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         if (alt_odb_tail)
 672                 return;
 673
 674         alt_odb_tail = &alt_odb_list;
 675         link_alt_odb_entries(the_repository->alternate_db, PATH_SEP, NULL, 0);
 676
 677         read_info_alternates(get_object_directory(), 0);
 678 }
 679
 680 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 681 static int freshen_file(const char *fn)
 682 {
 683         struct utimbuf t;
 684         t.actime = t.modtime = time(NULL);
 685         return !utime(fn, &t);
 686 }
 687
 688 /*
 689  * All of the check_and_freshen functions return 1 if the file exists and was
 690  * freshened (if freshening was requested), 0 otherwise. If they return
 691  * 0, you should not assume that it is safe to skip a write of the object (it
 692  * either does not exist on disk, or has a stale mtime and may be subject to
 693  * pruning).
 694  */
 695 int check_and_freshen_file(const char *fn, int freshen)
 696 {
 697         if (access(fn, F_OK))
 698                 return 0;
 699         if (freshen && !freshen_file(fn))
 700                 return 0;
 701         return 1;
 702 }
 703
 704 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
 705 {
 706         static struct strbuf buf = STRBUF_INIT;
 707
 708         strbuf_reset(&buf);
 709         sha1_file_name(&buf, sha1);
 710
 711         return check_and_freshen_file(buf.buf, freshen);
 712 }
 713
 714 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
 715 {
 716         struct alternate_object_database *alt;
 717         prepare_alt_odb();
 718         for (alt = alt_odb_list; alt; alt = alt->next) {
 719                 const char *path = alt_sha1_path(alt, sha1);
 720                 if (check_and_freshen_file(path, freshen))
 721                         return 1;
 722         }
 723         return 0;
 724 }
 725
 726 static int check_and_freshen(const unsigned char *sha1, int freshen)
 727 {
 728         return check_and_freshen_local(sha1, freshen) ||
 729                check_and_freshen_nonlocal(sha1, freshen);
 730 }
 731
 732 int has_loose_object_nonlocal(const unsigned char *sha1)
 733 {
 734         return check_and_freshen_nonlocal(sha1, 0);
 735 }
 736
 737 static int has_loose_object(const unsigned char *sha1)
 738 {
 739         return check_and_freshen(sha1, 0);
 740 }
 741
 742 static void mmap_limit_check(size_t length)
 743 {
 744         static size_t limit = 0;
 745         if (!limit) {
 746                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 747                 if (!limit)
 748                         limit = SIZE_MAX;
 749         }
 750         if (length > limit)
 751                 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
 752                     (uintmax_t)length, (uintmax_t)limit);
 753 }
 754
 755 void *xmmap_gently(void *start, size_t length,
 756                   int prot, int flags, int fd, off_t offset)
 757 {
 758         void *ret;
 759
 760         mmap_limit_check(length);
 761         ret = mmap(start, length, prot, flags, fd, offset);
 762         if (ret == MAP_FAILED) {
 763                 if (!length)
 764                         return NULL;
 765                 release_pack_memory(length);
 766                 ret = mmap(start, length, prot, flags, fd, offset);
 767         }
 768         return ret;
 769 }
 770
 771 void *xmmap(void *start, size_t length,
 772         int prot, int flags, int fd, off_t offset)
 773 {
 774         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 775         if (ret == MAP_FAILED)
 776                 die_errno("mmap failed");
 777         return ret;
 778 }
 779
 780 /*
 781  * With an in-core object data in "map", rehash it to make sure the
 782  * object name actually matches "sha1" to detect object corruption.
 783  * With "map" == NULL, try reading the object named with "sha1" using
 784  * the streaming interface and rehash it to do the same.
 785  */
 786 int check_object_signature(const struct object_id *oid, void *map,
 787                            unsigned long size, const char *type)
 788 {
 789         struct object_id real_oid;
 790         enum object_type obj_type;
 791         struct git_istream *st;
 792         git_hash_ctx c;
 793         char hdr[MAX_HEADER_LEN];
 794         int hdrlen;
 795
 796         if (map) {
 797                 hash_object_file(map, size, type, &real_oid);
 798                 return oidcmp(oid, &real_oid) ? -1 : 0;
 799         }
 800
 801         st = open_istream(oid, &obj_type, &size, NULL);
 802         if (!st)
 803                 return -1;
 804
 805         /* Generate the header */
 806         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
 807
 808         /* Sha1.. */
 809         the_hash_algo->init_fn(&c);
 810         the_hash_algo->update_fn(&c, hdr, hdrlen);
 811         for (;;) {
 812                 char buf[1024 * 16];
 813                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
 814
 815                 if (readlen < 0) {
 816                         close_istream(st);
 817                         return -1;
 818                 }
 819                 if (!readlen)
 820                         break;
 821                 the_hash_algo->update_fn(&c, buf, readlen);
 822         }
 823         the_hash_algo->final_fn(real_oid.hash, &c);
 824         close_istream(st);
 825         return oidcmp(oid, &real_oid) ? -1 : 0;
 826 }
 827
 828 int git_open_cloexec(const char *name, int flags)
 829 {
 830         int fd;
 831         static int o_cloexec = O_CLOEXEC;
 832
 833         fd = open(name, flags | o_cloexec);
 834         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 835                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
 836                 o_cloexec &= ~O_CLOEXEC;
 837                 fd = open(name, flags | o_cloexec);
 838         }
 839
 840 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 841         {
 842                 static int fd_cloexec = FD_CLOEXEC;
 843
 844                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
 845                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 846                         int flags = fcntl(fd, F_GETFD);
 847                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 848                                 fd_cloexec = 0;
 849                 }
 850         }
 851 #endif
 852         return fd;
 853 }
 854
 855 /*
 856  * Find "sha1" as a loose object in the local repository or in an alternate.
 857  * Returns 0 on success, negative on failure.
 858  *
 859  * The "path" out-parameter will give the path of the object we found (if any).
 860  * Note that it may point to static storage and is only valid until another
 861  * call to sha1_file_name(), etc.
