convert: convert to struct object_id
[alt-git.git] / sha1_file.c
blob581a9dc52223ceb6d2b21a1b674688c6788f8339
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"
33 /* The maximum size for an object header. */
34 #define MAX_HEADER_LEN 32
36 const unsigned char null_sha1[GIT_MAX_RAWSZ];
37 const struct object_id null_oid;
38 const struct object_id empty_tree_oid = {
39 EMPTY_TREE_SHA1_BIN_LITERAL
41 const struct object_id empty_blob_oid = {
42 EMPTY_BLOB_SHA1_BIN_LITERAL
45 static void git_hash_sha1_init(git_hash_ctx *ctx)
47 git_SHA1_Init(&ctx->sha1);
50 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
52 git_SHA1_Update(&ctx->sha1, data, len);
55 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
57 git_SHA1_Final(hash, &ctx->sha1);
60 static void git_hash_unknown_init(git_hash_ctx *ctx)
62 die("trying to init unknown hash");
65 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
67 die("trying to update unknown hash");
70 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
72 die("trying to finalize unknown hash");
75 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
77 NULL,
78 0x00000000,
81 git_hash_unknown_init,
82 git_hash_unknown_update,
83 git_hash_unknown_final,
84 NULL,
85 NULL,
88 "sha-1",
89 /* "sha1", big-endian */
90 0x73686131,
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 const char *alt;
673 if (alt_odb_tail)
674 return;
676 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
678 alt_odb_tail = &alt_odb_list;
679 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
681 read_info_alternates(get_object_directory(), 0);
684 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
685 static int freshen_file(const char *fn)
687 struct utimbuf t;
688 t.actime = t.modtime = time(NULL);
689 return !utime(fn, &t);
693 * All of the check_and_freshen functions return 1 if the file exists and was
694 * freshened (if freshening was requested), 0 otherwise. If they return
695 * 0, you should not assume that it is safe to skip a write of the object (it
696 * either does not exist on disk, or has a stale mtime and may be subject to
697 * pruning).
699 int check_and_freshen_file(const char *fn, int freshen)
701 if (access(fn, F_OK))
702 return 0;
703 if (freshen && !freshen_file(fn))
704 return 0;
705 return 1;
708 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
710 static struct strbuf buf = STRBUF_INIT;
712 strbuf_reset(&buf);
713 sha1_file_name(&buf, sha1);
715 return check_and_freshen_file(buf.buf, freshen);
718 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
720 struct alternate_object_database *alt;
721 prepare_alt_odb();
722 for (alt = alt_odb_list; alt; alt = alt->next) {
723 const char *path = alt_sha1_path(alt, sha1);
724 if (check_and_freshen_file(path, freshen))
725 return 1;
727 return 0;
730 static int check_and_freshen(const unsigned char *sha1, int freshen)
732 return check_and_freshen_local(sha1, freshen) ||
733 check_and_freshen_nonlocal(sha1, freshen);
736 int has_loose_object_nonlocal(const unsigned char *sha1)
738 return check_and_freshen_nonlocal(sha1, 0);
741 static int has_loose_object(const unsigned char *sha1)
743 return check_and_freshen(sha1, 0);
746 static void mmap_limit_check(size_t length)
748 static size_t limit = 0;
749 if (!limit) {
750 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
751 if (!limit)
752 limit = SIZE_MAX;
754 if (length > limit)
755 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
756 (uintmax_t)length, (uintmax_t)limit);
759 void *xmmap_gently(void *start, size_t length,
760 int prot, int flags, int fd, off_t offset)
762 void *ret;
764 mmap_limit_check(length);
765 ret = mmap(start, length, prot, flags, fd, offset);
766 if (ret == MAP_FAILED) {
767 if (!length)
768 return NULL;
769 release_pack_memory(length);
770 ret = mmap(start, length, prot, flags, fd, offset);
772 return ret;
775 void *xmmap(void *start, size_t length,
776 int prot, int flags, int fd, off_t offset)
778 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
779 if (ret == MAP_FAILED)
780 die_errno("mmap failed");
781 return ret;
785 * With an in-core object data in "map", rehash it to make sure the
786 * object name actually matches "sha1" to detect object corruption.
787 * With "map" == NULL, try reading the object named with "sha1" using
788 * the streaming interface and rehash it to do the same.
790 int check_object_signature(const struct object_id *oid, void *map,
791 unsigned long size, const char *type)
793 struct object_id real_oid;
794 enum object_type obj_type;
795 struct git_istream *st;
796 git_hash_ctx c;
797 char hdr[MAX_HEADER_LEN];
798 int hdrlen;
800 if (map) {
801 hash_object_file(map, size, type, &real_oid);
802 return oidcmp(oid, &real_oid) ? -1 : 0;
805 st = open_istream(oid, &obj_type, &size, NULL);
806 if (!st)
807 return -1;
809 /* Generate the header */
810 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
812 /* Sha1.. */
813 the_hash_algo->init_fn(&c);
814 the_hash_algo->update_fn(&c, hdr, hdrlen);
815 for (;;) {
816 char buf[1024 * 16];
817 ssize_t readlen = read_istream(st, buf, sizeof(buf));
819 if (readlen < 0) {
820 close_istream(st);
821 return -1;
823 if (!readlen)
824 break;
825 the_hash_algo->update_fn(&c, buf, readlen);
827 the_hash_algo->final_fn(real_oid.hash, &c);
828 close_istream(st);
829 return oidcmp(oid, &real_oid) ? -1 : 0;
832 int git_open_cloexec(const char *name, int flags)
834 int fd;
835 static int o_cloexec = O_CLOEXEC;
837 fd = open(name, flags | o_cloexec);
838 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
839 /* Try again w/o O_CLOEXEC: the kernel might not support it */
840 o_cloexec &= ~O_CLOEXEC;
841 fd = open(name, flags | o_cloexec);
844 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
846 static int fd_cloexec = FD_CLOEXEC;
848 if (!o_cloexec && 0 <= fd && fd_cloexec) {
849 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
850 int flags = fcntl(fd, F_GETFD);
851 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
852 fd_cloexec = 0;
855 #endif
856 return fd;
860 * Find "sha1" as a loose object in the local repository or in an alternate.
