fast-import: make find_marks work on any mark set
[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_clone(git_hash_ctx *dst, const git_hash_ctx *src)
78 {
79         git_SHA1_Clone(&dst->sha1, &src->sha1);
80 }
81
82 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
83 {
84         git_SHA1_Update(&ctx->sha1, data, len);
85 }
86
87 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
88 {
89         git_SHA1_Final(hash, &ctx->sha1);
90 }
91
92
93 static void git_hash_sha256_init(git_hash_ctx *ctx)
94 {
95         git_SHA256_Init(&ctx->sha256);
96 }
97
98 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
99 {
100         git_SHA256_Clone(&dst->sha256, &src->sha256);
101 }
102
103 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
104 {
105         git_SHA256_Update(&ctx->sha256, data, len);
106 }
107
108 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
109 {
110         git_SHA256_Final(hash, &ctx->sha256);
111 }
112
113 static void git_hash_unknown_init(git_hash_ctx *ctx)
114 {
115         BUG("trying to init unknown hash");
116 }
117
118 static void git_hash_unknown_clone(git_hash_ctx *dst, const git_hash_ctx *src)
119 {
120         BUG("trying to clone unknown hash");
121 }
122
123 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
124 {
125         BUG("trying to update unknown hash");
126 }
127
128 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
129 {
130         BUG("trying to finalize unknown hash");
131 }
132
133 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
134         {
135                 NULL,
136                 0x00000000,
137                 0,
138                 0,
139                 0,
140                 git_hash_unknown_init,
141                 git_hash_unknown_clone,
142                 git_hash_unknown_update,
143                 git_hash_unknown_final,
144                 NULL,
145                 NULL,
146         },
147         {
148                 "sha1",
149                 /* "sha1", big-endian */
150                 0x73686131,
151                 GIT_SHA1_RAWSZ,
152                 GIT_SHA1_HEXSZ,
153                 GIT_SHA1_BLKSZ,
154                 git_hash_sha1_init,
155                 git_hash_sha1_clone,
156                 git_hash_sha1_update,
157                 git_hash_sha1_final,
158                 &empty_tree_oid,
159                 &empty_blob_oid,
160         },
161         {
162                 "sha256",
163                 /* "s256", big-endian */
164                 0x73323536,
165                 GIT_SHA256_RAWSZ,
166                 GIT_SHA256_HEXSZ,
167                 GIT_SHA256_BLKSZ,
168                 git_hash_sha256_init,
169                 git_hash_sha256_clone,
170                 git_hash_sha256_update,
171                 git_hash_sha256_final,
172                 &empty_tree_oid_sha256,
173                 &empty_blob_oid_sha256,
174         }
175 };
176
177 const char *empty_tree_oid_hex(void)
178 {
179         static char buf[GIT_MAX_HEXSZ + 1];
180         return oid_to_hex_r(buf, the_hash_algo->empty_tree);
181 }
182
183 const char *empty_blob_oid_hex(void)
184 {
185         static char buf[GIT_MAX_HEXSZ + 1];
186         return oid_to_hex_r(buf, the_hash_algo->empty_blob);
187 }
188
189 int hash_algo_by_name(const char *name)
190 {
191         int i;
192         if (!name)
193                 return GIT_HASH_UNKNOWN;
194         for (i = 1; i < GIT_HASH_NALGOS; i++)
195                 if (!strcmp(name, hash_algos[i].name))
196                         return i;
197         return GIT_HASH_UNKNOWN;
198 }
199
200 int hash_algo_by_id(uint32_t format_id)
201 {
202         int i;
203         for (i = 1; i < GIT_HASH_NALGOS; i++)
204                 if (format_id == hash_algos[i].format_id)
205                         return i;
206         return GIT_HASH_UNKNOWN;
207 }
208
209 int hash_algo_by_length(int len)
210 {
211         int i;
212         for (i = 1; i < GIT_HASH_NALGOS; i++)
213                 if (len == hash_algos[i].rawsz)
214                         return i;
215         return GIT_HASH_UNKNOWN;
216 }
217
218 /*
219  * This is meant to hold a *small* number of objects that you would
220  * want read_object_file() to be able to return, but yet you do not want
221  * to write them into the object store (e.g. a browse-only
222  * application).
223  */
224 static struct cached_object {
225         struct object_id oid;
226         enum object_type type;
227         void *buf;
228         unsigned long size;
229 } *cached_objects;
230 static int cached_object_nr, cached_object_alloc;
231
232 static struct cached_object empty_tree = {
233         { EMPTY_TREE_SHA1_BIN_LITERAL },
234         OBJ_TREE,
235         "",
236         0
237 };
238
239 static struct cached_object *find_cached_object(const struct object_id *oid)
240 {
241         int i;
242         struct cached_object *co = cached_objects;
243
244         for (i = 0; i < cached_object_nr; i++, co++) {
245                 if (oideq(&co->oid, oid))
246                         return co;
247         }
248         if (oideq(oid, the_hash_algo->empty_tree))
249                 return &empty_tree;
250         return NULL;
251 }
252
253
254 static int get_conv_flags(unsigned flags)
255 {
256         if (flags & HASH_RENORMALIZE)
257                 return CONV_EOL_RENORMALIZE;
258         else if (flags & HASH_WRITE_OBJECT)
259                 return global_conv_flags_eol | CONV_WRITE_OBJECT;
260         else
261                 return 0;
262 }
263
264
265 int mkdir_in_gitdir(const char *path)
266 {
267         if (mkdir(path, 0777)) {
268                 int saved_errno = errno;
269                 struct stat st;
270                 struct strbuf sb = STRBUF_INIT;
271
272                 if (errno != EEXIST)
273                         return -1;
274                 /*
275                  * Are we looking at a path in a symlinked worktree
276                  * whose original repository does not yet have it?
277                  * e.g. .git/rr-cache pointing at its original
278                  * repository in which the user hasn't performed any
279                  * conflict resolution yet?
280                  */
281                 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
282                     strbuf_readlink(&sb, path, st.st_size) ||
283                     !is_absolute_path(sb.buf) ||
284                     mkdir(sb.buf, 0777)) {
285                         strbuf_release(&sb);
286                         errno = saved_errno;
287                         return -1;
288                 }
289                 strbuf_release(&sb);
290         }
291         return adjust_shared_perm(path);
292 }
293
294 enum scld_error safe_create_leading_directories(char *path)
295 {
296         char *next_component = path + offset_1st_component(path);
297         enum scld_error ret = SCLD_OK;
298
299         while (ret == SCLD_OK && next_component) {
300                 struct stat st;
301                 char *slash = next_component, slash_character;
302
303                 while (*slash && !is_dir_sep(*slash))
304                         slash++;
305
306                 if (!*slash)
307                         break;
308
309                 next_component = slash + 1;
310                 while (is_dir_sep(*next_component))
311                         next_component++;
312                 if (!*next_component)
313                         break;
314
315                 slash_character = *slash;
316                 *slash = '\0';
317                 if (!stat(path, &st)) {
318                         /* path exists */
319                         if (!S_ISDIR(st.st_mode)) {
320                                 errno = ENOTDIR;
321                                 ret = SCLD_EXISTS;
322                         }
323                 } else if (mkdir(path, 0777)) {
324                         if (errno == EEXIST &&
325                             !stat(path, &st) && S_ISDIR(st.st_mode))
326                                 ; /* somebody created it since we checked */
327                         else if (errno == ENOENT)
328                                 /*
329                                  * Either mkdir() failed because
330                                  * somebody just pruned the containing
331                                  * directory, or stat() failed because
332                                  * the file that was in our way was
333                                  * just removed.  Either way, inform
334                                  * the caller that it might be worth
335                                  * trying again:
336                                  */
337                                 ret = SCLD_VANISHED;
338                         else
339                                 ret = SCLD_FAILED;
340                 } else if (adjust_shared_perm(path)) {
341                         ret = SCLD_PERMS;
342                 }
343                 *slash = slash_character;
344         }
345         return ret;
346 }
347
348 enum scld_error safe_create_leading_directories_const(const char *path)
349 {
350         int save_errno;
351         /* path points to cache entries, so xstrdup before messing with it */
352         char *buf = xstrdup(path);
353         enum scld_error result = safe_create_leading_directories(buf);
354
355         save_errno = errno;
356         free(buf);
357         errno = save_errno;
358         return result;
359 }
360
361 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
362 {
363         /*
364          * The number of times we will try to remove empty directories
365          * in the way of path. This is only 1 because if another
366          * process is racily creating directories that conflict with
367          * us, we don't want to fight against them.
368          */
369         int remove_directories_remaining = 1;
370
371         /*
372          * The number of times that we will try to create the
373          * directories containing path. We are willing to attempt this
374          * more than once, because another process could be trying to
375          * clean up empty directories at the same time as we are
376          * trying to create them.
