Merge branch 'pb/submodule-doc-fix'
[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;
698         int seen_error = 0;
699
700         ref_git = real_pathdup(path, 0);
701         if (!ref_git) {
702                 seen_error = 1;
703                 strbuf_addf(err, _("path '%s' does not exist"), path);
704                 goto out;
705         }
706
707         repo = read_gitfile(ref_git);
708         if (!repo)
709                 repo = read_gitfile(mkpath("%s/.git", ref_git));
710         if (repo) {
711                 free(ref_git);
712                 ref_git = xstrdup(repo);
713         }
714
715         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
716                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
717                 free(ref_git);
718                 ref_git = ref_git_git;
719         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
720                 struct strbuf sb = STRBUF_INIT;
721                 seen_error = 1;
722                 if (get_common_dir(&sb, ref_git)) {
723                         strbuf_addf(err,
724                                     _("reference repository '%s' as a linked "
725                                       "checkout is not supported yet."),
726                                     path);
727                         goto out;
728                 }
729
730                 strbuf_addf(err, _("reference repository '%s' is not a "
731                                         "local repository."), path);
732                 goto out;
733         }
734
735         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
736                 strbuf_addf(err, _("reference repository '%s' is shallow"),
737                             path);
738                 seen_error = 1;
739                 goto out;
740         }
741
742         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
743                 strbuf_addf(err,
744                             _("reference repository '%s' is grafted"),
745                             path);
746                 seen_error = 1;
747                 goto out;
748         }
749
750 out:
751         if (seen_error) {
752                 FREE_AND_NULL(ref_git);
753         }
754
755         return ref_git;
756 }
757
758 static void fill_alternate_refs_command(struct child_process *cmd,
759                                         const char *repo_path)
760 {
761         const char *value;
762
763         if (!git_config_get_value("core.alternateRefsCommand", &value)) {
764                 cmd->use_shell = 1;
765
766                 strvec_push(&cmd->args, value);
767                 strvec_push(&cmd->args, repo_path);
768         } else {
769                 cmd->git_cmd = 1;
770
771                 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
772                 strvec_push(&cmd->args, "for-each-ref");
773                 strvec_push(&cmd->args, "--format=%(objectname)");
774
775                 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
776                         strvec_push(&cmd->args, "--");
777                         strvec_split(&cmd->args, value);
778                 }
779         }
780
781         cmd->env = local_repo_env;
782         cmd->out = -1;
783 }
784
785 static void read_alternate_refs(const char *path,
786                                 alternate_ref_fn *cb,
787                                 void *data)
788 {
789         struct child_process cmd = CHILD_PROCESS_INIT;
790         struct strbuf line = STRBUF_INIT;
791         FILE *fh;
792
793         fill_alternate_refs_command(&cmd, path);
794
795         if (start_command(&cmd))
796                 return;
797
798         fh = xfdopen(cmd.out, "r");
799         while (strbuf_getline_lf(&line, fh) != EOF) {
800                 struct object_id oid;
801                 const char *p;
802
803                 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
804                         warning(_("invalid line while parsing alternate refs: %s"),
805                                 line.buf);
806                         break;
807                 }
808
809                 cb(&oid, data);
810         }
811
812         fclose(fh);
813         finish_command(&cmd);
814         strbuf_release(&line);
815 }
816
817 struct alternate_refs_data {
818         alternate_ref_fn *fn;
819         void *data;
820 };
821
822 static int refs_from_alternate_cb(struct object_directory *e,
823                                   void *data)
824 {
825         struct strbuf path = STRBUF_INIT;
826         size_t base_len;
827         struct alternate_refs_data *cb = data;
828
829         if (!strbuf_realpath(&path, e->path, 0))
830                 goto out;
831         if (!strbuf_strip_suffix(&path, "/objects"))
832                 goto out;
833         base_len = path.len;
834
835         /* Is this a git repository with refs? */
836         strbuf_addstr(&path, "/refs");
837         if (!is_directory(path.buf))
838                 goto out;
839         strbuf_setlen(&path, base_len);
840
841         read_alternate_refs(path.buf, cb->fn, cb->data);
842
843 out:
844         strbuf_release(&path);
845         return 0;
846 }
847
848 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
849 {
850         struct alternate_refs_data cb;
851         cb.fn = fn;
852         cb.data = data;
853         foreach_alt_odb(refs_from_alternate_cb, &cb);
854 }
855
856 int foreach_alt_odb(alt_odb_fn fn, void *cb)
857 {
858         struct object_directory *ent;
859         int r = 0;
860
861         prepare_alt_odb(the_repository);
862         for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
863                 r = fn(ent, cb);
864                 if (r)
865                         break;
866         }
867         return r;
868 }
869
870 void prepare_alt_odb(struct repository *r)
871 {
872         if (r->objects->loaded_alternates)
873                 return;
874
875         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
876
877         read_info_alternates(r, r->objects->odb->path, 0);
878         r->objects->loaded_alternates = 1;
879 }
880
881 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
882 static int freshen_file(const char *fn)
883 {
884         return !utime(fn, NULL);
885 }
886
887 /*
888  * All of the check_and_freshen functions return 1 if the file exists and was
889  * freshened (if freshening was requested), 0 otherwise. If they return
890  * 0, you should not assume that it is safe to skip a write of the object (it
891  * either does not exist on disk, or has a stale mtime and may be subject to
892  * pruning).
893  */
894 int check_and_freshen_file(const char *fn, int freshen)
895 {
896         if (access(fn, F_OK))
897                 return 0;
898         if (freshen && !freshen_file(fn))
899                 return 0;
900         return 1;
901 }
902
903 static int check_and_freshen_odb(struct object_directory *odb,
904                                  const struct object_id *oid,
905                                  int freshen)
906 {
907         static struct strbuf path = STRBUF_INIT;
908         odb_loose_path(odb, &path, oid);
909         return check_and_freshen_file(path.buf, freshen);
910 }
911
912 static int check_and_freshen_local(const struct object_id *oid, int freshen)
913 {
914         return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
915 }
916
917 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
918 {
919         struct object_directory *odb;
920
921         prepare_alt_odb(the_repository);
922         for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
923                 if (check_and_freshen_odb(odb, oid, freshen))
924                         return 1;
925         }
926         return 0;
927 }
928
929 static int check_and_freshen(const struct object_id *oid, int freshen)
930 {
931         return check_and_freshen_local(oid, freshen) ||
932                check_and_freshen_nonlocal(oid, freshen);
933 }
934
935 int has_loose_object_nonlocal(const struct object_id *oid)
936 {
937         return check_and_freshen_nonlocal(oid, 0);
938 }
939
940 static int has_loose_object(const struct object_id *oid)
941 {
942         return check_and_freshen(oid, 0);
943 }
944
945 static void mmap_limit_check(size_t length)
946 {
947         static size_t limit = 0;
948         if (!limit) {
949                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
950                 if (!limit)
951                         limit = SIZE_MAX;
952         }
953         if (length > limit)
954                 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
955                     (uintmax_t)length, (uintmax_t)limit);
956 }
957
958 void *xmmap_gently(void *start, size_t length,
959                   int prot, int flags, int fd, off_t offset)
960 {
961         void *ret;
962
963         mmap_limit_check(length);
964         ret = mmap(start, length, prot, flags, fd, offset);
965         if (ret == MAP_FAILED && !length)
966                 ret = NULL;
967         return ret;
968 }
969
970 void *xmmap(void *start, size_t length,
971         int prot, int flags, int fd, off_t offset)
972 {
973         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
974         if (ret == MAP_FAILED)
975                 die_errno(_("mmap failed"));
976         return ret;
977 }
978
979 /*
980  * With an in-core object data in "map", rehash it to make sure the
981  * object name actually matches "oid" to detect object corruption.
