git.oblomov.eu Git - git/blob - sha1_file.c

   1 /*
   2  * GIT - The information manager from hell
   3  *
   4  * Copyright (C) Linus Torvalds, 2005
   5  *
   6  * This handles basic git sha1 object files - packing, unpacking,
   7  * creation etc.
   8  */
   9 #include "cache.h"
  10 #include "config.h"
  11 #include "string-list.h"
  12 #include "lockfile.h"
  13 #include "delta.h"
  14 #include "pack.h"
  15 #include "blob.h"
  16 #include "commit.h"
  17 #include "run-command.h"
  18 #include "tag.h"
  19 #include "tree.h"
  20 #include "tree-walk.h"
  21 #include "refs.h"
  22 #include "pack-revindex.h"
  23 #include "sha1-lookup.h"
  24 #include "bulk-checkin.h"
  25 #include "streaming.h"
  26 #include "dir.h"
  27 #include "list.h"
  28 #include "mergesort.h"
  29 #include "quote.h"
  30 #include "packfile.h"
  31 #include "fetch-object.h"
  32
  33 /* The maximum size for an object header. */
  34 #define MAX_HEADER_LEN 32
  35
  36 const unsigned char null_sha1[GIT_MAX_RAWSZ];
  37 const struct object_id null_oid;
  38 const struct object_id empty_tree_oid = {
  39         EMPTY_TREE_SHA1_BIN_LITERAL
  40 };
  41 const struct object_id empty_blob_oid = {
  42         EMPTY_BLOB_SHA1_BIN_LITERAL
  43 };
  44
  45 static void git_hash_sha1_init(git_hash_ctx *ctx)
  46 {
  47         git_SHA1_Init(&ctx->sha1);
  48 }
  49
  50 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
  51 {
  52         git_SHA1_Update(&ctx->sha1, data, len);
  53 }
  54
  55 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
  56 {
  57         git_SHA1_Final(hash, &ctx->sha1);
  58 }
  59
  60 static void git_hash_unknown_init(git_hash_ctx *ctx)
  61 {
  62         die("trying to init unknown hash");
  63 }
  64
  65 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
  66 {
  67         die("trying to update unknown hash");
  68 }
  69
  70 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
  71 {
  72         die("trying to finalize unknown hash");
  73 }
  74
  75 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
  76         {
  77                 NULL,
  78                 0x00000000,
  79                 0,
  80                 0,
  81                 git_hash_unknown_init,
  82                 git_hash_unknown_update,
  83                 git_hash_unknown_final,
  84                 NULL,
  85                 NULL,
  86         },
  87         {
  88                 "sha-1",
  89                 /* "sha1", big-endian */
  90                 0x73686131,
  91                 GIT_SHA1_RAWSZ,
  92                 GIT_SHA1_HEXSZ,
  93                 git_hash_sha1_init,
  94                 git_hash_sha1_update,
  95                 git_hash_sha1_final,
  96                 &empty_tree_oid,
  97                 &empty_blob_oid,
  98         },
  99 };
 100
 101 /*
 102  * This is meant to hold a *small* number of objects that you would
 103  * want read_sha1_file() to be able to return, but yet you do not want
 104  * to write them into the object store (e.g. a browse-only
 105  * application).
 106  */
 107 static struct cached_object {
 108         unsigned char sha1[20];
 109         enum object_type type;
 110         void *buf;
 111         unsigned long size;
 112 } *cached_objects;
 113 static int cached_object_nr, cached_object_alloc;
 114
 115 static struct cached_object empty_tree = {
 116         EMPTY_TREE_SHA1_BIN_LITERAL,
 117         OBJ_TREE,
 118         "",
 119         0
 120 };
 121
 122 static struct cached_object *find_cached_object(const unsigned char *sha1)
 123 {
 124         int i;
 125         struct cached_object *co = cached_objects;
 126
 127         for (i = 0; i < cached_object_nr; i++, co++) {
 128                 if (!hashcmp(co->sha1, sha1))
 129                         return co;
 130         }
 131         if (!hashcmp(sha1, empty_tree.sha1))
 132                 return &empty_tree;
 133         return NULL;
 134 }
 135
 136
 137 static int get_conv_flags(unsigned flags)
 138 {
 139         if (flags & HASH_RENORMALIZE)
 140                 return CONV_EOL_RENORMALIZE;
 141         else if (flags & HASH_WRITE_OBJECT)
 142           return global_conv_flags_eol;
 143         else
 144                 return 0;
 145 }
 146
 147
 148 int mkdir_in_gitdir(const char *path)
 149 {
 150         if (mkdir(path, 0777)) {
 151                 int saved_errno = errno;
 152                 struct stat st;
 153                 struct strbuf sb = STRBUF_INIT;
 154
 155                 if (errno != EEXIST)
 156                         return -1;
 157                 /*
 158                  * Are we looking at a path in a symlinked worktree
 159                  * whose original repository does not yet have it?
 160                  * e.g. .git/rr-cache pointing at its original
 161                  * repository in which the user hasn't performed any
 162                  * conflict resolution yet?
 163                  */
 164                 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
 165                     strbuf_readlink(&sb, path, st.st_size) ||
 166                     !is_absolute_path(sb.buf) ||
 167                     mkdir(sb.buf, 0777)) {
 168                         strbuf_release(&sb);
 169                         errno = saved_errno;
 170                         return -1;
 171                 }
 172                 strbuf_release(&sb);
 173         }
 174         return adjust_shared_perm(path);
 175 }
 176
 177 enum scld_error safe_create_leading_directories(char *path)
 178 {
 179         char *next_component = path + offset_1st_component(path);
 180         enum scld_error ret = SCLD_OK;
 181
 182         while (ret == SCLD_OK && next_component) {
 183                 struct stat st;
 184                 char *slash = next_component, slash_character;
 185
 186                 while (*slash && !is_dir_sep(*slash))
 187                         slash++;
 188
 189                 if (!*slash)
 190                         break;
 191
 192                 next_component = slash + 1;
 193                 while (is_dir_sep(*next_component))
 194                         next_component++;
 195                 if (!*next_component)
 196                         break;
 197
 198                 slash_character = *slash;
 199                 *slash = '\0';
 200                 if (!stat(path, &st)) {
 201                         /* path exists */
 202                         if (!S_ISDIR(st.st_mode)) {
 203                                 errno = ENOTDIR;
 204                                 ret = SCLD_EXISTS;
 205                         }
 206                 } else if (mkdir(path, 0777)) {
 207                         if (errno == EEXIST &&
 208                             !stat(path, &st) && S_ISDIR(st.st_mode))
 209                                 ; /* somebody created it since we checked */
 210                         else if (errno == ENOENT)
 211                                 /*
 212                                  * Either mkdir() failed because
 213                                  * somebody just pruned the containing
 214                                  * directory, or stat() failed because
 215                                  * the file that was in our way was
 216                                  * just removed.  Either way, inform
 217                                  * the caller that it might be worth
 218                                  * trying again:
 219                                  */
 220                                 ret = SCLD_VANISHED;
 221                         else
 222                                 ret = SCLD_FAILED;
 223                 } else if (adjust_shared_perm(path)) {
 224                         ret = SCLD_PERMS;
 225                 }
 226                 *slash = slash_character;
 227         }
 228         return ret;
 229 }
 230
 231 enum scld_error safe_create_leading_directories_const(const char *path)
 232 {
 233         int save_errno;
 234         /* path points to cache entries, so xstrdup before messing with it */
 235         char *buf = xstrdup(path);
 236         enum scld_error result = safe_create_leading_directories(buf);
 237
 238         save_errno = errno;
 239         free(buf);
 240         errno = save_errno;
 241         return result;
 242 }
 243
 244 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
 245 {
 246         /*
 247          * The number of times we will try to remove empty directories
 248          * in the way of path. This is only 1 because if another
 249          * process is racily creating directories that conflict with
 250          * us, we don't want to fight against them.
 251          */
 252         int remove_directories_remaining = 1;
 253
 254         /*
 255          * The number of times that we will try to create the
 256          * directories containing path. We are willing to attempt this
 257          * more than once, because another process could be trying to
 258          * clean up empty directories at the same time as we are
 259          * trying to create them.
