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 void loose_object_path(struct repository *r, struct strbuf *buf,
 350                        const unsigned char *sha1)
 351 {
 352         strbuf_reset(buf);
 353         strbuf_addstr(buf, r->objects->objectdir);
 354         strbuf_addch(buf, '/');
 355         fill_sha1_path(buf, sha1);
 356 }
 357
 358 struct strbuf *alt_scratch_buf(struct object_directory *odb)
 359 {
 360         strbuf_setlen(&odb->scratch, odb->base_len);
 361         return &odb->scratch;
 362 }
 363
 364 static const char *alt_sha1_path(struct object_directory *odb,
 365                                  const unsigned char *sha1)
 366 {
 367         struct strbuf *buf = alt_scratch_buf(odb);
 368         fill_sha1_path(buf, sha1);
 369         return buf->buf;
 370 }
 371
 372 /*
 373  * Return non-zero iff the path is usable as an alternate object database.
 374  */
 375 static int alt_odb_usable(struct raw_object_store *o,
 376                           struct strbuf *path,
 377                           const char *normalized_objdir)
 378 {
 379         struct object_directory *odb;
 380
 381         /* Detect cases where alternate disappeared */
 382         if (!is_directory(path->buf)) {
 383                 error(_("object directory %s does not exist; "
 384                         "check .git/objects/info/alternates"),
 385                       path->buf);
 386                 return 0;
 387         }
 388
 389         /*
 390          * Prevent the common mistake of listing the same
 391          * thing twice, or object directory itself.
 392          */
 393         for (odb = o->alt_odb_list; odb; odb = odb->next) {
 394                 if (!fspathcmp(path->buf, odb->path))
 395                         return 0;
 396         }
 397         if (!fspathcmp(path->buf, normalized_objdir))
 398                 return 0;
 399
 400         return 1;
 401 }
 402
 403 /*
 404  * Prepare alternate object database registry.
 405  *
 406  * The variable alt_odb_list points at the list of struct
 407  * object_directory.  The elements on this list come from
 408  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 409  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 410  * whose contents is similar to that environment variable but can be
 411  * LF separated.  Its base points at a statically allocated buffer that
 412  * contains "/the/directory/corresponding/to/.git/objects/...", while
 413  * its name points just after the slash at the end of ".git/objects/"
 414  * in the example above, and has enough space to hold 40-byte hex
 415  * SHA1, an extra slash for the first level indirection, and the
 416  * terminating NUL.
 417  */
 418 static void read_info_alternates(struct repository *r,
 419                                  const char *relative_base,
 420                                  int depth);
 421 static int link_alt_odb_entry(struct repository *r, const char *entry,
 422         const char *relative_base, int depth, const char *normalized_objdir)
 423 {
 424         struct object_directory *ent;
 425         struct strbuf pathbuf = STRBUF_INIT;
 426
 427         if (!is_absolute_path(entry) && relative_base) {
 428                 strbuf_realpath(&pathbuf, relative_base, 1);
 429                 strbuf_addch(&pathbuf, '/');
 430         }
 431         strbuf_addstr(&pathbuf, entry);
 432
 433         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 434                 error(_("unable to normalize alternate object path: %s"),
 435                       pathbuf.buf);
 436                 strbuf_release(&pathbuf);
 437                 return -1;
 438         }
 439
 440         /*
 441          * The trailing slash after the directory name is given by
 442          * this function at the end. Remove duplicates.
 443          */
 444         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 445                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
 446
 447         if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
 448                 strbuf_release(&pathbuf);
 449                 return -1;
 450         }
 451
 452         ent = alloc_alt_odb(pathbuf.buf);
 453
 454         /* add the alternate entry */
 455         *r->objects->alt_odb_tail = ent;
 456         r->objects->alt_odb_tail = &(ent->next);
 457         ent->next = NULL;
 458
 459         /* recursively add alternates */
 460         read_info_alternates(r, pathbuf.buf, depth + 1);
 461
 462         strbuf_release(&pathbuf);
 463         return 0;
 464 }
 465
 466 static const char *parse_alt_odb_entry(const char *string,
 467                                        int sep,
 468                                        struct strbuf *out)
 469 {
 470         const char *end;
 471
 472         strbuf_reset(out);
 473
 474         if (*string == '#') {
 475                 /* comment; consume up to next separator */
 476                 end = strchrnul(string, sep);
 477         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
 478                 /*
 479                  * quoted path; unquote_c_style has copied the
 480                  * data for us and set "end". Broken quoting (e.g.,
 481                  * an entry that doesn't end with a quote) falls
 482                  * back to the unquoted case below.
 483                  */
 484         } else {
 485                 /* normal, unquoted path */
 486                 end = strchrnul(string, sep);
 487                 strbuf_add(out, string, end - string);
 488         }
 489
 490         if (*end)
 491                 end++;
 492         return end;
 493 }
 494
 495 static void link_alt_odb_entries(struct repository *r, const char *alt,
 496                                  int sep, const char *relative_base, int depth)
 497 {
 498         struct strbuf objdirbuf = STRBUF_INIT;
 499         struct strbuf entry = STRBUF_INIT;
 500
 501         if (!alt || !*alt)
 502                 return;
 503
 504         if (depth > 5) {
 505                 error(_("%s: ignoring alternate object stores, nesting too deep"),
 506                                 relative_base);
 507                 return;
 508         }
 509
 510         strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
 511         if (strbuf_normalize_path(&objdirbuf) < 0)
 512                 die(_("unable to normalize object directory: %s"),
 513                     objdirbuf.buf);
 514
 515         while (*alt) {
 516                 alt = parse_alt_odb_entry(alt, sep, &entry);
 517                 if (!entry.len)
 518                         continue;
 519                 link_alt_odb_entry(r, entry.buf,
 520                                    relative_base, depth, objdirbuf.buf);
 521         }
 522         strbuf_release(&entry);
 523         strbuf_release(&objdirbuf);
 524 }
 525
 526 static void read_info_alternates(struct repository *r,
 527                                  const char *relative_base,
 528                                  int depth)
 529 {
 530         char *path;
 531         struct strbuf buf = STRBUF_INIT;
 532
 533         path = xstrfmt("%s/info/alternates", relative_base);
 534         if (strbuf_read_file(&buf, path, 1024) < 0) {
 535                 warn_on_fopen_errors(path);
 536                 free(path);
 537                 return;
 538         }
 539
 540         link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
 541         strbuf_release(&buf);
 542         free(path);
 543 }
 544
 545 struct object_directory *alloc_alt_odb(const char *dir)
 546 {
 547         struct object_directory *ent;
 548
 549         FLEX_ALLOC_STR(ent, path, dir);
 550         strbuf_init(&ent->scratch, 0);
 551         strbuf_addf(&ent->scratch, "%s/", dir);
 552         ent->base_len = ent->scratch.len;
 553
 554         return ent;
 555 }
 556
 557 void add_to_alternates_file(const char *reference)
 558 {
 559         struct lock_file lock = LOCK_INIT;
 560         char *alts = git_pathdup("objects/info/alternates");
 561         FILE *in, *out;
 562         int found = 0;
 563
 564         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
 565         out = fdopen_lock_file(&lock, "w");
 566         if (!out)
 567                 die_errno(_("unable to fdopen alternates lockfile"));
 568
 569         in = fopen(alts, "r");
 570         if (in) {
 571                 struct strbuf line = STRBUF_INIT;
 572
 573                 while (strbuf_getline(&line, in) != EOF) {
 574                         if (!strcmp(reference, line.buf)) {
 575                                 found = 1;
 576                                 break;
 577                         }
 578                         fprintf_or_die(out, "%s\n", line.buf);
 579                 }
 580
 581                 strbuf_release(&line);
 582                 fclose(in);
 583         }
 584         else if (errno != ENOENT)
 585                 die_errno(_("unable to read alternates file"));
 586
 587         if (found) {
 588                 rollback_lock_file(&lock);
 589         } else {
 590                 fprintf_or_die(out, "%s\n", reference);
 591                 if (commit_lock_file(&lock))
 592                         die_errno(_("unable to move new alternates file into place"));
 593                 if (the_repository->objects->alt_odb_tail)
 594                         link_alt_odb_entries(the_repository, reference,
 595                                              '\n', NULL, 0);
 596         }
 597         free(alts);
 598 }
 599
 600 void add_to_alternates_memory(const char *reference)
 601 {
 602         /*
 603          * Make sure alternates are initialized, or else our entry may be
 604          * overwritten when they are.