861 * Returns 0 on success, negative on failure.
863 * The "path" out-parameter will give the path of the object we found (if any).
864 * Note that it may point to static storage and is only valid until another
865 * call to sha1_file_name(), etc.
867 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
868 const char **path)
870 struct alternate_object_database *alt;
871 static struct strbuf buf = STRBUF_INIT;
873 strbuf_reset(&buf);
874 sha1_file_name(&buf, sha1);
875 *path = buf.buf;
877 if (!lstat(*path, st))
878 return 0;
880 prepare_alt_odb();
881 errno = ENOENT;
882 for (alt = alt_odb_list; alt; alt = alt->next) {
883 *path = alt_sha1_path(alt, sha1);
884 if (!lstat(*path, st))
885 return 0;
888 return -1;
892 * Like stat_sha1_file(), but actually open the object and return the
893 * descriptor. See the caveats on the "path" parameter above.
895 static int open_sha1_file(const unsigned char *sha1, const char **path)
897 int fd;
898 struct alternate_object_database *alt;
899 int most_interesting_errno;
900 static struct strbuf buf = STRBUF_INIT;
902 strbuf_reset(&buf);
903 sha1_file_name(&buf, sha1);
904 *path = buf.buf;
906 fd = git_open(*path);
907 if (fd >= 0)
908 return fd;
909 most_interesting_errno = errno;
911 prepare_alt_odb();
912 for (alt = alt_odb_list; alt; alt = alt->next) {
913 *path = alt_sha1_path(alt, sha1);
914 fd = git_open(*path);
915 if (fd >= 0)
916 return fd;
917 if (most_interesting_errno == ENOENT)
918 most_interesting_errno = errno;
920 errno = most_interesting_errno;
921 return -1;
925 * Map the loose object at "path" if it is not NULL, or the path found by
926 * searching for a loose object named "sha1".
928 static void *map_sha1_file_1(const char *path,
929 const unsigned char *sha1,
930 unsigned long *size)
932 void *map;
933 int fd;
935 if (path)
936 fd = git_open(path);
937 else
938 fd = open_sha1_file(sha1, &path);
939 map = NULL;
940 if (fd >= 0) {
941 struct stat st;
943 if (!fstat(fd, &st)) {
944 *size = xsize_t(st.st_size);
945 if (!*size) {
946 /* mmap() is forbidden on empty files */
947 error("object file %s is empty", path);
948 return NULL;
950 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
952 close(fd);
954 return map;
957 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
959 return map_sha1_file_1(NULL, sha1, size);
962 static int unpack_sha1_short_header(git_zstream *stream,
963 unsigned char *map, unsigned long mapsize,
964 void *buffer, unsigned long bufsiz)
966 /* Get the data stream */
967 memset(stream, 0, sizeof(*stream));
968 stream->next_in = map;
969 stream->avail_in = mapsize;
970 stream->next_out = buffer;
971 stream->avail_out = bufsiz;
973 git_inflate_init(stream);
974 return git_inflate(stream, 0);
977 int unpack_sha1_header(git_zstream *stream,
978 unsigned char *map, unsigned long mapsize,
979 void *buffer, unsigned long bufsiz)
981 int status = unpack_sha1_short_header(stream, map, mapsize,
982 buffer, bufsiz);
984 if (status < Z_OK)
985 return status;
987 /* Make sure we have the terminating NUL */
988 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
989 return -1;
990 return 0;
993 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
994 unsigned long mapsize, void *buffer,
995 unsigned long bufsiz, struct strbuf *header)
997 int status;
999 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1000 if (status < Z_OK)
1001 return -1;
1004 * Check if entire header is unpacked in the first iteration.
1006 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1007 return 0;
1010 * buffer[0..bufsiz] was not large enough. Copy the partial
1011 * result out to header, and then append the result of further
1012 * reading the stream.
1014 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1015 stream->next_out = buffer;
1016 stream->avail_out = bufsiz;
1018 do {
1019 status = git_inflate(stream, 0);
1020 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1021 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1022 return 0;
1023 stream->next_out = buffer;
1024 stream->avail_out = bufsiz;
1025 } while (status != Z_STREAM_END);
1026 return -1;
1029 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1031 int bytes = strlen(buffer) + 1;
1032 unsigned char *buf = xmallocz(size);
1033 unsigned long n;
1034 int status = Z_OK;
1036 n = stream->total_out - bytes;
1037 if (n > size)
1038 n = size;
1039 memcpy(buf, (char *) buffer + bytes, n);
1040 bytes = n;
1041 if (bytes <= size) {
1043 * The above condition must be (bytes <= size), not
1044 * (bytes < size). In other words, even though we
1045 * expect no more output and set avail_out to zero,
1046 * the input zlib stream may have bytes that express
1047 * "this concludes the stream", and we *do* want to
1048 * eat that input.
1050 * Otherwise we would not be able to test that we
1051 * consumed all the input to reach the expected size;
1052 * we also want to check that zlib tells us that all
1053 * went well with status == Z_STREAM_END at the end.
