Tests: clean up and document submodule helpers
[git/debian.git] / sha1_file.c
blobd7de8184edba767f0ffced8e3bd29db92b4e68af
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 (depth > 5) {
408 error("%s: ignoring alternate object stores, nesting too deep.",
409 relative_base);
410 return;
413 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
414 if (strbuf_normalize_path(&objdirbuf) < 0)
415 die("unable to normalize object directory: %s",
416 objdirbuf.buf);
418 while (*alt) {
419 alt = parse_alt_odb_entry(alt, sep, &entry);
420 if (!entry.len)
421 continue;
422 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
424 strbuf_release(&entry);
425 strbuf_release(&objdirbuf);
428 static void read_info_alternates(const char * relative_base, int depth)
430 char *path;
431 struct strbuf buf = STRBUF_INIT;
433 path = xstrfmt("%s/info/alternates", relative_base);
434 if (strbuf_read_file(&buf, path, 1024) < 0) {
435 warn_on_fopen_errors(path);
436 free(path);
437 return;
440 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
441 strbuf_release(&buf);
442 free(path);
445 struct alternate_object_database *alloc_alt_odb(const char *dir)
447 struct alternate_object_database *ent;
449 FLEX_ALLOC_STR(ent, path, dir);
450 strbuf_init(&ent->scratch, 0);
451 strbuf_addf(&ent->scratch, "%s/", dir);
452 ent->base_len = ent->scratch.len;
454 return ent;
457 void add_to_alternates_file(const char *reference)
459 struct lock_file lock = LOCK_INIT;
460 char *alts = git_pathdup("objects/info/alternates");
461 FILE *in, *out;
462 int found = 0;
464 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
465 out = fdopen_lock_file(&lock, "w");
466 if (!out)
467 die_errno("unable to fdopen alternates lockfile");
469 in = fopen(alts, "r");
470 if (in) {
471 struct strbuf line = STRBUF_INIT;
473 while (strbuf_getline(&line, in) != EOF) {
474 if (!strcmp(reference, line.buf)) {
475 found = 1;
476 break;
478 fprintf_or_die(out, "%s\n", line.buf);
481 strbuf_release(&line);
482 fclose(in);
484 else if (errno != ENOENT)
485 die_errno("unable to read alternates file");
487 if (found) {
488 rollback_lock_file(&lock);
489 } else {
490 fprintf_or_die(out, "%s\n", reference);
491 if (commit_lock_file(&lock))
492 die_errno("unable to move new alternates file into place");
493 if (alt_odb_tail)
494 link_alt_odb_entries(reference, '\n', NULL, 0);
496 free(alts);
499 void add_to_alternates_memory(const char *reference)
502 * Make sure alternates are initialized, or else our entry may be
503 * overwritten when they are.
505 prepare_alt_odb();
507 link_alt_odb_entries(reference, '\n', NULL, 0);
511 * Compute the exact path an alternate is at and returns it. In case of
512 * error NULL is returned and the human readable error is added to `err`
513 * `path` may be relative and should point to $GITDIR.
514 * `err` must not be null.
516 char *compute_alternate_path(const char *path, struct strbuf *err)
518 char *ref_git = NULL;
519 const char *repo, *ref_git_s;
520 int seen_error = 0;
522 ref_git_s = real_path_if_valid(path);
523 if (!ref_git_s) {
524 seen_error = 1;
525 strbuf_addf(err, _("path '%s' does not exist"), path);
526 goto out;
527 } else
529 * Beware: read_gitfile(), real_path() and mkpath()
530 * return static buffer
532 ref_git = xstrdup(ref_git_s);
534 repo = read_gitfile(ref_git);
535 if (!repo)
536 repo = read_gitfile(mkpath("%s/.git", ref_git));
537 if (repo) {
538 free(ref_git);
539 ref_git = xstrdup(repo);
542 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
543 char *ref_git_git = mkpathdup("%s/.git", ref_git);
544 free(ref_git);
545 ref_git = ref_git_git;
546 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
547 struct strbuf sb = STRBUF_INIT;
548 seen_error = 1;
549 if (get_common_dir(&sb, ref_git)) {
550 strbuf_addf(err,
551 _("reference repository '%s' as a linked "
552 "checkout is not supported yet."),
553 path);
554 goto out;
557 strbuf_addf(err, _("reference repository '%s' is not a "
558 "local repository."), path);
559 goto out;
562 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
563 strbuf_addf(err, _("reference repository '%s' is shallow"),
564 path);
565 seen_error = 1;
566 goto out;
569 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
570 strbuf_addf(err,
571 _("reference repository '%s' is grafted"),
572 path);
573 seen_error = 1;
574 goto out;
577 out:
578 if (seen_error) {
579 FREE_AND_NULL(ref_git);
582 return ref_git;
585 int foreach_alt_odb(alt_odb_fn fn, void *cb)
587 struct alternate_object_database *ent;
588 int r = 0;
590 prepare_alt_odb();
591 for (ent = alt_odb_list; ent; ent = ent->next) {
592 r = fn(ent, cb);
593 if (r)
594 break;
596 return r;
599 void prepare_alt_odb(void)
601 const char *alt;
603 if (alt_odb_tail)
604 return;
606 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
607 if (!alt) alt = "";
609 alt_odb_tail = &alt_odb_list;
610 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
612 read_info_alternates(get_object_directory(), 0);
615 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
616 static int freshen_file(const char *fn)
618 struct utimbuf t;
619 t.actime = t.modtime = time(NULL);
620 return !utime(fn, &t);
624 * All of the check_and_freshen functions return 1 if the file exists and was
625 * freshened (if freshening was requested), 0 otherwise. If they return
626 * 0, you should not assume that it is safe to skip a write of the object (it
627 * either does not exist on disk, or has a stale mtime and may be subject to
628 * pruning).
