pack-revindex.c

   1 #include "git-compat-util.h"
   2 #include "gettext.h"
   3 #include "pack-revindex.h"
   4 #include "odb.h"
   5 #include "packfile.h"
   6 #include "strbuf.h"
   7 #include "trace2.h"
   8 #include "parse.h"
   9 #include "repository.h"
  10 #include "midx.h"
  11 #include "csum-file.h"
  12
  13 struct revindex_entry {
  14         off_t offset;
  15         unsigned int nr;
  16 };
  17
  18 /*
  19  * Pack index for existing packs give us easy access to the offsets into
  20  * corresponding pack file where each object's data starts, but the entries
  21  * do not store the size of the compressed representation (uncompressed
  22  * size is easily available by examining the pack entry header).  It is
  23  * also rather expensive to find the sha1 for an object given its offset.
  24  *
  25  * The pack index file is sorted by object name mapping to offset;
  26  * this revindex array is a list of offset/index_nr pairs
  27  * ordered by offset, so if you know the offset of an object, next offset
  28  * is where its packed representation ends and the index_nr can be used to
  29  * get the object sha1 from the main index.
  30  */
  31
  32 /*
  33  * This is a least-significant-digit radix sort.
  34  *
  35  * It sorts each of the "n" items in "entries" by its offset field. The "max"
  36  * parameter must be at least as large as the largest offset in the array,
  37  * and lets us quit the sort early.
  38  */
  39 static void sort_revindex(struct revindex_entry *entries, unsigned n, off_t max)
  40 {
  41         /*
  42          * We use a "digit" size of 16 bits. That keeps our memory
  43          * usage reasonable, and we can generally (for a 4G or smaller
  44          * packfile) quit after two rounds of radix-sorting.
  45          */
  46 #define DIGIT_SIZE (16)
  47 #define BUCKETS (1 << DIGIT_SIZE)
  48         /*
  49          * We want to know the bucket that a[i] will go into when we are using
  50          * the digit that is N bits from the (least significant) end.
  51          */
  52 #define BUCKET_FOR(a, i, bits) (((a)[(i)].offset >> (bits)) & (BUCKETS-1))
  53
  54         /*
  55          * We need O(n) temporary storage. Rather than do an extra copy of the
  56          * partial results into "entries", we sort back and forth between the
  57          * real array and temporary storage. In each iteration of the loop, we
  58          * keep track of them with alias pointers, always sorting from "from"
  59          * to "to".
  60          */
  61         struct revindex_entry *tmp, *from, *to;
  62         int bits;
  63         unsigned *pos;
  64
  65         ALLOC_ARRAY(pos, BUCKETS);
  66         ALLOC_ARRAY(tmp, n);
  67         from = entries;
  68         to = tmp;
  69
  70         /*
  71          * If (max >> bits) is zero, then we know that the radix digit we are
  72          * on (and any higher) will be zero for all entries, and our loop will
  73          * be a no-op, as everybody lands in the same zero-th bucket.
  74          */
  75         for (bits = 0; max >> bits; bits += DIGIT_SIZE) {
  76                 unsigned i;
  77
  78                 memset(pos, 0, BUCKETS * sizeof(*pos));
  79
  80                 /*
  81                  * We want pos[i] to store the index of the last element that
  82                  * will go in bucket "i" (actually one past the last element).
  83                  * To do this, we first count the items that will go in each
  84                  * bucket, which gives us a relative offset from the last
  85                  * bucket. We can then cumulatively add the index from the
  86                  * previous bucket to get the true index.
  87                  */
  88                 for (i = 0; i < n; i++)
  89                         pos[BUCKET_FOR(from, i, bits)]++;
  90                 for (i = 1; i < BUCKETS; i++)
  91                         pos[i] += pos[i-1];
  92
  93                 /*
  94                  * Now we can drop the elements into their correct buckets (in
  95                  * our temporary array).  We iterate the pos counter backwards
  96                  * to avoid using an extra index to count up. And since we are
  97                  * going backwards there, we must also go backwards through the
  98                  * array itself, to keep the sort stable.
