Merge branch 'sg/travis-fixes' into maint
[git.git] / sha1_file.c
blob10c3a0083dedee98dc3d48267ad349589adef3f0
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 = xcalloc(1, sizeof(struct lock_file));
460 char *alts = git_pathdup("objects/info/alternates");
461 FILE *in, *out;
463 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
464 out = fdopen_lock_file(lock, "w");
465 if (!out)
466 die_errno("unable to fdopen alternates lockfile");
468 in = fopen(alts, "r");
469 if (in) {
470 struct strbuf line = STRBUF_INIT;
471 int found = 0;
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 if (found) {
485 rollback_lock_file(lock);
486 lock = NULL;
489 else if (errno != ENOENT)
490 die_errno("unable to read alternates file");
492 if (lock) {
493 fprintf_or_die(out, "%s\n", reference);
494 if (commit_lock_file(lock))
495 die_errno("unable to move new alternates file into place");
496 if (alt_odb_tail)
497 link_alt_odb_entries(reference, '\n', NULL, 0);
499 free(alts);
502 void add_to_alternates_memory(const char *reference)
505 * Make sure alternates are initialized, or else our entry may be
506 * overwritten when they are.
508 prepare_alt_odb();
510 link_alt_odb_entries(reference, '\n', NULL, 0);
514 * Compute the exact path an alternate is at and returns it. In case of
515 * error NULL is returned and the human readable error is added to `err`
516 * `path` may be relative and should point to $GITDIR.
517 * `err` must not be null.
519 char *compute_alternate_path(const char *path, struct strbuf *err)
521 char *ref_git = NULL;
522 const char *repo, *ref_git_s;
523 int seen_error = 0;
525 ref_git_s = real_path_if_valid(path);
526 if (!ref_git_s) {
527 seen_error = 1;
528 strbuf_addf(err, _("path '%s' does not exist"), path);
529 goto out;
530 } else
532 * Beware: read_gitfile(), real_path() and mkpath()
533 * return static buffer
535 ref_git = xstrdup(ref_git_s);
537 repo = read_gitfile(ref_git);
538 if (!repo)
539 repo = read_gitfile(mkpath("%s/.git", ref_git));
540 if (repo) {
541 free(ref_git);
542 ref_git = xstrdup(repo);
545 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
546 char *ref_git_git = mkpathdup("%s/.git", ref_git);
547 free(ref_git);
548 ref_git = ref_git_git;
549 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
550 struct strbuf sb = STRBUF_INIT;
551 seen_error = 1;
552 if (get_common_dir(&sb, ref_git)) {
553 strbuf_addf(err,
554 _("reference repository '%s' as a linked "
555 "checkout is not supported yet."),
556 path);
557 goto out;
560 strbuf_addf(err, _("reference repository '%s' is not a "
561 "local repository."), path);
562 goto out;
565 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
566 strbuf_addf(err, _("reference repository '%s' is shallow"),
567 path);
568 seen_error = 1;
569 goto out;
572 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
573 strbuf_addf(err,
574 _("reference repository '%s' is grafted"),
575 path);
576 seen_error = 1;
577 goto out;
580 out:
581 if (seen_error) {
582 FREE_AND_NULL(ref_git);
585 return ref_git;
588 int foreach_alt_odb(alt_odb_fn fn, void *cb)
590 struct alternate_object_database *ent;
591 int r = 0;
593 prepare_alt_odb();
594 for (ent = alt_odb_list; ent; ent = ent->next) {
595 r = fn(ent, cb);
596 if (r)
597 break;
599 return r;
602 void prepare_alt_odb(void)
604 const char *alt;
606 if (alt_odb_tail)
607 return;
609 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
610 if (!alt) alt = "";
612 alt_odb_tail = &alt_odb_list;
613 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
615 read_info_alternates(get_object_directory(), 0);
618 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
619 static int freshen_file(const char *fn)
621 struct utimbuf t;
622 t.actime = t.modtime = time(NULL);
623 return !utime(fn, &t);
627 * All of the check_and_freshen functions return 1 if the file exists and was
628 * freshened (if freshening was requested), 0 otherwise. If they return
629 * 0, you should not assume that it is safe to skip a write of the object (it
630 * either does not exist on disk, or has a stale mtime and may be subject to
631 * pruning).
633 int check_and_freshen_file(const char *fn, int freshen)
635 if (access(fn, F_OK))
636 return 0;
637 if (freshen && !freshen_file(fn))
638 return 0;
639 return 1;
642 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
644 return check_and_freshen_file(sha1_file_name(sha1), freshen);
647 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
649 struct alternate_object_database *alt;
650 prepare_alt_odb();
651 for (alt = alt_odb_list; alt; alt = alt->next) {
652 const char *path = alt_sha1_path(alt, sha1);
653 if (check_and_freshen_file(path, freshen))
654 return 1;
656 return 0;
659 static int check_and_freshen(const unsigned char *sha1, int freshen)
661 return check_and_freshen_local(sha1, freshen) ||
662 check_and_freshen_nonlocal(sha1, freshen);
665 int has_loose_object_nonlocal(const unsigned char *sha1)
667 return check_and_freshen_nonlocal(sha1, 0);
670 static int has_loose_object(const unsigned char *sha1)
672 return check_and_freshen(sha1, 0);
675 static void mmap_limit_check(size_t length)
677 static size_t limit = 0;
678 if (!limit) {
679 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
680 if (!limit)
681 limit = SIZE_MAX;
683 if (length > limit)
684 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
685 (uintmax_t)length, (uintmax_t)limit);
688 void *xmmap_gently(void *start, size_t length,
689 int prot, int flags, int fd, off_t offset)
691 void *ret;
693 mmap_limit_check(length);
694 ret = mmap(start, length, prot, flags, fd, offset);
695 if (ret == MAP_FAILED) {
696 if (!length)
697 return NULL;
698 release_pack_memory(length);
699 ret = mmap(start, length, prot, flags, fd, offset);
701 return ret;
704 void *xmmap(void *start, size_t length,
705 int prot, int flags, int fd, off_t offset)
707 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
708 if (ret == MAP_FAILED)
709 die_errno("mmap failed");
710 return ret;
714 * With an in-core object data in "map", rehash it to make sure the
715 * object name actually matches "sha1" to detect object corruption.
