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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
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/zfs_znode.h>
45 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
46 { byteswap_uint8_array, TRUE, "unallocated" },
47 { zap_byteswap, TRUE, "object directory" },
48 { byteswap_uint64_array, TRUE, "object array" },
49 { byteswap_uint8_array, TRUE, "packed nvlist" },
50 { byteswap_uint64_array, TRUE, "packed nvlist size" },
51 { byteswap_uint64_array, TRUE, "bplist" },
52 { byteswap_uint64_array, TRUE, "bplist header" },
53 { byteswap_uint64_array, TRUE, "SPA space map header" },
54 { byteswap_uint64_array, TRUE, "SPA space map" },
55 { byteswap_uint64_array, TRUE, "ZIL intent log" },
56 { dnode_buf_byteswap, TRUE, "DMU dnode" },
57 { dmu_objset_byteswap, TRUE, "DMU objset" },
58 { byteswap_uint64_array, TRUE, "DSL directory" },
59 { zap_byteswap, TRUE, "DSL directory child map"},
60 { zap_byteswap, TRUE, "DSL dataset snap map" },
61 { zap_byteswap, TRUE, "DSL props" },
62 { byteswap_uint64_array, TRUE, "DSL dataset" },
63 { zfs_znode_byteswap, TRUE, "ZFS znode" },
64 { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" },
65 { byteswap_uint8_array, FALSE, "ZFS plain file" },
66 { zap_byteswap, TRUE, "ZFS directory" },
67 { zap_byteswap, TRUE, "ZFS master node" },
68 { zap_byteswap, TRUE, "ZFS delete queue" },
69 { byteswap_uint8_array, FALSE, "zvol object" },
70 { zap_byteswap, TRUE, "zvol prop" },
71 { byteswap_uint8_array, FALSE, "other uint8[]" },
72 { byteswap_uint64_array, FALSE, "other uint64[]" },
73 { zap_byteswap, TRUE, "other ZAP" },
74 { zap_byteswap, TRUE, "persistent error log" },
75 { byteswap_uint8_array, TRUE, "SPA history" },
76 { byteswap_uint64_array, TRUE, "SPA history offsets" },
77 { zap_byteswap, TRUE, "Pool properties" },
78 { zap_byteswap, TRUE, "DSL permissions" },
79 { zfs_acl_byteswap, TRUE, "ZFS ACL" },
80 { byteswap_uint8_array, TRUE, "ZFS SYSACL" },
81 { byteswap_uint8_array, TRUE, "FUID table" },
82 { byteswap_uint64_array, TRUE, "FUID table size" },
83 { zap_byteswap, TRUE, "DSL dataset next clones"},
84 { zap_byteswap, TRUE, "scrub work queue" },
85 { zap_byteswap, TRUE, "ZFS user/group used" },
86 { zap_byteswap, TRUE, "ZFS user/group quota" },
90 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
91 void *tag, dmu_buf_t **dbp)
98 err = dnode_hold(os->os, object, FTAG, &dn);
101 blkid = dbuf_whichblock(dn, offset);
102 rw_enter(&dn->dn_struct_rwlock, RW_READER);
103 db = dbuf_hold(dn, blkid, tag);
104 rw_exit(&dn->dn_struct_rwlock);
108 err = dbuf_read(db, NULL, DB_RF_CANFAIL);
115 dnode_rele(dn, FTAG);
123 return (DN_MAX_BONUSLEN);
127 dmu_set_bonus(dmu_buf_t *db, int newsize, dmu_tx_t *tx)
129 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
131 if (dn->dn_bonus != (dmu_buf_impl_t *)db)
133 if (newsize < 0 || newsize > db->db_size)
135 dnode_setbonuslen(dn, newsize, tx);
140 * returns ENOENT, EIO, or 0.
143 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
149 error = dnode_hold(os->os, object, FTAG, &dn);
153 rw_enter(&dn->dn_struct_rwlock, RW_READER);
154 if (dn->dn_bonus == NULL) {
155 rw_exit(&dn->dn_struct_rwlock);
156 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
157 if (dn->dn_bonus == NULL)
158 dbuf_create_bonus(dn);
161 rw_exit(&dn->dn_struct_rwlock);
163 /* as long as the bonus buf is held, the dnode will be held */
164 if (refcount_add(&db->db_holds, tag) == 1)
165 VERIFY(dnode_add_ref(dn, db));
167 dnode_rele(dn, FTAG);
169 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED));
176 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
177 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
178 * and can induce severe lock contention when writing to several files
179 * whose dnodes are in the same block.