1055 stream->next_out = buf + bytes;
1056 stream->avail_out = size - bytes;
1057 while (status == Z_OK)
1058 status = git_inflate(stream, Z_FINISH);
1060 if (status == Z_STREAM_END && !stream->avail_in) {
1061 git_inflate_end(stream);
1062 return buf;
1065 if (status < 0)
1066 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1067 else if (stream->avail_in)
1068 error("garbage at end of loose object '%s'",
1069 sha1_to_hex(sha1));
1070 free(buf);
1071 return NULL;
1075 * We used to just use "sscanf()", but that's actually way
1076 * too permissive for what we want to check. So do an anal
1077 * object header parse by hand.
1079 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1080 unsigned int flags)
1082 const char *type_buf = hdr;
1083 unsigned long size;
1084 int type, type_len = 0;
1087 * The type can be of any size but is followed by
1088 * a space.
1090 for (;;) {
1091 char c = *hdr++;
1092 if (!c)
1093 return -1;
1094 if (c == ' ')
1095 break;
1096 type_len++;
1099 type = type_from_string_gently(type_buf, type_len, 1);
1100 if (oi->type_name)
1101 strbuf_add(oi->type_name, type_buf, type_len);
1103 * Set type to 0 if its an unknown object and
1104 * we're obtaining the type using '--allow-unknown-type'
1105 * option.
1107 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1108 type = 0;
1109 else if (type < 0)
1110 die("invalid object type");
1111 if (oi->typep)
1112 *oi->typep = type;
1115 * The length must follow immediately, and be in canonical
1116 * decimal format (ie "010" is not valid).
1118 size = *hdr++ - '0';
1119 if (size > 9)
1120 return -1;
1121 if (size) {
1122 for (;;) {
1123 unsigned long c = *hdr - '0';
1124 if (c > 9)
1125 break;
1126 hdr++;
1127 size = size * 10 + c;
1131 if (oi->sizep)
1132 *oi->sizep = size;
1135 * The length must be followed by a zero byte
1137 return *hdr ? -1 : type;
1140 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1142 struct object_info oi = OBJECT_INFO_INIT;
1144 oi.sizep = sizep;
1145 return parse_sha1_header_extended(hdr, &oi, 0);
1148 static int sha1_loose_object_info(const unsigned char *sha1,
1149 struct object_info *oi,
1150 int flags)
1152 int status = 0;
1153 unsigned long mapsize;
1154 void *map;
1155 git_zstream stream;
1156 char hdr[MAX_HEADER_LEN];
1157 struct strbuf hdrbuf = STRBUF_INIT;
1158 unsigned long size_scratch;
1160 if (oi->delta_base_sha1)
1161 hashclr(oi->delta_base_sha1);
1164 * If we don't care about type or size, then we don't
1165 * need to look inside the object at all. Note that we
1166 * do not optimize out the stat call, even if the
1167 * caller doesn't care about the disk-size, since our
1168 * return value implicitly indicates whether the
1169 * object even exists.
1171 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1172 const char *path;
1173 struct stat st;
1174 if (stat_sha1_file(sha1, &st, &path) < 0)
1175 return -1;
1176 if (oi->disk_sizep)
1177 *oi->disk_sizep = st.st_size;
1178 return 0;
1181 map = map_sha1_file(sha1, &mapsize);
1182 if (!map)
1183 return -1;
1185 if (!oi->sizep)
1186 oi->sizep = &size_scratch;
1188 if (oi->disk_sizep)
1189 *oi->disk_sizep = mapsize;
1190 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1191 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1192 status = error("unable to unpack %s header with --allow-unknown-type",
1193 sha1_to_hex(sha1));
1194 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1195 status = error("unable to unpack %s header",
1196 sha1_to_hex(sha1));
1197 if (status < 0)
1198 ; /* Do nothing */
1199 else if (hdrbuf.len) {
1200 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1201 status = error("unable to parse %s header with --allow-unknown-type",
1202 sha1_to_hex(sha1));
1203 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1204 status = error("unable to parse %s header", sha1_to_hex(sha1));
1206 if (status >= 0 && oi->contentp) {
1207 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1208 *oi->sizep, sha1);
1209 if (!*oi->contentp) {
1210 git_inflate_end(&stream);
1211 status = -1;
1213 } else
1214 git_inflate_end(&stream);
1216 munmap(map, mapsize);
1217 if (status && oi->typep)
1218 *oi->typep = status;
1219 if (oi->sizep == &size_scratch)
1220 oi->sizep = NULL;
1221 strbuf_release(&hdrbuf);
1222 oi->whence = OI_LOOSE;
1223 return (status < 0) ? status : 0;
1226 int fetch_if_missing = 1;
1228 int oid_object_info_extended(const struct object_id *oid, struct object_info *oi, unsigned flags)
1230 static struct object_info blank_oi = OBJECT_INFO_INIT;
1231 struct pack_entry e;
1232 int rtype;
1233 const struct object_id *real = oid;
1234 int already_retried = 0;
1236 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1237 real = lookup_replace_object(oid);
1239 if (is_null_oid(real))
1240 return -1;
1242 if (!oi)
1243 oi = &blank_oi;
1245 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1246 struct cached_object *co = find_cached_object(real->hash);
1247 if (co) {
1248 if (oi->typep)
1249 *(oi->typep) = co->type;
1250 if (oi->sizep)
1251 *(oi->sizep) = co->size;
1252 if (oi->disk_sizep)
1253 *(oi->disk_sizep) = 0;
1254 if (oi->delta_base_sha1)
1255 hashclr(oi->delta_base_sha1);
1256 if (oi->type_name)
1257 strbuf_addstr(oi->type_name, type_name(co->type));
1258 if (oi->contentp)
1259 *oi->contentp = xmemdupz(co->buf, co->size);
1260 oi->whence = OI_CACHED;
1261 return 0;
1265 while (1) {
1266 if (find_pack_entry(real->hash, &e))
1267 break;
1269 /* Most likely it's a loose object. */
1270 if (!sha1_loose_object_info(real->hash, oi, flags))
1271 return 0;
1273 /* Not a loose object; someone else may have just packed it. */
1274 reprepare_packed_git();
1275 if (find_pack_entry(real->hash, &e))
1276 break;
1278 /* Check if it is a missing object */
1279 if (fetch_if_missing && repository_format_partial_clone &&
1280 !already_retried) {
1282 * TODO Investigate haveing fetch_object() return
1283 * TODO error/success and stopping the music here.
