1789Sahrens /*
2789Sahrens * CDDL HEADER START
3789Sahrens *
4789Sahrens * The contents of this file are subject to the terms of the
51544Seschrock * Common Development and Distribution License (the "License").
61544Seschrock * You may not use this file except in compliance with the License.
7789Sahrens *
8789Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9789Sahrens * or http://www.opensolaris.org/os/licensing.
10789Sahrens * See the License for the specific language governing permissions
11789Sahrens * and limitations under the License.
12789Sahrens *
13789Sahrens * When distributing Covered Code, include this CDDL HEADER in each
14789Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15789Sahrens * If applicable, add the following below this CDDL HEADER, with the
16789Sahrens * fields enclosed by brackets "[]" replaced with your own identifying
17789Sahrens * information: Portions Copyright [yyyy] [name of copyright owner]
18789Sahrens *
19789Sahrens * CDDL HEADER END
20789Sahrens */
21789Sahrens /*
2212123STim.Haley@Sun.COM * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23789Sahrens */
24789Sahrens
25789Sahrens #include <sys/dmu.h>
26789Sahrens #include <sys/dmu_impl.h>
27789Sahrens #include <sys/dmu_tx.h>
28789Sahrens #include <sys/dbuf.h>
29789Sahrens #include <sys/dnode.h>
30789Sahrens #include <sys/zfs_context.h>
31789Sahrens #include <sys/dmu_objset.h>
32789Sahrens #include <sys/dmu_traverse.h>
33789Sahrens #include <sys/dsl_dataset.h>
34789Sahrens #include <sys/dsl_dir.h>
35789Sahrens #include <sys/dsl_pool.h>
362199Sahrens #include <sys/dsl_synctask.h>
372885Sahrens #include <sys/dsl_prop.h>
38789Sahrens #include <sys/dmu_zfetch.h>
39789Sahrens #include <sys/zfs_ioctl.h>
40789Sahrens #include <sys/zap.h>
411544Seschrock #include <sys/zio_checksum.h>
4211935SMark.Shellenbaum@Sun.COM #include <sys/sa.h>
432688Smaybee #ifdef _KERNEL
442688Smaybee #include <sys/vmsystm.h>
457315SJonathan.Adams@Sun.COM #include <sys/zfs_znode.h>
462688Smaybee #endif
47789Sahrens
48789Sahrens const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
49789Sahrens { byteswap_uint8_array, TRUE, "unallocated" },
50789Sahrens { zap_byteswap, TRUE, "object directory" },
51789Sahrens { byteswap_uint64_array, TRUE, "object array" },
52789Sahrens { byteswap_uint8_array, TRUE, "packed nvlist" },
53789Sahrens { byteswap_uint64_array, TRUE, "packed nvlist size" },
5412470SMatthew.Ahrens@Sun.COM { byteswap_uint64_array, TRUE, "bpobj" },
5512470SMatthew.Ahrens@Sun.COM { byteswap_uint64_array, TRUE, "bpobj header" },
56789Sahrens { byteswap_uint64_array, TRUE, "SPA space map header" },
57789Sahrens { byteswap_uint64_array, TRUE, "SPA space map" },
58789Sahrens { byteswap_uint64_array, TRUE, "ZIL intent log" },
59789Sahrens { dnode_buf_byteswap, TRUE, "DMU dnode" },
60789Sahrens { dmu_objset_byteswap, TRUE, "DMU objset" },
61789Sahrens { byteswap_uint64_array, TRUE, "DSL directory" },
62789Sahrens { zap_byteswap, TRUE, "DSL directory child map"},
63789Sahrens { zap_byteswap, TRUE, "DSL dataset snap map" },
64789Sahrens { zap_byteswap, TRUE, "DSL props" },
65789Sahrens { byteswap_uint64_array, TRUE, "DSL dataset" },
66789Sahrens { zfs_znode_byteswap, TRUE, "ZFS znode" },
675331Samw { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" },
68789Sahrens { byteswap_uint8_array, FALSE, "ZFS plain file" },
69789Sahrens { zap_byteswap, TRUE, "ZFS directory" },
70789Sahrens { zap_byteswap, TRUE, "ZFS master node" },
71789Sahrens { zap_byteswap, TRUE, "ZFS delete queue" },
72789Sahrens { byteswap_uint8_array, FALSE, "zvol object" },
73789Sahrens { zap_byteswap, TRUE, "zvol prop" },
74789Sahrens { byteswap_uint8_array, FALSE, "other uint8[]" },
75789Sahrens { byteswap_uint64_array, FALSE, "other uint64[]" },
76789Sahrens { zap_byteswap, TRUE, "other ZAP" },
771544Seschrock { zap_byteswap, TRUE, "persistent error log" },
782926Sek110237 { byteswap_uint8_array, TRUE, "SPA history" },
792926Sek110237 { byteswap_uint64_array, TRUE, "SPA history offsets" },
804543Smarks { zap_byteswap, TRUE, "Pool properties" },
815331Samw { zap_byteswap, TRUE, "DSL permissions" },
825331Samw { zfs_acl_byteswap, TRUE, "ZFS ACL" },
835331Samw { byteswap_uint8_array, TRUE, "ZFS SYSACL" },
845331Samw { byteswap_uint8_array, TRUE, "FUID table" },
855339Smarks { byteswap_uint64_array, TRUE, "FUID table size" },
867046Sahrens { zap_byteswap, TRUE, "DSL dataset next clones"},
8712296SLin.Ling@Sun.COM { zap_byteswap, TRUE, "scan work queue" },
889396SMatthew.Ahrens@Sun.COM { zap_byteswap, TRUE, "ZFS user/group used" },
899396SMatthew.Ahrens@Sun.COM { zap_byteswap, TRUE, "ZFS user/group quota" },
9010242Schris.kirby@sun.com { zap_byteswap, TRUE, "snapshot refcount tags"},
9110922SJeff.Bonwick@Sun.COM { zap_byteswap, TRUE, "DDT ZAP algorithm" },
9210922SJeff.Bonwick@Sun.COM { zap_byteswap, TRUE, "DDT statistics" },
9311935SMark.Shellenbaum@Sun.COM { byteswap_uint8_array, TRUE, "System attributes" },
9411935SMark.Shellenbaum@Sun.COM { zap_byteswap, TRUE, "SA master node" },
9511935SMark.Shellenbaum@Sun.COM { zap_byteswap, TRUE, "SA attr registration" },
9612296SLin.Ling@Sun.COM { zap_byteswap, TRUE, "SA attr layouts" },
9712296SLin.Ling@Sun.COM { zap_byteswap, TRUE, "scan translations" },
9812296SLin.Ling@Sun.COM { byteswap_uint8_array, FALSE, "deduplicated block" },
9912470SMatthew.Ahrens@Sun.COM { zap_byteswap, TRUE, "DSL deadlist map" },
10012470SMatthew.Ahrens@Sun.COM { byteswap_uint64_array, TRUE, "DSL deadlist map hdr" },
10112470SMatthew.Ahrens@Sun.COM { zap_byteswap, TRUE, "DSL dir clones" },
10212470SMatthew.Ahrens@Sun.COM { byteswap_uint64_array, TRUE, "bpobj subobj" },
10312296SLin.Ling@Sun.COM };
104789Sahrens
105789Sahrens int
dmu_buf_hold(objset_t * os,uint64_t object,uint64_t offset,void * tag,dmu_buf_t ** dbp,int flags)1061544Seschrock dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
10712285SJeff.Bonwick@Sun.COM void *tag, dmu_buf_t **dbp, int flags)
108789Sahrens {
109789Sahrens dnode_t *dn;
110789Sahrens uint64_t blkid;
111789Sahrens dmu_buf_impl_t *db;
1121544Seschrock int err;
11312285SJeff.Bonwick@Sun.COM int db_flags = DB_RF_CANFAIL;
11412285SJeff.Bonwick@Sun.COM
11512285SJeff.Bonwick@Sun.COM if (flags & DMU_READ_NO_PREFETCH)
11612285SJeff.Bonwick@Sun.COM db_flags |= DB_RF_NOPREFETCH;
117789Sahrens
11810298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
1191544Seschrock if (err)
1201544Seschrock return (err);
121789Sahrens blkid = dbuf_whichblock(dn, offset);
122789Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER);
1231544Seschrock db = dbuf_hold(dn, blkid, tag);
124789Sahrens rw_exit(&dn->dn_struct_rwlock);
1251544Seschrock if (db == NULL) {
1261544Seschrock err = EIO;
1271544Seschrock } else {
12812285SJeff.Bonwick@Sun.COM err = dbuf_read(db, NULL, db_flags);
1291544Seschrock if (err) {
1301544Seschrock dbuf_rele(db, tag);
1311544Seschrock db = NULL;
1321544Seschrock }
1331544Seschrock }
1341544Seschrock
135789Sahrens dnode_rele(dn, FTAG);
136*12684STom.Erickson@Sun.COM *dbp = &db->db; /* NULL db plus first field offset is NULL */
1371544Seschrock return (err);
138789Sahrens }
139789Sahrens
140789Sahrens int
dmu_bonus_max(void)141789Sahrens dmu_bonus_max(void)
142789Sahrens {
143789Sahrens return (DN_MAX_BONUSLEN);
144789Sahrens }
145789Sahrens
1464944Smaybee int
dmu_set_bonus(dmu_buf_t * db_fake,int newsize,dmu_tx_t * tx)147*12684STom.Erickson@Sun.COM dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
1484944Smaybee {
149*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
150*12684STom.Erickson@Sun.COM dnode_t *dn;
151*12684STom.Erickson@Sun.COM int error;
152*12684STom.Erickson@Sun.COM
153*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
154*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
1554944Smaybee
156*12684STom.Erickson@Sun.COM if (dn->dn_bonus != db) {
157*12684STom.Erickson@Sun.COM error = EINVAL;
158*12684STom.Erickson@Sun.COM } else if (newsize < 0 || newsize > db_fake->db_size) {
159*12684STom.Erickson@Sun.COM error = EINVAL;
160*12684STom.Erickson@Sun.COM } else {
161*12684STom.Erickson@Sun.COM dnode_setbonuslen(dn, newsize, tx);
162*12684STom.Erickson@Sun.COM error = 0;
163*12684STom.Erickson@Sun.COM }
164*12684STom.Erickson@Sun.COM
165*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
166*12684STom.Erickson@Sun.COM return (error);
1674944Smaybee }
1684944Smaybee
16911935SMark.Shellenbaum@Sun.COM int
dmu_set_bonustype(dmu_buf_t * db_fake,dmu_object_type_t type,dmu_tx_t * tx)170*12684STom.Erickson@Sun.COM dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
17111935SMark.Shellenbaum@Sun.COM {
172*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
173*12684STom.Erickson@Sun.COM dnode_t *dn;
174*12684STom.Erickson@Sun.COM int error;
175*12684STom.Erickson@Sun.COM
176*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
177*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
178*12684STom.Erickson@Sun.COM
179*12684STom.Erickson@Sun.COM if (type > DMU_OT_NUMTYPES) {
180*12684STom.Erickson@Sun.COM error = EINVAL;
181*12684STom.Erickson@Sun.COM } else if (dn->dn_bonus != db) {
182*12684STom.Erickson@Sun.COM error = EINVAL;
183*12684STom.Erickson@Sun.COM } else {
184*12684STom.Erickson@Sun.COM dnode_setbonus_type(dn, type, tx);
185*12684STom.Erickson@Sun.COM error = 0;
186*12684STom.Erickson@Sun.COM }
18711935SMark.Shellenbaum@Sun.COM
188*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
189*12684STom.Erickson@Sun.COM return (error);
190*12684STom.Erickson@Sun.COM }
19111935SMark.Shellenbaum@Sun.COM
192*12684STom.Erickson@Sun.COM dmu_object_type_t
dmu_get_bonustype(dmu_buf_t * db_fake)193*12684STom.Erickson@Sun.COM dmu_get_bonustype(dmu_buf_t *db_fake)
194*12684STom.Erickson@Sun.COM {
195*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
196*12684STom.Erickson@Sun.COM dnode_t *dn;
197*12684STom.Erickson@Sun.COM dmu_object_type_t type;
19811935SMark.Shellenbaum@Sun.COM
199*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
200*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
201*12684STom.Erickson@Sun.COM type = dn->dn_bonustype;
202*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
203*12684STom.Erickson@Sun.COM
204*12684STom.Erickson@Sun.COM return (type);
20511935SMark.Shellenbaum@Sun.COM }
20611935SMark.Shellenbaum@Sun.COM
20711935SMark.Shellenbaum@Sun.COM int
dmu_rm_spill(objset_t * os,uint64_t object,dmu_tx_t * tx)20811935SMark.Shellenbaum@Sun.COM dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
20911935SMark.Shellenbaum@Sun.COM {
21011935SMark.Shellenbaum@Sun.COM dnode_t *dn;
21111935SMark.Shellenbaum@Sun.COM int error;
21211935SMark.Shellenbaum@Sun.COM
21311935SMark.Shellenbaum@Sun.COM error = dnode_hold(os, object, FTAG, &dn);
21411935SMark.Shellenbaum@Sun.COM dbuf_rm_spill(dn, tx);
21512178SMark.Shellenbaum@Sun.COM rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
21612178SMark.Shellenbaum@Sun.COM dnode_rm_spill(dn, tx);
21712178SMark.Shellenbaum@Sun.COM rw_exit(&dn->dn_struct_rwlock);
21811935SMark.Shellenbaum@Sun.COM dnode_rele(dn, FTAG);
21911935SMark.Shellenbaum@Sun.COM return (error);
22011935SMark.Shellenbaum@Sun.COM }
22111935SMark.Shellenbaum@Sun.COM
222789Sahrens /*
2231544Seschrock * returns ENOENT, EIO, or 0.
