sha1-file: make pretend_object_file() not prefetch
[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 "object-store.h"
34 #include "promisor-remote.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 #define EMPTY_TREE_SHA256_BIN_LITERAL \
44         "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
45         "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
46         "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
47         "\x53\x21"
48
49 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
50         "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
51         "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
52 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
53         "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
54         "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
55         "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
56         "\x18\x13"
57
58 const struct object_id null_oid;
59 static const struct object_id empty_tree_oid = {
60         EMPTY_TREE_SHA1_BIN_LITERAL
61 };
62 static const struct object_id empty_blob_oid = {
63         EMPTY_BLOB_SHA1_BIN_LITERAL
64 };
65 static const struct object_id empty_tree_oid_sha256 = {
66         EMPTY_TREE_SHA256_BIN_LITERAL
67 };
68 static const struct object_id empty_blob_oid_sha256 = {
69         EMPTY_BLOB_SHA256_BIN_LITERAL
70 };
71
72 static void git_hash_sha1_init(git_hash_ctx *ctx)
73 {
74         git_SHA1_Init(&ctx->sha1);
75 }
76
77 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
78 {
79         git_SHA1_Update(&ctx->sha1, data, len);
80 }
81
82 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
83 {
84         git_SHA1_Final(hash, &ctx->sha1);
85 }
86
87
88 static void git_hash_sha256_init(git_hash_ctx *ctx)
89 {
90         git_SHA256_Init(&ctx->sha256);
91 }
92
93 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
94 {
95         git_SHA256_Update(&ctx->sha256, data, len);
96 }
97
98 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
99 {
100         git_SHA256_Final(hash, &ctx->sha256);
101 }
102
103 static void git_hash_unknown_init(git_hash_ctx *ctx)
104 {
105         BUG("trying to init unknown hash");
106 }
107
108 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
109 {
110         BUG("trying to update unknown hash");
111 }
112
113 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
114 {
115         BUG("trying to finalize unknown hash");
116 }
117
118 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
119         {
120                 NULL,
121                 0x00000000,
122                 0,
123                 0,
124                 0,
125                 git_hash_unknown_init,
126                 git_hash_unknown_update,
127                 git_hash_unknown_final,
128                 NULL,
129                 NULL,
130         },
131         {
132                 "sha1",
133                 /* "sha1", big-endian */
134                 0x73686131,
135                 GIT_SHA1_RAWSZ,
136                 GIT_SHA1_HEXSZ,
137                 GIT_SHA1_BLKSZ,
138                 git_hash_sha1_init,
139                 git_hash_sha1_update,
140                 git_hash_sha1_final,
141                 &empty_tree_oid,
142                 &empty_blob_oid,
143         },
144         {
145                 "sha256",
146                 /* "s256", big-endian */
147                 0x73323536,
148                 GIT_SHA256_RAWSZ,
149                 GIT_SHA256_HEXSZ,
150                 GIT_SHA256_BLKSZ,
151                 git_hash_sha256_init,
152                 git_hash_sha256_update,
153                 git_hash_sha256_final,
154                 &empty_tree_oid_sha256,
155                 &empty_blob_oid_sha256,
156         }
157 };
158
159 const char *empty_tree_oid_hex(void)
160 {
161         static char buf[GIT_MAX_HEXSZ + 1];
162         return oid_to_hex_r(buf, the_hash_algo->empty_tree);
163 }
164
165 const char *empty_blob_oid_hex(void)
166 {
167         static char buf[GIT_MAX_HEXSZ + 1];
168         return oid_to_hex_r(buf, the_hash_algo->empty_blob);
169 }
170
171 int hash_algo_by_name(const char *name)
172 {
173         int i;
174         if (!name)
175                 return GIT_HASH_UNKNOWN;
176         for (i = 1; i < GIT_HASH_NALGOS; i++)
177                 if (!strcmp(name, hash_algos[i].name))
178                         return i;
179         return GIT_HASH_UNKNOWN;
180 }
181
182 int hash_algo_by_id(uint32_t format_id)
183 {
184         int i;
185         for (i = 1; i < GIT_HASH_NALGOS; i++)
186                 if (format_id == hash_algos[i].format_id)
187                         return i;
188         return GIT_HASH_UNKNOWN;
189 }
190
191 int hash_algo_by_length(int len)
192 {
193         int i;
194         for (i = 1; i < GIT_HASH_NALGOS; i++)
195                 if (len == hash_algos[i].rawsz)
196                         return i;
197         return GIT_HASH_UNKNOWN;
198 }
199
200 /*
201  * This is meant to hold a *small* number of objects that you would
202  * want read_object_file() to be able to return, but yet you do not want
203  * to write them into the object store (e.g. a browse-only
204  * application).
205  */
206 static struct cached_object {
207         struct object_id oid;
208         enum object_type type;
209         void *buf;
210         unsigned long size;
211 } *cached_objects;
212 static int cached_object_nr, cached_object_alloc;
213
214 static struct cached_object empty_tree = {
215         { EMPTY_TREE_SHA1_BIN_LITERAL },
216         OBJ_TREE,
217         "",
218         0
219 };
220
221 static struct cached_object *find_cached_object(const struct object_id *oid)
222 {
223         int i;
224         struct cached_object *co = cached_objects;
225
226         for (i = 0; i < cached_object_nr; i++, co++) {
227                 if (oideq(&co->oid, oid))
228                         return co;
229         }
230         if (oideq(oid, the_hash_algo->empty_tree))
231                 return &empty_tree;
232         return NULL;
233 }
234
235
236 static int get_conv_flags(unsigned flags)
237 {
238         if (flags & HASH_RENORMALIZE)
239                 return CONV_EOL_RENORMALIZE;
240         else if (flags & HASH_WRITE_OBJECT)
241                 return global_conv_flags_eol | CONV_WRITE_OBJECT;
242         else
243                 return 0;
244 }
245
246
247 int mkdir_in_gitdir(const char *path)
248 {
249         if (mkdir(path, 0777)) {
250                 int saved_errno = errno;
251                 struct stat st;
252                 struct strbuf sb = STRBUF_INIT;
253
254                 if (errno != EEXIST)
255                         return -1;
256                 /*
257                  * Are we looking at a path in a symlinked worktree
258                  * whose original repository does not yet have it?
259                  * e.g. .git/rr-cache pointing at its original
260                  * repository in which the user hasn't performed any
261                  * conflict resolution yet?
262                  */
263                 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
264                     strbuf_readlink(&sb, path, st.st_size) ||
265                     !is_absolute_path(sb.buf) ||
266                     mkdir(sb.buf, 0777)) {
267                         strbuf_release(&sb);
268                         errno = saved_errno;
269                         return -1;
270                 }
271                 strbuf_release(&sb);
272         }
273         return adjust_shared_perm(path);
274 }
275
276 enum scld_error safe_create_leading_directories(char *path)
277 {
278         char *next_component = path + offset_1st_component(path);
279         enum scld_error ret = SCLD_OK;
280
281         while (ret == SCLD_OK && next_component) {
282                 struct stat st;
283                 char *slash = next_component, slash_character;
284
285                 while (*slash && !is_dir_sep(*slash))
286                         slash++;
287
288                 if (!*slash)
289                         break;
290
291                 next_component = slash + 1;
292                 while (is_dir_sep(*next_component))
293                         next_component++;
294                 if (!*next_component)
295                         break;
296
297                 slash_character = *slash;
298                 *slash = '\0';
299                 if (!stat(path, &st)) {
300                         /* path exists */
301                         if (!S_ISDIR(st.st_mode)) {
302                                 errno = ENOTDIR;
303                                 ret = SCLD_EXISTS;
304                         }
305                 } else if (mkdir(path, 0777)) {
306                         if (errno == EEXIST &&
307                             !stat(path, &st) && S_ISDIR(st.st_mode))
308                                 ; /* somebody created it since we checked */
309                         else if (errno == ENOENT)
310                                 /*
311                                  * Either mkdir() failed because
312                                  * somebody just pruned the containing
313                                  * directory, or stat() failed because
314                                  * the file that was in our way was
315                                  * just removed.  Either way, inform
316                                  * the caller that it might be worth
317                                  * trying again:
318                                  */
319                                 ret = SCLD_VANISHED;
320                         else
321                                 ret = SCLD_FAILED;
322                 } else if (adjust_shared_perm(path)) {
323                         ret = SCLD_PERMS;
324                 }
325                 *slash = slash_character;
326         }
327         return ret;
328 }
329
330 enum scld_error safe_create_leading_directories_const(const char *path)
331 {
332         int save_errno;
333         /* path points to cache entries, so xstrdup before messing with it */
334         char *buf = xstrdup(path);
335         enum scld_error result = safe_create_leading_directories(buf);
336
337         save_errno = errno;
338         free(buf);
339         errno = save_errno;
340         return result;
341 }
342
343 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
344 {
345         /*
346          * The number of times we will try to remove empty directories
347          * in the way of path. This is only 1 because if another
348          * process is racily creating directories that conflict with
349          * us, we don't want to fight against them.
350          */
351         int remove_directories_remaining = 1;
352
353         /*
354          * The number of times that we will try to create the
355          * directories containing path. We are willing to attempt this
356          * more than once, because another process could be trying to
357          * clean up empty directories at the same time as we are
358          * trying to create them.
