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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
27 #include <sys/dmu_impl.h>
28 #include <sys/dmu_tx.h>
30 #include <sys/dnode.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dmu_zfetch.h>
40 #include <sys/zfs_ioctl.h>
42 #include <sys/zio_checksum.h>
43 #include <sys/zio_compress.h>
46 #include <sys/zfs_znode.h>
50 * Enable/disable nopwrite feature.
52 int zfs_nopwrite_enabled = 1;
53 SYSCTL_DECL(_vfs_zfs);
54 TUNABLE_INT("vfs.zfs.nopwrite_enabled", &zfs_nopwrite_enabled);
55 SYSCTL_INT(_vfs_zfs, OID_AUTO, nopwrite_enabled, CTLFLAG_RDTUN,
56 &zfs_nopwrite_enabled, 0, "Enable nopwrite feature");
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" }
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" }
129 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
130 void *tag, dmu_buf_t **dbp, int flags)
136 int db_flags = DB_RF_CANFAIL;
138 if (flags & DMU_READ_NO_PREFETCH)
139 db_flags |= DB_RF_NOPREFETCH;
141 err = dnode_hold(os, object, FTAG, &dn);
144 blkid = dbuf_whichblock(dn, offset);
145 rw_enter(&dn->dn_struct_rwlock, RW_READER);
146 db = dbuf_hold(dn, blkid, tag);
147 rw_exit(&dn->dn_struct_rwlock);
151 err = dbuf_read(db, NULL, db_flags);
158 dnode_rele(dn, FTAG);
159 *dbp = &db->db; /* NULL db plus first field offset is NULL */
166 return (DN_MAX_BONUSLEN);
170 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
172 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
179 if (dn->dn_bonus != db) {
181 } else if (newsize < 0 || newsize > db_fake->db_size) {
184 dnode_setbonuslen(dn, newsize, tx);
193 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
195 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
202 if (!DMU_OT_IS_VALID(type)) {
204 } else if (dn->dn_bonus != db) {
207 dnode_setbonus_type(dn, type, tx);
216 dmu_get_bonustype(dmu_buf_t *db_fake)
218 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
220 dmu_object_type_t type;
224 type = dn->dn_bonustype;
231 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
236 error = dnode_hold(os, object, FTAG, &dn);
237 dbuf_rm_spill(dn, tx);
238 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
239 dnode_rm_spill(dn, tx);
240 rw_exit(&dn->dn_struct_rwlock);
241 dnode_rele(dn, FTAG);
246 * returns ENOENT, EIO, or 0.
249 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
255 error = dnode_hold(os, object, FTAG, &dn);
259 rw_enter(&dn->dn_struct_rwlock, RW_READER);
260 if (dn->dn_bonus == NULL) {
261 rw_exit(&dn->dn_struct_rwlock);
262 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
263 if (dn->dn_bonus == NULL)
264 dbuf_create_bonus(dn);
268 /* as long as the bonus buf is held, the dnode will be held */
269 if (refcount_add(&db->db_holds, tag) == 1) {
270 VERIFY(dnode_add_ref(dn, db));
271 (void) atomic_inc_32_nv(&dn->dn_dbufs_count);
275 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
276 * hold and incrementing the dbuf count to ensure that dnode_move() sees
277 * a dnode hold for every dbuf.
279 rw_exit(&dn->dn_struct_rwlock);
281 dnode_rele(dn, FTAG);
283 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
290 * returns ENOENT, EIO, or 0.
292 * This interface will allocate a blank spill dbuf when a spill blk
293 * doesn't already exist on the dnode.
295 * if you only want to find an already existing spill db, then
296 * dmu_spill_hold_existing() should be used.
