sha1_file.c: move delayed getenv(altdb) back to setup_git_env()
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "config.h"
11 #include "string-list.h"
12 #include "lockfile.h"
13 #include "delta.h"
14 #include "pack.h"
15 #include "blob.h"
16 #include "commit.h"
17 #include "run-command.h"
18 #include "tag.h"
19 #include "tree.h"
20 #include "tree-walk.h"
21 #include "refs.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "streaming.h"
26 #include "dir.h"
27 #include "list.h"
28 #include "mergesort.h"
29 #include "quote.h"
30 #include "packfile.h"
31 #include "fetch-object.h"
32
33 const unsigned char null_sha1[GIT_MAX_RAWSZ];
34 const struct object_id null_oid;
35 const struct object_id empty_tree_oid = {
36 EMPTY_TREE_SHA1_BIN_LITERAL
37 };
38 const struct object_id empty_blob_oid = {
39 EMPTY_BLOB_SHA1_BIN_LITERAL
40 };
41
42 static void git_hash_sha1_init(void *ctx)
43 {
44 git_SHA1_Init((git_SHA_CTX *)ctx);
45 }
46
47 static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
48 {
49 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
50 }
51
52 static void git_hash_sha1_final(unsigned char *hash, void *ctx)
53 {
54 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
55 }
56
57 static void git_hash_unknown_init(void *ctx)
58 {
59 die("trying to init unknown hash");
60 }
61
62 static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
63 {
64 die("trying to update unknown hash");
65 }
66
67 static void git_hash_unknown_final(unsigned char *hash, void *ctx)
68 {
69 die("trying to finalize unknown hash");
70 }
71
72 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
73 {
74 NULL,
75 0x00000000,
76 0,
77 0,
78 0,
79 git_hash_unknown_init,
80 git_hash_unknown_update,
81 git_hash_unknown_final,
82 NULL,
83 NULL,
84 },
85 {
86 "sha-1",
87 /* "sha1", big-endian */
88 0x73686131,
89 sizeof(git_SHA_CTX),
90 GIT_SHA1_RAWSZ,
91 GIT_SHA1_HEXSZ,
92 git_hash_sha1_init,
93 git_hash_sha1_update,
94 git_hash_sha1_final,
95 &empty_tree_oid,
96 &empty_blob_oid,
97 },
98 };
99
100 /*
101 * This is meant to hold a *small* number of objects that you would
102 * want read_sha1_file() to be able to return, but yet you do not want
103 * to write them into the object store (e.g. a browse-only
104 * application).
105 */
106 static struct cached_object {
107 unsigned char sha1[20];
108 enum object_type type;
109 void *buf;
110 unsigned long size;
111 } *cached_objects;
112 static int cached_object_nr, cached_object_alloc;
113
114 static struct cached_object empty_tree = {
115 EMPTY_TREE_SHA1_BIN_LITERAL,
116 OBJ_TREE,
117 "",
118 0
119 };
120
121 static struct cached_object *find_cached_object(const unsigned char *sha1)
122 {
123 int i;
124 struct cached_object *co = cached_objects;
125
126 for (i = 0; i < cached_object_nr; i++, co++) {
127 if (!hashcmp(co->sha1, sha1))
128 return co;
129 }
130 if (!hashcmp(sha1, empty_tree.sha1))
131 return &empty_tree;
132 return NULL;
133 }
134
135
136 static int get_conv_flags(unsigned flags)
137 {
138 if (flags & HASH_RENORMALIZE)
139 return CONV_EOL_RENORMALIZE;
140 else if (flags & HASH_WRITE_OBJECT)
141 return global_conv_flags_eol;
142 else
143 return 0;
144 }
145
146
147 int mkdir_in_gitdir(const char *path)
148 {
149 if (mkdir(path, 0777)) {
150 int saved_errno = errno;
151 struct stat st;
152 struct strbuf sb = STRBUF_INIT;
153
154 if (errno != EEXIST)
155 return -1;
156 /*
157 * Are we looking at a path in a symlinked worktree
158 * whose original repository does not yet have it?
159 * e.g. .git/rr-cache pointing at its original
160 * repository in which the user hasn't performed any
161 * conflict resolution yet?
162 */
163 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
164 strbuf_readlink(&sb, path, st.st_size) ||
165 !is_absolute_path(sb.buf) ||
166 mkdir(sb.buf, 0777)) {
167 strbuf_release(&sb);
168 errno = saved_errno;
169 return -1;
170 }
171 strbuf_release(&sb);
172 }
173 return adjust_shared_perm(path);
174 }
175
176 enum scld_error safe_create_leading_directories(char *path)
177 {
178 char *next_component = path + offset_1st_component(path);
179 enum scld_error ret = SCLD_OK;
180
181 while (ret == SCLD_OK && next_component) {
182 struct stat st;
183 char *slash = next_component, slash_character;
184
185 while (*slash && !is_dir_sep(*slash))
186 slash++;
187
188 if (!*slash)
189 break;
190
191 next_component = slash + 1;
192 while (is_dir_sep(*next_component))
193 next_component++;
194 if (!*next_component)
195 break;
196
197 slash_character = *slash;
198 *slash = '\0';
199 if (!stat(path, &st)) {
200 /* path exists */
201 if (!S_ISDIR(st.st_mode)) {
202 errno = ENOTDIR;
203 ret = SCLD_EXISTS;
204 }
205 } else if (mkdir(path, 0777)) {
206 if (errno == EEXIST &&
207 !stat(path, &st) && S_ISDIR(st.st_mode))
208 ; /* somebody created it since we checked */
209 else if (errno == ENOENT)
210 /*
211 * Either mkdir() failed because
212 * somebody just pruned the containing
213 * directory, or stat() failed because
214 * the file that was in our way was
215 * just removed. Either way, inform
216 * the caller that it might be worth
217 * trying again:
218 */
219 ret = SCLD_VANISHED;
220 else
221 ret = SCLD_FAILED;
222 } else if (adjust_shared_perm(path)) {
223 ret = SCLD_PERMS;
224 }
225 *slash = slash_character;
226 }
227 return ret;
228 }
229
230 enum scld_error safe_create_leading_directories_const(const char *path)
231 {
232 int save_errno;
233 /* path points to cache entries, so xstrdup before messing with it */
234 char *buf = xstrdup(path);
235 enum scld_error result = safe_create_leading_directories(buf);
236
237 save_errno = errno;
238 free(buf);
239 errno = save_errno;
240 return result;
241 }
242
243 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
244 {
245 /*
246 * The number of times we will try to remove empty directories
247 * in the way of path. This is only 1 because if another
248 * process is racily creating directories that conflict with
249 * us, we don't want to fight against them.
250 */
251 int remove_directories_remaining = 1;
252
253 /*
254 * The number of times that we will try to create the
255 * directories containing path. We are willing to attempt this
256 * more than once, because another process could be trying to
257 * clean up empty directories at the same time as we are
258 * trying to create them.
