Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/dmu.c @ 11670:1d964fb5d948
6595532 ZIL is too talkative
6388458 zil need not inflate blocksize that much
6738159 slog can probably pack 2X more data per lwb
6897432 ziltest_noslog fails after dedup putback
6911570 Shouldn't start an ereport on speculative read failure
6920442 ZIL: Should use the stronger checksum fletcher4
author | Neil Perrin <Neil.Perrin@Sun.COM> |
---|---|
date | Wed, 17 Feb 2010 15:13:37 -0700 |
parents | 10d35fc3d7fd |
children | 538c866aaac6 |
rev | line source |
---|---|
789 | 1 /* |
2 * CDDL HEADER START | |
3 * | |
4 * The contents of this file are subject to the terms of the | |
1544 | 5 * Common Development and Distribution License (the "License"). |
6 * You may not use this file except in compliance with the License. | |
789 | 7 * |
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 * or http://www.opensolaris.org/os/licensing. | |
10 * See the License for the specific language governing permissions | |
11 * and limitations under the License. | |
12 * | |
13 * When distributing Covered Code, include this CDDL HEADER in each | |
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 * If applicable, add the following below this CDDL HEADER, with the | |
16 * fields enclosed by brackets "[]" replaced with your own identifying | |
17 * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 * | |
19 * CDDL HEADER END | |
20 */ | |
21 /* | |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved. |
789 | 23 * Use is subject to license terms. |
24 */ | |
25 | |
26 #include <sys/dmu.h> | |
27 #include <sys/dmu_impl.h> | |
28 #include <sys/dmu_tx.h> | |
29 #include <sys/dbuf.h> | |
30 #include <sys/dnode.h> | |
31 #include <sys/zfs_context.h> | |
32 #include <sys/dmu_objset.h> | |
33 #include <sys/dmu_traverse.h> | |
34 #include <sys/dsl_dataset.h> | |
35 #include <sys/dsl_dir.h> | |
36 #include <sys/dsl_pool.h> | |
2199 | 37 #include <sys/dsl_synctask.h> |
2885 | 38 #include <sys/dsl_prop.h> |
789 | 39 #include <sys/dmu_zfetch.h> |
40 #include <sys/zfs_ioctl.h> | |
41 #include <sys/zap.h> | |
1544 | 42 #include <sys/zio_checksum.h> |
2688
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
43 #ifdef _KERNEL |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
44 #include <sys/vmsystm.h> |
7315
cdba25672122
6735592 "zio" vmem arenas are confusing, not reported separately
Jonathan W Adams <Jonathan.Adams@Sun.COM>
parents:
7237
diff
changeset
|
45 #include <sys/zfs_znode.h> |
2688
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
46 #endif |
789 | 47 |
48 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { | |
49 { byteswap_uint8_array, TRUE, "unallocated" }, | |
50 { zap_byteswap, TRUE, "object directory" }, | |
51 { byteswap_uint64_array, TRUE, "object array" }, | |
52 { byteswap_uint8_array, TRUE, "packed nvlist" }, | |
53 { byteswap_uint64_array, TRUE, "packed nvlist size" }, | |
54 { byteswap_uint64_array, TRUE, "bplist" }, | |
55 { byteswap_uint64_array, TRUE, "bplist header" }, | |
56 { byteswap_uint64_array, TRUE, "SPA space map header" }, | |
57 { byteswap_uint64_array, TRUE, "SPA space map" }, | |
58 { byteswap_uint64_array, TRUE, "ZIL intent log" }, | |
59 { dnode_buf_byteswap, TRUE, "DMU dnode" }, | |
60 { dmu_objset_byteswap, TRUE, "DMU objset" }, | |
61 { byteswap_uint64_array, TRUE, "DSL directory" }, | |
62 { zap_byteswap, TRUE, "DSL directory child map"}, | |
63 { zap_byteswap, TRUE, "DSL dataset snap map" }, | |
64 { zap_byteswap, TRUE, "DSL props" }, | |
65 { byteswap_uint64_array, TRUE, "DSL dataset" }, | |
66 { zfs_znode_byteswap, TRUE, "ZFS znode" }, | |
5331 | 67 { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" }, |
789 | 68 { byteswap_uint8_array, FALSE, "ZFS plain file" }, |
69 { zap_byteswap, TRUE, "ZFS directory" }, | |
70 { zap_byteswap, TRUE, "ZFS master node" }, | |
71 { zap_byteswap, TRUE, "ZFS delete queue" }, | |
72 { byteswap_uint8_array, FALSE, "zvol object" }, | |
73 { zap_byteswap, TRUE, "zvol prop" }, | |
74 { byteswap_uint8_array, FALSE, "other uint8[]" }, | |
75 { byteswap_uint64_array, FALSE, "other uint64[]" }, | |
76 { zap_byteswap, TRUE, "other ZAP" }, | |
1544 | 77 { zap_byteswap, TRUE, "persistent error log" }, |
2926 | 78 { byteswap_uint8_array, TRUE, "SPA history" }, |
79 { byteswap_uint64_array, TRUE, "SPA history offsets" }, | |
4543 | 80 { zap_byteswap, TRUE, "Pool properties" }, |
5331 | 81 { zap_byteswap, TRUE, "DSL permissions" }, |
82 { zfs_acl_byteswap, TRUE, "ZFS ACL" }, | |
83 { byteswap_uint8_array, TRUE, "ZFS SYSACL" }, | |
84 { byteswap_uint8_array, TRUE, "FUID table" }, | |
5339 | 85 { byteswap_uint64_array, TRUE, "FUID table size" }, |
7046
361307ae060d
6343667 scrub/resilver has to start over when a snapshot is taken
ahrens
parents:
6992
diff
changeset
|
86 { zap_byteswap, TRUE, "DSL dataset next clones"}, |
361307ae060d
6343667 scrub/resilver has to start over when a snapshot is taken
ahrens
parents:
6992
diff
changeset
|
87 { zap_byteswap, TRUE, "scrub work queue" }, |
9396
f41cf682d0d3
PSARC/2009/204 ZFS user/group quotas & space accounting
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
7872
diff
changeset
|
88 { zap_byteswap, TRUE, "ZFS user/group used" }, |
f41cf682d0d3
PSARC/2009/204 ZFS user/group quotas & space accounting
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
7872
diff
changeset
|
89 { zap_byteswap, TRUE, "ZFS user/group quota" }, |
10242
c40d075fbca6
PSARC/2009/297 zfs snapshot holds
Chris Kirby <chris.kirby@sun.com>
parents:
10209
diff
changeset
|
90 { zap_byteswap, TRUE, "snapshot refcount tags"}, |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
91 { zap_byteswap, TRUE, "DDT ZAP algorithm" }, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
92 { zap_byteswap, TRUE, "DDT statistics" }, |
789 | 93 }; |
94 | |
95 int | |
1544 | 96 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, |
97 void *tag, dmu_buf_t **dbp) | |
789 | 98 { |
99 dnode_t *dn; | |
100 uint64_t blkid; | |
101 dmu_buf_impl_t *db; | |
1544 | 102 int err; |
789 | 103 |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
104 err = dnode_hold(os, object, FTAG, &dn); |
1544 | 105 if (err) |
106 return (err); | |
789 | 107 blkid = dbuf_whichblock(dn, offset); |
108 rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
1544 | 109 db = dbuf_hold(dn, blkid, tag); |
789 | 110 rw_exit(&dn->dn_struct_rwlock); |
1544 | 111 if (db == NULL) { |
112 err = EIO; | |
113 } else { | |
114 err = dbuf_read(db, NULL, DB_RF_CANFAIL); | |
115 if (err) { | |
116 dbuf_rele(db, tag); | |
117 db = NULL; | |
118 } | |
119 } | |
120 | |
789 | 121 dnode_rele(dn, FTAG); |
1544 | 122 *dbp = &db->db; |
123 return (err); | |
789 | 124 } |
125 | |
126 int | |
127 dmu_bonus_max(void) | |
128 { | |
129 return (DN_MAX_BONUSLEN); | |
130 } | |
131 | |
4944
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
132 int |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
133 dmu_set_bonus(dmu_buf_t *db, int newsize, dmu_tx_t *tx) |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
134 { |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
135 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
136 |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
137 if (dn->dn_bonus != (dmu_buf_impl_t *)db) |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
138 return (EINVAL); |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
139 if (newsize < 0 || newsize > db->db_size) |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
140 return (EINVAL); |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
141 dnode_setbonuslen(dn, newsize, tx); |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
142 return (0); |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
143 } |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
144 |
789 | 145 /* |
1544 | 146 * returns ENOENT, EIO, or 0. |
789 | 147 */ |
1544 | 148 int |
149 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) | |
789 | 150 { |
1544 | 151 dnode_t *dn; |
789 | 152 dmu_buf_impl_t *db; |
4944
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
153 int error; |
789 | 154 |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
155 error = dnode_hold(os, object, FTAG, &dn); |
4944
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
156 if (error) |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
157 return (error); |
789 | 158 |
1544 | 159 rw_enter(&dn->dn_struct_rwlock, RW_READER); |
160 if (dn->dn_bonus == NULL) { | |
789 | 161 rw_exit(&dn->dn_struct_rwlock); |
1544 | 162 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
163 if (dn->dn_bonus == NULL) | |
4944
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
164 dbuf_create_bonus(dn); |
789 | 165 } |
1544 | 166 db = dn->dn_bonus; |
167 rw_exit(&dn->dn_struct_rwlock); | |
4944
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
168 |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
169 /* as long as the bonus buf is held, the dnode will be held */ |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
170 if (refcount_add(&db->db_holds, tag) == 1) |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
171 VERIFY(dnode_add_ref(dn, db)); |
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
172 |
789 | 173 dnode_rele(dn, FTAG); |
1544 | 174 |
175 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED)); | |
176 | |
177 *dbp = &db->db; | |
178 return (0); | |
789 | 179 } |
180 | |
2391 | 181 /* |
182 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
183 * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
184 * and can induce severe lock contention when writing to several files | |
185 * whose dnodes are in the same block. | |
186 */ | |
187 static int | |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
188 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
189 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) |
789 | 190 { |
7468
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
191 dsl_pool_t *dp = NULL; |
789 | 192 dmu_buf_t **dbp; |
193 uint64_t blkid, nblks, i; | |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
194 uint32_t dbuf_flags; |
1544 | 195 int err; |
196 zio_t *zio; | |
7468
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
197 hrtime_t start; |
1544 | 198 |
199 ASSERT(length <= DMU_MAX_ACCESS); | |
789 | 200 |
10209
91f47f0e7728
6830541 zfs_get_data trips on a verify
Mark J Musante <Mark.Musante@Sun.COM>
parents:
9951
diff
changeset
|
