xref: /illumos-gate/usr/src/uts/common/fs/zfs/dmu.c (revision 6d6587176200e3e6c76fd5176b2486ea0eca1a8f)
1fa9e4066Sahrens /*
2fa9e4066Sahrens  * CDDL HEADER START
3fa9e4066Sahrens  *
4fa9e4066Sahrens  * The contents of this file are subject to the terms of the
5ea8dc4b6Seschrock  * Common Development and Distribution License (the "License").
6ea8dc4b6Seschrock  * You may not use this file except in compliance with the License.
7fa9e4066Sahrens  *
8fa9e4066Sahrens  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9fa9e4066Sahrens  * or http://www.opensolaris.org/os/licensing.
10fa9e4066Sahrens  * See the License for the specific language governing permissions
11fa9e4066Sahrens  * and limitations under the License.
12fa9e4066Sahrens  *
13fa9e4066Sahrens  * When distributing Covered Code, include this CDDL HEADER in each
14fa9e4066Sahrens  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15fa9e4066Sahrens  * If applicable, add the following below this CDDL HEADER, with the
16fa9e4066Sahrens  * fields enclosed by brackets "[]" replaced with your own identifying
17fa9e4066Sahrens  * information: Portions Copyright [yyyy] [name of copyright owner]
18fa9e4066Sahrens  *
19fa9e4066Sahrens  * CDDL HEADER END
20fa9e4066Sahrens  */
21fa9e4066Sahrens /*
2294d1a210STim Haley  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23857c96d2SIgor Kozhukhov  */
24857c96d2SIgor Kozhukhov /*
25857c96d2SIgor Kozhukhov  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
26*6d658717SJohn Levon  * Copyright 2019 Joyent, Inc.
27857c96d2SIgor Kozhukhov  * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
28adaec86aSMatthew Ahrens  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
29857c96d2SIgor Kozhukhov  * Copyright (c) 2018 DilOS
30fa9e4066Sahrens  */
31aad02571SSaso Kiselkov 
32fa9e4066Sahrens #include <sys/dmu.h>
33fa9e4066Sahrens #include <sys/dmu_impl.h>
34fa9e4066Sahrens #include <sys/dmu_tx.h>
35fa9e4066Sahrens #include <sys/dbuf.h>
36fa9e4066Sahrens #include <sys/dnode.h>
37fa9e4066Sahrens #include <sys/zfs_context.h>
38fa9e4066Sahrens #include <sys/dmu_objset.h>
39fa9e4066Sahrens #include <sys/dmu_traverse.h>
40fa9e4066Sahrens #include <sys/dsl_dataset.h>
41fa9e4066Sahrens #include <sys/dsl_dir.h>
42fa9e4066Sahrens #include <sys/dsl_pool.h>
431d452cf5Sahrens #include <sys/dsl_synctask.h>
44a2eea2e1Sahrens #include <sys/dsl_prop.h>
45fa9e4066Sahrens #include <sys/dmu_zfetch.h>
46fa9e4066Sahrens #include <sys/zfs_ioctl.h>
47fa9e4066Sahrens #include <sys/zap.h>
48ea8dc4b6Seschrock #include <sys/zio_checksum.h>
4980901aeaSGeorge Wilson #include <sys/zio_compress.h>
500a586ceaSMark Shellenbaum #include <sys/sa.h>
51b8289d24SDaniil Lunev #include <sys/zfeature.h>
52770499e1SDan Kimmel #include <sys/abd.h>
5344eda4d7Smaybee #ifdef _KERNEL
5444eda4d7Smaybee #include <sys/vmsystm.h>
550fab61baSJonathan W Adams #include <sys/zfs_znode.h>
5644eda4d7Smaybee #endif
57fa9e4066Sahrens 
58857c96d2SIgor Kozhukhov static xuio_stats_t xuio_stats = {
59857c96d2SIgor Kozhukhov 	{ "onloan_read_buf",	KSTAT_DATA_UINT64 },
60857c96d2SIgor Kozhukhov 	{ "onloan_write_buf",	KSTAT_DATA_UINT64 },
61857c96d2SIgor Kozhukhov 	{ "read_buf_copied",	KSTAT_DATA_UINT64 },
62857c96d2SIgor Kozhukhov 	{ "read_buf_nocopy",	KSTAT_DATA_UINT64 },
63857c96d2SIgor Kozhukhov 	{ "write_buf_copied",	KSTAT_DATA_UINT64 },
64857c96d2SIgor Kozhukhov 	{ "write_buf_nocopy",	KSTAT_DATA_UINT64 }
65857c96d2SIgor Kozhukhov };
66857c96d2SIgor Kozhukhov 
67857c96d2SIgor Kozhukhov #define	XUIOSTAT_INCR(stat, val)	\
68857c96d2SIgor Kozhukhov 	atomic_add_64(&xuio_stats.stat.value.ui64, (val))
69857c96d2SIgor Kozhukhov #define	XUIOSTAT_BUMP(stat)	XUIOSTAT_INCR(stat, 1)
70857c96d2SIgor Kozhukhov 
7180901aeaSGeorge Wilson /*
7280901aeaSGeorge Wilson  * Enable/disable nopwrite feature.
7380901aeaSGeorge Wilson  */
7480901aeaSGeorge Wilson int zfs_nopwrite_enabled = 1;
7580901aeaSGeorge Wilson 
76ff5177eeSAlek Pinchuk /*
77ff5177eeSAlek Pinchuk  * Tunable to control percentage of dirtied blocks from frees in one TXG.
78ff5177eeSAlek Pinchuk  * After this threshold is crossed, additional dirty blocks from frees
79ff5177eeSAlek Pinchuk  * wait until the next TXG.
80ff5177eeSAlek Pinchuk  * A value of zero will disable this throttle.
81ff5177eeSAlek Pinchuk  */
82ff5177eeSAlek Pinchuk uint32_t zfs_per_txg_dirty_frees_percent = 30;
83ff5177eeSAlek Pinchuk 
845cabbc6bSPrashanth Sreenivasa /*
855cabbc6bSPrashanth Sreenivasa  * This can be used for testing, to ensure that certain actions happen
865cabbc6bSPrashanth Sreenivasa  * while in the middle of a remap (which might otherwise complete too
875cabbc6bSPrashanth Sreenivasa  * quickly).
885cabbc6bSPrashanth Sreenivasa  */
895cabbc6bSPrashanth Sreenivasa int zfs_object_remap_one_indirect_delay_ticks = 0;
905cabbc6bSPrashanth Sreenivasa 
9152abb70eSMatthew Ahrens /*
9252abb70eSMatthew Ahrens  * Limit the amount we can prefetch with one call to this amount.  This
9352abb70eSMatthew Ahrens  * helps to limit the amount of memory that can be used by prefetching.
9452abb70eSMatthew Ahrens  * Larger objects should be prefetched a bit at a time.
9552abb70eSMatthew Ahrens  */
9652abb70eSMatthew Ahrens uint64_t dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE;
9752abb70eSMatthew Ahrens 
98fa9e4066Sahrens const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
99eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  TRUE,  FALSE, FALSE, "unallocated"		},
100eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "object directory"	},
101eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  TRUE,  FALSE, "object array"		},
102eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  TRUE,  FALSE, FALSE, "packed nvlist"	},
103eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "packed nvlist size"	},
104eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "bpobj"	},
105eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "bpobj header"		},
106eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "SPA space map header"	},
107eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "SPA space map"	},
108eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, TRUE,  "ZIL intent log"	},
109eb633035STom Caputi 	{ DMU_BSWAP_DNODE,  TRUE,  FALSE, TRUE,  "DMU dnode"	},
110eb633035STom Caputi 	{ DMU_BSWAP_OBJSET, TRUE,  TRUE,  FALSE, "DMU objset"	},
111eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  TRUE,  FALSE, "DSL directory"	},
112eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "DSL directory child map" },
113eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "DSL dataset snap map"	},
114eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "DSL props"	},
115eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  TRUE,  FALSE, "DSL dataset"	},
116eb633035STom Caputi 	{ DMU_BSWAP_ZNODE,  TRUE,  FALSE, FALSE, "ZFS znode"	},
117eb633035STom Caputi 	{ DMU_BSWAP_OLDACL, TRUE,  FALSE, TRUE,  "ZFS V0 ACL"	},
118eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  FALSE, FALSE, TRUE,  "ZFS plain file"	},
119eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, TRUE,  "ZFS directory"	},
120eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "ZFS master node"	},
121eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, TRUE,  "ZFS delete queue"	},
122eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  FALSE, FALSE, TRUE,  "zvol object"	},
123eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "zvol prop"	},
124eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  FALSE, FALSE, TRUE,  "other uint8[]"	},
125eb633035STom Caputi 	{ DMU_BSWAP_UINT64, FALSE, FALSE, TRUE,  "other uint64[]"	},
126eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "other ZAP"	},
127eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "persistent error log"	},
128eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  TRUE,  FALSE, FALSE, "SPA history"	},
129eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "SPA history offsets"	},
130eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "Pool properties"	},
131eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "DSL permissions"	},
132eb633035STom Caputi 	{ DMU_BSWAP_ACL,    TRUE,  FALSE, TRUE,  "ZFS ACL"	},
133eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  TRUE,  FALSE, TRUE,  "ZFS SYSACL"	},
134eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  TRUE,  FALSE, TRUE,  "FUID table"	},
135eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "FUID table size"	},
136eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "DSL dataset next clones" },
137eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "scan work queue"	},
138eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, TRUE,  "ZFS user/group used"	},
139eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, TRUE,  "ZFS user/group quota"	},
140eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "snapshot refcount tags" },
141eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "DDT ZAP algorithm"	},
142eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "DDT statistics"	},
143eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  TRUE,  FALSE, TRUE,  "System attributes"	},
144eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, TRUE,  "SA master node"	},
145eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, TRUE,  "SA attr registration"	},
146eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, TRUE,  "SA attr layouts"	},
147eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  FALSE, FALSE, "scan translations"	},
148eb633035STom Caputi 	{ DMU_BSWAP_UINT8,  FALSE, FALSE, TRUE,  "deduplicated block"	},
149eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "DSL deadlist map" },
150eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  TRUE,  FALSE, "DSL deadlist map hdr"	},
151eb633035STom Caputi 	{ DMU_BSWAP_ZAP,    TRUE,  TRUE,  FALSE, "DSL dir clones"	},
152eb633035STom Caputi 	{ DMU_BSWAP_UINT64, TRUE,  FALSE, FALSE, "bpobj subobj"		}
153ad135b5dSChristopher Siden };
154ad135b5dSChristopher Siden 
155ad135b5dSChristopher Siden const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
156ad135b5dSChristopher Siden 	{	byteswap_uint8_array,	"uint8"		},
157ad135b5dSChristopher Siden 	{	byteswap_uint16_array,	"uint16"	},
158ad135b5dSChristopher Siden 	{	byteswap_uint32_array,	"uint32"	},
159ad135b5dSChristopher Siden 	{	byteswap_uint64_array,	"uint64"	},
160ad135b5dSChristopher Siden 	{	zap_byteswap,		"zap"		},
161ad135b5dSChristopher Siden 	{	dnode_buf_byteswap,	"dnode"		},
162ad135b5dSChristopher Siden 	{	dmu_objset_byteswap,	"objset"	},
163ad135b5dSChristopher Siden 	{	zfs_znode_byteswap,	"znode"		},
164ad135b5dSChristopher Siden 	{	zfs_oldacl_byteswap,	"oldacl"	},
165ad135b5dSChristopher Siden 	{	zfs_acl_byteswap,	"acl"		}
1663f9d6ad7SLin Ling };
167fa9e4066Sahrens 
16879d72832SMatthew Ahrens int
16979d72832SMatthew Ahrens dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset,
17079d72832SMatthew Ahrens     void *tag, dmu_buf_t **dbp)
17179d72832SMatthew Ahrens {
17279d72832SMatthew Ahrens 	uint64_t blkid;
17379d72832SMatthew Ahrens 	dmu_buf_impl_t *db;
17479d72832SMatthew Ahrens 
17579d72832SMatthew Ahrens 	blkid = dbuf_whichblock(dn, 0, offset);
17679d72832SMatthew Ahrens 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
17779d72832SMatthew Ahrens 	db = dbuf_hold(dn, blkid, tag);
17879d72832SMatthew Ahrens 	rw_exit(&dn->dn_struct_rwlock);
17979d72832SMatthew Ahrens 
18079d72832SMatthew Ahrens 	if (db == NULL) {
18179d72832SMatthew Ahrens 		*dbp = NULL;
18279d72832SMatthew Ahrens 		return (SET_ERROR(EIO));
18379d72832SMatthew Ahrens 	}
18479d72832SMatthew Ahrens 
18579d72832SMatthew Ahrens 	*dbp = &db->db;
18679d72832SMatthew Ahrens 	return (0);
18779d72832SMatthew Ahrens }
188fa9e4066Sahrens int
1895d7b4d43SMatthew Ahrens dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
1905d7b4d43SMatthew Ahrens     void *tag, dmu_buf_t **dbp)
191fa9e4066Sahrens {
192fa9e4066Sahrens 	dnode_t *dn;
193fa9e4066Sahrens 	uint64_t blkid;
194fa9e4066Sahrens 	dmu_buf_impl_t *db;
195ea8dc4b6Seschrock 	int err;
196fa9e4066Sahrens 
197503ad85cSMatthew Ahrens 	err = dnode_hold(os, object, FTAG, &dn);
198ea8dc4b6Seschrock 	if (err)
199ea8dc4b6Seschrock 		return (err);
200a2cdcdd2SPaul Dagnelie 	blkid = dbuf_whichblock(dn, 0, offset);
201fa9e4066Sahrens 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
202ea8dc4b6Seschrock 	db = dbuf_hold(dn, blkid, tag);
203fa9e4066Sahrens 	rw_exit(&dn->dn_struct_rwlock);
2045d7b4d43SMatthew Ahrens 	dnode_rele(dn, FTAG);
2055d7b4d43SMatthew Ahrens 
206ea8dc4b6Seschrock 	if (db == NULL) {
2075d7b4d43SMatthew Ahrens 		*dbp = NULL;
2085d7b4d43SMatthew Ahrens 		return (SET_ERROR(EIO));
2095d7b4d43SMatthew Ahrens 	}
2105d7b4d43SMatthew Ahrens 
2115d7b4d43SMatthew Ahrens 	*dbp = &db->db;
2125d7b4d43SMatthew Ahrens 	return (err);
2135d7b4d43SMatthew Ahrens }
2145d7b4d43SMatthew Ahrens 
21579d72832SMatthew Ahrens int
21679d72832SMatthew Ahrens dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
21779d72832SMatthew Ahrens     void *tag, dmu_buf_t **dbp, int flags)
21879d72832SMatthew Ahrens {
21979d72832SMatthew Ahrens 	int err;
22079d72832SMatthew Ahrens 	int db_flags = DB_RF_CANFAIL;
22179d72832SMatthew Ahrens 
22279d72832SMatthew Ahrens 	if (flags & DMU_READ_NO_PREFETCH)
22379d72832SMatthew Ahrens 		db_flags |= DB_RF_NOPREFETCH;
224eb633035STom Caputi 	if (flags & DMU_READ_NO_DECRYPT)
225eb633035STom Caputi 		db_flags |= DB_RF_NO_DECRYPT;
22679d72832SMatthew Ahrens 
22779d72832SMatthew Ahrens 	err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp);
22879d72832SMatthew Ahrens 	if (err == 0) {
22979d72832SMatthew Ahrens 		dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
23079d72832SMatthew Ahrens 		err = dbuf_read(db, NULL, db_flags);
23179d72832SMatthew Ahrens 		if (err != 0) {
23279d72832SMatthew Ahrens 			dbuf_rele(db, tag);
23379d72832SMatthew Ahrens 			*dbp = NULL;
23479d72832SMatthew Ahrens 		}
23579d72832SMatthew Ahrens 	}
23679d72832SMatthew Ahrens 
23779d72832SMatthew Ahrens 	return (err);
23879d72832SMatthew Ahrens }
23979d72832SMatthew Ahrens 
2405d7b4d43SMatthew Ahrens int
2415d7b4d43SMatthew Ahrens dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
2425d7b4d43SMatthew Ahrens     void *tag, dmu_buf_t **dbp, int flags)
2435d7b4d43SMatthew Ahrens {
2445d7b4d43SMatthew Ahrens 	int err;
2455d7b4d43SMatthew Ahrens 	int db_flags = DB_RF_CANFAIL;
2465d7b4d43SMatthew Ahrens 
2475d7b4d43SMatthew Ahrens 	if (flags & DMU_READ_NO_PREFETCH)
2485d7b4d43SMatthew Ahrens 		db_flags |= DB_RF_NOPREFETCH;
249eb633035STom Caputi 	if (flags & DMU_READ_NO_DECRYPT)
250eb633035STom Caputi 		db_flags |= DB_RF_NO_DECRYPT;
2515d7b4d43SMatthew Ahrens 
2525d7b4d43SMatthew Ahrens 	err = dmu_buf_hold_noread(os, object, offset, tag, dbp);
2535d7b4d43SMatthew Ahrens 	if (err == 0) {
2545d7b4d43SMatthew Ahrens 		dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
25547cb52daSJeff Bonwick 		err = dbuf_read(db, NULL, db_flags);
2565d7b4d43SMatthew Ahrens 		if (err != 0) {
257ea8dc4b6Seschrock 			dbuf_rele(db, tag);
2585d7b4d43SMatthew Ahrens 			*dbp = NULL;
259ea8dc4b6Seschrock 		}
260ea8dc4b6Seschrock 	}
261fa9e4066Sahrens 
262ea8dc4b6Seschrock 	return (err);
263fa9e4066Sahrens }
264fa9e4066Sahrens 
265fa9e4066Sahrens int
266fa9e4066Sahrens dmu_bonus_max(void)
267fa9e4066Sahrens {
26854811da5SToomas Soome 	return (DN_OLD_MAX_BONUSLEN);
269fa9e4066Sahrens }
270fa9e4066Sahrens 
2711934e92fSmaybee int
272744947dcSTom Erickson dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
2731934e92fSmaybee {
274744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
275744947dcSTom Erickson 	dnode_t *dn;
276744947dcSTom Erickson 	int error;
2771934e92fSmaybee 
278744947dcSTom Erickson 	DB_DNODE_ENTER(db);
279744947dcSTom Erickson 	dn = DB_DNODE(db);
280744947dcSTom Erickson 
281744947dcSTom Erickson 	if (dn->dn_bonus != db) {
282be6fd75aSMatthew Ahrens 		error = SET_ERROR(EINVAL);
283744947dcSTom Erickson 	} else if (newsize < 0 || newsize > db_fake->db_size) {
284be6fd75aSMatthew Ahrens 		error = SET_ERROR(EINVAL);
285744947dcSTom Erickson 	} else {
286744947dcSTom Erickson 		dnode_setbonuslen(dn, newsize, tx);
287744947dcSTom Erickson 		error = 0;
288744947dcSTom Erickson 	}
289744947dcSTom Erickson 
290744947dcSTom Erickson 	DB_DNODE_EXIT(db);
291744947dcSTom Erickson 	return (error);
2921934e92fSmaybee }
2931934e92fSmaybee 
2940a586ceaSMark Shellenbaum int
295744947dcSTom Erickson dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
2960a586ceaSMark Shellenbaum {
297744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
298744947dcSTom Erickson 	dnode_t *dn;
299744947dcSTom Erickson 	int error;
3000a586ceaSMark Shellenbaum 
301744947dcSTom Erickson 	DB_DNODE_ENTER(db);
302744947dcSTom Erickson 	dn = DB_DNODE(db);
303744947dcSTom Erickson 
304ad135b5dSChristopher Siden 	if (!DMU_OT_IS_VALID(type)) {
305be6fd75aSMatthew Ahrens 		error = SET_ERROR(EINVAL);
306744947dcSTom Erickson 	} else if (dn->dn_bonus != db) {
307be6fd75aSMatthew Ahrens 		error = SET_ERROR(EINVAL);
308744947dcSTom Erickson 	} else {
309744947dcSTom Erickson 		dnode_setbonus_type(dn, type, tx);
310744947dcSTom Erickson 		error = 0;
311744947dcSTom Erickson 	}
3120a586ceaSMark Shellenbaum 
313744947dcSTom Erickson 	DB_DNODE_EXIT(db);
314744947dcSTom Erickson 	return (error);
315744947dcSTom Erickson }
3160a586ceaSMark Shellenbaum 
317744947dcSTom Erickson dmu_object_type_t
318744947dcSTom Erickson dmu_get_bonustype(dmu_buf_t *db_fake)
319744947dcSTom Erickson {
320744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
321744947dcSTom Erickson 	dnode_t *dn;
322744947dcSTom Erickson 	dmu_object_type_t type;
323744947dcSTom Erickson 
324744947dcSTom Erickson 	DB_DNODE_ENTER(db);
325744947dcSTom Erickson 	dn = DB_DNODE(db);
326744947dcSTom Erickson 	type = dn->dn_bonustype;
327744947dcSTom Erickson 	DB_DNODE_EXIT(db);
328744947dcSTom Erickson 
329744947dcSTom Erickson 	return (type);
3300a586ceaSMark Shellenbaum }
3310a586ceaSMark Shellenbaum 
3320a586ceaSMark Shellenbaum int
3330a586ceaSMark Shellenbaum dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
3340a586ceaSMark Shellenbaum {
3350a586ceaSMark Shellenbaum 	dnode_t *dn;
3360a586ceaSMark Shellenbaum 	int error;
3370a586ceaSMark Shellenbaum 
3380a586ceaSMark Shellenbaum 	error = dnode_hold(os, object, FTAG, &dn);
3390a586ceaSMark Shellenbaum 	dbuf_rm_spill(dn, tx);
34006e0070dSMark Shellenbaum 	rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
34106e0070dSMark Shellenbaum 	dnode_rm_spill(dn, tx);
34206e0070dSMark Shellenbaum 	rw_exit(&dn->dn_struct_rwlock);
3430a586ceaSMark Shellenbaum 	dnode_rele(dn, FTAG);
3440a586ceaSMark Shellenbaum 	return (error);
3450a586ceaSMark Shellenbaum }
3460a586ceaSMark Shellenbaum 
347eb633035STom Caputi /*
348eb633035STom Caputi  * Lookup and hold the bonus buffer for the provided dnode.  If the dnode
349eb633035STom Caputi  * has not yet been allocated a new bonus dbuf a will be allocated.
