Merge branch 'ds/gc-doc-typofix'
[git] / 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 %"PRIuMAX , type_name(obj_type), (uintmax_t)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 static int quick_has_loose(struct repository *r,
925                            const unsigned char *sha1)
926 {
927         int subdir_nr = sha1[0];
928         struct object_id oid;
929         struct object_directory *odb;
930
931         hashcpy(oid.hash, sha1);
932
933         prepare_alt_odb(r);
934         for (odb = r->objects->odb; odb; odb = odb->next) {
935                 odb_load_loose_cache(odb, subdir_nr);
936                 if (oid_array_lookup(&odb->loose_objects_cache, &oid) >= 0)
937                         return 1;
938         }
939         return 0;
940 }
941
942 /*
943  * Map the loose object at "path" if it is not NULL, or the path found by
944  * searching for a loose object named "sha1".
945  */
946 static void *map_sha1_file_1(struct repository *r, const char *path,
947                              const unsigned char *sha1, unsigned long *size)
948 {
949         void *map;
950         int fd;
951
952         if (path)
953                 fd = git_open(path);
954         else
955                 fd = open_sha1_file(r, sha1, &path);
956         map = NULL;
957         if (fd >= 0) {
958                 struct stat st;
959
960                 if (!fstat(fd, &st)) {
961                         *size = xsize_t(st.st_size);
962                         if (!*size) {
963                                 /* mmap() is forbidden on empty files */
964                                 error(_("object file %s is empty"), path);
965                                 return NULL;
966                         }
967                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
968                 }
969                 close(fd);
970         }
971         return map;
972 }
973
974 void *map_sha1_file(struct repository *r,
975                     const unsigned char *sha1, unsigned long *size)
976 {
977         return map_sha1_file_1(r, NULL, sha1, size);
978 }
979
980 static int unpack_sha1_short_header(git_zstream *stream,
981                                     unsigned char *map, unsigned long mapsize,
982                                     void *buffer, unsigned long bufsiz)
983 {
984         /* Get the data stream */
985         memset(stream, 0, sizeof(*stream));
986         stream->next_in = map;
987         stream->avail_in = mapsize;
988         stream->next_out = buffer;
989         stream->avail_out = bufsiz;
990
991         git_inflate_init(stream);
992         return git_inflate(stream, 0);
993 }
994
995 int unpack_sha1_header(git_zstream *stream,
996                        unsigned char *map, unsigned long mapsize,
997                        void *buffer, unsigned long bufsiz)
998 {
999         int status = unpack_sha1_short_header(stream, map, mapsize,
1000                                               buffer, bufsiz);
1001
1002         if (status < Z_OK)
1003                 return status;
1004
1005         /* Make sure we have the terminating NUL */
1006         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1007                 return -1;
1008         return 0;
1009 }
1010
1011 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1012                                         unsigned long mapsize, void *buffer,
1013                                         unsigned long bufsiz, struct strbuf *header)
1014 {
1015         int status;
1016
1017         status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1018         if (status < Z_OK)
1019                 return -1;
1020
1021         /*
1022          * Check if entire header is unpacked in the first iteration.
1023          */
1024         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1025                 return 0;
1026
1027         /*
1028          * buffer[0..bufsiz] was not large enough.  Copy the partial
1029          * result out to header, and then append the result of further
1030          * reading the stream.
1031          */
1032         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1033         stream->next_out = buffer;
1034         stream->avail_out = bufsiz;
1035
1036         do {
1037                 status = git_inflate(stream, 0);
1038                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1039                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1040                         return 0;
1041                 stream->next_out = buffer;
1042                 stream->avail_out = bufsiz;
1043         } while (status != Z_STREAM_END);
1044         return -1;
1045 }
1046
1047 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1048 {
1049         int bytes = strlen(buffer) + 1;
1050         unsigned char *buf = xmallocz(size);
1051         unsigned long n;
1052         int status = Z_OK;
1053
1054         n = stream->total_out - bytes;
1055         if (n > size)
1056                 n = size;
1057         memcpy(buf, (char *) buffer + bytes, n);
1058         bytes = n;
1059         if (bytes <= size) {
1060                 /*
1061                  * The above condition must be (bytes <= size), not
1062                  * (bytes < size).  In other words, even though we
1063                  * expect no more output and set avail_out to zero,
1064                  * the input zlib stream may have bytes that express
1065                  * "this concludes the stream", and we *do* want to
1066                  * eat that input.
1067                  *
1068                  * Otherwise we would not be able to test that we
1069                  * consumed all the input to reach the expected size;
1070                  * we also want to check that zlib tells us that all
1071                  * went well with status == Z_STREAM_END at the end.
1072                  */
1073                 stream->next_out = buf + bytes;
1074                 stream->avail_out = size - bytes;
1075                 while (status == Z_OK)
1076                         status = git_inflate(stream, Z_FINISH);
1077         }
1078         if (status == Z_STREAM_END && !stream->avail_in) {
1079                 git_inflate_end(stream);
1080                 return buf;
1081         }
1082
1083         if (status < 0)
1084                 error(_("corrupt loose object '%s'"), sha1_to_hex(sha1));
1085         else if (stream->avail_in)
1086                 error(_("garbage at end of loose object '%s'"),
1087                       sha1_to_hex(sha1));
1088         free(buf);
1089         return NULL;
1090 }
1091
1092 /*
1093  * We used to just use "sscanf()", but that's actually way
1094  * too permissive for what we want to check. So do an anal
1095  * object header parse by hand.