201 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
202 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) |
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
203 dbuf_flags |= DB_RF_NOPREFETCH; |
1544 | 204 |
789 | 205 rw_enter(&dn->dn_struct_rwlock, RW_READER); |
206 if (dn->dn_datablkshift) { | |
207 int blkshift = dn->dn_datablkshift; | |
208 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - | |
3912 | 209 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; |
789 | 210 } else { |
3713 | 211 if (offset + length > dn->dn_datablksz) { |
212 zfs_panic_recover("zfs: accessing past end of object " | |
213 "%llx/%llx (size=%u access=%llu+%llu)", | |
214 (longlong_t)dn->dn_objset-> | |
215 os_dsl_dataset->ds_object, | |
216 (longlong_t)dn->dn_object, dn->dn_datablksz, | |
217 (longlong_t)offset, (longlong_t)length); | |
10209
91f47f0e7728
6830541 zfs_get_data trips on a verify
Mark J Musante <Mark.Musante@Sun.COM>
parents:
9951
diff
changeset
|
218 rw_exit(&dn->dn_struct_rwlock); |
3713 | 219 return (EIO); |
220 } | |
789 | 221 nblks = 1; |
222 } | |
1544 | 223 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); |
789 | 224 |
7468
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
225 if (dn->dn_objset->os_dsl_dataset) |
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
226 dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool; |
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
227 if (dp && dsl_pool_sync_context(dp)) |
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
228 start = gethrtime(); |
7754
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
229 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
789 | 230 blkid = dbuf_whichblock(dn, offset); |
231 for (i = 0; i < nblks; i++) { | |
1544 | 232 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); |
233 if (db == NULL) { | |
234 rw_exit(&dn->dn_struct_rwlock); | |
235 dmu_buf_rele_array(dbp, nblks, tag); | |
236 zio_nowait(zio); | |
237 return (EIO); | |
238 } | |
239 /* initiate async i/o */ | |
2391 | 240 if (read) { |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
241 (void) dbuf_read(db, zio, dbuf_flags); |
1544 | 242 } |
243 dbp[i] = &db->db; | |
789 | 244 } |
245 rw_exit(&dn->dn_struct_rwlock); | |
246 | |
1544 | 247 /* wait for async i/o */ |
248 err = zio_wait(zio); | |
7468
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
249 /* track read overhead when we are in sync context */ |
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
250 if (dp && dsl_pool_sync_context(dp)) |
23915842aa09
6687412 zfs stuck issuing small txg
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7385
diff
changeset
|
251 dp->dp_read_overhead += gethrtime() - start; |
1544 | 252 if (err) { |
253 dmu_buf_rele_array(dbp, nblks, tag); | |
254 return (err); | |
789 | 255 } |
256 | |
1544 | 257 /* wait for other io to complete */ |
258 if (read) { | |
259 for (i = 0; i < nblks; i++) { | |
260 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
261 mutex_enter(&db->db_mtx); | |
262 while (db->db_state == DB_READ || | |
263 db->db_state == DB_FILL) | |
264 cv_wait(&db->db_changed, &db->db_mtx); | |
265 if (db->db_state == DB_UNCACHED) | |
266 err = EIO; | |
267 mutex_exit(&db->db_mtx); | |
268 if (err) { | |
269 dmu_buf_rele_array(dbp, nblks, tag); | |
270 return (err); | |
271 } | |
272 } | |
273 } | |
789 | 274 |
1544 | 275 *numbufsp = nblks; |
276 *dbpp = dbp; | |
277 return (0); | |
789 | 278 } |
279 | |
2885 | 280 static int |
2391 | 281 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, |
282 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
283 { | |
284 dnode_t *dn; | |
285 int err; | |
286 | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
287 err = dnode_hold(os, object, FTAG, &dn); |
2391 | 288 if (err) |
289 return (err); | |
290 | |
291 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
292 numbufsp, dbpp, DMU_READ_PREFETCH); |
2391 | 293 |
294 dnode_rele(dn, FTAG); | |
295 | |
296 return (err); | |
297 } | |
298 | |
299 int | |
300 dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, | |
301 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
302 { | |
303 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; | |
304 int err; | |
305 | |
306 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
307 numbufsp, dbpp, DMU_READ_PREFETCH); |
2391 | 308 |
309 return (err); | |
310 } | |
311 | |
789 | 312 void |
1544 | 313 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) |
789 | 314 { |
315 int i; | |
316 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
317 | |
318 if (numbufs == 0) | |
319 return; | |
320 | |
1544 | 321 for (i = 0; i < numbufs; i++) { |
322 if (dbp[i]) | |
323 dbuf_rele(dbp[i], tag); | |
324 } | |
789 | 325 |
326 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
327 } | |
328 | |
329 void | |
330 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) | |
331 { | |
332 dnode_t *dn; | |
333 uint64_t blkid; | |
1544 | 334 int nblks, i, err; |
789 | 335 |
2986 | 336 if (zfs_prefetch_disable) |
337 return; | |
338 | |
789 | 339 if (len == 0) { /* they're interested in the bonus buffer */ |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
340 dn = os->os_meta_dnode; |
789 | 341 |
342 if (object == 0 || object >= DN_MAX_OBJECT) | |
343 return; | |
344 | |
345 rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
346 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); | |
347 dbuf_prefetch(dn, blkid); | |
348 rw_exit(&dn->dn_struct_rwlock); | |
349 return; | |
350 } | |
351 | |
352 /* | |
353 * XXX - Note, if the dnode for the requested object is not | |
354 * already cached, we will do a *synchronous* read in the | |
355 * dnode_hold() call. The same is true for any indirects. | |
356 */ | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
357 err = dnode_hold(os, object, FTAG, &dn); |
1544 | 358 if (err != 0) |
789 | 359 return; |
360 | |
361 rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
362 if (dn->dn_datablkshift) { | |
363 int blkshift = dn->dn_datablkshift; | |
364 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) - | |
3912 | 365 P2ALIGN(offset, 1<<blkshift)) >> blkshift; |
789 | 366 } else { |
367 nblks = (offset < dn->dn_datablksz); | |
368 } | |
369 | |
370 if (nblks != 0) { | |
371 blkid = dbuf_whichblock(dn, offset); | |
372 for (i = 0; i < nblks; i++) | |
373 dbuf_prefetch(dn, blkid+i); | |
374 } | |
375 | |
376 rw_exit(&dn->dn_struct_rwlock); | |
377 | |
378 dnode_rele(dn, FTAG); | |
379 } | |
380 | |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
381 /* |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
382 * Get the next "chunk" of file data to free. We traverse the file from |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
383 * the end so that the file gets shorter over time (if we crashes in the |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
384 * middle, this will leave us in a better state). We find allocated file |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
385 * data by simply searching the allocated level 1 indirects. |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
386 */ |
6992 | 387 static int |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
388 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit) |
6992 | 389 { |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
390 uint64_t len = *start - limit; |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
391 uint64_t blkcnt = 0; |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
392 uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1)); |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
393 uint64_t iblkrange = |
7385
f69ff8507427
6727817 ZFS assertion failure (zero length arc_buf_alloc) during NFS I/O
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7315
diff
changeset
|
394 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
6992 | 395 |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
396 ASSERT(limit <= *start); |
6992 | 397 |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
398 if (len <= iblkrange * maxblks) { |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
399 *start = limit; |
6992 | 400 return (0); |
401 } | |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
402 ASSERT(ISP2(iblkrange)); |
7385
f69ff8507427
6727817 ZFS assertion failure (zero length arc_buf_alloc) during NFS I/O
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7315
diff
changeset
|
403 |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
404 while (*start > limit && blkcnt < maxblks) { |
7385
f69ff8507427
6727817 ZFS assertion failure (zero length arc_buf_alloc) during NFS I/O
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7315
diff
changeset
|
405 int err; |
6992 | 406 |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
407 /* find next allocated L1 indirect */ |
6992 | 408 err = dnode_next_offset(dn, |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
409 DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
6992 | 410 |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
411 /* if there are no more, then we are done */ |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
412 if (err == ESRCH) { |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
413 *start = limit; |
6992 | 414 return (0); |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
415 } else if (err) { |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
416 return (err); |
6992 | 417 } |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
418 blkcnt += 1; |
6992 | 419 |
9950
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
420 /* reset offset to end of "next" block back */ |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
421 *start = P2ALIGN(*start, iblkrange); |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
422 if (*start <= limit) |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
423 *start = limit; |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
424 else |
78fc41aa9bc5
6792701 Removing large holey file does not free space
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
9512
diff
changeset
|
425 *start -= 1; |
6992 | 426 } |
427 return (0); | |
428 } | |
429 | |
430 static int | |
431 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
432 uint64_t length, boolean_t free_dnode) | |
433 { | |
434 dmu_tx_t *tx; | |
435 uint64_t object_size, start, end, len; | |
436 boolean_t trunc = (length == DMU_OBJECT_END); | |
437 int align, err; | |
