sha1_file: improve sha1_file_name() perfs
[alt-git.git] / sha1_file.c
blobd82cf1b4a4d4b2c4a344057c3441e93dcd0d199b
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 void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
256 strbuf_addstr(buf, get_object_directory());
257 strbuf_addch(buf, '/');
258 fill_sha1_path(buf, sha1);
261 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
263 strbuf_setlen(&alt->scratch, alt->base_len);
264 return &alt->scratch;
267 static const char *alt_sha1_path(struct alternate_object_database *alt,
268 const unsigned char *sha1)
270 struct strbuf *buf = alt_scratch_buf(alt);
271 fill_sha1_path(buf, sha1);
272 return buf->buf;
275 struct alternate_object_database *alt_odb_list;
276 static struct alternate_object_database **alt_odb_tail;
279 * Return non-zero iff the path is usable as an alternate object database.
281 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
283 struct alternate_object_database *alt;
285 /* Detect cases where alternate disappeared */
286 if (!is_directory(path->buf)) {
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
289 path->buf);
290 return 0;
294 * Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
297 for (alt = alt_odb_list; alt; alt = alt->next) {
298 if (!fspathcmp(path->buf, alt->path))
299 return 0;
301 if (!fspathcmp(path->buf, normalized_objdir))
302 return 0;
304 return 1;
308 * Prepare alternate object database registry.
310 * The variable alt_odb_list points at the list of struct
311 * alternate_object_database. The elements on this list come from
312 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
313 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
314 * whose contents is similar to that environment variable but can be
315 * LF separated. Its base points at a statically allocated buffer that
316 * contains "/the/directory/corresponding/to/.git/objects/...", while
317 * its name points just after the slash at the end of ".git/objects/"
318 * in the example above, and has enough space to hold 40-byte hex
319 * SHA1, an extra slash for the first level indirection, and the
320 * terminating NUL.
322 static void read_info_alternates(const char * relative_base, int depth);
323 static int link_alt_odb_entry(const char *entry, const char *relative_base,
324 int depth, const char *normalized_objdir)
326 struct alternate_object_database *ent;
327 struct strbuf pathbuf = STRBUF_INIT;
329 if (!is_absolute_path(entry) && relative_base) {
330 strbuf_realpath(&pathbuf, relative_base, 1);
331 strbuf_addch(&pathbuf, '/');
333 strbuf_addstr(&pathbuf, entry);
335 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
336 error("unable to normalize alternate object path: %s",
337 pathbuf.buf);
338 strbuf_release(&pathbuf);
339 return -1;
343 * The trailing slash after the directory name is given by
344 * this function at the end. Remove duplicates.
346 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
347 strbuf_setlen(&pathbuf, pathbuf.len - 1);
349 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
350 strbuf_release(&pathbuf);
351 return -1;
354 ent = alloc_alt_odb(pathbuf.buf);
356 /* add the alternate entry */
357 *alt_odb_tail = ent;
358 alt_odb_tail = &(ent->next);
359 ent->next = NULL;
361 /* recursively add alternates */
362 read_info_alternates(pathbuf.buf, depth + 1);
364 strbuf_release(&pathbuf);
365 return 0;
368 static const char *parse_alt_odb_entry(const char *string,
369 int sep,
370 struct strbuf *out)
372 const char *end;
374 strbuf_reset(out);
376 if (*string == '#') {
377 /* comment; consume up to next separator */
378 end = strchrnul(string, sep);
379 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
381 * quoted path; unquote_c_style has copied the
382 * data for us and set "end". Broken quoting (e.g.,
383 * an entry that doesn't end with a quote) falls
384 * back to the unquoted case below.
386 } else {
387 /* normal, unquoted path */
388 end = strchrnul(string, sep);
389 strbuf_add(out, string, end - string);
392 if (*end)
393 end++;
394 return end;
397 static void link_alt_odb_entries(const char *alt, int sep,
398 const char *relative_base, int depth)
400 struct strbuf objdirbuf = STRBUF_INIT;
401 struct strbuf entry = STRBUF_INIT;
403 if (!alt || !*alt)
404 return;
406 if (depth > 5) {
407 error("%s: ignoring alternate object stores, nesting too deep.",
408 relative_base);
409 return;
412 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
413 if (strbuf_normalize_path(&objdirbuf) < 0)
414 die("unable to normalize object directory: %s",
415 objdirbuf.buf);
417 while (*alt) {
418 alt = parse_alt_odb_entry(alt, sep, &entry);
419 if (!entry.len)
420 continue;
421 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
423 strbuf_release(&entry);
424 strbuf_release(&objdirbuf);
427 static void read_info_alternates(const char * relative_base, int depth)
429 char *path;
430 struct strbuf buf = STRBUF_INIT;
432 path = xstrfmt("%s/info/alternates", relative_base);
433 if (strbuf_read_file(&buf, path, 1024) < 0) {
434 warn_on_fopen_errors(path);
435 free(path);
436 return;
439 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
440 strbuf_release(&buf);
441 free(path);
444 struct alternate_object_database *alloc_alt_odb(const char *dir)
446 struct alternate_object_database *ent;
448 FLEX_ALLOC_STR(ent, path, dir);
449 strbuf_init(&ent->scratch, 0);
450 strbuf_addf(&ent->scratch, "%s/", dir);
451 ent->base_len = ent->scratch.len;
453 return ent;
456 void add_to_alternates_file(const char *reference)
458 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
459 char *alts = git_pathdup("objects/info/alternates");
460 FILE *in, *out;
462 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
463 out = fdopen_lock_file(lock, "w");
464 if (!out)
465 die_errno("unable to fdopen alternates lockfile");
467 in = fopen(alts, "r");
468 if (in) {
469 struct strbuf line = STRBUF_INIT;
470 int found = 0;
472 while (strbuf_getline(&line, in) != EOF) {
473 if (!strcmp(reference, line.buf)) {
474 found = 1;
475 break;
477 fprintf_or_die(out, "%s\n", line.buf);
480 strbuf_release(&line);
481 fclose(in);
483 if (found) {
484 rollback_lock_file(lock);
485 lock = NULL;
488 else if (errno != ENOENT)
489 die_errno("unable to read alternates file");
491 if (lock) {
492 fprintf_or_die(out, "%s\n", reference);
493 if (commit_lock_file(lock))
494 die_errno("unable to move new alternates file into place");
495 if (alt_odb_tail)
496 link_alt_odb_entries(reference, '\n', NULL, 0);
498 free(alts);
501 void add_to_alternates_memory(const char *reference)
504 * Make sure alternates are initialized, or else our entry may be
505 * overwritten when they are.