 862  */
 863 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
 864                           const char **path)
 865 {
 866         struct alternate_object_database *alt;
 867         static struct strbuf buf = STRBUF_INIT;
 868
 869         strbuf_reset(&buf);
 870         sha1_file_name(&buf, sha1);
 871         *path = buf.buf;
 872
 873         if (!lstat(*path, st))
 874                 return 0;
 875
 876         prepare_alt_odb();
 877         errno = ENOENT;
 878         for (alt = alt_odb_list; alt; alt = alt->next) {
 879                 *path = alt_sha1_path(alt, sha1);
 880                 if (!lstat(*path, st))
 881                         return 0;
 882         }
 883
 884         return -1;
 885 }
 886
 887 /*
 888  * Like stat_sha1_file(), but actually open the object and return the
 889  * descriptor. See the caveats on the "path" parameter above.
 890  */
 891 static int open_sha1_file(const unsigned char *sha1, const char **path)
 892 {
 893         int fd;
 894         struct alternate_object_database *alt;
 895         int most_interesting_errno;
 896         static struct strbuf buf = STRBUF_INIT;
 897
 898         strbuf_reset(&buf);
 899         sha1_file_name(&buf, sha1);
 900         *path = buf.buf;
 901
 902         fd = git_open(*path);
 903         if (fd >= 0)
 904                 return fd;
 905         most_interesting_errno = errno;
 906
 907         prepare_alt_odb();
 908         for (alt = alt_odb_list; alt; alt = alt->next) {
 909                 *path = alt_sha1_path(alt, sha1);
 910                 fd = git_open(*path);
 911                 if (fd >= 0)
 912                         return fd;
 913                 if (most_interesting_errno == ENOENT)
 914                         most_interesting_errno = errno;
 915         }
 916         errno = most_interesting_errno;
 917         return -1;
 918 }
 919
 920 /*
 921  * Map the loose object at "path" if it is not NULL, or the path found by
 922  * searching for a loose object named "sha1".
 923  */
 924 static void *map_sha1_file_1(const char *path,
 925                              const unsigned char *sha1,
 926                              unsigned long *size)
 927 {
 928         void *map;
 929         int fd;
 930
 931         if (path)
 932                 fd = git_open(path);
 933         else
 934                 fd = open_sha1_file(sha1, &path);
 935         map = NULL;
 936         if (fd >= 0) {
 937                 struct stat st;
 938
 939                 if (!fstat(fd, &st)) {
 940                         *size = xsize_t(st.st_size);
 941                         if (!*size) {
 942                                 /* mmap() is forbidden on empty files */
 943                                 error("object file %s is empty", path);
 944                                 return NULL;
 945                         }
 946                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 947                 }
 948                 close(fd);
 949         }
 950         return map;
 951 }
 952
 953 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 954 {
 955         return map_sha1_file_1(NULL, sha1, size);
 956 }
 957
 958 static int unpack_sha1_short_header(git_zstream *stream,
 959                                     unsigned char *map, unsigned long mapsize,
 960                                     void *buffer, unsigned long bufsiz)
 961 {
 962         /* Get the data stream */
 963         memset(stream, 0, sizeof(*stream));
 964         stream->next_in = map;
 965         stream->avail_in = mapsize;
 966         stream->next_out = buffer;
 967         stream->avail_out = bufsiz;
 968
 969         git_inflate_init(stream);
 970         return git_inflate(stream, 0);
 971 }
 972
 973 int unpack_sha1_header(git_zstream *stream,
 974                        unsigned char *map, unsigned long mapsize,
 975                        void *buffer, unsigned long bufsiz)
 976 {
 977         int status = unpack_sha1_short_header(stream, map, mapsize,
 978                                               buffer, bufsiz);
 979
 980         if (status < Z_OK)
 981                 return status;
 982
 983         /* Make sure we have the terminating NUL */
 984         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 985                 return -1;
 986         return 0;
 987 }
 988
 989 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
 990                                         unsigned long mapsize, void *buffer,
 991                                         unsigned long bufsiz, struct strbuf *header)
 992 {
 993         int status;
 994
 995         status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
 996         if (status < Z_OK)
 997                 return -1;
 998
 999         /*
1000          * Check if entire header is unpacked in the first iteration.
1001          */
1002         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1003                 return 0;
1004
1005         /*
1006          * buffer[0..bufsiz] was not large enough.  Copy the partial
1007          * result out to header, and then append the result of further
1008          * reading the stream.
1009          */
1010         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1011         stream->next_out = buffer;
1012         stream->avail_out = bufsiz;
1013
1014         do {
1015                 status = git_inflate(stream, 0);
1016                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1017                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1018                         return 0;
1019                 stream->next_out = buffer;
1020                 stream->avail_out = bufsiz;
1021         } while (status != Z_STREAM_END);
1022         return -1;
1023 }
1024
1025 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1026 {
1027         int bytes = strlen(buffer) + 1;
1028         unsigned char *buf = xmallocz(size);
1029         unsigned long n;
1030         int status = Z_OK;
1031
1032         n = stream->total_out - bytes;
1033         if (n > size)
1034                 n = size;
1035         memcpy(buf, (char *) buffer + bytes, n);
1036         bytes = n;
1037         if (bytes <= size) {
1038                 /*
1039                  * The above condition must be (bytes <= size), not
1040                  * (bytes < size).  In other words, even though we
1041                  * expect no more output and set avail_out to zero,
1042                  * the input zlib stream may have bytes that express
1043                  * "this concludes the stream", and we *do* want to
1044                  * eat that input.
1045                  *
1046                  * Otherwise we would not be able to test that we
1047                  * consumed all the input to reach the expected size;
1048                  * we also want to check that zlib tells us that all
1049                  * went well with status == Z_STREAM_END at the end.
1050                  */
1051                 stream->next_out = buf + bytes;
1052                 stream->avail_out = size - bytes;
1053                 while (status == Z_OK)
1054                         status = git_inflate(stream, Z_FINISH);
1055         }
1056         if (status == Z_STREAM_END && !stream->avail_in) {
1057                 git_inflate_end(stream);
1058                 return buf;
1059         }
1060
1061         if (status < 0)
1062                 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1063         else if (stream->avail_in)
1064                 error("garbage at end of loose object '%s'",
1065                       sha1_to_hex(sha1));
1066         free(buf);
1067         return NULL;
1068 }
1069
1070 /*
1071  * We used to just use "sscanf()", but that's actually way
1072  * too permissive for what we want to check. So do an anal
1073  * object header parse by hand.
