repository: introduce raw object store field
[alt-git.git] / sha1_file.c
blobaf5aa83db51a460581c86f82ca293de1b84d09e2
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 "list.h"
28 #include "mergesort.h"
29 #include "quote.h"
30 #include "packfile.h"
31 #include "fetch-object.h"
32 #include "object-store.h"
34 const unsigned char null_sha1[GIT_MAX_RAWSZ];
35 const struct object_id null_oid;
36 const struct object_id empty_tree_oid = {
37 EMPTY_TREE_SHA1_BIN_LITERAL
39 const struct object_id empty_blob_oid = {
40 EMPTY_BLOB_SHA1_BIN_LITERAL
43 static void git_hash_sha1_init(void *ctx)
45 git_SHA1_Init((git_SHA_CTX *)ctx);
48 static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
50 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
53 static void git_hash_sha1_final(unsigned char *hash, void *ctx)
55 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
58 static void git_hash_unknown_init(void *ctx)
60 die("trying to init unknown hash");
63 static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
65 die("trying to update unknown hash");
68 static void git_hash_unknown_final(unsigned char *hash, void *ctx)
70 die("trying to finalize unknown hash");
73 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
75 NULL,
76 0x00000000,
80 git_hash_unknown_init,
81 git_hash_unknown_update,
82 git_hash_unknown_final,
83 NULL,
84 NULL,
87 "sha-1",
88 /* "sha1", big-endian */
89 0x73686131,
90 sizeof(git_SHA_CTX),
91 GIT_SHA1_RAWSZ,
92 GIT_SHA1_HEXSZ,
93 git_hash_sha1_init,
94 git_hash_sha1_update,
95 git_hash_sha1_final,
96 &empty_tree_oid,
97 &empty_blob_oid,
102 * This is meant to hold a *small* number of objects that you would
103 * want read_sha1_file() to be able to return, but yet you do not want
104 * to write them into the object store (e.g. a browse-only
105 * application).
107 static struct cached_object {
108 unsigned char sha1[20];
109 enum object_type type;
110 void *buf;
111 unsigned long size;
112 } *cached_objects;
113 static int cached_object_nr, cached_object_alloc;
115 static struct cached_object empty_tree = {
116 EMPTY_TREE_SHA1_BIN_LITERAL,
117 OBJ_TREE,
122 static struct cached_object *find_cached_object(const unsigned char *sha1)
124 int i;
125 struct cached_object *co = cached_objects;
127 for (i = 0; i < cached_object_nr; i++, co++) {
128 if (!hashcmp(co->sha1, sha1))
129 return co;
131 if (!hashcmp(sha1, empty_tree.sha1))
132 return &empty_tree;
133 return NULL;
137 static int get_conv_flags(unsigned flags)
139 if (flags & HASH_RENORMALIZE)
140 return CONV_EOL_RENORMALIZE;
141 else if (flags & HASH_WRITE_OBJECT)
142 return global_conv_flags_eol;
143 else
144 return 0;
148 int mkdir_in_gitdir(const char *path)
150 if (mkdir(path, 0777)) {
151 int saved_errno = errno;
152 struct stat st;
153 struct strbuf sb = STRBUF_INIT;
155 if (errno != EEXIST)
156 return -1;
158 * Are we looking at a path in a symlinked worktree
159 * whose original repository does not yet have it?
160 * e.g. .git/rr-cache pointing at its original
161 * repository in which the user hasn't performed any
162 * conflict resolution yet?
164 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
165 strbuf_readlink(&sb, path, st.st_size) ||
166 !is_absolute_path(sb.buf) ||
167 mkdir(sb.buf, 0777)) {
168 strbuf_release(&sb);
169 errno = saved_errno;
170 return -1;
172 strbuf_release(&sb);
174 return adjust_shared_perm(path);
177 enum scld_error safe_create_leading_directories(char *path)
179 char *next_component = path + offset_1st_component(path);
180 enum scld_error ret = SCLD_OK;
182 while (ret == SCLD_OK && next_component) {
183 struct stat st;
184 char *slash = next_component, slash_character;
186 while (*slash && !is_dir_sep(*slash))
187 slash++;
189 if (!*slash)
190 break;
192 next_component = slash + 1;
193 while (is_dir_sep(*next_component))
194 next_component++;
195 if (!*next_component)
196 break;
198 slash_character = *slash;
199 *slash = '\0';
200 if (!stat(path, &st)) {
201 /* path exists */
202 if (!S_ISDIR(st.st_mode)) {
203 errno = ENOTDIR;
204 ret = SCLD_EXISTS;
206 } else if (mkdir(path, 0777)) {
207 if (errno == EEXIST &&
208 !stat(path, &st) && S_ISDIR(st.st_mode))
209 ; /* somebody created it since we checked */
210 else if (errno == ENOENT)
212 * Either mkdir() failed because
213 * somebody just pruned the containing
214 * directory, or stat() failed because
215 * the file that was in our way was
216 * just removed. Either way, inform
217 * the caller that it might be worth
218 * trying again:
220 ret = SCLD_VANISHED;
221 else
222 ret = SCLD_FAILED;
223 } else if (adjust_shared_perm(path)) {
224 ret = SCLD_PERMS;
226 *slash = slash_character;
228 return ret;
231 enum scld_error safe_create_leading_directories_const(const char *path)
233 int save_errno;
234 /* path points to cache entries, so xstrdup before messing with it */
235 char *buf = xstrdup(path);
236 enum scld_error result = safe_create_leading_directories(buf);
238 save_errno = errno;
239 free(buf);
240 errno = save_errno;
241 return result;
244 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
247 * The number of times we will try to remove empty directories
248 * in the way of path. This is only 1 because if another
249 * process is racily creating directories that conflict with
250 * us, we don't want to fight against them.
252 int remove_directories_remaining = 1;
255 * The number of times that we will try to create the
256 * directories containing path. We are willing to attempt this
257 * more than once, because another process could be trying to
258 * clean up empty directories at the same time as we are
259 * trying to create them.
261 int create_directories_remaining = 3;
263 /* A scratch copy of path, filled lazily if we need it: */
264 struct strbuf path_copy = STRBUF_INIT;
266 int ret, save_errno;
268 /* Sanity check: */
269 assert(*path);
271 retry_fn:
272 ret = fn(path, cb);
273 save_errno = errno;
274 if (!ret)
275 goto out;
277 if (errno == EISDIR && remove_directories_remaining-- > 0) {
279 * A directory is in the way. Maybe it is empty; try
280 * to remove it:
282 if (!path_copy.len)
283 strbuf_addstr(&path_copy, path);
285 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
286 goto retry_fn;
287 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
289 * Maybe the containing directory didn't exist, or
290 * maybe it was just deleted by a process that is
291 * racing with us to clean up empty directories. Try
292 * to create it:
294 enum scld_error scld_result;
296 if (!path_copy.len)
297 strbuf_addstr(&path_copy, path);
299 do {
300 scld_result = safe_create_leading_directories(path_copy.buf);
301 if (scld_result == SCLD_OK)
302 goto retry_fn;
303 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
306 out:
307 strbuf_release(&path_copy);
308 errno = save_errno;
309 return ret;
312 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
314 int i;
315 for (i = 0; i < 20; i++) {
316 static char hex[] = "0123456789abcdef";
317 unsigned int val = sha1[i];
318 strbuf_addch(buf, hex[val >> 4]);
319 strbuf_addch(buf, hex[val & 0xf]);
320 if (!i)
321 strbuf_addch(buf, '/');
325 void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
327 strbuf_addstr(buf, get_object_directory());
328 strbuf_addch(buf, '/');
329 fill_sha1_path(buf, sha1);
332 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
334 strbuf_setlen(&alt->scratch, alt->base_len);
335 return &alt->scratch;
338 static const char *alt_sha1_path(struct alternate_object_database *alt,
339 const unsigned char *sha1)
341 struct strbuf *buf = alt_scratch_buf(alt);
342 fill_sha1_path(buf, sha1);
343 return buf->buf;
346 struct alternate_object_database *alt_odb_list;
347 static struct alternate_object_database **alt_odb_tail;
350 * Return non-zero iff the path is usable as an alternate object database.
