git.ozlabs.org Git - ccan/blob - ccan/tdb2/test/layout.c

   1 /* TDB tools to create various canned database layouts. */
   2 #include "layout.h"
   3 #include <stdlib.h>
   4 #include <string.h>
   5 #include <assert.h>
   6 #include <err.h>
   7 #include "logging.h"
   8
   9 struct tdb_layout *new_tdb_layout(const char *filename)
  10 {
  11         struct tdb_layout *layout = malloc(sizeof(*layout));
  12         layout->filename = filename;
  13         layout->num_elems = 0;
  14         layout->elem = NULL;
  15         return layout;
  16 }
  17
  18 static void add(struct tdb_layout *layout, union tdb_layout_elem elem)
  19 {
  20         layout->elem = realloc(layout->elem,
  21                                sizeof(layout->elem[0])
  22                                * (layout->num_elems+1));
  23         layout->elem[layout->num_elems++] = elem;
  24 }
  25
  26 void tdb_layout_add_freetable(struct tdb_layout *layout)
  27 {
  28         union tdb_layout_elem elem;
  29         elem.base.type = FREETABLE;
  30         add(layout, elem);
  31 }
  32
  33 void tdb_layout_add_free(struct tdb_layout *layout, tdb_len_t len,
  34                          unsigned ftable)
  35 {
  36         union tdb_layout_elem elem;
  37         elem.base.type = FREE;
  38         elem.free.len = len;
  39         elem.free.ftable_num = ftable;
  40         add(layout, elem);
  41 }
  42
  43 static struct tdb_data dup_key(struct tdb_data key)
  44 {
  45         struct tdb_data ret;
  46         ret.dsize = key.dsize;
  47         ret.dptr = malloc(ret.dsize);
  48         memcpy(ret.dptr, key.dptr, ret.dsize);
  49         return ret;
  50 }
  51
  52 void tdb_layout_add_used(struct tdb_layout *layout,
  53                          TDB_DATA key, TDB_DATA data,
  54                          tdb_len_t extra)
  55 {
  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;
  61         add(layout, elem);
  62 }
  63
  64 static tdb_len_t free_record_len(tdb_len_t len)
  65 {
  66         return sizeof(struct tdb_used_record) + len;
  67 }
  68
  69 static tdb_len_t data_record_len(struct tle_used *used)
  70 {
  71         tdb_len_t len;
  72         len = sizeof(struct tdb_used_record)
  73                 + used->key.dsize + used->data.dsize + used->extra;
  74         assert(len >= sizeof(struct tdb_free_record));
  75         return len;
  76 }
  77
  78 static tdb_len_t hashtable_len(struct tle_hashtable *htable)
  79 {
  80         return sizeof(struct tdb_used_record)
  81                 + (sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS)
  82                 + htable->extra;
  83 }
  84
  85 static tdb_len_t freetable_len(struct tle_freetable *ftable)
  86 {
  87         return sizeof(struct tdb_freetable);
  88 }
  89
  90 static void set_free_record(void *mem, tdb_len_t len)
  91 {
  92         /* We do all the work in add_to_freetable */
  93 }
  94
  95 static void add_zero_pad(struct tdb_used_record *u, size_t len, size_t extra)
  96 {
  97         if (extra)
  98                 ((char *)(u + 1))[len] = '\0';
  99 }
 100
 101 static void set_data_record(void *mem, struct tdb_context *tdb,
 102                             struct tle_used *used)
 103 {
 104         struct tdb_used_record *u = mem;
 105
 106         set_header(tdb, u, TDB_USED_MAGIC, used->key.dsize, used->data.dsize,
 107                    used->key.dsize + used->data.dsize + used->extra,
 108                    tdb_hash(tdb, used->key.dptr, used->key.dsize));
 109         memcpy(u + 1, used->key.dptr, used->key.dsize);
 110         memcpy((char *)(u + 1) + used->key.dsize,
 111                used->data.dptr, used->data.dsize);
 112         add_zero_pad(u, used->key.dsize + used->data.dsize, used->extra);
 113 }
 114
 115 static void set_hashtable(void *mem, struct tdb_context *tdb,
 116                           struct tle_hashtable *htable)
 117 {
 118         struct tdb_used_record *u = mem;
 119         tdb_len_t len = sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS;
 120
 121         set_header(tdb, u, TDB_HTABLE_MAGIC, 0, len, len + htable->extra, 0);
 122         memset(u + 1, 0, len);
 123         add_zero_pad(u, len, htable->extra);
 124 }
 125
 126 static void set_freetable(void *mem, struct tdb_context *tdb,
 127                          struct tle_freetable *freetable, struct tdb_header *hdr,
 128                          tdb_off_t last_ftable)
 129 {
 130         struct tdb_freetable *ftable = mem;