438 | |
439 align = 1 << dn->dn_datablkshift; | |
440 ASSERT(align > 0); | |
441 object_size = align == 1 ? dn->dn_datablksz : | |
442 (dn->dn_maxblkid + 1) << dn->dn_datablkshift; | |
443 | |
9396
f41cf682d0d3
PSARC/2009/204 ZFS user/group quotas & space accounting
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
7872
diff
changeset
|
444 end = offset + length; |
f41cf682d0d3
PSARC/2009/204 ZFS user/group quotas & space accounting
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
7872
diff
changeset
|
445 if (trunc || end > object_size) |
6992 | 446 end = object_size; |
447 if (end <= offset) | |
448 return (0); | |
449 length = end - offset; | |
450 | |
451 while (length) { | |
452 start = end; | |
9396
f41cf682d0d3
PSARC/2009/204 ZFS user/group quotas & space accounting
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
7872
diff
changeset
|
453 /* assert(offset <= start) */ |
6992 | 454 err = get_next_chunk(dn, &start, offset); |
455 if (err) | |
456 return (err); | |
457 len = trunc ? DMU_OBJECT_END : end - start; | |
458 | |
459 tx = dmu_tx_create(os); | |
460 dmu_tx_hold_free(tx, dn->dn_object, start, len); | |
461 err = dmu_tx_assign(tx, TXG_WAIT); | |
462 if (err) { | |
463 dmu_tx_abort(tx); | |
464 return (err); | |
465 } | |
466 | |
467 dnode_free_range(dn, start, trunc ? -1 : len, tx); | |
468 | |
7385
f69ff8507427
6727817 ZFS assertion failure (zero length arc_buf_alloc) during NFS I/O
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7315
diff
changeset
|
469 if (start == 0 && free_dnode) { |
f69ff8507427
6727817 ZFS assertion failure (zero length arc_buf_alloc) during NFS I/O
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7315
diff
changeset
|
470 ASSERT(trunc); |
6992 | 471 dnode_free(dn, tx); |
7385
f69ff8507427
6727817 ZFS assertion failure (zero length arc_buf_alloc) during NFS I/O
Mark Maybee <Mark.Maybee@Sun.COM>
parents:
7315
diff
changeset
|
472 } |
6992 | 473 |
474 length -= end - start; | |
475 | |
476 dmu_tx_commit(tx); | |
477 end = start; | |
478 } | |
479 return (0); | |
480 } | |
481 | |
482 int | |
483 dmu_free_long_range(objset_t *os, uint64_t object, | |
484 uint64_t offset, uint64_t length) | |
485 { | |
486 dnode_t *dn; | |
487 int err; | |
488 | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
489 err = dnode_hold(os, object, FTAG, &dn); |
6992 | 490 if (err != 0) |
491 return (err); | |
492 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE); | |
493 dnode_rele(dn, FTAG); | |
494 return (err); | |
495 } | |
496 | |
497 int | |
498 dmu_free_object(objset_t *os, uint64_t object) | |
499 { | |
500 dnode_t *dn; | |
501 dmu_tx_t *tx; | |
502 int err; | |
503 | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
504 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, |
6992 | 505 FTAG, &dn); |
506 if (err != 0) | |
507 return (err); | |
508 if (dn->dn_nlevels == 1) { | |
509 tx = dmu_tx_create(os); | |
510 dmu_tx_hold_bonus(tx, object); | |
511 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END); | |
512 err = dmu_tx_assign(tx, TXG_WAIT); | |
513 if (err == 0) { | |
514 dnode_free_range(dn, 0, DMU_OBJECT_END, tx); | |
515 dnode_free(dn, tx); | |
516 dmu_tx_commit(tx); | |
517 } else { | |
518 dmu_tx_abort(tx); | |
519 } | |
520 } else { | |
521 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE); | |
522 } | |
523 dnode_rele(dn, FTAG); | |
524 return (err); | |
525 } | |
526 | |
1544 | 527 int |
789 | 528 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, |
529 uint64_t size, dmu_tx_t *tx) | |
530 { | |
1544 | 531 dnode_t *dn; |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
532 int err = dnode_hold(os, object, FTAG, &dn); |
1544 | 533 if (err) |
534 return (err); | |
789 | 535 ASSERT(offset < UINT64_MAX); |
536 ASSERT(size == -1ULL || size <= UINT64_MAX - offset); | |
537 dnode_free_range(dn, offset, size, tx); | |
538 dnode_rele(dn, FTAG); | |
1544 | 539 return (0); |
789 | 540 } |
541 | |
1544 | 542 int |
543 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
544 void *buf, uint32_t flags) |
789 | 545 { |
546 dnode_t *dn; | |
547 dmu_buf_t **dbp; | |
10209
91f47f0e7728
6830541 zfs_get_data trips on a verify
Mark J Musante <Mark.Musante@Sun.COM>
parents:
9951
diff
changeset
|
548 int numbufs, err; |
789 | 549 |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
550 err = dnode_hold(os, object, FTAG, &dn); |
1544 | 551 if (err) |
552 return (err); | |
3638
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
553 |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
554 /* |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
555 * Deal with odd block sizes, where there can't be data past the first |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
556 * block. If we ever do the tail block optimization, we will need to |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
557 * handle that here as well. |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
558 */ |
10209
91f47f0e7728
6830541 zfs_get_data trips on a verify
Mark J Musante <Mark.Musante@Sun.COM>
parents:
9951
diff
changeset
|
559 if (dn->dn_maxblkid == 0) { |
789 | 560 int newsz = offset > dn->dn_datablksz ? 0 : |
561 MIN(size, dn->dn_datablksz - offset); | |
562 bzero((char *)buf + newsz, size - newsz); | |
563 size = newsz; | |
564 } | |
565 | |
566 while (size > 0) { | |
567 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
10209
91f47f0e7728
6830541 zfs_get_data trips on a verify
Mark J Musante <Mark.Musante@Sun.COM>
parents:
9951
diff
changeset
|
568 int i; |
789 | 569 |
570 /* | |
571 * NB: we could do this block-at-a-time, but it's nice | |
572 * to be reading in parallel. | |
573 */ | |
2885 | 574 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, |
9512
64cafcbcc337
6801810 Commit of aligned streaming rewrites to ZIL device causes unwanted disk reads
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
9412
diff
changeset
|
575 TRUE, FTAG, &numbufs, &dbp, flags); |
1544 | 576 if (err) |
4944
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
577 break; |
789 | 578 |
579 for (i = 0; i < numbufs; i++) { | |
580 int tocpy; | |
581 int bufoff; | |
582 dmu_buf_t *db = dbp[i]; | |
583 | |
584 ASSERT(size > 0); | |
585 | |
586 bufoff = offset - db->db_offset; | |
587 tocpy = (int)MIN(db->db_size - bufoff, size); | |
588 | |
589 bcopy((char *)db->db_data + bufoff, buf, tocpy); | |
590 | |
591 offset += tocpy; | |
592 size -= tocpy; | |
593 buf = (char *)buf + tocpy; | |
594 } | |
1544 | 595 dmu_buf_rele_array(dbp, numbufs, FTAG); |
789 | 596 } |
2885 | 597 dnode_rele(dn, FTAG); |
4944
96d96f8de974
6569719 panic dangling dbufs (dn=ffffffff28814d30, dbuf=ffffffff20756008)
maybee
parents:
4543
diff
changeset
|
598 return (err); |
789 | 599 } |
600 | |
601 void | |
602 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
603 const void *buf, dmu_tx_t *tx) | |
604 { | |
605 dmu_buf_t **dbp; | |
606 int numbufs, i; | |
607 | |
2391 | 608 if (size == 0) |
609 return; | |
610 | |
1544 | 611 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, |
612 FALSE, FTAG, &numbufs, &dbp)); | |
789 | 613 |
614 for (i = 0; i < numbufs; i++) { | |
615 int tocpy; | |
616 int bufoff; | |
617 dmu_buf_t *db = dbp[i]; | |
618 | |
619 ASSERT(size > 0); | |
620 | |
621 bufoff = offset - db->db_offset; | |
622 tocpy = (int)MIN(db->db_size - bufoff, size); | |
623 | |
624 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
625 | |
626 if (tocpy == db->db_size) | |
627 dmu_buf_will_fill(db, tx); | |
628 else | |
629 dmu_buf_will_dirty(db, tx); | |
630 | |
631 bcopy(buf, (char *)db->db_data + bufoff, tocpy); | |
632 | |
633 if (tocpy == db->db_size) | |
634 dmu_buf_fill_done(db, tx); | |
635 | |
636 offset += tocpy; | |
637 size -= tocpy; | |
638 buf = (char *)buf + tocpy; | |
639 } | |
1544 | 640 dmu_buf_rele_array(dbp, numbufs, FTAG); |
789 | 641 } |
642 | |
7872
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
643 void |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
644 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
645 dmu_tx_t *tx) |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
646 { |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
647 dmu_buf_t **dbp; |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
648 int numbufs, i; |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
649 |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
650 if (size == 0) |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
651 return; |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
652 |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
653 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
654 FALSE, FTAG, &numbufs, &dbp)); |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
655 |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
656 for (i = 0; i < numbufs; i++) { |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
657 dmu_buf_t *db = dbp[i]; |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
658 |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
659 dmu_buf_will_not_fill(db, tx); |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
660 } |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
661 dmu_buf_rele_array(dbp, numbufs, FTAG); |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
662 } |
40a9434212f6
6646775 Speed up the dumpifying process for zvols
Tim Haley <Tim.Haley@Sun.COM>
parents:
7754
diff
changeset
|
663 |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
664 /* |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
665 * DMU support for xuio |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
666 */ |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
667 kstat_t *xuio_ksp = NULL; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
668 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
669 int |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
670 dmu_xuio_init(xuio_t *xuio, int nblk) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
671 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
672 dmu_xuio_t *priv; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
673 uio_t *uio = &xuio->xu_uio; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
674 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
675 uio->uio_iovcnt = nblk; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
676 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
677 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