299 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
301 dmu_buf_impl_t *db = NULL;
304 if ((flags & DB_RF_HAVESTRUCT) == 0)
305 rw_enter(&dn->dn_struct_rwlock, RW_READER);
307 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
309 if ((flags & DB_RF_HAVESTRUCT) == 0)
310 rw_exit(&dn->dn_struct_rwlock);
313 err = dbuf_read(db, NULL, flags);
322 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
324 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
331 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
334 rw_enter(&dn->dn_struct_rwlock, RW_READER);
336 if (!dn->dn_have_spill) {
339 err = dmu_spill_hold_by_dnode(dn,
340 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
343 rw_exit(&dn->dn_struct_rwlock);
351 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
353 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
359 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
366 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
367 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
368 * and can induce severe lock contention when writing to several files
369 * whose dnodes are in the same block.
372 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
373 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
375 dsl_pool_t *dp = NULL;
377 uint64_t blkid, nblks, i;
383 ASSERT(length <= DMU_MAX_ACCESS);
385 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
386 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
387 dbuf_flags |= DB_RF_NOPREFETCH;
389 rw_enter(&dn->dn_struct_rwlock, RW_READER);
390 if (dn->dn_datablkshift) {
391 int blkshift = dn->dn_datablkshift;
392 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
393 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
395 if (offset + length > dn->dn_datablksz) {
396 zfs_panic_recover("zfs: accessing past end of object "
397 "%llx/%llx (size=%u access=%llu+%llu)",
398 (longlong_t)dn->dn_objset->
399 os_dsl_dataset->ds_object,
400 (longlong_t)dn->dn_object, dn->dn_datablksz,
401 (longlong_t)offset, (longlong_t)length);
402 rw_exit(&dn->dn_struct_rwlock);
407 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
409 if (dn->dn_objset->os_dsl_dataset)
410 dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
411 if (dp && dsl_pool_sync_context(dp))
413 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
414 blkid = dbuf_whichblock(dn, offset);
415 for (i = 0; i < nblks; i++) {
416 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
418 rw_exit(&dn->dn_struct_rwlock);
419 dmu_buf_rele_array(dbp, nblks, tag);
423 /* initiate async i/o */
425 (void) dbuf_read(db, zio, dbuf_flags);
428 curthread->td_ru.ru_oublock++;
432 rw_exit(&dn->dn_struct_rwlock);
434 /* wait for async i/o */
436 /* track read overhead when we are in sync context */
437 if (dp && dsl_pool_sync_context(dp))
438 dp->dp_read_overhead += gethrtime() - start;
440 dmu_buf_rele_array(dbp, nblks, tag);
444 /* wait for other io to complete */
446 for (i = 0; i < nblks; i++) {
447 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
448 mutex_enter(&db->db_mtx);
449 while (db->db_state == DB_READ ||
450 db->db_state == DB_FILL)
451 cv_wait(&db->db_changed, &db->db_mtx);
452 if (db->db_state == DB_UNCACHED)
454 mutex_exit(&db->db_mtx);
456 dmu_buf_rele_array(dbp, nblks, tag);
468 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
469 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
474 err = dnode_hold(os, object, FTAG, &dn);
478 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
479 numbufsp, dbpp, DMU_READ_PREFETCH);
481 dnode_rele(dn, FTAG);
487 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
488 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
490 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
496 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
497 numbufsp, dbpp, DMU_READ_PREFETCH);
504 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
507 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
512 for (i = 0; i < numbufs; i++) {
514 dbuf_rele(dbp[i], tag);
517 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
521 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
527 if (zfs_prefetch_disable)
530 if (len == 0) { /* they're interested in the bonus buffer */
531 dn = DMU_META_DNODE(os);
533 if (object == 0 || object >= DN_MAX_OBJECT)
536 rw_enter(&dn->dn_struct_rwlock, RW_READER);
537 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
538 dbuf_prefetch(dn, blkid);
539 rw_exit(&dn->dn_struct_rwlock);
544 * XXX - Note, if the dnode for the requested object is not
545 * already cached, we will do a *synchronous* read in the
546 * dnode_hold() call. The same is true for any indirects.