982  * With "map" == NULL, try reading the object named with "oid" using
983  * the streaming interface and rehash it to do the same.
984  */
985 int check_object_signature(struct repository *r, const struct object_id *oid,
986                            void *map, unsigned long size, const char *type)
987 {
988         struct object_id real_oid;
989         enum object_type obj_type;
990         struct git_istream *st;
991         git_hash_ctx c;
992         char hdr[MAX_HEADER_LEN];
993         int hdrlen;
994
995         if (map) {
996                 hash_object_file(r->hash_algo, map, size, type, &real_oid);
997                 return !oideq(oid, &real_oid) ? -1 : 0;
998         }
999
1000         st = open_istream(r, oid, &obj_type, &size, NULL);
1001         if (!st)
1002                 return -1;
1003
1004         /* Generate the header */
1005         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1006
1007         /* Sha1.. */
1008         r->hash_algo->init_fn(&c);
1009         r->hash_algo->update_fn(&c, hdr, hdrlen);
1010         for (;;) {
1011                 char buf[1024 * 16];
1012                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1013
1014                 if (readlen < 0) {
1015                         close_istream(st);
1016                         return -1;
1017                 }
1018                 if (!readlen)
1019                         break;
1020                 r->hash_algo->update_fn(&c, buf, readlen);
1021         }
1022         r->hash_algo->final_fn(real_oid.hash, &c);
1023         close_istream(st);
1024         return !oideq(oid, &real_oid) ? -1 : 0;
1025 }
1026
1027 int git_open_cloexec(const char *name, int flags)
1028 {
1029         int fd;
1030         static int o_cloexec = O_CLOEXEC;
1031
1032         fd = open(name, flags | o_cloexec);
1033         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1034                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1035                 o_cloexec &= ~O_CLOEXEC;
1036                 fd = open(name, flags | o_cloexec);
1037         }
1038
1039 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1040         {
1041                 static int fd_cloexec = FD_CLOEXEC;
1042
1043                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1044                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
1045                         int flags = fcntl(fd, F_GETFD);
1046                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1047                                 fd_cloexec = 0;
1048                 }
1049         }
1050 #endif
1051         return fd;
1052 }
1053
1054 /*
1055  * Find "oid" as a loose object in the local repository or in an alternate.
1056  * Returns 0 on success, negative on failure.
1057  *
1058  * The "path" out-parameter will give the path of the object we found (if any).
1059  * Note that it may point to static storage and is only valid until another
1060  * call to stat_loose_object().
1061  */
1062 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1063                              struct stat *st, const char **path)
1064 {
1065         struct object_directory *odb;
1066         static struct strbuf buf = STRBUF_INIT;
1067
1068         prepare_alt_odb(r);
1069         for (odb = r->objects->odb; odb; odb = odb->next) {
1070                 *path = odb_loose_path(odb, &buf, oid);
1071                 if (!lstat(*path, st))
1072                         return 0;
1073         }
1074
1075         return -1;
1076 }
1077
1078 /*
1079  * Like stat_loose_object(), but actually open the object and return the
1080  * descriptor. See the caveats on the "path" parameter above.
1081  */
1082 static int open_loose_object(struct repository *r,
1083                              const struct object_id *oid, const char **path)
1084 {
1085         int fd;
1086         struct object_directory *odb;
1087         int most_interesting_errno = ENOENT;
1088         static struct strbuf buf = STRBUF_INIT;
1089
1090         prepare_alt_odb(r);
1091         for (odb = r->objects->odb; odb; odb = odb->next) {
1092                 *path = odb_loose_path(odb, &buf, oid);
1093                 fd = git_open(*path);
1094                 if (fd >= 0)
1095                         return fd;
1096
1097                 if (most_interesting_errno == ENOENT)
1098                         most_interesting_errno = errno;
1099         }
1100         errno = most_interesting_errno;
1101         return -1;
1102 }
1103
1104 static int quick_has_loose(struct repository *r,
1105                            const struct object_id *oid)
1106 {
1107         struct object_directory *odb;
1108
1109         prepare_alt_odb(r);
1110         for (odb = r->objects->odb; odb; odb = odb->next) {
1111                 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1112                         return 1;
1113         }
1114         return 0;
1115 }
1116
1117 /*
1118  * Map the loose object at "path" if it is not NULL, or the path found by
1119  * searching for a loose object named "oid".
1120  */
1121 static void *map_loose_object_1(struct repository *r, const char *path,
1122                              const struct object_id *oid, unsigned long *size)
1123 {
1124         void *map;
1125         int fd;
1126
1127         if (path)
1128                 fd = git_open(path);
1129         else
1130                 fd = open_loose_object(r, oid, &path);
1131         map = NULL;
1132         if (fd >= 0) {
1133                 struct stat st;
1134
1135                 if (!fstat(fd, &st)) {
1136                         *size = xsize_t(st.st_size);
1137                         if (!*size) {
1138                                 /* mmap() is forbidden on empty files */
1139                                 error(_("object file %s is empty"), path);
1140                                 close(fd);
1141                                 return NULL;
1142                         }
1143                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1144                 }
1145                 close(fd);
1146         }
1147         return map;
1148 }
1149
1150 void *map_loose_object(struct repository *r,
1151                        const struct object_id *oid,
1152                        unsigned long *size)
1153 {
1154         return map_loose_object_1(r, NULL, oid, size);
1155 }
1156
1157 static int unpack_loose_short_header(git_zstream *stream,
1158                                      unsigned char *map, unsigned long mapsize,
1159                                      void *buffer, unsigned long bufsiz)
1160 {
1161         int ret;
1162
1163         /* Get the data stream */
1164         memset(stream, 0, sizeof(*stream));
1165         stream->next_in = map;
1166         stream->avail_in = mapsize;
1167         stream->next_out = buffer;
1168         stream->avail_out = bufsiz;
1169
1170         git_inflate_init(stream);
1171         obj_read_unlock();
1172         ret = git_inflate(stream, 0);
1173         obj_read_lock();
1174
1175         return ret;
1176 }
1177
1178 int unpack_loose_header(git_zstream *stream,
1179                         unsigned char *map, unsigned long mapsize,
1180                         void *buffer, unsigned long bufsiz)
1181 {
1182         int status = unpack_loose_short_header(stream, map, mapsize,
1183                                                buffer, bufsiz);
1184
1185         if (status < Z_OK)
1186                 return status;
1187
1188         /* Make sure we have the terminating NUL */
1189         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1190                 return -1;
1191         return 0;
1192 }
1193
1194 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1195                                          unsigned long mapsize, void *buffer,
1196                                          unsigned long bufsiz, struct strbuf *header)
1197 {
1198         int status;
1199
1200         status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1201         if (status < Z_OK)
1202                 return -1;
1203
1204         /*
1205          * Check if entire header is unpacked in the first iteration.