359          */
360         int create_directories_remaining = 3;
361
362         /* A scratch copy of path, filled lazily if we need it: */
363         struct strbuf path_copy = STRBUF_INIT;
364
365         int ret, save_errno;
366
367         /* Sanity check: */
368         assert(*path);
369
370 retry_fn:
371         ret = fn(path, cb);
372         save_errno = errno;
373         if (!ret)
374                 goto out;
375
376         if (errno == EISDIR && remove_directories_remaining-- > 0) {
377                 /*
378                  * A directory is in the way. Maybe it is empty; try
379                  * to remove it:
380                  */
381                 if (!path_copy.len)
382                         strbuf_addstr(&path_copy, path);
383
384                 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
385                         goto retry_fn;
386         } else if (errno == ENOENT && create_directories_remaining-- > 0) {
387                 /*
388                  * Maybe the containing directory didn't exist, or
389                  * maybe it was just deleted by a process that is
390                  * racing with us to clean up empty directories. Try
391                  * to create it:
392                  */
393                 enum scld_error scld_result;
394
395                 if (!path_copy.len)
396                         strbuf_addstr(&path_copy, path);
397
398                 do {
399                         scld_result = safe_create_leading_directories(path_copy.buf);
400                         if (scld_result == SCLD_OK)
401                                 goto retry_fn;
402                 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
403         }
404
405 out:
406         strbuf_release(&path_copy);
407         errno = save_errno;
408         return ret;
409 }
410
411 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
412 {
413         int i;
414         for (i = 0; i < the_hash_algo->rawsz; i++) {
415                 static char hex[] = "0123456789abcdef";
416                 unsigned int val = oid->hash[i];
417                 strbuf_addch(buf, hex[val >> 4]);
418                 strbuf_addch(buf, hex[val & 0xf]);
419                 if (!i)
420                         strbuf_addch(buf, '/');
421         }
422 }
423
424 static const char *odb_loose_path(struct object_directory *odb,
425                                   struct strbuf *buf,
426                                   const struct object_id *oid)
427 {
428         strbuf_reset(buf);
429         strbuf_addstr(buf, odb->path);
430         strbuf_addch(buf, '/');
431         fill_loose_path(buf, oid);
432         return buf->buf;
433 }
434
435 const char *loose_object_path(struct repository *r, struct strbuf *buf,
436                               const struct object_id *oid)
437 {
438         return odb_loose_path(r->objects->odb, buf, oid);
439 }
440
441 /*
442  * Return non-zero iff the path is usable as an alternate object database.
443  */
444 static int alt_odb_usable(struct raw_object_store *o,
445                           struct strbuf *path,
446                           const char *normalized_objdir)
447 {
448         struct object_directory *odb;
449
450         /* Detect cases where alternate disappeared */
451         if (!is_directory(path->buf)) {
452                 error(_("object directory %s does not exist; "
453                         "check .git/objects/info/alternates"),
454                       path->buf);
455                 return 0;
456         }
457
458         /*
459          * Prevent the common mistake of listing the same
460          * thing twice, or object directory itself.
461          */
462         for (odb = o->odb; odb; odb = odb->next) {
463                 if (!fspathcmp(path->buf, odb->path))
464                         return 0;
465         }
466         if (!fspathcmp(path->buf, normalized_objdir))
467                 return 0;
468
469         return 1;
470 }
471
472 /*
473  * Prepare alternate object database registry.
474  *
475  * The variable alt_odb_list points at the list of struct
476  * object_directory.  The elements on this list come from
477  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
478  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
479  * whose contents is similar to that environment variable but can be
480  * LF separated.  Its base points at a statically allocated buffer that
481  * contains "/the/directory/corresponding/to/.git/objects/...", while
482  * its name points just after the slash at the end of ".git/objects/"
483  * in the example above, and has enough space to hold 40-byte hex
484  * SHA1, an extra slash for the first level indirection, and the
485  * terminating NUL.
486  */
487 static void read_info_alternates(struct repository *r,
488                                  const char *relative_base,
489                                  int depth);
490 static int link_alt_odb_entry(struct repository *r, const char *entry,
491         const char *relative_base, int depth, const char *normalized_objdir)
492 {
493         struct object_directory *ent;
494         struct strbuf pathbuf = STRBUF_INIT;
495
496         if (!is_absolute_path(entry) && relative_base) {
497                 strbuf_realpath(&pathbuf, relative_base, 1);
498                 strbuf_addch(&pathbuf, '/');
499         }
500         strbuf_addstr(&pathbuf, entry);
501
502         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
503                 error(_("unable to normalize alternate object path: %s"),
504                       pathbuf.buf);
505                 strbuf_release(&pathbuf);
506                 return -1;
507         }
508
509         /*
510          * The trailing slash after the directory name is given by
511          * this function at the end. Remove duplicates.
512          */
513         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
514                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
515
516         if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
517                 strbuf_release(&pathbuf);
518                 return -1;
519         }
520
521         ent = xcalloc(1, sizeof(*ent));
522         ent->path = xstrdup(pathbuf.buf);
523
524         /* add the alternate entry */
525         *r->objects->odb_tail = ent;
526         r->objects->odb_tail = &(ent->next);
527         ent->next = NULL;
528
529         /* recursively add alternates */
530         read_info_alternates(r, pathbuf.buf, depth + 1);
531
532         strbuf_release(&pathbuf);
533         return 0;
534 }
535
536 static const char *parse_alt_odb_entry(const char *string,
537                                        int sep,
538                                        struct strbuf *out)
539 {
540         const char *end;
541
542         strbuf_reset(out);
543
544         if (*string == '#') {
545                 /* comment; consume up to next separator */
546                 end = strchrnul(string, sep);
547         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
548                 /*
549                  * quoted path; unquote_c_style has copied the
550                  * data for us and set "end". Broken quoting (e.g.,
551                  * an entry that doesn't end with a quote) falls
552                  * back to the unquoted case below.
553                  */
554         } else {
555                 /* normal, unquoted path */
556                 end = strchrnul(string, sep);
557                 strbuf_add(out, string, end - string);
558         }
559
560         if (*end)
561                 end++;
562         return end;
563 }
564
565 static void link_alt_odb_entries(struct repository *r, const char *alt,
566                                  int sep, const char *relative_base, int depth)
567 {
568         struct strbuf objdirbuf = STRBUF_INIT;
569         struct strbuf entry = STRBUF_INIT;
570
571         if (!alt || !*alt)
572                 return;
573
574         if (depth > 5) {
575                 error(_("%s: ignoring alternate object stores, nesting too deep"),
576                                 relative_base);
577                 return;
578         }
579
580         strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
581         if (strbuf_normalize_path(&objdirbuf) < 0)
582                 die(_("unable to normalize object directory: %s"),
583                     objdirbuf.buf);
584
585         while (*alt) {
586                 alt = parse_alt_odb_entry(alt, sep, &entry);
587                 if (!entry.len)
588                         continue;
589                 link_alt_odb_entry(r, entry.buf,
590                                    relative_base, depth, objdirbuf.buf);
591         }
592         strbuf_release(&entry);
593         strbuf_release(&objdirbuf);
594 }
595
596 static void read_info_alternates(struct repository *r,
597                                  const char *relative_base,
598                                  int depth)
599 {
600         char *path;
601         struct strbuf buf = STRBUF_INIT;
602
603         path = xstrfmt("%s/info/alternates", relative_base);
604         if (strbuf_read_file(&buf, path, 1024) < 0) {
605                 warn_on_fopen_errors(path);
606                 free(path);
607                 return;
608         }
609
610         link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
611         strbuf_release(&buf);
612         free(path);
613 }
614
615 void add_to_alternates_file(const char *reference)
616 {
617         struct lock_file lock = LOCK_INIT;
618         char *alts = git_pathdup("objects/info/alternates");
619         FILE *in, *out;
620         int found = 0;
621
622         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
623         out = fdopen_lock_file(&lock, "w");
624         if (!out)
625                 die_errno(_("unable to fdopen alternates lockfile"));
626
627         in = fopen(alts, "r");
628         if (in) {
629                 struct strbuf line = STRBUF_INIT;
630
631                 while (strbuf_getline(&line, in) != EOF) {
632                         if (!strcmp(reference, line.buf)) {
633                                 found = 1;
634                                 break;
635                         }
636                         fprintf_or_die(out, "%s\n", line.buf);
637                 }
638
639                 strbuf_release(&line);
640                 fclose(in);
641         }
642         else if (errno != ENOENT)
643                 die_errno(_("unable to read alternates file"));
644
645         if (found) {
646                 rollback_lock_file(&lock);
647         } else {
648                 fprintf_or_die(out, "%s\n", reference);
649                 if (commit_lock_file(&lock))
650                         die_errno(_("unable to move new alternates file into place"));
651                 if (the_repository->objects->loaded_alternates)
652                         link_alt_odb_entries(the_repository, reference,
653                                              '\n', NULL, 0);
654         }
655         free(alts);
656 }
657
658 void add_to_alternates_memory(const char *reference)
659 {
660         /*
661          * Make sure alternates are initialized, or else our entry may be
662          * overwritten when they are.
663          */
664         prepare_alt_odb(the_repository);
665
666         link_alt_odb_entries(the_repository, reference,
667                              '\n', NULL, 0);
668 }
669
670 /*
671  * Compute the exact path an alternate is at and returns it. In case of
672  * error NULL is returned and the human readable error is added to `err`
673  * `path` may be relative and should point to $GIT_DIR.
674  * `err` must not be null.