352 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
354 struct alternate_object_database *alt;
356 /* Detect cases where alternate disappeared */
357 if (!is_directory(path->buf)) {
358 error("object directory %s does not exist; "
359 "check .git/objects/info/alternates.",
360 path->buf);
361 return 0;
365 * Prevent the common mistake of listing the same
366 * thing twice, or object directory itself.
368 for (alt = alt_odb_list; alt; alt = alt->next) {
369 if (!fspathcmp(path->buf, alt->path))
370 return 0;
372 if (!fspathcmp(path->buf, normalized_objdir))
373 return 0;
375 return 1;
379 * Prepare alternate object database registry.
381 * The variable alt_odb_list points at the list of struct
382 * alternate_object_database. The elements on this list come from
383 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
384 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
385 * whose contents is similar to that environment variable but can be
386 * LF separated. Its base points at a statically allocated buffer that
387 * contains "/the/directory/corresponding/to/.git/objects/...", while
388 * its name points just after the slash at the end of ".git/objects/"
389 * in the example above, and has enough space to hold 40-byte hex
390 * SHA1, an extra slash for the first level indirection, and the
391 * terminating NUL.
393 static void read_info_alternates(const char * relative_base, int depth);
394 static int link_alt_odb_entry(const char *entry, const char *relative_base,
395 int depth, const char *normalized_objdir)
397 struct alternate_object_database *ent;
398 struct strbuf pathbuf = STRBUF_INIT;
400 if (!is_absolute_path(entry) && relative_base) {
401 strbuf_realpath(&pathbuf, relative_base, 1);
402 strbuf_addch(&pathbuf, '/');
404 strbuf_addstr(&pathbuf, entry);
406 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
407 error("unable to normalize alternate object path: %s",
408 pathbuf.buf);
409 strbuf_release(&pathbuf);
410 return -1;
414 * The trailing slash after the directory name is given by
415 * this function at the end. Remove duplicates.
417 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
418 strbuf_setlen(&pathbuf, pathbuf.len - 1);
420 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
421 strbuf_release(&pathbuf);
422 return -1;
425 ent = alloc_alt_odb(pathbuf.buf);
427 /* add the alternate entry */
428 *alt_odb_tail = ent;
429 alt_odb_tail = &(ent->next);
430 ent->next = NULL;
432 /* recursively add alternates */
433 read_info_alternates(pathbuf.buf, depth + 1);
435 strbuf_release(&pathbuf);
436 return 0;
439 static const char *parse_alt_odb_entry(const char *string,
440 int sep,
441 struct strbuf *out)
443 const char *end;
445 strbuf_reset(out);
447 if (*string == '#') {
448 /* comment; consume up to next separator */
449 end = strchrnul(string, sep);
450 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
452 * quoted path; unquote_c_style has copied the
453 * data for us and set "end". Broken quoting (e.g.,
454 * an entry that doesn't end with a quote) falls
455 * back to the unquoted case below.
457 } else {
458 /* normal, unquoted path */
459 end = strchrnul(string, sep);
460 strbuf_add(out, string, end - string);
463 if (*end)
464 end++;
465 return end;
468 static void link_alt_odb_entries(const char *alt, int sep,
469 const char *relative_base, int depth)
471 struct strbuf objdirbuf = STRBUF_INIT;
472 struct strbuf entry = STRBUF_INIT;
474 if (!alt || !*alt)
475 return;
477 if (depth > 5) {
478 error("%s: ignoring alternate object stores, nesting too deep.",
479 relative_base);
480 return;
483 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
484 if (strbuf_normalize_path(&objdirbuf) < 0)
485 die("unable to normalize object directory: %s",
486 objdirbuf.buf);
488 while (*alt) {
489 alt = parse_alt_odb_entry(alt, sep, &entry);
490 if (!entry.len)
491 continue;
492 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
494 strbuf_release(&entry);
495 strbuf_release(&objdirbuf);
498 static void read_info_alternates(const char * relative_base, int depth)
500 char *path;
501 struct strbuf buf = STRBUF_INIT;
503 path = xstrfmt("%s/info/alternates", relative_base);
504 if (strbuf_read_file(&buf, path, 1024) < 0) {
505 warn_on_fopen_errors(path);
506 free(path);
507 return;
510 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
511 strbuf_release(&buf);
512 free(path);
515 struct alternate_object_database *alloc_alt_odb(const char *dir)
517 struct alternate_object_database *ent;
519 FLEX_ALLOC_STR(ent, path, dir);
520 strbuf_init(&ent->scratch, 0);
521 strbuf_addf(&ent->scratch, "%s/", dir);
522 ent->base_len = ent->scratch.len;
524 return ent;
527 void add_to_alternates_file(const char *reference)
529 struct lock_file lock = LOCK_INIT;
530 char *alts = git_pathdup("objects/info/alternates");
531 FILE *in, *out;
532 int found = 0;
534 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
535 out = fdopen_lock_file(&lock, "w");
536 if (!out)
537 die_errno("unable to fdopen alternates lockfile");
539 in = fopen(alts, "r");
540 if (in) {
541 struct strbuf line = STRBUF_INIT;
543 while (strbuf_getline(&line, in) != EOF) {
544 if (!strcmp(reference, line.buf)) {
545 found = 1;
546 break;
548 fprintf_or_die(out, "%s\n", line.buf);
551 strbuf_release(&line);
552 fclose(in);
554 else if (errno != ENOENT)
555 die_errno("unable to read alternates file");
557 if (found) {
558 rollback_lock_file(&lock);
559 } else {
560 fprintf_or_die(out, "%s\n", reference);
561 if (commit_lock_file(&lock))
562 die_errno("unable to move new alternates file into place");
563 if (alt_odb_tail)
564 link_alt_odb_entries(reference, '\n', NULL, 0);
566 free(alts);
569 void add_to_alternates_memory(const char *reference)
572 * Make sure alternates are initialized, or else our entry may be
573 * overwritten when they are.
575 prepare_alt_odb();
577 link_alt_odb_entries(reference, '\n', NULL, 0);
581 * Compute the exact path an alternate is at and returns it. In case of
582 * error NULL is returned and the human readable error is added to `err`
583 * `path` may be relative and should point to $GITDIR.
584 * `err` must not be null.