678 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
679 priv->cnt = nblk; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
680 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
681 priv->iovp = uio->uio_iov; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
682 XUIO_XUZC_PRIV(xuio) = priv; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
683 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
684 if (XUIO_XUZC_RW(xuio) == UIO_READ) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
685 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
686 else |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
687 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
688 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
689 return (0); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
690 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
691 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
692 void |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
693 dmu_xuio_fini(xuio_t *xuio) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
694 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
695 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
696 int nblk = priv->cnt; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
697 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
698 kmem_free(priv->iovp, nblk * sizeof (iovec_t)); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
699 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
700 kmem_free(priv, sizeof (dmu_xuio_t)); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
701 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
702 if (XUIO_XUZC_RW(xuio) == UIO_READ) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
703 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
704 else |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
705 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
706 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
707 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
708 /* |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
709 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
710 * and increase priv->next by 1. |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
711 */ |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
712 int |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
713 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
714 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
715 struct iovec *iov; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
716 uio_t *uio = &xuio->xu_uio; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
717 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
718 int i = priv->next++; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
719 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
720 ASSERT(i < priv->cnt); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
721 ASSERT(off + n <= arc_buf_size(abuf)); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
722 iov = uio->uio_iov + i; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
723 iov->iov_base = (char *)abuf->b_data + off; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
724 iov->iov_len = n; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
725 priv->bufs[i] = abuf; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
726 return (0); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
727 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
728 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
729 int |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
730 dmu_xuio_cnt(xuio_t *xuio) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
731 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
732 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
733 return (priv->cnt); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
734 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
735 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
736 arc_buf_t * |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
737 dmu_xuio_arcbuf(xuio_t *xuio, int i) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
738 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
739 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
740 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
741 ASSERT(i < priv->cnt); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
742 return (priv->bufs[i]); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
743 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
744 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
745 void |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
746 dmu_xuio_clear(xuio_t *xuio, int i) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
747 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
748 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
749 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
750 ASSERT(i < priv->cnt); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
751 priv->bufs[i] = NULL; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
752 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
753 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
754 static void |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
755 xuio_stat_init(void) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
756 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
757 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
758 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
759 KSTAT_FLAG_VIRTUAL); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
760 if (xuio_ksp != NULL) { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
761 xuio_ksp->ks_data = &xuio_stats; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
762 kstat_install(xuio_ksp); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
763 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
764 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
765 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
766 static void |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
767 xuio_stat_fini(void) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
768 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
769 if (xuio_ksp != NULL) { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
770 kstat_delete(xuio_ksp); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
771 xuio_ksp = NULL; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
772 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
773 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
774 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
775 void |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
776 xuio_stat_wbuf_copied() |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
777 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
778 XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
779 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
780 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
781 void |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
782 xuio_stat_wbuf_nocopy() |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
783 { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
784 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
785 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
786 |
789 | 787 #ifdef _KERNEL |
788 int | |
3638
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
789 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
790 { |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
791 dmu_buf_t **dbp; |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
792 int numbufs, i, err; |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
793 xuio_t *xuio = NULL; |
3638
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
794 |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
795 /* |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
796 * NB: we could do this block-at-a-time, but it's nice |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
797 * to be reading in parallel. |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
798 */ |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
799 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
800 &numbufs, &dbp); |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
801 if (err) |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
802 return (err); |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
803 |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
804 if (uio->uio_extflg == UIO_XUIO) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
805 xuio = (xuio_t *)uio; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
806 |
3638
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
807 for (i = 0; i < numbufs; i++) { |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
808 int tocpy; |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
809 int bufoff; |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
810 dmu_buf_t *db = dbp[i]; |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
811 |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
812 ASSERT(size > 0); |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
813 |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
814 bufoff = uio->uio_loffset - db->db_offset; |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
815 tocpy = (int)MIN(db->db_size - bufoff, size); |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
816 |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
817 if (xuio) { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
818 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
819 arc_buf_t *dbuf_abuf = dbi->db_buf; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
820 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
821 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
822 if (!err) { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
823 uio->uio_resid -= tocpy; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
824 uio->uio_loffset += tocpy; |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
825 } |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
826 |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
827 if (abuf == dbuf_abuf) |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
828 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
829 else |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