350eb633035STom Caputi  * Returns ENOENT, EIO, or 0.
351eb633035STom Caputi  */
352eb633035STom Caputi int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp,
353eb633035STom Caputi     uint32_t flags)
354eb633035STom Caputi {
355eb633035STom Caputi 	dmu_buf_impl_t *db;
356eb633035STom Caputi 	int error;
357eb633035STom Caputi 	uint32_t db_flags = DB_RF_MUST_SUCCEED;
358eb633035STom Caputi 
359eb633035STom Caputi 	if (flags & DMU_READ_NO_PREFETCH)
360eb633035STom Caputi 		db_flags |= DB_RF_NOPREFETCH;
361eb633035STom Caputi 	if (flags & DMU_READ_NO_DECRYPT)
362*6d658717SJohn Levon 		db_flags |= DB_RF_NO_DECRYPT;
363eb633035STom Caputi 
364eb633035STom Caputi 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
365eb633035STom Caputi 	if (dn->dn_bonus == NULL) {
366eb633035STom Caputi 		rw_exit(&dn->dn_struct_rwlock);
367eb633035STom Caputi 		rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
368eb633035STom Caputi 		if (dn->dn_bonus == NULL)
369eb633035STom Caputi 			dbuf_create_bonus(dn);
370eb633035STom Caputi 	}
371eb633035STom Caputi 	db = dn->dn_bonus;
372eb633035STom Caputi 
373eb633035STom Caputi 	/* as long as the bonus buf is held, the dnode will be held */
374eb633035STom Caputi 	if (zfs_refcount_add(&db->db_holds, tag) == 1) {
375eb633035STom Caputi 		VERIFY(dnode_add_ref(dn, db));
376eb633035STom Caputi 		atomic_inc_32(&dn->dn_dbufs_count);
377eb633035STom Caputi 	}
378eb633035STom Caputi 
379eb633035STom Caputi 	/*
380eb633035STom Caputi 	 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
381eb633035STom Caputi 	 * hold and incrementing the dbuf count to ensure that dnode_move() sees
382eb633035STom Caputi 	 * a dnode hold for every dbuf.
383eb633035STom Caputi 	 */
384eb633035STom Caputi 	rw_exit(&dn->dn_struct_rwlock);
385eb633035STom Caputi 
386eb633035STom Caputi 	error = dbuf_read(db, NULL, db_flags);
387eb633035STom Caputi 	if (error) {
388eb633035STom Caputi 		dnode_evict_bonus(dn);
389eb633035STom Caputi 		dbuf_rele(db, tag);
390eb633035STom Caputi 		*dbp = NULL;
391eb633035STom Caputi 		return (error);
392eb633035STom Caputi 	}
393eb633035STom Caputi 
394eb633035STom Caputi 	*dbp = &db->db;
395eb633035STom Caputi 	return (0);
396eb633035STom Caputi }
397eb633035STom Caputi 
398fa9e4066Sahrens /*
399ea8dc4b6Seschrock  * returns ENOENT, EIO, or 0.
400fa9e4066Sahrens  */
401ea8dc4b6Seschrock int
402eb633035STom Caputi dmu_bonus_hold_impl(objset_t *os, uint64_t object, void *tag, uint32_t flags,
403eb633035STom Caputi     dmu_buf_t **dbp)
404fa9e4066Sahrens {
405ea8dc4b6Seschrock 	dnode_t *dn;
406fa9e4066Sahrens 	dmu_buf_impl_t *db;
4071934e92fSmaybee 	int error;
408eb633035STom Caputi 	uint32_t db_flags = DB_RF_MUST_SUCCEED;
409eb633035STom Caputi 
410eb633035STom Caputi 	if (flags & DMU_READ_NO_PREFETCH)
411eb633035STom Caputi 		db_flags |= DB_RF_NOPREFETCH;
412eb633035STom Caputi 	if (flags & DMU_READ_NO_DECRYPT)
413eb633035STom Caputi 		db_flags |= DB_RF_NO_DECRYPT;
414fa9e4066Sahrens 
415503ad85cSMatthew Ahrens 	error = dnode_hold(os, object, FTAG, &dn);
4161934e92fSmaybee 	if (error)
4171934e92fSmaybee 		return (error);
418fa9e4066Sahrens 
419ea8dc4b6Seschrock 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
420ea8dc4b6Seschrock 	if (dn->dn_bonus == NULL) {
421fa9e4066Sahrens 		rw_exit(&dn->dn_struct_rwlock);
422ea8dc4b6Seschrock 		rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
423ea8dc4b6Seschrock 		if (dn->dn_bonus == NULL)
4241934e92fSmaybee 			dbuf_create_bonus(dn);
425fa9e4066Sahrens 	}
426ea8dc4b6Seschrock 	db = dn->dn_bonus;
4271934e92fSmaybee 
4281934e92fSmaybee 	/* as long as the bonus buf is held, the dnode will be held */
429e914ace2STim Schumacher 	if (zfs_refcount_add(&db->db_holds, tag) == 1) {
4301934e92fSmaybee 		VERIFY(dnode_add_ref(dn, db));
431640c1670SJosef 'Jeff' Sipek 		atomic_inc_32(&dn->dn_dbufs_count);
432744947dcSTom Erickson 	}
433744947dcSTom Erickson 
434744947dcSTom Erickson 	/*
435744947dcSTom Erickson 	 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
436744947dcSTom Erickson 	 * hold and incrementing the dbuf count to ensure that dnode_move() sees
437744947dcSTom Erickson 	 * a dnode hold for every dbuf.
438744947dcSTom Erickson 	 */
439744947dcSTom Erickson 	rw_exit(&dn->dn_struct_rwlock);
4401934e92fSmaybee 
441fa9e4066Sahrens 	dnode_rele(dn, FTAG);
442ea8dc4b6Seschrock 
443eb633035STom Caputi 	error = dbuf_read(db, NULL, db_flags);
444eb633035STom Caputi 	if (error) {
445eb633035STom Caputi 		dnode_evict_bonus(dn);
446eb633035STom Caputi 		dbuf_rele(db, tag);
447eb633035STom Caputi 		*dbp = NULL;
448eb633035STom Caputi 		return (error);
449eb633035STom Caputi 	}
450ea8dc4b6Seschrock 
451ea8dc4b6Seschrock 	*dbp = &db->db;
452ea8dc4b6Seschrock 	return (0);
453fa9e4066Sahrens }
454fa9e4066Sahrens 
455eb633035STom Caputi int
456eb633035STom Caputi dmu_bonus_hold(objset_t *os, uint64_t obj, void *tag, dmu_buf_t **dbp)
457eb633035STom Caputi {
458eb633035STom Caputi 	return (dmu_bonus_hold_impl(os, obj, tag, DMU_READ_NO_PREFETCH, dbp));
459eb633035STom Caputi }
460eb633035STom Caputi 
4610a586ceaSMark Shellenbaum /*
4620a586ceaSMark Shellenbaum  * returns ENOENT, EIO, or 0.
4630a586ceaSMark Shellenbaum  *
4640a586ceaSMark Shellenbaum  * This interface will allocate a blank spill dbuf when a spill blk
4650a586ceaSMark Shellenbaum  * doesn't already exist on the dnode.
4660a586ceaSMark Shellenbaum  *
4670a586ceaSMark Shellenbaum  * if you only want to find an already existing spill db, then
4680a586ceaSMark Shellenbaum  * dmu_spill_hold_existing() should be used.
4690a586ceaSMark Shellenbaum  */
4700a586ceaSMark Shellenbaum int
4710a586ceaSMark Shellenbaum dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
4720a586ceaSMark Shellenbaum {
4730a586ceaSMark Shellenbaum 	dmu_buf_impl_t *db = NULL;
4740a586ceaSMark Shellenbaum 	int err;
4750a586ceaSMark Shellenbaum 
4760a586ceaSMark Shellenbaum 	if ((flags & DB_RF_HAVESTRUCT) == 0)
4770a586ceaSMark Shellenbaum 		rw_enter(&dn->dn_struct_rwlock, RW_READER);
4780a586ceaSMark Shellenbaum 
4790a586ceaSMark Shellenbaum 	db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
4800a586ceaSMark Shellenbaum 
4810a586ceaSMark Shellenbaum 	if ((flags & DB_RF_HAVESTRUCT) == 0)
4820a586ceaSMark Shellenbaum 		rw_exit(&dn->dn_struct_rwlock);
4830a586ceaSMark Shellenbaum 
4840a586ceaSMark Shellenbaum 	ASSERT(db != NULL);
4851d8ccc7bSMark Shellenbaum 	err = dbuf_read(db, NULL, flags);
4861d8ccc7bSMark Shellenbaum 	if (err == 0)
4871d8ccc7bSMark Shellenbaum 		*dbp = &db->db;
4881d8ccc7bSMark Shellenbaum 	else
4891d8ccc7bSMark Shellenbaum 		dbuf_rele(db, tag);
4900a586ceaSMark Shellenbaum 	return (err);
4910a586ceaSMark Shellenbaum }
4920a586ceaSMark Shellenbaum 
4930a586ceaSMark Shellenbaum int
4940a586ceaSMark Shellenbaum dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
4950a586ceaSMark Shellenbaum {
496744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
497744947dcSTom Erickson 	dnode_t *dn;
4980a586ceaSMark Shellenbaum 	int err;
4990a586ceaSMark Shellenbaum 
500744947dcSTom Erickson 	DB_DNODE_ENTER(db);
501744947dcSTom Erickson 	dn = DB_DNODE(db);
502744947dcSTom Erickson 
503744947dcSTom Erickson 	if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
504be6fd75aSMatthew Ahrens 		err = SET_ERROR(EINVAL);
505744947dcSTom Erickson 	} else {
506744947dcSTom Erickson 		rw_enter(&dn->dn_struct_rwlock, RW_READER);
507744947dcSTom Erickson 
508744947dcSTom Erickson 		if (!dn->dn_have_spill) {
509be6fd75aSMatthew Ahrens 			err = SET_ERROR(ENOENT);
510744947dcSTom Erickson 		} else {
511744947dcSTom Erickson 			err = dmu_spill_hold_by_dnode(dn,
512744947dcSTom Erickson 			    DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
513744947dcSTom Erickson 		}
5140a586ceaSMark Shellenbaum 
5150a586ceaSMark Shellenbaum 		rw_exit(&dn->dn_struct_rwlock);
5160a586ceaSMark Shellenbaum 	}
517744947dcSTom Erickson 
518744947dcSTom Erickson 	DB_DNODE_EXIT(db);
5190a586ceaSMark Shellenbaum 	return (err);
5200a586ceaSMark Shellenbaum }
5210a586ceaSMark Shellenbaum 
5220a586ceaSMark Shellenbaum int
523eb633035STom Caputi dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag,
524eb633035STom Caputi     dmu_buf_t **dbp)
5250a586ceaSMark Shellenbaum {
526744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
527744947dcSTom Erickson 	dnode_t *dn;
528744947dcSTom Erickson 	int err;
529eb633035STom Caputi 	uint32_t db_flags = DB_RF_CANFAIL;
530eb633035STom Caputi 
531eb633035STom Caputi 	if (flags & DMU_READ_NO_DECRYPT)
532eb633035STom Caputi 		db_flags |= DB_RF_NO_DECRYPT;
533744947dcSTom Erickson 
534744947dcSTom Erickson 	DB_DNODE_ENTER(db);
535744947dcSTom Erickson 	dn = DB_DNODE(db);
536eb633035STom Caputi 	err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp);
537744947dcSTom Erickson 	DB_DNODE_EXIT(db);
538744947dcSTom Erickson 
539744947dcSTom Erickson 	return (err);
5400a586ceaSMark Shellenbaum }
5410a586ceaSMark Shellenbaum 
54213506d1eSmaybee /*
54313506d1eSmaybee  * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
54413506d1eSmaybee  * to take a held dnode rather than <os, object> -- the lookup is wasteful,
54513506d1eSmaybee  * and can induce severe lock contention when writing to several files
54613506d1eSmaybee  * whose dnodes are in the same block.
54713506d1eSmaybee  */
5488dfe5547SRichard Yao int
5497bfdf011SNeil Perrin dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
550cf6106c8SMatthew Ahrens     boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
551fa9e4066Sahrens {
552fa9e4066Sahrens 	dmu_buf_t **dbp;
553fa9e4066Sahrens 	uint64_t blkid, nblks, i;
5547bfdf011SNeil Perrin 	uint32_t dbuf_flags;
555ea8dc4b6Seschrock 	int err;
556ea8dc4b6Seschrock 	zio_t *zio;
557ea8dc4b6Seschrock 
558ea8dc4b6Seschrock 	ASSERT(length <= DMU_MAX_ACCESS);
559fa9e4066Sahrens 
560cf6106c8SMatthew Ahrens 	/*
561cf6106c8SMatthew Ahrens 	 * Note: We directly notify the prefetch code of this read, so that
562cf6106c8SMatthew Ahrens 	 * we can tell it about the multi-block read.  dbuf_read() only knows
563cf6106c8SMatthew Ahrens 	 * about the one block it is accessing.
564cf6106c8SMatthew Ahrens 	 */
565cf6106c8SMatthew Ahrens 	dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT |
566cf6106c8SMatthew Ahrens 	    DB_RF_NOPREFETCH;
567ea8dc4b6Seschrock 
568fa9e4066Sahrens 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
569fa9e4066Sahrens 	if (dn->dn_datablkshift) {
570fa9e4066Sahrens 		int blkshift = dn->dn_datablkshift;
571cf6106c8SMatthew Ahrens 		nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) -
572cf6106c8SMatthew Ahrens 		    P2ALIGN(offset, 1ULL << blkshift)) >> blkshift;
573fa9e4066Sahrens 	} else {
5740125049cSahrens 		if (offset + length > dn->dn_datablksz) {
5750125049cSahrens 			zfs_panic_recover("zfs: accessing past end of object "
5760125049cSahrens 			    "%llx/%llx (size=%u access=%llu+%llu)",
5770125049cSahrens 			    (longlong_t)dn->dn_objset->
5780125049cSahrens 			    os_dsl_dataset->ds_object,
5790125049cSahrens 			    (longlong_t)dn->dn_object, dn->dn_datablksz,
5800125049cSahrens 			    (longlong_t)offset, (longlong_t)length);
581c87b8fc5SMark J Musante 			rw_exit(&dn->dn_struct_rwlock);
582be6fd75aSMatthew Ahrens 			return (SET_ERROR(EIO));
5830125049cSahrens 		}
584fa9e4066Sahrens 		nblks = 1;
585fa9e4066Sahrens 	}
586ea8dc4b6Seschrock 	dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
587fa9e4066Sahrens 
588e14bb325SJeff Bonwick 	zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
589a2cdcdd2SPaul Dagnelie 	blkid = dbuf_whichblock(dn, 0, offset);
590fa9e4066Sahrens 	for (i = 0; i < nblks; i++) {
591cf6106c8SMatthew Ahrens 		dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag);
592ea8dc4b6Seschrock 		if (db == NULL) {
593ea8dc4b6Seschrock 			rw_exit(&dn->dn_struct_rwlock);
594ea8dc4b6Seschrock 			dmu_buf_rele_array(dbp, nblks, tag);
595ea8dc4b6Seschrock 			zio_nowait(zio);
596be6fd75aSMatthew Ahrens 			return (SET_ERROR(EIO));
597ea8dc4b6Seschrock 		}
598cf6106c8SMatthew Ahrens 
599ea8dc4b6Seschrock 		/* initiate async i/o */
600cf6106c8SMatthew Ahrens 		if (read)
6017bfdf011SNeil Perrin 			(void) dbuf_read(db, zio, dbuf_flags);
602ea8dc4b6Seschrock 		dbp[i] = &db->db;
603fa9e4066Sahrens 	}
604cf6106c8SMatthew Ahrens 
605cb92f413SAlexander Motin 	if ((flags & DMU_READ_NO_PREFETCH) == 0 &&
606cb92f413SAlexander Motin 	    DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) {
607cb92f413SAlexander Motin 		dmu_zfetch(&dn->dn_zfetch, blkid, nblks,
608cb92f413SAlexander Motin 		    read && DNODE_IS_CACHEABLE(dn));
609cf6106c8SMatthew Ahrens 	}
610fa9e4066Sahrens 	rw_exit(&dn->dn_struct_rwlock);
611fa9e4066Sahrens 
612ea8dc4b6Seschrock 	/* wait for async i/o */
613ea8dc4b6Seschrock 	err = zio_wait(zio);
614ea8dc4b6Seschrock 	if (err) {
615ea8dc4b6Seschrock 		dmu_buf_rele_array(dbp, nblks, tag);
616ea8dc4b6Seschrock 		return (err);
617ea8dc4b6Seschrock 	}
618fa9e4066Sahrens 
619ea8dc4b6Seschrock 	/* wait for other io to complete */
620ea8dc4b6Seschrock 	if (read) {
621ea8dc4b6Seschrock 		for (i = 0; i < nblks; i++) {
622ea8dc4b6Seschrock 			dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
623ea8dc4b6Seschrock 			mutex_enter(&db->db_mtx);
624ea8dc4b6Seschrock 			while (db->db_state == DB_READ ||
625ea8dc4b6Seschrock 			    db->db_state == DB_FILL)
626ea8dc4b6Seschrock 				cv_wait(&db->db_changed, &db->db_mtx);
627ea8dc4b6Seschrock 			if (db->db_state == DB_UNCACHED)
628be6fd75aSMatthew Ahrens 				err = SET_ERROR(EIO);
629ea8dc4b6Seschrock 			mutex_exit(&db->db_mtx);
630ea8dc4b6Seschrock 			if (err) {
631ea8dc4b6Seschrock 				dmu_buf_rele_array(dbp, nblks, tag);
632ea8dc4b6Seschrock 				return (err);
633ea8dc4b6Seschrock 			}
634ea8dc4b6Seschrock 		}
635ea8dc4b6Seschrock 	}
636fa9e4066Sahrens 
637ea8dc4b6Seschrock 	*numbufsp = nblks;
638ea8dc4b6Seschrock 	*dbpp = dbp;
639ea8dc4b6Seschrock 	return (0);
640fa9e4066Sahrens }
641fa9e4066Sahrens 
642a2eea2e1Sahrens static int
64313506d1eSmaybee dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
64413506d1eSmaybee     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
64513506d1eSmaybee {
64613506d1eSmaybee 	dnode_t *dn;
64713506d1eSmaybee 	int err;
64813506d1eSmaybee 
649503ad85cSMatthew Ahrens 	err = dnode_hold(os, object, FTAG, &dn);
65013506d1eSmaybee 	if (err)
65113506d1eSmaybee 		return (err);
65213506d1eSmaybee 
65313506d1eSmaybee 	err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
6547bfdf011SNeil Perrin 	    numbufsp, dbpp, DMU_READ_PREFETCH);
65513506d1eSmaybee 
65613506d1eSmaybee 	dnode_rele(dn, FTAG);
65713506d1eSmaybee 
65813506d1eSmaybee 	return (err);
65913506d1eSmaybee }
66013506d1eSmaybee 
66113506d1eSmaybee int
662744947dcSTom Erickson dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
663cf6106c8SMatthew Ahrens     uint64_t length, boolean_t read, void *tag, int *numbufsp,
664cf6106c8SMatthew Ahrens     dmu_buf_t ***dbpp)
66513506d1eSmaybee {
666744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
667744947dcSTom Erickson 	dnode_t *dn;
66813506d1eSmaybee 	int err;
66913506d1eSmaybee 
670744947dcSTom Erickson 	DB_DNODE_ENTER(db);
671744947dcSTom Erickson 	dn = DB_DNODE(db);
67213506d1eSmaybee 	err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
6737bfdf011SNeil Perrin 	    numbufsp, dbpp, DMU_READ_PREFETCH);
674744947dcSTom Erickson 	DB_DNODE_EXIT(db);
67513506d1eSmaybee 
67613506d1eSmaybee 	return (err);
67713506d1eSmaybee }
67813506d1eSmaybee 
679fa9e4066Sahrens void
680ea8dc4b6Seschrock dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
681fa9e4066Sahrens {
682fa9e4066Sahrens 	int i;
683fa9e4066Sahrens 	dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
684fa9e4066Sahrens 
685fa9e4066Sahrens 	if (numbufs == 0)
686fa9e4066Sahrens 		return;
687fa9e4066Sahrens 
688ea8dc4b6Seschrock 	for (i = 0; i < numbufs; i++) {
689ea8dc4b6Seschrock 		if (dbp[i])
690ea8dc4b6Seschrock 			dbuf_rele(dbp[i], tag);
691ea8dc4b6Seschrock 	}
692fa9e4066Sahrens 
693fa9e4066Sahrens 	kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
694fa9e4066Sahrens }
695fa9e4066Sahrens 
69669962b56SMatthew Ahrens /*
697a2cdcdd2SPaul Dagnelie  * Issue prefetch i/os for the given blocks.  If level is greater than 0, the
698a2cdcdd2SPaul Dagnelie  * indirect blocks prefeteched will be those that point to the blocks containing
699a2cdcdd2SPaul Dagnelie  * the data starting at offset, and continuing to offset + len.