548 err = dnode_hold(os, object, FTAG, &dn);
552 rw_enter(&dn->dn_struct_rwlock, RW_READER);
553 if (dn->dn_datablkshift) {
554 int blkshift = dn->dn_datablkshift;
555 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
556 P2ALIGN(offset, 1<<blkshift)) >> blkshift;
558 nblks = (offset < dn->dn_datablksz);
562 blkid = dbuf_whichblock(dn, offset);
563 for (i = 0; i < nblks; i++)
564 dbuf_prefetch(dn, blkid+i);
567 rw_exit(&dn->dn_struct_rwlock);
569 dnode_rele(dn, FTAG);
573 * Get the next "chunk" of file data to free. We traverse the file from
574 * the end so that the file gets shorter over time (if we crashes in the
575 * middle, this will leave us in a better state). We find allocated file
576 * data by simply searching the allocated level 1 indirects.
579 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
581 uint64_t len = *start - limit;
583 uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
585 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
587 ASSERT(limit <= *start);
589 if (len <= iblkrange * maxblks) {
593 ASSERT(ISP2(iblkrange));
595 while (*start > limit && blkcnt < maxblks) {
598 /* find next allocated L1 indirect */
599 err = dnode_next_offset(dn,
600 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
602 /* if there are no more, then we are done */
611 /* reset offset to end of "next" block back */
612 *start = P2ALIGN(*start, iblkrange);
622 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
623 uint64_t length, boolean_t free_dnode)
626 uint64_t object_size, start, end, len;
627 boolean_t trunc = (length == DMU_OBJECT_END);
630 align = 1 << dn->dn_datablkshift;
632 object_size = align == 1 ? dn->dn_datablksz :
633 (dn->dn_maxblkid + 1) << dn->dn_datablkshift;
635 end = offset + length;
636 if (trunc || end > object_size)
640 length = end - offset;
644 /* assert(offset <= start) */
645 err = get_next_chunk(dn, &start, offset);
648 len = trunc ? DMU_OBJECT_END : end - start;
650 tx = dmu_tx_create(os);
651 dmu_tx_hold_free(tx, dn->dn_object, start, len);
652 err = dmu_tx_assign(tx, TXG_WAIT);
658 dnode_free_range(dn, start, trunc ? -1 : len, tx);
660 if (start == 0 && free_dnode) {
665 length -= end - start;
674 dmu_free_long_range(objset_t *os, uint64_t object,
675 uint64_t offset, uint64_t length)
680 err = dnode_hold(os, object, FTAG, &dn);
683 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
684 dnode_rele(dn, FTAG);
689 dmu_free_object(objset_t *os, uint64_t object)
695 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
699 if (dn->dn_nlevels == 1) {
700 tx = dmu_tx_create(os);
701 dmu_tx_hold_bonus(tx, object);
702 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END);
703 err = dmu_tx_assign(tx, TXG_WAIT);
705 dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
712 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
714 dnode_rele(dn, FTAG);
719 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
720 uint64_t size, dmu_tx_t *tx)
723 int err = dnode_hold(os, object, FTAG, &dn);
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);
734 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
735 void *buf, uint32_t flags)
741 err = dnode_hold(os, object, FTAG, &dn);
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.
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);
758 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
762 * NB: we could do this block-at-a-time, but it's nice
763 * to be reading in parallel.
