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
  40 #define EMPTY_TREE_SHA1_BIN_LITERAL \
  41          "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
  42          "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
  43
  44 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
  45         "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
  46         "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
  47
  48 const unsigned char null_sha1[GIT_MAX_RAWSZ];
  49 const struct object_id null_oid;
  50 static const struct object_id empty_tree_oid = {
  51         EMPTY_TREE_SHA1_BIN_LITERAL
  52 };
  53 static const struct object_id empty_blob_oid = {
  54         EMPTY_BLOB_SHA1_BIN_LITERAL
  55 };
  56
  57 static void git_hash_sha1_init(git_hash_ctx *ctx)
  58 {
  59         git_SHA1_Init(&ctx->sha1);
  60 }
  61
  62 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
  63 {
  64         git_SHA1_Update(&ctx->sha1, data, len);
  65 }
  66
  67 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
  68 {
  69         git_SHA1_Final(hash, &ctx->sha1);
  70 }
  71
  72 static void git_hash_unknown_init(git_hash_ctx *ctx)
  73 {
  74         BUG("trying to init unknown hash");
  75 }
  76
  77 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
  78 {
  79         BUG("trying to update unknown hash");
  80 }
  81
  82 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
  83 {
  84         BUG("trying to finalize unknown hash");
  85 }
  86
  87 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
  88         {
  89                 NULL,
  90                 0x00000000,
  91                 0,
  92                 0,
  93                 git_hash_unknown_init,
  94                 git_hash_unknown_update,
  95                 git_hash_unknown_final,
  96                 NULL,
  97                 NULL,
  98         },
  99         {
 100                 "sha-1",
 101                 /* "sha1", big-endian */
 102                 0x73686131,
 103                 GIT_SHA1_RAWSZ,
 104                 GIT_SHA1_HEXSZ,
 105                 git_hash_sha1_init,
 106                 git_hash_sha1_update,
 107                 git_hash_sha1_final,
 108                 &empty_tree_oid,
 109                 &empty_blob_oid,
 110         },
 111 };
 112
 113 const char *empty_tree_oid_hex(void)
 114 {
 115         static char buf[GIT_MAX_HEXSZ + 1];
 116         return oid_to_hex_r(buf, the_hash_algo->empty_tree);
 117 }
 118
 119 const char *empty_blob_oid_hex(void)
 120 {
 121         static char buf[GIT_MAX_HEXSZ + 1];
 122         return oid_to_hex_r(buf, the_hash_algo->empty_blob);
 123 }
 124
 125 /*
 126  * This is meant to hold a *small* number of objects that you would
 127  * want read_sha1_file() to be able to return, but yet you do not want
 128  * to write them into the object store (e.g. a browse-only
 129  * application).
 130  */
 131 static struct cached_object {
 132         struct object_id oid;
 133         enum object_type type;
 134         void *buf;
 135         unsigned long size;
 136 } *cached_objects;
 137 static int cached_object_nr, cached_object_alloc;
 138
 139 static struct cached_object empty_tree = {
 140         { EMPTY_TREE_SHA1_BIN_LITERAL },
 141         OBJ_TREE,
 142         "",
 143         0
 144 };
 145
 146 static struct cached_object *find_cached_object(const struct object_id *oid)
 147 {
 148         int i;
 149         struct cached_object *co = cached_objects;
 150
 151         for (i = 0; i < cached_object_nr; i++, co++) {
 152                 if (oideq(&co->oid, oid))
 153                         return co;
 154         }
 155         if (oideq(oid, the_hash_algo->empty_tree))
 156                 return &empty_tree;
 157         return NULL;
 158 }
 159
 160
 161 static int get_conv_flags(unsigned flags)
 162 {
 163         if (flags & HASH_RENORMALIZE)
 164                 return CONV_EOL_RENORMALIZE;
 165         else if (flags & HASH_WRITE_OBJECT)
 166                 return global_conv_flags_eol | CONV_WRITE_OBJECT;
 167         else
 168                 return 0;
 169 }
 170
 171
 172 int mkdir_in_gitdir(const char *path)
 173 {
 174         if (mkdir(path, 0777)) {
 175                 int saved_errno = errno;
 176                 struct stat st;
 177                 struct strbuf sb = STRBUF_INIT;
 178
 179                 if (errno != EEXIST)
 180                         return -1;
 181                 /*
 182                  * Are we looking at a path in a symlinked worktree
 183                  * whose original repository does not yet have it?
 184                  * e.g. .git/rr-cache pointing at its original
 185                  * repository in which the user hasn't performed any
 186                  * conflict resolution yet?
 187                  */
 188                 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
 189                     strbuf_readlink(&sb, path, st.st_size) ||
 190                     !is_absolute_path(sb.buf) ||
 191                     mkdir(sb.buf, 0777)) {
 192                         strbuf_release(&sb);
 193                         errno = saved_errno;
 194                         return -1;
 195                 }
 196                 strbuf_release(&sb);
 197         }
 198         return adjust_shared_perm(path);
 199 }
 200
 201 enum scld_error safe_create_leading_directories(char *path)
 202 {
 203         char *next_component = path + offset_1st_component(path);
 204         enum scld_error ret = SCLD_OK;
 205
 206         while (ret == SCLD_OK && next_component) {
 207                 struct stat st;
 208                 char *slash = next_component, slash_character;
 209
 210                 while (*slash && !is_dir_sep(*slash))
 211                         slash++;
 212
 213                 if (!*slash)
 214                         break;
 215
 216                 next_component = slash + 1;
 217                 while (is_dir_sep(*next_component))
 218                         next_component++;
 219                 if (!*next_component)
 220                         break;
 221
 222                 slash_character = *slash;
 223                 *slash = '\0';
 224                 if (!stat(path, &st)) {
 225                         /* path exists */
 226                         if (!S_ISDIR(st.st_mode)) {
 227                                 errno = ENOTDIR;
 228                                 ret = SCLD_EXISTS;
 229                         }
 230                 } else if (mkdir(path, 0777)) {
 231                         if (errno == EEXIST &&
 232                             !stat(path, &st) && S_ISDIR(st.st_mode))
 233                                 ; /* somebody created it since we checked */
 234                         else if (errno == ENOENT)
 235                                 /*
 236                                  * Either mkdir() failed because
 237                                  * somebody just pruned the containing
 238                                  * directory, or stat() failed because
 239                                  * the file that was in our way was
 240                                  * just removed.  Either way, inform
 241                                  * the caller that it might be worth
 242                                  * trying again:
 243                                  */
 244                                 ret = SCLD_VANISHED;
 245                         else
 246                                 ret = SCLD_FAILED;
 247                 } else if (adjust_shared_perm(path)) {
 248                         ret = SCLD_PERMS;
 249                 }
 250                 *slash = slash_character;
 251         }
 252         return ret;
 253 }
 254
 255 enum scld_error safe_create_leading_directories_const(const char *path)
 256 {
 257         int save_errno;
 258         /* path points to cache entries, so xstrdup before messing with it */
 259         char *buf = xstrdup(path);
 260         enum scld_error result = safe_create_leading_directories(buf);
 261
 262         save_errno = errno;
 263         free(buf);
 264         errno = save_errno;
 265         return result;
 266 }
 267
 268 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
 269 {
 270         /*
 271          * The number of times we will try to remove empty directories
 272          * in the way of path. This is only 1 because if another
 273          * process is racily creating directories that conflict with
 274          * us, we don't want to fight against them.
 275          */
 276         int remove_directories_remaining = 1;
 277
 278         /*
 279          * The number of times that we will try to create the
 280          * directories containing path. We are willing to attempt this
 281          * more than once, because another process could be trying to
 282          * clean up empty directories at the same time as we are
 283          * trying to create them.