377          */
378         int create_directories_remaining = 3;
379
380         /* A scratch copy of path, filled lazily if we need it: */
381         struct strbuf path_copy = STRBUF_INIT;
382
383         int ret, save_errno;
384
385         /* Sanity check: */
386         assert(*path);
387
388 retry_fn:
389         ret = fn(path, cb);
390         save_errno = errno;
391         if (!ret)
392                 goto out;
393
394         if (errno == EISDIR && remove_directories_remaining-- > 0) {
395                 /*
396                  * A directory is in the way. Maybe it is empty; try
397                  * to remove it:
398                  */
399                 if (!path_copy.len)
400                         strbuf_addstr(&path_copy, path);
401
402                 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
403                         goto retry_fn;
404         } else if (errno == ENOENT && create_directories_remaining-- > 0) {
405                 /*
406                  * Maybe the containing directory didn't exist, or
407                  * maybe it was just deleted by a process that is
408                  * racing with us to clean up empty directories. Try
409                  * to create it:
410                  */
411                 enum scld_error scld_result;
412
413                 if (!path_copy.len)
414                         strbuf_addstr(&path_copy, path);
415
416                 do {
417                         scld_result = safe_create_leading_directories(path_copy.buf);
418                         if (scld_result == SCLD_OK)
419                                 goto retry_fn;
420                 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
421         }
422
423 out:
424         strbuf_release(&path_copy);
425         errno = save_errno;
426         return ret;
427 }
428
429 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
430 {
431         int i;
432         for (i = 0; i < the_hash_algo->rawsz; i++) {
433                 static char hex[] = "0123456789abcdef";
434                 unsigned int val = oid->hash[i];
435                 strbuf_addch(buf, hex[val >> 4]);
436                 strbuf_addch(buf, hex[val & 0xf]);
437                 if (!i)
438                         strbuf_addch(buf, '/');
439         }
440 }
441
442 static const char *odb_loose_path(struct object_directory *odb,
443                                   struct strbuf *buf,
444                                   const struct object_id *oid)
445 {
446         strbuf_reset(buf);
447         strbuf_addstr(buf, odb->path);
448         strbuf_addch(buf, '/');
449         fill_loose_path(buf, oid);
450         return buf->buf;
451 }
452
453 const char *loose_object_path(struct repository *r, struct strbuf *buf,
454                               const struct object_id *oid)
455 {
456         return odb_loose_path(r->objects->odb, buf, oid);
457 }
458
459 /*
460  * Return non-zero iff the path is usable as an alternate object database.
461  */
462 static int alt_odb_usable(struct raw_object_store *o,
463                           struct strbuf *path,
464                           const char *normalized_objdir)
465 {
466         struct object_directory *odb;
467
468         /* Detect cases where alternate disappeared */
469         if (!is_directory(path->buf)) {
470                 error(_("object directory %s does not exist; "
471                         "check .git/objects/info/alternates"),
472                       path->buf);
473                 return 0;
474         }
475
476         /*
477          * Prevent the common mistake of listing the same
478          * thing twice, or object directory itself.
479          */
480         for (odb = o->odb; odb; odb = odb->next) {
481                 if (!fspathcmp(path->buf, odb->path))
482                         return 0;
483         }
484         if (!fspathcmp(path->buf, normalized_objdir))
485                 return 0;
486
487         return 1;
488 }
489
490 /*
491  * Prepare alternate object database registry.
492  *
493  * The variable alt_odb_list points at the list of struct
494  * object_directory.  The elements on this list come from
495  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
496  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
497  * whose contents is similar to that environment variable but can be
498  * LF separated.  Its base points at a statically allocated buffer that
499  * contains "/the/directory/corresponding/to/.git/objects/...", while
500  * its name points just after the slash at the end of ".git/objects/"
501  * in the example above, and has enough space to hold 40-byte hex
502  * SHA1, an extra slash for the first level indirection, and the
503  * terminating NUL.
504  */
505 static void read_info_alternates(struct repository *r,
506                                  const char *relative_base,
507                                  int depth);
508 static int link_alt_odb_entry(struct repository *r, const char *entry,
509         const char *relative_base, int depth, const char *normalized_objdir)
510 {
511         struct object_directory *ent;
512         struct strbuf pathbuf = STRBUF_INIT;
513
514         if (!is_absolute_path(entry) && relative_base) {
515                 strbuf_realpath(&pathbuf, relative_base, 1);
516                 strbuf_addch(&pathbuf, '/');
517         }
518         strbuf_addstr(&pathbuf, entry);
519
520         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
521                 error(_("unable to normalize alternate object path: %s"),
522                       pathbuf.buf);
523                 strbuf_release(&pathbuf);
524                 return -1;
525         }
526
527         /*
528          * The trailing slash after the directory name is given by
529          * this function at the end. Remove duplicates.
530          */
531         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
532                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
533
534         if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
535                 strbuf_release(&pathbuf);
536                 return -1;
537         }
538
539         ent = xcalloc(1, sizeof(*ent));
540         ent->path = xstrdup(pathbuf.buf);
541
542         /* add the alternate entry */
543         *r->objects->odb_tail = ent;
544         r->objects->odb_tail = &(ent->next);
545         ent->next = NULL;
546
547         /* recursively add alternates */
548         read_info_alternates(r, pathbuf.buf, depth + 1);
549
550         strbuf_release(&pathbuf);
551         return 0;
552 }
553
554 static const char *parse_alt_odb_entry(const char *string,
555                                        int sep,
556                                        struct strbuf *out)
557 {
558         const char *end;
559
560         strbuf_reset(out);
561
562         if (*string == '#') {
563                 /* comment; consume up to next separator */
564                 end = strchrnul(string, sep);
565         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
566                 /*
567                  * quoted path; unquote_c_style has copied the
568                  * data for us and set "end". Broken quoting (e.g.,
569                  * an entry that doesn't end with a quote) falls
570                  * back to the unquoted case below.
571                  */
572         } else {
573                 /* normal, unquoted path */
574                 end = strchrnul(string, sep);
575                 strbuf_add(out, string, end - string);
576         }
577
578         if (*end)
579                 end++;
580         return end;
581 }
582
583 static void link_alt_odb_entries(struct repository *r, const char *alt,
584                                  int sep, const char *relative_base, int depth)
585 {
586         struct strbuf objdirbuf = STRBUF_INIT;
587         struct strbuf entry = STRBUF_INIT;
588
589         if (!alt || !*alt)
590                 return;
591
592         if (depth > 5) {
593                 error(_("%s: ignoring alternate object stores, nesting too deep"),
594                                 relative_base);
595                 return;
596         }
597
598         strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
599         if (strbuf_normalize_path(&objdirbuf) < 0)
600                 die(_("unable to normalize object directory: %s"),
601                     objdirbuf.buf);
602
603         while (*alt) {
604                 alt = parse_alt_odb_entry(alt, sep, &entry);
605                 if (!entry.len)
606                         continue;
607                 link_alt_odb_entry(r, entry.buf,
608                                    relative_base, depth, objdirbuf.buf);
609         }
610         strbuf_release(&entry);
611         strbuf_release(&objdirbuf);
612 }
613
614 static void read_info_alternates(struct repository *r,
615                                  const char *relative_base,
616                                  int depth)
617 {
618         char *path;
619         struct strbuf buf = STRBUF_INIT;
620
621         path = xstrfmt("%s/info/alternates", relative_base);
622         if (strbuf_read_file(&buf, path, 1024) < 0) {
623                 warn_on_fopen_errors(path);
624                 free(path);
625                 return;
626         }
627
628         link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
629         strbuf_release(&buf);
630         free(path);
631 }
632
633 void add_to_alternates_file(const char *reference)
634 {
635         struct lock_file lock = LOCK_INIT;
636         char *alts = git_pathdup("objects/info/alternates");
637         FILE *in, *out;
638         int found = 0;
639
640         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
641         out = fdopen_lock_file(&lock, "w");
642         if (!out)
643                 die_errno(_("unable to fdopen alternates lockfile"));
644
645         in = fopen(alts, "r");
646         if (in) {
647                 struct strbuf line = STRBUF_INIT;
648
649                 while (strbuf_getline(&line, in) != EOF) {
650                         if (!strcmp(reference, line.buf)) {
651                                 found = 1;
652                                 break;
653                         }
654                         fprintf_or_die(out, "%s\n", line.buf);
655                 }
656
657                 strbuf_release(&line);
658                 fclose(in);
659         }
660         else if (errno != ENOENT)
661                 die_errno(_("unable to read alternates file"));
662
663         if (found) {
664                 rollback_lock_file(&lock);
665         } else {
666                 fprintf_or_die(out, "%s\n", reference);
667                 if (commit_lock_file(&lock))
668                         die_errno(_("unable to move new alternates file into place"));
669                 if (the_repository->objects->loaded_alternates)
670                         link_alt_odb_entries(the_repository, reference,
671                                              '\n', NULL, 0);
672         }
673         free(alts);
674 }
675
676 void add_to_alternates_memory(const char *reference)
677 {
678         /*
679          * Make sure alternates are initialized, or else our entry may be
680          * overwritten when they are.
681          */
682         prepare_alt_odb(the_repository);
683
684         link_alt_odb_entries(the_repository, reference,
685                              '\n', NULL, 0);
686 }
687
688 /*
689  * Compute the exact path an alternate is at and returns it. In case of
690  * error NULL is returned and the human readable error is added to `err`
691  * `path` may be relative and should point to $GIT_DIR.
692  * `err` must not be null.
