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 "repository.h"
  26 #include "replace-object.h"
  27 #include "streaming.h"
  28 #include "dir.h"
  29 #include "list.h"
  30 #include "mergesort.h"
  31 #include "quote.h"
  32 #include "packfile.h"
  33 #include "fetch-object.h"
  34 #include "object-store.h"
  35
  36 /* The maximum size for an object header. */
  37 #define MAX_HEADER_LEN 32
  38
  39 const unsigned char null_sha1[GIT_MAX_RAWSZ];
  40 const struct object_id null_oid;
  41 const struct object_id empty_tree_oid = {
  42         EMPTY_TREE_SHA1_BIN_LITERAL
  43 };
  44 const struct object_id empty_blob_oid = {
  45         EMPTY_BLOB_SHA1_BIN_LITERAL
  46 };
  47
  48 static void git_hash_sha1_init(git_hash_ctx *ctx)
  49 {
  50         git_SHA1_Init(&ctx->sha1);
  51 }
  52
  53 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
  54 {
  55         git_SHA1_Update(&ctx->sha1, data, len);
  56 }
  57
  58 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
  59 {
  60         git_SHA1_Final(hash, &ctx->sha1);
  61 }
  62
  63 static void git_hash_unknown_init(git_hash_ctx *ctx)
  64 {
  65         die("trying to init unknown hash");
  66 }
  67
  68 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
  69 {
  70         die("trying to update unknown hash");
  71 }
  72
  73 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
  74 {
  75         die("trying to finalize unknown hash");
  76 }
  77
  78 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
  79         {
  80                 NULL,
  81                 0x00000000,
  82                 0,
  83                 0,
  84                 git_hash_unknown_init,
  85                 git_hash_unknown_update,
  86                 git_hash_unknown_final,
  87                 NULL,
  88                 NULL,
  89         },
  90         {
  91                 "sha-1",
  92                 /* "sha1", big-endian */
  93                 0x73686131,
  94                 GIT_SHA1_RAWSZ,
  95                 GIT_SHA1_HEXSZ,
  96                 git_hash_sha1_init,
  97                 git_hash_sha1_update,
  98                 git_hash_sha1_final,
  99                 &empty_tree_oid,
 100                 &empty_blob_oid,
 101         },
 102 };
 103
 104 /*
 105  * This is meant to hold a *small* number of objects that you would
 106  * want read_sha1_file() to be able to return, but yet you do not want
 107  * to write them into the object store (e.g. a browse-only
 108  * application).
 109  */
 110 static struct cached_object {
 111         unsigned char sha1[20];
 112         enum object_type type;
 113         void *buf;
 114         unsigned long size;
 115 } *cached_objects;
 116 static int cached_object_nr, cached_object_alloc;
 117
 118 static struct cached_object empty_tree = {
 119         EMPTY_TREE_SHA1_BIN_LITERAL,
 120         OBJ_TREE,
 121         "",
 122         0
 123 };
 124
 125 static struct cached_object *find_cached_object(const unsigned char *sha1)
 126 {
 127         int i;
 128         struct cached_object *co = cached_objects;
 129
 130         for (i = 0; i < cached_object_nr; i++, co++) {
 131                 if (!hashcmp(co->sha1, sha1))
 132                         return co;
 133         }
 134         if (!hashcmp(sha1, empty_tree.sha1))
 135                 return &empty_tree;
 136         return NULL;
 137 }
 138
 139
 140 static int get_conv_flags(unsigned flags)
 141 {
 142         if (flags & HASH_RENORMALIZE)
 143                 return CONV_EOL_RENORMALIZE;
 144         else if (flags & HASH_WRITE_OBJECT)
 145                 return global_conv_flags_eol | CONV_WRITE_OBJECT;
 146         else
 147                 return 0;
 148 }
 149
 150
 151 int mkdir_in_gitdir(const char *path)
 152 {
 153         if (mkdir(path, 0777)) {
 154                 int saved_errno = errno;
 155                 struct stat st;
 156                 struct strbuf sb = STRBUF_INIT;
 157
 158                 if (errno != EEXIST)
 159                         return -1;
 160                 /*
 161                  * Are we looking at a path in a symlinked worktree
 162                  * whose original repository does not yet have it?
 163                  * e.g. .git/rr-cache pointing at its original
 164                  * repository in which the user hasn't performed any
 165                  * conflict resolution yet?
 166                  */
 167                 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
 168                     strbuf_readlink(&sb, path, st.st_size) ||
 169                     !is_absolute_path(sb.buf) ||
 170                     mkdir(sb.buf, 0777)) {
 171                         strbuf_release(&sb);
 172                         errno = saved_errno;
 173                         return -1;
 174                 }
 175                 strbuf_release(&sb);
 176         }
 177         return adjust_shared_perm(path);
 178 }
 179
 180 enum scld_error safe_create_leading_directories(char *path)
 181 {
 182         char *next_component = path + offset_1st_component(path);
 183         enum scld_error ret = SCLD_OK;
 184
 185         while (ret == SCLD_OK && next_component) {
 186                 struct stat st;
 187                 char *slash = next_component, slash_character;
 188
 189                 while (*slash && !is_dir_sep(*slash))
 190                         slash++;
 191
 192                 if (!*slash)
 193                         break;
 194
 195                 next_component = slash + 1;
 196                 while (is_dir_sep(*next_component))
 197                         next_component++;
 198                 if (!*next_component)
 199                         break;
 200
 201                 slash_character = *slash;
 202                 *slash = '\0';
 203                 if (!stat(path, &st)) {
 204                         /* path exists */
 205                         if (!S_ISDIR(st.st_mode)) {
 206                                 errno = ENOTDIR;
 207                                 ret = SCLD_EXISTS;
 208                         }
 209                 } else if (mkdir(path, 0777)) {
 210                         if (errno == EEXIST &&
 211                             !stat(path, &st) && S_ISDIR(st.st_mode))
 212                                 ; /* somebody created it since we checked */
 213                         else if (errno == ENOENT)
 214                                 /*
 215                                  * Either mkdir() failed because
 216                                  * somebody just pruned the containing
 217                                  * directory, or stat() failed because
 218                                  * the file that was in our way was
 219                                  * just removed.  Either way, inform
 220                                  * the caller that it might be worth
 221                                  * trying again:
 222                                  */
 223                                 ret = SCLD_VANISHED;
 224                         else
 225                                 ret = SCLD_FAILED;
 226                 } else if (adjust_shared_perm(path)) {
 227                         ret = SCLD_PERMS;
 228                 }
 229                 *slash = slash_character;
 230         }
 231         return ret;
 232 }
 233
 234 enum scld_error safe_create_leading_directories_const(const char *path)
 235 {
 236         int save_errno;
 237         /* path points to cache entries, so xstrdup before messing with it */
 238         char *buf = xstrdup(path);
 239         enum scld_error result = safe_create_leading_directories(buf);
 240
 241         save_errno = errno;
 242         free(buf);
 243         errno = save_errno;
 244         return result;
 245 }
 246
 247 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
 248 {
 249         /*
 250          * The number of times we will try to remove empty directories
 251          * in the way of path. This is only 1 because if another
 252          * process is racily creating directories that conflict with
 253          * us, we don't want to fight against them.
 254          */
 255         int remove_directories_remaining = 1;
 256
 257         /*
 258          * The number of times that we will try to create the
 259          * directories containing path. We are willing to attempt this
 260          * more than once, because another process could be trying to
 261          * clean up empty directories at the same time as we are
 262          * trying to create them.