1074  */
1075 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1076                                unsigned int flags)
1077 {
1078         const char *type_buf = hdr;
1079         unsigned long size;
1080         int type, type_len = 0;
1081
1082         /*
1083          * The type can be of any size but is followed by
1084          * a space.
1085          */
1086         for (;;) {
1087                 char c = *hdr++;
1088                 if (!c)
1089                         return -1;
1090                 if (c == ' ')
1091                         break;
1092                 type_len++;
1093         }
1094
1095         type = type_from_string_gently(type_buf, type_len, 1);
1096         if (oi->type_name)
1097                 strbuf_add(oi->type_name, type_buf, type_len);
1098         /*
1099          * Set type to 0 if its an unknown object and
1100          * we're obtaining the type using '--allow-unknown-type'
1101          * option.
1102          */
1103         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1104                 type = 0;
1105         else if (type < 0)
1106                 die("invalid object type");
1107         if (oi->typep)
1108                 *oi->typep = type;
1109
1110         /*
1111          * The length must follow immediately, and be in canonical
1112          * decimal format (ie "010" is not valid).
1113          */
1114         size = *hdr++ - '0';
1115         if (size > 9)
1116                 return -1;
1117         if (size) {
1118                 for (;;) {
1119                         unsigned long c = *hdr - '0';
1120                         if (c > 9)
1121                                 break;
1122                         hdr++;
1123                         size = size * 10 + c;
1124                 }
1125         }
1126
1127         if (oi->sizep)
1128                 *oi->sizep = size;
1129
1130         /*
1131          * The length must be followed by a zero byte
1132          */
1133         return *hdr ? -1 : type;
1134 }
1135
1136 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1137 {
1138         struct object_info oi = OBJECT_INFO_INIT;
1139
1140         oi.sizep = sizep;
1141         return parse_sha1_header_extended(hdr, &oi, 0);
1142 }
1143
1144 static int sha1_loose_object_info(const unsigned char *sha1,
1145                                   struct object_info *oi,
1146                                   int flags)
1147 {
1148         int status = 0;
1149         unsigned long mapsize;
1150         void *map;
1151         git_zstream stream;
1152         char hdr[MAX_HEADER_LEN];
1153         struct strbuf hdrbuf = STRBUF_INIT;
1154         unsigned long size_scratch;
1155
1156         if (oi->delta_base_sha1)
1157                 hashclr(oi->delta_base_sha1);
1158
1159         /*
1160          * If we don't care about type or size, then we don't
1161          * need to look inside the object at all. Note that we
1162          * do not optimize out the stat call, even if the
1163          * caller doesn't care about the disk-size, since our
1164          * return value implicitly indicates whether the
1165          * object even exists.
1166          */
1167         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1168                 const char *path;
1169                 struct stat st;
1170                 if (stat_sha1_file(sha1, &st, &path) < 0)
1171                         return -1;
1172                 if (oi->disk_sizep)
1173                         *oi->disk_sizep = st.st_size;
1174                 return 0;
1175         }
1176
1177         map = map_sha1_file(sha1, &mapsize);
1178         if (!map)
1179                 return -1;
1180
1181         if (!oi->sizep)
1182                 oi->sizep = &size_scratch;
1183
1184         if (oi->disk_sizep)
1185                 *oi->disk_sizep = mapsize;
1186         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1187                 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1188                         status = error("unable to unpack %s header with --allow-unknown-type",
1189                                        sha1_to_hex(sha1));
1190         } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1191                 status = error("unable to unpack %s header",
1192                                sha1_to_hex(sha1));
1193         if (status < 0)
1194                 ; /* Do nothing */
1195         else if (hdrbuf.len) {
1196                 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1197                         status = error("unable to parse %s header with --allow-unknown-type",
1198                                        sha1_to_hex(sha1));
1199         } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1200                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1201
1202         if (status >= 0 && oi->contentp) {
1203                 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1204                                                  *oi->sizep, sha1);
1205                 if (!*oi->contentp) {
1206                         git_inflate_end(&stream);
1207                         status = -1;
1208                 }
1209         } else
1210                 git_inflate_end(&stream);
1211
1212         munmap(map, mapsize);
1213         if (status && oi->typep)
1214                 *oi->typep = status;
1215         if (oi->sizep == &size_scratch)
1216                 oi->sizep = NULL;
1217         strbuf_release(&hdrbuf);
1218         oi->whence = OI_LOOSE;
1219         return (status < 0) ? status : 0;
1220 }
1221
1222 int fetch_if_missing = 1;
1223
1224 int oid_object_info_extended(const struct object_id *oid, struct object_info *oi, unsigned flags)
1225 {
1226         static struct object_info blank_oi = OBJECT_INFO_INIT;
1227         struct pack_entry e;
1228         int rtype;
1229         const struct object_id *real = oid;
1230         int already_retried = 0;
1231
1232         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1233                 real = lookup_replace_object(oid);
1234
1235         if (is_null_oid(real))
1236                 return -1;
1237
1238         if (!oi)
1239                 oi = &blank_oi;
1240
1241         if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1242                 struct cached_object *co = find_cached_object(real->hash);
1243                 if (co) {
1244                         if (oi->typep)
1245                                 *(oi->typep) = co->type;
1246                         if (oi->sizep)
1247                                 *(oi->sizep) = co->size;
1248                         if (oi->disk_sizep)
1249                                 *(oi->disk_sizep) = 0;
1250                         if (oi->delta_base_sha1)
1251                                 hashclr(oi->delta_base_sha1);
1252                         if (oi->type_name)
1253                                 strbuf_addstr(oi->type_name, type_name(co->type));
1254                         if (oi->contentp)
1255                                 *oi->contentp = xmemdupz(co->buf, co->size);
1256                         oi->whence = OI_CACHED;
1257                         return 0;
1258                 }
1259         }
1260
1261         while (1) {
1262                 if (find_pack_entry(real->hash, &e))
1263                         break;
1264
1265                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1266                         return -1;
1267
1268                 /* Most likely it's a loose object. */
1269                 if (!sha1_loose_object_info(real->hash, oi, flags))
1270                         return 0;
1271
1272                 /* Not a loose object; someone else may have just packed it. */
1273                 if (!(flags & OBJECT_INFO_QUICK)) {
1274                         reprepare_packed_git();
1275                         if (find_pack_entry(real->hash, &e))
1276                                 break;
1277                 }
1278
1279                 /* Check if it is a missing object */
1280                 if (fetch_if_missing && repository_format_partial_clone &&
1281                     !already_retried) {
1282                         /*
1283                          * TODO Investigate haveing fetch_object() return
1284                          * TODO error/success and stopping the music here.