70069962b56SMatthew Ahrens  *
701eb633035STom Caputi  * Note that if the indirect blocks above the blocks being prefetched are not
702eb633035STom Caputi  * in cache, they will be asychronously read in.
70369962b56SMatthew Ahrens  */
704fa9e4066Sahrens void
705a2cdcdd2SPaul Dagnelie dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
706a2cdcdd2SPaul Dagnelie     uint64_t len, zio_priority_t pri)
707fa9e4066Sahrens {
708fa9e4066Sahrens 	dnode_t *dn;
709fa9e4066Sahrens 	uint64_t blkid;
71069962b56SMatthew Ahrens 	int nblks, err;
711fa9e4066Sahrens 
712fa9e4066Sahrens 	if (len == 0) {  /* they're interested in the bonus buffer */
713744947dcSTom Erickson 		dn = DMU_META_DNODE(os);
714fa9e4066Sahrens 
715fa9e4066Sahrens 		if (object == 0 || object >= DN_MAX_OBJECT)
716fa9e4066Sahrens 			return;
717fa9e4066Sahrens 
718fa9e4066Sahrens 		rw_enter(&dn->dn_struct_rwlock, RW_READER);
719a2cdcdd2SPaul Dagnelie 		blkid = dbuf_whichblock(dn, level,
720a2cdcdd2SPaul Dagnelie 		    object * sizeof (dnode_phys_t));
721a2cdcdd2SPaul Dagnelie 		dbuf_prefetch(dn, level, blkid, pri, 0);
722fa9e4066Sahrens 		rw_exit(&dn->dn_struct_rwlock);
723fa9e4066Sahrens 		return;
724fa9e4066Sahrens 	}
725fa9e4066Sahrens 
72652abb70eSMatthew Ahrens 	/*
72752abb70eSMatthew Ahrens 	 * See comment before the definition of dmu_prefetch_max.
72852abb70eSMatthew Ahrens 	 */
72952abb70eSMatthew Ahrens 	len = MIN(len, dmu_prefetch_max);
73052abb70eSMatthew Ahrens 
731fa9e4066Sahrens 	/*
732fa9e4066Sahrens 	 * XXX - Note, if the dnode for the requested object is not
733fa9e4066Sahrens 	 * already cached, we will do a *synchronous* read in the
734fa9e4066Sahrens 	 * dnode_hold() call.  The same is true for any indirects.
735fa9e4066Sahrens 	 */
736503ad85cSMatthew Ahrens 	err = dnode_hold(os, object, FTAG, &dn);
737ea8dc4b6Seschrock 	if (err != 0)
738fa9e4066Sahrens 		return;
739fa9e4066Sahrens 
740fa9e4066Sahrens 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
741a2cdcdd2SPaul Dagnelie 	/*
742a2cdcdd2SPaul Dagnelie 	 * offset + len - 1 is the last byte we want to prefetch for, and offset
743a2cdcdd2SPaul Dagnelie 	 * is the first.  Then dbuf_whichblk(dn, level, off + len - 1) is the
744a2cdcdd2SPaul Dagnelie 	 * last block we want to prefetch, and dbuf_whichblock(dn, level,
745a2cdcdd2SPaul Dagnelie 	 * offset)  is the first.  Then the number we need to prefetch is the
746a2cdcdd2SPaul Dagnelie 	 * last - first + 1.
747a2cdcdd2SPaul Dagnelie 	 */
748a2cdcdd2SPaul Dagnelie 	if (level > 0 || dn->dn_datablkshift != 0) {
749a2cdcdd2SPaul Dagnelie 		nblks = dbuf_whichblock(dn, level, offset + len - 1) -
750a2cdcdd2SPaul Dagnelie 		    dbuf_whichblock(dn, level, offset) + 1;
751fa9e4066Sahrens 	} else {
752fa9e4066Sahrens 		nblks = (offset < dn->dn_datablksz);
753fa9e4066Sahrens 	}
754fa9e4066Sahrens 
755fa9e4066Sahrens 	if (nblks != 0) {
756a2cdcdd2SPaul Dagnelie 		blkid = dbuf_whichblock(dn, level, offset);
75769962b56SMatthew Ahrens 		for (int i = 0; i < nblks; i++)
758a2cdcdd2SPaul Dagnelie 			dbuf_prefetch(dn, level, blkid + i, pri, 0);
759fa9e4066Sahrens 	}
760fa9e4066Sahrens 
761fa9e4066Sahrens 	rw_exit(&dn->dn_struct_rwlock);
762fa9e4066Sahrens 
763fa9e4066Sahrens 	dnode_rele(dn, FTAG);
764fa9e4066Sahrens }
765fa9e4066Sahrens 
76676256205SMark Maybee /*
76776256205SMark Maybee  * Get the next "chunk" of file data to free.  We traverse the file from
76876256205SMark Maybee  * the end so that the file gets shorter over time (if we crashes in the
76976256205SMark Maybee  * middle, this will leave us in a better state).  We find allocated file
77076256205SMark Maybee  * data by simply searching the allocated level 1 indirects.
771713d6c20SMatthew Ahrens  *
772713d6c20SMatthew Ahrens  * On input, *start should be the first offset that does not need to be
773713d6c20SMatthew Ahrens  * freed (e.g. "offset + length").  On return, *start will be the first
774713d6c20SMatthew Ahrens  * offset that should be freed.
77576256205SMark Maybee  */
776cdb0ab79Smaybee static int
777713d6c20SMatthew Ahrens get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum)
778cdb0ab79Smaybee {
779713d6c20SMatthew Ahrens 	uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
780713d6c20SMatthew Ahrens 	/* bytes of data covered by a level-1 indirect block */
78176256205SMark Maybee 	uint64_t iblkrange =
7821c8564a7SMark Maybee 	    dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
783cdb0ab79Smaybee 
784713d6c20SMatthew Ahrens 	ASSERT3U(minimum, <=, *start);
785cdb0ab79Smaybee 
786713d6c20SMatthew Ahrens 	if (*start - minimum <= iblkrange * maxblks) {
787713d6c20SMatthew Ahrens 		*start = minimum;
788cdb0ab79Smaybee 		return (0);
789cdb0ab79Smaybee 	}
79076256205SMark Maybee 	ASSERT(ISP2(iblkrange));
791cdb0ab79Smaybee 
792713d6c20SMatthew Ahrens 	for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) {
7931c8564a7SMark Maybee 		int err;
794cdb0ab79Smaybee 
795713d6c20SMatthew Ahrens 		/*
796713d6c20SMatthew Ahrens 		 * dnode_next_offset(BACKWARDS) will find an allocated L1
797713d6c20SMatthew Ahrens 		 * indirect block at or before the input offset.  We must
798713d6c20SMatthew Ahrens 		 * decrement *start so that it is at the end of the region
799713d6c20SMatthew Ahrens 		 * to search.
800713d6c20SMatthew Ahrens 		 */
801713d6c20SMatthew Ahrens 		(*start)--;
802cdb0ab79Smaybee 		err = dnode_next_offset(dn,
80376256205SMark Maybee 		    DNODE_FIND_BACKWARDS, start, 2, 1, 0);
804cdb0ab79Smaybee 
805713d6c20SMatthew Ahrens 		/* if there are no indirect blocks before start, we are done */
80676256205SMark Maybee 		if (err == ESRCH) {
807713d6c20SMatthew Ahrens 			*start = minimum;
808713d6c20SMatthew Ahrens 			break;
809713d6c20SMatthew Ahrens 		} else if (err != 0) {
810cdb0ab79Smaybee 			return (err);
81176256205SMark Maybee 		}
81276256205SMark Maybee 
813713d6c20SMatthew Ahrens 		/* set start to the beginning of this L1 indirect */
81476256205SMark Maybee 		*start = P2ALIGN(*start, iblkrange);
815cdb0ab79Smaybee 	}
816713d6c20SMatthew Ahrens 	if (*start < minimum)
817713d6c20SMatthew Ahrens 		*start = minimum;
818cdb0ab79Smaybee 	return (0);
819cdb0ab79Smaybee }
820cdb0ab79Smaybee 
821eb721827SAlek Pinchuk /*
822eb721827SAlek Pinchuk  * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set,
823eb721827SAlek Pinchuk  * otherwise return false.
824eb721827SAlek Pinchuk  * Used below in dmu_free_long_range_impl() to enable abort when unmounting
825eb721827SAlek Pinchuk  */
826eb721827SAlek Pinchuk /*ARGSUSED*/
827eb721827SAlek Pinchuk static boolean_t
828eb721827SAlek Pinchuk dmu_objset_zfs_unmounting(objset_t *os)
829eb721827SAlek Pinchuk {
830eb721827SAlek Pinchuk #ifdef _KERNEL
831eb721827SAlek Pinchuk 	if (dmu_objset_type(os) == DMU_OST_ZFS)
832eb721827SAlek Pinchuk 		return (zfs_get_vfs_flag_unmounted(os));
833eb721827SAlek Pinchuk #endif
834eb721827SAlek Pinchuk 	return (B_FALSE);
835eb721827SAlek Pinchuk }
836eb721827SAlek Pinchuk 
837cdb0ab79Smaybee static int
838cdb0ab79Smaybee dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
839713d6c20SMatthew Ahrens     uint64_t length)
840cdb0ab79Smaybee {
841713d6c20SMatthew Ahrens 	uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
842713d6c20SMatthew Ahrens 	int err;
843ff5177eeSAlek Pinchuk 	uint64_t dirty_frees_threshold;
844ff5177eeSAlek Pinchuk 	dsl_pool_t *dp = dmu_objset_pool(os);
845713d6c20SMatthew Ahrens 
846713d6c20SMatthew Ahrens 	if (offset >= object_size)
847cdb0ab79Smaybee 		return (0);
848cdb0ab79Smaybee 
849ff5177eeSAlek Pinchuk 	if (zfs_per_txg_dirty_frees_percent <= 100)
850ff5177eeSAlek Pinchuk 		dirty_frees_threshold =
851ff5177eeSAlek Pinchuk 		    zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100;
852ff5177eeSAlek Pinchuk 	else
853ff5177eeSAlek Pinchuk 		dirty_frees_threshold = zfs_dirty_data_max / 4;
854ff5177eeSAlek Pinchuk 
855713d6c20SMatthew Ahrens 	if (length == DMU_OBJECT_END || offset + length > object_size)
856713d6c20SMatthew Ahrens 		length = object_size - offset;
857713d6c20SMatthew Ahrens 
858713d6c20SMatthew Ahrens 	while (length != 0) {
859ff5177eeSAlek Pinchuk 		uint64_t chunk_end, chunk_begin, chunk_len;
860ff5177eeSAlek Pinchuk 		uint64_t long_free_dirty_all_txgs = 0;
861ff5177eeSAlek Pinchuk 		dmu_tx_t *tx;
862713d6c20SMatthew Ahrens 
863eb721827SAlek Pinchuk 		if (dmu_objset_zfs_unmounting(dn->dn_objset))
864eb721827SAlek Pinchuk 			return (SET_ERROR(EINTR));
865eb721827SAlek Pinchuk 
866713d6c20SMatthew Ahrens 		chunk_end = chunk_begin = offset + length;
867713d6c20SMatthew Ahrens 
868713d6c20SMatthew Ahrens 		/* move chunk_begin backwards to the beginning of this chunk */
869713d6c20SMatthew Ahrens 		err = get_next_chunk(dn, &chunk_begin, offset);
870cdb0ab79Smaybee 		if (err)
871cdb0ab79Smaybee 			return (err);
872713d6c20SMatthew Ahrens 		ASSERT3U(chunk_begin, >=, offset);
873713d6c20SMatthew Ahrens 		ASSERT3U(chunk_begin, <=, chunk_end);
874cdb0ab79Smaybee 
875ff5177eeSAlek Pinchuk 		chunk_len = chunk_end - chunk_begin;
876ff5177eeSAlek Pinchuk 
877ff5177eeSAlek Pinchuk 		mutex_enter(&dp->dp_lock);
878ff5177eeSAlek Pinchuk 		for (int t = 0; t < TXG_SIZE; t++) {
879ff5177eeSAlek Pinchuk 			long_free_dirty_all_txgs +=
880ff5177eeSAlek Pinchuk 			    dp->dp_long_free_dirty_pertxg[t];
881ff5177eeSAlek Pinchuk 		}
882ff5177eeSAlek Pinchuk 		mutex_exit(&dp->dp_lock);
883ff5177eeSAlek Pinchuk 
884ff5177eeSAlek Pinchuk 		/*
885ff5177eeSAlek Pinchuk 		 * To avoid filling up a TXG with just frees wait for
886ff5177eeSAlek Pinchuk 		 * the next TXG to open before freeing more chunks if
887ff5177eeSAlek Pinchuk 		 * we have reached the threshold of frees
888ff5177eeSAlek Pinchuk 		 */
889ff5177eeSAlek Pinchuk 		if (dirty_frees_threshold != 0 &&
890ff5177eeSAlek Pinchuk 		    long_free_dirty_all_txgs >= dirty_frees_threshold) {
891ff5177eeSAlek Pinchuk 			txg_wait_open(dp, 0);
892ff5177eeSAlek Pinchuk 			continue;
893ff5177eeSAlek Pinchuk 		}
894ff5177eeSAlek Pinchuk 
895ff5177eeSAlek Pinchuk 		tx = dmu_tx_create(os);
896ff5177eeSAlek Pinchuk 		dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len);
8974bb73804SMatthew Ahrens 
8984bb73804SMatthew Ahrens 		/*
8994bb73804SMatthew Ahrens 		 * Mark this transaction as typically resulting in a net
9004bb73804SMatthew Ahrens 		 * reduction in space used.
9014bb73804SMatthew Ahrens 		 */
9024bb73804SMatthew Ahrens 		dmu_tx_mark_netfree(tx);
903cdb0ab79Smaybee 		err = dmu_tx_assign(tx, TXG_WAIT);
904cdb0ab79Smaybee 		if (err) {
905cdb0ab79Smaybee 			dmu_tx_abort(tx);
906cdb0ab79Smaybee 			return (err);
907cdb0ab79Smaybee 		}
908ff5177eeSAlek Pinchuk 
909ff5177eeSAlek Pinchuk 		mutex_enter(&dp->dp_lock);
910ff5177eeSAlek Pinchuk 		dp->dp_long_free_dirty_pertxg[dmu_tx_get_txg(tx) & TXG_MASK] +=
911ff5177eeSAlek Pinchuk 		    chunk_len;
912ff5177eeSAlek Pinchuk 		mutex_exit(&dp->dp_lock);
913ff5177eeSAlek Pinchuk 		DTRACE_PROBE3(free__long__range,
914ff5177eeSAlek Pinchuk 		    uint64_t, long_free_dirty_all_txgs, uint64_t, chunk_len,
915ff5177eeSAlek Pinchuk 		    uint64_t, dmu_tx_get_txg(tx));
916ff5177eeSAlek Pinchuk 		dnode_free_range(dn, chunk_begin, chunk_len, tx);
917eb633035STom Caputi 
918cdb0ab79Smaybee 		dmu_tx_commit(tx);
919713d6c20SMatthew Ahrens 
920ff5177eeSAlek Pinchuk 		length -= chunk_len;
921cdb0ab79Smaybee 	}
922cdb0ab79Smaybee 	return (0);
923cdb0ab79Smaybee }
924cdb0ab79Smaybee 
925cdb0ab79Smaybee int
926cdb0ab79Smaybee dmu_free_long_range(objset_t *os, uint64_t object,
927cdb0ab79Smaybee     uint64_t offset, uint64_t length)
928cdb0ab79Smaybee {
929cdb0ab79Smaybee 	dnode_t *dn;
930cdb0ab79Smaybee 	int err;
931cdb0ab79Smaybee 
932503ad85cSMatthew Ahrens 	err = dnode_hold(os, object, FTAG, &dn);
933cdb0ab79Smaybee 	if (err != 0)
934cdb0ab79Smaybee 		return (err);
935713d6c20SMatthew Ahrens 	err = dmu_free_long_range_impl(os, dn, offset, length);
9365253393bSMatthew Ahrens 
9375253393bSMatthew Ahrens 	/*
9385253393bSMatthew Ahrens 	 * It is important to zero out the maxblkid when freeing the entire
9395253393bSMatthew Ahrens 	 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
9405253393bSMatthew Ahrens 	 * will take the fast path, and (b) dnode_reallocate() can verify
9415253393bSMatthew Ahrens 	 * that the entire file has been freed.
9425253393bSMatthew Ahrens 	 */
94343466aaeSMax Grossman 	if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
9445253393bSMatthew Ahrens 		dn->dn_maxblkid = 0;
9455253393bSMatthew Ahrens 
946cdb0ab79Smaybee 	dnode_rele(dn, FTAG);
947cdb0ab79Smaybee 	return (err);
948cdb0ab79Smaybee }
949cdb0ab79Smaybee 
950cdb0ab79Smaybee int
951713d6c20SMatthew Ahrens dmu_free_long_object(objset_t *os, uint64_t object)
952cdb0ab79Smaybee {
953cdb0ab79Smaybee 	dmu_tx_t *tx;
954cdb0ab79Smaybee 	int err;
955cdb0ab79Smaybee 
956713d6c20SMatthew Ahrens 	err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
957cdb0ab79Smaybee 	if (err != 0)
958cdb0ab79Smaybee 		return (err);
959713d6c20SMatthew Ahrens 
960713d6c20SMatthew Ahrens 	tx = dmu_tx_create(os);
961713d6c20SMatthew Ahrens 	dmu_tx_hold_bonus(tx, object);
962713d6c20SMatthew Ahrens 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
9634bb73804SMatthew Ahrens 	dmu_tx_mark_netfree(tx);
964713d6c20SMatthew Ahrens 	err = dmu_tx_assign(tx, TXG_WAIT);
965713d6c20SMatthew Ahrens 	if (err == 0) {
966eb633035STom Caputi 		if (err == 0)
967eb633035STom Caputi 			err = dmu_object_free(os, object, tx);
968eb633035STom Caputi 
969713d6c20SMatthew Ahrens 		dmu_tx_commit(tx);
970cdb0ab79Smaybee 	} else {
971713d6c20SMatthew Ahrens 		dmu_tx_abort(tx);
972cdb0ab79Smaybee 	}
973713d6c20SMatthew Ahrens 
974cdb0ab79Smaybee 	return (err);
975cdb0ab79Smaybee }
976cdb0ab79Smaybee 
977ea8dc4b6Seschrock int
978fa9e4066Sahrens dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
979fa9e4066Sahrens     uint64_t size, dmu_tx_t *tx)
980fa9e4066Sahrens {
981ea8dc4b6Seschrock 	dnode_t *dn;
982503ad85cSMatthew Ahrens 	int err = dnode_hold(os, object, FTAG, &dn);
983ea8dc4b6Seschrock 	if (err)
984ea8dc4b6Seschrock 		return (err);
985fa9e4066Sahrens 	ASSERT(offset < UINT64_MAX);
986eb633035STom Caputi 	ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset);
987fa9e4066Sahrens 	dnode_free_range(dn, offset, size, tx);
988fa9e4066Sahrens 	dnode_rele(dn, FTAG);
989ea8dc4b6Seschrock 	return (0);
990fa9e4066Sahrens }
991fa9e4066Sahrens 
992b0c42cd4Sbzzz static int
993b0c42cd4Sbzzz dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size,
9947bfdf011SNeil Perrin     void *buf, uint32_t flags)
995fa9e4066Sahrens {
996fa9e4066Sahrens 	dmu_buf_t **dbp;
997b0c42cd4Sbzzz 	int numbufs, err = 0;
998feb08c6bSbillm 
999feb08c6bSbillm 	/*
1000feb08c6bSbillm 	 * Deal with odd block sizes, where there can't be data past the first
1001feb08c6bSbillm 	 * block.  If we ever do the tail block optimization, we will need to
1002feb08c6bSbillm 	 * handle that here as well.