693  */
694 char *compute_alternate_path(const char *path, struct strbuf *err)
695 {
696         char *ref_git = NULL;
697         const char *repo, *ref_git_s;
698         int seen_error = 0;
699
700         ref_git_s = real_path_if_valid(path);
701         if (!ref_git_s) {
702                 seen_error = 1;
703                 strbuf_addf(err, _("path '%s' does not exist"), path);
704                 goto out;
705         } else
706                 /*
707                  * Beware: read_gitfile(), real_path() and mkpath()
708                  * return static buffer
709                  */
710                 ref_git = xstrdup(ref_git_s);
711
712         repo = read_gitfile(ref_git);
713         if (!repo)
714                 repo = read_gitfile(mkpath("%s/.git", ref_git));
715         if (repo) {
716                 free(ref_git);
717                 ref_git = xstrdup(repo);
718         }
719
720         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
721                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
722                 free(ref_git);
723                 ref_git = ref_git_git;
724         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
725                 struct strbuf sb = STRBUF_INIT;
726                 seen_error = 1;
727                 if (get_common_dir(&sb, ref_git)) {
728                         strbuf_addf(err,
729                                     _("reference repository '%s' as a linked "
730                                       "checkout is not supported yet."),
731                                     path);
732                         goto out;
733                 }
734
735                 strbuf_addf(err, _("reference repository '%s' is not a "
736                                         "local repository."), path);
737                 goto out;
738         }
739
740         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
741                 strbuf_addf(err, _("reference repository '%s' is shallow"),
742                             path);
743                 seen_error = 1;
744                 goto out;
745         }
746
747         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
748                 strbuf_addf(err,
749                             _("reference repository '%s' is grafted"),
750                             path);
751                 seen_error = 1;
752                 goto out;
753         }
754
755 out:
756         if (seen_error) {
757                 FREE_AND_NULL(ref_git);
758         }
759
760         return ref_git;
761 }
762
763 static void fill_alternate_refs_command(struct child_process *cmd,
764                                         const char *repo_path)
765 {
766         const char *value;
767
768         if (!git_config_get_value("core.alternateRefsCommand", &value)) {
769                 cmd->use_shell = 1;
770
771                 argv_array_push(&cmd->args, value);
772                 argv_array_push(&cmd->args, repo_path);
773         } else {
774                 cmd->git_cmd = 1;
775
776                 argv_array_pushf(&cmd->args, "--git-dir=%s", repo_path);
777                 argv_array_push(&cmd->args, "for-each-ref");
778                 argv_array_push(&cmd->args, "--format=%(objectname)");
779
780                 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
781                         argv_array_push(&cmd->args, "--");
782                         argv_array_split(&cmd->args, value);
783                 }
784         }
785
786         cmd->env = local_repo_env;
787         cmd->out = -1;
788 }
789
790 static void read_alternate_refs(const char *path,
791                                 alternate_ref_fn *cb,
792                                 void *data)
793 {
794         struct child_process cmd = CHILD_PROCESS_INIT;
795         struct strbuf line = STRBUF_INIT;
796         FILE *fh;
797
798         fill_alternate_refs_command(&cmd, path);
799
800         if (start_command(&cmd))
801                 return;
802
803         fh = xfdopen(cmd.out, "r");
804         while (strbuf_getline_lf(&line, fh) != EOF) {
805                 struct object_id oid;
806                 const char *p;
807
808                 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
809                         warning(_("invalid line while parsing alternate refs: %s"),
810                                 line.buf);
811                         break;
812                 }
813
814                 cb(&oid, data);
815         }
816
817         fclose(fh);
818         finish_command(&cmd);
819         strbuf_release(&line);
820 }
821
822 struct alternate_refs_data {
823         alternate_ref_fn *fn;
824         void *data;
825 };
826
827 static int refs_from_alternate_cb(struct object_directory *e,
828                                   void *data)
829 {
830         struct strbuf path = STRBUF_INIT;
831         size_t base_len;
832         struct alternate_refs_data *cb = data;
833
834         if (!strbuf_realpath(&path, e->path, 0))
835                 goto out;
836         if (!strbuf_strip_suffix(&path, "/objects"))
837                 goto out;
838         base_len = path.len;
839
840         /* Is this a git repository with refs? */
841         strbuf_addstr(&path, "/refs");
842         if (!is_directory(path.buf))
843                 goto out;
844         strbuf_setlen(&path, base_len);
845
846         read_alternate_refs(path.buf, cb->fn, cb->data);
847
848 out:
849         strbuf_release(&path);
850         return 0;
851 }
852
853 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
854 {
855         struct alternate_refs_data cb;
856         cb.fn = fn;
857         cb.data = data;
858         foreach_alt_odb(refs_from_alternate_cb, &cb);
859 }
860
861 int foreach_alt_odb(alt_odb_fn fn, void *cb)
862 {
863         struct object_directory *ent;
864         int r = 0;
865
866         prepare_alt_odb(the_repository);
867         for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
868                 r = fn(ent, cb);
869                 if (r)
870                         break;
871         }
872         return r;
873 }
874
875 void prepare_alt_odb(struct repository *r)
876 {
877         if (r->objects->loaded_alternates)
878                 return;
879
880         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
881
882         read_info_alternates(r, r->objects->odb->path, 0);
883         r->objects->loaded_alternates = 1;
884 }
885
886 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
887 static int freshen_file(const char *fn)
888 {
889         struct utimbuf t;
890         t.actime = t.modtime = time(NULL);
891         return !utime(fn, &t);
892 }
893
894 /*
895  * All of the check_and_freshen functions return 1 if the file exists and was
896  * freshened (if freshening was requested), 0 otherwise. If they return
897  * 0, you should not assume that it is safe to skip a write of the object (it
898  * either does not exist on disk, or has a stale mtime and may be subject to
899  * pruning).
900  */
901 int check_and_freshen_file(const char *fn, int freshen)
902 {
903         if (access(fn, F_OK))
904                 return 0;
905         if (freshen && !freshen_file(fn))
906                 return 0;
907         return 1;
908 }
909
910 static int check_and_freshen_odb(struct object_directory *odb,
911                                  const struct object_id *oid,
912                                  int freshen)
913 {
914         static struct strbuf path = STRBUF_INIT;
915         odb_loose_path(odb, &path, oid);
916         return check_and_freshen_file(path.buf, freshen);
917 }
918
919 static int check_and_freshen_local(const struct object_id *oid, int freshen)
920 {
921         return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
922 }
923
924 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
925 {
926         struct object_directory *odb;
927
928         prepare_alt_odb(the_repository);
929         for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
930                 if (check_and_freshen_odb(odb, oid, freshen))
931                         return 1;
932         }
933         return 0;
934 }
935
936 static int check_and_freshen(const struct object_id *oid, int freshen)
937 {
938         return check_and_freshen_local(oid, freshen) ||
939                check_and_freshen_nonlocal(oid, freshen);
940 }
941
942 int has_loose_object_nonlocal(const struct object_id *oid)
943 {
944         return check_and_freshen_nonlocal(oid, 0);
945 }
946
947 static int has_loose_object(const struct object_id *oid)
948 {
949         return check_and_freshen(oid, 0);
950 }
951
952 static void mmap_limit_check(size_t length)
953 {
954         static size_t limit = 0;
955         if (!limit) {
956                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
957                 if (!limit)
958                         limit = SIZE_MAX;
959         }
960         if (length > limit)
961                 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
962                     (uintmax_t)length, (uintmax_t)limit);
963 }
964
965 void *xmmap_gently(void *start, size_t length,
966                   int prot, int flags, int fd, off_t offset)
967 {
968         void *ret;
969
970         mmap_limit_check(length);
971         ret = mmap(start, length, prot, flags, fd, offset);
972         if (ret == MAP_FAILED && !length)
973                 ret = NULL;
974         return ret;
975 }
976
977 void *xmmap(void *start, size_t length,
978         int prot, int flags, int fd, off_t offset)
979 {
980         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
981         if (ret == MAP_FAILED)
982                 die_errno(_("mmap failed"));
983         return ret;
984 }
985
986 /*
987  * With an in-core object data in "map", rehash it to make sure the
988  * object name actually matches "oid" to detect object corruption.
989  * With "map" == NULL, try reading the object named with "oid" using
990  * the streaming interface and rehash it to do the same.
991  */
992 int check_object_signature(struct repository *r, const struct object_id *oid,
993                            void *map, unsigned long size, const char *type)
994 {
995         struct object_id real_oid;
996         enum object_type obj_type;
997         struct git_istream *st;
998         git_hash_ctx c;
999         char hdr[MAX_HEADER_LEN];
1000         int hdrlen;
1001
1002         if (map) {
1003                 hash_object_file(r->hash_algo, map, size, type, &real_oid);
1004                 return !oideq(oid, &real_oid) ? -1 : 0;
1005         }
1006
1007         st = open_istream(r, oid, &obj_type, &size, NULL);
1008         if (!st)
1009                 return -1;
1010
1011         /* Generate the header */
1012         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1013
1014         /* Sha1.. */
1015         r->hash_algo->init_fn(&c);
1016         r->hash_algo->update_fn(&c, hdr, hdrlen);
1017         for (;;) {
1018                 char buf[1024 * 16];
1019                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1020
1021                 if (readlen < 0) {
1022                         close_istream(st);
1023                         return -1;
1024                 }
1025                 if (!readlen)
1026                         break;
1027                 r->hash_algo->update_fn(&c, buf, readlen);
1028         }
1029         r->hash_algo->final_fn(real_oid.hash, &c);
1030         close_istream(st);
1031         return !oideq(oid, &real_oid) ? -1 : 0;
1032 }
1033
1034 int git_open_cloexec(const char *name, int flags)
1035 {
1036         int fd;
1037         static int o_cloexec = O_CLOEXEC;
1038
1039         fd = open(name, flags | o_cloexec);
1040         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1041                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1042                 o_cloexec &= ~O_CLOEXEC;
1043                 fd = open(name, flags | o_cloexec);
1044         }
1045
1046 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1047         {
1048                 static int fd_cloexec = FD_CLOEXEC;
1049
1050                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1051                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
1052                         int flags = fcntl(fd, F_GETFD);
1053                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1054                                 fd_cloexec = 0;
1055                 }
1056         }
1057 #endif
1058         return fd;
1059 }
1060
1061 /*
1062  * Find "oid" as a loose object in the local repository or in an alternate.
