Print this page
5056 ZFS deadlock on db_mtx and dn_holds
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/sys/zap_leaf.h
+++ new/usr/src/uts/common/fs/zfs/sys/zap_leaf.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 + * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
23 24 */
24 25
25 26 #ifndef _SYS_ZAP_LEAF_H
26 27 #define _SYS_ZAP_LEAF_H
27 28
28 29 #include <sys/zap.h>
29 30
30 31 #ifdef __cplusplus
31 32 extern "C" {
32 33 #endif
33 34
34 35 struct zap;
35 36 struct zap_name;
36 37 struct zap_stats;
37 38
38 39 #define ZAP_LEAF_MAGIC 0x2AB1EAF
39 40
40 41 /* chunk size = 24 bytes */
41 42 #define ZAP_LEAF_CHUNKSIZE 24
42 43
43 44 /*
44 45 * The amount of space available for chunks is:
45 46 * block size (1<<l->l_bs) - hash entry size (2) * number of hash
46 47 * entries - header space (2*chunksize)
47 48 */
48 49 #define ZAP_LEAF_NUMCHUNKS(l) \
49 50 (((1<<(l)->l_bs) - 2*ZAP_LEAF_HASH_NUMENTRIES(l)) / \
50 51 ZAP_LEAF_CHUNKSIZE - 2)
51 52
52 53 /*
53 54 * The amount of space within the chunk available for the array is:
54 55 * chunk size - space for type (1) - space for next pointer (2)
55 56 */
56 57 #define ZAP_LEAF_ARRAY_BYTES (ZAP_LEAF_CHUNKSIZE - 3)
57 58
58 59 #define ZAP_LEAF_ARRAY_NCHUNKS(bytes) \
59 60 (((bytes)+ZAP_LEAF_ARRAY_BYTES-1)/ZAP_LEAF_ARRAY_BYTES)
60 61
61 62 /*
62 63 * Low water mark: when there are only this many chunks free, start
63 64 * growing the ptrtbl. Ideally, this should be larger than a
64 65 * "reasonably-sized" entry. 20 chunks is more than enough for the
65 66 * largest directory entry (MAXNAMELEN (256) byte name, 8-byte value),
66 67 * while still being only around 3% for 16k blocks.
67 68 */
68 69 #define ZAP_LEAF_LOW_WATER (20)
69 70
70 71 /*
71 72 * The leaf hash table has block size / 2^5 (32) number of entries,
72 73 * which should be more than enough for the maximum number of entries,
73 74 * which is less than block size / CHUNKSIZE (24) / minimum number of
74 75 * chunks per entry (3).
75 76 */
76 77 #define ZAP_LEAF_HASH_SHIFT(l) ((l)->l_bs - 5)
77 78 #define ZAP_LEAF_HASH_NUMENTRIES(l) (1 << ZAP_LEAF_HASH_SHIFT(l))
78 79
79 80 /*
80 81 * The chunks start immediately after the hash table. The end of the
81 82 * hash table is at l_hash + HASH_NUMENTRIES, which we simply cast to a
82 83 * chunk_t.
83 84 */
84 85 #define ZAP_LEAF_CHUNK(l, idx) \
85 86 ((zap_leaf_chunk_t *) \
86 87 (zap_leaf_phys(l)->l_hash + ZAP_LEAF_HASH_NUMENTRIES(l)))[idx]
87 88 #define ZAP_LEAF_ENTRY(l, idx) (&ZAP_LEAF_CHUNK(l, idx).l_entry)
88 89
89 90 typedef enum zap_chunk_type {
90 91 ZAP_CHUNK_FREE = 253,
91 92 ZAP_CHUNK_ENTRY = 252,
92 93 ZAP_CHUNK_ARRAY = 251,
93 94 ZAP_CHUNK_TYPE_MAX = 250
94 95 } zap_chunk_type_t;
95 96
96 97 #define ZLF_ENTRIES_CDSORTED (1<<0)
97 98
98 99 /*
99 100 * TAKE NOTE:
100 101 * If zap_leaf_phys_t is modified, zap_leaf_byteswap() must be modified.
101 102 */
102 103 typedef struct zap_leaf_phys {
103 104 struct zap_leaf_header {
104 105 /* Public to ZAP */
105 106 uint64_t lh_block_type; /* ZBT_LEAF */
106 107 uint64_t lh_pad1;
107 108 uint64_t lh_prefix; /* hash prefix of this leaf */
108 109 uint32_t lh_magic; /* ZAP_LEAF_MAGIC */
109 110 uint16_t lh_nfree; /* number free chunks */
110 111 uint16_t lh_nentries; /* number of entries */
111 112 uint16_t lh_prefix_len; /* num bits used to id this */
112 113
113 114 /* Private to zap_leaf */
114 115 uint16_t lh_freelist; /* chunk head of free list */
115 116 uint8_t lh_flags; /* ZLF_* flags */
116 117 uint8_t lh_pad2[11];
117 118 } l_hdr; /* 2 24-byte chunks */
118 119
119 120 /*
120 121 * The header is followed by a hash table with
121 122 * ZAP_LEAF_HASH_NUMENTRIES(zap) entries. The hash table is
122 123 * followed by an array of ZAP_LEAF_NUMCHUNKS(zap)
123 124 * zap_leaf_chunk structures. These structures are accessed
124 125 * with the ZAP_LEAF_CHUNK() macro.