 605          */
 606         prepare_alt_odb(the_repository);
 607
 608         link_alt_odb_entries(the_repository, reference,
 609                              '\n', NULL, 0);
 610 }
 611
 612 /*
 613  * Compute the exact path an alternate is at and returns it. In case of
 614  * error NULL is returned and the human readable error is added to `err`
 615  * `path` may be relative and should point to $GIT_DIR.
 616  * `err` must not be null.
 617  */
 618 char *compute_alternate_path(const char *path, struct strbuf *err)
 619 {
 620         char *ref_git = NULL;
 621         const char *repo, *ref_git_s;
 622         int seen_error = 0;
 623
 624         ref_git_s = real_path_if_valid(path);
 625         if (!ref_git_s) {
 626                 seen_error = 1;
 627                 strbuf_addf(err, _("path '%s' does not exist"), path);
 628                 goto out;
 629         } else
 630                 /*
 631                  * Beware: read_gitfile(), real_path() and mkpath()
 632                  * return static buffer
 633                  */
 634                 ref_git = xstrdup(ref_git_s);
 635
 636         repo = read_gitfile(ref_git);
 637         if (!repo)
 638                 repo = read_gitfile(mkpath("%s/.git", ref_git));
 639         if (repo) {
 640                 free(ref_git);
 641                 ref_git = xstrdup(repo);
 642         }
 643
 644         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 645                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
 646                 free(ref_git);
 647                 ref_git = ref_git_git;
 648         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 649                 struct strbuf sb = STRBUF_INIT;
 650                 seen_error = 1;
 651                 if (get_common_dir(&sb, ref_git)) {
 652                         strbuf_addf(err,
 653                                     _("reference repository '%s' as a linked "
 654                                       "checkout is not supported yet."),
 655                                     path);
 656                         goto out;
 657                 }
 658
 659                 strbuf_addf(err, _("reference repository '%s' is not a "
 660                                         "local repository."), path);
 661                 goto out;
 662         }
 663
 664         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 665                 strbuf_addf(err, _("reference repository '%s' is shallow"),
 666                             path);
 667                 seen_error = 1;
 668                 goto out;
 669         }
 670
 671         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 672                 strbuf_addf(err,
 673                             _("reference repository '%s' is grafted"),
 674                             path);
 675                 seen_error = 1;
 676                 goto out;
 677         }
 678
 679 out:
 680         if (seen_error) {
 681                 FREE_AND_NULL(ref_git);
 682         }
 683
 684         return ref_git;
 685 }
 686
 687 int foreach_alt_odb(alt_odb_fn fn, void *cb)
 688 {
 689         struct object_directory *ent;
 690         int r = 0;
 691
 692         prepare_alt_odb(the_repository);
 693         for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
 694                 r = fn(ent, cb);
 695                 if (r)
 696                         break;
 697         }
 698         return r;
 699 }
 700
 701 void prepare_alt_odb(struct repository *r)
 702 {
 703         if (r->objects->alt_odb_tail)
 704                 return;
 705
 706         r->objects->alt_odb_tail = &r->objects->alt_odb_list;
 707         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
 708
 709         read_info_alternates(r, r->objects->objectdir, 0);
 710 }
 711
 712 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 713 static int freshen_file(const char *fn)
 714 {
 715         struct utimbuf t;
 716         t.actime = t.modtime = time(NULL);
 717         return !utime(fn, &t);
 718 }
 719
 720 /*
 721  * All of the check_and_freshen functions return 1 if the file exists and was
 722  * freshened (if freshening was requested), 0 otherwise. If they return
 723  * 0, you should not assume that it is safe to skip a write of the object (it
 724  * either does not exist on disk, or has a stale mtime and may be subject to
 725  * pruning).
 726  */
 727 int check_and_freshen_file(const char *fn, int freshen)
 728 {
 729         if (access(fn, F_OK))
 730                 return 0;
 731         if (freshen && !freshen_file(fn))
 732                 return 0;
 733         return 1;
 734 }
 735
 736 static int check_and_freshen_local(const struct object_id *oid, int freshen)
 737 {
 738         static struct strbuf buf = STRBUF_INIT;
 739
 740         loose_object_path(the_repository, &buf, oid->hash);
 741
 742         return check_and_freshen_file(buf.buf, freshen);
 743 }
 744
 745 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
 746 {
 747         struct object_directory *odb;
 748         prepare_alt_odb(the_repository);
 749         for (odb = the_repository->objects->alt_odb_list; odb; odb = odb->next) {
 750                 const char *path = alt_sha1_path(odb, oid->hash);
 751                 if (check_and_freshen_file(path, freshen))
 752                         return 1;
 753         }
 754         return 0;
 755 }
 756
 757 static int check_and_freshen(const struct object_id *oid, int freshen)
 758 {
 759         return check_and_freshen_local(oid, freshen) ||
 760                check_and_freshen_nonlocal(oid, freshen);
 761 }
 762
 763 int has_loose_object_nonlocal(const struct object_id *oid)
 764 {
 765         return check_and_freshen_nonlocal(oid, 0);
 766 }
 767
 768 static int has_loose_object(const struct object_id *oid)
 769 {
 770         return check_and_freshen(oid, 0);
 771 }
 772
 773 static void mmap_limit_check(size_t length)
 774 {
 775         static size_t limit = 0;
 776         if (!limit) {
 777                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 778                 if (!limit)
 779                         limit = SIZE_MAX;
 780         }
 781         if (length > limit)
 782                 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
 783                     (uintmax_t)length, (uintmax_t)limit);
 784 }
 785
 786 void *xmmap_gently(void *start, size_t length,
 787                   int prot, int flags, int fd, off_t offset)
 788 {
 789         void *ret;
 790
 791         mmap_limit_check(length);
 792         ret = mmap(start, length, prot, flags, fd, offset);
 793         if (ret == MAP_FAILED) {
 794                 if (!length)
 795                         return NULL;
 796                 release_pack_memory(length);
 797                 ret = mmap(start, length, prot, flags, fd, offset);
 798         }
 799         return ret;
 800 }
 801
 802 void *xmmap(void *start, size_t length,
 803         int prot, int flags, int fd, off_t offset)
 804 {
 805         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 806         if (ret == MAP_FAILED)
 807                 die_errno(_("mmap failed"));
 808         return ret;
 809 }
 810
 811 /*
 812  * With an in-core object data in "map", rehash it to make sure the
 813  * object name actually matches "sha1" to detect object corruption.
 814  * With "map" == NULL, try reading the object named with "sha1" using
 815  * the streaming interface and rehash it to do the same.