1063  * Returns 0 on success, negative on failure.
1064  *
1065  * The "path" out-parameter will give the path of the object we found (if any).
1066  * Note that it may point to static storage and is only valid until another
1067  * call to stat_loose_object().
1068  */
1069 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1070                              struct stat *st, const char **path)
1071 {
1072         struct object_directory *odb;
1073         static struct strbuf buf = STRBUF_INIT;
1074
1075         prepare_alt_odb(r);
1076         for (odb = r->objects->odb; odb; odb = odb->next) {
1077                 *path = odb_loose_path(odb, &buf, oid);
1078                 if (!lstat(*path, st))
1079                         return 0;
1080         }
1081
1082         return -1;
1083 }
1084
1085 /*
1086  * Like stat_loose_object(), but actually open the object and return the
1087  * descriptor. See the caveats on the "path" parameter above.
1088  */
1089 static int open_loose_object(struct repository *r,
1090                              const struct object_id *oid, const char **path)
1091 {
1092         int fd;
1093         struct object_directory *odb;
1094         int most_interesting_errno = ENOENT;
1095         static struct strbuf buf = STRBUF_INIT;
1096
1097         prepare_alt_odb(r);
1098         for (odb = r->objects->odb; odb; odb = odb->next) {
1099                 *path = odb_loose_path(odb, &buf, oid);
1100                 fd = git_open(*path);
1101                 if (fd >= 0)
1102                         return fd;
1103
1104                 if (most_interesting_errno == ENOENT)
1105                         most_interesting_errno = errno;
1106         }
1107         errno = most_interesting_errno;
1108         return -1;
1109 }
1110
1111 static int quick_has_loose(struct repository *r,
1112                            const struct object_id *oid)
1113 {
1114         struct object_directory *odb;
1115
1116         prepare_alt_odb(r);
1117         for (odb = r->objects->odb; odb; odb = odb->next) {
1118                 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1119                         return 1;
1120         }
1121         return 0;
1122 }
1123
1124 /*
1125  * Map the loose object at "path" if it is not NULL, or the path found by
1126  * searching for a loose object named "oid".
1127  */
1128 static void *map_loose_object_1(struct repository *r, const char *path,
1129                              const struct object_id *oid, unsigned long *size)
1130 {
1131         void *map;
1132         int fd;
1133
1134         if (path)
1135                 fd = git_open(path);
1136         else
1137                 fd = open_loose_object(r, oid, &path);
1138         map = NULL;
1139         if (fd >= 0) {
1140                 struct stat st;
1141
1142                 if (!fstat(fd, &st)) {
1143                         *size = xsize_t(st.st_size);
1144                         if (!*size) {
1145                                 /* mmap() is forbidden on empty files */
1146                                 error(_("object file %s is empty"), path);
1147                                 close(fd);
1148                                 return NULL;
1149                         }
1150                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1151                 }
1152                 close(fd);
1153         }
1154         return map;
1155 }
1156
1157 void *map_loose_object(struct repository *r,
1158                        const struct object_id *oid,
1159                        unsigned long *size)
1160 {
1161         return map_loose_object_1(r, NULL, oid, size);
1162 }
1163
1164 static int unpack_loose_short_header(git_zstream *stream,
1165                                      unsigned char *map, unsigned long mapsize,
1166                                      void *buffer, unsigned long bufsiz)
1167 {
1168         int ret;
1169
1170         /* Get the data stream */
1171         memset(stream, 0, sizeof(*stream));
1172         stream->next_in = map;
1173         stream->avail_in = mapsize;
1174         stream->next_out = buffer;
1175         stream->avail_out = bufsiz;
1176
1177         git_inflate_init(stream);
1178         obj_read_unlock();
1179         ret = git_inflate(stream, 0);
1180         obj_read_lock();
1181
1182         return ret;
1183 }
1184
1185 int unpack_loose_header(git_zstream *stream,
1186                         unsigned char *map, unsigned long mapsize,
1187                         void *buffer, unsigned long bufsiz)
1188 {
1189         int status = unpack_loose_short_header(stream, map, mapsize,
1190                                                buffer, bufsiz);
1191
1192         if (status < Z_OK)
1193                 return status;
1194
1195         /* Make sure we have the terminating NUL */
1196         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1197                 return -1;
1198         return 0;
1199 }
1200
1201 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1202                                          unsigned long mapsize, void *buffer,
1203                                          unsigned long bufsiz, struct strbuf *header)
1204 {
1205         int status;
1206
1207         status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1208         if (status < Z_OK)
1209                 return -1;
1210
1211         /*
1212          * Check if entire header is unpacked in the first iteration.
1213          */
1214         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1215                 return 0;
1216
1217         /*
1218          * buffer[0..bufsiz] was not large enough.  Copy the partial
1219          * result out to header, and then append the result of further
1220          * reading the stream.
1221          */
1222         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1223         stream->next_out = buffer;
1224         stream->avail_out = bufsiz;
1225
1226         do {
1227                 obj_read_unlock();
1228                 status = git_inflate(stream, 0);
1229                 obj_read_lock();
1230                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1231                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1232                         return 0;
1233                 stream->next_out = buffer;
1234                 stream->avail_out = bufsiz;
1235         } while (status != Z_STREAM_END);
1236         return -1;
1237 }
1238
1239 static void *unpack_loose_rest(git_zstream *stream,
1240                                void *buffer, unsigned long size,
1241                                const struct object_id *oid)
1242 {
1243         int bytes = strlen(buffer) + 1;
1244         unsigned char *buf = xmallocz(size);
1245         unsigned long n;
1246         int status = Z_OK;
1247
1248         n = stream->total_out - bytes;
1249         if (n > size)
1250                 n = size;
1251         memcpy(buf, (char *) buffer + bytes, n);
1252         bytes = n;
1253         if (bytes <= size) {
1254                 /*
1255                  * The above condition must be (bytes <= size), not
1256                  * (bytes < size).  In other words, even though we
1257                  * expect no more output and set avail_out to zero,
1258                  * the input zlib stream may have bytes that express
1259                  * "this concludes the stream", and we *do* want to
1260                  * eat that input.
1261                  *
1262                  * Otherwise we would not be able to test that we
1263                  * consumed all the input to reach the expected size;
1264                  * we also want to check that zlib tells us that all
1265                  * went well with status == Z_STREAM_END at the end.
1266                  */
1267                 stream->next_out = buf + bytes;
1268                 stream->avail_out = size - bytes;
1269                 while (status == Z_OK) {
1270                         obj_read_unlock();
1271                         status = git_inflate(stream, Z_FINISH);
1272                         obj_read_lock();
1273                 }
1274         }
1275         if (status == Z_STREAM_END && !stream->avail_in) {
1276                 git_inflate_end(stream);
1277                 return buf;
1278         }
1279
1280         if (status < 0)
1281                 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1282         else if (stream->avail_in)
1283                 error(_("garbage at end of loose object '%s'"),
1284                       oid_to_hex(oid));
1285         free(buf);
1286         return NULL;
1287 }
1288
1289 /*
1290  * We used to just use "sscanf()", but that's actually way
1291  * too permissive for what we want to check. So do an anal
1292  * object header parse by hand.
1293  */
1294 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1295                                        unsigned int flags)
1296 {
1297         const char *type_buf = hdr;
1298         unsigned long size;
1299         int type, type_len = 0;
1300
1301         /*
1302          * The type can be of any size but is followed by
1303          * a space.
1304          */
1305         for (;;) {
1306                 char c = *hdr++;
1307                 if (!c)
1308                         return -1;
1309                 if (c == ' ')
1310                         break;
1311                 type_len++;
1312         }
1313
1314         type = type_from_string_gently(type_buf, type_len, 1);
1315         if (oi->type_name)
1316                 strbuf_add(oi->type_name, type_buf, type_len);
1317         /*
1318          * Set type to 0 if its an unknown object and
1319          * we're obtaining the type using '--allow-unknown-type'
1320          * option.
1321          */
1322         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1323                 type = 0;
1324         else if (type < 0)
1325                 die(_("invalid object type"));
1326         if (oi->typep)
1327                 *oi->typep = type;
1328
1329         /*
1330          * The length must follow immediately, and be in canonical
1331          * decimal format (ie "010" is not valid).
