word diff: handle zero length matches
[git] / object-file.c
1 /*
2  * GIT - The information manager from hell
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  *
6  * This handles basic git 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 "hash-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 static enum scld_error safe_create_leading_directories_1(char *path, int share)
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 (share && 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(char *path)
349 {
350         return safe_create_leading_directories_1(path, 1);
351 }
352
353 enum scld_error safe_create_leading_directories_no_share(char *path)
354 {
355         return safe_create_leading_directories_1(path, 0);
356 }
357
358 enum scld_error safe_create_leading_directories_const(const char *path)
359 {
360         int save_errno;
361         /* path points to cache entries, so xstrdup before messing with it */
362         char *buf = xstrdup(path);
363         enum scld_error result = safe_create_leading_directories(buf);
364
365         save_errno = errno;
366         free(buf);
367         errno = save_errno;
368         return result;
369 }
370
371 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
372 {
373         /*
374          * The number of times we will try to remove empty directories
375          * in the way of path. This is only 1 because if another
376          * process is racily creating directories that conflict with
377          * us, we don't want to fight against them.
378          */
379         int remove_directories_remaining = 1;
380
381         /*
382          * The number of times that we will try to create the
383          * directories containing path. We are willing to attempt this
384          * more than once, because another process could be trying to
385          * clean up empty directories at the same time as we are
386          * trying to create them.
387          */
388         int create_directories_remaining = 3;
389
390         /* A scratch copy of path, filled lazily if we need it: */
391         struct strbuf path_copy = STRBUF_INIT;
392
393         int ret, save_errno;
394
395         /* Sanity check: */
396         assert(*path);
397
398 retry_fn:
399         ret = fn(path, cb);
400         save_errno = errno;
401         if (!ret)
402                 goto out;
403
404         if (errno == EISDIR && remove_directories_remaining-- > 0) {
405                 /*
406                  * A directory is in the way. Maybe it is empty; try
407                  * to remove it:
408                  */
409                 if (!path_copy.len)
410                         strbuf_addstr(&path_copy, path);
411
412                 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
413                         goto retry_fn;
414         } else if (errno == ENOENT && create_directories_remaining-- > 0) {
415                 /*
416                  * Maybe the containing directory didn't exist, or
417                  * maybe it was just deleted by a process that is
418                  * racing with us to clean up empty directories. Try
419                  * to create it:
420                  */
421                 enum scld_error scld_result;
422
423                 if (!path_copy.len)
424                         strbuf_addstr(&path_copy, path);
425
426                 do {
427                         scld_result = safe_create_leading_directories(path_copy.buf);
428                         if (scld_result == SCLD_OK)
429                                 goto retry_fn;
430                 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
431         }
432
433 out:
434         strbuf_release(&path_copy);
435         errno = save_errno;
436         return ret;
437 }
438
439 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
440 {
441         int i;
442         for (i = 0; i < the_hash_algo->rawsz; i++) {
443                 static char hex[] = "0123456789abcdef";
444                 unsigned int val = oid->hash[i];
445                 strbuf_addch(buf, hex[val >> 4]);
446                 strbuf_addch(buf, hex[val & 0xf]);
447                 if (!i)
448                         strbuf_addch(buf, '/');
449         }
450 }
451
452 static const char *odb_loose_path(struct object_directory *odb,
453                                   struct strbuf *buf,
454                                   const struct object_id *oid)
455 {
456         strbuf_reset(buf);
457         strbuf_addstr(buf, odb->path);
458         strbuf_addch(buf, '/');
459         fill_loose_path(buf, oid);
460         return buf->buf;
461 }
462
463 const char *loose_object_path(struct repository *r, struct strbuf *buf,
464                               const struct object_id *oid)
465 {
466         return odb_loose_path(r->objects->odb, buf, oid);
467 }
468
469 /*
470  * Return non-zero iff the path is usable as an alternate object database.
471  */
472 static int alt_odb_usable(struct raw_object_store *o,
473                           struct strbuf *path,
474                           const char *normalized_objdir)
475 {
476         struct object_directory *odb;
477
478         /* Detect cases where alternate disappeared */
479         if (!is_directory(path->buf)) {
480                 error(_("object directory %s does not exist; "
481                         "check .git/objects/info/alternates"),
482                       path->buf);
483                 return 0;
484         }
485
486         /*
487          * Prevent the common mistake of listing the same
488          * thing twice, or object directory itself.
489          */
490         for (odb = o->odb; odb; odb = odb->next) {
491                 if (!fspathcmp(path->buf, odb->path))
492                         return 0;
493         }
494         if (!fspathcmp(path->buf, normalized_objdir))
495                 return 0;
496
497         return 1;
498 }
499
500 /*
501  * Prepare alternate object database registry.
502  *
503  * The variable alt_odb_list points at the list of struct
504  * object_directory.  The elements on this list come from
505  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
506  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
507  * whose contents is similar to that environment variable but can be
508  * LF separated.  Its base points at a statically allocated buffer that
509  * contains "/the/directory/corresponding/to/.git/objects/...", while
510  * its name points just after the slash at the end of ".git/objects/"
511  * in the example above, and has enough space to hold all hex characters
512  * of the object ID, an extra slash for the first level indirection, and
513  * the terminating NUL.
514  */
515 static void read_info_alternates(struct repository *r,
516                                  const char *relative_base,
517                                  int depth);
518 static int link_alt_odb_entry(struct repository *r, const char *entry,
519         const char *relative_base, int depth, const char *normalized_objdir)
520 {
521         struct object_directory *ent;
522         struct strbuf pathbuf = STRBUF_INIT;
523
524         if (!is_absolute_path(entry) && relative_base) {
525                 strbuf_realpath(&pathbuf, relative_base, 1);
526                 strbuf_addch(&pathbuf, '/');
527         }
528         strbuf_addstr(&pathbuf, entry);
529
530         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
531                 error(_("unable to normalize alternate object path: %s"),
532                       pathbuf.buf);
533                 strbuf_release(&pathbuf);
534                 return -1;
535         }
536
537         /*
538          * The trailing slash after the directory name is given by
539          * this function at the end. Remove duplicates.
540          */
541         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
542                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
543
544         if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
545                 strbuf_release(&pathbuf);
546                 return -1;
547         }
548
549         CALLOC_ARRAY(ent, 1);
550         ent->path = xstrdup(pathbuf.buf);
551
552         /* add the alternate entry */
553         *r->objects->odb_tail = ent;
554         r->objects->odb_tail = &(ent->next);
555         ent->next = NULL;
556
557         /* recursively add alternates */
558         read_info_alternates(r, pathbuf.buf, depth + 1);
559
560         strbuf_release(&pathbuf);
561         return 0;
562 }
563
564 static const char *parse_alt_odb_entry(const char *string,
565                                        int sep,
566                                        struct strbuf *out)
567 {
568         const char *end;
569
570         strbuf_reset(out);
571
572         if (*string == '#') {
573                 /* comment; consume up to next separator */
574                 end = strchrnul(string, sep);
575         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
576                 /*
577                  * quoted path; unquote_c_style has copied the
578                  * data for us and set "end". Broken quoting (e.g.,
579                  * an entry that doesn't end with a quote) falls
580                  * back to the unquoted case below.
581                  */
582         } else {
583                 /* normal, unquoted path */
584                 end = strchrnul(string, sep);
585                 strbuf_add(out, string, end - string);
586         }
587
588         if (*end)
589                 end++;
590         return end;
591 }
592
593 static void link_alt_odb_entries(struct repository *r, const char *alt,
594                                  int sep, const char *relative_base, int depth)
595 {
596         struct strbuf objdirbuf = STRBUF_INIT;
597         struct strbuf entry = STRBUF_INIT;
598
599         if (!alt || !*alt)
600                 return;
601
602         if (depth > 5) {
603                 error(_("%s: ignoring alternate object stores, nesting too deep"),
604                                 relative_base);
605                 return;
606         }
607
608         strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
609         if (strbuf_normalize_path(&objdirbuf) < 0)
610                 die(_("unable to normalize object directory: %s"),
611                     objdirbuf.buf);
612
613         while (*alt) {
614                 alt = parse_alt_odb_entry(alt, sep, &entry);
615                 if (!entry.len)
616                         continue;
617                 link_alt_odb_entry(r, entry.buf,
618                                    relative_base, depth, objdirbuf.buf);
619         }
620         strbuf_release(&entry);
621         strbuf_release(&objdirbuf);
622 }
623
624 static void read_info_alternates(struct repository *r,
625                                  const char *relative_base,
626                                  int depth)
627 {
628         char *path;
629         struct strbuf buf = STRBUF_INIT;
630
631         path = xstrfmt("%s/info/alternates", relative_base);
632         if (strbuf_read_file(&buf, path, 1024) < 0) {
633                 warn_on_fopen_errors(path);
634                 free(path);
635                 return;
636         }
637
638         link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
639         strbuf_release(&buf);
640         free(path);
641 }
642
643 void add_to_alternates_file(const char *reference)
644 {
645         struct lock_file lock = LOCK_INIT;
646         char *alts = git_pathdup("objects/info/alternates");
647         FILE *in, *out;
648         int found = 0;
649
650         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
651         out = fdopen_lock_file(&lock, "w");
652         if (!out)
653                 die_errno(_("unable to fdopen alternates lockfile"));
654
655         in = fopen(alts, "r");
656         if (in) {
657                 struct strbuf line = STRBUF_INIT;
658
659                 while (strbuf_getline(&line, in) != EOF) {
660                         if (!strcmp(reference, line.buf)) {
661                                 found = 1;
662                                 break;
663                         }
664                         fprintf_or_die(out, "%s\n", line.buf);
665                 }
666
667                 strbuf_release(&line);
668                 fclose(in);
669         }
670         else if (errno != ENOENT)
671                 die_errno(_("unable to read alternates file"));
672
673         if (found) {
674                 rollback_lock_file(&lock);
675         } else {
676                 fprintf_or_die(out, "%s\n", reference);
677                 if (commit_lock_file(&lock))
678                         die_errno(_("unable to move new alternates file into place"));
679                 if (the_repository->objects->loaded_alternates)
680                         link_alt_odb_entries(the_repository, reference,
681                                              '\n', NULL, 0);
682         }
683         free(alts);
684 }
685
686 void add_to_alternates_memory(const char *reference)
687 {
688         /*
689          * Make sure alternates are initialized, or else our entry may be
690          * overwritten when they are.
691          */
692         prepare_alt_odb(the_repository);
693
694         link_alt_odb_entries(the_repository, reference,
695                              '\n', NULL, 0);
696 }
697
698 /*
699  * Compute the exact path an alternate is at and returns it. In case of
700  * error NULL is returned and the human readable error is added to `err`
701  * `path` may be relative and should point to $GIT_DIR.
702  * `err` must not be null.
703  */
704 char *compute_alternate_path(const char *path, struct strbuf *err)
705 {
706         char *ref_git = NULL;
707         const char *repo;
708         int seen_error = 0;
709
710         ref_git = real_pathdup(path, 0);
711         if (!ref_git) {
712                 seen_error = 1;
713                 strbuf_addf(err, _("path '%s' does not exist"), path);
714                 goto out;
715         }
716
717         repo = read_gitfile(ref_git);
718         if (!repo)
719                 repo = read_gitfile(mkpath("%s/.git", ref_git));
720         if (repo) {
721                 free(ref_git);
722                 ref_git = xstrdup(repo);
723         }
724
725         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
726                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
727                 free(ref_git);
728                 ref_git = ref_git_git;
729         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
730                 struct strbuf sb = STRBUF_INIT;
731                 seen_error = 1;
732                 if (get_common_dir(&sb, ref_git)) {
733                         strbuf_addf(err,
734                                     _("reference repository '%s' as a linked "
735                                       "checkout is not supported yet."),
736                                     path);
737                         goto out;
738                 }
739
740                 strbuf_addf(err, _("reference repository '%s' is not a "
741                                         "local repository."), path);
742                 goto out;
743         }
744
745         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
746                 strbuf_addf(err, _("reference repository '%s' is shallow"),
747                             path);
748                 seen_error = 1;
749                 goto out;
750         }
751
752         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
753                 strbuf_addf(err,
754                             _("reference repository '%s' is grafted"),
755                             path);
756                 seen_error = 1;
757                 goto out;
758         }
759
760 out:
761         if (seen_error) {
762                 FREE_AND_NULL(ref_git);
763         }
764
765         return ref_git;
766 }
767
768 static void fill_alternate_refs_command(struct child_process *cmd,
769                                         const char *repo_path)
770 {
771         const char *value;
772
773         if (!git_config_get_value("core.alternateRefsCommand", &value)) {
774                 cmd->use_shell = 1;
775
776                 strvec_push(&cmd->args, value);
777                 strvec_push(&cmd->args, repo_path);
778         } else {
779                 cmd->git_cmd = 1;
780
781                 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
782                 strvec_push(&cmd->args, "for-each-ref");
783                 strvec_push(&cmd->args, "--format=%(objectname)");
784
785                 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
786                         strvec_push(&cmd->args, "--");
787                         strvec_split(&cmd->args, value);
788                 }
789         }
790
791         cmd->env = local_repo_env;
792         cmd->out = -1;
793 }
794
795 static void read_alternate_refs(const char *path,
796                                 alternate_ref_fn *cb,
797                                 void *data)
798 {
799         struct child_process cmd = CHILD_PROCESS_INIT;
800         struct strbuf line = STRBUF_INIT;
801         FILE *fh;
802
803         fill_alternate_refs_command(&cmd, path);
804
805         if (start_command(&cmd))
806                 return;
807
808         fh = xfdopen(cmd.out, "r");
809         while (strbuf_getline_lf(&line, fh) != EOF) {
810                 struct object_id oid;
811                 const char *p;
812
813                 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
814                         warning(_("invalid line while parsing alternate refs: %s"),
815                                 line.buf);
816                         break;
817                 }
818
819                 cb(&oid, data);
820         }
821
822         fclose(fh);
823         finish_command(&cmd);
824         strbuf_release(&line);
825 }
826
827 struct alternate_refs_data {
828         alternate_ref_fn *fn;
829         void *data;
830 };
831
832 static int refs_from_alternate_cb(struct object_directory *e,
833                                   void *data)
834 {
835         struct strbuf path = STRBUF_INIT;
836         size_t base_len;
837         struct alternate_refs_data *cb = data;
838
839         if (!strbuf_realpath(&path, e->path, 0))
840                 goto out;
841         if (!strbuf_strip_suffix(&path, "/objects"))
842                 goto out;
843         base_len = path.len;
844
845         /* Is this a git repository with refs? */
846         strbuf_addstr(&path, "/refs");
847         if (!is_directory(path.buf))
848                 goto out;
849         strbuf_setlen(&path, base_len);
850
851         read_alternate_refs(path.buf, cb->fn, cb->data);
852
853 out:
854         strbuf_release(&path);
855         return 0;
856 }
857
858 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
859 {
860         struct alternate_refs_data cb;
861         cb.fn = fn;
862         cb.data = data;
863         foreach_alt_odb(refs_from_alternate_cb, &cb);
864 }
865
866 int foreach_alt_odb(alt_odb_fn fn, void *cb)
867 {
868         struct object_directory *ent;
869         int r = 0;
870
871         prepare_alt_odb(the_repository);
872         for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
873                 r = fn(ent, cb);
874                 if (r)
875                         break;
876         }
877         return r;
878 }
879
880 void prepare_alt_odb(struct repository *r)
881 {
882         if (r->objects->loaded_alternates)
883                 return;
884
885         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
886
887         read_info_alternates(r, r->objects->odb->path, 0);
888         r->objects->loaded_alternates = 1;
889 }
890
891 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
892 static int freshen_file(const char *fn)
893 {
894         return !utime(fn, NULL);
895 }
896
897 /*
898  * All of the check_and_freshen functions return 1 if the file exists and was
899  * freshened (if freshening was requested), 0 otherwise. If they return
900  * 0, you should not assume that it is safe to skip a write of the object (it
901  * either does not exist on disk, or has a stale mtime and may be subject to
902  * pruning).
903  */
904 int check_and_freshen_file(const char *fn, int freshen)
905 {
906         if (access(fn, F_OK))
907                 return 0;
908         if (freshen && !freshen_file(fn))
909                 return 0;
910         return 1;
911 }
912
913 static int check_and_freshen_odb(struct object_directory *odb,
914                                  const struct object_id *oid,
915                                  int freshen)
916 {
917         static struct strbuf path = STRBUF_INIT;
918         odb_loose_path(odb, &path, oid);
919         return check_and_freshen_file(path.buf, freshen);
920 }
921
922 static int check_and_freshen_local(const struct object_id *oid, int freshen)
923 {
924         return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
925 }
926
927 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
928 {
929         struct object_directory *odb;
930
931         prepare_alt_odb(the_repository);
932         for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
933                 if (check_and_freshen_odb(odb, oid, freshen))
934                         return 1;
935         }
936         return 0;
937 }
938
939 static int check_and_freshen(const struct object_id *oid, int freshen)
940 {
941         return check_and_freshen_local(oid, freshen) ||
942                check_and_freshen_nonlocal(oid, freshen);
943 }
944
945 int has_loose_object_nonlocal(const struct object_id *oid)
946 {
947         return check_and_freshen_nonlocal(oid, 0);
948 }
949
950 static int has_loose_object(const struct object_id *oid)
951 {
952         return check_and_freshen(oid, 0);
953 }
954
955 static void mmap_limit_check(size_t length)
956 {
957         static size_t limit = 0;
958         if (!limit) {
959                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
960                 if (!limit)
961                         limit = SIZE_MAX;
962         }
963         if (length > limit)
964                 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
965                     (uintmax_t)length, (uintmax_t)limit);
966 }
967
968 void *xmmap_gently(void *start, size_t length,
969                   int prot, int flags, int fd, off_t offset)
970 {
971         void *ret;
972
973         mmap_limit_check(length);
974         ret = mmap(start, length, prot, flags, fd, offset);
975         if (ret == MAP_FAILED && !length)
976                 ret = NULL;
977         return ret;
978 }
979
980 void *xmmap(void *start, size_t length,
981         int prot, int flags, int fd, off_t offset)
982 {
983         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
984         if (ret == MAP_FAILED)
985                 die_errno(_("mmap failed"));
986         return ret;
987 }
988
989 /*
990  * With an in-core object data in "map", rehash it to make sure the
991  * object name actually matches "oid" to detect object corruption.
992  * With "map" == NULL, try reading the object named with "oid" using
993  * the streaming interface and rehash it to do the same.
994  */
995 int check_object_signature(struct repository *r, const struct object_id *oid,
996                            void *map, unsigned long size, const char *type)
997 {
998         struct object_id real_oid;
999         enum object_type obj_type;
1000         struct git_istream *st;
1001         git_hash_ctx c;
1002         char hdr[MAX_HEADER_LEN];
1003         int hdrlen;
1004
1005         if (map) {
1006                 hash_object_file(r->hash_algo, map, size, type, &real_oid);
1007                 return !oideq(oid, &real_oid) ? -1 : 0;
1008         }
1009
1010         st = open_istream(r, oid, &obj_type, &size, NULL);
1011         if (!st)
1012                 return -1;
1013
1014         /* Generate the header */
1015         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1016
1017         /* Sha1.. */
1018         r->hash_algo->init_fn(&c);
1019         r->hash_algo->update_fn(&c, hdr, hdrlen);
1020         for (;;) {
1021                 char buf[1024 * 16];
1022                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1023
1024                 if (readlen < 0) {
1025                         close_istream(st);
1026                         return -1;
1027                 }
1028                 if (!readlen)
1029                         break;
1030                 r->hash_algo->update_fn(&c, buf, readlen);
1031         }
1032         r->hash_algo->final_fn(real_oid.hash, &c);
1033         close_istream(st);
1034         return !oideq(oid, &real_oid) ? -1 : 0;
1035 }
1036
1037 int git_open_cloexec(const char *name, int flags)
1038 {
1039         int fd;
1040         static int o_cloexec = O_CLOEXEC;
1041
1042         fd = open(name, flags | o_cloexec);
1043         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1044                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1045                 o_cloexec &= ~O_CLOEXEC;
1046                 fd = open(name, flags | o_cloexec);
1047         }
1048
1049 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1050         {
1051                 static int fd_cloexec = FD_CLOEXEC;
1052
1053                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1054                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
1055                         int flags = fcntl(fd, F_GETFD);
1056                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1057                                 fd_cloexec = 0;
1058                 }
1059         }
1060 #endif
1061         return fd;
1062 }
1063
1064 /*
1065  * Find "oid" as a loose object in the local repository or in an alternate.
1066  * Returns 0 on success, negative on failure.
1067  *
1068  * The "path" out-parameter will give the path of the object we found (if any).
1069  * Note that it may point to static storage and is only valid until another
1070  * call to stat_loose_object().
1071  */
1072 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1073                              struct stat *st, const char **path)
1074 {
1075         struct object_directory *odb;
1076         static struct strbuf buf = STRBUF_INIT;
1077
1078         prepare_alt_odb(r);
1079         for (odb = r->objects->odb; odb; odb = odb->next) {
1080                 *path = odb_loose_path(odb, &buf, oid);
1081                 if (!lstat(*path, st))
1082                         return 0;
1083         }
1084
1085         return -1;
1086 }
1087
1088 /*
1089  * Like stat_loose_object(), but actually open the object and return the
1090  * descriptor. See the caveats on the "path" parameter above.
1091  */
1092 static int open_loose_object(struct repository *r,
1093                              const struct object_id *oid, const char **path)
1094 {
1095         int fd;
1096         struct object_directory *odb;
1097         int most_interesting_errno = ENOENT;
1098         static struct strbuf buf = STRBUF_INIT;
1099
1100         prepare_alt_odb(r);
1101         for (odb = r->objects->odb; odb; odb = odb->next) {
1102                 *path = odb_loose_path(odb, &buf, oid);
1103                 fd = git_open(*path);
1104                 if (fd >= 0)
1105                         return fd;
1106
1107                 if (most_interesting_errno == ENOENT)
1108                         most_interesting_errno = errno;
1109         }
1110         errno = most_interesting_errno;
1111         return -1;
1112 }
1113
1114 static int quick_has_loose(struct repository *r,
1115                            const struct object_id *oid)
1116 {
1117         struct object_directory *odb;
1118
1119         prepare_alt_odb(r);
1120         for (odb = r->objects->odb; odb; odb = odb->next) {
1121                 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1122                         return 1;
1123         }
1124         return 0;
1125 }
1126
1127 /*
1128  * Map the loose object at "path" if it is not NULL, or the path found by
1129  * searching for a loose object named "oid".
1130  */
1131 static void *map_loose_object_1(struct repository *r, const char *path,
1132                              const struct object_id *oid, unsigned long *size)
1133 {
1134         void *map;
1135         int fd;
1136
1137         if (path)
1138                 fd = git_open(path);
1139         else
1140                 fd = open_loose_object(r, oid, &path);
1141         map = NULL;
1142         if (fd >= 0) {
1143                 struct stat st;
1144
1145                 if (!fstat(fd, &st)) {
1146                         *size = xsize_t(st.st_size);
1147                         if (!*size) {
1148                                 /* mmap() is forbidden on empty files */
1149                                 error(_("object file %s is empty"), path);
1150                                 close(fd);
1151                                 return NULL;
1152                         }
1153                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1154                 }
1155                 close(fd);
1156         }
1157         return map;
1158 }
1159
1160 void *map_loose_object(struct repository *r,
1161                        const struct object_id *oid,
1162                        unsigned long *size)
1163 {
1164         return map_loose_object_1(r, NULL, oid, size);
1165 }
1166
1167 static int unpack_loose_short_header(git_zstream *stream,
1168                                      unsigned char *map, unsigned long mapsize,
1169                                      void *buffer, unsigned long bufsiz)
1170 {
1171         int ret;
1172
1173         /* Get the data stream */
1174         memset(stream, 0, sizeof(*stream));
1175         stream->next_in = map;
1176         stream->avail_in = mapsize;
1177         stream->next_out = buffer;
1178         stream->avail_out = bufsiz;
1179
1180         git_inflate_init(stream);
1181         obj_read_unlock();
1182         ret = git_inflate(stream, 0);
1183         obj_read_lock();
1184
1185         return ret;
1186 }
1187
1188 int unpack_loose_header(git_zstream *stream,
1189                         unsigned char *map, unsigned long mapsize,
1190                         void *buffer, unsigned long bufsiz)
1191 {
1192         int status = unpack_loose_short_header(stream, map, mapsize,
1193                                                buffer, bufsiz);
1194
1195         if (status < Z_OK)
1196                 return status;
1197
1198         /* Make sure we have the terminating NUL */
1199         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1200                 return -1;
1201         return 0;
1202 }
1203
1204 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1205                                          unsigned long mapsize, void *buffer,
1206                                          unsigned long bufsiz, struct strbuf *header)
1207 {
1208         int status;
1209
1210         status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1211         if (status < Z_OK)
1212                 return -1;
1213
1214         /*
1215          * Check if entire header is unpacked in the first iteration.
1216          */
1217         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1218                 return 0;
1219
1220         /*
1221          * buffer[0..bufsiz] was not large enough.  Copy the partial
1222          * result out to header, and then append the result of further
1223          * reading the stream.
1224          */
1225         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1226         stream->next_out = buffer;
1227         stream->avail_out = bufsiz;
1228
1229         do {
1230                 obj_read_unlock();
1231                 status = git_inflate(stream, 0);
1232                 obj_read_lock();
1233                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1234                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1235                         return 0;
1236                 stream->next_out = buffer;
1237                 stream->avail_out = bufsiz;
1238         } while (status != Z_STREAM_END);
1239         return -1;
1240 }
1241
1242 static void *unpack_loose_rest(git_zstream *stream,
1243                                void *buffer, unsigned long size,
1244                                const struct object_id *oid)
1245 {
1246         int bytes = strlen(buffer) + 1;
1247         unsigned char *buf = xmallocz(size);
1248         unsigned long n;
1249         int status = Z_OK;
1250
1251         n = stream->total_out - bytes;
1252         if (n > size)
1253                 n = size;
1254         memcpy(buf, (char *) buffer + bytes, n);
1255         bytes = n;
1256         if (bytes <= size) {
1257                 /*
1258                  * The above condition must be (bytes <= size), not
1259                  * (bytes < size).  In other words, even though we
1260                  * expect no more output and set avail_out to zero,
1261                  * the input zlib stream may have bytes that express
1262                  * "this concludes the stream", and we *do* want to
1263                  * eat that input.
1264                  *
1265                  * Otherwise we would not be able to test that we
1266                  * consumed all the input to reach the expected size;
1267                  * we also want to check that zlib tells us that all
1268                  * went well with status == Z_STREAM_END at the end.
1269                  */
1270                 stream->next_out = buf + bytes;
1271                 stream->avail_out = size - bytes;
1272                 while (status == Z_OK) {
1273                         obj_read_unlock();
1274                         status = git_inflate(stream, Z_FINISH);
1275                         obj_read_lock();
1276                 }
1277         }
1278         if (status == Z_STREAM_END && !stream->avail_in) {
1279                 git_inflate_end(stream);
1280                 return buf;
1281         }
1282
1283         if (status < 0)
1284                 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1285         else if (stream->avail_in)
1286                 error(_("garbage at end of loose object '%s'"),
1287                       oid_to_hex(oid));
1288         free(buf);
1289         return NULL;
1290 }
1291
1292 /*
1293  * We used to just use "sscanf()", but that's actually way
1294  * too permissive for what we want to check. So do an anal
1295  * object header parse by hand.
1296  */
1297 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1298                                        unsigned int flags)
1299 {
1300         const char *type_buf = hdr;
1301         unsigned long size;
1302         int type, type_len = 0;
1303
1304         /*
1305          * The type can be of any size but is followed by
1306          * a space.
1307          */
1308         for (;;) {
1309                 char c = *hdr++;
1310                 if (!c)
1311                         return -1;
1312                 if (c == ' ')
1313                         break;
1314                 type_len++;
1315         }
1316
1317         type = type_from_string_gently(type_buf, type_len, 1);
1318         if (oi->type_name)
1319                 strbuf_add(oi->type_name, type_buf, type_len);
1320         /*
1321          * Set type to 0 if its an unknown object and
1322          * we're obtaining the type using '--allow-unknown-type'
1323          * option.
1324          */
1325         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1326                 type = 0;
1327         else if (type < 0)
1328                 die(_("invalid object type"));
1329         if (oi->typep)
1330                 *oi->typep = type;
1331
1332         /*
1333          * The length must follow immediately, and be in canonical
1334          * decimal format (ie "010" is not valid).
1335          */
1336         size = *hdr++ - '0';
1337         if (size > 9)
1338                 return -1;
1339         if (size) {
1340                 for (;;) {
1341                         unsigned long c = *hdr - '0';
1342                         if (c > 9)
1343                                 break;
1344                         hdr++;
1345                         size = size * 10 + c;
1346                 }
1347         }
1348
1349         if (oi->sizep)
1350                 *oi->sizep = size;
1351
1352         /*
1353          * The length must be followed by a zero byte
1354          */
1355         return *hdr ? -1 : type;
1356 }
1357
1358 int parse_loose_header(const char *hdr, unsigned long *sizep)
1359 {
1360         struct object_info oi = OBJECT_INFO_INIT;
1361
1362         oi.sizep = sizep;
1363         return parse_loose_header_extended(hdr, &oi, 0);
1364 }
1365
1366 static int loose_object_info(struct repository *r,
1367                              const struct object_id *oid,
1368                              struct object_info *oi, int flags)
1369 {
1370         int status = 0;
1371         unsigned long mapsize;
1372         void *map;
1373         git_zstream stream;
1374         char hdr[MAX_HEADER_LEN];
1375         struct strbuf hdrbuf = STRBUF_INIT;
1376         unsigned long size_scratch;
1377
1378         if (oi->delta_base_oid)
1379                 oidclr(oi->delta_base_oid);
1380
1381         /*
1382          * If we don't care about type or size, then we don't
1383          * need to look inside the object at all. Note that we
1384          * do not optimize out the stat call, even if the
1385          * caller doesn't care about the disk-size, since our
1386          * return value implicitly indicates whether the
1387          * object even exists.
1388          */
1389         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1390                 const char *path;
1391                 struct stat st;
1392                 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1393                         return quick_has_loose(r, oid) ? 0 : -1;
1394                 if (stat_loose_object(r, oid, &st, &path) < 0)
1395                         return -1;
1396                 if (oi->disk_sizep)
1397                         *oi->disk_sizep = st.st_size;
1398                 return 0;
1399         }
1400
1401         map = map_loose_object(r, oid, &mapsize);
1402         if (!map)
1403                 return -1;
1404
1405         if (!oi->sizep)
1406                 oi->sizep = &size_scratch;
1407
1408         if (oi->disk_sizep)
1409                 *oi->disk_sizep = mapsize;
1410         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1411                 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1412                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1413                                        oid_to_hex(oid));
1414         } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1415                 status = error(_("unable to unpack %s header"),
1416                                oid_to_hex(oid));
1417         if (status < 0)
1418                 ; /* Do nothing */
1419         else if (hdrbuf.len) {
1420                 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1421                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1422                                        oid_to_hex(oid));
1423         } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1424                 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1425
1426         if (status >= 0 && oi->contentp) {
1427                 *oi->contentp = unpack_loose_rest(&stream, hdr,
1428                                                   *oi->sizep, oid);
1429                 if (!*oi->contentp) {
1430                         git_inflate_end(&stream);
1431                         status = -1;
1432                 }
1433         } else
1434                 git_inflate_end(&stream);
1435
1436         munmap(map, mapsize);
1437         if (status && oi->typep)
1438                 *oi->typep = status;
1439         if (oi->sizep == &size_scratch)
1440                 oi->sizep = NULL;
1441         strbuf_release(&hdrbuf);
1442         oi->whence = OI_LOOSE;
1443         return (status < 0) ? status : 0;
1444 }
1445
1446 int obj_read_use_lock = 0;
1447 pthread_mutex_t obj_read_mutex;
1448
1449 void enable_obj_read_lock(void)
1450 {
1451         if (obj_read_use_lock)
1452                 return;
1453
1454         obj_read_use_lock = 1;
1455         init_recursive_mutex(&obj_read_mutex);
1456 }
1457
1458 void disable_obj_read_lock(void)
1459 {
1460         if (!obj_read_use_lock)
1461                 return;
1462
1463         obj_read_use_lock = 0;
1464         pthread_mutex_destroy(&obj_read_mutex);
1465 }
1466
1467 int fetch_if_missing = 1;
1468
1469 static int do_oid_object_info_extended(struct repository *r,
1470                                        const struct object_id *oid,
1471                                        struct object_info *oi, unsigned flags)
1472 {
1473         static struct object_info blank_oi = OBJECT_INFO_INIT;
1474         struct cached_object *co;
1475         struct pack_entry e;
1476         int rtype;
1477         const struct object_id *real = oid;
1478         int already_retried = 0;
1479
1480
1481         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1482                 real = lookup_replace_object(r, oid);
1483
1484         if (is_null_oid(real))
1485                 return -1;
1486
1487         if (!oi)
1488                 oi = &blank_oi;
1489
1490         co = find_cached_object(real);
1491         if (co) {
1492                 if (oi->typep)
1493                         *(oi->typep) = co->type;
1494                 if (oi->sizep)
1495                         *(oi->sizep) = co->size;
1496                 if (oi->disk_sizep)
1497                         *(oi->disk_sizep) = 0;
1498                 if (oi->delta_base_oid)
1499                         oidclr(oi->delta_base_oid);
1500                 if (oi->type_name)
1501                         strbuf_addstr(oi->type_name, type_name(co->type));
1502                 if (oi->contentp)
1503                         *oi->contentp = xmemdupz(co->buf, co->size);
1504                 oi->whence = OI_CACHED;
1505                 return 0;
1506         }
1507
1508         while (1) {
1509                 if (find_pack_entry(r, real, &e))
1510                         break;
1511
1512                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1513                         return -1;
1514
1515                 /* Most likely it's a loose object. */
1516                 if (!loose_object_info(r, real, oi, flags))
1517                         return 0;
1518
1519                 /* Not a loose object; someone else may have just packed it. */
1520                 if (!(flags & OBJECT_INFO_QUICK)) {
1521                         reprepare_packed_git(r);
1522                         if (find_pack_entry(r, real, &e))
1523                                 break;
1524                 }
1525
1526                 /* Check if it is a missing object */
1527                 if (fetch_if_missing && has_promisor_remote() &&
1528                     !already_retried && r == the_repository &&
1529                     !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1530                         /*
1531                          * TODO Investigate checking promisor_remote_get_direct()
1532                          * TODO return value and stopping on error here.
1533                          * TODO Pass a repository struct through
1534                          * promisor_remote_get_direct(), such that arbitrary
1535                          * repositories work.
1536                          */
1537                         promisor_remote_get_direct(r, real, 1);
1538                         already_retried = 1;
1539                         continue;
1540                 }
1541
1542                 return -1;
1543         }
1544
1545         if (oi == &blank_oi)
1546                 /*
1547                  * We know that the caller doesn't actually need the
1548                  * information below, so return early.
1549                  */
1550                 return 0;
1551         rtype = packed_object_info(r, e.p, e.offset, oi);
1552         if (rtype < 0) {
1553                 mark_bad_packed_object(e.p, real->hash);
1554                 return do_oid_object_info_extended(r, real, oi, 0);
1555         } else if (oi->whence == OI_PACKED) {
1556                 oi->u.packed.offset = e.offset;
1557                 oi->u.packed.pack = e.p;
1558                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1559                                          rtype == OBJ_OFS_DELTA);
1560         }
1561
1562         return 0;
1563 }
1564
1565 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1566                              struct object_info *oi, unsigned flags)
1567 {
1568         int ret;
1569         obj_read_lock();
1570         ret = do_oid_object_info_extended(r, oid, oi, flags);
1571         obj_read_unlock();
1572         return ret;
1573 }
1574
1575
1576 /* returns enum object_type or negative */
1577 int oid_object_info(struct repository *r,
1578                     const struct object_id *oid,
1579                     unsigned long *sizep)
1580 {
1581         enum object_type type;
1582         struct object_info oi = OBJECT_INFO_INIT;
1583
1584         oi.typep = &type;
1585         oi.sizep = sizep;
1586         if (oid_object_info_extended(r, oid, &oi,
1587                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1588                 return -1;
1589         return type;
1590 }
1591
1592 static void *read_object(struct repository *r,
1593                          const struct object_id *oid, enum object_type *type,
1594                          unsigned long *size)
1595 {
1596         struct object_info oi = OBJECT_INFO_INIT;
1597         void *content;
1598         oi.typep = type;
1599         oi.sizep = size;
1600         oi.contentp = &content;
1601
1602         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1603                 return NULL;
1604         return content;
1605 }
1606
1607 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1608                         struct object_id *oid)
1609 {
1610         struct cached_object *co;
1611
1612         hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1613         if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1614             find_cached_object(oid))
1615                 return 0;
1616         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1617         co = &cached_objects[cached_object_nr++];
1618         co->size = len;
1619         co->type = type;
1620         co->buf = xmalloc(len);
1621         memcpy(co->buf, buf, len);
1622         oidcpy(&co->oid, oid);
1623         return 0;
1624 }
1625
1626 /*
1627  * This function dies on corrupt objects; the callers who want to
1628  * deal with them should arrange to call read_object() and give error
1629  * messages themselves.
1630  */
1631 void *read_object_file_extended(struct repository *r,
1632                                 const struct object_id *oid,
1633                                 enum object_type *type,
1634                                 unsigned long *size,
1635                                 int lookup_replace)
1636 {
1637         void *data;
1638         const struct packed_git *p;
1639         const char *path;
1640         struct stat st;
1641         const struct object_id *repl = lookup_replace ?
1642                 lookup_replace_object(r, oid) : oid;
1643
1644         errno = 0;
1645         data = read_object(r, repl, type, size);
1646         if (data)
1647                 return data;
1648
1649         obj_read_lock();
1650         if (errno && errno != ENOENT)
1651                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1652
1653         /* die if we replaced an object with one that does not exist */
1654         if (repl != oid)
1655                 die(_("replacement %s not found for %s"),
1656                     oid_to_hex(repl), oid_to_hex(oid));
1657
1658         if (!stat_loose_object(r, repl, &st, &path))
1659                 die(_("loose object %s (stored in %s) is corrupt"),
1660                     oid_to_hex(repl), path);
1661
1662         if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1663                 die(_("packed object %s (stored in %s) is corrupt"),
1664                     oid_to_hex(repl), p->pack_name);
1665         obj_read_unlock();
1666
1667         return NULL;
1668 }
1669
1670 void *read_object_with_reference(struct repository *r,
1671                                  const struct object_id *oid,
1672                                  const char *required_type_name,
1673                                  unsigned long *size,
1674                                  struct object_id *actual_oid_return)
1675 {
1676         enum object_type type, required_type;
1677         void *buffer;
1678         unsigned long isize;
1679         struct object_id actual_oid;
1680
1681         required_type = type_from_string(required_type_name);
1682         oidcpy(&actual_oid, oid);
1683         while (1) {
1684                 int ref_length = -1;
1685                 const char *ref_type = NULL;
1686
1687                 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1688                 if (!buffer)
1689                         return NULL;
1690                 if (type == required_type) {
1691                         *size = isize;
1692                         if (actual_oid_return)
1693                                 oidcpy(actual_oid_return, &actual_oid);
1694                         return buffer;
1695                 }
1696                 /* Handle references */
1697                 else if (type == OBJ_COMMIT)
1698                         ref_type = "tree ";
1699                 else if (type == OBJ_TAG)
1700                         ref_type = "object ";
1701                 else {
1702                         free(buffer);
1703                         return NULL;
1704                 }
1705                 ref_length = strlen(ref_type);
1706
1707                 if (ref_length + the_hash_algo->hexsz > isize ||
1708                     memcmp(buffer, ref_type, ref_length) ||
1709                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1710                         free(buffer);
1711                         return NULL;
1712                 }
1713                 free(buffer);
1714                 /* Now we have the ID of the referred-to object in
1715                  * actual_oid.  Check again. */
1716         }
1717 }
1718
1719 static void write_object_file_prepare(const struct git_hash_algo *algo,
1720                                       const void *buf, unsigned long len,
1721                                       const char *type, struct object_id *oid,
1722                                       char *hdr, int *hdrlen)
1723 {
1724         git_hash_ctx c;
1725
1726         /* Generate the header */
1727         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1728
1729         /* Sha1.. */
1730         algo->init_fn(&c);
1731         algo->update_fn(&c, hdr, *hdrlen);
1732         algo->update_fn(&c, buf, len);
1733         algo->final_fn(oid->hash, &c);
1734 }
1735
1736 /*
1737  * Move the just written object into its final resting place.
1738  */
1739 int finalize_object_file(const char *tmpfile, const char *filename)
1740 {
1741         int ret = 0;
1742
1743         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1744                 goto try_rename;
1745         else if (link(tmpfile, filename))
1746                 ret = errno;
1747
1748         /*
1749          * Coda hack - coda doesn't like cross-directory links,
1750          * so we fall back to a rename, which will mean that it
1751          * won't be able to check collisions, but that's not a
1752          * big deal.
1753          *
1754          * The same holds for FAT formatted media.
1755          *
1756          * When this succeeds, we just return.  We have nothing
1757          * left to unlink.
1758          */
1759         if (ret && ret != EEXIST) {
1760         try_rename:
1761                 if (!rename(tmpfile, filename))
1762                         goto out;
1763                 ret = errno;
1764         }
1765         unlink_or_warn(tmpfile);
1766         if (ret) {
1767                 if (ret != EEXIST) {
1768                         return error_errno(_("unable to write file %s"), filename);
1769                 }
1770                 /* FIXME!!! Collision check here ? */
1771         }
1772
1773 out:
1774         if (adjust_shared_perm(filename))
1775                 return error(_("unable to set permission to '%s'"), filename);
1776         return 0;
1777 }
1778
1779 static int write_buffer(int fd, const void *buf, size_t len)
1780 {
1781         if (write_in_full(fd, buf, len) < 0)
1782                 return error_errno(_("file write error"));
1783         return 0;
1784 }
1785
1786 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1787                      unsigned long len, const char *type,
1788                      struct object_id *oid)
1789 {
1790         char hdr[MAX_HEADER_LEN];
1791         int hdrlen = sizeof(hdr);
1792         write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1793         return 0;
1794 }
1795
1796 /* Finalize a file on disk, and close it. */
1797 static void close_loose_object(int fd)
1798 {
1799         if (fsync_object_files)
1800                 fsync_or_die(fd, "loose object file");
1801         if (close(fd) != 0)
1802                 die_errno(_("error when closing loose object file"));
1803 }
1804
1805 /* Size of directory component, including the ending '/' */
1806 static inline int directory_size(const char *filename)
1807 {
1808         const char *s = strrchr(filename, '/');
1809         if (!s)
1810                 return 0;
1811         return s - filename + 1;
1812 }
1813
1814 /*
1815  * This creates a temporary file in the same directory as the final
1816  * 'filename'
1817  *
1818  * We want to avoid cross-directory filename renames, because those
1819  * can have problems on various filesystems (FAT, NFS, Coda).
1820  */
1821 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1822 {
1823         int fd, dirlen = directory_size(filename);
1824
1825         strbuf_reset(tmp);
1826         strbuf_add(tmp, filename, dirlen);
1827         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1828         fd = git_mkstemp_mode(tmp->buf, 0444);
1829         if (fd < 0 && dirlen && errno == ENOENT) {
1830                 /*
1831                  * Make sure the directory exists; note that the contents
1832                  * of the buffer are undefined after mkstemp returns an
1833                  * error, so we have to rewrite the whole buffer from
1834                  * scratch.
1835                  */
1836                 strbuf_reset(tmp);
1837                 strbuf_add(tmp, filename, dirlen - 1);
1838                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1839                         return -1;
1840                 if (adjust_shared_perm(tmp->buf))
1841                         return -1;
1842
1843                 /* Try again */
1844                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1845                 fd = git_mkstemp_mode(tmp->buf, 0444);
1846         }
1847         return fd;
1848 }
1849
1850 static int write_loose_object(const struct object_id *oid, char *hdr,
1851                               int hdrlen, const void *buf, unsigned long len,
1852                               time_t mtime)
1853 {
1854         int fd, ret;
1855         unsigned char compressed[4096];
1856         git_zstream stream;
1857         git_hash_ctx c;
1858         struct object_id parano_oid;
1859         static struct strbuf tmp_file = STRBUF_INIT;
1860         static struct strbuf filename = STRBUF_INIT;
1861
1862         loose_object_path(the_repository, &filename, oid);
1863
1864         fd = create_tmpfile(&tmp_file, filename.buf);
1865         if (fd < 0) {
1866                 if (errno == EACCES)
1867                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1868                 else
1869                         return error_errno(_("unable to create temporary file"));
1870         }
1871
1872         /* Set it up */
1873         git_deflate_init(&stream, zlib_compression_level);
1874         stream.next_out = compressed;
1875         stream.avail_out = sizeof(compressed);
1876         the_hash_algo->init_fn(&c);
1877
1878         /* First header.. */
1879         stream.next_in = (unsigned char *)hdr;
1880         stream.avail_in = hdrlen;
1881         while (git_deflate(&stream, 0) == Z_OK)
1882                 ; /* nothing */
1883         the_hash_algo->update_fn(&c, hdr, hdrlen);
1884
1885         /* Then the data itself.. */
1886         stream.next_in = (void *)buf;
1887         stream.avail_in = len;
1888         do {
1889                 unsigned char *in0 = stream.next_in;
1890                 ret = git_deflate(&stream, Z_FINISH);
1891                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1892                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1893                         die(_("unable to write loose object file"));
1894                 stream.next_out = compressed;
1895                 stream.avail_out = sizeof(compressed);
1896         } while (ret == Z_OK);
1897
1898         if (ret != Z_STREAM_END)
1899                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1900                     ret);
1901         ret = git_deflate_end_gently(&stream);
1902         if (ret != Z_OK)
1903                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1904                     ret);
1905         the_hash_algo->final_fn(parano_oid.hash, &c);
1906         if (!oideq(oid, &parano_oid))
1907                 die(_("confused by unstable object source data for %s"),
1908                     oid_to_hex(oid));
1909
1910         close_loose_object(fd);
1911
1912         if (mtime) {
1913                 struct utimbuf utb;
1914                 utb.actime = mtime;
1915                 utb.modtime = mtime;
1916                 if (utime(tmp_file.buf, &utb) < 0)
1917                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1918         }
1919
1920         return finalize_object_file(tmp_file.buf, filename.buf);
1921 }
1922
1923 static int freshen_loose_object(const struct object_id *oid)
1924 {
1925         return check_and_freshen(oid, 1);
1926 }
1927
1928 static int freshen_packed_object(const struct object_id *oid)
1929 {
1930         struct pack_entry e;
1931         if (!find_pack_entry(the_repository, oid, &e))
1932                 return 0;
1933         if (e.p->freshened)
1934                 return 1;
1935         if (!freshen_file(e.p->pack_name))
1936                 return 0;
1937         e.p->freshened = 1;
1938         return 1;
1939 }
1940
1941 int write_object_file(const void *buf, unsigned long len, const char *type,
1942                       struct object_id *oid)
1943 {
1944         char hdr[MAX_HEADER_LEN];
1945         int hdrlen = sizeof(hdr);
1946
1947         /* Normally if we have it in the pack then we do not bother writing
1948          * it out into .git/objects/??/?{38} file.
1949          */
1950         write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1951                                   &hdrlen);
1952         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1953                 return 0;
1954         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1955 }
1956
1957 int hash_object_file_literally(const void *buf, unsigned long len,
1958                                const char *type, struct object_id *oid,
1959                                unsigned flags)
1960 {
1961         char *header;
1962         int hdrlen, status = 0;
1963
1964         /* type string, SP, %lu of the length plus NUL must fit this */
1965         hdrlen = strlen(type) + MAX_HEADER_LEN;
1966         header = xmalloc(hdrlen);
1967         write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
1968                                   &hdrlen);
1969
1970         if (!(flags & HASH_WRITE_OBJECT))
1971                 goto cleanup;
1972         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1973                 goto cleanup;
1974         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1975
1976 cleanup:
1977         free(header);
1978         return status;
1979 }
1980
1981 int force_object_loose(const struct object_id *oid, time_t mtime)
1982 {
1983         void *buf;
1984         unsigned long len;
1985         enum object_type type;
1986         char hdr[MAX_HEADER_LEN];
1987         int hdrlen;
1988         int ret;
1989
1990         if (has_loose_object(oid))
1991                 return 0;
1992         buf = read_object(the_repository, oid, &type, &len);
1993         if (!buf)
1994                 return error(_("cannot read object for %s"), oid_to_hex(oid));
1995         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
1996         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1997         free(buf);
1998
1999         return ret;
2000 }
2001
2002 int has_object(struct repository *r, const struct object_id *oid,
2003                unsigned flags)
2004 {
2005         int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2006         unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2007                 (quick ? OBJECT_INFO_QUICK : 0);
2008
2009         if (!startup_info->have_repository)
2010                 return 0;
2011         return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2012 }
2013
2014 int repo_has_object_file_with_flags(struct repository *r,
2015                                     const struct object_id *oid, int flags)
2016 {
2017         if (!startup_info->have_repository)
2018                 return 0;
2019         return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2020 }
2021
2022 int repo_has_object_file(struct repository *r,
2023                          const struct object_id *oid)
2024 {
2025         return repo_has_object_file_with_flags(r, oid, 0);
2026 }
2027
2028 static void check_tree(const void *buf, size_t size)
2029 {
2030         struct tree_desc desc;
2031         struct name_entry entry;
2032
2033         init_tree_desc(&desc, buf, size);
2034         while (tree_entry(&desc, &entry))
2035                 /* do nothing
2036                  * tree_entry() will die() on malformed entries */
2037                 ;
2038 }
2039
2040 static void check_commit(const void *buf, size_t size)
2041 {
2042         struct commit c;
2043         memset(&c, 0, sizeof(c));
2044         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2045                 die(_("corrupt commit"));
2046 }
2047
2048 static void check_tag(const void *buf, size_t size)
2049 {
2050         struct tag t;
2051         memset(&t, 0, sizeof(t));
2052         if (parse_tag_buffer(the_repository, &t, buf, size))
2053                 die(_("corrupt tag"));
2054 }
2055
2056 static int index_mem(struct index_state *istate,
2057                      struct object_id *oid, void *buf, size_t size,
2058                      enum object_type type,
2059                      const char *path, unsigned flags)
2060 {
2061         int ret, re_allocated = 0;
2062         int write_object = flags & HASH_WRITE_OBJECT;
2063
2064         if (!type)
2065                 type = OBJ_BLOB;
2066
2067         /*
2068          * Convert blobs to git internal format
2069          */
2070         if ((type == OBJ_BLOB) && path) {
2071                 struct strbuf nbuf = STRBUF_INIT;
2072                 if (convert_to_git(istate, path, buf, size, &nbuf,
2073                                    get_conv_flags(flags))) {
2074                         buf = strbuf_detach(&nbuf, &size);
2075                         re_allocated = 1;
2076                 }
2077         }
2078         if (flags & HASH_FORMAT_CHECK) {
2079                 if (type == OBJ_TREE)
2080                         check_tree(buf, size);
2081                 if (type == OBJ_COMMIT)
2082                         check_commit(buf, size);
2083                 if (type == OBJ_TAG)
2084                         check_tag(buf, size);
2085         }
2086
2087         if (write_object)
2088                 ret = write_object_file(buf, size, type_name(type), oid);
2089         else
2090                 ret = hash_object_file(the_hash_algo, buf, size,
2091                                        type_name(type), oid);
2092         if (re_allocated)
2093                 free(buf);
2094         return ret;
2095 }
2096
2097 static int index_stream_convert_blob(struct index_state *istate,
2098                                      struct object_id *oid,
2099                                      int fd,
2100                                      const char *path,
2101                                      unsigned flags)
2102 {
2103         int ret;
2104         const int write_object = flags & HASH_WRITE_OBJECT;
2105         struct strbuf sbuf = STRBUF_INIT;
2106
2107         assert(path);
2108         assert(would_convert_to_git_filter_fd(istate, path));
2109
2110         convert_to_git_filter_fd(istate, path, fd, &sbuf,
2111                                  get_conv_flags(flags));
2112
2113         if (write_object)
2114                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2115                                         oid);
2116         else
2117                 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2118                                        type_name(OBJ_BLOB), oid);
2119         strbuf_release(&sbuf);
2120         return ret;
2121 }
2122
2123 static int index_pipe(struct index_state *istate, struct object_id *oid,
2124                       int fd, enum object_type type,
2125                       const char *path, unsigned flags)
2126 {
2127         struct strbuf sbuf = STRBUF_INIT;
2128         int ret;
2129
2130         if (strbuf_read(&sbuf, fd, 4096) >= 0)
2131                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2132         else
2133                 ret = -1;
2134         strbuf_release(&sbuf);
2135         return ret;
2136 }
2137
2138 #define SMALL_FILE_SIZE (32*1024)
2139
2140 static int index_core(struct index_state *istate,
2141                       struct object_id *oid, int fd, size_t size,
2142                       enum object_type type, const char *path,
2143                       unsigned flags)
2144 {
2145         int ret;
2146
2147         if (!size) {
2148                 ret = index_mem(istate, oid, "", size, type, path, flags);
2149         } else if (size <= SMALL_FILE_SIZE) {
2150                 char *buf = xmalloc(size);
2151                 ssize_t read_result = read_in_full(fd, buf, size);
2152                 if (read_result < 0)
2153                         ret = error_errno(_("read error while indexing %s"),
2154                                           path ? path : "<unknown>");
2155                 else if (read_result != size)
2156                         ret = error(_("short read while indexing %s"),
2157                                     path ? path : "<unknown>");
2158                 else
2159                         ret = index_mem(istate, oid, buf, size, type, path, flags);
2160                 free(buf);
2161         } else {
2162                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2163                 ret = index_mem(istate, oid, buf, size, type, path, flags);
2164                 munmap(buf, size);
2165         }
2166         return ret;
2167 }
2168
2169 /*
2170  * This creates one packfile per large blob unless bulk-checkin
2171  * machinery is "plugged".
2172  *
2173  * This also bypasses the usual "convert-to-git" dance, and that is on
2174  * purpose. We could write a streaming version of the converting
2175  * functions and insert that before feeding the data to fast-import
2176  * (or equivalent in-core API described above). However, that is
2177  * somewhat complicated, as we do not know the size of the filter
2178  * result, which we need to know beforehand when writing a git object.
2179  * Since the primary motivation for trying to stream from the working
2180  * tree file and to avoid mmaping it in core is to deal with large
2181  * binary blobs, they generally do not want to get any conversion, and
2182  * callers should avoid this code path when filters are requested.
2183  */
2184 static int index_stream(struct object_id *oid, int fd, size_t size,
2185                         enum object_type type, const char *path,
2186                         unsigned flags)
2187 {
2188         return index_bulk_checkin(oid, fd, size, type, path, flags);
2189 }
2190
2191 int index_fd(struct index_state *istate, struct object_id *oid,
2192              int fd, struct stat *st,
2193              enum object_type type, const char *path, unsigned flags)
2194 {
2195         int ret;
2196
2197         /*
2198          * Call xsize_t() only when needed to avoid potentially unnecessary
2199          * die() for large files.
2200          */
2201         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2202                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2203         else if (!S_ISREG(st->st_mode))
2204                 ret = index_pipe(istate, oid, fd, type, path, flags);
2205         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2206                  (path && would_convert_to_git(istate, path)))
2207                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2208                                  type, path, flags);
2209         else
2210                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2211                                    flags);
2212         close(fd);
2213         return ret;
2214 }
2215
2216 int index_path(struct index_state *istate, struct object_id *oid,
2217                const char *path, struct stat *st, unsigned flags)
2218 {
2219         int fd;
2220         struct strbuf sb = STRBUF_INIT;
2221         int rc = 0;
2222
2223         switch (st->st_mode & S_IFMT) {
2224         case S_IFREG:
2225                 fd = open(path, O_RDONLY);
2226                 if (fd < 0)
2227                         return error_errno("open(\"%s\")", path);
2228                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2229                         return error(_("%s: failed to insert into database"),
2230                                      path);
2231                 break;
2232         case S_IFLNK:
2233                 if (strbuf_readlink(&sb, path, st->st_size))
2234                         return error_errno("readlink(\"%s\")", path);
2235                 if (!(flags & HASH_WRITE_OBJECT))
2236                         hash_object_file(the_hash_algo, sb.buf, sb.len,
2237                                          blob_type, oid);
2238                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2239                         rc = error(_("%s: failed to insert into database"), path);
2240                 strbuf_release(&sb);
2241                 break;
2242         case S_IFDIR:
2243                 return resolve_gitlink_ref(path, "HEAD", oid);
2244         default:
2245                 return error(_("%s: unsupported file type"), path);
2246         }
2247         return rc;
2248 }
2249
2250 int read_pack_header(int fd, struct pack_header *header)
2251 {
2252         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2253                 /* "eof before pack header was fully read" */
2254                 return PH_ERROR_EOF;
2255
2256         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2257                 /* "protocol error (pack signature mismatch detected)" */
2258                 return PH_ERROR_PACK_SIGNATURE;
2259         if (!pack_version_ok(header->hdr_version))
2260                 /* "protocol error (pack version unsupported)" */
2261                 return PH_ERROR_PROTOCOL;
2262         return 0;
2263 }
2264
2265 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2266 {
2267         enum object_type type = oid_object_info(the_repository, oid, NULL);
2268         if (type < 0)
2269                 die(_("%s is not a valid object"), oid_to_hex(oid));
2270         if (type != expect)
2271                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2272                     type_name(expect));
2273 }
2274
2275 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2276                                 struct strbuf *path,
2277                                 each_loose_object_fn obj_cb,
2278                                 each_loose_cruft_fn cruft_cb,
2279                                 each_loose_subdir_fn subdir_cb,
2280                                 void *data)
2281 {
2282         size_t origlen, baselen;
2283         DIR *dir;
2284         struct dirent *de;
2285         int r = 0;
2286         struct object_id oid;
2287
2288         if (subdir_nr > 0xff)
2289                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2290
2291         origlen = path->len;
2292         strbuf_complete(path, '/');
2293         strbuf_addf(path, "%02x", subdir_nr);
2294
2295         dir = opendir(path->buf);
2296         if (!dir) {
2297                 if (errno != ENOENT)
2298                         r = error_errno(_("unable to open %s"), path->buf);
2299                 strbuf_setlen(path, origlen);
2300                 return r;
2301         }
2302
2303         oid.hash[0] = subdir_nr;
2304         strbuf_addch(path, '/');
2305         baselen = path->len;
2306
2307         while ((de = readdir(dir))) {
2308                 size_t namelen;
2309                 if (is_dot_or_dotdot(de->d_name))
2310                         continue;
2311
2312                 namelen = strlen(de->d_name);
2313                 strbuf_setlen(path, baselen);
2314                 strbuf_add(path, de->d_name, namelen);
2315                 if (namelen == the_hash_algo->hexsz - 2 &&
2316                     !hex_to_bytes(oid.hash + 1, de->d_name,
2317                                   the_hash_algo->rawsz - 1)) {
2318                         if (obj_cb) {
2319                                 r = obj_cb(&oid, path->buf, data);
2320                                 if (r)
2321                                         break;
2322                         }
2323                         continue;
2324                 }
2325
2326                 if (cruft_cb) {
2327                         r = cruft_cb(de->d_name, path->buf, data);
2328                         if (r)
2329                                 break;
2330                 }
2331         }
2332         closedir(dir);
2333
2334         strbuf_setlen(path, baselen - 1);
2335         if (!r && subdir_cb)
2336                 r = subdir_cb(subdir_nr, path->buf, data);
2337
2338         strbuf_setlen(path, origlen);
2339
2340         return r;
2341 }
2342
2343 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2344                             each_loose_object_fn obj_cb,
2345                             each_loose_cruft_fn cruft_cb,
2346                             each_loose_subdir_fn subdir_cb,
2347                             void *data)
2348 {
2349         int r = 0;
2350         int i;
2351
2352         for (i = 0; i < 256; i++) {
2353                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2354                                                 subdir_cb, data);
2355                 if (r)
2356                         break;
2357         }
2358
2359         return r;
2360 }
2361
2362 int for_each_loose_file_in_objdir(const char *path,
2363                                   each_loose_object_fn obj_cb,
2364                                   each_loose_cruft_fn cruft_cb,
2365                                   each_loose_subdir_fn subdir_cb,
2366                                   void *data)
2367 {
2368         struct strbuf buf = STRBUF_INIT;
2369         int r;
2370
2371         strbuf_addstr(&buf, path);
2372         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2373                                               subdir_cb, data);
2374         strbuf_release(&buf);
2375
2376         return r;
2377 }
2378
2379 int for_each_loose_object(each_loose_object_fn cb, void *data,
2380                           enum for_each_object_flags flags)
2381 {
2382         struct object_directory *odb;
2383
2384         prepare_alt_odb(the_repository);
2385         for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2386                 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2387                                                       NULL, data);
2388                 if (r)
2389                         return r;
2390
2391                 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2392                         break;
2393         }
2394
2395         return 0;
2396 }
2397
2398 static int append_loose_object(const struct object_id *oid, const char *path,
2399                                void *data)
2400 {
2401         oid_array_append(data, oid);
2402         return 0;
2403 }
2404
2405 struct oid_array *odb_loose_cache(struct object_directory *odb,
2406                                   const struct object_id *oid)
2407 {
2408         int subdir_nr = oid->hash[0];
2409         struct strbuf buf = STRBUF_INIT;
2410
2411         if (subdir_nr < 0 ||
2412             subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2413                 BUG("subdir_nr out of range");
2414
2415         if (odb->loose_objects_subdir_seen[subdir_nr])
2416                 return &odb->loose_objects_cache[subdir_nr];
2417
2418         strbuf_addstr(&buf, odb->path);
2419         for_each_file_in_obj_subdir(subdir_nr, &buf,
2420                                     append_loose_object,
2421                                     NULL, NULL,
2422                                     &odb->loose_objects_cache[subdir_nr]);
2423         odb->loose_objects_subdir_seen[subdir_nr] = 1;
2424         strbuf_release(&buf);
2425         return &odb->loose_objects_cache[subdir_nr];
2426 }
2427
2428 void odb_clear_loose_cache(struct object_directory *odb)
2429 {
2430         int i;
2431
2432         for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2433                 oid_array_clear(&odb->loose_objects_cache[i]);
2434         memset(&odb->loose_objects_subdir_seen, 0,
2435                sizeof(odb->loose_objects_subdir_seen));
2436 }
2437
2438 static int check_stream_oid(git_zstream *stream,
2439                             const char *hdr,
2440                             unsigned long size,
2441                             const char *path,
2442                             const struct object_id *expected_oid)
2443 {
2444         git_hash_ctx c;
2445         struct object_id real_oid;
2446         unsigned char buf[4096];
2447         unsigned long total_read;
2448         int status = Z_OK;
2449
2450         the_hash_algo->init_fn(&c);
2451         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2452
2453         /*
2454          * We already read some bytes into hdr, but the ones up to the NUL
2455          * do not count against the object's content size.
2456          */
2457         total_read = stream->total_out - strlen(hdr) - 1;
2458
2459         /*
2460          * This size comparison must be "<=" to read the final zlib packets;
2461          * see the comment in unpack_loose_rest for details.
2462          */
2463         while (total_read <= size &&
2464                (status == Z_OK ||
2465                 (status == Z_BUF_ERROR && !stream->avail_out))) {
2466                 stream->next_out = buf;
2467                 stream->avail_out = sizeof(buf);
2468                 if (size - total_read < stream->avail_out)
2469                         stream->avail_out = size - total_read;
2470                 status = git_inflate(stream, Z_FINISH);
2471                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2472                 total_read += stream->next_out - buf;
2473         }
2474         git_inflate_end(stream);
2475
2476         if (status != Z_STREAM_END) {
2477                 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2478                 return -1;
2479         }
2480         if (stream->avail_in) {
2481                 error(_("garbage at end of loose object '%s'"),
2482                       oid_to_hex(expected_oid));
2483                 return -1;
2484         }
2485
2486         the_hash_algo->final_fn(real_oid.hash, &c);
2487         if (!oideq(expected_oid, &real_oid)) {
2488                 error(_("hash mismatch for %s (expected %s)"), path,
2489                       oid_to_hex(expected_oid));
2490                 return -1;
2491         }
2492
2493         return 0;
2494 }
2495
2496 int read_loose_object(const char *path,
2497                       const struct object_id *expected_oid,
2498                       enum object_type *type,
2499                       unsigned long *size,
2500                       void **contents)
2501 {
2502         int ret = -1;
2503         void *map = NULL;
2504         unsigned long mapsize;
2505         git_zstream stream;
2506         char hdr[MAX_HEADER_LEN];
2507
2508         *contents = NULL;
2509
2510         map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2511         if (!map) {
2512                 error_errno(_("unable to mmap %s"), path);
2513                 goto out;
2514         }
2515
2516         if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2517                 error(_("unable to unpack header of %s"), path);
2518                 goto out;
2519         }
2520
2521         *type = parse_loose_header(hdr, size);
2522         if (*type < 0) {
2523                 error(_("unable to parse header of %s"), path);
2524                 git_inflate_end(&stream);
2525                 goto out;
2526         }
2527
2528         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2529                 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2530                         goto out;
2531         } else {
2532                 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2533                 if (!*contents) {
2534                         error(_("unable to unpack contents of %s"), path);
2535                         git_inflate_end(&stream);
2536                         goto out;
2537                 }
2538                 if (check_object_signature(the_repository, expected_oid,
2539                                            *contents, *size,
2540                                            type_name(*type))) {
2541                         error(_("hash mismatch for %s (expected %s)"), path,
2542                               oid_to_hex(expected_oid));
2543                         free(*contents);
2544                         goto out;
2545                 }
2546         }
2547
2548         ret = 0; /* everything checks out */
2549
2550 out:
2551         if (map)
2552                 munmap(map, mapsize);
2553         return ret;
2554 }