765 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
766 TRUE, FTAG, &numbufs, &dbp, flags);
770 for (i = 0; i < numbufs; i++) {
773 dmu_buf_t *db = dbp[i];
777 bufoff = offset - db->db_offset;
778 tocpy = (int)MIN(db->db_size - bufoff, size);
780 bcopy((char *)db->db_data + bufoff, buf, tocpy);
784 buf = (char *)buf + tocpy;
786 dmu_buf_rele_array(dbp, numbufs, FTAG);
788 dnode_rele(dn, FTAG);
793 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
794 const void *buf, dmu_tx_t *tx)
802 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
803 FALSE, FTAG, &numbufs, &dbp));
805 for (i = 0; i < numbufs; i++) {
808 dmu_buf_t *db = dbp[i];
812 bufoff = offset - db->db_offset;
813 tocpy = (int)MIN(db->db_size - bufoff, size);
815 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
817 if (tocpy == db->db_size)
818 dmu_buf_will_fill(db, tx);
820 dmu_buf_will_dirty(db, tx);
822 bcopy(buf, (char *)db->db_data + bufoff, tocpy);
824 if (tocpy == db->db_size)
825 dmu_buf_fill_done(db, tx);
829 buf = (char *)buf + tocpy;
831 dmu_buf_rele_array(dbp, numbufs, FTAG);
835 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
844 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
845 FALSE, FTAG, &numbufs, &dbp));
847 for (i = 0; i < numbufs; i++) {
848 dmu_buf_t *db = dbp[i];
850 dmu_buf_will_not_fill(db, tx);
852 dmu_buf_rele_array(dbp, numbufs, FTAG);
856 * DMU support for xuio
858 kstat_t *xuio_ksp = NULL;
861 dmu_xuio_init(xuio_t *xuio, int nblk)
864 uio_t *uio = &xuio->xu_uio;
866 uio->uio_iovcnt = nblk;
867 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
869 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
871 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
872 priv->iovp = uio->uio_iov;
873 XUIO_XUZC_PRIV(xuio) = priv;
875 if (XUIO_XUZC_RW(xuio) == UIO_READ)
876 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
878 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
884 dmu_xuio_fini(xuio_t *xuio)
886 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
887 int nblk = priv->cnt;
889 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
890 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
891 kmem_free(priv, sizeof (dmu_xuio_t));
893 if (XUIO_XUZC_RW(xuio) == UIO_READ)
894 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
896 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
900 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
901 * and increase priv->next by 1.
904 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
907 uio_t *uio = &xuio->xu_uio;
908 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
909 int i = priv->next++;
911 ASSERT(i < priv->cnt);
912 ASSERT(off + n <= arc_buf_size(abuf));
913 iov = uio->uio_iov + i;
914 iov->iov_base = (char *)abuf->b_data + off;
916 priv->bufs[i] = abuf;
921 dmu_xuio_cnt(xuio_t *xuio)
923 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
928 dmu_xuio_arcbuf(xuio_t *xuio, int i)
930 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
932 ASSERT(i < priv->cnt);
933 return (priv->bufs[i]);
937 dmu_xuio_clear(xuio_t *xuio, int i)
939 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
941 ASSERT(i < priv->cnt);
942 priv->bufs[i] = NULL;
948 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
949 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
951 if (xuio_ksp != NULL) {
952 xuio_ksp->ks_data = &xuio_stats;
953 kstat_install(xuio_ksp);
960 if (xuio_ksp != NULL) {
961 kstat_delete(xuio_ksp);
967 xuio_stat_wbuf_copied()
969 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
973 xuio_stat_wbuf_nocopy()
975 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
980 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
987 * NB: we could do this block-at-a-time, but it's nice
988 * to be reading in parallel.
990 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
996 if (uio->uio_extflg == UIO_XUIO)
997 xuio = (xuio_t *)uio;
1000 for (i = 0; i < numbufs; i++) {
1003 dmu_buf_t *db = dbp[i];
1007 bufoff = uio->uio_loffset - db->db_offset;
1008 tocpy = (int)MIN(db->db_size - bufoff, size);
1011 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1012 arc_buf_t *dbuf_abuf = dbi->db_buf;
1013 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1014 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1016 uio->uio_resid -= tocpy;