 263          */
 264         int create_directories_remaining = 3;
 265
 266         /* A scratch copy of path, filled lazily if we need it: */
 267         struct strbuf path_copy = STRBUF_INIT;
 268
 269         int ret, save_errno;
 270
 271         /* Sanity check: */
 272         assert(*path);
 273
 274 retry_fn:
 275         ret = fn(path, cb);
 276         save_errno = errno;
 277         if (!ret)
 278                 goto out;
 279
 280         if (errno == EISDIR && remove_directories_remaining-- > 0) {
 281                 /*
 282                  * A directory is in the way. Maybe it is empty; try
 283                  * to remove it:
 284                  */
 285                 if (!path_copy.len)
 286                         strbuf_addstr(&path_copy, path);
 287
 288                 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
 289                         goto retry_fn;
 290         } else if (errno == ENOENT && create_directories_remaining-- > 0) {
 291                 /*
 292                  * Maybe the containing directory didn't exist, or
 293                  * maybe it was just deleted by a process that is
 294                  * racing with us to clean up empty directories. Try
 295                  * to create it:
 296                  */
 297                 enum scld_error scld_result;
 298
 299                 if (!path_copy.len)
 300                         strbuf_addstr(&path_copy, path);
 301
 302                 do {
 303                         scld_result = safe_create_leading_directories(path_copy.buf);
 304                         if (scld_result == SCLD_OK)
 305                                 goto retry_fn;
 306                 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
 307         }
 308
 309 out:
 310         strbuf_release(&path_copy);
 311         errno = save_errno;
 312         return ret;
 313 }
 314
 315 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 316 {
 317         int i;
 318         for (i = 0; i < 20; i++) {
 319                 static char hex[] = "0123456789abcdef";
 320                 unsigned int val = sha1[i];
 321                 strbuf_addch(buf, hex[val >> 4]);
 322                 strbuf_addch(buf, hex[val & 0xf]);
 323                 if (!i)
 324                         strbuf_addch(buf, '/');
 325         }
 326 }
 327
 328 void sha1_file_name(struct repository *r, struct strbuf *buf, const unsigned char *sha1)
 329 {
 330         strbuf_addstr(buf, r->objects->objectdir);
 331         strbuf_addch(buf, '/');
 332         fill_sha1_path(buf, sha1);
 333 }
 334
 335 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
 336 {
 337         strbuf_setlen(&alt->scratch, alt->base_len);
 338         return &alt->scratch;
 339 }
 340
 341 static const char *alt_sha1_path(struct alternate_object_database *alt,
 342                                  const unsigned char *sha1)
 343 {
 344         struct strbuf *buf = alt_scratch_buf(alt);
 345         fill_sha1_path(buf, sha1);
 346         return buf->buf;
 347 }
 348
 349 /*
 350  * Return non-zero iff the path is usable as an alternate object database.
 351  */
 352 static int alt_odb_usable(struct raw_object_store *o,
 353                           struct strbuf *path,
 354                           const char *normalized_objdir)
 355 {
 356         struct alternate_object_database *alt;
 357
 358         /* Detect cases where alternate disappeared */
 359         if (!is_directory(path->buf)) {
 360                 error("object directory %s does not exist; "
 361                       "check .git/objects/info/alternates.",
 362                       path->buf);
 363                 return 0;
 364         }
 365
 366         /*
 367          * Prevent the common mistake of listing the same
 368          * thing twice, or object directory itself.
 369          */
 370         for (alt = o->alt_odb_list; alt; alt = alt->next) {
 371                 if (!fspathcmp(path->buf, alt->path))
 372                         return 0;
 373         }
 374         if (!fspathcmp(path->buf, normalized_objdir))
 375                 return 0;
 376
 377         return 1;
 378 }
 379
 380 /*
 381  * Prepare alternate object database registry.
 382  *
 383  * The variable alt_odb_list points at the list of struct
 384  * alternate_object_database.  The elements on this list come from
 385  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 386  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 387  * whose contents is similar to that environment variable but can be
 388  * LF separated.  Its base points at a statically allocated buffer that
 389  * contains "/the/directory/corresponding/to/.git/objects/...", while
 390  * its name points just after the slash at the end of ".git/objects/"
 391  * in the example above, and has enough space to hold 40-byte hex
 392  * SHA1, an extra slash for the first level indirection, and the
 393  * terminating NUL.
 394  */
 395 static void read_info_alternates(struct repository *r,
 396                                  const char *relative_base,
 397                                  int depth);
 398 static int link_alt_odb_entry(struct repository *r, const char *entry,
 399         const char *relative_base, int depth, const char *normalized_objdir)
 400 {
 401         struct alternate_object_database *ent;
 402         struct strbuf pathbuf = STRBUF_INIT;
 403
 404         if (!is_absolute_path(entry) && relative_base) {
 405                 strbuf_realpath(&pathbuf, relative_base, 1);
 406                 strbuf_addch(&pathbuf, '/');
 407         }
 408         strbuf_addstr(&pathbuf, entry);
 409
 410         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 411                 error("unable to normalize alternate object path: %s",
 412                       pathbuf.buf);
 413                 strbuf_release(&pathbuf);
 414                 return -1;
 415         }
 416
 417         /*
 418          * The trailing slash after the directory name is given by
 419          * this function at the end. Remove duplicates.
 420          */
 421         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 422                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
 423
 424         if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
 425                 strbuf_release(&pathbuf);
 426                 return -1;
 427         }
 428
 429         ent = alloc_alt_odb(pathbuf.buf);
 430
 431         /* add the alternate entry */
 432         *r->objects->alt_odb_tail = ent;
 433         r->objects->alt_odb_tail = &(ent->next);
 434         ent->next = NULL;
 435
 436         /* recursively add alternates */
 437         read_info_alternates(r, pathbuf.buf, depth + 1);
 438
 439         strbuf_release(&pathbuf);
 440         return 0;
 441 }
 442
 443 static const char *parse_alt_odb_entry(const char *string,
 444                                        int sep,
 445                                        struct strbuf *out)
 446 {
 447         const char *end;
 448
 449         strbuf_reset(out);
 450
 451         if (*string == '#') {
 452                 /* comment; consume up to next separator */
 453                 end = strchrnul(string, sep);
 454         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
 455                 /*
 456                  * quoted path; unquote_c_style has copied the
 457                  * data for us and set "end". Broken quoting (e.g.,
 458                  * an entry that doesn't end with a quote) falls
 459                  * back to the unquoted case below.
 460                  */
 461         } else {
 462                 /* normal, unquoted path */
 463                 end = strchrnul(string, sep);
 464                 strbuf_add(out, string, end - string);
 465         }
 466
 467         if (*end)
 468                 end++;
 469         return end;
 470 }
 471
 472 static void link_alt_odb_entries(struct repository *r, const char *alt,
 473                                  int sep, const char *relative_base, int depth)
 474 {
 475         struct strbuf objdirbuf = STRBUF_INIT;
 476         struct strbuf entry = STRBUF_INIT;
 477
 478         if (!alt || !*alt)
 479                 return;
 480
 481         if (depth > 5) {
 482                 error("%s: ignoring alternate object stores, nesting too deep.",
 483                                 relative_base);
 484                 return;
 485         }
 486
 487         strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
 488         if (strbuf_normalize_path(&objdirbuf) < 0)
 489                 die("unable to normalize object directory: %s",
 490                     objdirbuf.buf);
 491
 492         while (*alt) {
 493                 alt = parse_alt_odb_entry(alt, sep, &entry);
 494                 if (!entry.len)
 495                         continue;
 496                 link_alt_odb_entry(r, entry.buf,
 497                                    relative_base, depth, objdirbuf.buf);
 498         }
 499         strbuf_release(&entry);
 500         strbuf_release(&objdirbuf);
 501 }
 502
 503 static void read_info_alternates(struct repository *r,
 504                                  const char *relative_base,
 505                                  int depth)
 506 {
 507         char *path;
 508         struct strbuf buf = STRBUF_INIT;
 509
 510         path = xstrfmt("%s/info/alternates", relative_base);
 511         if (strbuf_read_file(&buf, path, 1024) < 0) {
 512                 warn_on_fopen_errors(path);
 513                 free(path);
 514                 return;
 515         }
 516
 517         link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
 518         strbuf_release(&buf);
 519         free(path);
 520 }
 521
 522 struct alternate_object_database *alloc_alt_odb(const char *dir)
 523 {
 524         struct alternate_object_database *ent;
 525
 526         FLEX_ALLOC_STR(ent, path, dir);
 527         strbuf_init(&ent->scratch, 0);
 528         strbuf_addf(&ent->scratch, "%s/", dir);
 529         ent->base_len = ent->scratch.len;
 530
 531         return ent;
 532 }
 533
 534 void add_to_alternates_file(const char *reference)
 535 {
 536         struct lock_file lock = LOCK_INIT;
 537         char *alts = git_pathdup("objects/info/alternates");
 538         FILE *in, *out;
 539         int found = 0;
 540
 541         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
 542         out = fdopen_lock_file(&lock, "w");
 543         if (!out)
 544                 die_errno("unable to fdopen alternates lockfile");
 545
 546         in = fopen(alts, "r");
 547         if (in) {
 548                 struct strbuf line = STRBUF_INIT;
 549
 550                 while (strbuf_getline(&line, in) != EOF) {
 551                         if (!strcmp(reference, line.buf)) {
 552                                 found = 1;
 553                                 break;
 554                         }
 555                         fprintf_or_die(out, "%s\n", line.buf);
 556                 }
 557
 558                 strbuf_release(&line);
 559                 fclose(in);
 560         }
 561         else if (errno != ENOENT)
 562                 die_errno("unable to read alternates file");
 563
 564         if (found) {
 565                 rollback_lock_file(&lock);
 566         } else {
 567                 fprintf_or_die(out, "%s\n", reference);
 568                 if (commit_lock_file(&lock))
 569                         die_errno("unable to move new alternates file into place");
 570                 if (the_repository->objects->alt_odb_tail)
 571                         link_alt_odb_entries(the_repository, reference,
 572                                              '\n', NULL, 0);
 573         }
 574         free(alts);
 575 }
 576
 577 void add_to_alternates_memory(const char *reference)
 578 {
 579         /*
 580          * Make sure alternates are initialized, or else our entry may be
 581          * overwritten when they are.