1285 fetch_object(repository_format_partial_clone, real->hash);
1286 already_retried = 1;
1287 continue;
1290 return -1;
1293 if (oi == &blank_oi)
1295 * We know that the caller doesn't actually need the
1296 * information below, so return early.
1298 return 0;
1299 rtype = packed_object_info(e.p, e.offset, oi);
1300 if (rtype < 0) {
1301 mark_bad_packed_object(e.p, real->hash);
1302 return oid_object_info_extended(real, oi, 0);
1303 } else if (oi->whence == OI_PACKED) {
1304 oi->u.packed.offset = e.offset;
1305 oi->u.packed.pack = e.p;
1306 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1307 rtype == OBJ_OFS_DELTA);
1310 return 0;
1313 /* returns enum object_type or negative */
1314 int oid_object_info(const struct object_id *oid, unsigned long *sizep)
1316 enum object_type type;
1317 struct object_info oi = OBJECT_INFO_INIT;
1319 oi.typep = &type;
1320 oi.sizep = sizep;
1321 if (oid_object_info_extended(oid, &oi,
1322 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1323 return -1;
1324 return type;
1327 static void *read_object(const unsigned char *sha1, enum object_type *type,
1328 unsigned long *size)
1330 struct object_id oid;
1331 struct object_info oi = OBJECT_INFO_INIT;
1332 void *content;
1333 oi.typep = type;
1334 oi.sizep = size;
1335 oi.contentp = &content;
1337 hashcpy(oid.hash, sha1);
1339 if (oid_object_info_extended(&oid, &oi, 0) < 0)
1340 return NULL;
1341 return content;
1344 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1345 struct object_id *oid)
1347 struct cached_object *co;
1349 hash_object_file(buf, len, type_name(type), oid);
1350 if (has_sha1_file(oid->hash) || find_cached_object(oid->hash))
1351 return 0;
1352 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1353 co = &cached_objects[cached_object_nr++];
1354 co->size = len;
1355 co->type = type;
1356 co->buf = xmalloc(len);
1357 memcpy(co->buf, buf, len);
1358 hashcpy(co->sha1, oid->hash);
1359 return 0;
1363 * This function dies on corrupt objects; the callers who want to
1364 * deal with them should arrange to call read_object() and give error
1365 * messages themselves.
1367 void *read_object_file_extended(const struct object_id *oid,
1368 enum object_type *type,
1369 unsigned long *size,
1370 int lookup_replace)
1372 void *data;
1373 const struct packed_git *p;
1374 const char *path;
1375 struct stat st;
1376 const struct object_id *repl = lookup_replace ? lookup_replace_object(oid)
1377 : oid;
1379 errno = 0;
1380 data = read_object(repl->hash, type, size);
1381 if (data)
1382 return data;
1384 if (errno && errno != ENOENT)
1385 die_errno("failed to read object %s", oid_to_hex(oid));
1387 /* die if we replaced an object with one that does not exist */
1388 if (repl != oid)
1389 die("replacement %s not found for %s",
1390 oid_to_hex(repl), oid_to_hex(oid));
1392 if (!stat_sha1_file(repl->hash, &st, &path))
1393 die("loose object %s (stored in %s) is corrupt",
1394 oid_to_hex(repl), path);
1396 if ((p = has_packed_and_bad(repl->hash)) != NULL)
1397 die("packed object %s (stored in %s) is corrupt",
1398 oid_to_hex(repl), p->pack_name);
1400 return NULL;
1403 void *read_object_with_reference(const struct object_id *oid,
1404 const char *required_type_name,
1405 unsigned long *size,
1406 struct object_id *actual_oid_return)
1408 enum object_type type, required_type;
1409 void *buffer;
1410 unsigned long isize;
1411 struct object_id actual_oid;
1413 required_type = type_from_string(required_type_name);
1414 oidcpy(&actual_oid, oid);
1415 while (1) {
1416 int ref_length = -1;
1417 const char *ref_type = NULL;
1419 buffer = read_object_file(&actual_oid, &type, &isize);
1420 if (!buffer)
1421 return NULL;
1422 if (type == required_type) {
1423 *size = isize;
1424 if (actual_oid_return)
1425 oidcpy(actual_oid_return, &actual_oid);
1426 return buffer;
1428 /* Handle references */
1429 else if (type == OBJ_COMMIT)
1430 ref_type = "tree ";
1431 else if (type == OBJ_TAG)
1432 ref_type = "object ";
1433 else {
1434 free(buffer);
1435 return NULL;
1437 ref_length = strlen(ref_type);
1439 if (ref_length + GIT_SHA1_HEXSZ > isize ||
1440 memcmp(buffer, ref_type, ref_length) ||
1441 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1442 free(buffer);
1443 return NULL;
1445 free(buffer);
1446 /* Now we have the ID of the referred-to object in
1447 * actual_oid. Check again. */
1451 static void write_object_file_prepare(const void *buf, unsigned long len,
1452 const char *type, struct object_id *oid,
1453 char *hdr, int *hdrlen)
1455 git_hash_ctx c;
1457 /* Generate the header */
1458 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1460 /* Sha1.. */
1461 the_hash_algo->init_fn(&c);
1462 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1463 the_hash_algo->update_fn(&c, buf, len);
1464 the_hash_algo->final_fn(oid->hash, &c);
1468 * Move the just written object into its final resting place.