182 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
183 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
185 dsl_pool_t *dp = NULL;
187 uint64_t blkid, nblks, i;
193 ASSERT(length <= DMU_MAX_ACCESS);
195 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT;
196 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
197 dbuf_flags |= DB_RF_NOPREFETCH;
199 rw_enter(&dn->dn_struct_rwlock, RW_READER);
200 if (dn->dn_datablkshift) {
201 int blkshift = dn->dn_datablkshift;
202 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
203 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
205 if (offset + length > dn->dn_datablksz) {
206 zfs_panic_recover("zfs: accessing past end of object "
207 "%llx/%llx (size=%u access=%llu+%llu)",
208 (longlong_t)dn->dn_objset->
209 os_dsl_dataset->ds_object,
210 (longlong_t)dn->dn_object, dn->dn_datablksz,
211 (longlong_t)offset, (longlong_t)length);
216 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
218 if (dn->dn_objset->os_dsl_dataset)
219 dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
220 if (dp && dsl_pool_sync_context(dp))
222 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
223 blkid = dbuf_whichblock(dn, offset);
224 for (i = 0; i < nblks; i++) {
225 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
227 rw_exit(&dn->dn_struct_rwlock);
228 dmu_buf_rele_array(dbp, nblks, tag);
232 /* initiate async i/o */
234 rw_exit(&dn->dn_struct_rwlock);
235 (void) dbuf_read(db, zio, dbuf_flags);
236 rw_enter(&dn->dn_struct_rwlock, RW_READER);
240 rw_exit(&dn->dn_struct_rwlock);
242 /* wait for async i/o */
244 /* track read overhead when we are in sync context */
245 if (dp && dsl_pool_sync_context(dp))
246 dp->dp_read_overhead += gethrtime() - start;
248 dmu_buf_rele_array(dbp, nblks, tag);
252 /* wait for other io to complete */
254 for (i = 0; i < nblks; i++) {
255 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
256 mutex_enter(&db->db_mtx);
257 while (db->db_state == DB_READ ||
258 db->db_state == DB_FILL)
259 cv_wait(&db->db_changed, &db->db_mtx);
260 if (db->db_state == DB_UNCACHED)
262 mutex_exit(&db->db_mtx);
264 dmu_buf_rele_array(dbp, nblks, tag);
276 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
277 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
282 err = dnode_hold(os->os, object, FTAG, &dn);
286 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
287 numbufsp, dbpp, DMU_READ_PREFETCH);
289 dnode_rele(dn, FTAG);
295 dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
296 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
298 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
301 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
302 numbufsp, dbpp, DMU_READ_PREFETCH);
308 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
311 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
316 for (i = 0; i < numbufs; i++) {
318 dbuf_rele(dbp[i], tag);
321 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
325 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
331 if (zfs_prefetch_disable)
334 if (len == 0) { /* they're interested in the bonus buffer */
335 dn = os->os->os_meta_dnode;
337 if (object == 0 || object >= DN_MAX_OBJECT)
340 rw_enter(&dn->dn_struct_rwlock, RW_READER);
341 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
342 dbuf_prefetch(dn, blkid);
343 rw_exit(&dn->dn_struct_rwlock);
348 * XXX - Note, if the dnode for the requested object is not
349 * already cached, we will do a *synchronous* read in the
350 * dnode_hold() call. The same is true for any indirects.
352 err = dnode_hold(os->os, object, FTAG, &dn);
356 rw_enter(&dn->dn_struct_rwlock, RW_READER);
357 if (dn->dn_datablkshift) {
358 int blkshift = dn->dn_datablkshift;
359 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
360 P2ALIGN(offset, 1<<blkshift)) >> blkshift;
362 nblks = (offset < dn->dn_datablksz);
366 blkid = dbuf_whichblock(dn, offset);
367 for (i = 0; i < nblks; i++)
368 dbuf_prefetch(dn, blkid+i);
371 rw_exit(&dn->dn_struct_rwlock);
373 dnode_rele(dn, FTAG);
377 * Get the next "chunk" of file data to free. We traverse the file from
378 * the end so that the file gets shorter over time (if we crashes in the
379 * middle, this will leave us in a better state). We find allocated file
380 * data by simply searching the allocated level 1 indirects.