716 * With "map" == NULL, try reading the object named with "sha1" using
717 * the streaming interface and rehash it to do the same.
719 int check_sha1_signature(const unsigned char *sha1, void *map,
720 unsigned long size, const char *type)
722 unsigned char real_sha1[20];
723 enum object_type obj_type;
724 struct git_istream *st;
725 git_SHA_CTX c;
726 char hdr[32];
727 int hdrlen;
729 if (map) {
730 hash_sha1_file(map, size, type, real_sha1);
731 return hashcmp(sha1, real_sha1) ? -1 : 0;
734 st = open_istream(sha1, &obj_type, &size, NULL);
735 if (!st)
736 return -1;
738 /* Generate the header */
739 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
741 /* Sha1.. */
742 git_SHA1_Init(&c);
743 git_SHA1_Update(&c, hdr, hdrlen);
744 for (;;) {
745 char buf[1024 * 16];
746 ssize_t readlen = read_istream(st, buf, sizeof(buf));
748 if (readlen < 0) {
749 close_istream(st);
750 return -1;
752 if (!readlen)
753 break;
754 git_SHA1_Update(&c, buf, readlen);
756 git_SHA1_Final(real_sha1, &c);
757 close_istream(st);
758 return hashcmp(sha1, real_sha1) ? -1 : 0;
761 int git_open_cloexec(const char *name, int flags)
763 int fd;
764 static int o_cloexec = O_CLOEXEC;
766 fd = open(name, flags | o_cloexec);
767 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
768 /* Try again w/o O_CLOEXEC: the kernel might not support it */
769 o_cloexec &= ~O_CLOEXEC;
770 fd = open(name, flags | o_cloexec);
773 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
775 static int fd_cloexec = FD_CLOEXEC;
777 if (!o_cloexec && 0 <= fd && fd_cloexec) {
778 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
779 int flags = fcntl(fd, F_GETFD);
780 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
781 fd_cloexec = 0;
784 #endif
785 return fd;
789 * Find "sha1" as a loose object in the local repository or in an alternate.
790 * Returns 0 on success, negative on failure.
792 * The "path" out-parameter will give the path of the object we found (if any).
793 * Note that it may point to static storage and is only valid until another
794 * call to sha1_file_name(), etc.
796 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
797 const char **path)
799 struct alternate_object_database *alt;
801 *path = sha1_file_name(sha1);
802 if (!lstat(*path, st))
803 return 0;
805 prepare_alt_odb();
806 errno = ENOENT;
807 for (alt = alt_odb_list; alt; alt = alt->next) {
808 *path = alt_sha1_path(alt, sha1);
809 if (!lstat(*path, st))
810 return 0;
813 return -1;
817 * Like stat_sha1_file(), but actually open the object and return the
818 * descriptor. See the caveats on the "path" parameter above.
820 static int open_sha1_file(const unsigned char *sha1, const char **path)
822 int fd;
823 struct alternate_object_database *alt;
824 int most_interesting_errno;
826 *path = sha1_file_name(sha1);
827 fd = git_open(*path);
828 if (fd >= 0)
829 return fd;
830 most_interesting_errno = errno;
832 prepare_alt_odb();
833 for (alt = alt_odb_list; alt; alt = alt->next) {
834 *path = alt_sha1_path(alt, sha1);
835 fd = git_open(*path);
836 if (fd >= 0)
837 return fd;
838 if (most_interesting_errno == ENOENT)
839 most_interesting_errno = errno;
841 errno = most_interesting_errno;
842 return -1;
846 * Map the loose object at "path" if it is not NULL, or the path found by
847 * searching for a loose object named "sha1".
849 static void *map_sha1_file_1(const char *path,
850 const unsigned char *sha1,
851 unsigned long *size)
853 void *map;
854 int fd;
856 if (path)
857 fd = git_open(path);
858 else
859 fd = open_sha1_file(sha1, &path);
860 map = NULL;
861 if (fd >= 0) {
862 struct stat st;
864 if (!fstat(fd, &st)) {
865 *size = xsize_t(st.st_size);
866 if (!*size) {
867 /* mmap() is forbidden on empty files */
868 error("object file %s is empty", path);
869 return NULL;
871 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
873 close(fd);
875 return map;
878 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
880 return map_sha1_file_1(NULL, sha1, size);
883 static int unpack_sha1_short_header(git_zstream *stream,
884 unsigned char *map, unsigned long mapsize,
885 void *buffer, unsigned long bufsiz)
887 /* Get the data stream */
888 memset(stream, 0, sizeof(*stream));
889 stream->next_in = map;
890 stream->avail_in = mapsize;
891 stream->next_out = buffer;
892 stream->avail_out = bufsiz;
894 git_inflate_init(stream);
895 return git_inflate(stream, 0);
898 int unpack_sha1_header(git_zstream *stream,
899 unsigned char *map, unsigned long mapsize,
900 void *buffer, unsigned long bufsiz)
902 int status = unpack_sha1_short_header(stream, map, mapsize,
903 buffer, bufsiz);
905 if (status < Z_OK)
906 return status;
908 /* Make sure we have the terminating NUL */
909 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
910 return -1;
911 return 0;
914 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
915 unsigned long mapsize, void *buffer,
916 unsigned long bufsiz, struct strbuf *header)
918 int status;
920 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
921 if (status < Z_OK)
922 return -1;
925 * Check if entire header is unpacked in the first iteration.