1470 int finalize_object_file(const char *tmpfile, const char *filename)
1472 int ret = 0;
1474 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1475 goto try_rename;
1476 else if (link(tmpfile, filename))
1477 ret = errno;
1480 * Coda hack - coda doesn't like cross-directory links,
1481 * so we fall back to a rename, which will mean that it
1482 * won't be able to check collisions, but that's not a
1483 * big deal.
1485 * The same holds for FAT formatted media.
1487 * When this succeeds, we just return. We have nothing
1488 * left to unlink.
1490 if (ret && ret != EEXIST) {
1491 try_rename:
1492 if (!rename(tmpfile, filename))
1493 goto out;
1494 ret = errno;
1496 unlink_or_warn(tmpfile);
1497 if (ret) {
1498 if (ret != EEXIST) {
1499 return error_errno("unable to write sha1 filename %s", filename);
1501 /* FIXME!!! Collision check here ? */
1504 out:
1505 if (adjust_shared_perm(filename))
1506 return error("unable to set permission to '%s'", filename);
1507 return 0;
1510 static int write_buffer(int fd, const void *buf, size_t len)
1512 if (write_in_full(fd, buf, len) < 0)
1513 return error_errno("file write error");
1514 return 0;
1517 int hash_object_file(const void *buf, unsigned long len, const char *type,
1518 struct object_id *oid)
1520 char hdr[MAX_HEADER_LEN];
1521 int hdrlen = sizeof(hdr);
1522 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1523 return 0;
1526 /* Finalize a file on disk, and close it. */
1527 static void close_sha1_file(int fd)
1529 if (fsync_object_files)
1530 fsync_or_die(fd, "sha1 file");
1531 if (close(fd) != 0)
1532 die_errno("error when closing sha1 file");
1535 /* Size of directory component, including the ending '/' */
1536 static inline int directory_size(const char *filename)
1538 const char *s = strrchr(filename, '/');
1539 if (!s)
1540 return 0;
1541 return s - filename + 1;
1545 * This creates a temporary file in the same directory as the final
1546 * 'filename'
1548 * We want to avoid cross-directory filename renames, because those
1549 * can have problems on various filesystems (FAT, NFS, Coda).
1551 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1553 int fd, dirlen = directory_size(filename);
1555 strbuf_reset(tmp);
1556 strbuf_add(tmp, filename, dirlen);
1557 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1558 fd = git_mkstemp_mode(tmp->buf, 0444);
1559 if (fd < 0 && dirlen && errno == ENOENT) {
1561 * Make sure the directory exists; note that the contents
1562 * of the buffer are undefined after mkstemp returns an
1563 * error, so we have to rewrite the whole buffer from
1564 * scratch.
1566 strbuf_reset(tmp);
1567 strbuf_add(tmp, filename, dirlen - 1);
1568 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1569 return -1;
1570 if (adjust_shared_perm(tmp->buf))
1571 return -1;
1573 /* Try again */
1574 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1575 fd = git_mkstemp_mode(tmp->buf, 0444);
1577 return fd;
1580 static int write_loose_object(const struct object_id *oid, char *hdr,
1581 int hdrlen, const void *buf, unsigned long len,
1582 time_t mtime)
1584 int fd, ret;
1585 unsigned char compressed[4096];
1586 git_zstream stream;
1587 git_hash_ctx c;
1588 struct object_id parano_oid;
1589 static struct strbuf tmp_file = STRBUF_INIT;
1590 static struct strbuf filename = STRBUF_INIT;
1592 strbuf_reset(&filename);
1593 sha1_file_name(&filename, oid->hash);
1595 fd = create_tmpfile(&tmp_file, filename.buf);
1596 if (fd < 0) {
1597 if (errno == EACCES)
1598 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1599 else
1600 return error_errno("unable to create temporary file");
1603 /* Set it up */
1604 git_deflate_init(&stream, zlib_compression_level);
1605 stream.next_out = compressed;
1606 stream.avail_out = sizeof(compressed);
1607 the_hash_algo->init_fn(&c);
1609 /* First header.. */
1610 stream.next_in = (unsigned char *)hdr;
1611 stream.avail_in = hdrlen;
1612 while (git_deflate(&stream, 0) == Z_OK)
1613 ; /* nothing */
1614 the_hash_algo->update_fn(&c, hdr, hdrlen);
1616 /* Then the data itself.. */
1617 stream.next_in = (void *)buf;
1618 stream.avail_in = len;
1619 do {
1620 unsigned char *in0 = stream.next_in;
1621 ret = git_deflate(&stream, Z_FINISH);
1622 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1623 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1624 die("unable to write sha1 file");
1625 stream.next_out = compressed;
1626 stream.avail_out = sizeof(compressed);
1627 } while (ret == Z_OK);
1629 if (ret != Z_STREAM_END)
1630 die("unable to deflate new object %s (%d)", oid_to_hex(oid),
1631 ret);
1632 ret = git_deflate_end_gently(&stream);
1633 if (ret != Z_OK)
1634 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid),
1635 ret);
1636 the_hash_algo->final_fn(parano_oid.hash, &c);
1637 if (oidcmp(oid, &parano_oid) != 0)
1638 die("confused by unstable object source data for %s",
1639 oid_to_hex(oid));
1641 close_sha1_file(fd);
1643 if (mtime) {
1644 struct utimbuf utb;
1645 utb.actime = mtime;
1646 utb.modtime = mtime;
1647 if (utime(tmp_file.buf, &utb) < 0)
1648 warning_errno("failed utime() on %s", tmp_file.buf);
1651 return finalize_object_file(tmp_file.buf, filename.buf);
1654 static int freshen_loose_object(const unsigned char *sha1)
1656 return check_and_freshen(sha1, 1);
1659 static int freshen_packed_object(const unsigned char *sha1)
1661 struct pack_entry e;
1662 if (!find_pack_entry(sha1, &e))
1663 return 0;
1664 if (e.p->freshened)
1665 return 1;
1666 if (!freshen_file(e.p->pack_name))
1667 return 0;
1668 e.p->freshened = 1;
1669 return 1;
1672 int write_object_file(const void *buf, unsigned long len, const char *type,
1673 struct object_id *oid)
1675 char hdr[MAX_HEADER_LEN];
1676 int hdrlen = sizeof(hdr);
1678 /* Normally if we have it in the pack then we do not bother writing
1679 * it out into .git/objects/??/?{38} file.
