Git 2.16-rc0
[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 "mru.h"
28 #include "list.h"
29 #include "mergesort.h"
30 #include "quote.h"
31 #include "packfile.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 enum safe_crlf get_safe_crlf(unsigned flags)
137 {
138 if (flags & HASH_RENORMALIZE)
139 return SAFE_CRLF_RENORMALIZE;
140 else if (flags & HASH_WRITE_OBJECT)
141 return safe_crlf;
142 else
143 return SAFE_CRLF_FALSE;
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 const char *sha1_file_name(const unsigned char *sha1)
325 {
326 static struct strbuf buf = STRBUF_INIT;
327
328 strbuf_reset(&buf);
329 strbuf_addf(&buf, "%s/", get_object_directory());
330
331 fill_sha1_path(&buf, sha1);
332 return buf.buf;
333 }
334
335 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
336 {
337 strbuf_setlen(&alt->scratch, alt->base_len);
338 return &alt->scratch;
339 }
340
341 static const char *alt_sha1_path(struct alternate_object_database *alt,
342 const unsigned char *sha1)
343 {
344 struct strbuf *buf = alt_scratch_buf(alt);
345 fill_sha1_path(buf, sha1);
346 return buf->buf;
347 }
348
349 struct alternate_object_database *alt_odb_list;
350 static struct alternate_object_database **alt_odb_tail;
351
352 /*
353 * Return non-zero iff the path is usable as an alternate object database.
354 */
355 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
356 {
357 struct alternate_object_database *alt;
358
359 /* Detect cases where alternate disappeared */
360 if (!is_directory(path->buf)) {
361 error("object directory %s does not exist; "
362 "check .git/objects/info/alternates.",
363 path->buf);
364 return 0;
365 }
366
367 /*
368 * Prevent the common mistake of listing the same
369 * thing twice, or object directory itself.
370 */
371 for (alt = alt_odb_list; alt; alt = alt->next) {
372 if (!fspathcmp(path->buf, alt->path))
373 return 0;
374 }
375 if (!fspathcmp(path->buf, normalized_objdir))
376 return 0;
377
378 return 1;
379 }
380
381 /*
382 * Prepare alternate object database registry.
383 *
384 * The variable alt_odb_list points at the list of struct
385 * alternate_object_database. The elements on this list come from
386 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
387 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
388 * whose contents is similar to that environment variable but can be
389 * LF separated. Its base points at a statically allocated buffer that
390 * contains "/the/directory/corresponding/to/.git/objects/...", while
391 * its name points just after the slash at the end of ".git/objects/"
392 * in the example above, and has enough space to hold 40-byte hex
393 * SHA1, an extra slash for the first level indirection, and the
394 * terminating NUL.
395 */
396 static void read_info_alternates(const char * relative_base, int depth);
397 static int link_alt_odb_entry(const char *entry, const char *relative_base,
398 int depth, const char *normalized_objdir)
399 {
400 struct alternate_object_database *ent;
401 struct strbuf pathbuf = STRBUF_INIT;
402
403 if (!is_absolute_path(entry) && relative_base) {
404 strbuf_realpath(&pathbuf, relative_base, 1);
405 strbuf_addch(&pathbuf, '/');
406 }
407 strbuf_addstr(&pathbuf, entry);
408
409 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
410 error("unable to normalize alternate object path: %s",
411 pathbuf.buf);
412 strbuf_release(&pathbuf);
413 return -1;
414 }
415
416 /*
417 * The trailing slash after the directory name is given by
418 * this function at the end. Remove duplicates.
419 */
420 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
421 strbuf_setlen(&pathbuf, pathbuf.len - 1);
422
423 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
424 strbuf_release(&pathbuf);
425 return -1;
426 }
427
428 ent = alloc_alt_odb(pathbuf.buf);
429
430 /* add the alternate entry */
431 *alt_odb_tail = ent;
432 alt_odb_tail = &(ent->next);
433 ent->next = NULL;
434
435 /* recursively add alternates */
436 read_info_alternates(pathbuf.buf, depth + 1);
437
438 strbuf_release(&pathbuf);
439 return 0;
440 }
441
442 static const char *parse_alt_odb_entry(const char *string,
443 int sep,
444 struct strbuf *out)
445 {
446 const char *end;
447
448 strbuf_reset(out);
449
450 if (*string == '#') {
451 /* comment; consume up to next separator */
452 end = strchrnul(string, sep);
453 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
454 /*
455 * quoted path; unquote_c_style has copied the
456 * data for us and set "end". Broken quoting (e.g.,
457 * an entry that doesn't end with a quote) falls
458 * back to the unquoted case below.
459 */
460 } else {
461 /* normal, unquoted path */
462 end = strchrnul(string, sep);
463 strbuf_add(out, string, end - string);
464 }
465
466 if (*end)
467 end++;
468 return end;
469 }
470
471 static void link_alt_odb_entries(const char *alt, int sep,
472 const char *relative_base, int depth)
473 {
474 struct strbuf objdirbuf = STRBUF_INIT;
475 struct strbuf entry = STRBUF_INIT;
476
477 if (!alt || !*alt)
478 return;
479
480 if (depth > 5) {
481 error("%s: ignoring alternate object stores, nesting too deep.",
482 relative_base);
483 return;
484 }
485
486 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
487 if (strbuf_normalize_path(&objdirbuf) < 0)
488 die("unable to normalize object directory: %s",
489 objdirbuf.buf);
490
491 while (*alt) {
492 alt = parse_alt_odb_entry(alt, sep, &entry);
493 if (!entry.len)
494 continue;
495 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
496 }
497 strbuf_release(&entry);
498 strbuf_release(&objdirbuf);
499 }
500
501 static void read_info_alternates(const char * relative_base, int depth)
502 {
503 char *path;
504 struct strbuf buf = STRBUF_INIT;
505
506 path = xstrfmt("%s/info/alternates", relative_base);
507 if (strbuf_read_file(&buf, path, 1024) < 0) {
508 warn_on_fopen_errors(path);
509 free(path);
510 return;
511 }
512
513 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
514 strbuf_release(&buf);
515 free(path);
516 }
517
518 struct alternate_object_database *alloc_alt_odb(const char *dir)
519 {
520 struct alternate_object_database *ent;
521
522 FLEX_ALLOC_STR(ent, path, dir);
523 strbuf_init(&ent->scratch, 0);
524 strbuf_addf(&ent->scratch, "%s/", dir);
525 ent->base_len = ent->scratch.len;
526
527 return ent;
528 }
529
530 void add_to_alternates_file(const char *reference)
531 {
532 struct lock_file lock = LOCK_INIT;
533 char *alts = git_pathdup("objects/info/alternates");
534 FILE *in, *out;
535 int found = 0;
536
537 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
538 out = fdopen_lock_file(&lock, "w");
539 if (!out)
540 die_errno("unable to fdopen alternates lockfile");
541
542 in = fopen(alts, "r");
543 if (in) {
544 struct strbuf line = STRBUF_INIT;
545
546 while (strbuf_getline(&line, in) != EOF) {
547 if (!strcmp(reference, line.buf)) {
548 found = 1;
549 break;
550 }
551 fprintf_or_die(out, "%s\n", line.buf);
552 }
553
554 strbuf_release(&line);
555 fclose(in);
556 }
557 else if (errno != ENOENT)
558 die_errno("unable to read alternates file");
559
560 if (found) {
561 rollback_lock_file(&lock);
562 } else {
563 fprintf_or_die(out, "%s\n", reference);
564 if (commit_lock_file(&lock))
565 die_errno("unable to move new alternates file into place");
566 if (alt_odb_tail)
567 link_alt_odb_entries(reference, '\n', NULL, 0);
568 }
569 free(alts);
570 }
571
572 void add_to_alternates_memory(const char *reference)
573 {
574 /*
575 * Make sure alternates are initialized, or else our entry may be
576 * overwritten when they are.