 284          */
 285         int create_directories_remaining = 3;
 286
 287         /* A scratch copy of path, filled lazily if we need it: */
 288         struct strbuf path_copy = STRBUF_INIT;
 289
 290         int ret, save_errno;
 291
 292         /* Sanity check: */
 293         assert(*path);
 294
 295 retry_fn:
 296         ret = fn(path, cb);
 297         save_errno = errno;
 298         if (!ret)
 299                 goto out;
 300
 301         if (errno == EISDIR && remove_directories_remaining-- > 0) {
 302                 /*
 303                  * A directory is in the way. Maybe it is empty; try
 304                  * to remove it:
 305                  */
 306                 if (!path_copy.len)
 307                         strbuf_addstr(&path_copy, path);
 308
 309                 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
 310                         goto retry_fn;
 311         } else if (errno == ENOENT && create_directories_remaining-- > 0) {
 312                 /*
 313                  * Maybe the containing directory didn't exist, or
 314                  * maybe it was just deleted by a process that is
 315                  * racing with us to clean up empty directories. Try
 316                  * to create it:
 317                  */
 318                 enum scld_error scld_result;
 319
 320                 if (!path_copy.len)
 321                         strbuf_addstr(&path_copy, path);
 322
 323                 do {
 324                         scld_result = safe_create_leading_directories(path_copy.buf);
 325                         if (scld_result == SCLD_OK)
 326                                 goto retry_fn;
 327                 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
 328         }
 329
 330 out:
 331         strbuf_release(&path_copy);
 332         errno = save_errno;
 333         return ret;
 334 }
 335
 336 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 337 {
 338         int i;
 339         for (i = 0; i < the_hash_algo->rawsz; i++) {
 340                 static char hex[] = "0123456789abcdef";
 341                 unsigned int val = sha1[i];
 342                 strbuf_addch(buf, hex[val >> 4]);
 343                 strbuf_addch(buf, hex[val & 0xf]);
 344                 if (!i)
 345                         strbuf_addch(buf, '/');
 346         }
 347 }
 348
 349 static const char *odb_loose_path(struct object_directory *odb,
 350                                   struct strbuf *buf,
 351                                   const unsigned char *sha1)
 352 {
 353         strbuf_reset(buf);
 354         strbuf_addstr(buf, odb->path);
 355         strbuf_addch(buf, '/');
 356         fill_sha1_path(buf, sha1);
 357         return buf->buf;
 358 }
 359
 360 const char *loose_object_path(struct repository *r, struct strbuf *buf,
 361                               const unsigned char *sha1)
 362 {
 363         return odb_loose_path(r->objects->odb, buf, sha1);
 364 }
 365
 366 /*
 367  * Return non-zero iff the path is usable as an alternate object database.
 368  */
 369 static int alt_odb_usable(struct raw_object_store *o,
 370                           struct strbuf *path,
 371                           const char *normalized_objdir)
 372 {
 373         struct object_directory *odb;
 374
 375         /* Detect cases where alternate disappeared */
 376         if (!is_directory(path->buf)) {
 377                 error(_("object directory %s does not exist; "
 378                         "check .git/objects/info/alternates"),
 379                       path->buf);
 380                 return 0;
 381         }
 382
 383         /*
 384          * Prevent the common mistake of listing the same
 385          * thing twice, or object directory itself.
 386          */
 387         for (odb = o->odb; odb; odb = odb->next) {
 388                 if (!fspathcmp(path->buf, odb->path))
 389                         return 0;
 390         }
 391         if (!fspathcmp(path->buf, normalized_objdir))
 392                 return 0;
 393
 394         return 1;
 395 }
 396
 397 /*
 398  * Prepare alternate object database registry.
 399  *
 400  * The variable alt_odb_list points at the list of struct
 401  * object_directory.  The elements on this list come from
 402  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 403  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 404  * whose contents is similar to that environment variable but can be
 405  * LF separated.  Its base points at a statically allocated buffer that
 406  * contains "/the/directory/corresponding/to/.git/objects/...", while
 407  * its name points just after the slash at the end of ".git/objects/"
 408  * in the example above, and has enough space to hold 40-byte hex
 409  * SHA1, an extra slash for the first level indirection, and the
 410  * terminating NUL.
 411  */
 412 static void read_info_alternates(struct repository *r,
 413                                  const char *relative_base,
 414                                  int depth);
 415 static int link_alt_odb_entry(struct repository *r, const char *entry,
 416         const char *relative_base, int depth, const char *normalized_objdir)
 417 {
 418         struct object_directory *ent;
 419         struct strbuf pathbuf = STRBUF_INIT;
 420
 421         if (!is_absolute_path(entry) && relative_base) {
 422                 strbuf_realpath(&pathbuf, relative_base, 1);
 423                 strbuf_addch(&pathbuf, '/');
 424         }
 425         strbuf_addstr(&pathbuf, entry);
 426
 427         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 428                 error(_("unable to normalize alternate object path: %s"),
 429                       pathbuf.buf);
 430                 strbuf_release(&pathbuf);
 431                 return -1;
 432         }
 433
 434         /*
 435          * The trailing slash after the directory name is given by
 436          * this function at the end. Remove duplicates.
 437          */
 438         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 439                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
 440
 441         if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
 442                 strbuf_release(&pathbuf);
 443                 return -1;
 444         }
 445
 446         ent = xcalloc(1, sizeof(*ent));
 447         ent->path = xstrdup(pathbuf.buf);
 448
 449         /* add the alternate entry */
 450         *r->objects->odb_tail = ent;
 451         r->objects->odb_tail = &(ent->next);
 452         ent->next = NULL;
 453
 454         /* recursively add alternates */
 455         read_info_alternates(r, pathbuf.buf, depth + 1);
 456
 457         strbuf_release(&pathbuf);
 458         return 0;
 459 }
 460
 461 static const char *parse_alt_odb_entry(const char *string,
 462                                        int sep,
 463                                        struct strbuf *out)
 464 {
 465         const char *end;
 466
 467         strbuf_reset(out);
 468
 469         if (*string == '#') {
 470                 /* comment; consume up to next separator */
 471                 end = strchrnul(string, sep);
 472         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
 473                 /*
 474                  * quoted path; unquote_c_style has copied the
 475                  * data for us and set "end". Broken quoting (e.g.,
 476                  * an entry that doesn't end with a quote) falls
 477                  * back to the unquoted case below.
 478                  */
 479         } else {
 480                 /* normal, unquoted path */
 481                 end = strchrnul(string, sep);
 482                 strbuf_add(out, string, end - string);
 483         }
 484
 485         if (*end)
 486                 end++;
 487         return end;
 488 }
 489
 490 static void link_alt_odb_entries(struct repository *r, const char *alt,
 491                                  int sep, const char *relative_base, int depth)
 492 {
 493         struct strbuf objdirbuf = STRBUF_INIT;
 494         struct strbuf entry = STRBUF_INIT;
 495
 496         if (!alt || !*alt)
 497                 return;
 498
 499         if (depth > 5) {
 500                 error(_("%s: ignoring alternate object stores, nesting too deep"),
 501                                 relative_base);
 502                 return;
 503         }
 504
 505         strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
 506         if (strbuf_normalize_path(&objdirbuf) < 0)
 507                 die(_("unable to normalize object directory: %s"),
 508                     objdirbuf.buf);
 509
 510         while (*alt) {
 511                 alt = parse_alt_odb_entry(alt, sep, &entry);
 512                 if (!entry.len)
 513                         continue;
 514                 link_alt_odb_entry(r, entry.buf,
 515                                    relative_base, depth, objdirbuf.buf);
 516         }
 517         strbuf_release(&entry);
 518         strbuf_release(&objdirbuf);
 519 }
 520
 521 static void read_info_alternates(struct repository *r,
 522                                  const char *relative_base,
 523                                  int depth)
 524 {
 525         char *path;
 526         struct strbuf buf = STRBUF_INIT;
 527
 528         path = xstrfmt("%s/info/alternates", relative_base);
 529         if (strbuf_read_file(&buf, path, 1024) < 0) {
 530                 warn_on_fopen_errors(path);
 531                 free(path);
 532                 return;
 533         }
 534
 535         link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
 536         strbuf_release(&buf);
 537         free(path);
 538 }
 539
 540 void add_to_alternates_file(const char *reference)
 541 {
 542         struct lock_file lock = LOCK_INIT;
 543         char *alts = git_pathdup("objects/info/alternates");
 544         FILE *in, *out;
 545         int found = 0;
 546
 547         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
 548         out = fdopen_lock_file(&lock, "w");
 549         if (!out)
 550                 die_errno(_("unable to fdopen alternates lockfile"));
 551
 552         in = fopen(alts, "r");
 553         if (in) {
 554                 struct strbuf line = STRBUF_INIT;
 555
 556                 while (strbuf_getline(&line, in) != EOF) {
 557                         if (!strcmp(reference, line.buf)) {
 558                                 found = 1;
 559                                 break;
 560                         }
 561                         fprintf_or_die(out, "%s\n", line.buf);
 562                 }
 563
 564                 strbuf_release(&line);
 565                 fclose(in);
 566         }
 567         else if (errno != ENOENT)
 568                 die_errno(_("unable to read alternates file"));
 569
 570         if (found) {
 571                 rollback_lock_file(&lock);
 572         } else {
 573                 fprintf_or_die(out, "%s\n", reference);
 574                 if (commit_lock_file(&lock))
 575                         die_errno(_("unable to move new alternates file into place"));
 576                 if (the_repository->objects->loaded_alternates)
 577                         link_alt_odb_entries(the_repository, reference,
 578                                              '\n', NULL, 0);
 579         }
 580         free(alts);
 581 }
 582
 583 void add_to_alternates_memory(const char *reference)
 584 {
 585         /*
 586          * Make sure alternates are initialized, or else our entry may be
 587          * overwritten when they are.