 582          */
 583         prepare_alt_odb(the_repository);
 584
 585         link_alt_odb_entries(the_repository, reference,
 586                              '\n', NULL, 0);
 587 }
 588
 589 /*
 590  * Compute the exact path an alternate is at and returns it. In case of
 591  * error NULL is returned and the human readable error is added to `err`
 592  * `path` may be relative and should point to $GITDIR.
 593  * `err` must not be null.
 594  */
 595 char *compute_alternate_path(const char *path, struct strbuf *err)
 596 {
 597         char *ref_git = NULL;
 598         const char *repo, *ref_git_s;
 599         int seen_error = 0;
 600
 601         ref_git_s = real_path_if_valid(path);
 602         if (!ref_git_s) {
 603                 seen_error = 1;
 604                 strbuf_addf(err, _("path '%s' does not exist"), path);
 605                 goto out;
 606         } else
 607                 /*
 608                  * Beware: read_gitfile(), real_path() and mkpath()
 609                  * return static buffer
 610                  */
 611                 ref_git = xstrdup(ref_git_s);
 612
 613         repo = read_gitfile(ref_git);
 614         if (!repo)
 615                 repo = read_gitfile(mkpath("%s/.git", ref_git));
 616         if (repo) {
 617                 free(ref_git);
 618                 ref_git = xstrdup(repo);
 619         }
 620
 621         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 622                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
 623                 free(ref_git);
 624                 ref_git = ref_git_git;
 625         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 626                 struct strbuf sb = STRBUF_INIT;
 627                 seen_error = 1;
 628                 if (get_common_dir(&sb, ref_git)) {
 629                         strbuf_addf(err,
 630                                     _("reference repository '%s' as a linked "
 631                                       "checkout is not supported yet."),
 632                                     path);
 633                         goto out;
 634                 }
 635
 636                 strbuf_addf(err, _("reference repository '%s' is not a "
 637                                         "local repository."), path);
 638                 goto out;
 639         }
 640
 641         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 642                 strbuf_addf(err, _("reference repository '%s' is shallow"),
 643                             path);
 644                 seen_error = 1;
 645                 goto out;
 646         }
 647
 648         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 649                 strbuf_addf(err,
 650                             _("reference repository '%s' is grafted"),
 651                             path);
 652                 seen_error = 1;
 653                 goto out;
 654         }
 655
 656 out:
 657         if (seen_error) {
 658                 FREE_AND_NULL(ref_git);
 659         }
 660
 661         return ref_git;
 662 }
 663
 664 int foreach_alt_odb(alt_odb_fn fn, void *cb)
 665 {
 666         struct alternate_object_database *ent;
 667         int r = 0;
 668
 669         prepare_alt_odb(the_repository);
 670         for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
 671                 r = fn(ent, cb);
 672                 if (r)
 673                         break;
 674         }
 675         return r;
 676 }
 677
 678 void prepare_alt_odb(struct repository *r)
 679 {
 680         if (r->objects->alt_odb_tail)
 681                 return;
 682
 683         r->objects->alt_odb_tail = &r->objects->alt_odb_list;
 684         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
 685
 686         read_info_alternates(r, r->objects->objectdir, 0);
 687 }
 688
 689 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 690 static int freshen_file(const char *fn)
 691 {
 692         struct utimbuf t;
 693         t.actime = t.modtime = time(NULL);
 694         return !utime(fn, &t);
 695 }
 696
 697 /*
 698  * All of the check_and_freshen functions return 1 if the file exists and was
 699  * freshened (if freshening was requested), 0 otherwise. If they return
 700  * 0, you should not assume that it is safe to skip a write of the object (it
 701  * either does not exist on disk, or has a stale mtime and may be subject to
 702  * pruning).
 703  */
 704 int check_and_freshen_file(const char *fn, int freshen)
 705 {
 706         if (access(fn, F_OK))
 707                 return 0;
 708         if (freshen && !freshen_file(fn))
 709                 return 0;
 710         return 1;
 711 }
 712
 713 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
 714 {
 715         static struct strbuf buf = STRBUF_INIT;
 716
 717         strbuf_reset(&buf);
 718         sha1_file_name(the_repository, &buf, sha1);
 719
 720         return check_and_freshen_file(buf.buf, freshen);
 721 }
 722
 723 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
 724 {
 725         struct alternate_object_database *alt;
 726         prepare_alt_odb(the_repository);
 727         for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
 728                 const char *path = alt_sha1_path(alt, sha1);
 729                 if (check_and_freshen_file(path, freshen))
 730                         return 1;
 731         }
 732         return 0;
 733 }
 734
 735 static int check_and_freshen(const unsigned char *sha1, int freshen)
 736 {
 737         return check_and_freshen_local(sha1, freshen) ||
 738                check_and_freshen_nonlocal(sha1, freshen);
 739 }
 740
 741 int has_loose_object_nonlocal(const unsigned char *sha1)
 742 {
 743         return check_and_freshen_nonlocal(sha1, 0);
 744 }
 745
 746 static int has_loose_object(const unsigned char *sha1)
 747 {
 748         return check_and_freshen(sha1, 0);
 749 }
 750
 751 static void mmap_limit_check(size_t length)
 752 {
 753         static size_t limit = 0;
 754         if (!limit) {
 755                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 756                 if (!limit)
 757                         limit = SIZE_MAX;
 758         }
 759         if (length > limit)
 760                 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
 761                     (uintmax_t)length, (uintmax_t)limit);
 762 }
 763
 764 void *xmmap_gently(void *start, size_t length,
 765                   int prot, int flags, int fd, off_t offset)
 766 {
 767         void *ret;
 768
 769         mmap_limit_check(length);
 770         ret = mmap(start, length, prot, flags, fd, offset);
 771         if (ret == MAP_FAILED) {
 772                 if (!length)
 773                         return NULL;
 774                 release_pack_memory(length);
 775                 ret = mmap(start, length, prot, flags, fd, offset);
 776         }
 777         return ret;
 778 }
 779
 780 void *xmmap(void *start, size_t length,
 781         int prot, int flags, int fd, off_t offset)
 782 {
 783         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 784         if (ret == MAP_FAILED)
 785                 die_errno("mmap failed");
 786         return ret;
 787 }
 788
 789 /*
 790  * With an in-core object data in "map", rehash it to make sure the
 791  * object name actually matches "sha1" to detect object corruption.
 792  * With "map" == NULL, try reading the object named with "sha1" using
 793  * the streaming interface and rehash it to do the same.
 794  */
 795 int check_object_signature(const struct object_id *oid, void *map,
 796                            unsigned long size, const char *type)
 797 {
 798         struct object_id real_oid;
 799         enum object_type obj_type;
 800         struct git_istream *st;
 801         git_hash_ctx c;
 802         char hdr[MAX_HEADER_LEN];
 803         int hdrlen;
 804
 805         if (map) {
 806                 hash_object_file(map, size, type, &real_oid);
 807                 return oidcmp(oid, &real_oid) ? -1 : 0;
 808         }
 809
 810         st = open_istream(oid, &obj_type, &size, NULL);
 811         if (!st)
 812                 return -1;
 813
 814         /* Generate the header */
 815         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
 816
 817         /* Sha1.. */
 818         the_hash_algo->init_fn(&c);
 819         the_hash_algo->update_fn(&c, hdr, hdrlen);
 820         for (;;) {
 821                 char buf[1024 * 16];
 822                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
 823
 824                 if (readlen < 0) {
 825                         close_istream(st);
 826                         return -1;
 827                 }
 828                 if (!readlen)
 829                         break;
 830                 the_hash_algo->update_fn(&c, buf, readlen);
 831         }
 832         the_hash_algo->final_fn(real_oid.hash, &c);
 833         close_istream(st);
 834         return oidcmp(oid, &real_oid) ? -1 : 0;
 835 }
 836
 837 int git_open_cloexec(const char *name, int flags)
 838 {
 839         int fd;
 840         static int o_cloexec = O_CLOEXEC;
 841
 842         fd = open(name, flags | o_cloexec);
 843         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 844                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
 845                 o_cloexec &= ~O_CLOEXEC;
 846                 fd = open(name, flags | o_cloexec);
 847         }
 848
 849 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 850         {
 851                 static int fd_cloexec = FD_CLOEXEC;
 852
 853                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
 854                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 855                         int flags = fcntl(fd, F_GETFD);
 856                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 857                                 fd_cloexec = 0;
 858                 }
 859         }
 860 #endif
 861         return fd;
 862 }
 863
 864 /*
 865  * Find "sha1" as a loose object in the local repository or in an alternate.
 866  * Returns 0 on success, negative on failure.
 867  *
 868  * The "path" out-parameter will give the path of the object we found (if any).
