1 /*
2 * Copyright (C) 2011 Red Hat, Inc.
3 *
4 * This file is released under the GPL.
5 */
6
7 #include "dm-btree-internal.h"
8 #include "dm-transaction-manager.h"
9
10 #include <linux/device-mapper.h>
11
12 #define DM_MSG_PREFIX "btree spine"
13
14 /*----------------------------------------------------------------*/
15
16 #define BTREE_CSUM_XOR 121107
17
18 static int node_check(struct dm_block_validator *v,
19 struct dm_block *b,
20 size_t block_size);
21
node_prepare_for_write(struct dm_block_validator * v,struct dm_block * b,size_t block_size)22 static void node_prepare_for_write(struct dm_block_validator *v,
23 struct dm_block *b,
24 size_t block_size)
25 {
26 struct btree_node *n = dm_block_data(b);
27 struct node_header *h = &n->header;
28
29 h->blocknr = cpu_to_le64(dm_block_location(b));
30 h->csum = cpu_to_le32(dm_bm_checksum(&h->flags,
31 block_size - sizeof(__le32),
32 BTREE_CSUM_XOR));
33 }
34
node_check(struct dm_block_validator * v,struct dm_block * b,size_t block_size)35 static int node_check(struct dm_block_validator *v,
36 struct dm_block *b,
37 size_t block_size)
38 {
39 struct btree_node *n = dm_block_data(b);
40 struct node_header *h = &n->header;
41 size_t value_size;
42 __le32 csum_disk;
43 uint32_t flags;
44
45 if (dm_block_location(b) != le64_to_cpu(h->blocknr)) {
46 DMERR_LIMIT("node_check failed: blocknr %llu != wanted %llu",
47 le64_to_cpu(h->blocknr), dm_block_location(b));
48 return -ENOTBLK;
49 }
50
51 csum_disk = cpu_to_le32(dm_bm_checksum(&h->flags,
52 block_size - sizeof(__le32),
53 BTREE_CSUM_XOR));
54 if (csum_disk != h->csum) {
55 DMERR_LIMIT("node_check failed: csum %u != wanted %u",
56 le32_to_cpu(csum_disk), le32_to_cpu(h->csum));
57 return -EILSEQ;
58 }
59
60 value_size = le32_to_cpu(h->value_size);
61
62 if (sizeof(struct node_header) +
63 (sizeof(__le64) + value_size) * le32_to_cpu(h->max_entries) > block_size) {
64 DMERR_LIMIT("node_check failed: max_entries too large");
65 return -EILSEQ;
66 }
67
68 if (le32_to_cpu(h->nr_entries) > le32_to_cpu(h->max_entries)) {
69 DMERR_LIMIT("node_check failed: too many entries");
70 return -EILSEQ;
71 }
72
73 /*
74 * The node must be either INTERNAL or LEAF.
75 */
76 flags = le32_to_cpu(h->flags);
77 if (!(flags & INTERNAL_NODE) && !(flags & LEAF_NODE)) {
78 DMERR_LIMIT("node_check failed: node is neither INTERNAL or LEAF");
79 return -EILSEQ;
80 }
81
82 return 0;
83 }
84
85 struct dm_block_validator btree_node_validator = {
86 .name = "btree_node",
87 .prepare_for_write = node_prepare_for_write,
88 .check = node_check
89 };
90
91 /*----------------------------------------------------------------*/
92
bn_read_lock(struct dm_btree_info * info,dm_block_t b,struct dm_block ** result)93 int bn_read_lock(struct dm_btree_info *info, dm_block_t b,
94 struct dm_block **result)
95 {
96 return dm_tm_read_lock(info->tm, b, &btree_node_validator, result);
97 }
98
bn_shadow(struct dm_btree_info * info,dm_block_t orig,struct dm_btree_value_type * vt,struct dm_block ** result)99 static int bn_shadow(struct dm_btree_info *info, dm_block_t orig,
100 struct dm_btree_value_type *vt,
101 struct dm_block **result)
102 {
103 int r, inc;
104
105 r = dm_tm_shadow_block(info->tm, orig, &btree_node_validator,
106 result, &inc);
107 if (!r && inc)
108 inc_children(info->tm, dm_block_data(*result), vt);
109
110 return r;
111 }
112
new_block(struct dm_btree_info * info,struct dm_block ** result)113 int new_block(struct dm_btree_info *info, struct dm_block **result)
114 {
115 return dm_tm_new_block(info->tm, &btree_node_validator, result);
116 }
117
unlock_block(struct dm_btree_info * info,struct dm_block * b)118 void unlock_block(struct dm_btree_info *info, struct dm_block *b)
119 {
120 dm_tm_unlock(info->tm, b);
121 }
122
123 /*----------------------------------------------------------------*/
124