927 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
928 return 0;
931 * buffer[0..bufsiz] was not large enough. Copy the partial
932 * result out to header, and then append the result of further
933 * reading the stream.
935 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
936 stream->next_out = buffer;
937 stream->avail_out = bufsiz;
939 do {
940 status = git_inflate(stream, 0);
941 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
942 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
943 return 0;
944 stream->next_out = buffer;
945 stream->avail_out = bufsiz;
946 } while (status != Z_STREAM_END);
947 return -1;
950 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
952 int bytes = strlen(buffer) + 1;
953 unsigned char *buf = xmallocz(size);
954 unsigned long n;
955 int status = Z_OK;
957 n = stream->total_out - bytes;
958 if (n > size)
959 n = size;
960 memcpy(buf, (char *) buffer + bytes, n);
961 bytes = n;
962 if (bytes <= size) {
964 * The above condition must be (bytes <= size), not
965 * (bytes < size). In other words, even though we
966 * expect no more output and set avail_out to zero,
967 * the input zlib stream may have bytes that express
968 * "this concludes the stream", and we *do* want to
969 * eat that input.
971 * Otherwise we would not be able to test that we
972 * consumed all the input to reach the expected size;
973 * we also want to check that zlib tells us that all
974 * went well with status == Z_STREAM_END at the end.
976 stream->next_out = buf + bytes;
977 stream->avail_out = size - bytes;
978 while (status == Z_OK)
979 status = git_inflate(stream, Z_FINISH);
981 if (status == Z_STREAM_END && !stream->avail_in) {
982 git_inflate_end(stream);
983 return buf;
986 if (status < 0)
987 error("corrupt loose object '%s'", sha1_to_hex(sha1));
988 else if (stream->avail_in)
989 error("garbage at end of loose object '%s'",
990 sha1_to_hex(sha1));
991 free(buf);
992 return NULL;
996 * We used to just use "sscanf()", but that's actually way
997 * too permissive for what we want to check. So do an anal
998 * object header parse by hand.
1000 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1001 unsigned int flags)
1003 const char *type_buf = hdr;
1004 unsigned long size;
1005 int type, type_len = 0;
1008 * The type can be of any size but is followed by
1009 * a space.
1011 for (;;) {
1012 char c = *hdr++;
1013 if (!c)
1014 return -1;
1015 if (c == ' ')
1016 break;
1017 type_len++;
1020 type = type_from_string_gently(type_buf, type_len, 1);
1021 if (oi->typename)
1022 strbuf_add(oi->typename, type_buf, type_len);
1024 * Set type to 0 if its an unknown object and
1025 * we're obtaining the type using '--allow-unknown-type'
1026 * option.
1028 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1029 type = 0;
1030 else if (type < 0)
1031 die("invalid object type");
1032 if (oi->typep)
1033 *oi->typep = type;
1036 * The length must follow immediately, and be in canonical
1037 * decimal format (ie "010" is not valid).
1039 size = *hdr++ - '0';
1040 if (size > 9)
1041 return -1;
1042 if (size) {
1043 for (;;) {
1044 unsigned long c = *hdr - '0';
1045 if (c > 9)
1046 break;
1047 hdr++;
1048 size = size * 10 + c;
1052 if (oi->sizep)
1053 *oi->sizep = size;
1056 * The length must be followed by a zero byte
1058 return *hdr ? -1 : type;
1061 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1063 struct object_info oi = OBJECT_INFO_INIT;
1065 oi.sizep = sizep;
1066 return parse_sha1_header_extended(hdr, &oi, 0);
1069 static int sha1_loose_object_info(const unsigned char *sha1,
1070 struct object_info *oi,
1071 int flags)
1073 int status = 0;
1074 unsigned long mapsize;
1075 void *map;
1076 git_zstream stream;
1077 char hdr[32];
1078 struct strbuf hdrbuf = STRBUF_INIT;
1079 unsigned long size_scratch;
1081 if (oi->delta_base_sha1)
1082 hashclr(oi->delta_base_sha1);
1085 * If we don't care about type or size, then we don't
1086 * need to look inside the object at all. Note that we
1087 * do not optimize out the stat call, even if the
1088 * caller doesn't care about the disk-size, since our
1089 * return value implicitly indicates whether the
1090 * object even exists.
