pack: move find_sha1_pack()
[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[20];
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 int mkdir_in_gitdir(const char *path)
78 {
79 if (mkdir(path, 0777)) {
80 int saved_errno = errno;
81 struct stat st;
82 struct strbuf sb = STRBUF_INIT;
83
84 if (errno != EEXIST)
85 return -1;
86 /*
87 * Are we looking at a path in a symlinked worktree
88 * whose original repository does not yet have it?
89 * e.g. .git/rr-cache pointing at its original
90 * repository in which the user hasn't performed any
91 * conflict resolution yet?
92 */
93 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
94 strbuf_readlink(&sb, path, st.st_size) ||
95 !is_absolute_path(sb.buf) ||
96 mkdir(sb.buf, 0777)) {
97 strbuf_release(&sb);
98 errno = saved_errno;
99 return -1;
100 }
101 strbuf_release(&sb);
102 }
103 return adjust_shared_perm(path);
104 }
105
106 enum scld_error safe_create_leading_directories(char *path)
107 {
108 char *next_component = path + offset_1st_component(path);
109 enum scld_error ret = SCLD_OK;
110
111 while (ret == SCLD_OK && next_component) {
112 struct stat st;
113 char *slash = next_component, slash_character;
114
115 while (*slash && !is_dir_sep(*slash))
116 slash++;
117
118 if (!*slash)
119 break;
120
121 next_component = slash + 1;
122 while (is_dir_sep(*next_component))
123 next_component++;
124 if (!*next_component)
125 break;
126
127 slash_character = *slash;
128 *slash = '\0';
129 if (!stat(path, &st)) {
130 /* path exists */
131 if (!S_ISDIR(st.st_mode)) {
132 errno = ENOTDIR;
133 ret = SCLD_EXISTS;
134 }
135 } else if (mkdir(path, 0777)) {
136 if (errno == EEXIST &&
137 !stat(path, &st) && S_ISDIR(st.st_mode))
138 ; /* somebody created it since we checked */
139 else if (errno == ENOENT)
140 /*
141 * Either mkdir() failed because
142 * somebody just pruned the containing
143 * directory, or stat() failed because
144 * the file that was in our way was
145 * just removed. Either way, inform
146 * the caller that it might be worth
147 * trying again:
148 */
149 ret = SCLD_VANISHED;
150 else
151 ret = SCLD_FAILED;
152 } else if (adjust_shared_perm(path)) {
153 ret = SCLD_PERMS;
154 }
155 *slash = slash_character;
156 }
157 return ret;
158 }
159
160 enum scld_error safe_create_leading_directories_const(const char *path)
161 {
162 int save_errno;
163 /* path points to cache entries, so xstrdup before messing with it */
164 char *buf = xstrdup(path);
165 enum scld_error result = safe_create_leading_directories(buf);
166
167 save_errno = errno;
168 free(buf);
169 errno = save_errno;
170 return result;
171 }
172
173 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
174 {
175 /*
176 * The number of times we will try to remove empty directories
177 * in the way of path. This is only 1 because if another
178 * process is racily creating directories that conflict with
179 * us, we don't want to fight against them.
180 */
181 int remove_directories_remaining = 1;
182
183 /*
184 * The number of times that we will try to create the
185 * directories containing path. We are willing to attempt this
186 * more than once, because another process could be trying to
187 * clean up empty directories at the same time as we are
188 * trying to create them.
189 */
190 int create_directories_remaining = 3;
191
192 /* A scratch copy of path, filled lazily if we need it: */
193 struct strbuf path_copy = STRBUF_INIT;
194
195 int ret, save_errno;
196
197 /* Sanity check: */
198 assert(*path);
199
200 retry_fn:
201 ret = fn(path, cb);
202 save_errno = errno;
203 if (!ret)
204 goto out;
205
206 if (errno == EISDIR && remove_directories_remaining-- > 0) {
207 /*
208 * A directory is in the way. Maybe it is empty; try
209 * to remove it:
210 */
211 if (!path_copy.len)
212 strbuf_addstr(&path_copy, path);
213
214 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
215 goto retry_fn;
216 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
217 /*
218 * Maybe the containing directory didn't exist, or
219 * maybe it was just deleted by a process that is
220 * racing with us to clean up empty directories. Try
221 * to create it:
222 */
223 enum scld_error scld_result;
224
225 if (!path_copy.len)
226 strbuf_addstr(&path_copy, path);
227
228 do {
229 scld_result = safe_create_leading_directories(path_copy.buf);
230 if (scld_result == SCLD_OK)
231 goto retry_fn;
232 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
233 }
234
235 out:
236 strbuf_release(&path_copy);
237 errno = save_errno;
238 return ret;
239 }
240
241 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
242 {
243 int i;
244 for (i = 0; i < 20; i++) {
245 static char hex[] = "0123456789abcdef";
246 unsigned int val = sha1[i];
247 strbuf_addch(buf, hex[val >> 4]);
248 strbuf_addch(buf, hex[val & 0xf]);
249 if (!i)
250 strbuf_addch(buf, '/');
251 }
252 }
253
254 const char *sha1_file_name(const unsigned char *sha1)
255 {
256 static struct strbuf buf = STRBUF_INIT;
257
258 strbuf_reset(&buf);
259 strbuf_addf(&buf, "%s/", get_object_directory());
260
261 fill_sha1_path(&buf, sha1);
262 return buf.buf;
263 }
264
265 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
266 {
267 strbuf_setlen(&alt->scratch, alt->base_len);
268 return &alt->scratch;
269 }
270
271 static const char *alt_sha1_path(struct alternate_object_database *alt,
272 const unsigned char *sha1)
273 {
274 struct strbuf *buf = alt_scratch_buf(alt);
275 fill_sha1_path(buf, sha1);
276 return buf->buf;
277 }
278
279 struct alternate_object_database *alt_odb_list;
280 static struct alternate_object_database **alt_odb_tail;
281
282 /*
283 * Return non-zero iff the path is usable as an alternate object database.
284 */
285 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
286 {
287 struct alternate_object_database *alt;
288
289 /* Detect cases where alternate disappeared */
290 if (!is_directory(path->buf)) {
291 error("object directory %s does not exist; "
292 "check .git/objects/info/alternates.",
293 path->buf);
294 return 0;
295 }
296
297 /*
298 * Prevent the common mistake of listing the same
299 * thing twice, or object directory itself.
300 */
301 for (alt = alt_odb_list; alt; alt = alt->next) {
302 if (!fspathcmp(path->buf, alt->path))
303 return 0;
304 }
305 if (!fspathcmp(path->buf, normalized_objdir))
306 return 0;
307
308 return 1;
309 }
310
311 /*
312 * Prepare alternate object database registry.
313 *
314 * The variable alt_odb_list points at the list of struct
315 * alternate_object_database. The elements on this list come from
316 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
317 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
318 * whose contents is similar to that environment variable but can be
319 * LF separated. Its base points at a statically allocated buffer that
320 * contains "/the/directory/corresponding/to/.git/objects/...", while
321 * its name points just after the slash at the end of ".git/objects/"
322 * in the example above, and has enough space to hold 40-byte hex
323 * SHA1, an extra slash for the first level indirection, and the
324 * terminating NUL.
