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, FSYNC_COMPONENT_PACK_METADATA,
 163                           CSUM_HASH_IN_STREAM | CSUM_CLOSE |
 164                           ((opts->flags & WRITE_IDX_VERIFY) ? 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         if (!(flags & WRITE_REV) && !(flags & WRITE_REV_VERIFY))
 226                 return NULL;
 227
 228         ALLOC_ARRAY(pack_order, nr_objects);
 229         for (i = 0; i < nr_objects; i++)
 230                 pack_order[i] = i;
 231         QSORT_S(pack_order, nr_objects, pack_order_cmp, objects);
 232
 233         ret = write_rev_file_order(rev_name, pack_order, nr_objects, hash,
 234                                    flags);
 235
 236         free(pack_order);
 237
 238         return ret;
 239 }
 240
 241 const char *write_rev_file_order(const char *rev_name,
 242                                  uint32_t *pack_order,
 243                                  uint32_t nr_objects,
 244                                  const unsigned char *hash,
 245                                  unsigned flags)
 246 {
 247         struct hashfile *f;
 248         int fd;
 249
 250         if ((flags & WRITE_REV) && (flags & WRITE_REV_VERIFY))
 251                 die(_("cannot both write and verify reverse index"));
 252
 253         if (flags & WRITE_REV) {
 254                 if (!rev_name) {
 255                         struct strbuf tmp_file = STRBUF_INIT;
 256                         fd = odb_mkstemp(&tmp_file, "pack/tmp_rev_XXXXXX");
 257                         rev_name = strbuf_detach(&tmp_file, NULL);
 258                 } else {
 259                         unlink(rev_name);
 260                         fd = xopen(rev_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
 261                 }
 262                 f = hashfd(fd, rev_name);
 263         } else if (flags & WRITE_REV_VERIFY) {
 264                 struct stat statbuf;
 265                 if (stat(rev_name, &statbuf)) {
 266                         if (errno == ENOENT) {
 267                                 /* .rev files are optional */
 268                                 return NULL;
 269                         } else
 270                                 die_errno(_("could not stat: %s"), rev_name);
 271                 }
 272                 f = hashfd_check(rev_name);
 273         } else
 274                 return NULL;
 275
 276         write_rev_header(f);
 277
 278         write_rev_index_positions(f, pack_order, nr_objects);
 279         write_rev_trailer(f, hash);
 280
 281         if (rev_name && adjust_shared_perm(rev_name) < 0)
 282                 die(_("failed to make %s readable"), rev_name);
 283
 284         finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,
 285                           CSUM_HASH_IN_STREAM | CSUM_CLOSE |
 286                           ((flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));
 287
 288         return rev_name;
 289 }
 290
 291 off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)
 292 {
 293         struct pack_header hdr;
 294
 295         hdr.hdr_signature = htonl(PACK_SIGNATURE);
 296         hdr.hdr_version = htonl(PACK_VERSION);
 297         hdr.hdr_entries = htonl(nr_entries);
 298         hashwrite(f, &hdr, sizeof(hdr));
 299         return sizeof(hdr);
 300 }
 301
 302 /*
 303  * Update pack header with object_count and compute new SHA1 for pack data
 304  * associated to pack_fd, and write that SHA1 at the end.  That new SHA1
 305  * is also returned in new_pack_sha1.
 306  *
 307  * If partial_pack_sha1 is non null, then the SHA1 of the existing pack
 308  * (without the header update) is computed and validated against the
 309  * one provided in partial_pack_sha1.  The validation is performed at
 310  * partial_pack_offset bytes in the pack file.  The SHA1 of the remaining
 311  * data (i.e. from partial_pack_offset to the end) is then computed and
 312  * returned in partial_pack_sha1.
 313  *
 314  * Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
 315  * partial_pack_sha1 can refer to the same buffer if the caller is not
 316  * interested in the resulting SHA1 of pack data above partial_pack_offset.
