43e1581530818a01f4516160e0c5b89d22c61be0
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
33const unsigned char null_sha1[20];
34const struct object_id null_oid;
35const struct object_id empty_tree_oid = {
36 EMPTY_TREE_SHA1_BIN_LITERAL
37};
38const 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 */
48static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53} *cached_objects;
54static int cached_object_nr, cached_object_alloc;
55
56static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
58 OBJ_TREE,
59 "",
60 0
61};
62
63static 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
77int 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
106enum 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
160enum 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
173int 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
200retry_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
235out:
236 strbuf_release(&path_copy);
237 errno = save_errno;
238 return ret;
239}
240
241static 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
254const 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
265struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
266{
267 strbuf_setlen(&alt->scratch, alt->base_len);
268 return &alt->scratch;
269}
270
271static 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
279struct alternate_object_database *alt_odb_list;
280static struct alternate_object_database **alt_odb_tail;
281
282/*
283 * Return non-zero iff the path is usable as an alternate object database.
284 */
285static 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 */
326static void read_info_alternates(const char * relative_base, int depth);
327static 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
372static 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
401static 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
428static 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
454struct 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
466void 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
511void 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 */
528char *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
589out:
590 if (seen_error) {
591 FREE_AND_NULL(ref_git);
592 }
593
594 return ref_git;
595}
596
597int 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
611void 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. */
628static 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 */
642int 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
651static 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
656static 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
668static 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
674int has_loose_object_nonlocal(const unsigned char *sha1)
675{
676 return check_and_freshen_nonlocal(sha1, 0);
677}
678
679static int has_loose_object(const unsigned char *sha1)
680{
681 return check_and_freshen(sha1, 0);
682}
683
684static 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
697void *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
713void *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 */
728int 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
770int 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 */
805static 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 */
829static 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 */
858static 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
887void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
888{
889 return map_sha1_file_1(NULL, sha1, size);
890}
891
892static 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
907int 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
923static 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
959static 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 */
1009static 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
1070int 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
1078const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1079 uint32_t n)
1080{
1081 const unsigned char *index = p->index_data;
1082 if (!index) {
1083 if (open_pack_index(p))
1084 return NULL;
1085 index = p->index_data;
1086 }
1087 if (n >= p->num_objects)
1088 return NULL;
1089 index += 4 * 256;
1090 if (p->index_version == 1) {
1091 return index + 24 * n + 4;
1092 } else {
1093 index += 8;
1094 return index + 20 * n;
1095 }
1096}
1097
1098const struct object_id *nth_packed_object_oid(struct object_id *oid,
1099 struct packed_git *p,
1100 uint32_t n)
1101{
1102 const unsigned char *hash = nth_packed_object_sha1(p, n);
1103 if (!hash)
1104 return NULL;
1105 hashcpy(oid->hash, hash);
1106 return oid;
1107}
1108
1109void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1110{
1111 const unsigned char *ptr = vptr;
1112 const unsigned char *start = p->index_data;
1113 const unsigned char *end = start + p->index_size;
1114 if (ptr < start)
1115 die(_("offset before start of pack index for %s (corrupt index?)"),
1116 p->pack_name);
1117 /* No need to check for underflow; .idx files must be at least 8 bytes */
1118 if (ptr >= end - 8)
1119 die(_("offset beyond end of pack index for %s (truncated index?)"),
1120 p->pack_name);
1121}
1122
1123off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1124{
1125 const unsigned char *index = p->index_data;
1126 index += 4 * 256;
1127 if (p->index_version == 1) {
1128 return ntohl(*((uint32_t *)(index + 24 * n)));
1129 } else {
1130 uint32_t off;
1131 index += 8 + p->num_objects * (20 + 4);
1132 off = ntohl(*((uint32_t *)(index + 4 * n)));
1133 if (!(off & 0x80000000))
1134 return off;
1135 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1136 check_pack_index_ptr(p, index);
1137 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1138 ntohl(*((uint32_t *)(index + 4)));
1139 }
1140}
1141
1142off_t find_pack_entry_one(const unsigned char *sha1,
1143 struct packed_git *p)
1144{
1145 const uint32_t *level1_ofs = p->index_data;
1146 const unsigned char *index = p->index_data;
1147 unsigned hi, lo, stride;
1148 static int debug_lookup = -1;
1149
1150 if (debug_lookup < 0)
1151 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1152
1153 if (!index) {
1154 if (open_pack_index(p))
1155 return 0;
1156 level1_ofs = p->index_data;
1157 index = p->index_data;
1158 }
1159 if (p->index_version > 1) {
1160 level1_ofs += 2;
1161 index += 8;
1162 }
1163 index += 4 * 256;
1164 hi = ntohl(level1_ofs[*sha1]);
1165 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1166 if (p->index_version > 1) {
1167 stride = 20;
1168 } else {
1169 stride = 24;
1170 index += 4;
1171 }
1172
1173 if (debug_lookup)
1174 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1175 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1176
1177 while (lo < hi) {
1178 unsigned mi = (lo + hi) / 2;
1179 int cmp = hashcmp(index + mi * stride, sha1);
1180
1181 if (debug_lookup)
1182 printf("lo %u hi %u rg %u mi %u\n",
1183 lo, hi, hi - lo, mi);
1184 if (!cmp)
1185 return nth_packed_object_offset(p, mi);
1186 if (cmp > 0)
1187 hi = mi;
1188 else
1189 lo = mi+1;
1190 }
1191 return 0;
1192}
1193
1194int is_pack_valid(struct packed_git *p)
1195{
1196 /* An already open pack is known to be valid. */
1197 if (p->pack_fd != -1)
1198 return 1;
1199
1200 /* If the pack has one window completely covering the
1201 * file size, the pack is known to be valid even if
1202 * the descriptor is not currently open.