1096  */
1097 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1098                                unsigned int flags)
1099 {
1100         const char *type_buf = hdr;
1101         unsigned long size;
1102         int type, type_len = 0;
1103
1104         /*
1105          * The type can be of any size but is followed by
1106          * a space.
1107          */
1108         for (;;) {
1109                 char c = *hdr++;
1110                 if (!c)
1111                         return -1;
1112                 if (c == ' ')
1113                         break;
1114                 type_len++;
1115         }
1116
1117         type = type_from_string_gently(type_buf, type_len, 1);
1118         if (oi->type_name)
1119                 strbuf_add(oi->type_name, type_buf, type_len);
1120         /*
1121          * Set type to 0 if its an unknown object and
1122          * we're obtaining the type using '--allow-unknown-type'
1123          * option.
1124          */
1125         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1126                 type = 0;
1127         else if (type < 0)
1128                 die(_("invalid object type"));
1129         if (oi->typep)
1130                 *oi->typep = type;
1131
1132         /*
1133          * The length must follow immediately, and be in canonical
1134          * decimal format (ie "010" is not valid).
1135          */
1136         size = *hdr++ - '0';
1137         if (size > 9)
1138                 return -1;
1139         if (size) {
1140                 for (;;) {
1141                         unsigned long c = *hdr - '0';
1142                         if (c > 9)
1143                                 break;
1144                         hdr++;
1145                         size = size * 10 + c;
1146                 }
1147         }
1148
1149         if (oi->sizep)
1150                 *oi->sizep = size;
1151
1152         /*
1153          * The length must be followed by a zero byte
1154          */
1155         return *hdr ? -1 : type;
1156 }
1157
1158 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1159 {
1160         struct object_info oi = OBJECT_INFO_INIT;
1161
1162         oi.sizep = sizep;
1163         return parse_sha1_header_extended(hdr, &oi, 0);
1164 }
1165
1166 static int sha1_loose_object_info(struct repository *r,
1167                                   const unsigned char *sha1,
1168                                   struct object_info *oi, int flags)
1169 {
1170         int status = 0;
1171         unsigned long mapsize;
1172         void *map;
1173         git_zstream stream;
1174         char hdr[MAX_HEADER_LEN];
1175         struct strbuf hdrbuf = STRBUF_INIT;
1176         unsigned long size_scratch;
1177
1178         if (oi->delta_base_sha1)
1179                 hashclr(oi->delta_base_sha1);
1180
1181         /*
1182          * If we don't care about type or size, then we don't
1183          * need to look inside the object at all. Note that we
1184          * do not optimize out the stat call, even if the
1185          * caller doesn't care about the disk-size, since our
1186          * return value implicitly indicates whether the
1187          * object even exists.
1188          */
1189         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1190                 const char *path;
1191                 struct stat st;
1192                 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1193                         return quick_has_loose(r, sha1) ? 0 : -1;
1194                 if (stat_sha1_file(r, sha1, &st, &path) < 0)
1195                         return -1;
1196                 if (oi->disk_sizep)
1197                         *oi->disk_sizep = st.st_size;
1198                 return 0;
1199         }
1200
1201         map = map_sha1_file(r, sha1, &mapsize);
1202         if (!map)
1203                 return -1;
1204
1205         if (!oi->sizep)
1206                 oi->sizep = &size_scratch;
1207
1208         if (oi->disk_sizep)
1209                 *oi->disk_sizep = mapsize;
1210         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1211                 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1212                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1213                                        sha1_to_hex(sha1));
1214         } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1215                 status = error(_("unable to unpack %s header"),
1216                                sha1_to_hex(sha1));
1217         if (status < 0)
1218                 ; /* Do nothing */
1219         else if (hdrbuf.len) {
1220                 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1221                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1222                                        sha1_to_hex(sha1));
1223         } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1224                 status = error(_("unable to parse %s header"), sha1_to_hex(sha1));
1225
1226         if (status >= 0 && oi->contentp) {
1227                 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1228                                                  *oi->sizep, sha1);
1229                 if (!*oi->contentp) {
1230                         git_inflate_end(&stream);
1231                         status = -1;
1232                 }
1233         } else
1234                 git_inflate_end(&stream);
1235
1236         munmap(map, mapsize);
1237         if (status && oi->typep)
1238                 *oi->typep = status;
1239         if (oi->sizep == &size_scratch)
1240                 oi->sizep = NULL;
1241         strbuf_release(&hdrbuf);
1242         oi->whence = OI_LOOSE;
1243         return (status < 0) ? status : 0;
1244 }
1245
1246 int fetch_if_missing = 1;
1247
1248 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1249                              struct object_info *oi, unsigned flags)
1250 {
1251         static struct object_info blank_oi = OBJECT_INFO_INIT;
1252         struct pack_entry e;
1253         int rtype;
1254         const struct object_id *real = oid;
1255         int already_retried = 0;
1256
1257         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1258                 real = lookup_replace_object(r, oid);
1259
1260         if (is_null_oid(real))
1261                 return -1;
1262
1263         if (!oi)
1264                 oi = &blank_oi;
1265
1266         if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1267                 struct cached_object *co = find_cached_object(real);
1268                 if (co) {
1269                         if (oi->typep)
1270                                 *(oi->typep) = co->type;
1271                         if (oi->sizep)
1272                                 *(oi->sizep) = co->size;
1273                         if (oi->disk_sizep)
1274                                 *(oi->disk_sizep) = 0;
1275                         if (oi->delta_base_sha1)
1276                                 hashclr(oi->delta_base_sha1);
1277                         if (oi->type_name)
1278                                 strbuf_addstr(oi->type_name, type_name(co->type));
1279                         if (oi->contentp)
1280                                 *oi->contentp = xmemdupz(co->buf, co->size);
1281                         oi->whence = OI_CACHED;
1282                         return 0;
1283                 }
1284         }
1285
1286         while (1) {
1287                 if (find_pack_entry(r, real, &e))
1288                         break;
1289
1290                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1291                         return -1;
1292
1293                 /* Most likely it's a loose object. */
1294                 if (!sha1_loose_object_info(r, real->hash, oi, flags))
1295                         return 0;
1296
1297                 /* Not a loose object; someone else may have just packed it. */
1298                 if (!(flags & OBJECT_INFO_QUICK)) {
1299                         reprepare_packed_git(r);
1300                         if (find_pack_entry(r, real, &e))
1301                                 break;
1302                 }
1303
1304                 /* Check if it is a missing object */
1305                 if (fetch_if_missing && repository_format_partial_clone &&
1306                     !already_retried && r == the_repository) {
1307                         /*
1308                          * TODO Investigate having fetch_object() return
1309                          * TODO error/success and stopping the music here.