 869  * Note that it may point to static storage and is only valid until another
 870  * call to sha1_file_name(), etc.
 871  */
 872 static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
 873                           struct stat *st, const char **path)
 874 {
 875         struct alternate_object_database *alt;
 876         static struct strbuf buf = STRBUF_INIT;
 877
 878         strbuf_reset(&buf);
 879         sha1_file_name(r, &buf, sha1);
 880         *path = buf.buf;
 881
 882         if (!lstat(*path, st))
 883                 return 0;
 884
 885         prepare_alt_odb(r);
 886         errno = ENOENT;
 887         for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
 888                 *path = alt_sha1_path(alt, sha1);
 889                 if (!lstat(*path, st))
 890                         return 0;
 891         }
 892
 893         return -1;
 894 }
 895
 896 /*
 897  * Like stat_sha1_file(), but actually open the object and return the
 898  * descriptor. See the caveats on the "path" parameter above.
 899  */
 900 static int open_sha1_file(struct repository *r,
 901                           const unsigned char *sha1, const char **path)
 902 {
 903         int fd;
 904         struct alternate_object_database *alt;
 905         int most_interesting_errno;
 906         static struct strbuf buf = STRBUF_INIT;
 907
 908         strbuf_reset(&buf);
 909         sha1_file_name(r, &buf, sha1);
 910         *path = buf.buf;
 911
 912         fd = git_open(*path);
 913         if (fd >= 0)
 914                 return fd;
 915         most_interesting_errno = errno;
 916
 917         prepare_alt_odb(r);
 918         for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
 919                 *path = alt_sha1_path(alt, sha1);
 920                 fd = git_open(*path);
 921                 if (fd >= 0)
 922                         return fd;
 923                 if (most_interesting_errno == ENOENT)
 924                         most_interesting_errno = errno;
 925         }
 926         errno = most_interesting_errno;
 927         return -1;
 928 }
 929
 930 /*
 931  * Map the loose object at "path" if it is not NULL, or the path found by
 932  * searching for a loose object named "sha1".
 933  */
 934 static void *map_sha1_file_1(struct repository *r, const char *path,
 935                              const unsigned char *sha1, unsigned long *size)
 936 {
 937         void *map;
 938         int fd;
 939
 940         if (path)
 941                 fd = git_open(path);
 942         else
 943                 fd = open_sha1_file(r, sha1, &path);
 944         map = NULL;
 945         if (fd >= 0) {
 946                 struct stat st;
 947
 948                 if (!fstat(fd, &st)) {
 949                         *size = xsize_t(st.st_size);
 950                         if (!*size) {
 951                                 /* mmap() is forbidden on empty files */
 952                                 error("object file %s is empty", path);
 953                                 return NULL;
 954                         }
 955                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 956                 }
 957                 close(fd);
 958         }
 959         return map;
 960 }
 961
 962 void *map_sha1_file(struct repository *r,
 963                     const unsigned char *sha1, unsigned long *size)
 964 {
 965         return map_sha1_file_1(r, NULL, sha1, size);
 966 }
 967
 968 static int unpack_sha1_short_header(git_zstream *stream,
 969                                     unsigned char *map, unsigned long mapsize,
 970                                     void *buffer, unsigned long bufsiz)
 971 {
 972         /* Get the data stream */
 973         memset(stream, 0, sizeof(*stream));
 974         stream->next_in = map;
 975         stream->avail_in = mapsize;
 976         stream->next_out = buffer;
 977         stream->avail_out = bufsiz;
 978
 979         git_inflate_init(stream);
 980         return git_inflate(stream, 0);
 981 }
 982
 983 int unpack_sha1_header(git_zstream *stream,
 984                        unsigned char *map, unsigned long mapsize,
 985                        void *buffer, unsigned long bufsiz)
 986 {
 987         int status = unpack_sha1_short_header(stream, map, mapsize,
 988                                               buffer, bufsiz);
 989
 990         if (status < Z_OK)
 991                 return status;
 992
 993         /* Make sure we have the terminating NUL */
 994         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 995                 return -1;
 996         return 0;
 997 }
 998
 999 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1000                                         unsigned long mapsize, void *buffer,
1001                                         unsigned long bufsiz, struct strbuf *header)
1002 {
1003         int status;
1004
1005         status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1006         if (status < Z_OK)
1007                 return -1;
1008
1009         /*
1010          * Check if entire header is unpacked in the first iteration.
1011          */
1012         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1013                 return 0;
1014
1015         /*
1016          * buffer[0..bufsiz] was not large enough.  Copy the partial
1017          * result out to header, and then append the result of further
1018          * reading the stream.
1019          */
1020         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1021         stream->next_out = buffer;
1022         stream->avail_out = bufsiz;
1023
1024         do {
1025                 status = git_inflate(stream, 0);
1026                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1027                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1028                         return 0;
1029                 stream->next_out = buffer;
1030                 stream->avail_out = bufsiz;
1031         } while (status != Z_STREAM_END);
1032         return -1;
1033 }
1034
1035 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1036 {
1037         int bytes = strlen(buffer) + 1;
1038         unsigned char *buf = xmallocz(size);
1039         unsigned long n;
1040         int status = Z_OK;
1041
1042         n = stream->total_out - bytes;
1043         if (n > size)
1044                 n = size;
1045         memcpy(buf, (char *) buffer + bytes, n);
1046         bytes = n;
1047         if (bytes <= size) {
1048                 /*
1049                  * The above condition must be (bytes <= size), not
1050                  * (bytes < size).  In other words, even though we
1051                  * expect no more output and set avail_out to zero,
1052                  * the input zlib stream may have bytes that express
1053                  * "this concludes the stream", and we *do* want to
1054                  * eat that input.
1055                  *
1056                  * Otherwise we would not be able to test that we
1057                  * consumed all the input to reach the expected size;
1058                  * we also want to check that zlib tells us that all
1059                  * went well with status == Z_STREAM_END at the end.
1060                  */
1061                 stream->next_out = buf + bytes;
1062                 stream->avail_out = size - bytes;
1063                 while (status == Z_OK)
1064                         status = git_inflate(stream, Z_FINISH);
1065         }
1066         if (status == Z_STREAM_END && !stream->avail_in) {
1067                 git_inflate_end(stream);
1068                 return buf;
1069         }
1070
1071         if (status < 0)
1072                 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1073         else if (stream->avail_in)
1074                 error("garbage at end of loose object '%s'",
1075                       sha1_to_hex(sha1));
1076         free(buf);
1077         return NULL;
1078 }
1079
1080 /*
1081  * We used to just use "sscanf()", but that's actually way
1082  * too permissive for what we want to check. So do an anal
1083  * object header parse by hand.
1084  */
1085 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1086                                unsigned int flags)
1087 {
1088         const char *type_buf = hdr;
1089         unsigned long size;
1090         int type, type_len = 0;
1091
1092         /*
1093          * The type can be of any size but is followed by
1094          * a space.
1095          */
1096         for (;;) {
1097                 char c = *hdr++;
1098                 if (!c)
1099                         return -1;
1100                 if (c == ' ')
1101                         break;
1102                 type_len++;
1103         }
1104
1105         type = type_from_string_gently(type_buf, type_len, 1);
1106         if (oi->type_name)
1107                 strbuf_add(oi->type_name, type_buf, type_len);
1108         /*
1109          * Set type to 0 if its an unknown object and
1110          * we're obtaining the type using '--allow-unknown-type'
1111          * option.
1112          */
1113         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1114                 type = 0;
1115         else if (type < 0)
1116                 die("invalid object type");
1117         if (oi->typep)
1118                 *oi->typep = type;
1119
1120         /*
1121          * The length must follow immediately, and be in canonical
1122          * decimal format (ie "010" is not valid).
1123          */
1124         size = *hdr++ - '0';
1125         if (size > 9)
1126                 return -1;
1127         if (size) {
1128                 for (;;) {
1129                         unsigned long c = *hdr - '0';
1130                         if (c > 9)
1131                                 break;
1132                         hdr++;
1133                         size = size * 10 + c;
1134                 }
1135         }
1136
1137         if (oi->sizep)
1138                 *oi->sizep = size;
1139
1140         /*
1141          * The length must be followed by a zero byte
1142          */
1143         return *hdr ? -1 : type;
1144 }
1145
1146 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1147 {
1148         struct object_info oi = OBJECT_INFO_INIT;
1149
1150         oi.sizep = sizep;
1151         return parse_sha1_header_extended(hdr, &oi, 0);
1152 }
1153
1154 static int sha1_loose_object_info(struct repository *r,
1155                                   const unsigned char *sha1,
1156                                   struct object_info *oi, int flags)
1157 {
1158         int status = 0;
1159         unsigned long mapsize;
1160         void *map;
1161         git_zstream stream;
1162         char hdr[MAX_HEADER_LEN];
1163         struct strbuf hdrbuf = STRBUF_INIT;
1164         unsigned long size_scratch;
1165
1166         if (oi->delta_base_sha1)
1167                 hashclr(oi->delta_base_sha1);
1168
1169         /*
1170          * If we don't care about type or size, then we don't
1171          * need to look inside the object at all. Note that we
1172          * do not optimize out the stat call, even if the
1173          * caller doesn't care about the disk-size, since our
1174          * return value implicitly indicates whether the
1175          * object even exists.