675  */
676 char *compute_alternate_path(const char *path, struct strbuf *err)
677 {
678         char *ref_git = NULL;
679         const char *repo, *ref_git_s;
680         int seen_error = 0;
681
682         ref_git_s = real_path_if_valid(path);
683         if (!ref_git_s) {
684                 seen_error = 1;
685                 strbuf_addf(err, _("path '%s' does not exist"), path);
686                 goto out;
687         } else
688                 /*
689                  * Beware: read_gitfile(), real_path() and mkpath()
690                  * return static buffer
691                  */
692                 ref_git = xstrdup(ref_git_s);
693
694         repo = read_gitfile(ref_git);
695         if (!repo)
696                 repo = read_gitfile(mkpath("%s/.git", ref_git));
697         if (repo) {
698                 free(ref_git);
699                 ref_git = xstrdup(repo);
700         }
701
702         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
703                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
704                 free(ref_git);
705                 ref_git = ref_git_git;
706         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
707                 struct strbuf sb = STRBUF_INIT;
708                 seen_error = 1;
709                 if (get_common_dir(&sb, ref_git)) {
710                         strbuf_addf(err,
711                                     _("reference repository '%s' as a linked "
712                                       "checkout is not supported yet."),
713                                     path);
714                         goto out;
715                 }
716
717                 strbuf_addf(err, _("reference repository '%s' is not a "
718                                         "local repository."), path);
719                 goto out;
720         }
721
722         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
723                 strbuf_addf(err, _("reference repository '%s' is shallow"),
724                             path);
725                 seen_error = 1;
726                 goto out;
727         }
728
729         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
730                 strbuf_addf(err,
731                             _("reference repository '%s' is grafted"),
732                             path);
733                 seen_error = 1;
734                 goto out;
735         }
736
737 out:
738         if (seen_error) {
739                 FREE_AND_NULL(ref_git);
740         }
741
742         return ref_git;
743 }
744
745 static void fill_alternate_refs_command(struct child_process *cmd,
746                                         const char *repo_path)
747 {
748         const char *value;
749
750         if (!git_config_get_value("core.alternateRefsCommand", &value)) {
751                 cmd->use_shell = 1;
752
753                 argv_array_push(&cmd->args, value);
754                 argv_array_push(&cmd->args, repo_path);
755         } else {
756                 cmd->git_cmd = 1;
757
758                 argv_array_pushf(&cmd->args, "--git-dir=%s", repo_path);
759                 argv_array_push(&cmd->args, "for-each-ref");
760                 argv_array_push(&cmd->args, "--format=%(objectname)");
761
762                 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
763                         argv_array_push(&cmd->args, "--");
764                         argv_array_split(&cmd->args, value);
765                 }
766         }
767
768         cmd->env = local_repo_env;
769         cmd->out = -1;
770 }
771
772 static void read_alternate_refs(const char *path,
773                                 alternate_ref_fn *cb,
774                                 void *data)
775 {
776         struct child_process cmd = CHILD_PROCESS_INIT;
777         struct strbuf line = STRBUF_INIT;
778         FILE *fh;
779
780         fill_alternate_refs_command(&cmd, path);
781
782         if (start_command(&cmd))
783                 return;
784
785         fh = xfdopen(cmd.out, "r");
786         while (strbuf_getline_lf(&line, fh) != EOF) {
787                 struct object_id oid;
788                 const char *p;
789
790                 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
791                         warning(_("invalid line while parsing alternate refs: %s"),
792                                 line.buf);
793                         break;
794                 }
795
796                 cb(&oid, data);
797         }
798
799         fclose(fh);
800         finish_command(&cmd);
801         strbuf_release(&line);
802 }
803
804 struct alternate_refs_data {
805         alternate_ref_fn *fn;
806         void *data;
807 };
808
809 static int refs_from_alternate_cb(struct object_directory *e,
810                                   void *data)
811 {
812         struct strbuf path = STRBUF_INIT;
813         size_t base_len;
814         struct alternate_refs_data *cb = data;
815
816         if (!strbuf_realpath(&path, e->path, 0))
817                 goto out;
818         if (!strbuf_strip_suffix(&path, "/objects"))
819                 goto out;
820         base_len = path.len;
821
822         /* Is this a git repository with refs? */
823         strbuf_addstr(&path, "/refs");
824         if (!is_directory(path.buf))
825                 goto out;
826         strbuf_setlen(&path, base_len);
827
828         read_alternate_refs(path.buf, cb->fn, cb->data);
829
830 out:
831         strbuf_release(&path);
832         return 0;
833 }
834
835 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
836 {
837         struct alternate_refs_data cb;
838         cb.fn = fn;
839         cb.data = data;
840         foreach_alt_odb(refs_from_alternate_cb, &cb);
841 }
842
843 int foreach_alt_odb(alt_odb_fn fn, void *cb)
844 {
845         struct object_directory *ent;
846         int r = 0;
847
848         prepare_alt_odb(the_repository);
849         for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
850                 r = fn(ent, cb);
851                 if (r)
852                         break;
853         }
854         return r;
855 }
856
857 void prepare_alt_odb(struct repository *r)
858 {
859         if (r->objects->loaded_alternates)
860                 return;
861
862         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
863
864         read_info_alternates(r, r->objects->odb->path, 0);
865         r->objects->loaded_alternates = 1;
866 }
867
868 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
869 static int freshen_file(const char *fn)
870 {
871         struct utimbuf t;
872         t.actime = t.modtime = time(NULL);
873         return !utime(fn, &t);
874 }
875
876 /*
877  * All of the check_and_freshen functions return 1 if the file exists and was
878  * freshened (if freshening was requested), 0 otherwise. If they return
879  * 0, you should not assume that it is safe to skip a write of the object (it
880  * either does not exist on disk, or has a stale mtime and may be subject to
881  * pruning).
882  */
883 int check_and_freshen_file(const char *fn, int freshen)
884 {
885         if (access(fn, F_OK))
886                 return 0;
887         if (freshen && !freshen_file(fn))
888                 return 0;
889         return 1;
890 }
891
892 static int check_and_freshen_odb(struct object_directory *odb,
893                                  const struct object_id *oid,
894                                  int freshen)
895 {
896         static struct strbuf path = STRBUF_INIT;
897         odb_loose_path(odb, &path, oid);
898         return check_and_freshen_file(path.buf, freshen);
899 }
900
901 static int check_and_freshen_local(const struct object_id *oid, int freshen)
902 {
903         return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
904 }
905
906 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
907 {
908         struct object_directory *odb;
909
910         prepare_alt_odb(the_repository);
911         for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
912                 if (check_and_freshen_odb(odb, oid, freshen))
913                         return 1;
914         }
915         return 0;
916 }
917
918 static int check_and_freshen(const struct object_id *oid, int freshen)
919 {
920         return check_and_freshen_local(oid, freshen) ||
921                check_and_freshen_nonlocal(oid, freshen);
922 }
923
924 int has_loose_object_nonlocal(const struct object_id *oid)
925 {
926         return check_and_freshen_nonlocal(oid, 0);
927 }
928
929 static int has_loose_object(const struct object_id *oid)
930 {
931         return check_and_freshen(oid, 0);
932 }
933
934 static void mmap_limit_check(size_t length)
935 {
936         static size_t limit = 0;
937         if (!limit) {
938                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
939                 if (!limit)
940                         limit = SIZE_MAX;
941         }
942         if (length > limit)
943                 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
944                     (uintmax_t)length, (uintmax_t)limit);
945 }
946
947 void *xmmap_gently(void *start, size_t length,
948                   int prot, int flags, int fd, off_t offset)
949 {
950         void *ret;
951
952         mmap_limit_check(length);
953         ret = mmap(start, length, prot, flags, fd, offset);
954         if (ret == MAP_FAILED && !length)
955                 ret = NULL;
956         return ret;
957 }
958
959 void *xmmap(void *start, size_t length,
960         int prot, int flags, int fd, off_t offset)
961 {
962         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
963         if (ret == MAP_FAILED)
964                 die_errno(_("mmap failed"));
965         return ret;
966 }
967
968 /*
969  * With an in-core object data in "map", rehash it to make sure the
970  * object name actually matches "oid" to detect object corruption.
971  * With "map" == NULL, try reading the object named with "oid" using
972  * the streaming interface and rehash it to do the same.
973  */
974 int check_object_signature(struct repository *r, const struct object_id *oid,
975                            void *map, unsigned long size, const char *type)
976 {
977         struct object_id real_oid;
978         enum object_type obj_type;
979         struct git_istream *st;
980         git_hash_ctx c;
981         char hdr[MAX_HEADER_LEN];
982         int hdrlen;
983
984         if (map) {
985                 hash_object_file(r->hash_algo, map, size, type, &real_oid);
986                 return !oideq(oid, &real_oid) ? -1 : 0;
987         }
988
989         st = open_istream(r, oid, &obj_type, &size, NULL);
990         if (!st)
991                 return -1;
992
993         /* Generate the header */
994         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
995
996         /* Sha1.. */
997         r->hash_algo->init_fn(&c);
998         r->hash_algo->update_fn(&c, hdr, hdrlen);
999         for (;;) {
1000                 char buf[1024 * 16];
1001                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1002
1003                 if (readlen < 0) {
1004                         close_istream(st);
1005                         return -1;
1006                 }
1007                 if (!readlen)
1008                         break;
1009                 r->hash_algo->update_fn(&c, buf, readlen);
1010         }
1011         r->hash_algo->final_fn(real_oid.hash, &c);
1012         close_istream(st);
1013         return !oideq(oid, &real_oid) ? -1 : 0;
1014 }
1015
1016 int git_open_cloexec(const char *name, int flags)
1017 {
1018         int fd;
1019         static int o_cloexec = O_CLOEXEC;
1020
1021         fd = open(name, flags | o_cloexec);
1022         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1023                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1024                 o_cloexec &= ~O_CLOEXEC;
1025                 fd = open(name, flags | o_cloexec);
1026         }
1027
1028 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1029         {
1030                 static int fd_cloexec = FD_CLOEXEC;
1031
1032                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1033                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
1034                         int flags = fcntl(fd, F_GETFD);
1035                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1036                                 fd_cloexec = 0;
1037                 }
1038         }
1039 #endif
1040         return fd;
1041 }
1042
1043 /*
1044  * Find "oid" as a loose object in the local repository or in an alternate.