630 int check_and_freshen_file(const char *fn, int freshen)
632 if (access(fn, F_OK))
633 return 0;
634 if (freshen && !freshen_file(fn))
635 return 0;
636 return 1;
639 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
641 return check_and_freshen_file(sha1_file_name(sha1), freshen);
644 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
646 struct alternate_object_database *alt;
647 prepare_alt_odb();
648 for (alt = alt_odb_list; alt; alt = alt->next) {
649 const char *path = alt_sha1_path(alt, sha1);
650 if (check_and_freshen_file(path, freshen))
651 return 1;
653 return 0;
656 static int check_and_freshen(const unsigned char *sha1, int freshen)
658 return check_and_freshen_local(sha1, freshen) ||
659 check_and_freshen_nonlocal(sha1, freshen);
662 int has_loose_object_nonlocal(const unsigned char *sha1)
664 return check_and_freshen_nonlocal(sha1, 0);
667 static int has_loose_object(const unsigned char *sha1)
669 return check_and_freshen(sha1, 0);
672 static void mmap_limit_check(size_t length)
674 static size_t limit = 0;
675 if (!limit) {
676 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
677 if (!limit)
678 limit = SIZE_MAX;
680 if (length > limit)
681 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
682 (uintmax_t)length, (uintmax_t)limit);
685 void *xmmap_gently(void *start, size_t length,
686 int prot, int flags, int fd, off_t offset)
688 void *ret;
690 mmap_limit_check(length);
691 ret = mmap(start, length, prot, flags, fd, offset);
692 if (ret == MAP_FAILED) {
693 if (!length)
694 return NULL;
695 release_pack_memory(length);
696 ret = mmap(start, length, prot, flags, fd, offset);
698 return ret;
701 void *xmmap(void *start, size_t length,
702 int prot, int flags, int fd, off_t offset)
704 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
705 if (ret == MAP_FAILED)
706 die_errno("mmap failed");
707 return ret;
711 * With an in-core object data in "map", rehash it to make sure the
712 * object name actually matches "sha1" to detect object corruption.
713 * With "map" == NULL, try reading the object named with "sha1" using
714 * the streaming interface and rehash it to do the same.
716 int check_sha1_signature(const unsigned char *sha1, void *map,
717 unsigned long size, const char *type)
719 unsigned char real_sha1[20];
720 enum object_type obj_type;
721 struct git_istream *st;
722 git_SHA_CTX c;
723 char hdr[32];
724 int hdrlen;
726 if (map) {
727 hash_sha1_file(map, size, type, real_sha1);
728 return hashcmp(sha1, real_sha1) ? -1 : 0;
731 st = open_istream(sha1, &obj_type, &size, NULL);
732 if (!st)
733 return -1;
735 /* Generate the header */
736 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
738 /* Sha1.. */
739 git_SHA1_Init(&c);
740 git_SHA1_Update(&c, hdr, hdrlen);
741 for (;;) {
742 char buf[1024 * 16];
743 ssize_t readlen = read_istream(st, buf, sizeof(buf));
745 if (readlen < 0) {
746 close_istream(st);
747 return -1;
749 if (!readlen)
750 break;
751 git_SHA1_Update(&c, buf, readlen);
753 git_SHA1_Final(real_sha1, &c);
754 close_istream(st);
755 return hashcmp(sha1, real_sha1) ? -1 : 0;
758 int git_open_cloexec(const char *name, int flags)
760 int fd;
761 static int o_cloexec = O_CLOEXEC;
763 fd = open(name, flags | o_cloexec);
764 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
765 /* Try again w/o O_CLOEXEC: the kernel might not support it */
766 o_cloexec &= ~O_CLOEXEC;
767 fd = open(name, flags | o_cloexec);
770 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
772 static int fd_cloexec = FD_CLOEXEC;
774 if (!o_cloexec && 0 <= fd && fd_cloexec) {
775 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
776 int flags = fcntl(fd, F_GETFD);
777 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
778 fd_cloexec = 0;
781 #endif
782 return fd;
786 * Find "sha1" as a loose object in the local repository or in an alternate.
787 * Returns 0 on success, negative on failure.
789 * The "path" out-parameter will give the path of the object we found (if any).
790 * Note that it may point to static storage and is only valid until another
791 * call to sha1_file_name(), etc.
793 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
794 const char **path)
796 struct alternate_object_database *alt;
798 *path = sha1_file_name(sha1);
799 if (!lstat(*path, st))
800 return 0;
802 prepare_alt_odb();
803 errno = ENOENT;
804 for (alt = alt_odb_list; alt; alt = alt->next) {
805 *path = alt_sha1_path(alt, sha1);
806 if (!lstat(*path, st))
807 return 0;
810 return -1;
814 * Like stat_sha1_file(), but actually open the object and return the
815 * descriptor. See the caveats on the "path" parameter above.
817 static int open_sha1_file(const unsigned char *sha1, const char **path)
819 int fd;
820 struct alternate_object_database *alt;
821 int most_interesting_errno;
823 *path = sha1_file_name(sha1);
824 fd = git_open(*path);
825 if (fd >= 0)
826 return fd;
827 most_interesting_errno = errno;
829 prepare_alt_odb();
830 for (alt = alt_odb_list; alt; alt = alt->next) {
831 *path = alt_sha1_path(alt, sha1);
832 fd = git_open(*path);
833 if (fd >= 0)
834 return fd;
835 if (most_interesting_errno == ENOENT)
836 most_interesting_errno = errno;
838 errno = most_interesting_errno;
839 return -1;
843 * Map the loose object at "path" if it is not NULL, or the path found by
844 * searching for a loose object named "sha1".