383 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
385 uint64_t len = *start - limit;
387 uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
389 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
391 ASSERT(limit <= *start);
393 if (len <= iblkrange * maxblks) {
397 ASSERT(ISP2(iblkrange));
399 while (*start > limit && blkcnt < maxblks) {
402 /* find next allocated L1 indirect */
403 err = dnode_next_offset(dn,
404 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
406 /* if there are no more, then we are done */
415 /* reset offset to end of "next" block back */
416 *start = P2ALIGN(*start, iblkrange);
426 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
427 uint64_t length, boolean_t free_dnode)
430 uint64_t object_size, start, end, len;
431 boolean_t trunc = (length == DMU_OBJECT_END);
434 align = 1 << dn->dn_datablkshift;
436 object_size = align == 1 ? dn->dn_datablksz :
437 (dn->dn_maxblkid + 1) << dn->dn_datablkshift;
439 end = offset + length;
440 if (trunc || end > object_size)
444 length = end - offset;
448 /* assert(offset <= start) */
449 err = get_next_chunk(dn, &start, offset);
452 len = trunc ? DMU_OBJECT_END : end - start;
454 tx = dmu_tx_create(os);
455 dmu_tx_hold_free(tx, dn->dn_object, start, len);
456 err = dmu_tx_assign(tx, TXG_WAIT);
462 dnode_free_range(dn, start, trunc ? -1 : len, tx);
464 if (start == 0 && free_dnode) {
469 length -= end - start;
478 dmu_free_long_range(objset_t *os, uint64_t object,
479 uint64_t offset, uint64_t length)
484 err = dnode_hold(os->os, object, FTAG, &dn);
487 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
488 dnode_rele(dn, FTAG);
493 dmu_free_object(objset_t *os, uint64_t object)
499 err = dnode_hold_impl(os->os, object, DNODE_MUST_BE_ALLOCATED,
503 if (dn->dn_nlevels == 1) {
504 tx = dmu_tx_create(os);
505 dmu_tx_hold_bonus(tx, object);
506 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END);
507 err = dmu_tx_assign(tx, TXG_WAIT);
509 dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
516 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
518 dnode_rele(dn, FTAG);
523 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
524 uint64_t size, dmu_tx_t *tx)
527 int err = dnode_hold(os->os, object, FTAG, &dn);
530 ASSERT(offset < UINT64_MAX);
531 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
532 dnode_free_range(dn, offset, size, tx);
533 dnode_rele(dn, FTAG);
538 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
539 void *buf, uint32_t flags)
545 err = dnode_hold(os->os, object, FTAG, &dn);
550 * Deal with odd block sizes, where there can't be data past the first
551 * block. If we ever do the tail block optimization, we will need to
552 * handle that here as well.
554 if (dn->dn_datablkshift == 0) {
555 int newsz = offset > dn->dn_datablksz ? 0 :
556 MIN(size, dn->dn_datablksz - offset);
557 bzero((char *)buf + newsz, size - newsz);
562 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
565 * NB: we could do this block-at-a-time, but it's nice
566 * to be reading in parallel.
568 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
569 TRUE, FTAG, &numbufs, &dbp, flags);
573 for (i = 0; i < numbufs; i++) {
576 dmu_buf_t *db = dbp[i];
580 bufoff = offset - db->db_offset;
581 tocpy = (int)MIN(db->db_size - bufoff, size);
583 bcopy((char *)db->db_data + bufoff, buf, tocpy);
587 buf = (char *)buf + tocpy;
589 dmu_buf_rele_array(dbp, numbufs, FTAG);
591 dnode_rele(dn, FTAG);
596 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
597 const void *buf, dmu_tx_t *tx)
605 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
606 FALSE, FTAG, &numbufs, &dbp));
608 for (i = 0; i < numbufs; i++) {
611 dmu_buf_t *db = dbp[i];
615 bufoff = offset - db->db_offset;
616 tocpy = (int)MIN(db->db_size - bufoff, size);
618 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
620 if (tocpy == db->db_size)
621 dmu_buf_will_fill(db, tx);
623 dmu_buf_will_dirty(db, tx);
625 bcopy(buf, (char *)db->db_data + bufoff, tocpy);
627 if (tocpy == db->db_size)
628 dmu_buf_fill_done(db, tx);
632 buf = (char *)buf + tocpy;
634 dmu_buf_rele_array(dbp, numbufs, FTAG);
639 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
645 * NB: we could do this block-at-a-time, but it's nice
646 * to be reading in parallel.