  99                  *
 100                  * Note that we use an unsigned iterator to make sure we can
 101                  * handle 2^32-1 objects, even on a 32-bit system. But this
 102                  * means we cannot use the more obvious "i >= 0" loop condition
 103                  * for counting backwards, and must instead check for
 104                  * wrap-around with UINT_MAX.
 105                  */
 106                 for (i = n - 1; i != UINT_MAX; i--)
 107                         to[--pos[BUCKET_FOR(from, i, bits)]] = from[i];
 108
 109                 /*
 110                  * Now "to" contains the most sorted list, so we swap "from" and
 111                  * "to" for the next iteration.
 112                  */
 113                 SWAP(from, to);
 114         }
 115
 116         /*
 117          * If we ended with our data in the original array, great. If not,
 118          * we have to move it back from the temporary storage.
 119          */
 120         if (from != entries)
 121                 COPY_ARRAY(entries, tmp, n);
 122         free(tmp);
 123         free(pos);
 124
 125 #undef BUCKET_FOR
 126 #undef BUCKETS
 127 #undef DIGIT_SIZE
 128 }
 129
 130 /*
 131  * Ordered list of offsets of objects in the pack.
 132  */
 133 static void create_pack_revindex(struct packed_git *p)
 134 {
 135         const unsigned num_ent = p->num_objects;
 136         unsigned i;
 137         const char *index = p->index_data;
 138         const unsigned hashsz = p->repo->hash_algo->rawsz;
 139
 140         ALLOC_ARRAY(p->revindex, num_ent + 1);
 141         index += 4 * 256;
 142
 143         if (p->index_version > 1) {
 144                 const uint32_t *off_32 =
 145                         (uint32_t *)(index + 8 + (size_t)p->num_objects * (hashsz + 4));
 146                 const uint32_t *off_64 = off_32 + p->num_objects;
 147                 for (i = 0; i < num_ent; i++) {
 148                         const uint32_t off = ntohl(*off_32++);
 149                         if (!(off & 0x80000000)) {
 150                                 p->revindex[i].offset = off;
 151                         } else {
 152                                 p->revindex[i].offset = get_be64(off_64);
 153                                 off_64 += 2;
 154                         }
 155                         p->revindex[i].nr = i;
 156                 }
 157         } else {
 158                 for (i = 0; i < num_ent; i++) {
 159                         const uint32_t hl = *((uint32_t *)(index + (hashsz + 4) * i));
 160                         p->revindex[i].offset = ntohl(hl);
 161                         p->revindex[i].nr = i;
 162                 }
 163         }
 164
 165         /*
 166          * This knows the pack format -- the hash trailer
 167          * follows immediately after the last object data.