830 XUIOSTAT_BUMP(xuiostat_rbuf_copied); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
831 } else { |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
832 err = uiomove((char *)db->db_data + bufoff, tocpy, |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
833 UIO_READ, uio); |
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
834 } |
3638
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
835 if (err) |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
836 break; |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
837 |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
838 size -= tocpy; |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
839 } |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
840 dmu_buf_rele_array(dbp, numbufs, FTAG); |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
841 |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
842 return (err); |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
843 } |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
844 |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
845 int |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
846 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, |
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
847 dmu_tx_t *tx) |
789 | 848 { |
849 dmu_buf_t **dbp; | |
850 int numbufs, i; | |
851 int err = 0; | |
852 | |
2391 | 853 if (size == 0) |
854 return (0); | |
855 | |
3638
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
856 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, |
1544 | 857 FALSE, FTAG, &numbufs, &dbp); |
858 if (err) | |
859 return (err); | |
789 | 860 |
861 for (i = 0; i < numbufs; i++) { | |
862 int tocpy; | |
863 int bufoff; | |
864 dmu_buf_t *db = dbp[i]; | |
865 | |
866 ASSERT(size > 0); | |
867 | |
3638
6b28ebc717aa
6496357 spec_fsync() is useless on devices that do write caching
billm
parents:
3547
diff
changeset
|
868 bufoff = uio->uio_loffset - db->db_offset; |
789 | 869 tocpy = (int)MIN(db->db_size - bufoff, size); |
870 | |
871 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
872 | |
873 if (tocpy == db->db_size) | |
874 dmu_buf_will_fill(db, tx); | |
875 else | |
876 dmu_buf_will_dirty(db, tx); | |
877 | |
878 /* | |
879 * XXX uiomove could block forever (eg. nfs-backed | |
880 * pages). There needs to be a uiolockdown() function | |
881 * to lock the pages in memory, so that uiomove won't | |
882 * block. | |
883 */ | |
884 err = uiomove((char *)db->db_data + bufoff, tocpy, | |
885 UIO_WRITE, uio); | |
886 | |
887 if (tocpy == db->db_size) | |
888 dmu_buf_fill_done(db, tx); | |
889 | |
890 if (err) | |
891 break; | |
892 | |
893 size -= tocpy; | |
894 } | |
1544 | 895 dmu_buf_rele_array(dbp, numbufs, FTAG); |
789 | 896 return (err); |
897 } | |
2688
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
898 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
899 int |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
900 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
901 page_t *pp, dmu_tx_t *tx) |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
902 { |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
903 dmu_buf_t **dbp; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
904 int numbufs, i; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
905 int err; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
906 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
907 if (size == 0) |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
908 return (0); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
909 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
910 err = dmu_buf_hold_array(os, object, offset, size, |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
911 FALSE, FTAG, &numbufs, &dbp); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
912 if (err) |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
913 return (err); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
914 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
915 for (i = 0; i < numbufs; i++) { |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
916 int tocpy, copied, thiscpy; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
917 int bufoff; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
918 dmu_buf_t *db = dbp[i]; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
919 caddr_t va; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
920 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
921 ASSERT(size > 0); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
922 ASSERT3U(db->db_size, >=, PAGESIZE); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
923 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
924 bufoff = offset - db->db_offset; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
925 tocpy = (int)MIN(db->db_size - bufoff, size); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
926 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
927 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
928 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
929 if (tocpy == db->db_size) |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
930 dmu_buf_will_fill(db, tx); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
931 else |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
932 dmu_buf_will_dirty(db, tx); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
933 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
934 for (copied = 0; copied < tocpy; copied += PAGESIZE) { |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
935 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
936 thiscpy = MIN(PAGESIZE, tocpy - copied); |
7315
cdba25672122
6735592 "zio" vmem arenas are confusing, not reported separately
Jonathan W Adams <Jonathan.Adams@Sun.COM>
parents:
7237
diff
changeset
|
937 va = zfs_map_page(pp, S_READ); |
2688
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
938 bcopy(va, (char *)db->db_data + bufoff, thiscpy); |
7315
cdba25672122
6735592 "zio" vmem arenas are confusing, not reported separately
Jonathan W Adams <Jonathan.Adams@Sun.COM>
parents:
7237
diff
changeset
|
939 zfs_unmap_page(pp, va); |
2688
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
940 pp = pp->p_next; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
941 bufoff += PAGESIZE; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
942 } |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
943 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
944 if (tocpy == db->db_size) |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
945 dmu_buf_fill_done(db, tx); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
946 |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
947 offset += tocpy; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
948 size -= tocpy; |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
949 } |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
950 dmu_buf_rele_array(dbp, numbufs, FTAG); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
951 return (err); |
12e2ad22b970
6447701 ZFS hangs when iSCSI Target attempts to initialize its backing store
maybee
parents:
2676
diff
changeset
|
952 } |
789 | 953 #endif |
954 | |
9412
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
955 /* |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
956 * Allocate a loaned anonymous arc buffer. |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
957 */ |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
958 arc_buf_t * |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
959 dmu_request_arcbuf(dmu_buf_t *handle, int size) |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
960 { |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
961 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode; |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
962 |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
963 return (arc_loan_buf(dn->dn_objset->os_spa, size)); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
964 } |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
965 |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
966 /* |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
967 * Free a loaned arc buffer. |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
968 */ |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
969 void |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
970 dmu_return_arcbuf(arc_buf_t *buf) |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
971 { |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
972 arc_return_buf(buf, FTAG); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
973 VERIFY(arc_buf_remove_ref(buf, FTAG) == 1); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
974 } |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
975 |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
976 /* |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
977 * When possible directly assign passed loaned arc buffer to a dbuf. |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
978 * If this is not possible copy the contents of passed arc buf via |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
979 * dmu_write(). |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
980 */ |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
981 void |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
982 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
983 dmu_tx_t *tx) |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
984 { |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
985 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode; |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
986 dmu_buf_impl_t *db; |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
987 uint32_t blksz = (uint32_t)arc_buf_size(buf); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
988 uint64_t blkid; |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
989 |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
990 rw_enter(&dn->dn_struct_rwlock, RW_READER); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
991 blkid = dbuf_whichblock(dn, offset); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
992 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
993 rw_exit(&dn->dn_struct_rwlock); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
994 |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
995 if (offset == db->db.db_offset && blksz == db->db.db_size) { |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