224789Sahrens */
2251544Seschrock int
dmu_bonus_hold(objset_t * os,uint64_t object,void * tag,dmu_buf_t ** dbp)2261544Seschrock dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
227789Sahrens {
2281544Seschrock dnode_t *dn;
229789Sahrens dmu_buf_impl_t *db;
2304944Smaybee int error;
231789Sahrens
23210298SMatthew.Ahrens@Sun.COM error = dnode_hold(os, object, FTAG, &dn);
2334944Smaybee if (error)
2344944Smaybee return (error);
235789Sahrens
2361544Seschrock rw_enter(&dn->dn_struct_rwlock, RW_READER);
2371544Seschrock if (dn->dn_bonus == NULL) {
238789Sahrens rw_exit(&dn->dn_struct_rwlock);
2391544Seschrock rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
2401544Seschrock if (dn->dn_bonus == NULL)
2414944Smaybee dbuf_create_bonus(dn);
242789Sahrens }
2431544Seschrock db = dn->dn_bonus;
2444944Smaybee
2454944Smaybee /* as long as the bonus buf is held, the dnode will be held */
246*12684STom.Erickson@Sun.COM if (refcount_add(&db->db_holds, tag) == 1) {
2474944Smaybee VERIFY(dnode_add_ref(dn, db));
248*12684STom.Erickson@Sun.COM (void) atomic_inc_32_nv(&dn->dn_dbufs_count);
249*12684STom.Erickson@Sun.COM }
250*12684STom.Erickson@Sun.COM
251*12684STom.Erickson@Sun.COM /*
252*12684STom.Erickson@Sun.COM * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
253*12684STom.Erickson@Sun.COM * hold and incrementing the dbuf count to ensure that dnode_move() sees
254*12684STom.Erickson@Sun.COM * a dnode hold for every dbuf.
255*12684STom.Erickson@Sun.COM */
256*12684STom.Erickson@Sun.COM rw_exit(&dn->dn_struct_rwlock);
2574944Smaybee
258789Sahrens dnode_rele(dn, FTAG);
2591544Seschrock
26012285SJeff.Bonwick@Sun.COM VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
2611544Seschrock
2621544Seschrock *dbp = &db->db;
2631544Seschrock return (0);
264789Sahrens }
265789Sahrens
2662391Smaybee /*
26711935SMark.Shellenbaum@Sun.COM * returns ENOENT, EIO, or 0.
26811935SMark.Shellenbaum@Sun.COM *
26911935SMark.Shellenbaum@Sun.COM * This interface will allocate a blank spill dbuf when a spill blk
27011935SMark.Shellenbaum@Sun.COM * doesn't already exist on the dnode.
27111935SMark.Shellenbaum@Sun.COM *
27211935SMark.Shellenbaum@Sun.COM * if you only want to find an already existing spill db, then
27311935SMark.Shellenbaum@Sun.COM * dmu_spill_hold_existing() should be used.
27411935SMark.Shellenbaum@Sun.COM */
27511935SMark.Shellenbaum@Sun.COM int
dmu_spill_hold_by_dnode(dnode_t * dn,uint32_t flags,void * tag,dmu_buf_t ** dbp)27611935SMark.Shellenbaum@Sun.COM dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
27711935SMark.Shellenbaum@Sun.COM {
27811935SMark.Shellenbaum@Sun.COM dmu_buf_impl_t *db = NULL;
27911935SMark.Shellenbaum@Sun.COM int err;
28011935SMark.Shellenbaum@Sun.COM
28111935SMark.Shellenbaum@Sun.COM if ((flags & DB_RF_HAVESTRUCT) == 0)
28211935SMark.Shellenbaum@Sun.COM rw_enter(&dn->dn_struct_rwlock, RW_READER);
28311935SMark.Shellenbaum@Sun.COM
28411935SMark.Shellenbaum@Sun.COM db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
28511935SMark.Shellenbaum@Sun.COM
28611935SMark.Shellenbaum@Sun.COM if ((flags & DB_RF_HAVESTRUCT) == 0)
28711935SMark.Shellenbaum@Sun.COM rw_exit(&dn->dn_struct_rwlock);
28811935SMark.Shellenbaum@Sun.COM
28911935SMark.Shellenbaum@Sun.COM ASSERT(db != NULL);
29012493SMark.Shellenbaum@Oracle.COM err = dbuf_read(db, NULL, flags);
29112493SMark.Shellenbaum@Oracle.COM if (err == 0)
29212493SMark.Shellenbaum@Oracle.COM *dbp = &db->db;
29312493SMark.Shellenbaum@Oracle.COM else
29412493SMark.Shellenbaum@Oracle.COM dbuf_rele(db, tag);
29511935SMark.Shellenbaum@Sun.COM return (err);
29611935SMark.Shellenbaum@Sun.COM }
29711935SMark.Shellenbaum@Sun.COM
29811935SMark.Shellenbaum@Sun.COM int
dmu_spill_hold_existing(dmu_buf_t * bonus,void * tag,dmu_buf_t ** dbp)29911935SMark.Shellenbaum@Sun.COM dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
30011935SMark.Shellenbaum@Sun.COM {
301*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
302*12684STom.Erickson@Sun.COM dnode_t *dn;
30311935SMark.Shellenbaum@Sun.COM int err;
30411935SMark.Shellenbaum@Sun.COM
305*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
306*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
307*12684STom.Erickson@Sun.COM
308*12684STom.Erickson@Sun.COM if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
309*12684STom.Erickson@Sun.COM err = EINVAL;
310*12684STom.Erickson@Sun.COM } else {
311*12684STom.Erickson@Sun.COM rw_enter(&dn->dn_struct_rwlock, RW_READER);
31211935SMark.Shellenbaum@Sun.COM
313*12684STom.Erickson@Sun.COM if (!dn->dn_have_spill) {
314*12684STom.Erickson@Sun.COM err = ENOENT;
315*12684STom.Erickson@Sun.COM } else {
316*12684STom.Erickson@Sun.COM err = dmu_spill_hold_by_dnode(dn,
317*12684STom.Erickson@Sun.COM DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
318*12684STom.Erickson@Sun.COM }
319*12684STom.Erickson@Sun.COM
32011935SMark.Shellenbaum@Sun.COM rw_exit(&dn->dn_struct_rwlock);
32111935SMark.Shellenbaum@Sun.COM }
322*12684STom.Erickson@Sun.COM
323*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
32411935SMark.Shellenbaum@Sun.COM return (err);
32511935SMark.Shellenbaum@Sun.COM }
32611935SMark.Shellenbaum@Sun.COM
32711935SMark.Shellenbaum@Sun.COM int
dmu_spill_hold_by_bonus(dmu_buf_t * bonus,void * tag,dmu_buf_t ** dbp)32811935SMark.Shellenbaum@Sun.COM dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
32911935SMark.Shellenbaum@Sun.COM {
330*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
331*12684STom.Erickson@Sun.COM dnode_t *dn;
332*12684STom.Erickson@Sun.COM int err;
333*12684STom.Erickson@Sun.COM
334*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
335*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
336*12684STom.Erickson@Sun.COM err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
337*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
338*12684STom.Erickson@Sun.COM
339*12684STom.Erickson@Sun.COM return (err);
34011935SMark.Shellenbaum@Sun.COM }
34111935SMark.Shellenbaum@Sun.COM
34211935SMark.Shellenbaum@Sun.COM /*
3432391Smaybee * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
3442391Smaybee * to take a held dnode rather than <os, object> -- the lookup is wasteful,
3452391Smaybee * and can induce severe lock contention when writing to several files
3462391Smaybee * whose dnodes are in the same block.
3472391Smaybee */
3482391Smaybee static int
dmu_buf_hold_array_by_dnode(dnode_t * dn,uint64_t offset,uint64_t length,int read,void * tag,int * numbufsp,dmu_buf_t *** dbpp,uint32_t flags)3499512SNeil.Perrin@Sun.COM dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
3509512SNeil.Perrin@Sun.COM int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
351789Sahrens {
3527468SMark.Maybee@Sun.COM dsl_pool_t *dp = NULL;
353789Sahrens dmu_buf_t **dbp;
354789Sahrens uint64_t blkid, nblks, i;
3559512SNeil.Perrin@Sun.COM uint32_t dbuf_flags;
3561544Seschrock int err;
3571544Seschrock zio_t *zio;
3587468SMark.Maybee@Sun.COM hrtime_t start;
3591544Seschrock
3601544Seschrock ASSERT(length <= DMU_MAX_ACCESS);
361789Sahrens
36210209SMark.Musante@Sun.COM dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
3639512SNeil.Perrin@Sun.COM if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
3649512SNeil.Perrin@Sun.COM dbuf_flags |= DB_RF_NOPREFETCH;
3651544Seschrock
366789Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER);
367789Sahrens if (dn->dn_datablkshift) {
368789Sahrens int blkshift = dn->dn_datablkshift;
369789Sahrens nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
3703912Slling P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
371789Sahrens } else {
3723713Sahrens if (offset + length > dn->dn_datablksz) {
3733713Sahrens zfs_panic_recover("zfs: accessing past end of object "
3743713Sahrens "%llx/%llx (size=%u access=%llu+%llu)",
3753713Sahrens (longlong_t)dn->dn_objset->
3763713Sahrens os_dsl_dataset->ds_object,
3773713Sahrens (longlong_t)dn->dn_object, dn->dn_datablksz,
3783713Sahrens (longlong_t)offset, (longlong_t)length);
37910209SMark.Musante@Sun.COM rw_exit(&dn->dn_struct_rwlock);
3803713Sahrens return (EIO);
3813713Sahrens }
382789Sahrens nblks = 1;
383789Sahrens }
3841544Seschrock dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
385789Sahrens
3867468SMark.Maybee@Sun.COM if (dn->dn_objset->os_dsl_dataset)
3877468SMark.Maybee@Sun.COM dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
3887468SMark.Maybee@Sun.COM if (dp && dsl_pool_sync_context(dp))
3897468SMark.Maybee@Sun.COM start = gethrtime();
3907754SJeff.Bonwick@Sun.COM zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
391789Sahrens blkid = dbuf_whichblock(dn, offset);
392789Sahrens for (i = 0; i < nblks; i++) {
3931544Seschrock dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
3941544Seschrock if (db == NULL) {
3951544Seschrock rw_exit(&dn->dn_struct_rwlock);
3961544Seschrock dmu_buf_rele_array(dbp, nblks, tag);
3971544Seschrock zio_nowait(zio);
3981544Seschrock return (EIO);
3991544Seschrock }
4001544Seschrock /* initiate async i/o */
4012391Smaybee if (read) {
4029512SNeil.Perrin@Sun.COM (void) dbuf_read(db, zio, dbuf_flags);
4031544Seschrock }
4041544Seschrock dbp[i] = &db->db;
405789Sahrens }
406789Sahrens rw_exit(&dn->dn_struct_rwlock);
407789Sahrens
4081544Seschrock /* wait for async i/o */
4091544Seschrock err = zio_wait(zio);
4107468SMark.Maybee@Sun.COM /* track read overhead when we are in sync context */
4117468SMark.Maybee@Sun.COM if (dp && dsl_pool_sync_context(dp))
4127468SMark.Maybee@Sun.COM dp->dp_read_overhead += gethrtime() - start;
4131544Seschrock if (err) {
4141544Seschrock dmu_buf_rele_array(dbp, nblks, tag);
4151544Seschrock return (err);
416789Sahrens }
417789Sahrens
4181544Seschrock /* wait for other io to complete */
4191544Seschrock if (read) {
4201544Seschrock for (i = 0; i < nblks; i++) {
4211544Seschrock dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
4221544Seschrock mutex_enter(&db->db_mtx);
4231544Seschrock while (db->db_state == DB_READ ||
4241544Seschrock db->db_state == DB_FILL)
4251544Seschrock cv_wait(&db->db_changed, &db->db_mtx);
4261544Seschrock if (db->db_state == DB_UNCACHED)
4271544Seschrock err = EIO;
4281544Seschrock mutex_exit(&db->db_mtx);
4291544Seschrock if (err) {
4301544Seschrock dmu_buf_rele_array(dbp, nblks, tag);
4311544Seschrock return (err);
4321544Seschrock }
4331544Seschrock }
4341544Seschrock }
435789Sahrens
4361544Seschrock *numbufsp = nblks;
4371544Seschrock *dbpp = dbp;
4381544Seschrock return (0);
439789Sahrens }
440789Sahrens
4412885Sahrens static int
dmu_buf_hold_array(objset_t * os,uint64_t object,uint64_t offset,uint64_t length,int read,void * tag,int * numbufsp,dmu_buf_t *** dbpp)4422391Smaybee dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
4432391Smaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
4442391Smaybee {
4452391Smaybee dnode_t *dn;
4462391Smaybee int err;
4472391Smaybee
44810298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
4492391Smaybee if (err)
4502391Smaybee return (err);
4512391Smaybee
4522391Smaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
