pack-revindex.c

   1 #include "cache.h"
   2 #include "pack-revindex.h"
   3 #include "object-store.h"
   4 #include "packfile.h"
   5 #include "config.h"
   6 #include "midx.h"
   7
   8 struct revindex_entry {
   9         off_t offset;
  10         unsigned int nr;
  11 };
  12
  13 /*
  14  * Pack index for existing packs give us easy access to the offsets into
  15  * corresponding pack file where each object's data starts, but the entries
  16  * do not store the size of the compressed representation (uncompressed
  17  * size is easily available by examining the pack entry header).  It is
  18  * also rather expensive to find the sha1 for an object given its offset.
  19  *
  20  * The pack index file is sorted by object name mapping to offset;
  21  * this revindex array is a list of offset/index_nr pairs
  22  * ordered by offset, so if you know the offset of an object, next offset
  23  * is where its packed representation ends and the index_nr can be used to
  24  * get the object sha1 from the main index.
  25  */
  26
  27 /*
  28  * This is a least-significant-digit radix sort.
  29  *
  30  * It sorts each of the "n" items in "entries" by its offset field. The "max"
  31  * parameter must be at least as large as the largest offset in the array,
  32  * and lets us quit the sort early.
  33  */
  34 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
  35 {
  36         /*
  37          * We use a "digit" size of 16 bits. That keeps our memory
  38          * usage reasonable, and we can generally (for a 4G or smaller
  39          * packfile) quit after two rounds of radix-sorting.
  40          */
  41 #define DIGIT_SIZE (16)
  42 #define BUCKETS (1 << DIGIT_SIZE)
  43         /*
  44          * We want to know the bucket that a[i] will go into when we are using
  45          * the digit that is N bits from the (least significant) end.
  46          */
  47 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
  48
  49         /*
  50          * We need O(n) temporary storage. Rather than do an extra copy of the
  51          * partial results into "entries", we sort back and forth between the
  52          * real array and temporary storage. In each iteration of the loop, we
  53          * keep track of them with alias pointers, always sorting from "from"
  54          * to "to".
  55          */
  56         struct revindex_entry *tmp, *from, *to;
  57         int bits;
  58         unsigned *pos;
  59
  60         ALLOC_ARRAY(pos, BUCKETS);
  61         ALLOC_ARRAY(tmp, n);
  62         from = entries;
  63         to = tmp;
  64
  65         /*
  66          * If (max >> bits) is zero, then we know that the radix digit we are
  67          * on (and any higher) will be zero for all entries, and our loop will
  68          * be a no-op, as everybody lands in the same zero-th bucket.
  69          */
  70         for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
  71                 unsigned i;
  72
  73                 memset(pos, 0, BUCKETS * sizeof(*pos));
  74
  75                 /*
  76                  * We want pos[i] to store the index of the last element that
  77                  * will go in bucket "i" (actually one past the last element).
  78                  * To do this, we first count the items that will go in each
  79                  * bucket, which gives us a relative offset from the last
  80                  * bucket. We can then cumulatively add the index from the
  81                  * previous bucket to get the true index.
  82                  */
  83                 for (i = 0; i < n; i++)
  84                         pos[BUCKET_FOR(from, i, bits)]++;
  85                 for (i = 1; i < BUCKETS; i++)
  86                         pos[i] += pos[i-1];
  87
  88                 /*
  89                  * Now we can drop the elements into their correct buckets (in
  90                  * our temporary array).  We iterate the pos counter backwards
  91                  * to avoid using an extra index to count up. And since we are
  92                  * going backwards there, we must also go backwards through the
  93                  * array itself, to keep the sort stable.
  94                  *
  95                  * Note that we use an unsigned iterator to make sure we can
  96                  * handle 2^32-1 objects, even on a 32-bit system. But this
  97                  * means we cannot use the more obvious "i >= 0" loop condition
  98                  * for counting backwards, and must instead check for
  99                  * wrap-around with UINT_MAX.
 100                  */
 101                 for (i = n - 1; i != UINT_MAX; i--)
 102                         to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
 103
 104                 /*
 105                  * Now "to" contains the most sorted list, so we swap "from" and
 106                  * "to" for the next iteration.
 107                  */
 108                 SWAP(from, to);
 109         }
 110
 111         /*
 112          * If we ended with our data in the original array, great. If not,
 113          * we have to move it back from the temporary storage.