259 */
260 int create_directories_remaining = 3;
261
262 /* A scratch copy of path, filled lazily if we need it: */
263 struct strbuf path_copy = STRBUF_INIT;
264
265 int ret, save_errno;
266
267 /* Sanity check: */
268 assert(*path);
269
270 retry_fn:
271 ret = fn(path, cb);
272 save_errno = errno;
273 if (!ret)
274 goto out;
275
276 if (errno == EISDIR && remove_directories_remaining-- > 0) {
277 /*
278 * A directory is in the way. Maybe it is empty; try
279 * to remove it:
280 */
281 if (!path_copy.len)
282 strbuf_addstr(&path_copy, path);
283
284 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
285 goto retry_fn;
286 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
287 /*
288 * Maybe the containing directory didn't exist, or
289 * maybe it was just deleted by a process that is
290 * racing with us to clean up empty directories. Try
291 * to create it:
292 */
293 enum scld_error scld_result;
294
295 if (!path_copy.len)
296 strbuf_addstr(&path_copy, path);
297
298 do {
299 scld_result = safe_create_leading_directories(path_copy.buf);
300 if (scld_result == SCLD_OK)
301 goto retry_fn;
302 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
303 }
304
305 out:
306 strbuf_release(&path_copy);
307 errno = save_errno;
308 return ret;
309 }
310
311 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
312 {
313 int i;
314 for (i = 0; i < 20; i++) {
315 static char hex[] = "0123456789abcdef";
316 unsigned int val = sha1[i];
317 strbuf_addch(buf, hex[val >> 4]);
318 strbuf_addch(buf, hex[val & 0xf]);
319 if (!i)
320 strbuf_addch(buf, '/');
321 }
322 }
323
324 void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
325 {
326 strbuf_addstr(buf, get_object_directory());
327 strbuf_addch(buf, '/');
328 fill_sha1_path(buf, sha1);
329 }
330
331 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
332 {
333 strbuf_setlen(&alt->scratch, alt->base_len);
334 return &alt->scratch;
335 }
336
337 static const char *alt_sha1_path(struct alternate_object_database *alt,
338 const unsigned char *sha1)
339 {
340 struct strbuf *buf = alt_scratch_buf(alt);
341 fill_sha1_path(buf, sha1);
342 return buf->buf;
343 }
344
345 struct alternate_object_database *alt_odb_list;
346 static struct alternate_object_database **alt_odb_tail;
347
348 /*
349 * Return non-zero iff the path is usable as an alternate object database.
350 */
351 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
352 {
353 struct alternate_object_database *alt;
354
355 /* Detect cases where alternate disappeared */
356 if (!is_directory(path->buf)) {
357 error("object directory %s does not exist; "
358 "check .git/objects/info/alternates.",
359 path->buf);
360 return 0;
361 }
362
363 /*
364 * Prevent the common mistake of listing the same
365 * thing twice, or object directory itself.
366 */
367 for (alt = alt_odb_list; alt; alt = alt->next) {
368 if (!fspathcmp(path->buf, alt->path))
369 return 0;
370 }
371 if (!fspathcmp(path->buf, normalized_objdir))
372 return 0;
373
374 return 1;
375 }
376
377 /*
378 * Prepare alternate object database registry.
379 *
380 * The variable alt_odb_list points at the list of struct
381 * alternate_object_database. The elements on this list come from
382 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
383 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
384 * whose contents is similar to that environment variable but can be
385 * LF separated. Its base points at a statically allocated buffer that
386 * contains "/the/directory/corresponding/to/.git/objects/...", while
387 * its name points just after the slash at the end of ".git/objects/"
388 * in the example above, and has enough space to hold 40-byte hex
389 * SHA1, an extra slash for the first level indirection, and the
390 * terminating NUL.
391 */
392 static void read_info_alternates(const char * relative_base, int depth);
393 static int link_alt_odb_entry(const char *entry, const char *relative_base,
394 int depth, const char *normalized_objdir)
395 {
396 struct alternate_object_database *ent;
397 struct strbuf pathbuf = STRBUF_INIT;
398
399 if (!is_absolute_path(entry) && relative_base) {
400 strbuf_realpath(&pathbuf, relative_base, 1);
401 strbuf_addch(&pathbuf, '/');
402 }
403 strbuf_addstr(&pathbuf, entry);
404
405 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
406 error("unable to normalize alternate object path: %s",
407 pathbuf.buf);
408 strbuf_release(&pathbuf);
409 return -1;
410 }
411
412 /*
413 * The trailing slash after the directory name is given by
414 * this function at the end. Remove duplicates.
415 */
416 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
417 strbuf_setlen(&pathbuf, pathbuf.len - 1);
418
419 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
420 strbuf_release(&pathbuf);
421 return -1;
422 }
423
424 ent = alloc_alt_odb(pathbuf.buf);
425
426 /* add the alternate entry */
427 *alt_odb_tail = ent;
428 alt_odb_tail = &(ent->next);
429 ent->next = NULL;
430
431 /* recursively add alternates */
432 read_info_alternates(pathbuf.buf, depth + 1);
433
434 strbuf_release(&pathbuf);
435 return 0;
436 }
437
438 static const char *parse_alt_odb_entry(const char *string,
439 int sep,
440 struct strbuf *out)
441 {
442 const char *end;
443
444 strbuf_reset(out);
445
446 if (*string == '#') {
447 /* comment; consume up to next separator */
448 end = strchrnul(string, sep);
449 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
450 /*
451 * quoted path; unquote_c_style has copied the
452 * data for us and set "end". Broken quoting (e.g.,
453 * an entry that doesn't end with a quote) falls
454 * back to the unquoted case below.
455 */
456 } else {
457 /* normal, unquoted path */
458 end = strchrnul(string, sep);
459 strbuf_add(out, string, end - string);
460 }
461
462 if (*end)
463 end++;
464 return end;
465 }
466
467 static void link_alt_odb_entries(const char *alt, int sep,
468 const char *relative_base, int depth)
469 {
470 struct strbuf objdirbuf = STRBUF_INIT;
471 struct strbuf entry = STRBUF_INIT;
472
473 if (!alt || !*alt)
474 return;
475
476 if (depth > 5) {
477 error("%s: ignoring alternate object stores, nesting too deep.",
478 relative_base);
479 return;
480 }
481
482 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
483 if (strbuf_normalize_path(&objdirbuf) < 0)
484 die("unable to normalize object directory: %s",
485 objdirbuf.buf);
486
487 while (*alt) {
488 alt = parse_alt_odb_entry(alt, sep, &entry);
489 if (!entry.len)
490 continue;
491 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
492 }
493 strbuf_release(&entry);
494 strbuf_release(&objdirbuf);
495 }
496
497 static void read_info_alternates(const char * relative_base, int depth)
498 {
499 char *path;
500 struct strbuf buf = STRBUF_INIT;
501
502 path = xstrfmt("%s/info/alternates", relative_base);
503 if (strbuf_read_file(&buf, path, 1024) < 0) {
504 warn_on_fopen_errors(path);
505 free(path);
506 return;
507 }
508
509 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
510 strbuf_release(&buf);
511 free(path);
512 }
513
514 struct alternate_object_database *alloc_alt_odb(const char *dir)
515 {
516 struct alternate_object_database *ent;
517
518 FLEX_ALLOC_STR(ent, path, dir);
519 strbuf_init(&ent->scratch, 0);
520 strbuf_addf(&ent->scratch, "%s/", dir);
521 ent->base_len = ent->scratch.len;
522
523 return ent;
524 }
525
526 void add_to_alternates_file(const char *reference)
527 {
528 struct lock_file lock = LOCK_INIT;
529 char *alts = git_pathdup("objects/info/alternates");
530 FILE *in, *out;
531 int found = 0;
532
533 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
534 out = fdopen_lock_file(&lock, "w");
535 if (!out)
536 die_errno("unable to fdopen alternates lockfile");
537
538 in = fopen(alts, "r");
539 if (in) {
540 struct strbuf line = STRBUF_INIT;
541
542 while (strbuf_getline(&line, in) != EOF) {
543 if (!strcmp(reference, line.buf)) {
544 found = 1;
545 break;
546 }
547 fprintf_or_die(out, "%s\n", line.buf);
548 }
549
550 strbuf_release(&line);
551 fclose(in);
552 }
553 else if (errno != ENOENT)
554 die_errno("unable to read alternates file");
555
556 if (found) {
557 rollback_lock_file(&lock);
558 } else {
559 fprintf_or_die(out, "%s\n", reference);
560 if (commit_lock_file(&lock))
561 die_errno("unable to move new alternates file into place");
562 if (alt_odb_tail)
563 link_alt_odb_entries(reference, '\n', NULL, 0);
564 }
565 free(alts);
566 }
567
568 void add_to_alternates_memory(const char *reference)
569 {
570 /*
571 * Make sure alternates are initialized, or else our entry may be
572 * overwritten when they are.
573 */
574 prepare_alt_odb();
575
576 link_alt_odb_entries(reference, '\n', NULL, 0);
577 }
578
579 /*
580 * Compute the exact path an alternate is at and returns it. In case of
581 * error NULL is returned and the human readable error is added to `err`
582 * `path` may be relative and should point to $GITDIR.
583 * `err` must not be null.