 168          */
 169         p->revindex[num_ent].offset = p->pack_size - hashsz;
 170         p->revindex[num_ent].nr = -1;
 171         sort_revindex(p->revindex, num_ent, p->pack_size);
 172 }
 173
 174 static int create_pack_revindex_in_memory(struct packed_git *p)
 175 {
 176         if (git_env_bool(GIT_TEST_REV_INDEX_DIE_IN_MEMORY, 0))
 177                 die("dying as requested by '%s'",
 178                     GIT_TEST_REV_INDEX_DIE_IN_MEMORY);
 179         if (open_pack_index(p))
 180                 return -1;
 181         create_pack_revindex(p);
 182         return 0;
 183 }
 184
 185 static char *pack_revindex_filename(struct packed_git *p)
 186 {
 187         size_t len;
 188         if (!strip_suffix(p->pack_name, ".pack", &len))
 189                 BUG("pack_name does not end in .pack");
 190         return xstrfmt("%.*s.rev", (int)len, p->pack_name);
 191 }
 192
 193 #define RIDX_HEADER_SIZE (12)
 194
 195 static size_t ridx_min_size(const struct git_hash_algo *algo)
 196 {
 197         return RIDX_HEADER_SIZE + (2 * algo->rawsz);
 198 }
 199
 200 struct revindex_header {
 201         uint32_t signature;
 202         uint32_t version;
 203         uint32_t hash_id;
 204 };
 205
 206 static int load_revindex_from_disk(const struct git_hash_algo *algo,
 207                                    char *revindex_name,
 208                                    uint32_t num_objects,
 209                                    const uint32_t **data_p, size_t *len_p)
 210 {
 211         int fd, ret = 0;
 212         struct stat st;
 213         void *data = NULL;
 214         size_t revindex_size;
 215         struct revindex_header *hdr;
 216
 217         if (git_env_bool(GIT_TEST_REV_INDEX_DIE_ON_DISK, 0))
 218                 die("dying as requested by '%s'", GIT_TEST_REV_INDEX_DIE_ON_DISK);
 219
 220         fd = git_open(revindex_name);
 221
 222         if (fd < 0) {
 223                 /* "No file" means return 1. */
 224                 ret = 1;
 225                 goto cleanup;
 226         }
 227         if (fstat(fd, &st)) {
 228                 ret = error_errno(_("failed to read %s"), revindex_name);
 229                 goto cleanup;
 230         }
 231
 232         revindex_size = xsize_t(st.st_size);
 233
 234         if (revindex_size < ridx_min_size(algo)) {
 235                 ret = error(_("reverse-index file %s is too small"), revindex_name);
 236                 goto cleanup;
 237         }
 238
 239         if (revindex_size - ridx_min_size(algo) != st_mult(sizeof(uint32_t), num_objects)) {
 240                 ret = error(_("reverse-index file %s is corrupt"), revindex_name);
 241                 goto cleanup;
 242         }
 243
 244         data = xmmap(NULL, revindex_size, PROT_READ, MAP_PRIVATE, fd, 0);
 245         hdr = data;
 246
 247         if (ntohl(hdr->signature) != RIDX_SIGNATURE) {
 248                 ret = error(_("reverse-index file %s has unknown signature"), revindex_name);
 249                 goto cleanup;
 250         }
 251         if (ntohl(hdr->version) != 1) {
 252                 ret = error(_("reverse-index file %s has unsupported version %"PRIu32),
 253                             revindex_name, ntohl(hdr->version));
 254                 goto cleanup;
 255         }
 256         if (!(ntohl(hdr->hash_id) == 1 || ntohl(hdr->hash_id) == 2)) {
 257                 ret = error(_("reverse-index file %s has unsupported hash id %"PRIu32),
 258                             revindex_name, ntohl(hdr->hash_id));
 259                 goto cleanup;
 260         }
 261
 262 cleanup:
 263         if (ret) {
 264                 if (data)
 265                         munmap(data, revindex_size);
 266         } else {
 267                 *len_p = revindex_size;
 268                 *data_p = (const uint32_t *)data;
 269         }
 270
 271         if (fd >= 0)
 272                 close(fd);
 273         return ret;
 274 }
 275
 276 int load_pack_revindex_from_disk(struct packed_git *p)
 277 {
 278         char *revindex_name;
 279         int ret;
 280         if (open_pack_index(p))
 281                 return -1;
 282
 283         revindex_name = pack_revindex_filename(p);
 284
 285         ret = load_revindex_from_disk(p->repo->hash_algo,
 286                                       revindex_name,
 287                                       p->num_objects,
 288                                       &p->revindex_map,
 289                                       &p->revindex_size);
 290         if (ret)
 291                 goto cleanup;
 292
 293         p->revindex_data = (const uint32_t *)((const char *)p->revindex_map + RIDX_HEADER_SIZE);
 294
 295 cleanup:
 296         free(revindex_name);
 297         return ret;
 298 }
 299
 300 int load_pack_revindex(struct repository *r, struct packed_git *p)
 301 {
 302         if (p->revindex || p->revindex_data)
 303                 return 0;
 304
 305         prepare_repo_settings(r);
 306
 307         if (r->settings.pack_read_reverse_index &&
 308             !load_pack_revindex_from_disk(p))
 309                 return 0;
 310         else if (!create_pack_revindex_in_memory(p))
 311                 return 0;
 312         return -1;
 313 }
 314
 315 /*
 316  * verify_pack_revindex verifies that the on-disk rev-index for the given
 317  * pack-file is the same that would be created if written from scratch.