325 */
326 static void read_info_alternates(const char * relative_base, int depth);
327 static int link_alt_odb_entry(const char *entry, const char *relative_base,
328 int depth, const char *normalized_objdir)
329 {
330 struct alternate_object_database *ent;
331 struct strbuf pathbuf = STRBUF_INIT;
332
333 if (!is_absolute_path(entry) && relative_base) {
334 strbuf_realpath(&pathbuf, relative_base, 1);
335 strbuf_addch(&pathbuf, '/');
336 }
337 strbuf_addstr(&pathbuf, entry);
338
339 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
340 error("unable to normalize alternate object path: %s",
341 pathbuf.buf);
342 strbuf_release(&pathbuf);
343 return -1;
344 }
345
346 /*
347 * The trailing slash after the directory name is given by
348 * this function at the end. Remove duplicates.
349 */
350 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
351 strbuf_setlen(&pathbuf, pathbuf.len - 1);
352
353 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
354 strbuf_release(&pathbuf);
355 return -1;
356 }
357
358 ent = alloc_alt_odb(pathbuf.buf);
359
360 /* add the alternate entry */
361 *alt_odb_tail = ent;
362 alt_odb_tail = &(ent->next);
363 ent->next = NULL;
364
365 /* recursively add alternates */
366 read_info_alternates(pathbuf.buf, depth + 1);
367
368 strbuf_release(&pathbuf);
369 return 0;
370 }
371
372 static const char *parse_alt_odb_entry(const char *string,
373 int sep,
374 struct strbuf *out)
375 {
376 const char *end;
377
378 strbuf_reset(out);
379
380 if (*string == '#') {
381 /* comment; consume up to next separator */
382 end = strchrnul(string, sep);
383 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
384 /*
385 * quoted path; unquote_c_style has copied the
386 * data for us and set "end". Broken quoting (e.g.,
387 * an entry that doesn't end with a quote) falls
388 * back to the unquoted case below.
389 */
390 } else {
391 /* normal, unquoted path */
392 end = strchrnul(string, sep);
393 strbuf_add(out, string, end - string);
394 }
395
396 if (*end)
397 end++;
398 return end;
399 }
400
401 static void link_alt_odb_entries(const char *alt, int len, int sep,
402 const char *relative_base, int depth)
403 {
404 struct strbuf objdirbuf = STRBUF_INIT;
405 struct strbuf entry = STRBUF_INIT;
406
407 if (depth > 5) {
408 error("%s: ignoring alternate object stores, nesting too deep.",
409 relative_base);
410 return;
411 }
412
413 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
414 if (strbuf_normalize_path(&objdirbuf) < 0)
415 die("unable to normalize object directory: %s",
416 objdirbuf.buf);
417
418 while (*alt) {
419 alt = parse_alt_odb_entry(alt, sep, &entry);
420 if (!entry.len)
421 continue;
422 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
423 }
424 strbuf_release(&entry);
425 strbuf_release(&objdirbuf);
426 }
427
428 static void read_info_alternates(const char * relative_base, int depth)
429 {
430 char *map;
431 size_t mapsz;
432 struct stat st;
433 char *path;
434 int fd;
435
436 path = xstrfmt("%s/info/alternates", relative_base);
437 fd = git_open(path);
438 free(path);
439 if (fd < 0)
440 return;
441 if (fstat(fd, &st) || (st.st_size == 0)) {
442 close(fd);
443 return;
444 }
445 mapsz = xsize_t(st.st_size);
446 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
447 close(fd);
448
449 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
450
451 munmap(map, mapsz);
452 }
453
454 struct alternate_object_database *alloc_alt_odb(const char *dir)
455 {
456 struct alternate_object_database *ent;
457
458 FLEX_ALLOC_STR(ent, path, dir);
459 strbuf_init(&ent->scratch, 0);
460 strbuf_addf(&ent->scratch, "%s/", dir);
461 ent->base_len = ent->scratch.len;
462
463 return ent;
464 }
465
466 void add_to_alternates_file(const char *reference)
467 {
468 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
469 char *alts = git_pathdup("objects/info/alternates");
470 FILE *in, *out;
471
472 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
473 out = fdopen_lock_file(lock, "w");
474 if (!out)
475 die_errno("unable to fdopen alternates lockfile");
476
477 in = fopen(alts, "r");
478 if (in) {
479 struct strbuf line = STRBUF_INIT;
480 int found = 0;
481
482 while (strbuf_getline(&line, in) != EOF) {
483 if (!strcmp(reference, line.buf)) {
484 found = 1;
485 break;
486 }
487 fprintf_or_die(out, "%s\n", line.buf);
488 }
489
490 strbuf_release(&line);
491 fclose(in);
492
493 if (found) {
494 rollback_lock_file(lock);
495 lock = NULL;
496 }
497 }
498 else if (errno != ENOENT)
499 die_errno("unable to read alternates file");
500
501 if (lock) {
502 fprintf_or_die(out, "%s\n", reference);
503 if (commit_lock_file(lock))
504 die_errno("unable to move new alternates file into place");
505 if (alt_odb_tail)
506 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
507 }
508 free(alts);
509 }
510
511 void add_to_alternates_memory(const char *reference)
512 {
513 /*
514 * Make sure alternates are initialized, or else our entry may be
515 * overwritten when they are.
516 */
517 prepare_alt_odb();
518
519 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
520 }
521
522 /*
523 * Compute the exact path an alternate is at and returns it. In case of
524 * error NULL is returned and the human readable error is added to `err`
525 * `path` may be relative and should point to $GITDIR.
526 * `err` must not be null.
527 */
528 char *compute_alternate_path(const char *path, struct strbuf *err)
529 {
530 char *ref_git = NULL;
531 const char *repo, *ref_git_s;
532 int seen_error = 0;
533
534 ref_git_s = real_path_if_valid(path);
535 if (!ref_git_s) {
536 seen_error = 1;
537 strbuf_addf(err, _("path '%s' does not exist"), path);
538 goto out;
539 } else
540 /*
541 * Beware: read_gitfile(), real_path() and mkpath()
542 * return static buffer
543 */
544 ref_git = xstrdup(ref_git_s);
545
546 repo = read_gitfile(ref_git);
547 if (!repo)
548 repo = read_gitfile(mkpath("%s/.git", ref_git));
549 if (repo) {
550 free(ref_git);
551 ref_git = xstrdup(repo);
552 }
553
554 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
555 char *ref_git_git = mkpathdup("%s/.git", ref_git);
556 free(ref_git);
557 ref_git = ref_git_git;
558 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
559 struct strbuf sb = STRBUF_INIT;
560 seen_error = 1;
561 if (get_common_dir(&sb, ref_git)) {
562 strbuf_addf(err,
563 _("reference repository '%s' as a linked "
564 "checkout is not supported yet."),
565 path);
566 goto out;
567 }
568
569 strbuf_addf(err, _("reference repository '%s' is not a "
570 "local repository."), path);
571 goto out;
572 }
573
574 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
575 strbuf_addf(err, _("reference repository '%s' is shallow"),
576 path);
577 seen_error = 1;
578 goto out;
579 }
580
581 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
582 strbuf_addf(err,
583 _("reference repository '%s' is grafted"),
584 path);
585 seen_error = 1;
586 goto out;
587 }
588
589 out:
590 if (seen_error) {
591 FREE_AND_NULL(ref_git);
592 }
593
594 return ref_git;
595 }
596
597 int foreach_alt_odb(alt_odb_fn fn, void *cb)
598 {
599 struct alternate_object_database *ent;
600 int r = 0;
601
602 prepare_alt_odb();
603 for (ent = alt_odb_list; ent; ent = ent->next) {
604 r = fn(ent, cb);
605 if (r)
606 break;
607 }
608 return r;
609 }
610
611 void prepare_alt_odb(void)
612 {
613 const char *alt;
614
615 if (alt_odb_tail)
616 return;
617
618 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
619 if (!alt) alt = "";
620
621 alt_odb_tail = &alt_odb_list;
622 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
623
624 read_info_alternates(get_object_directory(), 0);
625 }
626
627 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
628 static int freshen_file(const char *fn)
629 {
630 struct utimbuf t;
631 t.actime = t.modtime = time(NULL);
632 return !utime(fn, &t);
633 }
634
635 /*
636 * All of the check_and_freshen functions return 1 if the file exists and was
637 * freshened (if freshening was requested), 0 otherwise. If they return
638 * 0, you should not assume that it is safe to skip a write of the object (it
639 * either does not exist on disk, or has a stale mtime and may be subject to
640 * pruning).
