().
1572 */
1573 int
1574 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1575 {
1576 blkptr_t *bp = zgd->zgd_bp;
1577 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1578 objset_t *os = db->db_objset;
1579 dsl_dataset_t *ds = os->os_dsl_dataset;
1580 dbuf_dirty_record_t *dr;
1581 dmu_sync_arg_t *dsa;
1582 zbookmark_phys_t zb;
1583 zio_prop_t zp;
1584 dnode_t *dn;
1585
1586 ASSERT(pio != NULL);
1587 ASSERT(txg != 0);
1588
1589 SET_BOOKMARK(&zb, ds->ds_object,
1590 db->db.db_object, db->db_level, db->db_blkid);
1591
1592 DB_DNODE_ENTER(db);
1593 dn = DB_DNODE(db);
1594 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1595 DB_DNODE_EXIT(db);
1596
1597 /*
1598 * If we're frozen (running ziltest), we always need to generate a bp.
1599 */
1600 if (txg > spa_freeze_txg(os->os_spa))
1601 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1602
1603 /*
1604 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1605 * and us. If we determine that this txg is not yet syncing,
1606 * but it begins to sync a moment later, that's OK because the
1607 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1608 */
1609 mutex_enter(&db->db_mtx);
1610
1611 if (txg <= spa_last_synced_txg(os->os_spa)) {
1612 /*
1613 * This txg has already synced. There's nothing to do.
1614 */
1615 mutex_exit(&db->db_mtx);
1616 return (SET_ERROR(EEXIST));
1617 }
1618
1619 if (txg <= spa_syncing_txg(os->os_spa)) {
1620 /*
1621 * This txg is currently syncing, so we can't mess with
1622 * the dirty record anymore; just write a new log block.
1623 */
1624 mutex_exit(&db->db_mtx);
1625 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1626 }
1627
1628 dr = db->db_last_dirty;
1629 while (dr && dr->dr_txg != txg)
1630 dr = dr->dr_next;
1631
1632 if (dr == NULL) {
1633 /*
1634 * There's no dr for this dbuf, so it must have been freed.
1635 * There's no need to log writes to freed blocks, so we're done.
1636 */
1637 mutex_exit(&db->db_mtx);
1638 return (SET_ERROR(ENOENT));
1639 }
1640
1641 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
1642
1643 /*
1644 * Assume the on-disk data is X, the current syncing data is Y,
1645 * and the current in-memory data is Z (currently in dmu_sync).
1646 * X and Z are identical but Y is has been modified. Normally,
1647 * when X and Z are the same we will perform a nopwrite but if Y
1648 * is different we must disable nopwrite since the resulting write
1649 * of Y to disk can free the block containing X. If we allowed a
1650 * nopwrite to occur the block pointing to Z would reference a freed
1651 * block. Since this is a rare case we simplify this by disabling
1652 * nopwrite if the current dmu_sync-ing dbuf has been modified in
1653 * a previous transaction.
1654 */
1655 if (dr->dr_next)
1656 zp.zp_nopwrite = B_FALSE;
1657
1658 ASSERT(dr->dr_txg == txg);
1659 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1660 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1661 /*
1662 * We have already issued a sync write for this buffer,
1663 * or this buffer has already been synced. It could not
1664 * have been dirtied since, or we would have cleared the state.
1665 */
1666 mutex_exit(&db->db_mtx);
1667 return (SET_ERROR(EALREADY));
1668 }
1669
1670 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1671 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1672 mutex_exit(&db->db_mtx);
1673
1674 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE);
1675 dsa->dsa_dr = dr;
1676 dsa->dsa_done = done;
1677 dsa->dsa_zgd = zgd;
1678 dsa->dsa_tx = NULL;
1679
1680 zio_nowait(arc_write(pio, os->os_spa, txg,
1681 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
1682 DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready,
1683 NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE,
1684 ZIO_FLAG_CANFAIL, &zb));
1685
1686 return (0);
1687 }
1688
1689 int
1690 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1691 dmu_tx_t *tx)
1692 {
1693 dnode_t *dn;
1694 int err;
1695
1696 err = dnode_hold(os, object, FTAG, &dn);
1697 if (err)
1698 return (err);
1699 err = dnode_set_blksz(dn, size, ibs, tx);
1700 dnode_rele(dn, FTAG);
1701 return (err);
1702 }
1703
1704 void
1705 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1706 dmu_tx_t *tx)
1707 {
1708 dnode_t *dn;
1709
1710 /*
1711 * Send streams include each object's checksum function. This
1712 * check ensures that the receiving system can understand the
1713 * checksum function transmitted.