507 prepare_alt_odb();
509 link_alt_odb_entries(reference, '\n', NULL, 0);
513 * Compute the exact path an alternate is at and returns it. In case of
514 * error NULL is returned and the human readable error is added to `err`
515 * `path` may be relative and should point to $GITDIR.
516 * `err` must not be null.
518 char *compute_alternate_path(const char *path, struct strbuf *err)
520 char *ref_git = NULL;
521 const char *repo, *ref_git_s;
522 int seen_error = 0;
524 ref_git_s = real_path_if_valid(path);
525 if (!ref_git_s) {
526 seen_error = 1;
527 strbuf_addf(err, _("path '%s' does not exist"), path);
528 goto out;
529 } else
531 * Beware: read_gitfile(), real_path() and mkpath()
532 * return static buffer
534 ref_git = xstrdup(ref_git_s);
536 repo = read_gitfile(ref_git);
537 if (!repo)
538 repo = read_gitfile(mkpath("%s/.git", ref_git));
539 if (repo) {
540 free(ref_git);
541 ref_git = xstrdup(repo);
544 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
545 char *ref_git_git = mkpathdup("%s/.git", ref_git);
546 free(ref_git);
547 ref_git = ref_git_git;
548 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
549 struct strbuf sb = STRBUF_INIT;
550 seen_error = 1;
551 if (get_common_dir(&sb, ref_git)) {
552 strbuf_addf(err,
553 _("reference repository '%s' as a linked "
554 "checkout is not supported yet."),
555 path);
556 goto out;
559 strbuf_addf(err, _("reference repository '%s' is not a "
560 "local repository."), path);
561 goto out;
564 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
565 strbuf_addf(err, _("reference repository '%s' is shallow"),
566 path);
567 seen_error = 1;
568 goto out;
571 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
572 strbuf_addf(err,
573 _("reference repository '%s' is grafted"),
574 path);
575 seen_error = 1;
576 goto out;
579 out:
580 if (seen_error) {
581 FREE_AND_NULL(ref_git);
584 return ref_git;
587 int foreach_alt_odb(alt_odb_fn fn, void *cb)
589 struct alternate_object_database *ent;
590 int r = 0;
592 prepare_alt_odb();
593 for (ent = alt_odb_list; ent; ent = ent->next) {
594 r = fn(ent, cb);
595 if (r)
596 break;
598 return r;
601 void prepare_alt_odb(void)
603 const char *alt;
605 if (alt_odb_tail)
606 return;
608 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
610 alt_odb_tail = &alt_odb_list;
611 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
613 read_info_alternates(get_object_directory(), 0);
616 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
617 static int freshen_file(const char *fn)
619 struct utimbuf t;
620 t.actime = t.modtime = time(NULL);
621 return !utime(fn, &t);
625 * All of the check_and_freshen functions return 1 if the file exists and was
626 * freshened (if freshening was requested), 0 otherwise. If they return
627 * 0, you should not assume that it is safe to skip a write of the object (it
628 * either does not exist on disk, or has a stale mtime and may be subject to
629 * pruning).
631 int check_and_freshen_file(const char *fn, int freshen)
633 if (access(fn, F_OK))
634 return 0;
635 if (freshen && !freshen_file(fn))
636 return 0;
637 return 1;
640 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
642 static struct strbuf buf = STRBUF_INIT;
644 strbuf_reset(&buf);
645 sha1_file_name(&buf, sha1);
647 return check_and_freshen_file(buf.buf, freshen);
650 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
652 struct alternate_object_database *alt;
653 prepare_alt_odb();
654 for (alt = alt_odb_list; alt; alt = alt->next) {
655 const char *path = alt_sha1_path(alt, sha1);
656 if (check_and_freshen_file(path, freshen))
657 return 1;
659 return 0;
662 static int check_and_freshen(const unsigned char *sha1, int freshen)
664 return check_and_freshen_local(sha1, freshen) ||
665 check_and_freshen_nonlocal(sha1, freshen);
668 int has_loose_object_nonlocal(const unsigned char *sha1)
670 return check_and_freshen_nonlocal(sha1, 0);
673 static int has_loose_object(const unsigned char *sha1)
675 return check_and_freshen(sha1, 0);
678 static void mmap_limit_check(size_t length)
680 static size_t limit = 0;
681 if (!limit) {
682 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
683 if (!limit)
684 limit = SIZE_MAX;
686 if (length > limit)
687 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
688 (uintmax_t)length, (uintmax_t)limit);
691 void *xmmap_gently(void *start, size_t length,
692 int prot, int flags, int fd, off_t offset)
694 void *ret;
696 mmap_limit_check(length);
697 ret = mmap(start, length, prot, flags, fd, offset);
698 if (ret == MAP_FAILED) {
699 if (!length)
700 return NULL;
701 release_pack_memory(length);
702 ret = mmap(start, length, prot, flags, fd, offset);
704 return ret;
707 void *xmmap(void *start, size_t length,
708 int prot, int flags, int fd, off_t offset)
710 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
711 if (ret == MAP_FAILED)
712 die_errno("mmap failed");
713 return ret;
717 * With an in-core object data in "map", rehash it to make sure the
718 * object name actually matches "sha1" to detect object corruption.
719 * With "map" == NULL, try reading the object named with "sha1" using
720 * the streaming interface and rehash it to do the same.