1045  * Returns 0 on success, negative on failure.
1046  *
1047  * The "path" out-parameter will give the path of the object we found (if any).
1048  * Note that it may point to static storage and is only valid until another
1049  * call to stat_loose_object().
1050  */
1051 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1052                              struct stat *st, const char **path)
1053 {
1054         struct object_directory *odb;
1055         static struct strbuf buf = STRBUF_INIT;
1056
1057         prepare_alt_odb(r);
1058         for (odb = r->objects->odb; odb; odb = odb->next) {
1059                 *path = odb_loose_path(odb, &buf, oid);
1060                 if (!lstat(*path, st))
1061                         return 0;
1062         }
1063
1064         return -1;
1065 }
1066
1067 /*
1068  * Like stat_loose_object(), but actually open the object and return the
1069  * descriptor. See the caveats on the "path" parameter above.
1070  */
1071 static int open_loose_object(struct repository *r,
1072                              const struct object_id *oid, const char **path)
1073 {
1074         int fd;
1075         struct object_directory *odb;
1076         int most_interesting_errno = ENOENT;
1077         static struct strbuf buf = STRBUF_INIT;
1078
1079         prepare_alt_odb(r);
1080         for (odb = r->objects->odb; odb; odb = odb->next) {
1081                 *path = odb_loose_path(odb, &buf, oid);
1082                 fd = git_open(*path);
1083                 if (fd >= 0)
1084                         return fd;
1085
1086                 if (most_interesting_errno == ENOENT)
1087                         most_interesting_errno = errno;
1088         }
1089         errno = most_interesting_errno;
1090         return -1;
1091 }
1092
1093 static int quick_has_loose(struct repository *r,
1094                            const struct object_id *oid)
1095 {
1096         struct object_directory *odb;
1097
1098         prepare_alt_odb(r);
1099         for (odb = r->objects->odb; odb; odb = odb->next) {
1100                 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1101                         return 1;
1102         }
1103         return 0;
1104 }
1105
1106 /*
1107  * Map the loose object at "path" if it is not NULL, or the path found by
1108  * searching for a loose object named "oid".
1109  */
1110 static void *map_loose_object_1(struct repository *r, const char *path,
1111                              const struct object_id *oid, unsigned long *size)
1112 {
1113         void *map;
1114         int fd;
1115
1116         if (path)
1117                 fd = git_open(path);
1118         else
1119                 fd = open_loose_object(r, oid, &path);
1120         map = NULL;
1121         if (fd >= 0) {
1122                 struct stat st;
1123
1124                 if (!fstat(fd, &st)) {
1125                         *size = xsize_t(st.st_size);
1126                         if (!*size) {
1127                                 /* mmap() is forbidden on empty files */
1128                                 error(_("object file %s is empty"), path);
1129                                 close(fd);
1130                                 return NULL;
1131                         }
1132                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1133                 }
1134                 close(fd);
1135         }
1136         return map;
1137 }
1138
1139 void *map_loose_object(struct repository *r,
1140                        const struct object_id *oid,
1141                        unsigned long *size)
1142 {
1143         return map_loose_object_1(r, NULL, oid, size);
1144 }
1145
1146 static int unpack_loose_short_header(git_zstream *stream,
1147                                      unsigned char *map, unsigned long mapsize,
1148                                      void *buffer, unsigned long bufsiz)
1149 {
1150         int ret;
1151
1152         /* Get the data stream */
1153         memset(stream, 0, sizeof(*stream));
1154         stream->next_in = map;
1155         stream->avail_in = mapsize;
1156         stream->next_out = buffer;
1157         stream->avail_out = bufsiz;
1158
1159         git_inflate_init(stream);
1160         obj_read_unlock();
1161         ret = git_inflate(stream, 0);
1162         obj_read_lock();
1163
1164         return ret;
1165 }
1166
1167 int unpack_loose_header(git_zstream *stream,
1168                         unsigned char *map, unsigned long mapsize,
1169                         void *buffer, unsigned long bufsiz)
1170 {
1171         int status = unpack_loose_short_header(stream, map, mapsize,
1172                                                buffer, bufsiz);
1173
1174         if (status < Z_OK)
1175                 return status;
1176
1177         /* Make sure we have the terminating NUL */
1178         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1179                 return -1;
1180         return 0;
1181 }
1182
1183 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1184                                          unsigned long mapsize, void *buffer,
1185                                          unsigned long bufsiz, struct strbuf *header)
1186 {
1187         int status;
1188
1189         status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1190         if (status < Z_OK)
1191                 return -1;
1192
1193         /*
1194          * Check if entire header is unpacked in the first iteration.
1195          */
1196         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1197                 return 0;
1198
1199         /*
1200          * buffer[0..bufsiz] was not large enough.  Copy the partial
1201          * result out to header, and then append the result of further
1202          * reading the stream.
1203          */
1204         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1205         stream->next_out = buffer;
1206         stream->avail_out = bufsiz;
1207
1208         do {
1209                 obj_read_unlock();
1210                 status = git_inflate(stream, 0);
1211                 obj_read_lock();
1212                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1213                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1214                         return 0;
1215                 stream->next_out = buffer;
1216                 stream->avail_out = bufsiz;
1217         } while (status != Z_STREAM_END);
1218         return -1;
1219 }
1220
1221 static void *unpack_loose_rest(git_zstream *stream,
1222                                void *buffer, unsigned long size,
1223                                const struct object_id *oid)
1224 {
1225         int bytes = strlen(buffer) + 1;
1226         unsigned char *buf = xmallocz(size);
1227         unsigned long n;
1228         int status = Z_OK;
1229
1230         n = stream->total_out - bytes;
1231         if (n > size)
1232                 n = size;
1233         memcpy(buf, (char *) buffer + bytes, n);
1234         bytes = n;
1235         if (bytes <= size) {
1236                 /*
1237                  * The above condition must be (bytes <= size), not
1238                  * (bytes < size).  In other words, even though we
1239                  * expect no more output and set avail_out to zero,
1240                  * the input zlib stream may have bytes that express
1241                  * "this concludes the stream", and we *do* want to
1242                  * eat that input.
1243                  *
1244                  * Otherwise we would not be able to test that we
1245                  * consumed all the input to reach the expected size;
1246                  * we also want to check that zlib tells us that all
1247                  * went well with status == Z_STREAM_END at the end.
1248                  */
1249                 stream->next_out = buf + bytes;
1250                 stream->avail_out = size - bytes;
1251                 while (status == Z_OK) {
1252                         obj_read_unlock();
1253                         status = git_inflate(stream, Z_FINISH);
1254                         obj_read_lock();
1255                 }
1256         }
1257         if (status == Z_STREAM_END && !stream->avail_in) {
1258                 git_inflate_end(stream);
1259                 return buf;
1260         }
1261
1262         if (status < 0)
1263                 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1264         else if (stream->avail_in)
1265                 error(_("garbage at end of loose object '%s'"),
1266                       oid_to_hex(oid));
1267         free(buf);
1268         return NULL;
1269 }
1270
1271 /*
1272  * We used to just use "sscanf()", but that's actually way
1273  * too permissive for what we want to check. So do an anal
1274  * object header parse by hand.
1275  */
1276 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1277                                        unsigned int flags)
1278 {
1279         const char *type_buf = hdr;
1280         unsigned long size;
1281         int type, type_len = 0;
1282
1283         /*
1284          * The type can be of any size but is followed by
1285          * a space.
1286          */
1287         for (;;) {
1288                 char c = *hdr++;
1289                 if (!c)
1290                         return -1;
1291                 if (c == ' ')
1292                         break;
1293                 type_len++;
1294         }
1295
1296         type = type_from_string_gently(type_buf, type_len, 1);
1297         if (oi->type_name)
1298                 strbuf_add(oi->type_name, type_buf, type_len);
1299         /*
1300          * Set type to 0 if its an unknown object and
1301          * we're obtaining the type using '--allow-unknown-type'
1302          * option.
1303          */
1304         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1305                 type = 0;
1306         else if (type < 0)
1307                 die(_("invalid object type"));
1308         if (oi->typep)
1309                 *oi->typep = type;
1310
1311         /*
1312          * The length must follow immediately, and be in canonical
1313          * decimal format (ie "010" is not valid).