4539512SNeil.Perrin@Sun.COM numbufsp, dbpp, DMU_READ_PREFETCH);
4542391Smaybee
4552391Smaybee dnode_rele(dn, FTAG);
4562391Smaybee
4572391Smaybee return (err);
4582391Smaybee }
4592391Smaybee
4602391Smaybee int
dmu_buf_hold_array_by_bonus(dmu_buf_t * db_fake,uint64_t offset,uint64_t length,int read,void * tag,int * numbufsp,dmu_buf_t *** dbpp)461*12684STom.Erickson@Sun.COM dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
4622391Smaybee uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
4632391Smaybee {
464*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
465*12684STom.Erickson@Sun.COM dnode_t *dn;
4662391Smaybee int err;
4672391Smaybee
468*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
469*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
4702391Smaybee err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
4719512SNeil.Perrin@Sun.COM numbufsp, dbpp, DMU_READ_PREFETCH);
472*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
4732391Smaybee
4742391Smaybee return (err);
4752391Smaybee }
4762391Smaybee
477789Sahrens void
dmu_buf_rele_array(dmu_buf_t ** dbp_fake,int numbufs,void * tag)4781544Seschrock dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
479789Sahrens {
480789Sahrens int i;
481789Sahrens dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
482789Sahrens
483789Sahrens if (numbufs == 0)
484789Sahrens return;
485789Sahrens
4861544Seschrock for (i = 0; i < numbufs; i++) {
4871544Seschrock if (dbp[i])
4881544Seschrock dbuf_rele(dbp[i], tag);
4891544Seschrock }
490789Sahrens
491789Sahrens kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
492789Sahrens }
493789Sahrens
494789Sahrens void
dmu_prefetch(objset_t * os,uint64_t object,uint64_t offset,uint64_t len)495789Sahrens dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
496789Sahrens {
497789Sahrens dnode_t *dn;
498789Sahrens uint64_t blkid;
4991544Seschrock int nblks, i, err;
500789Sahrens
5012986Sek110237 if (zfs_prefetch_disable)
5022986Sek110237 return;
5032986Sek110237
504789Sahrens if (len == 0) { /* they're interested in the bonus buffer */
505*12684STom.Erickson@Sun.COM dn = DMU_META_DNODE(os);
506789Sahrens
507789Sahrens if (object == 0 || object >= DN_MAX_OBJECT)
508789Sahrens return;
509789Sahrens
510789Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER);
511789Sahrens blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
512789Sahrens dbuf_prefetch(dn, blkid);
513789Sahrens rw_exit(&dn->dn_struct_rwlock);
514789Sahrens return;
515789Sahrens }
516789Sahrens
517789Sahrens /*
518789Sahrens * XXX - Note, if the dnode for the requested object is not
519789Sahrens * already cached, we will do a *synchronous* read in the
520789Sahrens * dnode_hold() call. The same is true for any indirects.
521789Sahrens */
52210298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
5231544Seschrock if (err != 0)
524789Sahrens return;
525789Sahrens
526789Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER);
527789Sahrens if (dn->dn_datablkshift) {
528789Sahrens int blkshift = dn->dn_datablkshift;
529789Sahrens nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
5303912Slling P2ALIGN(offset, 1<<blkshift)) >> blkshift;
531789Sahrens } else {
532789Sahrens nblks = (offset < dn->dn_datablksz);
533789Sahrens }
534789Sahrens
535789Sahrens if (nblks != 0) {
536789Sahrens blkid = dbuf_whichblock(dn, offset);
537789Sahrens for (i = 0; i < nblks; i++)
538789Sahrens dbuf_prefetch(dn, blkid+i);
539789Sahrens }
540789Sahrens
541789Sahrens rw_exit(&dn->dn_struct_rwlock);
542789Sahrens
543789Sahrens dnode_rele(dn, FTAG);
544789Sahrens }
545789Sahrens
5469950SMark.Maybee@Sun.COM /*
5479950SMark.Maybee@Sun.COM * Get the next "chunk" of file data to free. We traverse the file from
5489950SMark.Maybee@Sun.COM * the end so that the file gets shorter over time (if we crashes in the
5499950SMark.Maybee@Sun.COM * middle, this will leave us in a better state). We find allocated file
5509950SMark.Maybee@Sun.COM * data by simply searching the allocated level 1 indirects.
5519950SMark.Maybee@Sun.COM */
5526992Smaybee static int
get_next_chunk(dnode_t * dn,uint64_t * start,uint64_t limit)5539950SMark.Maybee@Sun.COM get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
5546992Smaybee {
5559950SMark.Maybee@Sun.COM uint64_t len = *start - limit;
5569950SMark.Maybee@Sun.COM uint64_t blkcnt = 0;
5579950SMark.Maybee@Sun.COM uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
5589950SMark.Maybee@Sun.COM uint64_t iblkrange =
5597385SMark.Maybee@Sun.COM dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
5606992Smaybee
5619950SMark.Maybee@Sun.COM ASSERT(limit <= *start);
5626992Smaybee
5639950SMark.Maybee@Sun.COM if (len <= iblkrange * maxblks) {
5649950SMark.Maybee@Sun.COM *start = limit;
5656992Smaybee return (0);
5666992Smaybee }
5679950SMark.Maybee@Sun.COM ASSERT(ISP2(iblkrange));
5687385SMark.Maybee@Sun.COM
5699950SMark.Maybee@Sun.COM while (*start > limit && blkcnt < maxblks) {
5707385SMark.Maybee@Sun.COM int err;
5716992Smaybee
5729950SMark.Maybee@Sun.COM /* find next allocated L1 indirect */
5736992Smaybee err = dnode_next_offset(dn,
5749950SMark.Maybee@Sun.COM DNODE_FIND_BACKWARDS, start, 2, 1, 0);
5756992Smaybee
5769950SMark.Maybee@Sun.COM /* if there are no more, then we are done */
5779950SMark.Maybee@Sun.COM if (err == ESRCH) {
5789950SMark.Maybee@Sun.COM *start = limit;
5796992Smaybee return (0);
5809950SMark.Maybee@Sun.COM } else if (err) {
5819950SMark.Maybee@Sun.COM return (err);
5826992Smaybee }
5839950SMark.Maybee@Sun.COM blkcnt += 1;
5846992Smaybee
5859950SMark.Maybee@Sun.COM /* reset offset to end of "next" block back */
5869950SMark.Maybee@Sun.COM *start = P2ALIGN(*start, iblkrange);
5879950SMark.Maybee@Sun.COM if (*start <= limit)
5889950SMark.Maybee@Sun.COM *start = limit;
5899950SMark.Maybee@Sun.COM else
5909950SMark.Maybee@Sun.COM *start -= 1;
5916992Smaybee }
5926992Smaybee return (0);
5936992Smaybee }
5946992Smaybee
5956992Smaybee static int
dmu_free_long_range_impl(objset_t * os,dnode_t * dn,uint64_t offset,uint64_t length,boolean_t free_dnode)5966992Smaybee dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
5976992Smaybee uint64_t length, boolean_t free_dnode)
5986992Smaybee {
5996992Smaybee dmu_tx_t *tx;
6006992Smaybee uint64_t object_size, start, end, len;
6016992Smaybee boolean_t trunc = (length == DMU_OBJECT_END);
6026992Smaybee int align, err;
6036992Smaybee
6046992Smaybee align = 1 << dn->dn_datablkshift;
6056992Smaybee ASSERT(align > 0);
6066992Smaybee object_size = align == 1 ? dn->dn_datablksz :
6076992Smaybee (dn->dn_maxblkid + 1) << dn->dn_datablkshift;
6086992Smaybee
6099396SMatthew.Ahrens@Sun.COM end = offset + length;
6109396SMatthew.Ahrens@Sun.COM if (trunc || end > object_size)
6116992Smaybee end = object_size;
6126992Smaybee if (end <= offset)
6136992Smaybee return (0);
6146992Smaybee length = end - offset;
6156992Smaybee
6166992Smaybee while (length) {
6176992Smaybee start = end;
6189396SMatthew.Ahrens@Sun.COM /* assert(offset <= start) */
6196992Smaybee err = get_next_chunk(dn, &start, offset);
6206992Smaybee if (err)
6216992Smaybee return (err);
6226992Smaybee len = trunc ? DMU_OBJECT_END : end - start;
6236992Smaybee
6246992Smaybee tx = dmu_tx_create(os);
6256992Smaybee dmu_tx_hold_free(tx, dn->dn_object, start, len);
6266992Smaybee err = dmu_tx_assign(tx, TXG_WAIT);
6276992Smaybee if (err) {
6286992Smaybee dmu_tx_abort(tx);
6296992Smaybee return (err);
6306992Smaybee }
6316992Smaybee
6326992Smaybee dnode_free_range(dn, start, trunc ? -1 : len, tx);
6336992Smaybee
6347385SMark.Maybee@Sun.COM if (start == 0 && free_dnode) {
6357385SMark.Maybee@Sun.COM ASSERT(trunc);
6366992Smaybee dnode_free(dn, tx);
6377385SMark.Maybee@Sun.COM }
6386992Smaybee
6396992Smaybee length -= end - start;
6406992Smaybee
6416992Smaybee dmu_tx_commit(tx);
6426992Smaybee end = start;
6436992Smaybee }
6446992Smaybee return (0);
6456992Smaybee }
6466992Smaybee
6476992Smaybee int
dmu_free_long_range(objset_t * os,uint64_t object,uint64_t offset,uint64_t length)6486992Smaybee dmu_free_long_range(objset_t *os, uint64_t object,
6496992Smaybee uint64_t offset, uint64_t length)
6506992Smaybee {
6516992Smaybee dnode_t *dn;
6526992Smaybee int err;
6536992Smaybee
65410298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
6556992Smaybee if (err != 0)
6566992Smaybee return (err);
6576992Smaybee err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
6586992Smaybee dnode_rele(dn, FTAG);
6596992Smaybee return (err);
6606992Smaybee }
6616992Smaybee
6626992Smaybee int
dmu_free_object(objset_t * os,uint64_t object)6636992Smaybee dmu_free_object(objset_t *os, uint64_t object)
6646992Smaybee {
6656992Smaybee dnode_t *dn;
6666992Smaybee dmu_tx_t *tx;
6676992Smaybee int err;
6686992Smaybee
66910298SMatthew.Ahrens@Sun.COM err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
6706992Smaybee FTAG, &dn);
6716992Smaybee if (err != 0)
6726992Smaybee return (err);
6736992Smaybee if (dn->dn_nlevels == 1) {
6746992Smaybee tx = dmu_tx_create(os);
6756992Smaybee dmu_tx_hold_bonus(tx, object);
6766992Smaybee dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END);
6776992Smaybee err = dmu_tx_assign(tx, TXG_WAIT);
6786992Smaybee if (err == 0) {
6796992Smaybee dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
6806992Smaybee dnode_free(dn, tx);
6816992Smaybee dmu_tx_commit(tx);
6826992Smaybee } else {
6836992Smaybee dmu_tx_abort(tx);
6846992Smaybee }
6856992Smaybee } else {
6866992Smaybee err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
6876992Smaybee }
6886992Smaybee dnode_rele(dn, FTAG);
6896992Smaybee return (err);
6906992Smaybee }
6916992Smaybee
6921544Seschrock int
dmu_free_range(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,dmu_tx_t * tx)693789Sahrens dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
694789Sahrens uint64_t size, dmu_tx_t *tx)
695789Sahrens {
6961544Seschrock dnode_t *dn;
69710298SMatthew.Ahrens@Sun.COM int err = dnode_hold(os, object, FTAG, &dn);
6981544Seschrock if (err)
6991544Seschrock return (err);
700789Sahrens ASSERT(offset < UINT64_MAX);
701789Sahrens ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
702789Sahrens dnode_free_range(dn, offset, size, tx);
703789Sahrens dnode_rele(dn, FTAG);
7041544Seschrock return (0);
705789Sahrens }
706789Sahrens
7071544Seschrock int
dmu_read(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,void * buf,uint32_t flags)7081544Seschrock dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
7099512SNeil.Perrin@Sun.COM void *buf, uint32_t flags)
710789Sahrens {
711789Sahrens dnode_t *dn;
712789Sahrens dmu_buf_t **dbp;
71310209SMark.Musante@Sun.COM int numbufs, err;
714789Sahrens
71510298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
7161544Seschrock if (err)
7171544Seschrock return (err);
7183638Sbillm
7193638Sbillm /*
7203638Sbillm * Deal with odd block sizes, where there can't be data past the first
7213638Sbillm * block. If we ever do the tail block optimization, we will need to
7223638Sbillm * handle that here as well.