 131         memset(ftable, 0, sizeof(*ftable));
 132         set_header(tdb, &ftable->hdr, TDB_FTABLE_MAGIC, 0,
 133                         sizeof(*ftable) - sizeof(ftable->hdr),
 134                         sizeof(*ftable) - sizeof(ftable->hdr), 0);
 135
 136         if (last_ftable) {
 137                 ftable = (struct tdb_freetable *)((char *)hdr + last_ftable);
 138                 ftable->next = freetable->base.off;
 139         } else {
 140                 hdr->free_table = freetable->base.off;
 141         }
 142 }
 143
 144 static void add_to_freetable(struct tdb_context *tdb,
 145                              tdb_off_t eoff,
 146                              tdb_off_t elen,
 147                              unsigned ftable,
 148                              struct tle_freetable *freetable)
 149 {
 150         tdb->tdb2.ftable_off = freetable->base.off;
 151         tdb->tdb2.ftable = ftable;
 152         add_free_record(tdb, eoff, sizeof(struct tdb_used_record) + elen,
 153                         TDB_LOCK_WAIT, false);
 154 }
 155
 156 static tdb_off_t hbucket_off(tdb_off_t group_start, unsigned ingroup)
 157 {
 158         return group_start
 159                 + (ingroup % (1 << TDB_HASH_GROUP_BITS)) * sizeof(tdb_off_t);
 160 }
 161
 162 /* Get bits from a value. */
 163 static uint32_t bits(uint64_t val, unsigned start, unsigned num)
 164 {
 165         assert(num <= 32);
 166         return (val >> start) & ((1U << num) - 1);
 167 }
 168
 169 /* We take bits from the top: that way we can lock whole sections of the hash
 170  * by using lock ranges. */
 171 static uint32_t use_bits(uint64_t h, unsigned num, unsigned *used)
 172 {
 173         *used += num;
 174         return bits(h, 64 - *used, num);
 175 }
 176
 177 static tdb_off_t encode_offset(tdb_off_t new_off, unsigned bucket,
 178                                uint64_t h)
 179 {
 180         return bucket
 181                 | new_off
 182                 | ((uint64_t)bits(h, 64 - TDB_OFF_UPPER_STEAL_EXTRA,
 183                                   TDB_OFF_UPPER_STEAL_EXTRA)
 184                    << TDB_OFF_HASH_EXTRA_BIT);
 185 }
 186
 187 /* FIXME: Our hash table handling here is primitive: we don't expand! */
 188 static void add_to_hashtable(struct tdb_context *tdb,
 189                              tdb_off_t eoff,
 190                              struct tdb_data key)
 191 {
 192         uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
 193         tdb_off_t b_off, group_start;
 194         unsigned i, group, in_group;
 195         unsigned used = 0;
 196
 197         group = use_bits(h, TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS, &used);
 198         in_group = use_bits(h, TDB_HASH_GROUP_BITS, &used);
 199
 200         group_start = offsetof(struct tdb_header, hashtable)
 201                 + group * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
 202
 203         for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
 204                 unsigned bucket = (in_group + i) % (1 << TDB_HASH_GROUP_BITS);
 205
 206                 b_off = hbucket_off(group_start, bucket);
 207                 if (tdb_read_off(tdb, b_off) == 0) {
 208                         tdb_write_off(tdb, b_off,
 209                                       encode_offset(eoff, in_group, h));
 210                         return;
 211                 }
 212         }
 213         abort();
 214 }
 215
 216 static struct tle_freetable *find_ftable(struct tdb_layout *layout, unsigned num)
 217 {
 218         unsigned i;
 219
 220         for (i = 0; i < layout->num_elems; i++) {
 221                 if (layout->elem[i].base.type != FREETABLE)
 222                         continue;
 223                 if (num == 0)
 224                         return &layout->elem[i].ftable;
 225                 num--;
 226         }
 227         abort();
 228 }
 229
 230 /* FIXME: Support TDB_CONVERT */
 231 struct tdb_context *tdb_layout_get(struct tdb_layout *layout,
 232                                    union tdb_attribute *attr)
 233 {
 234         unsigned int i;
 235         tdb_off_t off, len, last_ftable;
 236         char *mem;
 237         struct tdb_context *tdb;
 238
 239         off = sizeof(struct tdb_header);
 240
 241         /* First pass of layout: calc lengths */
 242         for (i = 0; i < layout->num_elems; i++) {
 243                 union tdb_layout_elem *e = &layout->elem[i];