1176          */
1177         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1178                 const char *path;
1179                 struct stat st;
1180                 if (stat_sha1_file(r, sha1, &st, &path) < 0)
1181                         return -1;
1182                 if (oi->disk_sizep)
1183                         *oi->disk_sizep = st.st_size;
1184                 return 0;
1185         }
1186
1187         map = map_sha1_file(r, sha1, &mapsize);
1188         if (!map)
1189                 return -1;
1190
1191         if (!oi->sizep)
1192                 oi->sizep = &size_scratch;
1193
1194         if (oi->disk_sizep)
1195                 *oi->disk_sizep = mapsize;
1196         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1197                 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1198                         status = error("unable to unpack %s header with --allow-unknown-type",
1199                                        sha1_to_hex(sha1));
1200         } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1201                 status = error("unable to unpack %s header",
1202                                sha1_to_hex(sha1));
1203         if (status < 0)
1204                 ; /* Do nothing */
1205         else if (hdrbuf.len) {
1206                 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1207                         status = error("unable to parse %s header with --allow-unknown-type",
1208                                        sha1_to_hex(sha1));
1209         } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1210                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1211
1212         if (status >= 0 && oi->contentp) {
1213                 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1214                                                  *oi->sizep, sha1);
1215                 if (!*oi->contentp) {
1216                         git_inflate_end(&stream);
1217                         status = -1;
1218                 }
1219         } else
1220                 git_inflate_end(&stream);
1221
1222         munmap(map, mapsize);
1223         if (status && oi->typep)
1224                 *oi->typep = status;
1225         if (oi->sizep == &size_scratch)
1226                 oi->sizep = NULL;
1227         strbuf_release(&hdrbuf);
1228         oi->whence = OI_LOOSE;
1229         return (status < 0) ? status : 0;
1230 }
1231
1232 int fetch_if_missing = 1;
1233
1234 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1235                              struct object_info *oi, unsigned flags)
1236 {
1237         static struct object_info blank_oi = OBJECT_INFO_INIT;
1238         struct pack_entry e;
1239         int rtype;
1240         const struct object_id *real = oid;
1241         int already_retried = 0;
1242
1243         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1244                 real = lookup_replace_object(r, oid);
1245
1246         if (is_null_oid(real))
1247                 return -1;
1248
1249         if (!oi)
1250                 oi = &blank_oi;
1251
1252         if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1253                 struct cached_object *co = find_cached_object(real->hash);
1254                 if (co) {
1255                         if (oi->typep)
1256                                 *(oi->typep) = co->type;
1257                         if (oi->sizep)
1258                                 *(oi->sizep) = co->size;
1259                         if (oi->disk_sizep)
1260                                 *(oi->disk_sizep) = 0;
1261                         if (oi->delta_base_sha1)
1262                                 hashclr(oi->delta_base_sha1);
1263                         if (oi->type_name)
1264                                 strbuf_addstr(oi->type_name, type_name(co->type));
1265                         if (oi->contentp)
1266                                 *oi->contentp = xmemdupz(co->buf, co->size);
1267                         oi->whence = OI_CACHED;
1268                         return 0;
1269                 }
1270         }
1271
1272         while (1) {
1273                 if (find_pack_entry(r, real->hash, &e))
1274                         break;
1275
1276                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1277                         return -1;
1278
1279                 /* Most likely it's a loose object. */
1280                 if (!sha1_loose_object_info(r, real->hash, oi, flags))
1281                         return 0;
1282
1283                 /* Not a loose object; someone else may have just packed it. */
1284                 if (!(flags & OBJECT_INFO_QUICK)) {
1285                         reprepare_packed_git(r);
1286                         if (find_pack_entry(r, real->hash, &e))
1287                                 break;
1288                 }
1289
1290                 /* Check if it is a missing object */
1291                 if (fetch_if_missing && repository_format_partial_clone &&
1292                     !already_retried && r == the_repository) {
1293                         /*
1294                          * TODO Investigate having fetch_object() return
1295                          * TODO error/success and stopping the music here.
1296                          * TODO Pass a repository struct through fetch_object,
1297                          * such that arbitrary repositories work.
1298                          */
1299                         fetch_object(repository_format_partial_clone, real->hash);
1300                         already_retried = 1;
1301                         continue;
1302                 }
1303
1304                 return -1;
1305         }
1306
1307         if (oi == &blank_oi)
1308                 /*
1309                  * We know that the caller doesn't actually need the
1310                  * information below, so return early.
1311                  */
1312                 return 0;
1313         rtype = packed_object_info(r, e.p, e.offset, oi);
1314         if (rtype < 0) {
1315                 mark_bad_packed_object(e.p, real->hash);
1316                 return oid_object_info_extended(r, real, oi, 0);
1317         } else if (oi->whence == OI_PACKED) {
1318                 oi->u.packed.offset = e.offset;
1319                 oi->u.packed.pack = e.p;
1320                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1321                                          rtype == OBJ_OFS_DELTA);
1322         }
1323
1324         return 0;
1325 }
1326
1327 /* returns enum object_type or negative */
1328 int oid_object_info(struct repository *r,
1329                     const struct object_id *oid,
1330                     unsigned long *sizep)
1331 {
1332         enum object_type type;
1333         struct object_info oi = OBJECT_INFO_INIT;
1334
1335         oi.typep = &type;
1336         oi.sizep = sizep;
1337         if (oid_object_info_extended(r, oid, &oi,
1338                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1339                 return -1;
1340         return type;
1341 }
1342
1343 static void *read_object(const unsigned char *sha1, enum object_type *type,
1344                          unsigned long *size)
1345 {
1346         struct object_id oid;
1347         struct object_info oi = OBJECT_INFO_INIT;
1348         void *content;
1349         oi.typep = type;
1350         oi.sizep = size;
1351         oi.contentp = &content;
1352
1353         hashcpy(oid.hash, sha1);
1354
1355         if (oid_object_info_extended(the_repository, &oid, &oi, 0) < 0)
1356                 return NULL;
1357         return content;
1358 }
1359
1360 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1361                         struct object_id *oid)
1362 {
1363         struct cached_object *co;
1364
1365         hash_object_file(buf, len, type_name(type), oid);
1366         if (has_sha1_file(oid->hash) || find_cached_object(oid->hash))
1367                 return 0;
1368         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1369         co = &cached_objects[cached_object_nr++];
1370         co->size = len;
1371         co->type = type;
1372         co->buf = xmalloc(len);
1373         memcpy(co->buf, buf, len);
1374         hashcpy(co->sha1, oid->hash);
1375         return 0;
1376 }
1377
1378 /*
1379  * This function dies on corrupt objects; the callers who want to
1380  * deal with them should arrange to call read_object() and give error
1381  * messages themselves.
1382  */
1383 void *read_object_file_extended(const struct object_id *oid,
1384                                 enum object_type *type,
1385                                 unsigned long *size,
1386                                 int lookup_replace)
1387 {
1388         void *data;
1389         const struct packed_git *p;
1390         const char *path;
1391         struct stat st;
1392         const struct object_id *repl = lookup_replace ?