7233638Sbillm */
72410209SMark.Musante@Sun.COM if (dn->dn_maxblkid == 0) {
725789Sahrens int newsz = offset > dn->dn_datablksz ? 0 :
726789Sahrens MIN(size, dn->dn_datablksz - offset);
727789Sahrens bzero((char *)buf + newsz, size - newsz);
728789Sahrens size = newsz;
729789Sahrens }
730789Sahrens
731789Sahrens while (size > 0) {
732789Sahrens uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
73310209SMark.Musante@Sun.COM int i;
734789Sahrens
735789Sahrens /*
736789Sahrens * NB: we could do this block-at-a-time, but it's nice
737789Sahrens * to be reading in parallel.
738789Sahrens */
7392885Sahrens err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
7409512SNeil.Perrin@Sun.COM TRUE, FTAG, &numbufs, &dbp, flags);
7411544Seschrock if (err)
7424944Smaybee break;
743789Sahrens
744789Sahrens for (i = 0; i < numbufs; i++) {
745789Sahrens int tocpy;
746789Sahrens int bufoff;
747789Sahrens dmu_buf_t *db = dbp[i];
748789Sahrens
749789Sahrens ASSERT(size > 0);
750789Sahrens
751789Sahrens bufoff = offset - db->db_offset;
752789Sahrens tocpy = (int)MIN(db->db_size - bufoff, size);
753789Sahrens
754789Sahrens bcopy((char *)db->db_data + bufoff, buf, tocpy);
755789Sahrens
756789Sahrens offset += tocpy;
757789Sahrens size -= tocpy;
758789Sahrens buf = (char *)buf + tocpy;
759789Sahrens }
7601544Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG);
761789Sahrens }
7622885Sahrens dnode_rele(dn, FTAG);
7634944Smaybee return (err);
764789Sahrens }
765789Sahrens
766789Sahrens void
dmu_write(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,const void * buf,dmu_tx_t * tx)767789Sahrens dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
768789Sahrens const void *buf, dmu_tx_t *tx)
769789Sahrens {
770789Sahrens dmu_buf_t **dbp;
771789Sahrens int numbufs, i;
772789Sahrens
7732391Smaybee if (size == 0)
7742391Smaybee return;
7752391Smaybee
7761544Seschrock VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
7771544Seschrock FALSE, FTAG, &numbufs, &dbp));
778789Sahrens
779789Sahrens for (i = 0; i < numbufs; i++) {
780789Sahrens int tocpy;
781789Sahrens int bufoff;
782789Sahrens dmu_buf_t *db = dbp[i];
783789Sahrens
784789Sahrens ASSERT(size > 0);
785789Sahrens
786789Sahrens bufoff = offset - db->db_offset;
787789Sahrens tocpy = (int)MIN(db->db_size - bufoff, size);
788789Sahrens
789789Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
790789Sahrens
791789Sahrens if (tocpy == db->db_size)
792789Sahrens dmu_buf_will_fill(db, tx);
793789Sahrens else
794789Sahrens dmu_buf_will_dirty(db, tx);
795789Sahrens
796789Sahrens bcopy(buf, (char *)db->db_data + bufoff, tocpy);
797789Sahrens
798789Sahrens if (tocpy == db->db_size)
799789Sahrens dmu_buf_fill_done(db, tx);
800789Sahrens
801789Sahrens offset += tocpy;
802789Sahrens size -= tocpy;
803789Sahrens buf = (char *)buf + tocpy;
804789Sahrens }
8051544Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG);
806789Sahrens }
807789Sahrens
8087872STim.Haley@Sun.COM void
dmu_prealloc(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,dmu_tx_t * tx)8097872STim.Haley@Sun.COM dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
8107872STim.Haley@Sun.COM dmu_tx_t *tx)
8117872STim.Haley@Sun.COM {
8127872STim.Haley@Sun.COM dmu_buf_t **dbp;
8137872STim.Haley@Sun.COM int numbufs, i;
8147872STim.Haley@Sun.COM
8157872STim.Haley@Sun.COM if (size == 0)
8167872STim.Haley@Sun.COM return;
8177872STim.Haley@Sun.COM
8187872STim.Haley@Sun.COM VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
8197872STim.Haley@Sun.COM FALSE, FTAG, &numbufs, &dbp));
8207872STim.Haley@Sun.COM
8217872STim.Haley@Sun.COM for (i = 0; i < numbufs; i++) {
8227872STim.Haley@Sun.COM dmu_buf_t *db = dbp[i];
8237872STim.Haley@Sun.COM
8247872STim.Haley@Sun.COM dmu_buf_will_not_fill(db, tx);
8257872STim.Haley@Sun.COM }
8267872STim.Haley@Sun.COM dmu_buf_rele_array(dbp, numbufs, FTAG);
8277872STim.Haley@Sun.COM }
8287872STim.Haley@Sun.COM
82911539SChunli.Zhang@Sun.COM /*
83011539SChunli.Zhang@Sun.COM * DMU support for xuio
83111539SChunli.Zhang@Sun.COM */
83211539SChunli.Zhang@Sun.COM kstat_t *xuio_ksp = NULL;
83311539SChunli.Zhang@Sun.COM
83411539SChunli.Zhang@Sun.COM int
dmu_xuio_init(xuio_t * xuio,int nblk)83511539SChunli.Zhang@Sun.COM dmu_xuio_init(xuio_t *xuio, int nblk)
83611539SChunli.Zhang@Sun.COM {
83711539SChunli.Zhang@Sun.COM dmu_xuio_t *priv;
83811539SChunli.Zhang@Sun.COM uio_t *uio = &xuio->xu_uio;
83911539SChunli.Zhang@Sun.COM
84011539SChunli.Zhang@Sun.COM uio->uio_iovcnt = nblk;
84111539SChunli.Zhang@Sun.COM uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
84211539SChunli.Zhang@Sun.COM
84311539SChunli.Zhang@Sun.COM priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
84411539SChunli.Zhang@Sun.COM priv->cnt = nblk;
84511539SChunli.Zhang@Sun.COM priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
84611539SChunli.Zhang@Sun.COM priv->iovp = uio->uio_iov;
84711539SChunli.Zhang@Sun.COM XUIO_XUZC_PRIV(xuio) = priv;
84811539SChunli.Zhang@Sun.COM
84911539SChunli.Zhang@Sun.COM if (XUIO_XUZC_RW(xuio) == UIO_READ)
85011539SChunli.Zhang@Sun.COM XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
85111539SChunli.Zhang@Sun.COM else
85211539SChunli.Zhang@Sun.COM XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
85311539SChunli.Zhang@Sun.COM
85411539SChunli.Zhang@Sun.COM return (0);
85511539SChunli.Zhang@Sun.COM }
85611539SChunli.Zhang@Sun.COM
85711539SChunli.Zhang@Sun.COM void
dmu_xuio_fini(xuio_t * xuio)85811539SChunli.Zhang@Sun.COM dmu_xuio_fini(xuio_t *xuio)
85911539SChunli.Zhang@Sun.COM {
86011539SChunli.Zhang@Sun.COM dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
86111539SChunli.Zhang@Sun.COM int nblk = priv->cnt;
86211539SChunli.Zhang@Sun.COM
86311539SChunli.Zhang@Sun.COM kmem_free(priv->iovp, nblk * sizeof (iovec_t));
86411539SChunli.Zhang@Sun.COM kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
86511539SChunli.Zhang@Sun.COM kmem_free(priv, sizeof (dmu_xuio_t));
86611539SChunli.Zhang@Sun.COM
86711539SChunli.Zhang@Sun.COM if (XUIO_XUZC_RW(xuio) == UIO_READ)
86811539SChunli.Zhang@Sun.COM XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
86911539SChunli.Zhang@Sun.COM else
87011539SChunli.Zhang@Sun.COM XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
87111539SChunli.Zhang@Sun.COM }
87211539SChunli.Zhang@Sun.COM
87311539SChunli.Zhang@Sun.COM /*
87411539SChunli.Zhang@Sun.COM * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
87511539SChunli.Zhang@Sun.COM * and increase priv->next by 1.
87611539SChunli.Zhang@Sun.COM */
87711539SChunli.Zhang@Sun.COM int
dmu_xuio_add(xuio_t * xuio,arc_buf_t * abuf,offset_t off,size_t n)87811539SChunli.Zhang@Sun.COM dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
87911539SChunli.Zhang@Sun.COM {
88011539SChunli.Zhang@Sun.COM struct iovec *iov;
88111539SChunli.Zhang@Sun.COM uio_t *uio = &xuio->xu_uio;
88211539SChunli.Zhang@Sun.COM dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
88311539SChunli.Zhang@Sun.COM int i = priv->next++;
88411539SChunli.Zhang@Sun.COM
88511539SChunli.Zhang@Sun.COM ASSERT(i < priv->cnt);
88611539SChunli.Zhang@Sun.COM ASSERT(off + n <= arc_buf_size(abuf));
88711539SChunli.Zhang@Sun.COM iov = uio->uio_iov + i;
88811539SChunli.Zhang@Sun.COM iov->iov_base = (char *)abuf->b_data + off;
88911539SChunli.Zhang@Sun.COM iov->iov_len = n;
89011539SChunli.Zhang@Sun.COM priv->bufs[i] = abuf;
89111539SChunli.Zhang@Sun.COM return (0);
89211539SChunli.Zhang@Sun.COM }
89311539SChunli.Zhang@Sun.COM
89411539SChunli.Zhang@Sun.COM int
dmu_xuio_cnt(xuio_t * xuio)89511539SChunli.Zhang@Sun.COM dmu_xuio_cnt(xuio_t *xuio)
89611539SChunli.Zhang@Sun.COM {
89711539SChunli.Zhang@Sun.COM dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
89811539SChunli.Zhang@Sun.COM return (priv->cnt);
89911539SChunli.Zhang@Sun.COM }
90011539SChunli.Zhang@Sun.COM
90111539SChunli.Zhang@Sun.COM arc_buf_t *
dmu_xuio_arcbuf(xuio_t * xuio,int i)90211539SChunli.Zhang@Sun.COM dmu_xuio_arcbuf(xuio_t *xuio, int i)
90311539SChunli.Zhang@Sun.COM {
90411539SChunli.Zhang@Sun.COM dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
90511539SChunli.Zhang@Sun.COM
90611539SChunli.Zhang@Sun.COM ASSERT(i < priv->cnt);
90711539SChunli.Zhang@Sun.COM return (priv->bufs[i]);
90811539SChunli.Zhang@Sun.COM }
90911539SChunli.Zhang@Sun.COM
91011539SChunli.Zhang@Sun.COM void
dmu_xuio_clear(xuio_t * xuio,int i)91111539SChunli.Zhang@Sun.COM dmu_xuio_clear(xuio_t *xuio, int i)
91211539SChunli.Zhang@Sun.COM {
91311539SChunli.Zhang@Sun.COM dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
91411539SChunli.Zhang@Sun.COM
91511539SChunli.Zhang@Sun.COM ASSERT(i < priv->cnt);
91611539SChunli.Zhang@Sun.COM priv->bufs[i] = NULL;
91711539SChunli.Zhang@Sun.COM }
91811539SChunli.Zhang@Sun.COM
91911539SChunli.Zhang@Sun.COM static void
xuio_stat_init(void)92011539SChunli.Zhang@Sun.COM xuio_stat_init(void)
92111539SChunli.Zhang@Sun.COM {
92211539SChunli.Zhang@Sun.COM xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
92311539SChunli.Zhang@Sun.COM KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
92411539SChunli.Zhang@Sun.COM KSTAT_FLAG_VIRTUAL);
92511539SChunli.Zhang@Sun.COM if (xuio_ksp != NULL) {
92611539SChunli.Zhang@Sun.COM xuio_ksp->ks_data = &xuio_stats;
92711539SChunli.Zhang@Sun.COM kstat_install(xuio_ksp);
92811539SChunli.Zhang@Sun.COM }
92911539SChunli.Zhang@Sun.COM }
93011539SChunli.Zhang@Sun.COM
93111539SChunli.Zhang@Sun.COM static void
xuio_stat_fini(void)93211539SChunli.Zhang@Sun.COM xuio_stat_fini(void)
93311539SChunli.Zhang@Sun.COM {
93411539SChunli.Zhang@Sun.COM if (xuio_ksp != NULL) {
93511539SChunli.Zhang@Sun.COM kstat_delete(xuio_ksp);
93611539SChunli.Zhang@Sun.COM xuio_ksp = NULL;
93711539SChunli.Zhang@Sun.COM }
93811539SChunli.Zhang@Sun.COM }
93911539SChunli.Zhang@Sun.COM
94011539SChunli.Zhang@Sun.COM void
xuio_stat_wbuf_copied()94111539SChunli.Zhang@Sun.COM xuio_stat_wbuf_copied()
94211539SChunli.Zhang@Sun.COM {
94311539SChunli.Zhang@Sun.COM XUIOSTAT_BUMP(xuiostat_wbuf_copied);
94411539SChunli.Zhang@Sun.COM }
94511539SChunli.Zhang@Sun.COM
94611539SChunli.Zhang@Sun.COM void
xuio_stat_wbuf_nocopy()94711539SChunli.Zhang@Sun.COM xuio_stat_wbuf_nocopy()
94811539SChunli.Zhang@Sun.COM {
94911539SChunli.Zhang@Sun.COM XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
95011539SChunli.Zhang@Sun.COM }
95111539SChunli.Zhang@Sun.COM
952789Sahrens #ifdef _KERNEL
953789Sahrens int
dmu_read_uio(objset_t * os,uint64_t object,uio_t * uio,uint64_t size)9543638Sbillm dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
9553638Sbillm {
9563638Sbillm dmu_buf_t **dbp;
9573638Sbillm int numbufs, i, err;
95811539SChunli.Zhang@Sun.COM xuio_t *xuio = NULL;
9593638Sbillm
9603638Sbillm /*
9613638Sbillm * NB: we could do this block-at-a-time, but it's nice
9623638Sbillm * to be reading in parallel.