577 */
578 prepare_alt_odb();
579
580 link_alt_odb_entries(reference, '\n', NULL, 0);
581 }
582
583 /*
584 * Compute the exact path an alternate is at and returns it. In case of
585 * error NULL is returned and the human readable error is added to `err`
586 * `path` may be relative and should point to $GITDIR.
587 * `err` must not be null.
588 */
589 char *compute_alternate_path(const char *path, struct strbuf *err)
590 {
591 char *ref_git = NULL;
592 const char *repo, *ref_git_s;
593 int seen_error = 0;
594
595 ref_git_s = real_path_if_valid(path);
596 if (!ref_git_s) {
597 seen_error = 1;
598 strbuf_addf(err, _("path '%s' does not exist"), path);
599 goto out;
600 } else
601 /*
602 * Beware: read_gitfile(), real_path() and mkpath()
603 * return static buffer
604 */
605 ref_git = xstrdup(ref_git_s);
606
607 repo = read_gitfile(ref_git);
608 if (!repo)
609 repo = read_gitfile(mkpath("%s/.git", ref_git));
610 if (repo) {
611 free(ref_git);
612 ref_git = xstrdup(repo);
613 }
614
615 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
616 char *ref_git_git = mkpathdup("%s/.git", ref_git);
617 free(ref_git);
618 ref_git = ref_git_git;
619 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
620 struct strbuf sb = STRBUF_INIT;
621 seen_error = 1;
622 if (get_common_dir(&sb, ref_git)) {
623 strbuf_addf(err,
624 _("reference repository '%s' as a linked "
625 "checkout is not supported yet."),
626 path);
627 goto out;
628 }
629
630 strbuf_addf(err, _("reference repository '%s' is not a "
631 "local repository."), path);
632 goto out;
633 }
634
635 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
636 strbuf_addf(err, _("reference repository '%s' is shallow"),
637 path);
638 seen_error = 1;
639 goto out;
640 }
641
642 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
643 strbuf_addf(err,
644 _("reference repository '%s' is grafted"),
645 path);
646 seen_error = 1;
647 goto out;
648 }
649
650 out:
651 if (seen_error) {
652 FREE_AND_NULL(ref_git);
653 }
654
655 return ref_git;
656 }
657
658 int foreach_alt_odb(alt_odb_fn fn, void *cb)
659 {
660 struct alternate_object_database *ent;
661 int r = 0;
662
663 prepare_alt_odb();
664 for (ent = alt_odb_list; ent; ent = ent->next) {
665 r = fn(ent, cb);
666 if (r)
667 break;
668 }
669 return r;
670 }
671
672 void prepare_alt_odb(void)
673 {
674 const char *alt;
675
676 if (alt_odb_tail)
677 return;
678
679 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
680
681 alt_odb_tail = &alt_odb_list;
682 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
683
684 read_info_alternates(get_object_directory(), 0);
685 }
686
687 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
688 static int freshen_file(const char *fn)
689 {
690 struct utimbuf t;
691 t.actime = t.modtime = time(NULL);
692 return !utime(fn, &t);
693 }
694
695 /*
696 * All of the check_and_freshen functions return 1 if the file exists and was
697 * freshened (if freshening was requested), 0 otherwise. If they return
698 * 0, you should not assume that it is safe to skip a write of the object (it
699 * either does not exist on disk, or has a stale mtime and may be subject to
700 * pruning).
701 */
702 int check_and_freshen_file(const char *fn, int freshen)
703 {
704 if (access(fn, F_OK))
705 return 0;
706 if (freshen && !freshen_file(fn))
707 return 0;
708 return 1;
709 }
710
711 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
712 {
713 return check_and_freshen_file(sha1_file_name(sha1), freshen);
714 }
715
716 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
717 {
718 struct alternate_object_database *alt;
719 prepare_alt_odb();
720 for (alt = alt_odb_list; alt; alt = alt->next) {
721 const char *path = alt_sha1_path(alt, sha1);
722 if (check_and_freshen_file(path, freshen))
723 return 1;
724 }
725 return 0;
726 }
727
728 static int check_and_freshen(const unsigned char *sha1, int freshen)
729 {
730 return check_and_freshen_local(sha1, freshen) ||
731 check_and_freshen_nonlocal(sha1, freshen);
732 }
733
734 int has_loose_object_nonlocal(const unsigned char *sha1)
735 {
736 return check_and_freshen_nonlocal(sha1, 0);
737 }
738
739 static int has_loose_object(const unsigned char *sha1)
740 {
741 return check_and_freshen(sha1, 0);
742 }
743
744 static void mmap_limit_check(size_t length)
745 {
746 static size_t limit = 0;
747 if (!limit) {
748 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
749 if (!limit)
750 limit = SIZE_MAX;
751 }
752 if (length > limit)
753 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
754 (uintmax_t)length, (uintmax_t)limit);
755 }
756
757 void *xmmap_gently(void *start, size_t length,
758 int prot, int flags, int fd, off_t offset)
759 {
760 void *ret;
761
762 mmap_limit_check(length);
763 ret = mmap(start, length, prot, flags, fd, offset);
764 if (ret == MAP_FAILED) {
765 if (!length)
766 return NULL;
767 release_pack_memory(length);
768 ret = mmap(start, length, prot, flags, fd, offset);
769 }
770 return ret;
771 }
772
773 void *xmmap(void *start, size_t length,
774 int prot, int flags, int fd, off_t offset)
775 {
776 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
777 if (ret == MAP_FAILED)
778 die_errno("mmap failed");
779 return ret;
780 }
781
782 /*
783 * With an in-core object data in "map", rehash it to make sure the
784 * object name actually matches "sha1" to detect object corruption.
785 * With "map" == NULL, try reading the object named with "sha1" using
786 * the streaming interface and rehash it to do the same.