641 */
642 int check_and_freshen_file(const char *fn, int freshen)
643 {
644 if (access(fn, F_OK))
645 return 0;
646 if (freshen && !freshen_file(fn))
647 return 0;
648 return 1;
649 }
650
651 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
652 {
653 return check_and_freshen_file(sha1_file_name(sha1), freshen);
654 }
655
656 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
657 {
658 struct alternate_object_database *alt;
659 prepare_alt_odb();
660 for (alt = alt_odb_list; alt; alt = alt->next) {
661 const char *path = alt_sha1_path(alt, sha1);
662 if (check_and_freshen_file(path, freshen))
663 return 1;
664 }
665 return 0;
666 }
667
668 static int check_and_freshen(const unsigned char *sha1, int freshen)
669 {
670 return check_and_freshen_local(sha1, freshen) ||
671 check_and_freshen_nonlocal(sha1, freshen);
672 }
673
674 int has_loose_object_nonlocal(const unsigned char *sha1)
675 {
676 return check_and_freshen_nonlocal(sha1, 0);
677 }
678
679 static int has_loose_object(const unsigned char *sha1)
680 {
681 return check_and_freshen(sha1, 0);
682 }
683
684 static void mmap_limit_check(size_t length)
685 {
686 static size_t limit = 0;
687 if (!limit) {
688 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
689 if (!limit)
690 limit = SIZE_MAX;
691 }
692 if (length > limit)
693 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
694 (uintmax_t)length, (uintmax_t)limit);
695 }
696
697 void *xmmap_gently(void *start, size_t length,
698 int prot, int flags, int fd, off_t offset)
699 {
700 void *ret;
701
702 mmap_limit_check(length);
703 ret = mmap(start, length, prot, flags, fd, offset);
704 if (ret == MAP_FAILED) {
705 if (!length)
706 return NULL;
707 release_pack_memory(length);
708 ret = mmap(start, length, prot, flags, fd, offset);
709 }
710 return ret;
711 }
712
713 void *xmmap(void *start, size_t length,
714 int prot, int flags, int fd, off_t offset)
715 {
716 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
717 if (ret == MAP_FAILED)
718 die_errno("mmap failed");
719 return ret;
720 }
721
722 /*
723 * With an in-core object data in "map", rehash it to make sure the
724 * object name actually matches "sha1" to detect object corruption.
725 * With "map" == NULL, try reading the object named with "sha1" using
726 * the streaming interface and rehash it to do the same.
727 */
728 int check_sha1_signature(const unsigned char *sha1, void *map,
729 unsigned long size, const char *type)
730 {
731 unsigned char real_sha1[20];
732 enum object_type obj_type;
733 struct git_istream *st;
734 git_SHA_CTX c;
735 char hdr[32];
736 int hdrlen;
737
738 if (map) {
739 hash_sha1_file(map, size, type, real_sha1);
740 return hashcmp(sha1, real_sha1) ? -1 : 0;
741 }
742
743 st = open_istream(sha1, &obj_type, &size, NULL);
744 if (!st)
745 return -1;
746
747 /* Generate the header */
748 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
749
750 /* Sha1.. */
751 git_SHA1_Init(&c);
752 git_SHA1_Update(&c, hdr, hdrlen);
753 for (;;) {
754 char buf[1024 * 16];
755 ssize_t readlen = read_istream(st, buf, sizeof(buf));
756
757 if (readlen < 0) {
758 close_istream(st);
759 return -1;
760 }
761 if (!readlen)
762 break;
763 git_SHA1_Update(&c, buf, readlen);
764 }
765 git_SHA1_Final(real_sha1, &c);
766 close_istream(st);
767 return hashcmp(sha1, real_sha1) ? -1 : 0;
768 }
769
770 int git_open_cloexec(const char *name, int flags)
771 {
772 int fd;
773 static int o_cloexec = O_CLOEXEC;
774
775 fd = open(name, flags | o_cloexec);
776 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
777 /* Try again w/o O_CLOEXEC: the kernel might not support it */
778 o_cloexec &= ~O_CLOEXEC;
779 fd = open(name, flags | o_cloexec);
780 }
781
782 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
783 {
784 static int fd_cloexec = FD_CLOEXEC;
785
786 if (!o_cloexec && 0 <= fd && fd_cloexec) {
787 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
788 int flags = fcntl(fd, F_GETFD);
789 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
790 fd_cloexec = 0;
791 }
792 }
793 #endif
794 return fd;
795 }
796
797 /*
798 * Find "sha1" as a loose object in the local repository or in an alternate.
799 * Returns 0 on success, negative on failure.
800 *
801 * The "path" out-parameter will give the path of the object we found (if any).
802 * Note that it may point to static storage and is only valid until another
803 * call to sha1_file_name(), etc.
804 */
805 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
806 const char **path)
807 {
808 struct alternate_object_database *alt;
809
810 *path = sha1_file_name(sha1);
811 if (!lstat(*path, st))
812 return 0;
813
814 prepare_alt_odb();
815 errno = ENOENT;
816 for (alt = alt_odb_list; alt; alt = alt->next) {
817 *path = alt_sha1_path(alt, sha1);
818 if (!lstat(*path, st))
819 return 0;
820 }
821
822 return -1;
823 }
824
825 /*
826 * Like stat_sha1_file(), but actually open the object and return the
827 * descriptor. See the caveats on the "path" parameter above.
828 */
829 static int open_sha1_file(const unsigned char *sha1, const char **path)
830 {
831 int fd;
832 struct alternate_object_database *alt;
833 int most_interesting_errno;
834
835 *path = sha1_file_name(sha1);
836 fd = git_open(*path);
837 if (fd >= 0)
838 return fd;
839 most_interesting_errno = errno;
840
841 prepare_alt_odb();
842 for (alt = alt_odb_list; alt; alt = alt->next) {
843 *path = alt_sha1_path(alt, sha1);
844 fd = git_open(*path);
845 if (fd >= 0)
846 return fd;
847 if (most_interesting_errno == ENOENT)
848 most_interesting_errno = errno;
849 }
850 errno = most_interesting_errno;
851 return -1;
852 }
853
854 /*
855 * Map the loose object at "path" if it is not NULL, or the path found by
856 * searching for a loose object named "sha1".
