Merge branch 'rs/sequencer-rewrite-file-cleanup' into maint
[git/gitweb.git] / sha1_file.c
blob36f85455e6748efb7c2d593c7a619fa13150600b
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
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"
33 const unsigned char null_sha1[GIT_MAX_RAWSZ];
34 const struct object_id null_oid;
35 const struct object_id empty_tree_oid = {
36 EMPTY_TREE_SHA1_BIN_LITERAL
38 const struct object_id empty_blob_oid = {
39 EMPTY_BLOB_SHA1_BIN_LITERAL
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).
48 static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53 } *cached_objects;
54 static int cached_object_nr, cached_object_alloc;
56 static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
58 OBJ_TREE,
59 "",
63 static struct cached_object *find_cached_object(const unsigned char *sha1)
65 int i;
66 struct cached_object *co = cached_objects;
68 for (i = 0; i < cached_object_nr; i++, co++) {
69 if (!hashcmp(co->sha1, sha1))
70 return co;
72 if (!hashcmp(sha1, empty_tree.sha1))
73 return &empty_tree;
74 return NULL;
77 int mkdir_in_gitdir(const char *path)
79 if (mkdir(path, 0777)) {
80 int saved_errno = errno;
81 struct stat st;
82 struct strbuf sb = STRBUF_INIT;
84 if (errno != EEXIST)
85 return -1;
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?
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;
101 strbuf_release(&sb);
103 return adjust_shared_perm(path);
106 enum scld_error safe_create_leading_directories(char *path)
108 char *next_component = path + offset_1st_component(path);
109 enum scld_error ret = SCLD_OK;
111 while (ret == SCLD_OK && next_component) {
112 struct stat st;
113 char *slash = next_component, slash_character;
115 while (*slash && !is_dir_sep(*slash))
116 slash++;
118 if (!*slash)
119 break;
121 next_component = slash + 1;
122 while (is_dir_sep(*next_component))
123 next_component++;
124 if (!*next_component)
125 break;
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;
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)
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:
149 ret = SCLD_VANISHED;
150 else
151 ret = SCLD_FAILED;
152 } else if (adjust_shared_perm(path)) {
153 ret = SCLD_PERMS;
155 *slash = slash_character;
157 return ret;
160 enum scld_error safe_create_leading_directories_const(const char *path)
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);
167 save_errno = errno;
168 free(buf);
169 errno = save_errno;
170 return result;
173 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
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.
181 int remove_directories_remaining = 1;
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.
190 int create_directories_remaining = 3;
192 /* A scratch copy of path, filled lazily if we need it: */
193 struct strbuf path_copy = STRBUF_INIT;
195 int ret, save_errno;
197 /* Sanity check: */
198 assert(*path);
200 retry_fn:
201 ret = fn(path, cb);
202 save_errno = errno;
203 if (!ret)
204 goto out;
206 if (errno == EISDIR && remove_directories_remaining-- > 0) {
208 * A directory is in the way. Maybe it is empty; try
209 * to remove it:
211 if (!path_copy.len)
212 strbuf_addstr(&path_copy, path);
214 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
215 goto retry_fn;
216 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
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:
223 enum scld_error scld_result;
225 if (!path_copy.len)
226 strbuf_addstr(&path_copy, path);
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);
235 out:
236 strbuf_release(&path_copy);
237 errno = save_errno;
238 return ret;
241 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
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, '/');
254 const char *sha1_file_name(const unsigned char *sha1)
256 static struct strbuf buf = STRBUF_INIT;
258 strbuf_reset(&buf);
259 strbuf_addf(&buf, "%s/", get_object_directory());
261 fill_sha1_path(&buf, sha1);
262 return buf.buf;
265 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
267 strbuf_setlen(&alt->scratch, alt->base_len);
268 return &alt->scratch;
271 static const char *alt_sha1_path(struct alternate_object_database *alt,
272 const unsigned char *sha1)
274 struct strbuf *buf = alt_scratch_buf(alt);
275 fill_sha1_path(buf, sha1);
276 return buf->buf;
279 struct alternate_object_database *alt_odb_list;
280 static struct alternate_object_database **alt_odb_tail;
283 * Return non-zero iff the path is usable as an alternate object database.
285 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
287 struct alternate_object_database *alt;
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;
298 * Prevent the common mistake of listing the same
299 * thing twice, or object directory itself.
301 for (alt = alt_odb_list; alt; alt = alt->next) {
302 if (!fspathcmp(path->buf, alt->path))
303 return 0;
305 if (!fspathcmp(path->buf, normalized_objdir))
306 return 0;
308 return 1;
312 * Prepare alternate object database registry.
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.
326 static void read_info_alternates(const char * relative_base, int depth);
327 static int link_alt_odb_entry(const char *entry, const char *relative_base,
328 int depth, const char *normalized_objdir)
330 struct alternate_object_database *ent;
331 struct strbuf pathbuf = STRBUF_INIT;
333 if (!is_absolute_path(entry) && relative_base) {
334 strbuf_realpath(&pathbuf, relative_base, 1);
335 strbuf_addch(&pathbuf, '/');
337 strbuf_addstr(&pathbuf, entry);
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;
347 * The trailing slash after the directory name is given by
348 * this function at the end. Remove duplicates.
350 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
351 strbuf_setlen(&pathbuf, pathbuf.len - 1);
353 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
354 strbuf_release(&pathbuf);
355 return -1;
358 ent = alloc_alt_odb(pathbuf.buf);
360 /* add the alternate entry */
361 *alt_odb_tail = ent;
362 alt_odb_tail = &(ent->next);
363 ent->next = NULL;
365 /* recursively add alternates */
366 read_info_alternates(pathbuf.buf, depth + 1);
368 strbuf_release(&pathbuf);
369 return 0;
372 static const char *parse_alt_odb_entry(const char *string,
373 int sep,
374 struct strbuf *out)
376 const char *end;
378 strbuf_reset(out);
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)) {
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.