1092 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1093 const char *path;
1094 struct stat st;
1095 if (stat_sha1_file(sha1, &st, &path) < 0)
1096 return -1;
1097 if (oi->disk_sizep)
1098 *oi->disk_sizep = st.st_size;
1099 return 0;
1102 map = map_sha1_file(sha1, &mapsize);
1103 if (!map)
1104 return -1;
1106 if (!oi->sizep)
1107 oi->sizep = &size_scratch;
1109 if (oi->disk_sizep)
1110 *oi->disk_sizep = mapsize;
1111 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1112 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1113 status = error("unable to unpack %s header with --allow-unknown-type",
1114 sha1_to_hex(sha1));
1115 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1116 status = error("unable to unpack %s header",
1117 sha1_to_hex(sha1));
1118 if (status < 0)
1119 ; /* Do nothing */
1120 else if (hdrbuf.len) {
1121 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1122 status = error("unable to parse %s header with --allow-unknown-type",
1123 sha1_to_hex(sha1));
1124 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1125 status = error("unable to parse %s header", sha1_to_hex(sha1));
1127 if (status >= 0 && oi->contentp) {
1128 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1129 *oi->sizep, sha1);
1130 if (!*oi->contentp) {
1131 git_inflate_end(&stream);
1132 status = -1;
1134 } else
1135 git_inflate_end(&stream);
1137 munmap(map, mapsize);
1138 if (status && oi->typep)
1139 *oi->typep = status;
1140 if (oi->sizep == &size_scratch)
1141 oi->sizep = NULL;
1142 strbuf_release(&hdrbuf);
1143 oi->whence = OI_LOOSE;
1144 return (status < 0) ? status : 0;
1147 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1149 static struct object_info blank_oi = OBJECT_INFO_INIT;
1150 struct pack_entry e;
1151 int rtype;
1152 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1153 lookup_replace_object(sha1) :
1154 sha1;
1156 if (!oi)
1157 oi = &blank_oi;
1159 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1160 struct cached_object *co = find_cached_object(real);
1161 if (co) {
1162 if (oi->typep)
1163 *(oi->typep) = co->type;
1164 if (oi->sizep)
1165 *(oi->sizep) = co->size;
1166 if (oi->disk_sizep)
1167 *(oi->disk_sizep) = 0;
1168 if (oi->delta_base_sha1)
1169 hashclr(oi->delta_base_sha1);
1170 if (oi->typename)
1171 strbuf_addstr(oi->typename, typename(co->type));
1172 if (oi->contentp)
1173 *oi->contentp = xmemdupz(co->buf, co->size);
1174 oi->whence = OI_CACHED;
1175 return 0;
1179 if (!find_pack_entry(real, &e)) {
1180 /* Most likely it's a loose object. */
1181 if (!sha1_loose_object_info(real, oi, flags))
1182 return 0;
1184 /* Not a loose object; someone else may have just packed it. */
1185 if (flags & OBJECT_INFO_QUICK) {
1186 return -1;
1187 } else {
1188 reprepare_packed_git();
1189 if (!find_pack_entry(real, &e))
1190 return -1;
1194 if (oi == &blank_oi)
1196 * We know that the caller doesn't actually need the
1197 * information below, so return early.
1199 return 0;
1201 rtype = packed_object_info(e.p, e.offset, oi);
1202 if (rtype < 0) {
1203 mark_bad_packed_object(e.p, real);
1204 return sha1_object_info_extended(real, oi, 0);
1205 } else if (oi->whence == OI_PACKED) {
1206 oi->u.packed.offset = e.offset;
1207 oi->u.packed.pack = e.p;
1208 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1209 rtype == OBJ_OFS_DELTA);
1212 return 0;
1215 /* returns enum object_type or negative */
1216 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1218 enum object_type type;
1219 struct object_info oi = OBJECT_INFO_INIT;
1221 oi.typep = &type;
1222 oi.sizep = sizep;
1223 if (sha1_object_info_extended(sha1, &oi,
1224 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1225 return -1;
1226 return type;
1229 static void *read_object(const unsigned char *sha1, enum object_type *type,
1230 unsigned long *size)
1232 struct object_info oi = OBJECT_INFO_INIT;
1233 void *content;
1234 oi.typep = type;
1235 oi.sizep = size;
1236 oi.contentp = &content;
1238 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1239 return NULL;
1240 return content;
1243 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1244 unsigned char *sha1)
1246 struct cached_object *co;
1248 hash_sha1_file(buf, len, typename(type), sha1);
1249 if (has_sha1_file(sha1) || find_cached_object(sha1))
1250 return 0;
1251 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1252 co = &cached_objects[cached_object_nr++];
1253 co->size = len;
1254 co->type = type;
1255 co->buf = xmalloc(len);
1256 memcpy(co->buf, buf, len);
1257 hashcpy(co->sha1, sha1);
1258 return 0;
1262 * This function dies on corrupt objects; the callers who want to
1263 * deal with them should arrange to call read_object() and give error
1264 * messages themselves.
1266 void *read_sha1_file_extended(const unsigned char *sha1,
1267 enum object_type *type,
1268 unsigned long *size,
1269 int lookup_replace)
1271 void *data;
1272 const struct packed_git *p;
1273 const char *path;
1274 struct stat st;
1275 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1276 : sha1;
1278 errno = 0;
1279 data = read_object(repl, type, size);
1280 if (data)
1281 return data;
1283 if (errno && errno != ENOENT)
1284 die_errno("failed to read object %s", sha1_to_hex(sha1));
1286 /* die if we replaced an object with one that does not exist */
1287 if (repl != sha1)
1288 die("replacement %s not found for %s",
1289 sha1_to_hex(repl), sha1_to_hex(sha1));
1291 if (!stat_sha1_file(repl, &st, &path))
1292 die("loose object %s (stored in %s) is corrupt",
1293 sha1_to_hex(repl), path);
1295 if ((p = has_packed_and_bad(repl)) != NULL)
1296 die("packed object %s (stored in %s) is corrupt",
1297 sha1_to_hex(repl), p->pack_name);
1299 return NULL;
1302 void *read_object_with_reference(const unsigned char *sha1,
1303 const char *required_type_name,
1304 unsigned long *size,
1305 unsigned char *actual_sha1_return)
1307 enum object_type type, required_type;
1308 void *buffer;
1309 unsigned long isize;
1310 unsigned char actual_sha1[20];
1312 required_type = type_from_string(required_type_name);
1313 hashcpy(actual_sha1, sha1);
1314 while (1) {
1315 int ref_length = -1;
1316 const char *ref_type = NULL;
1318 buffer = read_sha1_file(actual_sha1, &type, &isize);
1319 if (!buffer)
1320 return NULL;
1321 if (type == required_type) {
1322 *size = isize;
1323 if (actual_sha1_return)
1324 hashcpy(actual_sha1_return, actual_sha1);
1325 return buffer;
1327 /* Handle references */
1328 else if (type == OBJ_COMMIT)
1329 ref_type = "tree ";
1330 else if (type == OBJ_TAG)
1331 ref_type = "object ";
1332 else {
1333 free(buffer);
1334 return NULL;
1336 ref_length = strlen(ref_type);
1338 if (ref_length + 40 > isize ||
1339 memcmp(buffer, ref_type, ref_length) ||
1340 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1341 free(buffer);
1342 return NULL;
1344 free(buffer);
1345 /* Now we have the ID of the referred-to object in
1346 * actual_sha1. Check again. */
1350 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1351 const char *type, unsigned char *sha1,
1352 char *hdr, int *hdrlen)
1354 git_SHA_CTX c;
1356 /* Generate the header */
1357 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1359 /* Sha1.. */
1360 git_SHA1_Init(&c);
1361 git_SHA1_Update(&c, hdr, *hdrlen);
1362 git_SHA1_Update(&c, buf, len);
1363 git_SHA1_Final(sha1, &c);
1367 * Move the just written object into its final resting place.