996 dbuf_assign_arcbuf(db, buf, tx); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
997 dbuf_rele(db, FTAG); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
998 } else { |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
999 dbuf_rele(db, FTAG); |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1000 dmu_write(dn->dn_objset, dn->dn_object, offset, blksz, |
9412
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
1001 buf->b_data, tx); |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
1002 dmu_return_arcbuf(buf); |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
1003 XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9412
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
1004 } |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
1005 } |
4aefd8704ce0
6717022 ZFS DMU needs zero-copy support
Aleksandr Guzovskiy <Aleksandr.Guzovskiy@Sun.COM>
parents:
9396
diff
changeset
|
1006 |
2237 | 1007 typedef struct { |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1008 dbuf_dirty_record_t *dsa_dr; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1009 dmu_sync_cb_t *dsa_done; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1010 zgd_t *dsa_zgd; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1011 dmu_tx_t *dsa_tx; |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1012 } dmu_sync_arg_t; |
2237 | 1013 |
1014 /* ARGSUSED */ | |
1015 static void | |
7754
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1016 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) |
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1017 { |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1018 dmu_sync_arg_t *dsa = varg; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1019 dmu_buf_t *db = dsa->dsa_zgd->zgd_db; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1020 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; |
7754
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1021 blkptr_t *bp = zio->io_bp; |
10800
469478b180d9
6880764 fsync on zfs is broken if writes are greater than 32kb on a hard crash and no log attached
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
10474
diff
changeset
|
1022 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1023 if (zio->io_error == 0) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1024 if (BP_IS_HOLE(bp)) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1025 /* |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1026 * A block of zeros may compress to a hole, but the |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1027 * block size still needs to be known for replay. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1028 */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1029 BP_SET_LSIZE(bp, db->db_size); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1030 } else { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1031 ASSERT(BP_GET_TYPE(bp) == dn->dn_type); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1032 ASSERT(BP_GET_LEVEL(bp) == 0); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1033 bp->blk_fill = 1; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1034 } |
7754
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1035 } |
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1036 } |
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1037 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1038 static void |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1039 dmu_sync_late_arrival_ready(zio_t *zio) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1040 { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1041 dmu_sync_ready(zio, NULL, zio->io_private); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1042 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1043 |
7754
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1044 /* ARGSUSED */ |
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1045 static void |
2237 | 1046 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) |
1047 { | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1048 dmu_sync_arg_t *dsa = varg; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1049 dbuf_dirty_record_t *dr = dsa->dsa_dr; |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1050 dmu_buf_impl_t *db = dr->dr_dbuf; |
2237 | 1051 |
10810
b6b161a6ae4a
6892298 buf->b_hdr->b_state != arc_anon, file: ../../common/fs/zfs/arc.c, line: 2849
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
10800
diff
changeset
|
1052 mutex_enter(&db->db_mtx); |
b6b161a6ae4a
6892298 buf->b_hdr->b_state != arc_anon, file: ../../common/fs/zfs/arc.c, line: 2849
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
10800
diff
changeset
|
1053 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1054 if (zio->io_error == 0) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1055 dr->dt.dl.dr_overridden_by = *zio->io_bp; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1056 dr->dt.dl.dr_override_state = DR_OVERRIDDEN; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1057 dr->dt.dl.dr_copies = zio->io_prop.zp_copies; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1058 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1059 BP_ZERO(&dr->dt.dl.dr_overridden_by); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1060 } else { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1061 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1062 } |
2237 | 1063 cv_broadcast(&db->db_changed); |
10810
b6b161a6ae4a
6892298 buf->b_hdr->b_state != arc_anon, file: ../../common/fs/zfs/arc.c, line: 2849
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
10800
diff
changeset
|
1064 mutex_exit(&db->db_mtx); |
b6b161a6ae4a
6892298 buf->b_hdr->b_state != arc_anon, file: ../../common/fs/zfs/arc.c, line: 2849
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
10800
diff
changeset
|
1065 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1066 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1067 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1068 kmem_free(dsa, sizeof (*dsa)); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1069 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1070 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1071 static void |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1072 dmu_sync_late_arrival_done(zio_t *zio) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1073 { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1074 blkptr_t *bp = zio->io_bp; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1075 dmu_sync_arg_t *dsa = zio->io_private; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1076 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1077 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1078 ASSERT(zio->io_bp->blk_birth == zio->io_txg); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1079 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1080 zio_free(zio->io_spa, zio->io_txg, zio->io_bp); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1081 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1082 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1083 dmu_tx_commit(dsa->dsa_tx); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1084 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1085 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1086 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1087 kmem_free(dsa, sizeof (*dsa)); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1088 } |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1089 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1090 static int |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1091 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1092 zio_prop_t *zp, zbookmark_t *zb) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1093 { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1094 dmu_sync_arg_t *dsa; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1095 dmu_tx_t *tx; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1096 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1097 tx = dmu_tx_create(os); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1098 dmu_tx_hold_space(tx, zgd->zgd_db->db_size); |
11670
1d964fb5d948
6595532 ZIL is too talkative
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
11539
diff
changeset
|
1099 if (dmu_tx_assign(tx, TXG_WAIT) != 0) { |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1100 dmu_tx_abort(tx); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1101 return (EIO); /* Make zl_get_data do txg_waited_synced() */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1102 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1103 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1104 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1105 dsa->dsa_dr = NULL; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1106 dsa->dsa_done = done; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1107 dsa->dsa_zgd = zgd; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1108 dsa->dsa_tx = tx; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1109 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1110 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1111 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1112 dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1113 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1114 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1115 return (0); |
2237 | 1116 } |
1117 | |
789 | 1118 /* |
2237 | 1119 * Intent log support: sync the block associated with db to disk. |
1120 * N.B. and XXX: the caller is responsible for making sure that the | |
1121 * data isn't changing while dmu_sync() is writing it. | |
789 | 1122 * |
1123 * Return values: | |
1124 * | |
2237 | 1125 * EEXIST: this txg has already been synced, so there's nothing to to. |
789 | 1126 * The caller should not log the write. |
1127 * | |
1128 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1129 * The caller should not log the write. | |
1130 * | |
2237 | 1131 * EALREADY: this block is already in the process of being synced. |
1132 * The caller should track its progress (somehow). | |
789 | 1133 * |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1134 * EIO: could not do the I/O. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1135 * The caller should do a txg_wait_synced(). |