1285                          */
1286                         fetch_object(repository_format_partial_clone, real->hash);
1287                         already_retried = 1;
1288                         continue;
1289                 }
1290
1291                 return -1;
1292         }
1293
1294         if (oi == &blank_oi)
1295                 /*
1296                  * We know that the caller doesn't actually need the
1297                  * information below, so return early.
1298                  */
1299                 return 0;
1300         rtype = packed_object_info(e.p, e.offset, oi);
1301         if (rtype < 0) {
1302                 mark_bad_packed_object(e.p, real->hash);
1303                 return oid_object_info_extended(real, oi, 0);
1304         } else if (oi->whence == OI_PACKED) {
1305                 oi->u.packed.offset = e.offset;
1306                 oi->u.packed.pack = e.p;
1307                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1308                                          rtype == OBJ_OFS_DELTA);
1309         }
1310
1311         return 0;
1312 }
1313
1314 /* returns enum object_type or negative */
1315 int oid_object_info(const struct object_id *oid, unsigned long *sizep)
1316 {
1317         enum object_type type;
1318         struct object_info oi = OBJECT_INFO_INIT;
1319
1320         oi.typep = &type;
1321         oi.sizep = sizep;
1322         if (oid_object_info_extended(oid, &oi,
1323                                      OBJECT_INFO_LOOKUP_REPLACE) < 0)
1324                 return -1;
1325         return type;
1326 }
1327
1328 static void *read_object(const unsigned char *sha1, enum object_type *type,
1329                          unsigned long *size)
1330 {
1331         struct object_id oid;
1332         struct object_info oi = OBJECT_INFO_INIT;
1333         void *content;
1334         oi.typep = type;
1335         oi.sizep = size;
1336         oi.contentp = &content;
1337
1338         hashcpy(oid.hash, sha1);
1339
1340         if (oid_object_info_extended(&oid, &oi, 0) < 0)
1341                 return NULL;
1342         return content;
1343 }
1344
1345 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1346                         struct object_id *oid)
1347 {
1348         struct cached_object *co;
1349
1350         hash_object_file(buf, len, type_name(type), oid);
1351         if (has_sha1_file(oid->hash) || find_cached_object(oid->hash))
1352                 return 0;
1353         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1354         co = &cached_objects[cached_object_nr++];
1355         co->size = len;
1356         co->type = type;
1357         co->buf = xmalloc(len);
1358         memcpy(co->buf, buf, len);
1359         hashcpy(co->sha1, oid->hash);
1360         return 0;
1361 }
1362
1363 /*
1364  * This function dies on corrupt objects; the callers who want to
1365  * deal with them should arrange to call read_object() and give error
1366  * messages themselves.
1367  */
1368 void *read_object_file_extended(const struct object_id *oid,
1369                                 enum object_type *type,
1370                                 unsigned long *size,
1371                                 int lookup_replace)
1372 {
1373         void *data;
1374         const struct packed_git *p;
1375         const char *path;
1376         struct stat st;
1377         const struct object_id *repl = lookup_replace ? lookup_replace_object(oid)
1378                                                       : oid;
1379
1380         errno = 0;
1381         data = read_object(repl->hash, type, size);
1382         if (data)
1383                 return data;
1384
1385         if (errno && errno != ENOENT)
1386                 die_errno("failed to read object %s", oid_to_hex(oid));
1387
1388         /* die if we replaced an object with one that does not exist */
1389         if (repl != oid)
1390                 die("replacement %s not found for %s",
1391                     oid_to_hex(repl), oid_to_hex(oid));
1392
1393         if (!stat_sha1_file(repl->hash, &st, &path))
1394                 die("loose object %s (stored in %s) is corrupt",
1395                     oid_to_hex(repl), path);
1396
1397         if ((p = has_packed_and_bad(repl->hash)) != NULL)
1398                 die("packed object %s (stored in %s) is corrupt",
1399                     oid_to_hex(repl), p->pack_name);
1400
1401         return NULL;
1402 }
1403
1404 void *read_object_with_reference(const struct object_id *oid,
1405                                  const char *required_type_name,
1406                                  unsigned long *size,
1407                                  struct object_id *actual_oid_return)
1408 {
1409         enum object_type type, required_type;
1410         void *buffer;
1411         unsigned long isize;
1412         struct object_id actual_oid;
1413
1414         required_type = type_from_string(required_type_name);
1415         oidcpy(&actual_oid, oid);
1416         while (1) {
1417                 int ref_length = -1;
1418                 const char *ref_type = NULL;
1419
1420                 buffer = read_object_file(&actual_oid, &type, &isize);
1421                 if (!buffer)
1422                         return NULL;
1423                 if (type == required_type) {
1424                         *size = isize;
1425                         if (actual_oid_return)
1426                                 oidcpy(actual_oid_return, &actual_oid);
1427                         return buffer;
1428                 }
1429                 /* Handle references */
1430                 else if (type == OBJ_COMMIT)
1431                         ref_type = "tree ";
1432                 else if (type == OBJ_TAG)
1433                         ref_type = "object ";
1434                 else {
1435                         free(buffer);
1436                         return NULL;
1437                 }
1438                 ref_length = strlen(ref_type);
1439
1440                 if (ref_length + GIT_SHA1_HEXSZ > isize ||
1441                     memcmp(buffer, ref_type, ref_length) ||
1442                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1443                         free(buffer);
1444                         return NULL;
1445                 }
1446                 free(buffer);
1447                 /* Now we have the ID of the referred-to object in
1448                  * actual_oid.  Check again. */
1449         }
1450 }
1451
1452 static void write_object_file_prepare(const void *buf, unsigned long len,
1453                                       const char *type, struct object_id *oid,
1454                                       char *hdr, int *hdrlen)
1455 {
1456         git_hash_ctx c;
1457
1458         /* Generate the header */
1459         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1460
1461         /* Sha1.. */
1462         the_hash_algo->init_fn(&c);
1463         the_hash_algo->update_fn(&c, hdr, *hdrlen);
1464         the_hash_algo->update_fn(&c, buf, len);
1465         the_hash_algo->final_fn(oid->hash, &c);
1466 }
1467
1468 /*
1469  * Move the just written object into its final resting place.