722 int check_sha1_signature(const unsigned char *sha1, void *map,
723 unsigned long size, const char *type)
725 unsigned char real_sha1[20];
726 enum object_type obj_type;
727 struct git_istream *st;
728 git_SHA_CTX c;
729 char hdr[32];
730 int hdrlen;
732 if (map) {
733 hash_sha1_file(map, size, type, real_sha1);
734 return hashcmp(sha1, real_sha1) ? -1 : 0;
737 st = open_istream(sha1, &obj_type, &size, NULL);
738 if (!st)
739 return -1;
741 /* Generate the header */
742 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
744 /* Sha1.. */
745 git_SHA1_Init(&c);
746 git_SHA1_Update(&c, hdr, hdrlen);
747 for (;;) {
748 char buf[1024 * 16];
749 ssize_t readlen = read_istream(st, buf, sizeof(buf));
751 if (readlen < 0) {
752 close_istream(st);
753 return -1;
755 if (!readlen)
756 break;
757 git_SHA1_Update(&c, buf, readlen);
759 git_SHA1_Final(real_sha1, &c);
760 close_istream(st);
761 return hashcmp(sha1, real_sha1) ? -1 : 0;
764 int git_open_cloexec(const char *name, int flags)
766 int fd;
767 static int o_cloexec = O_CLOEXEC;
769 fd = open(name, flags | o_cloexec);
770 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
771 /* Try again w/o O_CLOEXEC: the kernel might not support it */
772 o_cloexec &= ~O_CLOEXEC;
773 fd = open(name, flags | o_cloexec);
776 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
778 static int fd_cloexec = FD_CLOEXEC;
780 if (!o_cloexec && 0 <= fd && fd_cloexec) {
781 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
782 int flags = fcntl(fd, F_GETFD);
783 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
784 fd_cloexec = 0;
787 #endif
788 return fd;
792 * Find "sha1" as a loose object in the local repository or in an alternate.
793 * Returns 0 on success, negative on failure.
795 * The "path" out-parameter will give the path of the object we found (if any).
796 * Note that it may point to static storage and is only valid until another
797 * call to sha1_file_name(), etc.
799 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
800 const char **path)
802 struct alternate_object_database *alt;
803 static struct strbuf buf = STRBUF_INIT;
805 strbuf_reset(&buf);
806 sha1_file_name(&buf, sha1);
807 *path = buf.buf;
809 if (!lstat(*path, st))
810 return 0;
812 prepare_alt_odb();
813 errno = ENOENT;
814 for (alt = alt_odb_list; alt; alt = alt->next) {
815 *path = alt_sha1_path(alt, sha1);
816 if (!lstat(*path, st))
817 return 0;
820 return -1;
824 * Like stat_sha1_file(), but actually open the object and return the
825 * descriptor. See the caveats on the "path" parameter above.
827 static int open_sha1_file(const unsigned char *sha1, const char **path)
829 int fd;
830 struct alternate_object_database *alt;
831 int most_interesting_errno;
832 static struct strbuf buf = STRBUF_INIT;
834 strbuf_reset(&buf);
835 sha1_file_name(&buf, sha1);
836 *path = buf.buf;
838 fd = git_open(*path);
839 if (fd >= 0)
840 return fd;
841 most_interesting_errno = errno;
843 prepare_alt_odb();
844 for (alt = alt_odb_list; alt; alt = alt->next) {
845 *path = alt_sha1_path(alt, sha1);
846 fd = git_open(*path);
847 if (fd >= 0)
848 return fd;
849 if (most_interesting_errno == ENOENT)
850 most_interesting_errno = errno;
852 errno = most_interesting_errno;
853 return -1;
857 * Map the loose object at "path" if it is not NULL, or the path found by
858 * searching for a loose object named "sha1".
860 static void *map_sha1_file_1(const char *path,
861 const unsigned char *sha1,
862 unsigned long *size)
864 void *map;
865 int fd;
867 if (path)
868 fd = git_open(path);
869 else
870 fd = open_sha1_file(sha1, &path);
871 map = NULL;
872 if (fd >= 0) {
873 struct stat st;
875 if (!fstat(fd, &st)) {
876 *size = xsize_t(st.st_size);
877 if (!*size) {
878 /* mmap() is forbidden on empty files */
879 error("object file %s is empty", path);
880 return NULL;
882 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
884 close(fd);
886 return map;
889 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
891 return map_sha1_file_1(NULL, sha1, size);
894 static int unpack_sha1_short_header(git_zstream *stream,
895 unsigned char *map, unsigned long mapsize,
896 void *buffer, unsigned long bufsiz)
898 /* Get the data stream */
899 memset(stream, 0, sizeof(*stream));
900 stream->next_in = map;
901 stream->avail_in = mapsize;
902 stream->next_out = buffer;
903 stream->avail_out = bufsiz;
905 git_inflate_init(stream);
906 return git_inflate(stream, 0);
909 int unpack_sha1_header(git_zstream *stream,
910 unsigned char *map, unsigned long mapsize,
911 void *buffer, unsigned long bufsiz)
913 int status = unpack_sha1_short_header(stream, map, mapsize,
914 buffer, bufsiz);
916 if (status < Z_OK)
917 return status;
919 /* Make sure we have the terminating NUL */
920 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
921 return -1;
922 return 0;
925 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
926 unsigned long mapsize, void *buffer,
927 unsigned long bufsiz, struct strbuf *header)
929 int status;
931 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
932 if (status < Z_OK)
933 return -1;
936 * Check if entire header is unpacked in the first iteration.
938 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
939 return 0;
942 * buffer[0..bufsiz] was not large enough. Copy the partial
943 * result out to header, and then append the result of further
944 * reading the stream.
946 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
947 stream->next_out = buffer;
948 stream->avail_out = bufsiz;
950 do {
951 status = git_inflate(stream, 0);
952 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
953 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
954 return 0;
955 stream->next_out = buffer;
956 stream->avail_out = bufsiz;
957 } while (status != Z_STREAM_END);
958 return -1;
961 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
963 int bytes = strlen(buffer) + 1;
964 unsigned char *buf = xmallocz(size);
965 unsigned long n;
966 int status = Z_OK;
968 n = stream->total_out - bytes;
969 if (n > size)
970 n = size;
971 memcpy(buf, (char *) buffer + bytes, n);
972 bytes = n;
973 if (bytes <= size) {
975 * The above condition must be (bytes <= size), not
976 * (bytes < size). In other words, even though we
977 * expect no more output and set avail_out to zero,
978 * the input zlib stream may have bytes that express
979 * "this concludes the stream", and we *do* want to
980 * eat that input.
982 * Otherwise we would not be able to test that we
983 * consumed all the input to reach the expected size;
984 * we also want to check that zlib tells us that all
985 * went well with status == Z_STREAM_END at the end.
