pack-write.c

   1 #include "cache.h"
   2 #include "pack.h"
   3 #include "csum-file.h"
   4 #include "remote.h"
   5 #include "chunk-format.h"
   6 #include "pack-mtimes.h"
   7 #include "oidmap.h"
   8 #include "chunk-format.h"
   9 #include "pack-objects.h"
  10
  11 void reset_pack_idx_option(struct pack_idx_option *opts)
  12 {
  13         memset(opts, 0, sizeof(*opts));
  14         opts->version = 2;
  15         opts->off32_limit = 0x7fffffff;
  16 }
  17
  18 static int sha1_compare(const void *_a, const void *_b)
  19 {
  20         struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;
  21         struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;
  22         return oidcmp(&a->oid, &b->oid);
  23 }
  24
  25 static int cmp_uint32(const void *a_, const void *b_)
  26 {
  27         uint32_t a = *((uint32_t *)a_);
  28         uint32_t b = *((uint32_t *)b_);
  29
  30         return (a < b) ? -1 : (a != b);
  31 }
  32
  33 static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
  34 {
  35         uint32_t ofsval;
  36
  37         if ((offset >> 31) || (opts->off32_limit < offset))
  38                 return 1;
  39         if (!opts->anomaly_nr)
  40                 return 0;
  41         ofsval = offset;
  42         return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,
  43                          sizeof(ofsval), cmp_uint32);
  44 }
  45
  46 /*
  47  * The *sha1 contains the pack content SHA1 hash.
  48  * The objects array passed in will be sorted by SHA1 on exit.
  49  */
  50 const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,
  51                            int nr_objects, const struct pack_idx_option *opts,
  52                            const unsigned char *sha1)
  53 {
  54         struct hashfile *f;
  55         struct pack_idx_entry **sorted_by_sha, **list, **last;
  56         off_t last_obj_offset = 0;
  57         int i, fd;
  58         uint32_t index_version;
  59
  60         if (nr_objects) {
  61                 sorted_by_sha = objects;
  62                 list = sorted_by_sha;
  63                 last = sorted_by_sha + nr_objects;
  64                 for (i = 0; i < nr_objects; ++i) {
  65                         if (objects[i]->offset > last_obj_offset)
  66                                 last_obj_offset = objects[i]->offset;
  67                 }
  68                 QSORT(sorted_by_sha, nr_objects, sha1_compare);
  69         }
  70         else
  71                 sorted_by_sha = list = last = NULL;
  72
  73         if (opts->flags & WRITE_IDX_VERIFY) {
  74                 assert(index_name);
  75                 f = hashfd_check(index_name);
  76         } else {
  77                 if (!index_name) {
  78                         struct strbuf tmp_file = STRBUF_INIT;
  79                         fd = odb_mkstemp(&tmp_file, "pack/tmp_idx_XXXXXX");
  80                         index_name = strbuf_detach(&tmp_file, NULL);
  81                 } else {
  82                         unlink(index_name);
  83                         fd = xopen(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
  84                 }
  85                 f = hashfd(fd, index_name);
  86         }
  87
  88         /* if last object's offset is >= 2^31 we should use index V2 */
  89         index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
  90
  91         /* index versions 2 and above need a header */
  92         if (index_version >= 2) {
  93                 struct pack_idx_header hdr;
  94                 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
  95                 hdr.idx_version = htonl(index_version);
  96                 hashwrite(f, &hdr, sizeof(hdr));
  97         }
  98
  99         /*
 100          * Write the first-level table (the list is sorted,
 101          * but we use a 256-entry lookup to be able to avoid
 102          * having to do eight extra binary search iterations).
 103          */
 104         for (i = 0; i < 256; i++) {
 105                 struct pack_idx_entry **next = list;
 106                 while (next < last) {
 107                         struct pack_idx_entry *obj = *next;
 108                         if (obj->oid.hash[0] != i)
 109                                 break;
 110                         next++;
 111                 }
 112                 hashwrite_be32(f, next - sorted_by_sha);
 113                 list = next;
 114         }
 115
 116         /*
 117          * Write the actual SHA1 entries..