 260          */
 261         int create_directories_remaining = 3;
 262
 263         /* A scratch copy of path, filled lazily if we need it: */
 264         struct strbuf path_copy = STRBUF_INIT;
 265
 266         int ret, save_errno;
 267
 268         /* Sanity check: */
 269         assert(*path);
 270
 271 retry_fn:
 272         ret = fn(path, cb);
 273         save_errno = errno;
 274         if (!ret)
 275                 goto out;
 276
 277         if (errno == EISDIR && remove_directories_remaining-- > 0) {
 278                 /*
 279                  * A directory is in the way. Maybe it is empty; try
 280                  * to remove it:
 281                  */
 282                 if (!path_copy.len)
 283                         strbuf_addstr(&path_copy, path);
 284
 285                 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
 286                         goto retry_fn;
 287         } else if (errno == ENOENT && create_directories_remaining-- > 0) {
 288                 /*
 289                  * Maybe the containing directory didn't exist, or
 290                  * maybe it was just deleted by a process that is
 291                  * racing with us to clean up empty directories. Try
 292                  * to create it:
 293                  */
 294                 enum scld_error scld_result;
 295
 296                 if (!path_copy.len)
 297                         strbuf_addstr(&path_copy, path);
 298
 299                 do {
 300                         scld_result = safe_create_leading_directories(path_copy.buf);
 301                         if (scld_result == SCLD_OK)
 302                                 goto retry_fn;
 303                 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
 304         }
 305
 306 out:
 307         strbuf_release(&path_copy);
 308         errno = save_errno;
 309         return ret;
 310 }
 311
 312 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 313 {
 314         int i;
 315         for (i = 0; i < 20; i++) {
 316                 static char hex[] = "0123456789abcdef";
 317                 unsigned int val = sha1[i];
 318                 strbuf_addch(buf, hex[val >> 4]);
 319                 strbuf_addch(buf, hex[val & 0xf]);
 320                 if (!i)
 321                         strbuf_addch(buf, '/');
 322         }
 323 }
 324
 325 void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
 326 {
 327         strbuf_addstr(buf, get_object_directory());
 328         strbuf_addch(buf, '/');
 329         fill_sha1_path(buf, sha1);
 330 }
 331
 332 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
 333 {
 334         strbuf_setlen(&alt->scratch, alt->base_len);
 335         return &alt->scratch;
 336 }
 337
 338 static const char *alt_sha1_path(struct alternate_object_database *alt,
 339                                  const unsigned char *sha1)
 340 {
 341         struct strbuf *buf = alt_scratch_buf(alt);
 342         fill_sha1_path(buf, sha1);
 343         return buf->buf;
 344 }
 345
 346 struct alternate_object_database *alt_odb_list;
 347 static struct alternate_object_database **alt_odb_tail;
 348
 349 /*
 350  * Return non-zero iff the path is usable as an alternate object database.
 351  */
 352 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
 353 {
 354         struct alternate_object_database *alt;
 355
 356         /* Detect cases where alternate disappeared */
 357         if (!is_directory(path->buf)) {
 358                 error("object directory %s does not exist; "
 359                       "check .git/objects/info/alternates.",
 360                       path->buf);
 361                 return 0;
 362         }
 363
 364         /*
 365          * Prevent the common mistake of listing the same
 366          * thing twice, or object directory itself.
 367          */
 368         for (alt = alt_odb_list; alt; alt = alt->next) {
 369                 if (!fspathcmp(path->buf, alt->path))
 370                         return 0;
 371         }
 372         if (!fspathcmp(path->buf, normalized_objdir))
 373                 return 0;
 374
 375         return 1;
 376 }
 377
 378 /*
 379  * Prepare alternate object database registry.
 380  *
 381  * The variable alt_odb_list points at the list of struct
 382  * alternate_object_database.  The elements on this list come from
 383  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 384  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 385  * whose contents is similar to that environment variable but can be
 386  * LF separated.  Its base points at a statically allocated buffer that
 387  * contains "/the/directory/corresponding/to/.git/objects/...", while
 388  * its name points just after the slash at the end of ".git/objects/"
 389  * in the example above, and has enough space to hold 40-byte hex
 390  * SHA1, an extra slash for the first level indirection, and the
 391  * terminating NUL.
 392  */
 393 static void read_info_alternates(const char * relative_base, int depth);
 394 static int link_alt_odb_entry(const char *entry, const char *relative_base,
 395         int depth, const char *normalized_objdir)
 396 {
 397         struct alternate_object_database *ent;
 398         struct strbuf pathbuf = STRBUF_INIT;
 399
 400         if (!is_absolute_path(entry) && relative_base) {
 401                 strbuf_realpath(&pathbuf, relative_base, 1);
 402                 strbuf_addch(&pathbuf, '/');
 403         }
 404         strbuf_addstr(&pathbuf, entry);
 405
 406         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 407                 error("unable to normalize alternate object path: %s",
 408                       pathbuf.buf);
 409                 strbuf_release(&pathbuf);
 410                 return -1;
 411         }
 412
 413         /*
 414          * The trailing slash after the directory name is given by
 415          * this function at the end. Remove duplicates.
 416          */
 417         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 418                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
 419
 420         if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
 421                 strbuf_release(&pathbuf);
 422                 return -1;
 423         }
 424
 425         ent = alloc_alt_odb(pathbuf.buf);
 426
 427         /* add the alternate entry */
 428         *alt_odb_tail = ent;
 429         alt_odb_tail = &(ent->next);
 430         ent->next = NULL;
 431
 432         /* recursively add alternates */
 433         read_info_alternates(pathbuf.buf, depth + 1);
 434
 435         strbuf_release(&pathbuf);
 436         return 0;
 437 }
 438
 439 static const char *parse_alt_odb_entry(const char *string,
 440                                        int sep,
 441                                        struct strbuf *out)
 442 {
 443         const char *end;
 444
 445         strbuf_reset(out);
 446
 447         if (*string == '#') {
 448                 /* comment; consume up to next separator */
 449                 end = strchrnul(string, sep);
 450         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
 451                 /*
 452                  * quoted path; unquote_c_style has copied the
 453                  * data for us and set "end". Broken quoting (e.g.,
 454                  * an entry that doesn't end with a quote) falls
 455                  * back to the unquoted case below.
 456                  */
 457         } else {
 458                 /* normal, unquoted path */
 459                 end = strchrnul(string, sep);
 460                 strbuf_add(out, string, end - string);
 461         }
 462
 463         if (*end)
 464                 end++;
 465         return end;
 466 }
 467
 468 static void link_alt_odb_entries(const char *alt, int sep,
 469                                  const char *relative_base, int depth)
 470 {
 471         struct strbuf objdirbuf = STRBUF_INIT;
 472         struct strbuf entry = STRBUF_INIT;
 473
 474         if (!alt || !*alt)
 475                 return;
 476
 477         if (depth > 5) {
 478                 error("%s: ignoring alternate object stores, nesting too deep.",
 479                                 relative_base);
 480                 return;
 481         }
 482
 483         strbuf_add_absolute_path(&objdirbuf, get_object_directory());
 484         if (strbuf_normalize_path(&objdirbuf) < 0)
 485                 die("unable to normalize object directory: %s",
 486                     objdirbuf.buf);
 487
 488         while (*alt) {
 489                 alt = parse_alt_odb_entry(alt, sep, &entry);
 490                 if (!entry.len)
 491                         continue;
 492                 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
 493         }
 494         strbuf_release(&entry);
 495         strbuf_release(&objdirbuf);
 496 }
 497
 498 static void read_info_alternates(const char * relative_base, int depth)
 499 {
 500         char *path;
 501         struct strbuf buf = STRBUF_INIT;
 502
 503         path = xstrfmt("%s/info/alternates", relative_base);
 504         if (strbuf_read_file(&buf, path, 1024) < 0) {
 505                 warn_on_fopen_errors(path);
 506                 free(path);
 507                 return;
 508         }
 509
 510         link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
 511         strbuf_release(&buf);
 512         free(path);
 513 }
 514
 515 struct alternate_object_database *alloc_alt_odb(const char *dir)
 516 {
 517         struct alternate_object_database *ent;
 518
 519         FLEX_ALLOC_STR(ent, path, dir);
 520         strbuf_init(&ent->scratch, 0);
 521         strbuf_addf(&ent->scratch, "%s/", dir);
 522         ent->base_len = ent->scratch.len;
 523
 524         return ent;
 525 }
 526
 527 void add_to_alternates_file(const char *reference)
 528 {
 529         struct lock_file lock = LOCK_INIT;
 530         char *alts = git_pathdup("objects/info/alternates");
 531         FILE *in, *out;
 532         int found = 0;
 533
 534         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
 535         out = fdopen_lock_file(&lock, "w");
 536         if (!out)
 537                 die_errno("unable to fdopen alternates lockfile");
 538
 539         in = fopen(alts, "r");
 540         if (in) {
 541                 struct strbuf line = STRBUF_INIT;
 542
 543                 while (strbuf_getline(&line, in) != EOF) {
 544                         if (!strcmp(reference, line.buf)) {
 545                                 found = 1;
 546                                 break;
 547                         }
 548                         fprintf_or_die(out, "%s\n", line.buf);
 549                 }
 550
 551                 strbuf_release(&line);
 552                 fclose(in);
 553         }
 554         else if (errno != ENOENT)
 555                 die_errno("unable to read alternates file");
 556
 557         if (found) {
 558                 rollback_lock_file(&lock);
 559         } else {
 560                 fprintf_or_die(out, "%s\n", reference);
 561                 if (commit_lock_file(&lock))
 562                         die_errno("unable to move new alternates file into place");
 563                 if (alt_odb_tail)
 564                         link_alt_odb_entries(reference, '\n', NULL, 0);
 565         }
 566         free(alts);
 567 }
 568
 569 void add_to_alternates_memory(const char *reference)
 570 {
 571         /*
 572          * Make sure alternates are initialized, or else our entry may be
 573          * overwritten when they are.
