pack-write.c

   1 #include "cache.h"
   2 #include "pack.h"
   3 #include "csum-file.h"
   4 #include "remote.h"
   5
   6 void reset_pack_idx_option(struct pack_idx_option *opts)
   7 {
   8         memset(opts, 0, sizeof(*opts));
   9         opts->version = 2;
  10         opts->off32_limit = 0x7fffffff;
  11 }
  12
  13 static int sha1_compare(const void *_a, const void *_b)
  14 {
  15         struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;
  16         struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;
  17         return oidcmp(&a->oid, &b->oid);
  18 }
  19
  20 static int cmp_uint32(const void *a_, const void *b_)
  21 {
  22         uint32_t a = *((uint32_t *)a_);
  23         uint32_t b = *((uint32_t *)b_);
  24
  25         return (a < b) ? -1 : (a != b);
  26 }
  27
  28 static int need_large_offset(off_t offset, const struct pack_idx_option *opts)
  29 {
  30         uint32_t ofsval;
  31
  32         if ((offset >> 31) || (opts->off32_limit < offset))
  33                 return 1;
  34         if (!opts->anomaly_nr)
  35                 return 0;
  36         ofsval = offset;
  37         return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,
  38                          sizeof(ofsval), cmp_uint32);
  39 }
  40
  41 /*
  42  * The *sha1 contains the pack content SHA1 hash.
  43  * The objects array passed in will be sorted by SHA1 on exit.
  44  */
  45 const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,
  46                            int nr_objects, const struct pack_idx_option *opts,
  47                            const unsigned char *sha1)
  48 {
  49         struct hashfile *f;
  50         struct pack_idx_entry **sorted_by_sha, **list, **last;
  51         off_t last_obj_offset = 0;
  52         int i, fd;
  53         uint32_t index_version;
  54
  55         if (nr_objects) {
  56                 sorted_by_sha = objects;
  57                 list = sorted_by_sha;
  58                 last = sorted_by_sha + nr_objects;
  59                 for (i = 0; i < nr_objects; ++i) {
  60                         if (objects[i]->offset > last_obj_offset)
  61                                 last_obj_offset = objects[i]->offset;
  62                 }
  63                 QSORT(sorted_by_sha, nr_objects, sha1_compare);
  64         }
  65         else
  66                 sorted_by_sha = list = last = NULL;
  67
  68         if (opts->flags & WRITE_IDX_VERIFY) {
  69                 assert(index_name);
  70                 f = hashfd_check(index_name);
  71         } else {
  72                 if (!index_name) {
  73                         struct strbuf tmp_file = STRBUF_INIT;
  74                         fd = odb_mkstemp(&tmp_file, "pack/tmp_idx_XXXXXX");
  75                         index_name = strbuf_detach(&tmp_file, NULL);
  76                 } else {
  77                         unlink(index_name);
  78                         fd = xopen(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
  79                 }
  80                 f = hashfd(fd, index_name);
  81         }
  82
  83         /* if last object's offset is >= 2^31 we should use index V2 */
  84         index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;
  85
  86         /* index versions 2 and above need a header */
  87         if (index_version >= 2) {
  88                 struct pack_idx_header hdr;
  89                 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
  90                 hdr.idx_version = htonl(index_version);
  91                 hashwrite(f, &hdr, sizeof(hdr));
  92         }
  93
  94         /*
  95          * Write the first-level table (the list is sorted,
  96          * but we use a 256-entry lookup to be able to avoid
  97          * having to do eight extra binary search iterations).
  98          */
  99         for (i = 0; i < 256; i++) {
 100                 struct pack_idx_entry **next = list;
 101                 while (next < last) {
 102                         struct pack_idx_entry *obj = *next;
 103                         if (obj->oid.hash[0] != i)
 104                                 break;
 105                         next++;
 106                 }
 107                 hashwrite_be32(f, next - sorted_by_sha);
 108                 list = next;
 109         }
 110
 111         /*
 112          * Write the actual SHA1 entries..