init_ro_spine(struct ro_spine * s,struct dm_btree_info * info)125 void init_ro_spine(struct ro_spine *s, struct dm_btree_info *info)
126 {
127 s->info = info;
128 s->count = 0;
129 s->nodes[0] = NULL;
130 s->nodes[1] = NULL;
131 }
132
exit_ro_spine(struct ro_spine * s)133 void exit_ro_spine(struct ro_spine *s)
134 {
135 int i;
136
137 for (i = 0; i < s->count; i++) {
138 unlock_block(s->info, s->nodes[i]);
139 }
140 }
141
ro_step(struct ro_spine * s,dm_block_t new_child)142 int ro_step(struct ro_spine *s, dm_block_t new_child)
143 {
144 int r;
145
146 if (s->count == 2) {
147 unlock_block(s->info, s->nodes[0]);
148 s->nodes[0] = s->nodes[1];
149 s->count--;
150 }
151
152 r = bn_read_lock(s->info, new_child, s->nodes + s->count);
153 if (!r)
154 s->count++;
155
156 return r;
157 }
158
ro_pop(struct ro_spine * s)159 void ro_pop(struct ro_spine *s)
160 {
161 BUG_ON(!s->count);
162 --s->count;
163 unlock_block(s->info, s->nodes[s->count]);
164 }
165
ro_node(struct ro_spine * s)166 struct btree_node *ro_node(struct ro_spine *s)
167 {
168 struct dm_block *block;
169
170 BUG_ON(!s->count);
171 block = s->nodes[s->count - 1];
172
173 return dm_block_data(block);
174 }
175
176 /*----------------------------------------------------------------*/
177
init_shadow_spine(struct shadow_spine * s,struct dm_btree_info * info)178 void init_shadow_spine(struct shadow_spine *s, struct dm_btree_info *info)
179 {
180 s->info = info;
181 s->count = 0;
182 }
183
exit_shadow_spine(struct shadow_spine * s)184 void exit_shadow_spine(struct shadow_spine *s)
185 {
186 int i;
187
188 for (i = 0; i < s->count; i++) {
189 unlock_block(s->info, s->nodes[i]);
190 }
191 }
192
shadow_step(struct shadow_spine * s,dm_block_t b,struct dm_btree_value_type * vt)193 int shadow_step(struct shadow_spine *s, dm_block_t b,
194 struct dm_btree_value_type *vt)
195 {
196 int r;
197
198 if (s->count == 2) {
199 unlock_block(s->info, s->nodes[0]);
200 s->nodes[0] = s->nodes[1];
201 s->count--;
202 }
203
204 r = bn_shadow(s->info, b, vt, s->nodes + s->count);
205 if (!r) {
206 if (!s->count)
207 s->root = dm_block_location(s->nodes[0]);
208
209 s->count++;
210 }
211
212 return r;
213 }
214
shadow_current(struct shadow_spine * s)215 struct dm_block *shadow_current(struct shadow_spine *s)
216 {
217 BUG_ON(!s->count);
218
219 return s->nodes[s->count - 1];
220 }
221
shadow_parent(struct shadow_spine * s)222 struct dm_block *shadow_parent(struct shadow_spine *s)
223 {
224 BUG_ON(s->count != 2);
225
226 return s->count == 2 ? s->nodes[0] : NULL;
227 }
228
shadow_has_parent(struct shadow_spine * s)229 int shadow_has_parent(struct shadow_spine *s)
230 {
231 return s->count >= 2;
232 }
233
shadow_root(struct shadow_spine * s)234 dm_block_t shadow_root(struct shadow_spine *s)
235 {
236 return s->root;
237 }
238
le64_inc(void * context,const void * value_le,unsigned count)239 static void le64_inc(void *context, const void *value_le, unsigned count)
240 {
241 dm_tm_with_runs(context, value_le, count, dm_tm_inc_range);
242 }
243
le64_dec(void * context,const void * value_le,unsigned count)244 static void le64_dec(void *context, const void *value_le, unsigned count)
245 {
246 dm_tm_with_runs(context, value_le, count, dm_tm_dec_range);
247 }
248
le64_equal(void * context,const void * value1_le,const void * value2_le)249 static int le64_equal(void *context, const void *value1_le, const void *value2_le)
250 {
251 __le64 v1_le, v2_le;
252
253 memcpy(&v1_le, value1_le, sizeof(v1_le));
254 memcpy(&v2_le, value2_le, sizeof(v2_le));
255 return v1_le == v2_le;
256 }
257
init_le64_type(struct dm_transaction_manager * tm,struct dm_btree_value_type * vt)258 void init_le64_type(struct dm_transaction_manager *tm,
259 struct dm_btree_value_type *vt)
260 {
261 vt->context = tm;
262 vt->size = sizeof(__le64);
263 vt->inc = le64_inc;
264 vt->dec = le64_dec;
265 vt->equal = le64_equal;
266 }
267