787 */
788 int check_sha1_signature(const unsigned char *sha1, void *map,
789 unsigned long size, const char *type)
790 {
791 unsigned char real_sha1[20];
792 enum object_type obj_type;
793 struct git_istream *st;
794 git_SHA_CTX c;
795 char hdr[32];
796 int hdrlen;
797
798 if (map) {
799 hash_sha1_file(map, size, type, real_sha1);
800 return hashcmp(sha1, real_sha1) ? -1 : 0;
801 }
802
803 st = open_istream(sha1, &obj_type, &size, NULL);
804 if (!st)
805 return -1;
806
807 /* Generate the header */
808 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
809
810 /* Sha1.. */
811 git_SHA1_Init(&c);
812 git_SHA1_Update(&c, hdr, hdrlen);
813 for (;;) {
814 char buf[1024 * 16];
815 ssize_t readlen = read_istream(st, buf, sizeof(buf));
816
817 if (readlen < 0) {
818 close_istream(st);
819 return -1;
820 }
821 if (!readlen)
822 break;
823 git_SHA1_Update(&c, buf, readlen);
824 }
825 git_SHA1_Final(real_sha1, &c);
826 close_istream(st);
827 return hashcmp(sha1, real_sha1) ? -1 : 0;
828 }
829
830 int git_open_cloexec(const char *name, int flags)
831 {
832 int fd;
833 static int o_cloexec = O_CLOEXEC;
834
835 fd = open(name, flags | o_cloexec);
836 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
837 /* Try again w/o O_CLOEXEC: the kernel might not support it */
838 o_cloexec &= ~O_CLOEXEC;
839 fd = open(name, flags | o_cloexec);
840 }
841
842 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
843 {
844 static int fd_cloexec = FD_CLOEXEC;
845
846 if (!o_cloexec && 0 <= fd && fd_cloexec) {
847 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
848 int flags = fcntl(fd, F_GETFD);
849 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
850 fd_cloexec = 0;
851 }
852 }
853 #endif
854 return fd;
855 }
856
857 /*
858 * Find "sha1" as a loose object in the local repository or in an alternate.
859 * Returns 0 on success, negative on failure.
860 *
861 * The "path" out-parameter will give the path of the object we found (if any).
862 * Note that it may point to static storage and is only valid until another
863 * call to sha1_file_name(), etc.
864 */
865 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
866 const char **path)
867 {
868 struct alternate_object_database *alt;
869
870 *path = sha1_file_name(sha1);
871 if (!lstat(*path, st))
872 return 0;
873
874 prepare_alt_odb();
875 errno = ENOENT;
876 for (alt = alt_odb_list; alt; alt = alt->next) {
877 *path = alt_sha1_path(alt, sha1);
878 if (!lstat(*path, st))
879 return 0;
880 }
881
882 return -1;
883 }
884
885 /*
886 * Like stat_sha1_file(), but actually open the object and return the
887 * descriptor. See the caveats on the "path" parameter above.
888 */
889 static int open_sha1_file(const unsigned char *sha1, const char **path)
890 {
891 int fd;
892 struct alternate_object_database *alt;
893 int most_interesting_errno;
894
895 *path = sha1_file_name(sha1);
896 fd = git_open(*path);
897 if (fd >= 0)
898 return fd;
899 most_interesting_errno = errno;
900
901 prepare_alt_odb();
902 for (alt = alt_odb_list; alt; alt = alt->next) {
903 *path = alt_sha1_path(alt, sha1);
904 fd = git_open(*path);
905 if (fd >= 0)
906 return fd;
907 if (most_interesting_errno == ENOENT)
908 most_interesting_errno = errno;
909 }
910 errno = most_interesting_errno;
911 return -1;
912 }
913
914 /*
915 * Map the loose object at "path" if it is not NULL, or the path found by
916 * searching for a loose object named "sha1".
917 */
918 static void *map_sha1_file_1(const char *path,
919 const unsigned char *sha1,
920 unsigned long *size)
921 {
922 void *map;
923 int fd;
924
925 if (path)
926 fd = git_open(path);
927 else
928 fd = open_sha1_file(sha1, &path);
929 map = NULL;
930 if (fd >= 0) {
931 struct stat st;
932
933 if (!fstat(fd, &st)) {
934 *size = xsize_t(st.st_size);
935 if (!*size) {
936 /* mmap() is forbidden on empty files */
937 error("object file %s is empty", path);
938 return NULL;
939 }
940 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
941 }
942 close(fd);
943 }
944 return map;
945 }
946
947 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
948 {
949 return map_sha1_file_1(NULL, sha1, size);
950 }
951
952 static int unpack_sha1_short_header(git_zstream *stream,
953 unsigned char *map, unsigned long mapsize,
954 void *buffer, unsigned long bufsiz)
955 {
956 /* Get the data stream */
957 memset(stream, 0, sizeof(*stream));
958 stream->next_in = map;
959 stream->avail_in = mapsize;
960 stream->next_out = buffer;
961 stream->avail_out = bufsiz;
962
963 git_inflate_init(stream);
964 return git_inflate(stream, 0);
965 }
966
967 int unpack_sha1_header(git_zstream *stream,
968 unsigned char *map, unsigned long mapsize,
969 void *buffer, unsigned long bufsiz)
970 {
971 int status = unpack_sha1_short_header(stream, map, mapsize,
972 buffer, bufsiz);
973
974 if (status < Z_OK)
975 return status;
976
977 /* Make sure we have the terminating NUL */
978 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
979 return -1;
980 return 0;
981 }
982
983 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
984 unsigned long mapsize, void *buffer,
985 unsigned long bufsiz, struct strbuf *header)
986 {
987 int status;
988
989 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
990 if (status < Z_OK)
991 return -1;
992
993 /*
994 * Check if entire header is unpacked in the first iteration.
995 */
996 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
997 return 0;
998
999 /*
1000 * buffer[0..bufsiz] was not large enough. Copy the partial
1001 * result out to header, and then append the result of further
1002 * reading the stream.
1003 */
1004 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1005 stream->next_out = buffer;
1006 stream->avail_out = bufsiz;
1007
1008 do {
1009 status = git_inflate(stream, 0);
1010 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1011 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1012 return 0;
1013 stream->next_out = buffer;
1014 stream->avail_out = bufsiz;
1015 } while (status != Z_STREAM_END);
1016 return -1;
1017 }
1018
1019 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1020 {
1021 int bytes = strlen(buffer) + 1;
1022 unsigned char *buf = xmallocz(size);
1023 unsigned long n;
1024 int status = Z_OK;
1025
1026 n = stream->total_out - bytes;
1027 if (n > size)
1028 n = size;
1029 memcpy(buf, (char *) buffer + bytes, n);
1030 bytes = n;
1031 if (bytes <= size) {
1032 /*
1033 * The above condition must be (bytes <= size), not
1034 * (bytes < size). In other words, even though we
1035 * expect no more output and set avail_out to zero,
1036 * the input zlib stream may have bytes that express
1037 * "this concludes the stream", and we *do* want to
1038 * eat that input.
1039 *
1040 * Otherwise we would not be able to test that we
1041 * consumed all the input to reach the expected size;
1042 * we also want to check that zlib tells us that all
1043 * went well with status == Z_STREAM_END at the end.