 816  */
 817 int check_object_signature(const struct object_id *oid, void *map,
 818                            unsigned long size, const char *type)
 819 {
 820         struct object_id real_oid;
 821         enum object_type obj_type;
 822         struct git_istream *st;
 823         git_hash_ctx c;
 824         char hdr[MAX_HEADER_LEN];
 825         int hdrlen;
 826
 827         if (map) {
 828                 hash_object_file(map, size, type, &real_oid);
 829                 return !oideq(oid, &real_oid) ? -1 : 0;
 830         }
 831
 832         st = open_istream(oid, &obj_type, &size, NULL);
 833         if (!st)
 834                 return -1;
 835
 836         /* Generate the header */
 837         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
 838
 839         /* Sha1.. */
 840         the_hash_algo->init_fn(&c);
 841         the_hash_algo->update_fn(&c, hdr, hdrlen);
 842         for (;;) {
 843                 char buf[1024 * 16];
 844                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
 845
 846                 if (readlen < 0) {
 847                         close_istream(st);
 848                         return -1;
 849                 }
 850                 if (!readlen)
 851                         break;
 852                 the_hash_algo->update_fn(&c, buf, readlen);
 853         }
 854         the_hash_algo->final_fn(real_oid.hash, &c);
 855         close_istream(st);
 856         return !oideq(oid, &real_oid) ? -1 : 0;
 857 }
 858
 859 int git_open_cloexec(const char *name, int flags)
 860 {
 861         int fd;
 862         static int o_cloexec = O_CLOEXEC;
 863
 864         fd = open(name, flags | o_cloexec);
 865         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 866                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
 867                 o_cloexec &= ~O_CLOEXEC;
 868                 fd = open(name, flags | o_cloexec);
 869         }
 870
 871 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 872         {
 873                 static int fd_cloexec = FD_CLOEXEC;
 874
 875                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
 876                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 877                         int flags = fcntl(fd, F_GETFD);
 878                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 879                                 fd_cloexec = 0;
 880                 }
 881         }
 882 #endif
 883         return fd;
 884 }
 885
 886 /*
 887  * Find "sha1" as a loose object in the local repository or in an alternate.
 888  * Returns 0 on success, negative on failure.
 889  *
 890  * The "path" out-parameter will give the path of the object we found (if any).
 891  * Note that it may point to static storage and is only valid until another
 892  * call to loose_object_path(), etc.
 893  */
 894 static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
 895                           struct stat *st, const char **path)
 896 {
 897         struct object_directory *odb;
 898         static struct strbuf buf = STRBUF_INIT;
 899
 900         loose_object_path(r, &buf, sha1);
 901         *path = buf.buf;
 902
 903         if (!lstat(*path, st))
 904                 return 0;
 905
 906         prepare_alt_odb(r);
 907         errno = ENOENT;
 908         for (odb = r->objects->alt_odb_list; odb; odb = odb->next) {
 909                 *path = alt_sha1_path(odb, sha1);
 910                 if (!lstat(*path, st))
 911                         return 0;
 912         }
 913
 914         return -1;
 915 }
 916
 917 /*
 918  * Like stat_sha1_file(), but actually open the object and return the
 919  * descriptor. See the caveats on the "path" parameter above.
 920  */
 921 static int open_sha1_file(struct repository *r,
 922                           const unsigned char *sha1, const char **path)
 923 {
 924         int fd;
 925         struct object_directory *odb;
 926         int most_interesting_errno;
 927         static struct strbuf buf = STRBUF_INIT;
 928
 929         loose_object_path(r, &buf, sha1);
 930         *path = buf.buf;
 931
 932         fd = git_open(*path);
 933         if (fd >= 0)
 934                 return fd;
 935         most_interesting_errno = errno;
 936
 937         prepare_alt_odb(r);
 938         for (odb = r->objects->alt_odb_list; odb; odb = odb->next) {
 939                 *path = alt_sha1_path(odb, sha1);
 940                 fd = git_open(*path);
 941                 if (fd >= 0)
 942                         return fd;
 943                 if (most_interesting_errno == ENOENT)
 944                         most_interesting_errno = errno;
 945         }
 946         errno = most_interesting_errno;
 947         return -1;
 948 }
 949
 950 /*
 951  * Map the loose object at "path" if it is not NULL, or the path found by
 952  * searching for a loose object named "sha1".
 953  */
 954 static void *map_sha1_file_1(struct repository *r, const char *path,
 955                              const unsigned char *sha1, unsigned long *size)
 956 {
 957         void *map;
 958         int fd;
 959
 960         if (path)
 961                 fd = git_open(path);
 962         else
 963                 fd = open_sha1_file(r, sha1, &path);
 964         map = NULL;
 965         if (fd >= 0) {
 966                 struct stat st;
 967
 968                 if (!fstat(fd, &st)) {
 969                         *size = xsize_t(st.st_size);
 970                         if (!*size) {
 971                                 /* mmap() is forbidden on empty files */
 972                                 error(_("object file %s is empty"), path);
 973                                 return NULL;
 974                         }
 975                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 976                 }
 977                 close(fd);
 978         }
 979         return map;
 980 }
 981
 982 void *map_sha1_file(struct repository *r,
 983                     const unsigned char *sha1, unsigned long *size)
 984 {
 985         return map_sha1_file_1(r, NULL, sha1, size);
 986 }
 987
 988 static int unpack_sha1_short_header(git_zstream *stream,
 989                                     unsigned char *map, unsigned long mapsize,
 990                                     void *buffer, unsigned long bufsiz)
 991 {
 992         /* Get the data stream */
 993         memset(stream, 0, sizeof(*stream));
 994         stream->next_in = map;
 995         stream->avail_in = mapsize;
 996         stream->next_out = buffer;
 997         stream->avail_out = bufsiz;
 998
 999         git_inflate_init(stream);
1000         return git_inflate(stream, 0);
1001 }
1002
1003 int unpack_sha1_header(git_zstream *stream,
1004                        unsigned char *map, unsigned long mapsize,
1005                        void *buffer, unsigned long bufsiz)
1006 {
1007         int status = unpack_sha1_short_header(stream, map, mapsize,
1008                                               buffer, bufsiz);
1009
1010         if (status < Z_OK)
1011                 return status;
1012
1013         /* Make sure we have the terminating NUL */
1014         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1015                 return -1;
1016         return 0;
1017 }
1018
1019 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1020                                         unsigned long mapsize, void *buffer,
1021                                         unsigned long bufsiz, struct strbuf *header)
1022 {
1023         int status;
1024
1025         status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1026         if (status < Z_OK)
1027                 return -1;
1028
1029         /*
1030          * Check if entire header is unpacked in the first iteration.
1031          */
1032         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1033                 return 0;
1034
1035         /*
1036          * buffer[0..bufsiz] was not large enough.  Copy the partial
1037          * result out to header, and then append the result of further
1038          * reading the stream.
1039          */
1040         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1041         stream->next_out = buffer;
1042         stream->avail_out = bufsiz;
1043
1044         do {
1045                 status = git_inflate(stream, 0);
1046                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1047                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1048                         return 0;
1049                 stream->next_out = buffer;
1050                 stream->avail_out = bufsiz;
1051         } while (status != Z_STREAM_END);
1052         return -1;
1053 }
1054
1055 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1056 {
1057         int bytes = strlen(buffer) + 1;
1058         unsigned char *buf = xmallocz(size);
1059         unsigned long n;
1060         int status = Z_OK;
1061
1062         n = stream->total_out - bytes;
1063         if (n > size)
1064                 n = size;
1065         memcpy(buf, (char *) buffer + bytes, n);
1066         bytes = n;
1067         if (bytes <= size) {
1068                 /*
1069                  * The above condition must be (bytes <= size), not
1070                  * (bytes < size).  In other words, even though we
1071                  * expect no more output and set avail_out to zero,
1072                  * the input zlib stream may have bytes that express
1073                  * "this concludes the stream", and we *do* want to
1074                  * eat that input.
1075                  *
1076                  * Otherwise we would not be able to test that we
1077                  * consumed all the input to reach the expected size;
1078                  * we also want to check that zlib tells us that all
1079                  * went well with status == Z_STREAM_END at the end.