 588          */
 589         prepare_alt_odb(the_repository);
 590
 591         link_alt_odb_entries(the_repository, reference,
 592                              '\n', NULL, 0);
 593 }
 594
 595 /*
 596  * Compute the exact path an alternate is at and returns it. In case of
 597  * error NULL is returned and the human readable error is added to `err`
 598  * `path` may be relative and should point to $GIT_DIR.
 599  * `err` must not be null.
 600  */
 601 char *compute_alternate_path(const char *path, struct strbuf *err)
 602 {
 603         char *ref_git = NULL;
 604         const char *repo, *ref_git_s;
 605         int seen_error = 0;
 606
 607         ref_git_s = real_path_if_valid(path);
 608         if (!ref_git_s) {
 609                 seen_error = 1;
 610                 strbuf_addf(err, _("path '%s' does not exist"), path);
 611                 goto out;
 612         } else
 613                 /*
 614                  * Beware: read_gitfile(), real_path() and mkpath()
 615                  * return static buffer
 616                  */
 617                 ref_git = xstrdup(ref_git_s);
 618
 619         repo = read_gitfile(ref_git);
 620         if (!repo)
 621                 repo = read_gitfile(mkpath("%s/.git", ref_git));
 622         if (repo) {
 623                 free(ref_git);
 624                 ref_git = xstrdup(repo);
 625         }
 626
 627         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 628                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
 629                 free(ref_git);
 630                 ref_git = ref_git_git;
 631         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 632                 struct strbuf sb = STRBUF_INIT;
 633                 seen_error = 1;
 634                 if (get_common_dir(&sb, ref_git)) {
 635                         strbuf_addf(err,
 636                                     _("reference repository '%s' as a linked "
 637                                       "checkout is not supported yet."),
 638                                     path);
 639                         goto out;
 640                 }
 641
 642                 strbuf_addf(err, _("reference repository '%s' is not a "
 643                                         "local repository."), path);
 644                 goto out;
 645         }
 646
 647         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 648                 strbuf_addf(err, _("reference repository '%s' is shallow"),
 649                             path);
 650                 seen_error = 1;
 651                 goto out;
 652         }
 653
 654         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 655                 strbuf_addf(err,
 656                             _("reference repository '%s' is grafted"),
 657                             path);
 658                 seen_error = 1;
 659                 goto out;
 660         }
 661
 662 out:
 663         if (seen_error) {
 664                 FREE_AND_NULL(ref_git);
 665         }
 666
 667         return ref_git;
 668 }
 669
 670 int foreach_alt_odb(alt_odb_fn fn, void *cb)
 671 {
 672         struct object_directory *ent;
 673         int r = 0;
 674
 675         prepare_alt_odb(the_repository);
 676         for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
 677                 r = fn(ent, cb);
 678                 if (r)
 679                         break;
 680         }
 681         return r;
 682 }
 683
 684 void prepare_alt_odb(struct repository *r)
 685 {
 686         if (r->objects->loaded_alternates)
 687                 return;
 688
 689         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
 690
 691         read_info_alternates(r, r->objects->odb->path, 0);
 692         r->objects->loaded_alternates = 1;
 693 }
 694
 695 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 696 static int freshen_file(const char *fn)
 697 {
 698         struct utimbuf t;
 699         t.actime = t.modtime = time(NULL);
 700         return !utime(fn, &t);
 701 }
 702
 703 /*
 704  * All of the check_and_freshen functions return 1 if the file exists and was
 705  * freshened (if freshening was requested), 0 otherwise. If they return
 706  * 0, you should not assume that it is safe to skip a write of the object (it
 707  * either does not exist on disk, or has a stale mtime and may be subject to
 708  * pruning).
 709  */
 710 int check_and_freshen_file(const char *fn, int freshen)
 711 {
 712         if (access(fn, F_OK))
 713                 return 0;
 714         if (freshen && !freshen_file(fn))
 715                 return 0;
 716         return 1;
 717 }
 718
 719 static int check_and_freshen_odb(struct object_directory *odb,
 720                                  const struct object_id *oid,
 721                                  int freshen)
 722 {
 723         static struct strbuf path = STRBUF_INIT;
 724         odb_loose_path(odb, &path, oid->hash);
 725         return check_and_freshen_file(path.buf, freshen);
 726 }
 727
 728 static int check_and_freshen_local(const struct object_id *oid, int freshen)
 729 {
 730         return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
 731 }
 732
 733 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
 734 {
 735         struct object_directory *odb;
 736
 737         prepare_alt_odb(the_repository);
 738         for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
 739                 if (check_and_freshen_odb(odb, oid, freshen))
 740                         return 1;
 741         }
 742         return 0;
 743 }
 744
 745 static int check_and_freshen(const struct object_id *oid, int freshen)
 746 {
 747         return check_and_freshen_local(oid, freshen) ||
 748                check_and_freshen_nonlocal(oid, freshen);
 749 }
 750
 751 int has_loose_object_nonlocal(const struct object_id *oid)
 752 {
 753         return check_and_freshen_nonlocal(oid, 0);
 754 }
 755
 756 static int has_loose_object(const struct object_id *oid)
 757 {
 758         return check_and_freshen(oid, 0);
 759 }
 760
 761 static void mmap_limit_check(size_t length)
 762 {
 763         static size_t limit = 0;
 764         if (!limit) {
 765                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 766                 if (!limit)
 767                         limit = SIZE_MAX;
 768         }
 769         if (length > limit)
 770                 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
 771                     (uintmax_t)length, (uintmax_t)limit);
 772 }
 773
 774 void *xmmap_gently(void *start, size_t length,
 775                   int prot, int flags, int fd, off_t offset)
 776 {
 777         void *ret;
 778
 779         mmap_limit_check(length);
 780         ret = mmap(start, length, prot, flags, fd, offset);
 781         if (ret == MAP_FAILED) {
 782                 if (!length)
 783                         return NULL;
 784                 release_pack_memory(length);
 785                 ret = mmap(start, length, prot, flags, fd, offset);
 786         }
 787         return ret;
 788 }
 789
 790 void *xmmap(void *start, size_t length,
 791         int prot, int flags, int fd, off_t offset)
 792 {
 793         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 794         if (ret == MAP_FAILED)
 795                 die_errno(_("mmap failed"));
 796         return ret;
 797 }
 798
 799 /*
 800  * With an in-core object data in "map", rehash it to make sure the
 801  * object name actually matches "sha1" to detect object corruption.
 802  * With "map" == NULL, try reading the object named with "sha1" using
 803  * the streaming interface and rehash it to do the same.
 804  */
 805 int check_object_signature(const struct object_id *oid, void *map,
 806                            unsigned long size, const char *type)
 807 {
 808         struct object_id real_oid;
 809         enum object_type obj_type;
 810         struct git_istream *st;
 811         git_hash_ctx c;
 812         char hdr[MAX_HEADER_LEN];
 813         int hdrlen;
 814
 815         if (map) {
 816                 hash_object_file(map, size, type, &real_oid);
 817                 return !oideq(oid, &real_oid) ? -1 : 0;
 818         }
 819
 820         st = open_istream(oid, &obj_type, &size, NULL);
 821         if (!st)
 822                 return -1;
 823
 824         /* Generate the header */
 825         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
 826
 827         /* Sha1.. */
 828         the_hash_algo->init_fn(&c);
 829         the_hash_algo->update_fn(&c, hdr, hdrlen);
 830         for (;;) {
 831                 char buf[1024 * 16];
 832                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
 833
 834                 if (readlen < 0) {
 835                         close_istream(st);
 836                         return -1;
 837                 }
 838                 if (!readlen)
 839                         break;
 840                 the_hash_algo->update_fn(&c, buf, readlen);
 841         }
 842         the_hash_algo->final_fn(real_oid.hash, &c);
 843         close_istream(st);
 844         return !oideq(oid, &real_oid) ? -1 : 0;
 845 }
 846
 847 int git_open_cloexec(const char *name, int flags)
 848 {
 849         int fd;
 850         static int o_cloexec = O_CLOEXEC;
 851
 852         fd = open(name, flags | o_cloexec);
 853         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 854                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
 855                 o_cloexec &= ~O_CLOEXEC;
 856                 fd = open(name, flags | o_cloexec);
 857         }
 858
 859 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 860         {
 861                 static int fd_cloexec = FD_CLOEXEC;
 862
 863                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
 864                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 865                         int flags = fcntl(fd, F_GETFD);
 866                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 867                                 fd_cloexec = 0;
 868                 }
 869         }
 870 #endif
 871         return fd;
 872 }
 873
 874 /*
 875  * Find "sha1" as a loose object in the local repository or in an alternate.
 876  * Returns 0 on success, negative on failure.
 877  *
 878  * The "path" out-parameter will give the path of the object we found (if any).
 879  * Note that it may point to static storage and is only valid until another
 880  * call to stat_sha1_file().