586 char *compute_alternate_path(const char *path, struct strbuf *err)
588 char *ref_git = NULL;
589 const char *repo, *ref_git_s;
590 int seen_error = 0;
592 ref_git_s = real_path_if_valid(path);
593 if (!ref_git_s) {
594 seen_error = 1;
595 strbuf_addf(err, _("path '%s' does not exist"), path);
596 goto out;
597 } else
599 * Beware: read_gitfile(), real_path() and mkpath()
600 * return static buffer
602 ref_git = xstrdup(ref_git_s);
604 repo = read_gitfile(ref_git);
605 if (!repo)
606 repo = read_gitfile(mkpath("%s/.git", ref_git));
607 if (repo) {
608 free(ref_git);
609 ref_git = xstrdup(repo);
612 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
613 char *ref_git_git = mkpathdup("%s/.git", ref_git);
614 free(ref_git);
615 ref_git = ref_git_git;
616 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
617 struct strbuf sb = STRBUF_INIT;
618 seen_error = 1;
619 if (get_common_dir(&sb, ref_git)) {
620 strbuf_addf(err,
621 _("reference repository '%s' as a linked "
622 "checkout is not supported yet."),
623 path);
624 goto out;
627 strbuf_addf(err, _("reference repository '%s' is not a "
628 "local repository."), path);
629 goto out;
632 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
633 strbuf_addf(err, _("reference repository '%s' is shallow"),
634 path);
635 seen_error = 1;
636 goto out;
639 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
640 strbuf_addf(err,
641 _("reference repository '%s' is grafted"),
642 path);
643 seen_error = 1;
644 goto out;
647 out:
648 if (seen_error) {
649 FREE_AND_NULL(ref_git);
652 return ref_git;
655 int foreach_alt_odb(alt_odb_fn fn, void *cb)
657 struct alternate_object_database *ent;
658 int r = 0;
660 prepare_alt_odb();
661 for (ent = alt_odb_list; ent; ent = ent->next) {
662 r = fn(ent, cb);
663 if (r)
664 break;
666 return r;
669 void prepare_alt_odb(void)
671 if (alt_odb_tail)
672 return;
674 alt_odb_tail = &alt_odb_list;
675 link_alt_odb_entries(the_repository->objects->alternate_db,
676 PATH_SEP, NULL, 0);
678 read_info_alternates(get_object_directory(), 0);
681 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
682 static int freshen_file(const char *fn)
684 struct utimbuf t;
685 t.actime = t.modtime = time(NULL);
686 return !utime(fn, &t);
690 * All of the check_and_freshen functions return 1 if the file exists and was
691 * freshened (if freshening was requested), 0 otherwise. If they return
692 * 0, you should not assume that it is safe to skip a write of the object (it
693 * either does not exist on disk, or has a stale mtime and may be subject to
694 * pruning).
696 int check_and_freshen_file(const char *fn, int freshen)
698 if (access(fn, F_OK))
699 return 0;
700 if (freshen && !freshen_file(fn))
701 return 0;
702 return 1;
705 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
707 static struct strbuf buf = STRBUF_INIT;
709 strbuf_reset(&buf);
710 sha1_file_name(&buf, sha1);
712 return check_and_freshen_file(buf.buf, freshen);
715 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
717 struct alternate_object_database *alt;
718 prepare_alt_odb();
719 for (alt = alt_odb_list; alt; alt = alt->next) {
720 const char *path = alt_sha1_path(alt, sha1);
721 if (check_and_freshen_file(path, freshen))
722 return 1;
724 return 0;
727 static int check_and_freshen(const unsigned char *sha1, int freshen)
729 return check_and_freshen_local(sha1, freshen) ||
730 check_and_freshen_nonlocal(sha1, freshen);
733 int has_loose_object_nonlocal(const unsigned char *sha1)
735 return check_and_freshen_nonlocal(sha1, 0);
738 static int has_loose_object(const unsigned char *sha1)
740 return check_and_freshen(sha1, 0);
743 static void mmap_limit_check(size_t length)
745 static size_t limit = 0;
746 if (!limit) {
747 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
748 if (!limit)
749 limit = SIZE_MAX;
751 if (length > limit)
752 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
753 (uintmax_t)length, (uintmax_t)limit);
756 void *xmmap_gently(void *start, size_t length,
757 int prot, int flags, int fd, off_t offset)
759 void *ret;
761 mmap_limit_check(length);
762 ret = mmap(start, length, prot, flags, fd, offset);
763 if (ret == MAP_FAILED) {
764 if (!length)
765 return NULL;
766 release_pack_memory(length);
767 ret = mmap(start, length, prot, flags, fd, offset);
769 return ret;
772 void *xmmap(void *start, size_t length,
773 int prot, int flags, int fd, off_t offset)
775 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
776 if (ret == MAP_FAILED)
777 die_errno("mmap failed");
778 return ret;
782 * With an in-core object data in "map", rehash it to make sure the
783 * object name actually matches "sha1" to detect object corruption.
784 * With "map" == NULL, try reading the object named with "sha1" using
785 * the streaming interface and rehash it to do the same.
787 int check_sha1_signature(const unsigned char *sha1, void *map,
788 unsigned long size, const char *type)
790 unsigned char real_sha1[20];
791 enum object_type obj_type;
792 struct git_istream *st;
793 git_SHA_CTX c;
794 char hdr[32];
795 int hdrlen;
797 if (map) {
798 hash_sha1_file(map, size, type, real_sha1);
799 return hashcmp(sha1, real_sha1) ? -1 : 0;
802 st = open_istream(sha1, &obj_type, &size, NULL);
803 if (!st)
804 return -1;
806 /* Generate the header */
807 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
809 /* Sha1.. */
810 git_SHA1_Init(&c);
811 git_SHA1_Update(&c, hdr, hdrlen);
812 for (;;) {
813 char buf[1024 * 16];
814 ssize_t readlen = read_istream(st, buf, sizeof(buf));
816 if (readlen < 0) {
817 close_istream(st);
818 return -1;
820 if (!readlen)
821 break;
822 git_SHA1_Update(&c, buf, readlen);
824 git_SHA1_Final(real_sha1, &c);
825 close_istream(st);
826 return hashcmp(sha1, real_sha1) ? -1 : 0;
829 int git_open_cloexec(const char *name, int flags)
831 int fd;
832 static int o_cloexec = O_CLOEXEC;
834 fd = open(name, flags | o_cloexec);
835 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
836 /* Try again w/o O_CLOEXEC: the kernel might not support it */
837 o_cloexec &= ~O_CLOEXEC;
838 fd = open(name, flags | o_cloexec);
841 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
843 static int fd_cloexec = FD_CLOEXEC;
845 if (!o_cloexec && 0 <= fd && fd_cloexec) {
846 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
847 int flags = fcntl(fd, F_GETFD);
848 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
849 fd_cloexec = 0;
852 #endif
853 return fd;
857 * Find "sha1" as a loose object in the local repository or in an alternate.
858 * Returns 0 on success, negative on failure.
860 * The "path" out-parameter will give the path of the object we found (if any).
861 * Note that it may point to static storage and is only valid until another
862 * call to sha1_file_name(), etc.
864 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
865 const char **path)
867 struct alternate_object_database *alt;
868 static struct strbuf buf = STRBUF_INIT;
870 strbuf_reset(&buf);
871 sha1_file_name(&buf, sha1);
872 *path = buf.buf;
874 if (!lstat(*path, st))
875 return 0;
877 prepare_alt_odb();
878 errno = ENOENT;
879 for (alt = alt_odb_list; alt; alt = alt->next) {
880 *path = alt_sha1_path(alt, sha1);
881 if (!lstat(*path, st))
882 return 0;
885 return -1;
889 * Like stat_sha1_file(), but actually open the object and return the
890 * descriptor. See the caveats on the "path" parameter above.