1206          */
1207         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1208                 return 0;
1209
1210         /*
1211          * buffer[0..bufsiz] was not large enough.  Copy the partial
1212          * result out to header, and then append the result of further
1213          * reading the stream.
1214          */
1215         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1216         stream->next_out = buffer;
1217         stream->avail_out = bufsiz;
1218
1219         do {
1220                 obj_read_unlock();
1221                 status = git_inflate(stream, 0);
1222                 obj_read_lock();
1223                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1224                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1225                         return 0;
1226                 stream->next_out = buffer;
1227                 stream->avail_out = bufsiz;
1228         } while (status != Z_STREAM_END);
1229         return -1;
1230 }
1231
1232 static void *unpack_loose_rest(git_zstream *stream,
1233                                void *buffer, unsigned long size,
1234                                const struct object_id *oid)
1235 {
1236         int bytes = strlen(buffer) + 1;
1237         unsigned char *buf = xmallocz(size);
1238         unsigned long n;
1239         int status = Z_OK;
1240
1241         n = stream->total_out - bytes;
1242         if (n > size)
1243                 n = size;
1244         memcpy(buf, (char *) buffer + bytes, n);
1245         bytes = n;
1246         if (bytes <= size) {
1247                 /*
1248                  * The above condition must be (bytes <= size), not
1249                  * (bytes < size).  In other words, even though we
1250                  * expect no more output and set avail_out to zero,
1251                  * the input zlib stream may have bytes that express
1252                  * "this concludes the stream", and we *do* want to
1253                  * eat that input.
1254                  *
1255                  * Otherwise we would not be able to test that we
1256                  * consumed all the input to reach the expected size;
1257                  * we also want to check that zlib tells us that all
1258                  * went well with status == Z_STREAM_END at the end.
1259                  */
1260                 stream->next_out = buf + bytes;
1261                 stream->avail_out = size - bytes;
1262                 while (status == Z_OK) {
1263                         obj_read_unlock();
1264                         status = git_inflate(stream, Z_FINISH);
1265                         obj_read_lock();
1266                 }
1267         }
1268         if (status == Z_STREAM_END && !stream->avail_in) {
1269                 git_inflate_end(stream);
1270                 return buf;
1271         }
1272
1273         if (status < 0)
1274                 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1275         else if (stream->avail_in)
1276                 error(_("garbage at end of loose object '%s'"),
1277                       oid_to_hex(oid));
1278         free(buf);
1279         return NULL;
1280 }
1281
1282 /*
1283  * We used to just use "sscanf()", but that's actually way
1284  * too permissive for what we want to check. So do an anal
1285  * object header parse by hand.
1286  */
1287 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1288                                        unsigned int flags)
1289 {
1290         const char *type_buf = hdr;
1291         unsigned long size;
1292         int type, type_len = 0;
1293
1294         /*
1295          * The type can be of any size but is followed by
1296          * a space.
1297          */
1298         for (;;) {
1299                 char c = *hdr++;
1300                 if (!c)
1301                         return -1;
1302                 if (c == ' ')
1303                         break;
1304                 type_len++;
1305         }
1306
1307         type = type_from_string_gently(type_buf, type_len, 1);
1308         if (oi->type_name)
1309                 strbuf_add(oi->type_name, type_buf, type_len);
1310         /*
1311          * Set type to 0 if its an unknown object and
1312          * we're obtaining the type using '--allow-unknown-type'
1313          * option.
1314          */
1315         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1316                 type = 0;
1317         else if (type < 0)
1318                 die(_("invalid object type"));
1319         if (oi->typep)
1320                 *oi->typep = type;
1321
1322         /*
1323          * The length must follow immediately, and be in canonical
1324          * decimal format (ie "010" is not valid).
1325          */
1326         size = *hdr++ - '0';
1327         if (size > 9)
1328                 return -1;
1329         if (size) {
1330                 for (;;) {
1331                         unsigned long c = *hdr - '0';
1332                         if (c > 9)
1333                                 break;
1334                         hdr++;
1335                         size = size * 10 + c;
1336                 }
1337         }
1338
1339         if (oi->sizep)
1340                 *oi->sizep = size;
1341
1342         /*
1343          * The length must be followed by a zero byte
1344          */
1345         return *hdr ? -1 : type;
1346 }
1347
1348 int parse_loose_header(const char *hdr, unsigned long *sizep)
1349 {
1350         struct object_info oi = OBJECT_INFO_INIT;
1351
1352         oi.sizep = sizep;
1353         return parse_loose_header_extended(hdr, &oi, 0);
1354 }
1355
1356 static int loose_object_info(struct repository *r,
1357                              const struct object_id *oid,
1358                              struct object_info *oi, int flags)
1359 {
1360         int status = 0;
1361         unsigned long mapsize;
1362         void *map;
1363         git_zstream stream;
1364         char hdr[MAX_HEADER_LEN];
1365         struct strbuf hdrbuf = STRBUF_INIT;
1366         unsigned long size_scratch;
1367
1368         if (oi->delta_base_oid)
1369                 oidclr(oi->delta_base_oid);
1370
1371         /*
1372          * If we don't care about type or size, then we don't
1373          * need to look inside the object at all. Note that we
1374          * do not optimize out the stat call, even if the
1375          * caller doesn't care about the disk-size, since our
1376          * return value implicitly indicates whether the
1377          * object even exists.