1332          */
1333         size = *hdr++ - '0';
1334         if (size > 9)
1335                 return -1;
1336         if (size) {
1337                 for (;;) {
1338                         unsigned long c = *hdr - '0';
1339                         if (c > 9)
1340                                 break;
1341                         hdr++;
1342                         size = size * 10 + c;
1343                 }
1344         }
1345
1346         if (oi->sizep)
1347                 *oi->sizep = size;
1348
1349         /*
1350          * The length must be followed by a zero byte
1351          */
1352         return *hdr ? -1 : type;
1353 }
1354
1355 int parse_loose_header(const char *hdr, unsigned long *sizep)
1356 {
1357         struct object_info oi = OBJECT_INFO_INIT;
1358
1359         oi.sizep = sizep;
1360         return parse_loose_header_extended(hdr, &oi, 0);
1361 }
1362
1363 static int loose_object_info(struct repository *r,
1364                              const struct object_id *oid,
1365                              struct object_info *oi, int flags)
1366 {
1367         int status = 0;
1368         unsigned long mapsize;
1369         void *map;
1370         git_zstream stream;
1371         char hdr[MAX_HEADER_LEN];
1372         struct strbuf hdrbuf = STRBUF_INIT;
1373         unsigned long size_scratch;
1374
1375         if (oi->delta_base_sha1)
1376                 hashclr(oi->delta_base_sha1);
1377
1378         /*
1379          * If we don't care about type or size, then we don't
1380          * need to look inside the object at all. Note that we
1381          * do not optimize out the stat call, even if the
1382          * caller doesn't care about the disk-size, since our
1383          * return value implicitly indicates whether the
1384          * object even exists.
1385          */
1386         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1387                 const char *path;
1388                 struct stat st;
1389                 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1390                         return quick_has_loose(r, oid) ? 0 : -1;
1391                 if (stat_loose_object(r, oid, &st, &path) < 0)
1392                         return -1;
1393                 if (oi->disk_sizep)
1394                         *oi->disk_sizep = st.st_size;
1395                 return 0;
1396         }
1397
1398         map = map_loose_object(r, oid, &mapsize);
1399         if (!map)
1400                 return -1;
1401
1402         if (!oi->sizep)
1403                 oi->sizep = &size_scratch;
1404
1405         if (oi->disk_sizep)
1406                 *oi->disk_sizep = mapsize;
1407         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1408                 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1409                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1410                                        oid_to_hex(oid));
1411         } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1412                 status = error(_("unable to unpack %s header"),
1413                                oid_to_hex(oid));
1414         if (status < 0)
1415                 ; /* Do nothing */
1416         else if (hdrbuf.len) {
1417                 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1418                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1419                                        oid_to_hex(oid));
1420         } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1421                 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1422
1423         if (status >= 0 && oi->contentp) {
1424                 *oi->contentp = unpack_loose_rest(&stream, hdr,
1425                                                   *oi->sizep, oid);
1426                 if (!*oi->contentp) {
1427                         git_inflate_end(&stream);
1428                         status = -1;
1429                 }
1430         } else
1431                 git_inflate_end(&stream);
1432
1433         munmap(map, mapsize);
1434         if (status && oi->typep)
1435                 *oi->typep = status;
1436         if (oi->sizep == &size_scratch)
1437                 oi->sizep = NULL;
1438         strbuf_release(&hdrbuf);
1439         oi->whence = OI_LOOSE;
1440         return (status < 0) ? status : 0;
1441 }
1442
1443 int obj_read_use_lock = 0;
1444 pthread_mutex_t obj_read_mutex;
1445
1446 void enable_obj_read_lock(void)
1447 {
1448         if (obj_read_use_lock)
1449                 return;
1450
1451         obj_read_use_lock = 1;
1452         init_recursive_mutex(&obj_read_mutex);
1453 }
1454
1455 void disable_obj_read_lock(void)
1456 {
1457         if (!obj_read_use_lock)
1458                 return;
1459
1460         obj_read_use_lock = 0;
1461         pthread_mutex_destroy(&obj_read_mutex);
1462 }
1463
1464 int fetch_if_missing = 1;
1465
1466 static int do_oid_object_info_extended(struct repository *r,
1467                                        const struct object_id *oid,
1468                                        struct object_info *oi, unsigned flags)
1469 {
1470         static struct object_info blank_oi = OBJECT_INFO_INIT;
1471         struct cached_object *co;
1472         struct pack_entry e;
1473         int rtype;
1474         const struct object_id *real = oid;
1475         int already_retried = 0;
1476
1477
1478         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1479                 real = lookup_replace_object(r, oid);
1480
1481         if (is_null_oid(real))
1482                 return -1;
1483
1484         if (!oi)
1485                 oi = &blank_oi;
1486
1487         co = find_cached_object(real);
1488         if (co) {
1489                 if (oi->typep)
1490                         *(oi->typep) = co->type;
1491                 if (oi->sizep)
1492                         *(oi->sizep) = co->size;
1493                 if (oi->disk_sizep)
1494                         *(oi->disk_sizep) = 0;
1495                 if (oi->delta_base_sha1)
1496                         hashclr(oi->delta_base_sha1);
1497                 if (oi->type_name)
1498                         strbuf_addstr(oi->type_name, type_name(co->type));
1499                 if (oi->contentp)
1500                         *oi->contentp = xmemdupz(co->buf, co->size);
1501                 oi->whence = OI_CACHED;
1502                 return 0;
1503         }
1504
1505         while (1) {
1506                 if (find_pack_entry(r, real, &e))
1507                         break;
1508
1509                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1510                         return -1;
1511
1512                 /* Most likely it's a loose object. */
1513                 if (!loose_object_info(r, real, oi, flags))
1514                         return 0;
1515
1516                 /* Not a loose object; someone else may have just packed it. */
1517                 if (!(flags & OBJECT_INFO_QUICK)) {
1518                         reprepare_packed_git(r);
1519                         if (find_pack_entry(r, real, &e))
1520                                 break;
1521                 }
1522
1523                 /* Check if it is a missing object */
1524                 if (fetch_if_missing && has_promisor_remote() &&
1525                     !already_retried && r == the_repository &&
1526                     !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1527                         /*
1528                          * TODO Investigate checking promisor_remote_get_direct()
1529                          * TODO return value and stopping on error here.
1530                          * TODO Pass a repository struct through
1531                          * promisor_remote_get_direct(), such that arbitrary
1532                          * repositories work.
1533                          */
1534                         promisor_remote_get_direct(r, real, 1);
1535                         already_retried = 1;
1536                         continue;
1537                 }
1538
1539                 return -1;
1540         }
1541
1542         if (oi == &blank_oi)
1543                 /*
1544                  * We know that the caller doesn't actually need the
1545                  * information below, so return early.
1546                  */
1547                 return 0;
1548         rtype = packed_object_info(r, e.p, e.offset, oi);
1549         if (rtype < 0) {
1550                 mark_bad_packed_object(e.p, real->hash);
1551                 return do_oid_object_info_extended(r, real, oi, 0);
1552         } else if (oi->whence == OI_PACKED) {
1553                 oi->u.packed.offset = e.offset;
1554                 oi->u.packed.pack = e.p;
1555                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1556                                          rtype == OBJ_OFS_DELTA);
1557         }
1558
1559         return 0;
1560 }
1561
1562 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1563                              struct object_info *oi, unsigned flags)
1564 {
1565         int ret;
1566         obj_read_lock();
1567         ret = do_oid_object_info_extended(r, oid, oi, flags);
1568         obj_read_unlock();
1569         return ret;
1570 }
1571
1572
1573 /* returns enum object_type or negative */
1574 int oid_object_info(struct repository *r,
1575                     const struct object_id *oid,
1576                     unsigned long *sizep)
1577 {
1578         enum object_type type;
1579         struct object_info oi = OBJECT_INFO_INIT;
1580
1581         oi.typep = &type;
1582         oi.sizep = sizep;
1583         if (oid_object_info_extended(r, oid, &oi,
1584                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1585                 return -1;
1586         return type;
1587 }
1588
1589 static void *read_object(struct repository *r,
1590                          const struct object_id *oid, enum object_type *type,
1591                          unsigned long *size)
1592 {
1593         struct object_info oi = OBJECT_INFO_INIT;
1594         void *content;
1595         oi.typep = type;
1596         oi.sizep = size;
1597         oi.contentp = &content;
1598
1599         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1600                 return NULL;
1601         return content;
1602 }
1603
1604 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1605                         struct object_id *oid)
1606 {
1607         struct cached_object *co;
1608
1609         hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1610         if (has_object_file(oid) || find_cached_object(oid))
1611                 return 0;
1612         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1613         co = &cached_objects[cached_object_nr++];
1614         co->size = len;
1615         co->type = type;
1616         co->buf = xmalloc(len);
1617         memcpy(co->buf, buf, len);
1618         oidcpy(&co->oid, oid);
1619         return 0;
1620 }
1621
1622 /*
1623  * This function dies on corrupt objects; the callers who want to
1624  * deal with them should arrange to call read_object() and give error
1625  * messages themselves.
1626  */
1627 void *read_object_file_extended(struct repository *r,
1628                                 const struct object_id *oid,
1629                                 enum object_type *type,
1630                                 unsigned long *size,
1631                                 int lookup_replace)
1632 {
1633         void *data;
1634         const struct packed_git *p;
1635         const char *path;
1636         struct stat st;
1637         const struct object_id *repl = lookup_replace ?