892 static int open_sha1_file(const unsigned char *sha1, const char **path)
894 int fd;
895 struct alternate_object_database *alt;
896 int most_interesting_errno;
897 static struct strbuf buf = STRBUF_INIT;
899 strbuf_reset(&buf);
900 sha1_file_name(&buf, sha1);
901 *path = buf.buf;
903 fd = git_open(*path);
904 if (fd >= 0)
905 return fd;
906 most_interesting_errno = errno;
908 prepare_alt_odb();
909 for (alt = alt_odb_list; alt; alt = alt->next) {
910 *path = alt_sha1_path(alt, sha1);
911 fd = git_open(*path);
912 if (fd >= 0)
913 return fd;
914 if (most_interesting_errno == ENOENT)
915 most_interesting_errno = errno;
917 errno = most_interesting_errno;
918 return -1;
922 * Map the loose object at "path" if it is not NULL, or the path found by
923 * searching for a loose object named "sha1".
925 static void *map_sha1_file_1(const char *path,
926 const unsigned char *sha1,
927 unsigned long *size)
929 void *map;
930 int fd;
932 if (path)
933 fd = git_open(path);
934 else
935 fd = open_sha1_file(sha1, &path);
936 map = NULL;
937 if (fd >= 0) {
938 struct stat st;
940 if (!fstat(fd, &st)) {
941 *size = xsize_t(st.st_size);
942 if (!*size) {
943 /* mmap() is forbidden on empty files */
944 error("object file %s is empty", path);
945 return NULL;
947 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
949 close(fd);
951 return map;
954 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
956 return map_sha1_file_1(NULL, sha1, size);
959 static int unpack_sha1_short_header(git_zstream *stream,
960 unsigned char *map, unsigned long mapsize,
961 void *buffer, unsigned long bufsiz)
963 /* Get the data stream */
964 memset(stream, 0, sizeof(*stream));
965 stream->next_in = map;
966 stream->avail_in = mapsize;
967 stream->next_out = buffer;
968 stream->avail_out = bufsiz;
970 git_inflate_init(stream);
971 return git_inflate(stream, 0);
974 int unpack_sha1_header(git_zstream *stream,
975 unsigned char *map, unsigned long mapsize,
976 void *buffer, unsigned long bufsiz)
978 int status = unpack_sha1_short_header(stream, map, mapsize,
979 buffer, bufsiz);
981 if (status < Z_OK)
982 return status;
984 /* Make sure we have the terminating NUL */
985 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
986 return -1;
987 return 0;
990 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
991 unsigned long mapsize, void *buffer,
992 unsigned long bufsiz, struct strbuf *header)
994 int status;
996 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
997 if (status < Z_OK)
998 return -1;
1001 * Check if entire header is unpacked in the first iteration.
1003 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1004 return 0;
1007 * buffer[0..bufsiz] was not large enough. Copy the partial
1008 * result out to header, and then append the result of further
1009 * reading the stream.
1011 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1012 stream->next_out = buffer;
1013 stream->avail_out = bufsiz;
1015 do {
1016 status = git_inflate(stream, 0);
1017 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1018 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1019 return 0;
1020 stream->next_out = buffer;
1021 stream->avail_out = bufsiz;
1022 } while (status != Z_STREAM_END);
1023 return -1;
1026 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1028 int bytes = strlen(buffer) + 1;
1029 unsigned char *buf = xmallocz(size);
1030 unsigned long n;
1031 int status = Z_OK;
1033 n = stream->total_out - bytes;
1034 if (n > size)
1035 n = size;
1036 memcpy(buf, (char *) buffer + bytes, n);
1037 bytes = n;
1038 if (bytes <= size) {
1040 * The above condition must be (bytes <= size), not
1041 * (bytes < size). In other words, even though we
1042 * expect no more output and set avail_out to zero,
1043 * the input zlib stream may have bytes that express
1044 * "this concludes the stream", and we *do* want to
1045 * eat that input.
1047 * Otherwise we would not be able to test that we
1048 * consumed all the input to reach the expected size;
1049 * we also want to check that zlib tells us that all
1050 * went well with status == Z_STREAM_END at the end.
1052 stream->next_out = buf + bytes;
1053 stream->avail_out = size - bytes;
1054 while (status == Z_OK)
1055 status = git_inflate(stream, Z_FINISH);
1057 if (status == Z_STREAM_END && !stream->avail_in) {
1058 git_inflate_end(stream);
1059 return buf;
1062 if (status < 0)
1063 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1064 else if (stream->avail_in)
1065 error("garbage at end of loose object '%s'",
1066 sha1_to_hex(sha1));
1067 free(buf);
1068 return NULL;
1072 * We used to just use "sscanf()", but that's actually way
1073 * too permissive for what we want to check. So do an anal
1074 * object header parse by hand.
1076 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1077 unsigned int flags)
1079 const char *type_buf = hdr;
1080 unsigned long size;
1081 int type, type_len = 0;
1084 * The type can be of any size but is followed by
1085 * a space.
1087 for (;;) {
1088 char c = *hdr++;
1089 if (!c)
1090 return -1;
1091 if (c == ' ')
1092 break;
1093 type_len++;
1096 type = type_from_string_gently(type_buf, type_len, 1);
1097 if (oi->typename)
1098 strbuf_add(oi->typename, type_buf, type_len);
1100 * Set type to 0 if its an unknown object and
1101 * we're obtaining the type using '--allow-unknown-type'
1102 * option.
1104 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1105 type = 0;
1106 else if (type < 0)
1107 die("invalid object type");
1108 if (oi->typep)
1109 *oi->typep = type;
1112 * The length must follow immediately, and be in canonical
1113 * decimal format (ie "010" is not valid).
1115 size = *hdr++ - '0';
1116 if (size > 9)
1117 return -1;
1118 if (size) {
1119 for (;;) {
1120 unsigned long c = *hdr - '0';
1121 if (c > 9)
1122 break;
1123 hdr++;
1124 size = size * 10 + c;
1128 if (oi->sizep)
1129 *oi->sizep = size;
1132 * The length must be followed by a zero byte
1134 return *hdr ? -1 : type;
1137 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1139 struct object_info oi = OBJECT_INFO_INIT;
1141 oi.sizep = sizep;
1142 return parse_sha1_header_extended(hdr, &oi, 0);
1145 static int sha1_loose_object_info(const unsigned char *sha1,
1146 struct object_info *oi,
1147 int flags)
1149 int status = 0;
1150 unsigned long mapsize;
1151 void *map;
1152 git_zstream stream;
1153 char hdr[32];
1154 struct strbuf hdrbuf = STRBUF_INIT;
1155 unsigned long size_scratch;
1157 if (oi->delta_base_sha1)
1158 hashclr(oi->delta_base_sha1);
1161 * If we don't care about type or size, then we don't
1162 * need to look inside the object at all. Note that we
1163 * do not optimize out the stat call, even if the
1164 * caller doesn't care about the disk-size, since our
1165 * return value implicitly indicates whether the
1166 * object even exists.