1310                          * TODO Pass a repository struct through fetch_object,
1311                          * such that arbitrary repositories work.
1312                          */
1313                         fetch_objects(repository_format_partial_clone, real, 1);
1314                         already_retried = 1;
1315                         continue;
1316                 }
1317
1318                 return -1;
1319         }
1320
1321         if (oi == &blank_oi)
1322                 /*
1323                  * We know that the caller doesn't actually need the
1324                  * information below, so return early.
1325                  */
1326                 return 0;
1327         rtype = packed_object_info(r, e.p, e.offset, oi);
1328         if (rtype < 0) {
1329                 mark_bad_packed_object(e.p, real->hash);
1330                 return oid_object_info_extended(r, real, oi, 0);
1331         } else if (oi->whence == OI_PACKED) {
1332                 oi->u.packed.offset = e.offset;
1333                 oi->u.packed.pack = e.p;
1334                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1335                                          rtype == OBJ_OFS_DELTA);
1336         }
1337
1338         return 0;
1339 }
1340
1341 /* returns enum object_type or negative */
1342 int oid_object_info(struct repository *r,
1343                     const struct object_id *oid,
1344                     unsigned long *sizep)
1345 {
1346         enum object_type type;
1347         struct object_info oi = OBJECT_INFO_INIT;
1348
1349         oi.typep = &type;
1350         oi.sizep = sizep;
1351         if (oid_object_info_extended(r, oid, &oi,
1352                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1353                 return -1;
1354         return type;
1355 }
1356
1357 static void *read_object(const unsigned char *sha1, enum object_type *type,
1358                          unsigned long *size)
1359 {
1360         struct object_id oid;
1361         struct object_info oi = OBJECT_INFO_INIT;
1362         void *content;
1363         oi.typep = type;
1364         oi.sizep = size;
1365         oi.contentp = &content;
1366
1367         hashcpy(oid.hash, sha1);
1368
1369         if (oid_object_info_extended(the_repository, &oid, &oi, 0) < 0)
1370                 return NULL;
1371         return content;
1372 }
1373
1374 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1375                         struct object_id *oid)
1376 {
1377         struct cached_object *co;
1378
1379         hash_object_file(buf, len, type_name(type), oid);
1380         if (has_sha1_file(oid->hash) || find_cached_object(oid))
1381                 return 0;
1382         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1383         co = &cached_objects[cached_object_nr++];
1384         co->size = len;
1385         co->type = type;
1386         co->buf = xmalloc(len);
1387         memcpy(co->buf, buf, len);
1388         oidcpy(&co->oid, oid);
1389         return 0;
1390 }
1391
1392 /*
1393  * This function dies on corrupt objects; the callers who want to
1394  * deal with them should arrange to call read_object() and give error
1395  * messages themselves.
1396  */
1397 void *read_object_file_extended(const struct object_id *oid,
1398                                 enum object_type *type,
1399                                 unsigned long *size,
1400                                 int lookup_replace)
1401 {
1402         void *data;
1403         const struct packed_git *p;
1404         const char *path;
1405         struct stat st;
1406         const struct object_id *repl = lookup_replace ?