 118          */
 119         list = sorted_by_sha;
 120         for (i = 0; i < nr_objects; i++) {
 121                 struct pack_idx_entry *obj = *list++;
 122                 if (index_version < 2)
 123                         hashwrite_be32(f, obj->offset);
 124                 hashwrite(f, obj->oid.hash, the_hash_algo->rawsz);
 125                 if ((opts->flags & WRITE_IDX_STRICT) &&
 126                     (i && oideq(&list[-2]->oid, &obj->oid)))
 127                         die("The same object %s appears twice in the pack",
 128                             oid_to_hex(&obj->oid));
 129         }
 130
 131         if (index_version >= 2) {
 132                 unsigned int nr_large_offset = 0;
 133
 134                 /* write the crc32 table */
 135                 list = sorted_by_sha;
 136                 for (i = 0; i < nr_objects; i++) {
 137                         struct pack_idx_entry *obj = *list++;
 138                         hashwrite_be32(f, obj->crc32);
 139                 }
 140
 141                 /* write the 32-bit offset table */
 142                 list = sorted_by_sha;
 143                 for (i = 0; i < nr_objects; i++) {
 144                         struct pack_idx_entry *obj = *list++;
 145                         uint32_t offset;
 146
 147                         offset = (need_large_offset(obj->offset, opts)
 148                                   ? (0x80000000 | nr_large_offset++)
 149                                   : obj->offset);
 150                         hashwrite_be32(f, offset);
 151                 }
 152
 153                 /* write the large offset table */
 154                 list = sorted_by_sha;
 155                 while (nr_large_offset) {
 156                         struct pack_idx_entry *obj = *list++;
 157                         uint64_t offset = obj->offset;
 158
 159                         if (!need_large_offset(offset, opts))
 160                                 continue;
 161                         hashwrite_be64(f, offset);
 162                         nr_large_offset--;
 163                 }
 164         }
 165
 166         hashwrite(f, sha1, the_hash_algo->rawsz);
 167         finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,
 168                           CSUM_HASH_IN_STREAM | CSUM_CLOSE |
 169                           ((opts->flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));
 170         return index_name;
 171 }
 172
 173 static int pack_order_cmp(const void *va, const void *vb, void *ctx)
 174 {
 175         struct pack_idx_entry **objects = ctx;
 176
 177         off_t oa = objects[*(uint32_t*)va]->offset;
 178         off_t ob = objects[*(uint32_t*)vb]->offset;
 179
 180         if (oa < ob)
 181                 return -1;
 182         if (oa > ob)
 183                 return 1;
 184         return 0;
 185 }
 186
 187 static void write_rev_header(struct hashfile *f)
 188 {
 189         hashwrite_be32(f, RIDX_SIGNATURE);
 190         hashwrite_be32(f, RIDX_VERSION);
 191         hashwrite_be32(f, oid_version(the_hash_algo));
 192 }
 193
 194 static void write_rev_index_positions(struct hashfile *f,
 195                                       uint32_t *pack_order,
 196                                       uint32_t nr_objects)
 197 {
 198         uint32_t i;
 199         for (i = 0; i < nr_objects; i++)
 200                 hashwrite_be32(f, pack_order[i]);
 201 }
 202
 203 static void write_rev_trailer(struct hashfile *f, const unsigned char *hash)
 204 {
 205         hashwrite(f, hash, the_hash_algo->rawsz);
 206 }
 207
 208 const char *write_rev_file(const char *rev_name,
 209                            struct pack_idx_entry **objects,
 210                            uint32_t nr_objects,
 211                            const unsigned char *hash,
 212                            unsigned flags)
 213 {
 214         uint32_t *pack_order;
 215         uint32_t i;
 216         const char *ret;
 217
 218         if (!(flags & WRITE_REV) && !(flags & WRITE_REV_VERIFY))
 219                 return NULL;
 220
 221         ALLOC_ARRAY(pack_order, nr_objects);
 222         for (i = 0; i < nr_objects; i++)
 223                 pack_order[i] = i;
 224         QSORT_S(pack_order, nr_objects, pack_order_cmp, objects);
 225
 226         ret = write_rev_file_order(rev_name, pack_order, nr_objects, hash,
 227                                    flags);
 228
 229         free(pack_order);
 230
 231         return ret;
 232 }
 233
 234 const char *write_rev_file_order(const char *rev_name,
 235                                  uint32_t *pack_order,
 236                                  uint32_t nr_objects,
 237                                  const unsigned char *hash,
 238                                  unsigned flags)