390 } else {
391 /* normal, unquoted path */
392 end = strchrnul(string, sep);
393 strbuf_add(out, string, end - string);
396 if (*end)
397 end++;
398 return end;
401 static void link_alt_odb_entries(const char *alt, int sep,
402 const char *relative_base, int depth)
404 struct strbuf objdirbuf = STRBUF_INIT;
405 struct strbuf entry = STRBUF_INIT;
407 if (!alt || !*alt)
408 return;
410 if (depth > 5) {
411 error("%s: ignoring alternate object stores, nesting too deep.",
412 relative_base);
413 return;
416 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
417 if (strbuf_normalize_path(&objdirbuf) < 0)
418 die("unable to normalize object directory: %s",
419 objdirbuf.buf);
421 while (*alt) {
422 alt = parse_alt_odb_entry(alt, sep, &entry);
423 if (!entry.len)
424 continue;
425 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
427 strbuf_release(&entry);
428 strbuf_release(&objdirbuf);
431 static void read_info_alternates(const char * relative_base, int depth)
433 char *path;
434 struct strbuf buf = STRBUF_INIT;
436 path = xstrfmt("%s/info/alternates", relative_base);
437 if (strbuf_read_file(&buf, path, 1024) < 0) {
438 warn_on_fopen_errors(path);
439 free(path);
440 return;
443 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
444 strbuf_release(&buf);
445 free(path);
448 struct alternate_object_database *alloc_alt_odb(const char *dir)
450 struct alternate_object_database *ent;
452 FLEX_ALLOC_STR(ent, path, dir);
453 strbuf_init(&ent->scratch, 0);
454 strbuf_addf(&ent->scratch, "%s/", dir);
455 ent->base_len = ent->scratch.len;
457 return ent;
460 void add_to_alternates_file(const char *reference)
462 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
463 char *alts = git_pathdup("objects/info/alternates");
464 FILE *in, *out;
466 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
467 out = fdopen_lock_file(lock, "w");
468 if (!out)
469 die_errno("unable to fdopen alternates lockfile");
471 in = fopen(alts, "r");
472 if (in) {
473 struct strbuf line = STRBUF_INIT;
474 int found = 0;
476 while (strbuf_getline(&line, in) != EOF) {
477 if (!strcmp(reference, line.buf)) {
478 found = 1;
479 break;
481 fprintf_or_die(out, "%s\n", line.buf);
484 strbuf_release(&line);
485 fclose(in);
487 if (found) {
488 rollback_lock_file(lock);
489 lock = NULL;
492 else if (errno != ENOENT)
493 die_errno("unable to read alternates file");
495 if (lock) {
496 fprintf_or_die(out, "%s\n", reference);
497 if (commit_lock_file(lock))
498 die_errno("unable to move new alternates file into place");
499 if (alt_odb_tail)
500 link_alt_odb_entries(reference, '\n', NULL, 0);
502 free(alts);
505 void add_to_alternates_memory(const char *reference)
508 * Make sure alternates are initialized, or else our entry may be
509 * overwritten when they are.
511 prepare_alt_odb();
513 link_alt_odb_entries(reference, '\n', NULL, 0);
517 * Compute the exact path an alternate is at and returns it. In case of
518 * error NULL is returned and the human readable error is added to `err`
519 * `path` may be relative and should point to $GITDIR.
520 * `err` must not be null.
522 char *compute_alternate_path(const char *path, struct strbuf *err)
524 char *ref_git = NULL;
525 const char *repo, *ref_git_s;
526 int seen_error = 0;
528 ref_git_s = real_path_if_valid(path);
529 if (!ref_git_s) {
530 seen_error = 1;
531 strbuf_addf(err, _("path '%s' does not exist"), path);
532 goto out;
533 } else
535 * Beware: read_gitfile(), real_path() and mkpath()
536 * return static buffer
538 ref_git = xstrdup(ref_git_s);
540 repo = read_gitfile(ref_git);
541 if (!repo)
542 repo = read_gitfile(mkpath("%s/.git", ref_git));
543 if (repo) {
544 free(ref_git);
545 ref_git = xstrdup(repo);
548 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
549 char *ref_git_git = mkpathdup("%s/.git", ref_git);
550 free(ref_git);
551 ref_git = ref_git_git;
552 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
553 struct strbuf sb = STRBUF_INIT;
554 seen_error = 1;
555 if (get_common_dir(&sb, ref_git)) {
556 strbuf_addf(err,
557 _("reference repository '%s' as a linked "
558 "checkout is not supported yet."),
559 path);
560 goto out;
563 strbuf_addf(err, _("reference repository '%s' is not a "
564 "local repository."), path);
565 goto out;
568 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
569 strbuf_addf(err, _("reference repository '%s' is shallow"),
570 path);
571 seen_error = 1;
572 goto out;
575 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
576 strbuf_addf(err,
577 _("reference repository '%s' is grafted"),
578 path);
579 seen_error = 1;
580 goto out;
583 out:
584 if (seen_error) {
585 FREE_AND_NULL(ref_git);
588 return ref_git;
591 int foreach_alt_odb(alt_odb_fn fn, void *cb)
593 struct alternate_object_database *ent;
594 int r = 0;
596 prepare_alt_odb();
597 for (ent = alt_odb_list; ent; ent = ent->next) {
598 r = fn(ent, cb);
599 if (r)
600 break;
602 return r;
605 void prepare_alt_odb(void)
607 const char *alt;
609 if (alt_odb_tail)
610 return;
612 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
614 alt_odb_tail = &alt_odb_list;
615 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
617 read_info_alternates(get_object_directory(), 0);
620 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
621 static int freshen_file(const char *fn)
623 struct utimbuf t;
624 t.actime = t.modtime = time(NULL);
625 return !utime(fn, &t);
629 * All of the check_and_freshen functions return 1 if the file exists and was
630 * freshened (if freshening was requested), 0 otherwise. If they return
631 * 0, you should not assume that it is safe to skip a write of the object (it
632 * either does not exist on disk, or has a stale mtime and may be subject to
633 * pruning).
635 int check_and_freshen_file(const char *fn, int freshen)
637 if (access(fn, F_OK))
638 return 0;
639 if (freshen && !freshen_file(fn))
640 return 0;
641 return 1;
644 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
646 return check_and_freshen_file(sha1_file_name(sha1), freshen);
649 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
651 struct alternate_object_database *alt;
652 prepare_alt_odb();
653 for (alt = alt_odb_list; alt; alt = alt->next) {
654 const char *path = alt_sha1_path(alt, sha1);
655 if (check_and_freshen_file(path, freshen))
656 return 1;
658 return 0;
661 static int check_and_freshen(const unsigned char *sha1, int freshen)
663 return check_and_freshen_local(sha1, freshen) ||
664 check_and_freshen_nonlocal(sha1, freshen);
667 int has_loose_object_nonlocal(const unsigned char *sha1)
669 return check_and_freshen_nonlocal(sha1, 0);
672 static int has_loose_object(const unsigned char *sha1)
674 return check_and_freshen(sha1, 0);
677 static void mmap_limit_check(size_t length)
679 static size_t limit = 0;
680 if (!limit) {
681 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
682 if (!limit)
683 limit = SIZE_MAX;
685 if (length > limit)
686 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
687 (uintmax_t)length, (uintmax_t)limit);
690 void *xmmap_gently(void *start, size_t length,
691 int prot, int flags, int fd, off_t offset)
693 void *ret;
695 mmap_limit_check(length);
696 ret = mmap(start, length, prot, flags, fd, offset);
697 if (ret == MAP_FAILED) {
698 if (!length)
699 return NULL;
700 release_pack_memory(length);
701 ret = mmap(start, length, prot, flags, fd, offset);
703 return ret;
706 void *xmmap(void *start, size_t length,
707 int prot, int flags, int fd, off_t offset)
709 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
710 if (ret == MAP_FAILED)
711 die_errno("mmap failed");
712 return ret;
716 * With an in-core object data in "map", rehash it to make sure the
717 * object name actually matches "sha1" to detect object corruption.