1407                 lookup_replace_object(the_repository, oid) : oid;
1408
1409         errno = 0;
1410         data = read_object(repl->hash, type, size);
1411         if (data)
1412                 return data;
1413
1414         if (errno && errno != ENOENT)
1415                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1416
1417         /* die if we replaced an object with one that does not exist */
1418         if (repl != oid)
1419                 die(_("replacement %s not found for %s"),
1420                     oid_to_hex(repl), oid_to_hex(oid));
1421
1422         if (!stat_sha1_file(the_repository, repl->hash, &st, &path))
1423                 die(_("loose object %s (stored in %s) is corrupt"),
1424                     oid_to_hex(repl), path);
1425
1426         if ((p = has_packed_and_bad(repl->hash)) != NULL)
1427                 die(_("packed object %s (stored in %s) is corrupt"),
1428                     oid_to_hex(repl), p->pack_name);
1429
1430         return NULL;
1431 }
1432
1433 void *read_object_with_reference(const struct object_id *oid,
1434                                  const char *required_type_name,
1435                                  unsigned long *size,
1436                                  struct object_id *actual_oid_return)
1437 {
1438         enum object_type type, required_type;
1439         void *buffer;
1440         unsigned long isize;
1441         struct object_id actual_oid;
1442
1443         required_type = type_from_string(required_type_name);
1444         oidcpy(&actual_oid, oid);
1445         while (1) {
1446                 int ref_length = -1;
1447                 const char *ref_type = NULL;
1448
1449                 buffer = read_object_file(&actual_oid, &type, &isize);
1450                 if (!buffer)
1451                         return NULL;
1452                 if (type == required_type) {
1453                         *size = isize;
1454                         if (actual_oid_return)
1455                                 oidcpy(actual_oid_return, &actual_oid);
1456                         return buffer;
1457                 }
1458                 /* Handle references */
1459                 else if (type == OBJ_COMMIT)
1460                         ref_type = "tree ";
1461                 else if (type == OBJ_TAG)
1462                         ref_type = "object ";
1463                 else {
1464                         free(buffer);
1465                         return NULL;
1466                 }
1467                 ref_length = strlen(ref_type);
1468
1469                 if (ref_length + the_hash_algo->hexsz > isize ||
1470                     memcmp(buffer, ref_type, ref_length) ||
1471                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1472                         free(buffer);
1473                         return NULL;
1474                 }
1475                 free(buffer);
1476                 /* Now we have the ID of the referred-to object in
1477                  * actual_oid.  Check again. */
1478         }
1479 }
1480
1481 static void write_object_file_prepare(const void *buf, unsigned long len,
1482                                       const char *type, struct object_id *oid,
1483                                       char *hdr, int *hdrlen)
1484 {
1485         git_hash_ctx c;
1486
1487         /* Generate the header */
1488         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1489
1490         /* Sha1.. */
1491         the_hash_algo->init_fn(&c);
1492         the_hash_algo->update_fn(&c, hdr, *hdrlen);
1493         the_hash_algo->update_fn(&c, buf, len);
1494         the_hash_algo->final_fn(oid->hash, &c);
1495 }
1496
1497 /*
1498  * Move the just written object into its final resting place.
1499  */
1500 int finalize_object_file(const char *tmpfile, const char *filename)
1501 {
1502         int ret = 0;
1503
1504         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1505                 goto try_rename;
1506         else if (link(tmpfile, filename))
1507                 ret = errno;
1508
1509         /*
1510          * Coda hack - coda doesn't like cross-directory links,
1511          * so we fall back to a rename, which will mean that it
1512          * won't be able to check collisions, but that's not a
1513          * big deal.
1514          *
1515          * The same holds for FAT formatted media.
1516          *
1517          * When this succeeds, we just return.  We have nothing
1518          * left to unlink.
1519          */
1520         if (ret && ret != EEXIST) {
1521         try_rename:
1522                 if (!rename(tmpfile, filename))
1523                         goto out;
1524                 ret = errno;
1525         }
1526         unlink_or_warn(tmpfile);
1527         if (ret) {
1528                 if (ret != EEXIST) {
1529                         return error_errno(_("unable to write sha1 filename %s"), filename);
1530                 }
1531                 /* FIXME!!! Collision check here ? */
1532         }
1533
1534 out:
1535         if (adjust_shared_perm(filename))
1536                 return error(_("unable to set permission to '%s'"), filename);
1537         return 0;
1538 }
1539
1540 static int write_buffer(int fd, const void *buf, size_t len)
1541 {
1542         if (write_in_full(fd, buf, len) < 0)
1543                 return error_errno(_("file write error"));
1544         return 0;
1545 }
1546
1547 int hash_object_file(const void *buf, unsigned long len, const char *type,
1548                      struct object_id *oid)
1549 {
1550         char hdr[MAX_HEADER_LEN];
1551         int hdrlen = sizeof(hdr);
1552         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1553         return 0;
1554 }
1555
1556 /* Finalize a file on disk, and close it. */
1557 static void close_sha1_file(int fd)
1558 {
1559         if (fsync_object_files)
1560                 fsync_or_die(fd, "sha1 file");
1561         if (close(fd) != 0)
1562                 die_errno(_("error when closing sha1 file"));
1563 }
1564
1565 /* Size of directory component, including the ending '/' */
1566 static inline int directory_size(const char *filename)
1567 {
1568         const char *s = strrchr(filename, '/');
1569         if (!s)
1570                 return 0;
1571         return s - filename + 1;
1572 }
1573
1574 /*
1575  * This creates a temporary file in the same directory as the final
1576  * 'filename'
1577  *
1578  * We want to avoid cross-directory filename renames, because those
1579  * can have problems on various filesystems (FAT, NFS, Coda).
1580  */
1581 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1582 {
1583         int fd, dirlen = directory_size(filename);
1584
1585         strbuf_reset(tmp);
1586         strbuf_add(tmp, filename, dirlen);
1587         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1588         fd = git_mkstemp_mode(tmp->buf, 0444);
1589         if (fd < 0 && dirlen && errno == ENOENT) {
1590                 /*
1591                  * Make sure the directory exists; note that the contents
1592                  * of the buffer are undefined after mkstemp returns an
1593                  * error, so we have to rewrite the whole buffer from
1594                  * scratch.