789 | 1136 * |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1137 * 0: the I/O has been initiated. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1138 * The caller should log this blkptr in the done callback. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1139 * It is possible that the I/O will fail, in which case |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1140 * the error will be reported to the done callback and |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1141 * propagated to pio from zio_done(). |
789 | 1142 */ |
1143 int | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1144 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
789 | 1145 { |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1146 blkptr_t *bp = zgd->zgd_bp; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1147 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1148 objset_t *os = db->db_objset; |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1149 dsl_dataset_t *ds = os->os_dsl_dataset; |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1150 dbuf_dirty_record_t *dr; |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1151 dmu_sync_arg_t *dsa; |
2237 | 1152 zbookmark_t zb; |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1153 zio_prop_t zp; |
789 | 1154 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1155 ASSERT(pio != NULL); |
789 | 1156 ASSERT(BP_IS_HOLE(bp)); |
1157 ASSERT(txg != 0); | |
1158 | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1159 SET_BOOKMARK(&zb, ds->ds_object, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1160 db->db.db_object, db->db_level, db->db_blkid); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1161 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1162 dmu_write_policy(os, db->db_dnode, db->db_level, WP_DMU_SYNC, &zp); |
789 | 1163 |
1164 /* | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1165 * If we're frozen (running ziltest), we always need to generate a bp. |
1544 | 1166 */ |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1167 if (txg > spa_freeze_txg(os->os_spa)) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1168 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); |
1544 | 1169 |
1170 /* | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1171 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1172 * and us. If we determine that this txg is not yet syncing, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1173 * but it begins to sync a moment later, that's OK because the |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1174 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. |
789 | 1175 */ |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1176 mutex_enter(&db->db_mtx); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1177 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1178 if (txg <= spa_last_synced_txg(os->os_spa)) { |
789 | 1179 /* |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1180 * This txg has already synced. There's nothing to do. |
789 | 1181 */ |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1182 mutex_exit(&db->db_mtx); |
2237 | 1183 return (EEXIST); |
789 | 1184 } |
1185 | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1186 if (txg <= spa_syncing_txg(os->os_spa)) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1187 /* |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1188 * This txg is currently syncing, so we can't mess with |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1189 * the dirty record anymore; just write a new log block. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1190 */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1191 mutex_exit(&db->db_mtx); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1192 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); |
2237 | 1193 } |
1194 | |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1195 dr = db->db_last_dirty; |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1196 while (dr && dr->dr_txg != txg) |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1197 dr = dr->dr_next; |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1198 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1199 if (dr == NULL) { |
2237 | 1200 /* |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1201 * There's no dr for this dbuf, so it must have been freed. |
2237 | 1202 * There's no need to log writes to freed blocks, so we're done. |
1203 */ | |
1204 mutex_exit(&db->db_mtx); | |
789 | 1205 return (ENOENT); |
1206 } | |
1207 | |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1208 ASSERT(dr->dr_txg == txg); |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1209 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1210 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1211 /* |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1212 * We have already issued a sync write for this buffer, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1213 * or this buffer has already been synced. It could not |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1214 * have been dirtied since, or we would have cleared the state. |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1215 */ |
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1216 mutex_exit(&db->db_mtx); |
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1217 return (EALREADY); |
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1218 } |
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1219 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1220 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
3547
e396e0a440b1
6512391 DMU should leverage ZIO dependencies to achieve greater parallelism
maybee
parents:
3444
diff
changeset
|
1221 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
2237 | 1222 mutex_exit(&db->db_mtx); |
7754
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1223 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1224 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1225 dsa->dsa_dr = dr; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1226 dsa->dsa_done = done; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1227 dsa->dsa_zgd = zgd; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1228 dsa->dsa_tx = NULL; |
7754
b80e4842ad54
6754011 SPA 3.0: lock breakup, i/o pipeline refactoring, device failure handling
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
7468
diff
changeset
|
1229 |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1230 zio_nowait(arc_write(pio, os->os_spa, txg, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1231 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1232 dmu_sync_ready, dmu_sync_done, dsa, |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1233 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1234 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1235 return (0); |
789 | 1236 } |
1237 | |
1238 int | |
1239 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
1240 dmu_tx_t *tx) | |
1241 { | |
1544 | 1242 dnode_t *dn; |
1243 int err; | |
1244 | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1245 err = dnode_hold(os, object, FTAG, &dn); |
1544 | 1246 if (err) |
1247 return (err); | |
1248 err = dnode_set_blksz(dn, size, ibs, tx); | |
789 | 1249 dnode_rele(dn, FTAG); |
1250 return (err); | |
1251 } | |
1252 | |
1253 void | |
1254 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
1255 dmu_tx_t *tx) | |
1256 { | |
1544 | 1257 dnode_t *dn; |
1258 | |
1259 /* XXX assumes dnode_hold will not get an i/o error */ | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1260 (void) dnode_hold(os, object, FTAG, &dn); |
789 | 1261 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); |
1262 dn->dn_checksum = checksum; | |
1263 dnode_setdirty(dn, tx); | |
1264 dnode_rele(dn, FTAG); | |
1265 } | |
1266 | |
1267 void | |
1268 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
1269 dmu_tx_t *tx) | |
1270 { | |
1544 | 1271 dnode_t *dn; |
1272 | |
1273 /* XXX assumes dnode_hold will not get an i/o error */ | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1274 (void) dnode_hold(os, object, FTAG, &dn); |
789 | 1275 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); |
1276 dn->dn_compress = compress; | |
1277 dnode_setdirty(dn, tx); | |
1278 dnode_rele(dn, FTAG); | |
1279 } | |
1280 | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1281 int zfs_mdcomp_disable = 0; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1282 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1283 void |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1284 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1285 { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1286 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1287 boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1288 enum zio_checksum checksum = os->os_checksum; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1289 enum zio_compress compress = os->os_compress; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1290 enum zio_checksum dedup_checksum = os->os_dedup_checksum; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1291 boolean_t dedup; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1292 boolean_t dedup_verify = os->os_dedup_verify; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1293 int copies = os->os_copies; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1294 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1295 /* |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1296 * Determine checksum setting. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1297 */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1298 if (ismd) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1299 /* |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1300 * Metadata always gets checksummed. If the data |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1301 * checksum is multi-bit correctable, and it's not a |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1302 * ZBT-style checksum, then it's suitable for metadata |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1303 * as well. Otherwise, the metadata checksum defaults |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1304 * to fletcher4. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1305 */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1306 if (zio_checksum_table[checksum].ci_correctable < 1 || |