1369 int finalize_object_file(const char *tmpfile, const char *filename)
1371 int ret = 0;
1373 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1374 goto try_rename;
1375 else if (link(tmpfile, filename))
1376 ret = errno;
1379 * Coda hack - coda doesn't like cross-directory links,
1380 * so we fall back to a rename, which will mean that it
1381 * won't be able to check collisions, but that's not a
1382 * big deal.
1384 * The same holds for FAT formatted media.
1386 * When this succeeds, we just return. We have nothing
1387 * left to unlink.
1389 if (ret && ret != EEXIST) {
1390 try_rename:
1391 if (!rename(tmpfile, filename))
1392 goto out;
1393 ret = errno;
1395 unlink_or_warn(tmpfile);
1396 if (ret) {
1397 if (ret != EEXIST) {
1398 return error_errno("unable to write sha1 filename %s", filename);
1400 /* FIXME!!! Collision check here ? */
1403 out:
1404 if (adjust_shared_perm(filename))
1405 return error("unable to set permission to '%s'", filename);
1406 return 0;
1409 static int write_buffer(int fd, const void *buf, size_t len)
1411 if (write_in_full(fd, buf, len) < 0)
1412 return error_errno("file write error");
1413 return 0;
1416 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1417 unsigned char *sha1)
1419 char hdr[32];
1420 int hdrlen = sizeof(hdr);
1421 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1422 return 0;
1425 /* Finalize a file on disk, and close it. */
1426 static void close_sha1_file(int fd)
1428 if (fsync_object_files)
1429 fsync_or_die(fd, "sha1 file");
1430 if (close(fd) != 0)
1431 die_errno("error when closing sha1 file");
1434 /* Size of directory component, including the ending '/' */
1435 static inline int directory_size(const char *filename)
1437 const char *s = strrchr(filename, '/');
1438 if (!s)
1439 return 0;
1440 return s - filename + 1;
1444 * This creates a temporary file in the same directory as the final
1445 * 'filename'
1447 * We want to avoid cross-directory filename renames, because those
1448 * can have problems on various filesystems (FAT, NFS, Coda).
1450 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1452 int fd, dirlen = directory_size(filename);
1454 strbuf_reset(tmp);
1455 strbuf_add(tmp, filename, dirlen);
1456 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1457 fd = git_mkstemp_mode(tmp->buf, 0444);
1458 if (fd < 0 && dirlen && errno == ENOENT) {
1460 * Make sure the directory exists; note that the contents
1461 * of the buffer are undefined after mkstemp returns an
1462 * error, so we have to rewrite the whole buffer from
1463 * scratch.
1465 strbuf_reset(tmp);
1466 strbuf_add(tmp, filename, dirlen - 1);
1467 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1468 return -1;
1469 if (adjust_shared_perm(tmp->buf))
1470 return -1;
1472 /* Try again */
1473 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1474 fd = git_mkstemp_mode(tmp->buf, 0444);
1476 return fd;
1479 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1480 const void *buf, unsigned long len, time_t mtime)
1482 int fd, ret;
1483 unsigned char compressed[4096];
1484 git_zstream stream;
1485 git_SHA_CTX c;
1486 unsigned char parano_sha1[20];
1487 static struct strbuf tmp_file = STRBUF_INIT;
1488 const char *filename = sha1_file_name(sha1);
1490 fd = create_tmpfile(&tmp_file, filename);
1491 if (fd < 0) {
1492 if (errno == EACCES)
1493 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1494 else
1495 return error_errno("unable to create temporary file");
1498 /* Set it up */
1499 git_deflate_init(&stream, zlib_compression_level);
1500 stream.next_out = compressed;
1501 stream.avail_out = sizeof(compressed);
1502 git_SHA1_Init(&c);
1504 /* First header.. */
1505 stream.next_in = (unsigned char *)hdr;
1506 stream.avail_in = hdrlen;
1507 while (git_deflate(&stream, 0) == Z_OK)
1508 ; /* nothing */
1509 git_SHA1_Update(&c, hdr, hdrlen);
1511 /* Then the data itself.. */
1512 stream.next_in = (void *)buf;
1513 stream.avail_in = len;
1514 do {
1515 unsigned char *in0 = stream.next_in;
1516 ret = git_deflate(&stream, Z_FINISH);
1517 git_SHA1_Update(&c, in0, stream.next_in - in0);
1518 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1519 die("unable to write sha1 file");
1520 stream.next_out = compressed;
1521 stream.avail_out = sizeof(compressed);
1522 } while (ret == Z_OK);
1524 if (ret != Z_STREAM_END)
1525 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1526 ret = git_deflate_end_gently(&stream);
1527 if (ret != Z_OK)
1528 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1529 git_SHA1_Final(parano_sha1, &c);
1530 if (hashcmp(sha1, parano_sha1) != 0)
1531 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1533 close_sha1_file(fd);
1535 if (mtime) {
1536 struct utimbuf utb;
1537 utb.actime = mtime;
1538 utb.modtime = mtime;
1539 if (utime(tmp_file.buf, &utb) < 0)
1540 warning_errno("failed utime() on %s", tmp_file.buf);
1543 return finalize_object_file(tmp_file.buf, filename);
1546 static int freshen_loose_object(const unsigned char *sha1)
1548 return check_and_freshen(sha1, 1);
1551 static int freshen_packed_object(const unsigned char *sha1)
1553 struct pack_entry e;
1554 if (!find_pack_entry(sha1, &e))
1555 return 0;
1556 if (e.p->freshened)
1557 return 1;
1558 if (!freshen_file(e.p->pack_name))
1559 return 0;
1560 e.p->freshened = 1;
1561 return 1;
1564 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1566 char hdr[32];
1567 int hdrlen = sizeof(hdr);
1569 /* Normally if we have it in the pack then we do not bother writing
1570 * it out into .git/objects/??/?{38} file.