 317  */
 318 void fixup_pack_header_footer(int pack_fd,
 319                          unsigned char *new_pack_hash,
 320                          const char *pack_name,
 321                          uint32_t object_count,
 322                          unsigned char *partial_pack_hash,
 323                          off_t partial_pack_offset)
 324 {
 325         int aligned_sz, buf_sz = 8 * 1024;
 326         git_hash_ctx old_hash_ctx, new_hash_ctx;
 327         struct pack_header hdr;
 328         char *buf;
 329         ssize_t read_result;
 330
 331         the_hash_algo->init_fn(&old_hash_ctx);
 332         the_hash_algo->init_fn(&new_hash_ctx);
 333
 334         if (lseek(pack_fd, 0, SEEK_SET) != 0)
 335                 die_errno("Failed seeking to start of '%s'", pack_name);
 336         read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));
 337         if (read_result < 0)
 338                 die_errno("Unable to reread header of '%s'", pack_name);
 339         else if (read_result != sizeof(hdr))
 340                 die_errno("Unexpected short read for header of '%s'",
 341                           pack_name);
 342         if (lseek(pack_fd, 0, SEEK_SET) != 0)
 343                 die_errno("Failed seeking to start of '%s'", pack_name);
 344         the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));
 345         hdr.hdr_entries = htonl(object_count);
 346         the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));
 347         write_or_die(pack_fd, &hdr, sizeof(hdr));
 348         partial_pack_offset -= sizeof(hdr);
 349
 350         buf = xmalloc(buf_sz);
 351         aligned_sz = buf_sz - sizeof(hdr);
 352         for (;;) {
 353                 ssize_t m, n;
 354                 m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?
 355                         partial_pack_offset : aligned_sz;
 356                 n = xread(pack_fd, buf, m);
 357                 if (!n)
 358                         break;
 359                 if (n < 0)
 360                         die_errno("Failed to checksum '%s'", pack_name);
 361                 the_hash_algo->update_fn(&new_hash_ctx, buf, n);
 362
 363                 aligned_sz -= n;
 364                 if (!aligned_sz)
 365                         aligned_sz = buf_sz;
 366
 367                 if (!partial_pack_hash)
 368                         continue;
 369
 370                 the_hash_algo->update_fn(&old_hash_ctx, buf, n);
 371                 partial_pack_offset -= n;
 372                 if (partial_pack_offset == 0) {
 373                         unsigned char hash[GIT_MAX_RAWSZ];
 374                         the_hash_algo->final_fn(hash, &old_hash_ctx);
 375                         if (!hasheq(hash, partial_pack_hash))
 376                                 die("Unexpected checksum for %s "
 377                                     "(disk corruption?)", pack_name);
 378                         /*
 379                          * Now let's compute the SHA1 of the remainder of the
 380                          * pack, which also means making partial_pack_offset
 381                          * big enough not to matter anymore.
 382                          */
 383                         the_hash_algo->init_fn(&old_hash_ctx);
 384                         partial_pack_offset = ~partial_pack_offset;
 385                         partial_pack_offset -= MSB(partial_pack_offset, 1);
 386                 }
 387         }
 388         free(buf);
 389
 390         if (partial_pack_hash)
 391                 the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);
 392         the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);
 393         write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);
 394         fsync_component_or_die(FSYNC_COMPONENT_PACK, pack_fd, pack_name);
 395 }
 396
 397 char *index_pack_lockfile(int ip_out, int *is_well_formed)
 398 {
 399         char packname[GIT_MAX_HEXSZ + 6];
 400         const int len = the_hash_algo->hexsz + 6;
 401
 402         /*
 403          * The first thing we expect from index-pack's output
 404          * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
 405          * %40s is the newly created pack SHA1 name.  In the "keep"
 406          * case, we need it to remove the corresponding .keep file
 407          * later on.  If we don't get that then tough luck with it.