846 static void *map_sha1_file_1(const char *path,
847 const unsigned char *sha1,
848 unsigned long *size)
850 void *map;
851 int fd;
853 if (path)
854 fd = git_open(path);
855 else
856 fd = open_sha1_file(sha1, &path);
857 map = NULL;
858 if (fd >= 0) {
859 struct stat st;
861 if (!fstat(fd, &st)) {
862 *size = xsize_t(st.st_size);
863 if (!*size) {
864 /* mmap() is forbidden on empty files */
865 error("object file %s is empty", path);
866 return NULL;
868 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
870 close(fd);
872 return map;
875 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
877 return map_sha1_file_1(NULL, sha1, size);
880 static int unpack_sha1_short_header(git_zstream *stream,
881 unsigned char *map, unsigned long mapsize,
882 void *buffer, unsigned long bufsiz)
884 /* Get the data stream */
885 memset(stream, 0, sizeof(*stream));
886 stream->next_in = map;
887 stream->avail_in = mapsize;
888 stream->next_out = buffer;
889 stream->avail_out = bufsiz;
891 git_inflate_init(stream);
892 return git_inflate(stream, 0);
895 int unpack_sha1_header(git_zstream *stream,
896 unsigned char *map, unsigned long mapsize,
897 void *buffer, unsigned long bufsiz)
899 int status = unpack_sha1_short_header(stream, map, mapsize,
900 buffer, bufsiz);
902 if (status < Z_OK)
903 return status;
905 /* Make sure we have the terminating NUL */
906 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
907 return -1;
908 return 0;
911 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
912 unsigned long mapsize, void *buffer,
913 unsigned long bufsiz, struct strbuf *header)
915 int status;
917 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
918 if (status < Z_OK)
919 return -1;
922 * Check if entire header is unpacked in the first iteration.
924 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
925 return 0;
928 * buffer[0..bufsiz] was not large enough. Copy the partial
929 * result out to header, and then append the result of further
930 * reading the stream.
932 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
933 stream->next_out = buffer;
934 stream->avail_out = bufsiz;
936 do {
937 status = git_inflate(stream, 0);
938 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
939 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
940 return 0;
941 stream->next_out = buffer;
942 stream->avail_out = bufsiz;
943 } while (status != Z_STREAM_END);
944 return -1;
947 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
949 int bytes = strlen(buffer) + 1;
950 unsigned char *buf = xmallocz(size);
951 unsigned long n;
952 int status = Z_OK;
954 n = stream->total_out - bytes;
955 if (n > size)
956 n = size;
957 memcpy(buf, (char *) buffer + bytes, n);
958 bytes = n;
959 if (bytes <= size) {
961 * The above condition must be (bytes <= size), not
962 * (bytes < size). In other words, even though we
963 * expect no more output and set avail_out to zero,
964 * the input zlib stream may have bytes that express
965 * "this concludes the stream", and we *do* want to
966 * eat that input.
968 * Otherwise we would not be able to test that we
969 * consumed all the input to reach the expected size;
970 * we also want to check that zlib tells us that all
971 * went well with status == Z_STREAM_END at the end.
973 stream->next_out = buf + bytes;
974 stream->avail_out = size - bytes;
975 while (status == Z_OK)
976 status = git_inflate(stream, Z_FINISH);
978 if (status == Z_STREAM_END && !stream->avail_in) {
979 git_inflate_end(stream);
980 return buf;
983 if (status < 0)
984 error("corrupt loose object '%s'", sha1_to_hex(sha1));
985 else if (stream->avail_in)
986 error("garbage at end of loose object '%s'",
987 sha1_to_hex(sha1));
988 free(buf);
989 return NULL;
993 * We used to just use "sscanf()", but that's actually way
994 * too permissive for what we want to check. So do an anal
995 * object header parse by hand.
997 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
998 unsigned int flags)
1000 const char *type_buf = hdr;
1001 unsigned long size;
1002 int type, type_len = 0;
1005 * The type can be of any size but is followed by
1006 * a space.
1008 for (;;) {
1009 char c = *hdr++;
1010 if (!c)
1011 return -1;
1012 if (c == ' ')
1013 break;
1014 type_len++;
1017 type = type_from_string_gently(type_buf, type_len, 1);
1018 if (oi->typename)
1019 strbuf_add(oi->typename, type_buf, type_len);
1021 * Set type to 0 if its an unknown object and
1022 * we're obtaining the type using '--allow-unknown-type'
1023 * option.
1025 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1026 type = 0;
1027 else if (type < 0)
1028 die("invalid object type");
1029 if (oi->typep)
1030 *oi->typep = type;
1033 * The length must follow immediately, and be in canonical
1034 * decimal format (ie "010" is not valid).
1036 size = *hdr++ - '0';
1037 if (size > 9)
1038 return -1;
1039 if (size) {
1040 for (;;) {
1041 unsigned long c = *hdr - '0';
1042 if (c > 9)
1043 break;
1044 hdr++;
1045 size = size * 10 + c;
1049 if (oi->sizep)
1050 *oi->sizep = size;
1053 * The length must be followed by a zero byte
1055 return *hdr ? -1 : type;
1058 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1060 struct object_info oi = OBJECT_INFO_INIT;
1062 oi.sizep = sizep;
1063 return parse_sha1_header_extended(hdr, &oi, 0);
1066 static int sha1_loose_object_info(const unsigned char *sha1,
1067 struct object_info *oi,
1068 int flags)
1070 int status = 0;
1071 unsigned long mapsize;
1072 void *map;
1073 git_zstream stream;
1074 char hdr[32];
1075 struct strbuf hdrbuf = STRBUF_INIT;
1076 unsigned long size_scratch;
1078 if (oi->delta_base_sha1)
1079 hashclr(oi->delta_base_sha1);
1082 * If we don't care about type or size, then we don't
1083 * need to look inside the object at all. Note that we
1084 * do not optimize out the stat call, even if the
1085 * caller doesn't care about the disk-size, since our
1086 * return value implicitly indicates whether the
1087 * object even exists.