584 */
585 char *compute_alternate_path(const char *path, struct strbuf *err)
586 {
587 char *ref_git = NULL;
588 const char *repo, *ref_git_s;
589 int seen_error = 0;
590
591 ref_git_s = real_path_if_valid(path);
592 if (!ref_git_s) {
593 seen_error = 1;
594 strbuf_addf(err, _("path '%s' does not exist"), path);
595 goto out;
596 } else
597 /*
598 * Beware: read_gitfile(), real_path() and mkpath()
599 * return static buffer
600 */
601 ref_git = xstrdup(ref_git_s);
602
603 repo = read_gitfile(ref_git);
604 if (!repo)
605 repo = read_gitfile(mkpath("%s/.git", ref_git));
606 if (repo) {
607 free(ref_git);
608 ref_git = xstrdup(repo);
609 }
610
611 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
612 char *ref_git_git = mkpathdup("%s/.git", ref_git);
613 free(ref_git);
614 ref_git = ref_git_git;
615 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
616 struct strbuf sb = STRBUF_INIT;
617 seen_error = 1;
618 if (get_common_dir(&sb, ref_git)) {
619 strbuf_addf(err,
620 _("reference repository '%s' as a linked "
621 "checkout is not supported yet."),
622 path);
623 goto out;
624 }
625
626 strbuf_addf(err, _("reference repository '%s' is not a "
627 "local repository."), path);
628 goto out;
629 }
630
631 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
632 strbuf_addf(err, _("reference repository '%s' is shallow"),
633 path);
634 seen_error = 1;
635 goto out;
636 }
637
638 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
639 strbuf_addf(err,
640 _("reference repository '%s' is grafted"),
641 path);
642 seen_error = 1;
643 goto out;
644 }
645
646 out:
647 if (seen_error) {
648 FREE_AND_NULL(ref_git);
649 }
650
651 return ref_git;
652 }
653
654 int foreach_alt_odb(alt_odb_fn fn, void *cb)
655 {
656 struct alternate_object_database *ent;
657 int r = 0;
658
659 prepare_alt_odb();
660 for (ent = alt_odb_list; ent; ent = ent->next) {
661 r = fn(ent, cb);
662 if (r)
663 break;
664 }
665 return r;
666 }
667
668 void prepare_alt_odb(void)
669 {
670 if (alt_odb_tail)
671 return;
672
673 alt_odb_tail = &alt_odb_list;
674 link_alt_odb_entries(the_repository->alternate_db, PATH_SEP, NULL, 0);
675
676 read_info_alternates(get_object_directory(), 0);
677 }
678
679 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
680 static int freshen_file(const char *fn)
681 {
682 struct utimbuf t;
683 t.actime = t.modtime = time(NULL);
684 return !utime(fn, &t);
685 }
686
687 /*
688 * All of the check_and_freshen functions return 1 if the file exists and was
689 * freshened (if freshening was requested), 0 otherwise. If they return
690 * 0, you should not assume that it is safe to skip a write of the object (it
691 * either does not exist on disk, or has a stale mtime and may be subject to
692 * pruning).
693 */
694 int check_and_freshen_file(const char *fn, int freshen)
695 {
696 if (access(fn, F_OK))
697 return 0;
698 if (freshen && !freshen_file(fn))
699 return 0;
700 return 1;
701 }
702
703 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
704 {
705 static struct strbuf buf = STRBUF_INIT;
706
707 strbuf_reset(&buf);
708 sha1_file_name(&buf, sha1);
709
710 return check_and_freshen_file(buf.buf, freshen);
711 }
712
713 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
714 {
715 struct alternate_object_database *alt;
716 prepare_alt_odb();
717 for (alt = alt_odb_list; alt; alt = alt->next) {
718 const char *path = alt_sha1_path(alt, sha1);
719 if (check_and_freshen_file(path, freshen))
720 return 1;
721 }
722 return 0;
723 }
724
725 static int check_and_freshen(const unsigned char *sha1, int freshen)
726 {
727 return check_and_freshen_local(sha1, freshen) ||
728 check_and_freshen_nonlocal(sha1, freshen);
729 }
730
731 int has_loose_object_nonlocal(const unsigned char *sha1)
732 {
733 return check_and_freshen_nonlocal(sha1, 0);
734 }
735
736 static int has_loose_object(const unsigned char *sha1)
737 {
738 return check_and_freshen(sha1, 0);
739 }
740
741 static void mmap_limit_check(size_t length)
742 {
743 static size_t limit = 0;
744 if (!limit) {
745 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
746 if (!limit)
747 limit = SIZE_MAX;
748 }
749 if (length > limit)
750 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
751 (uintmax_t)length, (uintmax_t)limit);
752 }
753
754 void *xmmap_gently(void *start, size_t length,
755 int prot, int flags, int fd, off_t offset)
756 {
757 void *ret;
758
759 mmap_limit_check(length);
760 ret = mmap(start, length, prot, flags, fd, offset);
761 if (ret == MAP_FAILED) {
762 if (!length)
763 return NULL;
764 release_pack_memory(length);
765 ret = mmap(start, length, prot, flags, fd, offset);
766 }
767 return ret;
768 }
769
770 void *xmmap(void *start, size_t length,
771 int prot, int flags, int fd, off_t offset)
772 {
773 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
774 if (ret == MAP_FAILED)
775 die_errno("mmap failed");
776 return ret;
777 }
778
779 /*
780 * With an in-core object data in "map", rehash it to make sure the
781 * object name actually matches "sha1" to detect object corruption.
782 * With "map" == NULL, try reading the object named with "sha1" using
783 * the streaming interface and rehash it to do the same.
784 */
785 int check_sha1_signature(const unsigned char *sha1, void *map,
786 unsigned long size, const char *type)
787 {
788 unsigned char real_sha1[20];
789 enum object_type obj_type;
790 struct git_istream *st;
791 git_SHA_CTX c;
792 char hdr[32];
793 int hdrlen;
794
795 if (map) {
796 hash_sha1_file(map, size, type, real_sha1);
797 return hashcmp(sha1, real_sha1) ? -1 : 0;
798 }
799
800 st = open_istream(sha1, &obj_type, &size, NULL);
801 if (!st)
802 return -1;
803
804 /* Generate the header */
805 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
806
807 /* Sha1.. */
808 git_SHA1_Init(&c);
809 git_SHA1_Update(&c, hdr, hdrlen);
810 for (;;) {
811 char buf[1024 * 16];
812 ssize_t readlen = read_istream(st, buf, sizeof(buf));
813
814 if (readlen < 0) {
815 close_istream(st);
816 return -1;
817 }
818 if (!readlen)
819 break;
820 git_SHA1_Update(&c, buf, readlen);
821 }
822 git_SHA1_Final(real_sha1, &c);
823 close_istream(st);
824 return hashcmp(sha1, real_sha1) ? -1 : 0;
825 }
826
827 int git_open_cloexec(const char *name, int flags)
828 {
829 int fd;
830 static int o_cloexec = O_CLOEXEC;
831
832 fd = open(name, flags | o_cloexec);
833 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
834 /* Try again w/o O_CLOEXEC: the kernel might not support it */
835 o_cloexec &= ~O_CLOEXEC;
836 fd = open(name, flags | o_cloexec);
837 }
838
839 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
840 {
841 static int fd_cloexec = FD_CLOEXEC;
842
843 if (!o_cloexec && 0 <= fd && fd_cloexec) {
844 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
845 int flags = fcntl(fd, F_GETFD);
846 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
847 fd_cloexec = 0;
848 }
849 }
850 #endif
851 return fd;
852 }
853
854 /*
855 * Find "sha1" as a loose object in the local repository or in an alternate.
856 * Returns 0 on success, negative on failure.
857 *
858 * The "path" out-parameter will give the path of the object we found (if any).
859 * Note that it may point to static storage and is only valid until another
860 * call to sha1_file_name(), etc.