987 stream->next_out = buf + bytes;
988 stream->avail_out = size - bytes;
989 while (status == Z_OK)
990 status = git_inflate(stream, Z_FINISH);
992 if (status == Z_STREAM_END && !stream->avail_in) {
993 git_inflate_end(stream);
994 return buf;
997 if (status < 0)
998 error("corrupt loose object '%s'", sha1_to_hex(sha1));
999 else if (stream->avail_in)
1000 error("garbage at end of loose object '%s'",
1001 sha1_to_hex(sha1));
1002 free(buf);
1003 return NULL;
1007 * We used to just use "sscanf()", but that's actually way
1008 * too permissive for what we want to check. So do an anal
1009 * object header parse by hand.
1011 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1012 unsigned int flags)
1014 const char *type_buf = hdr;
1015 unsigned long size;
1016 int type, type_len = 0;
1019 * The type can be of any size but is followed by
1020 * a space.
1022 for (;;) {
1023 char c = *hdr++;
1024 if (!c)
1025 return -1;
1026 if (c == ' ')
1027 break;
1028 type_len++;
1031 type = type_from_string_gently(type_buf, type_len, 1);
1032 if (oi->typename)
1033 strbuf_add(oi->typename, type_buf, type_len);
1035 * Set type to 0 if its an unknown object and
1036 * we're obtaining the type using '--allow-unknown-type'
1037 * option.
1039 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1040 type = 0;
1041 else if (type < 0)
1042 die("invalid object type");
1043 if (oi->typep)
1044 *oi->typep = type;
1047 * The length must follow immediately, and be in canonical
1048 * decimal format (ie "010" is not valid).
1050 size = *hdr++ - '0';
1051 if (size > 9)
1052 return -1;
1053 if (size) {
1054 for (;;) {
1055 unsigned long c = *hdr - '0';
1056 if (c > 9)
1057 break;
1058 hdr++;
1059 size = size * 10 + c;
1063 if (oi->sizep)
1064 *oi->sizep = size;
1067 * The length must be followed by a zero byte
1069 return *hdr ? -1 : type;
1072 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1074 struct object_info oi = OBJECT_INFO_INIT;
1076 oi.sizep = sizep;
1077 return parse_sha1_header_extended(hdr, &oi, 0);
1080 static int sha1_loose_object_info(const unsigned char *sha1,
1081 struct object_info *oi,
1082 int flags)
1084 int status = 0;
1085 unsigned long mapsize;
1086 void *map;
1087 git_zstream stream;
1088 char hdr[32];
1089 struct strbuf hdrbuf = STRBUF_INIT;
1090 unsigned long size_scratch;
1092 if (oi->delta_base_sha1)
1093 hashclr(oi->delta_base_sha1);
1096 * If we don't care about type or size, then we don't
1097 * need to look inside the object at all. Note that we
1098 * do not optimize out the stat call, even if the
1099 * caller doesn't care about the disk-size, since our
1100 * return value implicitly indicates whether the
1101 * object even exists.
1103 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1104 const char *path;
1105 struct stat st;
1106 if (stat_sha1_file(sha1, &st, &path) < 0)
1107 return -1;
1108 if (oi->disk_sizep)
1109 *oi->disk_sizep = st.st_size;
1110 return 0;
1113 map = map_sha1_file(sha1, &mapsize);
1114 if (!map)
1115 return -1;
1117 if (!oi->sizep)
1118 oi->sizep = &size_scratch;
1120 if (oi->disk_sizep)
1121 *oi->disk_sizep = mapsize;
1122 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1123 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1124 status = error("unable to unpack %s header with --allow-unknown-type",
1125 sha1_to_hex(sha1));
1126 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1127 status = error("unable to unpack %s header",
1128 sha1_to_hex(sha1));
1129 if (status < 0)
1130 ; /* Do nothing */
1131 else if (hdrbuf.len) {
1132 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1133 status = error("unable to parse %s header with --allow-unknown-type",
1134 sha1_to_hex(sha1));
1135 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1136 status = error("unable to parse %s header", sha1_to_hex(sha1));
1138 if (status >= 0 && oi->contentp) {
1139 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1140 *oi->sizep, sha1);
1141 if (!*oi->contentp) {
1142 git_inflate_end(&stream);
1143 status = -1;
1145 } else
1146 git_inflate_end(&stream);
1148 munmap(map, mapsize);
1149 if (status && oi->typep)
1150 *oi->typep = status;
1151 if (oi->sizep == &size_scratch)
1152 oi->sizep = NULL;
1153 strbuf_release(&hdrbuf);
1154 oi->whence = OI_LOOSE;
1155 return (status < 0) ? status : 0;
1158 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1160 static struct object_info blank_oi = OBJECT_INFO_INIT;
1161 struct pack_entry e;
1162 int rtype;
1163 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1164 lookup_replace_object(sha1) :
1165 sha1;
1167 if (!oi)
1168 oi = &blank_oi;
1170 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1171 struct cached_object *co = find_cached_object(real);
1172 if (co) {
1173 if (oi->typep)
1174 *(oi->typep) = co->type;
1175 if (oi->sizep)
1176 *(oi->sizep) = co->size;
1177 if (oi->disk_sizep)
1178 *(oi->disk_sizep) = 0;
1179 if (oi->delta_base_sha1)
1180 hashclr(oi->delta_base_sha1);
1181 if (oi->typename)
1182 strbuf_addstr(oi->typename, typename(co->type));
1183 if (oi->contentp)
1184 *oi->contentp = xmemdupz(co->buf, co->size);
1185 oi->whence = OI_CACHED;
1186 return 0;
1190 if (!find_pack_entry(real, &e)) {
1191 /* Most likely it's a loose object. */
1192 if (!sha1_loose_object_info(real, oi, flags))
1193 return 0;
1195 /* Not a loose object; someone else may have just packed it. */
1196 if (flags & OBJECT_INFO_QUICK) {
1197 return -1;
1198 } else {
1199 reprepare_packed_git();
1200 if (!find_pack_entry(real, &e))
1201 return -1;
1205 if (oi == &blank_oi)
1207 * We know that the caller doesn't actually need the
1208 * information below, so return early.