1203 */
1204 if (p->windows) {
1205 struct pack_window *w = p->windows;
1206
1207 if (!w->offset && w->len == p->pack_size)
1208 return 1;
1209 }
1210
1211 /* Force the pack to open to prove its valid. */
1212 return !open_packed_git(p);
1213}
1214
1215static int fill_pack_entry(const unsigned char *sha1,
1216 struct pack_entry *e,
1217 struct packed_git *p)
1218{
1219 off_t offset;
1220
1221 if (p->num_bad_objects) {
1222 unsigned i;
1223 for (i = 0; i < p->num_bad_objects; i++)
1224 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1225 return 0;
1226 }
1227
1228 offset = find_pack_entry_one(sha1, p);
1229 if (!offset)
1230 return 0;
1231
1232 /*
1233 * We are about to tell the caller where they can locate the
1234 * requested object. We better make sure the packfile is
1235 * still here and can be accessed before supplying that
1236 * answer, as it may have been deleted since the index was
1237 * loaded!
1238 */
1239 if (!is_pack_valid(p))
1240 return 0;
1241 e->offset = offset;
1242 e->p = p;
1243 hashcpy(e->sha1, sha1);
1244 return 1;
1245}
1246
1247/*
1248 * Iff a pack file contains the object named by sha1, return true and
1249 * store its location to e.
1250 */
1251static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1252{
1253 struct mru_entry *p;
1254
1255 prepare_packed_git();
1256 if (!packed_git)
1257 return 0;
1258
1259 for (p = packed_git_mru->head; p; p = p->next) {
1260 if (fill_pack_entry(sha1, e, p->item)) {
1261 mru_mark(packed_git_mru, p);
1262 return 1;
1263 }
1264 }
1265 return 0;
1266}
1267
1268struct packed_git *find_sha1_pack(const unsigned char *sha1,
1269 struct packed_git *packs)
1270{
1271 struct packed_git *p;
1272
1273 for (p = packs; p; p = p->next) {
1274 if (find_pack_entry_one(sha1, p))
1275 return p;
1276 }
1277 return NULL;
1278
1279}
1280
1281static int sha1_loose_object_info(const unsigned char *sha1,
1282 struct object_info *oi,
1283 int flags)
1284{
1285 int status = 0;
1286 unsigned long mapsize;
1287 void *map;
1288 git_zstream stream;
1289 char hdr[32];
1290 struct strbuf hdrbuf = STRBUF_INIT;
1291 unsigned long size_scratch;
1292
1293 if (oi->delta_base_sha1)
1294 hashclr(oi->delta_base_sha1);
1295
1296 /*
1297 * If we don't care about type or size, then we don't
1298 * need to look inside the object at all. Note that we
1299 * do not optimize out the stat call, even if the
1300 * caller doesn't care about the disk-size, since our
1301 * return value implicitly indicates whether the
1302 * object even exists.
1303 */
1304 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1305 const char *path;
1306 struct stat st;
1307 if (stat_sha1_file(sha1, &st, &path) < 0)
1308 return -1;
1309 if (oi->disk_sizep)
1310 *oi->disk_sizep = st.st_size;
1311 return 0;
1312 }
1313
1314 map = map_sha1_file(sha1, &mapsize);
1315 if (!map)
1316 return -1;
1317
1318 if (!oi->sizep)
1319 oi->sizep = &size_scratch;
1320
1321 if (oi->disk_sizep)
1322 *oi->disk_sizep = mapsize;
1323 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1324 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1325 status = error("unable to unpack %s header with --allow-unknown-type",
1326 sha1_to_hex(sha1));
1327 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1328 status = error("unable to unpack %s header",
1329 sha1_to_hex(sha1));
1330 if (status < 0)
1331 ; /* Do nothing */
1332 else if (hdrbuf.len) {
1333 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1334 status = error("unable to parse %s header with --allow-unknown-type",
1335 sha1_to_hex(sha1));
1336 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1337 status = error("unable to parse %s header", sha1_to_hex(sha1));
1338
1339 if (status >= 0 && oi->contentp)
1340 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1341 *oi->sizep, sha1);
1342 else
1343 git_inflate_end(&stream);
1344
1345 munmap(map, mapsize);
1346 if (status && oi->typep)
1347 *oi->typep = status;
1348 if (oi->sizep == &size_scratch)
1349 oi->sizep = NULL;
1350 strbuf_release(&hdrbuf);
1351 oi->whence = OI_LOOSE;
1352 return (status < 0) ? status : 0;
1353}
1354
1355int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1356{
1357 static struct object_info blank_oi = OBJECT_INFO_INIT;
1358 struct pack_entry e;
1359 int rtype;
1360 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1361 lookup_replace_object(sha1) :