 239 {
 240         struct hashfile *f;
 241         int fd;
 242
 243         if ((flags & WRITE_REV) && (flags & WRITE_REV_VERIFY))
 244                 die(_("cannot both write and verify reverse index"));
 245
 246         if (flags & WRITE_REV) {
 247                 if (!rev_name) {
 248                         struct strbuf tmp_file = STRBUF_INIT;
 249                         fd = odb_mkstemp(&tmp_file, "pack/tmp_rev_XXXXXX");
 250                         rev_name = strbuf_detach(&tmp_file, NULL);
 251                 } else {
 252                         unlink(rev_name);
 253                         fd = xopen(rev_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
 254                 }
 255                 f = hashfd(fd, rev_name);
 256         } else if (flags & WRITE_REV_VERIFY) {
 257                 struct stat statbuf;
 258                 if (stat(rev_name, &statbuf)) {
 259                         if (errno == ENOENT) {
 260                                 /* .rev files are optional */
 261                                 return NULL;
 262                         } else
 263                                 die_errno(_("could not stat: %s"), rev_name);
 264                 }
 265                 f = hashfd_check(rev_name);
 266         } else
 267                 return NULL;
 268
 269         write_rev_header(f);
 270
 271         write_rev_index_positions(f, pack_order, nr_objects);
 272         write_rev_trailer(f, hash);
 273
 274         if (rev_name && adjust_shared_perm(rev_name) < 0)
 275                 die(_("failed to make %s readable"), rev_name);
 276
 277         finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,
 278                           CSUM_HASH_IN_STREAM | CSUM_CLOSE |
 279                           ((flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));
 280
 281         return rev_name;
 282 }
 283
 284 static void write_mtimes_header(struct hashfile *f)
 285 {
 286         hashwrite_be32(f, MTIMES_SIGNATURE);
 287         hashwrite_be32(f, MTIMES_VERSION);
 288         hashwrite_be32(f, oid_version(the_hash_algo));
 289 }
 290
 291 /*
 292  * Writes the object mtimes of "objects" for use in a .mtimes file.
 293  * Note that objects must be in lexicographic (index) order, which is
 294  * the expected ordering of these values in the .mtimes file.
 295  */
 296 static void write_mtimes_objects(struct hashfile *f,
 297                                  struct packing_data *to_pack,
 298                                  struct pack_idx_entry **objects,
 299                                  uint32_t nr_objects)
 300 {
 301         uint32_t i;
 302         for (i = 0; i < nr_objects; i++) {
 303                 struct object_entry *e = (struct object_entry*)objects[i];
 304                 hashwrite_be32(f, oe_cruft_mtime(to_pack, e));
 305         }
 306 }
 307
 308 static void write_mtimes_trailer(struct hashfile *f, const unsigned char *hash)
 309 {
 310         hashwrite(f, hash, the_hash_algo->rawsz);
 311 }
 312
 313 static const char *write_mtimes_file(struct packing_data *to_pack,
 314                                      struct pack_idx_entry **objects,
 315                                      uint32_t nr_objects,
 316                                      const unsigned char *hash)
 317 {
 318         struct strbuf tmp_file = STRBUF_INIT;
 319         const char *mtimes_name;
 320         struct hashfile *f;
 321         int fd;
 322
 323         if (!to_pack)
 324                 BUG("cannot call write_mtimes_file with NULL packing_data");
 325
 326         fd = odb_mkstemp(&tmp_file, "pack/tmp_mtimes_XXXXXX");
 327         mtimes_name = strbuf_detach(&tmp_file, NULL);
 328         f = hashfd(fd, mtimes_name);
 329
 330         write_mtimes_header(f);
 331         write_mtimes_objects(f, to_pack, objects, nr_objects);
 332         write_mtimes_trailer(f, hash);
 333
 334         if (adjust_shared_perm(mtimes_name) < 0)
 335                 die(_("failed to make %s readable"), mtimes_name);
 336
 337         finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,
 338                           CSUM_HASH_IN_STREAM | CSUM_CLOSE | CSUM_FSYNC);
 339
 340         return mtimes_name;
 341 }
 342
 343 off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
 344 {
 345         struct pack_header hdr;
 346
 347         hdr.hdr_signature = htonl(PACK_SIGNATURE);
 348         hdr.hdr_version = htonl(PACK_VERSION);
 349         hdr.hdr_entries = htonl(nr_entries);
 350         hashwrite(f, &hdr, sizeof(hdr));
 351         return sizeof(hdr);
 352 }
 353
 354 /*
 355  * Update pack header with object_count and compute new SHA1 for pack data
 356  * associated to pack_fd, and write that SHA1 at the end.  That new SHA1
 357  * is also returned in new_pack_sha1.