1595                  */
1596                 strbuf_reset(tmp);
1597                 strbuf_add(tmp, filename, dirlen - 1);
1598                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1599                         return -1;
1600                 if (adjust_shared_perm(tmp->buf))
1601                         return -1;
1602
1603                 /* Try again */
1604                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1605                 fd = git_mkstemp_mode(tmp->buf, 0444);
1606         }
1607         return fd;
1608 }
1609
1610 static int write_loose_object(const struct object_id *oid, char *hdr,
1611                               int hdrlen, const void *buf, unsigned long len,
1612                               time_t mtime)
1613 {
1614         int fd, ret;
1615         unsigned char compressed[4096];
1616         git_zstream stream;
1617         git_hash_ctx c;
1618         struct object_id parano_oid;
1619         static struct strbuf tmp_file = STRBUF_INIT;
1620         static struct strbuf filename = STRBUF_INIT;
1621
1622         loose_object_path(the_repository, &filename, oid->hash);
1623
1624         fd = create_tmpfile(&tmp_file, filename.buf);
1625         if (fd < 0) {
1626                 if (errno == EACCES)
1627                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1628                 else
1629                         return error_errno(_("unable to create temporary file"));
1630         }
1631
1632         /* Set it up */
1633         git_deflate_init(&stream, zlib_compression_level);
1634         stream.next_out = compressed;
1635         stream.avail_out = sizeof(compressed);
1636         the_hash_algo->init_fn(&c);
1637
1638         /* First header.. */
1639         stream.next_in = (unsigned char *)hdr;
1640         stream.avail_in = hdrlen;
1641         while (git_deflate(&stream, 0) == Z_OK)
1642                 ; /* nothing */
1643         the_hash_algo->update_fn(&c, hdr, hdrlen);
1644
1645         /* Then the data itself.. */
1646         stream.next_in = (void *)buf;
1647         stream.avail_in = len;
1648         do {
1649                 unsigned char *in0 = stream.next_in;
1650                 ret = git_deflate(&stream, Z_FINISH);
1651                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1652                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1653                         die(_("unable to write sha1 file"));
1654                 stream.next_out = compressed;
1655                 stream.avail_out = sizeof(compressed);
1656         } while (ret == Z_OK);
1657
1658         if (ret != Z_STREAM_END)
1659                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1660                     ret);
1661         ret = git_deflate_end_gently(&stream);
1662         if (ret != Z_OK)
1663                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1664                     ret);
1665         the_hash_algo->final_fn(parano_oid.hash, &c);
1666         if (!oideq(oid, &parano_oid))
1667                 die(_("confused by unstable object source data for %s"),
1668                     oid_to_hex(oid));
1669
1670         close_sha1_file(fd);
1671
1672         if (mtime) {
1673                 struct utimbuf utb;
1674                 utb.actime = mtime;
1675                 utb.modtime = mtime;
1676                 if (utime(tmp_file.buf, &utb) < 0)
1677                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1678         }
1679
1680         return finalize_object_file(tmp_file.buf, filename.buf);
1681 }
1682
1683 static int freshen_loose_object(const struct object_id *oid)
1684 {
1685         return check_and_freshen(oid, 1);
1686 }
1687
1688 static int freshen_packed_object(const struct object_id *oid)
1689 {
1690         struct pack_entry e;
1691         if (!find_pack_entry(the_repository, oid, &e))
1692                 return 0;
1693         if (e.p->freshened)
1694                 return 1;
1695         if (!freshen_file(e.p->pack_name))
1696                 return 0;
1697         e.p->freshened = 1;
1698         return 1;
1699 }
1700
1701 int write_object_file(const void *buf, unsigned long len, const char *type,
1702                       struct object_id *oid)
1703 {
1704         char hdr[MAX_HEADER_LEN];
1705         int hdrlen = sizeof(hdr);
1706
1707         /* Normally if we have it in the pack then we do not bother writing
1708          * it out into .git/objects/??/?{38} file.
1709          */
1710         write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1711         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1712                 return 0;
1713         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1714 }
1715
1716 int hash_object_file_literally(const void *buf, unsigned long len,
1717                                const char *type, struct object_id *oid,
1718                                unsigned flags)
1719 {
1720         char *header;
1721         int hdrlen, status = 0;
1722
1723         /* type string, SP, %lu of the length plus NUL must fit this */
1724         hdrlen = strlen(type) + MAX_HEADER_LEN;
1725         header = xmalloc(hdrlen);
1726         write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1727
1728         if (!(flags & HASH_WRITE_OBJECT))
1729                 goto cleanup;
1730         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1731                 goto cleanup;
1732         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1733
1734 cleanup:
1735         free(header);
1736         return status;
1737 }
1738
1739 int force_object_loose(const struct object_id *oid, time_t mtime)
1740 {
1741         void *buf;
1742         unsigned long len;
1743         enum object_type type;
1744         char hdr[MAX_HEADER_LEN];
1745         int hdrlen;
1746         int ret;
1747
1748         if (has_loose_object(oid))
1749                 return 0;
1750         buf = read_object(oid->hash, &type, &len);
1751         if (!buf)
1752                 return error(_("cannot read sha1_file for %s"), oid_to_hex(oid));
1753         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1754         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1755         free(buf);
1756
1757         return ret;
1758 }
1759
1760 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1761 {
1762         struct object_id oid;
1763         if (!startup_info->have_repository)
1764                 return 0;
1765         hashcpy(oid.hash, sha1);
1766         return oid_object_info_extended(the_repository, &oid, NULL,