11670
1d964fb5d948
6595532 ZIL is too talkative
Neil Perrin <Neil.Perrin@Sun.COM>
parents:
11539
diff
changeset
|
1307 zio_checksum_table[checksum].ci_eck) |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1308 checksum = ZIO_CHECKSUM_FLETCHER_4; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1309 } else { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1310 checksum = zio_checksum_select(dn->dn_checksum, checksum); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1311 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1312 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1313 /* |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1314 * Determine compression setting. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1315 */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1316 if (ismd) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1317 /* |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1318 * XXX -- we should design a compression algorithm |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1319 * that specializes in arrays of bps. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1320 */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1321 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1322 ZIO_COMPRESS_LZJB; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1323 } else { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1324 compress = zio_compress_select(dn->dn_compress, compress); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1325 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1326 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1327 /* |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1328 * Determine dedup setting. If we are in dmu_sync(), we won't |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1329 * actually dedup now because that's all done in syncing context; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1330 * but we do want to use the dedup checkum. If the checksum is not |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1331 * strong enough to ensure unique signatures, force dedup_verify. |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1332 */ |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1333 dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1334 if (dedup) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1335 checksum = dedup_checksum; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1336 if (!zio_checksum_table[checksum].ci_dedup) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1337 dedup_verify = 1; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1338 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1339 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1340 if (wp & WP_DMU_SYNC) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1341 dedup = 0; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1342 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1343 if (wp & WP_NOFILL) { |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1344 ASSERT(!ismd && level == 0); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1345 checksum = ZIO_CHECKSUM_OFF; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1346 compress = ZIO_COMPRESS_OFF; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1347 dedup = B_FALSE; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1348 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1349 |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1350 zp->zp_checksum = checksum; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1351 zp->zp_compress = compress; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1352 zp->zp_type = type; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1353 zp->zp_level = level; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1354 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa)); |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1355 zp->zp_dedup = dedup; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1356 zp->zp_dedup_verify = dedup && dedup_verify; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1357 } |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1358 |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1731
diff
changeset
|
1359 int |
789 | 1360 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) |
1361 { | |
1362 dnode_t *dn; | |
1363 int i, err; | |
1364 | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1365 err = dnode_hold(os, object, FTAG, &dn); |
1544 | 1366 if (err) |
1367 return (err); | |
789 | 1368 /* |
1369 * Sync any current changes before | |
1370 * we go trundling through the block pointers. | |
1371 */ | |
1372 for (i = 0; i < TXG_SIZE; i++) { | |
1596
2e2377ccbf85
6395371 ASSERT in dmu_tx_count_free: blkid + i < dn->dn_phys->dn_nblkptr
ahrens
parents:
1544
diff
changeset
|
1373 if (list_link_active(&dn->dn_dirty_link[i])) |
789 | 1374 break; |
1375 } | |
1376 if (i != TXG_SIZE) { | |
1377 dnode_rele(dn, FTAG); | |
1378 txg_wait_synced(dmu_objset_pool(os), 0); | |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1379 err = dnode_hold(os, object, FTAG, &dn); |
1544 | 1380 if (err) |
1381 return (err); | |
789 | 1382 } |
1383 | |
6992 | 1384 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); |
789 | 1385 dnode_rele(dn, FTAG); |
1386 | |
1387 return (err); | |
1388 } | |
1389 | |
1390 void | |
1391 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
1392 { | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1393 dnode_phys_t *dnp; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1394 |
789 | 1395 rw_enter(&dn->dn_struct_rwlock, RW_READER); |
1396 mutex_enter(&dn->dn_mtx); | |
1397 | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1398 dnp = dn->dn_phys; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1399 |
789 | 1400 doi->doi_data_block_size = dn->dn_datablksz; |
1401 doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
1402 1ULL << dn->dn_indblkshift : 0; | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1403 doi->doi_type = dn->dn_type; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1404 doi->doi_bonus_type = dn->dn_bonustype; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1405 doi->doi_bonus_size = dn->dn_bonuslen; |
789 | 1406 doi->doi_indirection = dn->dn_nlevels; |
1407 doi->doi_checksum = dn->dn_checksum; | |
1408 doi->doi_compress = dn->dn_compress; | |
10922
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1409 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1410 doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1411 doi->doi_fill_count = 0; |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1412 for (int i = 0; i < dnp->dn_nblkptr; i++) |
e2081f502306
PSARC 2009/571 ZFS Deduplication Properties
Jeff Bonwick <Jeff.Bonwick@Sun.COM>
parents:
10810
diff
changeset
|
1413 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill; |
789 | 1414 |
1415 mutex_exit(&dn->dn_mtx); | |
1416 rw_exit(&dn->dn_struct_rwlock); | |
1417 } | |
1418 | |
1419 /* | |
1420 * Get information on a DMU object. | |
1421 * If doi is NULL, just indicates whether the object exists. | |
1422 */ | |
1423 int | |
1424 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
1425 { | |
1544 | 1426 dnode_t *dn; |
10298
a0d52501437c
6860996 %temporary clones are not automatically destroyed on error
Matthew Ahrens <Matthew.Ahrens@Sun.COM>
parents:
10242
diff
changeset
|
1427 int err = dnode_hold(os, object, FTAG, &dn); |
789 | 1428 |
1544 | 1429 if (err) |
1430 return (err); | |
789 | 1431 |
1432 if (doi != NULL) | |
1433 dmu_object_info_from_dnode(dn, doi); | |
1434 | |
1435 dnode_rele(dn, FTAG); | |
1436 return (0); | |
1437 } | |
1438 | |
1439 /* | |
1440 * As above, but faster; can be used when you have a held dbuf in hand. | |
1441 */ | |
1442 void | |
1443 dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi) | |
1444 { | |
1445 dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi); | |
1446 } | |
1447 | |
1448 /* | |
1449 * Faster still when you only care about the size. | |
1450 * This is specifically optimized for zfs_getattr(). | |
1451 */ | |
1452 void | |
1453 dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512) | |
1454 { | |
1455 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; | |
1456 | |
1457 *blksize = dn->dn_datablksz; | |
2082 | 1458 /* add 1 for dnode space */ |
1459 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> | |
1460 SPA_MINBLOCKSHIFT) + 1; | |
789 | 1461 } |
1462 | |
1463 void | |
1464 byteswap_uint64_array(void *vbuf, size_t size) | |
1465 { | |
1466 uint64_t *buf = vbuf; | |
1467 size_t count = size >> 3; | |
1468 int i; | |
1469 | |
1470 ASSERT((size & 7) == 0); | |
1471 | |
1472 for (i = 0; i < count; i++) | |
1473 buf[i] = BSWAP_64(buf[i]); | |
1474 } | |
1475 | |
1476 void | |
1477 byteswap_uint32_array(void *vbuf, size_t size) | |
1478 { | |
1479 uint32_t *buf = vbuf; | |
1480 size_t count = size >> 2; | |
1481 int i; | |
1482 | |
1483 ASSERT((size & 3) == 0); | |
1484 | |
1485 for (i = 0; i < count; i++) | |
1486 buf[i] = BSWAP_32(buf[i]); | |
1487 } | |
1488 | |
1489 void | |
1490 byteswap_uint16_array(void *vbuf, size_t size) | |
1491 { | |
1492 uint16_t *buf = vbuf; | |
1493 size_t count = size >> 1; | |
1494 int i; | |
1495 | |
1496 ASSERT((size & 1) == 0); | |
1497 | |
1498 for (i = 0; i < count; i++) | |
1499 buf[i] = BSWAP_16(buf[i]); | |
1500 } | |
1501 | |
1502 /* ARGSUSED */ | |
1503 void | |
1504 byteswap_uint8_array(void *vbuf, size_t size) | |
1505 { | |
1506 } | |
1507 | |
1508 void | |
1509 dmu_init(void) | |
1510 { | |
1511 dbuf_init(); | |
1512 dnode_init(); | |
10474
0e96dd3b905a
6859997 zfs caching performance problem
Rich Morris <Richard.Morris@Sun.COM>
parents:
10298
diff
changeset
|
1513 zfetch_init(); |
789 | 1514 arc_init(); |
5450 | 1515 l2arc_init(); |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
1516 xuio_stat_init(); |
789 | 1517 } |
1518 | |
1519 void | |
1520 dmu_fini(void) | |
1521 { | |
1522 arc_fini(); | |
10474
0e96dd3b905a
6859997 zfs caching performance problem
Rich Morris <Richard.Morris@Sun.COM>
parents:
10298
diff
changeset
|
1523 zfetch_fini(); |
789 | 1524 dnode_fini(); |
1525 dbuf_fini(); | |
5450 | 1526 l2arc_fini(); |
11539
10d35fc3d7fd
6873106 Need a mechanism to share buffers between fs modules
chunli zhang - Sun Microsystems - Irvine United States <Chunli.Zhang@Sun.COM>
parents:
10922
diff
changeset
|
1527 xuio_stat_fini(); |
789 | 1528 } |