718 * With "map" == NULL, try reading the object named with "sha1" using
719 * the streaming interface and rehash it to do the same.
721 int check_sha1_signature(const unsigned char *sha1, void *map,
722 unsigned long size, const char *type)
724 unsigned char real_sha1[20];
725 enum object_type obj_type;
726 struct git_istream *st;
727 git_SHA_CTX c;
728 char hdr[32];
729 int hdrlen;
731 if (map) {
732 hash_sha1_file(map, size, type, real_sha1);
733 return hashcmp(sha1, real_sha1) ? -1 : 0;
736 st = open_istream(sha1, &obj_type, &size, NULL);
737 if (!st)
738 return -1;
740 /* Generate the header */
741 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
743 /* Sha1.. */
744 git_SHA1_Init(&c);
745 git_SHA1_Update(&c, hdr, hdrlen);
746 for (;;) {
747 char buf[1024 * 16];
748 ssize_t readlen = read_istream(st, buf, sizeof(buf));
750 if (readlen < 0) {
751 close_istream(st);
752 return -1;
754 if (!readlen)
755 break;
756 git_SHA1_Update(&c, buf, readlen);
758 git_SHA1_Final(real_sha1, &c);
759 close_istream(st);
760 return hashcmp(sha1, real_sha1) ? -1 : 0;
763 int git_open_cloexec(const char *name, int flags)
765 int fd;
766 static int o_cloexec = O_CLOEXEC;
768 fd = open(name, flags | o_cloexec);
769 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
770 /* Try again w/o O_CLOEXEC: the kernel might not support it */
771 o_cloexec &= ~O_CLOEXEC;
772 fd = open(name, flags | o_cloexec);
775 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
777 static int fd_cloexec = FD_CLOEXEC;
779 if (!o_cloexec && 0 <= fd && fd_cloexec) {
780 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
781 int flags = fcntl(fd, F_GETFD);
782 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
783 fd_cloexec = 0;
786 #endif
787 return fd;
791 * Find "sha1" as a loose object in the local repository or in an alternate.
792 * Returns 0 on success, negative on failure.
794 * The "path" out-parameter will give the path of the object we found (if any).
795 * Note that it may point to static storage and is only valid until another
796 * call to sha1_file_name(), etc.
798 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
799 const char **path)
801 struct alternate_object_database *alt;
803 *path = sha1_file_name(sha1);
804 if (!lstat(*path, st))
805 return 0;
807 prepare_alt_odb();
808 errno = ENOENT;
809 for (alt = alt_odb_list; alt; alt = alt->next) {
810 *path = alt_sha1_path(alt, sha1);
811 if (!lstat(*path, st))
812 return 0;
815 return -1;
819 * Like stat_sha1_file(), but actually open the object and return the
820 * descriptor. See the caveats on the "path" parameter above.
822 static int open_sha1_file(const unsigned char *sha1, const char **path)
824 int fd;
825 struct alternate_object_database *alt;
826 int most_interesting_errno;
828 *path = sha1_file_name(sha1);
829 fd = git_open(*path);
830 if (fd >= 0)
831 return fd;
832 most_interesting_errno = errno;
834 prepare_alt_odb();
835 for (alt = alt_odb_list; alt; alt = alt->next) {
836 *path = alt_sha1_path(alt, sha1);
837 fd = git_open(*path);
838 if (fd >= 0)
839 return fd;
840 if (most_interesting_errno == ENOENT)
841 most_interesting_errno = errno;
843 errno = most_interesting_errno;
844 return -1;
848 * Map the loose object at "path" if it is not NULL, or the path found by
849 * searching for a loose object named "sha1".
851 static void *map_sha1_file_1(const char *path,
852 const unsigned char *sha1,
853 unsigned long *size)
855 void *map;
856 int fd;
858 if (path)
859 fd = git_open(path);
860 else
861 fd = open_sha1_file(sha1, &path);
862 map = NULL;
863 if (fd >= 0) {
864 struct stat st;
866 if (!fstat(fd, &st)) {
867 *size = xsize_t(st.st_size);
868 if (!*size) {
869 /* mmap() is forbidden on empty files */
870 error("object file %s is empty", path);
871 return NULL;
873 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
875 close(fd);
877 return map;
880 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
882 return map_sha1_file_1(NULL, sha1, size);
885 static int unpack_sha1_short_header(git_zstream *stream,
886 unsigned char *map, unsigned long mapsize,
887 void *buffer, unsigned long bufsiz)
889 /* Get the data stream */
890 memset(stream, 0, sizeof(*stream));
891 stream->next_in = map;
892 stream->avail_in = mapsize;
893 stream->next_out = buffer;
894 stream->avail_out = bufsiz;
896 git_inflate_init(stream);
897 return git_inflate(stream, 0);
900 int unpack_sha1_header(git_zstream *stream,
901 unsigned char *map, unsigned long mapsize,
902 void *buffer, unsigned long bufsiz)
904 int status = unpack_sha1_short_header(stream, map, mapsize,
905 buffer, bufsiz);
907 if (status < Z_OK)
908 return status;
910 /* Make sure we have the terminating NUL */
911 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
912 return -1;
913 return 0;
916 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
917 unsigned long mapsize, void *buffer,
918 unsigned long bufsiz, struct strbuf *header)
920 int status;
922 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
923 if (status < Z_OK)
924 return -1;
927 * Check if entire header is unpacked in the first iteration.
929 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
930 return 0;
933 * buffer[0..bufsiz] was not large enough. Copy the partial
934 * result out to header, and then append the result of further
935 * reading the stream.