9633638Sbillm */
9643638Sbillm err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
9653638Sbillm &numbufs, &dbp);
9663638Sbillm if (err)
9673638Sbillm return (err);
9683638Sbillm
96911539SChunli.Zhang@Sun.COM if (uio->uio_extflg == UIO_XUIO)
97011539SChunli.Zhang@Sun.COM xuio = (xuio_t *)uio;
97111539SChunli.Zhang@Sun.COM
9723638Sbillm for (i = 0; i < numbufs; i++) {
9733638Sbillm int tocpy;
9743638Sbillm int bufoff;
9753638Sbillm dmu_buf_t *db = dbp[i];
9763638Sbillm
9773638Sbillm ASSERT(size > 0);
9783638Sbillm
9793638Sbillm bufoff = uio->uio_loffset - db->db_offset;
9803638Sbillm tocpy = (int)MIN(db->db_size - bufoff, size);
9813638Sbillm
98211539SChunli.Zhang@Sun.COM if (xuio) {
98311539SChunli.Zhang@Sun.COM dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
98411539SChunli.Zhang@Sun.COM arc_buf_t *dbuf_abuf = dbi->db_buf;
98511539SChunli.Zhang@Sun.COM arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
98611539SChunli.Zhang@Sun.COM err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
98711539SChunli.Zhang@Sun.COM if (!err) {
98811539SChunli.Zhang@Sun.COM uio->uio_resid -= tocpy;
98911539SChunli.Zhang@Sun.COM uio->uio_loffset += tocpy;
99011539SChunli.Zhang@Sun.COM }
99111539SChunli.Zhang@Sun.COM
99211539SChunli.Zhang@Sun.COM if (abuf == dbuf_abuf)
99311539SChunli.Zhang@Sun.COM XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
99411539SChunli.Zhang@Sun.COM else
99511539SChunli.Zhang@Sun.COM XUIOSTAT_BUMP(xuiostat_rbuf_copied);
99611539SChunli.Zhang@Sun.COM } else {
99711539SChunli.Zhang@Sun.COM err = uiomove((char *)db->db_data + bufoff, tocpy,
99811539SChunli.Zhang@Sun.COM UIO_READ, uio);
99911539SChunli.Zhang@Sun.COM }
10003638Sbillm if (err)
10013638Sbillm break;
10023638Sbillm
10033638Sbillm size -= tocpy;
10043638Sbillm }
10053638Sbillm dmu_buf_rele_array(dbp, numbufs, FTAG);
10063638Sbillm
10073638Sbillm return (err);
10083638Sbillm }
10093638Sbillm
101012123STim.Haley@Sun.COM static int
dmu_write_uio_dnode(dnode_t * dn,uio_t * uio,uint64_t size,dmu_tx_t * tx)101112123STim.Haley@Sun.COM dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1012789Sahrens {
1013789Sahrens dmu_buf_t **dbp;
101412123STim.Haley@Sun.COM int numbufs;
1015789Sahrens int err = 0;
101612123STim.Haley@Sun.COM int i;
1017789Sahrens
101812123STim.Haley@Sun.COM err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
101912123STim.Haley@Sun.COM FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
10201544Seschrock if (err)
10211544Seschrock return (err);
1022789Sahrens
1023789Sahrens for (i = 0; i < numbufs; i++) {
1024789Sahrens int tocpy;
1025789Sahrens int bufoff;
1026789Sahrens dmu_buf_t *db = dbp[i];
1027789Sahrens
1028789Sahrens ASSERT(size > 0);
1029789Sahrens
10303638Sbillm bufoff = uio->uio_loffset - db->db_offset;
1031789Sahrens tocpy = (int)MIN(db->db_size - bufoff, size);
1032789Sahrens
1033789Sahrens ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1034789Sahrens
1035789Sahrens if (tocpy == db->db_size)
1036789Sahrens dmu_buf_will_fill(db, tx);
1037789Sahrens else
1038789Sahrens dmu_buf_will_dirty(db, tx);
1039789Sahrens
1040789Sahrens /*
1041789Sahrens * XXX uiomove could block forever (eg. nfs-backed
1042789Sahrens * pages). There needs to be a uiolockdown() function
1043789Sahrens * to lock the pages in memory, so that uiomove won't
1044789Sahrens * block.
1045789Sahrens */
1046789Sahrens err = uiomove((char *)db->db_data + bufoff, tocpy,
1047789Sahrens UIO_WRITE, uio);
1048789Sahrens
1049789Sahrens if (tocpy == db->db_size)
1050789Sahrens dmu_buf_fill_done(db, tx);
1051789Sahrens
1052789Sahrens if (err)
1053789Sahrens break;
1054789Sahrens
1055789Sahrens size -= tocpy;
1056789Sahrens }
105712123STim.Haley@Sun.COM
10581544Seschrock dmu_buf_rele_array(dbp, numbufs, FTAG);
1059789Sahrens return (err);
1060789Sahrens }
10612688Smaybee
10622688Smaybee int
dmu_write_uio_dbuf(dmu_buf_t * zdb,uio_t * uio,uint64_t size,dmu_tx_t * tx)106312123STim.Haley@Sun.COM dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
106412123STim.Haley@Sun.COM dmu_tx_t *tx)
106512123STim.Haley@Sun.COM {
1066*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1067*12684STom.Erickson@Sun.COM dnode_t *dn;
1068*12684STom.Erickson@Sun.COM int err;
1069*12684STom.Erickson@Sun.COM
107012123STim.Haley@Sun.COM if (size == 0)
107112123STim.Haley@Sun.COM return (0);
107212123STim.Haley@Sun.COM
1073*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
1074*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
1075*12684STom.Erickson@Sun.COM err = dmu_write_uio_dnode(dn, uio, size, tx);
1076*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
1077*12684STom.Erickson@Sun.COM
1078*12684STom.Erickson@Sun.COM return (err);
107912123STim.Haley@Sun.COM }
108012123STim.Haley@Sun.COM
108112123STim.Haley@Sun.COM int
dmu_write_uio(objset_t * os,uint64_t object,uio_t * uio,uint64_t size,dmu_tx_t * tx)108212123STim.Haley@Sun.COM dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
108312123STim.Haley@Sun.COM dmu_tx_t *tx)
108412123STim.Haley@Sun.COM {
108512123STim.Haley@Sun.COM dnode_t *dn;
108612123STim.Haley@Sun.COM int err;
108712123STim.Haley@Sun.COM
108812123STim.Haley@Sun.COM if (size == 0)
108912123STim.Haley@Sun.COM return (0);
109012123STim.Haley@Sun.COM
109112123STim.Haley@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
109212123STim.Haley@Sun.COM if (err)
109312123STim.Haley@Sun.COM return (err);
109412123STim.Haley@Sun.COM
109512123STim.Haley@Sun.COM err = dmu_write_uio_dnode(dn, uio, size, tx);
109612123STim.Haley@Sun.COM
109712123STim.Haley@Sun.COM dnode_rele(dn, FTAG);
109812123STim.Haley@Sun.COM
109912123STim.Haley@Sun.COM return (err);
110012123STim.Haley@Sun.COM }
110112123STim.Haley@Sun.COM
110212123STim.Haley@Sun.COM int
dmu_write_pages(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,page_t * pp,dmu_tx_t * tx)11032688Smaybee dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
11042688Smaybee page_t *pp, dmu_tx_t *tx)
11052688Smaybee {
11062688Smaybee dmu_buf_t **dbp;
11072688Smaybee int numbufs, i;
11082688Smaybee int err;
11092688Smaybee
11102688Smaybee if (size == 0)
11112688Smaybee return (0);
11122688Smaybee
11132688Smaybee err = dmu_buf_hold_array(os, object, offset, size,
11142688Smaybee FALSE, FTAG, &numbufs, &dbp);
11152688Smaybee if (err)
11162688Smaybee return (err);
11172688Smaybee
11182688Smaybee for (i = 0; i < numbufs; i++) {
11192688Smaybee int tocpy, copied, thiscpy;
11202688Smaybee int bufoff;
11212688Smaybee dmu_buf_t *db = dbp[i];
11222688Smaybee caddr_t va;
11232688Smaybee
11242688Smaybee ASSERT(size > 0);
11252688Smaybee ASSERT3U(db->db_size, >=, PAGESIZE);
11262688Smaybee
11272688Smaybee bufoff = offset - db->db_offset;
11282688Smaybee tocpy = (int)MIN(db->db_size - bufoff, size);
11292688Smaybee
11302688Smaybee ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
11312688Smaybee
11322688Smaybee if (tocpy == db->db_size)
11332688Smaybee dmu_buf_will_fill(db, tx);
11342688Smaybee else
11352688Smaybee dmu_buf_will_dirty(db, tx);
11362688Smaybee
11372688Smaybee for (copied = 0; copied < tocpy; copied += PAGESIZE) {
11382688Smaybee ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
11392688Smaybee thiscpy = MIN(PAGESIZE, tocpy - copied);
11407315SJonathan.Adams@Sun.COM va = zfs_map_page(pp, S_READ);
11412688Smaybee bcopy(va, (char *)db->db_data + bufoff, thiscpy);
11427315SJonathan.Adams@Sun.COM zfs_unmap_page(pp, va);
11432688Smaybee pp = pp->p_next;
11442688Smaybee bufoff += PAGESIZE;
11452688Smaybee }
11462688Smaybee
11472688Smaybee if (tocpy == db->db_size)
11482688Smaybee dmu_buf_fill_done(db, tx);
11492688Smaybee
11502688Smaybee offset += tocpy;
11512688Smaybee size -= tocpy;
11522688Smaybee }
11532688Smaybee dmu_buf_rele_array(dbp, numbufs, FTAG);
11542688Smaybee return (err);
11552688Smaybee }
1156789Sahrens #endif
1157789Sahrens
11589412SAleksandr.Guzovskiy@Sun.COM /*
11599412SAleksandr.Guzovskiy@Sun.COM * Allocate a loaned anonymous arc buffer.