1168 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1169 const char *path;
1170 struct stat st;
1171 if (stat_sha1_file(sha1, &st, &path) < 0)
1172 return -1;
1173 if (oi->disk_sizep)
1174 *oi->disk_sizep = st.st_size;
1175 return 0;
1178 map = map_sha1_file(sha1, &mapsize);
1179 if (!map)
1180 return -1;
1182 if (!oi->sizep)
1183 oi->sizep = &size_scratch;
1185 if (oi->disk_sizep)
1186 *oi->disk_sizep = mapsize;
1187 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1188 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1189 status = error("unable to unpack %s header with --allow-unknown-type",
1190 sha1_to_hex(sha1));
1191 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1192 status = error("unable to unpack %s header",
1193 sha1_to_hex(sha1));
1194 if (status < 0)
1195 ; /* Do nothing */
1196 else if (hdrbuf.len) {
1197 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1198 status = error("unable to parse %s header with --allow-unknown-type",
1199 sha1_to_hex(sha1));
1200 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1201 status = error("unable to parse %s header", sha1_to_hex(sha1));
1203 if (status >= 0 && oi->contentp) {
1204 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1205 *oi->sizep, sha1);
1206 if (!*oi->contentp) {
1207 git_inflate_end(&stream);
1208 status = -1;
1210 } else
1211 git_inflate_end(&stream);
1213 munmap(map, mapsize);
1214 if (status && oi->typep)
1215 *oi->typep = status;
1216 if (oi->sizep == &size_scratch)
1217 oi->sizep = NULL;
1218 strbuf_release(&hdrbuf);
1219 oi->whence = OI_LOOSE;
1220 return (status < 0) ? status : 0;
1223 int fetch_if_missing = 1;
1225 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1227 static struct object_info blank_oi = OBJECT_INFO_INIT;
1228 struct pack_entry e;
1229 int rtype;
1230 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1231 lookup_replace_object(sha1) :
1232 sha1;
1233 int already_retried = 0;
1235 if (is_null_sha1(real))
1236 return -1;
1238 if (!oi)
1239 oi = &blank_oi;
1241 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1242 struct cached_object *co = find_cached_object(real);
1243 if (co) {
1244 if (oi->typep)
1245 *(oi->typep) = co->type;
1246 if (oi->sizep)
1247 *(oi->sizep) = co->size;
1248 if (oi->disk_sizep)
1249 *(oi->disk_sizep) = 0;
1250 if (oi->delta_base_sha1)
1251 hashclr(oi->delta_base_sha1);
1252 if (oi->typename)
1253 strbuf_addstr(oi->typename, typename(co->type));
1254 if (oi->contentp)
1255 *oi->contentp = xmemdupz(co->buf, co->size);
1256 oi->whence = OI_CACHED;
1257 return 0;
1261 while (1) {
1262 if (find_pack_entry(real, &e))
1263 break;
1265 /* Most likely it's a loose object. */
1266 if (!sha1_loose_object_info(real, oi, flags))
1267 return 0;
1269 /* Not a loose object; someone else may have just packed it. */
1270 reprepare_packed_git();
1271 if (find_pack_entry(real, &e))
1272 break;
1274 /* Check if it is a missing object */
1275 if (fetch_if_missing && repository_format_partial_clone &&
1276 !already_retried) {
1278 * TODO Investigate haveing fetch_object() return
1279 * TODO error/success and stopping the music here.
1281 fetch_object(repository_format_partial_clone, real);
1282 already_retried = 1;
1283 continue;
1286 return -1;
1289 if (oi == &blank_oi)
1291 * We know that the caller doesn't actually need the
1292 * information below, so return early.
1294 return 0;
1295 rtype = packed_object_info(e.p, e.offset, oi);
1296 if (rtype < 0) {
1297 mark_bad_packed_object(e.p, real);
1298 return sha1_object_info_extended(real, oi, 0);
1299 } else if (oi->whence == OI_PACKED) {
1300 oi->u.packed.offset = e.offset;
1301 oi->u.packed.pack = e.p;
1302 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1303 rtype == OBJ_OFS_DELTA);
1306 return 0;
1309 /* returns enum object_type or negative */
1310 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1312 enum object_type type;
1313 struct object_info oi = OBJECT_INFO_INIT;
1315 oi.typep = &type;
1316 oi.sizep = sizep;
1317 if (sha1_object_info_extended(sha1, &oi,
1318 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1319 return -1;
1320 return type;
1323 static void *read_object(const unsigned char *sha1, enum object_type *type,
1324 unsigned long *size)
1326 struct object_info oi = OBJECT_INFO_INIT;
1327 void *content;
1328 oi.typep = type;
1329 oi.sizep = size;
1330 oi.contentp = &content;
1332 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1333 return NULL;
1334 return content;
1337 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1338 unsigned char *sha1)
1340 struct cached_object *co;
1342 hash_sha1_file(buf, len, typename(type), sha1);
1343 if (has_sha1_file(sha1) || find_cached_object(sha1))
1344 return 0;
1345 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1346 co = &cached_objects[cached_object_nr++];
1347 co->size = len;
1348 co->type = type;
1349 co->buf = xmalloc(len);
1350 memcpy(co->buf, buf, len);
1351 hashcpy(co->sha1, sha1);
1352 return 0;
1356 * This function dies on corrupt objects; the callers who want to
1357 * deal with them should arrange to call read_object() and give error
1358 * messages themselves.
1360 void *read_sha1_file_extended(const unsigned char *sha1,
1361 enum object_type *type,
1362 unsigned long *size,
1363 int lookup_replace)
1365 void *data;
1366 const struct packed_git *p;
1367 const char *path;
1368 struct stat st;
1369 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1370 : sha1;
1372 errno = 0;
1373 data = read_object(repl, type, size);
1374 if (data)
1375 return data;
1377 if (errno && errno != ENOENT)
1378 die_errno("failed to read object %s", sha1_to_hex(sha1));
1380 /* die if we replaced an object with one that does not exist */
1381 if (repl != sha1)
1382 die("replacement %s not found for %s",
1383 sha1_to_hex(repl), sha1_to_hex(sha1));
1385 if (!stat_sha1_file(repl, &st, &path))
1386 die("loose object %s (stored in %s) is corrupt",
1387 sha1_to_hex(repl), path);
1389 if ((p = has_packed_and_bad(repl)) != NULL)
1390 die("packed object %s (stored in %s) is corrupt",
1391 sha1_to_hex(repl), p->pack_name);
1393 return NULL;
1396 void *read_object_with_reference(const unsigned char *sha1,
1397 const char *required_type_name,
1398 unsigned long *size,
1399 unsigned char *actual_sha1_return)
1401 enum object_type type, required_type;
1402 void *buffer;
1403 unsigned long isize;
1404 unsigned char actual_sha1[20];
1406 required_type = type_from_string(required_type_name);
1407 hashcpy(actual_sha1, sha1);
1408 while (1) {
1409 int ref_length = -1;
1410 const char *ref_type = NULL;
1412 buffer = read_sha1_file(actual_sha1, &type, &isize);
1413 if (!buffer)
1414 return NULL;
1415 if (type == required_type) {
1416 *size = isize;
1417 if (actual_sha1_return)
1418 hashcpy(actual_sha1_return, actual_sha1);
1419 return buffer;
1421 /* Handle references */
1422 else if (type == OBJ_COMMIT)
1423 ref_type = "tree ";
1424 else if (type == OBJ_TAG)
1425 ref_type = "object ";
1426 else {
1427 free(buffer);
1428 return NULL;
1430 ref_length = strlen(ref_type);
1432 if (ref_length + 40 > isize ||
1433 memcmp(buffer, ref_type, ref_length) ||
1434 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1435 free(buffer);
1436 return NULL;
1438 free(buffer);
1439 /* Now we have the ID of the referred-to object in
1440 * actual_sha1. Check again. */
1444 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1445 const char *type, unsigned char *sha1,
1446 char *hdr, int *hdrlen)
1448 git_SHA_CTX c;
1450 /* Generate the header */
1451 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1453 /* Sha1.. */
1454 git_SHA1_Init(&c);
1455 git_SHA1_Update(&c, hdr, *hdrlen);
1456 git_SHA1_Update(&c, buf, len);
1457 git_SHA1_Final(sha1, &c);
1461 * Move the just written object into its final resting place.