1210 return 0;
1212 rtype = packed_object_info(e.p, e.offset, oi);
1213 if (rtype < 0) {
1214 mark_bad_packed_object(e.p, real);
1215 return sha1_object_info_extended(real, oi, 0);
1216 } else if (oi->whence == OI_PACKED) {
1217 oi->u.packed.offset = e.offset;
1218 oi->u.packed.pack = e.p;
1219 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1220 rtype == OBJ_OFS_DELTA);
1223 return 0;
1226 /* returns enum object_type or negative */
1227 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1229 enum object_type type;
1230 struct object_info oi = OBJECT_INFO_INIT;
1232 oi.typep = &type;
1233 oi.sizep = sizep;
1234 if (sha1_object_info_extended(sha1, &oi,
1235 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1236 return -1;
1237 return type;
1240 static void *read_object(const unsigned char *sha1, enum object_type *type,
1241 unsigned long *size)
1243 struct object_info oi = OBJECT_INFO_INIT;
1244 void *content;
1245 oi.typep = type;
1246 oi.sizep = size;
1247 oi.contentp = &content;
1249 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1250 return NULL;
1251 return content;
1254 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1255 unsigned char *sha1)
1257 struct cached_object *co;
1259 hash_sha1_file(buf, len, typename(type), sha1);
1260 if (has_sha1_file(sha1) || find_cached_object(sha1))
1261 return 0;
1262 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1263 co = &cached_objects[cached_object_nr++];
1264 co->size = len;
1265 co->type = type;
1266 co->buf = xmalloc(len);
1267 memcpy(co->buf, buf, len);
1268 hashcpy(co->sha1, sha1);
1269 return 0;
1273 * This function dies on corrupt objects; the callers who want to
1274 * deal with them should arrange to call read_object() and give error
1275 * messages themselves.
1277 void *read_sha1_file_extended(const unsigned char *sha1,
1278 enum object_type *type,
1279 unsigned long *size,
1280 int lookup_replace)
1282 void *data;
1283 const struct packed_git *p;
1284 const char *path;
1285 struct stat st;
1286 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1287 : sha1;
1289 errno = 0;
1290 data = read_object(repl, type, size);
1291 if (data)
1292 return data;
1294 if (errno && errno != ENOENT)
1295 die_errno("failed to read object %s", sha1_to_hex(sha1));
1297 /* die if we replaced an object with one that does not exist */
1298 if (repl != sha1)
1299 die("replacement %s not found for %s",
1300 sha1_to_hex(repl), sha1_to_hex(sha1));
1302 if (!stat_sha1_file(repl, &st, &path))
1303 die("loose object %s (stored in %s) is corrupt",
1304 sha1_to_hex(repl), path);
1306 if ((p = has_packed_and_bad(repl)) != NULL)
1307 die("packed object %s (stored in %s) is corrupt",
1308 sha1_to_hex(repl), p->pack_name);
1310 return NULL;
1313 void *read_object_with_reference(const unsigned char *sha1,
1314 const char *required_type_name,
1315 unsigned long *size,
1316 unsigned char *actual_sha1_return)
1318 enum object_type type, required_type;
1319 void *buffer;
1320 unsigned long isize;
1321 unsigned char actual_sha1[20];
1323 required_type = type_from_string(required_type_name);
1324 hashcpy(actual_sha1, sha1);
1325 while (1) {
1326 int ref_length = -1;
1327 const char *ref_type = NULL;
1329 buffer = read_sha1_file(actual_sha1, &type, &isize);
1330 if (!buffer)
1331 return NULL;
1332 if (type == required_type) {
1333 *size = isize;
1334 if (actual_sha1_return)
1335 hashcpy(actual_sha1_return, actual_sha1);
1336 return buffer;
1338 /* Handle references */
1339 else if (type == OBJ_COMMIT)
1340 ref_type = "tree ";
1341 else if (type == OBJ_TAG)
1342 ref_type = "object ";
1343 else {
1344 free(buffer);
1345 return NULL;
1347 ref_length = strlen(ref_type);
1349 if (ref_length + 40 > isize ||
1350 memcmp(buffer, ref_type, ref_length) ||
1351 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1352 free(buffer);
1353 return NULL;
1355 free(buffer);
1356 /* Now we have the ID of the referred-to object in
1357 * actual_sha1. Check again. */
1361 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1362 const char *type, unsigned char *sha1,
1363 char *hdr, int *hdrlen)
1365 git_SHA_CTX c;
1367 /* Generate the header */
1368 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1370 /* Sha1.. */
1371 git_SHA1_Init(&c);
1372 git_SHA1_Update(&c, hdr, *hdrlen);
1373 git_SHA1_Update(&c, buf, len);
1374 git_SHA1_Final(sha1, &c);
1378 * Move the just written object into its final resting place.
1380 int finalize_object_file(const char *tmpfile, const char *filename)
1382 int ret = 0;
1384 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1385 goto try_rename;
1386 else if (link(tmpfile, filename))
1387 ret = errno;
1390 * Coda hack - coda doesn't like cross-directory links,
1391 * so we fall back to a rename, which will mean that it
1392 * won't be able to check collisions, but that's not a
1393 * big deal.
1395 * The same holds for FAT formatted media.
1397 * When this succeeds, we just return. We have nothing
1398 * left to unlink.
1400 if (ret && ret != EEXIST) {
1401 try_rename:
1402 if (!rename(tmpfile, filename))
1403 goto out;
1404 ret = errno;
1406 unlink_or_warn(tmpfile);
1407 if (ret) {
1408 if (ret != EEXIST) {
1409 return error_errno("unable to write sha1 filename %s", filename);
1411 /* FIXME!!! Collision check here ? */
1414 out:
1415 if (adjust_shared_perm(filename))
1416 return error("unable to set permission to '%s'", filename);
1417 return 0;
1420 static int write_buffer(int fd, const void *buf, size_t len)
1422 if (write_in_full(fd, buf, len) < 0)
1423 return error_errno("file write error");
1424 return 0;
1427 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1428 unsigned char *sha1)
1430 char hdr[32];
1431 int hdrlen = sizeof(hdr);
1432 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1433 return 0;
1436 /* Finalize a file on disk, and close it. */
1437 static void close_sha1_file(int fd)
1439 if (fsync_object_files)
1440 fsync_or_die(fd, "sha1 file");
1441 if (close(fd) != 0)
1442 die_errno("error when closing sha1 file");
1445 /* Size of directory component, including the ending '/' */
1446 static inline int directory_size(const char *filename)
1448 const char *s = strrchr(filename, '/');
1449 if (!s)
1450 return 0;
1451 return s - filename + 1;
1455 * This creates a temporary file in the same directory as the final
1456 * 'filename'
1458 * We want to avoid cross-directory filename renames, because those
1459 * can have problems on various filesystems (FAT, NFS, Coda).