857 */
858 static void *map_sha1_file_1(const char *path,
859 const unsigned char *sha1,
860 unsigned long *size)
861 {
862 void *map;
863 int fd;
864
865 if (path)
866 fd = git_open(path);
867 else
868 fd = open_sha1_file(sha1, &path);
869 map = NULL;
870 if (fd >= 0) {
871 struct stat st;
872
873 if (!fstat(fd, &st)) {
874 *size = xsize_t(st.st_size);
875 if (!*size) {
876 /* mmap() is forbidden on empty files */
877 error("object file %s is empty", path);
878 return NULL;
879 }
880 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
881 }
882 close(fd);
883 }
884 return map;
885 }
886
887 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
888 {
889 return map_sha1_file_1(NULL, sha1, size);
890 }
891
892 static int unpack_sha1_short_header(git_zstream *stream,
893 unsigned char *map, unsigned long mapsize,
894 void *buffer, unsigned long bufsiz)
895 {
896 /* Get the data stream */
897 memset(stream, 0, sizeof(*stream));
898 stream->next_in = map;
899 stream->avail_in = mapsize;
900 stream->next_out = buffer;
901 stream->avail_out = bufsiz;
902
903 git_inflate_init(stream);
904 return git_inflate(stream, 0);
905 }
906
907 int unpack_sha1_header(git_zstream *stream,
908 unsigned char *map, unsigned long mapsize,
909 void *buffer, unsigned long bufsiz)
910 {
911 int status = unpack_sha1_short_header(stream, map, mapsize,
912 buffer, bufsiz);
913
914 if (status < Z_OK)
915 return status;
916
917 /* Make sure we have the terminating NUL */
918 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
919 return -1;
920 return 0;
921 }
922
923 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
924 unsigned long mapsize, void *buffer,
925 unsigned long bufsiz, struct strbuf *header)
926 {
927 int status;
928
929 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
930 if (status < Z_OK)
931 return -1;
932
933 /*
934 * Check if entire header is unpacked in the first iteration.
935 */
936 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
937 return 0;
938
939 /*
940 * buffer[0..bufsiz] was not large enough. Copy the partial
941 * result out to header, and then append the result of further
942 * reading the stream.
943 */
944 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
945 stream->next_out = buffer;
946 stream->avail_out = bufsiz;
947
948 do {
949 status = git_inflate(stream, 0);
950 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
951 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
952 return 0;
953 stream->next_out = buffer;
954 stream->avail_out = bufsiz;
955 } while (status != Z_STREAM_END);
956 return -1;
957 }
958
959 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
960 {
961 int bytes = strlen(buffer) + 1;
962 unsigned char *buf = xmallocz(size);
963 unsigned long n;
964 int status = Z_OK;
965
966 n = stream->total_out - bytes;
967 if (n > size)
968 n = size;
969 memcpy(buf, (char *) buffer + bytes, n);
970 bytes = n;
971 if (bytes <= size) {
972 /*
973 * The above condition must be (bytes <= size), not
974 * (bytes < size). In other words, even though we
975 * expect no more output and set avail_out to zero,
976 * the input zlib stream may have bytes that express
977 * "this concludes the stream", and we *do* want to
978 * eat that input.
979 *
980 * Otherwise we would not be able to test that we
981 * consumed all the input to reach the expected size;
982 * we also want to check that zlib tells us that all
983 * went well with status == Z_STREAM_END at the end.
984 */
985 stream->next_out = buf + bytes;
986 stream->avail_out = size - bytes;
987 while (status == Z_OK)
988 status = git_inflate(stream, Z_FINISH);
989 }
990 if (status == Z_STREAM_END && !stream->avail_in) {
991 git_inflate_end(stream);
992 return buf;
993 }
994
995 if (status < 0)
996 error("corrupt loose object '%s'", sha1_to_hex(sha1));
997 else if (stream->avail_in)
998 error("garbage at end of loose object '%s'",
999 sha1_to_hex(sha1));
1000 free(buf);
1001 return NULL;
1002 }
1003
1004 /*
1005 * We used to just use "sscanf()", but that's actually way
1006 * too permissive for what we want to check. So do an anal
1007 * object header parse by hand.
1008 */
1009 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1010 unsigned int flags)
1011 {
1012 const char *type_buf = hdr;
1013 unsigned long size;
1014 int type, type_len = 0;
1015
1016 /*
1017 * The type can be of any size but is followed by
1018 * a space.
1019 */
1020 for (;;) {
1021 char c = *hdr++;
1022 if (!c)
1023 return -1;
1024 if (c == ' ')
1025 break;
1026 type_len++;
1027 }
1028
1029 type = type_from_string_gently(type_buf, type_len, 1);
1030 if (oi->typename)
1031 strbuf_add(oi->typename, type_buf, type_len);
1032 /*
1033 * Set type to 0 if its an unknown object and
1034 * we're obtaining the type using '--allow-unknown-type'
1035 * option.
1036 */
1037 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1038 type = 0;
1039 else if (type < 0)
1040 die("invalid object type");
1041 if (oi->typep)
1042 *oi->typep = type;
1043
1044 /*
1045 * The length must follow immediately, and be in canonical
1046 * decimal format (ie "010" is not valid).
1047 */
1048 size = *hdr++ - '0';
1049 if (size > 9)
1050 return -1;
1051 if (size) {
1052 for (;;) {
1053 unsigned long c = *hdr - '0';
1054 if (c > 9)
1055 break;
1056 hdr++;
1057 size = size * 10 + c;
1058 }
1059 }
1060
1061 if (oi->sizep)
1062 *oi->sizep = size;
1063
1064 /*
1065 * The length must be followed by a zero byte
1066 */
1067 return *hdr ? -1 : type;
1068 }
1069
1070 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1071 {
1072 struct object_info oi = OBJECT_INFO_INIT;
1073
1074 oi.sizep = sizep;
1075 return parse_sha1_header_extended(hdr, &oi, 0);
1076 }
1077
1078 static int fill_pack_entry(const unsigned char *sha1,
1079 struct pack_entry *e,
1080 struct packed_git *p)
1081 {
1082 off_t offset;
1083
1084 if (p->num_bad_objects) {
1085 unsigned i;
1086 for (i = 0; i < p->num_bad_objects; i++)
1087 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1088 return 0;
1089 }
1090
1091 offset = find_pack_entry_one(sha1, p);
1092 if (!offset)
1093 return 0;
1094
1095 /*
1096 * We are about to tell the caller where they can locate the
1097 * requested object. We better make sure the packfile is
1098 * still here and can be accessed before supplying that
1099 * answer, as it may have been deleted since the index was
1100 * loaded!
1101 */
1102 if (!is_pack_valid(p))
1103 return 0;
1104 e->offset = offset;
1105 e->p = p;
1106 hashcpy(e->sha1, sha1);
1107 return 1;
1108 }
1109
1110 /*
1111 * Iff a pack file contains the object named by sha1, return true and
1112 * store its location to e.
1113 */
1114 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1115 {
1116 struct mru_entry *p;
1117
1118 prepare_packed_git();
1119 if (!packed_git)
1120 return 0;
1121
1122 for (p = packed_git_mru->head; p; p = p->next) {
1123 if (fill_pack_entry(sha1, e, p->item)) {
1124 mru_mark(packed_git_mru, p);
1125 return 1;
1126 }
1127 }
1128 return 0;
1129 }
1130
1131 static int sha1_loose_object_info(const unsigned char *sha1,
1132 struct object_info *oi,
1133 int flags)
1134 {
1135 int status = 0;
1136 unsigned long mapsize;
1137 void *map;
1138 git_zstream stream;
1139 char hdr[32];
1140 struct strbuf hdrbuf = STRBUF_INIT;
1141 unsigned long size_scratch;
1142
1143 if (oi->delta_base_sha1)
1144 hashclr(oi->delta_base_sha1);
1145
1146 /*
1147 * If we don't care about type or size, then we don't
1148 * need to look inside the object at all. Note that we
1149 * do not optimize out the stat call, even if the
1150 * caller doesn't care about the disk-size, since our
1151 * return value implicitly indicates whether the
1152 * object even exists.