 574          */
 575         prepare_alt_odb();
 576
 577         link_alt_odb_entries(reference, '\n', NULL, 0);
 578 }
 579
 580 /*
 581  * Compute the exact path an alternate is at and returns it. In case of
 582  * error NULL is returned and the human readable error is added to `err`
 583  * `path` may be relative and should point to $GITDIR.
 584  * `err` must not be null.
 585  */
 586 char *compute_alternate_path(const char *path, struct strbuf *err)
 587 {
 588         char *ref_git = NULL;
 589         const char *repo, *ref_git_s;
 590         int seen_error = 0;
 591
 592         ref_git_s = real_path_if_valid(path);
 593         if (!ref_git_s) {
 594                 seen_error = 1;
 595                 strbuf_addf(err, _("path '%s' does not exist"), path);
 596                 goto out;
 597         } else
 598                 /*
 599                  * Beware: read_gitfile(), real_path() and mkpath()
 600                  * return static buffer
 601                  */
 602                 ref_git = xstrdup(ref_git_s);
 603
 604         repo = read_gitfile(ref_git);
 605         if (!repo)
 606                 repo = read_gitfile(mkpath("%s/.git", ref_git));
 607         if (repo) {
 608                 free(ref_git);
 609                 ref_git = xstrdup(repo);
 610         }
 611
 612         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 613                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
 614                 free(ref_git);
 615                 ref_git = ref_git_git;
 616         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 617                 struct strbuf sb = STRBUF_INIT;
 618                 seen_error = 1;
 619                 if (get_common_dir(&sb, ref_git)) {
 620                         strbuf_addf(err,
 621                                     _("reference repository '%s' as a linked "
 622                                       "checkout is not supported yet."),
 623                                     path);
 624                         goto out;
 625                 }
 626
 627                 strbuf_addf(err, _("reference repository '%s' is not a "
 628                                         "local repository."), path);
 629                 goto out;
 630         }
 631
 632         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 633                 strbuf_addf(err, _("reference repository '%s' is shallow"),
 634                             path);
 635                 seen_error = 1;
 636                 goto out;
 637         }
 638
 639         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 640                 strbuf_addf(err,
 641                             _("reference repository '%s' is grafted"),
 642                             path);
 643                 seen_error = 1;
 644                 goto out;
 645         }
 646
 647 out:
 648         if (seen_error) {
 649                 FREE_AND_NULL(ref_git);
 650         }
 651
 652         return ref_git;
 653 }
 654
 655 int foreach_alt_odb(alt_odb_fn fn, void *cb)
 656 {
 657         struct alternate_object_database *ent;
 658         int r = 0;
 659
 660         prepare_alt_odb();
 661         for (ent = alt_odb_list; ent; ent = ent->next) {
 662                 r = fn(ent, cb);
 663                 if (r)
 664                         break;
 665         }
 666         return r;
 667 }
 668
 669 void prepare_alt_odb(void)
 670 {
 671         const char *alt;
 672
 673         if (alt_odb_tail)
 674                 return;
 675
 676         alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 677
 678         alt_odb_tail = &alt_odb_list;
 679         link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
 680
 681         read_info_alternates(get_object_directory(), 0);
 682 }
 683
 684 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 685 static int freshen_file(const char *fn)
 686 {
 687         struct utimbuf t;
 688         t.actime = t.modtime = time(NULL);
 689         return !utime(fn, &t);
 690 }
 691
 692 /*
 693  * All of the check_and_freshen functions return 1 if the file exists and was
 694  * freshened (if freshening was requested), 0 otherwise. If they return
 695  * 0, you should not assume that it is safe to skip a write of the object (it
 696  * either does not exist on disk, or has a stale mtime and may be subject to
 697  * pruning).
 698  */
 699 int check_and_freshen_file(const char *fn, int freshen)
 700 {
 701         if (access(fn, F_OK))
 702                 return 0;
 703         if (freshen && !freshen_file(fn))
 704                 return 0;
 705         return 1;
 706 }
 707
 708 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
 709 {
 710         static struct strbuf buf = STRBUF_INIT;
 711
 712         strbuf_reset(&buf);
 713         sha1_file_name(&buf, sha1);
 714
 715         return check_and_freshen_file(buf.buf, freshen);
 716 }
 717
 718 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
 719 {
 720         struct alternate_object_database *alt;
 721         prepare_alt_odb();
 722         for (alt = alt_odb_list; alt; alt = alt->next) {
 723                 const char *path = alt_sha1_path(alt, sha1);
 724                 if (check_and_freshen_file(path, freshen))
 725                         return 1;
 726         }
 727         return 0;
 728 }
 729
 730 static int check_and_freshen(const unsigned char *sha1, int freshen)
 731 {
 732         return check_and_freshen_local(sha1, freshen) ||
 733                check_and_freshen_nonlocal(sha1, freshen);
 734 }
 735
 736 int has_loose_object_nonlocal(const unsigned char *sha1)
 737 {
 738         return check_and_freshen_nonlocal(sha1, 0);
 739 }
 740
 741 static int has_loose_object(const unsigned char *sha1)
 742 {
 743         return check_and_freshen(sha1, 0);
 744 }
 745
 746 static void mmap_limit_check(size_t length)
 747 {
 748         static size_t limit = 0;
 749         if (!limit) {
 750                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 751                 if (!limit)
 752                         limit = SIZE_MAX;
 753         }
 754         if (length > limit)
 755                 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
 756                     (uintmax_t)length, (uintmax_t)limit);
 757 }
 758
 759 void *xmmap_gently(void *start, size_t length,
 760                   int prot, int flags, int fd, off_t offset)
 761 {
 762         void *ret;
 763
 764         mmap_limit_check(length);
 765         ret = mmap(start, length, prot, flags, fd, offset);
 766         if (ret == MAP_FAILED) {
 767                 if (!length)
 768                         return NULL;
 769                 release_pack_memory(length);
 770                 ret = mmap(start, length, prot, flags, fd, offset);
 771         }
 772         return ret;
 773 }
 774
 775 void *xmmap(void *start, size_t length,
 776         int prot, int flags, int fd, off_t offset)
 777 {
 778         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 779         if (ret == MAP_FAILED)
 780                 die_errno("mmap failed");
 781         return ret;
 782 }
 783
 784 /*
 785  * With an in-core object data in "map", rehash it to make sure the
 786  * object name actually matches "sha1" to detect object corruption.
 787  * With "map" == NULL, try reading the object named with "sha1" using
 788  * the streaming interface and rehash it to do the same.
 789  */
 790 int check_object_signature(const struct object_id *oid, void *map,
 791                            unsigned long size, const char *type)
 792 {
 793         struct object_id real_oid;
 794         enum object_type obj_type;
 795         struct git_istream *st;
 796         git_hash_ctx c;
 797         char hdr[MAX_HEADER_LEN];
 798         int hdrlen;
 799
 800         if (map) {
 801                 hash_object_file(map, size, type, &real_oid);
 802                 return oidcmp(oid, &real_oid) ? -1 : 0;
 803         }
 804
 805         st = open_istream(oid, &obj_type, &size, NULL);
 806         if (!st)
 807                 return -1;
 808
 809         /* Generate the header */
 810         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
 811
 812         /* Sha1.. */
 813         the_hash_algo->init_fn(&c);
 814         the_hash_algo->update_fn(&c, hdr, hdrlen);
 815         for (;;) {
 816                 char buf[1024 * 16];
 817                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
 818
 819                 if (readlen < 0) {
 820                         close_istream(st);
 821                         return -1;
 822                 }
 823                 if (!readlen)
 824                         break;
 825                 the_hash_algo->update_fn(&c, buf, readlen);
 826         }
 827         the_hash_algo->final_fn(real_oid.hash, &c);
 828         close_istream(st);
 829         return oidcmp(oid, &real_oid) ? -1 : 0;
 830 }
 831
 832 int git_open_cloexec(const char *name, int flags)
 833 {
 834         int fd;
 835         static int o_cloexec = O_CLOEXEC;
 836
 837         fd = open(name, flags | o_cloexec);
 838         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 839                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
 840                 o_cloexec &= ~O_CLOEXEC;
 841                 fd = open(name, flags | o_cloexec);
 842         }
 843
 844 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 845         {
 846                 static int fd_cloexec = FD_CLOEXEC;
 847
 848                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
 849                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 850                         int flags = fcntl(fd, F_GETFD);
 851                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 852                                 fd_cloexec = 0;
 853                 }
 854         }
 855 #endif
 856         return fd;
 857 }
 858
 859 /*
 860  * Find "sha1" as a loose object in the local repository or in an alternate.