 881  */
 882 static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
 883                           struct stat *st, const char **path)
 884 {
 885         struct object_directory *odb;
 886         static struct strbuf buf = STRBUF_INIT;
 887
 888         prepare_alt_odb(r);
 889         for (odb = r->objects->odb; odb; odb = odb->next) {
 890                 *path = odb_loose_path(odb, &buf, sha1);
 891                 if (!lstat(*path, st))
 892                         return 0;
 893         }
 894
 895         return -1;
 896 }
 897
 898 /*
 899  * Like stat_sha1_file(), but actually open the object and return the
 900  * descriptor. See the caveats on the "path" parameter above.
 901  */
 902 static int open_sha1_file(struct repository *r,
 903                           const unsigned char *sha1, const char **path)
 904 {
 905         int fd;
 906         struct object_directory *odb;
 907         int most_interesting_errno = ENOENT;
 908         static struct strbuf buf = STRBUF_INIT;
 909
 910         prepare_alt_odb(r);
 911         for (odb = r->objects->odb; odb; odb = odb->next) {
 912                 *path = odb_loose_path(odb, &buf, sha1);
 913                 fd = git_open(*path);
 914                 if (fd >= 0)
 915                         return fd;
 916
 917                 if (most_interesting_errno == ENOENT)
 918                         most_interesting_errno = errno;
 919         }
 920         errno = most_interesting_errno;
 921         return -1;
 922 }
 923
 924 /*
 925  * Map the loose object at "path" if it is not NULL, or the path found by
 926  * searching for a loose object named "sha1".
 927  */
 928 static void *map_sha1_file_1(struct repository *r, const char *path,
 929                              const unsigned char *sha1, unsigned long *size)
 930 {
 931         void *map;
 932         int fd;
 933
 934         if (path)
 935                 fd = git_open(path);
 936         else
 937                 fd = open_sha1_file(r, sha1, &path);
 938         map = NULL;
 939         if (fd >= 0) {
 940                 struct stat st;
 941
 942                 if (!fstat(fd, &st)) {
 943                         *size = xsize_t(st.st_size);
 944                         if (!*size) {
 945                                 /* mmap() is forbidden on empty files */
 946                                 error(_("object file %s is empty"), path);
 947                                 return NULL;
 948                         }
 949                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 950                 }
 951                 close(fd);
 952         }
 953         return map;
 954 }
 955
 956 void *map_sha1_file(struct repository *r,
 957                     const unsigned char *sha1, unsigned long *size)
 958 {
 959         return map_sha1_file_1(r, NULL, sha1, size);
 960 }
 961
 962 static int unpack_sha1_short_header(git_zstream *stream,
 963                                     unsigned char *map, unsigned long mapsize,
 964                                     void *buffer, unsigned long bufsiz)
 965 {
 966         /* Get the data stream */
 967         memset(stream, 0, sizeof(*stream));
 968         stream->next_in = map;
 969         stream->avail_in = mapsize;
 970         stream->next_out = buffer;
 971         stream->avail_out = bufsiz;
 972
 973         git_inflate_init(stream);
 974         return git_inflate(stream, 0);
 975 }
 976
 977 int unpack_sha1_header(git_zstream *stream,
 978                        unsigned char *map, unsigned long mapsize,
 979                        void *buffer, unsigned long bufsiz)
 980 {
 981         int status = unpack_sha1_short_header(stream, map, mapsize,
 982                                               buffer, bufsiz);
 983
 984         if (status < Z_OK)
 985                 return status;
 986
 987         /* Make sure we have the terminating NUL */
 988         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 989                 return -1;
 990         return 0;
 991 }
 992
 993 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
 994                                         unsigned long mapsize, void *buffer,
 995                                         unsigned long bufsiz, struct strbuf *header)
 996 {
 997         int status;
 998
 999         status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1000         if (status < Z_OK)
1001                 return -1;
1002
1003         /*
1004          * Check if entire header is unpacked in the first iteration.
1005          */
1006         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1007                 return 0;
1008
1009         /*
1010          * buffer[0..bufsiz] was not large enough.  Copy the partial
1011          * result out to header, and then append the result of further
1012          * reading the stream.
1013          */
1014         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1015         stream->next_out = buffer;
1016         stream->avail_out = bufsiz;
1017
1018         do {
1019                 status = git_inflate(stream, 0);
1020                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1021                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1022                         return 0;
1023                 stream->next_out = buffer;
1024                 stream->avail_out = bufsiz;
1025         } while (status != Z_STREAM_END);
1026         return -1;
1027 }
1028
1029 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1030 {
1031         int bytes = strlen(buffer) + 1;
1032         unsigned char *buf = xmallocz(size);
1033         unsigned long n;
1034         int status = Z_OK;
1035
1036         n = stream->total_out - bytes;
1037         if (n > size)
1038                 n = size;
1039         memcpy(buf, (char *) buffer + bytes, n);
1040         bytes = n;
1041         if (bytes <= size) {
1042                 /*
1043                  * The above condition must be (bytes <= size), not
1044                  * (bytes < size).  In other words, even though we
1045                  * expect no more output and set avail_out to zero,
1046                  * the input zlib stream may have bytes that express
1047                  * "this concludes the stream", and we *do* want to
1048                  * eat that input.
1049                  *
1050                  * Otherwise we would not be able to test that we
1051                  * consumed all the input to reach the expected size;
1052                  * we also want to check that zlib tells us that all
1053                  * went well with status == Z_STREAM_END at the end.
1054                  */
1055                 stream->next_out = buf + bytes;
1056                 stream->avail_out = size - bytes;
1057                 while (status == Z_OK)
1058                         status = git_inflate(stream, Z_FINISH);
1059         }
1060         if (status == Z_STREAM_END && !stream->avail_in) {
1061                 git_inflate_end(stream);
1062                 return buf;
1063         }
1064
1065         if (status < 0)
1066                 error(_("corrupt loose object '%s'"), sha1_to_hex(sha1));
1067         else if (stream->avail_in)
1068                 error(_("garbage at end of loose object '%s'"),
1069                       sha1_to_hex(sha1));
1070         free(buf);
1071         return NULL;
1072 }
1073
1074 /*
1075  * We used to just use "sscanf()", but that's actually way
1076  * too permissive for what we want to check. So do an anal
1077  * object header parse by hand.
1078  */
1079 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1080                                unsigned int flags)
1081 {
1082         const char *type_buf = hdr;
1083         unsigned long size;
1084         int type, type_len = 0;
1085
1086         /*
1087          * The type can be of any size but is followed by
1088          * a space.
1089          */
1090         for (;;) {
1091                 char c = *hdr++;
1092                 if (!c)
1093                         return -1;
1094                 if (c == ' ')
1095                         break;
1096                 type_len++;
1097         }
1098
1099         type = type_from_string_gently(type_buf, type_len, 1);
1100         if (oi->type_name)
1101                 strbuf_add(oi->type_name, type_buf, type_len);
1102         /*
1103          * Set type to 0 if its an unknown object and
1104          * we're obtaining the type using '--allow-unknown-type'
1105          * option.
1106          */
1107         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1108                 type = 0;
1109         else if (type < 0)
1110                 die(_("invalid object type"));
1111         if (oi->typep)
1112                 *oi->typep = type;
1113
1114         /*
1115          * The length must follow immediately, and be in canonical
1116          * decimal format (ie "010" is not valid).
1117          */
1118         size = *hdr++ - '0';
1119         if (size > 9)
1120                 return -1;
1121         if (size) {
1122                 for (;;) {
1123                         unsigned long c = *hdr - '0';
1124                         if (c > 9)
1125                                 break;
1126                         hdr++;
1127                         size = size * 10 + c;
1128                 }
1129         }
1130
1131         if (oi->sizep)
1132                 *oi->sizep = size;
1133
1134         /*
1135          * The length must be followed by a zero byte
1136          */
1137         return *hdr ? -1 : type;
1138 }
1139
1140 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1141 {
1142         struct object_info oi = OBJECT_INFO_INIT;
1143
1144         oi.sizep = sizep;
1145         return parse_sha1_header_extended(hdr, &oi, 0);
1146 }
1147
1148 static int sha1_loose_object_info(struct repository *r,
1149                                   const unsigned char *sha1,
1150                                   struct object_info *oi, int flags)
1151 {
1152         int status = 0;
1153         unsigned long mapsize;
1154         void *map;
1155         git_zstream stream;
1156         char hdr[MAX_HEADER_LEN];
1157         struct strbuf hdrbuf = STRBUF_INIT;
1158         unsigned long size_scratch;
1159
1160         if (oi->delta_base_sha1)
1161                 hashclr(oi->delta_base_sha1);
1162
1163         /*
1164          * If we don't care about type or size, then we don't
1165          * need to look inside the object at all. Note that we
1166          * do not optimize out the stat call, even if the
1167          * caller doesn't care about the disk-size, since our
1168          * return value implicitly indicates whether the
1169          * object even exists.