1017 uio->uio_loffset += tocpy;
1020 if (abuf == dbuf_abuf)
1021 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1023 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1025 err = uiomove((char *)db->db_data + bufoff, tocpy,
1033 dmu_buf_rele_array(dbp, numbufs, FTAG);
1039 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1046 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1047 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1051 for (i = 0; i < numbufs; i++) {
1054 dmu_buf_t *db = dbp[i];
1058 bufoff = uio->uio_loffset - db->db_offset;
1059 tocpy = (int)MIN(db->db_size - bufoff, size);
1061 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1063 if (tocpy == db->db_size)
1064 dmu_buf_will_fill(db, tx);
1066 dmu_buf_will_dirty(db, tx);
1069 * XXX uiomove could block forever (eg. nfs-backed
1070 * pages). There needs to be a uiolockdown() function
1071 * to lock the pages in memory, so that uiomove won't
1074 err = uiomove((char *)db->db_data + bufoff, tocpy,
1077 if (tocpy == db->db_size)
1078 dmu_buf_fill_done(db, tx);
1086 dmu_buf_rele_array(dbp, numbufs, FTAG);
1091 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1094 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1103 err = dmu_write_uio_dnode(dn, uio, size, tx);
1110 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1119 err = dnode_hold(os, object, FTAG, &dn);
1123 err = dmu_write_uio_dnode(dn, uio, size, tx);
1125 dnode_rele(dn, FTAG);
1132 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1133 page_t *pp, dmu_tx_t *tx)
1142 err = dmu_buf_hold_array(os, object, offset, size,
1143 FALSE, FTAG, &numbufs, &dbp);
1147 for (i = 0; i < numbufs; i++) {
1148 int tocpy, copied, thiscpy;
1150 dmu_buf_t *db = dbp[i];
1154 ASSERT3U(db->db_size, >=, PAGESIZE);
1156 bufoff = offset - db->db_offset;
1157 tocpy = (int)MIN(db->db_size - bufoff, size);
1159 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1161 if (tocpy == db->db_size)
1162 dmu_buf_will_fill(db, tx);
1164 dmu_buf_will_dirty(db, tx);
1166 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1167 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
1168 thiscpy = MIN(PAGESIZE, tocpy - copied);
1169 va = zfs_map_page(pp, S_READ);
1170 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1171 zfs_unmap_page(pp, va);
1176 if (tocpy == db->db_size)
1177 dmu_buf_fill_done(db, tx);
1182 dmu_buf_rele_array(dbp, numbufs, FTAG);
1189 * Allocate a loaned anonymous arc buffer.
1192 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1194 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1197 DB_GET_SPA(&spa, db);
1198 return (arc_loan_buf(spa, size));
1202 * Free a loaned arc buffer.
1205 dmu_return_arcbuf(arc_buf_t *buf)
1207 arc_return_buf(buf, FTAG);
1208 VERIFY(arc_buf_remove_ref(buf, FTAG) == 1);
1212 * When possible directly assign passed loaned arc buffer to a dbuf.
1213 * If this is not possible copy the contents of passed arc buf via
1217 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1220 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1223 uint32_t blksz = (uint32_t)arc_buf_size(buf);
1226 DB_DNODE_ENTER(dbuf);
1227 dn = DB_DNODE(dbuf);
1228 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1229 blkid = dbuf_whichblock(dn, offset);
1230 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1231 rw_exit(&dn->dn_struct_rwlock);
1232 DB_DNODE_EXIT(dbuf);
1234 if (offset == db->db.db_offset && blksz == db->db.db_size) {
1235 dbuf_assign_arcbuf(db, buf, tx);
1236 dbuf_rele(db, FTAG);
1241 DB_DNODE_ENTER(dbuf);
1242 dn = DB_DNODE(dbuf);
1244 object = dn->dn_object;
1245 DB_DNODE_EXIT(dbuf);
1247 dbuf_rele(db, FTAG);
1248 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1249 dmu_return_arcbuf(buf);
1250 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1255 dbuf_dirty_record_t *dsa_dr;
1256 dmu_sync_cb_t *dsa_done;
1263 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1265 dmu_sync_arg_t *dsa = varg;
1266 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1267 blkptr_t *bp = zio->io_bp;
1269 if (zio->io_error == 0) {
1270 if (BP_IS_HOLE(bp)) {
1272 * A block of zeros may compress to a hole, but the
1273 * block size still needs to be known for replay.