1378          */
1379         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1380                 const char *path;
1381                 struct stat st;
1382                 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1383                         return quick_has_loose(r, oid) ? 0 : -1;
1384                 if (stat_loose_object(r, oid, &st, &path) < 0)
1385                         return -1;
1386                 if (oi->disk_sizep)
1387                         *oi->disk_sizep = st.st_size;
1388                 return 0;
1389         }
1390
1391         map = map_loose_object(r, oid, &mapsize);
1392         if (!map)
1393                 return -1;
1394
1395         if (!oi->sizep)
1396                 oi->sizep = &size_scratch;
1397
1398         if (oi->disk_sizep)
1399                 *oi->disk_sizep = mapsize;
1400         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1401                 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1402                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1403                                        oid_to_hex(oid));
1404         } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1405                 status = error(_("unable to unpack %s header"),
1406                                oid_to_hex(oid));
1407         if (status < 0)
1408                 ; /* Do nothing */
1409         else if (hdrbuf.len) {
1410                 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1411                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1412                                        oid_to_hex(oid));
1413         } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1414                 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1415
1416         if (status >= 0 && oi->contentp) {
1417                 *oi->contentp = unpack_loose_rest(&stream, hdr,
1418                                                   *oi->sizep, oid);
1419                 if (!*oi->contentp) {
1420                         git_inflate_end(&stream);
1421                         status = -1;
1422                 }
1423         } else
1424                 git_inflate_end(&stream);
1425
1426         munmap(map, mapsize);
1427         if (status && oi->typep)
1428                 *oi->typep = status;
1429         if (oi->sizep == &size_scratch)
1430                 oi->sizep = NULL;
1431         strbuf_release(&hdrbuf);
1432         oi->whence = OI_LOOSE;
1433         return (status < 0) ? status : 0;
1434 }
1435
1436 int obj_read_use_lock = 0;
1437 pthread_mutex_t obj_read_mutex;
1438
1439 void enable_obj_read_lock(void)
1440 {
1441         if (obj_read_use_lock)
1442                 return;
1443
1444         obj_read_use_lock = 1;
1445         init_recursive_mutex(&obj_read_mutex);
1446 }
1447
1448 void disable_obj_read_lock(void)
1449 {
1450         if (!obj_read_use_lock)
1451                 return;
1452
1453         obj_read_use_lock = 0;
1454         pthread_mutex_destroy(&obj_read_mutex);
1455 }
1456
1457 int fetch_if_missing = 1;
1458
1459 static int do_oid_object_info_extended(struct repository *r,
1460                                        const struct object_id *oid,
1461                                        struct object_info *oi, unsigned flags)
1462 {
1463         static struct object_info blank_oi = OBJECT_INFO_INIT;
1464         struct cached_object *co;
1465         struct pack_entry e;
1466         int rtype;
1467         const struct object_id *real = oid;
1468         int already_retried = 0;
1469
1470
1471         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1472                 real = lookup_replace_object(r, oid);
1473
1474         if (is_null_oid(real))
1475                 return -1;
1476
1477         if (!oi)
1478                 oi = &blank_oi;
1479
1480         co = find_cached_object(real);
1481         if (co) {
1482                 if (oi->typep)
1483                         *(oi->typep) = co->type;
1484                 if (oi->sizep)
1485                         *(oi->sizep) = co->size;
1486                 if (oi->disk_sizep)
1487                         *(oi->disk_sizep) = 0;
1488                 if (oi->delta_base_oid)
1489                         oidclr(oi->delta_base_oid);
1490                 if (oi->type_name)
1491                         strbuf_addstr(oi->type_name, type_name(co->type));
1492                 if (oi->contentp)
1493                         *oi->contentp = xmemdupz(co->buf, co->size);
1494                 oi->whence = OI_CACHED;
1495                 return 0;
1496         }
1497
1498         while (1) {
1499                 if (find_pack_entry(r, real, &e))
1500                         break;
1501
1502                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1503                         return -1;
1504
1505                 /* Most likely it's a loose object. */
1506                 if (!loose_object_info(r, real, oi, flags))
1507                         return 0;
1508
1509                 /* Not a loose object; someone else may have just packed it. */
1510                 if (!(flags & OBJECT_INFO_QUICK)) {
1511                         reprepare_packed_git(r);
1512                         if (find_pack_entry(r, real, &e))
1513                                 break;
1514                 }
1515
1516                 /* Check if it is a missing object */
1517                 if (fetch_if_missing && has_promisor_remote() &&
1518                     !already_retried && r == the_repository &&
1519                     !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1520                         /*
1521                          * TODO Investigate checking promisor_remote_get_direct()
1522                          * TODO return value and stopping on error here.
1523                          * TODO Pass a repository struct through
1524                          * promisor_remote_get_direct(), such that arbitrary
1525                          * repositories work.
1526                          */
1527                         promisor_remote_get_direct(r, real, 1);
1528                         already_retried = 1;
1529                         continue;
1530                 }
1531
1532                 return -1;
1533         }
1534
1535         if (oi == &blank_oi)
1536                 /*
1537                  * We know that the caller doesn't actually need the
1538                  * information below, so return early.
1539                  */
1540                 return 0;
1541         rtype = packed_object_info(r, e.p, e.offset, oi);
1542         if (rtype < 0) {
1543                 mark_bad_packed_object(e.p, real->hash);
1544                 return do_oid_object_info_extended(r, real, oi, 0);
1545         } else if (oi->whence == OI_PACKED) {
1546                 oi->u.packed.offset = e.offset;
1547                 oi->u.packed.pack = e.p;
1548                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1549                                          rtype == OBJ_OFS_DELTA);
1550         }
1551
1552         return 0;
1553 }
1554
1555 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1556                              struct object_info *oi, unsigned flags)
1557 {
1558         int ret;
1559         obj_read_lock();
1560         ret = do_oid_object_info_extended(r, oid, oi, flags);
1561         obj_read_unlock();
1562         return ret;
1563 }
1564
1565
1566 /* returns enum object_type or negative */
1567 int oid_object_info(struct repository *r,
1568                     const struct object_id *oid,
1569                     unsigned long *sizep)
1570 {
1571         enum object_type type;
1572         struct object_info oi = OBJECT_INFO_INIT;
1573
1574         oi.typep = &type;
1575         oi.sizep = sizep;
1576         if (oid_object_info_extended(r, oid, &oi,
1577                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1578                 return -1;
1579         return type;
1580 }
1581
1582 static void *read_object(struct repository *r,
1583                          const struct object_id *oid, enum object_type *type,
1584                          unsigned long *size)
1585 {
1586         struct object_info oi = OBJECT_INFO_INIT;
1587         void *content;
1588         oi.typep = type;
1589         oi.sizep = size;
1590         oi.contentp = &content;
1591
1592         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1593                 return NULL;
1594         return content;
1595 }
1596
1597 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1598                         struct object_id *oid)
1599 {
1600         struct cached_object *co;
1601
1602         hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1603         if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1604             find_cached_object(oid))
1605                 return 0;
1606         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1607         co = &cached_objects[cached_object_nr++];
1608         co->size = len;
1609         co->type = type;
1610         co->buf = xmalloc(len);
1611         memcpy(co->buf, buf, len);
1612         oidcpy(&co->oid, oid);
1613         return 0;
1614 }
1615
1616 /*
1617  * This function dies on corrupt objects; the callers who want to
1618  * deal with them should arrange to call read_object() and give error
1619  * messages themselves.
1620  */
1621 void *read_object_file_extended(struct repository *r,
1622                                 const struct object_id *oid,
1623                                 enum object_type *type,
1624                                 unsigned long *size,
1625                                 int lookup_replace)
1626 {
1627         void *data;
1628         const struct packed_git *p;
1629         const char *path;
1630         struct stat st;
1631         const struct object_id *repl = lookup_replace ?