1463 int finalize_object_file(const char *tmpfile, const char *filename)
1465 int ret = 0;
1467 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1468 goto try_rename;
1469 else if (link(tmpfile, filename))
1470 ret = errno;
1473 * Coda hack - coda doesn't like cross-directory links,
1474 * so we fall back to a rename, which will mean that it
1475 * won't be able to check collisions, but that's not a
1476 * big deal.
1478 * The same holds for FAT formatted media.
1480 * When this succeeds, we just return. We have nothing
1481 * left to unlink.
1483 if (ret && ret != EEXIST) {
1484 try_rename:
1485 if (!rename(tmpfile, filename))
1486 goto out;
1487 ret = errno;
1489 unlink_or_warn(tmpfile);
1490 if (ret) {
1491 if (ret != EEXIST) {
1492 return error_errno("unable to write sha1 filename %s", filename);
1494 /* FIXME!!! Collision check here ? */
1497 out:
1498 if (adjust_shared_perm(filename))
1499 return error("unable to set permission to '%s'", filename);
1500 return 0;
1503 static int write_buffer(int fd, const void *buf, size_t len)
1505 if (write_in_full(fd, buf, len) < 0)
1506 return error_errno("file write error");
1507 return 0;
1510 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1511 unsigned char *sha1)
1513 char hdr[32];
1514 int hdrlen = sizeof(hdr);
1515 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1516 return 0;
1519 /* Finalize a file on disk, and close it. */
1520 static void close_sha1_file(int fd)
1522 if (fsync_object_files)
1523 fsync_or_die(fd, "sha1 file");
1524 if (close(fd) != 0)
1525 die_errno("error when closing sha1 file");
1528 /* Size of directory component, including the ending '/' */
1529 static inline int directory_size(const char *filename)
1531 const char *s = strrchr(filename, '/');
1532 if (!s)
1533 return 0;
1534 return s - filename + 1;
1538 * This creates a temporary file in the same directory as the final
1539 * 'filename'
1541 * We want to avoid cross-directory filename renames, because those
1542 * can have problems on various filesystems (FAT, NFS, Coda).
1544 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1546 int fd, dirlen = directory_size(filename);
1548 strbuf_reset(tmp);
1549 strbuf_add(tmp, filename, dirlen);
1550 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1551 fd = git_mkstemp_mode(tmp->buf, 0444);
1552 if (fd < 0 && dirlen && errno == ENOENT) {
1554 * Make sure the directory exists; note that the contents
1555 * of the buffer are undefined after mkstemp returns an
1556 * error, so we have to rewrite the whole buffer from
1557 * scratch.
1559 strbuf_reset(tmp);
1560 strbuf_add(tmp, filename, dirlen - 1);
1561 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1562 return -1;
1563 if (adjust_shared_perm(tmp->buf))
1564 return -1;
1566 /* Try again */
1567 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1568 fd = git_mkstemp_mode(tmp->buf, 0444);
1570 return fd;
1573 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1574 const void *buf, unsigned long len, time_t mtime)
1576 int fd, ret;
1577 unsigned char compressed[4096];
1578 git_zstream stream;
1579 git_SHA_CTX c;
1580 unsigned char parano_sha1[20];
1581 static struct strbuf tmp_file = STRBUF_INIT;
1582 static struct strbuf filename = STRBUF_INIT;
1584 strbuf_reset(&filename);
1585 sha1_file_name(&filename, sha1);
1587 fd = create_tmpfile(&tmp_file, filename.buf);
1588 if (fd < 0) {
1589 if (errno == EACCES)
1590 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1591 else
1592 return error_errno("unable to create temporary file");
1595 /* Set it up */
1596 git_deflate_init(&stream, zlib_compression_level);
1597 stream.next_out = compressed;
1598 stream.avail_out = sizeof(compressed);
1599 git_SHA1_Init(&c);
1601 /* First header.. */
1602 stream.next_in = (unsigned char *)hdr;
1603 stream.avail_in = hdrlen;
1604 while (git_deflate(&stream, 0) == Z_OK)
1605 ; /* nothing */
1606 git_SHA1_Update(&c, hdr, hdrlen);
1608 /* Then the data itself.. */
1609 stream.next_in = (void *)buf;
1610 stream.avail_in = len;
1611 do {
1612 unsigned char *in0 = stream.next_in;
1613 ret = git_deflate(&stream, Z_FINISH);
1614 git_SHA1_Update(&c, in0, stream.next_in - in0);
1615 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1616 die("unable to write sha1 file");
1617 stream.next_out = compressed;
1618 stream.avail_out = sizeof(compressed);
1619 } while (ret == Z_OK);
1621 if (ret != Z_STREAM_END)
1622 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1623 ret = git_deflate_end_gently(&stream);
1624 if (ret != Z_OK)
1625 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1626 git_SHA1_Final(parano_sha1, &c);
1627 if (hashcmp(sha1, parano_sha1) != 0)
1628 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1630 close_sha1_file(fd);
1632 if (mtime) {
1633 struct utimbuf utb;
1634 utb.actime = mtime;
1635 utb.modtime = mtime;
1636 if (utime(tmp_file.buf, &utb) < 0)
1637 warning_errno("failed utime() on %s", tmp_file.buf);
1640 return finalize_object_file(tmp_file.buf, filename.buf);
1643 static int freshen_loose_object(const unsigned char *sha1)
1645 return check_and_freshen(sha1, 1);
1648 static int freshen_packed_object(const unsigned char *sha1)
1650 struct pack_entry e;
1651 if (!find_pack_entry(sha1, &e))
1652 return 0;
1653 if (e.p->freshened)
1654 return 1;
1655 if (!freshen_file(e.p->pack_name))
1656 return 0;
1657 e.p->freshened = 1;
1658 return 1;
1661 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1663 char hdr[32];
1664 int hdrlen = sizeof(hdr);
1666 /* Normally if we have it in the pack then we do not bother writing
1667 * it out into .git/objects/??/?{38} file.