1044 */
1045 stream->next_out = buf + bytes;
1046 stream->avail_out = size - bytes;
1047 while (status == Z_OK)
1048 status = git_inflate(stream, Z_FINISH);
1049 }
1050 if (status == Z_STREAM_END && !stream->avail_in) {
1051 git_inflate_end(stream);
1052 return buf;
1053 }
1054
1055 if (status < 0)
1056 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1057 else if (stream->avail_in)
1058 error("garbage at end of loose object '%s'",
1059 sha1_to_hex(sha1));
1060 free(buf);
1061 return NULL;
1062 }
1063
1064 /*
1065 * We used to just use "sscanf()", but that's actually way
1066 * too permissive for what we want to check. So do an anal
1067 * object header parse by hand.
1068 */
1069 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1070 unsigned int flags)
1071 {
1072 const char *type_buf = hdr;
1073 unsigned long size;
1074 int type, type_len = 0;
1075
1076 /*
1077 * The type can be of any size but is followed by
1078 * a space.
1079 */
1080 for (;;) {
1081 char c = *hdr++;
1082 if (!c)
1083 return -1;
1084 if (c == ' ')
1085 break;
1086 type_len++;
1087 }
1088
1089 type = type_from_string_gently(type_buf, type_len, 1);
1090 if (oi->typename)
1091 strbuf_add(oi->typename, type_buf, type_len);
1092 /*
1093 * Set type to 0 if its an unknown object and
1094 * we're obtaining the type using '--allow-unknown-type'
1095 * option.
1096 */
1097 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1098 type = 0;
1099 else if (type < 0)
1100 die("invalid object type");
1101 if (oi->typep)
1102 *oi->typep = type;
1103
1104 /*
1105 * The length must follow immediately, and be in canonical
1106 * decimal format (ie "010" is not valid).
1107 */
1108 size = *hdr++ - '0';
1109 if (size > 9)
1110 return -1;
1111 if (size) {
1112 for (;;) {
1113 unsigned long c = *hdr - '0';
1114 if (c > 9)
1115 break;
1116 hdr++;
1117 size = size * 10 + c;
1118 }
1119 }
1120
1121 if (oi->sizep)
1122 *oi->sizep = size;
1123
1124 /*
1125 * The length must be followed by a zero byte
1126 */
1127 return *hdr ? -1 : type;
1128 }
1129
1130 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1131 {
1132 struct object_info oi = OBJECT_INFO_INIT;
1133
1134 oi.sizep = sizep;
1135 return parse_sha1_header_extended(hdr, &oi, 0);
1136 }
1137
1138 static int sha1_loose_object_info(const unsigned char *sha1,
1139 struct object_info *oi,
1140 int flags)
1141 {
1142 int status = 0;
1143 unsigned long mapsize;
1144 void *map;
1145 git_zstream stream;
1146 char hdr[32];
1147 struct strbuf hdrbuf = STRBUF_INIT;
1148 unsigned long size_scratch;
1149
1150 if (oi->delta_base_sha1)
1151 hashclr(oi->delta_base_sha1);
1152
1153 /*
1154 * If we don't care about type or size, then we don't
1155 * need to look inside the object at all. Note that we
1156 * do not optimize out the stat call, even if the
1157 * caller doesn't care about the disk-size, since our
1158 * return value implicitly indicates whether the
1159 * object even exists.
1160 */
1161 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1162 const char *path;
1163 struct stat st;
1164 if (stat_sha1_file(sha1, &st, &path) < 0)
1165 return -1;
1166 if (oi->disk_sizep)
1167 *oi->disk_sizep = st.st_size;
1168 return 0;
1169 }
1170
1171 map = map_sha1_file(sha1, &mapsize);
1172 if (!map)
1173 return -1;
1174
1175 if (!oi->sizep)
1176 oi->sizep = &size_scratch;
1177
1178 if (oi->disk_sizep)
1179 *oi->disk_sizep = mapsize;
1180 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1181 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1182 status = error("unable to unpack %s header with --allow-unknown-type",
1183 sha1_to_hex(sha1));
1184 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1185 status = error("unable to unpack %s header",
1186 sha1_to_hex(sha1));
1187 if (status < 0)
1188 ; /* Do nothing */
1189 else if (hdrbuf.len) {
1190 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1191 status = error("unable to parse %s header with --allow-unknown-type",
1192 sha1_to_hex(sha1));
1193 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1194 status = error("unable to parse %s header", sha1_to_hex(sha1));
1195
1196 if (status >= 0 && oi->contentp) {
1197 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1198 *oi->sizep, sha1);
1199 if (!*oi->contentp) {
1200 git_inflate_end(&stream);
1201 status = -1;
1202 }
1203 } else
1204 git_inflate_end(&stream);
1205
1206 munmap(map, mapsize);
1207 if (status && oi->typep)
1208 *oi->typep = status;
1209 if (oi->sizep == &size_scratch)
1210 oi->sizep = NULL;
1211 strbuf_release(&hdrbuf);
1212 oi->whence = OI_LOOSE;
1213 return (status < 0) ? status : 0;
1214 }
1215
1216 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1217 {
1218 static struct object_info blank_oi = OBJECT_INFO_INIT;
1219 struct pack_entry e;
1220 int rtype;
1221 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1222 lookup_replace_object(sha1) :
1223 sha1;
1224
1225 if (is_null_sha1(real))
1226 return -1;
1227
1228 if (!oi)
1229 oi = &blank_oi;
1230
1231 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1232 struct cached_object *co = find_cached_object(real);
1233 if (co) {
1234 if (oi->typep)
1235 *(oi->typep) = co->type;
1236 if (oi->sizep)
1237 *(oi->sizep) = co->size;
1238 if (oi->disk_sizep)
1239 *(oi->disk_sizep) = 0;
1240 if (oi->delta_base_sha1)
1241 hashclr(oi->delta_base_sha1);
1242 if (oi->typename)
1243 strbuf_addstr(oi->typename, typename(co->type));
1244 if (oi->contentp)
1245 *oi->contentp = xmemdupz(co->buf, co->size);
1246 oi->whence = OI_CACHED;
1247 return 0;
1248 }
1249 }
1250
1251 if (!find_pack_entry(real, &e)) {
1252 /* Most likely it's a loose object. */
1253 if (!sha1_loose_object_info(real, oi, flags))
1254 return 0;
1255
1256 /* Not a loose object; someone else may have just packed it. */
1257 if (flags & OBJECT_INFO_QUICK) {
1258 return -1;
1259 } else {
1260 reprepare_packed_git();
1261 if (!find_pack_entry(real, &e))
1262 return -1;
1263 }
1264 }
1265
1266 if (oi == &blank_oi)
1267 /*
1268 * We know that the caller doesn't actually need the
1269 * information below, so return early.