648 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
653 for (i = 0; i < numbufs; i++) {
656 dmu_buf_t *db = dbp[i];
660 bufoff = uio->uio_loffset - db->db_offset;
661 tocpy = (int)MIN(db->db_size - bufoff, size);
663 err = uiomove((char *)db->db_data + bufoff, tocpy,
670 dmu_buf_rele_array(dbp, numbufs, FTAG);
676 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
686 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size,
687 FALSE, FTAG, &numbufs, &dbp);
691 for (i = 0; i < numbufs; i++) {
694 dmu_buf_t *db = dbp[i];
698 bufoff = uio->uio_loffset - db->db_offset;
699 tocpy = (int)MIN(db->db_size - bufoff, size);
701 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
703 if (tocpy == db->db_size)
704 dmu_buf_will_fill(db, tx);
706 dmu_buf_will_dirty(db, tx);
709 * XXX uiomove could block forever (eg. nfs-backed
710 * pages). There needs to be a uiolockdown() function
711 * to lock the pages in memory, so that uiomove won't
714 err = uiomove((char *)db->db_data + bufoff, tocpy,
717 if (tocpy == db->db_size)
718 dmu_buf_fill_done(db, tx);
725 dmu_buf_rele_array(dbp, numbufs, FTAG);
731 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
732 page_t *pp, dmu_tx_t *tx)
741 err = dmu_buf_hold_array(os, object, offset, size,
742 FALSE, FTAG, &numbufs, &dbp);
746 for (i = 0; i < numbufs; i++) {
747 int tocpy, copied, thiscpy;
749 dmu_buf_t *db = dbp[i];
753 ASSERT3U(db->db_size, >=, PAGESIZE);
755 bufoff = offset - db->db_offset;
756 tocpy = (int)MIN(db->db_size - bufoff, size);
758 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
760 if (tocpy == db->db_size)
761 dmu_buf_will_fill(db, tx);
763 dmu_buf_will_dirty(db, tx);
765 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
766 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
767 thiscpy = MIN(PAGESIZE, tocpy - copied);
768 va = zfs_map_page(pp, S_READ);
769 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
770 zfs_unmap_page(pp, va);
775 if (tocpy == db->db_size)
776 dmu_buf_fill_done(db, tx);
781 dmu_buf_rele_array(dbp, numbufs, FTAG);
784 #endif /* !__FreeBSD__ */
788 * Allocate a loaned anonymous arc buffer.
791 dmu_request_arcbuf(dmu_buf_t *handle, int size)
793 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode;
795 return (arc_loan_buf(dn->dn_objset->os_spa, size));
799 * Free a loaned arc buffer.
802 dmu_return_arcbuf(arc_buf_t *buf)
804 arc_return_buf(buf, FTAG);
805 VERIFY(arc_buf_remove_ref(buf, FTAG) == 1);
809 * When possible directly assign passed loaned arc buffer to a dbuf.
810 * If this is not possible copy the contents of passed arc buf via
814 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
817 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode;
819 uint32_t blksz = (uint32_t)arc_buf_size(buf);
822 rw_enter(&dn->dn_struct_rwlock, RW_READER);
823 blkid = dbuf_whichblock(dn, offset);
824 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
825 rw_exit(&dn->dn_struct_rwlock);
827 if (offset == db->db.db_offset && blksz == db->db.db_size) {
828 dbuf_assign_arcbuf(db, buf, tx);
832 ASSERT(dn->dn_objset->os.os == dn->dn_objset);
833 dmu_write(&dn->dn_objset->os, dn->dn_object, offset, blksz,
835 dmu_return_arcbuf(buf);
840 dbuf_dirty_record_t *dr;
847 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
849 blkptr_t *bp = zio->io_bp;
850 dmu_sync_arg_t *in = varg;
851 dbuf_dirty_record_t *dr = in->dr;
852 dmu_buf_impl_t *db = dr->dr_dbuf;
854 if (!BP_IS_HOLE(bp)) {
855 ASSERT(BP_GET_TYPE(bp) == db->db_dnode->dn_type);
856 ASSERT(BP_GET_LEVEL(bp) == 0);
860 * dmu_sync() can compress a block of zeros to a null blkptr
861 * but the block size still needs to be passed through to replay
863 BP_SET_LSIZE(bp, db->db.db_size);
869 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
871 dmu_sync_arg_t *in = varg;
872 dbuf_dirty_record_t *dr = in->dr;
873 dmu_buf_impl_t *db = dr->dr_dbuf;
874 dmu_sync_cb_t *done = in->done;
876 mutex_enter(&db->db_mtx);
877 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
878 dr->dt.dl.dr_overridden_by = *zio->io_bp; /* structure assignment */
879 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
880 BP_ZERO(&dr->dt.dl.dr_overridden_by);
881 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
882 cv_broadcast(&db->db_changed);
883 mutex_exit(&db->db_mtx);
886 done(&(db->db), in->arg);
888 kmem_free(in, sizeof (dmu_sync_arg_t));
892 * Intent log support: sync the block associated with db to disk.