1080                  */
1081                 stream->next_out = buf + bytes;
1082                 stream->avail_out = size - bytes;
1083                 while (status == Z_OK)
1084                         status = git_inflate(stream, Z_FINISH);
1085         }
1086         if (status == Z_STREAM_END && !stream->avail_in) {
1087                 git_inflate_end(stream);
1088                 return buf;
1089         }
1090
1091         if (status < 0)
1092                 error(_("corrupt loose object '%s'"), sha1_to_hex(sha1));
1093         else if (stream->avail_in)
1094                 error(_("garbage at end of loose object '%s'"),
1095                       sha1_to_hex(sha1));
1096         free(buf);
1097         return NULL;
1098 }
1099
1100 /*
1101  * We used to just use "sscanf()", but that's actually way
1102  * too permissive for what we want to check. So do an anal
1103  * object header parse by hand.
1104  */
1105 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1106                                unsigned int flags)
1107 {
1108         const char *type_buf = hdr;
1109         unsigned long size;
1110         int type, type_len = 0;
1111
1112         /*
1113          * The type can be of any size but is followed by
1114          * a space.
1115          */
1116         for (;;) {
1117                 char c = *hdr++;
1118                 if (!c)
1119                         return -1;
1120                 if (c == ' ')
1121                         break;
1122                 type_len++;
1123         }
1124
1125         type = type_from_string_gently(type_buf, type_len, 1);
1126         if (oi->type_name)
1127                 strbuf_add(oi->type_name, type_buf, type_len);
1128         /*
1129          * Set type to 0 if its an unknown object and
1130          * we're obtaining the type using '--allow-unknown-type'
1131          * option.
1132          */
1133         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1134                 type = 0;
1135         else if (type < 0)
1136                 die(_("invalid object type"));
1137         if (oi->typep)
1138                 *oi->typep = type;
1139
1140         /*
1141          * The length must follow immediately, and be in canonical
1142          * decimal format (ie "010" is not valid).
1143          */
1144         size = *hdr++ - '0';
1145         if (size > 9)
1146                 return -1;
1147         if (size) {
1148                 for (;;) {
1149                         unsigned long c = *hdr - '0';
1150                         if (c > 9)
1151                                 break;
1152                         hdr++;
1153                         size = size * 10 + c;
1154                 }
1155         }
1156
1157         if (oi->sizep)
1158                 *oi->sizep = size;
1159
1160         /*
1161          * The length must be followed by a zero byte
1162          */
1163         return *hdr ? -1 : type;
1164 }
1165
1166 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1167 {
1168         struct object_info oi = OBJECT_INFO_INIT;
1169
1170         oi.sizep = sizep;
1171         return parse_sha1_header_extended(hdr, &oi, 0);
1172 }
1173
1174 static int sha1_loose_object_info(struct repository *r,
1175                                   const unsigned char *sha1,
1176                                   struct object_info *oi, int flags)
1177 {
1178         int status = 0;
1179         unsigned long mapsize;
1180         void *map;
1181         git_zstream stream;
1182         char hdr[MAX_HEADER_LEN];
1183         struct strbuf hdrbuf = STRBUF_INIT;
1184         unsigned long size_scratch;
1185
1186         if (oi->delta_base_sha1)
1187                 hashclr(oi->delta_base_sha1);
1188
1189         /*
1190          * If we don't care about type or size, then we don't
1191          * need to look inside the object at all. Note that we
1192          * do not optimize out the stat call, even if the
1193          * caller doesn't care about the disk-size, since our
1194          * return value implicitly indicates whether the
1195          * object even exists.
1196          */
1197         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1198                 const char *path;
1199                 struct stat st;
1200                 if (stat_sha1_file(r, sha1, &st, &path) < 0)
1201                         return -1;
1202                 if (oi->disk_sizep)
1203                         *oi->disk_sizep = st.st_size;
1204                 return 0;
1205         }
1206
1207         map = map_sha1_file(r, sha1, &mapsize);
1208         if (!map)
1209                 return -1;
1210
1211         if (!oi->sizep)
1212                 oi->sizep = &size_scratch;
1213
1214         if (oi->disk_sizep)
1215                 *oi->disk_sizep = mapsize;
1216         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1217                 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1218                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1219                                        sha1_to_hex(sha1));
1220         } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1221                 status = error(_("unable to unpack %s header"),
1222                                sha1_to_hex(sha1));
1223         if (status < 0)
1224                 ; /* Do nothing */
1225         else if (hdrbuf.len) {
1226                 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1227                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1228                                        sha1_to_hex(sha1));
1229         } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1230                 status = error(_("unable to parse %s header"), sha1_to_hex(sha1));
1231
1232         if (status >= 0 && oi->contentp) {
1233                 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1234                                                  *oi->sizep, sha1);
1235                 if (!*oi->contentp) {
1236                         git_inflate_end(&stream);
1237                         status = -1;
1238                 }
1239         } else
1240                 git_inflate_end(&stream);
1241
1242         munmap(map, mapsize);
1243         if (status && oi->typep)
1244                 *oi->typep = status;
1245         if (oi->sizep == &size_scratch)
1246                 oi->sizep = NULL;
1247         strbuf_release(&hdrbuf);
1248         oi->whence = OI_LOOSE;
1249         return (status < 0) ? status : 0;
1250 }
1251
1252 int fetch_if_missing = 1;
1253
1254 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1255                              struct object_info *oi, unsigned flags)
1256 {
1257         static struct object_info blank_oi = OBJECT_INFO_INIT;
1258         struct pack_entry e;
1259         int rtype;
1260         const struct object_id *real = oid;
1261         int already_retried = 0;
1262
1263         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1264                 real = lookup_replace_object(r, oid);
1265
1266         if (is_null_oid(real))
1267                 return -1;
1268
1269         if (!oi)
1270                 oi = &blank_oi;
1271
1272         if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1273                 struct cached_object *co = find_cached_object(real);
1274                 if (co) {
1275                         if (oi->typep)
1276                                 *(oi->typep) = co->type;
1277                         if (oi->sizep)
1278                                 *(oi->sizep) = co->size;
1279                         if (oi->disk_sizep)
1280                                 *(oi->disk_sizep) = 0;
1281                         if (oi->delta_base_sha1)
1282                                 hashclr(oi->delta_base_sha1);
1283                         if (oi->type_name)
1284                                 strbuf_addstr(oi->type_name, type_name(co->type));
1285                         if (oi->contentp)
1286                                 *oi->contentp = xmemdupz(co->buf, co->size);
1287                         oi->whence = OI_CACHED;
1288                         return 0;
1289                 }
1290         }
1291
1292         while (1) {
1293                 if (find_pack_entry(r, real, &e))
1294                         break;
1295
1296                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1297                         return -1;
1298
1299                 /* Most likely it's a loose object. */
1300                 if (!sha1_loose_object_info(r, real->hash, oi, flags))
1301                         return 0;
1302
1303                 /* Not a loose object; someone else may have just packed it. */
1304                 if (!(flags & OBJECT_INFO_QUICK)) {
1305                         reprepare_packed_git(r);
1306                         if (find_pack_entry(r, real, &e))
1307                                 break;
1308                 }
1309
1310                 /* Check if it is a missing object */
1311                 if (fetch_if_missing && repository_format_partial_clone &&
1312                     !already_retried && r == the_repository) {
1313                         /*
1314                          * TODO Investigate having fetch_object() return
1315                          * TODO error/success and stopping the music here.
1316                          * TODO Pass a repository struct through fetch_object,
1317                          * such that arbitrary repositories work.
1318                          */
1319                         fetch_objects(repository_format_partial_clone, real, 1);
1320                         already_retried = 1;
1321                         continue;
1322                 }
1323
1324                 return -1;
1325         }
1326
1327         if (oi == &blank_oi)
1328                 /*
1329                  * We know that the caller doesn't actually need the
1330                  * information below, so return early.