 244                 e->base.off = off;
 245                 switch (e->base.type) {
 246                 case FREETABLE:
 247                         len = freetable_len(&e->ftable);
 248                         break;
 249                 case FREE:
 250                         len = free_record_len(e->free.len);
 251                         break;
 252                 case DATA:
 253                         len = data_record_len(&e->used);
 254                         break;
 255                 case HASHTABLE:
 256                         len = hashtable_len(&e->hashtable);
 257                         break;
 258                 default:
 259                         abort();
 260                 }
 261                 off += len;
 262         }
 263
 264         mem = malloc(off);
 265         /* Fill with some weird pattern. */
 266         memset(mem, 0x99, off);
 267         /* Now populate our header, cribbing from a real TDB header. */
 268         tdb = tdb_open(NULL, TDB_INTERNAL, O_RDWR, 0, attr);
 269         memcpy(mem, tdb->file->map_ptr, sizeof(struct tdb_header));
 270
 271         /* Mug the tdb we have to make it use this. */
 272         free(tdb->file->map_ptr);
 273         tdb->file->map_ptr = mem;
 274         tdb->file->map_size = off;
 275
 276         last_ftable = 0;
 277         for (i = 0; i < layout->num_elems; i++) {
 278                 union tdb_layout_elem *e = &layout->elem[i];
 279                 switch (e->base.type) {
 280                 case FREETABLE:
 281                         set_freetable(mem + e->base.off, tdb, &e->ftable,
 282                                      (struct tdb_header *)mem, last_ftable);
 283                         last_ftable = e->base.off;
 284                         break;
 285                 case FREE:
 286                         set_free_record(mem + e->base.off, e->free.len);
 287                         break;
 288                 case DATA:
 289                         set_data_record(mem + e->base.off, tdb, &e->used);
 290                         break;
 291                 case HASHTABLE:
 292                         set_hashtable(mem + e->base.off, tdb, &e->hashtable);
 293                         break;
 294                 }
 295         }
 296         /* Must have a free table! */
 297         assert(last_ftable);
 298
 299         /* Now fill the free and hash tables. */
 300         for (i = 0; i < layout->num_elems; i++) {
 301                 union tdb_layout_elem *e = &layout->elem[i];
 302                 switch (e->base.type) {
 303                 case FREE:
 304                         add_to_freetable(tdb, e->base.off, e->free.len,
 305                                          e->free.ftable_num,
 306                                          find_ftable(layout, e->free.ftable_num));
 307                         break;
 308                 case DATA:
 309                         add_to_hashtable(tdb, e->base.off, e->used.key);
 310                         break;
 311                 default:
 312                         break;
 313                 }
 314         }
 315
 316         tdb->tdb2.ftable_off = find_ftable(layout, 0)->base.off;
 317
 318         /* Get physical if they asked for it. */
 319         if (layout->filename) {
 320                 int fd = open(layout->filename, O_WRONLY|O_TRUNC|O_CREAT,
 321                               0600);
 322                 if (fd < 0)
 323                         err(1, "opening %s for writing", layout->filename);
 324                 if (write(fd, tdb->file->map_ptr, tdb->file->map_size)
 325                     != tdb->file->map_size)
 326                         err(1, "writing %s", layout->filename);
 327                 close(fd);
 328                 tdb_close(tdb);
 329                 /* NOMMAP is for lockcheck. */
 330                 tdb = tdb_open(layout->filename, TDB_NOMMAP, O_RDWR, 0,
 331                                &tap_log_attr);
 332         }
 333
 334         return tdb;
 335 }
 336
 337 void tdb_layout_free(struct tdb_layout *layout)
 338 {
 339         unsigned int i;
 340
 341         for (i = 0; i < layout->num_elems; i++) {
 342                 if (layout->elem[i].base.type == DATA) {
 343                         free(layout->elem[i].used.key.dptr);
 344                         free(layout->elem[i].used.data.dptr);
 345                 }
 346         }
 347         free(layout->elem);
 348         free(layout);
 349 }