 358  *
 359  * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
 360  * (without the header update) is computed and validated against the
 361  * one provided in partial_pack_sha1.  The validation is performed at
 362  * partial_pack_offset bytes in the pack file.  The SHA1 of the remaining
 363  * data (i.e. from partial_pack_offset to the end) is then computed and
 364  * returned in partial_pack_sha1.
 365  *
 366  * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
 367  * partial_pack_sha1 can refer to the same buffer if the caller is not
 368  * interested in the resulting SHA1 of pack data above partial_pack_offset.
 369  */
 370 void fixup_pack_header_footer(int pack_fd,
 371                          unsigned char *new_pack_hash,
 372                          const char *pack_name,
 373                          uint32_t object_count,
 374                          unsigned char *partial_pack_hash,
 375                          off_t partial_pack_offset)
 376 {
 377         int aligned_sz, buf_sz = 8 * 1024;
 378         git_hash_ctx old_hash_ctx, new_hash_ctx;
 379         struct pack_header hdr;
 380         char *buf;
 381         ssize_t read_result;
 382
 383         the_hash_algo->init_fn(&old_hash_ctx);
 384         the_hash_algo->init_fn(&new_hash_ctx);
 385
 386         if (lseek(pack_fd, 0, SEEK_SET) != 0)
 387                 die_errno("Failed seeking to start of '%s'", pack_name);
 388         read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
 389         if (read_result < 0)
 390                 die_errno("Unable to reread header of '%s'", pack_name);
 391         else if (read_result != sizeof(hdr))
 392                 die_errno("Unexpected short read for header of '%s'",
 393                           pack_name);
 394         if (lseek(pack_fd, 0, SEEK_SET) != 0)
 395                 die_errno("Failed seeking to start of '%s'", pack_name);
 396         the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));
 397         hdr.hdr_entries = htonl(object_count);
 398         the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));
 399         write_or_die(pack_fd, &hdr, sizeof(hdr));
 400         partial_pack_offset -= sizeof(hdr);
 401
 402         buf = xmalloc(buf_sz);
 403         aligned_sz = buf_sz - sizeof(hdr);
 404         for (;;) {
 405                 ssize_t m, n;
 406                 m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
 407                         partial_pack_offset : aligned_sz;
 408                 n = xread(pack_fd, buf, m);
 409                 if (!n)
 410                         break;
 411                 if (n < 0)
 412                         die_errno("Failed to checksum '%s'", pack_name);
 413                 the_hash_algo->update_fn(&new_hash_ctx, buf, n);
 414
 415                 aligned_sz -= n;
 416                 if (!aligned_sz)
 417                         aligned_sz = buf_sz;
 418
 419                 if (!partial_pack_hash)
 420                         continue;
 421
 422                 the_hash_algo->update_fn(&old_hash_ctx, buf, n);
 423                 partial_pack_offset -= n;
 424                 if (partial_pack_offset == 0) {
 425                         unsigned char hash[GIT_MAX_RAWSZ];
 426                         the_hash_algo->final_fn(hash, &old_hash_ctx);
 427                         if (!hasheq(hash, partial_pack_hash))
 428                                 die("Unexpected checksum for %s "
 429                                     "(disk corruption?)", pack_name);
 430                         /*
 431                          * Now let's compute the SHA1 of the remainder of the
 432                          * pack, which also means making partial_pack_offset
 433                          * big enough not to matter anymore.
 434                          */
 435                         the_hash_algo->init_fn(&old_hash_ctx);
 436                         partial_pack_offset = ~partial_pack_offset;
 437                         partial_pack_offset -= MSB(partial_pack_offset, 1);
 438                 }
 439         }
 440         free(buf);
 441
 442         if (partial_pack_hash)
 443                 the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);
 444         the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);
 445         write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);
 446         fsync_component_or_die(FSYNC_COMPONENT_PACK, pack_fd, pack_name);
 447 }
 448
 449 char *index_pack_lockfile(int ip_out, int *is_well_formed)
 450 {
 451         char packname[GIT_MAX_HEXSZ + 6];
 452         const int len = the_hash_algo->hexsz + 6;
 453
 454         /*
 455          * The first thing we expect from index-pack's output
 456          * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
 457          * %40s is the newly created pack SHA1 name.  In the "keep"
 458          * case, we need it to remove the corresponding .keep file
 459          * later on.  If we don't get that then tough luck with it.