1270 */
1271 return 0;
1272
1273 rtype = packed_object_info(e.p, e.offset, oi);
1274 if (rtype < 0) {
1275 mark_bad_packed_object(e.p, real);
1276 return sha1_object_info_extended(real, oi, 0);
1277 } else if (oi->whence == OI_PACKED) {
1278 oi->u.packed.offset = e.offset;
1279 oi->u.packed.pack = e.p;
1280 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1281 rtype == OBJ_OFS_DELTA);
1282 }
1283
1284 return 0;
1285 }
1286
1287 /* returns enum object_type or negative */
1288 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1289 {
1290 enum object_type type;
1291 struct object_info oi = OBJECT_INFO_INIT;
1292
1293 oi.typep = &type;
1294 oi.sizep = sizep;
1295 if (sha1_object_info_extended(sha1, &oi,
1296 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1297 return -1;
1298 return type;
1299 }
1300
1301 static void *read_object(const unsigned char *sha1, enum object_type *type,
1302 unsigned long *size)
1303 {
1304 struct object_info oi = OBJECT_INFO_INIT;
1305 void *content;
1306 oi.typep = type;
1307 oi.sizep = size;
1308 oi.contentp = &content;
1309
1310 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1311 return NULL;
1312 return content;
1313 }
1314
1315 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1316 unsigned char *sha1)
1317 {
1318 struct cached_object *co;
1319
1320 hash_sha1_file(buf, len, typename(type), sha1);
1321 if (has_sha1_file(sha1) || find_cached_object(sha1))
1322 return 0;
1323 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1324 co = &cached_objects[cached_object_nr++];
1325 co->size = len;
1326 co->type = type;
1327 co->buf = xmalloc(len);
1328 memcpy(co->buf, buf, len);
1329 hashcpy(co->sha1, sha1);
1330 return 0;
1331 }
1332
1333 /*
1334 * This function dies on corrupt objects; the callers who want to
1335 * deal with them should arrange to call read_object() and give error
1336 * messages themselves.
1337 */
1338 void *read_sha1_file_extended(const unsigned char *sha1,
1339 enum object_type *type,
1340 unsigned long *size,
1341 int lookup_replace)
1342 {
1343 void *data;
1344 const struct packed_git *p;
1345 const char *path;
1346 struct stat st;
1347 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1348 : sha1;
1349
1350 errno = 0;
1351 data = read_object(repl, type, size);
1352 if (data)
1353 return data;
1354
1355 if (errno && errno != ENOENT)
1356 die_errno("failed to read object %s", sha1_to_hex(sha1));
1357
1358 /* die if we replaced an object with one that does not exist */
1359 if (repl != sha1)
1360 die("replacement %s not found for %s",
1361 sha1_to_hex(repl), sha1_to_hex(sha1));
1362
1363 if (!stat_sha1_file(repl, &st, &path))
1364 die("loose object %s (stored in %s) is corrupt",
1365 sha1_to_hex(repl), path);
1366
1367 if ((p = has_packed_and_bad(repl)) != NULL)
1368 die("packed object %s (stored in %s) is corrupt",
1369 sha1_to_hex(repl), p->pack_name);
1370
1371 return NULL;
1372 }
1373
1374 void *read_object_with_reference(const unsigned char *sha1,
1375 const char *required_type_name,
1376 unsigned long *size,
1377 unsigned char *actual_sha1_return)
1378 {
1379 enum object_type type, required_type;
1380 void *buffer;
1381 unsigned long isize;
1382 unsigned char actual_sha1[20];
1383
1384 required_type = type_from_string(required_type_name);
1385 hashcpy(actual_sha1, sha1);
1386 while (1) {
1387 int ref_length = -1;
1388 const char *ref_type = NULL;
1389
1390 buffer = read_sha1_file(actual_sha1, &type, &isize);
1391 if (!buffer)
1392 return NULL;
1393 if (type == required_type) {
1394 *size = isize;
1395 if (actual_sha1_return)
1396 hashcpy(actual_sha1_return, actual_sha1);
1397 return buffer;
1398 }
1399 /* Handle references */
1400 else if (type == OBJ_COMMIT)
1401 ref_type = "tree ";
1402 else if (type == OBJ_TAG)
1403 ref_type = "object ";
1404 else {
1405 free(buffer);
1406 return NULL;
1407 }
1408 ref_length = strlen(ref_type);
1409
1410 if (ref_length + 40 > isize ||
1411 memcmp(buffer, ref_type, ref_length) ||
1412 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1413 free(buffer);
1414 return NULL;
1415 }
1416 free(buffer);
1417 /* Now we have the ID of the referred-to object in
1418 * actual_sha1. Check again. */
1419 }
1420 }
1421
1422 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1423 const char *type, unsigned char *sha1,
1424 char *hdr, int *hdrlen)
1425 {
1426 git_SHA_CTX c;
1427
1428 /* Generate the header */
1429 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1430
1431 /* Sha1.. */
1432 git_SHA1_Init(&c);
1433 git_SHA1_Update(&c, hdr, *hdrlen);
1434 git_SHA1_Update(&c, buf, len);
1435 git_SHA1_Final(sha1, &c);
1436 }
1437
1438 /*
1439 * Move the just written object into its final resting place.
1440 */
1441 int finalize_object_file(const char *tmpfile, const char *filename)
1442 {
1443 int ret = 0;
1444
1445 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1446 goto try_rename;
1447 else if (link(tmpfile, filename))
1448 ret = errno;
1449
1450 /*
1451 * Coda hack - coda doesn't like cross-directory links,
1452 * so we fall back to a rename, which will mean that it
1453 * won't be able to check collisions, but that's not a
1454 * big deal.
1455 *
1456 * The same holds for FAT formatted media.
1457 *
1458 * When this succeeds, we just return. We have nothing
1459 * left to unlink.
1460 */
1461 if (ret && ret != EEXIST) {
1462 try_rename:
1463 if (!rename(tmpfile, filename))
1464 goto out;
1465 ret = errno;
1466 }
1467 unlink_or_warn(tmpfile);
1468 if (ret) {
1469 if (ret != EEXIST) {
1470 return error_errno("unable to write sha1 filename %s", filename);
1471 }
1472 /* FIXME!!! Collision check here ? */
1473 }
1474
1475 out:
1476 if (adjust_shared_perm(filename))
1477 return error("unable to set permission to '%s'", filename);
1478 return 0;
1479 }
1480
1481 static int write_buffer(int fd, const void *buf, size_t len)
1482 {
1483 if (write_in_full(fd, buf, len) < 0)
1484 return error_errno("file write error");
1485 return 0;
1486 }
1487
1488 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1489 unsigned char *sha1)
1490 {
1491 char hdr[32];
1492 int hdrlen = sizeof(hdr);
1493 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1494 return 0;
1495 }
1496
1497 /* Finalize a file on disk, and close it. */
1498 static void close_sha1_file(int fd)
1499 {
1500 if (fsync_object_files)
1501 fsync_or_die(fd, "sha1 file");
1502 if (close(fd) != 0)
1503 die_errno("error when closing sha1 file");
1504 }
1505
1506 /* Size of directory component, including the ending '/' */
1507 static inline int directory_size(const char *filename)
1508 {
1509 const char *s = strrchr(filename, '/');
1510 if (!s)
1511 return 0;
1512 return s - filename + 1;
1513 }
1514
1515 /*
1516 * This creates a temporary file in the same directory as the final
1517 * 'filename'
1518 *
1519 * We want to avoid cross-directory filename renames, because those
1520 * can have problems on various filesystems (FAT, NFS, Coda).