1331                  */
1332                 return 0;
1333         rtype = packed_object_info(r, e.p, e.offset, oi);
1334         if (rtype < 0) {
1335                 mark_bad_packed_object(e.p, real->hash);
1336                 return oid_object_info_extended(r, real, oi, 0);
1337         } else if (oi->whence == OI_PACKED) {
1338                 oi->u.packed.offset = e.offset;
1339                 oi->u.packed.pack = e.p;
1340                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1341                                          rtype == OBJ_OFS_DELTA);
1342         }
1343
1344         return 0;
1345 }
1346
1347 /* returns enum object_type or negative */
1348 int oid_object_info(struct repository *r,
1349                     const struct object_id *oid,
1350                     unsigned long *sizep)
1351 {
1352         enum object_type type;
1353         struct object_info oi = OBJECT_INFO_INIT;
1354
1355         oi.typep = &type;
1356         oi.sizep = sizep;
1357         if (oid_object_info_extended(r, oid, &oi,
1358                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1359                 return -1;
1360         return type;
1361 }
1362
1363 static void *read_object(const unsigned char *sha1, enum object_type *type,
1364                          unsigned long *size)
1365 {
1366         struct object_id oid;
1367         struct object_info oi = OBJECT_INFO_INIT;
1368         void *content;
1369         oi.typep = type;
1370         oi.sizep = size;
1371         oi.contentp = &content;
1372
1373         hashcpy(oid.hash, sha1);
1374
1375         if (oid_object_info_extended(the_repository, &oid, &oi, 0) < 0)
1376                 return NULL;
1377         return content;
1378 }
1379
1380 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1381                         struct object_id *oid)
1382 {
1383         struct cached_object *co;
1384
1385         hash_object_file(buf, len, type_name(type), oid);
1386         if (has_sha1_file(oid->hash) || find_cached_object(oid))
1387                 return 0;
1388         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1389         co = &cached_objects[cached_object_nr++];
1390         co->size = len;
1391         co->type = type;
1392         co->buf = xmalloc(len);
1393         memcpy(co->buf, buf, len);
1394         oidcpy(&co->oid, oid);
1395         return 0;
1396 }
1397
1398 /*
1399  * This function dies on corrupt objects; the callers who want to
1400  * deal with them should arrange to call read_object() and give error
1401  * messages themselves.
1402  */
1403 void *read_object_file_extended(const struct object_id *oid,
1404                                 enum object_type *type,
1405                                 unsigned long *size,
1406                                 int lookup_replace)
1407 {
1408         void *data;
1409         const struct packed_git *p;
1410         const char *path;
1411         struct stat st;
1412         const struct object_id *repl = lookup_replace ?
1413                 lookup_replace_object(the_repository, oid) : oid;
1414
1415         errno = 0;
1416         data = read_object(repl->hash, type, size);
1417         if (data)
1418                 return data;
1419
1420         if (errno && errno != ENOENT)
1421                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1422
1423         /* die if we replaced an object with one that does not exist */
1424         if (repl != oid)
1425                 die(_("replacement %s not found for %s"),
1426                     oid_to_hex(repl), oid_to_hex(oid));
1427
1428         if (!stat_sha1_file(the_repository, repl->hash, &st, &path))
1429                 die(_("loose object %s (stored in %s) is corrupt"),
1430                     oid_to_hex(repl), path);
1431
1432         if ((p = has_packed_and_bad(repl->hash)) != NULL)
1433                 die(_("packed object %s (stored in %s) is corrupt"),
1434                     oid_to_hex(repl), p->pack_name);
1435
1436         return NULL;
1437 }
1438
1439 void *read_object_with_reference(const struct object_id *oid,
1440                                  const char *required_type_name,
1441                                  unsigned long *size,
1442                                  struct object_id *actual_oid_return)
1443 {
1444         enum object_type type, required_type;
1445         void *buffer;
1446         unsigned long isize;
1447         struct object_id actual_oid;
1448
1449         required_type = type_from_string(required_type_name);
1450         oidcpy(&actual_oid, oid);
1451         while (1) {
1452                 int ref_length = -1;
1453                 const char *ref_type = NULL;
1454
1455                 buffer = read_object_file(&actual_oid, &type, &isize);
1456                 if (!buffer)
1457                         return NULL;
1458                 if (type == required_type) {
1459                         *size = isize;
1460                         if (actual_oid_return)
1461                                 oidcpy(actual_oid_return, &actual_oid);
1462                         return buffer;
1463                 }
1464                 /* Handle references */
1465                 else if (type == OBJ_COMMIT)
1466                         ref_type = "tree ";
1467                 else if (type == OBJ_TAG)
1468                         ref_type = "object ";
1469                 else {
1470                         free(buffer);
1471                         return NULL;
1472                 }
1473                 ref_length = strlen(ref_type);
1474
1475                 if (ref_length + the_hash_algo->hexsz > isize ||
1476                     memcmp(buffer, ref_type, ref_length) ||
1477                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1478                         free(buffer);
1479                         return NULL;
1480                 }
1481                 free(buffer);
1482                 /* Now we have the ID of the referred-to object in
1483                  * actual_oid.  Check again. */
1484         }
1485 }
1486
1487 static void write_object_file_prepare(const void *buf, unsigned long len,
1488                                       const char *type, struct object_id *oid,
1489                                       char *hdr, int *hdrlen)
1490 {
1491         git_hash_ctx c;
1492
1493         /* Generate the header */
1494         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1495
1496         /* Sha1.. */
1497         the_hash_algo->init_fn(&c);
1498         the_hash_algo->update_fn(&c, hdr, *hdrlen);
1499         the_hash_algo->update_fn(&c, buf, len);
1500         the_hash_algo->final_fn(oid->hash, &c);
1501 }
1502
1503 /*
1504  * Move the just written object into its final resting place.
1505  */
1506 int finalize_object_file(const char *tmpfile, const char *filename)
1507 {
1508         int ret = 0;
1509
1510         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1511                 goto try_rename;
1512         else if (link(tmpfile, filename))
1513                 ret = errno;
1514
1515         /*
1516          * Coda hack - coda doesn't like cross-directory links,
1517          * so we fall back to a rename, which will mean that it
1518          * won't be able to check collisions, but that's not a
1519          * big deal.
1520          *
1521          * The same holds for FAT formatted media.
1522          *
1523          * When this succeeds, we just return.  We have nothing
1524          * left to unlink.
1525          */
1526         if (ret && ret != EEXIST) {
1527         try_rename:
1528                 if (!rename(tmpfile, filename))
1529                         goto out;
1530                 ret = errno;
1531         }
1532         unlink_or_warn(tmpfile);
1533         if (ret) {
1534                 if (ret != EEXIST) {
1535                         return error_errno(_("unable to write sha1 filename %s"), filename);
1536                 }
1537                 /* FIXME!!! Collision check here ? */
1538         }
1539
1540 out:
1541         if (adjust_shared_perm(filename))
1542                 return error(_("unable to set permission to '%s'"), filename);
1543         return 0;
1544 }
1545
1546 static int write_buffer(int fd, const void *buf, size_t len)
1547 {
1548         if (write_in_full(fd, buf, len) < 0)
1549                 return error_errno(_("file write error"));
1550         return 0;
1551 }
1552
1553 int hash_object_file(const void *buf, unsigned long len, const char *type,
1554                      struct object_id *oid)
1555 {
1556         char hdr[MAX_HEADER_LEN];
1557         int hdrlen = sizeof(hdr);
1558         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1559         return 0;
1560 }
1561
1562 /* Finalize a file on disk, and close it. */
1563 static void close_sha1_file(int fd)
1564 {
1565         if (fsync_object_files)
1566                 fsync_or_die(fd, "sha1 file");
1567         if (close(fd) != 0)
1568                 die_errno(_("error when closing sha1 file"));
1569 }
1570
1571 /* Size of directory component, including the ending '/' */
1572 static inline int directory_size(const char *filename)
1573 {
1574         const char *s = strrchr(filename, '/');
1575         if (!s)
1576                 return 0;
1577         return s - filename + 1;
1578 }
1579
1580 /*
1581  * This creates a temporary file in the same directory as the final
1582  * 'filename'
1583  *
1584  * We want to avoid cross-directory filename renames, because those
1585  * can have problems on various filesystems (FAT, NFS, Coda).
