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