1521 */
1522 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1523 {
1524 int fd, dirlen = directory_size(filename);
1525
1526 strbuf_reset(tmp);
1527 strbuf_add(tmp, filename, dirlen);
1528 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1529 fd = git_mkstemp_mode(tmp->buf, 0444);
1530 if (fd < 0 && dirlen && errno == ENOENT) {
1531 /*
1532 * Make sure the directory exists; note that the contents
1533 * of the buffer are undefined after mkstemp returns an
1534 * error, so we have to rewrite the whole buffer from
1535 * scratch.
1536 */
1537 strbuf_reset(tmp);
1538 strbuf_add(tmp, filename, dirlen - 1);
1539 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1540 return -1;
1541 if (adjust_shared_perm(tmp->buf))
1542 return -1;
1543
1544 /* Try again */
1545 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1546 fd = git_mkstemp_mode(tmp->buf, 0444);
1547 }
1548 return fd;
1549 }
1550
1551 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1552 const void *buf, unsigned long len, time_t mtime)
1553 {
1554 int fd, ret;
1555 unsigned char compressed[4096];
1556 git_zstream stream;
1557 git_SHA_CTX c;
1558 unsigned char parano_sha1[20];
1559 static struct strbuf tmp_file = STRBUF_INIT;
1560 const char *filename = sha1_file_name(sha1);
1561
1562 fd = create_tmpfile(&tmp_file, filename);
1563 if (fd < 0) {
1564 if (errno == EACCES)
1565 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1566 else
1567 return error_errno("unable to create temporary file");
1568 }
1569
1570 /* Set it up */
1571 git_deflate_init(&stream, zlib_compression_level);
1572 stream.next_out = compressed;
1573 stream.avail_out = sizeof(compressed);
1574 git_SHA1_Init(&c);
1575
1576 /* First header.. */
1577 stream.next_in = (unsigned char *)hdr;
1578 stream.avail_in = hdrlen;
1579 while (git_deflate(&stream, 0) == Z_OK)
1580 ; /* nothing */
1581 git_SHA1_Update(&c, hdr, hdrlen);
1582
1583 /* Then the data itself.. */
1584 stream.next_in = (void *)buf;
1585 stream.avail_in = len;
1586 do {
1587 unsigned char *in0 = stream.next_in;
1588 ret = git_deflate(&stream, Z_FINISH);
1589 git_SHA1_Update(&c, in0, stream.next_in - in0);
1590 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1591 die("unable to write sha1 file");
1592 stream.next_out = compressed;
1593 stream.avail_out = sizeof(compressed);
1594 } while (ret == Z_OK);
1595
1596 if (ret != Z_STREAM_END)
1597 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1598 ret = git_deflate_end_gently(&stream);
1599 if (ret != Z_OK)
1600 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1601 git_SHA1_Final(parano_sha1, &c);
1602 if (hashcmp(sha1, parano_sha1) != 0)
1603 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1604
1605 close_sha1_file(fd);
1606
1607 if (mtime) {
1608 struct utimbuf utb;
1609 utb.actime = mtime;
1610 utb.modtime = mtime;
1611 if (utime(tmp_file.buf, &utb) < 0)
1612 warning_errno("failed utime() on %s", tmp_file.buf);
1613 }
1614
1615 return finalize_object_file(tmp_file.buf, filename);
1616 }
1617
1618 static int freshen_loose_object(const unsigned char *sha1)
1619 {
1620 return check_and_freshen(sha1, 1);
1621 }
1622
1623 static int freshen_packed_object(const unsigned char *sha1)
1624 {
1625 struct pack_entry e;
1626 if (!find_pack_entry(sha1, &e))
1627 return 0;
1628 if (e.p->freshened)
1629 return 1;
1630 if (!freshen_file(e.p->pack_name))
1631 return 0;
1632 e.p->freshened = 1;
1633 return 1;
1634 }
1635
1636 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1637 {
1638 char hdr[32];
1639 int hdrlen = sizeof(hdr);
1640
1641 /* Normally if we have it in the pack then we do not bother writing
1642 * it out into .git/objects/??/?{38} file.
1643 */
1644 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1645 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1646 return 0;
1647 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1648 }
1649
1650 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1651 struct object_id *oid, unsigned flags)
1652 {
1653 char *header;
1654 int hdrlen, status = 0;
1655
1656 /* type string, SP, %lu of the length plus NUL must fit this */
1657 hdrlen = strlen(type) + 32;
1658 header = xmalloc(hdrlen);
1659 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1660
1661 if (!(flags & HASH_WRITE_OBJECT))
1662 goto cleanup;
1663 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1664 goto cleanup;
1665 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1666
1667 cleanup:
1668 free(header);
1669 return status;
1670 }
1671
1672 int force_object_loose(const unsigned char *sha1, time_t mtime)
1673 {
1674 void *buf;
1675 unsigned long len;
1676 enum object_type type;
1677 char hdr[32];
1678 int hdrlen;
1679 int ret;
1680
1681 if (has_loose_object(sha1))
1682 return 0;
1683 buf = read_object(sha1, &type, &len);
1684 if (!buf)
1685 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1686 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1687 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1688 free(buf);
1689
1690 return ret;
1691 }
1692
1693 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1694 {
1695 if (!startup_info->have_repository)
1696 return 0;
1697 return sha1_object_info_extended(sha1, NULL,
1698 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1699 }
1700
1701 int has_object_file(const struct object_id *oid)
1702 {
1703 return has_sha1_file(oid->hash);
1704 }
1705
1706 int has_object_file_with_flags(const struct object_id *oid, int flags)
1707 {
1708 return has_sha1_file_with_flags(oid->hash, flags);
1709 }
1710
1711 static void check_tree(const void *buf, size_t size)
1712 {
1713 struct tree_desc desc;
1714 struct name_entry entry;
1715
1716 init_tree_desc(&desc, buf, size);
1717 while (tree_entry(&desc, &entry))
1718 /* do nothing
1719 * tree_entry() will die() on malformed entries */
1720 ;
1721 }
1722
1723 static void check_commit(const void *buf, size_t size)
1724 {
1725 struct commit c;
1726 memset(&c, 0, sizeof(c));
1727 if (parse_commit_buffer(&c, buf, size))
1728 die("corrupt commit");
1729 }
1730
1731 static void check_tag(const void *buf, size_t size)
1732 {
1733 struct tag t;
1734 memset(&t, 0, sizeof(t));
1735 if (parse_tag_buffer(&t, buf, size))
1736 die("corrupt tag");
1737 }
1738
1739 static int index_mem(struct object_id *oid, void *buf, size_t size,
1740 enum object_type type,
1741 const char *path, unsigned flags)
1742 {
1743 int ret, re_allocated = 0;