1681 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1682 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1683 return 0;
1684 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1687 int hash_object_file_literally(const void *buf, unsigned long len,
1688 const char *type, struct object_id *oid,
1689 unsigned flags)
1691 char *header;
1692 int hdrlen, status = 0;
1694 /* type string, SP, %lu of the length plus NUL must fit this */
1695 hdrlen = strlen(type) + MAX_HEADER_LEN;
1696 header = xmalloc(hdrlen);
1697 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1699 if (!(flags & HASH_WRITE_OBJECT))
1700 goto cleanup;
1701 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1702 goto cleanup;
1703 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1705 cleanup:
1706 free(header);
1707 return status;
1710 int force_object_loose(const struct object_id *oid, time_t mtime)
1712 void *buf;
1713 unsigned long len;
1714 enum object_type type;
1715 char hdr[MAX_HEADER_LEN];
1716 int hdrlen;
1717 int ret;
1719 if (has_loose_object(oid->hash))
1720 return 0;
1721 buf = read_object(oid->hash, &type, &len);
1722 if (!buf)
1723 return error("cannot read sha1_file for %s", oid_to_hex(oid));
1724 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1725 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1726 free(buf);
1728 return ret;
1731 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1733 struct object_id oid;
1734 if (!startup_info->have_repository)
1735 return 0;
1736 hashcpy(oid.hash, sha1);
1737 return oid_object_info_extended(&oid, NULL,
1738 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1741 int has_object_file(const struct object_id *oid)
1743 return has_sha1_file(oid->hash);
1746 int has_object_file_with_flags(const struct object_id *oid, int flags)
1748 return has_sha1_file_with_flags(oid->hash, flags);
1751 static void check_tree(const void *buf, size_t size)
1753 struct tree_desc desc;
1754 struct name_entry entry;
1756 init_tree_desc(&desc, buf, size);
1757 while (tree_entry(&desc, &entry))
1758 /* do nothing
1759 * tree_entry() will die() on malformed entries */
1763 static void check_commit(const void *buf, size_t size)
1765 struct commit c;
1766 memset(&c, 0, sizeof(c));
1767 if (parse_commit_buffer(&c, buf, size))
1768 die("corrupt commit");
1771 static void check_tag(const void *buf, size_t size)
1773 struct tag t;
1774 memset(&t, 0, sizeof(t));
1775 if (parse_tag_buffer(&t, buf, size))
1776 die("corrupt tag");
1779 static int index_mem(struct object_id *oid, void *buf, size_t size,
1780 enum object_type type,
1781 const char *path, unsigned flags)
1783 int ret, re_allocated = 0;
1784 int write_object = flags & HASH_WRITE_OBJECT;
1786 if (!type)
1787 type = OBJ_BLOB;
1790 * Convert blobs to git internal format
1792 if ((type == OBJ_BLOB) && path) {
1793 struct strbuf nbuf = STRBUF_INIT;
1794 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1795 get_conv_flags(flags))) {
1796 buf = strbuf_detach(&nbuf, &size);
1797 re_allocated = 1;
1800 if (flags & HASH_FORMAT_CHECK) {
1801 if (type == OBJ_TREE)
1802 check_tree(buf, size);
1803 if (type == OBJ_COMMIT)
1804 check_commit(buf, size);
1805 if (type == OBJ_TAG)
1806 check_tag(buf, size);
1809 if (write_object)
1810 ret = write_object_file(buf, size, type_name(type), oid);
1811 else
1812 ret = hash_object_file(buf, size, type_name(type), oid);
1813 if (re_allocated)
1814 free(buf);
1815 return ret;
1818 static int index_stream_convert_blob(struct object_id *oid, int fd,
1819 const char *path, unsigned flags)
1821 int ret;
1822 const int write_object = flags & HASH_WRITE_OBJECT;
1823 struct strbuf sbuf = STRBUF_INIT;
1825 assert(path);
1826 assert(would_convert_to_git_filter_fd(path));
1828 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1829 get_conv_flags(flags));
1831 if (write_object)
1832 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1833 oid);
1834 else
1835 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1836 oid);
1837 strbuf_release(&sbuf);
1838 return ret;
1841 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1842 const char *path, unsigned flags)
1844 struct strbuf sbuf = STRBUF_INIT;
1845 int ret;
1847 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1848 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1849 else
1850 ret = -1;
1851 strbuf_release(&sbuf);
1852 return ret;
1855 #define SMALL_FILE_SIZE (32*1024)
1857 static int index_core(struct object_id *oid, int fd, size_t size,
1858 enum object_type type, const char *path,
1859 unsigned flags)
1861 int ret;
1863 if (!size) {
1864 ret = index_mem(oid, "", size, type, path, flags);
1865 } else if (size <= SMALL_FILE_SIZE) {
1866 char *buf = xmalloc(size);
1867 ssize_t read_result = read_in_full(fd, buf, size);
1868 if (read_result < 0)
1869 ret = error_errno("read error while indexing %s",
1870 path ? path : "<unknown>");
1871 else if (read_result != size)
1872 ret = error("short read while indexing %s",
1873 path ? path : "<unknown>");
1874 else
1875 ret = index_mem(oid, buf, size, type, path, flags);
1876 free(buf);
1877 } else {
1878 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1879 ret = index_mem(oid, buf, size, type, path, flags);
1880 munmap(buf, size);
1882 return ret;
1886 * This creates one packfile per large blob unless bulk-checkin
1887 * machinery is "plugged".