 113          */
 114         list = sorted_by_sha;
 115         for (i = 0; i < nr_objects; i++) {
 116                 struct pack_idx_entry *obj = *list++;
 117                 if (index_version < 2)
 118                         hashwrite_be32(f, obj->offset);
 119                 hashwrite(f, obj->oid.hash, the_hash_algo->rawsz);
 120                 if ((opts->flags & WRITE_IDX_STRICT) &&
 121                     (i && oideq(&list[-2]->oid, &obj->oid)))
 122                         die("The same object %s appears twice in the pack",
 123                             oid_to_hex(&obj->oid));
 124         }
 125
 126         if (index_version >= 2) {
 127                 unsigned int nr_large_offset = 0;
 128
 129                 /* write the crc32 table */
 130                 list = sorted_by_sha;
 131                 for (i = 0; i < nr_objects; i++) {
 132                         struct pack_idx_entry *obj = *list++;
 133                         hashwrite_be32(f, obj->crc32);
 134                 }
 135
 136                 /* write the 32-bit offset table */
 137                 list = sorted_by_sha;
 138                 for (i = 0; i < nr_objects; i++) {
 139                         struct pack_idx_entry *obj = *list++;
 140                         uint32_t offset;
 141
 142                         offset = (need_large_offset(obj->offset, opts)
 143                                   ? (0x80000000 | nr_large_offset++)
 144                                   : obj->offset);
 145                         hashwrite_be32(f, offset);
 146                 }
 147
 148                 /* write the large offset table */
 149                 list = sorted_by_sha;
 150                 while (nr_large_offset) {
 151                         struct pack_idx_entry *obj = *list++;
 152                         uint64_t offset = obj->offset;
 153
 154                         if (!need_large_offset(offset, opts))
 155                                 continue;
 156                         hashwrite_be64(f, offset);
 157                         nr_large_offset--;
 158                 }
 159         }
 160
 161         hashwrite(f, sha1, the_hash_algo->rawsz);
 162         finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_CLOSE |
 163                                     ((opts->flags & WRITE_IDX_VERIFY)
 164                                     ? 0 : CSUM_FSYNC));
 165         return index_name;
 166 }
 167
 168 static int pack_order_cmp(const void *va, const void *vb, void *ctx)
 169 {
 170         struct pack_idx_entry **objects = ctx;
 171
 172         off_t oa = objects[*(uint32_t*)va]->offset;
 173         off_t ob = objects[*(uint32_t*)vb]->offset;
 174
 175         if (oa < ob)
 176                 return -1;
 177         if (oa > ob)
 178                 return 1;
 179         return 0;
 180 }
 181
 182 static void write_rev_header(struct hashfile *f)
 183 {
 184         uint32_t oid_version;
 185         switch (hash_algo_by_ptr(the_hash_algo)) {
 186         case GIT_HASH_SHA1:
 187                 oid_version = 1;
 188                 break;
 189         case GIT_HASH_SHA256:
 190                 oid_version = 2;
 191                 break;
 192         default:
 193                 die("write_rev_header: unknown hash version");
 194         }
 195
 196         hashwrite_be32(f, RIDX_SIGNATURE);
 197         hashwrite_be32(f, RIDX_VERSION);
 198         hashwrite_be32(f, oid_version);
 199 }
 200
 201 static void write_rev_index_positions(struct hashfile *f,
 202                                       uint32_t *pack_order,
 203                                       uint32_t nr_objects)
 204 {
 205         uint32_t i;
 206         for (i = 0; i < nr_objects; i++)
 207                 hashwrite_be32(f, pack_order[i]);
 208 }
 209
 210 static void write_rev_trailer(struct hashfile *f, const unsigned char *hash)
 211 {
 212         hashwrite(f, hash, the_hash_algo->rawsz);
 213 }
 214
 215 const char *write_rev_file(const char *rev_name,
 216                            struct pack_idx_entry **objects,