861 */
862 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
863 const char **path)
864 {
865 struct alternate_object_database *alt;
866 static struct strbuf buf = STRBUF_INIT;
867
868 strbuf_reset(&buf);
869 sha1_file_name(&buf, sha1);
870 *path = buf.buf;
871
872 if (!lstat(*path, st))
873 return 0;
874
875 prepare_alt_odb();
876 errno = ENOENT;
877 for (alt = alt_odb_list; alt; alt = alt->next) {
878 *path = alt_sha1_path(alt, sha1);
879 if (!lstat(*path, st))
880 return 0;
881 }
882
883 return -1;
884 }
885
886 /*
887 * Like stat_sha1_file(), but actually open the object and return the
888 * descriptor. See the caveats on the "path" parameter above.
889 */
890 static int open_sha1_file(const unsigned char *sha1, const char **path)
891 {
892 int fd;
893 struct alternate_object_database *alt;
894 int most_interesting_errno;
895 static struct strbuf buf = STRBUF_INIT;
896
897 strbuf_reset(&buf);
898 sha1_file_name(&buf, sha1);
899 *path = buf.buf;
900
901 fd = git_open(*path);
902 if (fd >= 0)
903 return fd;
904 most_interesting_errno = errno;
905
906 prepare_alt_odb();
907 for (alt = alt_odb_list; alt; alt = alt->next) {
908 *path = alt_sha1_path(alt, sha1);
909 fd = git_open(*path);
910 if (fd >= 0)
911 return fd;
912 if (most_interesting_errno == ENOENT)
913 most_interesting_errno = errno;
914 }
915 errno = most_interesting_errno;
916 return -1;
917 }
918
919 /*
920 * Map the loose object at "path" if it is not NULL, or the path found by
921 * searching for a loose object named "sha1".
922 */
923 static void *map_sha1_file_1(const char *path,
924 const unsigned char *sha1,
925 unsigned long *size)
926 {
927 void *map;
928 int fd;
929
930 if (path)
931 fd = git_open(path);
932 else
933 fd = open_sha1_file(sha1, &path);
934 map = NULL;
935 if (fd >= 0) {
936 struct stat st;
937
938 if (!fstat(fd, &st)) {
939 *size = xsize_t(st.st_size);
940 if (!*size) {
941 /* mmap() is forbidden on empty files */
942 error("object file %s is empty", path);
943 return NULL;
944 }
945 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
946 }
947 close(fd);
948 }
949 return map;
950 }
951
952 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
953 {
954 return map_sha1_file_1(NULL, sha1, size);
955 }
956
957 static int unpack_sha1_short_header(git_zstream *stream,
958 unsigned char *map, unsigned long mapsize,
959 void *buffer, unsigned long bufsiz)
960 {
961 /* Get the data stream */
962 memset(stream, 0, sizeof(*stream));
963 stream->next_in = map;
964 stream->avail_in = mapsize;
965 stream->next_out = buffer;
966 stream->avail_out = bufsiz;
967
968 git_inflate_init(stream);
969 return git_inflate(stream, 0);
970 }
971
972 int unpack_sha1_header(git_zstream *stream,
973 unsigned char *map, unsigned long mapsize,
974 void *buffer, unsigned long bufsiz)
975 {
976 int status = unpack_sha1_short_header(stream, map, mapsize,
977 buffer, bufsiz);
978
979 if (status < Z_OK)
980 return status;
981
982 /* Make sure we have the terminating NUL */
983 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
984 return -1;
985 return 0;
986 }
987
988 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
989 unsigned long mapsize, void *buffer,
990 unsigned long bufsiz, struct strbuf *header)
991 {
992 int status;
993
994 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
995 if (status < Z_OK)
996 return -1;
997
998 /*
999 * Check if entire header is unpacked in the first iteration.
1000 */
1001 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1002 return 0;
1003
1004 /*
1005 * buffer[0..bufsiz] was not large enough. Copy the partial
1006 * result out to header, and then append the result of further
1007 * reading the stream.
1008 */
1009 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1010 stream->next_out = buffer;
1011 stream->avail_out = bufsiz;
1012
1013 do {
1014 status = git_inflate(stream, 0);
1015 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1016 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1017 return 0;
1018 stream->next_out = buffer;
1019 stream->avail_out = bufsiz;
1020 } while (status != Z_STREAM_END);
1021 return -1;
1022 }
1023
1024 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1025 {
1026 int bytes = strlen(buffer) + 1;
1027 unsigned char *buf = xmallocz(size);
1028 unsigned long n;
1029 int status = Z_OK;
1030
1031 n = stream->total_out - bytes;
1032 if (n > size)
1033 n = size;
1034 memcpy(buf, (char *) buffer + bytes, n);
1035 bytes = n;
1036 if (bytes <= size) {
1037 /*
1038 * The above condition must be (bytes <= size), not
1039 * (bytes < size). In other words, even though we
1040 * expect no more output and set avail_out to zero,
1041 * the input zlib stream may have bytes that express
1042 * "this concludes the stream", and we *do* want to
1043 * eat that input.
1044 *
1045 * Otherwise we would not be able to test that we
1046 * consumed all the input to reach the expected size;
1047 * we also want to check that zlib tells us that all
1048 * went well with status == Z_STREAM_END at the end.
1049 */
1050 stream->next_out = buf + bytes;
1051 stream->avail_out = size - bytes;
1052 while (status == Z_OK)
1053 status = git_inflate(stream, Z_FINISH);
1054 }
1055 if (status == Z_STREAM_END && !stream->avail_in) {
1056 git_inflate_end(stream);
1057 return buf;
1058 }
1059
1060 if (status < 0)
1061 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1062 else if (stream->avail_in)
1063 error("garbage at end of loose object '%s'",
1064 sha1_to_hex(sha1));
1065 free(buf);
1066 return NULL;
1067 }
1068
1069 /*
1070 * We used to just use "sscanf()", but that's actually way
1071 * too permissive for what we want to check. So do an anal
1072 * object header parse by hand.
1073 */
1074 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1075 unsigned int flags)
1076 {
1077 const char *type_buf = hdr;
1078 unsigned long size;
1079 int type, type_len = 0;
1080
1081 /*
1082 * The type can be of any size but is followed by
1083 * a space.
1084 */
1085 for (;;) {
1086 char c = *hdr++;
1087 if (!c)
1088 return -1;
1089 if (c == ' ')
1090 break;
1091 type_len++;
1092 }
1093
1094 type = type_from_string_gently(type_buf, type_len, 1);
1095 if (oi->typename)
1096 strbuf_add(oi->typename, type_buf, type_len);
1097 /*
1098 * Set type to 0 if its an unknown object and
1099 * we're obtaining the type using '--allow-unknown-type'
1100 * option.
1101 */
1102 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1103 type = 0;
1104 else if (type < 0)
1105 die("invalid object type");
1106 if (oi->typep)
1107 *oi->typep = type;
1108
1109 /*
1110 * The length must follow immediately, and be in canonical
1111 * decimal format (ie "010" is not valid).
1112 */
1113 size = *hdr++ - '0';
1114 if (size > 9)
1115 return -1;
1116 if (size) {
1117 for (;;) {
1118 unsigned long c = *hdr - '0';
1119 if (c > 9)
1120 break;
1121 hdr++;
1122 size = size * 10 + c;
1123 }
1124 }
1125
1126 if (oi->sizep)
1127 *oi->sizep = size;
1128
1129 /*
1130 * The length must be followed by a zero byte
1131 */
1132 return *hdr ? -1 : type;
1133 }
1134
1135 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1136 {
1137 struct object_info oi = OBJECT_INFO_INIT;
1138
1139 oi.sizep = sizep;
1140 return parse_sha1_header_extended(hdr, &oi, 0);
1141 }
1142
1143 static int sha1_loose_object_info(const unsigned char *sha1,
1144 struct object_info *oi,
1145 int flags)
1146 {
1147 int status = 0;
1148 unsigned long mapsize;
1149 void *map;
1150 git_zstream stream;
1151 char hdr[32];
1152 struct strbuf hdrbuf = STRBUF_INIT;
1153 unsigned long size_scratch;
1154
1155 if (oi->delta_base_sha1)
1156 hashclr(oi->delta_base_sha1);
1157
1158 /*
1159 * If we don't care about type or size, then we don't
1160 * need to look inside the object at all. Note that we
1161 * do not optimize out the stat call, even if the
1162 * caller doesn't care about the disk-size, since our
1163 * return value implicitly indicates whether the
1164 * object even exists.