1275 BP_SET_LSIZE(bp, db->db_size);
1277 ASSERT(BP_GET_LEVEL(bp) == 0);
1284 dmu_sync_late_arrival_ready(zio_t *zio)
1286 dmu_sync_ready(zio, NULL, zio->io_private);
1291 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1293 dmu_sync_arg_t *dsa = varg;
1294 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1295 dmu_buf_impl_t *db = dr->dr_dbuf;
1297 mutex_enter(&db->db_mtx);
1298 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1299 if (zio->io_error == 0) {
1300 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1301 if (dr->dt.dl.dr_nopwrite) {
1302 blkptr_t *bp = zio->io_bp;
1303 blkptr_t *bp_orig = &zio->io_bp_orig;
1304 uint8_t chksum = BP_GET_CHECKSUM(bp_orig);
1306 ASSERT(BP_EQUAL(bp, bp_orig));
1307 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1308 ASSERT(zio_checksum_table[chksum].ci_dedup);
1310 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1311 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1312 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1313 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
1314 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1316 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1318 cv_broadcast(&db->db_changed);
1319 mutex_exit(&db->db_mtx);
1321 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1323 kmem_free(dsa, sizeof (*dsa));
1327 dmu_sync_late_arrival_done(zio_t *zio)
1329 blkptr_t *bp = zio->io_bp;
1330 dmu_sync_arg_t *dsa = zio->io_private;
1331 blkptr_t *bp_orig = &zio->io_bp_orig;
1333 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1335 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1336 * then there is nothing to do here. Otherwise, free the
1337 * newly allocated block in this txg.
1339 if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
1340 ASSERT(BP_EQUAL(bp, bp_orig));
1342 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1343 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1344 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1345 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1349 dmu_tx_commit(dsa->dsa_tx);
1351 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1353 kmem_free(dsa, sizeof (*dsa));
1357 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1358 zio_prop_t *zp, zbookmark_t *zb)
1360 dmu_sync_arg_t *dsa;
1363 tx = dmu_tx_create(os);
1364 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1365 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1367 return (EIO); /* Make zl_get_data do txg_waited_synced() */
1370 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1372 dsa->dsa_done = done;
1376 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1377 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
1378 dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa,
1379 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
1385 * Intent log support: sync the block associated with db to disk.
1386 * N.B. and XXX: the caller is responsible for making sure that the
1387 * data isn't changing while dmu_sync() is writing it.
1391 * EEXIST: this txg has already been synced, so there's nothing to do.
1392 * The caller should not log the write.
1394 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1395 * The caller should not log the write.
1397 * EALREADY: this block is already in the process of being synced.
1398 * The caller should track its progress (somehow).
1400 * EIO: could not do the I/O.
1401 * The caller should do a txg_wait_synced().
1403 * 0: the I/O has been initiated.
1404 * The caller should log this blkptr in the done callback.
1405 * It is possible that the I/O will fail, in which case
1406 * the error will be reported to the done callback and
1407 * propagated to pio from zio_done().
1410 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1412 blkptr_t *bp = zgd->zgd_bp;
1413 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1414 objset_t *os = db->db_objset;
1415 dsl_dataset_t *ds = os->os_dsl_dataset;
1416 dbuf_dirty_record_t *dr;
1417 dmu_sync_arg_t *dsa;
1422 ASSERT(pio != NULL);
1425 SET_BOOKMARK(&zb, ds->ds_object,
1426 db->db.db_object, db->db_level, db->db_blkid);
1430 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1434 * If we're frozen (running ziltest), we always need to generate a bp.
1436 if (txg > spa_freeze_txg(os->os_spa))
1437 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1440 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1441 * and us. If we determine that this txg is not yet syncing,
1442 * but it begins to sync a moment later, that's OK because the
1443 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1445 mutex_enter(&db->db_mtx);
1447 if (txg <= spa_last_synced_txg(os->os_spa)) {
1449 * This txg has already synced. There's nothing to do.
1451 mutex_exit(&db->db_mtx);
1455 if (txg <= spa_syncing_txg(os->os_spa)) {
1457 * This txg is currently syncing, so we can't mess with
1458 * the dirty record anymore; just write a new log block.
1460 mutex_exit(&db->db_mtx);
1461 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1464 dr = db->db_last_dirty;
1465 while (dr && dr->dr_txg != txg)
1470 * There's no dr for this dbuf, so it must have been freed.
1471 * There's no need to log writes to freed blocks, so we're done.