1314          */
1315         size = *hdr++ - '0';
1316         if (size > 9)
1317                 return -1;
1318         if (size) {
1319                 for (;;) {
1320                         unsigned long c = *hdr - '0';
1321                         if (c > 9)
1322                                 break;
1323                         hdr++;
1324                         size = size * 10 + c;
1325                 }
1326         }
1327
1328         if (oi->sizep)
1329                 *oi->sizep = size;
1330
1331         /*
1332          * The length must be followed by a zero byte
1333          */
1334         return *hdr ? -1 : type;
1335 }
1336
1337 int parse_loose_header(const char *hdr, unsigned long *sizep)
1338 {
1339         struct object_info oi = OBJECT_INFO_INIT;
1340
1341         oi.sizep = sizep;
1342         return parse_loose_header_extended(hdr, &oi, 0);
1343 }
1344
1345 static int loose_object_info(struct repository *r,
1346                              const struct object_id *oid,
1347                              struct object_info *oi, int flags)
1348 {
1349         int status = 0;
1350         unsigned long mapsize;
1351         void *map;
1352         git_zstream stream;
1353         char hdr[MAX_HEADER_LEN];
1354         struct strbuf hdrbuf = STRBUF_INIT;
1355         unsigned long size_scratch;
1356
1357         if (oi->delta_base_oid)
1358                 oidclr(oi->delta_base_oid);
1359
1360         /*
1361          * If we don't care about type or size, then we don't
1362          * need to look inside the object at all. Note that we
1363          * do not optimize out the stat call, even if the
1364          * caller doesn't care about the disk-size, since our
1365          * return value implicitly indicates whether the
1366          * object even exists.
1367          */
1368         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1369                 const char *path;
1370                 struct stat st;
1371                 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1372                         return quick_has_loose(r, oid) ? 0 : -1;
1373                 if (stat_loose_object(r, oid, &st, &path) < 0)
1374                         return -1;
1375                 if (oi->disk_sizep)
1376                         *oi->disk_sizep = st.st_size;
1377                 return 0;
1378         }
1379
1380         map = map_loose_object(r, oid, &mapsize);
1381         if (!map)
1382                 return -1;
1383
1384         if (!oi->sizep)
1385                 oi->sizep = &size_scratch;
1386
1387         if (oi->disk_sizep)
1388                 *oi->disk_sizep = mapsize;
1389         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1390                 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1391                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1392                                        oid_to_hex(oid));
1393         } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1394                 status = error(_("unable to unpack %s header"),
1395                                oid_to_hex(oid));
1396         if (status < 0)
1397                 ; /* Do nothing */
1398         else if (hdrbuf.len) {
1399                 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1400                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1401                                        oid_to_hex(oid));
1402         } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1403                 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1404
1405         if (status >= 0 && oi->contentp) {
1406                 *oi->contentp = unpack_loose_rest(&stream, hdr,
1407                                                   *oi->sizep, oid);
1408                 if (!*oi->contentp) {
1409                         git_inflate_end(&stream);
1410                         status = -1;
1411                 }
1412         } else
1413                 git_inflate_end(&stream);
1414
1415         munmap(map, mapsize);
1416         if (status && oi->typep)
1417                 *oi->typep = status;
1418         if (oi->sizep == &size_scratch)
1419                 oi->sizep = NULL;
1420         strbuf_release(&hdrbuf);
1421         oi->whence = OI_LOOSE;
1422         return (status < 0) ? status : 0;
1423 }
1424
1425 int obj_read_use_lock = 0;
1426 pthread_mutex_t obj_read_mutex;
1427
1428 void enable_obj_read_lock(void)
1429 {
1430         if (obj_read_use_lock)
1431                 return;
1432
1433         obj_read_use_lock = 1;
1434         init_recursive_mutex(&obj_read_mutex);
1435 }
1436
1437 void disable_obj_read_lock(void)
1438 {
1439         if (!obj_read_use_lock)
1440                 return;
1441
1442         obj_read_use_lock = 0;
1443         pthread_mutex_destroy(&obj_read_mutex);
1444 }
1445
1446 int fetch_if_missing = 1;
1447
1448 static int do_oid_object_info_extended(struct repository *r,
1449                                        const struct object_id *oid,
1450                                        struct object_info *oi, unsigned flags)
1451 {
1452         static struct object_info blank_oi = OBJECT_INFO_INIT;
1453         struct cached_object *co;
1454         struct pack_entry e;
1455         int rtype;
1456         const struct object_id *real = oid;
1457         int already_retried = 0;
1458
1459
1460         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1461                 real = lookup_replace_object(r, oid);
1462
1463         if (is_null_oid(real))
1464                 return -1;
1465
1466         if (!oi)
1467                 oi = &blank_oi;
1468
1469         co = find_cached_object(real);
1470         if (co) {
1471                 if (oi->typep)
1472                         *(oi->typep) = co->type;
1473                 if (oi->sizep)
1474                         *(oi->sizep) = co->size;
1475                 if (oi->disk_sizep)
1476                         *(oi->disk_sizep) = 0;
1477                 if (oi->delta_base_oid)
1478                         oidclr(oi->delta_base_oid);
1479                 if (oi->type_name)
1480                         strbuf_addstr(oi->type_name, type_name(co->type));
1481                 if (oi->contentp)
1482                         *oi->contentp = xmemdupz(co->buf, co->size);
1483                 oi->whence = OI_CACHED;
1484                 return 0;
1485         }
1486
1487         while (1) {
1488                 if (find_pack_entry(r, real, &e))
1489                         break;
1490
1491                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1492                         return -1;
1493
1494                 /* Most likely it's a loose object. */
1495                 if (!loose_object_info(r, real, oi, flags))
1496                         return 0;
1497
1498                 /* Not a loose object; someone else may have just packed it. */
1499                 if (!(flags & OBJECT_INFO_QUICK)) {
1500                         reprepare_packed_git(r);
1501                         if (find_pack_entry(r, real, &e))
1502                                 break;
1503                 }
1504
1505                 /* Check if it is a missing object */
1506                 if (fetch_if_missing && has_promisor_remote() &&
1507                     !already_retried && r == the_repository &&
1508                     !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1509                         /*
1510                          * TODO Investigate checking promisor_remote_get_direct()
1511                          * TODO return value and stopping on error here.
1512                          * TODO Pass a repository struct through
1513                          * promisor_remote_get_direct(), such that arbitrary
1514                          * repositories work.
1515                          */
1516                         promisor_remote_get_direct(r, real, 1);
1517                         already_retried = 1;
1518                         continue;
1519                 }
1520
1521                 return -1;
1522         }
1523
1524         if (oi == &blank_oi)
1525                 /*
1526                  * We know that the caller doesn't actually need the
1527                  * information below, so return early.
1528                  */
1529                 return 0;
1530         rtype = packed_object_info(r, e.p, e.offset, oi);
1531         if (rtype < 0) {
1532                 mark_bad_packed_object(e.p, real->hash);
1533                 return do_oid_object_info_extended(r, real, oi, 0);
1534         } else if (oi->whence == OI_PACKED) {
1535                 oi->u.packed.offset = e.offset;
1536                 oi->u.packed.pack = e.p;
1537                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1538                                          rtype == OBJ_OFS_DELTA);
1539         }
1540
1541         return 0;
1542 }
1543
1544 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1545                              struct object_info *oi, unsigned flags)
1546 {
1547         int ret;
1548         obj_read_lock();
1549         ret = do_oid_object_info_extended(r, oid, oi, flags);
1550         obj_read_unlock();
1551         return ret;
1552 }
1553
1554
1555 /* returns enum object_type or negative */
1556 int oid_object_info(struct repository *r,
1557                     const struct object_id *oid,
1558                     unsigned long *sizep)
1559 {
1560         enum object_type type;
1561         struct object_info oi = OBJECT_INFO_INIT;
1562
1563         oi.typep = &type;
1564         oi.sizep = sizep;
1565         if (oid_object_info_extended(r, oid, &oi,
1566                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1567                 return -1;
1568         return type;
1569 }
1570
1571 static void *read_object(struct repository *r,
1572                          const struct object_id *oid, enum object_type *type,
1573                          unsigned long *size)
1574 {
1575         struct object_info oi = OBJECT_INFO_INIT;
1576         void *content;
1577         oi.typep = type;
1578         oi.sizep = size;
1579         oi.contentp = &content;
1580
1581         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1582                 return NULL;
1583         return content;
1584 }
1585
1586 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1587                         struct object_id *oid)
1588 {
1589         struct cached_object *co;
1590
1591         hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1592         if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1593             find_cached_object(oid))
1594                 return 0;
1595         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1596         co = &cached_objects[cached_object_nr++];
1597         co->size = len;
1598         co->type = type;
1599         co->buf = xmalloc(len);
1600         memcpy(co->buf, buf, len);
1601         oidcpy(&co->oid, oid);
1602         return 0;
1603 }
1604
1605 /*
1606  * This function dies on corrupt objects; the callers who want to
1607  * deal with them should arrange to call read_object() and give error
1608  * messages themselves.
1609  */
1610 void *read_object_file_extended(struct repository *r,
1611                                 const struct object_id *oid,
1612                                 enum object_type *type,
1613                                 unsigned long *size,
1614                                 int lookup_replace)
1615 {
1616         void *data;
1617         const struct packed_git *p;
1618         const char *path;
1619         struct stat st;
1620         const struct object_id *repl = lookup_replace ?
