sha1_file: allow link_alt_odb_entries to handle arbitrary repositories
[git/raj.git] / sha1_file.c
blobd38f5cdb0ea8df80cb2a56af0806cc47e12eb487
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 "repository.h"
26 #include "streaming.h"
27 #include "dir.h"
28 #include "list.h"
29 #include "mergesort.h"
30 #include "quote.h"
31 #include "packfile.h"
32 #include "fetch-object.h"
33 #include "object-store.h"
35 const unsigned char null_sha1[GIT_MAX_RAWSZ];
36 const struct object_id null_oid;
37 const struct object_id empty_tree_oid = {
38 EMPTY_TREE_SHA1_BIN_LITERAL
40 const struct object_id empty_blob_oid = {
41 EMPTY_BLOB_SHA1_BIN_LITERAL
44 static void git_hash_sha1_init(void *ctx)
46 git_SHA1_Init((git_SHA_CTX *)ctx);
49 static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
51 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
54 static void git_hash_sha1_final(unsigned char *hash, void *ctx)
56 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
59 static void git_hash_unknown_init(void *ctx)
61 die("trying to init unknown hash");
64 static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
66 die("trying to update unknown hash");
69 static void git_hash_unknown_final(unsigned char *hash, void *ctx)
71 die("trying to finalize unknown hash");
74 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
76 NULL,
77 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 sizeof(git_SHA_CTX),
92 GIT_SHA1_RAWSZ,
93 GIT_SHA1_HEXSZ,
94 git_hash_sha1_init,
95 git_hash_sha1_update,
96 git_hash_sha1_final,
97 &empty_tree_oid,
98 &empty_blob_oid,
103 * This is meant to hold a *small* number of objects that you would
104 * want read_sha1_file() to be able to return, but yet you do not want
105 * to write them into the object store (e.g. a browse-only
106 * application).
108 static struct cached_object {
109 unsigned char sha1[20];
110 enum object_type type;
111 void *buf;
112 unsigned long size;
113 } *cached_objects;
114 static int cached_object_nr, cached_object_alloc;
116 static struct cached_object empty_tree = {
117 EMPTY_TREE_SHA1_BIN_LITERAL,
118 OBJ_TREE,
123 static struct cached_object *find_cached_object(const unsigned char *sha1)
125 int i;
126 struct cached_object *co = cached_objects;
128 for (i = 0; i < cached_object_nr; i++, co++) {
129 if (!hashcmp(co->sha1, sha1))
130 return co;
132 if (!hashcmp(sha1, empty_tree.sha1))
133 return &empty_tree;
134 return NULL;
138 static int get_conv_flags(unsigned flags)
140 if (flags & HASH_RENORMALIZE)
141 return CONV_EOL_RENORMALIZE;
142 else if (flags & HASH_WRITE_OBJECT)
143 return global_conv_flags_eol;
144 else
145 return 0;
149 int mkdir_in_gitdir(const char *path)
151 if (mkdir(path, 0777)) {
152 int saved_errno = errno;
153 struct stat st;
154 struct strbuf sb = STRBUF_INIT;
156 if (errno != EEXIST)
157 return -1;
159 * Are we looking at a path in a symlinked worktree
160 * whose original repository does not yet have it?
161 * e.g. .git/rr-cache pointing at its original
162 * repository in which the user hasn't performed any
163 * conflict resolution yet?