1473 mutex_exit(&db->db_mtx);
1477 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
1480 * Assume the on-disk data is X, the current syncing data is Y,
1481 * and the current in-memory data is Z (currently in dmu_sync).
1482 * X and Z are identical but Y is has been modified. Normally,
1483 * when X and Z are the same we will perform a nopwrite but if Y
1484 * is different we must disable nopwrite since the resulting write
1485 * of Y to disk can free the block containing X. If we allowed a
1486 * nopwrite to occur the block pointing to Z would reference a freed
1487 * block. Since this is a rare case we simplify this by disabling
1488 * nopwrite if the current dmu_sync-ing dbuf has been modified in
1489 * a previous transaction.
1492 zp.zp_nopwrite = B_FALSE;
1494 ASSERT(dr->dr_txg == txg);
1495 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1496 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1498 * We have already issued a sync write for this buffer,
1499 * or this buffer has already been synced. It could not
1500 * have been dirtied since, or we would have cleared the state.
1502 mutex_exit(&db->db_mtx);
1506 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1507 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1508 mutex_exit(&db->db_mtx);
1510 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1512 dsa->dsa_done = done;
1516 zio_nowait(arc_write(pio, os->os_spa, txg,
1517 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp,
1518 dmu_sync_ready, dmu_sync_done, dsa,
1519 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
1525 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1531 err = dnode_hold(os, object, FTAG, &dn);
1534 err = dnode_set_blksz(dn, size, ibs, tx);
1535 dnode_rele(dn, FTAG);
1540 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1545 /* XXX assumes dnode_hold will not get an i/o error */
1546 (void) dnode_hold(os, object, FTAG, &dn);
1547 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
1548 dn->dn_checksum = checksum;
1549 dnode_setdirty(dn, tx);
1550 dnode_rele(dn, FTAG);
1554 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1559 /* XXX assumes dnode_hold will not get an i/o error */
1560 (void) dnode_hold(os, object, FTAG, &dn);
1561 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
1562 dn->dn_compress = compress;
1563 dnode_setdirty(dn, tx);
1564 dnode_rele(dn, FTAG);
1567 int zfs_mdcomp_disable = 0;
1568 TUNABLE_INT("vfs.zfs.mdcomp_disable", &zfs_mdcomp_disable);
1569 SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RW,
1570 &zfs_mdcomp_disable, 0, "Disable metadata compression");
1573 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1575 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1576 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1578 enum zio_checksum checksum = os->os_checksum;
1579 enum zio_compress compress = os->os_compress;
1580 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1581 boolean_t dedup = B_FALSE;
1582 boolean_t nopwrite = B_FALSE;
1583 boolean_t dedup_verify = os->os_dedup_verify;
1584 int copies = os->os_copies;
1587 * We maintain different write policies for each of the following
1590 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
1591 * 3. all other level 0 blocks
1595 * XXX -- we should design a compression algorithm
1596 * that specializes in arrays of bps.
1598 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
1602 * Metadata always gets checksummed. If the data
1603 * checksum is multi-bit correctable, and it's not a
1604 * ZBT-style checksum, then it's suitable for metadata
1605 * as well. Otherwise, the metadata checksum defaults
1608 if (zio_checksum_table[checksum].ci_correctable < 1 ||
1609 zio_checksum_table[checksum].ci_eck)
1610 checksum = ZIO_CHECKSUM_FLETCHER_4;
1611 } else if (wp & WP_NOFILL) {
1615 * If we're writing preallocated blocks, we aren't actually
1616 * writing them so don't set any policy properties. These
1617 * blocks are currently only used by an external subsystem
1618 * outside of zfs (i.e. dump) and not written by the zio
1621 compress = ZIO_COMPRESS_OFF;
1622 checksum = ZIO_CHECKSUM_OFF;
1624 compress = zio_compress_select(dn->dn_compress, compress);
1626 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
1627 zio_checksum_select(dn->dn_checksum, checksum) :
1631 * Determine dedup setting. If we are in dmu_sync(),
1632 * we won't actually dedup now because that's all
1633 * done in syncing context; but we do want to use the
1634 * dedup checkum. If the checksum is not strong
1635 * enough to ensure unique signatures, force
1638 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
1639 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
1640 if (!zio_checksum_table[checksum].ci_dedup)
1641 dedup_verify = B_TRUE;
1645 * Enable nopwrite if we have a cryptographically secure
1646 * checksum that has no known collisions (i.e. SHA-256)
1647 * and compression is enabled. We don't enable nopwrite if
1648 * dedup is enabled as the two features are mutually exclusive.