937 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
938 stream->next_out = buffer;
939 stream->avail_out = bufsiz;
941 do {
942 status = git_inflate(stream, 0);
943 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
944 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
945 return 0;
946 stream->next_out = buffer;
947 stream->avail_out = bufsiz;
948 } while (status != Z_STREAM_END);
949 return -1;
952 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
954 int bytes = strlen(buffer) + 1;
955 unsigned char *buf = xmallocz(size);
956 unsigned long n;
957 int status = Z_OK;
959 n = stream->total_out - bytes;
960 if (n > size)
961 n = size;
962 memcpy(buf, (char *) buffer + bytes, n);
963 bytes = n;
964 if (bytes <= size) {
966 * The above condition must be (bytes <= size), not
967 * (bytes < size). In other words, even though we
968 * expect no more output and set avail_out to zero,
969 * the input zlib stream may have bytes that express
970 * "this concludes the stream", and we *do* want to
971 * eat that input.
973 * Otherwise we would not be able to test that we
974 * consumed all the input to reach the expected size;
975 * we also want to check that zlib tells us that all
976 * went well with status == Z_STREAM_END at the end.
978 stream->next_out = buf + bytes;
979 stream->avail_out = size - bytes;
980 while (status == Z_OK)
981 status = git_inflate(stream, Z_FINISH);
983 if (status == Z_STREAM_END && !stream->avail_in) {
984 git_inflate_end(stream);
985 return buf;
988 if (status < 0)
989 error("corrupt loose object '%s'", sha1_to_hex(sha1));
990 else if (stream->avail_in)
991 error("garbage at end of loose object '%s'",
992 sha1_to_hex(sha1));
993 free(buf);
994 return NULL;
998 * We used to just use "sscanf()", but that's actually way
999 * too permissive for what we want to check. So do an anal
1000 * object header parse by hand.
1002 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1003 unsigned int flags)
1005 const char *type_buf = hdr;
1006 unsigned long size;
1007 int type, type_len = 0;
1010 * The type can be of any size but is followed by
1011 * a space.
1013 for (;;) {
1014 char c = *hdr++;
1015 if (!c)
1016 return -1;
1017 if (c == ' ')
1018 break;
1019 type_len++;
1022 type = type_from_string_gently(type_buf, type_len, 1);
1023 if (oi->typename)
1024 strbuf_add(oi->typename, type_buf, type_len);
1026 * Set type to 0 if its an unknown object and
1027 * we're obtaining the type using '--allow-unknown-type'
1028 * option.
1030 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1031 type = 0;
1032 else if (type < 0)
1033 die("invalid object type");
1034 if (oi->typep)
1035 *oi->typep = type;
1038 * The length must follow immediately, and be in canonical
1039 * decimal format (ie "010" is not valid).
1041 size = *hdr++ - '0';
1042 if (size > 9)
1043 return -1;
1044 if (size) {
1045 for (;;) {
1046 unsigned long c = *hdr - '0';
1047 if (c > 9)
1048 break;
1049 hdr++;
1050 size = size * 10 + c;
1054 if (oi->sizep)
1055 *oi->sizep = size;
1058 * The length must be followed by a zero byte
1060 return *hdr ? -1 : type;
1063 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1065 struct object_info oi = OBJECT_INFO_INIT;
1067 oi.sizep = sizep;
1068 return parse_sha1_header_extended(hdr, &oi, 0);
1071 static int sha1_loose_object_info(const unsigned char *sha1,
1072 struct object_info *oi,
1073 int flags)
1075 int status = 0;
1076 unsigned long mapsize;
1077 void *map;
1078 git_zstream stream;
1079 char hdr[32];
1080 struct strbuf hdrbuf = STRBUF_INIT;
1081 unsigned long size_scratch;
1083 if (oi->delta_base_sha1)
1084 hashclr(oi->delta_base_sha1);
1087 * If we don't care about type or size, then we don't
1088 * need to look inside the object at all. Note that we
1089 * do not optimize out the stat call, even if the
1090 * caller doesn't care about the disk-size, since our
1091 * return value implicitly indicates whether the
1092 * object even exists.
1094 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1095 const char *path;
1096 struct stat st;
1097 if (stat_sha1_file(sha1, &st, &path) < 0)
1098 return -1;
1099 if (oi->disk_sizep)
1100 *oi->disk_sizep = st.st_size;
1101 return 0;
1104 map = map_sha1_file(sha1, &mapsize);
1105 if (!map)
1106 return -1;
1108 if (!oi->sizep)
1109 oi->sizep = &size_scratch;
1111 if (oi->disk_sizep)
1112 *oi->disk_sizep = mapsize;
1113 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1114 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1115 status = error("unable to unpack %s header with --allow-unknown-type",
1116 sha1_to_hex(sha1));
1117 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1118 status = error("unable to unpack %s header",
1119 sha1_to_hex(sha1));
1120 if (status < 0)
1121 ; /* Do nothing */
1122 else if (hdrbuf.len) {
1123 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1124 status = error("unable to parse %s header with --allow-unknown-type",
1125 sha1_to_hex(sha1));
1126 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1127 status = error("unable to parse %s header", sha1_to_hex(sha1));
1129 if (status >= 0 && oi->contentp) {
1130 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1131 *oi->sizep, sha1);
1132 if (!*oi->contentp) {
1133 git_inflate_end(&stream);
1134 status = -1;
1136 } else
1137 git_inflate_end(&stream);
1139 munmap(map, mapsize);
1140 if (status && oi->typep)
1141 *oi->typep = status;
1142 if (oi->sizep == &size_scratch)
1143 oi->sizep = NULL;
1144 strbuf_release(&hdrbuf);
1145 oi->whence = OI_LOOSE;
1146 return (status < 0) ? status : 0;
1149 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1151 static struct object_info blank_oi = OBJECT_INFO_INIT;
1152 struct pack_entry e;
1153 int rtype;
1154 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1155 lookup_replace_object(sha1) :
1156 sha1;
1158 if (!oi)
1159 oi = &blank_oi;
1161 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1162 struct cached_object *co = find_cached_object(real);
1163 if (co) {
1164 if (oi->typep)
1165 *(oi->typep) = co->type;
1166 if (oi->sizep)
1167 *(oi->sizep) = co->size;
1168 if (oi->disk_sizep)
1169 *(oi->disk_sizep) = 0;
1170 if (oi->delta_base_sha1)
1171 hashclr(oi->delta_base_sha1);
1172 if (oi->typename)
1173 strbuf_addstr(oi->typename, typename(co->type));
1174 if (oi->contentp)
1175 *oi->contentp = xmemdupz(co->buf, co->size);
1176 oi->whence = OI_CACHED;
1177 return 0;
1181 if (!find_pack_entry(real, &e)) {
1182 /* Most likely it's a loose object. */
1183 if (!sha1_loose_object_info(real, oi, flags))
1184 return 0;
1186 /* Not a loose object; someone else may have just packed it. */
1187 if (flags & OBJECT_INFO_QUICK) {
1188 return -1;
1189 } else {
1190 reprepare_packed_git();
1191 if (!find_pack_entry(real, &e))
1192 return -1;
1196 if (oi == &blank_oi)
1198 * We know that the caller doesn't actually need the
1199 * information below, so return early.
