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