1461 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1463 int fd, dirlen = directory_size(filename);
1465 strbuf_reset(tmp);
1466 strbuf_add(tmp, filename, dirlen);
1467 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1468 fd = git_mkstemp_mode(tmp->buf, 0444);
1469 if (fd < 0 && dirlen && errno == ENOENT) {
1471 * Make sure the directory exists; note that the contents
1472 * of the buffer are undefined after mkstemp returns an
1473 * error, so we have to rewrite the whole buffer from
1474 * scratch.
1476 strbuf_reset(tmp);
1477 strbuf_add(tmp, filename, dirlen - 1);
1478 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1479 return -1;
1480 if (adjust_shared_perm(tmp->buf))
1481 return -1;
1483 /* Try again */
1484 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1485 fd = git_mkstemp_mode(tmp->buf, 0444);
1487 return fd;
1490 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1491 const void *buf, unsigned long len, time_t mtime)
1493 int fd, ret;
1494 unsigned char compressed[4096];
1495 git_zstream stream;
1496 git_SHA_CTX c;
1497 unsigned char parano_sha1[20];
1498 static struct strbuf tmp_file = STRBUF_INIT;
1499 static struct strbuf filename = STRBUF_INIT;
1501 strbuf_reset(&filename);
1502 sha1_file_name(&filename, sha1);
1504 fd = create_tmpfile(&tmp_file, filename.buf);
1505 if (fd < 0) {
1506 if (errno == EACCES)
1507 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1508 else
1509 return error_errno("unable to create temporary file");
1512 /* Set it up */
1513 git_deflate_init(&stream, zlib_compression_level);
1514 stream.next_out = compressed;
1515 stream.avail_out = sizeof(compressed);
1516 git_SHA1_Init(&c);
1518 /* First header.. */
1519 stream.next_in = (unsigned char *)hdr;
1520 stream.avail_in = hdrlen;
1521 while (git_deflate(&stream, 0) == Z_OK)
1522 ; /* nothing */
1523 git_SHA1_Update(&c, hdr, hdrlen);
1525 /* Then the data itself.. */
1526 stream.next_in = (void *)buf;
1527 stream.avail_in = len;
1528 do {
1529 unsigned char *in0 = stream.next_in;
1530 ret = git_deflate(&stream, Z_FINISH);
1531 git_SHA1_Update(&c, in0, stream.next_in - in0);
1532 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1533 die("unable to write sha1 file");
1534 stream.next_out = compressed;
1535 stream.avail_out = sizeof(compressed);
1536 } while (ret == Z_OK);
1538 if (ret != Z_STREAM_END)
1539 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1540 ret = git_deflate_end_gently(&stream);
1541 if (ret != Z_OK)
1542 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1543 git_SHA1_Final(parano_sha1, &c);
1544 if (hashcmp(sha1, parano_sha1) != 0)
1545 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1547 close_sha1_file(fd);
1549 if (mtime) {
1550 struct utimbuf utb;
1551 utb.actime = mtime;
1552 utb.modtime = mtime;
1553 if (utime(tmp_file.buf, &utb) < 0)
1554 warning_errno("failed utime() on %s", tmp_file.buf);
1557 return finalize_object_file(tmp_file.buf, filename.buf);
1560 static int freshen_loose_object(const unsigned char *sha1)
1562 return check_and_freshen(sha1, 1);
1565 static int freshen_packed_object(const unsigned char *sha1)
1567 struct pack_entry e;
1568 if (!find_pack_entry(sha1, &e))
1569 return 0;
1570 if (e.p->freshened)
1571 return 1;
1572 if (!freshen_file(e.p->pack_name))
1573 return 0;
1574 e.p->freshened = 1;
1575 return 1;
1578 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1580 char hdr[32];
1581 int hdrlen = sizeof(hdr);
1583 /* Normally if we have it in the pack then we do not bother writing
1584 * it out into .git/objects/??/?{38} file.
1586 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1587 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1588 return 0;
1589 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1592 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1593 struct object_id *oid, unsigned flags)
1595 char *header;
1596 int hdrlen, status = 0;
1598 /* type string, SP, %lu of the length plus NUL must fit this */
1599 hdrlen = strlen(type) + 32;
1600 header = xmalloc(hdrlen);
1601 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1603 if (!(flags & HASH_WRITE_OBJECT))
1604 goto cleanup;
1605 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1606 goto cleanup;
1607 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1609 cleanup:
1610 free(header);
1611 return status;
1614 int force_object_loose(const unsigned char *sha1, time_t mtime)
1616 void *buf;
1617 unsigned long len;
1618 enum object_type type;
1619 char hdr[32];
1620 int hdrlen;
1621 int ret;
1623 if (has_loose_object(sha1))
1624 return 0;
1625 buf = read_object(sha1, &type, &len);
1626 if (!buf)
1627 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1628 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1629 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1630 free(buf);
1632 return ret;
1635 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1637 if (!startup_info->have_repository)
1638 return 0;
1639 return sha1_object_info_extended(sha1, NULL,
1640 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1643 int has_object_file(const struct object_id *oid)
1645 return has_sha1_file(oid->hash);
1648 int has_object_file_with_flags(const struct object_id *oid, int flags)
1650 return has_sha1_file_with_flags(oid->hash, flags);
1653 static void check_tree(const void *buf, size_t size)
1655 struct tree_desc desc;
1656 struct name_entry entry;
1658 init_tree_desc(&desc, buf, size);
1659 while (tree_entry(&desc, &entry))
1660 /* do nothing
1661 * tree_entry() will die() on malformed entries */
1665 static void check_commit(const void *buf, size_t size)
1667 struct commit c;
1668 memset(&c, 0, sizeof(c));
1669 if (parse_commit_buffer(&c, buf, size))
1670 die("corrupt commit");
1673 static void check_tag(const void *buf, size_t size)
1675 struct tag t;
1676 memset(&t, 0, sizeof(t));
1677 if (parse_tag_buffer(&t, buf, size))
1678 die("corrupt tag");
1681 static int index_mem(unsigned char *sha1, void *buf, size_t size,
1682 enum object_type type,
1683 const char *path, unsigned flags)