1170          */
1171         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1172                 const char *path;
1173                 struct stat st;
1174                 if (stat_sha1_file(r, sha1, &st, &path) < 0)
1175                         return -1;
1176                 if (oi->disk_sizep)
1177                         *oi->disk_sizep = st.st_size;
1178                 return 0;
1179         }
1180
1181         map = map_sha1_file(r, sha1, &mapsize);
1182         if (!map)
1183                 return -1;
1184
1185         if (!oi->sizep)
1186                 oi->sizep = &size_scratch;
1187
1188         if (oi->disk_sizep)
1189                 *oi->disk_sizep = mapsize;
1190         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1191                 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1192                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1193                                        sha1_to_hex(sha1));
1194         } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1195                 status = error(_("unable to unpack %s header"),
1196                                sha1_to_hex(sha1));
1197         if (status < 0)
1198                 ; /* Do nothing */
1199         else if (hdrbuf.len) {
1200                 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1201                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1202                                        sha1_to_hex(sha1));
1203         } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1204                 status = error(_("unable to parse %s header"), sha1_to_hex(sha1));
1205
1206         if (status >= 0 && oi->contentp) {
1207                 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1208                                                  *oi->sizep, sha1);
1209                 if (!*oi->contentp) {
1210                         git_inflate_end(&stream);
1211                         status = -1;
1212                 }
1213         } else
1214                 git_inflate_end(&stream);
1215
1216         munmap(map, mapsize);
1217         if (status && oi->typep)
1218                 *oi->typep = status;
1219         if (oi->sizep == &size_scratch)
1220                 oi->sizep = NULL;
1221         strbuf_release(&hdrbuf);
1222         oi->whence = OI_LOOSE;
1223         return (status < 0) ? status : 0;
1224 }
1225
1226 int fetch_if_missing = 1;
1227
1228 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1229                              struct object_info *oi, unsigned flags)
1230 {
1231         static struct object_info blank_oi = OBJECT_INFO_INIT;
1232         struct pack_entry e;
1233         int rtype;
1234         const struct object_id *real = oid;
1235         int already_retried = 0;
1236
1237         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1238                 real = lookup_replace_object(r, oid);
1239
1240         if (is_null_oid(real))
1241                 return -1;
1242
1243         if (!oi)
1244                 oi = &blank_oi;
1245
1246         if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1247                 struct cached_object *co = find_cached_object(real);
1248                 if (co) {
1249                         if (oi->typep)
1250                                 *(oi->typep) = co->type;
1251                         if (oi->sizep)
1252                                 *(oi->sizep) = co->size;
1253                         if (oi->disk_sizep)
1254                                 *(oi->disk_sizep) = 0;
1255                         if (oi->delta_base_sha1)
1256                                 hashclr(oi->delta_base_sha1);
1257                         if (oi->type_name)
1258                                 strbuf_addstr(oi->type_name, type_name(co->type));
1259                         if (oi->contentp)
1260                                 *oi->contentp = xmemdupz(co->buf, co->size);
1261                         oi->whence = OI_CACHED;
1262                         return 0;
1263                 }
1264         }
1265
1266         while (1) {
1267                 if (find_pack_entry(r, real, &e))
1268                         break;
1269
1270                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1271                         return -1;
1272
1273                 /* Most likely it's a loose object. */
1274                 if (!sha1_loose_object_info(r, real->hash, oi, flags))
1275                         return 0;
1276
1277                 /* Not a loose object; someone else may have just packed it. */
1278                 if (!(flags & OBJECT_INFO_QUICK)) {
1279                         reprepare_packed_git(r);
1280                         if (find_pack_entry(r, real, &e))
1281                                 break;
1282                 }
1283
1284                 /* Check if it is a missing object */
1285                 if (fetch_if_missing && repository_format_partial_clone &&
1286                     !already_retried && r == the_repository) {
1287                         /*
1288                          * TODO Investigate having fetch_object() return
1289                          * TODO error/success and stopping the music here.
1290                          * TODO Pass a repository struct through fetch_object,
1291                          * such that arbitrary repositories work.
1292                          */
1293                         fetch_objects(repository_format_partial_clone, real, 1);
1294                         already_retried = 1;
1295                         continue;
1296                 }
1297
1298                 return -1;
1299         }
1300
1301         if (oi == &blank_oi)
1302                 /*
1303                  * We know that the caller doesn't actually need the
1304                  * information below, so return early.
1305                  */
1306                 return 0;
1307         rtype = packed_object_info(r, e.p, e.offset, oi);
1308         if (rtype < 0) {
1309                 mark_bad_packed_object(e.p, real->hash);
1310                 return oid_object_info_extended(r, real, oi, 0);
1311         } else if (oi->whence == OI_PACKED) {
1312                 oi->u.packed.offset = e.offset;
1313                 oi->u.packed.pack = e.p;
1314                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1315                                          rtype == OBJ_OFS_DELTA);
1316         }
1317
1318         return 0;
1319 }
1320
1321 /* returns enum object_type or negative */
1322 int oid_object_info(struct repository *r,
1323                     const struct object_id *oid,
1324                     unsigned long *sizep)
1325 {
1326         enum object_type type;
1327         struct object_info oi = OBJECT_INFO_INIT;
1328
1329         oi.typep = &type;
1330         oi.sizep = sizep;
1331         if (oid_object_info_extended(r, oid, &oi,
1332                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1333                 return -1;
1334         return type;
1335 }
1336
1337 static void *read_object(const unsigned char *sha1, enum object_type *type,
1338                          unsigned long *size)
1339 {
1340         struct object_id oid;
1341         struct object_info oi = OBJECT_INFO_INIT;
1342         void *content;
1343         oi.typep = type;
1344         oi.sizep = size;
1345         oi.contentp = &content;
1346
1347         hashcpy(oid.hash, sha1);
1348
1349         if (oid_object_info_extended(the_repository, &oid, &oi, 0) < 0)
1350                 return NULL;
1351         return content;
1352 }
1353
1354 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1355                         struct object_id *oid)
1356 {
1357         struct cached_object *co;
1358
1359         hash_object_file(buf, len, type_name(type), oid);
1360         if (has_sha1_file(oid->hash) || find_cached_object(oid))
1361                 return 0;
1362         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1363         co = &cached_objects[cached_object_nr++];
1364         co->size = len;
1365         co->type = type;
1366         co->buf = xmalloc(len);
1367         memcpy(co->buf, buf, len);
1368         oidcpy(&co->oid, oid);
1369         return 0;
1370 }
1371
1372 /*
1373  * This function dies on corrupt objects; the callers who want to
1374  * deal with them should arrange to call read_object() and give error
1375  * messages themselves.
1376  */
1377 void *read_object_file_extended(const struct object_id *oid,
1378                                 enum object_type *type,
1379                                 unsigned long *size,
1380                                 int lookup_replace)
1381 {
1382         void *data;
1383         const struct packed_git *p;
1384         const char *path;
1385         struct stat st;
1386         const struct object_id *repl = lookup_replace ?
1387                 lookup_replace_object(the_repository, oid) : oid;
1388
1389         errno = 0;
1390         data = read_object(repl->hash, type, size);
1391         if (data)
1392                 return data;
1393
1394         if (errno && errno != ENOENT)
1395                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1396
1397         /* die if we replaced an object with one that does not exist */
1398         if (repl != oid)
1399                 die(_("replacement %s not found for %s"),
1400                     oid_to_hex(repl), oid_to_hex(oid));
1401
1402         if (!stat_sha1_file(the_repository, repl->hash, &st, &path))
1403                 die(_("loose object %s (stored in %s) is corrupt"),
1404                     oid_to_hex(repl), path);
1405
1406         if ((p = has_packed_and_bad(repl->hash)) != NULL)
1407                 die(_("packed object %s (stored in %s) is corrupt"),
1408                     oid_to_hex(repl), p->pack_name);
1409
1410         return NULL;
1411 }
1412
1413 void *read_object_with_reference(const struct object_id *oid,
1414                                  const char *required_type_name,
1415                                  unsigned long *size,
1416                                  struct object_id *actual_oid_return)
1417 {
1418         enum object_type type, required_type;
1419         void *buffer;
1420         unsigned long isize;
1421         struct object_id actual_oid;
1422
1423         required_type = type_from_string(required_type_name);
1424         oidcpy(&actual_oid, oid);
1425         while (1) {
1426                 int ref_length = -1;
1427                 const char *ref_type = NULL;
1428
1429                 buffer = read_object_file(&actual_oid, &type, &isize);
1430                 if (!buffer)
1431                         return NULL;
1432                 if (type == required_type) {
1433                         *size = isize;
1434                         if (actual_oid_return)
1435                                 oidcpy(actual_oid_return, &actual_oid);
1436                         return buffer;
1437                 }
1438                 /* Handle references */
1439                 else if (type == OBJ_COMMIT)
1440                         ref_type = "tree ";
1441                 else if (type == OBJ_TAG)
1442                         ref_type = "object ";
1443                 else {
1444                         free(buffer);
1445                         return NULL;
1446                 }
1447                 ref_length = strlen(ref_type);
1448
1449                 if (ref_length + the_hash_algo->hexsz > isize ||
1450                     memcmp(buffer, ref_type, ref_length) ||
1451                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1452                         free(buffer);
1453                         return NULL;
1454                 }
1455                 free(buffer);
1456                 /* Now we have the ID of the referred-to object in
1457                  * actual_oid.  Check again. */
1458         }
1459 }
1460
1461 static void write_object_file_prepare(const void *buf, unsigned long len,
1462                                       const char *type, struct object_id *oid,
1463                                       char *hdr, int *hdrlen)
1464 {
1465         git_hash_ctx c;
1466
1467         /* Generate the header */
1468         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1469
1470         /* Sha1.. */
1471         the_hash_algo->init_fn(&c);
1472         the_hash_algo->update_fn(&c, hdr, *hdrlen);
1473         the_hash_algo->update_fn(&c, buf, len);
1474         the_hash_algo->final_fn(oid->hash, &c);
1475 }
1476
1477 /*
1478  * Move the just written object into its final resting place.