1632                 lookup_replace_object(r, oid) : oid;
1633
1634         errno = 0;
1635         data = read_object(r, repl, type, size);
1636         if (data)
1637                 return data;
1638
1639         obj_read_lock();
1640         if (errno && errno != ENOENT)
1641                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1642
1643         /* die if we replaced an object with one that does not exist */
1644         if (repl != oid)
1645                 die(_("replacement %s not found for %s"),
1646                     oid_to_hex(repl), oid_to_hex(oid));
1647
1648         if (!stat_loose_object(r, repl, &st, &path))
1649                 die(_("loose object %s (stored in %s) is corrupt"),
1650                     oid_to_hex(repl), path);
1651
1652         if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1653                 die(_("packed object %s (stored in %s) is corrupt"),
1654                     oid_to_hex(repl), p->pack_name);
1655         obj_read_unlock();
1656
1657         return NULL;
1658 }
1659
1660 void *read_object_with_reference(struct repository *r,
1661                                  const struct object_id *oid,
1662                                  const char *required_type_name,
1663                                  unsigned long *size,
1664                                  struct object_id *actual_oid_return)
1665 {
1666         enum object_type type, required_type;
1667         void *buffer;
1668         unsigned long isize;
1669         struct object_id actual_oid;
1670
1671         required_type = type_from_string(required_type_name);
1672         oidcpy(&actual_oid, oid);
1673         while (1) {
1674                 int ref_length = -1;
1675                 const char *ref_type = NULL;
1676
1677                 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1678                 if (!buffer)
1679                         return NULL;
1680                 if (type == required_type) {
1681                         *size = isize;
1682                         if (actual_oid_return)
1683                                 oidcpy(actual_oid_return, &actual_oid);
1684                         return buffer;
1685                 }
1686                 /* Handle references */
1687                 else if (type == OBJ_COMMIT)
1688                         ref_type = "tree ";
1689                 else if (type == OBJ_TAG)
1690                         ref_type = "object ";
1691                 else {
1692                         free(buffer);
1693                         return NULL;
1694                 }
1695                 ref_length = strlen(ref_type);
1696
1697                 if (ref_length + the_hash_algo->hexsz > isize ||
1698                     memcmp(buffer, ref_type, ref_length) ||
1699                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1700                         free(buffer);
1701                         return NULL;
1702                 }
1703                 free(buffer);
1704                 /* Now we have the ID of the referred-to object in
1705                  * actual_oid.  Check again. */
1706         }
1707 }
1708
1709 static void write_object_file_prepare(const struct git_hash_algo *algo,
1710                                       const void *buf, unsigned long len,
1711                                       const char *type, struct object_id *oid,
1712                                       char *hdr, int *hdrlen)
1713 {
1714         git_hash_ctx c;
1715
1716         /* Generate the header */
1717         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1718
1719         /* Sha1.. */
1720         algo->init_fn(&c);
1721         algo->update_fn(&c, hdr, *hdrlen);
1722         algo->update_fn(&c, buf, len);
1723         algo->final_fn(oid->hash, &c);
1724 }
1725
1726 /*
1727  * Move the just written object into its final resting place.
1728  */
1729 int finalize_object_file(const char *tmpfile, const char *filename)
1730 {
1731         int ret = 0;
1732
1733         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1734                 goto try_rename;
1735         else if (link(tmpfile, filename))
1736                 ret = errno;
1737
1738         /*
1739          * Coda hack - coda doesn't like cross-directory links,
1740          * so we fall back to a rename, which will mean that it
1741          * won't be able to check collisions, but that's not a
1742          * big deal.
1743          *
1744          * The same holds for FAT formatted media.
1745          *
1746          * When this succeeds, we just return.  We have nothing
1747          * left to unlink.
1748          */
1749         if (ret && ret != EEXIST) {
1750         try_rename:
1751                 if (!rename(tmpfile, filename))
1752                         goto out;
1753                 ret = errno;
1754         }
1755         unlink_or_warn(tmpfile);
1756         if (ret) {
1757                 if (ret != EEXIST) {
1758                         return error_errno(_("unable to write file %s"), filename);
1759                 }
1760                 /* FIXME!!! Collision check here ? */
1761         }
1762
1763 out:
1764         if (adjust_shared_perm(filename))
1765                 return error(_("unable to set permission to '%s'"), filename);
1766         return 0;
1767 }
1768
1769 static int write_buffer(int fd, const void *buf, size_t len)
1770 {
1771         if (write_in_full(fd, buf, len) < 0)
1772                 return error_errno(_("file write error"));
1773         return 0;
1774 }
1775
1776 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1777                      unsigned long len, const char *type,
1778                      struct object_id *oid)
1779 {
1780         char hdr[MAX_HEADER_LEN];
1781         int hdrlen = sizeof(hdr);
1782         write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1783         return 0;
1784 }
1785
1786 /* Finalize a file on disk, and close it. */
1787 static void close_loose_object(int fd)
1788 {
1789         if (fsync_object_files)
1790                 fsync_or_die(fd, "loose object file");
1791         if (close(fd) != 0)
1792                 die_errno(_("error when closing loose object file"));
1793 }
1794
1795 /* Size of directory component, including the ending '/' */
1796 static inline int directory_size(const char *filename)
1797 {
1798         const char *s = strrchr(filename, '/');
1799         if (!s)
1800                 return 0;
1801         return s - filename + 1;
1802 }
1803
1804 /*
1805  * This creates a temporary file in the same directory as the final
1806  * 'filename'
1807  *
1808  * We want to avoid cross-directory filename renames, because those
1809  * can have problems on various filesystems (FAT, NFS, Coda).
1810  */
1811 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1812 {
1813         int fd, dirlen = directory_size(filename);
1814
1815         strbuf_reset(tmp);
1816         strbuf_add(tmp, filename, dirlen);
1817         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1818         fd = git_mkstemp_mode(tmp->buf, 0444);
1819         if (fd < 0 && dirlen && errno == ENOENT) {
1820                 /*
1821                  * Make sure the directory exists; note that the contents
1822                  * of the buffer are undefined after mkstemp returns an
1823                  * error, so we have to rewrite the whole buffer from
1824                  * scratch.