1714 */
1715 ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
1716
1717 VERIFY0(dnode_hold(os, object, FTAG, &dn));
1718 ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
1719 dn->dn_checksum = checksum;
1720 dnode_setdirty(dn, tx);
1721 dnode_rele(dn, FTAG);
1722 }
1723
1724 void
1725 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1726 dmu_tx_t *tx)
1727 {
1728 dnode_t *dn;
1729
1730 /*
1731 * Send streams include each object's compression function. This
1732 * check ensures that the receiving system can understand the
1733 * compression function transmitted.
1734 */
1735 ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
1736
1737 VERIFY0(dnode_hold(os, object, FTAG, &dn));
1738 dn->dn_compress = compress;
1739 dnode_setdirty(dn, tx);
1740 dnode_rele(dn, FTAG);
1741 }
1742
1743 int zfs_mdcomp_disable = 0;
1744
1745 /*
1746 * When the "redundant_metadata" property is set to "most", only indirect
1747 * blocks of this level and higher will have an additional ditto block.
1748 */
1749 int zfs_redundant_metadata_most_ditto_level = 2;
1750
1751 void
1752 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1753 {
1754 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1755 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1756 (wp & WP_SPILL));
1757 enum zio_checksum checksum = os->os_checksum;
1758 enum zio_compress compress = os->os_compress;
1759 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1760 boolean_t dedup = B_FALSE;
1761 boolean_t nopwrite = B_FALSE;
1762 boolean_t dedup_verify = os->os_dedup_verify;
1763 int copies = os->os_copies;
1764
1765 /*
1766 * We maintain different write policies for each of the following
1767 * types of data:
1768 * 1. metadata
1769 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
1770 * 3. all other level 0 blocks
1771 */
1772 if (ismd) {
1773 /*
1774 * XXX -- we should design a compression algorithm
1775 * that specializes in arrays of bps.
1776 */
1777 boolean_t lz4_ac = spa_feature_is_active(os->os_spa,
1778 SPA_FEATURE_LZ4_COMPRESS);
1779
1780 if (zfs_mdcomp_disable) {
1781 compress = ZIO_COMPRESS_EMPTY;
1782 } else if (lz4_ac) {
1783 compress = ZIO_COMPRESS_LZ4;
1784 } else {
1785 compress = ZIO_COMPRESS_LZJB;
1786 }
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
1799 if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
1800 (os->os_redundant_metadata ==
1801 ZFS_REDUNDANT_METADATA_MOST &&
1802 (level >= zfs_redundant_metadata_most_ditto_level ||
1803 DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))))
1804 copies++;
1805 } else if (wp & WP_NOFILL) {
1806 ASSERT(level == 0);
1807
1808 /*
1809 * If we're writing preallocated blocks, we aren't actually
1810 * writing them so don't set any policy properties. These
1811 * blocks are currently only used by an external subsystem
1812 * outside of zfs (i.e. dump) and not written by the zio
1813 * pipeline.
1814 */
1815 compress = ZIO_COMPRESS_OFF;
1816 checksum = ZIO_CHECKSUM_OFF;
1817 } else {
1818 compress = zio_compress_select(dn->dn_compress, compress);
1819
1820 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
1821 zio_checksum_select(dn->dn_checksum, checksum) :
1822 dedup_checksum;
1823
1824 /*
1825 * Determine dedup setting. If we are in dmu_sync(),
1826 * we won't actually dedup now because that's all
1827 * done in syncing context; but we do want to use the
1828 * dedup checkum. If the checksum is not strong
1829 * enough to ensure unique signatures, force
1830 * dedup_verify.