1889 * This also bypasses the usual "convert-to-git" dance, and that is on
1890 * purpose. We could write a streaming version of the converting
1891 * functions and insert that before feeding the data to fast-import
1892 * (or equivalent in-core API described above). However, that is
1893 * somewhat complicated, as we do not know the size of the filter
1894 * result, which we need to know beforehand when writing a git object.
1895 * Since the primary motivation for trying to stream from the working
1896 * tree file and to avoid mmaping it in core is to deal with large
1897 * binary blobs, they generally do not want to get any conversion, and
1898 * callers should avoid this code path when filters are requested.
1900 static int index_stream(struct object_id *oid, int fd, size_t size,
1901 enum object_type type, const char *path,
1902 unsigned flags)
1904 return index_bulk_checkin(oid, fd, size, type, path, flags);
1907 int index_fd(struct object_id *oid, int fd, struct stat *st,
1908 enum object_type type, const char *path, unsigned flags)
1910 int ret;
1913 * Call xsize_t() only when needed to avoid potentially unnecessary
1914 * die() for large files.
1916 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1917 ret = index_stream_convert_blob(oid, fd, path, flags);
1918 else if (!S_ISREG(st->st_mode))
1919 ret = index_pipe(oid, fd, type, path, flags);
1920 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1921 (path && would_convert_to_git(&the_index, path)))
1922 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1923 flags);
1924 else
1925 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1926 flags);
1927 close(fd);
1928 return ret;
1931 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1933 int fd;
1934 struct strbuf sb = STRBUF_INIT;
1935 int rc = 0;
1937 switch (st->st_mode & S_IFMT) {
1938 case S_IFREG:
1939 fd = open(path, O_RDONLY);
1940 if (fd < 0)
1941 return error_errno("open(\"%s\")", path);
1942 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1943 return error("%s: failed to insert into database",
1944 path);
1945 break;
1946 case S_IFLNK:
1947 if (strbuf_readlink(&sb, path, st->st_size))
1948 return error_errno("readlink(\"%s\")", path);
1949 if (!(flags & HASH_WRITE_OBJECT))
1950 hash_object_file(sb.buf, sb.len, blob_type, oid);
1951 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1952 rc = error("%s: failed to insert into database", path);
1953 strbuf_release(&sb);
1954 break;
1955 case S_IFDIR:
1956 return resolve_gitlink_ref(path, "HEAD", oid);
1957 default:
1958 return error("%s: unsupported file type", path);
1960 return rc;
1963 int read_pack_header(int fd, struct pack_header *header)
1965 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1966 /* "eof before pack header was fully read" */
1967 return PH_ERROR_EOF;
1969 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1970 /* "protocol error (pack signature mismatch detected)" */
1971 return PH_ERROR_PACK_SIGNATURE;
1972 if (!pack_version_ok(header->hdr_version))
1973 /* "protocol error (pack version unsupported)" */
1974 return PH_ERROR_PROTOCOL;
1975 return 0;
1978 void assert_oid_type(const struct object_id *oid, enum object_type expect)
1980 enum object_type type = oid_object_info(oid, NULL);
1981 if (type < 0)
1982 die("%s is not a valid object", oid_to_hex(oid));
1983 if (type != expect)
1984 die("%s is not a valid '%s' object", oid_to_hex(oid),
1985 type_name(expect));
1988 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1989 struct strbuf *path,
1990 each_loose_object_fn obj_cb,
1991 each_loose_cruft_fn cruft_cb,
1992 each_loose_subdir_fn subdir_cb,
1993 void *data)
1995 size_t origlen, baselen;
1996 DIR *dir;
1997 struct dirent *de;
1998 int r = 0;
1999 struct object_id oid;
2001 if (subdir_nr > 0xff)
2002 BUG("invalid loose object subdirectory: %x", subdir_nr);
2004 origlen = path->len;
2005 strbuf_complete(path, '/');
2006 strbuf_addf(path, "%02x", subdir_nr);
2008 dir = opendir(path->buf);
2009 if (!dir) {
2010 if (errno != ENOENT)
2011 r = error_errno("unable to open %s", path->buf);
2012 strbuf_setlen(path, origlen);
2013 return r;
2016 oid.hash[0] = subdir_nr;
2017 strbuf_addch(path, '/');
2018 baselen = path->len;
2020 while ((de = readdir(dir))) {
2021 size_t namelen;
2022 if (is_dot_or_dotdot(de->d_name))
2023 continue;
2025 namelen = strlen(de->d_name);
2026 strbuf_setlen(path, baselen);
2027 strbuf_add(path, de->d_name, namelen);
2028 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2029 !hex_to_bytes(oid.hash + 1, de->d_name,
2030 GIT_SHA1_RAWSZ - 1)) {
2031 if (obj_cb) {
2032 r = obj_cb(&oid, path->buf, data);
2033 if (r)
2034 break;
2036 continue;
2039 if (cruft_cb) {
2040 r = cruft_cb(de->d_name, path->buf, data);