1767                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1768 }
1769
1770 int has_object_file(const struct object_id *oid)
1771 {
1772         return has_sha1_file(oid->hash);
1773 }
1774
1775 int has_object_file_with_flags(const struct object_id *oid, int flags)
1776 {
1777         return has_sha1_file_with_flags(oid->hash, flags);
1778 }
1779
1780 static void check_tree(const void *buf, size_t size)
1781 {
1782         struct tree_desc desc;
1783         struct name_entry entry;
1784
1785         init_tree_desc(&desc, buf, size);
1786         while (tree_entry(&desc, &entry))
1787                 /* do nothing
1788                  * tree_entry() will die() on malformed entries */
1789                 ;
1790 }
1791
1792 static void check_commit(const void *buf, size_t size)
1793 {
1794         struct commit c;
1795         memset(&c, 0, sizeof(c));
1796         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1797                 die(_("corrupt commit"));
1798 }
1799
1800 static void check_tag(const void *buf, size_t size)
1801 {
1802         struct tag t;
1803         memset(&t, 0, sizeof(t));
1804         if (parse_tag_buffer(the_repository, &t, buf, size))
1805                 die(_("corrupt tag"));
1806 }
1807
1808 static int index_mem(struct index_state *istate,
1809                      struct object_id *oid, void *buf, size_t size,
1810                      enum object_type type,
1811                      const char *path, unsigned flags)
1812 {
1813         int ret, re_allocated = 0;
1814         int write_object = flags & HASH_WRITE_OBJECT;
1815
1816         if (!type)
1817                 type = OBJ_BLOB;
1818
1819         /*
1820          * Convert blobs to git internal format
1821          */
1822         if ((type == OBJ_BLOB) && path) {
1823                 struct strbuf nbuf = STRBUF_INIT;
1824                 if (convert_to_git(istate, path, buf, size, &nbuf,
1825                                    get_conv_flags(flags))) {
1826                         buf = strbuf_detach(&nbuf, &size);
1827                         re_allocated = 1;
1828                 }
1829         }
1830         if (flags & HASH_FORMAT_CHECK) {
1831                 if (type == OBJ_TREE)
1832                         check_tree(buf, size);
1833                 if (type == OBJ_COMMIT)
1834                         check_commit(buf, size);
1835                 if (type == OBJ_TAG)
1836                         check_tag(buf, size);
1837         }
1838
1839         if (write_object)
1840                 ret = write_object_file(buf, size, type_name(type), oid);
1841         else
1842                 ret = hash_object_file(buf, size, type_name(type), oid);
1843         if (re_allocated)
1844                 free(buf);
1845         return ret;
1846 }
1847
1848 static int index_stream_convert_blob(struct index_state *istate,
1849                                      struct object_id *oid,
1850                                      int fd,
1851                                      const char *path,
1852                                      unsigned flags)
1853 {
1854         int ret;
1855         const int write_object = flags & HASH_WRITE_OBJECT;
1856         struct strbuf sbuf = STRBUF_INIT;
1857
1858         assert(path);
1859         assert(would_convert_to_git_filter_fd(istate, path));
1860
1861         convert_to_git_filter_fd(istate, path, fd, &sbuf,
1862                                  get_conv_flags(flags));
1863
1864         if (write_object)
1865                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1866                                         oid);
1867         else
1868                 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1869                                        oid);
1870         strbuf_release(&sbuf);
1871         return ret;
1872 }
1873
1874 static int index_pipe(struct index_state *istate, struct object_id *oid,
1875                       int fd, enum object_type type,
1876                       const char *path, unsigned flags)
1877 {
1878         struct strbuf sbuf = STRBUF_INIT;
1879         int ret;
1880
1881         if (strbuf_read(&sbuf, fd, 4096) >= 0)
1882                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
1883         else
1884                 ret = -1;
1885         strbuf_release(&sbuf);
1886         return ret;
1887 }
1888
1889 #define SMALL_FILE_SIZE (32*1024)
1890
1891 static int index_core(struct index_state *istate,
1892                       struct object_id *oid, int fd, size_t size,
1893                       enum object_type type, const char *path,
1894                       unsigned flags)
1895 {
1896         int ret;
1897
1898         if (!size) {
1899                 ret = index_mem(istate, oid, "", size, type, path, flags);
1900         } else if (size <= SMALL_FILE_SIZE) {
1901                 char *buf = xmalloc(size);
1902                 ssize_t read_result = read_in_full(fd, buf, size);
1903                 if (read_result < 0)
1904                         ret = error_errno(_("read error while indexing %s"),
1905                                           path ? path : "<unknown>");
1906                 else if (read_result != size)
1907                         ret = error(_("short read while indexing %s"),
1908                                     path ? path : "<unknown>");
1909                 else
1910                         ret = index_mem(istate, oid, buf, size, type, path, flags);
1911                 free(buf);
1912         } else {
1913                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1914                 ret = index_mem(istate, oid, buf, size, type, path, flags);
1915                 munmap(buf, size);
1916         }
1917         return ret;
1918 }
1919
1920 /*
1921  * This creates one packfile per large blob unless bulk-checkin
1922  * machinery is "plugged".
1923  *
1924  * This also bypasses the usual "convert-to-git" dance, and that is on
1925  * purpose. We could write a streaming version of the converting
1926  * functions and insert that before feeding the data to fast-import
1927  * (or equivalent in-core API described above). However, that is
1928  * somewhat complicated, as we do not know the size of the filter
1929  * result, which we need to know beforehand when writing a git object.
1930  * Since the primary motivation for trying to stream from the working
1931  * tree file and to avoid mmaping it in core is to deal with large
1932  * binary blobs, they generally do not want to get any conversion, and
1933  * callers should avoid this code path when filters are requested.