 861  * Returns 0 on success, negative on failure.
 862  *
 863  * The "path" out-parameter will give the path of the object we found (if any).
 864  * Note that it may point to static storage and is only valid until another
 865  * call to sha1_file_name(), etc.
 866  */
 867 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
 868                           const char **path)
 869 {
 870         struct alternate_object_database *alt;
 871         static struct strbuf buf = STRBUF_INIT;
 872
 873         strbuf_reset(&buf);
 874         sha1_file_name(&buf, sha1);
 875         *path = buf.buf;
 876
 877         if (!lstat(*path, st))
 878                 return 0;
 879
 880         prepare_alt_odb();
 881         errno = ENOENT;
 882         for (alt = alt_odb_list; alt; alt = alt->next) {
 883                 *path = alt_sha1_path(alt, sha1);
 884                 if (!lstat(*path, st))
 885                         return 0;
 886         }
 887
 888         return -1;
 889 }
 890
 891 /*
 892  * Like stat_sha1_file(), but actually open the object and return the
 893  * descriptor. See the caveats on the "path" parameter above.
 894  */
 895 static int open_sha1_file(const unsigned char *sha1, const char **path)
 896 {
 897         int fd;
 898         struct alternate_object_database *alt;
 899         int most_interesting_errno;
 900         static struct strbuf buf = STRBUF_INIT;
 901
 902         strbuf_reset(&buf);
 903         sha1_file_name(&buf, sha1);
 904         *path = buf.buf;
 905
 906         fd = git_open(*path);
 907         if (fd >= 0)
 908                 return fd;
 909         most_interesting_errno = errno;
 910
 911         prepare_alt_odb();
 912         for (alt = alt_odb_list; alt; alt = alt->next) {
 913                 *path = alt_sha1_path(alt, sha1);
 914                 fd = git_open(*path);
 915                 if (fd >= 0)
 916                         return fd;
 917                 if (most_interesting_errno == ENOENT)
 918                         most_interesting_errno = errno;
 919         }
 920         errno = most_interesting_errno;
 921         return -1;
 922 }
 923
 924 /*
 925  * Map the loose object at "path" if it is not NULL, or the path found by
 926  * searching for a loose object named "sha1".
 927  */
 928 static void *map_sha1_file_1(const char *path,
 929                              const unsigned char *sha1,
 930                              unsigned long *size)
 931 {
 932         void *map;
 933         int fd;
 934
 935         if (path)
 936                 fd = git_open(path);
 937         else
 938                 fd = open_sha1_file(sha1, &path);
 939         map = NULL;
 940         if (fd >= 0) {
 941                 struct stat st;
 942
 943                 if (!fstat(fd, &st)) {
 944                         *size = xsize_t(st.st_size);
 945                         if (!*size) {
 946                                 /* mmap() is forbidden on empty files */
 947                                 error("object file %s is empty", path);
 948                                 return NULL;
 949                         }
 950                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 951                 }
 952                 close(fd);
 953         }
 954         return map;
 955 }
 956
 957 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 958 {
 959         return map_sha1_file_1(NULL, sha1, size);
 960 }
 961
 962 static int unpack_sha1_short_header(git_zstream *stream,
 963                                     unsigned char *map, unsigned long mapsize,
 964                                     void *buffer, unsigned long bufsiz)
 965 {
 966         /* Get the data stream */
 967         memset(stream, 0, sizeof(*stream));
 968         stream->next_in = map;
 969         stream->avail_in = mapsize;
 970         stream->next_out = buffer;
 971         stream->avail_out = bufsiz;
 972
 973         git_inflate_init(stream);
 974         return git_inflate(stream, 0);
 975 }
 976
 977 int unpack_sha1_header(git_zstream *stream,
 978                        unsigned char *map, unsigned long mapsize,
 979                        void *buffer, unsigned long bufsiz)
 980 {
 981         int status = unpack_sha1_short_header(stream, map, mapsize,
 982                                               buffer, bufsiz);
 983
 984         if (status < Z_OK)
 985                 return status;
 986
 987         /* Make sure we have the terminating NUL */
 988         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 989                 return -1;
 990         return 0;
 991 }
 992
 993 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
 994                                         unsigned long mapsize, void *buffer,
 995                                         unsigned long bufsiz, struct strbuf *header)
 996 {
 997         int status;
 998
 999         status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1000         if (status < Z_OK)
1001                 return -1;
1002
1003         /*
1004          * Check if entire header is unpacked in the first iteration.
1005          */
1006         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1007                 return 0;
1008
1009         /*
1010          * buffer[0..bufsiz] was not large enough.  Copy the partial
1011          * result out to header, and then append the result of further
1012          * reading the stream.
1013          */
1014         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1015         stream->next_out = buffer;
1016         stream->avail_out = bufsiz;
1017
1018         do {
1019                 status = git_inflate(stream, 0);
1020                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1021                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1022                         return 0;
1023                 stream->next_out = buffer;
1024                 stream->avail_out = bufsiz;
1025         } while (status != Z_STREAM_END);
1026         return -1;
1027 }
1028
1029 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1030 {
1031         int bytes = strlen(buffer) + 1;
1032         unsigned char *buf = xmallocz(size);
1033         unsigned long n;
1034         int status = Z_OK;
1035
1036         n = stream->total_out - bytes;
1037         if (n > size)
1038                 n = size;
1039         memcpy(buf, (char *) buffer + bytes, n);
1040         bytes = n;
1041         if (bytes <= size) {
1042                 /*
1043                  * The above condition must be (bytes <= size), not
1044                  * (bytes < size).  In other words, even though we
1045                  * expect no more output and set avail_out to zero,
1046                  * the input zlib stream may have bytes that express
1047                  * "this concludes the stream", and we *do* want to
1048                  * eat that input.
1049                  *
1050                  * Otherwise we would not be able to test that we
1051                  * consumed all the input to reach the expected size;
1052                  * we also want to check that zlib tells us that all
1053                  * went well with status == Z_STREAM_END at the end.
1054                  */
1055                 stream->next_out = buf + bytes;
1056                 stream->avail_out = size - bytes;
1057                 while (status == Z_OK)
1058                         status = git_inflate(stream, Z_FINISH);
1059         }
1060         if (status == Z_STREAM_END && !stream->avail_in) {
1061                 git_inflate_end(stream);
1062                 return buf;
1063         }
1064
1065         if (status < 0)
1066                 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1067         else if (stream->avail_in)
1068                 error("garbage at end of loose object '%s'",
1069                       sha1_to_hex(sha1));
1070         free(buf);
1071         return NULL;
1072 }
1073
1074 /*
1075  * We used to just use "sscanf()", but that's actually way
1076  * too permissive for what we want to check. So do an anal
1077  * object header parse by hand.
1078  */
1079 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1080                                unsigned int flags)
1081 {
1082         const char *type_buf = hdr;
1083         unsigned long size;
1084         int type, type_len = 0;
1085
1086         /*
1087          * The type can be of any size but is followed by
1088          * a space.
1089          */
1090         for (;;) {
1091                 char c = *hdr++;
1092                 if (!c)
1093                         return -1;
1094                 if (c == ' ')
1095                         break;
1096                 type_len++;
1097         }
1098
1099         type = type_from_string_gently(type_buf, type_len, 1);
1100         if (oi->type_name)
1101                 strbuf_add(oi->type_name, type_buf, type_len);
1102         /*
1103          * Set type to 0 if its an unknown object and
1104          * we're obtaining the type using '--allow-unknown-type'
1105          * option.
1106          */
1107         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1108                 type = 0;
1109         else if (type < 0)
1110                 die("invalid object type");
1111         if (oi->typep)
1112                 *oi->typep = type;
1113
1114         /*
1115          * The length must follow immediately, and be in canonical
1116          * decimal format (ie "010" is not valid).
1117          */
1118         size = *hdr++ - '0';
1119         if (size > 9)
1120                 return -1;
1121         if (size) {
1122                 for (;;) {
1123                         unsigned long c = *hdr - '0';
1124                         if (c > 9)
1125                                 break;
1126                         hdr++;
1127                         size = size * 10 + c;
1128                 }
1129         }
1130
1131         if (oi->sizep)
1132                 *oi->sizep = size;
1133
1134         /*
1135          * The length must be followed by a zero byte
1136          */
1137         return *hdr ? -1 : type;
1138 }
1139
1140 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1141 {
1142         struct object_info oi = OBJECT_INFO_INIT;
1143
1144         oi.sizep = sizep;
1145         return parse_sha1_header_extended(hdr, &oi, 0);
1146 }
1147
1148 static int sha1_loose_object_info(const unsigned char *sha1,
1149                                   struct object_info *oi,
1150                                   int flags)
1151 {
1152         int status = 0;
1153         unsigned long mapsize;
1154         void *map;
1155         git_zstream stream;
1156         char hdr[MAX_HEADER_LEN];
1157         struct strbuf hdrbuf = STRBUF_INIT;
1158         unsigned long size_scratch;
1159
1160         if (oi->delta_base_sha1)
1161                 hashclr(oi->delta_base_sha1);
1162
1163         /*
1164          * If we don't care about type or size, then we don't
1165          * need to look inside the object at all. Note that we
1166          * do not optimize out the stat call, even if the
1167          * caller doesn't care about the disk-size, since our
1168          * return value implicitly indicates whether the
1169          * object even exists.