1744 int write_object = flags & HASH_WRITE_OBJECT;
1745
1746 if (!type)
1747 type = OBJ_BLOB;
1748
1749 /*
1750 * Convert blobs to git internal format
1751 */
1752 if ((type == OBJ_BLOB) && path) {
1753 struct strbuf nbuf = STRBUF_INIT;
1754 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1755 get_safe_crlf(flags))) {
1756 buf = strbuf_detach(&nbuf, &size);
1757 re_allocated = 1;
1758 }
1759 }
1760 if (flags & HASH_FORMAT_CHECK) {
1761 if (type == OBJ_TREE)
1762 check_tree(buf, size);
1763 if (type == OBJ_COMMIT)
1764 check_commit(buf, size);
1765 if (type == OBJ_TAG)
1766 check_tag(buf, size);
1767 }
1768
1769 if (write_object)
1770 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1771 else
1772 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1773 if (re_allocated)
1774 free(buf);
1775 return ret;
1776 }
1777
1778 static int index_stream_convert_blob(struct object_id *oid, int fd,
1779 const char *path, unsigned flags)
1780 {
1781 int ret;
1782 const int write_object = flags & HASH_WRITE_OBJECT;
1783 struct strbuf sbuf = STRBUF_INIT;
1784
1785 assert(path);
1786 assert(would_convert_to_git_filter_fd(path));
1787
1788 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1789 get_safe_crlf(flags));
1790
1791 if (write_object)
1792 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1793 oid->hash);
1794 else
1795 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1796 oid->hash);
1797 strbuf_release(&sbuf);
1798 return ret;
1799 }
1800
1801 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1802 const char *path, unsigned flags)
1803 {
1804 struct strbuf sbuf = STRBUF_INIT;
1805 int ret;
1806
1807 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1808 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1809 else
1810 ret = -1;
1811 strbuf_release(&sbuf);
1812 return ret;
1813 }
1814
1815 #define SMALL_FILE_SIZE (32*1024)
1816
1817 static int index_core(struct object_id *oid, int fd, size_t size,
1818 enum object_type type, const char *path,
1819 unsigned flags)
1820 {
1821 int ret;
1822
1823 if (!size) {
1824 ret = index_mem(oid, "", size, type, path, flags);
1825 } else if (size <= SMALL_FILE_SIZE) {
1826 char *buf = xmalloc(size);
1827 ssize_t read_result = read_in_full(fd, buf, size);
1828 if (read_result < 0)
1829 ret = error_errno("read error while indexing %s",
1830 path ? path : "<unknown>");
1831 else if (read_result != size)
1832 ret = error("short read while indexing %s",
1833 path ? path : "<unknown>");
1834 else
1835 ret = index_mem(oid, buf, size, type, path, flags);
1836 free(buf);
1837 } else {
1838 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1839 ret = index_mem(oid, buf, size, type, path, flags);
1840 munmap(buf, size);
1841 }
1842 return ret;
1843 }
1844
1845 /*
1846 * This creates one packfile per large blob unless bulk-checkin
1847 * machinery is "plugged".
1848 *
1849 * This also bypasses the usual "convert-to-git" dance, and that is on
1850 * purpose. We could write a streaming version of the converting
1851 * functions and insert that before feeding the data to fast-import
1852 * (or equivalent in-core API described above). However, that is
1853 * somewhat complicated, as we do not know the size of the filter
1854 * result, which we need to know beforehand when writing a git object.
1855 * Since the primary motivation for trying to stream from the working
1856 * tree file and to avoid mmaping it in core is to deal with large
1857 * binary blobs, they generally do not want to get any conversion, and
1858 * callers should avoid this code path when filters are requested.
1859 */
1860 static int index_stream(struct object_id *oid, int fd, size_t size,
1861 enum object_type type, const char *path,
1862 unsigned flags)
1863 {
1864 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1865 }
1866
1867 int index_fd(struct object_id *oid, int fd, struct stat *st,
1868 enum object_type type, const char *path, unsigned flags)
1869 {
1870 int ret;
1871
1872 /*
1873 * Call xsize_t() only when needed to avoid potentially unnecessary
1874 * die() for large files.
1875 */
1876 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1877 ret = index_stream_convert_blob(oid, fd, path, flags);
1878 else if (!S_ISREG(st->st_mode))
1879 ret = index_pipe(oid, fd, type, path, flags);
1880 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1881 (path && would_convert_to_git(&the_index, path)))
1882 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1883 flags);
1884 else
1885 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1886 flags);
1887 close(fd);
1888 return ret;
1889 }
1890
1891 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1892 {
1893 int fd;
1894 struct strbuf sb = STRBUF_INIT;
1895 int rc = 0;
1896
1897 switch (st->st_mode & S_IFMT) {
1898 case S_IFREG:
1899 fd = open(path, O_RDONLY);
1900 if (fd < 0)
1901 return error_errno("open(\"%s\")", path);
1902 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1903 return error("%s: failed to insert into database",
1904 path);
1905 break;
1906 case S_IFLNK:
1907 if (strbuf_readlink(&sb, path, st->st_size))
1908 return error_errno("readlink(\"%s\")", path);
1909 if (!(flags & HASH_WRITE_OBJECT))
1910 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1911 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1912 rc = error("%s: failed to insert into database", path);
1913 strbuf_release(&sb);
1914 break;
1915 case S_IFDIR:
1916 return resolve_gitlink_ref(path, "HEAD", oid);
1917 default:
1918 return error("%s: unsupported file type", path);
1919 }
1920 return rc;
1921 }
1922
1923 int read_pack_header(int fd, struct pack_header *header)
1924 {
1925 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1926 /* "eof before pack header was fully read" */
1927 return PH_ERROR_EOF;
1928
1929 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1930 /* "protocol error (pack signature mismatch detected)" */
1931 return PH_ERROR_PACK_SIGNATURE;
1932 if (!pack_version_ok(header->hdr_version))
1933 /* "protocol error (pack version unsupported)" */
1934 return PH_ERROR_PROTOCOL;
1935 return 0;
1936 }
1937
1938 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1939 {
1940 enum object_type type = sha1_object_info(sha1, NULL);
1941 if (type < 0)
1942 die("%s is not a valid object", sha1_to_hex(sha1));
1943 if (type != expect)
1944 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1945 typename(expect));