1479  */
1480 int finalize_object_file(const char *tmpfile, const char *filename)
1481 {
1482         int ret = 0;
1483
1484         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1485                 goto try_rename;
1486         else if (link(tmpfile, filename))
1487                 ret = errno;
1488
1489         /*
1490          * Coda hack - coda doesn't like cross-directory links,
1491          * so we fall back to a rename, which will mean that it
1492          * won't be able to check collisions, but that's not a
1493          * big deal.
1494          *
1495          * The same holds for FAT formatted media.
1496          *
1497          * When this succeeds, we just return.  We have nothing
1498          * left to unlink.
1499          */
1500         if (ret && ret != EEXIST) {
1501         try_rename:
1502                 if (!rename(tmpfile, filename))
1503                         goto out;
1504                 ret = errno;
1505         }
1506         unlink_or_warn(tmpfile);
1507         if (ret) {
1508                 if (ret != EEXIST) {
1509                         return error_errno(_("unable to write sha1 filename %s"), filename);
1510                 }
1511                 /* FIXME!!! Collision check here ? */
1512         }
1513
1514 out:
1515         if (adjust_shared_perm(filename))
1516                 return error(_("unable to set permission to '%s'"), filename);
1517         return 0;
1518 }
1519
1520 static int write_buffer(int fd, const void *buf, size_t len)
1521 {
1522         if (write_in_full(fd, buf, len) < 0)
1523                 return error_errno(_("file write error"));
1524         return 0;
1525 }
1526
1527 int hash_object_file(const void *buf, unsigned long len, const char *type,
1528                      struct object_id *oid)
1529 {
1530         char hdr[MAX_HEADER_LEN];
1531         int hdrlen = sizeof(hdr);
1532         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1533         return 0;
1534 }
1535
1536 /* Finalize a file on disk, and close it. */
1537 static void close_sha1_file(int fd)
1538 {
1539         if (fsync_object_files)
1540                 fsync_or_die(fd, "sha1 file");
1541         if (close(fd) != 0)
1542                 die_errno(_("error when closing sha1 file"));
1543 }
1544
1545 /* Size of directory component, including the ending '/' */
1546 static inline int directory_size(const char *filename)
1547 {
1548         const char *s = strrchr(filename, '/');
1549         if (!s)
1550                 return 0;
1551         return s - filename + 1;
1552 }
1553
1554 /*
1555  * This creates a temporary file in the same directory as the final
1556  * 'filename'
1557  *
1558  * We want to avoid cross-directory filename renames, because those
1559  * can have problems on various filesystems (FAT, NFS, Coda).
1560  */
1561 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1562 {
1563         int fd, dirlen = directory_size(filename);
1564
1565         strbuf_reset(tmp);
1566         strbuf_add(tmp, filename, dirlen);
1567         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1568         fd = git_mkstemp_mode(tmp->buf, 0444);
1569         if (fd < 0 && dirlen && errno == ENOENT) {
1570                 /*
1571                  * Make sure the directory exists; note that the contents
1572                  * of the buffer are undefined after mkstemp returns an
1573                  * error, so we have to rewrite the whole buffer from
1574                  * scratch.
1575                  */
1576                 strbuf_reset(tmp);
1577                 strbuf_add(tmp, filename, dirlen - 1);
1578                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1579                         return -1;
1580                 if (adjust_shared_perm(tmp->buf))
1581                         return -1;
1582
1583                 /* Try again */
1584                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1585                 fd = git_mkstemp_mode(tmp->buf, 0444);
1586         }
1587         return fd;
1588 }
1589
1590 static int write_loose_object(const struct object_id *oid, char *hdr,
1591                               int hdrlen, const void *buf, unsigned long len,
1592                               time_t mtime)
1593 {
1594         int fd, ret;
1595         unsigned char compressed[4096];
1596         git_zstream stream;
1597         git_hash_ctx c;
1598         struct object_id parano_oid;
1599         static struct strbuf tmp_file = STRBUF_INIT;
1600         static struct strbuf filename = STRBUF_INIT;
1601
1602         loose_object_path(the_repository, &filename, oid->hash);
1603
1604         fd = create_tmpfile(&tmp_file, filename.buf);
1605         if (fd < 0) {
1606                 if (errno == EACCES)
1607                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1608                 else
1609                         return error_errno(_("unable to create temporary file"));
1610         }
1611
1612         /* Set it up */
1613         git_deflate_init(&stream, zlib_compression_level);
1614         stream.next_out = compressed;
1615         stream.avail_out = sizeof(compressed);
1616         the_hash_algo->init_fn(&c);
1617
1618         /* First header.. */
1619         stream.next_in = (unsigned char *)hdr;
1620         stream.avail_in = hdrlen;
1621         while (git_deflate(&stream, 0) == Z_OK)
1622                 ; /* nothing */
1623         the_hash_algo->update_fn(&c, hdr, hdrlen);
1624
1625         /* Then the data itself.. */
1626         stream.next_in = (void *)buf;
1627         stream.avail_in = len;
1628         do {
1629                 unsigned char *in0 = stream.next_in;
1630                 ret = git_deflate(&stream, Z_FINISH);
1631                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1632                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1633                         die(_("unable to write sha1 file"));
1634                 stream.next_out = compressed;
1635                 stream.avail_out = sizeof(compressed);
1636         } while (ret == Z_OK);
1637
1638         if (ret != Z_STREAM_END)
1639                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1640                     ret);
1641         ret = git_deflate_end_gently(&stream);
1642         if (ret != Z_OK)
1643                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1644                     ret);
1645         the_hash_algo->final_fn(parano_oid.hash, &c);
1646         if (!oideq(oid, &parano_oid))
1647                 die(_("confused by unstable object source data for %s"),
1648                     oid_to_hex(oid));
1649
1650         close_sha1_file(fd);
1651
1652         if (mtime) {
1653                 struct utimbuf utb;
1654                 utb.actime = mtime;
1655                 utb.modtime = mtime;
1656                 if (utime(tmp_file.buf, &utb) < 0)
1657                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1658         }
1659
1660         return finalize_object_file(tmp_file.buf, filename.buf);
1661 }
1662
1663 static int freshen_loose_object(const struct object_id *oid)
1664 {
1665         return check_and_freshen(oid, 1);
1666 }
1667
1668 static int freshen_packed_object(const struct object_id *oid)
1669 {
1670         struct pack_entry e;
1671         if (!find_pack_entry(the_repository, oid, &e))
1672                 return 0;
1673         if (e.p->freshened)
1674                 return 1;
1675         if (!freshen_file(e.p->pack_name))
1676                 return 0;
1677         e.p->freshened = 1;
1678         return 1;
1679 }
1680
1681 int write_object_file(const void *buf, unsigned long len, const char *type,
1682                       struct object_id *oid)
1683 {
1684         char hdr[MAX_HEADER_LEN];
1685         int hdrlen = sizeof(hdr);
1686
1687         /* Normally if we have it in the pack then we do not bother writing
1688          * it out into .git/objects/??/?{38} file.