1165 */
1166 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1167 const char *path;
1168 struct stat st;
1169 if (stat_sha1_file(sha1, &st, &path) < 0)
1170 return -1;
1171 if (oi->disk_sizep)
1172 *oi->disk_sizep = st.st_size;
1173 return 0;
1174 }
1175
1176 map = map_sha1_file(sha1, &mapsize);
1177 if (!map)
1178 return -1;
1179
1180 if (!oi->sizep)
1181 oi->sizep = &size_scratch;
1182
1183 if (oi->disk_sizep)
1184 *oi->disk_sizep = mapsize;
1185 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1186 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1187 status = error("unable to unpack %s header with --allow-unknown-type",
1188 sha1_to_hex(sha1));
1189 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1190 status = error("unable to unpack %s header",
1191 sha1_to_hex(sha1));
1192 if (status < 0)
1193 ; /* Do nothing */
1194 else if (hdrbuf.len) {
1195 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1196 status = error("unable to parse %s header with --allow-unknown-type",
1197 sha1_to_hex(sha1));
1198 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1199 status = error("unable to parse %s header", sha1_to_hex(sha1));
1200
1201 if (status >= 0 && oi->contentp) {
1202 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1203 *oi->sizep, sha1);
1204 if (!*oi->contentp) {
1205 git_inflate_end(&stream);
1206 status = -1;
1207 }
1208 } else
1209 git_inflate_end(&stream);
1210
1211 munmap(map, mapsize);
1212 if (status && oi->typep)
1213 *oi->typep = status;
1214 if (oi->sizep == &size_scratch)
1215 oi->sizep = NULL;
1216 strbuf_release(&hdrbuf);
1217 oi->whence = OI_LOOSE;
1218 return (status < 0) ? status : 0;
1219 }
1220
1221 int fetch_if_missing = 1;
1222
1223 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1224 {
1225 static struct object_info blank_oi = OBJECT_INFO_INIT;
1226 struct pack_entry e;
1227 int rtype;
1228 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1229 lookup_replace_object(sha1) :
1230 sha1;
1231 int already_retried = 0;
1232
1233 if (is_null_sha1(real))
1234 return -1;
1235
1236 if (!oi)
1237 oi = &blank_oi;
1238
1239 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1240 struct cached_object *co = find_cached_object(real);
1241 if (co) {
1242 if (oi->typep)
1243 *(oi->typep) = co->type;
1244 if (oi->sizep)
1245 *(oi->sizep) = co->size;
1246 if (oi->disk_sizep)
1247 *(oi->disk_sizep) = 0;
1248 if (oi->delta_base_sha1)
1249 hashclr(oi->delta_base_sha1);
1250 if (oi->typename)
1251 strbuf_addstr(oi->typename, typename(co->type));
1252 if (oi->contentp)
1253 *oi->contentp = xmemdupz(co->buf, co->size);
1254 oi->whence = OI_CACHED;
1255 return 0;
1256 }
1257 }
1258
1259 while (1) {
1260 if (find_pack_entry(real, &e))
1261 break;
1262
1263 /* Most likely it's a loose object. */
1264 if (!sha1_loose_object_info(real, oi, flags))
1265 return 0;
1266
1267 /* Not a loose object; someone else may have just packed it. */
1268 reprepare_packed_git();
1269 if (find_pack_entry(real, &e))
1270 break;
1271
1272 /* Check if it is a missing object */
1273 if (fetch_if_missing && repository_format_partial_clone &&
1274 !already_retried) {
1275 /*
1276 * TODO Investigate haveing fetch_object() return
1277 * TODO error/success and stopping the music here.
1278 */
1279 fetch_object(repository_format_partial_clone, real);
1280 already_retried = 1;
1281 continue;
1282 }
1283
1284 return -1;
1285 }
1286
1287 if (oi == &blank_oi)
1288 /*
1289 * We know that the caller doesn't actually need the
1290 * information below, so return early.
1291 */
1292 return 0;
1293 rtype = packed_object_info(e.p, e.offset, oi);
1294 if (rtype < 0) {
1295 mark_bad_packed_object(e.p, real);
1296 return sha1_object_info_extended(real, oi, 0);
1297 } else if (oi->whence == OI_PACKED) {
1298 oi->u.packed.offset = e.offset;
1299 oi->u.packed.pack = e.p;
1300 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1301 rtype == OBJ_OFS_DELTA);
1302 }
1303
1304 return 0;
1305 }
1306
1307 /* returns enum object_type or negative */
1308 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1309 {
1310 enum object_type type;
1311 struct object_info oi = OBJECT_INFO_INIT;
1312
1313 oi.typep = &type;
1314 oi.sizep = sizep;
1315 if (sha1_object_info_extended(sha1, &oi,
1316 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1317 return -1;
1318 return type;
1319 }
1320
1321 static void *read_object(const unsigned char *sha1, enum object_type *type,
1322 unsigned long *size)
1323 {
1324 struct object_info oi = OBJECT_INFO_INIT;
1325 void *content;
1326 oi.typep = type;
1327 oi.sizep = size;
1328 oi.contentp = &content;
1329
1330 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1331 return NULL;
1332 return content;
1333 }
1334
1335 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1336 unsigned char *sha1)
1337 {
1338 struct cached_object *co;
1339
1340 hash_sha1_file(buf, len, typename(type), sha1);
1341 if (has_sha1_file(sha1) || find_cached_object(sha1))
1342 return 0;
1343 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1344 co = &cached_objects[cached_object_nr++];
1345 co->size = len;
1346 co->type = type;
1347 co->buf = xmalloc(len);
1348 memcpy(co->buf, buf, len);
1349 hashcpy(co->sha1, sha1);
1350 return 0;
1351 }
1352
1353 /*
1354 * This function dies on corrupt objects; the callers who want to
1355 * deal with them should arrange to call read_object() and give error
1356 * messages themselves.
1357 */
1358 void *read_sha1_file_extended(const unsigned char *sha1,
1359 enum object_type *type,
1360 unsigned long *size,
1361 int lookup_replace)
1362 {
1363 void *data;
1364 const struct packed_git *p;
1365 const char *path;
1366 struct stat st;
1367 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1368 : sha1;
1369
1370 errno = 0;
1371 data = read_object(repl, type, size);
1372 if (data)
1373 return data;
1374
1375 if (errno && errno != ENOENT)
1376 die_errno("failed to read object %s", sha1_to_hex(sha1));
1377
1378 /* die if we replaced an object with one that does not exist */
1379 if (repl != sha1)
1380 die("replacement %s not found for %s",
1381 sha1_to_hex(repl), sha1_to_hex(sha1));
1382
1383 if (!stat_sha1_file(repl, &st, &path))
1384 die("loose object %s (stored in %s) is corrupt",
1385 sha1_to_hex(repl), path);
1386
1387 if ((p = has_packed_and_bad(repl)) != NULL)
1388 die("packed object %s (stored in %s) is corrupt",
1389 sha1_to_hex(repl), p->pack_name);
1390
1391 return NULL;
1392 }
1393
1394 void *read_object_with_reference(const unsigned char *sha1,
1395 const char *required_type_name,
1396 unsigned long *size,
1397 unsigned char *actual_sha1_return)
1398 {
1399 enum object_type type, required_type;
1400 void *buffer;
1401 unsigned long isize;
1402 unsigned char actual_sha1[20];
1403
1404 required_type = type_from_string(required_type_name);
1405 hashcpy(actual_sha1, sha1);
1406 while (1) {
1407 int ref_length = -1;
1408 const char *ref_type = NULL;
1409
1410 buffer = read_sha1_file(actual_sha1, &type, &isize);
1411 if (!buffer)
1412 return NULL;
1413 if (type == required_type) {
1414 *size = isize;
1415 if (actual_sha1_return)
1416 hashcpy(actual_sha1_return, actual_sha1);
1417 return buffer;
1418 }
1419 /* Handle references */
1420 else if (type == OBJ_COMMIT)
1421 ref_type = "tree ";
1422 else if (type == OBJ_TAG)
1423 ref_type = "object ";
1424 else {
1425 free(buffer);
1426 return NULL;
1427 }
1428 ref_length = strlen(ref_type);
1429
1430 if (ref_length + 40 > isize ||
1431 memcmp(buffer, ref_type, ref_length) ||
1432 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1433 free(buffer);
1434 return NULL;
1435 }
1436 free(buffer);
1437 /* Now we have the ID of the referred-to object in
1438 * actual_sha1. Check again. */
1439 }
1440 }
1441
1442 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1443 const char *type, unsigned char *sha1,
1444 char *hdr, int *hdrlen)
1445 {
1446 git_SHA_CTX c;
1447
1448 /* Generate the header */
1449 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1450
1451 /* Sha1.. */
1452 git_SHA1_Init(&c);
1453 git_SHA1_Update(&c, hdr, *hdrlen);
1454 git_SHA1_Update(&c, buf, len);
1455 git_SHA1_Final(sha1, &c);
1456 }
1457
1458 /*
1459 * Move the just written object into its final resting place.