893 * N.B. and XXX: the caller is responsible for making sure that the
894 * data isn't changing while dmu_sync() is writing it.
898 * EEXIST: this txg has already been synced, so there's nothing to to.
899 * The caller should not log the write.
901 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
902 * The caller should not log the write.
904 * EALREADY: this block is already in the process of being synced.
905 * The caller should track its progress (somehow).
907 * EINPROGRESS: the IO has been initiated.
908 * The caller should log this blkptr in the callback.
910 * 0: completed. Sets *bp to the blkptr just written.
911 * The caller should log this blkptr immediately.
914 dmu_sync(zio_t *pio, dmu_buf_t *db_fake,
915 blkptr_t *bp, uint64_t txg, dmu_sync_cb_t *done, void *arg)
917 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
918 objset_impl_t *os = db->db_objset;
919 dsl_pool_t *dp = os->os_dsl_dataset->ds_dir->dd_pool;
920 tx_state_t *tx = &dp->dp_tx;
921 dbuf_dirty_record_t *dr;
924 writeprops_t wp = { 0 };
928 ASSERT(BP_IS_HOLE(bp));
931 dprintf("dmu_sync txg=%llu, s,o,q %llu %llu %llu\n",
932 txg, tx->tx_synced_txg, tx->tx_open_txg, tx->tx_quiesced_txg);
935 * XXX - would be nice if we could do this without suspending...
940 * If this txg already synced, there's nothing to do.
942 if (txg <= tx->tx_synced_txg) {
945 * If we're running ziltest, we need the blkptr regardless.
947 if (txg > spa_freeze_txg(dp->dp_spa)) {
948 /* if db_blkptr == NULL, this was an empty write */
950 *bp = *db->db_blkptr; /* structure assignment */
956 mutex_enter(&db->db_mtx);
958 if (txg == tx->tx_syncing_txg) {
959 while (db->db_data_pending) {
961 * IO is in-progress. Wait for it to finish.
962 * XXX - would be nice to be able to somehow "attach"
963 * this zio to the parent zio passed in.
965 cv_wait(&db->db_changed, &db->db_mtx);
966 if (!db->db_data_pending &&
967 db->db_blkptr && BP_IS_HOLE(db->db_blkptr)) {
969 * IO was compressed away
971 *bp = *db->db_blkptr; /* structure assignment */
972 mutex_exit(&db->db_mtx);
976 ASSERT(db->db_data_pending ||
977 (db->db_blkptr && db->db_blkptr->blk_birth == txg));
980 if (db->db_blkptr && db->db_blkptr->blk_birth == txg) {
982 * IO is already completed.
984 *bp = *db->db_blkptr; /* structure assignment */
985 mutex_exit(&db->db_mtx);
991 dr = db->db_last_dirty;
992 while (dr && dr->dr_txg > txg)
994 if (dr == NULL || dr->dr_txg < txg) {
996 * This dbuf isn't dirty, must have been free_range'd.
997 * There's no need to log writes to freed blocks, so we're done.
999 mutex_exit(&db->db_mtx);
1004 ASSERT(dr->dr_txg == txg);
1005 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) {
1007 * We have already issued a sync write for this buffer.
1009 mutex_exit(&db->db_mtx);
1012 } else if (dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1014 * This buffer has already been synced. It could not
1015 * have been dirtied since, or we would have cleared the state.