1572 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1573 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1574 return 0;
1575 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1578 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1579 struct object_id *oid, unsigned flags)
1581 char *header;
1582 int hdrlen, status = 0;
1584 /* type string, SP, %lu of the length plus NUL must fit this */
1585 hdrlen = strlen(type) + 32;
1586 header = xmalloc(hdrlen);
1587 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1589 if (!(flags & HASH_WRITE_OBJECT))
1590 goto cleanup;
1591 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1592 goto cleanup;
1593 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1595 cleanup:
1596 free(header);
1597 return status;
1600 int force_object_loose(const unsigned char *sha1, time_t mtime)
1602 void *buf;
1603 unsigned long len;
1604 enum object_type type;
1605 char hdr[32];
1606 int hdrlen;
1607 int ret;
1609 if (has_loose_object(sha1))
1610 return 0;
1611 buf = read_object(sha1, &type, &len);
1612 if (!buf)
1613 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1614 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1615 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1616 free(buf);
1618 return ret;
1621 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1623 if (!startup_info->have_repository)
1624 return 0;
1625 return sha1_object_info_extended(sha1, NULL,
1626 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1629 int has_object_file(const struct object_id *oid)
1631 return has_sha1_file(oid->hash);
1634 int has_object_file_with_flags(const struct object_id *oid, int flags)
1636 return has_sha1_file_with_flags(oid->hash, flags);
1639 static void check_tree(const void *buf, size_t size)
1641 struct tree_desc desc;
1642 struct name_entry entry;
1644 init_tree_desc(&desc, buf, size);
1645 while (tree_entry(&desc, &entry))
1646 /* do nothing
1647 * tree_entry() will die() on malformed entries */
1651 static void check_commit(const void *buf, size_t size)
1653 struct commit c;
1654 memset(&c, 0, sizeof(c));
1655 if (parse_commit_buffer(&c, buf, size))
1656 die("corrupt commit");
1659 static void check_tag(const void *buf, size_t size)
1661 struct tag t;
1662 memset(&t, 0, sizeof(t));
1663 if (parse_tag_buffer(&t, buf, size))
1664 die("corrupt tag");
1667 static int index_mem(unsigned char *sha1, void *buf, size_t size,
1668 enum object_type type,
1669 const char *path, unsigned flags)
1671 int ret, re_allocated = 0;
1672 int write_object = flags & HASH_WRITE_OBJECT;
1674 if (!type)
1675 type = OBJ_BLOB;
1678 * Convert blobs to git internal format
1680 if ((type == OBJ_BLOB) && path) {
1681 struct strbuf nbuf = STRBUF_INIT;
1682 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1683 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1684 buf = strbuf_detach(&nbuf, &size);
1685 re_allocated = 1;
1688 if (flags & HASH_FORMAT_CHECK) {
1689 if (type == OBJ_TREE)
1690 check_tree(buf, size);
1691 if (type == OBJ_COMMIT)
1692 check_commit(buf, size);
1693 if (type == OBJ_TAG)
1694 check_tag(buf, size);
1697 if (write_object)
1698 ret = write_sha1_file(buf, size, typename(type), sha1);
1699 else
1700 ret = hash_sha1_file(buf, size, typename(type), sha1);
1701 if (re_allocated)
1702 free(buf);
1703 return ret;
1706 static int index_stream_convert_blob(unsigned char *sha1, int fd,
1707 const char *path, unsigned flags)
1709 int ret;
1710 const int write_object = flags & HASH_WRITE_OBJECT;
1711 struct strbuf sbuf = STRBUF_INIT;
1713 assert(path);
1714 assert(would_convert_to_git_filter_fd(path));
1716 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1717 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1719 if (write_object)
1720 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1721 sha1);
1722 else
1723 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1724 sha1);
1725 strbuf_release(&sbuf);
1726 return ret;
1729 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
1730 const char *path, unsigned flags)
1732 struct strbuf sbuf = STRBUF_INIT;
1733 int ret;
1735 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1736 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
1737 else
1738 ret = -1;
1739 strbuf_release(&sbuf);
1740 return ret;
1743 #define SMALL_FILE_SIZE (32*1024)
1745 static int index_core(unsigned char *sha1, int fd, size_t size,
1746 enum object_type type, const char *path,
1747 unsigned flags)
1749 int ret;
1751 if (!size) {
1752 ret = index_mem(sha1, "", size, type, path, flags);
1753 } else if (size <= SMALL_FILE_SIZE) {
1754 char *buf = xmalloc(size);
1755 ssize_t read_result = read_in_full(fd, buf, size);
1756 if (read_result < 0)
1757 ret = error_errno("read error while indexing %s",
1758 path ? path : "<unknown>");
1759 else if (read_result != size)
1760 ret = error("short read while indexing %s",
1761 path ? path : "<unknown>");
1762 else
1763 ret = index_mem(sha1, buf, size, type, path, flags);
1764 free(buf);
1765 } else {
1766 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1767 ret = index_mem(sha1, buf, size, type, path, flags);
1768 munmap(buf, size);
1770 return ret;
1774 * This creates one packfile per large blob unless bulk-checkin
1775 * machinery is "plugged".
