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