1 /* TDB tools to create various canned database layouts. */
9 struct tdb_layout *new_tdb_layout(const char *filename)
11 struct tdb_layout *layout = malloc(sizeof(*layout));
12 layout->filename = filename;
13 layout->num_elems = 0;
18 static void add(struct tdb_layout *layout, union tdb_layout_elem elem)
20 layout->elem = realloc(layout->elem,
21 sizeof(layout->elem[0])
22 * (layout->num_elems+1));
23 layout->elem[layout->num_elems++] = elem;
26 void tdb_layout_add_freelist(struct tdb_layout *layout)
28 union tdb_layout_elem elem;
29 elem.base.type = FREELIST;
33 void tdb_layout_add_free(struct tdb_layout *layout, tdb_len_t len,
36 union tdb_layout_elem elem;
37 elem.base.type = FREE;
39 elem.free.flist_num = flist;
43 static struct tdb_data dup_key(struct tdb_data key)
46 ret.dsize = key.dsize;
47 ret.dptr = malloc(ret.dsize);
48 memcpy(ret.dptr, key.dptr, ret.dsize);
52 void tdb_layout_add_used(struct tdb_layout *layout,
53 TDB_DATA key, TDB_DATA data,
56 union tdb_layout_elem elem;
57 elem.base.type = DATA;
58 elem.used.key = dup_key(key);
59 elem.used.data = dup_key(data);
60 elem.used.extra = extra;
64 static tdb_len_t free_record_len(tdb_len_t len)
66 return sizeof(struct tdb_used_record) + len;
69 static tdb_len_t data_record_len(struct tle_used *used)
72 len = sizeof(struct tdb_used_record)
73 + used->key.dsize + used->data.dsize + used->extra;
74 assert(len >= sizeof(struct tdb_free_record));
78 static tdb_len_t hashtable_len(struct tle_hashtable *htable)
80 return sizeof(struct tdb_used_record)
81 + (sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS)
85 static tdb_len_t freelist_len(struct tle_freelist *flist)
87 return sizeof(struct tdb_freelist);
90 static void set_free_record(void *mem, tdb_len_t len)
92 /* We do all the work in add_to_freetable */
95 static void set_data_record(void *mem, struct tdb_context *tdb,
96 struct tle_used *used)
98 struct tdb_used_record *u = mem;
100 set_header(tdb, u, used->key.dsize, used->data.dsize,
101 used->key.dsize + used->data.dsize + used->extra,
102 tdb_hash(tdb, used->key.dptr, used->key.dsize));
103 memcpy(u + 1, used->key.dptr, used->key.dsize);
104 memcpy((char *)(u + 1) + used->key.dsize,
105 used->data.dptr, used->data.dsize);
108 static void set_hashtable(void *mem, struct tdb_context *tdb,
109 struct tle_hashtable *htable)
111 struct tdb_used_record *u = mem;
112 tdb_len_t len = sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS;
114 set_header(tdb, u, 0, len, len + htable->extra, 0);
115 memset(u + 1, 0, len);
118 static void set_freelist(void *mem, struct tdb_context *tdb,
119 struct tle_freelist *freelist, struct tdb_header *hdr,
120 tdb_off_t last_flist)
122 struct tdb_freelist *flist = mem;
123 memset(flist, 0, sizeof(*flist));
124 set_header(tdb, &flist->hdr, 0,
125 sizeof(*flist) - sizeof(flist->hdr),
126 sizeof(*flist) - sizeof(flist->hdr), 1);
129 flist = (struct tdb_freelist *)((char *)hdr + last_flist);
130 flist->next = freelist->base.off;
132 hdr->free_list = freelist->base.off;
136 static void add_to_freetable(struct tdb_context *tdb,
139 struct tle_freelist *freelist)
141 tdb->flist_off = freelist->base.off;
142 add_free_record(tdb, eoff, sizeof(struct tdb_used_record) + elen);
145 static tdb_off_t hbucket_off(tdb_off_t group_start, unsigned ingroup)
148 + (ingroup % (1 << TDB_HASH_GROUP_BITS)) * sizeof(tdb_off_t);
151 /* Get bits from a value. */
152 static uint32_t bits(uint64_t val, unsigned start, unsigned num)