1201 return 0;
1203 rtype = packed_object_info(e.p, e.offset, oi);
1204 if (rtype < 0) {
1205 mark_bad_packed_object(e.p, real);
1206 return sha1_object_info_extended(real, oi, 0);
1207 } else if (oi->whence == OI_PACKED) {
1208 oi->u.packed.offset = e.offset;
1209 oi->u.packed.pack = e.p;
1210 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1211 rtype == OBJ_OFS_DELTA);
1214 return 0;
1217 /* returns enum object_type or negative */
1218 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1220 enum object_type type;
1221 struct object_info oi = OBJECT_INFO_INIT;
1223 oi.typep = &type;
1224 oi.sizep = sizep;
1225 if (sha1_object_info_extended(sha1, &oi,
1226 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1227 return -1;
1228 return type;
1231 static void *read_object(const unsigned char *sha1, enum object_type *type,
1232 unsigned long *size)
1234 struct object_info oi = OBJECT_INFO_INIT;
1235 void *content;
1236 oi.typep = type;
1237 oi.sizep = size;
1238 oi.contentp = &content;
1240 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1241 return NULL;
1242 return content;
1245 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1246 unsigned char *sha1)
1248 struct cached_object *co;
1250 hash_sha1_file(buf, len, typename(type), sha1);
1251 if (has_sha1_file(sha1) || find_cached_object(sha1))
1252 return 0;
1253 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1254 co = &cached_objects[cached_object_nr++];
1255 co->size = len;
1256 co->type = type;
1257 co->buf = xmalloc(len);
1258 memcpy(co->buf, buf, len);
1259 hashcpy(co->sha1, sha1);
1260 return 0;
1264 * This function dies on corrupt objects; the callers who want to
1265 * deal with them should arrange to call read_object() and give error
1266 * messages themselves.
1268 void *read_sha1_file_extended(const unsigned char *sha1,
1269 enum object_type *type,
1270 unsigned long *size,
1271 int lookup_replace)
1273 void *data;
1274 const struct packed_git *p;
1275 const char *path;
1276 struct stat st;
1277 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1278 : sha1;
1280 errno = 0;
1281 data = read_object(repl, type, size);
1282 if (data)
1283 return data;
1285 if (errno && errno != ENOENT)
1286 die_errno("failed to read object %s", sha1_to_hex(sha1));
1288 /* die if we replaced an object with one that does not exist */
1289 if (repl != sha1)
1290 die("replacement %s not found for %s",
1291 sha1_to_hex(repl), sha1_to_hex(sha1));
1293 if (!stat_sha1_file(repl, &st, &path))
1294 die("loose object %s (stored in %s) is corrupt",
1295 sha1_to_hex(repl), path);
1297 if ((p = has_packed_and_bad(repl)) != NULL)
1298 die("packed object %s (stored in %s) is corrupt",
1299 sha1_to_hex(repl), p->pack_name);
1301 return NULL;
1304 void *read_object_with_reference(const unsigned char *sha1,
1305 const char *required_type_name,
1306 unsigned long *size,
1307 unsigned char *actual_sha1_return)
1309 enum object_type type, required_type;
1310 void *buffer;
1311 unsigned long isize;
1312 unsigned char actual_sha1[20];
1314 required_type = type_from_string(required_type_name);
1315 hashcpy(actual_sha1, sha1);
1316 while (1) {
1317 int ref_length = -1;
1318 const char *ref_type = NULL;
1320 buffer = read_sha1_file(actual_sha1, &type, &isize);
1321 if (!buffer)
1322 return NULL;
1323 if (type == required_type) {
1324 *size = isize;
1325 if (actual_sha1_return)
1326 hashcpy(actual_sha1_return, actual_sha1);
1327 return buffer;
1329 /* Handle references */
1330 else if (type == OBJ_COMMIT)
1331 ref_type = "tree ";
1332 else if (type == OBJ_TAG)
1333 ref_type = "object ";
1334 else {
1335 free(buffer);
1336 return NULL;
1338 ref_length = strlen(ref_type);
1340 if (ref_length + 40 > isize ||
1341 memcmp(buffer, ref_type, ref_length) ||
1342 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1343 free(buffer);
1344 return NULL;
1346 free(buffer);
1347 /* Now we have the ID of the referred-to object in
1348 * actual_sha1. Check again. */
1352 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1353 const char *type, unsigned char *sha1,
1354 char *hdr, int *hdrlen)
1356 git_SHA_CTX c;
1358 /* Generate the header */
1359 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1361 /* Sha1.. */
1362 git_SHA1_Init(&c);
1363 git_SHA1_Update(&c, hdr, *hdrlen);
1364 git_SHA1_Update(&c, buf, len);
1365 git_SHA1_Final(sha1, &c);
1369 * Move the just written object into its final resting place.
1371 int finalize_object_file(const char *tmpfile, const char *filename)
1373 int ret = 0;
1375 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1376 goto try_rename;
1377 else if (link(tmpfile, filename))
1378 ret = errno;
1381 * Coda hack - coda doesn't like cross-directory links,
1382 * so we fall back to a rename, which will mean that it
1383 * won't be able to check collisions, but that's not a
1384 * big deal.
1386 * The same holds for FAT formatted media.
1388 * When this succeeds, we just return. We have nothing
1389 * left to unlink.
1391 if (ret && ret != EEXIST) {
1392 try_rename:
1393 if (!rename(tmpfile, filename))
1394 goto out;
1395 ret = errno;
1397 unlink_or_warn(tmpfile);
1398 if (ret) {
1399 if (ret != EEXIST) {
1400 return error_errno("unable to write sha1 filename %s", filename);
1402 /* FIXME!!! Collision check here ? */
1405 out:
1406 if (adjust_shared_perm(filename))
1407 return error("unable to set permission to '%s'", filename);
1408 return 0;
1411 static int write_buffer(int fd, const void *buf, size_t len)
1413 if (write_in_full(fd, buf, len) < 0)
1414 return error_errno("file write error");
1415 return 0;
1418 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1419 unsigned char *sha1)
1421 char hdr[32];
1422 int hdrlen = sizeof(hdr);
1423 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1424 return 0;
1427 /* Finalize a file on disk, and close it. */
1428 static void close_sha1_file(int fd)
1430 if (fsync_object_files)
1431 fsync_or_die(fd, "sha1 file");
1432 if (close(fd) != 0)
1433 die_errno("error when closing sha1 file");
1436 /* Size of directory component, including the ending '/' */
1437 static inline int directory_size(const char *filename)
1439 const char *s = strrchr(filename, '/');
1440 if (!s)
1441 return 0;
1442 return s - filename + 1;
1446 * This creates a temporary file in the same directory as the final
1447 * 'filename'
1449 * We want to avoid cross-directory filename renames, because those
1450 * can have problems on various filesystems (FAT, NFS, Coda).