1586  */
1587 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1588 {
1589         int fd, dirlen = directory_size(filename);
1590
1591         strbuf_reset(tmp);
1592         strbuf_add(tmp, filename, dirlen);
1593         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1594         fd = git_mkstemp_mode(tmp->buf, 0444);
1595         if (fd < 0 && dirlen && errno == ENOENT) {
1596                 /*
1597                  * Make sure the directory exists; note that the contents
1598                  * of the buffer are undefined after mkstemp returns an
1599                  * error, so we have to rewrite the whole buffer from
1600                  * scratch.
1601                  */
1602                 strbuf_reset(tmp);
1603                 strbuf_add(tmp, filename, dirlen - 1);
1604                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1605                         return -1;
1606                 if (adjust_shared_perm(tmp->buf))
1607                         return -1;
1608
1609                 /* Try again */
1610                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1611                 fd = git_mkstemp_mode(tmp->buf, 0444);
1612         }
1613         return fd;
1614 }
1615
1616 static int write_loose_object(const struct object_id *oid, char *hdr,
1617                               int hdrlen, const void *buf, unsigned long len,
1618                               time_t mtime)
1619 {
1620         int fd, ret;
1621         unsigned char compressed[4096];
1622         git_zstream stream;
1623         git_hash_ctx c;
1624         struct object_id parano_oid;
1625         static struct strbuf tmp_file = STRBUF_INIT;
1626         static struct strbuf filename = STRBUF_INIT;
1627
1628         loose_object_path(the_repository, &filename, oid->hash);
1629
1630         fd = create_tmpfile(&tmp_file, filename.buf);
1631         if (fd < 0) {
1632                 if (errno == EACCES)
1633                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1634                 else
1635                         return error_errno(_("unable to create temporary file"));
1636         }
1637
1638         /* Set it up */
1639         git_deflate_init(&stream, zlib_compression_level);
1640         stream.next_out = compressed;
1641         stream.avail_out = sizeof(compressed);
1642         the_hash_algo->init_fn(&c);
1643
1644         /* First header.. */
1645         stream.next_in = (unsigned char *)hdr;
1646         stream.avail_in = hdrlen;
1647         while (git_deflate(&stream, 0) == Z_OK)
1648                 ; /* nothing */
1649         the_hash_algo->update_fn(&c, hdr, hdrlen);
1650
1651         /* Then the data itself.. */
1652         stream.next_in = (void *)buf;
1653         stream.avail_in = len;
1654         do {
1655                 unsigned char *in0 = stream.next_in;
1656                 ret = git_deflate(&stream, Z_FINISH);
1657                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1658                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1659                         die(_("unable to write sha1 file"));
1660                 stream.next_out = compressed;
1661                 stream.avail_out = sizeof(compressed);
1662         } while (ret == Z_OK);
1663
1664         if (ret != Z_STREAM_END)
1665                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1666                     ret);
1667         ret = git_deflate_end_gently(&stream);
1668         if (ret != Z_OK)
1669                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1670                     ret);
1671         the_hash_algo->final_fn(parano_oid.hash, &c);
1672         if (!oideq(oid, &parano_oid))
1673                 die(_("confused by unstable object source data for %s"),
1674                     oid_to_hex(oid));
1675
1676         close_sha1_file(fd);
1677
1678         if (mtime) {
1679                 struct utimbuf utb;
1680                 utb.actime = mtime;
1681                 utb.modtime = mtime;
1682                 if (utime(tmp_file.buf, &utb) < 0)
1683                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1684         }
1685
1686         return finalize_object_file(tmp_file.buf, filename.buf);
1687 }
1688
1689 static int freshen_loose_object(const struct object_id *oid)
1690 {
1691         return check_and_freshen(oid, 1);
1692 }
1693
1694 static int freshen_packed_object(const struct object_id *oid)
1695 {
1696         struct pack_entry e;
1697         if (!find_pack_entry(the_repository, oid, &e))
1698                 return 0;
1699         if (e.p->freshened)
1700                 return 1;
1701         if (!freshen_file(e.p->pack_name))
1702                 return 0;
1703         e.p->freshened = 1;
1704         return 1;
1705 }
1706
1707 int write_object_file(const void *buf, unsigned long len, const char *type,
1708                       struct object_id *oid)
1709 {
1710         char hdr[MAX_HEADER_LEN];
1711         int hdrlen = sizeof(hdr);
1712
1713         /* Normally if we have it in the pack then we do not bother writing
1714          * it out into .git/objects/??/?{38} file.
1715          */
1716         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1717         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1718                 return 0;
1719         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1720 }
1721
1722 int hash_object_file_literally(const void *buf, unsigned long len,
1723                                const char *type, struct object_id *oid,
1724                                unsigned flags)
1725 {
1726         char *header;
1727         int hdrlen, status = 0;
1728
1729         /* type string, SP, %lu of the length plus NUL must fit this */
1730         hdrlen = strlen(type) + MAX_HEADER_LEN;
1731         header = xmalloc(hdrlen);
1732         write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1733
1734         if (!(flags & HASH_WRITE_OBJECT))
1735                 goto cleanup;
1736         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1737                 goto cleanup;
1738         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1739
1740 cleanup:
1741         free(header);
1742         return status;
1743 }
1744
1745 int force_object_loose(const struct object_id *oid, time_t mtime)
1746 {
1747         void *buf;
1748         unsigned long len;
1749         enum object_type type;
1750         char hdr[MAX_HEADER_LEN];
1751         int hdrlen;
1752         int ret;
1753
1754         if (has_loose_object(oid))
1755                 return 0;
1756         buf = read_object(oid->hash, &type, &len);
1757         if (!buf)
1758                 return error(_("cannot read sha1_file for %s"), oid_to_hex(oid));
1759         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1760         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1761         free(buf);
1762
1763         return ret;
1764 }
1765
1766 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1767 {
1768         struct object_id oid;
1769         if (!startup_info->have_repository)
1770                 return 0;
1771         hashcpy(oid.hash, sha1);
1772         return oid_object_info_extended(the_repository, &oid, NULL,
1773                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1774 }
1775
1776 int has_object_file(const struct object_id *oid)
1777 {
1778         return has_sha1_file(oid->hash);
1779 }
1780
1781 int has_object_file_with_flags(const struct object_id *oid, int flags)
1782 {
1783         return has_sha1_file_with_flags(oid->hash, flags);
1784 }
1785
1786 static void check_tree(const void *buf, size_t size)
1787 {
1788         struct tree_desc desc;
1789         struct name_entry entry;
1790
1791         init_tree_desc(&desc, buf, size);
1792         while (tree_entry(&desc, &entry))
1793                 /* do nothing
1794                  * tree_entry() will die() on malformed entries */
1795                 ;
1796 }
1797
1798 static void check_commit(const void *buf, size_t size)
1799 {
1800         struct commit c;
1801         memset(&c, 0, sizeof(c));
1802         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1803                 die(_("corrupt commit"));
1804 }
1805
1806 static void check_tag(const void *buf, size_t size)
1807 {
1808         struct tag t;
1809         memset(&t, 0, sizeof(t));
1810         if (parse_tag_buffer(the_repository, &t, buf, size))
1811                 die(_("corrupt tag"));
1812 }
1813
1814 static int index_mem(struct index_state *istate,
1815                      struct object_id *oid, void *buf, size_t size,
1816                      enum object_type type,
1817                      const char *path, unsigned flags)
1818 {
1819         int ret, re_allocated = 0;
1820         int write_object = flags & HASH_WRITE_OBJECT;
1821
1822         if (!type)
1823                 type = OBJ_BLOB;
1824
1825         /*
1826          * Convert blobs to git internal format
1827          */
1828         if ((type == OBJ_BLOB) && path) {
1829                 struct strbuf nbuf = STRBUF_INIT;
1830                 if (convert_to_git(istate, path, buf, size, &nbuf,