 408          */
 409         if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
 410                 const char *name;
 411
 412                 if (is_well_formed)
 413                         *is_well_formed = 1;
 414                 packname[len-1] = 0;
 415                 if (skip_prefix(packname, "keep\t", &name))
 416                         return xstrfmt("%s/pack/pack-%s.keep",
 417                                        get_object_directory(), name);
 418                 return NULL;
 419         }
 420         if (is_well_formed)
 421                 *is_well_formed = 0;
 422         return NULL;
 423 }
 424
 425 /*
 426  * The per-object header is a pretty dense thing, which is
 427  *  - first byte: low four bits are "size", then three bits of "type",
 428  *    and the high bit is "size continues".
 429  *  - each byte afterwards: low seven bits are size continuation,
 430  *    with the high bit being "size continues"
 431  */
 432 int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,
 433                                  enum object_type type, uintmax_t size)
 434 {
 435         int n = 1;
 436         unsigned char c;
 437
 438         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 439                 die("bad type %d", type);
 440
 441         c = (type << 4) | (size & 15);
 442         size >>= 4;
 443         while (size) {
 444                 if (n == hdr_len)
 445                         die("object size is too enormous to format");
 446                 *hdr++ = c | 0x80;
 447                 c = size & 0x7f;
 448                 size >>= 7;
 449                 n++;
 450         }
 451         *hdr = c;
 452         return n;
 453 }
 454
 455 struct hashfile *create_tmp_packfile(char **pack_tmp_name)
 456 {
 457         struct strbuf tmpname = STRBUF_INIT;
 458         int fd;
 459
 460         fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");
 461         *pack_tmp_name = strbuf_detach(&tmpname, NULL);
 462         return hashfd(fd, *pack_tmp_name);
 463 }
 464
 465 static void rename_tmp_packfile(struct strbuf *name_prefix, const char *source,
 466                                 const char *ext)
 467 {
 468         size_t name_prefix_len = name_prefix->len;
 469
 470         strbuf_addstr(name_prefix, ext);
 471         if (rename(source, name_prefix->buf))
 472                 die_errno("unable to rename temporary file to '%s'",
 473                           name_prefix->buf);
 474         strbuf_setlen(name_prefix, name_prefix_len);
 475 }
 476
 477 void rename_tmp_packfile_idx(struct strbuf *name_buffer,
 478                              char **idx_tmp_name)
 479 {
 480         rename_tmp_packfile(name_buffer, *idx_tmp_name, "idx");
 481 }
 482
 483 void stage_tmp_packfiles(struct strbuf *name_buffer,
 484                          const char *pack_tmp_name,
 485                          struct pack_idx_entry **written_list,
 486                          uint32_t nr_written,
 487                          struct pack_idx_option *pack_idx_opts,
 488                          unsigned char hash[],
 489                          char **idx_tmp_name)
 490 {
 491         const char *rev_tmp_name = NULL;
 492
 493         if (adjust_shared_perm(pack_tmp_name))
 494                 die_errno("unable to make temporary pack file readable");
 495
 496         *idx_tmp_name = (char *)write_idx_file(NULL, written_list, nr_written,
 497                                                pack_idx_opts, hash);
 498         if (adjust_shared_perm(*idx_tmp_name))
 499                 die_errno("unable to make temporary index file readable");
 500
 501         rev_tmp_name = write_rev_file(NULL, written_list, nr_written, hash,
 502                                       pack_idx_opts->flags);
 503
 504         rename_tmp_packfile(name_buffer, pack_tmp_name, "pack");
 505         if (rev_tmp_name)
 506                 rename_tmp_packfile(name_buffer, rev_tmp_name, "rev");
 507 }
 508
 509 void write_promisor_file(const char *promisor_name, struct ref **sought, int nr_sought)
 510 {
 511         int i, err;
 512         FILE *output = xfopen(promisor_name, "w");
 513
 514         for (i = 0; i < nr_sought; i++)
 515                 fprintf(output, "%s %s\n", oid_to_hex(&sought[i]->old_oid),
 516                         sought[i]->name);
 517
 518         err = ferror(output);
 519         err |= fclose(output);
 520         if (err)
 521                 die(_("could not write '%s' promisor file"), promisor_name);
 522 }