1089 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1090 const char *path;
1091 struct stat st;
1092 if (stat_sha1_file(sha1, &st, &path) < 0)
1093 return -1;
1094 if (oi->disk_sizep)
1095 *oi->disk_sizep = st.st_size;
1096 return 0;
1099 map = map_sha1_file(sha1, &mapsize);
1100 if (!map)
1101 return -1;
1103 if (!oi->sizep)
1104 oi->sizep = &size_scratch;
1106 if (oi->disk_sizep)
1107 *oi->disk_sizep = mapsize;
1108 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1109 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1110 status = error("unable to unpack %s header with --allow-unknown-type",
1111 sha1_to_hex(sha1));
1112 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1113 status = error("unable to unpack %s header",
1114 sha1_to_hex(sha1));
1115 if (status < 0)
1116 ; /* Do nothing */
1117 else if (hdrbuf.len) {
1118 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1119 status = error("unable to parse %s header with --allow-unknown-type",
1120 sha1_to_hex(sha1));
1121 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1122 status = error("unable to parse %s header", sha1_to_hex(sha1));
1124 if (status >= 0 && oi->contentp) {
1125 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1126 *oi->sizep, sha1);
1127 if (!*oi->contentp) {
1128 git_inflate_end(&stream);
1129 status = -1;
1131 } else
1132 git_inflate_end(&stream);
1134 munmap(map, mapsize);
1135 if (status && oi->typep)
1136 *oi->typep = status;
1137 if (oi->sizep == &size_scratch)
1138 oi->sizep = NULL;
1139 strbuf_release(&hdrbuf);
1140 oi->whence = OI_LOOSE;
1141 return (status < 0) ? status : 0;
1144 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1146 static struct object_info blank_oi = OBJECT_INFO_INIT;
1147 struct pack_entry e;
1148 int rtype;
1149 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1150 lookup_replace_object(sha1) :
1151 sha1;
1153 if (!oi)
1154 oi = &blank_oi;
1156 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1157 struct cached_object *co = find_cached_object(real);
1158 if (co) {
1159 if (oi->typep)
1160 *(oi->typep) = co->type;
1161 if (oi->sizep)
1162 *(oi->sizep) = co->size;
1163 if (oi->disk_sizep)
1164 *(oi->disk_sizep) = 0;
1165 if (oi->delta_base_sha1)
1166 hashclr(oi->delta_base_sha1);
1167 if (oi->typename)
1168 strbuf_addstr(oi->typename, typename(co->type));
1169 if (oi->contentp)
1170 *oi->contentp = xmemdupz(co->buf, co->size);
1171 oi->whence = OI_CACHED;
1172 return 0;
1176 if (!find_pack_entry(real, &e)) {
1177 /* Most likely it's a loose object. */
1178 if (!sha1_loose_object_info(real, oi, flags))
1179 return 0;
1181 /* Not a loose object; someone else may have just packed it. */
1182 if (flags & OBJECT_INFO_QUICK) {
1183 return -1;
1184 } else {
1185 reprepare_packed_git();
1186 if (!find_pack_entry(real, &e))
1187 return -1;
1191 if (oi == &blank_oi)
1193 * We know that the caller doesn't actually need the
1194 * information below, so return early.
1196 return 0;
1198 rtype = packed_object_info(e.p, e.offset, oi);
1199 if (rtype < 0) {
1200 mark_bad_packed_object(e.p, real);
1201 return sha1_object_info_extended(real, oi, 0);
1202 } else if (oi->whence == OI_PACKED) {
1203 oi->u.packed.offset = e.offset;
1204 oi->u.packed.pack = e.p;
1205 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1206 rtype == OBJ_OFS_DELTA);
1209 return 0;
1212 /* returns enum object_type or negative */
1213 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1215 enum object_type type;
1216 struct object_info oi = OBJECT_INFO_INIT;
1218 oi.typep = &type;
1219 oi.sizep = sizep;
1220 if (sha1_object_info_extended(sha1, &oi,
1221 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1222 return -1;
1223 return type;
1226 static void *read_object(const unsigned char *sha1, enum object_type *type,
1227 unsigned long *size)
1229 struct object_info oi = OBJECT_INFO_INIT;
1230 void *content;
1231 oi.typep = type;
1232 oi.sizep = size;
1233 oi.contentp = &content;
1235 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1236 return NULL;
1237 return content;
1240 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1241 unsigned char *sha1)
1243 struct cached_object *co;
1245 hash_sha1_file(buf, len, typename(type), sha1);
1246 if (has_sha1_file(sha1) || find_cached_object(sha1))
1247 return 0;
1248 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1249 co = &cached_objects[cached_object_nr++];
1250 co->size = len;
1251 co->type = type;
1252 co->buf = xmalloc(len);
1253 memcpy(co->buf, buf, len);
1254 hashcpy(co->sha1, sha1);
1255 return 0;
1259 * This function dies on corrupt objects; the callers who want to
1260 * deal with them should arrange to call read_object() and give error
1261 * messages themselves.