1831                                    get_conv_flags(flags))) {
1832                         buf = strbuf_detach(&nbuf, &size);
1833                         re_allocated = 1;
1834                 }
1835         }
1836         if (flags & HASH_FORMAT_CHECK) {
1837                 if (type == OBJ_TREE)
1838                         check_tree(buf, size);
1839                 if (type == OBJ_COMMIT)
1840                         check_commit(buf, size);
1841                 if (type == OBJ_TAG)
1842                         check_tag(buf, size);
1843         }
1844
1845         if (write_object)
1846                 ret = write_object_file(buf, size, type_name(type), oid);
1847         else
1848                 ret = hash_object_file(buf, size, type_name(type), oid);
1849         if (re_allocated)
1850                 free(buf);
1851         return ret;
1852 }
1853
1854 static int index_stream_convert_blob(struct index_state *istate,
1855                                      struct object_id *oid,
1856                                      int fd,
1857                                      const char *path,
1858                                      unsigned flags)
1859 {
1860         int ret;
1861         const int write_object = flags & HASH_WRITE_OBJECT;
1862         struct strbuf sbuf = STRBUF_INIT;
1863
1864         assert(path);
1865         assert(would_convert_to_git_filter_fd(istate, path));
1866
1867         convert_to_git_filter_fd(istate, path, fd, &sbuf,
1868                                  get_conv_flags(flags));
1869
1870         if (write_object)
1871                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1872                                         oid);
1873         else
1874                 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1875                                        oid);
1876         strbuf_release(&sbuf);
1877         return ret;
1878 }
1879
1880 static int index_pipe(struct index_state *istate, struct object_id *oid,
1881                       int fd, enum object_type type,
1882                       const char *path, unsigned flags)
1883 {
1884         struct strbuf sbuf = STRBUF_INIT;
1885         int ret;
1886
1887         if (strbuf_read(&sbuf, fd, 4096) >= 0)
1888                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
1889         else
1890                 ret = -1;
1891         strbuf_release(&sbuf);
1892         return ret;
1893 }
1894
1895 #define SMALL_FILE_SIZE (32*1024)
1896
1897 static int index_core(struct index_state *istate,
1898                       struct object_id *oid, int fd, size_t size,
1899                       enum object_type type, const char *path,
1900                       unsigned flags)
1901 {
1902         int ret;
1903
1904         if (!size) {
1905                 ret = index_mem(istate, oid, "", size, type, path, flags);
1906         } else if (size <= SMALL_FILE_SIZE) {
1907                 char *buf = xmalloc(size);
1908                 ssize_t read_result = read_in_full(fd, buf, size);
1909                 if (read_result < 0)
1910                         ret = error_errno(_("read error while indexing %s"),
1911                                           path ? path : "<unknown>");
1912                 else if (read_result != size)
1913                         ret = error(_("short read while indexing %s"),
1914                                     path ? path : "<unknown>");
1915                 else
1916                         ret = index_mem(istate, oid, buf, size, type, path, flags);
1917                 free(buf);
1918         } else {
1919                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1920                 ret = index_mem(istate, oid, buf, size, type, path, flags);
1921                 munmap(buf, size);
1922         }
1923         return ret;
1924 }
1925
1926 /*
1927  * This creates one packfile per large blob unless bulk-checkin
1928  * machinery is "plugged".
1929  *
1930  * This also bypasses the usual "convert-to-git" dance, and that is on
1931  * purpose. We could write a streaming version of the converting
1932  * functions and insert that before feeding the data to fast-import
1933  * (or equivalent in-core API described above). However, that is
1934  * somewhat complicated, as we do not know the size of the filter
1935  * result, which we need to know beforehand when writing a git object.
1936  * Since the primary motivation for trying to stream from the working
1937  * tree file and to avoid mmaping it in core is to deal with large
1938  * binary blobs, they generally do not want to get any conversion, and
1939  * callers should avoid this code path when filters are requested.
1940  */
1941 static int index_stream(struct object_id *oid, int fd, size_t size,
1942                         enum object_type type, const char *path,
1943                         unsigned flags)
1944 {
1945         return index_bulk_checkin(oid, fd, size, type, path, flags);
1946 }
1947
1948 int index_fd(struct index_state *istate, struct object_id *oid,
1949              int fd, struct stat *st,
1950              enum object_type type, const char *path, unsigned flags)
1951 {
1952         int ret;
1953
1954         /*
1955          * Call xsize_t() only when needed to avoid potentially unnecessary
1956          * die() for large files.
1957          */
1958         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
1959                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
1960         else if (!S_ISREG(st->st_mode))
1961                 ret = index_pipe(istate, oid, fd, type, path, flags);
1962         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1963                  (path && would_convert_to_git(istate, path)))
1964                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
1965                                  type, path, flags);
1966         else
1967                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1968                                    flags);
1969         close(fd);
1970         return ret;
1971 }
1972
1973 int index_path(struct index_state *istate, struct object_id *oid,
1974                const char *path, struct stat *st, unsigned flags)
1975 {
1976         int fd;
1977         struct strbuf sb = STRBUF_INIT;
1978         int rc = 0;
1979
1980         switch (st->st_mode & S_IFMT) {
1981         case S_IFREG:
1982                 fd = open(path, O_RDONLY);
1983                 if (fd < 0)
1984                         return error_errno("open(\"%s\")", path);
1985                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
1986                         return error(_("%s: failed to insert into database"),
1987                                      path);
1988                 break;
1989         case S_IFLNK:
1990                 if (strbuf_readlink(&sb, path, st->st_size))
1991                         return error_errno("readlink(\"%s\")", path);
1992                 if (!(flags & HASH_WRITE_OBJECT))
1993                         hash_object_file(sb.buf, sb.len, blob_type, oid);
1994                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1995                         rc = error(_("%s: failed to insert into database"), path);
1996                 strbuf_release(&sb);
1997                 break;
1998         case S_IFDIR:
1999                 return resolve_gitlink_ref(path, "HEAD", oid);
2000         default:
2001                 return error(_("%s: unsupported file type"), path);
2002         }
2003         return rc;
2004 }
2005
2006 int read_pack_header(int fd, struct pack_header *header)
2007 {
2008         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2009                 /* "eof before pack header was fully read" */
2010                 return PH_ERROR_EOF;
2011
2012         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2013                 /* "protocol error (pack signature mismatch detected)" */
2014                 return PH_ERROR_PACK_SIGNATURE;
2015         if (!pack_version_ok(header->hdr_version))
2016                 /* "protocol error (pack version unsupported)" */
2017                 return PH_ERROR_PROTOCOL;
2018         return 0;
2019 }
2020
2021 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2022 {
2023         enum object_type type = oid_object_info(the_repository, oid, NULL);
2024         if (type < 0)
2025                 die(_("%s is not a valid object"), oid_to_hex(oid));
2026         if (type != expect)
2027                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2028                     type_name(expect));
2029 }
2030
2031 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2032                                 struct strbuf *path,
2033                                 each_loose_object_fn obj_cb,
2034                                 each_loose_cruft_fn cruft_cb,
2035                                 each_loose_subdir_fn subdir_cb,
2036                                 void *data)
2037 {
2038         size_t origlen, baselen;
2039         DIR *dir;
2040         struct dirent *de;
2041         int r = 0;