1003feb08c6bSbillm 	 */
1004c87b8fc5SMark J Musante 	if (dn->dn_maxblkid == 0) {
1005fa9e4066Sahrens 		int newsz = offset > dn->dn_datablksz ? 0 :
1006fa9e4066Sahrens 		    MIN(size, dn->dn_datablksz - offset);
1007fa9e4066Sahrens 		bzero((char *)buf + newsz, size - newsz);
1008fa9e4066Sahrens 		size = newsz;
1009fa9e4066Sahrens 	}
1010fa9e4066Sahrens 
1011fa9e4066Sahrens 	while (size > 0) {
1012fa9e4066Sahrens 		uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
1013c87b8fc5SMark J Musante 		int i;
1014fa9e4066Sahrens 
1015fa9e4066Sahrens 		/*
1016fa9e4066Sahrens 		 * NB: we could do this block-at-a-time, but it's nice
1017fa9e4066Sahrens 		 * to be reading in parallel.
1018fa9e4066Sahrens 		 */
1019a2eea2e1Sahrens 		err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
10207bfdf011SNeil Perrin 		    TRUE, FTAG, &numbufs, &dbp, flags);
1021ea8dc4b6Seschrock 		if (err)
10221934e92fSmaybee 			break;
1023fa9e4066Sahrens 
1024fa9e4066Sahrens 		for (i = 0; i < numbufs; i++) {
1025fa9e4066Sahrens 			int tocpy;
1026fa9e4066Sahrens 			int bufoff;
1027fa9e4066Sahrens 			dmu_buf_t *db = dbp[i];
1028fa9e4066Sahrens 
1029fa9e4066Sahrens 			ASSERT(size > 0);
1030fa9e4066Sahrens 
1031fa9e4066Sahrens 			bufoff = offset - db->db_offset;
1032fa9e4066Sahrens 			tocpy = (int)MIN(db->db_size - bufoff, size);
1033fa9e4066Sahrens 
1034fa9e4066Sahrens 			bcopy((char *)db->db_data + bufoff, buf, tocpy);
1035fa9e4066Sahrens 
1036fa9e4066Sahrens 			offset += tocpy;
1037fa9e4066Sahrens 			size -= tocpy;
1038fa9e4066Sahrens 			buf = (char *)buf + tocpy;
1039fa9e4066Sahrens 		}
1040ea8dc4b6Seschrock 		dmu_buf_rele_array(dbp, numbufs, FTAG);
1041fa9e4066Sahrens 	}
10421934e92fSmaybee 	return (err);
1043fa9e4066Sahrens }
1044fa9e4066Sahrens 
1045b0c42cd4Sbzzz int
1046b0c42cd4Sbzzz dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1047b0c42cd4Sbzzz     void *buf, uint32_t flags)
1048fa9e4066Sahrens {
1049b0c42cd4Sbzzz 	dnode_t *dn;
1050b0c42cd4Sbzzz 	int err;
1051fa9e4066Sahrens 
1052b0c42cd4Sbzzz 	err = dnode_hold(os, object, FTAG, &dn);
1053b0c42cd4Sbzzz 	if (err != 0)
1054b0c42cd4Sbzzz 		return (err);
105513506d1eSmaybee 
1056b0c42cd4Sbzzz 	err = dmu_read_impl(dn, offset, size, buf, flags);
1057b0c42cd4Sbzzz 	dnode_rele(dn, FTAG);
1058b0c42cd4Sbzzz 	return (err);
1059b0c42cd4Sbzzz }
1060b0c42cd4Sbzzz 
1061b0c42cd4Sbzzz int
1062b0c42cd4Sbzzz dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
1063b0c42cd4Sbzzz     uint32_t flags)
1064b0c42cd4Sbzzz {
1065b0c42cd4Sbzzz 	return (dmu_read_impl(dn, offset, size, buf, flags));
1066b0c42cd4Sbzzz }
1067b0c42cd4Sbzzz 
1068b0c42cd4Sbzzz static void
1069b0c42cd4Sbzzz dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size,
1070b0c42cd4Sbzzz     const void *buf, dmu_tx_t *tx)
1071b0c42cd4Sbzzz {
1072b0c42cd4Sbzzz 	int i;
1073fa9e4066Sahrens 
1074fa9e4066Sahrens 	for (i = 0; i < numbufs; i++) {
1075fa9e4066Sahrens 		int tocpy;
1076fa9e4066Sahrens 		int bufoff;
1077fa9e4066Sahrens 		dmu_buf_t *db = dbp[i];
1078fa9e4066Sahrens 
1079fa9e4066Sahrens 		ASSERT(size > 0);
1080fa9e4066Sahrens 
1081fa9e4066Sahrens 		bufoff = offset - db->db_offset;
1082fa9e4066Sahrens 		tocpy = (int)MIN(db->db_size - bufoff, size);
1083fa9e4066Sahrens 
1084fa9e4066Sahrens 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1085fa9e4066Sahrens 
1086fa9e4066Sahrens 		if (tocpy == db->db_size)
1087fa9e4066Sahrens 			dmu_buf_will_fill(db, tx);
1088fa9e4066Sahrens 		else
1089fa9e4066Sahrens 			dmu_buf_will_dirty(db, tx);
1090fa9e4066Sahrens 
1091fa9e4066Sahrens 		bcopy(buf, (char *)db->db_data + bufoff, tocpy);
1092fa9e4066Sahrens 
1093fa9e4066Sahrens 		if (tocpy == db->db_size)
1094fa9e4066Sahrens 			dmu_buf_fill_done(db, tx);
1095fa9e4066Sahrens 
1096fa9e4066Sahrens 		offset += tocpy;
1097fa9e4066Sahrens 		size -= tocpy;
1098fa9e4066Sahrens 		buf = (char *)buf + tocpy;
1099fa9e4066Sahrens 	}
1100b0c42cd4Sbzzz }
1101b0c42cd4Sbzzz 
1102b0c42cd4Sbzzz void
1103b0c42cd4Sbzzz dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1104b0c42cd4Sbzzz     const void *buf, dmu_tx_t *tx)
1105b0c42cd4Sbzzz {
1106b0c42cd4Sbzzz 	dmu_buf_t **dbp;
1107b0c42cd4Sbzzz 	int numbufs;
1108b0c42cd4Sbzzz 
1109b0c42cd4Sbzzz 	if (size == 0)
1110b0c42cd4Sbzzz 		return;
1111b0c42cd4Sbzzz 
1112b0c42cd4Sbzzz 	VERIFY0(dmu_buf_hold_array(os, object, offset, size,
1113b0c42cd4Sbzzz 	    FALSE, FTAG, &numbufs, &dbp));
1114b0c42cd4Sbzzz 	dmu_write_impl(dbp, numbufs, offset, size, buf, tx);
1115b0c42cd4Sbzzz 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1116b0c42cd4Sbzzz }
1117b0c42cd4Sbzzz 
1118b0c42cd4Sbzzz void
1119b0c42cd4Sbzzz dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
1120b0c42cd4Sbzzz     const void *buf, dmu_tx_t *tx)
1121b0c42cd4Sbzzz {
1122b0c42cd4Sbzzz 	dmu_buf_t **dbp;
1123b0c42cd4Sbzzz 	int numbufs;
1124b0c42cd4Sbzzz 
1125b0c42cd4Sbzzz 	if (size == 0)
1126b0c42cd4Sbzzz 		return;
1127b0c42cd4Sbzzz 
1128b0c42cd4Sbzzz 	VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size,
1129b0c42cd4Sbzzz 	    FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH));
1130b0c42cd4Sbzzz 	dmu_write_impl(dbp, numbufs, offset, size, buf, tx);
1131ea8dc4b6Seschrock 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1132fa9e4066Sahrens }
1133fa9e4066Sahrens 
11345cabbc6bSPrashanth Sreenivasa static int
11355cabbc6bSPrashanth Sreenivasa dmu_object_remap_one_indirect(objset_t *os, dnode_t *dn,
11365cabbc6bSPrashanth Sreenivasa     uint64_t last_removal_txg, uint64_t offset)
11375cabbc6bSPrashanth Sreenivasa {
11385cabbc6bSPrashanth Sreenivasa 	uint64_t l1blkid = dbuf_whichblock(dn, 1, offset);
11395cabbc6bSPrashanth Sreenivasa 	int err = 0;
11405cabbc6bSPrashanth Sreenivasa 
11415cabbc6bSPrashanth Sreenivasa 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
11425cabbc6bSPrashanth Sreenivasa 	dmu_buf_impl_t *dbuf = dbuf_hold_level(dn, 1, l1blkid, FTAG);
11435cabbc6bSPrashanth Sreenivasa 	ASSERT3P(dbuf, !=, NULL);
11445cabbc6bSPrashanth Sreenivasa 
11455cabbc6bSPrashanth Sreenivasa 	/*
11465cabbc6bSPrashanth Sreenivasa 	 * If the block hasn't been written yet, this default will ensure
11475cabbc6bSPrashanth Sreenivasa 	 * we don't try to remap it.
11485cabbc6bSPrashanth Sreenivasa 	 */
11495cabbc6bSPrashanth Sreenivasa 	uint64_t birth = UINT64_MAX;
11505cabbc6bSPrashanth Sreenivasa 	ASSERT3U(last_removal_txg, !=, UINT64_MAX);
11515cabbc6bSPrashanth Sreenivasa 	if (dbuf->db_blkptr != NULL)
11525cabbc6bSPrashanth Sreenivasa 		birth = dbuf->db_blkptr->blk_birth;
11535cabbc6bSPrashanth Sreenivasa 	rw_exit(&dn->dn_struct_rwlock);
11545cabbc6bSPrashanth Sreenivasa 
11555cabbc6bSPrashanth Sreenivasa 	/*
11565cabbc6bSPrashanth Sreenivasa 	 * If this L1 was already written after the last removal, then we've
11575cabbc6bSPrashanth Sreenivasa 	 * already tried to remap it.
11585cabbc6bSPrashanth Sreenivasa 	 */
11595cabbc6bSPrashanth Sreenivasa 	if (birth <= last_removal_txg &&
11605cabbc6bSPrashanth Sreenivasa 	    dbuf_read(dbuf, NULL, DB_RF_MUST_SUCCEED) == 0 &&
11615cabbc6bSPrashanth Sreenivasa 	    dbuf_can_remap(dbuf)) {
11625cabbc6bSPrashanth Sreenivasa 		dmu_tx_t *tx = dmu_tx_create(os);
11635cabbc6bSPrashanth Sreenivasa 		dmu_tx_hold_remap_l1indirect(tx, dn->dn_object);
11645cabbc6bSPrashanth Sreenivasa 		err = dmu_tx_assign(tx, TXG_WAIT);
11655cabbc6bSPrashanth Sreenivasa 		if (err == 0) {
11665cabbc6bSPrashanth Sreenivasa 			(void) dbuf_dirty(dbuf, tx);
11675cabbc6bSPrashanth Sreenivasa 			dmu_tx_commit(tx);
11685cabbc6bSPrashanth Sreenivasa 		} else {
11695cabbc6bSPrashanth Sreenivasa 			dmu_tx_abort(tx);
11705cabbc6bSPrashanth Sreenivasa 		}
11715cabbc6bSPrashanth Sreenivasa 	}
11725cabbc6bSPrashanth Sreenivasa 
11735cabbc6bSPrashanth Sreenivasa 	dbuf_rele(dbuf, FTAG);
11745cabbc6bSPrashanth Sreenivasa 
11755cabbc6bSPrashanth Sreenivasa 	delay(zfs_object_remap_one_indirect_delay_ticks);
11765cabbc6bSPrashanth Sreenivasa 
11775cabbc6bSPrashanth Sreenivasa 	return (err);
11785cabbc6bSPrashanth Sreenivasa }
11795cabbc6bSPrashanth Sreenivasa 
11805cabbc6bSPrashanth Sreenivasa /*
11815cabbc6bSPrashanth Sreenivasa  * Remap all blockpointers in the object, if possible, so that they reference
11825cabbc6bSPrashanth Sreenivasa  * only concrete vdevs.
11835cabbc6bSPrashanth Sreenivasa  *
11845cabbc6bSPrashanth Sreenivasa  * To do this, iterate over the L0 blockpointers and remap any that reference
11855cabbc6bSPrashanth Sreenivasa  * an indirect vdev. Note that we only examine L0 blockpointers; since we
11865cabbc6bSPrashanth Sreenivasa  * cannot guarantee that we can remap all blockpointer anyways (due to split
11875cabbc6bSPrashanth Sreenivasa  * blocks), we do not want to make the code unnecessarily complicated to
11885cabbc6bSPrashanth Sreenivasa  * catch the unlikely case that there is an L1 block on an indirect vdev that
11895cabbc6bSPrashanth Sreenivasa  * contains no indirect blockpointers.
11905cabbc6bSPrashanth Sreenivasa  */
11915cabbc6bSPrashanth Sreenivasa int
11925cabbc6bSPrashanth Sreenivasa dmu_object_remap_indirects(objset_t *os, uint64_t object,
11935cabbc6bSPrashanth Sreenivasa     uint64_t last_removal_txg)
11945cabbc6bSPrashanth Sreenivasa {
11955cabbc6bSPrashanth Sreenivasa 	uint64_t offset, l1span;
11965cabbc6bSPrashanth Sreenivasa 	int err;
11975cabbc6bSPrashanth Sreenivasa 	dnode_t *dn;
11985cabbc6bSPrashanth Sreenivasa 
11995cabbc6bSPrashanth Sreenivasa 	err = dnode_hold(os, object, FTAG, &dn);
12005cabbc6bSPrashanth Sreenivasa 	if (err != 0) {
12015cabbc6bSPrashanth Sreenivasa 		return (err);
12025cabbc6bSPrashanth Sreenivasa 	}
12035cabbc6bSPrashanth Sreenivasa 
12045cabbc6bSPrashanth Sreenivasa 	if (dn->dn_nlevels <= 1) {
12055cabbc6bSPrashanth Sreenivasa 		if (issig(JUSTLOOKING) && issig(FORREAL)) {
12065cabbc6bSPrashanth Sreenivasa 			err = SET_ERROR(EINTR);
12075cabbc6bSPrashanth Sreenivasa 		}
12085cabbc6bSPrashanth Sreenivasa 
12095cabbc6bSPrashanth Sreenivasa 		/*
12105cabbc6bSPrashanth Sreenivasa 		 * If the dnode has no indirect blocks, we cannot dirty them.
12115cabbc6bSPrashanth Sreenivasa 		 * We still want to remap the blkptr(s) in the dnode if
12125cabbc6bSPrashanth Sreenivasa 		 * appropriate, so mark it as dirty.
12135cabbc6bSPrashanth Sreenivasa 		 */
12145cabbc6bSPrashanth Sreenivasa 		if (err == 0 && dnode_needs_remap(dn)) {
12155cabbc6bSPrashanth Sreenivasa 			dmu_tx_t *tx = dmu_tx_create(os);
12165cabbc6bSPrashanth Sreenivasa 			dmu_tx_hold_bonus(tx, dn->dn_object);
12175cabbc6bSPrashanth Sreenivasa 			if ((err = dmu_tx_assign(tx, TXG_WAIT)) == 0) {
12185cabbc6bSPrashanth Sreenivasa 				dnode_setdirty(dn, tx);
12195cabbc6bSPrashanth Sreenivasa 				dmu_tx_commit(tx);
12205cabbc6bSPrashanth Sreenivasa 			} else {
12215cabbc6bSPrashanth Sreenivasa 				dmu_tx_abort(tx);
12225cabbc6bSPrashanth Sreenivasa 			}
12235cabbc6bSPrashanth Sreenivasa 		}
12245cabbc6bSPrashanth Sreenivasa 
12255cabbc6bSPrashanth Sreenivasa 		dnode_rele(dn, FTAG);
12265cabbc6bSPrashanth Sreenivasa 		return (err);
12275cabbc6bSPrashanth Sreenivasa 	}
12285cabbc6bSPrashanth Sreenivasa 
12295cabbc6bSPrashanth Sreenivasa 	offset = 0;
12305cabbc6bSPrashanth Sreenivasa 	l1span = 1ULL << (dn->dn_indblkshift - SPA_BLKPTRSHIFT +
12315cabbc6bSPrashanth Sreenivasa 	    dn->dn_datablkshift);
12325cabbc6bSPrashanth Sreenivasa 	/*
12335cabbc6bSPrashanth Sreenivasa 	 * Find the next L1 indirect that is not a hole.
12345cabbc6bSPrashanth Sreenivasa 	 */
12355cabbc6bSPrashanth Sreenivasa 	while (dnode_next_offset(dn, 0, &offset, 2, 1, 0) == 0) {
12365cabbc6bSPrashanth Sreenivasa 		if (issig(JUSTLOOKING) && issig(FORREAL)) {
12375cabbc6bSPrashanth Sreenivasa 			err = SET_ERROR(EINTR);
12385cabbc6bSPrashanth Sreenivasa 			break;
12395cabbc6bSPrashanth Sreenivasa 		}
12405cabbc6bSPrashanth Sreenivasa 		if ((err = dmu_object_remap_one_indirect(os, dn,
12415cabbc6bSPrashanth Sreenivasa 		    last_removal_txg, offset)) != 0) {
12425cabbc6bSPrashanth Sreenivasa 			break;
12435cabbc6bSPrashanth Sreenivasa 		}
12445cabbc6bSPrashanth Sreenivasa 		offset += l1span;
12455cabbc6bSPrashanth Sreenivasa 	}
12465cabbc6bSPrashanth Sreenivasa 
12475cabbc6bSPrashanth Sreenivasa 	dnode_rele(dn, FTAG);
12485cabbc6bSPrashanth Sreenivasa 	return (err);
12495cabbc6bSPrashanth Sreenivasa }
12505cabbc6bSPrashanth Sreenivasa 
125182c9918fSTim Haley void
125282c9918fSTim Haley dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
125382c9918fSTim Haley     dmu_tx_t *tx)
125482c9918fSTim Haley {
125582c9918fSTim Haley 	dmu_buf_t **dbp;
125682c9918fSTim Haley 	int numbufs, i;
125782c9918fSTim Haley 
125882c9918fSTim Haley 	if (size == 0)
125982c9918fSTim Haley 		return;
126082c9918fSTim Haley 
126182c9918fSTim Haley 	VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
126282c9918fSTim Haley 	    FALSE, FTAG, &numbufs, &dbp));
126382c9918fSTim Haley 
126482c9918fSTim Haley 	for (i = 0; i < numbufs; i++) {
126582c9918fSTim Haley 		dmu_buf_t *db = dbp[i];
126682c9918fSTim Haley 
126782c9918fSTim Haley 		dmu_buf_will_not_fill(db, tx);
126882c9918fSTim Haley 	}
126982c9918fSTim Haley 	dmu_buf_rele_array(dbp, numbufs, FTAG);
127082c9918fSTim Haley }
127182c9918fSTim Haley 
12725d7b4d43SMatthew Ahrens void
12735d7b4d43SMatthew Ahrens dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
12745d7b4d43SMatthew Ahrens     void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
12755d7b4d43SMatthew Ahrens     int compressed_size, int byteorder, dmu_tx_t *tx)
12765d7b4d43SMatthew Ahrens {
12775d7b4d43SMatthew Ahrens 	dmu_buf_t *db;
12785d7b4d43SMatthew Ahrens 
12795d7b4d43SMatthew Ahrens 	ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
12805d7b4d43SMatthew Ahrens 	ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
12815d7b4d43SMatthew Ahrens 	VERIFY0(dmu_buf_hold_noread(os, object, offset,
12825d7b4d43SMatthew Ahrens 	    FTAG, &db));
12835d7b4d43SMatthew Ahrens 
12845d7b4d43SMatthew Ahrens 	dmu_buf_write_embedded(db,
12855d7b4d43SMatthew Ahrens 	    data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
12865d7b4d43SMatthew Ahrens 	    uncompressed_size, compressed_size, byteorder, tx);
12875d7b4d43SMatthew Ahrens 
12885d7b4d43SMatthew Ahrens 	dmu_buf_rele(db, FTAG);
12895d7b4d43SMatthew Ahrens }
12905d7b4d43SMatthew Ahrens 
1291c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /*
1292c242f9a0Schunli zhang - Sun Microsystems - Irvine United States  * DMU support for xuio
1293c242f9a0Schunli zhang - Sun Microsystems - Irvine United States  */
1294c242f9a0Schunli zhang - Sun Microsystems - Irvine United States kstat_t *xuio_ksp = NULL;
1295c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1296c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int
1297c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_init(xuio_t *xuio, int nblk)
1298c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1299c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	dmu_xuio_t *priv;
1300c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	uio_t *uio = &xuio->xu_uio;
1301c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1302c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	uio->uio_iovcnt = nblk;
1303c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
1304c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1305c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
1306c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	priv->cnt = nblk;
1307c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
1308c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	priv->iovp = uio->uio_iov;
1309c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	XUIO_XUZC_PRIV(xuio) = priv;
1310c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1311c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	if (XUIO_XUZC_RW(xuio) == UIO_READ)
1312c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
1313c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	else
1314c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
1315c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1316c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	return (0);
1317c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1318c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1319c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void
1320c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_fini(xuio_t *xuio)
1321c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1322c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1323c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	int nblk = priv->cnt;
1324c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1325c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	kmem_free(priv->iovp, nblk * sizeof (iovec_t));
1326c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
1327c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	kmem_free(priv, sizeof (dmu_xuio_t));
1328c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1329c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	if (XUIO_XUZC_RW(xuio) == UIO_READ)
1330c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
1331c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	else
1332c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
1333c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1334c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1335c242f9a0Schunli zhang - Sun Microsystems - Irvine United States /*
1336c242f9a0Schunli zhang - Sun Microsystems - Irvine United States  * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
1337c242f9a0Schunli zhang - Sun Microsystems - Irvine United States  * and increase priv->next by 1.