 318  *
 319  * A negative number is returned on error.
 320  */
 321 int verify_pack_revindex(struct packed_git *p)
 322 {
 323         int res = 0;
 324
 325         /* Do not bother checking if not initialized. */
 326         if (!p->revindex_map || !p->revindex_data)
 327                 return res;
 328
 329         if (!hashfile_checksum_valid(p->repo->hash_algo,
 330                                      (const unsigned char *)p->revindex_map, p->revindex_size)) {
 331                 error(_("invalid checksum"));
 332                 res = -1;
 333         }
 334
 335         /* This may fail due to a broken .idx. */
 336         if (create_pack_revindex_in_memory(p))
 337                 return res;
 338
 339         for (size_t i = 0; i < p->num_objects; i++) {
 340                 uint32_t nr = p->revindex[i].nr;
 341                 uint32_t rev_val = get_be32(p->revindex_data + i);
 342
 343                 if (nr != rev_val) {
 344                         error(_("invalid rev-index position at %"PRIu64": %"PRIu32" != %"PRIu32""),
 345                               (uint64_t)i, nr, rev_val);
 346                         res = -1;
 347                 }
 348         }
 349
 350         return res;
 351 }
 352
 353 static int can_use_midx_ridx_chunk(struct multi_pack_index *m)
 354 {
 355         if (!m->chunk_revindex)
 356                 return 0;
 357         if (m->chunk_revindex_len != st_mult(sizeof(uint32_t), m->num_objects)) {
 358                 error(_("multi-pack-index reverse-index chunk is the wrong size"));
 359                 return 0;
 360         }
 361         return 1;
 362 }
 363
 364 int load_midx_revindex(struct multi_pack_index *m)
 365 {
 366         struct strbuf revindex_name = STRBUF_INIT;
 367         int ret;
 368
 369         if (m->revindex_data)
 370                 return 0;
 371
 372         if (can_use_midx_ridx_chunk(m)) {
 373                 /*
 374                  * If the MIDX `m` has a `RIDX` chunk, then use its contents for
 375                  * the reverse index instead of trying to load a separate `.rev`
 376                  * file.
 377                  *
 378                  * Note that we do *not* set `m->revindex_map` here, since we do
 379                  * not want to accidentally call munmap() in the middle of the
 380                  * MIDX.