1934  */
1935 static int index_stream(struct object_id *oid, int fd, size_t size,
1936                         enum object_type type, const char *path,
1937                         unsigned flags)
1938 {
1939         return index_bulk_checkin(oid, fd, size, type, path, flags);
1940 }
1941
1942 int index_fd(struct index_state *istate, struct object_id *oid,
1943              int fd, struct stat *st,
1944              enum object_type type, const char *path, unsigned flags)
1945 {
1946         int ret;
1947
1948         /*
1949          * Call xsize_t() only when needed to avoid potentially unnecessary
1950          * die() for large files.
1951          */
1952         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
1953                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
1954         else if (!S_ISREG(st->st_mode))
1955                 ret = index_pipe(istate, oid, fd, type, path, flags);
1956         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1957                  (path && would_convert_to_git(istate, path)))
1958                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
1959                                  type, path, flags);
1960         else
1961                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1962                                    flags);
1963         close(fd);
1964         return ret;
1965 }
1966
1967 int index_path(struct index_state *istate, struct object_id *oid,
1968                const char *path, struct stat *st, unsigned flags)
1969 {
1970         int fd;
1971         struct strbuf sb = STRBUF_INIT;
1972         int rc = 0;
1973
1974         switch (st->st_mode & S_IFMT) {
1975         case S_IFREG:
1976                 fd = open(path, O_RDONLY);
1977                 if (fd < 0)
1978                         return error_errno("open(\"%s\")", path);
1979                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
1980                         return error(_("%s: failed to insert into database"),
1981                                      path);
1982                 break;
1983         case S_IFLNK:
1984                 if (strbuf_readlink(&sb, path, st->st_size))
1985                         return error_errno("readlink(\"%s\")", path);
1986                 if (!(flags & HASH_WRITE_OBJECT))
1987                         hash_object_file(sb.buf, sb.len, blob_type, oid);
1988                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1989                         rc = error(_("%s: failed to insert into database"), path);
1990                 strbuf_release(&sb);
1991                 break;
1992         case S_IFDIR:
1993                 return resolve_gitlink_ref(path, "HEAD", oid);
1994         default:
1995                 return error(_("%s: unsupported file type"), path);
1996         }
1997         return rc;
1998 }
1999
2000 int read_pack_header(int fd, struct pack_header *header)
2001 {
2002         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2003                 /* "eof before pack header was fully read" */
2004                 return PH_ERROR_EOF;
2005
2006         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2007                 /* "protocol error (pack signature mismatch detected)" */
2008                 return PH_ERROR_PACK_SIGNATURE;
2009         if (!pack_version_ok(header->hdr_version))
2010                 /* "protocol error (pack version unsupported)" */
2011                 return PH_ERROR_PROTOCOL;
2012         return 0;
2013 }
2014
2015 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2016 {
2017         enum object_type type = oid_object_info(the_repository, oid, NULL);
2018         if (type < 0)
2019                 die(_("%s is not a valid object"), oid_to_hex(oid));
2020         if (type != expect)
2021                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2022                     type_name(expect));
2023 }
2024
2025 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2026                                 struct strbuf *path,
2027                                 each_loose_object_fn obj_cb,
2028                                 each_loose_cruft_fn cruft_cb,
2029                                 each_loose_subdir_fn subdir_cb,
2030                                 void *data)
2031 {
2032         size_t origlen, baselen;
2033         DIR *dir;
2034         struct dirent *de;
2035         int r = 0;
2036         struct object_id oid;
2037
2038         if (subdir_nr > 0xff)
2039                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2040
2041         origlen = path->len;
2042         strbuf_complete(path, '/');
2043         strbuf_addf(path, "%02x", subdir_nr);
2044
2045         dir = opendir(path->buf);
2046         if (!dir) {
2047                 if (errno != ENOENT)
2048                         r = error_errno(_("unable to open %s"), path->buf);
2049                 strbuf_setlen(path, origlen);
2050                 return r;
2051         }
2052
2053         oid.hash[0] = subdir_nr;
2054         strbuf_addch(path, '/');
2055         baselen = path->len;
2056
2057         while ((de = readdir(dir))) {
2058                 size_t namelen;
2059                 if (is_dot_or_dotdot(de->d_name))
2060                         continue;
2061
2062                 namelen = strlen(de->d_name);
2063                 strbuf_setlen(path, baselen);
2064                 strbuf_add(path, de->d_name, namelen);
2065                 if (namelen == the_hash_algo->hexsz - 2 &&
2066                     !hex_to_bytes(oid.hash + 1, de->d_name,
2067                                   the_hash_algo->rawsz - 1)) {
2068                         if (obj_cb) {
2069                                 r = obj_cb(&oid, path->buf, data);
2070                                 if (r)
2071                                         break;
2072                         }
2073                         continue;
2074                 }
2075
2076                 if (cruft_cb) {
2077                         r = cruft_cb(de->d_name, path->buf, data);
2078                         if (r)
2079                                 break;
2080                 }
2081         }
2082         closedir(dir);
2083
2084         strbuf_setlen(path, baselen - 1);
2085         if (!r && subdir_cb)
2086                 r = subdir_cb(subdir_nr, path->buf, data);
2087
2088         strbuf_setlen(path, origlen);
2089
2090         return r;
2091 }
2092
2093 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2094                             each_loose_object_fn obj_cb,
2095                             each_loose_cruft_fn cruft_cb,
2096                             each_loose_subdir_fn subdir_cb,
2097                             void *data)
2098 {
2099         int r = 0;
2100         int i;
2101
2102         for (i = 0; i < 256; i++) {
2103                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2104                                                 subdir_cb, data);
2105                 if (r)