1777 * This also bypasses the usual "convert-to-git" dance, and that is on
1778 * purpose. We could write a streaming version of the converting
1779 * functions and insert that before feeding the data to fast-import
1780 * (or equivalent in-core API described above). However, that is
1781 * somewhat complicated, as we do not know the size of the filter
1782 * result, which we need to know beforehand when writing a git object.
1783 * Since the primary motivation for trying to stream from the working
1784 * tree file and to avoid mmaping it in core is to deal with large
1785 * binary blobs, they generally do not want to get any conversion, and
1786 * callers should avoid this code path when filters are requested.
1788 static int index_stream(struct object_id *oid, int fd, size_t size,
1789 enum object_type type, const char *path,
1790 unsigned flags)
1792 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1795 int index_fd(struct object_id *oid, int fd, struct stat *st,
1796 enum object_type type, const char *path, unsigned flags)
1798 int ret;
1801 * Call xsize_t() only when needed to avoid potentially unnecessary
1802 * die() for large files.
1804 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1805 ret = index_stream_convert_blob(oid->hash, fd, path, flags);
1806 else if (!S_ISREG(st->st_mode))
1807 ret = index_pipe(oid->hash, fd, type, path, flags);
1808 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1809 (path && would_convert_to_git(&the_index, path)))
1810 ret = index_core(oid->hash, fd, xsize_t(st->st_size), type, path,
1811 flags);
1812 else
1813 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1814 flags);
1815 close(fd);
1816 return ret;
1819 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1821 int fd;
1822 struct strbuf sb = STRBUF_INIT;
1823 int rc = 0;
1825 switch (st->st_mode & S_IFMT) {
1826 case S_IFREG:
1827 fd = open(path, O_RDONLY);
1828 if (fd < 0)
1829 return error_errno("open(\"%s\")", path);
1830 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1831 return error("%s: failed to insert into database",
1832 path);
1833 break;
1834 case S_IFLNK:
1835 if (strbuf_readlink(&sb, path, st->st_size))
1836 return error_errno("readlink(\"%s\")", path);
1837 if (!(flags & HASH_WRITE_OBJECT))
1838 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1839 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1840 rc = error("%s: failed to insert into database", path);
1841 strbuf_release(&sb);
1842 break;
1843 case S_IFDIR:
1844 return resolve_gitlink_ref(path, "HEAD", oid->hash);
1845 default:
1846 return error("%s: unsupported file type", path);
1848 return rc;
1851 int read_pack_header(int fd, struct pack_header *header)
1853 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1854 /* "eof before pack header was fully read" */
1855 return PH_ERROR_EOF;
1857 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1858 /* "protocol error (pack signature mismatch detected)" */
1859 return PH_ERROR_PACK_SIGNATURE;
1860 if (!pack_version_ok(header->hdr_version))
1861 /* "protocol error (pack version unsupported)" */
1862 return PH_ERROR_PROTOCOL;
1863 return 0;
1866 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1868 enum object_type type = sha1_object_info(sha1, NULL);
1869 if (type < 0)
1870 die("%s is not a valid object", sha1_to_hex(sha1));
1871 if (type != expect)
1872 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1873 typename(expect));
1876 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1877 struct strbuf *path,
1878 each_loose_object_fn obj_cb,
1879 each_loose_cruft_fn cruft_cb,
1880 each_loose_subdir_fn subdir_cb,
1881 void *data)
1883 size_t origlen, baselen;
1884 DIR *dir;
1885 struct dirent *de;
1886 int r = 0;
1888 if (subdir_nr > 0xff)
1889 BUG("invalid loose object subdirectory: %x", subdir_nr);
1891 origlen = path->len;
1892 strbuf_complete(path, '/');
1893 strbuf_addf(path, "%02x", subdir_nr);
1894 baselen = path->len;
1896 dir = opendir(path->buf);
1897 if (!dir) {
1898 if (errno != ENOENT)
1899 r = error_errno("unable to open %s", path->buf);
1900 strbuf_setlen(path, origlen);
1901 return r;
1904 while ((de = readdir(dir))) {
1905 if (is_dot_or_dotdot(de->d_name))
1906 continue;
1908 strbuf_setlen(path, baselen);
1909 strbuf_addf(path, "/%s", de->d_name);
1911 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
1912 char hex[GIT_MAX_HEXSZ+1];
1913 struct object_id oid;
1915 xsnprintf(hex, sizeof(hex), "%02x%s",
1916 subdir_nr, de->d_name);
1917 if (!get_oid_hex(hex, &oid)) {
1918 if (obj_cb) {
1919 r = obj_cb(&oid, path->buf, data);
1920 if (r)
1921 break;
1923 continue;
1927 if (cruft_cb) {
1928 r = cruft_cb(de->d_name, path->buf, data);