1638                 lookup_replace_object(r, oid) : oid;
1639
1640         errno = 0;
1641         data = read_object(r, repl, type, size);
1642         if (data)
1643                 return data;
1644
1645         obj_read_lock();
1646         if (errno && errno != ENOENT)
1647                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1648
1649         /* die if we replaced an object with one that does not exist */
1650         if (repl != oid)
1651                 die(_("replacement %s not found for %s"),
1652                     oid_to_hex(repl), oid_to_hex(oid));
1653
1654         if (!stat_loose_object(r, repl, &st, &path))
1655                 die(_("loose object %s (stored in %s) is corrupt"),
1656                     oid_to_hex(repl), path);
1657
1658         if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1659                 die(_("packed object %s (stored in %s) is corrupt"),
1660                     oid_to_hex(repl), p->pack_name);
1661         obj_read_unlock();
1662
1663         return NULL;
1664 }
1665
1666 void *read_object_with_reference(struct repository *r,
1667                                  const struct object_id *oid,
1668                                  const char *required_type_name,
1669                                  unsigned long *size,
1670                                  struct object_id *actual_oid_return)
1671 {
1672         enum object_type type, required_type;
1673         void *buffer;
1674         unsigned long isize;
1675         struct object_id actual_oid;
1676
1677         required_type = type_from_string(required_type_name);
1678         oidcpy(&actual_oid, oid);
1679         while (1) {
1680                 int ref_length = -1;
1681                 const char *ref_type = NULL;
1682
1683                 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1684                 if (!buffer)
1685                         return NULL;
1686                 if (type == required_type) {
1687                         *size = isize;
1688                         if (actual_oid_return)
1689                                 oidcpy(actual_oid_return, &actual_oid);
1690                         return buffer;
1691                 }
1692                 /* Handle references */
1693                 else if (type == OBJ_COMMIT)
1694                         ref_type = "tree ";
1695                 else if (type == OBJ_TAG)
1696                         ref_type = "object ";
1697                 else {
1698                         free(buffer);
1699                         return NULL;
1700                 }
1701                 ref_length = strlen(ref_type);
1702
1703                 if (ref_length + the_hash_algo->hexsz > isize ||
1704                     memcmp(buffer, ref_type, ref_length) ||
1705                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1706                         free(buffer);
1707                         return NULL;
1708                 }
1709                 free(buffer);
1710                 /* Now we have the ID of the referred-to object in
1711                  * actual_oid.  Check again. */
1712         }
1713 }
1714
1715 static void write_object_file_prepare(const struct git_hash_algo *algo,
1716                                       const void *buf, unsigned long len,
1717                                       const char *type, struct object_id *oid,
1718                                       char *hdr, int *hdrlen)
1719 {
1720         git_hash_ctx c;
1721
1722         /* Generate the header */
1723         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1724
1725         /* Sha1.. */
1726         algo->init_fn(&c);
1727         algo->update_fn(&c, hdr, *hdrlen);
1728         algo->update_fn(&c, buf, len);
1729         algo->final_fn(oid->hash, &c);
1730 }
1731
1732 /*
1733  * Move the just written object into its final resting place.
1734  */
1735 int finalize_object_file(const char *tmpfile, const char *filename)
1736 {
1737         int ret = 0;
1738
1739         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1740                 goto try_rename;
1741         else if (link(tmpfile, filename))
1742                 ret = errno;
1743
1744         /*
1745          * Coda hack - coda doesn't like cross-directory links,
1746          * so we fall back to a rename, which will mean that it
1747          * won't be able to check collisions, but that's not a
1748          * big deal.
1749          *
1750          * The same holds for FAT formatted media.
1751          *
1752          * When this succeeds, we just return.  We have nothing
1753          * left to unlink.
1754          */
1755         if (ret && ret != EEXIST) {
1756         try_rename:
1757                 if (!rename(tmpfile, filename))
1758                         goto out;
1759                 ret = errno;
1760         }
1761         unlink_or_warn(tmpfile);
1762         if (ret) {
1763                 if (ret != EEXIST) {
1764                         return error_errno(_("unable to write file %s"), filename);
1765                 }
1766                 /* FIXME!!! Collision check here ? */
1767         }
1768
1769 out:
1770         if (adjust_shared_perm(filename))
1771                 return error(_("unable to set permission to '%s'"), filename);
1772         return 0;
1773 }
1774
1775 static int write_buffer(int fd, const void *buf, size_t len)
1776 {
1777         if (write_in_full(fd, buf, len) < 0)
1778                 return error_errno(_("file write error"));
1779         return 0;
1780 }
1781
1782 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1783                      unsigned long len, const char *type,
1784                      struct object_id *oid)
1785 {
1786         char hdr[MAX_HEADER_LEN];
1787         int hdrlen = sizeof(hdr);
1788         write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1789         return 0;
1790 }
1791
1792 /* Finalize a file on disk, and close it. */
1793 static void close_loose_object(int fd)
1794 {
1795         if (fsync_object_files)
1796                 fsync_or_die(fd, "loose object file");
1797         if (close(fd) != 0)
1798                 die_errno(_("error when closing loose object file"));
1799 }
1800
1801 /* Size of directory component, including the ending '/' */
1802 static inline int directory_size(const char *filename)
1803 {
1804         const char *s = strrchr(filename, '/');
1805         if (!s)
1806                 return 0;
1807         return s - filename + 1;
1808 }
1809
1810 /*
1811  * This creates a temporary file in the same directory as the final
1812  * 'filename'
1813  *
1814  * We want to avoid cross-directory filename renames, because those
1815  * can have problems on various filesystems (FAT, NFS, Coda).
1816  */
1817 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1818 {
1819         int fd, dirlen = directory_size(filename);
1820
1821         strbuf_reset(tmp);
1822         strbuf_add(tmp, filename, dirlen);
1823         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1824         fd = git_mkstemp_mode(tmp->buf, 0444);
1825         if (fd < 0 && dirlen && errno == ENOENT) {
1826                 /*
1827                  * Make sure the directory exists; note that the contents
1828                  * of the buffer are undefined after mkstemp returns an
1829                  * error, so we have to rewrite the whole buffer from
1830                  * scratch.
1831                  */
1832                 strbuf_reset(tmp);
1833                 strbuf_add(tmp, filename, dirlen - 1);
1834                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1835                         return -1;
1836                 if (adjust_shared_perm(tmp->buf))
1837                         return -1;
1838
1839                 /* Try again */
1840                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1841                 fd = git_mkstemp_mode(tmp->buf, 0444);
1842         }
1843         return fd;
1844 }
1845
1846 static int write_loose_object(const struct object_id *oid, char *hdr,
1847                               int hdrlen, const void *buf, unsigned long len,
1848                               time_t mtime)
1849 {
1850         int fd, ret;
1851         unsigned char compressed[4096];
1852         git_zstream stream;
1853         git_hash_ctx c;
1854         struct object_id parano_oid;
1855         static struct strbuf tmp_file = STRBUF_INIT;
1856         static struct strbuf filename = STRBUF_INIT;
1857
1858         loose_object_path(the_repository, &filename, oid);
1859
1860         fd = create_tmpfile(&tmp_file, filename.buf);
1861         if (fd < 0) {
1862                 if (errno == EACCES)
1863                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1864                 else
1865                         return error_errno(_("unable to create temporary file"));
1866         }
1867
1868         /* Set it up */
1869         git_deflate_init(&stream, zlib_compression_level);
1870         stream.next_out = compressed;
1871         stream.avail_out = sizeof(compressed);
1872         the_hash_algo->init_fn(&c);
1873
1874         /* First header.. */
1875         stream.next_in = (unsigned char *)hdr;
1876         stream.avail_in = hdrlen;
1877         while (git_deflate(&stream, 0) == Z_OK)
1878                 ; /* nothing */
1879         the_hash_algo->update_fn(&c, hdr, hdrlen);
1880
1881         /* Then the data itself.. */
1882         stream.next_in = (void *)buf;
1883         stream.avail_in = len;
1884         do {
1885                 unsigned char *in0 = stream.next_in;
1886                 ret = git_deflate(&stream, Z_FINISH);
1887                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1888                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1889                         die(_("unable to write loose object file"));
1890                 stream.next_out = compressed;
1891                 stream.avail_out = sizeof(compressed);
1892         } while (ret == Z_OK);
1893
1894         if (ret != Z_STREAM_END)
1895                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1896                     ret);
1897         ret = git_deflate_end_gently(&stream);
1898         if (ret != Z_OK)
1899                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1900                     ret);
1901         the_hash_algo->final_fn(parano_oid.hash, &c);
1902         if (!oideq(oid, &parano_oid))
1903                 die(_("confused by unstable object source data for %s"),
1904                     oid_to_hex(oid));
1905
1906         close_loose_object(fd);
1907
1908         if (mtime) {
1909                 struct utimbuf utb;
1910                 utb.actime = mtime;
1911                 utb.modtime = mtime;
1912                 if (utime(tmp_file.buf, &utb) < 0)
1913                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1914         }
1915
1916         return finalize_object_file(tmp_file.buf, filename.buf);
1917 }
1918
1919 static int freshen_loose_object(const struct object_id *oid)
1920 {
1921         return check_and_freshen(oid, 1);
1922 }
1923
1924 static int freshen_packed_object(const struct object_id *oid)
1925 {
1926         struct pack_entry e;
1927         if (!find_pack_entry(the_repository, oid, &e))
1928                 return 0;
1929         if (e.p->freshened)
1930                 return 1;
1931         if (!freshen_file(e.p->pack_name))
1932                 return 0;
1933         e.p->freshened = 1;
1934         return 1;
1935 }
1936
1937 int write_object_file(const void *buf, unsigned long len, const char *type,
1938                       struct object_id *oid)
1939 {
1940         char hdr[MAX_HEADER_LEN];
1941         int hdrlen = sizeof(hdr);
1942
1943         /* Normally if we have it in the pack then we do not bother writing
1944          * it out into .git/objects/??/?{38} file.