1669 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1670 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1671 return 0;
1672 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1675 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1676 struct object_id *oid, unsigned flags)
1678 char *header;
1679 int hdrlen, status = 0;
1681 /* type string, SP, %lu of the length plus NUL must fit this */
1682 hdrlen = strlen(type) + 32;
1683 header = xmalloc(hdrlen);
1684 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1686 if (!(flags & HASH_WRITE_OBJECT))
1687 goto cleanup;
1688 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1689 goto cleanup;
1690 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1692 cleanup:
1693 free(header);
1694 return status;
1697 int force_object_loose(const unsigned char *sha1, time_t mtime)
1699 void *buf;
1700 unsigned long len;
1701 enum object_type type;
1702 char hdr[32];
1703 int hdrlen;
1704 int ret;
1706 if (has_loose_object(sha1))
1707 return 0;
1708 buf = read_object(sha1, &type, &len);
1709 if (!buf)
1710 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1711 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1712 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1713 free(buf);
1715 return ret;
1718 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1720 if (!startup_info->have_repository)
1721 return 0;
1722 return sha1_object_info_extended(sha1, NULL,
1723 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1726 int has_object_file(const struct object_id *oid)
1728 return has_sha1_file(oid->hash);
1731 int has_object_file_with_flags(const struct object_id *oid, int flags)
1733 return has_sha1_file_with_flags(oid->hash, flags);
1736 static void check_tree(const void *buf, size_t size)
1738 struct tree_desc desc;
1739 struct name_entry entry;
1741 init_tree_desc(&desc, buf, size);
1742 while (tree_entry(&desc, &entry))
1743 /* do nothing
1744 * tree_entry() will die() on malformed entries */
1748 static void check_commit(const void *buf, size_t size)
1750 struct commit c;
1751 memset(&c, 0, sizeof(c));
1752 if (parse_commit_buffer(&c, buf, size))
1753 die("corrupt commit");
1756 static void check_tag(const void *buf, size_t size)
1758 struct tag t;
1759 memset(&t, 0, sizeof(t));
1760 if (parse_tag_buffer(&t, buf, size))
1761 die("corrupt tag");
1764 static int index_mem(struct object_id *oid, void *buf, size_t size,
1765 enum object_type type,
1766 const char *path, unsigned flags)
1768 int ret, re_allocated = 0;
1769 int write_object = flags & HASH_WRITE_OBJECT;
1771 if (!type)
1772 type = OBJ_BLOB;
1775 * Convert blobs to git internal format
1777 if ((type == OBJ_BLOB) && path) {
1778 struct strbuf nbuf = STRBUF_INIT;
1779 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1780 get_conv_flags(flags))) {
1781 buf = strbuf_detach(&nbuf, &size);
1782 re_allocated = 1;
1785 if (flags & HASH_FORMAT_CHECK) {
1786 if (type == OBJ_TREE)
1787 check_tree(buf, size);
1788 if (type == OBJ_COMMIT)
1789 check_commit(buf, size);
1790 if (type == OBJ_TAG)
1791 check_tag(buf, size);
1794 if (write_object)
1795 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1796 else
1797 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1798 if (re_allocated)
1799 free(buf);
1800 return ret;
1803 static int index_stream_convert_blob(struct object_id *oid, int fd,
1804 const char *path, unsigned flags)
1806 int ret;
1807 const int write_object = flags & HASH_WRITE_OBJECT;
1808 struct strbuf sbuf = STRBUF_INIT;
1810 assert(path);
1811 assert(would_convert_to_git_filter_fd(path));
1813 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1814 get_conv_flags(flags));
1816 if (write_object)
1817 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1818 oid->hash);
1819 else
1820 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1821 oid->hash);
1822 strbuf_release(&sbuf);
1823 return ret;
1826 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1827 const char *path, unsigned flags)
1829 struct strbuf sbuf = STRBUF_INIT;
1830 int ret;
1832 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1833 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1834 else
1835 ret = -1;
1836 strbuf_release(&sbuf);
1837 return ret;
1840 #define SMALL_FILE_SIZE (32*1024)
1842 static int index_core(struct object_id *oid, int fd, size_t size,
1843 enum object_type type, const char *path,
1844 unsigned flags)
1846 int ret;
1848 if (!size) {
1849 ret = index_mem(oid, "", size, type, path, flags);
1850 } else if (size <= SMALL_FILE_SIZE) {
1851 char *buf = xmalloc(size);
1852 ssize_t read_result = read_in_full(fd, buf, size);
1853 if (read_result < 0)
1854 ret = error_errno("read error while indexing %s",
1855 path ? path : "<unknown>");
1856 else if (read_result != size)
1857 ret = error("short read while indexing %s",
1858 path ? path : "<unknown>");
1859 else
1860 ret = index_mem(oid, buf, size, type, path, flags);
1861 free(buf);
1862 } else {
1863 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1864 ret = index_mem(oid, buf, size, type, path, flags);
1865 munmap(buf, size);
1867 return ret;
1871 * This creates one packfile per large blob unless bulk-checkin
1872 * machinery is "plugged".
1874 * This also bypasses the usual "convert-to-git" dance, and that is on
1875 * purpose. We could write a streaming version of the converting
1876 * functions and insert that before feeding the data to fast-import
1877 * (or equivalent in-core API described above). However, that is
1878 * somewhat complicated, as we do not know the size of the filter
1879 * result, which we need to know beforehand when writing a git object.
1880 * Since the primary motivation for trying to stream from the working
1881 * tree file and to avoid mmaping it in core is to deal with large
1882 * binary blobs, they generally do not want to get any conversion, and
1883 * callers should avoid this code path when filters are requested.
1885 static int index_stream(struct object_id *oid, int fd, size_t size,
1886 enum object_type type, const char *path,
1887 unsigned flags)
1889 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1892 int index_fd(struct object_id *oid, int fd, struct stat *st,
1893 enum object_type type, const char *path, unsigned flags)
1895 int ret;
1898 * Call xsize_t() only when needed to avoid potentially unnecessary
1899 * die() for large files.