1470  */
1471 int finalize_object_file(const char *tmpfile, const char *filename)
1472 {
1473         int ret = 0;
1474
1475         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1476                 goto try_rename;
1477         else if (link(tmpfile, filename))
1478                 ret = errno;
1479
1480         /*
1481          * Coda hack - coda doesn't like cross-directory links,
1482          * so we fall back to a rename, which will mean that it
1483          * won't be able to check collisions, but that's not a
1484          * big deal.
1485          *
1486          * The same holds for FAT formatted media.
1487          *
1488          * When this succeeds, we just return.  We have nothing
1489          * left to unlink.
1490          */
1491         if (ret && ret != EEXIST) {
1492         try_rename:
1493                 if (!rename(tmpfile, filename))
1494                         goto out;
1495                 ret = errno;
1496         }
1497         unlink_or_warn(tmpfile);
1498         if (ret) {
1499                 if (ret != EEXIST) {
1500                         return error_errno("unable to write sha1 filename %s", filename);
1501                 }
1502                 /* FIXME!!! Collision check here ? */
1503         }
1504
1505 out:
1506         if (adjust_shared_perm(filename))
1507                 return error("unable to set permission to '%s'", filename);
1508         return 0;
1509 }
1510
1511 static int write_buffer(int fd, const void *buf, size_t len)
1512 {
1513         if (write_in_full(fd, buf, len) < 0)
1514                 return error_errno("file write error");
1515         return 0;
1516 }
1517
1518 int hash_object_file(const void *buf, unsigned long len, const char *type,
1519                      struct object_id *oid)
1520 {
1521         char hdr[MAX_HEADER_LEN];
1522         int hdrlen = sizeof(hdr);
1523         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1524         return 0;
1525 }
1526
1527 /* Finalize a file on disk, and close it. */
1528 static void close_sha1_file(int fd)
1529 {
1530         if (fsync_object_files)
1531                 fsync_or_die(fd, "sha1 file");
1532         if (close(fd) != 0)
1533                 die_errno("error when closing sha1 file");
1534 }
1535
1536 /* Size of directory component, including the ending '/' */
1537 static inline int directory_size(const char *filename)
1538 {
1539         const char *s = strrchr(filename, '/');
1540         if (!s)
1541                 return 0;
1542         return s - filename + 1;
1543 }
1544
1545 /*
1546  * This creates a temporary file in the same directory as the final
1547  * 'filename'
1548  *
1549  * We want to avoid cross-directory filename renames, because those
1550  * can have problems on various filesystems (FAT, NFS, Coda).
1551  */
1552 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1553 {
1554         int fd, dirlen = directory_size(filename);
1555
1556         strbuf_reset(tmp);
1557         strbuf_add(tmp, filename, dirlen);
1558         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1559         fd = git_mkstemp_mode(tmp->buf, 0444);
1560         if (fd < 0 && dirlen && errno == ENOENT) {
1561                 /*
1562                  * Make sure the directory exists; note that the contents
1563                  * of the buffer are undefined after mkstemp returns an
1564                  * error, so we have to rewrite the whole buffer from
1565                  * scratch.
1566                  */
1567                 strbuf_reset(tmp);
1568                 strbuf_add(tmp, filename, dirlen - 1);
1569                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1570                         return -1;
1571                 if (adjust_shared_perm(tmp->buf))
1572                         return -1;
1573
1574                 /* Try again */
1575                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1576                 fd = git_mkstemp_mode(tmp->buf, 0444);
1577         }
1578         return fd;
1579 }
1580
1581 static int write_loose_object(const struct object_id *oid, char *hdr,
1582                               int hdrlen, const void *buf, unsigned long len,
1583                               time_t mtime)
1584 {
1585         int fd, ret;
1586         unsigned char compressed[4096];
1587         git_zstream stream;
1588         git_hash_ctx c;
1589         struct object_id parano_oid;
1590         static struct strbuf tmp_file = STRBUF_INIT;
1591         static struct strbuf filename = STRBUF_INIT;
1592
1593         strbuf_reset(&filename);
1594         sha1_file_name(&filename, oid->hash);
1595
1596         fd = create_tmpfile(&tmp_file, filename.buf);
1597         if (fd < 0) {
1598                 if (errno == EACCES)
1599                         return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1600                 else
1601                         return error_errno("unable to create temporary file");
1602         }
1603
1604         /* Set it up */
1605         git_deflate_init(&stream, zlib_compression_level);
1606         stream.next_out = compressed;
1607         stream.avail_out = sizeof(compressed);
1608         the_hash_algo->init_fn(&c);
1609
1610         /* First header.. */
1611         stream.next_in = (unsigned char *)hdr;
1612         stream.avail_in = hdrlen;
1613         while (git_deflate(&stream, 0) == Z_OK)
1614                 ; /* nothing */
1615         the_hash_algo->update_fn(&c, hdr, hdrlen);
1616
1617         /* Then the data itself.. */
1618         stream.next_in = (void *)buf;
1619         stream.avail_in = len;
1620         do {
1621                 unsigned char *in0 = stream.next_in;
1622                 ret = git_deflate(&stream, Z_FINISH);
1623                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1624                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1625                         die("unable to write sha1 file");
1626                 stream.next_out = compressed;
1627                 stream.avail_out = sizeof(compressed);
1628         } while (ret == Z_OK);
1629
1630         if (ret != Z_STREAM_END)
1631                 die("unable to deflate new object %s (%d)", oid_to_hex(oid),
1632                     ret);
1633         ret = git_deflate_end_gently(&stream);
1634         if (ret != Z_OK)
1635                 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid),
1636                     ret);
1637         the_hash_algo->final_fn(parano_oid.hash, &c);
1638         if (oidcmp(oid, &parano_oid) != 0)
1639                 die("confused by unstable object source data for %s",
1640                     oid_to_hex(oid));
1641
1642         close_sha1_file(fd);
1643
1644         if (mtime) {
1645                 struct utimbuf utb;
1646                 utb.actime = mtime;
1647                 utb.modtime = mtime;
1648                 if (utime(tmp_file.buf, &utb) < 0)
1649                         warning_errno("failed utime() on %s", tmp_file.buf);
1650         }
1651
1652         return finalize_object_file(tmp_file.buf, filename.buf);
1653 }
1654
1655 static int freshen_loose_object(const unsigned char *sha1)
1656 {
1657         return check_and_freshen(sha1, 1);
1658 }
1659
1660 static int freshen_packed_object(const unsigned char *sha1)
1661 {
1662         struct pack_entry e;
1663         if (!find_pack_entry(sha1, &e))
1664                 return 0;
1665         if (e.p->freshened)
1666                 return 1;
1667         if (!freshen_file(e.p->pack_name))
1668                 return 0;
1669         e.p->freshened = 1;
1670         return 1;
1671 }
1672
1673 int write_object_file(const void *buf, unsigned long len, const char *type,
1674                       struct object_id *oid)
1675 {
1676         char hdr[MAX_HEADER_LEN];
1677         int hdrlen = sizeof(hdr);
1678
1679         /* Normally if we have it in the pack then we do not bother writing
1680          * it out into .git/objects/??/?{38} file.