 381                  */
 382                 trace2_data_string("load_midx_revindex", m->repo,
 383                                    "source", "midx");
 384                 m->revindex_data = (const uint32_t *)m->chunk_revindex;
 385                 return 0;
 386         }
 387
 388         trace2_data_string("load_midx_revindex", m->repo,
 389                            "source", "rev");
 390
 391         if (m->has_chain)
 392                 get_split_midx_filename_ext(m->repo->hash_algo, &revindex_name,
 393                                             m->object_dir, get_midx_checksum(m),
 394                                             MIDX_EXT_REV);
 395         else
 396                 get_midx_filename_ext(m->repo->hash_algo, &revindex_name,
 397                                       m->object_dir, get_midx_checksum(m),
 398                                       MIDX_EXT_REV);
 399
 400         ret = load_revindex_from_disk(m->repo->hash_algo,
 401                                       revindex_name.buf,
 402                                       m->num_objects,
 403                                       &m->revindex_map,
 404                                       &m->revindex_len);
 405         if (ret)
 406                 goto cleanup;
 407
 408         m->revindex_data = (const uint32_t *)((const char *)m->revindex_map + RIDX_HEADER_SIZE);
 409
 410 cleanup:
 411         strbuf_release(&revindex_name);
 412         return ret;
 413 }
 414
 415 int close_midx_revindex(struct multi_pack_index *m)
 416 {
 417         if (!m || !m->revindex_map)
 418                 return 0;
 419
 420         munmap((void*)m->revindex_map, m->revindex_len);
 421
 422         m->revindex_map = NULL;
 423         m->revindex_data = NULL;
 424         m->revindex_len = 0;
 425
 426         return 0;
 427 }
 428
 429 int offset_to_pack_pos(struct packed_git *p, off_t ofs, uint32_t *pos)
 430 {
 431         unsigned lo, hi;
 432
 433         if (load_pack_revindex(p->repo, p) < 0)
 434                 return -1;
 435
 436         lo = 0;
 437         hi = p->num_objects + 1;
 438
 439         do {
 440                 const unsigned mi = lo + (hi - lo) / 2;
 441                 off_t got = pack_pos_to_offset(p, mi);
 442
 443                 if (got == ofs) {
 444                         *pos = mi;
 445                         return 0;
 446                 } else if (ofs < got)
 447                         hi = mi;
 448                 else
 449                         lo = mi + 1;
 450         } while (lo < hi);
 451
 452         error("bad offset for revindex");
 453         return -1;
 454 }
 455
 456 uint32_t pack_pos_to_index(struct packed_git *p, uint32_t pos)
 457 {
 458         if (!(p->revindex || p->revindex_data))
 459                 BUG("pack_pos_to_index: reverse index not yet loaded");
 460         if (p->num_objects <= pos)
 461                 BUG("pack_pos_to_index: out-of-bounds object at %"PRIu32, pos);
 462
 463         if (p->revindex)
 464                 return p->revindex[pos].nr;
 465         else
 466                 return get_be32(p->revindex_data + pos);
 467 }
 468
 469 off_t pack_pos_to_offset(struct packed_git *p, uint32_t pos)
 470 {
 471         if (!(p->revindex || p->revindex_data))
 472                 BUG("pack_pos_to_index: reverse index not yet loaded");
 473         if (p->num_objects < pos)
 474                 BUG("pack_pos_to_offset: out-of-bounds object at %"PRIu32, pos);
 475
 476         if (p->revindex)
 477                 return p->revindex[pos].offset;
 478         else if (pos == p->num_objects)
 479                 return p->pack_size - p->repo->hash_algo->rawsz;
 480         else
 481                 return nth_packed_object_offset(p, pack_pos_to_index(p, pos));
 482 }
 483
 484 uint32_t pack_pos_to_midx(struct multi_pack_index *m, uint32_t pos)
 485 {
 486         while (m && pos < m->num_objects_in_base)
 487                 m = m->base_midx;
 488         if (!m)
 489                 BUG("NULL multi-pack-index for object position: %"PRIu32, pos);
 490         if (!m->revindex_data)
 491                 BUG("pack_pos_to_midx: reverse index not yet loaded");
 492         if (m->num_objects + m->num_objects_in_base <= pos)
 493                 BUG("pack_pos_to_midx: out-of-bounds object at %"PRIu32, pos);
 494         return get_be32(m->revindex_data + pos - m->num_objects_in_base);
 495 }
 496
 497 struct midx_pack_key {
 498         uint32_t pack;
 499         off_t offset;
 500
 501         uint32_t preferred_pack;
 502         struct multi_pack_index *midx;
 503 };
 504
 505 static int midx_pack_order_cmp(const void *va, const void *vb)
 506 {
 507         const struct midx_pack_key *key = va;
 508         struct multi_pack_index *midx = key->midx;
 509
 510         size_t pos = (uint32_t *)vb - (const uint32_t *)midx->revindex_data;
 511         uint32_t versus = pack_pos_to_midx(midx, pos + midx->num_objects_in_base);
 512         uint32_t versus_pack = nth_midxed_pack_int_id(midx, versus);
 513         off_t versus_offset;
 514
 515         uint32_t key_preferred = key->pack == key->preferred_pack;
 516         uint32_t versus_preferred = versus_pack == key->preferred_pack;
 517
 518         /*
 519          * First, compare the preferred-ness, noting that the preferred pack
 520          * comes first.