1362 sha1;
1363
1364 if (!oi)
1365 oi = &blank_oi;
1366
1367 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1368 struct cached_object *co = find_cached_object(real);
1369 if (co) {
1370 if (oi->typep)
1371 *(oi->typep) = co->type;
1372 if (oi->sizep)
1373 *(oi->sizep) = co->size;
1374 if (oi->disk_sizep)
1375 *(oi->disk_sizep) = 0;
1376 if (oi->delta_base_sha1)
1377 hashclr(oi->delta_base_sha1);
1378 if (oi->typename)
1379 strbuf_addstr(oi->typename, typename(co->type));
1380 if (oi->contentp)
1381 *oi->contentp = xmemdupz(co->buf, co->size);
1382 oi->whence = OI_CACHED;
1383 return 0;
1384 }
1385 }
1386
1387 if (!find_pack_entry(real, &e)) {
1388 /* Most likely it's a loose object. */
1389 if (!sha1_loose_object_info(real, oi, flags))
1390 return 0;
1391
1392 /* Not a loose object; someone else may have just packed it. */
1393 if (flags & OBJECT_INFO_QUICK) {
1394 return -1;
1395 } else {
1396 reprepare_packed_git();
1397 if (!find_pack_entry(real, &e))
1398 return -1;
1399 }
1400 }
1401
1402 if (oi == &blank_oi)
1403 /*
1404 * We know that the caller doesn't actually need the
1405 * information below, so return early.
1406 */
1407 return 0;
1408
1409 rtype = packed_object_info(e.p, e.offset, oi);
1410 if (rtype < 0) {
1411 mark_bad_packed_object(e.p, real);
1412 return sha1_object_info_extended(real, oi, 0);
1413 } else if (oi->whence == OI_PACKED) {
1414 oi->u.packed.offset = e.offset;
1415 oi->u.packed.pack = e.p;
1416 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1417 rtype == OBJ_OFS_DELTA);
1418 }
1419
1420 return 0;
1421}
1422
1423/* returns enum object_type or negative */
1424int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1425{
1426 enum object_type type;
1427 struct object_info oi = OBJECT_INFO_INIT;
1428
1429 oi.typep = &type;
1430 oi.sizep = sizep;
1431 if (sha1_object_info_extended(sha1, &oi,
1432 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1433 return -1;
1434 return type;
1435}
1436
1437static void *read_object(const unsigned char *sha1, enum object_type *type,
1438 unsigned long *size)
1439{
1440 struct object_info oi = OBJECT_INFO_INIT;
1441 void *content;
1442 oi.typep = type;
1443 oi.sizep = size;
1444 oi.contentp = &content;
1445
1446 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1447 return NULL;
1448 return content;
1449}
1450
1451int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1452 unsigned char *sha1)
1453{
1454 struct cached_object *co;
1455
1456 hash_sha1_file(buf, len, typename(type), sha1);
1457 if (has_sha1_file(sha1) || find_cached_object(sha1))
1458 return 0;
1459 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1460 co = &cached_objects[cached_object_nr++];
1461 co->size = len;
1462 co->type = type;
1463 co->buf = xmalloc(len);
1464 memcpy(co->buf, buf, len);
1465 hashcpy(co->sha1, sha1);
1466 return 0;
1467}
1468
1469/*
1470 * This function dies on corrupt objects; the callers who want to
1471 * deal with them should arrange to call read_object() and give error
1472 * messages themselves.
1473 */
1474void *read_sha1_file_extended(const unsigned char *sha1,
1475 enum object_type *type,
1476 unsigned long *size,
1477 int lookup_replace)
1478{
1479 void *data;
1480 const struct packed_git *p;
1481 const char *path;
1482 struct stat st;
1483 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1484 : sha1;
1485
1486 errno = 0;
1487 data = read_object(repl, type, size);
1488 if (data)
1489 return data;
1490
1491 if (errno && errno != ENOENT)
1492 die_errno("failed to read object %s", sha1_to_hex(sha1));
1493
1494 /* die if we replaced an object with one that does not exist */
1495 if (repl != sha1)
1496 die("replacement %s not found for %s",
1497 sha1_to_hex(repl), sha1_to_hex(sha1));
1498
1499 if (!stat_sha1_file(repl, &st, &path))
1500 die("loose object %s (stored in %s) is corrupt",
1501 sha1_to_hex(repl), path);
1502
1503 if ((p = has_packed_and_bad(repl)) != NULL)
1504 die("packed object %s (stored in %s) is corrupt",
1505 sha1_to_hex(repl), p->pack_name);
1506
1507 return NULL;
1508}
1509
1510void *read_object_with_reference(const unsigned char *sha1,
1511 const char *required_type_name,
1512 unsigned long *size,
1513 unsigned char *actual_sha1_return)
1514{
1515 enum object_type type, required_type;
1516 void *buffer;