1831 */
1832 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
1833 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
1834 if (!zio_checksum_table[checksum].ci_dedup)
1835 dedup_verify = B_TRUE;
1836 }
1837
1838 /*
1839 * Enable nopwrite if we have a cryptographically secure
1840 * checksum that has no known collisions (i.e. SHA-256)
1841 * and compression is enabled. We don't enable nopwrite if
1842 * dedup is enabled as the two features are mutually exclusive.
1843 */
1844 nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup &&
1845 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
1846 }
1847
1848 zp->zp_checksum = checksum;
1849 zp->zp_compress = compress;
1850 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
1851 zp->zp_level = level;
1852 zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
1853 zp->zp_dedup = dedup;
1854 zp->zp_dedup_verify = dedup && dedup_verify;
1855 zp->zp_nopwrite = nopwrite;
1856 }
1857
1858 int
1859 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1860 {
1861 dnode_t *dn;
1862 int i, err;
1863
1864 err = dnode_hold(os, object, FTAG, &dn);
1865 if (err)
1866 return (err);
1867 /*
1868 * Sync any current changes before
1869 * we go trundling through the block pointers.
1870 */
1871 for (i = 0; i < TXG_SIZE; i++) {
1872 if (list_link_active(&dn->dn_dirty_link[i]))
1873 break;
1874 }
1875 if (i != TXG_SIZE) {
1876 dnode_rele(dn, FTAG);
1877 txg_wait_synced(dmu_objset_pool(os), 0);
1878 err = dnode_hold(os, object, FTAG, &dn);
1879 if (err)
1880 return (err);
1881 }
1882
1883 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1884 dnode_rele(dn, FTAG);
1885
1886 return (err);
1887 }
1888
1889 void
1890 __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1891 {
1892 dnode_phys_t *dnp = dn->dn_phys;
1893 int i;
1894
1895 doi->doi_data_block_size = dn->dn_datablksz;
1896 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1897 1ULL << dn->dn_indblkshift : 0;
1898 doi->doi_type = dn->dn_type;
1899 doi->doi_bonus_type = dn->dn_bonustype;
1900 doi->doi_bonus_size = dn->dn_bonuslen;
1901 doi->doi_indirection = dn->dn_nlevels;
1902 doi->doi_checksum = dn->dn_checksum;
1903 doi->doi_compress = dn->dn_compress;
1904 doi->doi_nblkptr = dn->dn_nblkptr;
1905 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1906 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
1907 doi->doi_fill_count = 0;
1908 for (i = 0; i < dnp->dn_nblkptr; i++)
1909 doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
1910 }
1911
1912 void
1913 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1914 {
1915 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1916 mutex_enter(&dn->dn_mtx);
1917
1918 __dmu_object_info_from_dnode(dn, doi);
1919
1920 mutex_exit(&dn->dn_mtx);
1921 rw_exit(&dn->dn_struct_rwlock);
1922 }
1923
1924 /*
1925 * Get information on a DMU object.
1926 * If doi is NULL, just indicates whether the object exists.
1927 */
1928 int
1929 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1930 {
1931 dnode_t *dn;
1932 int err = dnode_hold(os, object, FTAG, &dn);
1933
1934 if (err)
1935 return (err);
1936
1937 if (doi != NULL)
1938 dmu_object_info_from_dnode(dn, doi);
1939
1940 dnode_rele(dn, FTAG);
1941 return (0);
1942 }
1943
1944 /*
1945 * As above, but faster; can be used when you have a held dbuf in hand.
1946 */
1947 void
1948 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1949 {
1950 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1951
1952 DB_DNODE_ENTER(db);
1953 dmu_object_info_from_dnode(DB_DNODE(db), doi);
1954 DB_DNODE_EXIT(db);
1955 }
1956
1957 /*
1958 * Faster still when you only care about the size.
1959 * This is specifically optimized for zfs_getattr().