1460 */
1461 int finalize_object_file(const char *tmpfile, const char *filename)
1462 {
1463 int ret = 0;
1464
1465 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1466 goto try_rename;
1467 else if (link(tmpfile, filename))
1468 ret = errno;
1469
1470 /*
1471 * Coda hack - coda doesn't like cross-directory links,
1472 * so we fall back to a rename, which will mean that it
1473 * won't be able to check collisions, but that's not a
1474 * big deal.
1475 *
1476 * The same holds for FAT formatted media.
1477 *
1478 * When this succeeds, we just return. We have nothing
1479 * left to unlink.
1480 */
1481 if (ret && ret != EEXIST) {
1482 try_rename:
1483 if (!rename(tmpfile, filename))
1484 goto out;
1485 ret = errno;
1486 }
1487 unlink_or_warn(tmpfile);
1488 if (ret) {
1489 if (ret != EEXIST) {
1490 return error_errno("unable to write sha1 filename %s", filename);
1491 }
1492 /* FIXME!!! Collision check here ? */
1493 }
1494
1495 out:
1496 if (adjust_shared_perm(filename))
1497 return error("unable to set permission to '%s'", filename);
1498 return 0;
1499 }
1500
1501 static int write_buffer(int fd, const void *buf, size_t len)
1502 {
1503 if (write_in_full(fd, buf, len) < 0)
1504 return error_errno("file write error");
1505 return 0;
1506 }
1507
1508 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1509 unsigned char *sha1)
1510 {
1511 char hdr[32];
1512 int hdrlen = sizeof(hdr);
1513 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1514 return 0;
1515 }
1516
1517 /* Finalize a file on disk, and close it. */
1518 static void close_sha1_file(int fd)
1519 {
1520 if (fsync_object_files)
1521 fsync_or_die(fd, "sha1 file");
1522 if (close(fd) != 0)
1523 die_errno("error when closing sha1 file");
1524 }
1525
1526 /* Size of directory component, including the ending '/' */
1527 static inline int directory_size(const char *filename)
1528 {
1529 const char *s = strrchr(filename, '/');
1530 if (!s)
1531 return 0;
1532 return s - filename + 1;
1533 }
1534
1535 /*
1536 * This creates a temporary file in the same directory as the final
1537 * 'filename'
1538 *
1539 * We want to avoid cross-directory filename renames, because those
1540 * can have problems on various filesystems (FAT, NFS, Coda).
1541 */
1542 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1543 {
1544 int fd, dirlen = directory_size(filename);
1545
1546 strbuf_reset(tmp);
1547 strbuf_add(tmp, filename, dirlen);
1548 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1549 fd = git_mkstemp_mode(tmp->buf, 0444);
1550 if (fd < 0 && dirlen && errno == ENOENT) {
1551 /*
1552 * Make sure the directory exists; note that the contents
1553 * of the buffer are undefined after mkstemp returns an
1554 * error, so we have to rewrite the whole buffer from
1555 * scratch.
1556 */
1557 strbuf_reset(tmp);
1558 strbuf_add(tmp, filename, dirlen - 1);
1559 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1560 return -1;
1561 if (adjust_shared_perm(tmp->buf))
1562 return -1;
1563
1564 /* Try again */
1565 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1566 fd = git_mkstemp_mode(tmp->buf, 0444);
1567 }
1568 return fd;
1569 }
1570
1571 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1572 const void *buf, unsigned long len, time_t mtime)
1573 {
1574 int fd, ret;
1575 unsigned char compressed[4096];
1576 git_zstream stream;
1577 git_SHA_CTX c;
1578 unsigned char parano_sha1[20];
1579 static struct strbuf tmp_file = STRBUF_INIT;
1580 static struct strbuf filename = STRBUF_INIT;
1581
1582 strbuf_reset(&filename);
1583 sha1_file_name(&filename, sha1);
1584
1585 fd = create_tmpfile(&tmp_file, filename.buf);
1586 if (fd < 0) {
1587 if (errno == EACCES)
1588 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1589 else
1590 return error_errno("unable to create temporary file");
1591 }
1592
1593 /* Set it up */
1594 git_deflate_init(&stream, zlib_compression_level);
1595 stream.next_out = compressed;
1596 stream.avail_out = sizeof(compressed);
1597 git_SHA1_Init(&c);
1598
1599 /* First header.. */
1600 stream.next_in = (unsigned char *)hdr;
1601 stream.avail_in = hdrlen;
1602 while (git_deflate(&stream, 0) == Z_OK)
1603 ; /* nothing */
1604 git_SHA1_Update(&c, hdr, hdrlen);
1605
1606 /* Then the data itself.. */
1607 stream.next_in = (void *)buf;
1608 stream.avail_in = len;
1609 do {
1610 unsigned char *in0 = stream.next_in;
1611 ret = git_deflate(&stream, Z_FINISH);
1612 git_SHA1_Update(&c, in0, stream.next_in - in0);
1613 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1614 die("unable to write sha1 file");
1615 stream.next_out = compressed;
1616 stream.avail_out = sizeof(compressed);
1617 } while (ret == Z_OK);
1618
1619 if (ret != Z_STREAM_END)
1620 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1621 ret = git_deflate_end_gently(&stream);
1622 if (ret != Z_OK)
1623 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1624 git_SHA1_Final(parano_sha1, &c);
1625 if (hashcmp(sha1, parano_sha1) != 0)
1626 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1627
1628 close_sha1_file(fd);
1629
1630 if (mtime) {
1631 struct utimbuf utb;
1632 utb.actime = mtime;
1633 utb.modtime = mtime;
1634 if (utime(tmp_file.buf, &utb) < 0)
1635 warning_errno("failed utime() on %s", tmp_file.buf);
1636 }
1637
1638 return finalize_object_file(tmp_file.buf, filename.buf);
1639 }
1640
1641 static int freshen_loose_object(const unsigned char *sha1)
1642 {
1643 return check_and_freshen(sha1, 1);
1644 }
1645
1646 static int freshen_packed_object(const unsigned char *sha1)
1647 {
1648 struct pack_entry e;
1649 if (!find_pack_entry(sha1, &e))
1650 return 0;
1651 if (e.p->freshened)
1652 return 1;
1653 if (!freshen_file(e.p->pack_name))
1654 return 0;
1655 e.p->freshened = 1;
1656 return 1;
1657 }
1658
1659 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1660 {
1661 char hdr[32];
1662 int hdrlen = sizeof(hdr);
1663
1664 /* Normally if we have it in the pack then we do not bother writing
1665 * it out into .git/objects/??/?{38} file.