 114          */
 115         if (from != entries)
 116                 COPY_ARRAY(entries, tmp, n);
 117         free(tmp);
 118         free(pos);
 119
 120 #undef BUCKET_FOR
 121 #undef BUCKETS
 122 #undef DIGIT_SIZE
 123 }
 124
 125 /*
 126  * Ordered list of offsets of objects in the pack.
 127  */
 128 static void create_pack_revindex(struct packed_git *p)
 129 {
 130         const unsigned num_ent = p->num_objects;
 131         unsigned i;
 132         const char *index = p->index_data;
 133         const unsigned hashsz = the_hash_algo->rawsz;
 134
 135         ALLOC_ARRAY(p->revindex, num_ent + 1);
 136         index += 4 * 256;
 137
 138         if (p->index_version > 1) {
 139                 const uint32_t *off_32 =
 140                         (uint32_t *)(index + 8 + (size_t)p->num_objects * (hashsz + 4));
 141                 const uint32_t *off_64 = off_32 + p->num_objects;
 142                 for (i = 0; i < num_ent; i++) {
 143                         const uint32_t off = ntohl(*off_32++);
 144                         if (!(off & 0x80000000)) {
 145                                 p->revindex[i].offset = off;
 146                         } else {
 147                                 p->revindex[i].offset = get_be64(off_64);
 148                                 off_64 += 2;
 149                         }
 150                         p->revindex[i].nr = i;
 151                 }
 152         } else {
 153                 for (i = 0; i < num_ent; i++) {
 154                         const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i));
 155                         p->revindex[i].offset = ntohl(hl);
 156                         p->revindex[i].nr = i;
 157                 }
 158         }
 159
 160         /*
 161          * This knows the pack format -- the hash trailer
 162          * follows immediately after the last object data.
 163          */
 164         p->revindex[num_ent].offset = p->pack_size - hashsz;
 165         p->revindex[num_ent].nr = -1;
 166         sort_revindex(p->revindex, num_ent, p->pack_size);
 167 }
 168
 169 static int create_pack_revindex_in_memory(struct packed_git *p)
 170 {
 171         if (git_env_bool(GIT_TEST_REV_INDEX_DIE_IN_MEMORY, 0))
 172                 die("dying as requested by '%s'",
 173                     GIT_TEST_REV_INDEX_DIE_IN_MEMORY);
 174         if (open_pack_index(p))
 175                 return -1;
 176         create_pack_revindex(p);
 177         return 0;
 178 }
 179
 180 static char *pack_revindex_filename(struct packed_git *p)
 181 {
 182         size_t len;
 183         if (!strip_suffix(p->pack_name, ".pack", &len))
 184                 BUG("pack_name does not end in .pack");
 185         return xstrfmt("%.*s.rev", (int)len, p->pack_name);
 186 }
 187
 188 #define RIDX_HEADER_SIZE (12)
 189 #define RIDX_MIN_SIZE (RIDX_HEADER_SIZE + (2 * the_hash_algo->rawsz))
 190
 191 struct revindex_header {
 192         uint32_t signature;
 193         uint32_t version;
 194         uint32_t hash_id;
 195 };
 196
 197 static int load_revindex_from_disk(char *revindex_name,
 198                                    uint32_t num_objects,
 199                                    const uint32_t **data_p, size_t *len_p)
 200 {
 201         int fd, ret = 0;
 202         struct stat st;
 203         void *data = NULL;
 204         size_t revindex_size;
 205         struct revindex_header *hdr;
 206
 207         fd = git_open(revindex_name);
 208
 209         if (fd < 0) {
 210                 ret = -1;
 211                 goto cleanup;
 212         }
 213         if (fstat(fd, &st)) {
 214                 ret = error_errno(_("failed to read %s"), revindex_name);
 215                 goto cleanup;
 216         }
 217
 218         revindex_size = xsize_t(st.st_size);
 219
 220         if (revindex_size < RIDX_MIN_SIZE) {
 221                 ret = error(_("reverse-index file %s is too small"), revindex_name);
 222                 goto cleanup;
 223         }
 224
 225         if (revindex_size - RIDX_MIN_SIZE != st_mult(sizeof(uint32_t), num_objects)) {
 226                 ret = error(_("reverse-index file %s is corrupt"), revindex_name);
 227                 goto cleanup;
 228         }
 229
 230         data = xmmap(NULL, revindex_size, PROT_READ, MAP_PRIVATE, fd, 0);
 231         hdr = data;
 232