1170          */
1171         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1172                 const char *path;
1173                 struct stat st;
1174                 if (stat_sha1_file(sha1, &st, &path) < 0)
1175                         return -1;
1176                 if (oi->disk_sizep)
1177                         *oi->disk_sizep = st.st_size;
1178                 return 0;
1179         }
1180
1181         map = map_sha1_file(sha1, &mapsize);
1182         if (!map)
1183                 return -1;
1184
1185         if (!oi->sizep)
1186                 oi->sizep = &size_scratch;
1187
1188         if (oi->disk_sizep)
1189                 *oi->disk_sizep = mapsize;
1190         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1191                 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1192                         status = error("unable to unpack %s header with --allow-unknown-type",
1193                                        sha1_to_hex(sha1));
1194         } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1195                 status = error("unable to unpack %s header",
1196                                sha1_to_hex(sha1));
1197         if (status < 0)
1198                 ; /* Do nothing */
1199         else if (hdrbuf.len) {
1200                 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1201                         status = error("unable to parse %s header with --allow-unknown-type",
1202                                        sha1_to_hex(sha1));
1203         } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1204                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1205
1206         if (status >= 0 && oi->contentp) {
1207                 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1208                                                  *oi->sizep, sha1);
1209                 if (!*oi->contentp) {
1210                         git_inflate_end(&stream);
1211                         status = -1;
1212                 }
1213         } else
1214                 git_inflate_end(&stream);
1215
1216         munmap(map, mapsize);
1217         if (status && oi->typep)
1218                 *oi->typep = status;
1219         if (oi->sizep == &size_scratch)
1220                 oi->sizep = NULL;
1221         strbuf_release(&hdrbuf);
1222         oi->whence = OI_LOOSE;
1223         return (status < 0) ? status : 0;
1224 }
1225
1226 int fetch_if_missing = 1;
1227
1228 int oid_object_info_extended(const struct object_id *oid, struct object_info *oi, unsigned flags)
1229 {
1230         static struct object_info blank_oi = OBJECT_INFO_INIT;
1231         struct pack_entry e;
1232         int rtype;
1233         const struct object_id *real = oid;
1234         int already_retried = 0;
1235
1236         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1237                 real = lookup_replace_object(oid);
1238
1239         if (is_null_oid(real))
1240                 return -1;
1241
1242         if (!oi)
1243                 oi = &blank_oi;
1244
1245         if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1246                 struct cached_object *co = find_cached_object(real->hash);
1247                 if (co) {
1248                         if (oi->typep)
1249                                 *(oi->typep) = co->type;
1250                         if (oi->sizep)
1251                                 *(oi->sizep) = co->size;
1252                         if (oi->disk_sizep)
1253                                 *(oi->disk_sizep) = 0;
1254                         if (oi->delta_base_sha1)
1255                                 hashclr(oi->delta_base_sha1);
1256                         if (oi->type_name)
1257                                 strbuf_addstr(oi->type_name, type_name(co->type));
1258                         if (oi->contentp)
1259                                 *oi->contentp = xmemdupz(co->buf, co->size);
1260                         oi->whence = OI_CACHED;
1261                         return 0;
1262                 }
1263         }
1264
1265         while (1) {
1266                 if (find_pack_entry(real->hash, &e))
1267                         break;
1268
1269                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1270                         return -1;
1271
1272                 /* Most likely it's a loose object. */
1273                 if (!sha1_loose_object_info(real->hash, oi, flags))
1274                         return 0;
1275
1276                 /* Not a loose object; someone else may have just packed it. */
1277                 if (!(flags & OBJECT_INFO_QUICK)) {
1278                         reprepare_packed_git();
1279                         if (find_pack_entry(real->hash, &e))
1280                                 break;
1281                 }
1282
1283                 /* Check if it is a missing object */
1284                 if (fetch_if_missing && repository_format_partial_clone &&
1285                     !already_retried) {
1286                         /*
1287                          * TODO Investigate haveing fetch_object() return
1288                          * TODO error/success and stopping the music here.
1289                          */
1290                         fetch_object(repository_format_partial_clone, real->hash);
1291                         already_retried = 1;
1292                         continue;
1293                 }
1294
1295                 return -1;
1296         }
1297
1298         if (oi == &blank_oi)
1299                 /*
1300                  * We know that the caller doesn't actually need the
1301                  * information below, so return early.
1302                  */
1303                 return 0;
1304         rtype = packed_object_info(e.p, e.offset, oi);
1305         if (rtype < 0) {
1306                 mark_bad_packed_object(e.p, real->hash);
1307                 return oid_object_info_extended(real, oi, 0);
1308         } else if (oi->whence == OI_PACKED) {
1309                 oi->u.packed.offset = e.offset;
1310                 oi->u.packed.pack = e.p;
1311                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1312                                          rtype == OBJ_OFS_DELTA);
1313         }
1314
1315         return 0;
1316 }
1317
1318 /* returns enum object_type or negative */
1319 int oid_object_info(const struct object_id *oid, unsigned long *sizep)
1320 {
1321         enum object_type type;
1322         struct object_info oi = OBJECT_INFO_INIT;
1323
1324         oi.typep = &type;
1325         oi.sizep = sizep;
1326         if (oid_object_info_extended(oid, &oi,
1327                                      OBJECT_INFO_LOOKUP_REPLACE) < 0)
1328                 return -1;
1329         return type;
1330 }
1331
1332 static void *read_object(const unsigned char *sha1, enum object_type *type,
1333                          unsigned long *size)
1334 {
1335         struct object_id oid;
1336         struct object_info oi = OBJECT_INFO_INIT;
1337         void *content;
1338         oi.typep = type;
1339         oi.sizep = size;
1340         oi.contentp = &content;
1341
1342         hashcpy(oid.hash, sha1);
1343
1344         if (oid_object_info_extended(&oid, &oi, 0) < 0)
1345                 return NULL;
1346         return content;
1347 }
1348
1349 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1350                         struct object_id *oid)
1351 {
1352         struct cached_object *co;
1353
1354         hash_object_file(buf, len, type_name(type), oid);
1355         if (has_sha1_file(oid->hash) || find_cached_object(oid->hash))
1356                 return 0;
1357         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1358         co = &cached_objects[cached_object_nr++];
1359         co->size = len;
1360         co->type = type;
1361         co->buf = xmalloc(len);
1362         memcpy(co->buf, buf, len);
1363         hashcpy(co->sha1, oid->hash);
1364         return 0;
1365 }
1366
1367 /*
1368  * This function dies on corrupt objects; the callers who want to
1369  * deal with them should arrange to call read_object() and give error
1370  * messages themselves.