165 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
166 strbuf_readlink(&sb, path, st.st_size) ||
167 !is_absolute_path(sb.buf) ||
168 mkdir(sb.buf, 0777)) {
169 strbuf_release(&sb);
170 errno = saved_errno;
171 return -1;
173 strbuf_release(&sb);
175 return adjust_shared_perm(path);
178 enum scld_error safe_create_leading_directories(char *path)
180 char *next_component = path + offset_1st_component(path);
181 enum scld_error ret = SCLD_OK;
183 while (ret == SCLD_OK && next_component) {
184 struct stat st;
185 char *slash = next_component, slash_character;
187 while (*slash && !is_dir_sep(*slash))
188 slash++;
190 if (!*slash)
191 break;
193 next_component = slash + 1;
194 while (is_dir_sep(*next_component))
195 next_component++;
196 if (!*next_component)
197 break;
199 slash_character = *slash;
200 *slash = '\0';
201 if (!stat(path, &st)) {
202 /* path exists */
203 if (!S_ISDIR(st.st_mode)) {
204 errno = ENOTDIR;
205 ret = SCLD_EXISTS;
207 } else if (mkdir(path, 0777)) {
208 if (errno == EEXIST &&
209 !stat(path, &st) && S_ISDIR(st.st_mode))
210 ; /* somebody created it since we checked */
211 else if (errno == ENOENT)
213 * Either mkdir() failed because
214 * somebody just pruned the containing
215 * directory, or stat() failed because
216 * the file that was in our way was
217 * just removed. Either way, inform
218 * the caller that it might be worth
219 * trying again:
221 ret = SCLD_VANISHED;
222 else
223 ret = SCLD_FAILED;
224 } else if (adjust_shared_perm(path)) {
225 ret = SCLD_PERMS;
227 *slash = slash_character;
229 return ret;
232 enum scld_error safe_create_leading_directories_const(const char *path)
234 int save_errno;
235 /* path points to cache entries, so xstrdup before messing with it */
236 char *buf = xstrdup(path);
237 enum scld_error result = safe_create_leading_directories(buf);
239 save_errno = errno;
240 free(buf);
241 errno = save_errno;
242 return result;
245 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
248 * The number of times we will try to remove empty directories
249 * in the way of path. This is only 1 because if another
250 * process is racily creating directories that conflict with
251 * us, we don't want to fight against them.
253 int remove_directories_remaining = 1;
256 * The number of times that we will try to create the
257 * directories containing path. We are willing to attempt this
258 * more than once, because another process could be trying to
259 * clean up empty directories at the same time as we are
260 * trying to create them.
262 int create_directories_remaining = 3;
264 /* A scratch copy of path, filled lazily if we need it: */
265 struct strbuf path_copy = STRBUF_INIT;
267 int ret, save_errno;
269 /* Sanity check: */
270 assert(*path);
272 retry_fn:
273 ret = fn(path, cb);
274 save_errno = errno;
275 if (!ret)
276 goto out;
278 if (errno == EISDIR && remove_directories_remaining-- > 0) {
280 * A directory is in the way. Maybe it is empty; try
281 * to remove it:
283 if (!path_copy.len)
284 strbuf_addstr(&path_copy, path);
286 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
287 goto retry_fn;
288 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
290 * Maybe the containing directory didn't exist, or
291 * maybe it was just deleted by a process that is
292 * racing with us to clean up empty directories. Try
293 * to create it:
295 enum scld_error scld_result;
297 if (!path_copy.len)
298 strbuf_addstr(&path_copy, path);
300 do {
301 scld_result = safe_create_leading_directories(path_copy.buf);
302 if (scld_result == SCLD_OK)
303 goto retry_fn;
304 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
307 out:
308 strbuf_release(&path_copy);
309 errno = save_errno;
310 return ret;
313 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
315 int i;
316 for (i = 0; i < 20; i++) {
317 static char hex[] = "0123456789abcdef";
318 unsigned int val = sha1[i];
319 strbuf_addch(buf, hex[val >> 4]);
320 strbuf_addch(buf, hex[val & 0xf]);
321 if (!i)
322 strbuf_addch(buf, '/');
326 void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
328 strbuf_addstr(buf, get_object_directory());
329 strbuf_addch(buf, '/');
330 fill_sha1_path(buf, sha1);
333 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
335 strbuf_setlen(&alt->scratch, alt->base_len);
336 return &alt->scratch;
339 static const char *alt_sha1_path(struct alternate_object_database *alt,
340 const unsigned char *sha1)
342 struct strbuf *buf = alt_scratch_buf(alt);
343 fill_sha1_path(buf, sha1);
344 return buf->buf;
348 * Return non-zero iff the path is usable as an alternate object database.
350 static int alt_odb_usable(struct raw_object_store *o,
351 struct strbuf *path,
352 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 = o->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(struct repository *r,
394 const char *relative_base,
395 int depth);
396 static int link_alt_odb_entry(struct repository *r, const char *entry,
397 const char *relative_base, int depth, const char *normalized_objdir)
399 struct alternate_object_database *ent;
400 struct strbuf pathbuf = STRBUF_INIT;
402 if (!is_absolute_path(entry) && relative_base) {
403 strbuf_realpath(&pathbuf, relative_base, 1);
404 strbuf_addch(&pathbuf, '/');
406 strbuf_addstr(&pathbuf, entry);
408 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
409 error("unable to normalize alternate object path: %s",
410 pathbuf.buf);
411 strbuf_release(&pathbuf);
412 return -1;
416 * The trailing slash after the directory name is given by
417 * this function at the end. Remove duplicates.