 233         if (ntohl(hdr->signature) != RIDX_SIGNATURE) {
 234                 ret = error(_("reverse-index file %s has unknown signature"), revindex_name);
 235                 goto cleanup;
 236         }
 237         if (ntohl(hdr->version) != 1) {
 238                 ret = error(_("reverse-index file %s has unsupported version %"PRIu32),
 239                             revindex_name, ntohl(hdr->version));
 240                 goto cleanup;
 241         }
 242         if (!(ntohl(hdr->hash_id) == 1 || ntohl(hdr->hash_id) == 2)) {
 243                 ret = error(_("reverse-index file %s has unsupported hash id %"PRIu32),
 244                             revindex_name, ntohl(hdr->hash_id));
 245                 goto cleanup;
 246         }
 247
 248 cleanup:
 249         if (ret) {
 250                 if (data)
 251                         munmap(data, revindex_size);
 252         } else {
 253                 *len_p = revindex_size;
 254                 *data_p = (const uint32_t *)data;
 255         }
 256
 257         if (fd >= 0)
 258                 close(fd);
 259         return ret;
 260 }
 261
 262 static int load_pack_revindex_from_disk(struct packed_git *p)
 263 {
 264         char *revindex_name;
 265         int ret;
 266         if (open_pack_index(p))
 267                 return -1;
 268
 269         revindex_name = pack_revindex_filename(p);
 270
 271         ret = load_revindex_from_disk(revindex_name,
 272                                       p->num_objects,
 273                                       &p->revindex_map,
 274                                       &p->revindex_size);
 275         if (ret)
 276                 goto cleanup;
 277
 278         p->revindex_data = (const uint32_t *)((const char *)p->revindex_map + RIDX_HEADER_SIZE);
 279
 280 cleanup:
 281         free(revindex_name);
 282         return ret;
 283 }
 284
 285 int load_pack_revindex(struct packed_git *p)
 286 {
 287         if (p->revindex || p->revindex_data)
 288                 return 0;
 289
 290         if (!load_pack_revindex_from_disk(p))
 291                 return 0;
 292         else if (!create_pack_revindex_in_memory(p))
 293                 return 0;
 294         return -1;
 295 }
 296
 297 int load_midx_revindex(struct multi_pack_index *m)
 298 {
 299         char *revindex_name;
 300         int ret;
 301         if (m->revindex_data)
 302                 return 0;
 303
 304         revindex_name = get_midx_rev_filename(m);
 305
 306         ret = load_revindex_from_disk(revindex_name,
 307                                       m->num_objects,
 308                                       &m->revindex_map,
 309                                       &m->revindex_len);
 310         if (ret)
 311                 goto cleanup;
 312
 313         m->revindex_data = (const uint32_t *)((const char *)m->revindex_map + RIDX_HEADER_SIZE);
 314
 315 cleanup:
 316         free(revindex_name);
 317         return ret;
 318 }
 319
 320 int close_midx_revindex(struct multi_pack_index *m)
 321 {
 322         if (!m || !m->revindex_map)
 323                 return 0;
 324
 325         munmap((void*)m->revindex_map, m->revindex_len);
 326
 327         m->revindex_map = NULL;
 328         m->revindex_data = NULL;
 329         m->revindex_len = 0;
 330
 331         return 0;
 332 }
 333
 334 int offset_to_pack_pos(struct packed_git *p, off_t ofs, uint32_t *pos)
 335 {
 336         unsigned lo, hi;
 337
 338         if (load_pack_revindex(p) < 0)
 339                 return -1;
 340
 341         lo = 0;
 342         hi = p->num_objects + 1;
 343
 344         do {
 345                 const unsigned mi = lo + (hi - lo) / 2;
 346                 off_t got = pack_pos_to_offset(p, mi);
 347
 348                 if (got == ofs) {
 349                         *pos = mi;
 350                         return 0;
 351                 } else if (ofs < got)
 352                         hi = mi;
 353                 else
 354                         lo = mi + 1;
 355         } while (lo < hi);
 356
 357         error("bad offset for revindex");
 358         return -1;
 359 }
 360
 361 uint32_t pack_pos_to_index(struct packed_git *p, uint32_t pos)
 362 {
 363         if (!(p->revindex || p->revindex_data))
 364                 BUG("pack_pos_to_index: reverse index not yet loaded");
 365         if (p->num_objects <= pos)