1681          */
1682         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1683         if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1684                 return 0;
1685         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1686 }
1687
1688 int hash_object_file_literally(const void *buf, unsigned long len,
1689                                const char *type, struct object_id *oid,
1690                                unsigned flags)
1691 {
1692         char *header;
1693         int hdrlen, status = 0;
1694
1695         /* type string, SP, %lu of the length plus NUL must fit this */
1696         hdrlen = strlen(type) + MAX_HEADER_LEN;
1697         header = xmalloc(hdrlen);
1698         write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1699
1700         if (!(flags & HASH_WRITE_OBJECT))
1701                 goto cleanup;
1702         if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1703                 goto cleanup;
1704         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1705
1706 cleanup:
1707         free(header);
1708         return status;
1709 }
1710
1711 int force_object_loose(const struct object_id *oid, time_t mtime)
1712 {
1713         void *buf;
1714         unsigned long len;
1715         enum object_type type;
1716         char hdr[MAX_HEADER_LEN];
1717         int hdrlen;
1718         int ret;
1719
1720         if (has_loose_object(oid->hash))
1721                 return 0;
1722         buf = read_object(oid->hash, &type, &len);
1723         if (!buf)
1724                 return error("cannot read sha1_file for %s", oid_to_hex(oid));
1725         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1726         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1727         free(buf);
1728
1729         return ret;
1730 }
1731
1732 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1733 {
1734         struct object_id oid;
1735         if (!startup_info->have_repository)
1736                 return 0;
1737         hashcpy(oid.hash, sha1);
1738         return oid_object_info_extended(&oid, NULL,
1739                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1740 }
1741
1742 int has_object_file(const struct object_id *oid)
1743 {
1744         return has_sha1_file(oid->hash);
1745 }
1746
1747 int has_object_file_with_flags(const struct object_id *oid, int flags)
1748 {
1749         return has_sha1_file_with_flags(oid->hash, flags);
1750 }
1751
1752 static void check_tree(const void *buf, size_t size)
1753 {
1754         struct tree_desc desc;
1755         struct name_entry entry;
1756
1757         init_tree_desc(&desc, buf, size);
1758         while (tree_entry(&desc, &entry))
1759                 /* do nothing
1760                  * tree_entry() will die() on malformed entries */
1761                 ;
1762 }
1763
1764 static void check_commit(const void *buf, size_t size)
1765 {
1766         struct commit c;
1767         memset(&c, 0, sizeof(c));
1768         if (parse_commit_buffer(&c, buf, size))
1769                 die("corrupt commit");
1770 }
1771
1772 static void check_tag(const void *buf, size_t size)
1773 {
1774         struct tag t;
1775         memset(&t, 0, sizeof(t));
1776         if (parse_tag_buffer(&t, buf, size))
1777                 die("corrupt tag");
1778 }
1779
1780 static int index_mem(struct object_id *oid, void *buf, size_t size,
1781                      enum object_type type,
1782                      const char *path, unsigned flags)
1783 {
1784         int ret, re_allocated = 0;
1785         int write_object = flags & HASH_WRITE_OBJECT;
1786
1787         if (!type)
1788                 type = OBJ_BLOB;
1789
1790         /*
1791          * Convert blobs to git internal format
1792          */
1793         if ((type == OBJ_BLOB) && path) {
1794                 struct strbuf nbuf = STRBUF_INIT;
1795                 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1796                                    get_conv_flags(flags))) {
1797                         buf = strbuf_detach(&nbuf, &size);
1798                         re_allocated = 1;
1799                 }
1800         }
1801         if (flags & HASH_FORMAT_CHECK) {
1802                 if (type == OBJ_TREE)
1803                         check_tree(buf, size);
1804                 if (type == OBJ_COMMIT)
1805                         check_commit(buf, size);
1806                 if (type == OBJ_TAG)
1807                         check_tag(buf, size);
1808         }
1809
1810         if (write_object)
1811                 ret = write_object_file(buf, size, type_name(type), oid);
1812         else
1813                 ret = hash_object_file(buf, size, type_name(type), oid);
1814         if (re_allocated)
1815                 free(buf);
1816         return ret;
1817 }
1818
1819 static int index_stream_convert_blob(struct object_id *oid, int fd,
1820                                      const char *path, unsigned flags)
1821 {
1822         int ret;
1823         const int write_object = flags & HASH_WRITE_OBJECT;
1824         struct strbuf sbuf = STRBUF_INIT;
1825
1826         assert(path);
1827         assert(would_convert_to_git_filter_fd(path));
1828
1829         convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1830                                  get_conv_flags(flags));
1831
1832         if (write_object)
1833                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1834                                         oid);
1835         else
1836                 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1837                                        oid);
1838         strbuf_release(&sbuf);
1839         return ret;
1840 }
1841
1842 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1843                       const char *path, unsigned flags)
1844 {
1845         struct strbuf sbuf = STRBUF_INIT;
1846         int ret;
1847
1848         if (strbuf_read(&sbuf, fd, 4096) >= 0)
1849                 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1850         else
1851                 ret = -1;
1852         strbuf_release(&sbuf);
1853         return ret;
1854 }
1855
1856 #define SMALL_FILE_SIZE (32*1024)
1857
1858 static int index_core(struct object_id *oid, int fd, size_t size,
1859                       enum object_type type, const char *path,
1860                       unsigned flags)
1861 {
1862         int ret;
1863
1864         if (!size) {
1865                 ret = index_mem(oid, "", size, type, path, flags);
1866         } else if (size <= SMALL_FILE_SIZE) {
1867                 char *buf = xmalloc(size);
1868                 ssize_t read_result = read_in_full(fd, buf, size);
1869                 if (read_result < 0)
1870                         ret = error_errno("read error while indexing %s",
1871                                           path ? path : "<unknown>");
1872                 else if (read_result != size)
1873                         ret = error("short read while indexing %s",
1874                                     path ? path : "<unknown>");
1875                 else
1876                         ret = index_mem(oid, buf, size, type, path, flags);
1877                 free(buf);
1878         } else {
1879                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1880                 ret = index_mem(oid, buf, size, type, path, flags);
1881                 munmap(buf, size);
1882         }
1883         return ret;
1884 }
1885
1886 /*
1887  * This creates one packfile per large blob unless bulk-checkin
1888  * machinery is "plugged".