1825                  */
1826                 strbuf_reset(tmp);
1827                 strbuf_add(tmp, filename, dirlen - 1);
1828                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1829                         return -1;
1830                 if (adjust_shared_perm(tmp->buf))
1831                         return -1;
1832
1833                 /* Try again */
1834                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1835                 fd = git_mkstemp_mode(tmp->buf, 0444);
1836         }
1837         return fd;
1838 }
1839
1840 static int write_loose_object(const struct object_id *oid, char *hdr,
1841                               int hdrlen, const void *buf, unsigned long len,
1842                               time_t mtime)
1843 {
1844         int fd, ret;
1845         unsigned char compressed[4096];
1846         git_zstream stream;
1847         git_hash_ctx c;
1848         struct object_id parano_oid;
1849         static struct strbuf tmp_file = STRBUF_INIT;
1850         static struct strbuf filename = STRBUF_INIT;
1851
1852         loose_object_path(the_repository, &filename, oid);
1853
1854         fd = create_tmpfile(&tmp_file, filename.buf);
1855         if (fd < 0) {
1856                 if (errno == EACCES)
1857                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1858                 else
1859                         return error_errno(_("unable to create temporary file"));
1860         }
1861
1862         /* Set it up */
1863         git_deflate_init(&stream, zlib_compression_level);
1864         stream.next_out = compressed;
1865         stream.avail_out = sizeof(compressed);
1866         the_hash_algo->init_fn(&c);
1867
1868         /* First header.. */
1869         stream.next_in = (unsigned char *)hdr;
1870         stream.avail_in = hdrlen;
1871         while (git_deflate(&stream, 0) == Z_OK)
1872                 ; /* nothing */
1873         the_hash_algo->update_fn(&c, hdr, hdrlen);
1874
1875         /* Then the data itself.. */
1876         stream.next_in = (void *)buf;
1877         stream.avail_in = len;
1878         do {
1879                 unsigned char *in0 = stream.next_in;
1880                 ret = git_deflate(&stream, Z_FINISH);
1881                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1882                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1883                         die(_("unable to write loose object file"));
1884                 stream.next_out = compressed;
1885                 stream.avail_out = sizeof(compressed);
1886         } while (ret == Z_OK);
1887
1888         if (ret != Z_STREAM_END)
1889                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1890                     ret);
1891         ret = git_deflate_end_gently(&stream);
1892         if (ret != Z_OK)
1893                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1894                     ret);
1895         the_hash_algo->final_fn(parano_oid.hash, &c);
1896         if (!oideq(oid, &parano_oid))
1897                 die(_("confused by unstable object source data for %s"),
1898                     oid_to_hex(oid));
1899
1900         close_loose_object(fd);
1901
1902         if (mtime) {
1903                 struct utimbuf utb;
1904                 utb.actime = mtime;
1905                 utb.modtime = mtime;
1906                 if (utime(tmp_file.buf, &utb) < 0)
1907                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1908         }
1909
1910         return finalize_object_file(tmp_file.buf, filename.buf);
1911 }
1912
1913 static int freshen_loose_object(const struct object_id *oid)
1914 {
1915         return check_and_freshen(oid, 1);
1916 }
1917
1918 static int freshen_packed_object(const struct object_id *oid)
1919 {
1920         struct pack_entry e;
1921         if (!find_pack_entry(the_repository, oid, &e))
1922                 return 0;
1923         if (e.p->freshened)
1924                 return 1;
1925         if (!freshen_file(e.p->pack_name))
1926                 return 0;
1927         e.p->freshened = 1;
1928         return 1;
1929 }
1930
1931 int write_object_file(const void *buf, unsigned long len, const char *type,
1932                       struct object_id *oid)
1933 {
1934         char hdr[MAX_HEADER_LEN];
1935         int hdrlen = sizeof(hdr);
1936
1937         /* Normally if we have it in the pack then we do not bother writing
1938          * it out into .git/objects/??/?{38} file.
1939          */
1940         write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1941                                   &hdrlen);
1942         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1943                 return 0;
1944         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1945 }
1946
1947 int hash_object_file_literally(const void *buf, unsigned long len,
1948                                const char *type, struct object_id *oid,
1949                                unsigned flags)
1950 {
1951         char *header;
1952         int hdrlen, status = 0;
1953
1954         /* type string, SP, %lu of the length plus NUL must fit this */
1955         hdrlen = strlen(type) + MAX_HEADER_LEN;
1956         header = xmalloc(hdrlen);
1957         write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1958                                   &hdrlen);
1959
1960         if (!(flags & HASH_WRITE_OBJECT))
1961                 goto cleanup;
1962         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1963                 goto cleanup;
1964         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1965
1966 cleanup:
1967         free(header);
1968         return status;
1969 }
1970
1971 int force_object_loose(const struct object_id *oid, time_t mtime)
1972 {
1973         void *buf;
1974         unsigned long len;
1975         enum object_type type;
1976         char hdr[MAX_HEADER_LEN];
1977         int hdrlen;
1978         int ret;
1979
1980         if (has_loose_object(oid))
1981                 return 0;
1982         buf = read_object(the_repository, oid, &type, &len);
1983         if (!buf)
1984                 return error(_("cannot read object for %s"), oid_to_hex(oid));
1985         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1986         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1987         free(buf);
1988
1989         return ret;
1990 }
1991
1992 int has_object(struct repository *r, const struct object_id *oid,
1993                unsigned flags)
1994 {
1995         int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
1996         unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
1997                 (quick ? OBJECT_INFO_QUICK : 0);
1998
1999         if (!startup_info->have_repository)
2000                 return 0;
2001         return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2002 }
2003
2004 int repo_has_object_file_with_flags(struct repository *r,
2005                                     const struct object_id *oid, int flags)
2006 {
2007         if (!startup_info->have_repository)
2008                 return 0;
2009         return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2010 }
2011
2012 int repo_has_object_file(struct repository *r,
2013                          const struct object_id *oid)
2014 {
2015         return repo_has_object_file_with_flags(r, oid, 0);
2016 }
2017
2018 static void check_tree(const void *buf, size_t size)
2019 {
2020         struct tree_desc desc;
2021         struct name_entry entry;
2022
2023         init_tree_desc(&desc, buf, size);
2024         while (tree_entry(&desc, &entry))
2025                 /* do nothing
2026                  * tree_entry() will die() on malformed entries */
2027                 ;
2028 }
2029
2030 static void check_commit(const void *buf, size_t size)
2031 {
2032         struct commit c;
2033         memset(&c, 0, sizeof(c));
2034         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2035                 die(_("corrupt commit"));
2036 }
2037
2038 static void check_tag(const void *buf, size_t size)
2039 {
2040         struct tag t;
2041         memset(&t, 0, sizeof(t));
2042         if (parse_tag_buffer(the_repository, &t, buf, size))
2043                 die(_("corrupt tag"));
2044 }
2045
2046 static int index_mem(struct index_state *istate,
2047                      struct object_id *oid, void *buf, size_t size,
2048                      enum object_type type,
2049                      const char *path, unsigned flags)
2050 {
2051         int ret, re_allocated = 0;
2052         int write_object = flags & HASH_WRITE_OBJECT;
2053
2054         if (!type)
2055                 type = OBJ_BLOB;
2056
2057         /*
2058          * Convert blobs to git internal format
2059          */
2060         if ((type == OBJ_BLOB) && path) {
2061                 struct strbuf nbuf = STRBUF_INIT;
2062                 if (convert_to_git(istate, path, buf, size, &nbuf,
2063                                    get_conv_flags(flags))) {
2064                         buf = strbuf_detach(&nbuf, &size);
2065                         re_allocated = 1;
2066                 }
2067         }
2068         if (flags & HASH_FORMAT_CHECK) {
2069                 if (type == OBJ_TREE)
2070                         check_tree(buf, size);
2071                 if (type == OBJ_COMMIT)
2072                         check_commit(buf, size);
2073                 if (type == OBJ_TAG)
2074                         check_tag(buf, size);
2075         }
2076
2077         if (write_object)
2078                 ret = write_object_file(buf, size, type_name(type), oid);
2079         else
2080                 ret = hash_object_file(the_hash_algo, buf, size,
2081                                        type_name(type), oid);
2082         if (re_allocated)
2083                 free(buf);
2084         return ret;
2085 }
2086
2087 static int index_stream_convert_blob(struct index_state *istate,
2088                                      struct object_id *oid,
2089                                      int fd,
2090                                      const char *path,
2091                                      unsigned flags)
2092 {
2093         int ret;
2094         const int write_object = flags & HASH_WRITE_OBJECT;
2095         struct strbuf sbuf = STRBUF_INIT;
2096
2097         assert(path);
2098         assert(would_convert_to_git_filter_fd(istate, path));
2099
2100         convert_to_git_filter_fd(istate, path, fd, &sbuf,
2101                                  get_conv_flags(flags));
2102
2103         if (write_object)
2104                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2105                                         oid);
2106         else
2107                 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2108                                        type_name(OBJ_BLOB), oid);
2109         strbuf_release(&sbuf);
2110         return ret;
2111 }
2112
2113 static int index_pipe(struct index_state *istate, struct object_id *oid,
2114                       int fd, enum object_type type,
2115                       const char *path, unsigned flags)
2116 {
2117         struct strbuf sbuf = STRBUF_INIT;
2118         int ret;
2119
2120         if (strbuf_read(&sbuf, fd, 4096) >= 0)
2121                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2122         else
2123                 ret = -1;
2124         strbuf_release(&sbuf);
2125         return ret;
2126 }
2127
2128 #define SMALL_FILE_SIZE (32*1024)
2129
2130 static int index_core(struct index_state *istate,
2131                       struct object_id *oid, int fd, size_t size,
2132                       enum object_type type, const char *path,
2133                       unsigned flags)
2134 {
2135         int ret;
2136
2137         if (!size) {
2138                 ret = index_mem(istate, oid, "", size, type, path, flags);
2139         } else if (size <= SMALL_FILE_SIZE) {
2140                 char *buf = xmalloc(size);
2141                 ssize_t read_result = read_in_full(fd, buf, size);
2142                 if (read_result < 0)
2143                         ret = error_errno(_("read error while indexing %s"),
2144                                           path ? path : "<unknown>");
2145                 else if (read_result != size)
2146                         ret = error(_("short read while indexing %s"),
2147                                     path ? path : "<unknown>");
2148                 else
2149                         ret = index_mem(istate, oid, buf, size, type, path, flags);
2150                 free(buf);
2151         } else {
2152                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2153                 ret = index_mem(istate, oid, buf, size, type, path, flags);
2154                 munmap(buf, size);
2155         }
2156         return ret;
2157 }
2158
2159 /*
2160  * This creates one packfile per large blob unless bulk-checkin
2161  * machinery is "plugged".