11609412SAleksandr.Guzovskiy@Sun.COM */
11619412SAleksandr.Guzovskiy@Sun.COM arc_buf_t *
dmu_request_arcbuf(dmu_buf_t * handle,int size)11629412SAleksandr.Guzovskiy@Sun.COM dmu_request_arcbuf(dmu_buf_t *handle, int size)
11639412SAleksandr.Guzovskiy@Sun.COM {
1164*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1165*12684STom.Erickson@Sun.COM spa_t *spa;
11669412SAleksandr.Guzovskiy@Sun.COM
1167*12684STom.Erickson@Sun.COM DB_GET_SPA(&spa, db);
1168*12684STom.Erickson@Sun.COM return (arc_loan_buf(spa, size));
11699412SAleksandr.Guzovskiy@Sun.COM }
11709412SAleksandr.Guzovskiy@Sun.COM
11719412SAleksandr.Guzovskiy@Sun.COM /*
11729412SAleksandr.Guzovskiy@Sun.COM * Free a loaned arc buffer.
11739412SAleksandr.Guzovskiy@Sun.COM */
11749412SAleksandr.Guzovskiy@Sun.COM void
dmu_return_arcbuf(arc_buf_t * buf)11759412SAleksandr.Guzovskiy@Sun.COM dmu_return_arcbuf(arc_buf_t *buf)
11769412SAleksandr.Guzovskiy@Sun.COM {
11779412SAleksandr.Guzovskiy@Sun.COM arc_return_buf(buf, FTAG);
11789412SAleksandr.Guzovskiy@Sun.COM VERIFY(arc_buf_remove_ref(buf, FTAG) == 1);
11799412SAleksandr.Guzovskiy@Sun.COM }
11809412SAleksandr.Guzovskiy@Sun.COM
11819412SAleksandr.Guzovskiy@Sun.COM /*
11829412SAleksandr.Guzovskiy@Sun.COM * When possible directly assign passed loaned arc buffer to a dbuf.
11839412SAleksandr.Guzovskiy@Sun.COM * If this is not possible copy the contents of passed arc buf via
11849412SAleksandr.Guzovskiy@Sun.COM * dmu_write().
11859412SAleksandr.Guzovskiy@Sun.COM */
11869412SAleksandr.Guzovskiy@Sun.COM void
dmu_assign_arcbuf(dmu_buf_t * handle,uint64_t offset,arc_buf_t * buf,dmu_tx_t * tx)11879412SAleksandr.Guzovskiy@Sun.COM dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
11889412SAleksandr.Guzovskiy@Sun.COM dmu_tx_t *tx)
11899412SAleksandr.Guzovskiy@Sun.COM {
1190*12684STom.Erickson@Sun.COM dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1191*12684STom.Erickson@Sun.COM dnode_t *dn;
11929412SAleksandr.Guzovskiy@Sun.COM dmu_buf_impl_t *db;
11939412SAleksandr.Guzovskiy@Sun.COM uint32_t blksz = (uint32_t)arc_buf_size(buf);
11949412SAleksandr.Guzovskiy@Sun.COM uint64_t blkid;
11959412SAleksandr.Guzovskiy@Sun.COM
1196*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(dbuf);
1197*12684STom.Erickson@Sun.COM dn = DB_DNODE(dbuf);
11989412SAleksandr.Guzovskiy@Sun.COM rw_enter(&dn->dn_struct_rwlock, RW_READER);
11999412SAleksandr.Guzovskiy@Sun.COM blkid = dbuf_whichblock(dn, offset);
12009412SAleksandr.Guzovskiy@Sun.COM VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
12019412SAleksandr.Guzovskiy@Sun.COM rw_exit(&dn->dn_struct_rwlock);
1202*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(dbuf);
12039412SAleksandr.Guzovskiy@Sun.COM
12049412SAleksandr.Guzovskiy@Sun.COM if (offset == db->db.db_offset && blksz == db->db.db_size) {
12059412SAleksandr.Guzovskiy@Sun.COM dbuf_assign_arcbuf(db, buf, tx);
12069412SAleksandr.Guzovskiy@Sun.COM dbuf_rele(db, FTAG);
12079412SAleksandr.Guzovskiy@Sun.COM } else {
1208*12684STom.Erickson@Sun.COM objset_t *os;
1209*12684STom.Erickson@Sun.COM uint64_t object;
1210*12684STom.Erickson@Sun.COM
1211*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(dbuf);
1212*12684STom.Erickson@Sun.COM dn = DB_DNODE(dbuf);
1213*12684STom.Erickson@Sun.COM os = dn->dn_objset;
1214*12684STom.Erickson@Sun.COM object = dn->dn_object;
1215*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(dbuf);
1216*12684STom.Erickson@Sun.COM
12179412SAleksandr.Guzovskiy@Sun.COM dbuf_rele(db, FTAG);
1218*12684STom.Erickson@Sun.COM dmu_write(os, object, offset, blksz, buf->b_data, tx);
12199412SAleksandr.Guzovskiy@Sun.COM dmu_return_arcbuf(buf);
122011539SChunli.Zhang@Sun.COM XUIOSTAT_BUMP(xuiostat_wbuf_copied);
12219412SAleksandr.Guzovskiy@Sun.COM }
12229412SAleksandr.Guzovskiy@Sun.COM }
12239412SAleksandr.Guzovskiy@Sun.COM
12242237Smaybee typedef struct {
122510922SJeff.Bonwick@Sun.COM dbuf_dirty_record_t *dsa_dr;
122610922SJeff.Bonwick@Sun.COM dmu_sync_cb_t *dsa_done;
122710922SJeff.Bonwick@Sun.COM zgd_t *dsa_zgd;
122810922SJeff.Bonwick@Sun.COM dmu_tx_t *dsa_tx;
12293547Smaybee } dmu_sync_arg_t;
12302237Smaybee
12312237Smaybee /* ARGSUSED */
12322237Smaybee static void
dmu_sync_ready(zio_t * zio,arc_buf_t * buf,void * varg)12337754SJeff.Bonwick@Sun.COM dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
12347754SJeff.Bonwick@Sun.COM {
123510922SJeff.Bonwick@Sun.COM dmu_sync_arg_t *dsa = varg;
123610922SJeff.Bonwick@Sun.COM dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
12377754SJeff.Bonwick@Sun.COM blkptr_t *bp = zio->io_bp;
123810800SNeil.Perrin@Sun.COM
123910922SJeff.Bonwick@Sun.COM if (zio->io_error == 0) {
124010922SJeff.Bonwick@Sun.COM if (BP_IS_HOLE(bp)) {
124110922SJeff.Bonwick@Sun.COM /*
124210922SJeff.Bonwick@Sun.COM * A block of zeros may compress to a hole, but the
124310922SJeff.Bonwick@Sun.COM * block size still needs to be known for replay.
124410922SJeff.Bonwick@Sun.COM */
124510922SJeff.Bonwick@Sun.COM BP_SET_LSIZE(bp, db->db_size);
124610922SJeff.Bonwick@Sun.COM } else {
124710922SJeff.Bonwick@Sun.COM ASSERT(BP_GET_LEVEL(bp) == 0);
124810922SJeff.Bonwick@Sun.COM bp->blk_fill = 1;
124910922SJeff.Bonwick@Sun.COM }
12507754SJeff.Bonwick@Sun.COM }
12517754SJeff.Bonwick@Sun.COM }
12527754SJeff.Bonwick@Sun.COM
125310922SJeff.Bonwick@Sun.COM static void
dmu_sync_late_arrival_ready(zio_t * zio)125410922SJeff.Bonwick@Sun.COM dmu_sync_late_arrival_ready(zio_t *zio)
125510922SJeff.Bonwick@Sun.COM {
125610922SJeff.Bonwick@Sun.COM dmu_sync_ready(zio, NULL, zio->io_private);
125710922SJeff.Bonwick@Sun.COM }
125810922SJeff.Bonwick@Sun.COM
12597754SJeff.Bonwick@Sun.COM /* ARGSUSED */
12607754SJeff.Bonwick@Sun.COM static void
dmu_sync_done(zio_t * zio,arc_buf_t * buf,void * varg)12612237Smaybee dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
12622237Smaybee {
126310922SJeff.Bonwick@Sun.COM dmu_sync_arg_t *dsa = varg;
126410922SJeff.Bonwick@Sun.COM dbuf_dirty_record_t *dr = dsa->dsa_dr;
12653547Smaybee dmu_buf_impl_t *db = dr->dr_dbuf;
12662237Smaybee
126710810SNeil.Perrin@Sun.COM mutex_enter(&db->db_mtx);
126810810SNeil.Perrin@Sun.COM ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
126910922SJeff.Bonwick@Sun.COM if (zio->io_error == 0) {
127010922SJeff.Bonwick@Sun.COM dr->dt.dl.dr_overridden_by = *zio->io_bp;
127110922SJeff.Bonwick@Sun.COM dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
127210922SJeff.Bonwick@Sun.COM dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
127310922SJeff.Bonwick@Sun.COM if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
127410922SJeff.Bonwick@Sun.COM BP_ZERO(&dr->dt.dl.dr_overridden_by);
127510922SJeff.Bonwick@Sun.COM } else {
127610922SJeff.Bonwick@Sun.COM dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
127710922SJeff.Bonwick@Sun.COM }
12782237Smaybee cv_broadcast(&db->db_changed);
127910810SNeil.Perrin@Sun.COM mutex_exit(&db->db_mtx);
128010810SNeil.Perrin@Sun.COM
128110922SJeff.Bonwick@Sun.COM dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
128210922SJeff.Bonwick@Sun.COM
128310922SJeff.Bonwick@Sun.COM kmem_free(dsa, sizeof (*dsa));
128410922SJeff.Bonwick@Sun.COM }
128510922SJeff.Bonwick@Sun.COM
128610922SJeff.Bonwick@Sun.COM static void
dmu_sync_late_arrival_done(zio_t * zio)128710922SJeff.Bonwick@Sun.COM dmu_sync_late_arrival_done(zio_t *zio)
128810922SJeff.Bonwick@Sun.COM {
128910922SJeff.Bonwick@Sun.COM blkptr_t *bp = zio->io_bp;
129010922SJeff.Bonwick@Sun.COM dmu_sync_arg_t *dsa = zio->io_private;
129110922SJeff.Bonwick@Sun.COM
129210922SJeff.Bonwick@Sun.COM if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
129310922SJeff.Bonwick@Sun.COM ASSERT(zio->io_bp->blk_birth == zio->io_txg);
129410922SJeff.Bonwick@Sun.COM ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
129510922SJeff.Bonwick@Sun.COM zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
129610922SJeff.Bonwick@Sun.COM }
129710922SJeff.Bonwick@Sun.COM
129810922SJeff.Bonwick@Sun.COM dmu_tx_commit(dsa->dsa_tx);
129910922SJeff.Bonwick@Sun.COM
130010922SJeff.Bonwick@Sun.COM dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
130110922SJeff.Bonwick@Sun.COM
130210922SJeff.Bonwick@Sun.COM kmem_free(dsa, sizeof (*dsa));
130310922SJeff.Bonwick@Sun.COM }
13043547Smaybee
130510922SJeff.Bonwick@Sun.COM static int
dmu_sync_late_arrival(zio_t * pio,objset_t * os,dmu_sync_cb_t * done,zgd_t * zgd,zio_prop_t * zp,zbookmark_t * zb)130610922SJeff.Bonwick@Sun.COM dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
130710922SJeff.Bonwick@Sun.COM zio_prop_t *zp, zbookmark_t *zb)
130810922SJeff.Bonwick@Sun.COM {
130910922SJeff.Bonwick@Sun.COM dmu_sync_arg_t *dsa;
131010922SJeff.Bonwick@Sun.COM dmu_tx_t *tx;
131110922SJeff.Bonwick@Sun.COM
131210922SJeff.Bonwick@Sun.COM tx = dmu_tx_create(os);
131310922SJeff.Bonwick@Sun.COM dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
131411670SNeil.Perrin@Sun.COM if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
131510922SJeff.Bonwick@Sun.COM dmu_tx_abort(tx);
131610922SJeff.Bonwick@Sun.COM return (EIO); /* Make zl_get_data do txg_waited_synced() */
131710922SJeff.Bonwick@Sun.COM }
131810922SJeff.Bonwick@Sun.COM
131910922SJeff.Bonwick@Sun.COM dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
132010922SJeff.Bonwick@Sun.COM dsa->dsa_dr = NULL;
132110922SJeff.Bonwick@Sun.COM dsa->dsa_done = done;
132210922SJeff.Bonwick@Sun.COM dsa->dsa_zgd = zgd;
132310922SJeff.Bonwick@Sun.COM dsa->dsa_tx = tx;
132410922SJeff.Bonwick@Sun.COM
132510922SJeff.Bonwick@Sun.COM zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
132610922SJeff.Bonwick@Sun.COM zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
132710922SJeff.Bonwick@Sun.COM dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa,
132810922SJeff.Bonwick@Sun.COM ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
132910922SJeff.Bonwick@Sun.COM
133010922SJeff.Bonwick@Sun.COM return (0);
13312237Smaybee }
13322237Smaybee
1333789Sahrens /*
13342237Smaybee * Intent log support: sync the block associated with db to disk.