1338c242f9a0Schunli zhang - Sun Microsystems - Irvine United States  */
1339c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int
1340c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
1341c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1342c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	struct iovec *iov;
1343c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	uio_t *uio = &xuio->xu_uio;
1344c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1345c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	int i = priv->next++;
1346c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1347c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	ASSERT(i < priv->cnt);
13485602294fSDan Kimmel 	ASSERT(off + n <= arc_buf_lsize(abuf));
1349c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	iov = uio->uio_iov + i;
1350c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	iov->iov_base = (char *)abuf->b_data + off;
1351c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	iov->iov_len = n;
1352c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	priv->bufs[i] = abuf;
1353c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	return (0);
1354c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1355c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1356c242f9a0Schunli zhang - Sun Microsystems - Irvine United States int
1357c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_cnt(xuio_t *xuio)
1358c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1359c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1360c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	return (priv->cnt);
1361c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1362c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1363c242f9a0Schunli zhang - Sun Microsystems - Irvine United States arc_buf_t *
1364c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_arcbuf(xuio_t *xuio, int i)
1365c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1366c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1367c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1368c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	ASSERT(i < priv->cnt);
1369c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	return (priv->bufs[i]);
1370c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1371c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1372c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void
1373c242f9a0Schunli zhang - Sun Microsystems - Irvine United States dmu_xuio_clear(xuio_t *xuio, int i)
1374c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1375c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1376c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1377c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	ASSERT(i < priv->cnt);
1378c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	priv->bufs[i] = NULL;
1379c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1380c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1381c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void
1382c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_init(void)
1383c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1384c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
1385c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	    KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
1386c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	    KSTAT_FLAG_VIRTUAL);
1387c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	if (xuio_ksp != NULL) {
1388c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		xuio_ksp->ks_data = &xuio_stats;
1389c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		kstat_install(xuio_ksp);
1390c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	}
1391c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1392c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1393c242f9a0Schunli zhang - Sun Microsystems - Irvine United States static void
1394c242f9a0Schunli zhang - Sun Microsystems - Irvine United States xuio_stat_fini(void)
1395c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1396c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	if (xuio_ksp != NULL) {
1397c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		kstat_delete(xuio_ksp);
1398c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		xuio_ksp = NULL;
1399c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	}
1400c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1401c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1402c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void
140399aa8b55SPrashanth Sreenivasa xuio_stat_wbuf_copied(void)
1404c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1405c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1406c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1407c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1408c242f9a0Schunli zhang - Sun Microsystems - Irvine United States void
140999aa8b55SPrashanth Sreenivasa xuio_stat_wbuf_nocopy(void)
1410c242f9a0Schunli zhang - Sun Microsystems - Irvine United States {
1411c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1412c242f9a0Schunli zhang - Sun Microsystems - Irvine United States }
1413c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1414fa9e4066Sahrens #ifdef _KERNEL
14158dfe5547SRichard Yao int
1416f8554bb9SMatthew Ahrens dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
1417feb08c6bSbillm {
1418feb08c6bSbillm 	dmu_buf_t **dbp;
1419feb08c6bSbillm 	int numbufs, i, err;
1420c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	xuio_t *xuio = NULL;
1421feb08c6bSbillm 
1422feb08c6bSbillm 	/*
1423feb08c6bSbillm 	 * NB: we could do this block-at-a-time, but it's nice
1424feb08c6bSbillm 	 * to be reading in parallel.
1425feb08c6bSbillm 	 */
1426f8554bb9SMatthew Ahrens 	err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1427f8554bb9SMatthew Ahrens 	    TRUE, FTAG, &numbufs, &dbp, 0);
1428feb08c6bSbillm 	if (err)
1429feb08c6bSbillm 		return (err);
1430feb08c6bSbillm 
1431c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	if (uio->uio_extflg == UIO_XUIO)
1432c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		xuio = (xuio_t *)uio;
1433c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1434feb08c6bSbillm 	for (i = 0; i < numbufs; i++) {
1435feb08c6bSbillm 		int tocpy;
1436feb08c6bSbillm 		int bufoff;
1437feb08c6bSbillm 		dmu_buf_t *db = dbp[i];
1438feb08c6bSbillm 
1439feb08c6bSbillm 		ASSERT(size > 0);
1440feb08c6bSbillm 
1441feb08c6bSbillm 		bufoff = uio->uio_loffset - db->db_offset;
1442feb08c6bSbillm 		tocpy = (int)MIN(db->db_size - bufoff, size);
1443feb08c6bSbillm 
1444c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		if (xuio) {
1445c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1446c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			arc_buf_t *dbuf_abuf = dbi->db_buf;
1447c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1448c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1449c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			if (!err) {
1450c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 				uio->uio_resid -= tocpy;
1451c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 				uio->uio_loffset += tocpy;
1452c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			}
1453c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 
1454c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			if (abuf == dbuf_abuf)
1455c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 				XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1456c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			else
1457c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 				XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1458c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		} else {
1459c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			err = uiomove((char *)db->db_data + bufoff, tocpy,
1460c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 			    UIO_READ, uio);
1461c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		}
1462feb08c6bSbillm 		if (err)
1463feb08c6bSbillm 			break;
1464feb08c6bSbillm 
1465feb08c6bSbillm 		size -= tocpy;
1466feb08c6bSbillm 	}
1467feb08c6bSbillm 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1468feb08c6bSbillm 
1469feb08c6bSbillm 	return (err);
1470feb08c6bSbillm }
1471feb08c6bSbillm 
1472f8554bb9SMatthew Ahrens /*
1473f8554bb9SMatthew Ahrens  * Read 'size' bytes into the uio buffer.
1474f8554bb9SMatthew Ahrens  * From object zdb->db_object.
1475f8554bb9SMatthew Ahrens  * Starting at offset uio->uio_loffset.
1476f8554bb9SMatthew Ahrens  *
1477f8554bb9SMatthew Ahrens  * If the caller already has a dbuf in the target object
1478f8554bb9SMatthew Ahrens  * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1479f8554bb9SMatthew Ahrens  * because we don't have to find the dnode_t for the object.
1480f8554bb9SMatthew Ahrens  */
1481f8554bb9SMatthew Ahrens int
1482f8554bb9SMatthew Ahrens dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size)
1483f8554bb9SMatthew Ahrens {
1484f8554bb9SMatthew Ahrens 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1485f8554bb9SMatthew Ahrens 	dnode_t *dn;
1486f8554bb9SMatthew Ahrens 	int err;
1487f8554bb9SMatthew Ahrens 
1488f8554bb9SMatthew Ahrens 	if (size == 0)
1489f8554bb9SMatthew Ahrens 		return (0);
1490f8554bb9SMatthew Ahrens 
1491f8554bb9SMatthew Ahrens 	DB_DNODE_ENTER(db);
1492f8554bb9SMatthew Ahrens 	dn = DB_DNODE(db);
1493f8554bb9SMatthew Ahrens 	err = dmu_read_uio_dnode(dn, uio, size);
1494f8554bb9SMatthew Ahrens 	DB_DNODE_EXIT(db);
1495f8554bb9SMatthew Ahrens 
1496f8554bb9SMatthew Ahrens 	return (err);
1497f8554bb9SMatthew Ahrens }
1498f8554bb9SMatthew Ahrens 
1499f8554bb9SMatthew Ahrens /*
1500f8554bb9SMatthew Ahrens  * Read 'size' bytes into the uio buffer.
1501f8554bb9SMatthew Ahrens  * From the specified object
1502f8554bb9SMatthew Ahrens  * Starting at offset uio->uio_loffset.
1503f8554bb9SMatthew Ahrens  */
1504f8554bb9SMatthew Ahrens int
1505f8554bb9SMatthew Ahrens dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1506f8554bb9SMatthew Ahrens {
1507f8554bb9SMatthew Ahrens 	dnode_t *dn;
1508f8554bb9SMatthew Ahrens 	int err;
1509f8554bb9SMatthew Ahrens 
1510f8554bb9SMatthew Ahrens 	if (size == 0)
1511f8554bb9SMatthew Ahrens 		return (0);
1512f8554bb9SMatthew Ahrens 
1513f8554bb9SMatthew Ahrens 	err = dnode_hold(os, object, FTAG, &dn);
1514f8554bb9SMatthew Ahrens 	if (err)
1515f8554bb9SMatthew Ahrens 		return (err);
1516f8554bb9SMatthew Ahrens 
1517f8554bb9SMatthew Ahrens 	err = dmu_read_uio_dnode(dn, uio, size);
1518f8554bb9SMatthew Ahrens 
1519f8554bb9SMatthew Ahrens 	dnode_rele(dn, FTAG);
1520f8554bb9SMatthew Ahrens 
1521f8554bb9SMatthew Ahrens 	return (err);
1522f8554bb9SMatthew Ahrens }
1523f8554bb9SMatthew Ahrens 
15248dfe5547SRichard Yao int
152594d1a210STim Haley dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1526fa9e4066Sahrens {
1527fa9e4066Sahrens 	dmu_buf_t **dbp;
152894d1a210STim Haley 	int numbufs;
1529fa9e4066Sahrens 	int err = 0;
153094d1a210STim Haley 	int i;
1531fa9e4066Sahrens 
153294d1a210STim Haley 	err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
153394d1a210STim Haley 	    FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1534ea8dc4b6Seschrock 	if (err)
1535ea8dc4b6Seschrock 		return (err);
1536fa9e4066Sahrens 
1537fa9e4066Sahrens 	for (i = 0; i < numbufs; i++) {
1538fa9e4066Sahrens 		int tocpy;
1539fa9e4066Sahrens 		int bufoff;
1540fa9e4066Sahrens 		dmu_buf_t *db = dbp[i];
1541fa9e4066Sahrens 
1542fa9e4066Sahrens 		ASSERT(size > 0);
1543fa9e4066Sahrens 
1544feb08c6bSbillm 		bufoff = uio->uio_loffset - db->db_offset;
1545fa9e4066Sahrens 		tocpy = (int)MIN(db->db_size - bufoff, size);
1546fa9e4066Sahrens 
1547fa9e4066Sahrens 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1548fa9e4066Sahrens 
1549fa9e4066Sahrens 		if (tocpy == db->db_size)
1550fa9e4066Sahrens 			dmu_buf_will_fill(db, tx);
1551fa9e4066Sahrens 		else
1552fa9e4066Sahrens 			dmu_buf_will_dirty(db, tx);
1553fa9e4066Sahrens 
1554fa9e4066Sahrens 		/*
1555fa9e4066Sahrens 		 * XXX uiomove could block forever (eg. nfs-backed
1556fa9e4066Sahrens 		 * pages).  There needs to be a uiolockdown() function
1557fa9e4066Sahrens 		 * to lock the pages in memory, so that uiomove won't
1558fa9e4066Sahrens 		 * block.
1559fa9e4066Sahrens 		 */
1560fa9e4066Sahrens 		err = uiomove((char *)db->db_data + bufoff, tocpy,
1561fa9e4066Sahrens 		    UIO_WRITE, uio);
1562fa9e4066Sahrens 
1563fa9e4066Sahrens 		if (tocpy == db->db_size)
1564fa9e4066Sahrens 			dmu_buf_fill_done(db, tx);
1565fa9e4066Sahrens 
1566fa9e4066Sahrens 		if (err)
1567fa9e4066Sahrens 			break;
1568fa9e4066Sahrens 
1569fa9e4066Sahrens 		size -= tocpy;
1570fa9e4066Sahrens 	}
157194d1a210STim Haley 
1572ea8dc4b6Seschrock 	dmu_buf_rele_array(dbp, numbufs, FTAG);
1573fa9e4066Sahrens 	return (err);
1574fa9e4066Sahrens }
157544eda4d7Smaybee 
1576f8554bb9SMatthew Ahrens /*
1577f8554bb9SMatthew Ahrens  * Write 'size' bytes from the uio buffer.
1578f8554bb9SMatthew Ahrens  * To object zdb->db_object.
1579f8554bb9SMatthew Ahrens  * Starting at offset uio->uio_loffset.
1580f8554bb9SMatthew Ahrens  *
1581f8554bb9SMatthew Ahrens  * If the caller already has a dbuf in the target object
1582f8554bb9SMatthew Ahrens  * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1583f8554bb9SMatthew Ahrens  * because we don't have to find the dnode_t for the object.
1584f8554bb9SMatthew Ahrens  */
158594d1a210STim Haley int
158694d1a210STim Haley dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
158794d1a210STim Haley     dmu_tx_t *tx)
158894d1a210STim Haley {
1589744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1590744947dcSTom Erickson 	dnode_t *dn;
1591744947dcSTom Erickson 	int err;
1592744947dcSTom Erickson 
159394d1a210STim Haley 	if (size == 0)
159494d1a210STim Haley 		return (0);
159594d1a210STim Haley 
1596744947dcSTom Erickson 	DB_DNODE_ENTER(db);
1597744947dcSTom Erickson 	dn = DB_DNODE(db);
1598744947dcSTom Erickson 	err = dmu_write_uio_dnode(dn, uio, size, tx);
1599744947dcSTom Erickson 	DB_DNODE_EXIT(db);
1600744947dcSTom Erickson 
1601744947dcSTom Erickson 	return (err);
160294d1a210STim Haley }
160394d1a210STim Haley 
1604f8554bb9SMatthew Ahrens /*
1605f8554bb9SMatthew Ahrens  * Write 'size' bytes from the uio buffer.
1606f8554bb9SMatthew Ahrens  * To the specified object.
1607f8554bb9SMatthew Ahrens  * Starting at offset uio->uio_loffset.
1608f8554bb9SMatthew Ahrens  */
160994d1a210STim Haley int
161094d1a210STim Haley dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
161194d1a210STim Haley     dmu_tx_t *tx)
161294d1a210STim Haley {
161394d1a210STim Haley 	dnode_t *dn;
161494d1a210STim Haley 	int err;
161594d1a210STim Haley 
161694d1a210STim Haley 	if (size == 0)
161794d1a210STim Haley 		return (0);
161894d1a210STim Haley 
161994d1a210STim Haley 	err = dnode_hold(os, object, FTAG, &dn);
162094d1a210STim Haley 	if (err)
162194d1a210STim Haley 		return (err);
162294d1a210STim Haley 
162394d1a210STim Haley 	err = dmu_write_uio_dnode(dn, uio, size, tx);
162494d1a210STim Haley 
162594d1a210STim Haley 	dnode_rele(dn, FTAG);
162694d1a210STim Haley 
162794d1a210STim Haley 	return (err);
162894d1a210STim Haley }
162994d1a210STim Haley 
163044eda4d7Smaybee int
163144eda4d7Smaybee dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
163244eda4d7Smaybee     page_t *pp, dmu_tx_t *tx)
163344eda4d7Smaybee {
163444eda4d7Smaybee 	dmu_buf_t **dbp;
163544eda4d7Smaybee 	int numbufs, i;
163644eda4d7Smaybee 	int err;
163744eda4d7Smaybee 
163844eda4d7Smaybee 	if (size == 0)
163944eda4d7Smaybee 		return (0);
164044eda4d7Smaybee 
164144eda4d7Smaybee 	err = dmu_buf_hold_array(os, object, offset, size,
164244eda4d7Smaybee 	    FALSE, FTAG, &numbufs, &dbp);
164344eda4d7Smaybee 	if (err)
164444eda4d7Smaybee 		return (err);
164544eda4d7Smaybee 
164644eda4d7Smaybee 	for (i = 0; i < numbufs; i++) {
164744eda4d7Smaybee 		int tocpy, copied, thiscpy;
164844eda4d7Smaybee 		int bufoff;
164944eda4d7Smaybee 		dmu_buf_t *db = dbp[i];
165044eda4d7Smaybee 		caddr_t va;
165144eda4d7Smaybee 
165244eda4d7Smaybee 		ASSERT(size > 0);
165344eda4d7Smaybee 		ASSERT3U(db->db_size, >=, PAGESIZE);
165444eda4d7Smaybee 
165544eda4d7Smaybee 		bufoff = offset - db->db_offset;
165644eda4d7Smaybee 		tocpy = (int)MIN(db->db_size - bufoff, size);
165744eda4d7Smaybee 
165844eda4d7Smaybee 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
165944eda4d7Smaybee 
166044eda4d7Smaybee 		if (tocpy == db->db_size)
166144eda4d7Smaybee 			dmu_buf_will_fill(db, tx);
166244eda4d7Smaybee 		else
166344eda4d7Smaybee 			dmu_buf_will_dirty(db, tx);
166444eda4d7Smaybee 
166544eda4d7Smaybee 		for (copied = 0; copied < tocpy; copied += PAGESIZE) {
166644eda4d7Smaybee 			ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
166744eda4d7Smaybee 			thiscpy = MIN(PAGESIZE, tocpy - copied);
16680fab61baSJonathan W Adams 			va = zfs_map_page(pp, S_READ);
166944eda4d7Smaybee 			bcopy(va, (char *)db->db_data + bufoff, thiscpy);
16700fab61baSJonathan W Adams 			zfs_unmap_page(pp, va);
167144eda4d7Smaybee 			pp = pp->p_next;
167244eda4d7Smaybee 			bufoff += PAGESIZE;
167344eda4d7Smaybee 		}
167444eda4d7Smaybee 
167544eda4d7Smaybee 		if (tocpy == db->db_size)
167644eda4d7Smaybee 			dmu_buf_fill_done(db, tx);
167744eda4d7Smaybee 
167844eda4d7Smaybee 		offset += tocpy;
167944eda4d7Smaybee 		size -= tocpy;
168044eda4d7Smaybee 	}
168144eda4d7Smaybee 	dmu_buf_rele_array(dbp, numbufs, FTAG);
168244eda4d7Smaybee 	return (err);
168344eda4d7Smaybee }
1684fa9e4066Sahrens #endif
1685fa9e4066Sahrens 
16862fdbea25SAleksandr Guzovskiy /*
16872fdbea25SAleksandr Guzovskiy  * Allocate a loaned anonymous arc buffer.