1517 unsigned long isize;
1518 unsigned char actual_sha1[20];
1519
1520 required_type = type_from_string(required_type_name);
1521 hashcpy(actual_sha1, sha1);
1522 while (1) {
1523 int ref_length = -1;
1524 const char *ref_type = NULL;
1525
1526 buffer = read_sha1_file(actual_sha1, &type, &isize);
1527 if (!buffer)
1528 return NULL;
1529 if (type == required_type) {
1530 *size = isize;
1531 if (actual_sha1_return)
1532 hashcpy(actual_sha1_return, actual_sha1);
1533 return buffer;
1534 }
1535 /* Handle references */
1536 else if (type == OBJ_COMMIT)
1537 ref_type = "tree ";
1538 else if (type == OBJ_TAG)
1539 ref_type = "object ";
1540 else {
1541 free(buffer);
1542 return NULL;
1543 }
1544 ref_length = strlen(ref_type);
1545
1546 if (ref_length + 40 > isize ||
1547 memcmp(buffer, ref_type, ref_length) ||
1548 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1549 free(buffer);
1550 return NULL;
1551 }
1552 free(buffer);
1553 /* Now we have the ID of the referred-to object in
1554 * actual_sha1. Check again. */
1555 }
1556}
1557
1558static void write_sha1_file_prepare(const void *buf, unsigned long len,
1559 const char *type, unsigned char *sha1,
1560 char *hdr, int *hdrlen)
1561{
1562 git_SHA_CTX c;
1563
1564 /* Generate the header */
1565 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1566
1567 /* Sha1.. */
1568 git_SHA1_Init(&c);
1569 git_SHA1_Update(&c, hdr, *hdrlen);
1570 git_SHA1_Update(&c, buf, len);
1571 git_SHA1_Final(sha1, &c);
1572}
1573
1574/*
1575 * Move the just written object into its final resting place.
1576 */
1577int finalize_object_file(const char *tmpfile, const char *filename)
1578{
1579 int ret = 0;
1580
1581 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1582 goto try_rename;
1583 else if (link(tmpfile, filename))
1584 ret = errno;
1585
1586 /*
1587 * Coda hack - coda doesn't like cross-directory links,
1588 * so we fall back to a rename, which will mean that it
1589 * won't be able to check collisions, but that's not a
1590 * big deal.
1591 *
1592 * The same holds for FAT formatted media.
1593 *
1594 * When this succeeds, we just return. We have nothing
1595 * left to unlink.
1596 */
1597 if (ret && ret != EEXIST) {
1598 try_rename:
1599 if (!rename(tmpfile, filename))
1600 goto out;
1601 ret = errno;
1602 }
1603 unlink_or_warn(tmpfile);
1604 if (ret) {
1605 if (ret != EEXIST) {
1606 return error_errno("unable to write sha1 filename %s", filename);
1607 }
1608 /* FIXME!!! Collision check here ? */
1609 }
1610
1611out:
1612 if (adjust_shared_perm(filename))
1613 return error("unable to set permission to '%s'", filename);
1614 return 0;
1615}
1616
1617static int write_buffer(int fd, const void *buf, size_t len)
1618{
1619 if (write_in_full(fd, buf, len) < 0)
1620 return error_errno("file write error");
1621 return 0;
1622}
1623
1624int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1625 unsigned char *sha1)
1626{
1627 char hdr[32];
1628 int hdrlen = sizeof(hdr);
1629 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1630 return 0;
1631}
1632
1633/* Finalize a file on disk, and close it. */
1634static void close_sha1_file(int fd)
1635{
1636 if (fsync_object_files)
1637 fsync_or_die(fd, "sha1 file");
1638 if (close(fd) != 0)
1639 die_errno("error when closing sha1 file");
1640}
1641
1642/* Size of directory component, including the ending '/' */
1643static inline int directory_size(const char *filename)
1644{
1645 const char *s = strrchr(filename, '/');
1646 if (!s)
1647 return 0;
1648 return s - filename + 1;
1649}
1650
1651/*
1652 * This creates a temporary file in the same directory as the final
1653 * 'filename'
1654 *
1655 * We want to avoid cross-directory filename renames, because those
1656 * can have problems on various filesystems (FAT, NFS, Coda).
1657 */
1658static int create_tmpfile(struct strbuf *tmp, const char *filename)
1659{
1660 int fd, dirlen = directory_size(filename);
1661
1662 strbuf_reset(tmp);
1663 strbuf_add(tmp, filename, dirlen);
1664 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1665 fd = git_mkstemp_mode(tmp->buf, 0444);
1666 if (fd < 0 && dirlen && errno == ENOENT) {
1667 /*
1668 * Make sure the directory exists; note that the contents
1669 * of the buffer are undefined after mkstemp returns an
1670 * error, so we have to rewrite the whole buffer from
1671 * scratch.