1689          */
1690         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1691         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1692                 return 0;
1693         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1694 }
1695
1696 int hash_object_file_literally(const void *buf, unsigned long len,
1697                                const char *type, struct object_id *oid,
1698                                unsigned flags)
1699 {
1700         char *header;
1701         int hdrlen, status = 0;
1702
1703         /* type string, SP, %lu of the length plus NUL must fit this */
1704         hdrlen = strlen(type) + MAX_HEADER_LEN;
1705         header = xmalloc(hdrlen);
1706         write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1707
1708         if (!(flags & HASH_WRITE_OBJECT))
1709                 goto cleanup;
1710         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1711                 goto cleanup;
1712         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1713
1714 cleanup:
1715         free(header);
1716         return status;
1717 }
1718
1719 int force_object_loose(const struct object_id *oid, time_t mtime)
1720 {
1721         void *buf;
1722         unsigned long len;
1723         enum object_type type;
1724         char hdr[MAX_HEADER_LEN];
1725         int hdrlen;
1726         int ret;
1727
1728         if (has_loose_object(oid))
1729                 return 0;
1730         buf = read_object(oid->hash, &type, &len);
1731         if (!buf)
1732                 return error(_("cannot read sha1_file for %s"), oid_to_hex(oid));
1733         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1734         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1735         free(buf);
1736
1737         return ret;
1738 }
1739
1740 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1741 {
1742         struct object_id oid;
1743         if (!startup_info->have_repository)
1744                 return 0;
1745         hashcpy(oid.hash, sha1);
1746         return oid_object_info_extended(the_repository, &oid, NULL,
1747                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1748 }
1749
1750 int has_object_file(const struct object_id *oid)
1751 {
1752         return has_sha1_file(oid->hash);
1753 }
1754
1755 int has_object_file_with_flags(const struct object_id *oid, int flags)
1756 {
1757         return has_sha1_file_with_flags(oid->hash, flags);
1758 }
1759
1760 static void check_tree(const void *buf, size_t size)
1761 {
1762         struct tree_desc desc;
1763         struct name_entry entry;
1764
1765         init_tree_desc(&desc, buf, size);
1766         while (tree_entry(&desc, &entry))
1767                 /* do nothing
1768                  * tree_entry() will die() on malformed entries */
1769                 ;
1770 }
1771
1772 static void check_commit(const void *buf, size_t size)
1773 {
1774         struct commit c;
1775         memset(&c, 0, sizeof(c));
1776         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1777                 die(_("corrupt commit"));
1778 }
1779
1780 static void check_tag(const void *buf, size_t size)
1781 {
1782         struct tag t;
1783         memset(&t, 0, sizeof(t));
1784         if (parse_tag_buffer(the_repository, &t, buf, size))
1785                 die(_("corrupt tag"));
1786 }
1787
1788 static int index_mem(struct index_state *istate,
1789                      struct object_id *oid, void *buf, size_t size,
1790                      enum object_type type,
1791                      const char *path, unsigned flags)
1792 {
1793         int ret, re_allocated = 0;
1794         int write_object = flags & HASH_WRITE_OBJECT;
1795
1796         if (!type)
1797                 type = OBJ_BLOB;
1798
1799         /*
1800          * Convert blobs to git internal format
1801          */
1802         if ((type == OBJ_BLOB) && path) {
1803                 struct strbuf nbuf = STRBUF_INIT;
1804                 if (convert_to_git(istate, path, buf, size, &nbuf,
1805                                    get_conv_flags(flags))) {
1806                         buf = strbuf_detach(&nbuf, &size);
1807                         re_allocated = 1;
1808                 }
1809         }
1810         if (flags & HASH_FORMAT_CHECK) {
1811                 if (type == OBJ_TREE)
1812                         check_tree(buf, size);
1813                 if (type == OBJ_COMMIT)
1814                         check_commit(buf, size);
1815                 if (type == OBJ_TAG)
1816                         check_tag(buf, size);
1817         }
1818
1819         if (write_object)
1820                 ret = write_object_file(buf, size, type_name(type), oid);
1821         else
1822                 ret = hash_object_file(buf, size, type_name(type), oid);
1823         if (re_allocated)
1824                 free(buf);
1825         return ret;
1826 }
1827
1828 static int index_stream_convert_blob(struct index_state *istate,
1829                                      struct object_id *oid,
1830                                      int fd,
1831                                      const char *path,
1832                                      unsigned flags)
1833 {
1834         int ret;
1835         const int write_object = flags & HASH_WRITE_OBJECT;
1836         struct strbuf sbuf = STRBUF_INIT;
1837
1838         assert(path);
1839         assert(would_convert_to_git_filter_fd(istate, path));
1840
1841         convert_to_git_filter_fd(istate, path, fd, &sbuf,
1842                                  get_conv_flags(flags));
1843
1844         if (write_object)
1845                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1846                                         oid);
1847         else
1848                 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1849                                        oid);
1850         strbuf_release(&sbuf);
1851         return ret;
1852 }
1853
1854 static int index_pipe(struct index_state *istate, struct object_id *oid,
1855                       int fd, enum object_type type,
1856                       const char *path, unsigned flags)
1857 {
1858         struct strbuf sbuf = STRBUF_INIT;
1859         int ret;
1860
1861         if (strbuf_read(&sbuf, fd, 4096) >= 0)
1862                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
1863         else
1864                 ret = -1;
1865         strbuf_release(&sbuf);
1866         return ret;
1867 }
1868
1869 #define SMALL_FILE_SIZE (32*1024)
1870
1871 static int index_core(struct index_state *istate,
1872                       struct object_id *oid, int fd, size_t size,
1873                       enum object_type type, const char *path,
1874                       unsigned flags)
1875 {
1876         int ret;
1877
1878         if (!size) {
1879                 ret = index_mem(istate, oid, "", size, type, path, flags);
1880         } else if (size <= SMALL_FILE_SIZE) {
1881                 char *buf = xmalloc(size);
1882                 ssize_t read_result = read_in_full(fd, buf, size);
1883                 if (read_result < 0)
1884                         ret = error_errno(_("read error while indexing %s"),
1885                                           path ? path : "<unknown>");
1886                 else if (read_result != size)
1887                         ret = error(_("short read while indexing %s"),
1888                                     path ? path : "<unknown>");
1889                 else
1890                         ret = index_mem(istate, oid, buf, size, type, path, flags);
1891                 free(buf);
1892         } else {
1893                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1894                 ret = index_mem(istate, oid, buf, size, type, path, flags);
1895                 munmap(buf, size);
1896         }
1897         return ret;
1898 }
1899
1900 /*
1901  * This creates one packfile per large blob unless bulk-checkin
1902  * machinery is "plugged".
1903  *
1904  * This also bypasses the usual "convert-to-git" dance, and that is on
1905  * purpose. We could write a streaming version of the converting
1906  * functions and insert that before feeding the data to fast-import
1907  * (or equivalent in-core API described above). However, that is
1908  * somewhat complicated, as we do not know the size of the filter
1909  * result, which we need to know beforehand when writing a git object.
1910  * Since the primary motivation for trying to stream from the working
1911  * tree file and to avoid mmaping it in core is to deal with large
1912  * binary blobs, they generally do not want to get any conversion, and
1913  * callers should avoid this code path when filters are requested.