16882fdbea25SAleksandr Guzovskiy  */
16892fdbea25SAleksandr Guzovskiy arc_buf_t *
16902fdbea25SAleksandr Guzovskiy dmu_request_arcbuf(dmu_buf_t *handle, int size)
16912fdbea25SAleksandr Guzovskiy {
1692744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
16932fdbea25SAleksandr Guzovskiy 
16945602294fSDan Kimmel 	return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size));
16952fdbea25SAleksandr Guzovskiy }
16962fdbea25SAleksandr Guzovskiy 
16972fdbea25SAleksandr Guzovskiy /*
16982fdbea25SAleksandr Guzovskiy  * Free a loaned arc buffer.
16992fdbea25SAleksandr Guzovskiy  */
17002fdbea25SAleksandr Guzovskiy void
17012fdbea25SAleksandr Guzovskiy dmu_return_arcbuf(arc_buf_t *buf)
17022fdbea25SAleksandr Guzovskiy {
17032fdbea25SAleksandr Guzovskiy 	arc_return_buf(buf, FTAG);
1704dcbf3bd6SGeorge Wilson 	arc_buf_destroy(buf, FTAG);
17052fdbea25SAleksandr Guzovskiy }
17062fdbea25SAleksandr Guzovskiy 
1707eb633035STom Caputi void
1708eb633035STom Caputi dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset,
1709eb633035STom Caputi     dmu_buf_t *handle, dmu_tx_t *tx)
1710eb633035STom Caputi {
1711eb633035STom Caputi 	dmu_buf_t *dst_handle;
1712eb633035STom Caputi 	dmu_buf_impl_t *dstdb;
1713eb633035STom Caputi 	dmu_buf_impl_t *srcdb = (dmu_buf_impl_t *)handle;
1714a60ca23dSTom Caputi 	dmu_object_type_t type;
1715eb633035STom Caputi 	arc_buf_t *abuf;
1716eb633035STom Caputi 	uint64_t datalen;
1717eb633035STom Caputi 	boolean_t byteorder;
1718eb633035STom Caputi 	uint8_t salt[ZIO_DATA_SALT_LEN];
1719eb633035STom Caputi 	uint8_t iv[ZIO_DATA_IV_LEN];
1720eb633035STom Caputi 	uint8_t mac[ZIO_DATA_MAC_LEN];
1721eb633035STom Caputi 
1722eb633035STom Caputi 	ASSERT3P(srcdb->db_buf, !=, NULL);
1723eb633035STom Caputi 
1724eb633035STom Caputi 	/* hold the db that we want to write to */
1725eb633035STom Caputi 	VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &dst_handle,
1726eb633035STom Caputi 	    DMU_READ_NO_DECRYPT));
1727eb633035STom Caputi 	dstdb = (dmu_buf_impl_t *)dst_handle;
1728eb633035STom Caputi 	datalen = arc_buf_size(srcdb->db_buf);
1729eb633035STom Caputi 
1730a60ca23dSTom Caputi 	DB_DNODE_ENTER(dstdb);
1731a60ca23dSTom Caputi 	type = DB_DNODE(dstdb)->dn_type;
1732a60ca23dSTom Caputi 	DB_DNODE_EXIT(dstdb);
1733a60ca23dSTom Caputi 
1734eb633035STom Caputi 	/* allocated an arc buffer that matches the type of srcdb->db_buf */
1735eb633035STom Caputi 	if (arc_is_encrypted(srcdb->db_buf)) {
1736eb633035STom Caputi 		arc_get_raw_params(srcdb->db_buf, &byteorder, salt, iv, mac);
1737eb633035STom Caputi 		abuf = arc_loan_raw_buf(os->os_spa, dmu_objset_id(os),
1738a60ca23dSTom Caputi 		    byteorder, salt, iv, mac, type,
1739eb633035STom Caputi 		    datalen, arc_buf_lsize(srcdb->db_buf),
1740eb633035STom Caputi 		    arc_get_compression(srcdb->db_buf));
1741eb633035STom Caputi 	} else {
1742eb633035STom Caputi 		/* we won't get a compressed db back from dmu_buf_hold() */
1743eb633035STom Caputi 		ASSERT3U(arc_get_compression(srcdb->db_buf),
1744eb633035STom Caputi 		    ==, ZIO_COMPRESS_OFF);
1745eb633035STom Caputi 		abuf = arc_loan_buf(os->os_spa,
1746a60ca23dSTom Caputi 		    DMU_OT_IS_METADATA(type), datalen);
1747eb633035STom Caputi 	}
1748eb633035STom Caputi 
1749eb633035STom Caputi 	ASSERT3U(datalen, ==, arc_buf_size(abuf));
1750eb633035STom Caputi 
1751eb633035STom Caputi 	/* copy the data to the new buffer and assign it to the dstdb */
1752eb633035STom Caputi 	bcopy(srcdb->db_buf->b_data, abuf->b_data, datalen);
1753eb633035STom Caputi 	dbuf_assign_arcbuf(dstdb, abuf, tx);
1754eb633035STom Caputi 	dmu_buf_rele(dst_handle, FTAG);
1755eb633035STom Caputi }
1756eb633035STom Caputi 
17572fdbea25SAleksandr Guzovskiy /*
17582fdbea25SAleksandr Guzovskiy  * When possible directly assign passed loaned arc buffer to a dbuf.
17592fdbea25SAleksandr Guzovskiy  * If this is not possible copy the contents of passed arc buf via
17602fdbea25SAleksandr Guzovskiy  * dmu_write().
17612fdbea25SAleksandr Guzovskiy  */
1762eb633035STom Caputi int
1763eb633035STom Caputi dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf,
17642fdbea25SAleksandr Guzovskiy     dmu_tx_t *tx)
17652fdbea25SAleksandr Guzovskiy {
17662fdbea25SAleksandr Guzovskiy 	dmu_buf_impl_t *db;
1767eb633035STom Caputi 	objset_t *os = dn->dn_objset;
1768eb633035STom Caputi 	uint64_t object = dn->dn_object;
17695602294fSDan Kimmel 	uint32_t blksz = (uint32_t)arc_buf_lsize(buf);
17702fdbea25SAleksandr Guzovskiy 	uint64_t blkid;
17712fdbea25SAleksandr Guzovskiy 
17722fdbea25SAleksandr Guzovskiy 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
1773a2cdcdd2SPaul Dagnelie 	blkid = dbuf_whichblock(dn, 0, offset);
1774eb633035STom Caputi 	db = dbuf_hold(dn, blkid, FTAG);
1775eb633035STom Caputi 	if (db == NULL)
1776eb633035STom Caputi 		return (SET_ERROR(EIO));
17772fdbea25SAleksandr Guzovskiy 	rw_exit(&dn->dn_struct_rwlock);
17782fdbea25SAleksandr Guzovskiy 
17798a904709SMatthew Ahrens 	/*
17808a904709SMatthew Ahrens 	 * We can only assign if the offset is aligned, the arc buf is the
17815602294fSDan Kimmel 	 * same size as the dbuf, and the dbuf is not metadata.
17828a904709SMatthew Ahrens 	 */
17835602294fSDan Kimmel 	if (offset == db->db.db_offset && blksz == db->db.db_size) {
17842fdbea25SAleksandr Guzovskiy 		dbuf_assign_arcbuf(db, buf, tx);
17852fdbea25SAleksandr Guzovskiy 		dbuf_rele(db, FTAG);
17862fdbea25SAleksandr Guzovskiy 	} else {
17875602294fSDan Kimmel 		/* compressed bufs must always be assignable to their dbuf */
17885602294fSDan Kimmel 		ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF);
17895602294fSDan Kimmel 		ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED));
17905602294fSDan Kimmel 
1791744947dcSTom Erickson 		os = dn->dn_objset;
1792744947dcSTom Erickson 		object = dn->dn_object;
17932fdbea25SAleksandr Guzovskiy 		dbuf_rele(db, FTAG);
1794744947dcSTom Erickson 		dmu_write(os, object, offset, blksz, buf->b_data, tx);
17952fdbea25SAleksandr Guzovskiy 		dmu_return_arcbuf(buf);
1796c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 		XUIOSTAT_BUMP(xuiostat_wbuf_copied);
17972fdbea25SAleksandr Guzovskiy 	}
1798eb633035STom Caputi 
1799eb633035STom Caputi 	return (0);
18002fdbea25SAleksandr Guzovskiy }
18012fdbea25SAleksandr Guzovskiy 
1802eb633035STom Caputi int
1803eb633035STom Caputi dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
18048dfe5547SRichard Yao     dmu_tx_t *tx)
18058dfe5547SRichard Yao {
1806eb633035STom Caputi 	int err;
18078dfe5547SRichard Yao 	dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
18088dfe5547SRichard Yao 
18098dfe5547SRichard Yao 	DB_DNODE_ENTER(dbuf);
1810eb633035STom Caputi 	err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx);
18118dfe5547SRichard Yao 	DB_DNODE_EXIT(dbuf);
1812eb633035STom Caputi 
1813eb633035STom Caputi 	return (err);
18148dfe5547SRichard Yao }
18158dfe5547SRichard Yao 
1816c5c6ffa0Smaybee typedef struct {
1817b24ab676SJeff Bonwick 	dbuf_dirty_record_t	*dsa_dr;
1818b24ab676SJeff Bonwick 	dmu_sync_cb_t		*dsa_done;
1819b24ab676SJeff Bonwick 	zgd_t			*dsa_zgd;
1820b24ab676SJeff Bonwick 	dmu_tx_t		*dsa_tx;
1821c717a561Smaybee } dmu_sync_arg_t;
1822c5c6ffa0Smaybee 
1823e14bb325SJeff Bonwick /* ARGSUSED */
1824e14bb325SJeff Bonwick static void
1825e14bb325SJeff Bonwick dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1826e14bb325SJeff Bonwick {
1827b24ab676SJeff Bonwick 	dmu_sync_arg_t *dsa = varg;
1828b24ab676SJeff Bonwick 	dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1829e14bb325SJeff Bonwick 	blkptr_t *bp = zio->io_bp;
1830975c32a0SNeil Perrin 
1831b24ab676SJeff Bonwick 	if (zio->io_error == 0) {
1832b24ab676SJeff Bonwick 		if (BP_IS_HOLE(bp)) {
1833b24ab676SJeff Bonwick 			/*
1834b24ab676SJeff Bonwick 			 * A block of zeros may compress to a hole, but the
1835b24ab676SJeff Bonwick 			 * block size still needs to be known for replay.
1836b24ab676SJeff Bonwick 			 */
1837b24ab676SJeff Bonwick 			BP_SET_LSIZE(bp, db->db_size);
18385d7b4d43SMatthew Ahrens 		} else if (!BP_IS_EMBEDDED(bp)) {
1839b24ab676SJeff Bonwick 			ASSERT(BP_GET_LEVEL(bp) == 0);
1840eb633035STom Caputi 			BP_SET_FILL(bp, 1);
1841b24ab676SJeff Bonwick 		}
1842e14bb325SJeff Bonwick 	}
1843e14bb325SJeff Bonwick }
1844e14bb325SJeff Bonwick 
1845b24ab676SJeff Bonwick static void
1846b24ab676SJeff Bonwick dmu_sync_late_arrival_ready(zio_t *zio)
1847b24ab676SJeff Bonwick {
1848b24ab676SJeff Bonwick 	dmu_sync_ready(zio, NULL, zio->io_private);
1849b24ab676SJeff Bonwick }
1850b24ab676SJeff Bonwick 
1851c5c6ffa0Smaybee /* ARGSUSED */
1852c5c6ffa0Smaybee static void
1853c5c6ffa0Smaybee dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1854c5c6ffa0Smaybee {
1855b24ab676SJeff Bonwick 	dmu_sync_arg_t *dsa = varg;
1856b24ab676SJeff Bonwick 	dbuf_dirty_record_t *dr = dsa->dsa_dr;
1857c717a561Smaybee 	dmu_buf_impl_t *db = dr->dr_dbuf;
1858cab3a55eSPrakash Surya 	zgd_t *zgd = dsa->dsa_zgd;
1859cab3a55eSPrakash Surya 
1860cab3a55eSPrakash Surya 	/*
1861cab3a55eSPrakash Surya 	 * Record the vdev(s) backing this blkptr so they can be flushed after
1862cab3a55eSPrakash Surya 	 * the writes for the lwb have completed.
1863cab3a55eSPrakash Surya 	 */
1864cab3a55eSPrakash Surya 	if (zio->io_error == 0) {
1865cab3a55eSPrakash Surya 		zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1866cab3a55eSPrakash Surya 	}
1867c5c6ffa0Smaybee 
1868b50a0fe0SNeil Perrin 	mutex_enter(&db->db_mtx);
1869b50a0fe0SNeil Perrin 	ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1870b24ab676SJeff Bonwick 	if (zio->io_error == 0) {
187180901aeaSGeorge Wilson 		dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
187280901aeaSGeorge Wilson 		if (dr->dt.dl.dr_nopwrite) {
187380901aeaSGeorge Wilson 			blkptr_t *bp = zio->io_bp;
187480901aeaSGeorge Wilson 			blkptr_t *bp_orig = &zio->io_bp_orig;
187580901aeaSGeorge Wilson 			uint8_t chksum = BP_GET_CHECKSUM(bp_orig);
187680901aeaSGeorge Wilson 
187780901aeaSGeorge Wilson 			ASSERT(BP_EQUAL(bp, bp_orig));
1878b7edcb94SMatthew Ahrens 			VERIFY(BP_EQUAL(bp, db->db_blkptr));
187980901aeaSGeorge Wilson 			ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
188045818ee1SMatthew Ahrens 			ASSERT(zio_checksum_table[chksum].ci_flags &
188145818ee1SMatthew Ahrens 			    ZCHECKSUM_FLAG_NOPWRITE);
188280901aeaSGeorge Wilson 		}
1883b24ab676SJeff Bonwick 		dr->dt.dl.dr_overridden_by = *zio->io_bp;
1884b24ab676SJeff Bonwick 		dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1885b24ab676SJeff Bonwick 		dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
188670163ac5SPrakash Surya 
188770163ac5SPrakash Surya 		/*
188870163ac5SPrakash Surya 		 * Old style holes are filled with all zeros, whereas
188970163ac5SPrakash Surya 		 * new-style holes maintain their lsize, type, level,
189070163ac5SPrakash Surya 		 * and birth time (see zio_write_compress). While we
189170163ac5SPrakash Surya 		 * need to reset the BP_SET_LSIZE() call that happened
189270163ac5SPrakash Surya 		 * in dmu_sync_ready for old style holes, we do *not*
189370163ac5SPrakash Surya 		 * want to wipe out the information contained in new
189470163ac5SPrakash Surya 		 * style holes. Thus, only zero out the block pointer if
189570163ac5SPrakash Surya 		 * it's an old style hole.
189670163ac5SPrakash Surya 		 */
189770163ac5SPrakash Surya 		if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) &&
189870163ac5SPrakash Surya 		    dr->dt.dl.dr_overridden_by.blk_birth == 0)
1899b24ab676SJeff Bonwick 			BP_ZERO(&dr->dt.dl.dr_overridden_by);
1900b24ab676SJeff Bonwick 	} else {
1901b24ab676SJeff Bonwick 		dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1902b24ab676SJeff Bonwick 	}
1903c5c6ffa0Smaybee 	cv_broadcast(&db->db_changed);
1904b50a0fe0SNeil Perrin 	mutex_exit(&db->db_mtx);
1905b50a0fe0SNeil Perrin 
1906b24ab676SJeff Bonwick 	dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1907b24ab676SJeff Bonwick 
1908b24ab676SJeff Bonwick 	kmem_free(dsa, sizeof (*dsa));
1909b24ab676SJeff Bonwick }
1910b24ab676SJeff Bonwick 
1911b24ab676SJeff Bonwick static void
1912b24ab676SJeff Bonwick dmu_sync_late_arrival_done(zio_t *zio)
1913b24ab676SJeff Bonwick {
1914b24ab676SJeff Bonwick 	blkptr_t *bp = zio->io_bp;
1915b24ab676SJeff Bonwick 	dmu_sync_arg_t *dsa = zio->io_private;
191680901aeaSGeorge Wilson 	blkptr_t *bp_orig = &zio->io_bp_orig;
1917cab3a55eSPrakash Surya 	zgd_t *zgd = dsa->dsa_zgd;
1918b24ab676SJeff Bonwick 
1919cab3a55eSPrakash Surya 	if (zio->io_error == 0) {
1920cab3a55eSPrakash Surya 		/*
1921cab3a55eSPrakash Surya 		 * Record the vdev(s) backing this blkptr so they can be
1922cab3a55eSPrakash Surya 		 * flushed after the writes for the lwb have completed.
1923cab3a55eSPrakash Surya 		 */
1924cab3a55eSPrakash Surya 		zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1925cab3a55eSPrakash Surya 
1926cab3a55eSPrakash Surya 		if (!BP_IS_HOLE(bp)) {
1927cab3a55eSPrakash Surya 			ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE));
1928cab3a55eSPrakash Surya 			ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1929cab3a55eSPrakash Surya 			ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1930cab3a55eSPrakash Surya 			ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1931cab3a55eSPrakash Surya 			zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1932cab3a55eSPrakash Surya 		}
1933b24ab676SJeff Bonwick 	}
1934b24ab676SJeff Bonwick 
1935b24ab676SJeff Bonwick 	dmu_tx_commit(dsa->dsa_tx);
1936b24ab676SJeff Bonwick 
1937b24ab676SJeff Bonwick 	dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1938b24ab676SJeff Bonwick 
1939770499e1SDan Kimmel 	abd_put(zio->io_abd);
1940b24ab676SJeff Bonwick 	kmem_free(dsa, sizeof (*dsa));
1941b24ab676SJeff Bonwick }
1942b24ab676SJeff Bonwick 
1943b24ab676SJeff Bonwick static int
1944b24ab676SJeff Bonwick dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
19457802d7bfSMatthew Ahrens     zio_prop_t *zp, zbookmark_phys_t *zb)
1946b24ab676SJeff Bonwick {
1947b24ab676SJeff Bonwick 	dmu_sync_arg_t *dsa;
1948b24ab676SJeff Bonwick 	dmu_tx_t *tx;
1949b24ab676SJeff Bonwick 
1950b24ab676SJeff Bonwick 	tx = dmu_tx_create(os);
1951b24ab676SJeff Bonwick 	dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
19526e1f5caaSNeil Perrin 	if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1953b24ab676SJeff Bonwick 		dmu_tx_abort(tx);
1954be6fd75aSMatthew Ahrens 		/* Make zl_get_data do txg_waited_synced() */
1955be6fd75aSMatthew Ahrens 		return (SET_ERROR(EIO));
1956b24ab676SJeff Bonwick 	}
1957b24ab676SJeff Bonwick 
19581271e4b1SPrakash Surya 	/*
19591271e4b1SPrakash Surya 	 * In order to prevent the zgd's lwb from being free'd prior to
19601271e4b1SPrakash Surya 	 * dmu_sync_late_arrival_done() being called, we have to ensure
19611271e4b1SPrakash Surya 	 * the lwb's "max txg" takes this tx's txg into account.
19621271e4b1SPrakash Surya 	 */
19631271e4b1SPrakash Surya 	zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx));
19641271e4b1SPrakash Surya 
1965b24ab676SJeff Bonwick 	dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1966b24ab676SJeff Bonwick 	dsa->dsa_dr = NULL;
1967b24ab676SJeff Bonwick 	dsa->dsa_done = done;
1968b24ab676SJeff Bonwick 	dsa->dsa_zgd = zgd;
1969b24ab676SJeff Bonwick 	dsa->dsa_tx = tx;
1970c717a561Smaybee 
1971b7edcb94SMatthew Ahrens 	/*
1972b7edcb94SMatthew Ahrens 	 * Since we are currently syncing this txg, it's nontrivial to
1973b7edcb94SMatthew Ahrens 	 * determine what BP to nopwrite against, so we disable nopwrite.
1974b7edcb94SMatthew Ahrens 	 *
1975b7edcb94SMatthew Ahrens 	 * When syncing, the db_blkptr is initially the BP of the previous
1976b7edcb94SMatthew Ahrens 	 * txg.  We can not nopwrite against it because it will be changed
1977b7edcb94SMatthew Ahrens 	 * (this is similar to the non-late-arrival case where the dbuf is
1978b7edcb94SMatthew Ahrens 	 * dirty in a future txg).
1979b7edcb94SMatthew Ahrens 	 *
1980b7edcb94SMatthew Ahrens 	 * Then dbuf_write_ready() sets bp_blkptr to the location we will write.
1981b7edcb94SMatthew Ahrens 	 * We can not nopwrite against it because although the BP will not
1982b7edcb94SMatthew Ahrens 	 * (typically) be changed, the data has not yet been persisted to this
1983b7edcb94SMatthew Ahrens 	 * location.