2041 if (r)
2042 break;
2045 closedir(dir);
2047 strbuf_setlen(path, baselen - 1);
2048 if (!r && subdir_cb)
2049 r = subdir_cb(subdir_nr, path->buf, data);
2051 strbuf_setlen(path, origlen);
2053 return r;
2056 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2057 each_loose_object_fn obj_cb,
2058 each_loose_cruft_fn cruft_cb,
2059 each_loose_subdir_fn subdir_cb,
2060 void *data)
2062 int r = 0;
2063 int i;
2065 for (i = 0; i < 256; i++) {
2066 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2067 subdir_cb, data);
2068 if (r)
2069 break;
2072 return r;
2075 int for_each_loose_file_in_objdir(const char *path,
2076 each_loose_object_fn obj_cb,
2077 each_loose_cruft_fn cruft_cb,
2078 each_loose_subdir_fn subdir_cb,
2079 void *data)
2081 struct strbuf buf = STRBUF_INIT;
2082 int r;
2084 strbuf_addstr(&buf, path);
2085 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2086 subdir_cb, data);
2087 strbuf_release(&buf);
2089 return r;
2092 struct loose_alt_odb_data {
2093 each_loose_object_fn *cb;
2094 void *data;
2097 static int loose_from_alt_odb(struct alternate_object_database *alt,
2098 void *vdata)
2100 struct loose_alt_odb_data *data = vdata;
2101 struct strbuf buf = STRBUF_INIT;
2102 int r;
2104 strbuf_addstr(&buf, alt->path);
2105 r = for_each_loose_file_in_objdir_buf(&buf,
2106 data->cb, NULL, NULL,
2107 data->data);
2108 strbuf_release(&buf);
2109 return r;
2112 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2114 struct loose_alt_odb_data alt;
2115 int r;
2117 r = for_each_loose_file_in_objdir(get_object_directory(),
2118 cb, NULL, NULL, data);
2119 if (r)
2120 return r;
2122 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2123 return 0;
2125 alt.cb = cb;
2126 alt.data = data;
2127 return foreach_alt_odb(loose_from_alt_odb, &alt);
2130 static int check_stream_sha1(git_zstream *stream,
2131 const char *hdr,
2132 unsigned long size,
2133 const char *path,
2134 const unsigned char *expected_sha1)
2136 git_hash_ctx c;
2137 unsigned char real_sha1[GIT_MAX_RAWSZ];
2138 unsigned char buf[4096];
2139 unsigned long total_read;
2140 int status = Z_OK;
2142 the_hash_algo->init_fn(&c);
2143 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2146 * We already read some bytes into hdr, but the ones up to the NUL
2147 * do not count against the object's content size.
2149 total_read = stream->total_out - strlen(hdr) - 1;
2152 * This size comparison must be "<=" to read the final zlib packets;
2153 * see the comment in unpack_sha1_rest for details.
2155 while (total_read <= size &&
2156 (status == Z_OK || status == Z_BUF_ERROR)) {
2157 stream->next_out = buf;
2158 stream->avail_out = sizeof(buf);
2159 if (size - total_read < stream->avail_out)
2160 stream->avail_out = size - total_read;
2161 status = git_inflate(stream, Z_FINISH);
2162 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2163 total_read += stream->next_out - buf;
2165 git_inflate_end(stream);
2167 if (status != Z_STREAM_END) {
2168 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2169 return -1;
2171 if (stream->avail_in) {
2172 error("garbage at end of loose object '%s'",
2173 sha1_to_hex(expected_sha1));
2174 return -1;
2177 the_hash_algo->final_fn(real_sha1, &c);
2178 if (hashcmp(expected_sha1, real_sha1)) {
2179 error("sha1 mismatch for %s (expected %s)", path,
2180 sha1_to_hex(expected_sha1));
2181 return -1;
2184 return 0;
2187 int read_loose_object(const char *path,
2188 const struct object_id *expected_oid,
2189 enum object_type *type,
2190 unsigned long *size,
2191 void **contents)
2193 int ret = -1;
2194 void *map = NULL;
2195 unsigned long mapsize;
2196 git_zstream stream;
2197 char hdr[MAX_HEADER_LEN];
2199 *contents = NULL;
2201 map = map_sha1_file_1(path, NULL, &mapsize);
2202 if (!map) {
2203 error_errno("unable to mmap %s", path);
2204 goto out;
2207 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2208 error("unable to unpack header of %s", path);
2209 goto out;
2212 *type = parse_sha1_header(hdr, size);
2213 if (*type < 0) {
2214 error("unable to parse header of %s", path);
2215 git_inflate_end(&stream);
2216 goto out;
2219 if (*type == OBJ_BLOB) {
2220 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2221 goto out;
2222 } else {
2223 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2224 if (!*contents) {
2225 error("unable to unpack contents of %s", path);
2226 git_inflate_end(&stream);
2227 goto out;
2229 if (check_object_signature(expected_oid, *contents,
2230 *size, type_name(*type))) {
2231 error("sha1 mismatch for %s (expected %s)", path,
2232 oid_to_hex(expected_oid));
2233 free(*contents);
2234 goto out;
2238 ret = 0; /* everything checks out */
2240 out:
2241 if (map)
2242 munmap(map, mapsize);
2243 return ret;