1153 */
1154 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1155 const char *path;
1156 struct stat st;
1157 if (stat_sha1_file(sha1, &st, &path) < 0)
1158 return -1;
1159 if (oi->disk_sizep)
1160 *oi->disk_sizep = st.st_size;
1161 return 0;
1162 }
1163
1164 map = map_sha1_file(sha1, &mapsize);
1165 if (!map)
1166 return -1;
1167
1168 if (!oi->sizep)
1169 oi->sizep = &size_scratch;
1170
1171 if (oi->disk_sizep)
1172 *oi->disk_sizep = mapsize;
1173 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1174 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1175 status = error("unable to unpack %s header with --allow-unknown-type",
1176 sha1_to_hex(sha1));
1177 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1178 status = error("unable to unpack %s header",
1179 sha1_to_hex(sha1));
1180 if (status < 0)
1181 ; /* Do nothing */
1182 else if (hdrbuf.len) {
1183 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1184 status = error("unable to parse %s header with --allow-unknown-type",
1185 sha1_to_hex(sha1));
1186 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1187 status = error("unable to parse %s header", sha1_to_hex(sha1));
1188
1189 if (status >= 0 && oi->contentp)
1190 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1191 *oi->sizep, sha1);
1192 else
1193 git_inflate_end(&stream);
1194
1195 munmap(map, mapsize);
1196 if (status && oi->typep)
1197 *oi->typep = status;
1198 if (oi->sizep == &size_scratch)
1199 oi->sizep = NULL;
1200 strbuf_release(&hdrbuf);
1201 oi->whence = OI_LOOSE;
1202 return (status < 0) ? status : 0;
1203 }
1204
1205 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1206 {
1207 static struct object_info blank_oi = OBJECT_INFO_INIT;
1208 struct pack_entry e;
1209 int rtype;
1210 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1211 lookup_replace_object(sha1) :
1212 sha1;
1213
1214 if (!oi)
1215 oi = &blank_oi;
1216
1217 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1218 struct cached_object *co = find_cached_object(real);
1219 if (co) {
1220 if (oi->typep)
1221 *(oi->typep) = co->type;
1222 if (oi->sizep)
1223 *(oi->sizep) = co->size;
1224 if (oi->disk_sizep)
1225 *(oi->disk_sizep) = 0;
1226 if (oi->delta_base_sha1)
1227 hashclr(oi->delta_base_sha1);
1228 if (oi->typename)
1229 strbuf_addstr(oi->typename, typename(co->type));
1230 if (oi->contentp)
1231 *oi->contentp = xmemdupz(co->buf, co->size);
1232 oi->whence = OI_CACHED;
1233 return 0;
1234 }
1235 }
1236
1237 if (!find_pack_entry(real, &e)) {
1238 /* Most likely it's a loose object. */
1239 if (!sha1_loose_object_info(real, oi, flags))
1240 return 0;
1241
1242 /* Not a loose object; someone else may have just packed it. */
1243 if (flags & OBJECT_INFO_QUICK) {
1244 return -1;
1245 } else {
1246 reprepare_packed_git();
1247 if (!find_pack_entry(real, &e))
1248 return -1;
1249 }
1250 }
1251
1252 if (oi == &blank_oi)
1253 /*
1254 * We know that the caller doesn't actually need the
1255 * information below, so return early.
1256 */
1257 return 0;
1258
1259 rtype = packed_object_info(e.p, e.offset, oi);
1260 if (rtype < 0) {
1261 mark_bad_packed_object(e.p, real);
1262 return sha1_object_info_extended(real, oi, 0);
1263 } else if (oi->whence == OI_PACKED) {
1264 oi->u.packed.offset = e.offset;
1265 oi->u.packed.pack = e.p;
1266 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1267 rtype == OBJ_OFS_DELTA);
1268 }
1269
1270 return 0;
1271 }
1272
1273 /* returns enum object_type or negative */
1274 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1275 {
1276 enum object_type type;
1277 struct object_info oi = OBJECT_INFO_INIT;
1278
1279 oi.typep = &type;
1280 oi.sizep = sizep;
1281 if (sha1_object_info_extended(sha1, &oi,
1282 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1283 return -1;
1284 return type;
1285 }
1286
1287 static void *read_object(const unsigned char *sha1, enum object_type *type,
1288 unsigned long *size)
1289 {
1290 struct object_info oi = OBJECT_INFO_INIT;
1291 void *content;
1292 oi.typep = type;
1293 oi.sizep = size;
1294 oi.contentp = &content;
1295
1296 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1297 return NULL;
1298 return content;
1299 }
1300
1301 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1302 unsigned char *sha1)
1303 {
1304 struct cached_object *co;
1305
1306 hash_sha1_file(buf, len, typename(type), sha1);
1307 if (has_sha1_file(sha1) || find_cached_object(sha1))
1308 return 0;
1309 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1310 co = &cached_objects[cached_object_nr++];
1311 co->size = len;
1312 co->type = type;
1313 co->buf = xmalloc(len);
1314 memcpy(co->buf, buf, len);
1315 hashcpy(co->sha1, sha1);
1316 return 0;
1317 }
1318
1319 /*
1320 * This function dies on corrupt objects; the callers who want to
1321 * deal with them should arrange to call read_object() and give error
1322 * messages themselves.
1323 */
1324 void *read_sha1_file_extended(const unsigned char *sha1,
1325 enum object_type *type,
1326 unsigned long *size,
1327 int lookup_replace)
1328 {
1329 void *data;
1330 const struct packed_git *p;
1331 const char *path;
1332 struct stat st;
1333 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1334 : sha1;
1335
1336 errno = 0;
1337 data = read_object(repl, type, size);
1338 if (data)
1339 return data;
1340
1341 if (errno && errno != ENOENT)
1342 die_errno("failed to read object %s", sha1_to_hex(sha1));
1343
1344 /* die if we replaced an object with one that does not exist */
1345 if (repl != sha1)
1346 die("replacement %s not found for %s",
1347 sha1_to_hex(repl), sha1_to_hex(sha1));
1348
1349 if (!stat_sha1_file(repl, &st, &path))
1350 die("loose object %s (stored in %s) is corrupt",
1351 sha1_to_hex(repl), path);
1352
1353 if ((p = has_packed_and_bad(repl)) != NULL)
1354 die("packed object %s (stored in %s) is corrupt",
1355 sha1_to_hex(repl), p->pack_name);
1356
1357 return NULL;
1358 }
1359
1360 void *read_object_with_reference(const unsigned char *sha1,
1361 const char *required_type_name,
1362 unsigned long *size,
1363 unsigned char *actual_sha1_return)
1364 {
1365 enum object_type type, required_type;
1366 void *buffer;
1367 unsigned long isize;
1368 unsigned char actual_sha1[20];
1369
1370 required_type = type_from_string(required_type_name);
1371 hashcpy(actual_sha1, sha1);
1372 while (1) {
1373 int ref_length = -1;
1374 const char *ref_type = NULL;
1375
1376 buffer = read_sha1_file(actual_sha1, &type, &isize);
1377 if (!buffer)
1378 return NULL;
1379 if (type == required_type) {
1380 *size = isize;
1381 if (actual_sha1_return)
1382 hashcpy(actual_sha1_return, actual_sha1);
1383 return buffer;
1384 }
1385 /* Handle references */
1386 else if (type == OBJ_COMMIT)
1387 ref_type = "tree ";
1388 else if (type == OBJ_TAG)
1389 ref_type = "object ";
1390 else {
1391 free(buffer);
1392 return NULL;
1393 }
1394 ref_length = strlen(ref_type);
1395
1396 if (ref_length + 40 > isize ||
1397 memcmp(buffer, ref_type, ref_length) ||
1398 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1399 free(buffer);
1400 return NULL;
1401 }
1402 free(buffer);
1403 /* Now we have the ID of the referred-to object in
1404 * actual_sha1. Check again. */
1405 }
1406 }
1407
1408 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1409 const char *type, unsigned char *sha1,
1410 char *hdr, int *hdrlen)
1411 {
1412 git_SHA_CTX c;
1413
1414 /* Generate the header */
1415 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1416
1417 /* Sha1.. */
1418 git_SHA1_Init(&c);
1419 git_SHA1_Update(&c, hdr, *hdrlen);
1420 git_SHA1_Update(&c, buf, len);
1421 git_SHA1_Final(sha1, &c);
1422 }
1423
1424 /*
1425 * Move the just written object into its final resting place.