1946 }
1947
1948 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1949 struct strbuf *path,
1950 each_loose_object_fn obj_cb,
1951 each_loose_cruft_fn cruft_cb,
1952 each_loose_subdir_fn subdir_cb,
1953 void *data)
1954 {
1955 size_t origlen, baselen;
1956 DIR *dir;
1957 struct dirent *de;
1958 int r = 0;
1959 struct object_id oid;
1960
1961 if (subdir_nr > 0xff)
1962 BUG("invalid loose object subdirectory: %x", subdir_nr);
1963
1964 origlen = path->len;
1965 strbuf_complete(path, '/');
1966 strbuf_addf(path, "%02x", subdir_nr);
1967
1968 dir = opendir(path->buf);
1969 if (!dir) {
1970 if (errno != ENOENT)
1971 r = error_errno("unable to open %s", path->buf);
1972 strbuf_setlen(path, origlen);
1973 return r;
1974 }
1975
1976 oid.hash[0] = subdir_nr;
1977 strbuf_addch(path, '/');
1978 baselen = path->len;
1979
1980 while ((de = readdir(dir))) {
1981 size_t namelen;
1982 if (is_dot_or_dotdot(de->d_name))
1983 continue;
1984
1985 namelen = strlen(de->d_name);
1986 strbuf_setlen(path, baselen);
1987 strbuf_add(path, de->d_name, namelen);
1988 if (namelen == GIT_SHA1_HEXSZ - 2 &&
1989 !hex_to_bytes(oid.hash + 1, de->d_name,
1990 GIT_SHA1_RAWSZ - 1)) {
1991 if (obj_cb) {
1992 r = obj_cb(&oid, path->buf, data);
1993 if (r)
1994 break;
1995 }
1996 continue;
1997 }
1998
1999 if (cruft_cb) {
2000 r = cruft_cb(de->d_name, path->buf, data);
2001 if (r)
2002 break;
2003 }
2004 }
2005 closedir(dir);
2006
2007 strbuf_setlen(path, baselen - 1);
2008 if (!r && subdir_cb)
2009 r = subdir_cb(subdir_nr, path->buf, data);
2010
2011 strbuf_setlen(path, origlen);
2012
2013 return r;
2014 }
2015
2016 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2017 each_loose_object_fn obj_cb,
2018 each_loose_cruft_fn cruft_cb,
2019 each_loose_subdir_fn subdir_cb,
2020 void *data)
2021 {
2022 int r = 0;
2023 int i;
2024
2025 for (i = 0; i < 256; i++) {
2026 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2027 subdir_cb, data);
2028 if (r)
2029 break;
2030 }
2031
2032 return r;
2033 }
2034
2035 int for_each_loose_file_in_objdir(const char *path,
2036 each_loose_object_fn obj_cb,
2037 each_loose_cruft_fn cruft_cb,
2038 each_loose_subdir_fn subdir_cb,
2039 void *data)
2040 {
2041 struct strbuf buf = STRBUF_INIT;
2042 int r;
2043
2044 strbuf_addstr(&buf, path);
2045 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2046 subdir_cb, data);
2047 strbuf_release(&buf);
2048
2049 return r;
2050 }
2051
2052 struct loose_alt_odb_data {
2053 each_loose_object_fn *cb;
2054 void *data;
2055 };
2056
2057 static int loose_from_alt_odb(struct alternate_object_database *alt,
2058 void *vdata)
2059 {
2060 struct loose_alt_odb_data *data = vdata;
2061 struct strbuf buf = STRBUF_INIT;
2062 int r;
2063
2064 strbuf_addstr(&buf, alt->path);
2065 r = for_each_loose_file_in_objdir_buf(&buf,
2066 data->cb, NULL, NULL,
2067 data->data);
2068 strbuf_release(&buf);
2069 return r;
2070 }
2071
2072 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2073 {
2074 struct loose_alt_odb_data alt;
2075 int r;
2076
2077 r = for_each_loose_file_in_objdir(get_object_directory(),
2078 cb, NULL, NULL, data);
2079 if (r)
2080 return r;
2081
2082 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2083 return 0;
2084
2085 alt.cb = cb;
2086 alt.data = data;
2087 return foreach_alt_odb(loose_from_alt_odb, &alt);
2088 }
2089
2090 static int check_stream_sha1(git_zstream *stream,
2091 const char *hdr,
2092 unsigned long size,
2093 const char *path,
2094 const unsigned char *expected_sha1)
2095 {
2096 git_SHA_CTX c;
2097 unsigned char real_sha1[GIT_MAX_RAWSZ];
2098 unsigned char buf[4096];
2099 unsigned long total_read;
2100 int status = Z_OK;
2101
2102 git_SHA1_Init(&c);
2103 git_SHA1_Update(&c, hdr, stream->total_out);
2104
2105 /*
2106 * We already read some bytes into hdr, but the ones up to the NUL
2107 * do not count against the object's content size.
2108 */
2109 total_read = stream->total_out - strlen(hdr) - 1;
2110
2111 /*
2112 * This size comparison must be "<=" to read the final zlib packets;
2113 * see the comment in unpack_sha1_rest for details.
2114 */
2115 while (total_read <= size &&
2116 (status == Z_OK || status == Z_BUF_ERROR)) {
2117 stream->next_out = buf;
2118 stream->avail_out = sizeof(buf);
2119 if (size - total_read < stream->avail_out)
2120 stream->avail_out = size - total_read;
2121 status = git_inflate(stream, Z_FINISH);
2122 git_SHA1_Update(&c, buf, stream->next_out - buf);
2123 total_read += stream->next_out - buf;
2124 }
2125 git_inflate_end(stream);
2126
2127 if (status != Z_STREAM_END) {
2128 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2129 return -1;
2130 }
2131 if (stream->avail_in) {
2132 error("garbage at end of loose object '%s'",
2133 sha1_to_hex(expected_sha1));
2134 return -1;
2135 }
2136
2137 git_SHA1_Final(real_sha1, &c);
2138 if (hashcmp(expected_sha1, real_sha1)) {
2139 error("sha1 mismatch for %s (expected %s)", path,
2140 sha1_to_hex(expected_sha1));
2141 return -1;
2142 }
2143
2144 return 0;
2145 }
2146
2147 int read_loose_object(const char *path,
2148 const unsigned char *expected_sha1,
2149 enum object_type *type,
2150 unsigned long *size,
2151 void **contents)
2152 {
2153 int ret = -1;
2154 void *map = NULL;
2155 unsigned long mapsize;
2156 git_zstream stream;
2157 char hdr[32];
2158
2159 *contents = NULL;
2160
2161 map = map_sha1_file_1(path, NULL, &mapsize);
2162 if (!map) {
2163 error_errno("unable to mmap %s", path);
2164 goto out;
2165 }
2166
2167 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2168 error("unable to unpack header of %s", path);
2169 goto out;
2170 }
2171
2172 *type = parse_sha1_header(hdr, size);
2173 if (*type < 0) {
2174 error("unable to parse header of %s", path);
2175 git_inflate_end(&stream);
2176 goto out;
2177 }
2178
2179 if (*type == OBJ_BLOB) {
2180 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2181 goto out;
2182 } else {
2183 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2184 if (!*contents) {
2185 error("unable to unpack contents of %s", path);
2186 git_inflate_end(&stream);
2187 goto out;
2188 }
2189 if (check_sha1_signature(expected_sha1, *contents,
2190 *size, typename(*type))) {
2191 error("sha1 mismatch for %s (expected %s)", path,
2192 sha1_to_hex(expected_sha1));
2193 free(*contents);
2194 goto out;
2195 }
2196 }
2197
2198 ret = 0; /* everything checks out */
2199
2200 out:
2201 if (map)
2202 munmap(map, mapsize);
2203 return ret;
2204 }