13352237Smaybee * N.B. and XXX: the caller is responsible for making sure that the
13362237Smaybee * data isn't changing while dmu_sync() is writing it.
1337789Sahrens *
1338789Sahrens * Return values:
1339789Sahrens *
13402237Smaybee * EEXIST: this txg has already been synced, so there's nothing to to.
1341789Sahrens * The caller should not log the write.
1342789Sahrens *
1343789Sahrens * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1344789Sahrens * The caller should not log the write.
1345789Sahrens *
13462237Smaybee * EALREADY: this block is already in the process of being synced.
13472237Smaybee * The caller should track its progress (somehow).
1348789Sahrens *
134910922SJeff.Bonwick@Sun.COM * EIO: could not do the I/O.
135010922SJeff.Bonwick@Sun.COM * The caller should do a txg_wait_synced().
1351789Sahrens *
135210922SJeff.Bonwick@Sun.COM * 0: the I/O has been initiated.
135310922SJeff.Bonwick@Sun.COM * The caller should log this blkptr in the done callback.
135410922SJeff.Bonwick@Sun.COM * It is possible that the I/O will fail, in which case
135510922SJeff.Bonwick@Sun.COM * the error will be reported to the done callback and
135610922SJeff.Bonwick@Sun.COM * propagated to pio from zio_done().
1357789Sahrens */
1358789Sahrens int
dmu_sync(zio_t * pio,uint64_t txg,dmu_sync_cb_t * done,zgd_t * zgd)135910922SJeff.Bonwick@Sun.COM dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1360789Sahrens {
136110922SJeff.Bonwick@Sun.COM blkptr_t *bp = zgd->zgd_bp;
136210922SJeff.Bonwick@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
136310298SMatthew.Ahrens@Sun.COM objset_t *os = db->db_objset;
136410922SJeff.Bonwick@Sun.COM dsl_dataset_t *ds = os->os_dsl_dataset;
13653547Smaybee dbuf_dirty_record_t *dr;
136610922SJeff.Bonwick@Sun.COM dmu_sync_arg_t *dsa;
13672237Smaybee zbookmark_t zb;
136810922SJeff.Bonwick@Sun.COM zio_prop_t zp;
1369*12684STom.Erickson@Sun.COM dnode_t *dn;
1370789Sahrens
137110922SJeff.Bonwick@Sun.COM ASSERT(pio != NULL);
1372789Sahrens ASSERT(BP_IS_HOLE(bp));
1373789Sahrens ASSERT(txg != 0);
1374789Sahrens
137510922SJeff.Bonwick@Sun.COM SET_BOOKMARK(&zb, ds->ds_object,
137610922SJeff.Bonwick@Sun.COM db->db.db_object, db->db_level, db->db_blkid);
137710922SJeff.Bonwick@Sun.COM
1378*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
1379*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
1380*12684STom.Erickson@Sun.COM dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1381*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
1382789Sahrens
1383789Sahrens /*
138410922SJeff.Bonwick@Sun.COM * If we're frozen (running ziltest), we always need to generate a bp.
13851544Seschrock */
138610922SJeff.Bonwick@Sun.COM if (txg > spa_freeze_txg(os->os_spa))
138710922SJeff.Bonwick@Sun.COM return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
13881544Seschrock
13891544Seschrock /*
139010922SJeff.Bonwick@Sun.COM * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
139110922SJeff.Bonwick@Sun.COM * and us. If we determine that this txg is not yet syncing,
139210922SJeff.Bonwick@Sun.COM * but it begins to sync a moment later, that's OK because the
139310922SJeff.Bonwick@Sun.COM * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1394789Sahrens */
139510922SJeff.Bonwick@Sun.COM mutex_enter(&db->db_mtx);
139610922SJeff.Bonwick@Sun.COM
139710922SJeff.Bonwick@Sun.COM if (txg <= spa_last_synced_txg(os->os_spa)) {
1398789Sahrens /*
139910922SJeff.Bonwick@Sun.COM * This txg has already synced. There's nothing to do.
1400789Sahrens */
140110922SJeff.Bonwick@Sun.COM mutex_exit(&db->db_mtx);
14022237Smaybee return (EEXIST);
1403789Sahrens }
1404789Sahrens
140510922SJeff.Bonwick@Sun.COM if (txg <= spa_syncing_txg(os->os_spa)) {
140610922SJeff.Bonwick@Sun.COM /*
140710922SJeff.Bonwick@Sun.COM * This txg is currently syncing, so we can't mess with
140810922SJeff.Bonwick@Sun.COM * the dirty record anymore; just write a new log block.
140910922SJeff.Bonwick@Sun.COM */
141010922SJeff.Bonwick@Sun.COM mutex_exit(&db->db_mtx);
141110922SJeff.Bonwick@Sun.COM return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
14122237Smaybee }
14132237Smaybee
14143547Smaybee dr = db->db_last_dirty;
141510922SJeff.Bonwick@Sun.COM while (dr && dr->dr_txg != txg)
14163547Smaybee dr = dr->dr_next;
141710922SJeff.Bonwick@Sun.COM
141810922SJeff.Bonwick@Sun.COM if (dr == NULL) {
14192237Smaybee /*
142010922SJeff.Bonwick@Sun.COM * There's no dr for this dbuf, so it must have been freed.
14212237Smaybee * There's no need to log writes to freed blocks, so we're done.
14222237Smaybee */
14232237Smaybee mutex_exit(&db->db_mtx);
1424789Sahrens return (ENOENT);
1425789Sahrens }
1426789Sahrens
14273547Smaybee ASSERT(dr->dr_txg == txg);
142810922SJeff.Bonwick@Sun.COM if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
142910922SJeff.Bonwick@Sun.COM dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
14303547Smaybee /*
143110922SJeff.Bonwick@Sun.COM * We have already issued a sync write for this buffer,
143210922SJeff.Bonwick@Sun.COM * or this buffer has already been synced. It could not
143310922SJeff.Bonwick@Sun.COM * have been dirtied since, or we would have cleared the state.
14343547Smaybee */
14353547Smaybee mutex_exit(&db->db_mtx);
14363547Smaybee return (EALREADY);
14373547Smaybee }
14383547Smaybee
143910922SJeff.Bonwick@Sun.COM ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
14403547Smaybee dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
14412237Smaybee mutex_exit(&db->db_mtx);
14427754SJeff.Bonwick@Sun.COM
144310922SJeff.Bonwick@Sun.COM dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
144410922SJeff.Bonwick@Sun.COM dsa->dsa_dr = dr;
144510922SJeff.Bonwick@Sun.COM dsa->dsa_done = done;
144610922SJeff.Bonwick@Sun.COM dsa->dsa_zgd = zgd;
144710922SJeff.Bonwick@Sun.COM dsa->dsa_tx = NULL;
14487754SJeff.Bonwick@Sun.COM
144910922SJeff.Bonwick@Sun.COM zio_nowait(arc_write(pio, os->os_spa, txg,
145010922SJeff.Bonwick@Sun.COM bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp,
145110922SJeff.Bonwick@Sun.COM dmu_sync_ready, dmu_sync_done, dsa,
145210922SJeff.Bonwick@Sun.COM ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
145310922SJeff.Bonwick@Sun.COM
145410922SJeff.Bonwick@Sun.COM return (0);
1455789Sahrens }
1456789Sahrens
1457789Sahrens int
dmu_object_set_blocksize(objset_t * os,uint64_t object,uint64_t size,int ibs,dmu_tx_t * tx)1458789Sahrens dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1459789Sahrens dmu_tx_t *tx)
1460789Sahrens {
14611544Seschrock dnode_t *dn;
14621544Seschrock int err;
14631544Seschrock
146410298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
14651544Seschrock if (err)
14661544Seschrock return (err);
14671544Seschrock err = dnode_set_blksz(dn, size, ibs, tx);
1468789Sahrens dnode_rele(dn, FTAG);
1469789Sahrens return (err);
1470789Sahrens }
1471789Sahrens
1472789Sahrens void
dmu_object_set_checksum(objset_t * os,uint64_t object,uint8_t checksum,dmu_tx_t * tx)1473789Sahrens dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1474789Sahrens dmu_tx_t *tx)
1475789Sahrens {
14761544Seschrock dnode_t *dn;
14771544Seschrock
14781544Seschrock /* XXX assumes dnode_hold will not get an i/o error */
147910298SMatthew.Ahrens@Sun.COM (void) dnode_hold(os, object, FTAG, &dn);
1480789Sahrens ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
1481789Sahrens dn->dn_checksum = checksum;
1482789Sahrens dnode_setdirty(dn, tx);
1483789Sahrens dnode_rele(dn, FTAG);
1484789Sahrens }
1485789Sahrens
1486789Sahrens void
dmu_object_set_compress(objset_t * os,uint64_t object,uint8_t compress,dmu_tx_t * tx)1487789Sahrens dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1488789Sahrens dmu_tx_t *tx)
1489789Sahrens {
14901544Seschrock dnode_t *dn;
14911544Seschrock
14921544Seschrock /* XXX assumes dnode_hold will not get an i/o error */
149310298SMatthew.Ahrens@Sun.COM (void) dnode_hold(os, object, FTAG, &dn);
1494789Sahrens ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
1495789Sahrens dn->dn_compress = compress;
1496789Sahrens dnode_setdirty(dn, tx);
1497789Sahrens dnode_rele(dn, FTAG);
1498789Sahrens }
1499789Sahrens
150010922SJeff.Bonwick@Sun.COM int zfs_mdcomp_disable = 0;
150110922SJeff.Bonwick@Sun.COM
150210922SJeff.Bonwick@Sun.COM void
dmu_write_policy(objset_t * os,dnode_t * dn,int level,int wp,zio_prop_t * zp)150310922SJeff.Bonwick@Sun.COM dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
150410922SJeff.Bonwick@Sun.COM {
150510922SJeff.Bonwick@Sun.COM dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
150612493SMark.Shellenbaum@Oracle.COM boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata ||
150712493SMark.Shellenbaum@Oracle.COM (wp & WP_SPILL));
150810922SJeff.Bonwick@Sun.COM enum zio_checksum checksum = os->os_checksum;
150910922SJeff.Bonwick@Sun.COM enum zio_compress compress = os->os_compress;
151010922SJeff.Bonwick@Sun.COM enum zio_checksum dedup_checksum = os->os_dedup_checksum;
151110922SJeff.Bonwick@Sun.COM boolean_t dedup;
151210922SJeff.Bonwick@Sun.COM boolean_t dedup_verify = os->os_dedup_verify;
151310922SJeff.Bonwick@Sun.COM int copies = os->os_copies;
151410922SJeff.Bonwick@Sun.COM
151510922SJeff.Bonwick@Sun.COM /*
151610922SJeff.Bonwick@Sun.COM * Determine checksum setting.
151710922SJeff.Bonwick@Sun.COM */
151810922SJeff.Bonwick@Sun.COM if (ismd) {
151910922SJeff.Bonwick@Sun.COM /*
152010922SJeff.Bonwick@Sun.COM * Metadata always gets checksummed. If the data
152110922SJeff.Bonwick@Sun.COM * checksum is multi-bit correctable, and it's not a
152210922SJeff.Bonwick@Sun.COM * ZBT-style checksum, then it's suitable for metadata
152310922SJeff.Bonwick@Sun.COM * as well. Otherwise, the metadata checksum defaults
152410922SJeff.Bonwick@Sun.COM * to fletcher4.
152510922SJeff.Bonwick@Sun.COM */
152610922SJeff.Bonwick@Sun.COM if (zio_checksum_table[checksum].ci_correctable < 1 ||
152711670SNeil.Perrin@Sun.COM zio_checksum_table[checksum].ci_eck)
152810922SJeff.Bonwick@Sun.COM checksum = ZIO_CHECKSUM_FLETCHER_4;
152910922SJeff.Bonwick@Sun.COM } else {
153010922SJeff.Bonwick@Sun.COM checksum = zio_checksum_select(dn->dn_checksum, checksum);
153110922SJeff.Bonwick@Sun.COM }
153210922SJeff.Bonwick@Sun.COM
153310922SJeff.Bonwick@Sun.COM /*
153410922SJeff.Bonwick@Sun.COM * Determine compression setting.