2162  *
2163  * This also bypasses the usual "convert-to-git" dance, and that is on
2164  * purpose. We could write a streaming version of the converting
2165  * functions and insert that before feeding the data to fast-import
2166  * (or equivalent in-core API described above). However, that is
2167  * somewhat complicated, as we do not know the size of the filter
2168  * result, which we need to know beforehand when writing a git object.
2169  * Since the primary motivation for trying to stream from the working
2170  * tree file and to avoid mmaping it in core is to deal with large
2171  * binary blobs, they generally do not want to get any conversion, and
2172  * callers should avoid this code path when filters are requested.
2173  */
2174 static int index_stream(struct object_id *oid, int fd, size_t size,
2175                         enum object_type type, const char *path,
2176                         unsigned flags)
2177 {
2178         return index_bulk_checkin(oid, fd, size, type, path, flags);
2179 }
2180
2181 int index_fd(struct index_state *istate, struct object_id *oid,
2182              int fd, struct stat *st,
2183              enum object_type type, const char *path, unsigned flags)
2184 {
2185         int ret;
2186
2187         /*
2188          * Call xsize_t() only when needed to avoid potentially unnecessary
2189          * die() for large files.
2190          */
2191         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2192                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2193         else if (!S_ISREG(st->st_mode))
2194                 ret = index_pipe(istate, oid, fd, type, path, flags);
2195         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2196                  (path && would_convert_to_git(istate, path)))
2197                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2198                                  type, path, flags);
2199         else
2200                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2201                                    flags);
2202         close(fd);
2203         return ret;
2204 }
2205
2206 int index_path(struct index_state *istate, struct object_id *oid,
2207                const char *path, struct stat *st, unsigned flags)
2208 {
2209         int fd;
2210         struct strbuf sb = STRBUF_INIT;
2211         int rc = 0;
2212
2213         switch (st->st_mode & S_IFMT) {
2214         case S_IFREG:
2215                 fd = open(path, O_RDONLY);
2216                 if (fd < 0)
2217                         return error_errno("open(\"%s\")", path);
2218                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2219                         return error(_("%s: failed to insert into database"),
2220                                      path);
2221                 break;
2222         case S_IFLNK:
2223                 if (strbuf_readlink(&sb, path, st->st_size))
2224                         return error_errno("readlink(\"%s\")", path);
2225                 if (!(flags & HASH_WRITE_OBJECT))
2226                         hash_object_file(the_hash_algo, sb.buf, sb.len,
2227                                          blob_type, oid);
2228                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2229                         rc = error(_("%s: failed to insert into database"), path);
2230                 strbuf_release(&sb);
2231                 break;
2232         case S_IFDIR:
2233                 return resolve_gitlink_ref(path, "HEAD", oid);
2234         default:
2235                 return error(_("%s: unsupported file type"), path);
2236         }
2237         return rc;
2238 }
2239
2240 int read_pack_header(int fd, struct pack_header *header)
2241 {
2242         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2243                 /* "eof before pack header was fully read" */
2244                 return PH_ERROR_EOF;
2245
2246         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2247                 /* "protocol error (pack signature mismatch detected)" */
2248                 return PH_ERROR_PACK_SIGNATURE;
2249         if (!pack_version_ok(header->hdr_version))
2250                 /* "protocol error (pack version unsupported)" */
2251                 return PH_ERROR_PROTOCOL;
2252         return 0;
2253 }
2254
2255 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2256 {
2257         enum object_type type = oid_object_info(the_repository, oid, NULL);
2258         if (type < 0)
2259                 die(_("%s is not a valid object"), oid_to_hex(oid));
2260         if (type != expect)
2261                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2262                     type_name(expect));
2263 }
2264
2265 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2266                                 struct strbuf *path,
2267                                 each_loose_object_fn obj_cb,
2268                                 each_loose_cruft_fn cruft_cb,
2269                                 each_loose_subdir_fn subdir_cb,
2270                                 void *data)
2271 {
2272         size_t origlen, baselen;
2273         DIR *dir;
2274         struct dirent *de;
2275         int r = 0;
2276         struct object_id oid;
2277
2278         if (subdir_nr > 0xff)
2279                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2280
2281         origlen = path->len;
2282         strbuf_complete(path, '/');
2283         strbuf_addf(path, "%02x", subdir_nr);
2284
2285         dir = opendir(path->buf);
2286         if (!dir) {
2287                 if (errno != ENOENT)
2288                         r = error_errno(_("unable to open %s"), path->buf);
2289                 strbuf_setlen(path, origlen);
2290                 return r;
2291         }
2292
2293         oid.hash[0] = subdir_nr;
2294         strbuf_addch(path, '/');
2295         baselen = path->len;
2296
2297         while ((de = readdir(dir))) {
2298                 size_t namelen;
2299                 if (is_dot_or_dotdot(de->d_name))
2300                         continue;
2301
2302                 namelen = strlen(de->d_name);
2303                 strbuf_setlen(path, baselen);
2304                 strbuf_add(path, de->d_name, namelen);
2305                 if (namelen == the_hash_algo->hexsz - 2 &&
2306                     !hex_to_bytes(oid.hash + 1, de->d_name,
2307                                   the_hash_algo->rawsz - 1)) {
2308                         if (obj_cb) {
2309                                 r = obj_cb(&oid, path->buf, data);
2310                                 if (r)
2311                                         break;
2312                         }
2313                         continue;
2314                 }
2315
2316                 if (cruft_cb) {
2317                         r = cruft_cb(de->d_name, path->buf, data);
2318                         if (r)
2319                                 break;
2320                 }
2321         }
2322         closedir(dir);
2323
2324         strbuf_setlen(path, baselen - 1);
2325         if (!r && subdir_cb)
2326                 r = subdir_cb(subdir_nr, path->buf, data);
2327
2328         strbuf_setlen(path, origlen);
2329
2330         return r;
2331 }
2332
2333 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2334                             each_loose_object_fn obj_cb,
2335                             each_loose_cruft_fn cruft_cb,
2336                             each_loose_subdir_fn subdir_cb,
2337                             void *data)
2338 {
2339         int r = 0;
2340         int i;
2341
2342         for (i = 0; i < 256; i++) {