1263 void *read_sha1_file_extended(const unsigned char *sha1,
1264 enum object_type *type,
1265 unsigned long *size,
1266 int lookup_replace)
1268 void *data;
1269 const struct packed_git *p;
1270 const char *path;
1271 struct stat st;
1272 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1273 : sha1;
1275 errno = 0;
1276 data = read_object(repl, type, size);
1277 if (data)
1278 return data;
1280 if (errno && errno != ENOENT)
1281 die_errno("failed to read object %s", sha1_to_hex(sha1));
1283 /* die if we replaced an object with one that does not exist */
1284 if (repl != sha1)
1285 die("replacement %s not found for %s",
1286 sha1_to_hex(repl), sha1_to_hex(sha1));
1288 if (!stat_sha1_file(repl, &st, &path))
1289 die("loose object %s (stored in %s) is corrupt",
1290 sha1_to_hex(repl), path);
1292 if ((p = has_packed_and_bad(repl)) != NULL)
1293 die("packed object %s (stored in %s) is corrupt",
1294 sha1_to_hex(repl), p->pack_name);
1296 return NULL;
1299 void *read_object_with_reference(const unsigned char *sha1,
1300 const char *required_type_name,
1301 unsigned long *size,
1302 unsigned char *actual_sha1_return)
1304 enum object_type type, required_type;
1305 void *buffer;
1306 unsigned long isize;
1307 unsigned char actual_sha1[20];
1309 required_type = type_from_string(required_type_name);
1310 hashcpy(actual_sha1, sha1);
1311 while (1) {
1312 int ref_length = -1;
1313 const char *ref_type = NULL;
1315 buffer = read_sha1_file(actual_sha1, &type, &isize);
1316 if (!buffer)
1317 return NULL;
1318 if (type == required_type) {
1319 *size = isize;
1320 if (actual_sha1_return)
1321 hashcpy(actual_sha1_return, actual_sha1);
1322 return buffer;
1324 /* Handle references */
1325 else if (type == OBJ_COMMIT)
1326 ref_type = "tree ";
1327 else if (type == OBJ_TAG)
1328 ref_type = "object ";
1329 else {
1330 free(buffer);
1331 return NULL;
1333 ref_length = strlen(ref_type);
1335 if (ref_length + 40 > isize ||
1336 memcmp(buffer, ref_type, ref_length) ||
1337 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1338 free(buffer);
1339 return NULL;
1341 free(buffer);
1342 /* Now we have the ID of the referred-to object in
1343 * actual_sha1. Check again. */
1347 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1348 const char *type, unsigned char *sha1,
1349 char *hdr, int *hdrlen)
1351 git_SHA_CTX c;
1353 /* Generate the header */
1354 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1356 /* Sha1.. */
1357 git_SHA1_Init(&c);
1358 git_SHA1_Update(&c, hdr, *hdrlen);
1359 git_SHA1_Update(&c, buf, len);
1360 git_SHA1_Final(sha1, &c);
1364 * Move the just written object into its final resting place.
1366 int finalize_object_file(const char *tmpfile, const char *filename)
1368 int ret = 0;
1370 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1371 goto try_rename;
1372 else if (link(tmpfile, filename))
1373 ret = errno;
1376 * Coda hack - coda doesn't like cross-directory links,
1377 * so we fall back to a rename, which will mean that it
1378 * won't be able to check collisions, but that's not a
1379 * big deal.
1381 * The same holds for FAT formatted media.
1383 * When this succeeds, we just return. We have nothing
1384 * left to unlink.
1386 if (ret && ret != EEXIST) {
1387 try_rename:
1388 if (!rename(tmpfile, filename))
1389 goto out;
1390 ret = errno;
1392 unlink_or_warn(tmpfile);
1393 if (ret) {
1394 if (ret != EEXIST) {
1395 return error_errno("unable to write sha1 filename %s", filename);
1397 /* FIXME!!! Collision check here ? */
1400 out:
1401 if (adjust_shared_perm(filename))
1402 return error("unable to set permission to '%s'", filename);
1403 return 0;
1406 static int write_buffer(int fd, const void *buf, size_t len)
1408 if (write_in_full(fd, buf, len) < 0)
1409 return error_errno("file write error");
1410 return 0;
1413 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1414 unsigned char *sha1)
1416 char hdr[32];
1417 int hdrlen = sizeof(hdr);
1418 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1419 return 0;
1422 /* Finalize a file on disk, and close it. */
1423 static void close_sha1_file(int fd)
1425 if (fsync_object_files)
1426 fsync_or_die(fd, "sha1 file");
1427 if (close(fd) != 0)
1428 die_errno("error when closing sha1 file");
1431 /* Size of directory component, including the ending '/' */
1432 static inline int directory_size(const char *filename)
1434 const char *s = strrchr(filename, '/');
1435 if (!s)
1436 return 0;
1437 return s - filename + 1;
1441 * This creates a temporary file in the same directory as the final
1442 * 'filename'
1444 * We want to avoid cross-directory filename renames, because those
1445 * can have problems on various filesystems (FAT, NFS, Coda).
1447 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1449 int fd, dirlen = directory_size(filename);
1451 strbuf_reset(tmp);
1452 strbuf_add(tmp, filename, dirlen);
1453 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1454 fd = git_mkstemp_mode(tmp->buf, 0444);
1455 if (fd < 0 && dirlen && errno == ENOENT) {
1457 * Make sure the directory exists; note that the contents
1458 * of the buffer are undefined after mkstemp returns an
1459 * error, so we have to rewrite the whole buffer from
1460 * scratch.