125 126 */
126 127
127 128 uint16_t l_hash[1];
128 129 } zap_leaf_phys_t;
129 130
130 131 typedef union zap_leaf_chunk {
131 132 struct zap_leaf_entry {
132 133 uint8_t le_type; /* always ZAP_CHUNK_ENTRY */
133 134 uint8_t le_value_intlen; /* size of value's ints */
134 135 uint16_t le_next; /* next entry in hash chain */
135 136 uint16_t le_name_chunk; /* first chunk of the name */
136 137 uint16_t le_name_numints; /* ints in name (incl null) */
137 138 uint16_t le_value_chunk; /* first chunk of the value */
138 139 uint16_t le_value_numints; /* value length in ints */
139 140 uint32_t le_cd; /* collision differentiator */
140 141 uint64_t le_hash; /* hash value of the name */
141 142 } l_entry;
142 143 struct zap_leaf_array {
143 144 uint8_t la_type; /* always ZAP_CHUNK_ARRAY */
144 145 uint8_t la_array[ZAP_LEAF_ARRAY_BYTES];
|
↓ open down ↓ |
112 lines elided |
↑ open up ↑ |
145 146 uint16_t la_next; /* next blk or CHAIN_END */
146 147 } l_array;
147 148 struct zap_leaf_free {
148 149 uint8_t lf_type; /* always ZAP_CHUNK_FREE */
149 150 uint8_t lf_pad[ZAP_LEAF_ARRAY_BYTES];
150 151 uint16_t lf_next; /* next in free list, or CHAIN_END */
151 152 } l_free;
152 153 } zap_leaf_chunk_t;
153 154
154 155 typedef struct zap_leaf {
156 + dmu_buf_user_t l_dbu;
155 157 krwlock_t l_rwlock;
156 158 uint64_t l_blkid; /* 1<<ZAP_BLOCK_SHIFT byte block off */
157 159 int l_bs; /* block size shift */
158 160 dmu_buf_t *l_dbuf;
159 161 } zap_leaf_t;
160 162
161 163 inline zap_leaf_phys_t *
162 164 zap_leaf_phys(zap_leaf_t *l)
163 165 {
164 166 return (l->l_dbuf->db_data);
165 167 }
166 168
167 169 typedef struct zap_entry_handle {
168 170 /* Set by zap_leaf and public to ZAP */
169 171 uint64_t zeh_num_integers;
170 172 uint64_t zeh_hash;
171 173 uint32_t zeh_cd;
172 174 uint8_t zeh_integer_size;
173 175
174 176 /* Private to zap_leaf */
175 177 uint16_t zeh_fakechunk;
176 178 uint16_t *zeh_chunkp;
177 179 zap_leaf_t *zeh_leaf;
178 180 } zap_entry_handle_t;
179 181
180 182 /*
181 183 * Return a handle to the named entry, or ENOENT if not found. The hash
182 184 * value must equal zap_hash(name).
183 185 */
184 186 extern int zap_leaf_lookup(zap_leaf_t *l,
185 187 struct zap_name *zn, zap_entry_handle_t *zeh);
186 188
187 189 /*
188 190 * Return a handle to the entry with this hash+cd, or the entry with the
189 191 * next closest hash+cd.
190 192 */
191 193 extern int zap_leaf_lookup_closest(zap_leaf_t *l,
192 194 uint64_t hash, uint32_t cd, zap_entry_handle_t *zeh);
193 195
194 196 /*
195 197 * Read the first num_integers in the attribute. Integer size
196 198 * conversion will be done without sign extension. Return EINVAL if
197 199 * integer_size is too small. Return EOVERFLOW if there are more than
198 200 * num_integers in the attribute.
199 201 */
200 202 extern int zap_entry_read(const zap_entry_handle_t *zeh,
201 203 uint8_t integer_size, uint64_t num_integers, void *buf);
202 204
203 205 extern int zap_entry_read_name(struct zap *zap, const zap_entry_handle_t *zeh,
204 206 uint16_t buflen, char *buf);
205 207
206 208 /*
207 209 * Replace the value of an existing entry.
208 210 *
209 211 * May fail if it runs out of space (ENOSPC).
210 212 */
211 213 extern int zap_entry_update(zap_entry_handle_t *zeh,
212 214 uint8_t integer_size, uint64_t num_integers, const void *buf);
213 215
214 216 /*
215 217 * Remove an entry.
216 218 */
217 219 extern void zap_entry_remove(zap_entry_handle_t *zeh);
218 220
219 221 /*
220 222 * Create an entry. An equal entry must not exist, and this entry must
221 223 * belong in this leaf (according to its hash value). Fills in the
222 224 * entry handle on success. Returns 0 on success or ENOSPC on failure.
223 225 */
224 226 extern int zap_entry_create(zap_leaf_t *l, struct zap_name *zn, uint32_t cd,
225 227 uint8_t integer_size, uint64_t num_integers, const void *buf,
226 228 zap_entry_handle_t *zeh);
227 229
228 230 /* Determine whether there is another entry with the same normalized form. */
229 231 extern boolean_t zap_entry_normalization_conflict(zap_entry_handle_t *zeh,
230 232 struct zap_name *zn, const char *name, struct zap *zap);
231 233
232 234 /*
233 235 * Other stuff.
234 236 */
235 237
236 238 extern void zap_leaf_init(zap_leaf_t *l, boolean_t sort);
237 239 extern void zap_leaf_byteswap(zap_leaf_phys_t *buf, int len);
238 240 extern void zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort);
239 241 extern void zap_leaf_stats(struct zap *zap, zap_leaf_t *l,
240 242 struct zap_stats *zs);
241 243
242 244 #ifdef __cplusplus
243 245 }
244 246 #endif
245 247
246 248 #endif /* _SYS_ZAP_LEAF_H */
|
↓ open down ↓ |
82 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX