Merge branch 'ab/config-based-hooks-base' into seen
[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 static const struct object_id empty_tree_oid = {
59         .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
60         .algo = GIT_HASH_SHA1,
61 };
62 static const struct object_id empty_blob_oid = {
63         .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
64         .algo = GIT_HASH_SHA1,
65 };
66 static const struct object_id null_oid_sha1 = {
67         .hash = {0},
68         .algo = GIT_HASH_SHA1,
69 };
70 static const struct object_id empty_tree_oid_sha256 = {
71         .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
72         .algo = GIT_HASH_SHA256,
73 };
74 static const struct object_id empty_blob_oid_sha256 = {
75         .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
76         .algo = GIT_HASH_SHA256,
77 };
78 static const struct object_id null_oid_sha256 = {
79         .hash = {0},
80         .algo = GIT_HASH_SHA256,
81 };
82
83 static void git_hash_sha1_init(git_hash_ctx *ctx)
84 {
85         git_SHA1_Init(&ctx->sha1);
86 }
87
88 static void git_hash_sha1_clone(git_hash_ctx *dst, const git_hash_ctx *src)
89 {
90         git_SHA1_Clone(&dst->sha1, &src->sha1);
91 }
92
93 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
94 {
95         git_SHA1_Update(&ctx->sha1, data, len);
96 }
97
98 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
99 {
100         git_SHA1_Final(hash, &ctx->sha1);
101 }
102
103 static void git_hash_sha1_final_oid(struct object_id *oid, git_hash_ctx *ctx)
104 {
105         git_SHA1_Final(oid->hash, &ctx->sha1);
106         memset(oid->hash + GIT_SHA1_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA1_RAWSZ);
107         oid->algo = GIT_HASH_SHA1;
108 }
109
110
111 static void git_hash_sha256_init(git_hash_ctx *ctx)
112 {
113         git_SHA256_Init(&ctx->sha256);
114 }
115
116 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
117 {
118         git_SHA256_Clone(&dst->sha256, &src->sha256);
119 }
120
121 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
122 {
123         git_SHA256_Update(&ctx->sha256, data, len);
124 }
125
126 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
127 {
128         git_SHA256_Final(hash, &ctx->sha256);
129 }
130
131 static void git_hash_sha256_final_oid(struct object_id *oid, git_hash_ctx *ctx)
132 {
133         git_SHA256_Final(oid->hash, &ctx->sha256);
134         /*
135          * This currently does nothing, so the compiler should optimize it out,
136          * but keep it in case we extend the hash size again.
137          */
138         memset(oid->hash + GIT_SHA256_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA256_RAWSZ);
139         oid->algo = GIT_HASH_SHA256;
140 }
141
142 static void git_hash_unknown_init(git_hash_ctx *ctx)
143 {
144         BUG("trying to init unknown hash");
145 }
146
147 static void git_hash_unknown_clone(git_hash_ctx *dst, const git_hash_ctx *src)
148 {
149         BUG("trying to clone unknown hash");
150 }
151
152 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
153 {
154         BUG("trying to update unknown hash");
155 }
156
157 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
158 {
159         BUG("trying to finalize unknown hash");
160 }
161
162 static void git_hash_unknown_final_oid(struct object_id *oid, git_hash_ctx *ctx)
163 {
164         BUG("trying to finalize unknown hash");
165 }
166
167
168 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
169         {
170                 NULL,
171                 0x00000000,
172                 0,
173                 0,
174                 0,
175                 git_hash_unknown_init,
176                 git_hash_unknown_clone,
177                 git_hash_unknown_update,
178                 git_hash_unknown_final,
179                 git_hash_unknown_final_oid,
180                 NULL,
181                 NULL,
182                 NULL,
183         },
184         {
185                 "sha1",
186                 /* "sha1", big-endian */
187                 0x73686131,
188                 GIT_SHA1_RAWSZ,
189                 GIT_SHA1_HEXSZ,
190                 GIT_SHA1_BLKSZ,
191                 git_hash_sha1_init,
192                 git_hash_sha1_clone,
193                 git_hash_sha1_update,
194                 git_hash_sha1_final,
195                 git_hash_sha1_final_oid,
196                 &empty_tree_oid,
197                 &empty_blob_oid,
198                 &null_oid_sha1,
199         },
200         {
201                 "sha256",
202                 /* "s256", big-endian */
203                 0x73323536,
204                 GIT_SHA256_RAWSZ,
205                 GIT_SHA256_HEXSZ,
206                 GIT_SHA256_BLKSZ,
207                 git_hash_sha256_init,
208                 git_hash_sha256_clone,
209                 git_hash_sha256_update,
210                 git_hash_sha256_final,
211                 git_hash_sha256_final_oid,
212                 &empty_tree_oid_sha256,
213                 &empty_blob_oid_sha256,
214                 &null_oid_sha256,
215         }
216 };
217
218 const struct object_id *null_oid(void)
219 {
220         return the_hash_algo->null_oid;
221 }
222
223 const char *empty_tree_oid_hex(void)
224 {
225         static char buf[GIT_MAX_HEXSZ + 1];
226         return oid_to_hex_r(buf, the_hash_algo->empty_tree);
227 }
228
229 const char *empty_blob_oid_hex(void)
230 {
231         static char buf[GIT_MAX_HEXSZ + 1];
232         return oid_to_hex_r(buf, the_hash_algo->empty_blob);
233 }
234
235 int hash_algo_by_name(const char *name)
236 {
237         int i;
238         if (!name)
239                 return GIT_HASH_UNKNOWN;
240         for (i = 1; i < GIT_HASH_NALGOS; i++)
241                 if (!strcmp(name, hash_algos[i].name))
242                         return i;
243         return GIT_HASH_UNKNOWN;
244 }
245
246 int hash_algo_by_id(uint32_t format_id)
247 {
248         int i;
249         for (i = 1; i < GIT_HASH_NALGOS; i++)
250                 if (format_id == hash_algos[i].format_id)
251                         return i;
252         return GIT_HASH_UNKNOWN;
253 }
254
255 int hash_algo_by_length(int len)
256 {
257         int i;
258         for (i = 1; i < GIT_HASH_NALGOS; i++)
259                 if (len == hash_algos[i].rawsz)
260                         return i;
261         return GIT_HASH_UNKNOWN;
262 }
263
264 /*
265  * This is meant to hold a *small* number of objects that you would
266  * want read_object_file() to be able to return, but yet you do not want
267  * to write them into the object store (e.g. a browse-only
268  * application).
269  */
270 static struct cached_object {
271         struct object_id oid;
272         enum object_type type;
273         void *buf;
274         unsigned long size;
275 } *cached_objects;
276 static int cached_object_nr, cached_object_alloc;
277
278 static struct cached_object empty_tree = {
279         { EMPTY_TREE_SHA1_BIN_LITERAL },
280         OBJ_TREE,
281         "",
282         0
283 };
284
285 static struct cached_object *find_cached_object(const struct object_id *oid)
286 {
287         int i;
288         struct cached_object *co = cached_objects;
289
290         for (i = 0; i < cached_object_nr; i++, co++) {
291                 if (oideq(&co->oid, oid))
292                         return co;
293         }
294         if (oideq(oid, the_hash_algo->empty_tree))
295                 return &empty_tree;
296         return NULL;
297 }
298
299
300 static int get_conv_flags(unsigned flags)
301 {
302         if (flags & HASH_RENORMALIZE)
303                 return CONV_EOL_RENORMALIZE;
304         else if (flags & HASH_WRITE_OBJECT)
305                 return global_conv_flags_eol | CONV_WRITE_OBJECT;
306         else
307                 return 0;
308 }
309
310
311 int mkdir_in_gitdir(const char *path)
312 {
313         if (mkdir(path, 0777)) {
314                 int saved_errno = errno;
315                 struct stat st;
316                 struct strbuf sb = STRBUF_INIT;
317
318                 if (errno != EEXIST)
319                         return -1;
320                 /*
321                  * Are we looking at a path in a symlinked worktree
322                  * whose original repository does not yet have it?
323                  * e.g. .git/rr-cache pointing at its original
324                  * repository in which the user hasn't performed any
325                  * conflict resolution yet?
326                  */
327                 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
328                     strbuf_readlink(&sb, path, st.st_size) ||
329                     !is_absolute_path(sb.buf) ||
330                     mkdir(sb.buf, 0777)) {
331                         strbuf_release(&sb);
332                         errno = saved_errno;
333                         return -1;
334                 }
335                 strbuf_release(&sb);
336         }
337         return adjust_shared_perm(path);
338 }
339
340 static enum scld_error safe_create_leading_directories_1(char *path, int share)
341 {
342         char *next_component = path + offset_1st_component(path);
343         enum scld_error ret = SCLD_OK;
344
345         while (ret == SCLD_OK && next_component) {
346                 struct stat st;
347                 char *slash = next_component, slash_character;
348
349                 while (*slash && !is_dir_sep(*slash))
350                         slash++;
351
352                 if (!*slash)
353                         break;
354
355                 next_component = slash + 1;
356                 while (is_dir_sep(*next_component))
357                         next_component++;
358                 if (!*next_component)
359                         break;
360
361                 slash_character = *slash;
362                 *slash = '\0';
363                 if (!stat(path, &st)) {
364                         /* path exists */
365                         if (!S_ISDIR(st.st_mode)) {
366                                 errno = ENOTDIR;
367                                 ret = SCLD_EXISTS;
368                         }
369                 } else if (mkdir(path, 0777)) {
370                         if (errno == EEXIST &&
371                             !stat(path, &st) && S_ISDIR(st.st_mode))
372                                 ; /* somebody created it since we checked */
373                         else if (errno == ENOENT)
374                                 /*
375                                  * Either mkdir() failed because
376                                  * somebody just pruned the containing
377                                  * directory, or stat() failed because
378                                  * the file that was in our way was
379                                  * just removed.  Either way, inform
380                                  * the caller that it might be worth
381                                  * trying again:
382                                  */
383                                 ret = SCLD_VANISHED;
384                         else
385                                 ret = SCLD_FAILED;
386                 } else if (share && adjust_shared_perm(path)) {
387                         ret = SCLD_PERMS;
388                 }
389                 *slash = slash_character;
390         }
391         return ret;
392 }
393
394 enum scld_error safe_create_leading_directories(char *path)
395 {
396         return safe_create_leading_directories_1(path, 1);
397 }
398
399 enum scld_error safe_create_leading_directories_no_share(char *path)
400 {
401         return safe_create_leading_directories_1(path, 0);
402 }
403
404 enum scld_error safe_create_leading_directories_const(const char *path)
405 {
406         int save_errno;
407         /* path points to cache entries, so xstrdup before messing with it */
408         char *buf = xstrdup(path);
409         enum scld_error result = safe_create_leading_directories(buf);
410
411         save_errno = errno;
412         free(buf);
413         errno = save_errno;
414         return result;
415 }
416
417 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
418 {
419         /*
420          * The number of times we will try to remove empty directories
421          * in the way of path. This is only 1 because if another
422          * process is racily creating directories that conflict with
423          * us, we don't want to fight against them.
424          */
425         int remove_directories_remaining = 1;
426
427         /*
428          * The number of times that we will try to create the
429          * directories containing path. We are willing to attempt this
430          * more than once, because another process could be trying to
431          * clean up empty directories at the same time as we are
432          * trying to create them.
433          */
434         int create_directories_remaining = 3;
435
436         /* A scratch copy of path, filled lazily if we need it: */
437         struct strbuf path_copy = STRBUF_INIT;
438
439         int ret, save_errno;
440
441         /* Sanity check: */
442         assert(*path);
443
444 retry_fn:
445         ret = fn(path, cb);
446         save_errno = errno;
447         if (!ret)
448                 goto out;
449
450         if (errno == EISDIR && remove_directories_remaining-- > 0) {
451                 /*
452                  * A directory is in the way. Maybe it is empty; try
453                  * to remove it:
454                  */
455                 if (!path_copy.len)
456                         strbuf_addstr(&path_copy, path);
457
458                 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
459                         goto retry_fn;
460         } else if (errno == ENOENT && create_directories_remaining-- > 0) {
461                 /*
462                  * Maybe the containing directory didn't exist, or
463                  * maybe it was just deleted by a process that is
464                  * racing with us to clean up empty directories. Try
465                  * to create it:
466                  */
467                 enum scld_error scld_result;
468
469                 if (!path_copy.len)
470                         strbuf_addstr(&path_copy, path);
471
472                 do {
473                         scld_result = safe_create_leading_directories(path_copy.buf);
474                         if (scld_result == SCLD_OK)
475                                 goto retry_fn;
476                 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
477         }
478
479 out:
480         strbuf_release(&path_copy);
481         errno = save_errno;
482         return ret;
483 }
484
485 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
486 {
487         int i;
488         for (i = 0; i < the_hash_algo->rawsz; i++) {
489                 static char hex[] = "0123456789abcdef";
490                 unsigned int val = oid->hash[i];
491                 strbuf_addch(buf, hex[val >> 4]);
492                 strbuf_addch(buf, hex[val & 0xf]);
493                 if (!i)
494                         strbuf_addch(buf, '/');
495         }
496 }
497
498 static const char *odb_loose_path(struct object_directory *odb,
499                                   struct strbuf *buf,
500                                   const struct object_id *oid)
501 {
502         strbuf_reset(buf);
503         strbuf_addstr(buf, odb->path);
504         strbuf_addch(buf, '/');
505         fill_loose_path(buf, oid);
506         return buf->buf;
507 }
508
509 const char *loose_object_path(struct repository *r, struct strbuf *buf,
510                               const struct object_id *oid)
511 {
512         return odb_loose_path(r->objects->odb, buf, oid);
513 }
514
515 /*
516  * Return non-zero iff the path is usable as an alternate object database.
517  */
518 static int alt_odb_usable(struct raw_object_store *o,
519                           struct strbuf *path,
520                           const char *normalized_objdir)
521 {
522         struct object_directory *odb;
523
524         /* Detect cases where alternate disappeared */
525         if (!is_directory(path->buf)) {
526                 error(_("object directory %s does not exist; "
527                         "check .git/objects/info/alternates"),
528                       path->buf);
529                 return 0;
530         }
531
532         /*
533          * Prevent the common mistake of listing the same
534          * thing twice, or object directory itself.
535          */
536         for (odb = o->odb; odb; odb = odb->next) {
537                 if (!fspathcmp(path->buf, odb->path))
538                         return 0;
539         }
540         if (!fspathcmp(path->buf, normalized_objdir))
541                 return 0;
542
543         return 1;
544 }
545
546 /*
547  * Prepare alternate object database registry.
548  *
549  * The variable alt_odb_list points at the list of struct
550  * object_directory.  The elements on this list come from
551  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
552  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
553  * whose contents is similar to that environment variable but can be
554  * LF separated.  Its base points at a statically allocated buffer that
555  * contains "/the/directory/corresponding/to/.git/objects/...", while
556  * its name points just after the slash at the end of ".git/objects/"
557  * in the example above, and has enough space to hold all hex characters
558  * of the object ID, an extra slash for the first level indirection, and
559  * the terminating NUL.
560  */
561 static void read_info_alternates(struct repository *r,
562                                  const char *relative_base,
563                                  int depth);
564 static int link_alt_odb_entry(struct repository *r, const char *entry,
565         const char *relative_base, int depth, const char *normalized_objdir)
566 {
567         struct object_directory *ent;
568         struct strbuf pathbuf = STRBUF_INIT;
569
570         if (!is_absolute_path(entry) && relative_base) {
571                 strbuf_realpath(&pathbuf, relative_base, 1);
572                 strbuf_addch(&pathbuf, '/');
573         }
574         strbuf_addstr(&pathbuf, entry);
575
576         if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
577                 error(_("unable to normalize alternate object path: %s"),
578                       pathbuf.buf);
579                 strbuf_release(&pathbuf);
580                 return -1;
581         }
582
583         /*
584          * The trailing slash after the directory name is given by
585          * this function at the end. Remove duplicates.
586          */
587         while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
588                 strbuf_setlen(&pathbuf, pathbuf.len - 1);
589
590         if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
591                 strbuf_release(&pathbuf);
592                 return -1;
593         }
594
595         CALLOC_ARRAY(ent, 1);
596         ent->path = xstrdup(pathbuf.buf);
597
598         /* add the alternate entry */
599         *r->objects->odb_tail = ent;
600         r->objects->odb_tail = &(ent->next);
601         ent->next = NULL;
602
603         /* recursively add alternates */
604         read_info_alternates(r, pathbuf.buf, depth + 1);
605
606         strbuf_release(&pathbuf);
607         return 0;
608 }
609
610 static const char *parse_alt_odb_entry(const char *string,
611                                        int sep,
612                                        struct strbuf *out)
613 {
614         const char *end;
615
616         strbuf_reset(out);
617
618         if (*string == '#') {
619                 /* comment; consume up to next separator */
620                 end = strchrnul(string, sep);
621         } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
622                 /*
623                  * quoted path; unquote_c_style has copied the
624                  * data for us and set "end". Broken quoting (e.g.,
625                  * an entry that doesn't end with a quote) falls
626                  * back to the unquoted case below.
627                  */
628         } else {
629                 /* normal, unquoted path */
630                 end = strchrnul(string, sep);
631                 strbuf_add(out, string, end - string);
632         }
633
634         if (*end)
635                 end++;
636         return end;
637 }
638
639 static void link_alt_odb_entries(struct repository *r, const char *alt,
640                                  int sep, const char *relative_base, int depth)
641 {
642         struct strbuf objdirbuf = STRBUF_INIT;
643         struct strbuf entry = STRBUF_INIT;
644
645         if (!alt || !*alt)
646                 return;
647
648         if (depth > 5) {
649                 error(_("%s: ignoring alternate object stores, nesting too deep"),
650                                 relative_base);
651                 return;
652         }
653
654         strbuf_add_absolute_path(&objdirbuf, r->objects->odb->path);
655         if (strbuf_normalize_path(&objdirbuf) < 0)
656                 die(_("unable to normalize object directory: %s"),
657                     objdirbuf.buf);
658
659         while (*alt) {
660                 alt = parse_alt_odb_entry(alt, sep, &entry);
661                 if (!entry.len)
662                         continue;
663                 link_alt_odb_entry(r, entry.buf,
664                                    relative_base, depth, objdirbuf.buf);
665         }
666         strbuf_release(&entry);
667         strbuf_release(&objdirbuf);
668 }
669
670 static void read_info_alternates(struct repository *r,
671                                  const char *relative_base,
672                                  int depth)
673 {
674         char *path;
675         struct strbuf buf = STRBUF_INIT;
676
677         path = xstrfmt("%s/info/alternates", relative_base);
678         if (strbuf_read_file(&buf, path, 1024) < 0) {
679                 warn_on_fopen_errors(path);
680                 free(path);
681                 return;
682         }
683
684         link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
685         strbuf_release(&buf);
686         free(path);
687 }
688
689 void add_to_alternates_file(const char *reference)
690 {
691         struct lock_file lock = LOCK_INIT;
692         char *alts = git_pathdup("objects/info/alternates");
693         FILE *in, *out;
694         int found = 0;
695
696         hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
697         out = fdopen_lock_file(&lock, "w");
698         if (!out)
699                 die_errno(_("unable to fdopen alternates lockfile"));
700
701         in = fopen(alts, "r");
702         if (in) {
703                 struct strbuf line = STRBUF_INIT;
704
705                 while (strbuf_getline(&line, in) != EOF) {
706                         if (!strcmp(reference, line.buf)) {
707                                 found = 1;
708                                 break;
709                         }
710                         fprintf_or_die(out, "%s\n", line.buf);
711                 }
712
713                 strbuf_release(&line);
714                 fclose(in);
715         }
716         else if (errno != ENOENT)
717                 die_errno(_("unable to read alternates file"));
718
719         if (found) {
720                 rollback_lock_file(&lock);
721         } else {
722                 fprintf_or_die(out, "%s\n", reference);
723                 if (commit_lock_file(&lock))
724                         die_errno(_("unable to move new alternates file into place"));
725                 if (the_repository->objects->loaded_alternates)
726                         link_alt_odb_entries(the_repository, reference,
727                                              '\n', NULL, 0);
728         }
729         free(alts);
730 }
731
732 void add_to_alternates_memory(const char *reference)
733 {
734         /*
735          * Make sure alternates are initialized, or else our entry may be
736          * overwritten when they are.
737          */
738         prepare_alt_odb(the_repository);
739
740         link_alt_odb_entries(the_repository, reference,
741                              '\n', NULL, 0);
742 }
743
744 /*
745  * Compute the exact path an alternate is at and returns it. In case of
746  * error NULL is returned and the human readable error is added to `err`
747  * `path` may be relative and should point to $GIT_DIR.
748  * `err` must not be null.
749  */
750 char *compute_alternate_path(const char *path, struct strbuf *err)
751 {
752         char *ref_git = NULL;
753         const char *repo;
754         int seen_error = 0;
755
756         ref_git = real_pathdup(path, 0);
757         if (!ref_git) {
758                 seen_error = 1;
759                 strbuf_addf(err, _("path '%s' does not exist"), path);
760                 goto out;
761         }
762
763         repo = read_gitfile(ref_git);
764         if (!repo)
765                 repo = read_gitfile(mkpath("%s/.git", ref_git));
766         if (repo) {
767                 free(ref_git);
768                 ref_git = xstrdup(repo);
769         }
770
771         if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
772                 char *ref_git_git = mkpathdup("%s/.git", ref_git);
773                 free(ref_git);
774                 ref_git = ref_git_git;
775         } else if (!is_directory(mkpath("%s/objects", ref_git))) {
776                 struct strbuf sb = STRBUF_INIT;
777                 seen_error = 1;
778                 if (get_common_dir(&sb, ref_git)) {
779                         strbuf_addf(err,
780                                     _("reference repository '%s' as a linked "
781                                       "checkout is not supported yet."),
782                                     path);
783                         goto out;
784                 }
785
786                 strbuf_addf(err, _("reference repository '%s' is not a "
787                                         "local repository."), path);
788                 goto out;
789         }
790
791         if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
792                 strbuf_addf(err, _("reference repository '%s' is shallow"),
793                             path);
794                 seen_error = 1;
795                 goto out;
796         }
797
798         if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
799                 strbuf_addf(err,
800                             _("reference repository '%s' is grafted"),
801                             path);
802                 seen_error = 1;
803                 goto out;
804         }
805
806 out:
807         if (seen_error) {
808                 FREE_AND_NULL(ref_git);
809         }
810
811         return ref_git;
812 }
813
814 static void fill_alternate_refs_command(struct child_process *cmd,
815                                         const char *repo_path)
816 {
817         const char *value;
818
819         if (!git_config_get_value("core.alternateRefsCommand", &value)) {
820                 cmd->use_shell = 1;
821
822                 strvec_push(&cmd->args, value);
823                 strvec_push(&cmd->args, repo_path);
824         } else {
825                 cmd->git_cmd = 1;
826
827                 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
828                 strvec_push(&cmd->args, "for-each-ref");
829                 strvec_push(&cmd->args, "--format=%(objectname)");
830
831                 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
832                         strvec_push(&cmd->args, "--");
833                         strvec_split(&cmd->args, value);
834                 }
835         }
836
837         cmd->env = local_repo_env;
838         cmd->out = -1;
839 }
840
841 static void read_alternate_refs(const char *path,
842                                 alternate_ref_fn *cb,
843                                 void *data)
844 {
845         struct child_process cmd = CHILD_PROCESS_INIT;
846         struct strbuf line = STRBUF_INIT;
847         FILE *fh;
848
849         fill_alternate_refs_command(&cmd, path);
850
851         if (start_command(&cmd))
852                 return;
853
854         fh = xfdopen(cmd.out, "r");
855         while (strbuf_getline_lf(&line, fh) != EOF) {
856                 struct object_id oid;
857                 const char *p;
858
859                 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
860                         warning(_("invalid line while parsing alternate refs: %s"),
861                                 line.buf);
862                         break;
863                 }
864
865                 cb(&oid, data);
866         }
867
868         fclose(fh);
869         finish_command(&cmd);
870         strbuf_release(&line);
871 }
872
873 struct alternate_refs_data {
874         alternate_ref_fn *fn;
875         void *data;
876 };
877
878 static int refs_from_alternate_cb(struct object_directory *e,
879                                   void *data)
880 {
881         struct strbuf path = STRBUF_INIT;
882         size_t base_len;
883         struct alternate_refs_data *cb = data;
884
885         if (!strbuf_realpath(&path, e->path, 0))
886                 goto out;
887         if (!strbuf_strip_suffix(&path, "/objects"))
888                 goto out;
889         base_len = path.len;
890
891         /* Is this a git repository with refs? */
892         strbuf_addstr(&path, "/refs");
893         if (!is_directory(path.buf))
894                 goto out;
895         strbuf_setlen(&path, base_len);
896
897         read_alternate_refs(path.buf, cb->fn, cb->data);
898
899 out:
900         strbuf_release(&path);
901         return 0;
902 }
903
904 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
905 {
906         struct alternate_refs_data cb;
907         cb.fn = fn;
908         cb.data = data;
909         foreach_alt_odb(refs_from_alternate_cb, &cb);
910 }
911
912 int foreach_alt_odb(alt_odb_fn fn, void *cb)
913 {
914         struct object_directory *ent;
915         int r = 0;
916
917         prepare_alt_odb(the_repository);
918         for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
919                 r = fn(ent, cb);
920                 if (r)
921                         break;
922         }
923         return r;
924 }
925
926 void prepare_alt_odb(struct repository *r)
927 {
928         if (r->objects->loaded_alternates)
929                 return;
930
931         link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
932
933         read_info_alternates(r, r->objects->odb->path, 0);
934         r->objects->loaded_alternates = 1;
935 }
936
937 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
938 static int freshen_file(const char *fn)
939 {
940         return !utime(fn, NULL);
941 }
942
943 /*
944  * All of the check_and_freshen functions return 1 if the file exists and was
945  * freshened (if freshening was requested), 0 otherwise. If they return
946  * 0, you should not assume that it is safe to skip a write of the object (it
947  * either does not exist on disk, or has a stale mtime and may be subject to
948  * pruning).
949  */
950 int check_and_freshen_file(const char *fn, int freshen)
951 {
952         if (access(fn, F_OK))
953                 return 0;
954         if (freshen && !freshen_file(fn))
955                 return 0;
956         return 1;
957 }
958
959 static int check_and_freshen_odb(struct object_directory *odb,
960                                  const struct object_id *oid,
961                                  int freshen)
962 {
963         static struct strbuf path = STRBUF_INIT;
964         odb_loose_path(odb, &path, oid);
965         return check_and_freshen_file(path.buf, freshen);
966 }
967
968 static int check_and_freshen_local(const struct object_id *oid, int freshen)
969 {
970         return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
971 }
972
973 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
974 {
975         struct object_directory *odb;
976
977         prepare_alt_odb(the_repository);
978         for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
979                 if (check_and_freshen_odb(odb, oid, freshen))
980                         return 1;
981         }
982         return 0;
983 }
984
985 static int check_and_freshen(const struct object_id *oid, int freshen)
986 {
987         return check_and_freshen_local(oid, freshen) ||
988                check_and_freshen_nonlocal(oid, freshen);
989 }
990
991 int has_loose_object_nonlocal(const struct object_id *oid)
992 {
993         return check_and_freshen_nonlocal(oid, 0);
994 }
995
996 static int has_loose_object(const struct object_id *oid)
997 {
998         return check_and_freshen(oid, 0);
999 }
1000
1001 static void mmap_limit_check(size_t length)
1002 {
1003         static size_t limit = 0;
1004         if (!limit) {
1005                 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1006                 if (!limit)
1007                         limit = SIZE_MAX;
1008         }
1009         if (length > limit)
1010                 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
1011                     (uintmax_t)length, (uintmax_t)limit);
1012 }
1013
1014 void *xmmap_gently(void *start, size_t length,
1015                   int prot, int flags, int fd, off_t offset)
1016 {
1017         void *ret;
1018
1019         mmap_limit_check(length);
1020         ret = mmap(start, length, prot, flags, fd, offset);
1021         if (ret == MAP_FAILED && !length)
1022                 ret = NULL;
1023         return ret;
1024 }
1025
1026 void *xmmap(void *start, size_t length,
1027         int prot, int flags, int fd, off_t offset)
1028 {
1029         void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1030         if (ret == MAP_FAILED)
1031                 die_errno(_("mmap failed"));
1032         return ret;
1033 }
1034
1035 /*
1036  * With an in-core object data in "map", rehash it to make sure the
1037  * object name actually matches "oid" to detect object corruption.
1038  * With "map" == NULL, try reading the object named with "oid" using
1039  * the streaming interface and rehash it to do the same.
1040  */
1041 int check_object_signature(struct repository *r, const struct object_id *oid,
1042                            void *map, unsigned long size, const char *type)
1043 {
1044         struct object_id real_oid;
1045         enum object_type obj_type;
1046         struct git_istream *st;
1047         git_hash_ctx c;
1048         char hdr[MAX_HEADER_LEN];
1049         int hdrlen;
1050
1051         if (map) {
1052                 hash_object_file(r->hash_algo, map, size, type, &real_oid);
1053                 return !oideq(oid, &real_oid) ? -1 : 0;
1054         }
1055
1056         st = open_istream(r, oid, &obj_type, &size, NULL);
1057         if (!st)
1058                 return -1;
1059
1060         /* Generate the header */
1061         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(obj_type), (uintmax_t)size) + 1;
1062
1063         /* Sha1.. */
1064         r->hash_algo->init_fn(&c);
1065         r->hash_algo->update_fn(&c, hdr, hdrlen);
1066         for (;;) {
1067                 char buf[1024 * 16];
1068                 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1069
1070                 if (readlen < 0) {
1071                         close_istream(st);
1072                         return -1;
1073                 }
1074                 if (!readlen)
1075                         break;
1076                 r->hash_algo->update_fn(&c, buf, readlen);
1077         }
1078         r->hash_algo->final_oid_fn(&real_oid, &c);
1079         close_istream(st);
1080         return !oideq(oid, &real_oid) ? -1 : 0;
1081 }
1082
1083 int git_open_cloexec(const char *name, int flags)
1084 {
1085         int fd;
1086         static int o_cloexec = O_CLOEXEC;
1087
1088         fd = open(name, flags | o_cloexec);
1089         if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1090                 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1091                 o_cloexec &= ~O_CLOEXEC;
1092                 fd = open(name, flags | o_cloexec);
1093         }
1094
1095 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1096         {
1097                 static int fd_cloexec = FD_CLOEXEC;
1098
1099                 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1100                         /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
1101                         int flags = fcntl(fd, F_GETFD);
1102                         if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1103                                 fd_cloexec = 0;
1104                 }
1105         }
1106 #endif
1107         return fd;
1108 }
1109
1110 /*
1111  * Find "oid" as a loose object in the local repository or in an alternate.
1112  * Returns 0 on success, negative on failure.
1113  *
1114  * The "path" out-parameter will give the path of the object we found (if any).
1115  * Note that it may point to static storage and is only valid until another
1116  * call to stat_loose_object().
1117  */
1118 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1119                              struct stat *st, const char **path)
1120 {
1121         struct object_directory *odb;
1122         static struct strbuf buf = STRBUF_INIT;
1123
1124         prepare_alt_odb(r);
1125         for (odb = r->objects->odb; odb; odb = odb->next) {
1126                 *path = odb_loose_path(odb, &buf, oid);
1127                 if (!lstat(*path, st))
1128                         return 0;
1129         }
1130
1131         return -1;
1132 }
1133
1134 /*
1135  * Like stat_loose_object(), but actually open the object and return the
1136  * descriptor. See the caveats on the "path" parameter above.
1137  */
1138 static int open_loose_object(struct repository *r,
1139                              const struct object_id *oid, const char **path)
1140 {
1141         int fd;
1142         struct object_directory *odb;
1143         int most_interesting_errno = ENOENT;
1144         static struct strbuf buf = STRBUF_INIT;
1145
1146         prepare_alt_odb(r);
1147         for (odb = r->objects->odb; odb; odb = odb->next) {
1148                 *path = odb_loose_path(odb, &buf, oid);
1149                 fd = git_open(*path);
1150                 if (fd >= 0)
1151                         return fd;
1152
1153                 if (most_interesting_errno == ENOENT)
1154                         most_interesting_errno = errno;
1155         }
1156         errno = most_interesting_errno;
1157         return -1;
1158 }
1159
1160 static int quick_has_loose(struct repository *r,
1161                            const struct object_id *oid)
1162 {
1163         struct object_directory *odb;
1164
1165         prepare_alt_odb(r);
1166         for (odb = r->objects->odb; odb; odb = odb->next) {
1167                 if (oid_array_lookup(odb_loose_cache(odb, oid), oid) >= 0)
1168                         return 1;
1169         }
1170         return 0;
1171 }
1172
1173 /*
1174  * Map the loose object at "path" if it is not NULL, or the path found by
1175  * searching for a loose object named "oid".
1176  */
1177 static void *map_loose_object_1(struct repository *r, const char *path,
1178                              const struct object_id *oid, unsigned long *size)
1179 {
1180         void *map;
1181         int fd;
1182
1183         if (path)
1184                 fd = git_open(path);
1185         else
1186                 fd = open_loose_object(r, oid, &path);
1187         map = NULL;
1188         if (fd >= 0) {
1189                 struct stat st;
1190
1191                 if (!fstat(fd, &st)) {
1192                         *size = xsize_t(st.st_size);
1193                         if (!*size) {
1194                                 /* mmap() is forbidden on empty files */
1195                                 error(_("object file %s is empty"), path);
1196                                 close(fd);
1197                                 return NULL;
1198                         }
1199                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1200                 }
1201                 close(fd);
1202         }
1203         return map;
1204 }
1205
1206 void *map_loose_object(struct repository *r,
1207                        const struct object_id *oid,
1208                        unsigned long *size)
1209 {
1210         return map_loose_object_1(r, NULL, oid, size);
1211 }
1212
1213 static int unpack_loose_short_header(git_zstream *stream,
1214                                      unsigned char *map, unsigned long mapsize,
1215                                      void *buffer, unsigned long bufsiz)
1216 {
1217         int ret;
1218
1219         /* Get the data stream */
1220         memset(stream, 0, sizeof(*stream));
1221         stream->next_in = map;
1222         stream->avail_in = mapsize;
1223         stream->next_out = buffer;
1224         stream->avail_out = bufsiz;
1225
1226         git_inflate_init(stream);
1227         obj_read_unlock();
1228         ret = git_inflate(stream, 0);
1229         obj_read_lock();
1230
1231         return ret;
1232 }
1233
1234 int unpack_loose_header(git_zstream *stream,
1235                         unsigned char *map, unsigned long mapsize,
1236                         void *buffer, unsigned long bufsiz)
1237 {
1238         int status = unpack_loose_short_header(stream, map, mapsize,
1239                                                buffer, bufsiz);
1240
1241         if (status < Z_OK)
1242                 return status;
1243
1244         /* Make sure we have the terminating NUL */
1245         if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1246                 return -1;
1247         return 0;
1248 }
1249
1250 static int unpack_loose_header_to_strbuf(git_zstream *stream, unsigned char *map,
1251                                          unsigned long mapsize, void *buffer,
1252                                          unsigned long bufsiz, struct strbuf *header)
1253 {
1254         int status;
1255
1256         status = unpack_loose_short_header(stream, map, mapsize, buffer, bufsiz);
1257         if (status < Z_OK)
1258                 return -1;
1259
1260         /*
1261          * Check if entire header is unpacked in the first iteration.
1262          */
1263         if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1264                 return 0;
1265
1266         /*
1267          * buffer[0..bufsiz] was not large enough.  Copy the partial
1268          * result out to header, and then append the result of further
1269          * reading the stream.
1270          */
1271         strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1272         stream->next_out = buffer;
1273         stream->avail_out = bufsiz;
1274
1275         do {
1276                 obj_read_unlock();
1277                 status = git_inflate(stream, 0);
1278                 obj_read_lock();
1279                 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1280                 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1281                         return 0;
1282                 stream->next_out = buffer;
1283                 stream->avail_out = bufsiz;
1284         } while (status != Z_STREAM_END);
1285         return -1;
1286 }
1287
1288 static void *unpack_loose_rest(git_zstream *stream,
1289                                void *buffer, unsigned long size,
1290                                const struct object_id *oid)
1291 {
1292         int bytes = strlen(buffer) + 1;
1293         unsigned char *buf = xmallocz(size);
1294         unsigned long n;
1295         int status = Z_OK;
1296
1297         n = stream->total_out - bytes;
1298         if (n > size)
1299                 n = size;
1300         memcpy(buf, (char *) buffer + bytes, n);
1301         bytes = n;
1302         if (bytes <= size) {
1303                 /*
1304                  * The above condition must be (bytes <= size), not
1305                  * (bytes < size).  In other words, even though we
1306                  * expect no more output and set avail_out to zero,
1307                  * the input zlib stream may have bytes that express
1308                  * "this concludes the stream", and we *do* want to
1309                  * eat that input.
1310                  *
1311                  * Otherwise we would not be able to test that we
1312                  * consumed all the input to reach the expected size;
1313                  * we also want to check that zlib tells us that all
1314                  * went well with status == Z_STREAM_END at the end.
1315                  */
1316                 stream->next_out = buf + bytes;
1317                 stream->avail_out = size - bytes;
1318                 while (status == Z_OK) {
1319                         obj_read_unlock();
1320                         status = git_inflate(stream, Z_FINISH);
1321                         obj_read_lock();
1322                 }
1323         }
1324         if (status == Z_STREAM_END && !stream->avail_in) {
1325                 git_inflate_end(stream);
1326                 return buf;
1327         }
1328
1329         if (status < 0)
1330                 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1331         else if (stream->avail_in)
1332                 error(_("garbage at end of loose object '%s'"),
1333                       oid_to_hex(oid));
1334         free(buf);
1335         return NULL;
1336 }
1337
1338 /*
1339  * We used to just use "sscanf()", but that's actually way
1340  * too permissive for what we want to check. So do an anal
1341  * object header parse by hand.
1342  */
1343 static int parse_loose_header_extended(const char *hdr, struct object_info *oi,
1344                                        unsigned int flags)
1345 {
1346         const char *type_buf = hdr;
1347         unsigned long size;
1348         int type, type_len = 0;
1349
1350         /*
1351          * The type can be of any size but is followed by
1352          * a space.
1353          */
1354         for (;;) {
1355                 char c = *hdr++;
1356                 if (!c)
1357                         return -1;
1358                 if (c == ' ')
1359                         break;
1360                 type_len++;
1361         }
1362
1363         type = type_from_string_gently(type_buf, type_len, 1);
1364         if (oi->type_name)
1365                 strbuf_add(oi->type_name, type_buf, type_len);
1366         /*
1367          * Set type to 0 if its an unknown object and
1368          * we're obtaining the type using '--allow-unknown-type'
1369          * option.
1370          */
1371         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1372                 type = 0;
1373         else if (type < 0)
1374                 die(_("invalid object type"));
1375         if (oi->typep)
1376                 *oi->typep = type;
1377
1378         /*
1379          * The length must follow immediately, and be in canonical
1380          * decimal format (ie "010" is not valid).
1381          */
1382         size = *hdr++ - '0';
1383         if (size > 9)
1384                 return -1;
1385         if (size) {
1386                 for (;;) {
1387                         unsigned long c = *hdr - '0';
1388                         if (c > 9)
1389                                 break;
1390                         hdr++;
1391                         size = size * 10 + c;
1392                 }
1393         }
1394
1395         if (oi->sizep)
1396                 *oi->sizep = size;
1397
1398         /*
1399          * The length must be followed by a zero byte
1400          */
1401         return *hdr ? -1 : type;
1402 }
1403
1404 int parse_loose_header(const char *hdr, unsigned long *sizep)
1405 {
1406         struct object_info oi = OBJECT_INFO_INIT;
1407
1408         oi.sizep = sizep;
1409         return parse_loose_header_extended(hdr, &oi, 0);
1410 }
1411
1412 static int loose_object_info(struct repository *r,
1413                              const struct object_id *oid,
1414                              struct object_info *oi, int flags)
1415 {
1416         int status = 0;
1417         unsigned long mapsize;
1418         void *map;
1419         git_zstream stream;
1420         char hdr[MAX_HEADER_LEN];
1421         struct strbuf hdrbuf = STRBUF_INIT;
1422         unsigned long size_scratch;
1423
1424         if (oi->delta_base_oid)
1425                 oidclr(oi->delta_base_oid);
1426
1427         /*
1428          * If we don't care about type or size, then we don't
1429          * need to look inside the object at all. Note that we
1430          * do not optimize out the stat call, even if the
1431          * caller doesn't care about the disk-size, since our
1432          * return value implicitly indicates whether the
1433          * object even exists.
1434          */
1435         if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1436                 const char *path;
1437                 struct stat st;
1438                 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1439                         return quick_has_loose(r, oid) ? 0 : -1;
1440                 if (stat_loose_object(r, oid, &st, &path) < 0)
1441                         return -1;
1442                 if (oi->disk_sizep)
1443                         *oi->disk_sizep = st.st_size;
1444                 return 0;
1445         }
1446
1447         map = map_loose_object(r, oid, &mapsize);
1448         if (!map)
1449                 return -1;
1450
1451         if (!oi->sizep)
1452                 oi->sizep = &size_scratch;
1453
1454         if (oi->disk_sizep)
1455                 *oi->disk_sizep = mapsize;
1456         if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1457                 if (unpack_loose_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1458                         status = error(_("unable to unpack %s header with --allow-unknown-type"),
1459                                        oid_to_hex(oid));
1460         } else if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1461                 status = error(_("unable to unpack %s header"),
1462                                oid_to_hex(oid));
1463         if (status < 0)
1464                 ; /* Do nothing */
1465         else if (hdrbuf.len) {
1466                 if ((status = parse_loose_header_extended(hdrbuf.buf, oi, flags)) < 0)
1467                         status = error(_("unable to parse %s header with --allow-unknown-type"),
1468                                        oid_to_hex(oid));
1469         } else if ((status = parse_loose_header_extended(hdr, oi, flags)) < 0)
1470                 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1471
1472         if (status >= 0 && oi->contentp) {
1473                 *oi->contentp = unpack_loose_rest(&stream, hdr,
1474                                                   *oi->sizep, oid);
1475                 if (!*oi->contentp) {
1476                         git_inflate_end(&stream);
1477                         status = -1;
1478                 }
1479         } else
1480                 git_inflate_end(&stream);
1481
1482         munmap(map, mapsize);
1483         if (status && oi->typep)
1484                 *oi->typep = status;
1485         if (oi->sizep == &size_scratch)
1486                 oi->sizep = NULL;
1487         strbuf_release(&hdrbuf);
1488         oi->whence = OI_LOOSE;
1489         return (status < 0) ? status : 0;
1490 }
1491
1492 int obj_read_use_lock = 0;
1493 pthread_mutex_t obj_read_mutex;
1494
1495 void enable_obj_read_lock(void)
1496 {
1497         if (obj_read_use_lock)
1498                 return;
1499
1500         obj_read_use_lock = 1;
1501         init_recursive_mutex(&obj_read_mutex);
1502 }
1503
1504 void disable_obj_read_lock(void)
1505 {
1506         if (!obj_read_use_lock)
1507                 return;
1508
1509         obj_read_use_lock = 0;
1510         pthread_mutex_destroy(&obj_read_mutex);
1511 }
1512
1513 int fetch_if_missing = 1;
1514
1515 static int do_oid_object_info_extended(struct repository *r,
1516                                        const struct object_id *oid,
1517                                        struct object_info *oi, unsigned flags)
1518 {
1519         static struct object_info blank_oi = OBJECT_INFO_INIT;
1520         struct cached_object *co;
1521         struct pack_entry e;
1522         int rtype;
1523         const struct object_id *real = oid;
1524         int already_retried = 0;
1525
1526
1527         if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1528                 real = lookup_replace_object(r, oid);
1529
1530         if (is_null_oid(real))
1531                 return -1;
1532
1533         if (!oi)
1534                 oi = &blank_oi;
1535
1536         co = find_cached_object(real);
1537         if (co) {
1538                 if (oi->typep)
1539                         *(oi->typep) = co->type;
1540                 if (oi->sizep)
1541                         *(oi->sizep) = co->size;
1542                 if (oi->disk_sizep)
1543                         *(oi->disk_sizep) = 0;
1544                 if (oi->delta_base_oid)
1545                         oidclr(oi->delta_base_oid);
1546                 if (oi->type_name)
1547                         strbuf_addstr(oi->type_name, type_name(co->type));
1548                 if (oi->contentp)
1549                         *oi->contentp = xmemdupz(co->buf, co->size);
1550                 oi->whence = OI_CACHED;
1551                 return 0;
1552         }
1553
1554         while (1) {
1555                 if (find_pack_entry(r, real, &e))
1556                         break;
1557
1558                 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1559                         return -1;
1560
1561                 /* Most likely it's a loose object. */
1562                 if (!loose_object_info(r, real, oi, flags))
1563                         return 0;
1564
1565                 /* Not a loose object; someone else may have just packed it. */
1566                 if (!(flags & OBJECT_INFO_QUICK)) {
1567                         reprepare_packed_git(r);
1568                         if (find_pack_entry(r, real, &e))
1569                                 break;
1570                 }
1571
1572                 /* Check if it is a missing object */
1573                 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1574                     !already_retried &&
1575                     !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1576                         /*
1577                          * TODO Investigate checking promisor_remote_get_direct()
1578                          * TODO return value and stopping on error here.
1579                          */
1580                         promisor_remote_get_direct(r, real, 1);
1581                         already_retried = 1;
1582                         continue;
1583                 }
1584
1585                 return -1;
1586         }
1587
1588         if (oi == &blank_oi)
1589                 /*
1590                  * We know that the caller doesn't actually need the
1591                  * information below, so return early.
1592                  */
1593                 return 0;
1594         rtype = packed_object_info(r, e.p, e.offset, oi);
1595         if (rtype < 0) {
1596                 mark_bad_packed_object(e.p, real->hash);
1597                 return do_oid_object_info_extended(r, real, oi, 0);
1598         } else if (oi->whence == OI_PACKED) {
1599                 oi->u.packed.offset = e.offset;
1600                 oi->u.packed.pack = e.p;
1601                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1602                                          rtype == OBJ_OFS_DELTA);
1603         }
1604
1605         return 0;
1606 }
1607
1608 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1609                              struct object_info *oi, unsigned flags)
1610 {
1611         int ret;
1612         obj_read_lock();
1613         ret = do_oid_object_info_extended(r, oid, oi, flags);
1614         obj_read_unlock();
1615         return ret;
1616 }
1617
1618
1619 /* returns enum object_type or negative */
1620 int oid_object_info(struct repository *r,
1621                     const struct object_id *oid,
1622                     unsigned long *sizep)
1623 {
1624         enum object_type type;
1625         struct object_info oi = OBJECT_INFO_INIT;
1626
1627         oi.typep = &type;
1628         oi.sizep = sizep;
1629         if (oid_object_info_extended(r, oid, &oi,
1630                                       OBJECT_INFO_LOOKUP_REPLACE) < 0)
1631                 return -1;
1632         return type;
1633 }
1634
1635 static void *read_object(struct repository *r,
1636                          const struct object_id *oid, enum object_type *type,
1637                          unsigned long *size)
1638 {
1639         struct object_info oi = OBJECT_INFO_INIT;
1640         void *content;
1641         oi.typep = type;
1642         oi.sizep = size;
1643         oi.contentp = &content;
1644
1645         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1646                 return NULL;
1647         return content;
1648 }
1649
1650 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1651                         struct object_id *oid)
1652 {
1653         struct cached_object *co;
1654
1655         hash_object_file(the_hash_algo, buf, len, type_name(type), oid);
1656         if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1657             find_cached_object(oid))
1658                 return 0;
1659         ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1660         co = &cached_objects[cached_object_nr++];
1661         co->size = len;
1662         co->type = type;
1663         co->buf = xmalloc(len);
1664         memcpy(co->buf, buf, len);
1665         oidcpy(&co->oid, oid);
1666         return 0;
1667 }
1668
1669 /*
1670  * This function dies on corrupt objects; the callers who want to
1671  * deal with them should arrange to call read_object() and give error
1672  * messages themselves.
1673  */
1674 void *read_object_file_extended(struct repository *r,
1675                                 const struct object_id *oid,
1676                                 enum object_type *type,
1677                                 unsigned long *size,
1678                                 int lookup_replace)
1679 {
1680         void *data;
1681         const struct packed_git *p;
1682         const char *path;
1683         struct stat st;
1684         const struct object_id *repl = lookup_replace ?
1685                 lookup_replace_object(r, oid) : oid;
1686
1687         errno = 0;
1688         data = read_object(r, repl, type, size);
1689         if (data)
1690                 return data;
1691
1692         obj_read_lock();
1693         if (errno && errno != ENOENT)
1694                 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1695
1696         /* die if we replaced an object with one that does not exist */
1697         if (repl != oid)
1698                 die(_("replacement %s not found for %s"),
1699                     oid_to_hex(repl), oid_to_hex(oid));
1700
1701         if (!stat_loose_object(r, repl, &st, &path))
1702                 die(_("loose object %s (stored in %s) is corrupt"),
1703                     oid_to_hex(repl), path);
1704
1705         if ((p = has_packed_and_bad(r, repl->hash)) != NULL)
1706                 die(_("packed object %s (stored in %s) is corrupt"),
1707                     oid_to_hex(repl), p->pack_name);
1708         obj_read_unlock();
1709
1710         return NULL;
1711 }
1712
1713 void *read_object_with_reference(struct repository *r,
1714                                  const struct object_id *oid,
1715                                  const char *required_type_name,
1716                                  unsigned long *size,
1717                                  struct object_id *actual_oid_return)
1718 {
1719         enum object_type type, required_type;
1720         void *buffer;
1721         unsigned long isize;
1722         struct object_id actual_oid;
1723
1724         required_type = type_from_string(required_type_name);
1725         oidcpy(&actual_oid, oid);
1726         while (1) {
1727                 int ref_length = -1;
1728                 const char *ref_type = NULL;
1729
1730                 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1731                 if (!buffer)
1732                         return NULL;
1733                 if (type == required_type) {
1734                         *size = isize;
1735                         if (actual_oid_return)
1736                                 oidcpy(actual_oid_return, &actual_oid);
1737                         return buffer;
1738                 }
1739                 /* Handle references */
1740                 else if (type == OBJ_COMMIT)
1741                         ref_type = "tree ";
1742                 else if (type == OBJ_TAG)
1743                         ref_type = "object ";
1744                 else {
1745                         free(buffer);
1746                         return NULL;
1747                 }
1748                 ref_length = strlen(ref_type);
1749
1750                 if (ref_length + the_hash_algo->hexsz > isize ||
1751                     memcmp(buffer, ref_type, ref_length) ||
1752                     get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1753                         free(buffer);
1754                         return NULL;
1755                 }
1756                 free(buffer);
1757                 /* Now we have the ID of the referred-to object in
1758                  * actual_oid.  Check again. */
1759         }
1760 }
1761
1762 static void write_object_file_prepare(const struct git_hash_algo *algo,
1763                                       const void *buf, unsigned long len,
1764                                       const char *type, struct object_id *oid,
1765                                       char *hdr, int *hdrlen)
1766 {
1767         git_hash_ctx c;
1768
1769         /* Generate the header */
1770         *hdrlen = xsnprintf(hdr, *hdrlen, "%s %"PRIuMAX , type, (uintmax_t)len)+1;
1771
1772         /* Sha1.. */
1773         algo->init_fn(&c);
1774         algo->update_fn(&c, hdr, *hdrlen);
1775         algo->update_fn(&c, buf, len);
1776         algo->final_oid_fn(oid, &c);
1777 }
1778
1779 /*
1780  * Move the just written object into its final resting place.
1781  */
1782 int finalize_object_file(const char *tmpfile, const char *filename)
1783 {
1784         int ret = 0;
1785
1786         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1787                 goto try_rename;
1788         else if (link(tmpfile, filename))
1789                 ret = errno;
1790
1791         /*
1792          * Coda hack - coda doesn't like cross-directory links,
1793          * so we fall back to a rename, which will mean that it
1794          * won't be able to check collisions, but that's not a
1795          * big deal.
1796          *
1797          * The same holds for FAT formatted media.
1798          *
1799          * When this succeeds, we just return.  We have nothing
1800          * left to unlink.
1801          */
1802         if (ret && ret != EEXIST) {
1803         try_rename:
1804                 if (!rename(tmpfile, filename))
1805                         goto out;
1806                 ret = errno;
1807         }
1808         unlink_or_warn(tmpfile);
1809         if (ret) {
1810                 if (ret != EEXIST) {
1811                         return error_errno(_("unable to write file %s"), filename);
1812                 }
1813                 /* FIXME!!! Collision check here ? */
1814         }
1815
1816 out:
1817         if (adjust_shared_perm(filename))
1818                 return error(_("unable to set permission to '%s'"), filename);
1819         return 0;
1820 }
1821
1822 static int write_buffer(int fd, const void *buf, size_t len)
1823 {
1824         if (write_in_full(fd, buf, len) < 0)
1825                 return error_errno(_("file write error"));
1826         return 0;
1827 }
1828
1829 int hash_object_file(const struct git_hash_algo *algo, const void *buf,
1830                      unsigned long len, const char *type,
1831                      struct object_id *oid)
1832 {
1833         char hdr[MAX_HEADER_LEN];
1834         int hdrlen = sizeof(hdr);
1835         write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
1836         return 0;
1837 }
1838
1839 /* Finalize a file on disk, and close it. */
1840 static void close_loose_object(int fd)
1841 {
1842         if (fsync_object_files)
1843                 fsync_or_die(fd, "loose object file");
1844         if (close(fd) != 0)
1845                 die_errno(_("error when closing loose object file"));
1846 }
1847
1848 /* Size of directory component, including the ending '/' */
1849 static inline int directory_size(const char *filename)
1850 {
1851         const char *s = strrchr(filename, '/');
1852         if (!s)
1853                 return 0;
1854         return s - filename + 1;
1855 }
1856
1857 /*
1858  * This creates a temporary file in the same directory as the final
1859  * 'filename'
1860  *
1861  * We want to avoid cross-directory filename renames, because those
1862  * can have problems on various filesystems (FAT, NFS, Coda).
1863  */
1864 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1865 {
1866         int fd, dirlen = directory_size(filename);
1867
1868         strbuf_reset(tmp);
1869         strbuf_add(tmp, filename, dirlen);
1870         strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1871         fd = git_mkstemp_mode(tmp->buf, 0444);
1872         if (fd < 0 && dirlen && errno == ENOENT) {
1873                 /*
1874                  * Make sure the directory exists; note that the contents
1875                  * of the buffer are undefined after mkstemp returns an
1876                  * error, so we have to rewrite the whole buffer from
1877                  * scratch.
1878                  */
1879                 strbuf_reset(tmp);
1880                 strbuf_add(tmp, filename, dirlen - 1);
1881                 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1882                         return -1;
1883                 if (adjust_shared_perm(tmp->buf))
1884                         return -1;
1885
1886                 /* Try again */
1887                 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1888                 fd = git_mkstemp_mode(tmp->buf, 0444);
1889         }
1890         return fd;
1891 }
1892
1893 static int write_loose_object(const struct object_id *oid, char *hdr,
1894                               int hdrlen, const void *buf, unsigned long len,
1895                               time_t mtime)
1896 {
1897         int fd, ret;
1898         unsigned char compressed[4096];
1899         git_zstream stream;
1900         git_hash_ctx c;
1901         struct object_id parano_oid;
1902         static struct strbuf tmp_file = STRBUF_INIT;
1903         static struct strbuf filename = STRBUF_INIT;
1904
1905         loose_object_path(the_repository, &filename, oid);
1906
1907         fd = create_tmpfile(&tmp_file, filename.buf);
1908         if (fd < 0) {
1909                 if (errno == EACCES)
1910                         return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1911                 else
1912                         return error_errno(_("unable to create temporary file"));
1913         }
1914
1915         /* Set it up */
1916         git_deflate_init(&stream, zlib_compression_level);
1917         stream.next_out = compressed;
1918         stream.avail_out = sizeof(compressed);
1919         the_hash_algo->init_fn(&c);
1920
1921         /* First header.. */
1922         stream.next_in = (unsigned char *)hdr;
1923         stream.avail_in = hdrlen;
1924         while (git_deflate(&stream, 0) == Z_OK)
1925                 ; /* nothing */
1926         the_hash_algo->update_fn(&c, hdr, hdrlen);
1927
1928         /* Then the data itself.. */
1929         stream.next_in = (void *)buf;
1930         stream.avail_in = len;
1931         do {
1932                 unsigned char *in0 = stream.next_in;
1933                 ret = git_deflate(&stream, Z_FINISH);
1934                 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1935                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1936                         die(_("unable to write loose object file"));
1937                 stream.next_out = compressed;
1938                 stream.avail_out = sizeof(compressed);
1939         } while (ret == Z_OK);
1940
1941         if (ret != Z_STREAM_END)
1942                 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1943                     ret);
1944         ret = git_deflate_end_gently(&stream);
1945         if (ret != Z_OK)
1946                 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1947                     ret);
1948         the_hash_algo->final_oid_fn(&parano_oid, &c);
1949         if (!oideq(oid, &parano_oid))
1950                 die(_("confused by unstable object source data for %s"),
1951                     oid_to_hex(oid));
1952
1953         close_loose_object(fd);
1954
1955         if (mtime) {
1956                 struct utimbuf utb;
1957                 utb.actime = mtime;
1958                 utb.modtime = mtime;
1959                 if (utime(tmp_file.buf, &utb) < 0)
1960                         warning_errno(_("failed utime() on %s"), tmp_file.buf);
1961         }
1962
1963         return finalize_object_file(tmp_file.buf, filename.buf);
1964 }
1965
1966 static int freshen_loose_object(const struct object_id *oid)
1967 {
1968         return check_and_freshen(oid, 1);
1969 }
1970
1971 static int freshen_packed_object(const struct object_id *oid)
1972 {
1973         struct pack_entry e;
1974         if (!find_pack_entry(the_repository, oid, &e))
1975                 return 0;
1976         if (e.p->freshened)
1977                 return 1;
1978         if (!freshen_file(e.p->pack_name))
1979                 return 0;
1980         e.p->freshened = 1;
1981         return 1;
1982 }
1983
1984 int write_object_file(const void *buf, unsigned long len, const char *type,
1985                       struct object_id *oid)
1986 {
1987         char hdr[MAX_HEADER_LEN];
1988         int hdrlen = sizeof(hdr);
1989
1990         /* Normally if we have it in the pack then we do not bother writing
1991          * it out into .git/objects/??/?{38} file.
1992          */
1993         write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
1994                                   &hdrlen);
1995         if (freshen_packed_object(oid) || freshen_loose_object(oid))
1996                 return 0;
1997         return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1998 }
1999
2000 int hash_object_file_literally(const void *buf, unsigned long len,
2001                                const char *type, struct object_id *oid,
2002                                unsigned flags)
2003 {
2004         char *header;
2005         int hdrlen, status = 0;
2006
2007         /* type string, SP, %lu of the length plus NUL must fit this */
2008         hdrlen = strlen(type) + MAX_HEADER_LEN;
2009         header = xmalloc(hdrlen);
2010         write_object_file_prepare(the_hash_algo, buf, len, type, oid, header,
2011                                   &hdrlen);
2012
2013         if (!(flags & HASH_WRITE_OBJECT))
2014                 goto cleanup;
2015         if (freshen_packed_object(oid) || freshen_loose_object(oid))
2016                 goto cleanup;
2017         status = write_loose_object(oid, header, hdrlen, buf, len, 0);
2018
2019 cleanup:
2020         free(header);
2021         return status;
2022 }
2023
2024 int force_object_loose(const struct object_id *oid, time_t mtime)
2025 {
2026         void *buf;
2027         unsigned long len;
2028         enum object_type type;
2029         char hdr[MAX_HEADER_LEN];
2030         int hdrlen;
2031         int ret;
2032
2033         if (has_loose_object(oid))
2034                 return 0;
2035         buf = read_object(the_repository, oid, &type, &len);
2036         if (!buf)
2037                 return error(_("cannot read object for %s"), oid_to_hex(oid));
2038         hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %"PRIuMAX , type_name(type), (uintmax_t)len) + 1;
2039         ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
2040         free(buf);
2041
2042         return ret;
2043 }
2044
2045 int has_object(struct repository *r, const struct object_id *oid,
2046                unsigned flags)
2047 {
2048         int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2049         unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2050                 (quick ? OBJECT_INFO_QUICK : 0);
2051
2052         if (!startup_info->have_repository)
2053                 return 0;
2054         return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2055 }
2056
2057 int repo_has_object_file_with_flags(struct repository *r,
2058                                     const struct object_id *oid, int flags)
2059 {
2060         if (!startup_info->have_repository)
2061                 return 0;
2062         return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2063 }
2064
2065 int repo_has_object_file(struct repository *r,
2066                          const struct object_id *oid)
2067 {
2068         return repo_has_object_file_with_flags(r, oid, 0);
2069 }
2070
2071 static void check_tree(const void *buf, size_t size)
2072 {
2073         struct tree_desc desc;
2074         struct name_entry entry;
2075
2076         init_tree_desc(&desc, buf, size);
2077         while (tree_entry(&desc, &entry))
2078                 /* do nothing
2079                  * tree_entry() will die() on malformed entries */
2080                 ;
2081 }
2082
2083 static void check_commit(const void *buf, size_t size)
2084 {
2085         struct commit c;
2086         memset(&c, 0, sizeof(c));
2087         if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2088                 die(_("corrupt commit"));
2089 }
2090
2091 static void check_tag(const void *buf, size_t size)
2092 {
2093         struct tag t;
2094         memset(&t, 0, sizeof(t));
2095         if (parse_tag_buffer(the_repository, &t, buf, size))
2096                 die(_("corrupt tag"));
2097 }
2098
2099 static int index_mem(struct index_state *istate,
2100                      struct object_id *oid, void *buf, size_t size,
2101                      enum object_type type,
2102                      const char *path, unsigned flags)
2103 {
2104         int ret, re_allocated = 0;
2105         int write_object = flags & HASH_WRITE_OBJECT;
2106
2107         if (!type)
2108                 type = OBJ_BLOB;
2109
2110         /*
2111          * Convert blobs to git internal format
2112          */
2113         if ((type == OBJ_BLOB) && path) {
2114                 struct strbuf nbuf = STRBUF_INIT;
2115                 if (convert_to_git(istate, path, buf, size, &nbuf,
2116                                    get_conv_flags(flags))) {
2117                         buf = strbuf_detach(&nbuf, &size);
2118                         re_allocated = 1;
2119                 }
2120         }
2121         if (flags & HASH_FORMAT_CHECK) {
2122                 if (type == OBJ_TREE)
2123                         check_tree(buf, size);
2124                 if (type == OBJ_COMMIT)
2125                         check_commit(buf, size);
2126                 if (type == OBJ_TAG)
2127                         check_tag(buf, size);
2128         }
2129
2130         if (write_object)
2131                 ret = write_object_file(buf, size, type_name(type), oid);
2132         else
2133                 ret = hash_object_file(the_hash_algo, buf, size,
2134                                        type_name(type), oid);
2135         if (re_allocated)
2136                 free(buf);
2137         return ret;
2138 }
2139
2140 static int index_stream_convert_blob(struct index_state *istate,
2141                                      struct object_id *oid,
2142                                      int fd,
2143                                      const char *path,
2144                                      unsigned flags)
2145 {
2146         int ret;
2147         const int write_object = flags & HASH_WRITE_OBJECT;
2148         struct strbuf sbuf = STRBUF_INIT;
2149
2150         assert(path);
2151         assert(would_convert_to_git_filter_fd(istate, path));
2152
2153         convert_to_git_filter_fd(istate, path, fd, &sbuf,
2154                                  get_conv_flags(flags));
2155
2156         if (write_object)
2157                 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
2158                                         oid);
2159         else
2160                 ret = hash_object_file(the_hash_algo, sbuf.buf, sbuf.len,
2161                                        type_name(OBJ_BLOB), oid);
2162         strbuf_release(&sbuf);
2163         return ret;
2164 }
2165
2166 static int index_pipe(struct index_state *istate, struct object_id *oid,
2167                       int fd, enum object_type type,
2168                       const char *path, unsigned flags)
2169 {
2170         struct strbuf sbuf = STRBUF_INIT;
2171         int ret;
2172
2173         if (strbuf_read(&sbuf, fd, 4096) >= 0)
2174                 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2175         else
2176                 ret = -1;
2177         strbuf_release(&sbuf);
2178         return ret;
2179 }
2180
2181 #define SMALL_FILE_SIZE (32*1024)
2182
2183 static int index_core(struct index_state *istate,
2184                       struct object_id *oid, int fd, size_t size,
2185                       enum object_type type, const char *path,
2186                       unsigned flags)
2187 {
2188         int ret;
2189
2190         if (!size) {
2191                 ret = index_mem(istate, oid, "", size, type, path, flags);
2192         } else if (size <= SMALL_FILE_SIZE) {
2193                 char *buf = xmalloc(size);
2194                 ssize_t read_result = read_in_full(fd, buf, size);
2195                 if (read_result < 0)
2196                         ret = error_errno(_("read error while indexing %s"),
2197                                           path ? path : "<unknown>");
2198                 else if (read_result != size)
2199                         ret = error(_("short read while indexing %s"),
2200                                     path ? path : "<unknown>");
2201                 else
2202                         ret = index_mem(istate, oid, buf, size, type, path, flags);
2203                 free(buf);
2204         } else {
2205                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2206                 ret = index_mem(istate, oid, buf, size, type, path, flags);
2207                 munmap(buf, size);
2208         }
2209         return ret;
2210 }
2211
2212 /*
2213  * This creates one packfile per large blob unless bulk-checkin
2214  * machinery is "plugged".
2215  *
2216  * This also bypasses the usual "convert-to-git" dance, and that is on
2217  * purpose. We could write a streaming version of the converting
2218  * functions and insert that before feeding the data to fast-import
2219  * (or equivalent in-core API described above). However, that is
2220  * somewhat complicated, as we do not know the size of the filter
2221  * result, which we need to know beforehand when writing a git object.
2222  * Since the primary motivation for trying to stream from the working
2223  * tree file and to avoid mmaping it in core is to deal with large
2224  * binary blobs, they generally do not want to get any conversion, and
2225  * callers should avoid this code path when filters are requested.
2226  */
2227 static int index_stream(struct object_id *oid, int fd, size_t size,
2228                         enum object_type type, const char *path,
2229                         unsigned flags)
2230 {
2231         return index_bulk_checkin(oid, fd, size, type, path, flags);
2232 }
2233
2234 int index_fd(struct index_state *istate, struct object_id *oid,
2235              int fd, struct stat *st,
2236              enum object_type type, const char *path, unsigned flags)
2237 {
2238         int ret;
2239
2240         /*
2241          * Call xsize_t() only when needed to avoid potentially unnecessary
2242          * die() for large files.
2243          */
2244         if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2245                 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2246         else if (!S_ISREG(st->st_mode))
2247                 ret = index_pipe(istate, oid, fd, type, path, flags);
2248         else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2249                  (path && would_convert_to_git(istate, path)))
2250                 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2251                                  type, path, flags);
2252         else
2253                 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2254                                    flags);
2255         close(fd);
2256         return ret;
2257 }
2258
2259 int index_path(struct index_state *istate, struct object_id *oid,
2260                const char *path, struct stat *st, unsigned flags)
2261 {
2262         int fd;
2263         struct strbuf sb = STRBUF_INIT;
2264         int rc = 0;
2265
2266         switch (st->st_mode & S_IFMT) {
2267         case S_IFREG:
2268                 fd = open(path, O_RDONLY);
2269                 if (fd < 0)
2270                         return error_errno("open(\"%s\")", path);
2271                 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2272                         return error(_("%s: failed to insert into database"),
2273                                      path);
2274                 break;
2275         case S_IFLNK:
2276                 if (strbuf_readlink(&sb, path, st->st_size))
2277                         return error_errno("readlink(\"%s\")", path);
2278                 if (!(flags & HASH_WRITE_OBJECT))
2279                         hash_object_file(the_hash_algo, sb.buf, sb.len,
2280                                          blob_type, oid);
2281                 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2282                         rc = error(_("%s: failed to insert into database"), path);
2283                 strbuf_release(&sb);
2284                 break;
2285         case S_IFDIR:
2286                 return resolve_gitlink_ref(path, "HEAD", oid);
2287         default:
2288                 return error(_("%s: unsupported file type"), path);
2289         }
2290         return rc;
2291 }
2292
2293 int read_pack_header(int fd, struct pack_header *header)
2294 {
2295         if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2296                 /* "eof before pack header was fully read" */
2297                 return PH_ERROR_EOF;
2298
2299         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2300                 /* "protocol error (pack signature mismatch detected)" */
2301                 return PH_ERROR_PACK_SIGNATURE;
2302         if (!pack_version_ok(header->hdr_version))
2303                 /* "protocol error (pack version unsupported)" */
2304                 return PH_ERROR_PROTOCOL;
2305         return 0;
2306 }
2307
2308 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2309 {
2310         enum object_type type = oid_object_info(the_repository, oid, NULL);
2311         if (type < 0)
2312                 die(_("%s is not a valid object"), oid_to_hex(oid));
2313         if (type != expect)
2314                 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2315                     type_name(expect));
2316 }
2317
2318 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2319                                 struct strbuf *path,
2320                                 each_loose_object_fn obj_cb,
2321                                 each_loose_cruft_fn cruft_cb,
2322                                 each_loose_subdir_fn subdir_cb,
2323                                 void *data)
2324 {
2325         size_t origlen, baselen;
2326         DIR *dir;
2327         struct dirent *de;
2328         int r = 0;
2329         struct object_id oid;
2330
2331         if (subdir_nr > 0xff)
2332                 BUG("invalid loose object subdirectory: %x", subdir_nr);
2333
2334         origlen = path->len;
2335         strbuf_complete(path, '/');
2336         strbuf_addf(path, "%02x", subdir_nr);
2337
2338         dir = opendir(path->buf);
2339         if (!dir) {
2340                 if (errno != ENOENT)
2341                         r = error_errno(_("unable to open %s"), path->buf);
2342                 strbuf_setlen(path, origlen);
2343                 return r;
2344         }
2345
2346         oid.hash[0] = subdir_nr;
2347         strbuf_addch(path, '/');
2348         baselen = path->len;
2349
2350         while ((de = readdir_skip_dot_and_dotdot(dir))) {
2351                 size_t namelen;
2352
2353                 namelen = strlen(de->d_name);
2354                 strbuf_setlen(path, baselen);
2355                 strbuf_add(path, de->d_name, namelen);
2356                 if (namelen == the_hash_algo->hexsz - 2 &&
2357                     !hex_to_bytes(oid.hash + 1, de->d_name,
2358                                   the_hash_algo->rawsz - 1)) {
2359                         oid_set_algo(&oid, the_hash_algo);
2360                         if (obj_cb) {
2361                                 r = obj_cb(&oid, path->buf, data);
2362                                 if (r)
2363                                         break;
2364                         }
2365                         continue;
2366                 }
2367
2368                 if (cruft_cb) {
2369                         r = cruft_cb(de->d_name, path->buf, data);
2370                         if (r)
2371                                 break;
2372                 }
2373         }
2374         closedir(dir);
2375
2376         strbuf_setlen(path, baselen - 1);
2377         if (!r && subdir_cb)
2378                 r = subdir_cb(subdir_nr, path->buf, data);
2379
2380         strbuf_setlen(path, origlen);
2381
2382         return r;
2383 }
2384
2385 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2386                             each_loose_object_fn obj_cb,
2387                             each_loose_cruft_fn cruft_cb,
2388                             each_loose_subdir_fn subdir_cb,
2389                             void *data)
2390 {
2391         int r = 0;
2392         int i;
2393
2394         for (i = 0; i < 256; i++) {
2395                 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2396                                                 subdir_cb, data);
2397                 if (r)
2398                         break;
2399         }
2400
2401         return r;
2402 }
2403
2404 int for_each_loose_file_in_objdir(const char *path,
2405                                   each_loose_object_fn obj_cb,
2406                                   each_loose_cruft_fn cruft_cb,
2407                                   each_loose_subdir_fn subdir_cb,
2408                                   void *data)
2409 {
2410         struct strbuf buf = STRBUF_INIT;
2411         int r;
2412
2413         strbuf_addstr(&buf, path);
2414         r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2415                                               subdir_cb, data);
2416         strbuf_release(&buf);
2417
2418         return r;
2419 }
2420
2421 int for_each_loose_object(each_loose_object_fn cb, void *data,
2422                           enum for_each_object_flags flags)
2423 {
2424         struct object_directory *odb;
2425
2426         prepare_alt_odb(the_repository);
2427         for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2428                 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2429                                                       NULL, data);
2430                 if (r)
2431                         return r;
2432
2433                 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2434                         break;
2435         }
2436
2437         return 0;
2438 }
2439
2440 static int append_loose_object(const struct object_id *oid, const char *path,
2441                                void *data)
2442 {
2443         oid_array_append(data, oid);
2444         return 0;
2445 }
2446
2447 struct oid_array *odb_loose_cache(struct object_directory *odb,
2448                                   const struct object_id *oid)
2449 {
2450         int subdir_nr = oid->hash[0];
2451         struct strbuf buf = STRBUF_INIT;
2452
2453         if (subdir_nr < 0 ||
2454             subdir_nr >= ARRAY_SIZE(odb->loose_objects_subdir_seen))
2455                 BUG("subdir_nr out of range");
2456
2457         if (odb->loose_objects_subdir_seen[subdir_nr])
2458                 return &odb->loose_objects_cache[subdir_nr];
2459
2460         strbuf_addstr(&buf, odb->path);
2461         for_each_file_in_obj_subdir(subdir_nr, &buf,
2462                                     append_loose_object,
2463                                     NULL, NULL,
2464                                     &odb->loose_objects_cache[subdir_nr]);
2465         odb->loose_objects_subdir_seen[subdir_nr] = 1;
2466         strbuf_release(&buf);
2467         return &odb->loose_objects_cache[subdir_nr];
2468 }
2469
2470 void odb_clear_loose_cache(struct object_directory *odb)
2471 {
2472         int i;
2473
2474         for (i = 0; i < ARRAY_SIZE(odb->loose_objects_cache); i++)
2475                 oid_array_clear(&odb->loose_objects_cache[i]);
2476         memset(&odb->loose_objects_subdir_seen, 0,
2477                sizeof(odb->loose_objects_subdir_seen));
2478 }
2479
2480 static int check_stream_oid(git_zstream *stream,
2481                             const char *hdr,
2482                             unsigned long size,
2483                             const char *path,
2484                             const struct object_id *expected_oid)
2485 {
2486         git_hash_ctx c;
2487         struct object_id real_oid;
2488         unsigned char buf[4096];
2489         unsigned long total_read;
2490         int status = Z_OK;
2491
2492         the_hash_algo->init_fn(&c);
2493         the_hash_algo->update_fn(&c, hdr, stream->total_out);
2494
2495         /*
2496          * We already read some bytes into hdr, but the ones up to the NUL
2497          * do not count against the object's content size.
2498          */
2499         total_read = stream->total_out - strlen(hdr) - 1;
2500
2501         /*
2502          * This size comparison must be "<=" to read the final zlib packets;
2503          * see the comment in unpack_loose_rest for details.
2504          */
2505         while (total_read <= size &&
2506                (status == Z_OK ||
2507                 (status == Z_BUF_ERROR && !stream->avail_out))) {
2508                 stream->next_out = buf;
2509                 stream->avail_out = sizeof(buf);
2510                 if (size - total_read < stream->avail_out)
2511                         stream->avail_out = size - total_read;
2512                 status = git_inflate(stream, Z_FINISH);
2513                 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2514                 total_read += stream->next_out - buf;
2515         }
2516         git_inflate_end(stream);
2517
2518         if (status != Z_STREAM_END) {
2519                 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2520                 return -1;
2521         }
2522         if (stream->avail_in) {
2523                 error(_("garbage at end of loose object '%s'"),
2524                       oid_to_hex(expected_oid));
2525                 return -1;
2526         }
2527
2528         the_hash_algo->final_oid_fn(&real_oid, &c);
2529         if (!oideq(expected_oid, &real_oid)) {
2530                 error(_("hash mismatch for %s (expected %s)"), path,
2531                       oid_to_hex(expected_oid));
2532                 return -1;
2533         }
2534
2535         return 0;
2536 }
2537
2538 int read_loose_object(const char *path,
2539                       const struct object_id *expected_oid,
2540                       enum object_type *type,
2541                       unsigned long *size,
2542                       void **contents)
2543 {
2544         int ret = -1;
2545         void *map = NULL;
2546         unsigned long mapsize;
2547         git_zstream stream;
2548         char hdr[MAX_HEADER_LEN];
2549
2550         *contents = NULL;
2551
2552         map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2553         if (!map) {
2554                 error_errno(_("unable to mmap %s"), path);
2555                 goto out;
2556         }
2557
2558         if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2559                 error(_("unable to unpack header of %s"), path);
2560                 goto out;
2561         }
2562
2563         *type = parse_loose_header(hdr, size);
2564         if (*type < 0) {
2565                 error(_("unable to parse header of %s"), path);
2566                 git_inflate_end(&stream);
2567                 goto out;
2568         }
2569
2570         if (*type == OBJ_BLOB && *size > big_file_threshold) {
2571                 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2572                         goto out;
2573         } else {
2574                 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2575                 if (!*contents) {
2576                         error(_("unable to unpack contents of %s"), path);
2577                         git_inflate_end(&stream);
2578                         goto out;
2579                 }
2580                 if (check_object_signature(the_repository, expected_oid,
2581                                            *contents, *size,
2582                                            type_name(*type))) {
2583                         error(_("hash mismatch for %s (expected %s)"), path,
2584                               oid_to_hex(expected_oid));
2585                         free(*contents);
2586                         goto out;
2587                 }
2588         }
2589
2590         ret = 0; /* everything checks out */
2591
2592 out:
2593         if (map)
2594                 munmap(map, mapsize);
2595         return ret;
2596 }