1452 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1454 int fd, dirlen = directory_size(filename);
1456 strbuf_reset(tmp);
1457 strbuf_add(tmp, filename, dirlen);
1458 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1459 fd = git_mkstemp_mode(tmp->buf, 0444);
1460 if (fd < 0 && dirlen && errno == ENOENT) {
1462 * Make sure the directory exists; note that the contents
1463 * of the buffer are undefined after mkstemp returns an
1464 * error, so we have to rewrite the whole buffer from
1465 * scratch.
1467 strbuf_reset(tmp);
1468 strbuf_add(tmp, filename, dirlen - 1);
1469 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1470 return -1;
1471 if (adjust_shared_perm(tmp->buf))
1472 return -1;
1474 /* Try again */
1475 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1476 fd = git_mkstemp_mode(tmp->buf, 0444);
1478 return fd;
1481 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1482 const void *buf, unsigned long len, time_t mtime)
1484 int fd, ret;
1485 unsigned char compressed[4096];
1486 git_zstream stream;
1487 git_SHA_CTX c;
1488 unsigned char parano_sha1[20];
1489 static struct strbuf tmp_file = STRBUF_INIT;
1490 const char *filename = sha1_file_name(sha1);
1492 fd = create_tmpfile(&tmp_file, filename);
1493 if (fd < 0) {
1494 if (errno == EACCES)
1495 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1496 else
1497 return error_errno("unable to create temporary file");
1500 /* Set it up */
1501 git_deflate_init(&stream, zlib_compression_level);
1502 stream.next_out = compressed;
1503 stream.avail_out = sizeof(compressed);
1504 git_SHA1_Init(&c);
1506 /* First header.. */
1507 stream.next_in = (unsigned char *)hdr;
1508 stream.avail_in = hdrlen;
1509 while (git_deflate(&stream, 0) == Z_OK)
1510 ; /* nothing */
1511 git_SHA1_Update(&c, hdr, hdrlen);
1513 /* Then the data itself.. */
1514 stream.next_in = (void *)buf;
1515 stream.avail_in = len;
1516 do {
1517 unsigned char *in0 = stream.next_in;
1518 ret = git_deflate(&stream, Z_FINISH);
1519 git_SHA1_Update(&c, in0, stream.next_in - in0);
1520 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1521 die("unable to write sha1 file");
1522 stream.next_out = compressed;
1523 stream.avail_out = sizeof(compressed);
1524 } while (ret == Z_OK);
1526 if (ret != Z_STREAM_END)
1527 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1528 ret = git_deflate_end_gently(&stream);
1529 if (ret != Z_OK)
1530 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1531 git_SHA1_Final(parano_sha1, &c);
1532 if (hashcmp(sha1, parano_sha1) != 0)
1533 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1535 close_sha1_file(fd);
1537 if (mtime) {
1538 struct utimbuf utb;
1539 utb.actime = mtime;
1540 utb.modtime = mtime;
1541 if (utime(tmp_file.buf, &utb) < 0)
1542 warning_errno("failed utime() on %s", tmp_file.buf);
1545 return finalize_object_file(tmp_file.buf, filename);
1548 static int freshen_loose_object(const unsigned char *sha1)
1550 return check_and_freshen(sha1, 1);
1553 static int freshen_packed_object(const unsigned char *sha1)
1555 struct pack_entry e;
1556 if (!find_pack_entry(sha1, &e))
1557 return 0;
1558 if (e.p->freshened)
1559 return 1;
1560 if (!freshen_file(e.p->pack_name))
1561 return 0;
1562 e.p->freshened = 1;
1563 return 1;
1566 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1568 char hdr[32];
1569 int hdrlen = sizeof(hdr);
1571 /* Normally if we have it in the pack then we do not bother writing
1572 * it out into .git/objects/??/?{38} file.
1574 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1575 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1576 return 0;
1577 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1580 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1581 struct object_id *oid, unsigned flags)
1583 char *header;
1584 int hdrlen, status = 0;
1586 /* type string, SP, %lu of the length plus NUL must fit this */
1587 hdrlen = strlen(type) + 32;
1588 header = xmalloc(hdrlen);
1589 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1591 if (!(flags & HASH_WRITE_OBJECT))
1592 goto cleanup;
1593 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1594 goto cleanup;
1595 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1597 cleanup:
1598 free(header);
1599 return status;
1602 int force_object_loose(const unsigned char *sha1, time_t mtime)
1604 void *buf;
1605 unsigned long len;
1606 enum object_type type;
1607 char hdr[32];
1608 int hdrlen;
1609 int ret;
1611 if (has_loose_object(sha1))
1612 return 0;
1613 buf = read_object(sha1, &type, &len);
1614 if (!buf)
1615 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1616 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1617 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1618 free(buf);
1620 return ret;
1623 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1625 if (!startup_info->have_repository)
1626 return 0;
1627 return sha1_object_info_extended(sha1, NULL,
1628 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1631 int has_object_file(const struct object_id *oid)
1633 return has_sha1_file(oid->hash);
1636 int has_object_file_with_flags(const struct object_id *oid, int flags)
1638 return has_sha1_file_with_flags(oid->hash, flags);
1641 static void check_tree(const void *buf, size_t size)
1643 struct tree_desc desc;
1644 struct name_entry entry;
1646 init_tree_desc(&desc, buf, size);
1647 while (tree_entry(&desc, &entry))
1648 /* do nothing
1649 * tree_entry() will die() on malformed entries */
1653 static void check_commit(const void *buf, size_t size)
1655 struct commit c;
1656 memset(&c, 0, sizeof(c));
1657 if (parse_commit_buffer(&c, buf, size))
1658 die("corrupt commit");
1661 static void check_tag(const void *buf, size_t size)
1663 struct tag t;
1664 memset(&t, 0, sizeof(t));
1665 if (parse_tag_buffer(&t, buf, size))
1666 die("corrupt tag");
1669 static int index_mem(unsigned char *sha1, void *buf, size_t size,
1670 enum object_type type,
1671 const char *path, unsigned flags)
1673 int ret, re_allocated = 0;
1674 int write_object = flags & HASH_WRITE_OBJECT;