 366                 BUG("pack_pos_to_index: out-of-bounds object at %"PRIu32, pos);
 367
 368         if (p->revindex)
 369                 return p->revindex[pos].nr;
 370         else
 371                 return get_be32(p->revindex_data + pos);
 372 }
 373
 374 off_t pack_pos_to_offset(struct packed_git *p, uint32_t pos)
 375 {
 376         if (!(p->revindex || p->revindex_data))
 377                 BUG("pack_pos_to_index: reverse index not yet loaded");
 378         if (p->num_objects < pos)
 379                 BUG("pack_pos_to_offset: out-of-bounds object at %"PRIu32, pos);
 380
 381         if (p->revindex)
 382                 return p->revindex[pos].offset;
 383         else if (pos == p->num_objects)
 384                 return p->pack_size - the_hash_algo->rawsz;
 385         else
 386                 return nth_packed_object_offset(p, pack_pos_to_index(p, pos));
 387 }
 388
 389 uint32_t pack_pos_to_midx(struct multi_pack_index *m, uint32_t pos)
 390 {
 391         if (!m->revindex_data)
 392                 BUG("pack_pos_to_midx: reverse index not yet loaded");
 393         if (m->num_objects <= pos)
 394                 BUG("pack_pos_to_midx: out-of-bounds object at %"PRIu32, pos);
 395         return get_be32(m->revindex_data + pos);
 396 }
 397
 398 struct midx_pack_key {
 399         uint32_t pack;
 400         off_t offset;
 401
 402         uint32_t preferred_pack;
 403         struct multi_pack_index *midx;
 404 };
 405
 406 static int midx_pack_order_cmp(const void *va, const void *vb)
 407 {
 408         const struct midx_pack_key *key = va;
 409         struct multi_pack_index *midx = key->midx;
 410
 411         uint32_t versus = pack_pos_to_midx(midx, (uint32_t*)vb - (const uint32_t *)midx->revindex_data);
 412         uint32_t versus_pack = nth_midxed_pack_int_id(midx, versus);
 413         off_t versus_offset;
 414
 415         uint32_t key_preferred = key->pack == key->preferred_pack;
 416         uint32_t versus_preferred = versus_pack == key->preferred_pack;
 417
 418         /*
 419          * First, compare the preferred-ness, noting that the preferred pack
 420          * comes first.
 421          */
 422         if (key_preferred && !versus_preferred)
 423                 return -1;
 424         else if (!key_preferred && versus_preferred)
 425                 return 1;
 426
 427         /* Then, break ties first by comparing the pack IDs. */
 428         if (key->pack < versus_pack)
 429                 return -1;
 430         else if (key->pack > versus_pack)
 431                 return 1;
 432
 433         /* Finally, break ties by comparing offsets within a pack. */
 434         versus_offset = nth_midxed_offset(midx, versus);
 435         if (key->offset < versus_offset)
 436                 return -1;
 437         else if (key->offset > versus_offset)
 438                 return 1;
 439
 440         return 0;
 441 }
 442
 443 int midx_to_pack_pos(struct multi_pack_index *m, uint32_t at, uint32_t *pos)
 444 {
 445         struct midx_pack_key key;
 446         uint32_t *found;
 447
 448         if (!m->revindex_data)
 449                 BUG("midx_to_pack_pos: reverse index not yet loaded");
 450         if (m->num_objects <= at)
 451                 BUG("midx_to_pack_pos: out-of-bounds object at %"PRIu32, at);
 452
 453         key.pack = nth_midxed_pack_int_id(m, at);
 454         key.offset = nth_midxed_offset(m, at);
 455         key.midx = m;
 456         /*
 457          * The preferred pack sorts first, so determine its identifier by
 458          * looking at the first object in pseudo-pack order.
 459          *
 460          * Note that if no --preferred-pack is explicitly given when writing a
 461          * multi-pack index, then whichever pack has the lowest identifier
 462          * implicitly is preferred (and includes all its objects, since ties are
 463          * broken first by pack identifier).
 464          */
 465         key.preferred_pack = nth_midxed_pack_int_id(m, pack_pos_to_midx(m, 0));
 466
 467         found = bsearch(&key, m->revindex_data, m->num_objects,
 468                         sizeof(*m->revindex_data), midx_pack_order_cmp);
 469
 470         if (!found)
 471                 return error("bad offset for revindex");
 472
 473         *pos = found - m->revindex_data;
 474         return 0;
 475 }