1672 */
1673 strbuf_reset(tmp);
1674 strbuf_add(tmp, filename, dirlen - 1);
1675 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1676 return -1;
1677 if (adjust_shared_perm(tmp->buf))
1678 return -1;
1679
1680 /* Try again */
1681 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1682 fd = git_mkstemp_mode(tmp->buf, 0444);
1683 }
1684 return fd;
1685}
1686
1687static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1688 const void *buf, unsigned long len, time_t mtime)
1689{
1690 int fd, ret;
1691 unsigned char compressed[4096];
1692 git_zstream stream;
1693 git_SHA_CTX c;
1694 unsigned char parano_sha1[20];
1695 static struct strbuf tmp_file = STRBUF_INIT;
1696 const char *filename = sha1_file_name(sha1);
1697
1698 fd = create_tmpfile(&tmp_file, filename);
1699 if (fd < 0) {
1700 if (errno == EACCES)
1701 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1702 else
1703 return error_errno("unable to create temporary file");
1704 }
1705
1706 /* Set it up */
1707 git_deflate_init(&stream, zlib_compression_level);
1708 stream.next_out = compressed;
1709 stream.avail_out = sizeof(compressed);
1710 git_SHA1_Init(&c);
1711
1712 /* First header.. */
1713 stream.next_in = (unsigned char *)hdr;
1714 stream.avail_in = hdrlen;
1715 while (git_deflate(&stream, 0) == Z_OK)
1716 ; /* nothing */
1717 git_SHA1_Update(&c, hdr, hdrlen);
1718
1719 /* Then the data itself.. */
1720 stream.next_in = (void *)buf;
1721 stream.avail_in = len;
1722 do {
1723 unsigned char *in0 = stream.next_in;
1724 ret = git_deflate(&stream, Z_FINISH);
1725 git_SHA1_Update(&c, in0, stream.next_in - in0);
1726 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1727 die("unable to write sha1 file");
1728 stream.next_out = compressed;
1729 stream.avail_out = sizeof(compressed);
1730 } while (ret == Z_OK);
1731
1732 if (ret != Z_STREAM_END)
1733 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1734 ret = git_deflate_end_gently(&stream);
1735 if (ret != Z_OK)
1736 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1737 git_SHA1_Final(parano_sha1, &c);
1738 if (hashcmp(sha1, parano_sha1) != 0)
1739 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1740
1741 close_sha1_file(fd);
1742
1743 if (mtime) {
1744 struct utimbuf utb;
1745 utb.actime = mtime;
1746 utb.modtime = mtime;
1747 if (utime(tmp_file.buf, &utb) < 0)
1748 warning_errno("failed utime() on %s", tmp_file.buf);
1749 }
1750
1751 return finalize_object_file(tmp_file.buf, filename);
1752}
1753
1754static int freshen_loose_object(const unsigned char *sha1)
1755{
1756 return check_and_freshen(sha1, 1);
1757}
1758
1759static int freshen_packed_object(const unsigned char *sha1)
1760{
1761 struct pack_entry e;
1762 if (!find_pack_entry(sha1, &e))
1763 return 0;
1764 if (e.p->freshened)
1765 return 1;
1766 if (!freshen_file(e.p->pack_name))
1767 return 0;
1768 e.p->freshened = 1;
1769 return 1;
1770}
1771
1772int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1773{
1774 char hdr[32];
1775 int hdrlen = sizeof(hdr);
1776
1777 /* Normally if we have it in the pack then we do not bother writing
1778 * it out into .git/objects/??/?{38} file.
1779 */
1780 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1781 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1782 return 0;
1783 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1784}
1785
1786int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1787 unsigned char *sha1, unsigned flags)
1788{
1789 char *header;
1790 int hdrlen, status = 0;
1791
1792 /* type string, SP, %lu of the length plus NUL must fit this */
1793 hdrlen = strlen(type) + 32;
1794 header = xmalloc(hdrlen);
1795 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
1796
1797 if (!(flags & HASH_WRITE_OBJECT))
1798 goto cleanup;
1799 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1800 goto cleanup;
1801 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
1802
1803cleanup:
1804 free(header);
1805 return status;
1806}
1807
1808int force_object_loose(const unsigned char *sha1, time_t mtime)
1809{
1810 void *buf;
1811 unsigned long len;
1812 enum object_type type;
1813 char hdr[32];
1814 int hdrlen;
1815 int ret;
1816
1817 if (has_loose_object(sha1))
1818 return 0;
1819 buf = read_object(sha1, &type, &len);
1820 if (!buf)
1821 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1822 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1823 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1824 free(buf);
1825
1826 return ret;
1827}
1828
1829int has_pack_index(const unsigned char *sha1)
1830{
1831 struct stat st;
1832 if (stat(sha1_pack_index_name(sha1), &st))
1833 return 0;
1834 return 1;
1835}
1836
1837int has_sha1_pack(const unsigned char *sha1)
1838{
1839 struct pack_entry e;
1840 return find_pack_entry(sha1, &e);
1841}
1842
1843int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1844{
1845 if (!startup_info->have_repository)
1846 return 0;
1847 return sha1_object_info_extended(sha1, NULL,
1848 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1849}
1850