1393                 lookup_replace_object(the_repository, oid) : oid;
1394
1395         errno = 0;
1396         data = read_object(repl->hash, type, size);
1397         if (data)
1398                 return data;
1399
1400         if (errno && errno != ENOENT)
1401                 die_errno("failed to read object %s", oid_to_hex(oid));
1402
1403         /* die if we replaced an object with one that does not exist */
1404         if (repl != oid)
1405                 die("replacement %s not found for %s",
1406                     oid_to_hex(repl), oid_to_hex(oid));
1407
1408         if (!stat_sha1_file(the_repository, repl->hash, &st, &path))
1409                 die("loose object %s (stored in %s) is corrupt",
1410                     oid_to_hex(repl), path);
1411
1412         if ((p = has_packed_and_bad(repl->hash)) != NULL)
1413                 die("packed object %s (stored in %s) is corrupt",
1414                     oid_to_hex(repl), p->pack_name);
1415
1416         return NULL;
1417 }
1418
1419 void *read_object_with_reference(const struct object_id *oid,
1420                                  const char *required_type_name,
1421                                  unsigned long *size,
1422                                  struct object_id *actual_oid_return)
1423 {
1424         enum object_type type, required_type;
1425         void *buffer;
1426         unsigned long isize;
1427         struct object_id actual_oid;
1428
1429         required_type = type_from_string(required_type_name);
1430         oidcpy(&actual_oid, oid);
1431         while (1) {
1432                 int ref_length = -1;
1433                 const char *ref_type = NULL;
1434
1435                 buffer = read_object_file(&actual_oid, &type, &isize);
1436                 if (!buffer)
1437                         return NULL;
1438                 if (type == required_type) {
1439                         *size = isize;
1440                         if (actual_oid_return)
1441                                 oidcpy(actual_oid_return, &actual_oid);
1442                         return buffer;
1443                 }
1444                 /* Handle references */
1445                 else if (type == OBJ_COMMIT)
1446                         ref_type = "tree ";
1447                 else if (type == OBJ_TAG)
1448                         ref_type = "object ";
1449                 else {
1450                         free(buffer);
1451                         return NULL;
1452                 }
1453                 ref_length = strlen(ref_type);
1454
1455                 if (ref_length + GIT_SHA1_HEXSZ > isize ||
1456                     memcmp(buffer, ref_type, ref_length) ||
1457                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1458                         free(buffer);
1459                         return NULL;
1460                 }
1461                 free(buffer);
1462                 /* Now we have the ID of the referred-to object in
1463                  * actual_oid.  Check again. */
1464         }
1465 }
1466
1467 static void write_object_file_prepare(const void *buf, unsigned long len,
1468                                       const char *type, struct object_id *oid,
1469                                       char *hdr, int *hdrlen)
1470 {
1471         git_hash_ctx c;
1472
1473         /* Generate the header */
1474         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1475
1476         /* Sha1.. */
1477         the_hash_algo->init_fn(&c);
1478         the_hash_algo->update_fn(&c, hdr, *hdrlen);
1479         the_hash_algo->update_fn(&c, buf, len);
1480         the_hash_algo->final_fn(oid->hash, &c);
1481 }
1482
1483 /*
1484  * Move the just written object into its final resting place.
1485  */
1486 int finalize_object_file(const char *tmpfile, const char *filename)
1487 {
1488         int ret = 0;
1489
1490         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1491                 goto try_rename;
1492         else if (link(tmpfile, filename))
1493                 ret = errno;
1494
1495         /*
1496          * Coda hack - coda doesn't like cross-directory links,
1497          * so we fall back to a rename, which will mean that it
1498          * won't be able to check collisions, but that's not a
1499          * big deal.
1500          *
1501          * The same holds for FAT formatted media.
1502          *
1503          * When this succeeds, we just return.  We have nothing
1504          * left to unlink.
1505          */
1506         if (ret && ret != EEXIST) {
1507         try_rename:
1508                 if (!rename(tmpfile, filename))
1509                         goto out;
1510                 ret = errno;
1511         }
1512         unlink_or_warn(tmpfile);
1513         if (ret) {
1514                 if (ret != EEXIST) {
1515                         return error_errno("unable to write sha1 filename %s", filename);
1516                 }
1517                 /* FIXME!!! Collision check here ? */
1518         }
1519
1520 out:
1521         if (adjust_shared_perm(filename))
1522                 return error("unable to set permission to '%s'", filename);
1523         return 0;
1524 }
1525
1526 static int write_buffer(int fd, const void *buf, size_t len)
1527 {
1528         if (write_in_full(fd, buf, len) < 0)
1529                 return error_errno("file write error");
1530         return 0;
1531 }
1532
1533 int hash_object_file(const void *buf, unsigned long len, const char *type,
1534                      struct object_id *oid)
1535 {
1536         char hdr[MAX_HEADER_LEN];
1537         int hdrlen = sizeof(hdr);
1538         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1539         return 0;
1540 }
1541
1542 /* Finalize a file on disk, and close it. */
1543 static void close_sha1_file(int fd)
1544 {
1545         if (fsync_object_files)
1546                 fsync_or_die(fd, "sha1 file");
1547         if (close(fd) != 0)
1548                 die_errno("error when closing sha1 file");
1549 }
1550
1551 /* Size of directory component, including the ending '/' */
1552 static inline int directory_size(const char *filename)
1553 {
1554         const char *s = strrchr(filename, '/');
1555         if (!s)
1556                 return 0;
1557         return s - filename + 1;
1558 }
1559
1560 /*
1561  * This creates a temporary file in the same directory as the final
1562  * 'filename'
1563  *
1564  * We want to avoid cross-directory filename renames, because those
1565  * can have problems on various filesystems (FAT, NFS, Coda).
1566  */
1567 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1568 {
1569         int fd, dirlen = directory_size(filename);
1570
1571         strbuf_reset(tmp);
1572         strbuf_add(tmp, filename, dirlen);
1573         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1574         fd = git_mkstemp_mode(tmp->buf, 0444);
1575         if (fd < 0 && dirlen && errno == ENOENT) {
1576                 /*
1577                  * Make sure the directory exists; note that the contents
1578                  * of the buffer are undefined after mkstemp returns an
1579                  * error, so we have to rewrite the whole buffer from
1580                  * scratch.
1581                  */
1582                 strbuf_reset(tmp);
1583                 strbuf_add(tmp, filename, dirlen - 1);
1584                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1585                         return -1;
1586                 if (adjust_shared_perm(tmp->buf))
1587                         return -1;
1588
1589                 /* Try again */
1590                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1591                 fd = git_mkstemp_mode(tmp->buf, 0444);
1592         }
1593         return fd;
1594 }
1595
1596 static int write_loose_object(const struct object_id *oid, char *hdr,
1597                               int hdrlen, const void *buf, unsigned long len,
1598                               time_t mtime)
1599 {
1600         int fd, ret;
1601         unsigned char compressed[4096];
1602         git_zstream stream;
1603         git_hash_ctx c;
1604         struct object_id parano_oid;
1605         static struct strbuf tmp_file = STRBUF_INIT;
1606         static struct strbuf filename = STRBUF_INIT;
1607
1608         strbuf_reset(&filename);
1609         sha1_file_name(the_repository, &filename, oid->hash);
1610
1611         fd = create_tmpfile(&tmp_file, filename.buf);
1612         if (fd < 0) {
1613                 if (errno == EACCES)
1614                         return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1615                 else
1616                         return error_errno("unable to create temporary file");
1617         }
1618
1619         /* Set it up */
1620         git_deflate_init(&stream, zlib_compression_level);
1621         stream.next_out = compressed;
1622         stream.avail_out = sizeof(compressed);
1623         the_hash_algo->init_fn(&c);
1624
1625         /* First header.. */
1626         stream.next_in = (unsigned char *)hdr;
1627         stream.avail_in = hdrlen;
1628         while (git_deflate(&stream, 0) == Z_OK)
1629                 ; /* nothing */
1630         the_hash_algo->update_fn(&c, hdr, hdrlen);
1631
1632         /* Then the data itself.. */
1633         stream.next_in = (void *)buf;
1634         stream.avail_in = len;
1635         do {
1636                 unsigned char *in0 = stream.next_in;
1637                 ret = git_deflate(&stream, Z_FINISH);
1638                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1639                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1640                         die("unable to write sha1 file");
1641                 stream.next_out = compressed;
1642                 stream.avail_out = sizeof(compressed);
1643         } while (ret == Z_OK);
1644
1645         if (ret != Z_STREAM_END)
1646                 die("unable to deflate new object %s (%d)", oid_to_hex(oid),
1647                     ret);
1648         ret = git_deflate_end_gently(&stream);
1649         if (ret != Z_OK)
1650                 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid),
1651                     ret);
1652         the_hash_algo->final_fn(parano_oid.hash, &c);
1653         if (oidcmp(oid, &parano_oid) != 0)
1654                 die("confused by unstable object source data for %s",
1655                     oid_to_hex(oid));
1656
1657         close_sha1_file(fd);
1658
1659         if (mtime) {
1660                 struct utimbuf utb;
1661                 utb.actime = mtime;
1662                 utb.modtime = mtime;
1663                 if (utime(tmp_file.buf, &utb) < 0)
1664                         warning_errno("failed utime() on %s", tmp_file.buf);
1665         }
1666
1667         return finalize_object_file(tmp_file.buf, filename.buf);
1668 }
1669
1670 static int freshen_loose_object(const unsigned char *sha1)
1671 {
1672         return check_and_freshen(sha1, 1);
1673 }
1674
1675 static int freshen_packed_object(const unsigned char *sha1)
1676 {
1677         struct pack_entry e;
1678         if (!find_pack_entry(the_repository, sha1, &e))
1679                 return 0;
1680         if (e.p->freshened)
1681                 return 1;
1682         if (!freshen_file(e.p->pack_name))
1683                 return 0;
1684         e.p->freshened = 1;
1685         return 1;
1686 }
1687
1688 int write_object_file(const void *buf, unsigned long len, const char *type,
1689                       struct object_id *oid)
1690 {
1691         char hdr[MAX_HEADER_LEN];
1692         int hdrlen = sizeof(hdr);
1693
1694         /* Normally if we have it in the pack then we do not bother writing
1695          * it out into .git/objects/??/?{38} file.