 217                            uint32_t nr_objects,
 218                            const unsigned char *hash,
 219                            unsigned flags)
 220 {
 221         uint32_t *pack_order;
 222         uint32_t i;
 223         const char *ret;
 224
 225         ALLOC_ARRAY(pack_order, nr_objects);
 226         for (i = 0; i < nr_objects; i++)
 227                 pack_order[i] = i;
 228         QSORT_S(pack_order, nr_objects, pack_order_cmp, objects);
 229
 230         ret = write_rev_file_order(rev_name, pack_order, nr_objects, hash,
 231                                    flags);
 232
 233         free(pack_order);
 234
 235         return ret;
 236 }
 237
 238 const char *write_rev_file_order(const char *rev_name,
 239                                  uint32_t *pack_order,
 240                                  uint32_t nr_objects,
 241                                  const unsigned char *hash,
 242                                  unsigned flags)
 243 {
 244         struct hashfile *f;
 245         int fd;
 246
 247         if ((flags & WRITE_REV) && (flags & WRITE_REV_VERIFY))
 248                 die(_("cannot both write and verify reverse index"));
 249
 250         if (flags & WRITE_REV) {
 251                 if (!rev_name) {
 252                         struct strbuf tmp_file = STRBUF_INIT;
 253                         fd = odb_mkstemp(&tmp_file, "pack/tmp_rev_XXXXXX");
 254                         rev_name = strbuf_detach(&tmp_file, NULL);
 255                 } else {
 256                         unlink(rev_name);
 257                         fd = xopen(rev_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
 258                 }
 259                 f = hashfd(fd, rev_name);
 260         } else if (flags & WRITE_REV_VERIFY) {
 261                 struct stat statbuf;
 262                 if (stat(rev_name, &statbuf)) {
 263                         if (errno == ENOENT) {
 264                                 /* .rev files are optional */
 265                                 return NULL;
 266                         } else
 267                                 die_errno(_("could not stat: %s"), rev_name);
 268                 }
 269                 f = hashfd_check(rev_name);
 270         } else
 271                 return NULL;
 272
 273         write_rev_header(f);
 274
 275         write_rev_index_positions(f, pack_order, nr_objects);
 276         write_rev_trailer(f, hash);
 277
 278         if (rev_name && adjust_shared_perm(rev_name) < 0)
 279                 die(_("failed to make %s readable"), rev_name);
 280
 281         finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_CLOSE |
 282                                     ((flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));
 283
 284         return rev_name;
 285 }
 286
 287 off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
 288 {
 289         struct pack_header hdr;
 290
 291         hdr.hdr_signature = htonl(PACK_SIGNATURE);
 292         hdr.hdr_version = htonl(PACK_VERSION);
 293         hdr.hdr_entries = htonl(nr_entries);
 294         hashwrite(f, &hdr, sizeof(hdr));
 295         return sizeof(hdr);
 296 }
 297
 298 /*
 299  * Update pack header with object_count and compute new SHA1 for pack data
 300  * associated to pack_fd, and write that SHA1 at the end.  That new SHA1
 301  * is also returned in new_pack_sha1.
 302  *
 303  * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
 304  * (without the header update) is computed and validated against the
 305  * one provided in partial_pack_sha1.  The validation is performed at
 306  * partial_pack_offset bytes in the pack file.  The SHA1 of the remaining
 307  * data (i.e. from partial_pack_offset to the end) is then computed and
 308  * returned in partial_pack_sha1.
 309  *
 310  * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
 311  * partial_pack_sha1 can refer to the same buffer if the caller is not
 312  * interested in the resulting SHA1 of pack data above partial_pack_offset.