1426 */
1427 int finalize_object_file(const char *tmpfile, const char *filename)
1428 {
1429 int ret = 0;
1430
1431 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1432 goto try_rename;
1433 else if (link(tmpfile, filename))
1434 ret = errno;
1435
1436 /*
1437 * Coda hack - coda doesn't like cross-directory links,
1438 * so we fall back to a rename, which will mean that it
1439 * won't be able to check collisions, but that's not a
1440 * big deal.
1441 *
1442 * The same holds for FAT formatted media.
1443 *
1444 * When this succeeds, we just return. We have nothing
1445 * left to unlink.
1446 */
1447 if (ret && ret != EEXIST) {
1448 try_rename:
1449 if (!rename(tmpfile, filename))
1450 goto out;
1451 ret = errno;
1452 }
1453 unlink_or_warn(tmpfile);
1454 if (ret) {
1455 if (ret != EEXIST) {
1456 return error_errno("unable to write sha1 filename %s", filename);
1457 }
1458 /* FIXME!!! Collision check here ? */
1459 }
1460
1461 out:
1462 if (adjust_shared_perm(filename))
1463 return error("unable to set permission to '%s'", filename);
1464 return 0;
1465 }
1466
1467 static int write_buffer(int fd, const void *buf, size_t len)
1468 {
1469 if (write_in_full(fd, buf, len) < 0)
1470 return error_errno("file write error");
1471 return 0;
1472 }
1473
1474 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1475 unsigned char *sha1)
1476 {
1477 char hdr[32];
1478 int hdrlen = sizeof(hdr);
1479 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1480 return 0;
1481 }
1482
1483 /* Finalize a file on disk, and close it. */
1484 static void close_sha1_file(int fd)
1485 {
1486 if (fsync_object_files)
1487 fsync_or_die(fd, "sha1 file");
1488 if (close(fd) != 0)
1489 die_errno("error when closing sha1 file");
1490 }
1491
1492 /* Size of directory component, including the ending '/' */
1493 static inline int directory_size(const char *filename)
1494 {
1495 const char *s = strrchr(filename, '/');
1496 if (!s)
1497 return 0;
1498 return s - filename + 1;
1499 }
1500
1501 /*
1502 * This creates a temporary file in the same directory as the final
1503 * 'filename'
1504 *
1505 * We want to avoid cross-directory filename renames, because those
1506 * can have problems on various filesystems (FAT, NFS, Coda).
1507 */
1508 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1509 {
1510 int fd, dirlen = directory_size(filename);
1511
1512 strbuf_reset(tmp);
1513 strbuf_add(tmp, filename, dirlen);
1514 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1515 fd = git_mkstemp_mode(tmp->buf, 0444);
1516 if (fd < 0 && dirlen && errno == ENOENT) {
1517 /*
1518 * Make sure the directory exists; note that the contents
1519 * of the buffer are undefined after mkstemp returns an
1520 * error, so we have to rewrite the whole buffer from
1521 * scratch.
1522 */
1523 strbuf_reset(tmp);
1524 strbuf_add(tmp, filename, dirlen - 1);
1525 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1526 return -1;
1527 if (adjust_shared_perm(tmp->buf))
1528 return -1;
1529
1530 /* Try again */
1531 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1532 fd = git_mkstemp_mode(tmp->buf, 0444);
1533 }
1534 return fd;
1535 }
1536
1537 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1538 const void *buf, unsigned long len, time_t mtime)
1539 {
1540 int fd, ret;
1541 unsigned char compressed[4096];
1542 git_zstream stream;
1543 git_SHA_CTX c;
1544 unsigned char parano_sha1[20];
1545 static struct strbuf tmp_file = STRBUF_INIT;
1546 const char *filename = sha1_file_name(sha1);
1547
1548 fd = create_tmpfile(&tmp_file, filename);
1549 if (fd < 0) {
1550 if (errno == EACCES)
1551 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1552 else
1553 return error_errno("unable to create temporary file");
1554 }
1555
1556 /* Set it up */
1557 git_deflate_init(&stream, zlib_compression_level);
1558 stream.next_out = compressed;
1559 stream.avail_out = sizeof(compressed);
1560 git_SHA1_Init(&c);
1561
1562 /* First header.. */
1563 stream.next_in = (unsigned char *)hdr;
1564 stream.avail_in = hdrlen;
1565 while (git_deflate(&stream, 0) == Z_OK)
1566 ; /* nothing */
1567 git_SHA1_Update(&c, hdr, hdrlen);
1568
1569 /* Then the data itself.. */
1570 stream.next_in = (void *)buf;
1571 stream.avail_in = len;
1572 do {
1573 unsigned char *in0 = stream.next_in;
1574 ret = git_deflate(&stream, Z_FINISH);
1575 git_SHA1_Update(&c, in0, stream.next_in - in0);
1576 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1577 die("unable to write sha1 file");
1578 stream.next_out = compressed;
1579 stream.avail_out = sizeof(compressed);
1580 } while (ret == Z_OK);
1581
1582 if (ret != Z_STREAM_END)
1583 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1584 ret = git_deflate_end_gently(&stream);
1585 if (ret != Z_OK)
1586 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1587 git_SHA1_Final(parano_sha1, &c);
1588 if (hashcmp(sha1, parano_sha1) != 0)
1589 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1590
1591 close_sha1_file(fd);
1592
1593 if (mtime) {
1594 struct utimbuf utb;
1595 utb.actime = mtime;
1596 utb.modtime = mtime;
1597 if (utime(tmp_file.buf, &utb) < 0)
1598 warning_errno("failed utime() on %s", tmp_file.buf);
1599 }
1600
1601 return finalize_object_file(tmp_file.buf, filename);
1602 }
1603
1604 static int freshen_loose_object(const unsigned char *sha1)
1605 {
1606 return check_and_freshen(sha1, 1);
1607 }
1608
1609 static int freshen_packed_object(const unsigned char *sha1)
1610 {
1611 struct pack_entry e;
1612 if (!find_pack_entry(sha1, &e))
1613 return 0;
1614 if (e.p->freshened)
1615 return 1;
1616 if (!freshen_file(e.p->pack_name))
1617 return 0;
1618 e.p->freshened = 1;
1619 return 1;
1620 }
1621
1622 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1623 {
1624 char hdr[32];
1625 int hdrlen = sizeof(hdr);
1626
1627 /* Normally if we have it in the pack then we do not bother writing
1628 * it out into .git/objects/??/?{38} file.