 460          */
 461         if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
 462                 const char *name;
 463
 464                 if (is_well_formed)
 465                         *is_well_formed = 1;
 466                 packname[len-1] = 0;
 467                 if (skip_prefix(packname, "keep\t", &name))
 468                         return xstrfmt("%s/pack/pack-%s.keep",
 469                                        get_object_directory(), name);
 470                 return NULL;
 471         }
 472         if (is_well_formed)
 473                 *is_well_formed = 0;
 474         return NULL;
 475 }
 476
 477 /*
 478  * The per-object header is a pretty dense thing, which is
 479  *  - first byte: low four bits are "size", then three bits of "type",
 480  *    and the high bit is "size continues".
 481  *  - each byte afterwards: low seven bits are size continuation,
 482  *    with the high bit being "size continues"
 483  */
 484 int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
 485                                  enum object_type type, uintmax_t size)
 486 {
 487         int n = 1;
 488         unsigned char c;
 489
 490         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 491                 die("bad type %d", type);
 492
 493         c = (type << 4) | (size & 15);
 494         size >>= 4;
 495         while (size) {
 496                 if (n == hdr_len)
 497                         die("object size is too enormous to format");
 498                 *hdr++ = c | 0x80;
 499                 c = size & 0x7f;
 500                 size >>= 7;
 501                 n++;
 502         }
 503         *hdr = c;
 504         return n;
 505 }
 506
 507 struct hashfile *create_tmp_packfile(char **pack_tmp_name)
 508 {
 509         struct strbuf tmpname = STRBUF_INIT;
 510         int fd;
 511
 512         fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
 513         *pack_tmp_name = strbuf_detach(&tmpname, NULL);
 514         return hashfd(fd, *pack_tmp_name);
 515 }
 516
 517 static void rename_tmp_packfile(struct strbuf *name_prefix, const char *source,
 518                                 const char *ext)
 519 {
 520         size_t name_prefix_len = name_prefix->len;
 521
 522         strbuf_addstr(name_prefix, ext);
 523         if (rename(source, name_prefix->buf))
 524                 die_errno("unable to rename temporary file to '%s'",
 525                           name_prefix->buf);
 526         strbuf_setlen(name_prefix, name_prefix_len);
 527 }
 528
 529 void rename_tmp_packfile_idx(struct strbuf *name_buffer,
 530                              char **idx_tmp_name)
 531 {
 532         rename_tmp_packfile(name_buffer, *idx_tmp_name, "idx");
 533 }
 534
 535 void stage_tmp_packfiles(struct strbuf *name_buffer,
 536                          const char *pack_tmp_name,
 537                          struct pack_idx_entry **written_list,
 538                          uint32_t nr_written,
 539                          struct packing_data *to_pack,
 540                          struct pack_idx_option *pack_idx_opts,
 541                          unsigned char hash[],
 542                          char **idx_tmp_name)
 543 {
 544         const char *rev_tmp_name = NULL;
 545         const char *mtimes_tmp_name = NULL;
 546
 547         if (adjust_shared_perm(pack_tmp_name))
 548                 die_errno("unable to make temporary pack file readable");
 549
 550         *idx_tmp_name = (char *)write_idx_file(NULL, written_list, nr_written,
 551                                                pack_idx_opts, hash);
 552         if (adjust_shared_perm(*idx_tmp_name))
 553                 die_errno("unable to make temporary index file readable");
 554
 555         rev_tmp_name = write_rev_file(NULL, written_list, nr_written, hash,
 556                                       pack_idx_opts->flags);
 557
 558         if (pack_idx_opts->flags & WRITE_MTIMES) {
 559                 mtimes_tmp_name = write_mtimes_file(to_pack, written_list,
 560                                                     nr_written,
 561                                                     hash);
 562         }
 563
 564         rename_tmp_packfile(name_buffer, pack_tmp_name, "pack");
 565         if (rev_tmp_name)
 566                 rename_tmp_packfile(name_buffer, rev_tmp_name, "rev");
 567         if (mtimes_tmp_name)
 568                 rename_tmp_packfile(name_buffer, mtimes_tmp_name, "mtimes");
 569 }
 570
 571 void write_promisor_file(const char *promisor_name, struct ref **sought, int nr_sought)
 572 {
 573         int i, err;
 574         FILE *output = xfopen(promisor_name, "w");
 575
 576         for (i = 0; i < nr_sought; i++)
 577                 fprintf(output, "%s %s\n", oid_to_hex(&sought[i]->old_oid),
 578                         sought[i]->name);
 579
 580         err = ferror(output);
 581         err |= fclose(output);
 582         if (err)
 583                 die(_("could not write '%s' promisor file"), promisor_name);
 584 }