1462 strbuf_reset(tmp);
1463 strbuf_add(tmp, filename, dirlen - 1);
1464 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1465 return -1;
1466 if (adjust_shared_perm(tmp->buf))
1467 return -1;
1469 /* Try again */
1470 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1471 fd = git_mkstemp_mode(tmp->buf, 0444);
1473 return fd;
1476 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1477 const void *buf, unsigned long len, time_t mtime)
1479 int fd, ret;
1480 unsigned char compressed[4096];
1481 git_zstream stream;
1482 git_SHA_CTX c;
1483 unsigned char parano_sha1[20];
1484 static struct strbuf tmp_file = STRBUF_INIT;
1485 const char *filename = sha1_file_name(sha1);
1487 fd = create_tmpfile(&tmp_file, filename);
1488 if (fd < 0) {
1489 if (errno == EACCES)
1490 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1491 else
1492 return error_errno("unable to create temporary file");
1495 /* Set it up */
1496 git_deflate_init(&stream, zlib_compression_level);
1497 stream.next_out = compressed;
1498 stream.avail_out = sizeof(compressed);
1499 git_SHA1_Init(&c);
1501 /* First header.. */
1502 stream.next_in = (unsigned char *)hdr;
1503 stream.avail_in = hdrlen;
1504 while (git_deflate(&stream, 0) == Z_OK)
1505 ; /* nothing */
1506 git_SHA1_Update(&c, hdr, hdrlen);
1508 /* Then the data itself.. */
1509 stream.next_in = (void *)buf;
1510 stream.avail_in = len;
1511 do {
1512 unsigned char *in0 = stream.next_in;
1513 ret = git_deflate(&stream, Z_FINISH);
1514 git_SHA1_Update(&c, in0, stream.next_in - in0);
1515 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1516 die("unable to write sha1 file");
1517 stream.next_out = compressed;
1518 stream.avail_out = sizeof(compressed);
1519 } while (ret == Z_OK);
1521 if (ret != Z_STREAM_END)
1522 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1523 ret = git_deflate_end_gently(&stream);
1524 if (ret != Z_OK)
1525 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1526 git_SHA1_Final(parano_sha1, &c);
1527 if (hashcmp(sha1, parano_sha1) != 0)
1528 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1530 close_sha1_file(fd);
1532 if (mtime) {
1533 struct utimbuf utb;
1534 utb.actime = mtime;
1535 utb.modtime = mtime;
1536 if (utime(tmp_file.buf, &utb) < 0)
1537 warning_errno("failed utime() on %s", tmp_file.buf);
1540 return finalize_object_file(tmp_file.buf, filename);
1543 static int freshen_loose_object(const unsigned char *sha1)
1545 return check_and_freshen(sha1, 1);
1548 static int freshen_packed_object(const unsigned char *sha1)
1550 struct pack_entry e;
1551 if (!find_pack_entry(sha1, &e))
1552 return 0;
1553 if (e.p->freshened)
1554 return 1;
1555 if (!freshen_file(e.p->pack_name))
1556 return 0;
1557 e.p->freshened = 1;
1558 return 1;
1561 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1563 char hdr[32];
1564 int hdrlen = sizeof(hdr);
1566 /* Normally if we have it in the pack then we do not bother writing
1567 * it out into .git/objects/??/?{38} file.
1569 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1570 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1571 return 0;
1572 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1575 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1576 struct object_id *oid, unsigned flags)
1578 char *header;
1579 int hdrlen, status = 0;
1581 /* type string, SP, %lu of the length plus NUL must fit this */
1582 hdrlen = strlen(type) + 32;
1583 header = xmalloc(hdrlen);
1584 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1586 if (!(flags & HASH_WRITE_OBJECT))
1587 goto cleanup;
1588 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1589 goto cleanup;
1590 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1592 cleanup:
1593 free(header);
1594 return status;
1597 int force_object_loose(const unsigned char *sha1, time_t mtime)
1599 void *buf;
1600 unsigned long len;
1601 enum object_type type;
1602 char hdr[32];
1603 int hdrlen;
1604 int ret;
1606 if (has_loose_object(sha1))
1607 return 0;
1608 buf = read_object(sha1, &type, &len);
1609 if (!buf)
1610 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1611 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1612 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1613 free(buf);
1615 return ret;
1618 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1620 if (!startup_info->have_repository)
1621 return 0;
1622 return sha1_object_info_extended(sha1, NULL,
1623 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1626 int has_object_file(const struct object_id *oid)
1628 return has_sha1_file(oid->hash);
1631 int has_object_file_with_flags(const struct object_id *oid, int flags)
1633 return has_sha1_file_with_flags(oid->hash, flags);
1636 static void check_tree(const void *buf, size_t size)
1638 struct tree_desc desc;
1639 struct name_entry entry;
1641 init_tree_desc(&desc, buf, size);
1642 while (tree_entry(&desc, &entry))
1643 /* do nothing
1644 * tree_entry() will die() on malformed entries */
1648 static void check_commit(const void *buf, size_t size)
1650 struct commit c;
1651 memset(&c, 0, sizeof(c));
1652 if (parse_commit_buffer(&c, buf, size))
1653 die("corrupt commit");
1656 static void check_tag(const void *buf, size_t size)
1658 struct tag t;
1659 memset(&t, 0, sizeof(t));
1660 if (parse_tag_buffer(&t, buf, size))
1661 die("corrupt tag");
1664 static int index_mem(struct object_id *oid, void *buf, size_t size,
1665 enum object_type type,
1666 const char *path, unsigned flags)
1668 int ret, re_allocated = 0;
1669 int write_object = flags & HASH_WRITE_OBJECT;
1671 if (!type)
1672 type = OBJ_BLOB;
1675 * Convert blobs to git internal format
1677 if ((type == OBJ_BLOB) && path) {
1678 struct strbuf nbuf = STRBUF_INIT;
1679 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1680 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1681 buf = strbuf_detach(&nbuf, &size);
1682 re_allocated = 1;
1685 if (flags & HASH_FORMAT_CHECK) {
1686 if (type == OBJ_TREE)
1687 check_tree(buf, size);
1688 if (type == OBJ_COMMIT)
1689 check_commit(buf, size);
1690 if (type == OBJ_TAG)
1691 check_tag(buf, size);
1694 if (write_object)
1695 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1696 else
1697 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1698 if (re_allocated)
1699 free(buf);
1700 return ret;
1703 static int index_stream_convert_blob(struct object_id *oid, int fd,
1704 const char *path, unsigned flags)
1706 int ret;
1707 const int write_object = flags & HASH_WRITE_OBJECT;
1708 struct strbuf sbuf = STRBUF_INIT;
1710 assert(path);
1711 assert(would_convert_to_git_filter_fd(path));
1713 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1714 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1716 if (write_object)
1717 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1718 oid->hash);
1719 else
1720 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1721 oid->hash);
1722 strbuf_release(&sbuf);
1723 return ret;
1726 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1727 const char *path, unsigned flags)
1729 struct strbuf sbuf = STRBUF_INIT;
1730 int ret;
1732 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1733 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1734 else
1735 ret = -1;
1736 strbuf_release(&sbuf);
1737 return ret;
1740 #define SMALL_FILE_SIZE (32*1024)
1742 static int index_core(struct object_id *oid, int fd, size_t size,
1743 enum object_type type, const char *path,
1744 unsigned flags)
1746 int ret;
1748 if (!size) {
1749 ret = index_mem(oid, "", size, type, path, flags);
1750 } else if (size <= SMALL_FILE_SIZE) {
1751 char *buf = xmalloc(size);
1752 ssize_t read_result = read_in_full(fd, buf, size);
1753 if (read_result < 0)
1754 ret = error_errno("read error while indexing %s",
1755 path ? path : "<unknown>");
1756 else if (read_result != size)
1757 ret = error("short read while indexing %s",
1758 path ? path : "<unknown>");
1759 else
1760 ret = index_mem(oid, buf, size, type, path, flags);
1761 free(buf);
1762 } else {
1763 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1764 ret = index_mem(oid, buf, size, type, path, flags);
1765 munmap(buf, size);
1767 return ret;
1771 * This creates one packfile per large blob unless bulk-checkin
1772 * machinery is "plugged".