1621                 lookup_replace_object(r, oid) : oid;
1622
1623         errno = 0;
1624         data = read_object(r, repl, type, size);
1625         if (data)
1626                 return data;
1627
1628         obj_read_lock();
1629         if (errno && errno != ENOENT)
1630                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1631
1632         /* die if we replaced an object with one that does not exist */
1633         if (repl != oid)
1634                 die(_("replacement %s not found for %s"),
1635                     oid_to_hex(repl), oid_to_hex(oid));
1636
1637         if (!stat_loose_object(r, repl, &st, &path))
1638                 die(_("loose object %s (stored in %s) is corrupt"),
1639                     oid_to_hex(repl), path);
1640
1641         if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1642                 die(_("packed object %s (stored in %s) is corrupt"),
1643                     oid_to_hex(repl), p->pack_name);
1644         obj_read_unlock();
1645
1646         return NULL;
1647 }
1648
1649 void *read_object_with_reference(struct repository *r,
1650                                  const struct object_id *oid,
1651                                  const char *required_type_name,
1652                                  unsigned long *size,
1653                                  struct object_id *actual_oid_return)
1654 {
1655         enum object_type type, required_type;
1656         void *buffer;
1657         unsigned long isize;
1658         struct object_id actual_oid;
1659
1660         required_type = type_from_string(required_type_name);
1661         oidcpy(&actual_oid, oid);
1662         while (1) {
1663                 int ref_length = -1;
1664                 const char *ref_type = NULL;
1665
1666                 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1667                 if (!buffer)
1668                         return NULL;
1669                 if (type == required_type) {
1670                         *size = isize;
1671                         if (actual_oid_return)
1672                                 oidcpy(actual_oid_return, &actual_oid);
1673                         return buffer;
1674                 }
1675                 /* Handle references */
1676                 else if (type == OBJ_COMMIT)
1677                         ref_type = "tree ";
1678                 else if (type == OBJ_TAG)
1679                         ref_type = "object ";
1680                 else {
1681                         free(buffer);
1682                         return NULL;
1683                 }
1684                 ref_length = strlen(ref_type);
1685
1686                 if (ref_length + the_hash_algo->hexsz > isize ||
1687                     memcmp(buffer, ref_type, ref_length) ||
1688                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1689                         free(buffer);
1690                         return NULL;
1691                 }
1692                 free(buffer);
1693                 /* Now we have the ID of the referred-to object in
1694                  * actual_oid.  Check again. */
1695         }
1696 }
1697
1698 static void write_object_file_prepare(const struct git_hash_algo *algo,
1699                                       const void *buf, unsigned long len,
1700                                       const char *type, struct object_id *oid,
1701                                       char *hdr, int *hdrlen)
1702 {
1703         git_hash_ctx c;
1704
1705         /* Generate the header */
1706         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1707
1708         /* Sha1.. */
1709         algo->init_fn(&c);
1710         algo->update_fn(&c, hdr, *hdrlen);
1711         algo->update_fn(&c, buf, len);
1712         algo->final_fn(oid->hash, &c);
1713 }
1714
1715 /*
1716  * Move the just written object into its final resting place.
1717  */
1718 int finalize_object_file(const char *tmpfile, const char *filename)
1719 {
1720         int ret = 0;
1721
1722         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1723                 goto try_rename;
1724         else if (link(tmpfile, filename))
1725                 ret = errno;
1726
1727         /*
1728          * Coda hack - coda doesn't like cross-directory links,
1729          * so we fall back to a rename, which will mean that it
1730          * won't be able to check collisions, but that's not a
1731          * big deal.
1732          *
1733          * The same holds for FAT formatted media.
1734          *
1735          * When this succeeds, we just return.  We have nothing
1736          * left to unlink.
1737          */
1738         if (ret && ret != EEXIST) {
1739         try_rename:
1740                 if (!rename(tmpfile, filename))
1741                         goto out;
1742                 ret = errno;
1743         }
1744         unlink_or_warn(tmpfile);
1745         if (ret) {
1746                 if (ret != EEXIST) {
1747                         return error_errno(_("unable to write file %s"), filename);
1748                 }
1749                 /* FIXME!!! Collision check here ? */
1750         }
1751
1752 out:
1753         if (adjust_shared_perm(filename))
1754                 return error(_("unable to set permission to '%s'"), filename);
1755         return 0;
1756 }
1757
1758 static int write_buffer(int fd, const void *buf, size_t len)
1759 {
1760         if (write_in_full(fd, buf, len) < 0)
1761                 return error_errno(_("file write error"));
1762         return 0;
1763 }
1764
1765 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1766                      unsigned long len, const char *type,
1767                      struct object_id *oid)
1768 {
1769         char hdr[MAX_HEADER_LEN];
1770         int hdrlen = sizeof(hdr);
1771         write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1772         return 0;
1773 }
1774
1775 /* Finalize a file on disk, and close it. */
1776 static void close_loose_object(int fd)
1777 {
1778         if (fsync_object_files)
1779                 fsync_or_die(fd, "loose object file");
1780         if (close(fd) != 0)
1781                 die_errno(_("error when closing loose object file"));
1782 }
1783
1784 /* Size of directory component, including the ending '/' */
1785 static inline int directory_size(const char *filename)
1786 {
1787         const char *s = strrchr(filename, '/');
1788         if (!s)
1789                 return 0;
1790         return s - filename + 1;
1791 }
1792
1793 /*
1794  * This creates a temporary file in the same directory as the final
1795  * 'filename'
1796  *
1797  * We want to avoid cross-directory filename renames, because those
1798  * can have problems on various filesystems (FAT, NFS, Coda).
1799  */
1800 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1801 {
1802         int fd, dirlen = directory_size(filename);
1803
1804         strbuf_reset(tmp);
1805         strbuf_add(tmp, filename, dirlen);
1806         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1807         fd = git_mkstemp_mode(tmp->buf, 0444);
1808         if (fd < 0 && dirlen && errno == ENOENT) {
1809                 /*
1810                  * Make sure the directory exists; note that the contents
1811                  * of the buffer are undefined after mkstemp returns an
1812                  * error, so we have to rewrite the whole buffer from
1813                  * scratch.
1814                  */
1815                 strbuf_reset(tmp);
1816                 strbuf_add(tmp, filename, dirlen - 1);
1817                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1818                         return -1;
1819                 if (adjust_shared_perm(tmp->buf))
1820                         return -1;
1821
1822                 /* Try again */
1823                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1824                 fd = git_mkstemp_mode(tmp->buf, 0444);
1825         }
1826         return fd;
1827 }
1828
1829 static int write_loose_object(const struct object_id *oid, char *hdr,
1830                               int hdrlen, const void *buf, unsigned long len,
1831                               time_t mtime)
1832 {
1833         int fd, ret;
1834         unsigned char compressed[4096];
1835         git_zstream stream;
1836         git_hash_ctx c;
1837         struct object_id parano_oid;
1838         static struct strbuf tmp_file = STRBUF_INIT;
1839         static struct strbuf filename = STRBUF_INIT;
1840
1841         loose_object_path(the_repository, &filename, oid);
1842
1843         fd = create_tmpfile(&tmp_file, filename.buf);
1844         if (fd < 0) {
1845                 if (errno == EACCES)
1846                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1847                 else
1848                         return error_errno(_("unable to create temporary file"));
1849         }
1850
1851         /* Set it up */
1852         git_deflate_init(&stream, zlib_compression_level);
1853         stream.next_out = compressed;
1854         stream.avail_out = sizeof(compressed);
1855         the_hash_algo->init_fn(&c);
1856
1857         /* First header.. */
1858         stream.next_in = (unsigned char *)hdr;
1859         stream.avail_in = hdrlen;
1860         while (git_deflate(&stream, 0) == Z_OK)
1861                 ; /* nothing */
1862         the_hash_algo->update_fn(&c, hdr, hdrlen);
1863
1864         /* Then the data itself.. */
1865         stream.next_in = (void *)buf;
1866         stream.avail_in = len;
1867         do {
1868                 unsigned char *in0 = stream.next_in;
1869                 ret = git_deflate(&stream, Z_FINISH);
1870                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1871                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1872                         die(_("unable to write loose object file"));
1873                 stream.next_out = compressed;
1874                 stream.avail_out = sizeof(compressed);
1875         } while (ret == Z_OK);
1876
1877         if (ret != Z_STREAM_END)
1878                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1879                     ret);
1880         ret = git_deflate_end_gently(&stream);
1881         if (ret != Z_OK)
1882                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1883                     ret);
1884         the_hash_algo->final_fn(parano_oid.hash, &c);
1885         if (!oideq(oid, &parano_oid))
1886                 die(_("confused by unstable object source data for %s"),
1887                     oid_to_hex(oid));
1888
1889         close_loose_object(fd);
1890
1891         if (mtime) {
1892                 struct utimbuf utb;
1893                 utb.actime = mtime;
1894                 utb.modtime = mtime;
1895                 if (utime(tmp_file.buf, &utb) < 0)
1896                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1897         }
1898
1899         return finalize_object_file(tmp_file.buf, filename.buf);
1900 }
1901
1902 static int freshen_loose_object(const struct object_id *oid)
1903 {
1904         return check_and_freshen(oid, 1);
1905 }
1906
1907 static int freshen_packed_object(const struct object_id *oid)
1908 {
1909         struct pack_entry e;
1910         if (!find_pack_entry(the_repository, oid, &e))
1911                 return 0;
1912         if (e.p->freshened)
1913                 return 1;
1914         if (!freshen_file(e.p->pack_name))
1915                 return 0;
1916         e.p->freshened = 1;
1917         return 1;
1918 }
1919
1920 int write_object_file(const void *buf, unsigned long len, const char *type,
1921                       struct object_id *oid)
1922 {
1923         char hdr[MAX_HEADER_LEN];
1924         int hdrlen = sizeof(hdr);
1925
1926         /* Normally if we have it in the pack then we do not bother writing
1927          * it out into .git/objects/??/?{38} file.