1676 if (!type)
1677 type = OBJ_BLOB;
1680 * Convert blobs to git internal format
1682 if ((type == OBJ_BLOB) && path) {
1683 struct strbuf nbuf = STRBUF_INIT;
1684 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1685 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1686 buf = strbuf_detach(&nbuf, &size);
1687 re_allocated = 1;
1690 if (flags & HASH_FORMAT_CHECK) {
1691 if (type == OBJ_TREE)
1692 check_tree(buf, size);
1693 if (type == OBJ_COMMIT)
1694 check_commit(buf, size);
1695 if (type == OBJ_TAG)
1696 check_tag(buf, size);
1699 if (write_object)
1700 ret = write_sha1_file(buf, size, typename(type), sha1);
1701 else
1702 ret = hash_sha1_file(buf, size, typename(type), sha1);
1703 if (re_allocated)
1704 free(buf);
1705 return ret;
1708 static int index_stream_convert_blob(unsigned char *sha1, int fd,
1709 const char *path, unsigned flags)
1711 int ret;
1712 const int write_object = flags & HASH_WRITE_OBJECT;
1713 struct strbuf sbuf = STRBUF_INIT;
1715 assert(path);
1716 assert(would_convert_to_git_filter_fd(path));
1718 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1719 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1721 if (write_object)
1722 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1723 sha1);
1724 else
1725 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1726 sha1);
1727 strbuf_release(&sbuf);
1728 return ret;
1731 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
1732 const char *path, unsigned flags)
1734 struct strbuf sbuf = STRBUF_INIT;
1735 int ret;
1737 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1738 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
1739 else
1740 ret = -1;
1741 strbuf_release(&sbuf);
1742 return ret;
1745 #define SMALL_FILE_SIZE (32*1024)
1747 static int index_core(unsigned char *sha1, int fd, size_t size,
1748 enum object_type type, const char *path,
1749 unsigned flags)
1751 int ret;
1753 if (!size) {
1754 ret = index_mem(sha1, "", size, type, path, flags);
1755 } else if (size <= SMALL_FILE_SIZE) {
1756 char *buf = xmalloc(size);
1757 ssize_t read_result = read_in_full(fd, buf, size);
1758 if (read_result < 0)
1759 ret = error_errno("read error while indexing %s",
1760 path ? path : "<unknown>");
1761 else if (read_result != size)
1762 ret = error("short read while indexing %s",
1763 path ? path : "<unknown>");
1764 else
1765 ret = index_mem(sha1, buf, size, type, path, flags);
1766 free(buf);
1767 } else {
1768 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1769 ret = index_mem(sha1, buf, size, type, path, flags);
1770 munmap(buf, size);
1772 return ret;
1776 * This creates one packfile per large blob unless bulk-checkin
1777 * machinery is "plugged".
1779 * This also bypasses the usual "convert-to-git" dance, and that is on
1780 * purpose. We could write a streaming version of the converting
1781 * functions and insert that before feeding the data to fast-import
1782 * (or equivalent in-core API described above). However, that is
1783 * somewhat complicated, as we do not know the size of the filter
1784 * result, which we need to know beforehand when writing a git object.
1785 * Since the primary motivation for trying to stream from the working
1786 * tree file and to avoid mmaping it in core is to deal with large
1787 * binary blobs, they generally do not want to get any conversion, and
1788 * callers should avoid this code path when filters are requested.
1790 static int index_stream(struct object_id *oid, int fd, size_t size,
1791 enum object_type type, const char *path,
1792 unsigned flags)
1794 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1797 int index_fd(struct object_id *oid, int fd, struct stat *st,
1798 enum object_type type, const char *path, unsigned flags)
1800 int ret;
1803 * Call xsize_t() only when needed to avoid potentially unnecessary
1804 * die() for large files.
1806 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1807 ret = index_stream_convert_blob(oid->hash, fd, path, flags);
1808 else if (!S_ISREG(st->st_mode))
1809 ret = index_pipe(oid->hash, fd, type, path, flags);
1810 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1811 (path && would_convert_to_git(&the_index, path)))
1812 ret = index_core(oid->hash, fd, xsize_t(st->st_size), type, path,
1813 flags);
1814 else
1815 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1816 flags);
1817 close(fd);
1818 return ret;
1821 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1823 int fd;
1824 struct strbuf sb = STRBUF_INIT;
1825 int rc = 0;
1827 switch (st->st_mode & S_IFMT) {
1828 case S_IFREG:
1829 fd = open(path, O_RDONLY);
1830 if (fd < 0)
1831 return error_errno("open(\"%s\")", path);
1832 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1833 return error("%s: failed to insert into database",
1834 path);
1835 break;
1836 case S_IFLNK:
1837 if (strbuf_readlink(&sb, path, st->st_size))
1838 return error_errno("readlink(\"%s\")", path);
1839 if (!(flags & HASH_WRITE_OBJECT))
1840 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1841 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1842 rc = error("%s: failed to insert into database", path);
1843 strbuf_release(&sb);
1844 break;
1845 case S_IFDIR:
1846 return resolve_gitlink_ref(path, "HEAD", oid->hash);
1847 default:
1848 return error("%s: unsupported file type", path);
1850 return rc;
1853 int read_pack_header(int fd, struct pack_header *header)
1855 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1856 /* "eof before pack header was fully read" */
1857 return PH_ERROR_EOF;
1859 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1860 /* "protocol error (pack signature mismatch detected)" */
1861 return PH_ERROR_PACK_SIGNATURE;
1862 if (!pack_version_ok(header->hdr_version))
1863 /* "protocol error (pack version unsupported)" */
1864 return PH_ERROR_PROTOCOL;
1865 return 0;
1868 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1870 enum object_type type = sha1_object_info(sha1, NULL);
1871 if (type < 0)
1872 die("%s is not a valid object", sha1_to_hex(sha1));