 313  */
 314 void fixup_pack_header_footer(int pack_fd,
 315                          unsigned char *new_pack_hash,
 316                          const char *pack_name,
 317                          uint32_t object_count,
 318                          unsigned char *partial_pack_hash,
 319                          off_t partial_pack_offset)
 320 {
 321         int aligned_sz, buf_sz = 8 * 1024;
 322         git_hash_ctx old_hash_ctx, new_hash_ctx;
 323         struct pack_header hdr;
 324         char *buf;
 325         ssize_t read_result;
 326
 327         the_hash_algo->init_fn(&old_hash_ctx);
 328         the_hash_algo->init_fn(&new_hash_ctx);
 329
 330         if (lseek(pack_fd, 0, SEEK_SET) != 0)
 331                 die_errno("Failed seeking to start of '%s'", pack_name);
 332         read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
 333         if (read_result < 0)
 334                 die_errno("Unable to reread header of '%s'", pack_name);
 335         else if (read_result != sizeof(hdr))
 336                 die_errno("Unexpected short read for header of '%s'",
 337                           pack_name);
 338         if (lseek(pack_fd, 0, SEEK_SET) != 0)
 339                 die_errno("Failed seeking to start of '%s'", pack_name);
 340         the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));
 341         hdr.hdr_entries = htonl(object_count);
 342         the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));
 343         write_or_die(pack_fd, &hdr, sizeof(hdr));
 344         partial_pack_offset -= sizeof(hdr);
 345
 346         buf = xmalloc(buf_sz);
 347         aligned_sz = buf_sz - sizeof(hdr);
 348         for (;;) {
 349                 ssize_t m, n;
 350                 m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
 351                         partial_pack_offset : aligned_sz;
 352                 n = xread(pack_fd, buf, m);
 353                 if (!n)
 354                         break;
 355                 if (n < 0)
 356                         die_errno("Failed to checksum '%s'", pack_name);
 357                 the_hash_algo->update_fn(&new_hash_ctx, buf, n);
 358
 359                 aligned_sz -= n;
 360                 if (!aligned_sz)
 361                         aligned_sz = buf_sz;
 362
 363                 if (!partial_pack_hash)
 364                         continue;
 365
 366                 the_hash_algo->update_fn(&old_hash_ctx, buf, n);
 367                 partial_pack_offset -= n;
 368                 if (partial_pack_offset == 0) {
 369                         unsigned char hash[GIT_MAX_RAWSZ];
 370                         the_hash_algo->final_fn(hash, &old_hash_ctx);
 371                         if (!hasheq(hash, partial_pack_hash))
 372                                 die("Unexpected checksum for %s "
 373                                     "(disk corruption?)", pack_name);
 374                         /*
 375                          * Now let's compute the SHA1 of the remainder of the
 376                          * pack, which also means making partial_pack_offset
 377                          * big enough not to matter anymore.
 378                          */
 379                         the_hash_algo->init_fn(&old_hash_ctx);
 380                         partial_pack_offset = ~partial_pack_offset;
 381                         partial_pack_offset -= MSB(partial_pack_offset, 1);
 382                 }
 383         }
 384         free(buf);
 385
 386         if (partial_pack_hash)
 387                 the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);
 388         the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);
 389         write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);
 390         fsync_or_die(pack_fd, pack_name);
 391 }
 392
 393 char *index_pack_lockfile(int ip_out, int *is_well_formed)
 394 {
 395         char packname[GIT_MAX_HEXSZ + 6];
 396         const int len = the_hash_algo->hexsz + 6;
 397
 398         /*
 399          * The first thing we expect from index-pack's output
 400          * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
 401          * %40s is the newly created pack SHA1 name.  In the "keep"
 402          * case, we need it to remove the corresponding .keep file
 403          * later on.  If we don't get that then tough luck with it.