1889  *
1890  * This also bypasses the usual "convert-to-git" dance, and that is on
1891  * purpose. We could write a streaming version of the converting
1892  * functions and insert that before feeding the data to fast-import
1893  * (or equivalent in-core API described above). However, that is
1894  * somewhat complicated, as we do not know the size of the filter
1895  * result, which we need to know beforehand when writing a git object.
1896  * Since the primary motivation for trying to stream from the working
1897  * tree file and to avoid mmaping it in core is to deal with large
1898  * binary blobs, they generally do not want to get any conversion, and
1899  * callers should avoid this code path when filters are requested.
1900  */
1901 static int index_stream(struct object_id *oid, int fd, size_t size,
1902                         enum object_type type, const char *path,
1903                         unsigned flags)
1904 {
1905         return index_bulk_checkin(oid, fd, size, type, path, flags);
1906 }
1907
1908 int index_fd(struct object_id *oid, int fd, struct stat *st,
1909              enum object_type type, const char *path, unsigned flags)
1910 {
1911         int ret;
1912
1913         /*
1914          * Call xsize_t() only when needed to avoid potentially unnecessary
1915          * die() for large files.
1916          */
1917         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1918                 ret = index_stream_convert_blob(oid, fd, path, flags);
1919         else if (!S_ISREG(st->st_mode))
1920                 ret = index_pipe(oid, fd, type, path, flags);
1921         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1922                  (path && would_convert_to_git(&the_index, path)))
1923                 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1924                                  flags);
1925         else
1926                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1927                                    flags);
1928         close(fd);
1929         return ret;
1930 }
1931
1932 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1933 {
1934         int fd;
1935         struct strbuf sb = STRBUF_INIT;
1936         int rc = 0;
1937
1938         switch (st->st_mode & S_IFMT) {
1939         case S_IFREG:
1940                 fd = open(path, O_RDONLY);
1941                 if (fd < 0)
1942                         return error_errno("open(\"%s\")", path);
1943                 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1944                         return error("%s: failed to insert into database",
1945                                      path);
1946                 break;
1947         case S_IFLNK:
1948                 if (strbuf_readlink(&sb, path, st->st_size))
1949                         return error_errno("readlink(\"%s\")", path);
1950                 if (!(flags & HASH_WRITE_OBJECT))
1951                         hash_object_file(sb.buf, sb.len, blob_type, oid);
1952                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1953                         rc = error("%s: failed to insert into database", path);
1954                 strbuf_release(&sb);
1955                 break;
1956         case S_IFDIR:
1957                 return resolve_gitlink_ref(path, "HEAD", oid);
1958         default:
1959                 return error("%s: unsupported file type", path);
1960         }
1961         return rc;
1962 }
1963
1964 int read_pack_header(int fd, struct pack_header *header)
1965 {
1966         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1967                 /* "eof before pack header was fully read" */
1968                 return PH_ERROR_EOF;
1969
1970         if (header->hdr_signature != htonl(PACK_SIGNATURE))
1971                 /* "protocol error (pack signature mismatch detected)" */
1972                 return PH_ERROR_PACK_SIGNATURE;
1973         if (!pack_version_ok(header->hdr_version))
1974                 /* "protocol error (pack version unsupported)" */
1975                 return PH_ERROR_PROTOCOL;
1976         return 0;
1977 }
1978
1979 void assert_oid_type(const struct object_id *oid, enum object_type expect)
1980 {
1981         enum object_type type = oid_object_info(oid, NULL);
1982         if (type < 0)
1983                 die("%s is not a valid object", oid_to_hex(oid));
1984         if (type != expect)
1985                 die("%s is not a valid '%s' object", oid_to_hex(oid),
1986                     type_name(expect));
1987 }
1988
1989 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1990                                 struct strbuf *path,
1991                                 each_loose_object_fn obj_cb,
1992                                 each_loose_cruft_fn cruft_cb,
1993                                 each_loose_subdir_fn subdir_cb,
1994                                 void *data)
1995 {
1996         size_t origlen, baselen;
1997         DIR *dir;
1998         struct dirent *de;
1999         int r = 0;
2000         struct object_id oid;
2001
2002         if (subdir_nr > 0xff)
2003                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2004
2005         origlen = path->len;
2006         strbuf_complete(path, '/');
2007         strbuf_addf(path, "%02x", subdir_nr);
2008
2009         dir = opendir(path->buf);
2010         if (!dir) {
2011                 if (errno != ENOENT)
2012                         r = error_errno("unable to open %s", path->buf);
2013                 strbuf_setlen(path, origlen);
2014                 return r;
2015         }
2016
2017         oid.hash[0] = subdir_nr;
2018         strbuf_addch(path, '/');
2019         baselen = path->len;
2020
2021         while ((de = readdir(dir))) {
2022                 size_t namelen;
2023                 if (is_dot_or_dotdot(de->d_name))
2024                         continue;
2025
2026                 namelen = strlen(de->d_name);
2027                 strbuf_setlen(path, baselen);
2028                 strbuf_add(path, de->d_name, namelen);
2029                 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2030                     !hex_to_bytes(oid.hash + 1, de->d_name,
2031                                   GIT_SHA1_RAWSZ - 1)) {
2032                         if (obj_cb) {
2033                                 r = obj_cb(&oid, path->buf, data);
2034                                 if (r)
2035                                         break;
2036                         }
2037                         continue;
2038                 }
2039
2040                 if (cruft_cb) {
2041                         r = cruft_cb(de->d_name, path->buf, data);
2042                         if (r)
2043                                 break;
2044                 }
2045         }
2046         closedir(dir);
2047
2048         strbuf_setlen(path, baselen - 1);
2049         if (!r && subdir_cb)
2050                 r = subdir_cb(subdir_nr, path->buf, data);
2051
2052         strbuf_setlen(path, origlen);
2053