1851int has_object_file(const struct object_id *oid)
1852{
1853 return has_sha1_file(oid->hash);
1854}
1855
1856int has_object_file_with_flags(const struct object_id *oid, int flags)
1857{
1858 return has_sha1_file_with_flags(oid->hash, flags);
1859}
1860
1861static void check_tree(const void *buf, size_t size)
1862{
1863 struct tree_desc desc;
1864 struct name_entry entry;
1865
1866 init_tree_desc(&desc, buf, size);
1867 while (tree_entry(&desc, &entry))
1868 /* do nothing
1869 * tree_entry() will die() on malformed entries */
1870 ;
1871}
1872
1873static void check_commit(const void *buf, size_t size)
1874{
1875 struct commit c;
1876 memset(&c, 0, sizeof(c));
1877 if (parse_commit_buffer(&c, buf, size))
1878 die("corrupt commit");
1879}
1880
1881static void check_tag(const void *buf, size_t size)
1882{
1883 struct tag t;
1884 memset(&t, 0, sizeof(t));
1885 if (parse_tag_buffer(&t, buf, size))
1886 die("corrupt tag");
1887}
1888
1889static int index_mem(unsigned char *sha1, void *buf, size_t size,
1890 enum object_type type,
1891 const char *path, unsigned flags)
1892{
1893 int ret, re_allocated = 0;
1894 int write_object = flags & HASH_WRITE_OBJECT;
1895
1896 if (!type)
1897 type = OBJ_BLOB;
1898
1899 /*
1900 * Convert blobs to git internal format
1901 */
1902 if ((type == OBJ_BLOB) && path) {
1903 struct strbuf nbuf = STRBUF_INIT;
1904 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1905 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1906 buf = strbuf_detach(&nbuf, &size);
1907 re_allocated = 1;
1908 }
1909 }
1910 if (flags & HASH_FORMAT_CHECK) {
1911 if (type == OBJ_TREE)
1912 check_tree(buf, size);
1913 if (type == OBJ_COMMIT)
1914 check_commit(buf, size);
1915 if (type == OBJ_TAG)
1916 check_tag(buf, size);
1917 }
1918
1919 if (write_object)
1920 ret = write_sha1_file(buf, size, typename(type), sha1);
1921 else
1922 ret = hash_sha1_file(buf, size, typename(type), sha1);
1923 if (re_allocated)
1924 free(buf);
1925 return ret;
1926}
1927
1928static int index_stream_convert_blob(unsigned char *sha1, int fd,
1929 const char *path, unsigned flags)
1930{
1931 int ret;
1932 const int write_object = flags & HASH_WRITE_OBJECT;
1933 struct strbuf sbuf = STRBUF_INIT;
1934
1935 assert(path);
1936 assert(would_convert_to_git_filter_fd(path));
1937
1938 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1939 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1940
1941 if (write_object)
1942 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1943 sha1);
1944 else
1945 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1946 sha1);
1947 strbuf_release(&sbuf);
1948 return ret;
1949}
1950
1951static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
1952 const char *path, unsigned flags)
1953{
1954 struct strbuf sbuf = STRBUF_INIT;
1955 int ret;
1956
1957 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1958 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
1959 else
1960 ret = -1;
1961 strbuf_release(&sbuf);
1962 return ret;
1963}
1964
1965#define SMALL_FILE_SIZE (32*1024)
1966
1967static int index_core(unsigned char *sha1, int fd, size_t size,
1968 enum object_type type, const char *path,
1969 unsigned flags)
1970{
1971 int ret;
1972
1973 if (!size) {
1974 ret = index_mem(sha1, "", size, type, path, flags);
1975 } else if (size <= SMALL_FILE_SIZE) {
1976 char *buf = xmalloc(size);
1977 if (size == read_in_full(fd, buf, size))
1978 ret = index_mem(sha1, buf, size, type, path, flags);
1979 else
1980 ret = error_errno("short read");
1981 free(buf);
1982 } else {
1983 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1984 ret = index_mem(sha1, buf, size, type, path, flags);
1985 munmap(buf, size);
1986 }
1987 return ret;
1988}
1989
1990/*
1991 * This creates one packfile per large blob unless bulk-checkin
1992 * machinery is "plugged".
1993 *
1994 * This also bypasses the usual "convert-to-git" dance, and that is on
1995 * purpose. We could write a streaming version of the converting
1996 * functions and insert that before feeding the data to fast-import
1997 * (or equivalent in-core API described above). However, that is
1998 * somewhat complicated, as we do not know the size of the filter
1999 * result, which we need to know beforehand when writing a git object.
2000 * Since the primary motivation for trying to stream from the working
2001 * tree file and to avoid mmaping it in core is to deal with large
2002 * binary blobs, they generally do not want to get any conversion, and
2003 * callers should avoid this code path when filters are requested.
2004 */
2005static int index_stream(unsigned char *sha1, int fd, size_t size,
2006 enum object_type type, const char *path,
2007 unsigned flags)
2008{
2009 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2010}
2011
2012int index_fd(unsigned char *sha1, int fd, struct stat *st,
2013 enum object_type type, const char *path, unsigned flags)
2014{
2015 int ret;
2016
2017 /*
2018 * Call xsize_t() only when needed to avoid potentially unnecessary
2019 * die() for large files.