1685 int ret, re_allocated = 0;
1686 int write_object = flags & HASH_WRITE_OBJECT;
1688 if (!type)
1689 type = OBJ_BLOB;
1692 * Convert blobs to git internal format
1694 if ((type == OBJ_BLOB) && path) {
1695 struct strbuf nbuf = STRBUF_INIT;
1696 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1697 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1698 buf = strbuf_detach(&nbuf, &size);
1699 re_allocated = 1;
1702 if (flags & HASH_FORMAT_CHECK) {
1703 if (type == OBJ_TREE)
1704 check_tree(buf, size);
1705 if (type == OBJ_COMMIT)
1706 check_commit(buf, size);
1707 if (type == OBJ_TAG)
1708 check_tag(buf, size);
1711 if (write_object)
1712 ret = write_sha1_file(buf, size, typename(type), sha1);
1713 else
1714 ret = hash_sha1_file(buf, size, typename(type), sha1);
1715 if (re_allocated)
1716 free(buf);
1717 return ret;
1720 static int index_stream_convert_blob(unsigned char *sha1, int fd,
1721 const char *path, unsigned flags)
1723 int ret;
1724 const int write_object = flags & HASH_WRITE_OBJECT;
1725 struct strbuf sbuf = STRBUF_INIT;
1727 assert(path);
1728 assert(would_convert_to_git_filter_fd(path));
1730 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1731 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1733 if (write_object)
1734 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1735 sha1);
1736 else
1737 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1738 sha1);
1739 strbuf_release(&sbuf);
1740 return ret;
1743 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
1744 const char *path, unsigned flags)
1746 struct strbuf sbuf = STRBUF_INIT;
1747 int ret;
1749 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1750 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
1751 else
1752 ret = -1;
1753 strbuf_release(&sbuf);
1754 return ret;
1757 #define SMALL_FILE_SIZE (32*1024)
1759 static int index_core(unsigned char *sha1, int fd, size_t size,
1760 enum object_type type, const char *path,
1761 unsigned flags)
1763 int ret;
1765 if (!size) {
1766 ret = index_mem(sha1, "", size, type, path, flags);
1767 } else if (size <= SMALL_FILE_SIZE) {
1768 char *buf = xmalloc(size);
1769 ssize_t read_result = read_in_full(fd, buf, size);
1770 if (read_result < 0)
1771 ret = error_errno("read error while indexing %s",
1772 path ? path : "<unknown>");
1773 else if (read_result != size)
1774 ret = error("short read while indexing %s",
1775 path ? path : "<unknown>");
1776 else
1777 ret = index_mem(sha1, buf, size, type, path, flags);
1778 free(buf);
1779 } else {
1780 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1781 ret = index_mem(sha1, buf, size, type, path, flags);
1782 munmap(buf, size);
1784 return ret;
1788 * This creates one packfile per large blob unless bulk-checkin
1789 * machinery is "plugged".
1791 * This also bypasses the usual "convert-to-git" dance, and that is on
1792 * purpose. We could write a streaming version of the converting
1793 * functions and insert that before feeding the data to fast-import
1794 * (or equivalent in-core API described above). However, that is
1795 * somewhat complicated, as we do not know the size of the filter
1796 * result, which we need to know beforehand when writing a git object.
1797 * Since the primary motivation for trying to stream from the working
1798 * tree file and to avoid mmaping it in core is to deal with large
1799 * binary blobs, they generally do not want to get any conversion, and
1800 * callers should avoid this code path when filters are requested.
1802 static int index_stream(struct object_id *oid, int fd, size_t size,
1803 enum object_type type, const char *path,
1804 unsigned flags)
1806 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1809 int index_fd(struct object_id *oid, int fd, struct stat *st,
1810 enum object_type type, const char *path, unsigned flags)
1812 int ret;
1815 * Call xsize_t() only when needed to avoid potentially unnecessary
1816 * die() for large files.
1818 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1819 ret = index_stream_convert_blob(oid->hash, fd, path, flags);
1820 else if (!S_ISREG(st->st_mode))
1821 ret = index_pipe(oid->hash, fd, type, path, flags);
1822 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1823 (path && would_convert_to_git(&the_index, path)))
1824 ret = index_core(oid->hash, fd, xsize_t(st->st_size), type, path,
1825 flags);
1826 else
1827 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1828 flags);
1829 close(fd);
1830 return ret;
1833 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1835 int fd;
1836 struct strbuf sb = STRBUF_INIT;
1837 int rc = 0;
1839 switch (st->st_mode & S_IFMT) {
1840 case S_IFREG:
1841 fd = open(path, O_RDONLY);
1842 if (fd < 0)
1843 return error_errno("open(\"%s\")", path);
1844 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1845 return error("%s: failed to insert into database",
1846 path);
1847 break;
1848 case S_IFLNK:
1849 if (strbuf_readlink(&sb, path, st->st_size))
1850 return error_errno("readlink(\"%s\")", path);
1851 if (!(flags & HASH_WRITE_OBJECT))
1852 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1853 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1854 rc = error("%s: failed to insert into database", path);
1855 strbuf_release(&sb);
1856 break;
1857 case S_IFDIR:
1858 return resolve_gitlink_ref(path, "HEAD", oid->hash);
1859 default:
1860 return error("%s: unsupported file type", path);
1862 return rc;
1865 int read_pack_header(int fd, struct pack_header *header)
1867 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1868 /* "eof before pack header was fully read" */
1869 return PH_ERROR_EOF;
1871 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1872 /* "protocol error (pack signature mismatch detected)" */
1873 return PH_ERROR_PACK_SIGNATURE;
1874 if (!pack_version_ok(header->hdr_version))
1875 /* "protocol error (pack version unsupported)" */
1876 return PH_ERROR_PROTOCOL;
1877 return 0;
1880 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1882 enum object_type type = sha1_object_info(sha1, NULL);
1883 if (type < 0)