1945          */
1946         write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1947                                   &hdrlen);
1948         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1949                 return 0;
1950         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1951 }
1952
1953 int hash_object_file_literally(const void *buf, unsigned long len,
1954                                const char *type, struct object_id *oid,
1955                                unsigned flags)
1956 {
1957         char *header;
1958         int hdrlen, status = 0;
1959
1960         /* type string, SP, %lu of the length plus NUL must fit this */
1961         hdrlen = strlen(type) + MAX_HEADER_LEN;
1962         header = xmalloc(hdrlen);
1963         write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1964                                   &hdrlen);
1965
1966         if (!(flags & HASH_WRITE_OBJECT))
1967                 goto cleanup;
1968         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1969                 goto cleanup;
1970         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1971
1972 cleanup:
1973         free(header);
1974         return status;
1975 }
1976
1977 int force_object_loose(const struct object_id *oid, time_t mtime)
1978 {
1979         void *buf;
1980         unsigned long len;
1981         enum object_type type;
1982         char hdr[MAX_HEADER_LEN];
1983         int hdrlen;
1984         int ret;
1985
1986         if (has_loose_object(oid))
1987                 return 0;
1988         buf = read_object(the_repository, oid, &type, &len);
1989         if (!buf)
1990                 return error(_("cannot read object for %s"), oid_to_hex(oid));
1991         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1992         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1993         free(buf);
1994
1995         return ret;
1996 }
1997
1998 int repo_has_object_file_with_flags(struct repository *r,
1999                                     const struct object_id *oid, int flags)
2000 {
2001         if (!startup_info->have_repository)
2002                 return 0;
2003         return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2004 }
2005
2006 int repo_has_object_file(struct repository *r,
2007                          const struct object_id *oid)
2008 {
2009         return repo_has_object_file_with_flags(r, oid, 0);
2010 }
2011
2012 static void check_tree(const void *buf, size_t size)
2013 {
2014         struct tree_desc desc;
2015         struct name_entry entry;
2016
2017         init_tree_desc(&desc, buf, size);
2018         while (tree_entry(&desc, &entry))
2019                 /* do nothing
2020                  * tree_entry() will die() on malformed entries */
2021                 ;
2022 }
2023
2024 static void check_commit(const void *buf, size_t size)
2025 {
2026         struct commit c;
2027         memset(&c, 0, sizeof(c));
2028         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2029                 die(_("corrupt commit"));
2030 }
2031
2032 static void check_tag(const void *buf, size_t size)
2033 {
2034         struct tag t;
2035         memset(&t, 0, sizeof(t));
2036         if (parse_tag_buffer(the_repository, &t, buf, size))
2037                 die(_("corrupt tag"));
2038 }
2039
2040 static int index_mem(struct index_state *istate,
2041                      struct object_id *oid, void *buf, size_t size,
2042                      enum object_type type,
2043                      const char *path, unsigned flags)
2044 {
2045         int ret, re_allocated = 0;
2046         int write_object = flags & HASH_WRITE_OBJECT;
2047
2048         if (!type)
2049                 type = OBJ_BLOB;
2050
2051         /*
2052          * Convert blobs to git internal format
2053          */
2054         if ((type == OBJ_BLOB) && path) {
2055                 struct strbuf nbuf = STRBUF_INIT;
2056                 if (convert_to_git(istate, path, buf, size, &nbuf,
2057                                    get_conv_flags(flags))) {
2058                         buf = strbuf_detach(&nbuf, &size);
2059                         re_allocated = 1;
2060                 }
2061         }
2062         if (flags & HASH_FORMAT_CHECK) {
2063                 if (type == OBJ_TREE)
2064                         check_tree(buf, size);
2065                 if (type == OBJ_COMMIT)
2066                         check_commit(buf, size);
2067                 if (type == OBJ_TAG)
2068                         check_tag(buf, size);
2069         }
2070
2071         if (write_object)
2072                 ret = write_object_file(buf, size, type_name(type), oid);
2073         else
2074                 ret = hash_object_file(the_hash_algo, buf, size,
2075                                        type_name(type), oid);
2076         if (re_allocated)
2077                 free(buf);
2078         return ret;
2079 }
2080
2081 static int index_stream_convert_blob(struct index_state *istate,
2082                                      struct object_id *oid,
2083                                      int fd,
2084                                      const char *path,
2085                                      unsigned flags)
2086 {
2087         int ret;
2088         const int write_object = flags & HASH_WRITE_OBJECT;
2089         struct strbuf sbuf = STRBUF_INIT;
2090
2091         assert(path);
2092         assert(would_convert_to_git_filter_fd(istate, path));
2093
2094         convert_to_git_filter_fd(istate, path, fd, &sbuf,
2095                                  get_conv_flags(flags));
2096
2097         if (write_object)
2098                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2099                                         oid);
2100         else
2101                 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2102                                        type_name(OBJ_BLOB), oid);
2103         strbuf_release(&sbuf);
2104         return ret;
2105 }
2106
2107 static int index_pipe(struct index_state *istate, struct object_id *oid,
2108                       int fd, enum object_type type,
2109                       const char *path, unsigned flags)
2110 {
2111         struct strbuf sbuf = STRBUF_INIT;
2112         int ret;
2113
2114         if (strbuf_read(&sbuf, fd, 4096) >= 0)
2115                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2116         else
2117                 ret = -1;
2118         strbuf_release(&sbuf);
2119         return ret;
2120 }
2121
2122 #define SMALL_FILE_SIZE (32*1024)
2123
2124 static int index_core(struct index_state *istate,
2125                       struct object_id *oid, int fd, size_t size,
2126                       enum object_type type, const char *path,
2127                       unsigned flags)
2128 {
2129         int ret;
2130
2131         if (!size) {
2132                 ret = index_mem(istate, oid, "", size, type, path, flags);
2133         } else if (size <= SMALL_FILE_SIZE) {
2134                 char *buf = xmalloc(size);
2135                 ssize_t read_result = read_in_full(fd, buf, size);
2136                 if (read_result < 0)
2137                         ret = error_errno(_("read error while indexing %s"),
2138                                           path ? path : "<unknown>");
2139                 else if (read_result != size)
2140                         ret = error(_("short read while indexing %s"),
2141                                     path ? path : "<unknown>");
2142                 else
2143                         ret = index_mem(istate, oid, buf, size, type, path, flags);
2144                 free(buf);
2145         } else {
2146                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2147                 ret = index_mem(istate, oid, buf, size, type, path, flags);
2148                 munmap(buf, size);
2149         }
2150         return ret;
2151 }
2152
2153 /*
2154  * This creates one packfile per large blob unless bulk-checkin
2155  * machinery is "plugged".
2156  *
2157  * This also bypasses the usual "convert-to-git" dance, and that is on
2158  * purpose. We could write a streaming version of the converting
2159  * functions and insert that before feeding the data to fast-import
2160  * (or equivalent in-core API described above). However, that is
2161  * somewhat complicated, as we do not know the size of the filter
2162  * result, which we need to know beforehand when writing a git object.
2163  * Since the primary motivation for trying to stream from the working
2164  * tree file and to avoid mmaping it in core is to deal with large
2165  * binary blobs, they generally do not want to get any conversion, and
2166  * callers should avoid this code path when filters are requested.
2167  */
2168 static int index_stream(struct object_id *oid, int fd, size_t size,
2169                         enum object_type type, const char *path,
2170                         unsigned flags)
2171 {
2172         return index_bulk_checkin(oid, fd, size, type, path, flags);
2173 }
2174
2175 int index_fd(struct index_state *istate, struct object_id *oid,
2176              int fd, struct stat *st,
2177              enum object_type type, const char *path, unsigned flags)
2178 {
2179         int ret;
2180
2181         /*
2182          * Call xsize_t() only when needed to avoid potentially unnecessary
2183          * die() for large files.