155 return (val >> start) & ((1U << num) - 1);
158 /* We take bits from the top: that way we can lock whole sections of the hash
159 * by using lock ranges. */
160 static uint32_t use_bits(uint64_t h, unsigned num, unsigned *used)
163 return bits(h, 64 - *used, num);
166 static tdb_off_t encode_offset(tdb_off_t new_off, unsigned bucket,
171 | ((uint64_t)bits(h, 64 - TDB_OFF_UPPER_STEAL_EXTRA,
172 TDB_OFF_UPPER_STEAL_EXTRA)
173 << TDB_OFF_HASH_EXTRA_BIT);
176 /* FIXME: Our hash table handling here is primitive: we don't expand! */
177 static void add_to_hashtable(struct tdb_context *tdb,
181 uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
182 tdb_off_t b_off, group_start;
183 unsigned i, group, in_group;
186 group = use_bits(h, TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS, &used);
187 in_group = use_bits(h, TDB_HASH_GROUP_BITS, &used);
189 group_start = offsetof(struct tdb_header, hashtable)
190 + group * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
192 for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
193 unsigned bucket = (in_group + i) % (1 << TDB_HASH_GROUP_BITS);
195 b_off = hbucket_off(group_start, bucket);
196 if (tdb_read_off(tdb, b_off) == 0) {
197 tdb_write_off(tdb, b_off,
198 encode_offset(eoff, bucket, h));
205 static struct tle_freelist *find_flist(struct tdb_layout *layout, unsigned num)
209 for (i = 0; i < layout->num_elems; i++) {
210 if (layout->elem[i].base.type != FREELIST)
213 return &layout->elem[i].flist;
219 /* FIXME: Support TDB_CONVERT */
220 struct tdb_context *tdb_layout_get(struct tdb_layout *layout)
223 tdb_off_t off, len, last_flist;
225 struct tdb_context *tdb;
227 off = sizeof(struct tdb_header);
229 /* First pass of layout: calc lengths */
230 for (i = 0; i < layout->num_elems; i++) {
231 union tdb_layout_elem *e = &layout->elem[i];
233 switch (e->base.type) {
235 len = freelist_len(&e->flist);
238 len = free_record_len(e->free.len);
241 len = data_record_len(&e->used);
244 len = hashtable_len(&e->hashtable);
253 /* Now populate our header, cribbing from a real TDB header. */
254 tdb = tdb_open(NULL, TDB_INTERNAL, O_RDWR, 0, &tap_log_attr);
255 memcpy(mem, tdb->map_ptr, sizeof(struct tdb_header));
257 /* Mug the tdb we have to make it use this. */
263 for (i = 0; i < layout->num_elems; i++) {
264 union tdb_layout_elem *e = &layout->elem[i];
265 switch (e->base.type) {
267 set_freelist(mem + e->base.off, tdb, &e->flist,
268 (struct tdb_header *)mem, last_flist);
269 last_flist = e->base.off;
272 set_free_record(mem + e->base.off, e->free.len);
275 set_data_record(mem + e->base.off, tdb, &e->used);
278 set_hashtable(mem + e->base.off, tdb, &e->hashtable);
282 /* Must have a free list! */
285 /* Now fill the free and hash tables. */
286 for (i = 0; i < layout->num_elems; i++) {
287 union tdb_layout_elem *e = &layout->elem[i];
288 switch (e->base.type) {
290 add_to_freetable(tdb, e->base.off, e->free.len,
291 find_flist(layout, e->free.flist_num));
294 add_to_hashtable(tdb, e->base.off, e->used.key);
301 tdb->flist_off = find_flist(layout, 0)->base.off;
303 /* Get physical if they asked for it. */
304 if (layout->filename) {
305 int fd = open(layout->filename, O_WRONLY|O_TRUNC|O_CREAT,
308 err(1, "opening %s for writing", layout->filename);
309 if (write(fd, tdb->map_ptr, tdb->map_size) != tdb->map_size)
310 err(1, "writing %s", layout->filename);
313 /* NOMMAP is for lockcheck. */
314 tdb = tdb_open(layout->filename, TDB_NOMMAP, O_RDWR, 0,