1984b7edcb94SMatthew Ahrens 	 *
1985b7edcb94SMatthew Ahrens 	 * Finally, when dbuf_write_done() is called, it is theoretically
1986b7edcb94SMatthew Ahrens 	 * possible to always nopwrite, because the data that was written in
1987b7edcb94SMatthew Ahrens 	 * this txg is the same data that we are trying to write.  However we
1988b7edcb94SMatthew Ahrens 	 * would need to check that this dbuf is not dirty in any future
1989b7edcb94SMatthew Ahrens 	 * txg's (as we do in the normal dmu_sync() path). For simplicity, we
1990b7edcb94SMatthew Ahrens 	 * don't nopwrite in this case.
1991b7edcb94SMatthew Ahrens 	 */
1992b7edcb94SMatthew Ahrens 	zp->zp_nopwrite = B_FALSE;
1993b7edcb94SMatthew Ahrens 
19945602294fSDan Kimmel 	zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1995770499e1SDan Kimmel 	    abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size),
1996770499e1SDan Kimmel 	    zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp,
1997770499e1SDan Kimmel 	    dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done,
1998770499e1SDan Kimmel 	    dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
1999b24ab676SJeff Bonwick 
2000b24ab676SJeff Bonwick 	return (0);
2001c5c6ffa0Smaybee }
2002c5c6ffa0Smaybee 
2003fa9e4066Sahrens /*
2004c5c6ffa0Smaybee  * Intent log support: sync the block associated with db to disk.
2005c5c6ffa0Smaybee  * N.B. and XXX: the caller is responsible for making sure that the
2006c5c6ffa0Smaybee  * data isn't changing while dmu_sync() is writing it.
2007fa9e4066Sahrens  *
2008fa9e4066Sahrens  * Return values:
2009fa9e4066Sahrens  *
201080901aeaSGeorge Wilson  *	EEXIST: this txg has already been synced, so there's nothing to do.
2011fa9e4066Sahrens  *		The caller should not log the write.
2012fa9e4066Sahrens  *
2013fa9e4066Sahrens  *	ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
2014fa9e4066Sahrens  *		The caller should not log the write.
2015fa9e4066Sahrens  *
2016c5c6ffa0Smaybee  *	EALREADY: this block is already in the process of being synced.
2017c5c6ffa0Smaybee  *		The caller should track its progress (somehow).
2018fa9e4066Sahrens  *
2019b24ab676SJeff Bonwick  *	EIO: could not do the I/O.
2020b24ab676SJeff Bonwick  *		The caller should do a txg_wait_synced().
2021fa9e4066Sahrens  *
2022b24ab676SJeff Bonwick  *	0: the I/O has been initiated.
2023b24ab676SJeff Bonwick  *		The caller should log this blkptr in the done callback.
2024b24ab676SJeff Bonwick  *		It is possible that the I/O will fail, in which case
2025b24ab676SJeff Bonwick  *		the error will be reported to the done callback and
2026b24ab676SJeff Bonwick  *		propagated to pio from zio_done().
2027fa9e4066Sahrens  */
2028fa9e4066Sahrens int
2029b24ab676SJeff Bonwick dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
2030fa9e4066Sahrens {
2031b24ab676SJeff Bonwick 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
2032503ad85cSMatthew Ahrens 	objset_t *os = db->db_objset;
2033b24ab676SJeff Bonwick 	dsl_dataset_t *ds = os->os_dsl_dataset;
2034c717a561Smaybee 	dbuf_dirty_record_t *dr;
2035b24ab676SJeff Bonwick 	dmu_sync_arg_t *dsa;
20367802d7bfSMatthew Ahrens 	zbookmark_phys_t zb;
2037b24ab676SJeff Bonwick 	zio_prop_t zp;
2038744947dcSTom Erickson 	dnode_t *dn;
2039fa9e4066Sahrens 
2040b24ab676SJeff Bonwick 	ASSERT(pio != NULL);
2041fa9e4066Sahrens 	ASSERT(txg != 0);
2042fa9e4066Sahrens 
2043b24ab676SJeff Bonwick 	SET_BOOKMARK(&zb, ds->ds_object,
2044b24ab676SJeff Bonwick 	    db->db.db_object, db->db_level, db->db_blkid);
2045b24ab676SJeff Bonwick 
2046744947dcSTom Erickson 	DB_DNODE_ENTER(db);
2047744947dcSTom Erickson 	dn = DB_DNODE(db);
2048adaec86aSMatthew Ahrens 	dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
2049744947dcSTom Erickson 	DB_DNODE_EXIT(db);
2050fa9e4066Sahrens 
2051ea8dc4b6Seschrock 	/*
2052b24ab676SJeff Bonwick 	 * If we're frozen (running ziltest), we always need to generate a bp.
2053ea8dc4b6Seschrock 	 */
2054b24ab676SJeff Bonwick 	if (txg > spa_freeze_txg(os->os_spa))
2055b24ab676SJeff Bonwick 		return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
2056ea8dc4b6Seschrock 
2057fa9e4066Sahrens 	/*
2058b24ab676SJeff Bonwick 	 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
2059b24ab676SJeff Bonwick 	 * and us.  If we determine that this txg is not yet syncing,
2060b24ab676SJeff Bonwick 	 * but it begins to sync a moment later, that's OK because the
2061b24ab676SJeff Bonwick 	 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
2062fa9e4066Sahrens 	 */
2063b24ab676SJeff Bonwick 	mutex_enter(&db->db_mtx);
2064b24ab676SJeff Bonwick 
2065b24ab676SJeff Bonwick 	if (txg <= spa_last_synced_txg(os->os_spa)) {
2066fa9e4066Sahrens 		/*
2067b24ab676SJeff Bonwick 		 * This txg has already synced.  There's nothing to do.
2068fa9e4066Sahrens 		 */
2069b24ab676SJeff Bonwick 		mutex_exit(&db->db_mtx);
2070be6fd75aSMatthew Ahrens 		return (SET_ERROR(EEXIST));
2071fa9e4066Sahrens 	}
2072fa9e4066Sahrens 
2073b24ab676SJeff Bonwick 	if (txg <= spa_syncing_txg(os->os_spa)) {
2074b24ab676SJeff Bonwick 		/*
2075b24ab676SJeff Bonwick 		 * This txg is currently syncing, so we can't mess with
2076b24ab676SJeff Bonwick 		 * the dirty record anymore; just write a new log block.
2077b24ab676SJeff Bonwick 		 */
2078b24ab676SJeff Bonwick 		mutex_exit(&db->db_mtx);
2079b24ab676SJeff Bonwick 		return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
2080fa9e4066Sahrens 	}
2081fa9e4066Sahrens 
2082c717a561Smaybee 	dr = db->db_last_dirty;
2083b24ab676SJeff Bonwick 	while (dr && dr->dr_txg != txg)
2084c717a561Smaybee 		dr = dr->dr_next;
2085b24ab676SJeff Bonwick 
2086b24ab676SJeff Bonwick 	if (dr == NULL) {
2087c5c6ffa0Smaybee 		/*
2088b24ab676SJeff Bonwick 		 * There's no dr for this dbuf, so it must have been freed.
2089c5c6ffa0Smaybee 		 * There's no need to log writes to freed blocks, so we're done.
2090c5c6ffa0Smaybee 		 */
2091c5c6ffa0Smaybee 		mutex_exit(&db->db_mtx);
2092be6fd75aSMatthew Ahrens 		return (SET_ERROR(ENOENT));
2093c5c6ffa0Smaybee 	}
2094c5c6ffa0Smaybee 
209580901aeaSGeorge Wilson 	ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
209680901aeaSGeorge Wilson 
2097b7edcb94SMatthew Ahrens 	if (db->db_blkptr != NULL) {
2098b7edcb94SMatthew Ahrens 		/*
2099b7edcb94SMatthew Ahrens 		 * We need to fill in zgd_bp with the current blkptr so that
2100b7edcb94SMatthew Ahrens 		 * the nopwrite code can check if we're writing the same
2101b7edcb94SMatthew Ahrens 		 * data that's already on disk.  We can only nopwrite if we
2102b7edcb94SMatthew Ahrens 		 * are sure that after making the copy, db_blkptr will not
2103b7edcb94SMatthew Ahrens 		 * change until our i/o completes.  We ensure this by
2104b7edcb94SMatthew Ahrens 		 * holding the db_mtx, and only allowing nopwrite if the
2105b7edcb94SMatthew Ahrens 		 * block is not already dirty (see below).  This is verified
2106b7edcb94SMatthew Ahrens 		 * by dmu_sync_done(), which VERIFYs that the db_blkptr has
2107b7edcb94SMatthew Ahrens 		 * not changed.
2108b7edcb94SMatthew Ahrens 		 */
2109b7edcb94SMatthew Ahrens 		*zgd->zgd_bp = *db->db_blkptr;
2110b7edcb94SMatthew Ahrens 	}
2111b7edcb94SMatthew Ahrens 
211280901aeaSGeorge Wilson 	/*
211334e8acefSMatthew Ahrens 	 * Assume the on-disk data is X, the current syncing data (in
211434e8acefSMatthew Ahrens 	 * txg - 1) is Y, and the current in-memory data is Z (currently
211534e8acefSMatthew Ahrens 	 * in dmu_sync).
211634e8acefSMatthew Ahrens 	 *
211734e8acefSMatthew Ahrens 	 * We usually want to perform a nopwrite if X and Z are the
211834e8acefSMatthew Ahrens 	 * same.  However, if Y is different (i.e. the BP is going to
211934e8acefSMatthew Ahrens 	 * change before this write takes effect), then a nopwrite will
212034e8acefSMatthew Ahrens 	 * be incorrect - we would override with X, which could have
212134e8acefSMatthew Ahrens 	 * been freed when Y was written.
212234e8acefSMatthew Ahrens 	 *
212334e8acefSMatthew Ahrens 	 * (Note that this is not a concern when we are nop-writing from
212434e8acefSMatthew Ahrens 	 * syncing context, because X and Y must be identical, because
212534e8acefSMatthew Ahrens 	 * all previous txgs have been synced.)
212634e8acefSMatthew Ahrens 	 *
212734e8acefSMatthew Ahrens 	 * Therefore, we disable nopwrite if the current BP could change
212834e8acefSMatthew Ahrens 	 * before this TXG.  There are two ways it could change: by
212934e8acefSMatthew Ahrens 	 * being dirty (dr_next is non-NULL), or by being freed
213034e8acefSMatthew Ahrens 	 * (dnode_block_freed()).  This behavior is verified by
213134e8acefSMatthew Ahrens 	 * zio_done(), which VERIFYs that the override BP is identical
213234e8acefSMatthew Ahrens 	 * to the on-disk BP.
213380901aeaSGeorge Wilson 	 */
213434e8acefSMatthew Ahrens 	DB_DNODE_ENTER(db);
213534e8acefSMatthew Ahrens 	dn = DB_DNODE(db);
213634e8acefSMatthew Ahrens 	if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid))
213780901aeaSGeorge Wilson 		zp.zp_nopwrite = B_FALSE;
213834e8acefSMatthew Ahrens 	DB_DNODE_EXIT(db);
213980901aeaSGeorge Wilson 
2140c717a561Smaybee 	ASSERT(dr->dr_txg == txg);
2141b24ab676SJeff Bonwick 	if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
2142b24ab676SJeff Bonwick 	    dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
2143c717a561Smaybee 		/*
2144b24ab676SJeff Bonwick 		 * We have already issued a sync write for this buffer,
2145b24ab676SJeff Bonwick 		 * or this buffer has already been synced.  It could not
2146c717a561Smaybee 		 * have been dirtied since, or we would have cleared the state.
2147c717a561Smaybee 		 */
2148c717a561Smaybee 		mutex_exit(&db->db_mtx);
2149be6fd75aSMatthew Ahrens 		return (SET_ERROR(EALREADY));
2150c717a561Smaybee 	}
2151c717a561Smaybee 
2152b24ab676SJeff Bonwick 	ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
2153c717a561Smaybee 	dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
2154fa9e4066Sahrens 	mutex_exit(&db->db_mtx);
2155e14bb325SJeff Bonwick 
2156b24ab676SJeff Bonwick 	dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
2157b24ab676SJeff Bonwick 	dsa->dsa_dr = dr;
2158b24ab676SJeff Bonwick 	dsa->dsa_done = done;
2159b24ab676SJeff Bonwick 	dsa->dsa_zgd = zgd;
2160b24ab676SJeff Bonwick 	dsa->dsa_tx = NULL;
2161e14bb325SJeff Bonwick 
2162b24ab676SJeff Bonwick 	zio_nowait(arc_write(pio, os->os_spa, txg,
2163b7edcb94SMatthew Ahrens 	    zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
2164dcbf3bd6SGeorge Wilson 	    &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa,
21658df0bcf0SPaul Dagnelie 	    ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
2166e14bb325SJeff Bonwick 
2167b24ab676SJeff Bonwick 	return (0);
2168fa9e4066Sahrens }
2169fa9e4066Sahrens 
2170eb633035STom Caputi int
2171eb633035STom Caputi dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx)
2172eb633035STom Caputi {
2173eb633035STom Caputi 	dnode_t *dn;
2174eb633035STom Caputi 	int err;
2175eb633035STom Caputi 
2176eb633035STom Caputi 	err = dnode_hold(os, object, FTAG, &dn);
2177eb633035STom Caputi 	if (err)
2178eb633035STom Caputi 		return (err);
2179eb633035STom Caputi 	err = dnode_set_nlevels(dn, nlevels, tx);
2180eb633035STom Caputi 	dnode_rele(dn, FTAG);
2181eb633035STom Caputi 	return (err);
2182eb633035STom Caputi }
2183eb633035STom Caputi 
2184fa9e4066Sahrens int
2185fa9e4066Sahrens dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
21869a686fbcSPaul Dagnelie     dmu_tx_t *tx)
2187fa9e4066Sahrens {
2188ea8dc4b6Seschrock 	dnode_t *dn;
2189ea8dc4b6Seschrock 	int err;
2190ea8dc4b6Seschrock 
2191503ad85cSMatthew Ahrens 	err = dnode_hold(os, object, FTAG, &dn);
2192ea8dc4b6Seschrock 	if (err)
2193ea8dc4b6Seschrock 		return (err);
2194ea8dc4b6Seschrock 	err = dnode_set_blksz(dn, size, ibs, tx);
2195fa9e4066Sahrens 	dnode_rele(dn, FTAG);
2196fa9e4066Sahrens 	return (err);
2197fa9e4066Sahrens }
2198fa9e4066Sahrens 
2199eb633035STom Caputi int
2200eb633035STom Caputi dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid,
2201eb633035STom Caputi     dmu_tx_t *tx)
2202eb633035STom Caputi {
2203eb633035STom Caputi 	dnode_t *dn;
2204eb633035STom Caputi 	int err;
2205eb633035STom Caputi 
2206eb633035STom Caputi 	err = dnode_hold(os, object, FTAG, &dn);
2207eb633035STom Caputi 	if (err)
2208eb633035STom Caputi 		return (err);
2209eb633035STom Caputi 	rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
2210eb633035STom Caputi 	dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE);
2211eb633035STom Caputi 	rw_exit(&dn->dn_struct_rwlock);
2212eb633035STom Caputi 	dnode_rele(dn, FTAG);
2213eb633035STom Caputi 	return (0);
2214eb633035STom Caputi }
2215eb633035STom Caputi 
2216fa9e4066Sahrens void
2217fa9e4066Sahrens dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
22189a686fbcSPaul Dagnelie     dmu_tx_t *tx)
2219fa9e4066Sahrens {
2220ea8dc4b6Seschrock 	dnode_t *dn;
2221ea8dc4b6Seschrock 
22225d7b4d43SMatthew Ahrens 	/*
22235d7b4d43SMatthew Ahrens 	 * Send streams include each object's checksum function.  This
22245d7b4d43SMatthew Ahrens 	 * check ensures that the receiving system can understand the
22255d7b4d43SMatthew Ahrens 	 * checksum function transmitted.
22265d7b4d43SMatthew Ahrens 	 */
22275d7b4d43SMatthew Ahrens 	ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
22285d7b4d43SMatthew Ahrens 
22295d7b4d43SMatthew Ahrens 	VERIFY0(dnode_hold(os, object, FTAG, &dn));
22305d7b4d43SMatthew Ahrens 	ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
2231fa9e4066Sahrens 	dn->dn_checksum = checksum;
2232fa9e4066Sahrens 	dnode_setdirty(dn, tx);
2233fa9e4066Sahrens 	dnode_rele(dn, FTAG);
2234fa9e4066Sahrens }
2235fa9e4066Sahrens 
2236fa9e4066Sahrens void
2237fa9e4066Sahrens dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
22389a686fbcSPaul Dagnelie     dmu_tx_t *tx)
2239fa9e4066Sahrens {
2240ea8dc4b6Seschrock 	dnode_t *dn;
2241ea8dc4b6Seschrock 
22425d7b4d43SMatthew Ahrens 	/*
22435d7b4d43SMatthew Ahrens 	 * Send streams include each object's compression function.  This
22445d7b4d43SMatthew Ahrens 	 * check ensures that the receiving system can understand the
22455d7b4d43SMatthew Ahrens 	 * compression function transmitted.
22465d7b4d43SMatthew Ahrens 	 */
22475d7b4d43SMatthew Ahrens 	ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
22485d7b4d43SMatthew Ahrens 
22495d7b4d43SMatthew Ahrens 	VERIFY0(dnode_hold(os, object, FTAG, &dn));
2250fa9e4066Sahrens 	dn->dn_compress = compress;
2251fa9e4066Sahrens 	dnode_setdirty(dn, tx);
2252fa9e4066Sahrens 	dnode_rele(dn, FTAG);
2253fa9e4066Sahrens }
2254fa9e4066Sahrens 
2255edf345e6SMatthew Ahrens /*
2256edf345e6SMatthew Ahrens  * When the "redundant_metadata" property is set to "most", only indirect
2257edf345e6SMatthew Ahrens  * blocks of this level and higher will have an additional ditto block.
2258edf345e6SMatthew Ahrens  */
2259edf345e6SMatthew Ahrens int zfs_redundant_metadata_most_ditto_level = 2;
2260edf345e6SMatthew Ahrens 
2261b24ab676SJeff Bonwick void
2262adaec86aSMatthew Ahrens dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
2263b24ab676SJeff Bonwick {
2264b24ab676SJeff Bonwick 	dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
2265ad135b5dSChristopher Siden 	boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
22661d8ccc7bSMark Shellenbaum 	    (wp & WP_SPILL));
2267b24ab676SJeff Bonwick 	enum zio_checksum checksum = os->os_checksum;
2268b24ab676SJeff Bonwick 	enum zio_compress compress = os->os_compress;
2269b24ab676SJeff Bonwick 	enum zio_checksum dedup_checksum = os->os_dedup_checksum;
22707540df39SGeorge Wilson 	boolean_t dedup = B_FALSE;
22717540df39SGeorge Wilson 	boolean_t nopwrite = B_FALSE;
2272b24ab676SJeff Bonwick 	boolean_t dedup_verify = os->os_dedup_verify;
2273eb633035STom Caputi 	boolean_t encrypt = B_FALSE;
2274b24ab676SJeff Bonwick 	int copies = os->os_copies;
2275b24ab676SJeff Bonwick 
2276b24ab676SJeff Bonwick 	/*
227780901aeaSGeorge Wilson 	 * We maintain different write policies for each of the following
227880901aeaSGeorge Wilson 	 * types of data:
227980901aeaSGeorge Wilson 	 *	 1. metadata
228080901aeaSGeorge Wilson 	 *	 2. preallocated blocks (i.e. level-0 blocks of a dump device)
228180901aeaSGeorge Wilson 	 *	 3. all other level 0 blocks
2282b24ab676SJeff Bonwick 	 */
2283b24ab676SJeff Bonwick 	if (ismd) {
2284eb633035STom Caputi 		/*
2285eb633035STom Caputi 		 * XXX -- we should design a compression algorithm
2286eb633035STom Caputi 		 * that specializes in arrays of bps.
2287eb633035STom Caputi 		 */
2288eb633035STom Caputi 		compress = zio_compress_select(os->os_spa,
2289eb633035STom Caputi 		    ZIO_COMPRESS_ON, ZIO_COMPRESS_ON);
229080901aeaSGeorge Wilson 
2291b24ab676SJeff Bonwick 		/*
2292b24ab676SJeff Bonwick 		 * Metadata always gets checksummed.  If the data
2293b24ab676SJeff Bonwick 		 * checksum is multi-bit correctable, and it's not a
2294b24ab676SJeff Bonwick 		 * ZBT-style checksum, then it's suitable for metadata
2295b24ab676SJeff Bonwick 		 * as well.  Otherwise, the metadata checksum defaults
2296b24ab676SJeff Bonwick 		 * to fletcher4.