1774 * This also bypasses the usual "convert-to-git" dance, and that is on
1775 * purpose. We could write a streaming version of the converting
1776 * functions and insert that before feeding the data to fast-import
1777 * (or equivalent in-core API described above). However, that is
1778 * somewhat complicated, as we do not know the size of the filter
1779 * result, which we need to know beforehand when writing a git object.
1780 * Since the primary motivation for trying to stream from the working
1781 * tree file and to avoid mmaping it in core is to deal with large
1782 * binary blobs, they generally do not want to get any conversion, and
1783 * callers should avoid this code path when filters are requested.
1785 static int index_stream(struct object_id *oid, int fd, size_t size,
1786 enum object_type type, const char *path,
1787 unsigned flags)
1789 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1792 int index_fd(struct object_id *oid, int fd, struct stat *st,
1793 enum object_type type, const char *path, unsigned flags)
1795 int ret;
1798 * Call xsize_t() only when needed to avoid potentially unnecessary
1799 * die() for large files.
1801 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1802 ret = index_stream_convert_blob(oid, fd, path, flags);
1803 else if (!S_ISREG(st->st_mode))
1804 ret = index_pipe(oid, fd, type, path, flags);
1805 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1806 (path && would_convert_to_git(&the_index, path)))
1807 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1808 flags);
1809 else
1810 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1811 flags);
1812 close(fd);
1813 return ret;
1816 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1818 int fd;
1819 struct strbuf sb = STRBUF_INIT;
1820 int rc = 0;
1822 switch (st->st_mode & S_IFMT) {
1823 case S_IFREG:
1824 fd = open(path, O_RDONLY);
1825 if (fd < 0)
1826 return error_errno("open(\"%s\")", path);
1827 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1828 return error("%s: failed to insert into database",
1829 path);
1830 break;
1831 case S_IFLNK:
1832 if (strbuf_readlink(&sb, path, st->st_size))
1833 return error_errno("readlink(\"%s\")", path);
1834 if (!(flags & HASH_WRITE_OBJECT))
1835 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1836 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1837 rc = error("%s: failed to insert into database", path);
1838 strbuf_release(&sb);
1839 break;
1840 case S_IFDIR:
1841 return resolve_gitlink_ref(path, "HEAD", oid);
1842 default:
1843 return error("%s: unsupported file type", path);
1845 return rc;
1848 int read_pack_header(int fd, struct pack_header *header)
1850 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1851 /* "eof before pack header was fully read" */
1852 return PH_ERROR_EOF;
1854 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1855 /* "protocol error (pack signature mismatch detected)" */
1856 return PH_ERROR_PACK_SIGNATURE;
1857 if (!pack_version_ok(header->hdr_version))
1858 /* "protocol error (pack version unsupported)" */
1859 return PH_ERROR_PROTOCOL;
1860 return 0;
1863 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1865 enum object_type type = sha1_object_info(sha1, NULL);
1866 if (type < 0)
1867 die("%s is not a valid object", sha1_to_hex(sha1));
1868 if (type != expect)
1869 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1870 typename(expect));
1873 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1874 struct strbuf *path,
1875 each_loose_object_fn obj_cb,
1876 each_loose_cruft_fn cruft_cb,
1877 each_loose_subdir_fn subdir_cb,
1878 void *data)
1880 size_t origlen, baselen;
1881 DIR *dir;
1882 struct dirent *de;
1883 int r = 0;
1885 if (subdir_nr > 0xff)
1886 BUG("invalid loose object subdirectory: %x", subdir_nr);
1888 origlen = path->len;
1889 strbuf_complete(path, '/');
1890 strbuf_addf(path, "%02x", subdir_nr);
1891 baselen = path->len;
1893 dir = opendir(path->buf);
1894 if (!dir) {
1895 if (errno != ENOENT)
1896 r = error_errno("unable to open %s", path->buf);
1897 strbuf_setlen(path, origlen);
1898 return r;
1901 while ((de = readdir(dir))) {
1902 if (is_dot_or_dotdot(de->d_name))
1903 continue;
1905 strbuf_setlen(path, baselen);
1906 strbuf_addf(path, "/%s", de->d_name);
1908 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
1909 char hex[GIT_MAX_HEXSZ+1];
1910 struct object_id oid;
1912 xsnprintf(hex, sizeof(hex), "%02x%s",
1913 subdir_nr, de->d_name);
1914 if (!get_oid_hex(hex, &oid)) {
1915 if (obj_cb) {
1916 r = obj_cb(&oid, path->buf, data);
1917 if (r)
1918 break;
1920 continue;
1924 if (cruft_cb) {
1925 r = cruft_cb(de->d_name, path->buf, data);
1926 if (r)