153510922SJeff.Bonwick@Sun.COM */
153610922SJeff.Bonwick@Sun.COM if (ismd) {
153710922SJeff.Bonwick@Sun.COM /*
153810922SJeff.Bonwick@Sun.COM * XXX -- we should design a compression algorithm
153910922SJeff.Bonwick@Sun.COM * that specializes in arrays of bps.
154010922SJeff.Bonwick@Sun.COM */
154110922SJeff.Bonwick@Sun.COM compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
154210922SJeff.Bonwick@Sun.COM ZIO_COMPRESS_LZJB;
154310922SJeff.Bonwick@Sun.COM } else {
154410922SJeff.Bonwick@Sun.COM compress = zio_compress_select(dn->dn_compress, compress);
154510922SJeff.Bonwick@Sun.COM }
154610922SJeff.Bonwick@Sun.COM
154710922SJeff.Bonwick@Sun.COM /*
154810922SJeff.Bonwick@Sun.COM * Determine dedup setting. If we are in dmu_sync(), we won't
154910922SJeff.Bonwick@Sun.COM * actually dedup now because that's all done in syncing context;
155010922SJeff.Bonwick@Sun.COM * but we do want to use the dedup checkum. If the checksum is not
155110922SJeff.Bonwick@Sun.COM * strong enough to ensure unique signatures, force dedup_verify.
155210922SJeff.Bonwick@Sun.COM */
155310922SJeff.Bonwick@Sun.COM dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF);
155410922SJeff.Bonwick@Sun.COM if (dedup) {
155510922SJeff.Bonwick@Sun.COM checksum = dedup_checksum;
155610922SJeff.Bonwick@Sun.COM if (!zio_checksum_table[checksum].ci_dedup)
155710922SJeff.Bonwick@Sun.COM dedup_verify = 1;
155810922SJeff.Bonwick@Sun.COM }
155910922SJeff.Bonwick@Sun.COM
156010922SJeff.Bonwick@Sun.COM if (wp & WP_DMU_SYNC)
156110922SJeff.Bonwick@Sun.COM dedup = 0;
156210922SJeff.Bonwick@Sun.COM
156310922SJeff.Bonwick@Sun.COM if (wp & WP_NOFILL) {
156410922SJeff.Bonwick@Sun.COM ASSERT(!ismd && level == 0);
156510922SJeff.Bonwick@Sun.COM checksum = ZIO_CHECKSUM_OFF;
156610922SJeff.Bonwick@Sun.COM compress = ZIO_COMPRESS_OFF;
156710922SJeff.Bonwick@Sun.COM dedup = B_FALSE;
156810922SJeff.Bonwick@Sun.COM }
156910922SJeff.Bonwick@Sun.COM
157010922SJeff.Bonwick@Sun.COM zp->zp_checksum = checksum;
157110922SJeff.Bonwick@Sun.COM zp->zp_compress = compress;
157211935SMark.Shellenbaum@Sun.COM zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
157310922SJeff.Bonwick@Sun.COM zp->zp_level = level;
157410922SJeff.Bonwick@Sun.COM zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa));
157510922SJeff.Bonwick@Sun.COM zp->zp_dedup = dedup;
157610922SJeff.Bonwick@Sun.COM zp->zp_dedup_verify = dedup && dedup_verify;
157710922SJeff.Bonwick@Sun.COM }
157810922SJeff.Bonwick@Sun.COM
15791775Sbillm int
dmu_offset_next(objset_t * os,uint64_t object,boolean_t hole,uint64_t * off)1580789Sahrens dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1581789Sahrens {
1582789Sahrens dnode_t *dn;
1583789Sahrens int i, err;
1584789Sahrens
158510298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
15861544Seschrock if (err)
15871544Seschrock return (err);
1588789Sahrens /*
1589789Sahrens * Sync any current changes before
1590789Sahrens * we go trundling through the block pointers.
1591789Sahrens */
1592789Sahrens for (i = 0; i < TXG_SIZE; i++) {
15931596Sahrens if (list_link_active(&dn->dn_dirty_link[i]))
1594789Sahrens break;
1595789Sahrens }
1596789Sahrens if (i != TXG_SIZE) {
1597789Sahrens dnode_rele(dn, FTAG);
1598789Sahrens txg_wait_synced(dmu_objset_pool(os), 0);
159910298SMatthew.Ahrens@Sun.COM err = dnode_hold(os, object, FTAG, &dn);
16001544Seschrock if (err)
16011544Seschrock return (err);
1602789Sahrens }
1603789Sahrens
16046992Smaybee err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1605789Sahrens dnode_rele(dn, FTAG);
1606789Sahrens
1607789Sahrens return (err);
1608789Sahrens }
1609789Sahrens
1610789Sahrens void
dmu_object_info_from_dnode(dnode_t * dn,dmu_object_info_t * doi)1611789Sahrens dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1612789Sahrens {
161310922SJeff.Bonwick@Sun.COM dnode_phys_t *dnp;
161410922SJeff.Bonwick@Sun.COM
1615789Sahrens rw_enter(&dn->dn_struct_rwlock, RW_READER);
1616789Sahrens mutex_enter(&dn->dn_mtx);
1617789Sahrens
161810922SJeff.Bonwick@Sun.COM dnp = dn->dn_phys;
161910922SJeff.Bonwick@Sun.COM
1620789Sahrens doi->doi_data_block_size = dn->dn_datablksz;
1621789Sahrens doi->doi_metadata_block_size = dn->dn_indblkshift ?
1622789Sahrens 1ULL << dn->dn_indblkshift : 0;
162310922SJeff.Bonwick@Sun.COM doi->doi_type = dn->dn_type;
162410922SJeff.Bonwick@Sun.COM doi->doi_bonus_type = dn->dn_bonustype;
162510922SJeff.Bonwick@Sun.COM doi->doi_bonus_size = dn->dn_bonuslen;
1626789Sahrens doi->doi_indirection = dn->dn_nlevels;
1627789Sahrens doi->doi_checksum = dn->dn_checksum;
1628789Sahrens doi->doi_compress = dn->dn_compress;
162910922SJeff.Bonwick@Sun.COM doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
163010922SJeff.Bonwick@Sun.COM doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz;
163110922SJeff.Bonwick@Sun.COM doi->doi_fill_count = 0;
163210922SJeff.Bonwick@Sun.COM for (int i = 0; i < dnp->dn_nblkptr; i++)
163310922SJeff.Bonwick@Sun.COM doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
1634789Sahrens
1635789Sahrens mutex_exit(&dn->dn_mtx);
1636789Sahrens rw_exit(&dn->dn_struct_rwlock);
1637789Sahrens }
1638789Sahrens
1639789Sahrens /*
1640789Sahrens * Get information on a DMU object.
1641789Sahrens * If doi is NULL, just indicates whether the object exists.
1642789Sahrens */
1643789Sahrens int
dmu_object_info(objset_t * os,uint64_t object,dmu_object_info_t * doi)1644789Sahrens dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1645789Sahrens {
16461544Seschrock dnode_t *dn;
164710298SMatthew.Ahrens@Sun.COM int err = dnode_hold(os, object, FTAG, &dn);
1648789Sahrens
16491544Seschrock if (err)
16501544Seschrock return (err);
1651789Sahrens
1652789Sahrens if (doi != NULL)
1653789Sahrens dmu_object_info_from_dnode(dn, doi);
1654789Sahrens
1655789Sahrens dnode_rele(dn, FTAG);
1656789Sahrens return (0);
1657789Sahrens }
1658789Sahrens
1659789Sahrens /*
1660789Sahrens * As above, but faster; can be used when you have a held dbuf in hand.
1661789Sahrens */
1662789Sahrens void
dmu_object_info_from_db(dmu_buf_t * db_fake,dmu_object_info_t * doi)1663*12684STom.Erickson@Sun.COM dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1664789Sahrens {
1665*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1666*12684STom.Erickson@Sun.COM
1667*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
1668*12684STom.Erickson@Sun.COM dmu_object_info_from_dnode(DB_DNODE(db), doi);
1669*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
1670789Sahrens }
1671789Sahrens
1672789Sahrens /*
1673789Sahrens * Faster still when you only care about the size.
1674789Sahrens * This is specifically optimized for zfs_getattr().
1675789Sahrens */
1676789Sahrens void
dmu_object_size_from_db(dmu_buf_t * db_fake,uint32_t * blksize,u_longlong_t * nblk512)1677*12684STom.Erickson@Sun.COM dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1678*12684STom.Erickson@Sun.COM u_longlong_t *nblk512)
1679789Sahrens {
1680*12684STom.Erickson@Sun.COM dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1681*12684STom.Erickson@Sun.COM dnode_t *dn;
1682*12684STom.Erickson@Sun.COM
1683*12684STom.Erickson@Sun.COM DB_DNODE_ENTER(db);
1684*12684STom.Erickson@Sun.COM dn = DB_DNODE(db);
1685789Sahrens
1686789Sahrens *blksize = dn->dn_datablksz;
16872082Seschrock /* add 1 for dnode space */
16882082Seschrock *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
16892082Seschrock SPA_MINBLOCKSHIFT) + 1;
1690*12684STom.Erickson@Sun.COM DB_DNODE_EXIT(db);
1691789Sahrens }
1692789Sahrens
1693789Sahrens void
byteswap_uint64_array(void * vbuf,size_t size)1694789Sahrens byteswap_uint64_array(void *vbuf, size_t size)
1695789Sahrens {
1696789Sahrens uint64_t *buf = vbuf;
1697789Sahrens size_t count = size >> 3;
1698789Sahrens int i;
1699789Sahrens
1700789Sahrens ASSERT((size & 7) == 0);
1701789Sahrens
1702789Sahrens for (i = 0; i < count; i++)
1703789Sahrens buf[i] = BSWAP_64(buf[i]);
1704789Sahrens }
1705789Sahrens
1706789Sahrens void
byteswap_uint32_array(void * vbuf,size_t size)1707789Sahrens byteswap_uint32_array(void *vbuf, size_t size)
1708789Sahrens {
1709789Sahrens uint32_t *buf = vbuf;
1710789Sahrens size_t count = size >> 2;
1711789Sahrens int i;
1712789Sahrens
1713789Sahrens ASSERT((size & 3) == 0);
1714789Sahrens
1715789Sahrens for (i = 0; i < count; i++)
1716789Sahrens buf[i] = BSWAP_32(buf[i]);
1717789Sahrens }
1718789Sahrens
1719789Sahrens void
byteswap_uint16_array(void * vbuf,size_t size)1720789Sahrens byteswap_uint16_array(void *vbuf, size_t size)
1721789Sahrens {
1722789Sahrens uint16_t *buf = vbuf;
1723789Sahrens size_t count = size >> 1;
1724789Sahrens int i;
1725789Sahrens
1726789Sahrens ASSERT((size & 1) == 0);
1727789Sahrens
1728789Sahrens for (i = 0; i < count; i++)
1729789Sahrens buf[i] = BSWAP_16(buf[i]);
1730789Sahrens }
1731789Sahrens
1732789Sahrens /* ARGSUSED */
1733789Sahrens void
byteswap_uint8_array(void * vbuf,size_t size)1734789Sahrens byteswap_uint8_array(void *vbuf, size_t size)
1735789Sahrens {
1736789Sahrens }
1737789Sahrens
1738789Sahrens void
dmu_init(void)1739789Sahrens dmu_init(void)
1740789Sahrens {
174112296SLin.Ling@Sun.COM zfs_dbgmsg_init();
1742*12684STom.Erickson@Sun.COM sa_cache_init();
1743*12684STom.Erickson@Sun.COM xuio_stat_init();
1744*12684STom.Erickson@Sun.COM dmu_objset_init();
1745*12684STom.Erickson@Sun.COM dnode_init();
1746789Sahrens dbuf_init();
174710474SRichard.Morris@Sun.COM zfetch_init();
1748789Sahrens arc_init();
17495450Sbrendan l2arc_init();
1750789Sahrens }
1751789Sahrens
1752789Sahrens void
dmu_fini(void)1753789Sahrens dmu_fini(void)
1754789Sahrens {
1755*12684STom.Erickson@Sun.COM l2arc_fini();
1756789Sahrens arc_fini();
175710474SRichard.Morris@Sun.COM zfetch_fini();
1758*12684STom.Erickson@Sun.COM dbuf_fini();
1759789Sahrens dnode_fini();
1760*12684STom.Erickson@Sun.COM dmu_objset_fini();
176111539SChunli.Zhang@Sun.COM xuio_stat_fini();
176211935SMark.Shellenbaum@Sun.COM sa_cache_fini();
176312296SLin.Ling@Sun.COM zfs_dbgmsg_fini();
1764789Sahrens }
1765