1696          */
1697         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1698         if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1699                 return 0;
1700         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1701 }
1702
1703 int hash_object_file_literally(const void *buf, unsigned long len,
1704                                const char *type, struct object_id *oid,
1705                                unsigned flags)
1706 {
1707         char *header;
1708         int hdrlen, status = 0;
1709
1710         /* type string, SP, %lu of the length plus NUL must fit this */
1711         hdrlen = strlen(type) + MAX_HEADER_LEN;
1712         header = xmalloc(hdrlen);
1713         write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1714
1715         if (!(flags & HASH_WRITE_OBJECT))
1716                 goto cleanup;
1717         if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1718                 goto cleanup;
1719         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1720
1721 cleanup:
1722         free(header);
1723         return status;
1724 }
1725
1726 int force_object_loose(const struct object_id *oid, time_t mtime)
1727 {
1728         void *buf;
1729         unsigned long len;
1730         enum object_type type;
1731         char hdr[MAX_HEADER_LEN];
1732         int hdrlen;
1733         int ret;
1734
1735         if (has_loose_object(oid->hash))
1736                 return 0;
1737         buf = read_object(oid->hash, &type, &len);
1738         if (!buf)
1739                 return error("cannot read sha1_file for %s", oid_to_hex(oid));
1740         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1741         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1742         free(buf);
1743
1744         return ret;
1745 }
1746
1747 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1748 {
1749         struct object_id oid;
1750         if (!startup_info->have_repository)
1751                 return 0;
1752         hashcpy(oid.hash, sha1);
1753         return oid_object_info_extended(the_repository, &oid, NULL,
1754                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1755 }
1756
1757 int has_object_file(const struct object_id *oid)
1758 {
1759         return has_sha1_file(oid->hash);
1760 }
1761
1762 int has_object_file_with_flags(const struct object_id *oid, int flags)
1763 {
1764         return has_sha1_file_with_flags(oid->hash, flags);
1765 }
1766
1767 static void check_tree(const void *buf, size_t size)
1768 {
1769         struct tree_desc desc;
1770         struct name_entry entry;
1771
1772         init_tree_desc(&desc, buf, size);
1773         while (tree_entry(&desc, &entry))
1774                 /* do nothing
1775                  * tree_entry() will die() on malformed entries */
1776                 ;
1777 }
1778
1779 static void check_commit(const void *buf, size_t size)
1780 {
1781         struct commit c;
1782         memset(&c, 0, sizeof(c));
1783         if (parse_commit_buffer(&c, buf, size))
1784                 die("corrupt commit");
1785 }
1786
1787 static void check_tag(const void *buf, size_t size)
1788 {
1789         struct tag t;
1790         memset(&t, 0, sizeof(t));
1791         if (parse_tag_buffer(&t, buf, size))
1792                 die("corrupt tag");
1793 }
1794
1795 static int index_mem(struct object_id *oid, void *buf, size_t size,
1796                      enum object_type type,
1797                      const char *path, unsigned flags)
1798 {
1799         int ret, re_allocated = 0;
1800         int write_object = flags & HASH_WRITE_OBJECT;
1801
1802         if (!type)
1803                 type = OBJ_BLOB;
1804
1805         /*
1806          * Convert blobs to git internal format
1807          */
1808         if ((type == OBJ_BLOB) && path) {
1809                 struct strbuf nbuf = STRBUF_INIT;
1810                 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1811                                    get_conv_flags(flags))) {
1812                         buf = strbuf_detach(&nbuf, &size);
1813                         re_allocated = 1;
1814                 }
1815         }
1816         if (flags & HASH_FORMAT_CHECK) {
1817                 if (type == OBJ_TREE)
1818                         check_tree(buf, size);
1819                 if (type == OBJ_COMMIT)
1820                         check_commit(buf, size);
1821                 if (type == OBJ_TAG)
1822                         check_tag(buf, size);
1823         }
1824
1825         if (write_object)
1826                 ret = write_object_file(buf, size, type_name(type), oid);
1827         else
1828                 ret = hash_object_file(buf, size, type_name(type), oid);
1829         if (re_allocated)
1830                 free(buf);
1831         return ret;
1832 }
1833
1834 static int index_stream_convert_blob(struct object_id *oid, int fd,
1835                                      const char *path, unsigned flags)
1836 {
1837         int ret;
1838         const int write_object = flags & HASH_WRITE_OBJECT;
1839         struct strbuf sbuf = STRBUF_INIT;
1840
1841         assert(path);
1842         assert(would_convert_to_git_filter_fd(path));
1843
1844         convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1845                                  get_conv_flags(flags));
1846
1847         if (write_object)
1848                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1849                                         oid);
1850         else
1851                 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1852                                        oid);
1853         strbuf_release(&sbuf);
1854         return ret;
1855 }
1856
1857 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1858                       const char *path, unsigned flags)
1859 {
1860         struct strbuf sbuf = STRBUF_INIT;
1861         int ret;
1862
1863         if (strbuf_read(&sbuf, fd, 4096) >= 0)
1864                 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1865         else
1866                 ret = -1;
1867         strbuf_release(&sbuf);
1868         return ret;
1869 }
1870
1871 #define SMALL_FILE_SIZE (32*1024)
1872
1873 static int index_core(struct object_id *oid, int fd, size_t size,
1874                       enum object_type type, const char *path,
1875                       unsigned flags)
1876 {
1877         int ret;
1878
1879         if (!size) {
1880                 ret = index_mem(oid, "", size, type, path, flags);
1881         } else if (size <= SMALL_FILE_SIZE) {
1882                 char *buf = xmalloc(size);
1883                 ssize_t read_result = read_in_full(fd, buf, size);
1884                 if (read_result < 0)
1885                         ret = error_errno("read error while indexing %s",
1886                                           path ? path : "<unknown>");
1887                 else if (read_result != size)
1888                         ret = error("short read while indexing %s",
1889                                     path ? path : "<unknown>");
1890                 else
1891                         ret = index_mem(oid, buf, size, type, path, flags);
1892                 free(buf);
1893         } else {
1894                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1895                 ret = index_mem(oid, buf, size, type, path, flags);
1896                 munmap(buf, size);
1897         }
1898         return ret;
1899 }
1900
1901 /*
1902  * This creates one packfile per large blob unless bulk-checkin
1903  * machinery is "plugged".
1904  *
1905  * This also bypasses the usual "convert-to-git" dance, and that is on
1906  * purpose. We could write a streaming version of the converting
1907  * functions and insert that before feeding the data to fast-import
1908  * (or equivalent in-core API described above). However, that is
1909  * somewhat complicated, as we do not know the size of the filter
1910  * result, which we need to know beforehand when writing a git object.
1911  * Since the primary motivation for trying to stream from the working
1912  * tree file and to avoid mmaping it in core is to deal with large
1913  * binary blobs, they generally do not want to get any conversion, and
1914  * callers should avoid this code path when filters are requested.
1915  */
1916 static int index_stream(struct object_id *oid, int fd, size_t size,
1917                         enum object_type type, const char *path,
1918                         unsigned flags)
1919 {
1920         return index_bulk_checkin(oid, fd, size, type, path, flags);
1921 }
1922
1923 int index_fd(struct object_id *oid, int fd, struct stat *st,
1924              enum object_type type, const char *path, unsigned flags)
1925 {
1926         int ret;
1927
1928         /*
1929          * Call xsize_t() only when needed to avoid potentially unnecessary
1930          * die() for large files.