1371  */
1372 void *read_object_file_extended(const struct object_id *oid,
1373                                 enum object_type *type,
1374                                 unsigned long *size,
1375                                 int lookup_replace)
1376 {
1377         void *data;
1378         const struct packed_git *p;
1379         const char *path;
1380         struct stat st;
1381         const struct object_id *repl = lookup_replace ? lookup_replace_object(oid)
1382                                                       : oid;
1383
1384         errno = 0;
1385         data = read_object(repl->hash, type, size);
1386         if (data)
1387                 return data;
1388
1389         if (errno && errno != ENOENT)
1390                 die_errno("failed to read object %s", oid_to_hex(oid));
1391
1392         /* die if we replaced an object with one that does not exist */
1393         if (repl != oid)
1394                 die("replacement %s not found for %s",
1395                     oid_to_hex(repl), oid_to_hex(oid));
1396
1397         if (!stat_sha1_file(repl->hash, &st, &path))
1398                 die("loose object %s (stored in %s) is corrupt",
1399                     oid_to_hex(repl), path);
1400
1401         if ((p = has_packed_and_bad(repl->hash)) != NULL)
1402                 die("packed object %s (stored in %s) is corrupt",
1403                     oid_to_hex(repl), p->pack_name);
1404
1405         return NULL;
1406 }
1407
1408 void *read_object_with_reference(const struct object_id *oid,
1409                                  const char *required_type_name,
1410                                  unsigned long *size,
1411                                  struct object_id *actual_oid_return)
1412 {
1413         enum object_type type, required_type;
1414         void *buffer;
1415         unsigned long isize;
1416         struct object_id actual_oid;
1417
1418         required_type = type_from_string(required_type_name);
1419         oidcpy(&actual_oid, oid);
1420         while (1) {
1421                 int ref_length = -1;
1422                 const char *ref_type = NULL;
1423
1424                 buffer = read_object_file(&actual_oid, &type, &isize);
1425                 if (!buffer)
1426                         return NULL;
1427                 if (type == required_type) {
1428                         *size = isize;
1429                         if (actual_oid_return)
1430                                 oidcpy(actual_oid_return, &actual_oid);
1431                         return buffer;
1432                 }
1433                 /* Handle references */
1434                 else if (type == OBJ_COMMIT)
1435                         ref_type = "tree ";
1436                 else if (type == OBJ_TAG)
1437                         ref_type = "object ";
1438                 else {
1439                         free(buffer);
1440                         return NULL;
1441                 }
1442                 ref_length = strlen(ref_type);
1443
1444                 if (ref_length + GIT_SHA1_HEXSZ > isize ||
1445                     memcmp(buffer, ref_type, ref_length) ||
1446                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1447                         free(buffer);
1448                         return NULL;
1449                 }
1450                 free(buffer);
1451                 /* Now we have the ID of the referred-to object in
1452                  * actual_oid.  Check again. */
1453         }
1454 }
1455
1456 static void write_object_file_prepare(const void *buf, unsigned long len,
1457                                       const char *type, struct object_id *oid,
1458                                       char *hdr, int *hdrlen)
1459 {
1460         git_hash_ctx c;
1461
1462         /* Generate the header */
1463         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1464
1465         /* Sha1.. */
1466         the_hash_algo->init_fn(&c);
1467         the_hash_algo->update_fn(&c, hdr, *hdrlen);
1468         the_hash_algo->update_fn(&c, buf, len);
1469         the_hash_algo->final_fn(oid->hash, &c);
1470 }
1471
1472 /*
1473  * Move the just written object into its final resting place.
1474  */
1475 int finalize_object_file(const char *tmpfile, const char *filename)
1476 {
1477         int ret = 0;
1478
1479         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1480                 goto try_rename;
1481         else if (link(tmpfile, filename))
1482                 ret = errno;
1483
1484         /*
1485          * Coda hack - coda doesn't like cross-directory links,
1486          * so we fall back to a rename, which will mean that it
1487          * won't be able to check collisions, but that's not a
1488          * big deal.
1489          *
1490          * The same holds for FAT formatted media.
1491          *
1492          * When this succeeds, we just return.  We have nothing
1493          * left to unlink.
1494          */
1495         if (ret && ret != EEXIST) {
1496         try_rename:
1497                 if (!rename(tmpfile, filename))
1498                         goto out;
1499                 ret = errno;
1500         }
1501         unlink_or_warn(tmpfile);
1502         if (ret) {
1503                 if (ret != EEXIST) {
1504                         return error_errno("unable to write sha1 filename %s", filename);
1505                 }
1506                 /* FIXME!!! Collision check here ? */
1507         }
1508
1509 out:
1510         if (adjust_shared_perm(filename))
1511                 return error("unable to set permission to '%s'", filename);
1512         return 0;
1513 }
1514
1515 static int write_buffer(int fd, const void *buf, size_t len)
1516 {
1517         if (write_in_full(fd, buf, len) < 0)
1518                 return error_errno("file write error");
1519         return 0;
1520 }
1521
1522 int hash_object_file(const void *buf, unsigned long len, const char *type,
1523                      struct object_id *oid)
1524 {
1525         char hdr[MAX_HEADER_LEN];
1526         int hdrlen = sizeof(hdr);
1527         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1528         return 0;
1529 }
1530
1531 /* Finalize a file on disk, and close it. */
1532 static void close_sha1_file(int fd)
1533 {
1534         if (fsync_object_files)
1535                 fsync_or_die(fd, "sha1 file");
1536         if (close(fd) != 0)
1537                 die_errno("error when closing sha1 file");
1538 }
1539
1540 /* Size of directory component, including the ending '/' */
1541 static inline int directory_size(const char *filename)
1542 {
1543         const char *s = strrchr(filename, '/');
1544         if (!s)
1545                 return 0;
1546         return s - filename + 1;
1547 }
1548
1549 /*
1550  * This creates a temporary file in the same directory as the final
1551  * 'filename'
1552  *
1553  * We want to avoid cross-directory filename renames, because those
1554  * can have problems on various filesystems (FAT, NFS, Coda).
1555  */
1556 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1557 {
1558         int fd, dirlen = directory_size(filename);
1559
1560         strbuf_reset(tmp);
1561         strbuf_add(tmp, filename, dirlen);
1562         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1563         fd = git_mkstemp_mode(tmp->buf, 0444);
1564         if (fd < 0 && dirlen && errno == ENOENT) {
1565                 /*
1566                  * Make sure the directory exists; note that the contents
1567                  * of the buffer are undefined after mkstemp returns an
1568                  * error, so we have to rewrite the whole buffer from
1569                  * scratch.
1570                  */
1571                 strbuf_reset(tmp);
1572                 strbuf_add(tmp, filename, dirlen - 1);
1573                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1574                         return -1;
1575                 if (adjust_shared_perm(tmp->buf))
1576                         return -1;
1577
1578                 /* Try again */
1579                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1580                 fd = git_mkstemp_mode(tmp->buf, 0444);
1581         }
1582         return fd;
1583 }
1584
1585 static int write_loose_object(const struct object_id *oid, char *hdr,
1586                               int hdrlen, const void *buf, unsigned long len,
1587                               time_t mtime)
1588 {
1589         int fd, ret;
1590         unsigned char compressed[4096];
1591         git_zstream stream;
1592         git_hash_ctx c;
1593         struct object_id parano_oid;
1594         static struct strbuf tmp_file = STRBUF_INIT;
1595         static struct strbuf filename = STRBUF_INIT;
1596
1597         strbuf_reset(&filename);
1598         sha1_file_name(&filename, oid->hash);
1599
1600         fd = create_tmpfile(&tmp_file, filename.buf);
1601         if (fd < 0) {
1602                 if (errno == EACCES)
1603                         return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1604                 else
1605                         return error_errno("unable to create temporary file");
1606         }
1607
1608         /* Set it up */
1609         git_deflate_init(&stream, zlib_compression_level);
1610         stream.next_out = compressed;
1611         stream.avail_out = sizeof(compressed);
1612         the_hash_algo->init_fn(&c);
1613
1614         /* First header.. */
1615         stream.next_in = (unsigned char *)hdr;
1616         stream.avail_in = hdrlen;
1617         while (git_deflate(&stream, 0) == Z_OK)
1618                 ; /* nothing */
1619         the_hash_algo->update_fn(&c, hdr, hdrlen);
1620
1621         /* Then the data itself.. */
1622         stream.next_in = (void *)buf;
1623         stream.avail_in = len;
1624         do {
1625                 unsigned char *in0 = stream.next_in;
1626                 ret = git_deflate(&stream, Z_FINISH);
1627                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1628                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1629                         die("unable to write sha1 file");
1630                 stream.next_out = compressed;
1631                 stream.avail_out = sizeof(compressed);
1632         } while (ret == Z_OK);
1633
1634         if (ret != Z_STREAM_END)
1635                 die("unable to deflate new object %s (%d)", oid_to_hex(oid),
1636                     ret);
1637         ret = git_deflate_end_gently(&stream);
1638         if (ret != Z_OK)
1639                 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid),
1640                     ret);
1641         the_hash_algo->final_fn(parano_oid.hash, &c);
1642         if (oidcmp(oid, &parano_oid) != 0)
1643                 die("confused by unstable object source data for %s",
1644                     oid_to_hex(oid));
1645
1646         close_sha1_file(fd);
1647
1648         if (mtime) {
1649                 struct utimbuf utb;
1650                 utb.actime = mtime;
1651                 utb.modtime = mtime;
1652                 if (utime(tmp_file.buf, &utb) < 0)
1653                         warning_errno("failed utime() on %s", tmp_file.buf);
1654         }
1655
1656         return finalize_object_file(tmp_file.buf, filename.buf);
1657 }
1658
1659 static int freshen_loose_object(const unsigned char *sha1)
1660 {
1661         return check_and_freshen(sha1, 1);
1662 }
1663
1664 static int freshen_packed_object(const unsigned char *sha1)
1665 {
1666         struct pack_entry e;
1667         if (!find_pack_entry(sha1, &e))
1668                 return 0;
1669         if (e.p->freshened)
1670                 return 1;
1671         if (!freshen_file(e.p->pack_name))
1672                 return 0;
1673         e.p->freshened = 1;
1674         return 1;
1675 }
1676
1677 int write_object_file(const void *buf, unsigned long len, const char *type,
1678                       struct object_id *oid)
1679 {
1680         char hdr[MAX_HEADER_LEN];
1681         int hdrlen = sizeof(hdr);
1682
1683         /* Normally if we have it in the pack then we do not bother writing
1684          * it out into .git/objects/??/?{38} file.