2343                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2344                                                 subdir_cb, data);
2345                 if (r)
2346                         break;
2347         }
2348
2349         return r;
2350 }
2351
2352 int for_each_loose_file_in_objdir(const char *path,
2353                                   each_loose_object_fn obj_cb,
2354                                   each_loose_cruft_fn cruft_cb,
2355                                   each_loose_subdir_fn subdir_cb,
2356                                   void *data)
2357 {
2358         struct strbuf buf = STRBUF_INIT;
2359         int r;
2360
2361         strbuf_addstr(&buf, path);
2362         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2363                                               subdir_cb, data);
2364         strbuf_release(&buf);
2365
2366         return r;
2367 }
2368
2369 int for_each_loose_object(each_loose_object_fn cb, void *data,
2370                           enum for_each_object_flags flags)
2371 {
2372         struct object_directory *odb;
2373
2374         prepare_alt_odb(the_repository);
2375         for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2376                 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2377                                                       NULL, data);
2378                 if (r)
2379                         return r;
2380
2381                 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2382                         break;
2383         }
2384
2385         return 0;
2386 }
2387
2388 static int append_loose_object(const struct object_id *oid, const char *path,
2389                                void *data)
2390 {
2391         oid_array_append(data, oid);
2392         return 0;
2393 }
2394
2395 struct oid_array *odb_loose_cache(struct object_directory *odb,
2396                                   const struct object_id *oid)
2397 {
2398         int subdir_nr = oid->hash[0];
2399         struct strbuf buf = STRBUF_INIT;
2400
2401         if (subdir_nr < 0 ||
2402             subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2403                 BUG("subdir_nr out of range");
2404
2405         if (odb->loose_objects_subdir_seen[subdir_nr])
2406                 return &odb->loose_objects_cache[subdir_nr];
2407
2408         strbuf_addstr(&buf, odb->path);
2409         for_each_file_in_obj_subdir(subdir_nr, &buf,
2410                                     append_loose_object,
2411                                     NULL, NULL,
2412                                     &odb->loose_objects_cache[subdir_nr]);
2413         odb->loose_objects_subdir_seen[subdir_nr] = 1;
2414         strbuf_release(&buf);
2415         return &odb->loose_objects_cache[subdir_nr];
2416 }
2417
2418 void odb_clear_loose_cache(struct object_directory *odb)
2419 {
2420         int i;
2421
2422         for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2423                 oid_array_clear(&odb->loose_objects_cache[i]);
2424         memset(&odb->loose_objects_subdir_seen, 0,
2425                sizeof(odb->loose_objects_subdir_seen));
2426 }
2427
2428 static int check_stream_oid(git_zstream *stream,
2429                             const char *hdr,
2430                             unsigned long size,
2431                             const char *path,
2432                             const struct object_id *expected_oid)
2433 {
2434         git_hash_ctx c;
2435         struct object_id real_oid;
2436         unsigned char buf[4096];
2437         unsigned long total_read;
2438         int status = Z_OK;
2439
2440         the_hash_algo->init_fn(&c);
2441         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2442
2443         /*
2444          * We already read some bytes into hdr, but the ones up to the NUL
2445          * do not count against the object's content size.
2446          */
2447         total_read = stream->total_out - strlen(hdr) - 1;
2448
2449         /*
2450          * This size comparison must be "<=" to read the final zlib packets;
2451          * see the comment in unpack_loose_rest for details.
2452          */
2453         while (total_read <= size &&
2454                (status == Z_OK ||
2455                 (status == Z_BUF_ERROR && !stream->avail_out))) {
2456                 stream->next_out = buf;
2457                 stream->avail_out = sizeof(buf);
2458                 if (size - total_read < stream->avail_out)
2459                         stream->avail_out = size - total_read;
2460                 status = git_inflate(stream, Z_FINISH);
2461                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2462                 total_read += stream->next_out - buf;
2463         }
2464         git_inflate_end(stream);
2465
2466         if (status != Z_STREAM_END) {
2467                 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2468                 return -1;
2469         }
2470         if (stream->avail_in) {
2471                 error(_("garbage at end of loose object '%s'"),
2472                       oid_to_hex(expected_oid));
2473                 return -1;
2474         }
2475
2476         the_hash_algo->final_fn(real_oid.hash, &c);
2477         if (!oideq(expected_oid, &real_oid)) {
2478                 error(_("hash mismatch for %s (expected %s)"), path,
2479                       oid_to_hex(expected_oid));
2480                 return -1;
2481         }
2482
2483         return 0;
2484 }
2485
2486 int read_loose_object(const char *path,
2487                       const struct object_id *expected_oid,
2488                       enum object_type *type,
2489                       unsigned long *size,
2490                       void **contents)
2491 {
2492         int ret = -1;
2493         void *map = NULL;
2494         unsigned long mapsize;
2495         git_zstream stream;
2496         char hdr[MAX_HEADER_LEN];
2497
2498         *contents = NULL;
2499
2500         map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2501         if (!map) {
2502                 error_errno(_("unable to mmap %s"), path);
2503                 goto out;
2504         }
2505
2506         if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2507                 error(_("unable to unpack header of %s"), path);
2508                 goto out;
2509         }
2510
2511         *type = parse_loose_header(hdr, size);
2512         if (*type < 0) {
2513                 error(_("unable to parse header of %s"), path);
2514                 git_inflate_end(&stream);
2515                 goto out;
2516         }
2517
2518         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2519                 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2520                         goto out;
2521         } else {
2522                 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2523                 if (!*contents) {
2524                         error(_("unable to unpack contents of %s"), path);
2525                         git_inflate_end(&stream);
2526                         goto out;
2527                 }
2528                 if (check_object_signature(the_repository, expected_oid,
2529                                            *contents, *size,
2530                                            type_name(*type))) {
2531                         error(_("hash mismatch for %s (expected %s)"), path,
2532                               oid_to_hex(expected_oid));
2533                         free(*contents);
2534                         goto out;
2535                 }
2536         }
2537
2538         ret = 0; /* everything checks out */
2539
2540 out:
2541         if (map)
2542                 munmap(map, mapsize);
2543         return ret;
2544 }