1928          */
1929         write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1930                                   &hdrlen);
1931         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1932                 return 0;
1933         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1934 }
1935
1936 int hash_object_file_literally(const void *buf, unsigned long len,
1937                                const char *type, struct object_id *oid,
1938                                unsigned flags)
1939 {
1940         char *header;
1941         int hdrlen, status = 0;
1942
1943         /* type string, SP, %lu of the length plus NUL must fit this */
1944         hdrlen = strlen(type) + MAX_HEADER_LEN;
1945         header = xmalloc(hdrlen);
1946         write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1947                                   &hdrlen);
1948
1949         if (!(flags & HASH_WRITE_OBJECT))
1950                 goto cleanup;
1951         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1952                 goto cleanup;
1953         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1954
1955 cleanup:
1956         free(header);
1957         return status;
1958 }
1959
1960 int force_object_loose(const struct object_id *oid, time_t mtime)
1961 {
1962         void *buf;
1963         unsigned long len;
1964         enum object_type type;
1965         char hdr[MAX_HEADER_LEN];
1966         int hdrlen;
1967         int ret;
1968
1969         if (has_loose_object(oid))
1970                 return 0;
1971         buf = read_object(the_repository, oid, &type, &len);
1972         if (!buf)
1973                 return error(_("cannot read object for %s"), oid_to_hex(oid));
1974         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1975         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1976         free(buf);
1977
1978         return ret;
1979 }
1980
1981 int repo_has_object_file_with_flags(struct repository *r,
1982                                     const struct object_id *oid, int flags)
1983 {
1984         if (!startup_info->have_repository)
1985                 return 0;
1986         return oid_object_info_extended(r, oid, NULL, flags) >= 0;
1987 }
1988
1989 int repo_has_object_file(struct repository *r,
1990                          const struct object_id *oid)
1991 {
1992         return repo_has_object_file_with_flags(r, oid, 0);
1993 }
1994
1995 static void check_tree(const void *buf, size_t size)
1996 {
1997         struct tree_desc desc;
1998         struct name_entry entry;
1999
2000         init_tree_desc(&desc, buf, size);
2001         while (tree_entry(&desc, &entry))
2002                 /* do nothing
2003                  * tree_entry() will die() on malformed entries */
2004                 ;
2005 }
2006
2007 static void check_commit(const void *buf, size_t size)
2008 {
2009         struct commit c;
2010         memset(&c, 0, sizeof(c));
2011         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2012                 die(_("corrupt commit"));
2013 }
2014
2015 static void check_tag(const void *buf, size_t size)
2016 {
2017         struct tag t;
2018         memset(&t, 0, sizeof(t));
2019         if (parse_tag_buffer(the_repository, &t, buf, size))
2020                 die(_("corrupt tag"));
2021 }
2022
2023 static int index_mem(struct index_state *istate,
2024                      struct object_id *oid, void *buf, size_t size,
2025                      enum object_type type,
2026                      const char *path, unsigned flags)
2027 {
2028         int ret, re_allocated = 0;
2029         int write_object = flags & HASH_WRITE_OBJECT;
2030
2031         if (!type)
2032                 type = OBJ_BLOB;
2033
2034         /*
2035          * Convert blobs to git internal format
2036          */
2037         if ((type == OBJ_BLOB) && path) {
2038                 struct strbuf nbuf = STRBUF_INIT;
2039                 if (convert_to_git(istate, path, buf, size, &nbuf,
2040                                    get_conv_flags(flags))) {
2041                         buf = strbuf_detach(&nbuf, &size);
2042                         re_allocated = 1;
2043                 }
2044         }
2045         if (flags & HASH_FORMAT_CHECK) {
2046                 if (type == OBJ_TREE)
2047                         check_tree(buf, size);
2048                 if (type == OBJ_COMMIT)
2049                         check_commit(buf, size);
2050                 if (type == OBJ_TAG)
2051                         check_tag(buf, size);
2052         }
2053
2054         if (write_object)
2055                 ret = write_object_file(buf, size, type_name(type), oid);
2056         else
2057                 ret = hash_object_file(the_hash_algo, buf, size,
2058                                        type_name(type), oid);
2059         if (re_allocated)
2060                 free(buf);
2061         return ret;
2062 }
2063
2064 static int index_stream_convert_blob(struct index_state *istate,
2065                                      struct object_id *oid,
2066                                      int fd,
2067                                      const char *path,
2068                                      unsigned flags)
2069 {
2070         int ret;
2071         const int write_object = flags & HASH_WRITE_OBJECT;
2072         struct strbuf sbuf = STRBUF_INIT;
2073
2074         assert(path);
2075         assert(would_convert_to_git_filter_fd(istate, path));
2076
2077         convert_to_git_filter_fd(istate, path, fd, &sbuf,
2078                                  get_conv_flags(flags));
2079
2080         if (write_object)
2081                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2082                                         oid);
2083         else
2084                 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2085                                        type_name(OBJ_BLOB), oid);
2086         strbuf_release(&sbuf);
2087         return ret;
2088 }
2089
2090 static int index_pipe(struct index_state *istate, struct object_id *oid,
2091                       int fd, enum object_type type,
2092                       const char *path, unsigned flags)
2093 {
2094         struct strbuf sbuf = STRBUF_INIT;
2095         int ret;
2096
2097         if (strbuf_read(&sbuf, fd, 4096) >= 0)
2098                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2099         else
2100                 ret = -1;
2101         strbuf_release(&sbuf);
2102         return ret;
2103 }
2104
2105 #define SMALL_FILE_SIZE (32*1024)
2106
2107 static int index_core(struct index_state *istate,
2108                       struct object_id *oid, int fd, size_t size,
2109                       enum object_type type, const char *path,
2110                       unsigned flags)
2111 {
2112         int ret;
2113
2114         if (!size) {
2115                 ret = index_mem(istate, oid, "", size, type, path, flags);
2116         } else if (size <= SMALL_FILE_SIZE) {
2117                 char *buf = xmalloc(size);
2118                 ssize_t read_result = read_in_full(fd, buf, size);
2119                 if (read_result < 0)
2120                         ret = error_errno(_("read error while indexing %s"),
2121                                           path ? path : "<unknown>");
2122                 else if (read_result != size)
2123                         ret = error(_("short read while indexing %s"),
2124                                     path ? path : "<unknown>");
2125                 else
2126                         ret = index_mem(istate, oid, buf, size, type, path, flags);
2127                 free(buf);
2128         } else {
2129                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2130                 ret = index_mem(istate, oid, buf, size, type, path, flags);
2131                 munmap(buf, size);
2132         }
2133         return ret;
2134 }
2135
2136 /*
2137  * This creates one packfile per large blob unless bulk-checkin
2138  * machinery is "plugged".
2139  *
2140  * This also bypasses the usual "convert-to-git" dance, and that is on
2141  * purpose. We could write a streaming version of the converting
2142  * functions and insert that before feeding the data to fast-import
2143  * (or equivalent in-core API described above). However, that is
2144  * somewhat complicated, as we do not know the size of the filter
2145  * result, which we need to know beforehand when writing a git object.
2146  * Since the primary motivation for trying to stream from the working
2147  * tree file and to avoid mmaping it in core is to deal with large
2148  * binary blobs, they generally do not want to get any conversion, and
2149  * callers should avoid this code path when filters are requested.
2150  */
2151 static int index_stream(struct object_id *oid, int fd, size_t size,
2152                         enum object_type type, const char *path,
2153                         unsigned flags)
2154 {
2155         return index_bulk_checkin(oid, fd, size, type, path, flags);
2156 }
2157
2158 int index_fd(struct index_state *istate, struct object_id *oid,
2159              int fd, struct stat *st,
2160              enum object_type type, const char *path, unsigned flags)
2161 {
2162         int ret;
2163
2164         /*
2165          * Call xsize_t() only when needed to avoid potentially unnecessary
2166          * die() for large files.