1017 *bp = dr->dt.dl.dr_overridden_by; /* structure assignment */
1018 mutex_exit(&db->db_mtx);
1023 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1024 in = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1028 mutex_exit(&db->db_mtx);
1031 zb.zb_objset = os->os_dsl_dataset->ds_object;
1032 zb.zb_object = db->db.db_object;
1033 zb.zb_level = db->db_level;
1034 zb.zb_blkid = db->db_blkid;
1036 wp.wp_type = db->db_dnode->dn_type;
1037 wp.wp_level = db->db_level;
1038 wp.wp_copies = os->os_copies;
1039 wp.wp_dnchecksum = db->db_dnode->dn_checksum;
1040 wp.wp_oschecksum = os->os_checksum;
1041 wp.wp_dncompress = db->db_dnode->dn_compress;
1042 wp.wp_oscompress = os->os_compress;
1044 ASSERT(BP_IS_HOLE(bp));
1046 zio = arc_write(pio, os->os_spa, &wp, DBUF_IS_L2CACHEABLE(db),
1047 txg, bp, dr->dt.dl.dr_data, dmu_sync_ready, dmu_sync_done, in,
1048 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1054 err = zio_wait(zio);
1061 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1067 err = dnode_hold(os->os, object, FTAG, &dn);
1070 err = dnode_set_blksz(dn, size, ibs, tx);
1071 dnode_rele(dn, FTAG);
1076 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1081 /* XXX assumes dnode_hold will not get an i/o error */
1082 (void) dnode_hold(os->os, object, FTAG, &dn);
1083 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
1084 dn->dn_checksum = checksum;
1085 dnode_setdirty(dn, tx);
1086 dnode_rele(dn, FTAG);
1090 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1095 /* XXX assumes dnode_hold will not get an i/o error */
1096 (void) dnode_hold(os->os, object, FTAG, &dn);
1097 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
1098 dn->dn_compress = compress;
1099 dnode_setdirty(dn, tx);
1100 dnode_rele(dn, FTAG);
1104 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1109 err = dnode_hold(os->os, object, FTAG, &dn);
1113 * Sync any current changes before
1114 * we go trundling through the block pointers.
1116 for (i = 0; i < TXG_SIZE; i++) {
1117 if (list_link_active(&dn->dn_dirty_link[i]))
1120 if (i != TXG_SIZE) {
1121 dnode_rele(dn, FTAG);
1122 txg_wait_synced(dmu_objset_pool(os), 0);
1123 err = dnode_hold(os->os, object, FTAG, &dn);
1128 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1129 dnode_rele(dn, FTAG);
1135 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1137 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1138 mutex_enter(&dn->dn_mtx);
1140 doi->doi_data_block_size = dn->dn_datablksz;
1141 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1142 1ULL << dn->dn_indblkshift : 0;
1143 doi->doi_indirection = dn->dn_nlevels;
1144 doi->doi_checksum = dn->dn_checksum;
1145 doi->doi_compress = dn->dn_compress;
1146 doi->doi_physical_blks = (DN_USED_BYTES(dn->dn_phys) +
1147 SPA_MINBLOCKSIZE/2) >> SPA_MINBLOCKSHIFT;
1148 doi->doi_max_block_offset = dn->dn_phys->dn_maxblkid;
1149 doi->doi_type = dn->dn_type;
1150 doi->doi_bonus_size = dn->dn_bonuslen;
1151 doi->doi_bonus_type = dn->dn_bonustype;
1153 mutex_exit(&dn->dn_mtx);
1154 rw_exit(&dn->dn_struct_rwlock);
1158 * Get information on a DMU object.
1159 * If doi is NULL, just indicates whether the object exists.
1162 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1165 int err = dnode_hold(os->os, object, FTAG, &dn);
1171 dmu_object_info_from_dnode(dn, doi);
1173 dnode_rele(dn, FTAG);
1178 * As above, but faster; can be used when you have a held dbuf in hand.
1181 dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi)
1183 dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi);
1187 * Faster still when you only care about the size.
1188 * This is specifically optimized for zfs_getattr().
1191 dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512)
1193 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
1195 *blksize = dn->dn_datablksz;
1196 /* add 1 for dnode space */
1197 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1198 SPA_MINBLOCKSHIFT) + 1;
1202 byteswap_uint64_array(void *vbuf, size_t size)
1204 uint64_t *buf = vbuf;
1205 size_t count = size >> 3;
1208 ASSERT((size & 7) == 0);
1210 for (i = 0; i < count; i++)
1211 buf[i] = BSWAP_64(buf[i]);
1215 byteswap_uint32_array(void *vbuf, size_t size)
1217 uint32_t *buf = vbuf;
1218 size_t count = size >> 2;
1221 ASSERT((size & 3) == 0);
1223 for (i = 0; i < count; i++)
1224 buf[i] = BSWAP_32(buf[i]);
1228 byteswap_uint16_array(void *vbuf, size_t size)
1230 uint16_t *buf = vbuf;
1231 size_t count = size >> 1;
1234 ASSERT((size & 1) == 0);
1236 for (i = 0; i < count; i++)
1237 buf[i] = BSWAP_16(buf[i]);
1242 byteswap_uint8_array(void *vbuf, size_t size)