1884 die("%s is not a valid object", sha1_to_hex(sha1));
1885 if (type != expect)
1886 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1887 typename(expect));
1890 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1891 struct strbuf *path,
1892 each_loose_object_fn obj_cb,
1893 each_loose_cruft_fn cruft_cb,
1894 each_loose_subdir_fn subdir_cb,
1895 void *data)
1897 size_t origlen, baselen;
1898 DIR *dir;
1899 struct dirent *de;
1900 int r = 0;
1902 if (subdir_nr > 0xff)
1903 BUG("invalid loose object subdirectory: %x", subdir_nr);
1905 origlen = path->len;
1906 strbuf_complete(path, '/');
1907 strbuf_addf(path, "%02x", subdir_nr);
1908 baselen = path->len;
1910 dir = opendir(path->buf);
1911 if (!dir) {
1912 if (errno != ENOENT)
1913 r = error_errno("unable to open %s", path->buf);
1914 strbuf_setlen(path, origlen);
1915 return r;
1918 while ((de = readdir(dir))) {
1919 if (is_dot_or_dotdot(de->d_name))
1920 continue;
1922 strbuf_setlen(path, baselen);
1923 strbuf_addf(path, "/%s", de->d_name);
1925 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
1926 char hex[GIT_MAX_HEXSZ+1];
1927 struct object_id oid;
1929 xsnprintf(hex, sizeof(hex), "%02x%s",
1930 subdir_nr, de->d_name);
1931 if (!get_oid_hex(hex, &oid)) {
1932 if (obj_cb) {
1933 r = obj_cb(&oid, path->buf, data);
1934 if (r)
1935 break;
1937 continue;
1941 if (cruft_cb) {
1942 r = cruft_cb(de->d_name, path->buf, data);
1943 if (r)
1944 break;
1947 closedir(dir);
1949 strbuf_setlen(path, baselen);
1950 if (!r && subdir_cb)
1951 r = subdir_cb(subdir_nr, path->buf, data);
1953 strbuf_setlen(path, origlen);
1955 return r;
1958 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
1959 each_loose_object_fn obj_cb,
1960 each_loose_cruft_fn cruft_cb,
1961 each_loose_subdir_fn subdir_cb,
1962 void *data)
1964 int r = 0;
1965 int i;
1967 for (i = 0; i < 256; i++) {
1968 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
1969 subdir_cb, data);
1970 if (r)
1971 break;
1974 return r;
1977 int for_each_loose_file_in_objdir(const char *path,
1978 each_loose_object_fn obj_cb,
1979 each_loose_cruft_fn cruft_cb,
1980 each_loose_subdir_fn subdir_cb,
1981 void *data)
1983 struct strbuf buf = STRBUF_INIT;
1984 int r;
1986 strbuf_addstr(&buf, path);
1987 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
1988 subdir_cb, data);
1989 strbuf_release(&buf);
1991 return r;
1994 struct loose_alt_odb_data {
1995 each_loose_object_fn *cb;
1996 void *data;
1999 static int loose_from_alt_odb(struct alternate_object_database *alt,
2000 void *vdata)
2002 struct loose_alt_odb_data *data = vdata;
2003 struct strbuf buf = STRBUF_INIT;
2004 int r;
2006 strbuf_addstr(&buf, alt->path);
2007 r = for_each_loose_file_in_objdir_buf(&buf,
2008 data->cb, NULL, NULL,
2009 data->data);
2010 strbuf_release(&buf);
2011 return r;
2014 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2016 struct loose_alt_odb_data alt;
2017 int r;
2019 r = for_each_loose_file_in_objdir(get_object_directory(),
2020 cb, NULL, NULL, data);
2021 if (r)
2022 return r;
2024 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2025 return 0;
2027 alt.cb = cb;
2028 alt.data = data;
2029 return foreach_alt_odb(loose_from_alt_odb, &alt);
2032 static int check_stream_sha1(git_zstream *stream,
2033 const char *hdr,
2034 unsigned long size,
2035 const char *path,
2036 const unsigned char *expected_sha1)
2038 git_SHA_CTX c;
2039 unsigned char real_sha1[GIT_MAX_RAWSZ];
2040 unsigned char buf[4096];
2041 unsigned long total_read;
2042 int status = Z_OK;
2044 git_SHA1_Init(&c);
2045 git_SHA1_Update(&c, hdr, stream->total_out);
2048 * We already read some bytes into hdr, but the ones up to the NUL
2049 * do not count against the object's content size.
2051 total_read = stream->total_out - strlen(hdr) - 1;
2054 * This size comparison must be "<=" to read the final zlib packets;
2055 * see the comment in unpack_sha1_rest for details.
2057 while (total_read <= size &&
2058 (status == Z_OK || status == Z_BUF_ERROR)) {
2059 stream->next_out = buf;
2060 stream->avail_out = sizeof(buf);
2061 if (size - total_read < stream->avail_out)
2062 stream->avail_out = size - total_read;
2063 status = git_inflate(stream, Z_FINISH);
2064 git_SHA1_Update(&c, buf, stream->next_out - buf);
2065 total_read += stream->next_out - buf;
2067 git_inflate_end(stream);
2069 if (status != Z_STREAM_END) {
2070 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2071 return -1;
2073 if (stream->avail_in) {
2074 error("garbage at end of loose object '%s'",
2075 sha1_to_hex(expected_sha1));
2076 return -1;
2079 git_SHA1_Final(real_sha1, &c);
2080 if (hashcmp(expected_sha1, real_sha1)) {
2081 error("sha1 mismatch for %s (expected %s)", path,
2082 sha1_to_hex(expected_sha1));
2083 return -1;
2086 return 0;
2089 int read_loose_object(const char *path,
2090 const unsigned char *expected_sha1,
2091 enum object_type *type,
2092 unsigned long *size,
2093 void **contents)
2095 int ret = -1;
2096 void *map = NULL;
2097 unsigned long mapsize;
2098 git_zstream stream;
2099 char hdr[32];
2101 *contents = NULL;
2103 map = map_sha1_file_1(path, NULL, &mapsize);
2104 if (!map) {
2105 error_errno("unable to mmap %s", path);
2106 goto out;
2109 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2110 error("unable to unpack header of %s", path);
2111 goto out;
2114 *type = parse_sha1_header(hdr, size);
2115 if (*type < 0) {
2116 error("unable to parse header of %s", path);
2117 git_inflate_end(&stream);
2118 goto out;
2121 if (*type == OBJ_BLOB) {
2122 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2123 goto out;
2124 } else {
2125 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2126 if (!*contents) {
2127 error("unable to unpack contents of %s", path);
2128 git_inflate_end(&stream);
2129 goto out;
2131 if (check_sha1_signature(expected_sha1, *contents,
2132 *size, typename(*type))) {
2133 error("sha1 mismatch for %s (expected %s)", path,
2134 sha1_to_hex(expected_sha1));
2135 free(*contents);
2136 goto out;
2140 ret = 0; /* everything checks out */
2142 out:
2143 if (map)
2144 munmap(map, mapsize);
2145 return ret;