2167          */
2168         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2169                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2170         else if (!S_ISREG(st->st_mode))
2171                 ret = index_pipe(istate, oid, fd, type, path, flags);
2172         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2173                  (path && would_convert_to_git(istate, path)))
2174                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2175                                  type, path, flags);
2176         else
2177                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2178                                    flags);
2179         close(fd);
2180         return ret;
2181 }
2182
2183 int index_path(struct index_state *istate, struct object_id *oid,
2184                const char *path, struct stat *st, unsigned flags)
2185 {
2186         int fd;
2187         struct strbuf sb = STRBUF_INIT;
2188         int rc = 0;
2189
2190         switch (st->st_mode & S_IFMT) {
2191         case S_IFREG:
2192                 fd = open(path, O_RDONLY);
2193                 if (fd < 0)
2194                         return error_errno("open(\"%s\")", path);
2195                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2196                         return error(_("%s: failed to insert into database"),
2197                                      path);
2198                 break;
2199         case S_IFLNK:
2200                 if (strbuf_readlink(&sb, path, st->st_size))
2201                         return error_errno("readlink(\"%s\")", path);
2202                 if (!(flags & HASH_WRITE_OBJECT))
2203                         hash_object_file(the_hash_algo, sb.buf, sb.len,
2204                                          blob_type, oid);
2205                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2206                         rc = error(_("%s: failed to insert into database"), path);
2207                 strbuf_release(&sb);
2208                 break;
2209         case S_IFDIR:
2210                 return resolve_gitlink_ref(path, "HEAD", oid);
2211         default:
2212                 return error(_("%s: unsupported file type"), path);
2213         }
2214         return rc;
2215 }
2216
2217 int read_pack_header(int fd, struct pack_header *header)
2218 {
2219         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2220                 /* "eof before pack header was fully read" */
2221                 return PH_ERROR_EOF;
2222
2223         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2224                 /* "protocol error (pack signature mismatch detected)" */
2225                 return PH_ERROR_PACK_SIGNATURE;
2226         if (!pack_version_ok(header->hdr_version))
2227                 /* "protocol error (pack version unsupported)" */
2228                 return PH_ERROR_PROTOCOL;
2229         return 0;
2230 }
2231
2232 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2233 {
2234         enum object_type type = oid_object_info(the_repository, oid, NULL);
2235         if (type < 0)
2236                 die(_("%s is not a valid object"), oid_to_hex(oid));
2237         if (type != expect)
2238                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2239                     type_name(expect));
2240 }
2241
2242 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2243                                 struct strbuf *path,
2244                                 each_loose_object_fn obj_cb,
2245                                 each_loose_cruft_fn cruft_cb,
2246                                 each_loose_subdir_fn subdir_cb,
2247                                 void *data)
2248 {
2249         size_t origlen, baselen;
2250         DIR *dir;
2251         struct dirent *de;
2252         int r = 0;
2253         struct object_id oid;
2254
2255         if (subdir_nr > 0xff)
2256                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2257
2258         origlen = path->len;
2259         strbuf_complete(path, '/');
2260         strbuf_addf(path, "%02x", subdir_nr);
2261
2262         dir = opendir(path->buf);
2263         if (!dir) {
2264                 if (errno != ENOENT)
2265                         r = error_errno(_("unable to open %s"), path->buf);
2266                 strbuf_setlen(path, origlen);
2267                 return r;
2268         }
2269
2270         oid.hash[0] = subdir_nr;
2271         strbuf_addch(path, '/');
2272         baselen = path->len;
2273
2274         while ((de = readdir(dir))) {
2275                 size_t namelen;
2276                 if (is_dot_or_dotdot(de->d_name))
2277                         continue;
2278
2279                 namelen = strlen(de->d_name);
2280                 strbuf_setlen(path, baselen);
2281                 strbuf_add(path, de->d_name, namelen);
2282                 if (namelen == the_hash_algo->hexsz - 2 &&
2283                     !hex_to_bytes(oid.hash + 1, de->d_name,
2284                                   the_hash_algo->rawsz - 1)) {
2285                         if (obj_cb) {
2286                                 r = obj_cb(&oid, path->buf, data);
2287                                 if (r)
2288                                         break;
2289                         }
2290                         continue;
2291                 }
2292
2293                 if (cruft_cb) {
2294                         r = cruft_cb(de->d_name, path->buf, data);
2295                         if (r)
2296                                 break;
2297                 }
2298         }
2299         closedir(dir);
2300
2301         strbuf_setlen(path, baselen - 1);
2302         if (!r && subdir_cb)
2303                 r = subdir_cb(subdir_nr, path->buf, data);
2304
2305         strbuf_setlen(path, origlen);
2306
2307         return r;
2308 }
2309
2310 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2311                             each_loose_object_fn obj_cb,
2312                             each_loose_cruft_fn cruft_cb,
2313                             each_loose_subdir_fn subdir_cb,
2314                             void *data)
2315 {
2316         int r = 0;
2317         int i;
2318
2319         for (i = 0; i < 256; i++) {
2320                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2321                                                 subdir_cb, data);
2322                 if (r)
2323                         break;
2324         }
2325
2326         return r;
2327 }
2328
2329 int for_each_loose_file_in_objdir(const char *path,
2330                                   each_loose_object_fn obj_cb,
2331                                   each_loose_cruft_fn cruft_cb,
2332                                   each_loose_subdir_fn subdir_cb,
2333                                   void *data)
2334 {
2335         struct strbuf buf = STRBUF_INIT;
2336         int r;
2337
2338         strbuf_addstr(&buf, path);
2339         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2340                                               subdir_cb, data);
2341         strbuf_release(&buf);
2342
2343         return r;
2344 }
2345
2346 int for_each_loose_object(each_loose_object_fn cb, void *data,
2347                           enum for_each_object_flags flags)
2348 {
2349         struct object_directory *odb;
2350
2351         prepare_alt_odb(the_repository);
2352         for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2353                 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2354                                                       NULL, data);
2355                 if (r)
2356                         return r;
2357
2358                 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2359                         break;
2360         }
2361
2362         return 0;
2363 }
2364
2365 static int append_loose_object(const struct object_id *oid, const char *path,
2366                                void *data)
2367 {
2368         oid_array_append(data, oid);
2369         return 0;
2370 }
2371
2372 struct oid_array *odb_loose_cache(struct object_directory *odb,
2373                                   const struct object_id *oid)
2374 {
2375         int subdir_nr = oid->hash[0];
2376         struct strbuf buf = STRBUF_INIT;
2377
2378         if (subdir_nr < 0 ||
2379             subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2380                 BUG("subdir_nr out of range");
2381
2382         if (odb->loose_objects_subdir_seen[subdir_nr])
2383                 return &odb->loose_objects_cache[subdir_nr];
2384
2385         strbuf_addstr(&buf, odb->path);
2386         for_each_file_in_obj_subdir(subdir_nr, &buf,
2387                                     append_loose_object,
2388                                     NULL, NULL,
2389                                     &odb->loose_objects_cache[subdir_nr]);
2390         odb->loose_objects_subdir_seen[subdir_nr] = 1;
2391         strbuf_release(&buf);
2392         return &odb->loose_objects_cache[subdir_nr];
2393 }
2394
2395 void odb_clear_loose_cache(struct object_directory *odb)
2396 {
2397         int i;
2398
2399         for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2400                 oid_array_clear(&odb->loose_objects_cache[i]);
2401         memset(&odb->loose_objects_subdir_seen, 0,
2402                sizeof(odb->loose_objects_subdir_seen));
2403 }
2404
2405 static int check_stream_oid(git_zstream *stream,
2406                             const char *hdr,
2407                             unsigned long size,
2408                             const char *path,
2409                             const struct object_id *expected_oid)
2410 {
2411         git_hash_ctx c;
2412         struct object_id real_oid;
2413         unsigned char buf[4096];
2414         unsigned long total_read;
2415         int status = Z_OK;
2416
2417         the_hash_algo->init_fn(&c);
2418         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2419
2420         /*
2421          * We already read some bytes into hdr, but the ones up to the NUL
2422          * do not count against the object's content size.
2423          */
2424         total_read = stream->total_out - strlen(hdr) - 1;
2425
2426         /*
2427          * This size comparison must be "<=" to read the final zlib packets;
2428          * see the comment in unpack_loose_rest for details.
2429          */
2430         while (total_read <= size &&
2431                (status == Z_OK ||
2432                 (status == Z_BUF_ERROR && !stream->avail_out))) {
2433                 stream->next_out = buf;
2434                 stream->avail_out = sizeof(buf);
2435                 if (size - total_read < stream->avail_out)
2436                         stream->avail_out = size - total_read;
2437                 status = git_inflate(stream, Z_FINISH);
2438                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2439                 total_read += stream->next_out - buf;
2440         }
2441         git_inflate_end(stream);
2442
2443         if (status != Z_STREAM_END) {
2444                 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2445                 return -1;
2446         }
2447         if (stream->avail_in) {
2448                 error(_("garbage at end of loose object '%s'"),
2449                       oid_to_hex(expected_oid));
2450                 return -1;
2451         }
2452
2453         the_hash_algo->final_fn(real_oid.hash, &c);
2454         if (!oideq(expected_oid, &real_oid)) {
2455                 error(_("hash mismatch for %s (expected %s)"), path,
2456                       oid_to_hex(expected_oid));
2457                 return -1;
2458         }
2459
2460         return 0;
2461 }
2462
2463 int read_loose_object(const char *path,
2464                       const struct object_id *expected_oid,
2465                       enum object_type *type,
2466                       unsigned long *size,
2467                       void **contents)
2468 {
2469         int ret = -1;
2470         void *map = NULL;
2471         unsigned long mapsize;
2472         git_zstream stream;
2473         char hdr[MAX_HEADER_LEN];
2474
2475         *contents = NULL;
2476
2477         map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2478         if (!map) {
2479                 error_errno(_("unable to mmap %s"), path);
2480                 goto out;
2481         }
2482
2483         if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2484                 error(_("unable to unpack header of %s"), path);
2485                 goto out;
2486         }
2487
2488         *type = parse_loose_header(hdr, size);
2489         if (*type < 0) {
2490                 error(_("unable to parse header of %s"), path);
2491                 git_inflate_end(&stream);
2492                 goto out;
2493         }
2494
2495         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2496                 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2497                         goto out;
2498         } else {
2499                 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2500                 if (!*contents) {
2501                         error(_("unable to unpack contents of %s"), path);
2502                         git_inflate_end(&stream);
2503                         goto out;
2504                 }
2505                 if (check_object_signature(the_repository, expected_oid,
2506                                            *contents, *size,
2507                                            type_name(*type))) {
2508                         error(_("hash mismatch for %s (expected %s)"), path,
2509                               oid_to_hex(expected_oid));
2510                         free(*contents);
2511                         goto out;
2512                 }
2513         }
2514
2515         ret = 0; /* everything checks out */
2516
2517 out:
2518         if (map)
2519                 munmap(map, mapsize);
2520         return ret;
2521 }