419 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
420 strbuf_setlen(&pathbuf, pathbuf.len - 1);
422 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
423 strbuf_release(&pathbuf);
424 return -1;
427 ent = alloc_alt_odb(pathbuf.buf);
429 /* add the alternate entry */
430 *r->objects->alt_odb_tail = ent;
431 r->objects->alt_odb_tail = &(ent->next);
432 ent->next = NULL;
434 /* recursively add alternates */
435 read_info_alternates(r, pathbuf.buf, depth + 1);
437 strbuf_release(&pathbuf);
438 return 0;
441 static const char *parse_alt_odb_entry(const char *string,
442 int sep,
443 struct strbuf *out)
445 const char *end;
447 strbuf_reset(out);
449 if (*string == '#') {
450 /* comment; consume up to next separator */
451 end = strchrnul(string, sep);
452 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
454 * quoted path; unquote_c_style has copied the
455 * data for us and set "end". Broken quoting (e.g.,
456 * an entry that doesn't end with a quote) falls
457 * back to the unquoted case below.
459 } else {
460 /* normal, unquoted path */
461 end = strchrnul(string, sep);
462 strbuf_add(out, string, end - string);
465 if (*end)
466 end++;
467 return end;
470 static void link_alt_odb_entries(struct repository *r, const char *alt,
471 int sep, const char *relative_base, int depth)
473 struct strbuf objdirbuf = STRBUF_INIT;
474 struct strbuf entry = STRBUF_INIT;
476 if (!alt || !*alt)
477 return;
479 if (depth > 5) {
480 error("%s: ignoring alternate object stores, nesting too deep.",
481 relative_base);
482 return;
485 strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
486 if (strbuf_normalize_path(&objdirbuf) < 0)
487 die("unable to normalize object directory: %s",
488 objdirbuf.buf);
490 while (*alt) {
491 alt = parse_alt_odb_entry(alt, sep, &entry);
492 if (!entry.len)
493 continue;
494 link_alt_odb_entry(r, entry.buf,
495 relative_base, depth, objdirbuf.buf);
497 strbuf_release(&entry);
498 strbuf_release(&objdirbuf);
501 static void read_info_alternates(struct repository *r,
502 const char *relative_base,
503 int depth)
505 char *path;
506 struct strbuf buf = STRBUF_INIT;
508 path = xstrfmt("%s/info/alternates", relative_base);
509 if (strbuf_read_file(&buf, path, 1024) < 0) {
510 warn_on_fopen_errors(path);
511 free(path);
512 return;
515 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
516 strbuf_release(&buf);
517 free(path);
520 struct alternate_object_database *alloc_alt_odb(const char *dir)
522 struct alternate_object_database *ent;
524 FLEX_ALLOC_STR(ent, path, dir);
525 strbuf_init(&ent->scratch, 0);
526 strbuf_addf(&ent->scratch, "%s/", dir);
527 ent->base_len = ent->scratch.len;
529 return ent;
532 void add_to_alternates_file(const char *reference)
534 struct lock_file lock = LOCK_INIT;
535 char *alts = git_pathdup("objects/info/alternates");
536 FILE *in, *out;
537 int found = 0;
539 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
540 out = fdopen_lock_file(&lock, "w");
541 if (!out)
542 die_errno("unable to fdopen alternates lockfile");
544 in = fopen(alts, "r");
545 if (in) {
546 struct strbuf line = STRBUF_INIT;
548 while (strbuf_getline(&line, in) != EOF) {
549 if (!strcmp(reference, line.buf)) {
550 found = 1;
551 break;
553 fprintf_or_die(out, "%s\n", line.buf);
556 strbuf_release(&line);
557 fclose(in);
559 else if (errno != ENOENT)
560 die_errno("unable to read alternates file");
562 if (found) {
563 rollback_lock_file(&lock);
564 } else {
565 fprintf_or_die(out, "%s\n", reference);
566 if (commit_lock_file(&lock))
567 die_errno("unable to move new alternates file into place");
568 if (the_repository->objects->alt_odb_tail)
569 link_alt_odb_entries(the_repository, reference,
570 '\n', NULL, 0);
572 free(alts);
575 void add_to_alternates_memory(const char *reference)
578 * Make sure alternates are initialized, or else our entry may be
579 * overwritten when they are.