1629 */
1630 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1631 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1632 return 0;
1633 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1634 }
1635
1636 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1637 unsigned char *sha1, unsigned flags)
1638 {
1639 char *header;
1640 int hdrlen, status = 0;
1641
1642 /* type string, SP, %lu of the length plus NUL must fit this */
1643 hdrlen = strlen(type) + 32;
1644 header = xmalloc(hdrlen);
1645 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
1646
1647 if (!(flags & HASH_WRITE_OBJECT))
1648 goto cleanup;
1649 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1650 goto cleanup;
1651 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
1652
1653 cleanup:
1654 free(header);
1655 return status;
1656 }
1657
1658 int force_object_loose(const unsigned char *sha1, time_t mtime)
1659 {
1660 void *buf;
1661 unsigned long len;
1662 enum object_type type;
1663 char hdr[32];
1664 int hdrlen;
1665 int ret;
1666
1667 if (has_loose_object(sha1))
1668 return 0;
1669 buf = read_object(sha1, &type, &len);
1670 if (!buf)
1671 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1672 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1673 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1674 free(buf);
1675
1676 return ret;
1677 }
1678
1679 int has_pack_index(const unsigned char *sha1)
1680 {
1681 struct stat st;
1682 if (stat(sha1_pack_index_name(sha1), &st))
1683 return 0;
1684 return 1;
1685 }
1686
1687 int has_sha1_pack(const unsigned char *sha1)
1688 {
1689 struct pack_entry e;
1690 return find_pack_entry(sha1, &e);
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(unsigned char *sha1, 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 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
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), sha1);
1771 else
1772 ret = hash_sha1_file(buf, size, typename(type), sha1);
1773 if (re_allocated)
1774 free(buf);
1775 return ret;
1776 }
1777
1778 static int index_stream_convert_blob(unsigned char *sha1, 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 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1790
1791 if (write_object)
1792 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1793 sha1);
1794 else
1795 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1796 sha1);
1797 strbuf_release(&sbuf);
1798 return ret;
1799 }
1800
1801 static int index_pipe(unsigned char *sha1, 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(sha1, 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(unsigned char *sha1, 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(sha1, "", size, type, path, flags);
1825 } else if (size <= SMALL_FILE_SIZE) {
1826 char *buf = xmalloc(size);
1827 if (size == read_in_full(fd, buf, size))
1828 ret = index_mem(sha1, buf, size, type, path, flags);
1829 else
1830 ret = error_errno("short read");
1831 free(buf);
1832 } else {
1833 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1834 ret = index_mem(sha1, buf, size, type, path, flags);
1835 munmap(buf, size);
1836 }
1837 return ret;
1838 }
1839
1840 /*
1841 * This creates one packfile per large blob unless bulk-checkin
1842 * machinery is "plugged".
1843 *
1844 * This also bypasses the usual "convert-to-git" dance, and that is on
1845 * purpose. We could write a streaming version of the converting
1846 * functions and insert that before feeding the data to fast-import
1847 * (or equivalent in-core API described above). However, that is
1848 * somewhat complicated, as we do not know the size of the filter
1849 * result, which we need to know beforehand when writing a git object.
1850 * Since the primary motivation for trying to stream from the working
1851 * tree file and to avoid mmaping it in core is to deal with large
1852 * binary blobs, they generally do not want to get any conversion, and
1853 * callers should avoid this code path when filters are requested.
1854 */
1855 static int index_stream(unsigned char *sha1, int fd, size_t size,
1856 enum object_type type, const char *path,
1857 unsigned flags)
1858 {
1859 return index_bulk_checkin(sha1, fd, size, type, path, flags);
1860 }
1861
1862 int index_fd(unsigned char *sha1, int fd, struct stat *st,
1863 enum object_type type, const char *path, unsigned flags)
1864 {
1865 int ret;
1866
1867 /*
1868 * Call xsize_t() only when needed to avoid potentially unnecessary
1869 * die() for large files.
1870 */
1871 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1872 ret = index_stream_convert_blob(sha1, fd, path, flags);
1873 else if (!S_ISREG(st->st_mode))
1874 ret = index_pipe(sha1, fd, type, path, flags);
1875 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1876 (path && would_convert_to_git(&the_index, path)))
1877 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
1878 flags);
1879 else
1880 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
1881 flags);
1882 close(fd);
1883 return ret;
1884 }
1885
1886 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
1887 {
1888 int fd;
1889 struct strbuf sb = STRBUF_INIT;
1890
1891 switch (st->st_mode & S_IFMT) {
1892 case S_IFREG:
1893 fd = open(path, O_RDONLY);
1894 if (fd < 0)
1895 return error_errno("open(\"%s\")", path);
1896 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
1897 return error("%s: failed to insert into database",
1898 path);
1899 break;
1900 case S_IFLNK:
1901 if (strbuf_readlink(&sb, path, st->st_size))
1902 return error_errno("readlink(\"%s\")", path);
1903 if (!(flags & HASH_WRITE_OBJECT))
1904 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
1905 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
1906 return error("%s: failed to insert into database",
1907 path);
1908 strbuf_release(&sb);
1909 break;
1910 case S_IFDIR:
1911 return resolve_gitlink_ref(path, "HEAD", sha1);
1912 default:
1913 return error("%s: unsupported file type", path);
1914 }
1915 return 0;
1916 }
1917
1918 int read_pack_header(int fd, struct pack_header *header)
1919 {
1920 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
1921 /* "eof before pack header was fully read" */
1922 return PH_ERROR_EOF;
1923
1924 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1925 /* "protocol error (pack signature mismatch detected)" */
1926 return PH_ERROR_PACK_SIGNATURE;
1927 if (!pack_version_ok(header->hdr_version))
1928 /* "protocol error (pack version unsupported)" */
1929 return PH_ERROR_PROTOCOL;
1930 return 0;
1931 }
1932
1933 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1934 {
1935 enum object_type type = sha1_object_info(sha1, NULL);
1936 if (type < 0)
1937 die("%s is not a valid object", sha1_to_hex(sha1));
1938 if (type != expect)
1939 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1940 typename(expect));
1941 }
1942
1943 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1944 struct strbuf *path,
1945 each_loose_object_fn obj_cb,
1946 each_loose_cruft_fn cruft_cb,
1947 each_loose_subdir_fn subdir_cb,
1948 void *data)
1949 {