1666 */
1667 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1668 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1669 return 0;
1670 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1671 }
1672
1673 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1674 struct object_id *oid, unsigned flags)
1675 {
1676 char *header;
1677 int hdrlen, status = 0;
1678
1679 /* type string, SP, %lu of the length plus NUL must fit this */
1680 hdrlen = strlen(type) + 32;
1681 header = xmalloc(hdrlen);
1682 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1683
1684 if (!(flags & HASH_WRITE_OBJECT))
1685 goto cleanup;
1686 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1687 goto cleanup;
1688 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1689
1690 cleanup:
1691 free(header);
1692 return status;
1693 }
1694
1695 int force_object_loose(const unsigned char *sha1, time_t mtime)
1696 {
1697 void *buf;
1698 unsigned long len;
1699 enum object_type type;
1700 char hdr[32];
1701 int hdrlen;
1702 int ret;
1703
1704 if (has_loose_object(sha1))
1705 return 0;
1706 buf = read_object(sha1, &type, &len);
1707 if (!buf)
1708 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1709 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1710 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1711 free(buf);
1712
1713 return ret;
1714 }
1715
1716 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1717 {
1718 if (!startup_info->have_repository)
1719 return 0;
1720 return sha1_object_info_extended(sha1, NULL,
1721 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1722 }
1723
1724 int has_object_file(const struct object_id *oid)
1725 {
1726 return has_sha1_file(oid->hash);
1727 }
1728
1729 int has_object_file_with_flags(const struct object_id *oid, int flags)
1730 {
1731 return has_sha1_file_with_flags(oid->hash, flags);
1732 }
1733
1734 static void check_tree(const void *buf, size_t size)
1735 {
1736 struct tree_desc desc;
1737 struct name_entry entry;
1738
1739 init_tree_desc(&desc, buf, size);
1740 while (tree_entry(&desc, &entry))
1741 /* do nothing
1742 * tree_entry() will die() on malformed entries */
1743 ;
1744 }
1745
1746 static void check_commit(const void *buf, size_t size)
1747 {
1748 struct commit c;
1749 memset(&c, 0, sizeof(c));
1750 if (parse_commit_buffer(&c, buf, size))
1751 die("corrupt commit");
1752 }
1753
1754 static void check_tag(const void *buf, size_t size)
1755 {
1756 struct tag t;
1757 memset(&t, 0, sizeof(t));
1758 if (parse_tag_buffer(&t, buf, size))
1759 die("corrupt tag");
1760 }
1761
1762 static int index_mem(struct object_id *oid, void *buf, size_t size,
1763 enum object_type type,
1764 const char *path, unsigned flags)
1765 {
1766 int ret, re_allocated = 0;
1767 int write_object = flags & HASH_WRITE_OBJECT;
1768
1769 if (!type)
1770 type = OBJ_BLOB;
1771
1772 /*
1773 * Convert blobs to git internal format
1774 */
1775 if ((type == OBJ_BLOB) && path) {
1776 struct strbuf nbuf = STRBUF_INIT;
1777 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1778 get_conv_flags(flags))) {
1779 buf = strbuf_detach(&nbuf, &size);
1780 re_allocated = 1;
1781 }
1782 }
1783 if (flags & HASH_FORMAT_CHECK) {
1784 if (type == OBJ_TREE)
1785 check_tree(buf, size);
1786 if (type == OBJ_COMMIT)
1787 check_commit(buf, size);
1788 if (type == OBJ_TAG)
1789 check_tag(buf, size);
1790 }
1791
1792 if (write_object)
1793 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1794 else
1795 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1796 if (re_allocated)
1797 free(buf);
1798 return ret;
1799 }
1800
1801 static int index_stream_convert_blob(struct object_id *oid, int fd,
1802 const char *path, unsigned flags)
1803 {
1804 int ret;
1805 const int write_object = flags & HASH_WRITE_OBJECT;
1806 struct strbuf sbuf = STRBUF_INIT;
1807
1808 assert(path);
1809 assert(would_convert_to_git_filter_fd(path));
1810
1811 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1812 get_conv_flags(flags));
1813
1814 if (write_object)
1815 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1816 oid->hash);
1817 else
1818 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1819 oid->hash);
1820 strbuf_release(&sbuf);
1821 return ret;
1822 }
1823
1824 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1825 const char *path, unsigned flags)
1826 {
1827 struct strbuf sbuf = STRBUF_INIT;
1828 int ret;
1829
1830 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1831 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1832 else
1833 ret = -1;
1834 strbuf_release(&sbuf);
1835 return ret;
1836 }
1837
1838 #define SMALL_FILE_SIZE (32*1024)
1839
1840 static int index_core(struct object_id *oid, int fd, size_t size,
1841 enum object_type type, const char *path,
1842 unsigned flags)
1843 {
1844 int ret;
1845
1846 if (!size) {
1847 ret = index_mem(oid, "", size, type, path, flags);
1848 } else if (size <= SMALL_FILE_SIZE) {
1849 char *buf = xmalloc(size);
1850 ssize_t read_result = read_in_full(fd, buf, size);
1851 if (read_result < 0)
1852 ret = error_errno("read error while indexing %s",
1853 path ? path : "<unknown>");
1854 else if (read_result != size)
1855 ret = error("short read while indexing %s",
1856 path ? path : "<unknown>");
1857 else
1858 ret = index_mem(oid, buf, size, type, path, flags);
1859 free(buf);
1860 } else {
1861 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1862 ret = index_mem(oid, buf, size, type, path, flags);
1863 munmap(buf, size);
1864 }
1865 return ret;
1866 }
1867
1868 /*
1869 * This creates one packfile per large blob unless bulk-checkin
1870 * machinery is "plugged".
1871 *
1872 * This also bypasses the usual "convert-to-git" dance, and that is on
1873 * purpose. We could write a streaming version of the converting
1874 * functions and insert that before feeding the data to fast-import
1875 * (or equivalent in-core API described above). However, that is
1876 * somewhat complicated, as we do not know the size of the filter
1877 * result, which we need to know beforehand when writing a git object.
1878 * Since the primary motivation for trying to stream from the working
1879 * tree file and to avoid mmaping it in core is to deal with large
1880 * binary blobs, they generally do not want to get any conversion, and
1881 * callers should avoid this code path when filters are requested.
1882 */
1883 static int index_stream(struct object_id *oid, int fd, size_t size,
1884 enum object_type type, const char *path,
1885 unsigned flags)
1886 {
1887 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1888 }
1889
1890 int index_fd(struct object_id *oid, int fd, struct stat *st,
1891 enum object_type type, const char *path, unsigned flags)
1892 {
1893 int ret;
1894
1895 /*
1896 * Call xsize_t() only when needed to avoid potentially unnecessary
1897 * die() for large files.