1873 if (type != expect)
1874 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1875 typename(expect));
1878 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1879 struct strbuf *path,
1880 each_loose_object_fn obj_cb,
1881 each_loose_cruft_fn cruft_cb,
1882 each_loose_subdir_fn subdir_cb,
1883 void *data)
1885 size_t origlen, baselen;
1886 DIR *dir;
1887 struct dirent *de;
1888 int r = 0;
1890 if (subdir_nr > 0xff)
1891 BUG("invalid loose object subdirectory: %x", subdir_nr);
1893 origlen = path->len;
1894 strbuf_complete(path, '/');
1895 strbuf_addf(path, "%02x", subdir_nr);
1896 baselen = path->len;
1898 dir = opendir(path->buf);
1899 if (!dir) {
1900 if (errno != ENOENT)
1901 r = error_errno("unable to open %s", path->buf);
1902 strbuf_setlen(path, origlen);
1903 return r;
1906 while ((de = readdir(dir))) {
1907 if (is_dot_or_dotdot(de->d_name))
1908 continue;
1910 strbuf_setlen(path, baselen);
1911 strbuf_addf(path, "/%s", de->d_name);
1913 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
1914 char hex[GIT_MAX_HEXSZ+1];
1915 struct object_id oid;
1917 xsnprintf(hex, sizeof(hex), "%02x%s",
1918 subdir_nr, de->d_name);
1919 if (!get_oid_hex(hex, &oid)) {
1920 if (obj_cb) {
1921 r = obj_cb(&oid, path->buf, data);
1922 if (r)
1923 break;
1925 continue;
1929 if (cruft_cb) {
1930 r = cruft_cb(de->d_name, path->buf, data);
1931 if (r)
1932 break;
1935 closedir(dir);
1937 strbuf_setlen(path, baselen);
1938 if (!r && subdir_cb)
1939 r = subdir_cb(subdir_nr, path->buf, data);
1941 strbuf_setlen(path, origlen);
1943 return r;
1946 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
1947 each_loose_object_fn obj_cb,
1948 each_loose_cruft_fn cruft_cb,
1949 each_loose_subdir_fn subdir_cb,
1950 void *data)
1952 int r = 0;
1953 int i;
1955 for (i = 0; i < 256; i++) {
1956 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
1957 subdir_cb, data);
1958 if (r)
1959 break;
1962 return r;
1965 int for_each_loose_file_in_objdir(const char *path,
1966 each_loose_object_fn obj_cb,
1967 each_loose_cruft_fn cruft_cb,
1968 each_loose_subdir_fn subdir_cb,
1969 void *data)
1971 struct strbuf buf = STRBUF_INIT;
1972 int r;
1974 strbuf_addstr(&buf, path);
1975 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
1976 subdir_cb, data);
1977 strbuf_release(&buf);
1979 return r;
1982 struct loose_alt_odb_data {
1983 each_loose_object_fn *cb;
1984 void *data;
1987 static int loose_from_alt_odb(struct alternate_object_database *alt,
1988 void *vdata)
1990 struct loose_alt_odb_data *data = vdata;
1991 struct strbuf buf = STRBUF_INIT;
1992 int r;
1994 strbuf_addstr(&buf, alt->path);
1995 r = for_each_loose_file_in_objdir_buf(&buf,
1996 data->cb, NULL, NULL,
1997 data->data);
1998 strbuf_release(&buf);
1999 return r;
2002 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2004 struct loose_alt_odb_data alt;
2005 int r;
2007 r = for_each_loose_file_in_objdir(get_object_directory(),
2008 cb, NULL, NULL, data);
2009 if (r)
2010 return r;
2012 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2013 return 0;
2015 alt.cb = cb;
2016 alt.data = data;
2017 return foreach_alt_odb(loose_from_alt_odb, &alt);
2020 static int check_stream_sha1(git_zstream *stream,
2021 const char *hdr,
2022 unsigned long size,
2023 const char *path,
2024 const unsigned char *expected_sha1)
2026 git_SHA_CTX c;
2027 unsigned char real_sha1[GIT_MAX_RAWSZ];
2028 unsigned char buf[4096];
2029 unsigned long total_read;
2030 int status = Z_OK;
2032 git_SHA1_Init(&c);
2033 git_SHA1_Update(&c, hdr, stream->total_out);
2036 * We already read some bytes into hdr, but the ones up to the NUL
2037 * do not count against the object's content size.
2039 total_read = stream->total_out - strlen(hdr) - 1;
2042 * This size comparison must be "<=" to read the final zlib packets;
2043 * see the comment in unpack_sha1_rest for details.
2045 while (total_read <= size &&
2046 (status == Z_OK || status == Z_BUF_ERROR)) {
2047 stream->next_out = buf;
2048 stream->avail_out = sizeof(buf);
2049 if (size - total_read < stream->avail_out)
2050 stream->avail_out = size - total_read;
2051 status = git_inflate(stream, Z_FINISH);
2052 git_SHA1_Update(&c, buf, stream->next_out - buf);
2053 total_read += stream->next_out - buf;
2055 git_inflate_end(stream);
2057 if (status != Z_STREAM_END) {
2058 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2059 return -1;
2061 if (stream->avail_in) {
2062 error("garbage at end of loose object '%s'",
2063 sha1_to_hex(expected_sha1));
2064 return -1;
2067 git_SHA1_Final(real_sha1, &c);
2068 if (hashcmp(expected_sha1, real_sha1)) {
2069 error("sha1 mismatch for %s (expected %s)", path,
2070 sha1_to_hex(expected_sha1));
2071 return -1;
2074 return 0;
2077 int read_loose_object(const char *path,
2078 const unsigned char *expected_sha1,
2079 enum object_type *type,
2080 unsigned long *size,
2081 void **contents)
2083 int ret = -1;
2084 void *map = NULL;
2085 unsigned long mapsize;
2086 git_zstream stream;
2087 char hdr[32];
2089 *contents = NULL;
2091 map = map_sha1_file_1(path, NULL, &mapsize);
2092 if (!map) {
2093 error_errno("unable to mmap %s", path);
2094 goto out;
2097 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2098 error("unable to unpack header of %s", path);
2099 goto out;
2102 *type = parse_sha1_header(hdr, size);
2103 if (*type < 0) {
2104 error("unable to parse header of %s", path);
2105 git_inflate_end(&stream);
2106 goto out;
2109 if (*type == OBJ_BLOB) {
2110 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2111 goto out;
2112 } else {
2113 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2114 if (!*contents) {
2115 error("unable to unpack contents of %s", path);
2116 git_inflate_end(&stream);
2117 goto out;
2119 if (check_sha1_signature(expected_sha1, *contents,
2120 *size, typename(*type))) {
2121 error("sha1 mismatch for %s (expected %s)", path,
2122 sha1_to_hex(expected_sha1));
2123 free(*contents);
2124 goto out;
2128 ret = 0; /* everything checks out */
2130 out:
2131 if (map)
2132 munmap(map, mapsize);
2133 return ret;