1901 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1902 ret = index_stream_convert_blob(oid, fd, path, flags);
1903 else if (!S_ISREG(st->st_mode))
1904 ret = index_pipe(oid, fd, type, path, flags);
1905 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1906 (path && would_convert_to_git(&the_index, path)))
1907 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1908 flags);
1909 else
1910 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1911 flags);
1912 close(fd);
1913 return ret;
1916 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1918 int fd;
1919 struct strbuf sb = STRBUF_INIT;
1920 int rc = 0;
1922 switch (st->st_mode & S_IFMT) {
1923 case S_IFREG:
1924 fd = open(path, O_RDONLY);
1925 if (fd < 0)
1926 return error_errno("open(\"%s\")", path);
1927 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1928 return error("%s: failed to insert into database",
1929 path);
1930 break;
1931 case S_IFLNK:
1932 if (strbuf_readlink(&sb, path, st->st_size))
1933 return error_errno("readlink(\"%s\")", path);
1934 if (!(flags & HASH_WRITE_OBJECT))
1935 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1936 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1937 rc = error("%s: failed to insert into database", path);
1938 strbuf_release(&sb);
1939 break;
1940 case S_IFDIR:
1941 return resolve_gitlink_ref(path, "HEAD", oid);
1942 default:
1943 return error("%s: unsupported file type", path);
1945 return rc;
1948 int read_pack_header(int fd, struct pack_header *header)
1950 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1951 /* "eof before pack header was fully read" */
1952 return PH_ERROR_EOF;
1954 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1955 /* "protocol error (pack signature mismatch detected)" */
1956 return PH_ERROR_PACK_SIGNATURE;
1957 if (!pack_version_ok(header->hdr_version))
1958 /* "protocol error (pack version unsupported)" */
1959 return PH_ERROR_PROTOCOL;
1960 return 0;
1963 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1965 enum object_type type = sha1_object_info(sha1, NULL);
1966 if (type < 0)
1967 die("%s is not a valid object", sha1_to_hex(sha1));
1968 if (type != expect)
1969 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1970 typename(expect));
1973 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1974 struct strbuf *path,
1975 each_loose_object_fn obj_cb,
1976 each_loose_cruft_fn cruft_cb,
1977 each_loose_subdir_fn subdir_cb,
1978 void *data)
1980 size_t origlen, baselen;
1981 DIR *dir;
1982 struct dirent *de;
1983 int r = 0;
1984 struct object_id oid;
1986 if (subdir_nr > 0xff)
1987 BUG("invalid loose object subdirectory: %x", subdir_nr);
1989 origlen = path->len;
1990 strbuf_complete(path, '/');
1991 strbuf_addf(path, "%02x", subdir_nr);
1993 dir = opendir(path->buf);
1994 if (!dir) {
1995 if (errno != ENOENT)
1996 r = error_errno("unable to open %s", path->buf);
1997 strbuf_setlen(path, origlen);
1998 return r;
2001 oid.hash[0] = subdir_nr;
2002 strbuf_addch(path, '/');
2003 baselen = path->len;
2005 while ((de = readdir(dir))) {
2006 size_t namelen;
2007 if (is_dot_or_dotdot(de->d_name))
2008 continue;
2010 namelen = strlen(de->d_name);
2011 strbuf_setlen(path, baselen);
2012 strbuf_add(path, de->d_name, namelen);
2013 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2014 !hex_to_bytes(oid.hash + 1, de->d_name,
2015 GIT_SHA1_RAWSZ - 1)) {
2016 if (obj_cb) {
2017 r = obj_cb(&oid, path->buf, data);
2018 if (r)
2019 break;
2021 continue;
2024 if (cruft_cb) {
2025 r = cruft_cb(de->d_name, path->buf, data);
2026 if (r)
2027 break;
2030 closedir(dir);
2032 strbuf_setlen(path, baselen - 1);
2033 if (!r && subdir_cb)
2034 r = subdir_cb(subdir_nr, path->buf, data);
2036 strbuf_setlen(path, origlen);
2038 return r;
2041 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2042 each_loose_object_fn obj_cb,
2043 each_loose_cruft_fn cruft_cb,
2044 each_loose_subdir_fn subdir_cb,
2045 void *data)
2047 int r = 0;
2048 int i;
2050 for (i = 0; i < 256; i++) {
2051 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2052 subdir_cb, data);
2053 if (r)
2054 break;
2057 return r;
2060 int for_each_loose_file_in_objdir(const char *path,
2061 each_loose_object_fn obj_cb,
2062 each_loose_cruft_fn cruft_cb,
2063 each_loose_subdir_fn subdir_cb,
2064 void *data)
2066 struct strbuf buf = STRBUF_INIT;
2067 int r;
2069 strbuf_addstr(&buf, path);
2070 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2071 subdir_cb, data);
2072 strbuf_release(&buf);
2074 return r;
2077 struct loose_alt_odb_data {
2078 each_loose_object_fn *cb;
2079 void *data;
2082 static int loose_from_alt_odb(struct alternate_object_database *alt,
2083 void *vdata)
2085 struct loose_alt_odb_data *data = vdata;
2086 struct strbuf buf = STRBUF_INIT;
2087 int r;
2089 strbuf_addstr(&buf, alt->path);
2090 r = for_each_loose_file_in_objdir_buf(&buf,
2091 data->cb, NULL, NULL,
2092 data->data);
2093 strbuf_release(&buf);
2094 return r;
2097 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2099 struct loose_alt_odb_data alt;
2100 int r;
2102 r = for_each_loose_file_in_objdir(get_object_directory(),
2103 cb, NULL, NULL, data);
2104 if (r)
2105 return r;
2107 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2108 return 0;
2110 alt.cb = cb;
2111 alt.data = data;
2112 return foreach_alt_odb(loose_from_alt_odb, &alt);
2115 static int check_stream_sha1(git_zstream *stream,
2116 const char *hdr,
2117 unsigned long size,
2118 const char *path,
2119 const unsigned char *expected_sha1)
2121 git_SHA_CTX c;
2122 unsigned char real_sha1[GIT_MAX_RAWSZ];
2123 unsigned char buf[4096];
2124 unsigned long total_read;
2125 int status = Z_OK;
2127 git_SHA1_Init(&c);
2128 git_SHA1_Update(&c, hdr, stream->total_out);
2131 * We already read some bytes into hdr, but the ones up to the NUL
2132 * do not count against the object's content size.
2134 total_read = stream->total_out - strlen(hdr) - 1;
2137 * This size comparison must be "<=" to read the final zlib packets;
2138 * see the comment in unpack_sha1_rest for details.
2140 while (total_read <= size &&
2141 (status == Z_OK || status == Z_BUF_ERROR)) {
2142 stream->next_out = buf;
2143 stream->avail_out = sizeof(buf);
2144 if (size - total_read < stream->avail_out)
2145 stream->avail_out = size - total_read;
2146 status = git_inflate(stream, Z_FINISH);
2147 git_SHA1_Update(&c, buf, stream->next_out - buf);
2148 total_read += stream->next_out - buf;
2150 git_inflate_end(stream);
2152 if (status != Z_STREAM_END) {
2153 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2154 return -1;
2156 if (stream->avail_in) {
2157 error("garbage at end of loose object '%s'",
2158 sha1_to_hex(expected_sha1));
2159 return -1;
2162 git_SHA1_Final(real_sha1, &c);
2163 if (hashcmp(expected_sha1, real_sha1)) {
2164 error("sha1 mismatch for %s (expected %s)", path,
2165 sha1_to_hex(expected_sha1));
2166 return -1;
2169 return 0;
2172 int read_loose_object(const char *path,
2173 const unsigned char *expected_sha1,
2174 enum object_type *type,
2175 unsigned long *size,
2176 void **contents)
2178 int ret = -1;
2179 void *map = NULL;
2180 unsigned long mapsize;
2181 git_zstream stream;
2182 char hdr[32];
2184 *contents = NULL;
2186 map = map_sha1_file_1(path, NULL, &mapsize);
2187 if (!map) {
2188 error_errno("unable to mmap %s", path);
2189 goto out;
2192 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2193 error("unable to unpack header of %s", path);
2194 goto out;
2197 *type = parse_sha1_header(hdr, size);
2198 if (*type < 0) {
2199 error("unable to parse header of %s", path);
2200 git_inflate_end(&stream);
2201 goto out;
2204 if (*type == OBJ_BLOB) {
2205 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2206 goto out;
2207 } else {
2208 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2209 if (!*contents) {
2210 error("unable to unpack contents of %s", path);
2211 git_inflate_end(&stream);
2212 goto out;
2214 if (check_sha1_signature(expected_sha1, *contents,
2215 *size, typename(*type))) {
2216 error("sha1 mismatch for %s (expected %s)", path,
2217 sha1_to_hex(expected_sha1));
2218 free(*contents);
2219 goto out;
2223 ret = 0; /* everything checks out */
2225 out:
2226 if (map)
2227 munmap(map, mapsize);
2228 return ret;