1898 */
1899 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1900 ret = index_stream_convert_blob(oid, fd, path, flags);
1901 else if (!S_ISREG(st->st_mode))
1902 ret = index_pipe(oid, fd, type, path, flags);
1903 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1904 (path && would_convert_to_git(&the_index, path)))
1905 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1906 flags);
1907 else
1908 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1909 flags);
1910 close(fd);
1911 return ret;
1912 }
1913
1914 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1915 {
1916 int fd;
1917 struct strbuf sb = STRBUF_INIT;
1918 int rc = 0;
1919
1920 switch (st->st_mode & S_IFMT) {
1921 case S_IFREG:
1922 fd = open(path, O_RDONLY);
1923 if (fd < 0)
1924 return error_errno("open(\"%s\")", path);
1925 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1926 return error("%s: failed to insert into database",
1927 path);
1928 break;
1929 case S_IFLNK:
1930 if (strbuf_readlink(&sb, path, st->st_size))
1931 return error_errno("readlink(\"%s\")", path);
1932 if (!(flags & HASH_WRITE_OBJECT))
1933 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1934 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1935 rc = error("%s: failed to insert into database", path);
1936 strbuf_release(&sb);
1937 break;
1938 case S_IFDIR:
1939 return resolve_gitlink_ref(path, "HEAD", oid);
1940 default:
1941 return error("%s: unsupported file type", path);
1942 }
1943 return rc;
1944 }
1945
1946 int read_pack_header(int fd, struct pack_header *header)
1947 {
1948 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1949 /* "eof before pack header was fully read" */
1950 return PH_ERROR_EOF;
1951
1952 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1953 /* "protocol error (pack signature mismatch detected)" */
1954 return PH_ERROR_PACK_SIGNATURE;
1955 if (!pack_version_ok(header->hdr_version))
1956 /* "protocol error (pack version unsupported)" */
1957 return PH_ERROR_PROTOCOL;
1958 return 0;
1959 }
1960
1961 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1962 {
1963 enum object_type type = sha1_object_info(sha1, NULL);
1964 if (type < 0)
1965 die("%s is not a valid object", sha1_to_hex(sha1));
1966 if (type != expect)
1967 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1968 typename(expect));
1969 }
1970
1971 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1972 struct strbuf *path,
1973 each_loose_object_fn obj_cb,
1974 each_loose_cruft_fn cruft_cb,
1975 each_loose_subdir_fn subdir_cb,
1976 void *data)
1977 {
1978 size_t origlen, baselen;
1979 DIR *dir;
1980 struct dirent *de;
1981 int r = 0;
1982 struct object_id oid;
1983
1984 if (subdir_nr > 0xff)
1985 BUG("invalid loose object subdirectory: %x", subdir_nr);
1986
1987 origlen = path->len;
1988 strbuf_complete(path, '/');
1989 strbuf_addf(path, "%02x", subdir_nr);
1990
1991 dir = opendir(path->buf);
1992 if (!dir) {
1993 if (errno != ENOENT)
1994 r = error_errno("unable to open %s", path->buf);
1995 strbuf_setlen(path, origlen);
1996 return r;
1997 }
1998
1999 oid.hash[0] = subdir_nr;
2000 strbuf_addch(path, '/');
2001 baselen = path->len;
2002
2003 while ((de = readdir(dir))) {
2004 size_t namelen;
2005 if (is_dot_or_dotdot(de->d_name))
2006 continue;
2007
2008 namelen = strlen(de->d_name);
2009 strbuf_setlen(path, baselen);
2010 strbuf_add(path, de->d_name, namelen);
2011 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2012 !hex_to_bytes(oid.hash + 1, de->d_name,
2013 GIT_SHA1_RAWSZ - 1)) {
2014 if (obj_cb) {
2015 r = obj_cb(&oid, path->buf, data);
2016 if (r)
2017 break;
2018 }
2019 continue;
2020 }
2021
2022 if (cruft_cb) {
2023 r = cruft_cb(de->d_name, path->buf, data);
2024 if (r)
2025 break;
2026 }
2027 }
2028 closedir(dir);
2029
2030 strbuf_setlen(path, baselen - 1);
2031 if (!r && subdir_cb)
2032 r = subdir_cb(subdir_nr, path->buf, data);
2033
2034 strbuf_setlen(path, origlen);
2035
2036 return r;
2037 }
2038
2039 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2040 each_loose_object_fn obj_cb,
2041 each_loose_cruft_fn cruft_cb,
2042 each_loose_subdir_fn subdir_cb,
2043 void *data)
2044 {
2045 int r = 0;
2046 int i;
2047
2048 for (i = 0; i < 256; i++) {
2049 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2050 subdir_cb, data);
2051 if (r)
2052 break;
2053 }
2054
2055 return r;
2056 }
2057
2058 int for_each_loose_file_in_objdir(const char *path,
2059 each_loose_object_fn obj_cb,
2060 each_loose_cruft_fn cruft_cb,
2061 each_loose_subdir_fn subdir_cb,
2062 void *data)
2063 {
2064 struct strbuf buf = STRBUF_INIT;
2065 int r;
2066
2067 strbuf_addstr(&buf, path);
2068 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2069 subdir_cb, data);
2070 strbuf_release(&buf);
2071
2072 return r;
2073 }
2074
2075 struct loose_alt_odb_data {
2076 each_loose_object_fn *cb;
2077 void *data;
2078 };
2079
2080 static int loose_from_alt_odb(struct alternate_object_database *alt,
2081 void *vdata)
2082 {
2083 struct loose_alt_odb_data *data = vdata;
2084 struct strbuf buf = STRBUF_INIT;
2085 int r;
2086
2087 strbuf_addstr(&buf, alt->path);
2088 r = for_each_loose_file_in_objdir_buf(&buf,
2089 data->cb, NULL, NULL,
2090 data->data);
2091 strbuf_release(&buf);
2092 return r;
2093 }
2094
2095 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2096 {
2097 struct loose_alt_odb_data alt;
2098 int r;
2099
2100 r = for_each_loose_file_in_objdir(get_object_directory(),
2101 cb, NULL, NULL, data);
2102 if (r)
2103 return r;
2104
2105 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2106 return 0;
2107
2108 alt.cb = cb;
2109 alt.data = data;
2110 return foreach_alt_odb(loose_from_alt_odb, &alt);
2111 }
2112
2113 static int check_stream_sha1(git_zstream *stream,
2114 const char *hdr,
2115 unsigned long size,
2116 const char *path,
2117 const unsigned char *expected_sha1)
2118 {
2119 git_SHA_CTX c;
2120 unsigned char real_sha1[GIT_MAX_RAWSZ];
2121 unsigned char buf[4096];
2122 unsigned long total_read;
2123 int status = Z_OK;
2124
2125 git_SHA1_Init(&c);
2126 git_SHA1_Update(&c, hdr, stream->total_out);
2127
2128 /*
2129 * We already read some bytes into hdr, but the ones up to the NUL
2130 * do not count against the object's content size.
2131 */
2132 total_read = stream->total_out - strlen(hdr) - 1;
2133
2134 /*
2135 * This size comparison must be "<=" to read the final zlib packets;
2136 * see the comment in unpack_sha1_rest for details.
2137 */
2138 while (total_read <= size &&
2139 (status == Z_OK || status == Z_BUF_ERROR)) {
2140 stream->next_out = buf;
2141 stream->avail_out = sizeof(buf);
2142 if (size - total_read < stream->avail_out)
2143 stream->avail_out = size - total_read;
2144 status = git_inflate(stream, Z_FINISH);
2145 git_SHA1_Update(&c, buf, stream->next_out - buf);
2146 total_read += stream->next_out - buf;
2147 }
2148 git_inflate_end(stream);
2149
2150 if (status != Z_STREAM_END) {
2151 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2152 return -1;
2153 }
2154 if (stream->avail_in) {
2155 error("garbage at end of loose object '%s'",
2156 sha1_to_hex(expected_sha1));
2157 return -1;
2158 }
2159
2160 git_SHA1_Final(real_sha1, &c);
2161 if (hashcmp(expected_sha1, real_sha1)) {
2162 error("sha1 mismatch for %s (expected %s)", path,
2163 sha1_to_hex(expected_sha1));
2164 return -1;
2165 }
2166
2167 return 0;
2168 }
2169
2170 int read_loose_object(const char *path,
2171 const unsigned char *expected_sha1,
2172 enum object_type *type,
2173 unsigned long *size,
2174 void **contents)
2175 {
2176 int ret = -1;
2177 void *map = NULL;
2178 unsigned long mapsize;
2179 git_zstream stream;
2180 char hdr[32];
2181
2182 *contents = NULL;
2183
2184 map = map_sha1_file_1(path, NULL, &mapsize);
2185 if (!map) {
2186 error_errno("unable to mmap %s", path);
2187 goto out;
2188 }
2189
2190 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2191 error("unable to unpack header of %s", path);
2192 goto out;
2193 }
2194
2195 *type = parse_sha1_header(hdr, size);
2196 if (*type < 0) {
2197 error("unable to parse header of %s", path);
2198 git_inflate_end(&stream);
2199 goto out;
2200 }
2201
2202 if (*type == OBJ_BLOB) {
2203 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2204 goto out;
2205 } else {
2206 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2207 if (!*contents) {
2208 error("unable to unpack contents of %s", path);
2209 git_inflate_end(&stream);
2210 goto out;
2211 }
2212 if (check_sha1_signature(expected_sha1, *contents,
2213 *size, typename(*type))) {
2214 error("sha1 mismatch for %s (expected %s)", path,
2215 sha1_to_hex(expected_sha1));
2216 free(*contents);
2217 goto out;
2218 }
2219 }
2220
2221 ret = 0; /* everything checks out */
2222
2223 out:
2224 if (map)
2225 munmap(map, mapsize);
2226 return ret;
2227 }