1960 */
1961 void
1962 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1963 u_longlong_t *nblk512)
1964 {
1965 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1966 dnode_t *dn;
1967
1968 DB_DNODE_ENTER(db);
1969 dn = DB_DNODE(db);
1970
1971 *blksize = dn->dn_datablksz;
1972 /* add 1 for dnode space */
1973 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1974 SPA_MINBLOCKSHIFT) + 1;
1975 DB_DNODE_EXIT(db);
1976 }
1977
1978 void
1979 byteswap_uint64_array(void *vbuf, size_t size)
1980 {
1981 uint64_t *buf = vbuf;
1982 size_t count = size >> 3;
1983 int i;
1984
1985 ASSERT((size & 7) == 0);
1986
1987 for (i = 0; i < count; i++)
1988 buf[i] = BSWAP_64(buf[i]);
1989 }
1990
1991 void
1992 byteswap_uint32_array(void *vbuf, size_t size)
1993 {
1994 uint32_t *buf = vbuf;
1995 size_t count = size >> 2;
1996 int i;
1997
1998 ASSERT((size & 3) == 0);
1999
2000 for (i = 0; i < count; i++)
2001 buf[i] = BSWAP_32(buf[i]);
2002 }
2003
2004 void
2005 byteswap_uint16_array(void *vbuf, size_t size)
2006 {
2007 uint16_t *buf = vbuf;
2008 size_t count = size >> 1;
2009 int i;
2010
2011 ASSERT((size & 1) == 0);
2012
2013 for (i = 0; i < count; i++)
2014 buf[i] = BSWAP_16(buf[i]);
2015 }
2016
2017 /* ARGSUSED */
2018 void
2019 byteswap_uint8_array(void *vbuf, size_t size)
2020 {
2021 }
2022
2023 void
2024 dmu_init(void)
2025 {
2026 zfs_dbgmsg_init();
2027 sa_cache_init();
2028 xuio_stat_init();
2029 dmu_objset_init();
2030 dnode_init();
2031 dbuf_init();
2032 zfetch_init();
2033 dmu_tx_init();
2034 l2arc_init();
2035 arc_init();
2036 }
2037
2038 void
2039 dmu_fini(void)
2040 {
2041 arc_fini(); /* arc depends on l2arc, so arc must go first */
2042 l2arc_fini();
2043 dmu_tx_fini();
2044 zfetch_fini();
2045 dbuf_fini();
2046 dnode_fini();
2047 dmu_objset_fini();
2048 xuio_stat_fini();
2049 sa_cache_fini();
2050 zfs_dbgmsg_fini();
2051 }
2052
2053 #if defined(_KERNEL) && defined(HAVE_SPL)
2054 EXPORT_SYMBOL(dmu_bonus_hold);
2055 EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus);
2056 EXPORT_SYMBOL(dmu_buf_rele_array);
2057 EXPORT_SYMBOL(dmu_prefetch);
2058 EXPORT_SYMBOL(dmu_free_range);
2059 EXPORT_SYMBOL(dmu_free_long_range);
2060 EXPORT_SYMBOL(dmu_free_long_object);
2061 EXPORT_SYMBOL(dmu_read);
2062 EXPORT_SYMBOL(dmu_write);
2063 EXPORT_SYMBOL(dmu_prealloc);
2064 EXPORT_SYMBOL(dmu_object_info);
2065 EXPORT_SYMBOL(dmu_object_info_from_dnode);
2066 EXPORT_SYMBOL(dmu_object_info_from_db);
2067 EXPORT_SYMBOL(dmu_object_size_from_db);
2068 EXPORT_SYMBOL(dmu_object_set_blocksize);
2069 EXPORT_SYMBOL(dmu_object_set_checksum);
2070 EXPORT_SYMBOL(dmu_object_set_compress);
2071 EXPORT_SYMBOL(dmu_write_policy);
2072 EXPORT_SYMBOL(dmu_sync);
2073 EXPORT_SYMBOL(dmu_request_arcbuf);
2074 EXPORT_SYMBOL(dmu_return_arcbuf);
2075 EXPORT_SYMBOL(dmu_assign_arcbuf);
2076 EXPORT_SYMBOL(dmu_buf_hold);
2077 EXPORT_SYMBOL(dmu_ot);
2078
2079 module_param(zfs_mdcomp_disable, int, 0644);
2080 MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression");
2081
2082 module_param(zfs_nopwrite_enabled, int, 0644);
2083 MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes");
2084
2085 #endif
ZFS On Linux mirror for PVE