 404          */
 405         if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
 406                 const char *name;
 407
 408                 if (is_well_formed)
 409                         *is_well_formed = 1;
 410                 packname[len-1] = 0;
 411                 if (skip_prefix(packname, "keep\t", &name))
 412                         return xstrfmt("%s/pack/pack-%s.keep",
 413                                        get_object_directory(), name);
 414                 return NULL;
 415         }
 416         if (is_well_formed)
 417                 *is_well_formed = 0;
 418         return NULL;
 419 }
 420
 421 /*
 422  * The per-object header is a pretty dense thing, which is
 423  *  - first byte: low four bits are "size", then three bits of "type",
 424  *    and the high bit is "size continues".
 425  *  - each byte afterwards: low seven bits are size continuation,
 426  *    with the high bit being "size continues"
 427  */
 428 int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
 429                                  enum object_type type, uintmax_t size)
 430 {
 431         int n = 1;
 432         unsigned char c;
 433
 434         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 435                 die("bad type %d", type);
 436
 437         c = (type << 4) | (size & 15);
 438         size >>= 4;
 439         while (size) {
 440                 if (n == hdr_len)
 441                         die("object size is too enormous to format");
 442                 *hdr++ = c | 0x80;
 443                 c = size & 0x7f;
 444                 size >>= 7;
 445                 n++;
 446         }
 447         *hdr = c;
 448         return n;
 449 }
 450
 451 struct hashfile *create_tmp_packfile(char **pack_tmp_name)
 452 {
 453         struct strbuf tmpname = STRBUF_INIT;
 454         int fd;
 455
 456         fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
 457         *pack_tmp_name = strbuf_detach(&tmpname, NULL);
 458         return hashfd(fd, *pack_tmp_name);
 459 }
 460
 461 void finish_tmp_packfile(struct strbuf *name_buffer,
 462                          const char *pack_tmp_name,
 463                          struct pack_idx_entry **written_list,
 464                          uint32_t nr_written,
 465                          struct pack_idx_option *pack_idx_opts,
 466                          unsigned char hash[])
 467 {
 468         const char *idx_tmp_name, *rev_tmp_name = NULL;
 469         int basename_len = name_buffer->len;
 470
 471         if (adjust_shared_perm(pack_tmp_name))
 472                 die_errno("unable to make temporary pack file readable");
 473
 474         idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
 475                                       pack_idx_opts, hash);
 476         if (adjust_shared_perm(idx_tmp_name))
 477                 die_errno("unable to make temporary index file readable");
 478
 479         rev_tmp_name = write_rev_file(NULL, written_list, nr_written, hash,
 480                                       pack_idx_opts->flags);
 481
 482         strbuf_addf(name_buffer, "%s.pack", hash_to_hex(hash));
 483
 484         if (rename(pack_tmp_name, name_buffer->buf))
 485                 die_errno("unable to rename temporary pack file");
 486
 487         strbuf_setlen(name_buffer, basename_len);
 488
 489         strbuf_addf(name_buffer, "%s.idx", hash_to_hex(hash));
 490         if (rename(idx_tmp_name, name_buffer->buf))
 491                 die_errno("unable to rename temporary index file");
 492
 493         strbuf_setlen(name_buffer, basename_len);
 494
 495         if (rev_tmp_name) {
 496                 strbuf_addf(name_buffer, "%s.rev", hash_to_hex(hash));
 497                 if (rename(rev_tmp_name, name_buffer->buf))
 498                         die_errno("unable to rename temporary reverse-index file");
 499         }
 500
 501         strbuf_setlen(name_buffer, basename_len);
 502
 503         free((void *)idx_tmp_name);
 504 }
 505
 506 void write_promisor_file(const char *promisor_name, struct ref **sought, int nr_sought)
 507 {
 508         int i, err;
 509         FILE *output = xfopen(promisor_name, "w");
 510
 511         for (i = 0; i < nr_sought; i++)
 512                 fprintf(output, "%s %s\n", oid_to_hex(&sought[i]->old_oid),
 513                         sought[i]->name);
 514
 515         err = ferror(output);
 516         err |= fclose(output);
 517         if (err)
 518                 die(_("could not write '%s' promisor file"), promisor_name);
 519 }