1929 if (r)
1930 break;
1933 closedir(dir);
1935 strbuf_setlen(path, baselen);
1936 if (!r && subdir_cb)
1937 r = subdir_cb(subdir_nr, path->buf, data);
1939 strbuf_setlen(path, origlen);
1941 return r;
1944 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
1945 each_loose_object_fn obj_cb,
1946 each_loose_cruft_fn cruft_cb,
1947 each_loose_subdir_fn subdir_cb,
1948 void *data)
1950 int r = 0;
1951 int i;
1953 for (i = 0; i < 256; i++) {
1954 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
1955 subdir_cb, data);
1956 if (r)
1957 break;
1960 return r;
1963 int for_each_loose_file_in_objdir(const char *path,
1964 each_loose_object_fn obj_cb,
1965 each_loose_cruft_fn cruft_cb,
1966 each_loose_subdir_fn subdir_cb,
1967 void *data)
1969 struct strbuf buf = STRBUF_INIT;
1970 int r;
1972 strbuf_addstr(&buf, path);
1973 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
1974 subdir_cb, data);
1975 strbuf_release(&buf);
1977 return r;
1980 struct loose_alt_odb_data {
1981 each_loose_object_fn *cb;
1982 void *data;
1985 static int loose_from_alt_odb(struct alternate_object_database *alt,
1986 void *vdata)
1988 struct loose_alt_odb_data *data = vdata;
1989 struct strbuf buf = STRBUF_INIT;
1990 int r;
1992 strbuf_addstr(&buf, alt->path);
1993 r = for_each_loose_file_in_objdir_buf(&buf,
1994 data->cb, NULL, NULL,
1995 data->data);
1996 strbuf_release(&buf);
1997 return r;
2000 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2002 struct loose_alt_odb_data alt;
2003 int r;
2005 r = for_each_loose_file_in_objdir(get_object_directory(),
2006 cb, NULL, NULL, data);
2007 if (r)
2008 return r;
2010 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2011 return 0;
2013 alt.cb = cb;
2014 alt.data = data;
2015 return foreach_alt_odb(loose_from_alt_odb, &alt);
2018 static int check_stream_sha1(git_zstream *stream,
2019 const char *hdr,
2020 unsigned long size,
2021 const char *path,
2022 const unsigned char *expected_sha1)
2024 git_SHA_CTX c;
2025 unsigned char real_sha1[GIT_MAX_RAWSZ];
2026 unsigned char buf[4096];
2027 unsigned long total_read;
2028 int status = Z_OK;
2030 git_SHA1_Init(&c);
2031 git_SHA1_Update(&c, hdr, stream->total_out);
2034 * We already read some bytes into hdr, but the ones up to the NUL
2035 * do not count against the object's content size.
2037 total_read = stream->total_out - strlen(hdr) - 1;
2040 * This size comparison must be "<=" to read the final zlib packets;
2041 * see the comment in unpack_sha1_rest for details.
2043 while (total_read <= size &&
2044 (status == Z_OK || status == Z_BUF_ERROR)) {
2045 stream->next_out = buf;
2046 stream->avail_out = sizeof(buf);
2047 if (size - total_read < stream->avail_out)
2048 stream->avail_out = size - total_read;
2049 status = git_inflate(stream, Z_FINISH);
2050 git_SHA1_Update(&c, buf, stream->next_out - buf);
2051 total_read += stream->next_out - buf;
2053 git_inflate_end(stream);
2055 if (status != Z_STREAM_END) {
2056 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2057 return -1;
2059 if (stream->avail_in) {
2060 error("garbage at end of loose object '%s'",
2061 sha1_to_hex(expected_sha1));
2062 return -1;
2065 git_SHA1_Final(real_sha1, &c);
2066 if (hashcmp(expected_sha1, real_sha1)) {
2067 error("sha1 mismatch for %s (expected %s)", path,
2068 sha1_to_hex(expected_sha1));
2069 return -1;
2072 return 0;
2075 int read_loose_object(const char *path,
2076 const unsigned char *expected_sha1,
2077 enum object_type *type,
2078 unsigned long *size,
2079 void **contents)
2081 int ret = -1;
2082 void *map = NULL;
2083 unsigned long mapsize;
2084 git_zstream stream;
2085 char hdr[32];
2087 *contents = NULL;
2089 map = map_sha1_file_1(path, NULL, &mapsize);
2090 if (!map) {
2091 error_errno("unable to mmap %s", path);
2092 goto out;
2095 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2096 error("unable to unpack header of %s", path);
2097 goto out;
2100 *type = parse_sha1_header(hdr, size);
2101 if (*type < 0) {
2102 error("unable to parse header of %s", path);
2103 git_inflate_end(&stream);
2104 goto out;
2107 if (*type == OBJ_BLOB) {
2108 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2109 goto out;
2110 } else {
2111 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2112 if (!*contents) {
2113 error("unable to unpack contents of %s", path);
2114 git_inflate_end(&stream);
2115 goto out;
2117 if (check_sha1_signature(expected_sha1, *contents,
2118 *size, typename(*type))) {
2119 error("sha1 mismatch for %s (expected %s)", path,
2120 sha1_to_hex(expected_sha1));
2121 free(*contents);
2122 goto out;
2126 ret = 0; /* everything checks out */
2128 out:
2129 if (map)
2130 munmap(map, mapsize);
2131 return ret;