1685          */
1686         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1687         if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1688                 return 0;
1689         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1690 }
1691
1692 int hash_object_file_literally(const void *buf, unsigned long len,
1693                                const char *type, struct object_id *oid,
1694                                unsigned flags)
1695 {
1696         char *header;
1697         int hdrlen, status = 0;
1698
1699         /* type string, SP, %lu of the length plus NUL must fit this */
1700         hdrlen = strlen(type) + MAX_HEADER_LEN;
1701         header = xmalloc(hdrlen);
1702         write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1703
1704         if (!(flags & HASH_WRITE_OBJECT))
1705                 goto cleanup;
1706         if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1707                 goto cleanup;
1708         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1709
1710 cleanup:
1711         free(header);
1712         return status;
1713 }
1714
1715 int force_object_loose(const struct object_id *oid, time_t mtime)
1716 {
1717         void *buf;
1718         unsigned long len;
1719         enum object_type type;
1720         char hdr[MAX_HEADER_LEN];
1721         int hdrlen;
1722         int ret;
1723
1724         if (has_loose_object(oid->hash))
1725                 return 0;
1726         buf = read_object(oid->hash, &type, &len);
1727         if (!buf)
1728                 return error("cannot read sha1_file for %s", oid_to_hex(oid));
1729         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1730         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1731         free(buf);
1732
1733         return ret;
1734 }
1735
1736 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1737 {
1738         struct object_id oid;
1739         if (!startup_info->have_repository)
1740                 return 0;
1741         hashcpy(oid.hash, sha1);
1742         return oid_object_info_extended(&oid, NULL,
1743                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1744 }
1745
1746 int has_object_file(const struct object_id *oid)
1747 {
1748         return has_sha1_file(oid->hash);
1749 }
1750
1751 int has_object_file_with_flags(const struct object_id *oid, int flags)
1752 {
1753         return has_sha1_file_with_flags(oid->hash, flags);
1754 }
1755
1756 static void check_tree(const void *buf, size_t size)
1757 {
1758         struct tree_desc desc;
1759         struct name_entry entry;
1760
1761         init_tree_desc(&desc, buf, size);
1762         while (tree_entry(&desc, &entry))
1763                 /* do nothing
1764                  * tree_entry() will die() on malformed entries */
1765                 ;
1766 }
1767
1768 static void check_commit(const void *buf, size_t size)
1769 {
1770         struct commit c;
1771         memset(&c, 0, sizeof(c));
1772         if (parse_commit_buffer(&c, buf, size))
1773                 die("corrupt commit");
1774 }
1775
1776 static void check_tag(const void *buf, size_t size)
1777 {
1778         struct tag t;
1779         memset(&t, 0, sizeof(t));
1780         if (parse_tag_buffer(&t, buf, size))
1781                 die("corrupt tag");
1782 }
1783
1784 static int index_mem(struct object_id *oid, void *buf, size_t size,
1785                      enum object_type type,
1786                      const char *path, unsigned flags)
1787 {
1788         int ret, re_allocated = 0;
1789         int write_object = flags & HASH_WRITE_OBJECT;
1790
1791         if (!type)
1792                 type = OBJ_BLOB;
1793
1794         /*
1795          * Convert blobs to git internal format
1796          */
1797         if ((type == OBJ_BLOB) && path) {
1798                 struct strbuf nbuf = STRBUF_INIT;
1799                 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1800                                    get_conv_flags(flags))) {
1801                         buf = strbuf_detach(&nbuf, &size);
1802                         re_allocated = 1;
1803                 }
1804         }
1805         if (flags & HASH_FORMAT_CHECK) {
1806                 if (type == OBJ_TREE)
1807                         check_tree(buf, size);
1808                 if (type == OBJ_COMMIT)
1809                         check_commit(buf, size);
1810                 if (type == OBJ_TAG)
1811                         check_tag(buf, size);
1812         }
1813
1814         if (write_object)
1815                 ret = write_object_file(buf, size, type_name(type), oid);
1816         else
1817                 ret = hash_object_file(buf, size, type_name(type), oid);
1818         if (re_allocated)
1819                 free(buf);
1820         return ret;
1821 }
1822
1823 static int index_stream_convert_blob(struct object_id *oid, int fd,
1824                                      const char *path, unsigned flags)
1825 {
1826         int ret;
1827         const int write_object = flags & HASH_WRITE_OBJECT;
1828         struct strbuf sbuf = STRBUF_INIT;
1829
1830         assert(path);
1831         assert(would_convert_to_git_filter_fd(path));
1832
1833         convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1834                                  get_conv_flags(flags));
1835
1836         if (write_object)
1837                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1838                                         oid);
1839         else
1840                 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1841                                        oid);
1842         strbuf_release(&sbuf);
1843         return ret;
1844 }
1845
1846 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1847                       const char *path, unsigned flags)
1848 {
1849         struct strbuf sbuf = STRBUF_INIT;
1850         int ret;
1851
1852         if (strbuf_read(&sbuf, fd, 4096) >= 0)
1853                 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1854         else
1855                 ret = -1;
1856         strbuf_release(&sbuf);
1857         return ret;
1858 }
1859
1860 #define SMALL_FILE_SIZE (32*1024)
1861
1862 static int index_core(struct object_id *oid, int fd, size_t size,
1863                       enum object_type type, const char *path,
1864                       unsigned flags)
1865 {
1866         int ret;
1867
1868         if (!size) {
1869                 ret = index_mem(oid, "", size, type, path, flags);
1870         } else if (size <= SMALL_FILE_SIZE) {
1871                 char *buf = xmalloc(size);
1872                 ssize_t read_result = read_in_full(fd, buf, size);
1873                 if (read_result < 0)
1874                         ret = error_errno("read error while indexing %s",
1875                                           path ? path : "<unknown>");
1876                 else if (read_result != size)
1877                         ret = error("short read while indexing %s",
1878                                     path ? path : "<unknown>");
1879                 else
1880                         ret = index_mem(oid, buf, size, type, path, flags);
1881                 free(buf);
1882         } else {
1883                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1884                 ret = index_mem(oid, buf, size, type, path, flags);
1885                 munmap(buf, size);
1886         }
1887         return ret;
1888 }
1889
1890 /*
1891  * This creates one packfile per large blob unless bulk-checkin
1892  * machinery is "plugged".
1893  *
1894  * This also bypasses the usual "convert-to-git" dance, and that is on
1895  * purpose. We could write a streaming version of the converting
1896  * functions and insert that before feeding the data to fast-import
1897  * (or equivalent in-core API described above). However, that is
1898  * somewhat complicated, as we do not know the size of the filter
1899  * result, which we need to know beforehand when writing a git object.
1900  * Since the primary motivation for trying to stream from the working
1901  * tree file and to avoid mmaping it in core is to deal with large
1902  * binary blobs, they generally do not want to get any conversion, and
1903  * callers should avoid this code path when filters are requested.
1904  */
1905 static int index_stream(struct object_id *oid, int fd, size_t size,
1906                         enum object_type type, const char *path,
1907                         unsigned flags)
1908 {
1909         return index_bulk_checkin(oid, fd, size, type, path, flags);
1910 }
1911
1912 int index_fd(struct object_id *oid, int fd, struct stat *st,
1913              enum object_type type, const char *path, unsigned flags)
1914 {
1915         int ret;
1916
1917         /*
1918          * Call xsize_t() only when needed to avoid potentially unnecessary
1919          * die() for large files.