2020 */
2021 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
2022 ret = index_stream_convert_blob(sha1, fd, path, flags);
2023 else if (!S_ISREG(st->st_mode))
2024 ret = index_pipe(sha1, fd, type, path, flags);
2025 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2026 (path && would_convert_to_git(&the_index, path)))
2027 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
2028 flags);
2029 else
2030 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
2031 flags);
2032 close(fd);
2033 return ret;
2034}
2035
2036int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2037{
2038 int fd;
2039 struct strbuf sb = STRBUF_INIT;
2040
2041 switch (st->st_mode & S_IFMT) {
2042 case S_IFREG:
2043 fd = open(path, O_RDONLY);
2044 if (fd < 0)
2045 return error_errno("open(\"%s\")", path);
2046 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2047 return error("%s: failed to insert into database",
2048 path);
2049 break;
2050 case S_IFLNK:
2051 if (strbuf_readlink(&sb, path, st->st_size))
2052 return error_errno("readlink(\"%s\")", path);
2053 if (!(flags & HASH_WRITE_OBJECT))
2054 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2055 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2056 return error("%s: failed to insert into database",
2057 path);
2058 strbuf_release(&sb);
2059 break;
2060 case S_IFDIR:
2061 return resolve_gitlink_ref(path, "HEAD", sha1);
2062 default:
2063 return error("%s: unsupported file type", path);
2064 }
2065 return 0;
2066}
2067
2068int read_pack_header(int fd, struct pack_header *header)
2069{
2070 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2071 /* "eof before pack header was fully read" */
2072 return PH_ERROR_EOF;
2073
2074 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2075 /* "protocol error (pack signature mismatch detected)" */
2076 return PH_ERROR_PACK_SIGNATURE;
2077 if (!pack_version_ok(header->hdr_version))
2078 /* "protocol error (pack version unsupported)" */
2079 return PH_ERROR_PROTOCOL;
2080 return 0;
2081}
2082
2083void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2084{
2085 enum object_type type = sha1_object_info(sha1, NULL);
2086 if (type < 0)
2087 die("%s is not a valid object", sha1_to_hex(sha1));
2088 if (type != expect)
2089 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2090 typename(expect));
2091}
2092
2093int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2094 struct strbuf *path,
2095 each_loose_object_fn obj_cb,
2096 each_loose_cruft_fn cruft_cb,
2097 each_loose_subdir_fn subdir_cb,
2098 void *data)
2099{
2100 size_t origlen, baselen;
2101 DIR *dir;
2102 struct dirent *de;
2103 int r = 0;
2104
2105 if (subdir_nr > 0xff)
2106 BUG("invalid loose object subdirectory: %x", subdir_nr);
2107
2108 origlen = path->len;
2109 strbuf_complete(path, '/');
2110 strbuf_addf(path, "%02x", subdir_nr);
2111 baselen = path->len;
2112
2113 dir = opendir(path->buf);
2114 if (!dir) {
2115 if (errno != ENOENT)
2116 r = error_errno("unable to open %s", path->buf);
2117 strbuf_setlen(path, origlen);
2118 return r;
2119 }
2120
2121 while ((de = readdir(dir))) {
2122 if (is_dot_or_dotdot(de->d_name))
2123 continue;
2124
2125 strbuf_setlen(path, baselen);
2126 strbuf_addf(path, "/%s", de->d_name);
2127
2128 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
2129 char hex[GIT_MAX_HEXSZ+1];
2130 struct object_id oid;
2131
2132 xsnprintf(hex, sizeof(hex), "%02x%s",
2133 subdir_nr, de->d_name);
2134 if (!get_oid_hex(hex, &oid)) {
2135 if (obj_cb) {
2136 r = obj_cb(&oid, path->buf, data);
2137 if (r)
2138 break;
2139 }
2140 continue;
2141 }
2142 }
2143
2144 if (cruft_cb) {
2145 r = cruft_cb(de->d_name, path->buf, data);
2146 if (r)
2147 break;
2148 }
2149 }
2150 closedir(dir);
2151
2152 strbuf_setlen(path, baselen);
2153 if (!r && subdir_cb)
2154 r = subdir_cb(subdir_nr, path->buf, data);
2155
2156 strbuf_setlen(path, origlen);
2157
2158 return r;
2159}
2160
2161int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2162 each_loose_object_fn obj_cb,
2163 each_loose_cruft_fn cruft_cb,
2164 each_loose_subdir_fn subdir_cb,
2165 void *data)
2166{
2167 int r = 0;
2168 int i;
2169
2170 for (i = 0; i < 256; i++) {
2171 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2172 subdir_cb, data);
2173 if (r)
2174 break;
2175 }
2176
2177 return r;
2178}
2179
2180int for_each_loose_file_in_objdir(const char *path,
2181 each_loose_object_fn obj_cb,
2182 each_loose_cruft_fn cruft_cb,
2183 each_loose_subdir_fn subdir_cb,
2184 void *data)
2185{
2186 struct strbuf buf = STRBUF_INIT;
2187 int r;
2188
2189 strbuf_addstr(&buf, path);
2190 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2191 subdir_cb, data);
2192 strbuf_release(&buf);
2193
2194 return r;
2195}
2196
2197struct loose_alt_odb_data {
2198 each_loose_object_fn *cb;
2199 void *data;