2106                         break;
2107         }
2108
2109         return r;
2110 }
2111
2112 int for_each_loose_file_in_objdir(const char *path,
2113                                   each_loose_object_fn obj_cb,
2114                                   each_loose_cruft_fn cruft_cb,
2115                                   each_loose_subdir_fn subdir_cb,
2116                                   void *data)
2117 {
2118         struct strbuf buf = STRBUF_INIT;
2119         int r;
2120
2121         strbuf_addstr(&buf, path);
2122         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2123                                               subdir_cb, data);
2124         strbuf_release(&buf);
2125
2126         return r;
2127 }
2128
2129 int for_each_loose_object(each_loose_object_fn cb, void *data,
2130                           enum for_each_object_flags flags)
2131 {
2132         struct object_directory *odb;
2133
2134         prepare_alt_odb(the_repository);
2135         for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2136                 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2137                                                       NULL, data);
2138                 if (r)
2139                         return r;
2140
2141                 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2142                         break;
2143         }
2144
2145         return 0;
2146 }
2147
2148 static int append_loose_object(const struct object_id *oid, const char *path,
2149                                void *data)
2150 {
2151         oid_array_append(data, oid);
2152         return 0;
2153 }
2154
2155 void odb_load_loose_cache(struct object_directory *odb, int subdir_nr)
2156 {
2157         struct strbuf buf = STRBUF_INIT;
2158
2159         if (subdir_nr < 0 ||
2160             subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2161                 BUG("subdir_nr out of range");
2162
2163         if (odb->loose_objects_subdir_seen[subdir_nr])
2164                 return;
2165
2166         strbuf_addstr(&buf, odb->path);
2167         for_each_file_in_obj_subdir(subdir_nr, &buf,
2168                                     append_loose_object,
2169                                     NULL, NULL,
2170                                     &odb->loose_objects_cache);
2171         odb->loose_objects_subdir_seen[subdir_nr] = 1;
2172         strbuf_release(&buf);
2173 }
2174
2175 static int check_stream_sha1(git_zstream *stream,
2176                              const char *hdr,
2177                              unsigned long size,
2178                              const char *path,
2179                              const unsigned char *expected_sha1)
2180 {
2181         git_hash_ctx c;
2182         unsigned char real_sha1[GIT_MAX_RAWSZ];
2183         unsigned char buf[4096];
2184         unsigned long total_read;
2185         int status = Z_OK;
2186
2187         the_hash_algo->init_fn(&c);
2188         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2189
2190         /*
2191          * We already read some bytes into hdr, but the ones up to the NUL
2192          * do not count against the object's content size.
2193          */
2194         total_read = stream->total_out - strlen(hdr) - 1;
2195
2196         /*
2197          * This size comparison must be "<=" to read the final zlib packets;
2198          * see the comment in unpack_sha1_rest for details.
2199          */
2200         while (total_read <= size &&
2201                (status == Z_OK ||
2202                 (status == Z_BUF_ERROR && !stream->avail_out))) {
2203                 stream->next_out = buf;
2204                 stream->avail_out = sizeof(buf);
2205                 if (size - total_read < stream->avail_out)
2206                         stream->avail_out = size - total_read;
2207                 status = git_inflate(stream, Z_FINISH);
2208                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2209                 total_read += stream->next_out - buf;
2210         }
2211         git_inflate_end(stream);
2212
2213         if (status != Z_STREAM_END) {
2214                 error(_("corrupt loose object '%s'"), sha1_to_hex(expected_sha1));
2215                 return -1;
2216         }
2217         if (stream->avail_in) {
2218                 error(_("garbage at end of loose object '%s'"),
2219                       sha1_to_hex(expected_sha1));
2220                 return -1;
2221         }
2222
2223         the_hash_algo->final_fn(real_sha1, &c);
2224         if (!hasheq(expected_sha1, real_sha1)) {
2225                 error(_("sha1 mismatch for %s (expected %s)"), path,
2226                       sha1_to_hex(expected_sha1));
2227                 return -1;
2228         }
2229
2230         return 0;
2231 }
2232
2233 int read_loose_object(const char *path,
2234                       const struct object_id *expected_oid,
2235                       enum object_type *type,
2236                       unsigned long *size,
2237                       void **contents)
2238 {
2239         int ret = -1;
2240         void *map = NULL;
2241         unsigned long mapsize;
2242         git_zstream stream;
2243         char hdr[MAX_HEADER_LEN];
2244
2245         *contents = NULL;
2246
2247         map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2248         if (!map) {
2249                 error_errno(_("unable to mmap %s"), path);
2250                 goto out;
2251         }
2252
2253         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2254                 error(_("unable to unpack header of %s"), path);
2255                 goto out;
2256         }
2257
2258         *type = parse_sha1_header(hdr, size);
2259         if (*type < 0) {
2260                 error(_("unable to parse header of %s"), path);
2261                 git_inflate_end(&stream);
2262                 goto out;
2263         }
2264
2265         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2266                 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2267                         goto out;
2268         } else {
2269                 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2270                 if (!*contents) {
2271                         error(_("unable to unpack contents of %s"), path);
2272                         git_inflate_end(&stream);
2273                         goto out;
2274                 }
2275                 if (check_object_signature(expected_oid, *contents,
2276                                          *size, type_name(*type))) {
2277                         error(_("sha1 mismatch for %s (expected %s)"), path,
2278                               oid_to_hex(expected_oid));
2279                         free(*contents);
2280                         goto out;
2281                 }
2282         }
2283
2284         ret = 0; /* everything checks out */
2285
2286 out:
2287         if (map)
2288                 munmap(map, mapsize);
2289         return ret;
2290 }