 521          */
 522         if (key_preferred && !versus_preferred)
 523                 return -1;
 524         else if (!key_preferred && versus_preferred)
 525                 return 1;
 526
 527         /* Then, break ties first by comparing the pack IDs. */
 528         if (key->pack < versus_pack)
 529                 return -1;
 530         else if (key->pack > versus_pack)
 531                 return 1;
 532
 533         /* Finally, break ties by comparing offsets within a pack. */
 534         versus_offset = nth_midxed_offset(midx, versus);
 535         if (key->offset < versus_offset)
 536                 return -1;
 537         else if (key->offset > versus_offset)
 538                 return 1;
 539
 540         return 0;
 541 }
 542
 543 static int midx_key_to_pack_pos(struct multi_pack_index *m,
 544                                 struct midx_pack_key *key,
 545                                 uint32_t *pos)
 546 {
 547         uint32_t *found;
 548
 549         if (key->pack >= m->num_packs + m->num_packs_in_base)
 550                 BUG("MIDX pack lookup out of bounds (%"PRIu32" >= %"PRIu32")",
 551                     key->pack, m->num_packs + m->num_packs_in_base);
 552         /*
 553          * The preferred pack sorts first, so determine its identifier by
 554          * looking at the first object in pseudo-pack order.
 555          *
 556          * Note that if no --preferred-pack is explicitly given when writing a
 557          * multi-pack index, then whichever pack has the lowest identifier
 558          * implicitly is preferred (and includes all its objects, since ties are
 559          * broken first by pack identifier).
 560          */
 561         if (midx_preferred_pack(key->midx, &key->preferred_pack) < 0)
 562                 return error(_("could not determine preferred pack"));
 563
 564         found = bsearch(key, m->revindex_data, m->num_objects,
 565                         sizeof(*m->revindex_data),
 566                         midx_pack_order_cmp);
 567
 568         if (!found)
 569                 return -1;
 570
 571         *pos = (found - m->revindex_data) + m->num_objects_in_base;
 572
 573         return 0;
 574 }
 575
 576 int midx_to_pack_pos(struct multi_pack_index *m, uint32_t at, uint32_t *pos)
 577 {
 578         struct midx_pack_key key;
 579
 580         while (m && at < m->num_objects_in_base)
 581                 m = m->base_midx;
 582         if (!m)
 583                 BUG("NULL multi-pack-index for object position: %"PRIu32, at);
 584         if (!m->revindex_data)
 585                 BUG("midx_to_pack_pos: reverse index not yet loaded");
 586         if (m->num_objects + m->num_objects_in_base <= at)
 587                 BUG("midx_to_pack_pos: out-of-bounds object at %"PRIu32, at);
 588
 589         key.pack = nth_midxed_pack_int_id(m, at);
 590         key.offset = nth_midxed_offset(m, at);
 591         key.midx = m;
 592
 593         return midx_key_to_pack_pos(m, &key, pos);
 594 }
 595
 596 int midx_pair_to_pack_pos(struct multi_pack_index *m, uint32_t pack_int_id,
 597                           off_t ofs, uint32_t *pos)
 598 {
 599         struct midx_pack_key key = {
 600                 .pack = pack_int_id,
 601                 .offset = ofs,
 602                 .midx = m,
 603         };
 604         return midx_key_to_pack_pos(m, &key, pos);
 605 }