1931          */
1932         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1933                 ret = index_stream_convert_blob(oid, fd, path, flags);
1934         else if (!S_ISREG(st->st_mode))
1935                 ret = index_pipe(oid, fd, type, path, flags);
1936         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1937                  (path && would_convert_to_git(&the_index, path)))
1938                 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1939                                  flags);
1940         else
1941                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1942                                    flags);
1943         close(fd);
1944         return ret;
1945 }
1946
1947 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1948 {
1949         int fd;
1950         struct strbuf sb = STRBUF_INIT;
1951         int rc = 0;
1952
1953         switch (st->st_mode & S_IFMT) {
1954         case S_IFREG:
1955                 fd = open(path, O_RDONLY);
1956                 if (fd < 0)
1957                         return error_errno("open(\"%s\")", path);
1958                 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1959                         return error("%s: failed to insert into database",
1960                                      path);
1961                 break;
1962         case S_IFLNK:
1963                 if (strbuf_readlink(&sb, path, st->st_size))
1964                         return error_errno("readlink(\"%s\")", path);
1965                 if (!(flags & HASH_WRITE_OBJECT))
1966                         hash_object_file(sb.buf, sb.len, blob_type, oid);
1967                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1968                         rc = error("%s: failed to insert into database", path);
1969                 strbuf_release(&sb);
1970                 break;
1971         case S_IFDIR:
1972                 return resolve_gitlink_ref(path, "HEAD", oid);
1973         default:
1974                 return error("%s: unsupported file type", path);
1975         }
1976         return rc;
1977 }
1978
1979 int read_pack_header(int fd, struct pack_header *header)
1980 {
1981         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1982                 /* "eof before pack header was fully read" */
1983                 return PH_ERROR_EOF;
1984
1985         if (header->hdr_signature != htonl(PACK_SIGNATURE))
1986                 /* "protocol error (pack signature mismatch detected)" */
1987                 return PH_ERROR_PACK_SIGNATURE;
1988         if (!pack_version_ok(header->hdr_version))
1989                 /* "protocol error (pack version unsupported)" */
1990                 return PH_ERROR_PROTOCOL;
1991         return 0;
1992 }
1993
1994 void assert_oid_type(const struct object_id *oid, enum object_type expect)
1995 {
1996         enum object_type type = oid_object_info(the_repository, oid, NULL);
1997         if (type < 0)
1998                 die("%s is not a valid object", oid_to_hex(oid));
1999         if (type != expect)
2000                 die("%s is not a valid '%s' object", oid_to_hex(oid),
2001                     type_name(expect));
2002 }
2003
2004 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2005                                 struct strbuf *path,
2006                                 each_loose_object_fn obj_cb,
2007                                 each_loose_cruft_fn cruft_cb,
2008                                 each_loose_subdir_fn subdir_cb,
2009                                 void *data)
2010 {
2011         size_t origlen, baselen;
2012         DIR *dir;
2013         struct dirent *de;
2014         int r = 0;
2015         struct object_id oid;
2016
2017         if (subdir_nr > 0xff)
2018                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2019
2020         origlen = path->len;
2021         strbuf_complete(path, '/');
2022         strbuf_addf(path, "%02x", subdir_nr);
2023
2024         dir = opendir(path->buf);
2025         if (!dir) {
2026                 if (errno != ENOENT)
2027                         r = error_errno("unable to open %s", path->buf);
2028                 strbuf_setlen(path, origlen);
2029                 return r;
2030         }
2031
2032         oid.hash[0] = subdir_nr;
2033         strbuf_addch(path, '/');
2034         baselen = path->len;
2035
2036         while ((de = readdir(dir))) {
2037                 size_t namelen;
2038                 if (is_dot_or_dotdot(de->d_name))
2039                         continue;
2040
2041                 namelen = strlen(de->d_name);
2042                 strbuf_setlen(path, baselen);
2043                 strbuf_add(path, de->d_name, namelen);
2044                 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2045                     !hex_to_bytes(oid.hash + 1, de->d_name,
2046                                   GIT_SHA1_RAWSZ - 1)) {
2047                         if (obj_cb) {
2048                                 r = obj_cb(&oid, path->buf, data);
2049                                 if (r)
2050                                         break;
2051                         }
2052                         continue;
2053                 }
2054
2055                 if (cruft_cb) {
2056                         r = cruft_cb(de->d_name, path->buf, data);
2057                         if (r)
2058                                 break;
2059                 }
2060         }
2061         closedir(dir);
2062
2063         strbuf_setlen(path, baselen - 1);
2064         if (!r && subdir_cb)
2065                 r = subdir_cb(subdir_nr, path->buf, data);
2066
2067         strbuf_setlen(path, origlen);
2068
2069         return r;
2070 }
2071
2072 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2073                             each_loose_object_fn obj_cb,
2074                             each_loose_cruft_fn cruft_cb,
2075                             each_loose_subdir_fn subdir_cb,
2076                             void *data)
2077 {
2078         int r = 0;
2079         int i;
2080
2081         for (i = 0; i < 256; i++) {
2082                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2083                                                 subdir_cb, data);
2084                 if (r)
2085                         break;
2086         }
2087
2088         return r;
2089 }
2090
2091 int for_each_loose_file_in_objdir(const char *path,
2092                                   each_loose_object_fn obj_cb,
2093                                   each_loose_cruft_fn cruft_cb,
2094                                   each_loose_subdir_fn subdir_cb,
2095                                   void *data)
2096 {
2097         struct strbuf buf = STRBUF_INIT;
2098         int r;
2099
2100         strbuf_addstr(&buf, path);
2101         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2102                                               subdir_cb, data);
2103         strbuf_release(&buf);
2104
2105         return r;
2106 }
2107
2108 struct loose_alt_odb_data {
2109         each_loose_object_fn *cb;
2110         void *data;
2111 };
2112
2113 static int loose_from_alt_odb(struct alternate_object_database *alt,
2114                               void *vdata)
2115 {
2116         struct loose_alt_odb_data *data = vdata;
2117         struct strbuf buf = STRBUF_INIT;
2118         int r;
2119
2120         strbuf_addstr(&buf, alt->path);
2121         r = for_each_loose_file_in_objdir_buf(&buf,
2122                                               data->cb, NULL, NULL,
2123                                               data->data);
2124         strbuf_release(&buf);
2125         return r;
2126 }
2127
2128 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2129 {
2130         struct loose_alt_odb_data alt;
2131         int r;
2132
2133         r = for_each_loose_file_in_objdir(get_object_directory(),
2134                                           cb, NULL, NULL, data);
2135         if (r)
2136                 return r;
2137
2138         if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2139                 return 0;
2140
2141         alt.cb = cb;
2142         alt.data = data;
2143         return foreach_alt_odb(loose_from_alt_odb, &alt);
2144 }
2145
2146 static int check_stream_sha1(git_zstream *stream,
2147                              const char *hdr,
2148                              unsigned long size,
2149                              const char *path,
2150                              const unsigned char *expected_sha1)
2151 {
2152         git_hash_ctx c;
2153         unsigned char real_sha1[GIT_MAX_RAWSZ];
2154         unsigned char buf[4096];
2155         unsigned long total_read;
2156         int status = Z_OK;
2157
2158         the_hash_algo->init_fn(&c);
2159         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2160
2161         /*
2162          * We already read some bytes into hdr, but the ones up to the NUL
2163          * do not count against the object's content size.
2164          */
2165         total_read = stream->total_out - strlen(hdr) - 1;
2166
2167         /*
2168          * This size comparison must be "<=" to read the final zlib packets;
2169          * see the comment in unpack_sha1_rest for details.
2170          */
2171         while (total_read <= size &&
2172                (status == Z_OK || status == Z_BUF_ERROR)) {
2173                 stream->next_out = buf;
2174                 stream->avail_out = sizeof(buf);
2175                 if (size - total_read < stream->avail_out)
2176                         stream->avail_out = size - total_read;
2177                 status = git_inflate(stream, Z_FINISH);
2178                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2179                 total_read += stream->next_out - buf;
2180         }
2181         git_inflate_end(stream);
2182
2183         if (status != Z_STREAM_END) {
2184                 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2185                 return -1;
2186         }
2187         if (stream->avail_in) {
2188                 error("garbage at end of loose object '%s'",
2189                       sha1_to_hex(expected_sha1));
2190                 return -1;
2191         }
2192
2193         the_hash_algo->final_fn(real_sha1, &c);
2194         if (hashcmp(expected_sha1, real_sha1)) {
2195                 error("sha1 mismatch for %s (expected %s)", path,
2196                       sha1_to_hex(expected_sha1));
2197                 return -1;
2198         }
2199
2200         return 0;
2201 }
2202
2203 int read_loose_object(const char *path,
2204                       const struct object_id *expected_oid,
2205                       enum object_type *type,
2206                       unsigned long *size,
2207                       void **contents)
2208 {
2209         int ret = -1;
2210         void *map = NULL;
2211         unsigned long mapsize;
2212         git_zstream stream;
2213         char hdr[MAX_HEADER_LEN];
2214
2215         *contents = NULL;
2216
2217         map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2218         if (!map) {
2219                 error_errno("unable to mmap %s", path);
2220                 goto out;
2221         }
2222
2223         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2224                 error("unable to unpack header of %s", path);
2225                 goto out;
2226         }
2227
2228         *type = parse_sha1_header(hdr, size);
2229         if (*type < 0) {
2230                 error("unable to parse header of %s", path);
2231                 git_inflate_end(&stream);
2232                 goto out;
2233         }
2234
2235         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2236                 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2237                         goto out;
2238         } else {
2239                 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2240                 if (!*contents) {
2241                         error("unable to unpack contents of %s", path);
2242                         git_inflate_end(&stream);
2243                         goto out;
2244                 }
2245                 if (check_object_signature(expected_oid, *contents,
2246                                          *size, type_name(*type))) {
2247                         error("sha1 mismatch for %s (expected %s)", path,
2248                               oid_to_hex(expected_oid));
2249                         free(*contents);
2250                         goto out;
2251                 }
2252         }
2253
2254         ret = 0; /* everything checks out */
2255
2256 out:
2257         if (map)
2258                 munmap(map, mapsize);
2259         return ret;
2260 }