1950 size_t origlen, baselen;
1951 DIR *dir;
1952 struct dirent *de;
1953 int r = 0;
1954
1955 if (subdir_nr > 0xff)
1956 BUG("invalid loose object subdirectory: %x", subdir_nr);
1957
1958 origlen = path->len;
1959 strbuf_complete(path, '/');
1960 strbuf_addf(path, "%02x", subdir_nr);
1961 baselen = path->len;
1962
1963 dir = opendir(path->buf);
1964 if (!dir) {
1965 if (errno != ENOENT)
1966 r = error_errno("unable to open %s", path->buf);
1967 strbuf_setlen(path, origlen);
1968 return r;
1969 }
1970
1971 while ((de = readdir(dir))) {
1972 if (is_dot_or_dotdot(de->d_name))
1973 continue;
1974
1975 strbuf_setlen(path, baselen);
1976 strbuf_addf(path, "/%s", de->d_name);
1977
1978 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
1979 char hex[GIT_MAX_HEXSZ+1];
1980 struct object_id oid;
1981
1982 xsnprintf(hex, sizeof(hex), "%02x%s",
1983 subdir_nr, de->d_name);
1984 if (!get_oid_hex(hex, &oid)) {
1985 if (obj_cb) {
1986 r = obj_cb(&oid, path->buf, data);
1987 if (r)
1988 break;
1989 }
1990 continue;
1991 }
1992 }
1993
1994 if (cruft_cb) {
1995 r = cruft_cb(de->d_name, path->buf, data);
1996 if (r)
1997 break;
1998 }
1999 }
2000 closedir(dir);
2001
2002 strbuf_setlen(path, baselen);
2003 if (!r && subdir_cb)
2004 r = subdir_cb(subdir_nr, path->buf, data);
2005
2006 strbuf_setlen(path, origlen);
2007
2008 return r;
2009 }
2010
2011 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2012 each_loose_object_fn obj_cb,
2013 each_loose_cruft_fn cruft_cb,
2014 each_loose_subdir_fn subdir_cb,
2015 void *data)
2016 {
2017 int r = 0;
2018 int i;
2019
2020 for (i = 0; i < 256; i++) {
2021 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2022 subdir_cb, data);
2023 if (r)
2024 break;
2025 }
2026
2027 return r;
2028 }
2029
2030 int for_each_loose_file_in_objdir(const char *path,
2031 each_loose_object_fn obj_cb,
2032 each_loose_cruft_fn cruft_cb,
2033 each_loose_subdir_fn subdir_cb,
2034 void *data)
2035 {
2036 struct strbuf buf = STRBUF_INIT;
2037 int r;
2038
2039 strbuf_addstr(&buf, path);
2040 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2041 subdir_cb, data);
2042 strbuf_release(&buf);
2043
2044 return r;
2045 }
2046
2047 struct loose_alt_odb_data {
2048 each_loose_object_fn *cb;
2049 void *data;
2050 };
2051
2052 static int loose_from_alt_odb(struct alternate_object_database *alt,
2053 void *vdata)
2054 {
2055 struct loose_alt_odb_data *data = vdata;
2056 struct strbuf buf = STRBUF_INIT;
2057 int r;
2058
2059 strbuf_addstr(&buf, alt->path);
2060 r = for_each_loose_file_in_objdir_buf(&buf,
2061 data->cb, NULL, NULL,
2062 data->data);
2063 strbuf_release(&buf);
2064 return r;
2065 }
2066
2067 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2068 {
2069 struct loose_alt_odb_data alt;
2070 int r;
2071
2072 r = for_each_loose_file_in_objdir(get_object_directory(),
2073 cb, NULL, NULL, data);
2074 if (r)
2075 return r;
2076
2077 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2078 return 0;
2079
2080 alt.cb = cb;
2081 alt.data = data;
2082 return foreach_alt_odb(loose_from_alt_odb, &alt);
2083 }
2084
2085 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
2086 {
2087 uint32_t i;
2088 int r = 0;
2089
2090 for (i = 0; i < p->num_objects; i++) {
2091 struct object_id oid;
2092
2093 if (!nth_packed_object_oid(&oid, p, i))
2094 return error("unable to get sha1 of object %u in %s",
2095 i, p->pack_name);
2096
2097 r = cb(&oid, p, i, data);
2098 if (r)
2099 break;
2100 }
2101 return r;
2102 }
2103
2104 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
2105 {
2106 struct packed_git *p;
2107 int r = 0;
2108 int pack_errors = 0;
2109
2110 prepare_packed_git();
2111 for (p = packed_git; p; p = p->next) {
2112 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2113 continue;
2114 if (open_pack_index(p)) {
2115 pack_errors = 1;
2116 continue;
2117 }
2118 r = for_each_object_in_pack(p, cb, data);
2119 if (r)
2120 break;
2121 }
2122 return r ? r : pack_errors;
2123 }
2124
2125 static int check_stream_sha1(git_zstream *stream,
2126 const char *hdr,
2127 unsigned long size,
2128 const char *path,
2129 const unsigned char *expected_sha1)
2130 {
2131 git_SHA_CTX c;
2132 unsigned char real_sha1[GIT_MAX_RAWSZ];
2133 unsigned char buf[4096];
2134 unsigned long total_read;
2135 int status = Z_OK;
2136
2137 git_SHA1_Init(&c);
2138 git_SHA1_Update(&c, hdr, stream->total_out);
2139
2140 /*
2141 * We already read some bytes into hdr, but the ones up to the NUL
2142 * do not count against the object's content size.
2143 */
2144 total_read = stream->total_out - strlen(hdr) - 1;
2145
2146 /*
2147 * This size comparison must be "<=" to read the final zlib packets;
2148 * see the comment in unpack_sha1_rest for details.
2149 */
2150 while (total_read <= size &&
2151 (status == Z_OK || status == Z_BUF_ERROR)) {
2152 stream->next_out = buf;
2153 stream->avail_out = sizeof(buf);
2154 if (size - total_read < stream->avail_out)
2155 stream->avail_out = size - total_read;
2156 status = git_inflate(stream, Z_FINISH);
2157 git_SHA1_Update(&c, buf, stream->next_out - buf);
2158 total_read += stream->next_out - buf;
2159 }
2160 git_inflate_end(stream);
2161
2162 if (status != Z_STREAM_END) {
2163 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2164 return -1;
2165 }
2166 if (stream->avail_in) {
2167 error("garbage at end of loose object '%s'",
2168 sha1_to_hex(expected_sha1));
2169 return -1;
2170 }
2171
2172 git_SHA1_Final(real_sha1, &c);
2173 if (hashcmp(expected_sha1, real_sha1)) {
2174 error("sha1 mismatch for %s (expected %s)", path,
2175 sha1_to_hex(expected_sha1));
2176 return -1;
2177 }
2178
2179 return 0;
2180 }
2181
2182 int read_loose_object(const char *path,
2183 const unsigned char *expected_sha1,
2184 enum object_type *type,
2185 unsigned long *size,
2186 void **contents)
2187 {
2188 int ret = -1;
2189 void *map = NULL;
2190 unsigned long mapsize;
2191 git_zstream stream;
2192 char hdr[32];
2193
2194 *contents = NULL;
2195
2196 map = map_sha1_file_1(path, NULL, &mapsize);
2197 if (!map) {
2198 error_errno("unable to mmap %s", path);
2199 goto out;
2200 }
2201
2202 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2203 error("unable to unpack header of %s", path);
2204 goto out;
2205 }
2206
2207 *type = parse_sha1_header(hdr, size);
2208 if (*type < 0) {
2209 error("unable to parse header of %s", path);
2210 git_inflate_end(&stream);
2211 goto out;
2212 }
2213
2214 if (*type == OBJ_BLOB) {
2215 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2216 goto out;
2217 } else {
2218 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2219 if (!*contents) {
2220 error("unable to unpack contents of %s", path);
2221 git_inflate_end(&stream);
2222 goto out;
2223 }
2224 if (check_sha1_signature(expected_sha1, *contents,
2225 *size, typename(*type))) {
2226 error("sha1 mismatch for %s (expected %s)", path,
2227 sha1_to_hex(expected_sha1));
2228 free(*contents);
2229 goto out;
2230 }
2231 }
2232
2233 ret = 0; /* everything checks out */
2234
2235 out:
2236 if (map)
2237 munmap(map, mapsize);
2238 return ret;
2239 }