2042         struct object_id oid;
2043
2044         if (subdir_nr > 0xff)
2045                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2046
2047         origlen = path->len;
2048         strbuf_complete(path, '/');
2049         strbuf_addf(path, "%02x", subdir_nr);
2050
2051         dir = opendir(path->buf);
2052         if (!dir) {
2053                 if (errno != ENOENT)
2054                         r = error_errno(_("unable to open %s"), path->buf);
2055                 strbuf_setlen(path, origlen);
2056                 return r;
2057         }
2058
2059         oid.hash[0] = subdir_nr;
2060         strbuf_addch(path, '/');
2061         baselen = path->len;
2062
2063         while ((de = readdir(dir))) {
2064                 size_t namelen;
2065                 if (is_dot_or_dotdot(de->d_name))
2066                         continue;
2067
2068                 namelen = strlen(de->d_name);
2069                 strbuf_setlen(path, baselen);
2070                 strbuf_add(path, de->d_name, namelen);
2071                 if (namelen == the_hash_algo->hexsz - 2 &&
2072                     !hex_to_bytes(oid.hash + 1, de->d_name,
2073                                   the_hash_algo->rawsz - 1)) {
2074                         if (obj_cb) {
2075                                 r = obj_cb(&oid, path->buf, data);
2076                                 if (r)
2077                                         break;
2078                         }
2079                         continue;
2080                 }
2081
2082                 if (cruft_cb) {
2083                         r = cruft_cb(de->d_name, path->buf, data);
2084                         if (r)
2085                                 break;
2086                 }
2087         }
2088         closedir(dir);
2089
2090         strbuf_setlen(path, baselen - 1);
2091         if (!r && subdir_cb)
2092                 r = subdir_cb(subdir_nr, path->buf, data);
2093
2094         strbuf_setlen(path, origlen);
2095
2096         return r;
2097 }
2098
2099 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2100                             each_loose_object_fn obj_cb,
2101                             each_loose_cruft_fn cruft_cb,
2102                             each_loose_subdir_fn subdir_cb,
2103                             void *data)
2104 {
2105         int r = 0;
2106         int i;
2107
2108         for (i = 0; i < 256; i++) {
2109                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2110                                                 subdir_cb, data);
2111                 if (r)
2112                         break;
2113         }
2114
2115         return r;
2116 }
2117
2118 int for_each_loose_file_in_objdir(const char *path,
2119                                   each_loose_object_fn obj_cb,
2120                                   each_loose_cruft_fn cruft_cb,
2121                                   each_loose_subdir_fn subdir_cb,
2122                                   void *data)
2123 {
2124         struct strbuf buf = STRBUF_INIT;
2125         int r;
2126
2127         strbuf_addstr(&buf, path);
2128         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2129                                               subdir_cb, data);
2130         strbuf_release(&buf);
2131
2132         return r;
2133 }
2134
2135 struct loose_alt_odb_data {
2136         each_loose_object_fn *cb;
2137         void *data;
2138 };
2139
2140 static int loose_from_alt_odb(struct object_directory *odb,
2141                               void *vdata)
2142 {
2143         struct loose_alt_odb_data *data = vdata;
2144         struct strbuf buf = STRBUF_INIT;
2145         int r;
2146
2147         strbuf_addstr(&buf, odb->path);
2148         r = for_each_loose_file_in_objdir_buf(&buf,
2149                                               data->cb, NULL, NULL,
2150                                               data->data);
2151         strbuf_release(&buf);
2152         return r;
2153 }
2154
2155 int for_each_loose_object(each_loose_object_fn cb, void *data,
2156                           enum for_each_object_flags flags)
2157 {
2158         struct loose_alt_odb_data alt;
2159         int r;
2160
2161         r = for_each_loose_file_in_objdir(get_object_directory(),
2162                                           cb, NULL, NULL, data);
2163         if (r)
2164                 return r;
2165
2166         if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2167                 return 0;
2168
2169         alt.cb = cb;
2170         alt.data = data;
2171         return foreach_alt_odb(loose_from_alt_odb, &alt);
2172 }
2173
2174 static int check_stream_sha1(git_zstream *stream,
2175                              const char *hdr,
2176                              unsigned long size,
2177                              const char *path,
2178                              const unsigned char *expected_sha1)
2179 {
2180         git_hash_ctx c;
2181         unsigned char real_sha1[GIT_MAX_RAWSZ];
2182         unsigned char buf[4096];
2183         unsigned long total_read;
2184         int status = Z_OK;
2185
2186         the_hash_algo->init_fn(&c);
2187         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2188
2189         /*
2190          * We already read some bytes into hdr, but the ones up to the NUL
2191          * do not count against the object's content size.
2192          */
2193         total_read = stream->total_out - strlen(hdr) - 1;
2194
2195         /*
2196          * This size comparison must be "<=" to read the final zlib packets;
2197          * see the comment in unpack_sha1_rest for details.
2198          */
2199         while (total_read <= size &&
2200                (status == Z_OK || status == Z_BUF_ERROR)) {
2201                 stream->next_out = buf;
2202                 stream->avail_out = sizeof(buf);
2203                 if (size - total_read < stream->avail_out)
2204                         stream->avail_out = size - total_read;
2205                 status = git_inflate(stream, Z_FINISH);
2206                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2207                 total_read += stream->next_out - buf;
2208         }
2209         git_inflate_end(stream);
2210
2211         if (status != Z_STREAM_END) {
2212                 error(_("corrupt loose object '%s'"), sha1_to_hex(expected_sha1));
2213                 return -1;
2214         }
2215         if (stream->avail_in) {
2216                 error(_("garbage at end of loose object '%s'"),
2217                       sha1_to_hex(expected_sha1));
2218                 return -1;
2219         }
2220
2221         the_hash_algo->final_fn(real_sha1, &c);
2222         if (!hasheq(expected_sha1, real_sha1)) {
2223                 error(_("sha1 mismatch for %s (expected %s)"), path,
2224                       sha1_to_hex(expected_sha1));
2225                 return -1;
2226         }
2227
2228         return 0;
2229 }
2230
2231 int read_loose_object(const char *path,
2232                       const struct object_id *expected_oid,
2233                       enum object_type *type,
2234                       unsigned long *size,
2235                       void **contents)
2236 {
2237         int ret = -1;
2238         void *map = NULL;
2239         unsigned long mapsize;
2240         git_zstream stream;
2241         char hdr[MAX_HEADER_LEN];
2242
2243         *contents = NULL;
2244
2245         map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2246         if (!map) {
2247                 error_errno(_("unable to mmap %s"), path);
2248                 goto out;
2249         }
2250
2251         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2252                 error(_("unable to unpack header of %s"), path);
2253                 goto out;
2254         }
2255
2256         *type = parse_sha1_header(hdr, size);
2257         if (*type < 0) {
2258                 error(_("unable to parse header of %s"), path);
2259                 git_inflate_end(&stream);
2260                 goto out;
2261         }
2262
2263         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2264                 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2265                         goto out;
2266         } else {
2267                 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2268                 if (!*contents) {
2269                         error(_("unable to unpack contents of %s"), path);
2270                         git_inflate_end(&stream);
2271                         goto out;
2272                 }
2273                 if (check_object_signature(expected_oid, *contents,
2274                                          *size, type_name(*type))) {
2275                         error(_("sha1 mismatch for %s (expected %s)"), path,
2276                               oid_to_hex(expected_oid));
2277                         free(*contents);
2278                         goto out;
2279                 }
2280         }
2281
2282         ret = 0; /* everything checks out */
2283
2284 out:
2285         if (map)
2286                 munmap(map, mapsize);
2287         return ret;
2288 }