2054         return r;
2055 }
2056
2057 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2058                             each_loose_object_fn obj_cb,
2059                             each_loose_cruft_fn cruft_cb,
2060                             each_loose_subdir_fn subdir_cb,
2061                             void *data)
2062 {
2063         int r = 0;
2064         int i;
2065
2066         for (i = 0; i < 256; i++) {
2067                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2068                                                 subdir_cb, data);
2069                 if (r)
2070                         break;
2071         }
2072
2073         return r;
2074 }
2075
2076 int for_each_loose_file_in_objdir(const char *path,
2077                                   each_loose_object_fn obj_cb,
2078                                   each_loose_cruft_fn cruft_cb,
2079                                   each_loose_subdir_fn subdir_cb,
2080                                   void *data)
2081 {
2082         struct strbuf buf = STRBUF_INIT;
2083         int r;
2084
2085         strbuf_addstr(&buf, path);
2086         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2087                                               subdir_cb, data);
2088         strbuf_release(&buf);
2089
2090         return r;
2091 }
2092
2093 struct loose_alt_odb_data {
2094         each_loose_object_fn *cb;
2095         void *data;
2096 };
2097
2098 static int loose_from_alt_odb(struct alternate_object_database *alt,
2099                               void *vdata)
2100 {
2101         struct loose_alt_odb_data *data = vdata;
2102         struct strbuf buf = STRBUF_INIT;
2103         int r;
2104
2105         strbuf_addstr(&buf, alt->path);
2106         r = for_each_loose_file_in_objdir_buf(&buf,
2107                                               data->cb, NULL, NULL,
2108                                               data->data);
2109         strbuf_release(&buf);
2110         return r;
2111 }
2112
2113 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2114 {
2115         struct loose_alt_odb_data alt;
2116         int r;
2117
2118         r = for_each_loose_file_in_objdir(get_object_directory(),
2119                                           cb, NULL, NULL, data);
2120         if (r)
2121                 return r;
2122
2123         if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2124                 return 0;
2125
2126         alt.cb = cb;
2127         alt.data = data;
2128         return foreach_alt_odb(loose_from_alt_odb, &alt);
2129 }
2130
2131 static int check_stream_sha1(git_zstream *stream,
2132                              const char *hdr,
2133                              unsigned long size,
2134                              const char *path,
2135                              const unsigned char *expected_sha1)
2136 {
2137         git_hash_ctx c;
2138         unsigned char real_sha1[GIT_MAX_RAWSZ];
2139         unsigned char buf[4096];
2140         unsigned long total_read;
2141         int status = Z_OK;
2142
2143         the_hash_algo->init_fn(&c);
2144         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2145
2146         /*
2147          * We already read some bytes into hdr, but the ones up to the NUL
2148          * do not count against the object's content size.
2149          */
2150         total_read = stream->total_out - strlen(hdr) - 1;
2151
2152         /*
2153          * This size comparison must be "<=" to read the final zlib packets;
2154          * see the comment in unpack_sha1_rest for details.
2155          */
2156         while (total_read <= size &&
2157                (status == Z_OK || status == Z_BUF_ERROR)) {
2158                 stream->next_out = buf;
2159                 stream->avail_out = sizeof(buf);
2160                 if (size - total_read < stream->avail_out)
2161                         stream->avail_out = size - total_read;
2162                 status = git_inflate(stream, Z_FINISH);
2163                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2164                 total_read += stream->next_out - buf;
2165         }
2166         git_inflate_end(stream);
2167
2168         if (status != Z_STREAM_END) {
2169                 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2170                 return -1;
2171         }
2172         if (stream->avail_in) {
2173                 error("garbage at end of loose object '%s'",
2174                       sha1_to_hex(expected_sha1));
2175                 return -1;
2176         }
2177
2178         the_hash_algo->final_fn(real_sha1, &c);
2179         if (hashcmp(expected_sha1, real_sha1)) {
2180                 error("sha1 mismatch for %s (expected %s)", path,
2181                       sha1_to_hex(expected_sha1));
2182                 return -1;
2183         }
2184
2185         return 0;
2186 }
2187
2188 int read_loose_object(const char *path,
2189                       const struct object_id *expected_oid,
2190                       enum object_type *type,
2191                       unsigned long *size,
2192                       void **contents)
2193 {
2194         int ret = -1;
2195         void *map = NULL;
2196         unsigned long mapsize;
2197         git_zstream stream;
2198         char hdr[MAX_HEADER_LEN];
2199
2200         *contents = NULL;
2201
2202         map = map_sha1_file_1(path, NULL, &mapsize);
2203         if (!map) {
2204                 error_errno("unable to mmap %s", path);
2205                 goto out;
2206         }
2207
2208         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2209                 error("unable to unpack header of %s", path);
2210                 goto out;
2211         }
2212
2213         *type = parse_sha1_header(hdr, size);
2214         if (*type < 0) {
2215                 error("unable to parse header of %s", path);
2216                 git_inflate_end(&stream);
2217                 goto out;
2218         }
2219
2220         if (*type == OBJ_BLOB) {
2221                 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2222                         goto out;
2223         } else {
2224                 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2225                 if (!*contents) {
2226                         error("unable to unpack contents of %s", path);
2227                         git_inflate_end(&stream);
2228                         goto out;
2229                 }
2230                 if (check_object_signature(expected_oid, *contents,
2231                                          *size, type_name(*type))) {
2232                         error("sha1 mismatch for %s (expected %s)", path,
2233                               oid_to_hex(expected_oid));
2234                         free(*contents);
2235                         goto out;
2236                 }
2237         }
2238
2239         ret = 0; /* everything checks out */
2240
2241 out:
2242         if (map)
2243                 munmap(map, mapsize);
2244         return ret;
2245 }