1650 nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup &&
1651 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
1654 zp->zp_checksum = checksum;
1655 zp->zp_compress = compress;
1656 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
1657 zp->zp_level = level;
1658 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa));
1659 zp->zp_dedup = dedup;
1660 zp->zp_dedup_verify = dedup && dedup_verify;
1661 zp->zp_nopwrite = nopwrite;
1665 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1670 err = dnode_hold(os, object, FTAG, &dn);
1674 * Sync any current changes before
1675 * we go trundling through the block pointers.
1677 for (i = 0; i < TXG_SIZE; i++) {
1678 if (list_link_active(&dn->dn_dirty_link[i]))
1681 if (i != TXG_SIZE) {
1682 dnode_rele(dn, FTAG);
1683 txg_wait_synced(dmu_objset_pool(os), 0);
1684 err = dnode_hold(os, object, FTAG, &dn);
1689 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1690 dnode_rele(dn, FTAG);
1696 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1700 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1701 mutex_enter(&dn->dn_mtx);
1705 doi->doi_data_block_size = dn->dn_datablksz;
1706 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1707 1ULL << dn->dn_indblkshift : 0;
1708 doi->doi_type = dn->dn_type;
1709 doi->doi_bonus_type = dn->dn_bonustype;
1710 doi->doi_bonus_size = dn->dn_bonuslen;
1711 doi->doi_indirection = dn->dn_nlevels;
1712 doi->doi_checksum = dn->dn_checksum;
1713 doi->doi_compress = dn->dn_compress;
1714 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1715 doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz;
1716 doi->doi_fill_count = 0;
1717 for (int i = 0; i < dnp->dn_nblkptr; i++)
1718 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
1720 mutex_exit(&dn->dn_mtx);
1721 rw_exit(&dn->dn_struct_rwlock);
1725 * Get information on a DMU object.
1726 * If doi is NULL, just indicates whether the object exists.
1729 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1732 int err = dnode_hold(os, object, FTAG, &dn);
1738 dmu_object_info_from_dnode(dn, doi);
1740 dnode_rele(dn, FTAG);
1745 * As above, but faster; can be used when you have a held dbuf in hand.
1748 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1750 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1753 dmu_object_info_from_dnode(DB_DNODE(db), doi);
1758 * Faster still when you only care about the size.
1759 * This is specifically optimized for zfs_getattr().
1762 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1763 u_longlong_t *nblk512)
1765 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1771 *blksize = dn->dn_datablksz;
1772 /* add 1 for dnode space */
1773 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1774 SPA_MINBLOCKSHIFT) + 1;
1779 byteswap_uint64_array(void *vbuf, size_t size)
1781 uint64_t *buf = vbuf;
1782 size_t count = size >> 3;
1785 ASSERT((size & 7) == 0);
1787 for (i = 0; i < count; i++)
1788 buf[i] = BSWAP_64(buf[i]);
1792 byteswap_uint32_array(void *vbuf, size_t size)
1794 uint32_t *buf = vbuf;
1795 size_t count = size >> 2;
1798 ASSERT((size & 3) == 0);
1800 for (i = 0; i < count; i++)
1801 buf[i] = BSWAP_32(buf[i]);
1805 byteswap_uint16_array(void *vbuf, size_t size)
1807 uint16_t *buf = vbuf;
1808 size_t count = size >> 1;
1811 ASSERT((size & 1) == 0);
1813 for (i = 0; i < count; i++)
1814 buf[i] = BSWAP_16(buf[i]);
1819 byteswap_uint8_array(void *vbuf, size_t size)