2200};
2201
2202static int loose_from_alt_odb(struct alternate_object_database *alt,
2203 void *vdata)
2204{
2205 struct loose_alt_odb_data *data = vdata;
2206 struct strbuf buf = STRBUF_INIT;
2207 int r;
2208
2209 strbuf_addstr(&buf, alt->path);
2210 r = for_each_loose_file_in_objdir_buf(&buf,
2211 data->cb, NULL, NULL,
2212 data->data);
2213 strbuf_release(&buf);
2214 return r;
2215}
2216
2217int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2218{
2219 struct loose_alt_odb_data alt;
2220 int r;
2221
2222 r = for_each_loose_file_in_objdir(get_object_directory(),
2223 cb, NULL, NULL, data);
2224 if (r)
2225 return r;
2226
2227 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2228 return 0;
2229
2230 alt.cb = cb;
2231 alt.data = data;
2232 return foreach_alt_odb(loose_from_alt_odb, &alt);
2233}
2234
2235static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
2236{
2237 uint32_t i;
2238 int r = 0;
2239
2240 for (i = 0; i < p->num_objects; i++) {
2241 struct object_id oid;
2242
2243 if (!nth_packed_object_oid(&oid, p, i))
2244 return error("unable to get sha1 of object %u in %s",
2245 i, p->pack_name);
2246
2247 r = cb(&oid, p, i, data);
2248 if (r)
2249 break;
2250 }
2251 return r;
2252}
2253
2254int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
2255{
2256 struct packed_git *p;
2257 int r = 0;
2258 int pack_errors = 0;
2259
2260 prepare_packed_git();
2261 for (p = packed_git; p; p = p->next) {
2262 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2263 continue;
2264 if (open_pack_index(p)) {
2265 pack_errors = 1;
2266 continue;
2267 }
2268 r = for_each_object_in_pack(p, cb, data);
2269 if (r)
2270 break;
2271 }
2272 return r ? r : pack_errors;
2273}
2274
2275static int check_stream_sha1(git_zstream *stream,
2276 const char *hdr,
2277 unsigned long size,
2278 const char *path,
2279 const unsigned char *expected_sha1)
2280{
2281 git_SHA_CTX c;
2282 unsigned char real_sha1[GIT_MAX_RAWSZ];
2283 unsigned char buf[4096];
2284 unsigned long total_read;
2285 int status = Z_OK;
2286
2287 git_SHA1_Init(&c);
2288 git_SHA1_Update(&c, hdr, stream->total_out);
2289
2290 /*
2291 * We already read some bytes into hdr, but the ones up to the NUL
2292 * do not count against the object's content size.
2293 */
2294 total_read = stream->total_out - strlen(hdr) - 1;
2295
2296 /*
2297 * This size comparison must be "<=" to read the final zlib packets;
2298 * see the comment in unpack_sha1_rest for details.
2299 */
2300 while (total_read <= size &&
2301 (status == Z_OK || status == Z_BUF_ERROR)) {
2302 stream->next_out = buf;
2303 stream->avail_out = sizeof(buf);
2304 if (size - total_read < stream->avail_out)
2305 stream->avail_out = size - total_read;
2306 status = git_inflate(stream, Z_FINISH);
2307 git_SHA1_Update(&c, buf, stream->next_out - buf);
2308 total_read += stream->next_out - buf;
2309 }
2310 git_inflate_end(stream);
2311
2312 if (status != Z_STREAM_END) {
2313 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2314 return -1;
2315 }
2316 if (stream->avail_in) {
2317 error("garbage at end of loose object '%s'",
2318 sha1_to_hex(expected_sha1));
2319 return -1;
2320 }
2321
2322 git_SHA1_Final(real_sha1, &c);
2323 if (hashcmp(expected_sha1, real_sha1)) {
2324 error("sha1 mismatch for %s (expected %s)", path,
2325 sha1_to_hex(expected_sha1));
2326 return -1;
2327 }
2328
2329 return 0;
2330}
2331
2332int read_loose_object(const char *path,
2333 const unsigned char *expected_sha1,
2334 enum object_type *type,
2335 unsigned long *size,
2336 void **contents)
2337{
2338 int ret = -1;
2339 void *map = NULL;
2340 unsigned long mapsize;
2341 git_zstream stream;
2342 char hdr[32];
2343
2344 *contents = NULL;
2345
2346 map = map_sha1_file_1(path, NULL, &mapsize);
2347 if (!map) {
2348 error_errno("unable to mmap %s", path);
2349 goto out;
2350 }
2351
2352 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2353 error("unable to unpack header of %s", path);
2354 goto out;
2355 }
2356
2357 *type = parse_sha1_header(hdr, size);
2358 if (*type < 0) {
2359 error("unable to parse header of %s", path);
2360 git_inflate_end(&stream);
2361 goto out;
2362 }
2363
2364 if (*type == OBJ_BLOB) {
2365 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2366 goto out;
2367 } else {
2368 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2369 if (!*contents) {
2370 error("unable to unpack contents of %s", path);
2371 git_inflate_end(&stream);
2372 goto out;
2373 }
2374 if (check_sha1_signature(expected_sha1, *contents,
2375 *size, typename(*type))) {
2376 error("sha1 mismatch for %s (expected %s)", path,
2377 sha1_to_hex(expected_sha1));
2378 free(*contents);
2379 goto out;
2380 }
2381 }
2382
2383 ret = 0; /* everything checks out */
2384
2385out:
2386 if (map)
2387 munmap(map, mapsize);
2388 return ret;
2389}