1914  */
1915 static int index_stream(struct object_id *oid, int fd, size_t size,
1916                         enum object_type type, const char *path,
1917                         unsigned flags)
1918 {
1919         return index_bulk_checkin(oid, fd, size, type, path, flags);
1920 }
1921
1922 int index_fd(struct index_state *istate, struct object_id *oid,
1923              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(istate, path))
1933                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
1934         else if (!S_ISREG(st->st_mode))
1935                 ret = index_pipe(istate, oid, fd, type, path, flags);
1936         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1937                  (path && would_convert_to_git(istate, path)))
1938                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
1939                                  type, path, 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 index_state *istate, struct object_id *oid,
1948                const char *path, struct stat *st, unsigned flags)
1949 {
1950         int fd;
1951         struct strbuf sb = STRBUF_INIT;
1952         int rc = 0;
1953
1954         switch (st->st_mode & S_IFMT) {
1955         case S_IFREG:
1956                 fd = open(path, O_RDONLY);
1957                 if (fd < 0)
1958                         return error_errno("open(\"%s\")", path);
1959                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
1960                         return error(_("%s: failed to insert into database"),
1961                                      path);
1962                 break;
1963         case S_IFLNK:
1964                 if (strbuf_readlink(&sb, path, st->st_size))
1965                         return error_errno("readlink(\"%s\")", path);
1966                 if (!(flags & HASH_WRITE_OBJECT))
1967                         hash_object_file(sb.buf, sb.len, blob_type, oid);
1968                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1969                         rc = error(_("%s: failed to insert into database"), path);
1970                 strbuf_release(&sb);
1971                 break;
1972         case S_IFDIR:
1973                 return resolve_gitlink_ref(path, "HEAD", oid);
1974         default:
1975                 return error(_("%s: unsupported file type"), path);
1976         }
1977         return rc;
1978 }
1979
1980 int read_pack_header(int fd, struct pack_header *header)
1981 {
1982         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1983                 /* "eof before pack header was fully read" */
1984                 return PH_ERROR_EOF;
1985
1986         if (header->hdr_signature != htonl(PACK_SIGNATURE))
1987                 /* "protocol error (pack signature mismatch detected)" */
1988                 return PH_ERROR_PACK_SIGNATURE;
1989         if (!pack_version_ok(header->hdr_version))
1990                 /* "protocol error (pack version unsupported)" */
1991                 return PH_ERROR_PROTOCOL;
1992         return 0;
1993 }
1994
1995 void assert_oid_type(const struct object_id *oid, enum object_type expect)
1996 {
1997         enum object_type type = oid_object_info(the_repository, oid, NULL);
1998         if (type < 0)
1999                 die(_("%s is not a valid object"), oid_to_hex(oid));
2000         if (type != expect)
2001                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2002                     type_name(expect));
2003 }
2004
2005 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2006                                 struct strbuf *path,
2007                                 each_loose_object_fn obj_cb,
2008                                 each_loose_cruft_fn cruft_cb,
2009                                 each_loose_subdir_fn subdir_cb,
2010                                 void *data)
2011 {
2012         size_t origlen, baselen;
2013         DIR *dir;
2014         struct dirent *de;
2015         int r = 0;
2016         struct object_id oid;
2017
2018         if (subdir_nr > 0xff)
2019                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2020
2021         origlen = path->len;
2022         strbuf_complete(path, '/');
2023         strbuf_addf(path, "%02x", subdir_nr);
2024
2025         dir = opendir(path->buf);
2026         if (!dir) {
2027                 if (errno != ENOENT)
2028                         r = error_errno(_("unable to open %s"), path->buf);
2029                 strbuf_setlen(path, origlen);
2030                 return r;
2031         }
2032
2033         oid.hash[0] = subdir_nr;
2034         strbuf_addch(path, '/');
2035         baselen = path->len;
2036
2037         while ((de = readdir(dir))) {
2038                 size_t namelen;
2039                 if (is_dot_or_dotdot(de->d_name))
2040                         continue;
2041
2042                 namelen = strlen(de->d_name);
2043                 strbuf_setlen(path, baselen);
2044                 strbuf_add(path, de->d_name, namelen);
2045                 if (namelen == the_hash_algo->hexsz - 2 &&
2046                     !hex_to_bytes(oid.hash + 1, de->d_name,
2047                                   the_hash_algo->rawsz - 1)) {
2048                         if (obj_cb) {
2049                                 r = obj_cb(&oid, path->buf, data);
2050                                 if (r)
2051                                         break;
2052                         }
2053                         continue;
2054                 }
2055
2056                 if (cruft_cb) {
2057                         r = cruft_cb(de->d_name, path->buf, data);
2058                         if (r)
2059                                 break;
2060                 }
2061         }
2062         closedir(dir);
2063
2064         strbuf_setlen(path, baselen - 1);
2065         if (!r && subdir_cb)
2066                 r = subdir_cb(subdir_nr, path->buf, data);
2067
2068         strbuf_setlen(path, origlen);
2069
2070         return r;
2071 }
2072
2073 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2074                             each_loose_object_fn obj_cb,
2075                             each_loose_cruft_fn cruft_cb,
2076                             each_loose_subdir_fn subdir_cb,
2077                             void *data)
2078 {
2079         int r = 0;
2080         int i;
2081
2082         for (i = 0; i < 256; i++) {
2083                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2084                                                 subdir_cb, data);
2085                 if (r)
2086                         break;
2087         }
2088
2089         return r;
2090 }
2091
2092 int for_each_loose_file_in_objdir(const char *path,
2093                                   each_loose_object_fn obj_cb,
2094                                   each_loose_cruft_fn cruft_cb,
2095                                   each_loose_subdir_fn subdir_cb,
2096                                   void *data)
2097 {
2098         struct strbuf buf = STRBUF_INIT;
2099         int r;
2100
2101         strbuf_addstr(&buf, path);
2102         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2103                                               subdir_cb, data);
2104         strbuf_release(&buf);
2105
2106         return r;
2107 }
2108
2109 int for_each_loose_object(each_loose_object_fn cb, void *data,
2110                           enum for_each_object_flags flags)
2111 {
2112         struct object_directory *odb;
2113
2114         prepare_alt_odb(the_repository);
2115         for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2116                 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2117                                                       NULL, data);
2118                 if (r)
2119                         return r;
2120
2121                 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2122                         break;
2123         }
2124
2125         return 0;
2126 }
2127
2128 static int append_loose_object(const struct object_id *oid, const char *path,
2129                                void *data)
2130 {
2131         oid_array_append(data, oid);
2132         return 0;
2133 }
2134
2135 void odb_load_loose_cache(struct object_directory *odb, int subdir_nr)
2136 {
2137         struct strbuf buf = STRBUF_INIT;
2138
2139         if (subdir_nr < 0 ||
2140             subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2141                 BUG("subdir_nr out of range");
2142
2143         if (odb->loose_objects_subdir_seen[subdir_nr])
2144                 return;
2145
2146         strbuf_addstr(&buf, odb->path);
2147         for_each_file_in_obj_subdir(subdir_nr, &buf,
2148                                     append_loose_object,
2149                                     NULL, NULL,
2150                                     &odb->loose_objects_cache);
2151         odb->loose_objects_subdir_seen[subdir_nr] = 1;
2152 }
2153
2154 static int check_stream_sha1(git_zstream *stream,
2155                              const char *hdr,
2156                              unsigned long size,
2157                              const char *path,
2158                              const unsigned char *expected_sha1)
2159 {
2160         git_hash_ctx c;
2161         unsigned char real_sha1[GIT_MAX_RAWSZ];
2162         unsigned char buf[4096];
2163         unsigned long total_read;
2164         int status = Z_OK;
2165
2166         the_hash_algo->init_fn(&c);
2167         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2168
2169         /*
2170          * We already read some bytes into hdr, but the ones up to the NUL
2171          * do not count against the object's content size.
2172          */
2173         total_read = stream->total_out - strlen(hdr) - 1;
2174
2175         /*
2176          * This size comparison must be "<=" to read the final zlib packets;
2177          * see the comment in unpack_sha1_rest for details.
2178          */
2179         while (total_read <= size &&
2180                (status == Z_OK || status == Z_BUF_ERROR)) {
2181                 stream->next_out = buf;
2182                 stream->avail_out = sizeof(buf);
2183                 if (size - total_read < stream->avail_out)
2184                         stream->avail_out = size - total_read;
2185                 status = git_inflate(stream, Z_FINISH);
2186                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2187                 total_read += stream->next_out - buf;
2188         }
2189         git_inflate_end(stream);
2190
2191         if (status != Z_STREAM_END) {
2192                 error(_("corrupt loose object '%s'"), sha1_to_hex(expected_sha1));
2193                 return -1;
2194         }
2195         if (stream->avail_in) {
2196                 error(_("garbage at end of loose object '%s'"),
2197                       sha1_to_hex(expected_sha1));
2198                 return -1;
2199         }
2200
2201         the_hash_algo->final_fn(real_sha1, &c);
2202         if (!hasheq(expected_sha1, real_sha1)) {
2203                 error(_("sha1 mismatch for %s (expected %s)"), path,
2204                       sha1_to_hex(expected_sha1));
2205                 return -1;
2206         }
2207
2208         return 0;
2209 }
2210
2211 int read_loose_object(const char *path,
2212                       const struct object_id *expected_oid,
2213                       enum object_type *type,
2214                       unsigned long *size,
2215                       void **contents)
2216 {
2217         int ret = -1;
2218         void *map = NULL;
2219         unsigned long mapsize;
2220         git_zstream stream;
2221         char hdr[MAX_HEADER_LEN];
2222
2223         *contents = NULL;
2224
2225         map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2226         if (!map) {
2227                 error_errno(_("unable to mmap %s"), path);
2228                 goto out;
2229         }
2230
2231         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2232                 error(_("unable to unpack header of %s"), path);
2233                 goto out;
2234         }
2235
2236         *type = parse_sha1_header(hdr, size);
2237         if (*type < 0) {
2238                 error(_("unable to parse header of %s"), path);
2239                 git_inflate_end(&stream);
2240                 goto out;
2241         }
2242
2243         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2244                 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2245                         goto out;
2246         } else {
2247                 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2248                 if (!*contents) {
2249                         error(_("unable to unpack contents of %s"), path);
2250                         git_inflate_end(&stream);
2251                         goto out;
2252                 }
2253                 if (check_object_signature(expected_oid, *contents,
2254                                          *size, type_name(*type))) {
2255                         error(_("sha1 mismatch for %s (expected %s)"), path,
2256                               oid_to_hex(expected_oid));
2257                         free(*contents);
2258                         goto out;
2259                 }
2260         }
2261
2262         ret = 0; /* everything checks out */
2263
2264 out:
2265         if (map)
2266                 munmap(map, mapsize);
2267         return ret;
2268 }