1927 break;
1930 closedir(dir);
1932 strbuf_setlen(path, baselen);
1933 if (!r && subdir_cb)
1934 r = subdir_cb(subdir_nr, path->buf, data);
1936 strbuf_setlen(path, origlen);
1938 return r;
1941 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
1942 each_loose_object_fn obj_cb,
1943 each_loose_cruft_fn cruft_cb,
1944 each_loose_subdir_fn subdir_cb,
1945 void *data)
1947 int r = 0;
1948 int i;
1950 for (i = 0; i < 256; i++) {
1951 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
1952 subdir_cb, data);
1953 if (r)
1954 break;
1957 return r;
1960 int for_each_loose_file_in_objdir(const char *path,
1961 each_loose_object_fn obj_cb,
1962 each_loose_cruft_fn cruft_cb,
1963 each_loose_subdir_fn subdir_cb,
1964 void *data)
1966 struct strbuf buf = STRBUF_INIT;
1967 int r;
1969 strbuf_addstr(&buf, path);
1970 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
1971 subdir_cb, data);
1972 strbuf_release(&buf);
1974 return r;
1977 struct loose_alt_odb_data {
1978 each_loose_object_fn *cb;
1979 void *data;
1982 static int loose_from_alt_odb(struct alternate_object_database *alt,
1983 void *vdata)
1985 struct loose_alt_odb_data *data = vdata;
1986 struct strbuf buf = STRBUF_INIT;
1987 int r;
1989 strbuf_addstr(&buf, alt->path);
1990 r = for_each_loose_file_in_objdir_buf(&buf,
1991 data->cb, NULL, NULL,
1992 data->data);
1993 strbuf_release(&buf);
1994 return r;
1997 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
1999 struct loose_alt_odb_data alt;
2000 int r;
2002 r = for_each_loose_file_in_objdir(get_object_directory(),
2003 cb, NULL, NULL, data);
2004 if (r)
2005 return r;
2007 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2008 return 0;
2010 alt.cb = cb;
2011 alt.data = data;
2012 return foreach_alt_odb(loose_from_alt_odb, &alt);
2015 static int check_stream_sha1(git_zstream *stream,
2016 const char *hdr,
2017 unsigned long size,
2018 const char *path,
2019 const unsigned char *expected_sha1)
2021 git_SHA_CTX c;
2022 unsigned char real_sha1[GIT_MAX_RAWSZ];
2023 unsigned char buf[4096];
2024 unsigned long total_read;
2025 int status = Z_OK;
2027 git_SHA1_Init(&c);
2028 git_SHA1_Update(&c, hdr, stream->total_out);
2031 * We already read some bytes into hdr, but the ones up to the NUL
2032 * do not count against the object's content size.
2034 total_read = stream->total_out - strlen(hdr) - 1;
2037 * This size comparison must be "<=" to read the final zlib packets;
2038 * see the comment in unpack_sha1_rest for details.
2040 while (total_read <= size &&
2041 (status == Z_OK || status == Z_BUF_ERROR)) {
2042 stream->next_out = buf;
2043 stream->avail_out = sizeof(buf);
2044 if (size - total_read < stream->avail_out)
2045 stream->avail_out = size - total_read;
2046 status = git_inflate(stream, Z_FINISH);
2047 git_SHA1_Update(&c, buf, stream->next_out - buf);
2048 total_read += stream->next_out - buf;
2050 git_inflate_end(stream);
2052 if (status != Z_STREAM_END) {
2053 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2054 return -1;
2056 if (stream->avail_in) {
2057 error("garbage at end of loose object '%s'",
2058 sha1_to_hex(expected_sha1));
2059 return -1;
2062 git_SHA1_Final(real_sha1, &c);
2063 if (hashcmp(expected_sha1, real_sha1)) {
2064 error("sha1 mismatch for %s (expected %s)", path,
2065 sha1_to_hex(expected_sha1));
2066 return -1;
2069 return 0;
2072 int read_loose_object(const char *path,
2073 const unsigned char *expected_sha1,
2074 enum object_type *type,
2075 unsigned long *size,
2076 void **contents)
2078 int ret = -1;
2079 void *map = NULL;
2080 unsigned long mapsize;
2081 git_zstream stream;
2082 char hdr[32];
2084 *contents = NULL;
2086 map = map_sha1_file_1(path, NULL, &mapsize);
2087 if (!map) {
2088 error_errno("unable to mmap %s", path);
2089 goto out;
2092 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2093 error("unable to unpack header of %s", path);
2094 goto out;
2097 *type = parse_sha1_header(hdr, size);
2098 if (*type < 0) {
2099 error("unable to parse header of %s", path);
2100 git_inflate_end(&stream);
2101 goto out;
2104 if (*type == OBJ_BLOB) {
2105 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2106 goto out;
2107 } else {
2108 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2109 if (!*contents) {
2110 error("unable to unpack contents of %s", path);
2111 git_inflate_end(&stream);
2112 goto out;
2114 if (check_sha1_signature(expected_sha1, *contents,
2115 *size, typename(*type))) {
2116 error("sha1 mismatch for %s (expected %s)", path,
2117 sha1_to_hex(expected_sha1));
2118 free(*contents);
2119 goto out;
2123 ret = 0; /* everything checks out */
2125 out:
2126 if (map)
2127 munmap(map, mapsize);
2128 return ret;