2184          */
2185         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2186                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2187         else if (!S_ISREG(st->st_mode))
2188                 ret = index_pipe(istate, oid, fd, type, path, flags);
2189         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2190                  (path && would_convert_to_git(istate, path)))
2191                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2192                                  type, path, flags);
2193         else
2194                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2195                                    flags);
2196         close(fd);
2197         return ret;
2198 }
2199
2200 int index_path(struct index_state *istate, struct object_id *oid,
2201                const char *path, struct stat *st, unsigned flags)
2202 {
2203         int fd;
2204         struct strbuf sb = STRBUF_INIT;
2205         int rc = 0;
2206
2207         switch (st->st_mode & S_IFMT) {
2208         case S_IFREG:
2209                 fd = open(path, O_RDONLY);
2210                 if (fd < 0)
2211                         return error_errno("open(\"%s\")", path);
2212                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2213                         return error(_("%s: failed to insert into database"),
2214                                      path);
2215                 break;
2216         case S_IFLNK:
2217                 if (strbuf_readlink(&sb, path, st->st_size))
2218                         return error_errno("readlink(\"%s\")", path);
2219                 if (!(flags & HASH_WRITE_OBJECT))
2220                         hash_object_file(the_hash_algo, sb.buf, sb.len,
2221                                          blob_type, oid);
2222                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2223                         rc = error(_("%s: failed to insert into database"), path);
2224                 strbuf_release(&sb);
2225                 break;
2226         case S_IFDIR:
2227                 return resolve_gitlink_ref(path, "HEAD", oid);
2228         default:
2229                 return error(_("%s: unsupported file type"), path);
2230         }
2231         return rc;
2232 }
2233
2234 int read_pack_header(int fd, struct pack_header *header)
2235 {
2236         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2237                 /* "eof before pack header was fully read" */
2238                 return PH_ERROR_EOF;
2239
2240         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2241                 /* "protocol error (pack signature mismatch detected)" */
2242                 return PH_ERROR_PACK_SIGNATURE;
2243         if (!pack_version_ok(header->hdr_version))
2244                 /* "protocol error (pack version unsupported)" */
2245                 return PH_ERROR_PROTOCOL;
2246         return 0;
2247 }
2248
2249 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2250 {
2251         enum object_type type = oid_object_info(the_repository, oid, NULL);
2252         if (type < 0)
2253                 die(_("%s is not a valid object"), oid_to_hex(oid));
2254         if (type != expect)
2255                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2256                     type_name(expect));
2257 }
2258
2259 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2260                                 struct strbuf *path,
2261                                 each_loose_object_fn obj_cb,
2262                                 each_loose_cruft_fn cruft_cb,
2263                                 each_loose_subdir_fn subdir_cb,
2264                                 void *data)
2265 {
2266         size_t origlen, baselen;
2267         DIR *dir;
2268         struct dirent *de;
2269         int r = 0;
2270         struct object_id oid;
2271
2272         if (subdir_nr > 0xff)
2273                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2274
2275         origlen = path->len;
2276         strbuf_complete(path, '/');
2277         strbuf_addf(path, "%02x", subdir_nr);
2278
2279         dir = opendir(path->buf);
2280         if (!dir) {
2281                 if (errno != ENOENT)
2282                         r = error_errno(_("unable to open %s"), path->buf);
2283                 strbuf_setlen(path, origlen);
2284                 return r;
2285         }
2286
2287         oid.hash[0] = subdir_nr;
2288         strbuf_addch(path, '/');
2289         baselen = path->len;
2290
2291         while ((de = readdir(dir))) {
2292                 size_t namelen;
2293                 if (is_dot_or_dotdot(de->d_name))
2294                         continue;
2295
2296                 namelen = strlen(de->d_name);
2297                 strbuf_setlen(path, baselen);
2298                 strbuf_add(path, de->d_name, namelen);
2299                 if (namelen == the_hash_algo->hexsz - 2 &&
2300                     !hex_to_bytes(oid.hash + 1, de->d_name,
2301                                   the_hash_algo->rawsz - 1)) {
2302                         if (obj_cb) {
2303                                 r = obj_cb(&oid, path->buf, data);
2304                                 if (r)
2305                                         break;
2306                         }
2307                         continue;
2308                 }
2309
2310                 if (cruft_cb) {
2311                         r = cruft_cb(de->d_name, path->buf, data);
2312                         if (r)
2313                                 break;
2314                 }
2315         }
2316         closedir(dir);
2317
2318         strbuf_setlen(path, baselen - 1);
2319         if (!r && subdir_cb)
2320                 r = subdir_cb(subdir_nr, path->buf, data);
2321
2322         strbuf_setlen(path, origlen);
2323
2324         return r;
2325 }
2326
2327 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2328                             each_loose_object_fn obj_cb,
2329                             each_loose_cruft_fn cruft_cb,
2330                             each_loose_subdir_fn subdir_cb,
2331                             void *data)
2332 {
2333         int r = 0;
2334         int i;
2335
2336         for (i = 0; i < 256; i++) {
2337                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2338                                                 subdir_cb, data);
2339                 if (r)
2340                         break;
2341         }
2342
2343         return r;
2344 }
2345
2346 int for_each_loose_file_in_objdir(const char *path,
2347                                   each_loose_object_fn obj_cb,
2348                                   each_loose_cruft_fn cruft_cb,
2349                                   each_loose_subdir_fn subdir_cb,
2350                                   void *data)
2351 {
2352         struct strbuf buf = STRBUF_INIT;
2353         int r;
2354
2355         strbuf_addstr(&buf, path);
2356         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2357                                               subdir_cb, data);
2358         strbuf_release(&buf);
2359
2360         return r;
2361 }
2362
2363 int for_each_loose_object(each_loose_object_fn cb, void *data,
2364                           enum for_each_object_flags flags)
2365 {
2366         struct object_directory *odb;
2367
2368         prepare_alt_odb(the_repository);
2369         for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2370                 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2371                                                       NULL, data);
2372                 if (r)
2373                         return r;
2374
2375                 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2376                         break;
2377         }
2378
2379         return 0;
2380 }
2381
2382 static int append_loose_object(const struct object_id *oid, const char *path,
2383                                void *data)
2384 {
2385         oid_array_append(data, oid);
2386         return 0;
2387 }
2388
2389 struct oid_array *odb_loose_cache(struct object_directory *odb,
2390                                   const struct object_id *oid)
2391 {
2392         int subdir_nr = oid->hash[0];
2393         struct strbuf buf = STRBUF_INIT;
2394
2395         if (subdir_nr < 0 ||
2396             subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2397                 BUG("subdir_nr out of range");
2398
2399         if (odb->loose_objects_subdir_seen[subdir_nr])
2400                 return &odb->loose_objects_cache[subdir_nr];
2401
2402         strbuf_addstr(&buf, odb->path);
2403         for_each_file_in_obj_subdir(subdir_nr, &buf,
2404                                     append_loose_object,
2405                                     NULL, NULL,
2406                                     &odb->loose_objects_cache[subdir_nr]);
2407         odb->loose_objects_subdir_seen[subdir_nr] = 1;
2408         strbuf_release(&buf);
2409         return &odb->loose_objects_cache[subdir_nr];
2410 }
2411
2412 void odb_clear_loose_cache(struct object_directory *odb)
2413 {
2414         int i;
2415
2416         for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2417                 oid_array_clear(&odb->loose_objects_cache[i]);
2418         memset(&odb->loose_objects_subdir_seen, 0,
2419                sizeof(odb->loose_objects_subdir_seen));
2420 }
2421
2422 static int check_stream_oid(git_zstream *stream,
2423                             const char *hdr,
2424                             unsigned long size,
2425                             const char *path,
2426                             const struct object_id *expected_oid)
2427 {
2428         git_hash_ctx c;
2429         struct object_id real_oid;
2430         unsigned char buf[4096];
2431         unsigned long total_read;
2432         int status = Z_OK;
2433
2434         the_hash_algo->init_fn(&c);
2435         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2436
2437         /*
2438          * We already read some bytes into hdr, but the ones up to the NUL
2439          * do not count against the object's content size.
2440          */
2441         total_read = stream->total_out - strlen(hdr) - 1;
2442
2443         /*
2444          * This size comparison must be "<=" to read the final zlib packets;
2445          * see the comment in unpack_loose_rest for details.
2446          */
2447         while (total_read <= size &&
2448                (status == Z_OK ||
2449                 (status == Z_BUF_ERROR && !stream->avail_out))) {
2450                 stream->next_out = buf;
2451                 stream->avail_out = sizeof(buf);
2452                 if (size - total_read < stream->avail_out)
2453                         stream->avail_out = size - total_read;
2454                 status = git_inflate(stream, Z_FINISH);
2455                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2456                 total_read += stream->next_out - buf;
2457         }
2458         git_inflate_end(stream);
2459
2460         if (status != Z_STREAM_END) {
2461                 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2462                 return -1;
2463         }
2464         if (stream->avail_in) {
2465                 error(_("garbage at end of loose object '%s'"),
2466                       oid_to_hex(expected_oid));
2467                 return -1;
2468         }
2469
2470         the_hash_algo->final_fn(real_oid.hash, &c);
2471         if (!oideq(expected_oid, &real_oid)) {
2472                 error(_("hash mismatch for %s (expected %s)"), path,
2473                       oid_to_hex(expected_oid));
2474                 return -1;
2475         }
2476
2477         return 0;
2478 }
2479
2480 int read_loose_object(const char *path,
2481                       const struct object_id *expected_oid,
2482                       enum object_type *type,
2483                       unsigned long *size,
2484                       void **contents)
2485 {
2486         int ret = -1;
2487         void *map = NULL;
2488         unsigned long mapsize;
2489         git_zstream stream;
2490         char hdr[MAX_HEADER_LEN];
2491
2492         *contents = NULL;
2493
2494         map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2495         if (!map) {
2496                 error_errno(_("unable to mmap %s"), path);
2497                 goto out;
2498         }
2499
2500         if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2501                 error(_("unable to unpack header of %s"), path);
2502                 goto out;
2503         }
2504
2505         *type = parse_loose_header(hdr, size);
2506         if (*type < 0) {
2507                 error(_("unable to parse header of %s"), path);
2508                 git_inflate_end(&stream);
2509                 goto out;
2510         }
2511
2512         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2513                 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2514                         goto out;
2515         } else {
2516                 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2517                 if (!*contents) {
2518                         error(_("unable to unpack contents of %s"), path);
2519                         git_inflate_end(&stream);
2520                         goto out;
2521                 }
2522                 if (check_object_signature(the_repository, expected_oid,
2523                                            *contents, *size,
2524                                            type_name(*type))) {
2525                         error(_("hash mismatch for %s (expected %s)"), path,
2526                               oid_to_hex(expected_oid));
2527                         free(*contents);
2528                         goto out;
2529                 }
2530         }
2531
2532         ret = 0; /* everything checks out */
2533
2534 out:
2535         if (map)
2536                 munmap(map, mapsize);
2537         return ret;
2538 }