2297b24ab676SJeff Bonwick 		 */
229845818ee1SMatthew Ahrens 		if (!(zio_checksum_table[checksum].ci_flags &
229945818ee1SMatthew Ahrens 		    ZCHECKSUM_FLAG_METADATA) ||
230045818ee1SMatthew Ahrens 		    (zio_checksum_table[checksum].ci_flags &
230145818ee1SMatthew Ahrens 		    ZCHECKSUM_FLAG_EMBEDDED))
2302b24ab676SJeff Bonwick 			checksum = ZIO_CHECKSUM_FLETCHER_4;
2303edf345e6SMatthew Ahrens 
2304edf345e6SMatthew Ahrens 		if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
2305edf345e6SMatthew Ahrens 		    (os->os_redundant_metadata ==
2306edf345e6SMatthew Ahrens 		    ZFS_REDUNDANT_METADATA_MOST &&
2307edf345e6SMatthew Ahrens 		    (level >= zfs_redundant_metadata_most_ditto_level ||
2308edf345e6SMatthew Ahrens 		    DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))))
2309edf345e6SMatthew Ahrens 			copies++;
231080901aeaSGeorge Wilson 	} else if (wp & WP_NOFILL) {
231180901aeaSGeorge Wilson 		ASSERT(level == 0);
231280901aeaSGeorge Wilson 
2313b24ab676SJeff Bonwick 		/*
231480901aeaSGeorge Wilson 		 * If we're writing preallocated blocks, we aren't actually
231580901aeaSGeorge Wilson 		 * writing them so don't set any policy properties.  These
231680901aeaSGeorge Wilson 		 * blocks are currently only used by an external subsystem
231780901aeaSGeorge Wilson 		 * outside of zfs (i.e. dump) and not written by the zio
231880901aeaSGeorge Wilson 		 * pipeline.
2319b24ab676SJeff Bonwick 		 */
232080901aeaSGeorge Wilson 		compress = ZIO_COMPRESS_OFF;
2321810e43b2SBill Pijewski 		checksum = ZIO_CHECKSUM_NOPARITY;
2322b24ab676SJeff Bonwick 	} else {
2323db1741f5SJustin T. Gibbs 		compress = zio_compress_select(os->os_spa, dn->dn_compress,
2324db1741f5SJustin T. Gibbs 		    compress);
2325b24ab676SJeff Bonwick 
232680901aeaSGeorge Wilson 		checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
232780901aeaSGeorge Wilson 		    zio_checksum_select(dn->dn_checksum, checksum) :
232880901aeaSGeorge Wilson 		    dedup_checksum;
2329b24ab676SJeff Bonwick 
233080901aeaSGeorge Wilson 		/*
233180901aeaSGeorge Wilson 		 * Determine dedup setting.  If we are in dmu_sync(),
233280901aeaSGeorge Wilson 		 * we won't actually dedup now because that's all
233380901aeaSGeorge Wilson 		 * done in syncing context; but we do want to use the
233480901aeaSGeorge Wilson 		 * dedup checkum.  If the checksum is not strong
233580901aeaSGeorge Wilson 		 * enough to ensure unique signatures, force
233680901aeaSGeorge Wilson 		 * dedup_verify.
233780901aeaSGeorge Wilson 		 */
233880901aeaSGeorge Wilson 		if (dedup_checksum != ZIO_CHECKSUM_OFF) {
233980901aeaSGeorge Wilson 			dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
234045818ee1SMatthew Ahrens 			if (!(zio_checksum_table[checksum].ci_flags &
234145818ee1SMatthew Ahrens 			    ZCHECKSUM_FLAG_DEDUP))
234280901aeaSGeorge Wilson 				dedup_verify = B_TRUE;
234380901aeaSGeorge Wilson 		}
2344b24ab676SJeff Bonwick 
234580901aeaSGeorge Wilson 		/*
234645818ee1SMatthew Ahrens 		 * Enable nopwrite if we have secure enough checksum
234745818ee1SMatthew Ahrens 		 * algorithm (see comment in zio_nop_write) and
234845818ee1SMatthew Ahrens 		 * compression is enabled.  We don't enable nopwrite if
234945818ee1SMatthew Ahrens 		 * dedup is enabled as the two features are mutually
235045818ee1SMatthew Ahrens 		 * exclusive.
235180901aeaSGeorge Wilson 		 */
235245818ee1SMatthew Ahrens 		nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags &
235345818ee1SMatthew Ahrens 		    ZCHECKSUM_FLAG_NOPWRITE) &&
235480901aeaSGeorge Wilson 		    compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
2355b24ab676SJeff Bonwick 	}
2356b24ab676SJeff Bonwick 
2357eb633035STom Caputi 	/*
2358eb633035STom Caputi 	 * All objects in an encrypted objset are protected from modification
2359eb633035STom Caputi 	 * via a MAC. Encrypted objects store their IV and salt in the last DVA
2360eb633035STom Caputi 	 * in the bp, so we cannot use all copies. Encrypted objects are also
2361eb633035STom Caputi 	 * not subject to nopwrite since writing the same data will still
2362eb633035STom Caputi 	 * result in a new ciphertext. Only encrypted blocks can be dedup'd
2363eb633035STom Caputi 	 * to avoid ambiguity in the dedup code since the DDT does not store
2364eb633035STom Caputi 	 * object types.
2365eb633035STom Caputi 	 */
2366eb633035STom Caputi 	if (os->os_encrypted && (wp & WP_NOFILL) == 0) {
2367eb633035STom Caputi 		encrypt = B_TRUE;
23685602294fSDan Kimmel 
2369eb633035STom Caputi 		if (DMU_OT_IS_ENCRYPTED(type)) {
2370eb633035STom Caputi 			copies = MIN(copies, SPA_DVAS_PER_BP - 1);
2371eb633035STom Caputi 			nopwrite = B_FALSE;
2372eb633035STom Caputi 		} else {
2373eb633035STom Caputi 			dedup = B_FALSE;
2374eb633035STom Caputi 		}
2375eb633035STom Caputi 
2376eb633035STom Caputi 		if (level <= 0 &&
2377eb633035STom Caputi 		    (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) {
2378eb633035STom Caputi 			compress = ZIO_COMPRESS_EMPTY;
2379eb633035STom Caputi 		}
2380eb633035STom Caputi 	}
2381eb633035STom Caputi 
2382eb633035STom Caputi 	zp->zp_compress = compress;
2383eb633035STom Caputi 	zp->zp_checksum = checksum;
23840a586ceaSMark Shellenbaum 	zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
2385b24ab676SJeff Bonwick 	zp->zp_level = level;
2386edf345e6SMatthew Ahrens 	zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
2387b24ab676SJeff Bonwick 	zp->zp_dedup = dedup;
2388b24ab676SJeff Bonwick 	zp->zp_dedup_verify = dedup && dedup_verify;
238980901aeaSGeorge Wilson 	zp->zp_nopwrite = nopwrite;
2390663207adSDon Brady 	zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ?
2391663207adSDon Brady 	    os->os_zpl_special_smallblock : 0;
2392eb633035STom Caputi 	zp->zp_encrypt = encrypt;
2393eb633035STom Caputi 	zp->zp_byteorder = ZFS_HOST_BYTEORDER;
2394eb633035STom Caputi 	bzero(zp->zp_salt, ZIO_DATA_SALT_LEN);
2395eb633035STom Caputi 	bzero(zp->zp_iv, ZIO_DATA_IV_LEN);
2396eb633035STom Caputi 	bzero(zp->zp_mac, ZIO_DATA_MAC_LEN);
2397b24ab676SJeff Bonwick }
2398b24ab676SJeff Bonwick 
2399fa9e4066Sahrens int
2400fa9e4066Sahrens dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
2401fa9e4066Sahrens {
2402fa9e4066Sahrens 	dnode_t *dn;
24032bcf0248SMax Grossman 	int err;
2404fa9e4066Sahrens 
2405fa9e4066Sahrens 	/*
2406fa9e4066Sahrens 	 * Sync any current changes before
2407fa9e4066Sahrens 	 * we go trundling through the block pointers.
2408fa9e4066Sahrens 	 */
24092bcf0248SMax Grossman 	err = dmu_object_wait_synced(os, object);
24102bcf0248SMax Grossman 	if (err) {
24112bcf0248SMax Grossman 		return (err);
2412fa9e4066Sahrens 	}
24132bcf0248SMax Grossman 
24142bcf0248SMax Grossman 	err = dnode_hold(os, object, FTAG, &dn);
24152bcf0248SMax Grossman 	if (err) {
24162bcf0248SMax Grossman 		return (err);
2417fa9e4066Sahrens 	}
2418fa9e4066Sahrens 
2419cdb0ab79Smaybee 	err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
2420fa9e4066Sahrens 	dnode_rele(dn, FTAG);
2421fa9e4066Sahrens 
2422fa9e4066Sahrens 	return (err);
2423fa9e4066Sahrens }
2424fa9e4066Sahrens 
24252bcf0248SMax Grossman /*
24262bcf0248SMax Grossman  * Given the ZFS object, if it contains any dirty nodes
24272bcf0248SMax Grossman  * this function flushes all dirty blocks to disk. This
24282bcf0248SMax Grossman  * ensures the DMU object info is updated. A more efficient
24292bcf0248SMax Grossman  * future version might just find the TXG with the maximum
24302bcf0248SMax Grossman  * ID and wait for that to be synced.
24312bcf0248SMax Grossman  */
24322bcf0248SMax Grossman int
24339a686fbcSPaul Dagnelie dmu_object_wait_synced(objset_t *os, uint64_t object)
24349a686fbcSPaul Dagnelie {
24352bcf0248SMax Grossman 	dnode_t *dn;
24362bcf0248SMax Grossman 	int error, i;
24372bcf0248SMax Grossman 
24382bcf0248SMax Grossman 	error = dnode_hold(os, object, FTAG, &dn);
24392bcf0248SMax Grossman 	if (error) {
24402bcf0248SMax Grossman 		return (error);
24412bcf0248SMax Grossman 	}
24422bcf0248SMax Grossman 
24432bcf0248SMax Grossman 	for (i = 0; i < TXG_SIZE; i++) {
24442bcf0248SMax Grossman 		if (list_link_active(&dn->dn_dirty_link[i])) {
24452bcf0248SMax Grossman 			break;
24462bcf0248SMax Grossman 		}
24472bcf0248SMax Grossman 	}
24482bcf0248SMax Grossman 	dnode_rele(dn, FTAG);
24492bcf0248SMax Grossman 	if (i != TXG_SIZE) {
24502bcf0248SMax Grossman 		txg_wait_synced(dmu_objset_pool(os), 0);
24512bcf0248SMax Grossman 	}
24522bcf0248SMax Grossman 
24532bcf0248SMax Grossman 	return (0);
24542bcf0248SMax Grossman }
24552bcf0248SMax Grossman 
2456fa9e4066Sahrens void
2457fa9e4066Sahrens dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
2458fa9e4066Sahrens {
2459b24ab676SJeff Bonwick 	dnode_phys_t *dnp;
2460b24ab676SJeff Bonwick 
2461fa9e4066Sahrens 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
2462fa9e4066Sahrens 	mutex_enter(&dn->dn_mtx);
2463fa9e4066Sahrens 
2464b24ab676SJeff Bonwick 	dnp = dn->dn_phys;
2465b24ab676SJeff Bonwick 
2466fa9e4066Sahrens 	doi->doi_data_block_size = dn->dn_datablksz;
2467fa9e4066Sahrens 	doi->doi_metadata_block_size = dn->dn_indblkshift ?
2468fa9e4066Sahrens 	    1ULL << dn->dn_indblkshift : 0;
2469b24ab676SJeff Bonwick 	doi->doi_type = dn->dn_type;
2470b24ab676SJeff Bonwick 	doi->doi_bonus_type = dn->dn_bonustype;
2471b24ab676SJeff Bonwick 	doi->doi_bonus_size = dn->dn_bonuslen;
247254811da5SToomas Soome 	doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT;
2473fa9e4066Sahrens 	doi->doi_indirection = dn->dn_nlevels;
2474fa9e4066Sahrens 	doi->doi_checksum = dn->dn_checksum;
2475fa9e4066Sahrens 	doi->doi_compress = dn->dn_compress;
2476e77d42eaSMatthew Ahrens 	doi->doi_nblkptr = dn->dn_nblkptr;
2477b24ab676SJeff Bonwick 	doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
2478d0475637SMatthew Ahrens 	doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
2479b24ab676SJeff Bonwick 	doi->doi_fill_count = 0;
2480b24ab676SJeff Bonwick 	for (int i = 0; i < dnp->dn_nblkptr; i++)
24815d7b4d43SMatthew Ahrens 		doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
2482fa9e4066Sahrens 
2483fa9e4066Sahrens 	mutex_exit(&dn->dn_mtx);
2484fa9e4066Sahrens 	rw_exit(&dn->dn_struct_rwlock);
2485fa9e4066Sahrens }
2486fa9e4066Sahrens 
2487fa9e4066Sahrens /*
2488fa9e4066Sahrens  * Get information on a DMU object.
2489fa9e4066Sahrens  * If doi is NULL, just indicates whether the object exists.
2490fa9e4066Sahrens  */
2491fa9e4066Sahrens int
2492fa9e4066Sahrens dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
2493fa9e4066Sahrens {
2494ea8dc4b6Seschrock 	dnode_t *dn;
2495503ad85cSMatthew Ahrens 	int err = dnode_hold(os, object, FTAG, &dn);
2496fa9e4066Sahrens 
2497ea8dc4b6Seschrock 	if (err)
2498ea8dc4b6Seschrock 		return (err);
2499fa9e4066Sahrens 
2500fa9e4066Sahrens 	if (doi != NULL)
2501fa9e4066Sahrens 		dmu_object_info_from_dnode(dn, doi);
2502fa9e4066Sahrens 
2503fa9e4066Sahrens 	dnode_rele(dn, FTAG);
2504fa9e4066Sahrens 	return (0);
2505fa9e4066Sahrens }
2506fa9e4066Sahrens 
2507fa9e4066Sahrens /*
2508fa9e4066Sahrens  * As above, but faster; can be used when you have a held dbuf in hand.
2509fa9e4066Sahrens  */
2510fa9e4066Sahrens void
2511744947dcSTom Erickson dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
2512fa9e4066Sahrens {
2513744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2514744947dcSTom Erickson 
2515744947dcSTom Erickson 	DB_DNODE_ENTER(db);
2516744947dcSTom Erickson 	dmu_object_info_from_dnode(DB_DNODE(db), doi);
2517744947dcSTom Erickson 	DB_DNODE_EXIT(db);
2518fa9e4066Sahrens }
2519fa9e4066Sahrens 
2520fa9e4066Sahrens /*
2521fa9e4066Sahrens  * Faster still when you only care about the size.
2522fa9e4066Sahrens  * This is specifically optimized for zfs_getattr().
2523fa9e4066Sahrens  */
2524fa9e4066Sahrens void
2525744947dcSTom Erickson dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
2526744947dcSTom Erickson     u_longlong_t *nblk512)
2527fa9e4066Sahrens {
2528744947dcSTom Erickson 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2529744947dcSTom Erickson 	dnode_t *dn;
2530744947dcSTom Erickson 
2531744947dcSTom Erickson 	DB_DNODE_ENTER(db);
2532744947dcSTom Erickson 	dn = DB_DNODE(db);
2533fa9e4066Sahrens 
2534fa9e4066Sahrens 	*blksize = dn->dn_datablksz;
253554811da5SToomas Soome 	/* add in number of slots used for the dnode itself */
253699653d4eSeschrock 	*nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
253754811da5SToomas Soome 	    SPA_MINBLOCKSHIFT) + dn->dn_num_slots;
253854811da5SToomas Soome 	DB_DNODE_EXIT(db);
253954811da5SToomas Soome }
254054811da5SToomas Soome 
254154811da5SToomas Soome void
254254811da5SToomas Soome dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize)
254354811da5SToomas Soome {
254454811da5SToomas Soome 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
254554811da5SToomas Soome 	dnode_t *dn;
254654811da5SToomas Soome 
254754811da5SToomas Soome 	DB_DNODE_ENTER(db);
254854811da5SToomas Soome 	dn = DB_DNODE(db);
254954811da5SToomas Soome 	*dnsize = dn->dn_num_slots << DNODE_SHIFT;
2550744947dcSTom Erickson 	DB_DNODE_EXIT(db);
2551fa9e4066Sahrens }
2552fa9e4066Sahrens 
2553fa9e4066Sahrens void
2554fa9e4066Sahrens byteswap_uint64_array(void *vbuf, size_t size)
2555fa9e4066Sahrens {
2556fa9e4066Sahrens 	uint64_t *buf = vbuf;
2557fa9e4066Sahrens 	size_t count = size >> 3;
2558fa9e4066Sahrens 	int i;
2559fa9e4066Sahrens 
2560fa9e4066Sahrens 	ASSERT((size & 7) == 0);
2561fa9e4066Sahrens 
2562fa9e4066Sahrens 	for (i = 0; i < count; i++)
2563fa9e4066Sahrens 		buf[i] = BSWAP_64(buf[i]);
2564fa9e4066Sahrens }
2565fa9e4066Sahrens 
2566fa9e4066Sahrens void
2567fa9e4066Sahrens byteswap_uint32_array(void *vbuf, size_t size)
2568fa9e4066Sahrens {
2569fa9e4066Sahrens 	uint32_t *buf = vbuf;
2570fa9e4066Sahrens 	size_t count = size >> 2;
2571fa9e4066Sahrens 	int i;
2572fa9e4066Sahrens 
2573fa9e4066Sahrens 	ASSERT((size & 3) == 0);
2574fa9e4066Sahrens 
2575fa9e4066Sahrens 	for (i = 0; i < count; i++)
2576fa9e4066Sahrens 		buf[i] = BSWAP_32(buf[i]);
2577fa9e4066Sahrens }
2578fa9e4066Sahrens 
2579fa9e4066Sahrens void
2580fa9e4066Sahrens byteswap_uint16_array(void *vbuf, size_t size)
2581fa9e4066Sahrens {
2582fa9e4066Sahrens 	uint16_t *buf = vbuf;
2583fa9e4066Sahrens 	size_t count = size >> 1;
2584fa9e4066Sahrens 	int i;
2585fa9e4066Sahrens 
2586fa9e4066Sahrens 	ASSERT((size & 1) == 0);
2587fa9e4066Sahrens 
2588fa9e4066Sahrens 	for (i = 0; i < count; i++)
2589fa9e4066Sahrens 		buf[i] = BSWAP_16(buf[i]);
2590fa9e4066Sahrens }
2591fa9e4066Sahrens 
2592fa9e4066Sahrens /* ARGSUSED */
2593fa9e4066Sahrens void
2594fa9e4066Sahrens byteswap_uint8_array(void *vbuf, size_t size)
2595fa9e4066Sahrens {
2596fa9e4066Sahrens }
2597fa9e4066Sahrens 
2598fa9e4066Sahrens void
2599fa9e4066Sahrens dmu_init(void)
2600fa9e4066Sahrens {
2601770499e1SDan Kimmel 	abd_init();
26023f9d6ad7SLin Ling 	zfs_dbgmsg_init();
2603744947dcSTom Erickson 	sa_cache_init();
2604744947dcSTom Erickson 	xuio_stat_init();
2605744947dcSTom Erickson 	dmu_objset_init();
2606fa9e4066Sahrens 	dnode_init();
26077cbf8b43SRich Morris 	zfetch_init();
2608fa94a07fSbrendan 	l2arc_init();
2609ce636f8bSMatthew Ahrens 	arc_init();
2610dcbf3bd6SGeorge Wilson 	dbuf_init();
2611fa9e4066Sahrens }
2612fa9e4066Sahrens 
2613fa9e4066Sahrens void
2614fa9e4066Sahrens dmu_fini(void)
2615fa9e4066Sahrens {
26163e30c24aSWill Andrews 	arc_fini(); /* arc depends on l2arc, so arc must go first */
2617ce636f8bSMatthew Ahrens 	l2arc_fini();
26187cbf8b43SRich Morris 	zfetch_fini();
2619fa9e4066Sahrens 	dbuf_fini();
2620744947dcSTom Erickson 	dnode_fini();
2621744947dcSTom Erickson 	dmu_objset_fini();
2622c242f9a0Schunli zhang - Sun Microsystems - Irvine United States 	xuio_stat_fini();
26230a586ceaSMark Shellenbaum 	sa_cache_fini();
26243f9d6ad7SLin Ling 	zfs_dbgmsg_fini();
2625770499e1SDan Kimmel 	abd_fini();
2626fa9e4066Sahrens }
2627