1920          */
1921         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1922                 ret = index_stream_convert_blob(oid, fd, path, flags);
1923         else if (!S_ISREG(st->st_mode))
1924                 ret = index_pipe(oid, fd, type, path, flags);
1925         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1926                  (path && would_convert_to_git(&the_index, path)))
1927                 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1928                                  flags);
1929         else
1930                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1931                                    flags);
1932         close(fd);
1933         return ret;
1934 }
1935
1936 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1937 {
1938         int fd;
1939         struct strbuf sb = STRBUF_INIT;
1940         int rc = 0;
1941
1942         switch (st->st_mode & S_IFMT) {
1943         case S_IFREG:
1944                 fd = open(path, O_RDONLY);
1945                 if (fd < 0)
1946                         return error_errno("open(\"%s\")", path);
1947                 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1948                         return error("%s: failed to insert into database",
1949                                      path);
1950                 break;
1951         case S_IFLNK:
1952                 if (strbuf_readlink(&sb, path, st->st_size))
1953                         return error_errno("readlink(\"%s\")", path);
1954                 if (!(flags & HASH_WRITE_OBJECT))
1955                         hash_object_file(sb.buf, sb.len, blob_type, oid);
1956                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1957                         rc = error("%s: failed to insert into database", path);
1958                 strbuf_release(&sb);
1959                 break;
1960         case S_IFDIR:
1961                 return resolve_gitlink_ref(path, "HEAD", oid);
1962         default:
1963                 return error("%s: unsupported file type", path);
1964         }
1965         return rc;
1966 }
1967
1968 int read_pack_header(int fd, struct pack_header *header)
1969 {
1970         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1971                 /* "eof before pack header was fully read" */
1972                 return PH_ERROR_EOF;
1973
1974         if (header->hdr_signature != htonl(PACK_SIGNATURE))
1975                 /* "protocol error (pack signature mismatch detected)" */
1976                 return PH_ERROR_PACK_SIGNATURE;
1977         if (!pack_version_ok(header->hdr_version))
1978                 /* "protocol error (pack version unsupported)" */
1979                 return PH_ERROR_PROTOCOL;
1980         return 0;
1981 }
1982
1983 void assert_oid_type(const struct object_id *oid, enum object_type expect)
1984 {
1985         enum object_type type = oid_object_info(oid, NULL);
1986         if (type < 0)
1987                 die("%s is not a valid object", oid_to_hex(oid));
1988         if (type != expect)
1989                 die("%s is not a valid '%s' object", oid_to_hex(oid),
1990                     type_name(expect));
1991 }
1992
1993 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1994                                 struct strbuf *path,
1995                                 each_loose_object_fn obj_cb,
1996                                 each_loose_cruft_fn cruft_cb,
1997                                 each_loose_subdir_fn subdir_cb,
1998                                 void *data)
1999 {
2000         size_t origlen, baselen;
2001         DIR *dir;
2002         struct dirent *de;
2003         int r = 0;
2004         struct object_id oid;
2005
2006         if (subdir_nr > 0xff)
2007                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2008
2009         origlen = path->len;
2010         strbuf_complete(path, '/');
2011         strbuf_addf(path, "%02x", subdir_nr);
2012
2013         dir = opendir(path->buf);
2014         if (!dir) {
2015                 if (errno != ENOENT)
2016                         r = error_errno("unable to open %s", path->buf);
2017                 strbuf_setlen(path, origlen);
2018                 return r;
2019         }
2020
2021         oid.hash[0] = subdir_nr;
2022         strbuf_addch(path, '/');
2023         baselen = path->len;
2024
2025         while ((de = readdir(dir))) {
2026                 size_t namelen;
2027                 if (is_dot_or_dotdot(de->d_name))
2028                         continue;
2029
2030                 namelen = strlen(de->d_name);
2031                 strbuf_setlen(path, baselen);
2032                 strbuf_add(path, de->d_name, namelen);
2033                 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2034                     !hex_to_bytes(oid.hash + 1, de->d_name,
2035                                   GIT_SHA1_RAWSZ - 1)) {
2036                         if (obj_cb) {
2037                                 r = obj_cb(&oid, path->buf, data);
2038                                 if (r)
2039                                         break;
2040                         }
2041                         continue;
2042                 }
2043
2044                 if (cruft_cb) {
2045                         r = cruft_cb(de->d_name, path->buf, data);
2046                         if (r)
2047                                 break;
2048                 }
2049         }
2050         closedir(dir);
2051
2052         strbuf_setlen(path, baselen - 1);
2053         if (!r && subdir_cb)
2054                 r = subdir_cb(subdir_nr, path->buf, data);
2055
2056         strbuf_setlen(path, origlen);
2057
2058         return r;
2059 }
2060
2061 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2062                             each_loose_object_fn obj_cb,
2063                             each_loose_cruft_fn cruft_cb,
2064                             each_loose_subdir_fn subdir_cb,
2065                             void *data)
2066 {
2067         int r = 0;
2068         int i;
2069
2070         for (i = 0; i < 256; i++) {
2071                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2072                                                 subdir_cb, data);
2073                 if (r)
2074                         break;
2075         }
2076
2077         return r;
2078 }
2079
2080 int for_each_loose_file_in_objdir(const char *path,
2081                                   each_loose_object_fn obj_cb,
2082                                   each_loose_cruft_fn cruft_cb,
2083                                   each_loose_subdir_fn subdir_cb,
2084                                   void *data)
2085 {
2086         struct strbuf buf = STRBUF_INIT;
2087         int r;
2088
2089         strbuf_addstr(&buf, path);
2090         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2091                                               subdir_cb, data);
2092         strbuf_release(&buf);
2093
2094         return r;
2095 }
2096
2097 struct loose_alt_odb_data {
2098         each_loose_object_fn *cb;
2099         void *data;
2100 };
2101
2102 static int loose_from_alt_odb(struct alternate_object_database *alt,
2103                               void *vdata)
2104 {
2105         struct loose_alt_odb_data *data = vdata;
2106         struct strbuf buf = STRBUF_INIT;
2107         int r;
2108
2109         strbuf_addstr(&buf, alt->path);
2110         r = for_each_loose_file_in_objdir_buf(&buf,
2111                                               data->cb, NULL, NULL,
2112                                               data->data);
2113         strbuf_release(&buf);
2114         return r;
2115 }
2116
2117 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2118 {
2119         struct loose_alt_odb_data alt;
2120         int r;
2121
2122         r = for_each_loose_file_in_objdir(get_object_directory(),
2123                                           cb, NULL, NULL, data);
2124         if (r)
2125                 return r;
2126
2127         if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2128                 return 0;
2129
2130         alt.cb = cb;
2131         alt.data = data;
2132         return foreach_alt_odb(loose_from_alt_odb, &alt);
2133 }
2134
2135 static int check_stream_sha1(git_zstream *stream,
2136                              const char *hdr,
2137                              unsigned long size,
2138                              const char *path,
2139                              const unsigned char *expected_sha1)
2140 {
2141         git_hash_ctx c;
2142         unsigned char real_sha1[GIT_MAX_RAWSZ];
2143         unsigned char buf[4096];
2144         unsigned long total_read;
2145         int status = Z_OK;
2146
2147         the_hash_algo->init_fn(&c);
2148         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2149
2150         /*
2151          * We already read some bytes into hdr, but the ones up to the NUL
2152          * do not count against the object's content size.
2153          */
2154         total_read = stream->total_out - strlen(hdr) - 1;
2155
2156         /*
2157          * This size comparison must be "<=" to read the final zlib packets;
2158          * see the comment in unpack_sha1_rest for details.
2159          */
2160         while (total_read <= size &&
2161                (status == Z_OK || status == Z_BUF_ERROR)) {
2162                 stream->next_out = buf;
2163                 stream->avail_out = sizeof(buf);
2164                 if (size - total_read < stream->avail_out)
2165                         stream->avail_out = size - total_read;
2166                 status = git_inflate(stream, Z_FINISH);
2167                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2168                 total_read += stream->next_out - buf;
2169         }
2170         git_inflate_end(stream);
2171
2172         if (status != Z_STREAM_END) {
2173                 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2174                 return -1;
2175         }
2176         if (stream->avail_in) {
2177                 error("garbage at end of loose object '%s'",
2178                       sha1_to_hex(expected_sha1));
2179                 return -1;
2180         }
2181
2182         the_hash_algo->final_fn(real_sha1, &c);
2183         if (hashcmp(expected_sha1, real_sha1)) {
2184                 error("sha1 mismatch for %s (expected %s)", path,
2185                       sha1_to_hex(expected_sha1));
2186                 return -1;
2187         }
2188
2189         return 0;
2190 }
2191
2192 int read_loose_object(const char *path,
2193                       const struct object_id *expected_oid,
2194                       enum object_type *type,
2195                       unsigned long *size,
2196                       void **contents)
2197 {
2198         int ret = -1;
2199         void *map = NULL;
2200         unsigned long mapsize;
2201         git_zstream stream;
2202         char hdr[MAX_HEADER_LEN];
2203
2204         *contents = NULL;
2205
2206         map = map_sha1_file_1(path, NULL, &mapsize);
2207         if (!map) {
2208                 error_errno("unable to mmap %s", path);
2209                 goto out;
2210         }
2211
2212         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2213                 error("unable to unpack header of %s", path);
2214                 goto out;
2215         }
2216
2217         *type = parse_sha1_header(hdr, size);
2218         if (*type < 0) {
2219                 error("unable to parse header of %s", path);
2220                 git_inflate_end(&stream);
2221                 goto out;
2222         }
2223
2224         if (*type == OBJ_BLOB) {
2225                 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2226                         goto out;
2227         } else {
2228                 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2229                 if (!*contents) {
2230                         error("unable to unpack contents of %s", path);
2231                         git_inflate_end(&stream);
2232                         goto out;
2233                 }
2234                 if (check_object_signature(expected_oid, *contents,
2235                                          *size, type_name(*type))) {
2236                         error("sha1 mismatch for %s (expected %s)", path,
2237                               oid_to_hex(expected_oid));
2238                         free(*contents);
2239                         goto out;
2240                 }
2241         }
2242
2243         ret = 0; /* everything checks out */
2244
2245 out:
2246         if (map)
2247                 munmap(map, mapsize);
2248         return ret;
2249 }