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