581 prepare_alt_odb(the_repository);
583 link_alt_odb_entries(the_repository, reference,
584 '\n', NULL, 0);
588 * Compute the exact path an alternate is at and returns it. In case of
589 * error NULL is returned and the human readable error is added to `err`
590 * `path` may be relative and should point to $GITDIR.
591 * `err` must not be null.
593 char *compute_alternate_path(const char *path, struct strbuf *err)
595 char *ref_git = NULL;
596 const char *repo, *ref_git_s;
597 int seen_error = 0;
599 ref_git_s = real_path_if_valid(path);
600 if (!ref_git_s) {
601 seen_error = 1;
602 strbuf_addf(err, _("path '%s' does not exist"), path);
603 goto out;
604 } else
606 * Beware: read_gitfile(), real_path() and mkpath()
607 * return static buffer
609 ref_git = xstrdup(ref_git_s);
611 repo = read_gitfile(ref_git);
612 if (!repo)
613 repo = read_gitfile(mkpath("%s/.git", ref_git));
614 if (repo) {
615 free(ref_git);
616 ref_git = xstrdup(repo);
619 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
620 char *ref_git_git = mkpathdup("%s/.git", ref_git);
621 free(ref_git);
622 ref_git = ref_git_git;
623 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
624 struct strbuf sb = STRBUF_INIT;
625 seen_error = 1;
626 if (get_common_dir(&sb, ref_git)) {
627 strbuf_addf(err,
628 _("reference repository '%s' as a linked "
629 "checkout is not supported yet."),
630 path);
631 goto out;
634 strbuf_addf(err, _("reference repository '%s' is not a "
635 "local repository."), path);
636 goto out;
639 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
640 strbuf_addf(err, _("reference repository '%s' is shallow"),
641 path);
642 seen_error = 1;
643 goto out;
646 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
647 strbuf_addf(err,
648 _("reference repository '%s' is grafted"),
649 path);
650 seen_error = 1;
651 goto out;
654 out:
655 if (seen_error) {
656 FREE_AND_NULL(ref_git);
659 return ref_git;
662 int foreach_alt_odb(alt_odb_fn fn, void *cb)
664 struct alternate_object_database *ent;
665 int r = 0;
667 prepare_alt_odb(the_repository);
668 for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
669 r = fn(ent, cb);
670 if (r)
671 break;
673 return r;
676 void prepare_alt_odb_the_repository(void)
678 if (the_repository->objects->alt_odb_tail)
679 return;
681 the_repository->objects->alt_odb_tail =
682 &the_repository->objects->alt_odb_list;
683 link_alt_odb_entries(the_repository,
684 the_repository->objects->alternate_db,
685 PATH_SEP, NULL, 0);
687 read_info_alternates(the_repository, get_object_directory(), 0);
690 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
691 static int freshen_file(const char *fn)
693 struct utimbuf t;
694 t.actime = t.modtime = time(NULL);
695 return !utime(fn, &t);
699 * All of the check_and_freshen functions return 1 if the file exists and was
700 * freshened (if freshening was requested), 0 otherwise. If they return
701 * 0, you should not assume that it is safe to skip a write of the object (it
702 * either does not exist on disk, or has a stale mtime and may be subject to
703 * pruning).
705 int check_and_freshen_file(const char *fn, int freshen)
707 if (access(fn, F_OK))
708 return 0;
709 if (freshen && !freshen_file(fn))
710 return 0;
711 return 1;
714 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
716 static struct strbuf buf = STRBUF_INIT;
718 strbuf_reset(&buf);
719 sha1_file_name(&buf, sha1);
721 return check_and_freshen_file(buf.buf, freshen);
724 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
726 struct alternate_object_database *alt;
727 prepare_alt_odb(the_repository);
728 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
729 const char *path = alt_sha1_path(alt, sha1);
730 if (check_and_freshen_file(path, freshen))
731 return 1;
733 return 0;
736 static int check_and_freshen(const unsigned char *sha1, int freshen)
738 return check_and_freshen_local(sha1, freshen) ||
739 check_and_freshen_nonlocal(sha1, freshen);
742 int has_loose_object_nonlocal(const unsigned char *sha1)
744 return check_and_freshen_nonlocal(sha1, 0);
747 static int has_loose_object(const unsigned char *sha1)
749 return check_and_freshen(sha1, 0);
752 static void mmap_limit_check(size_t length)
754 static size_t limit = 0;
755 if (!limit) {
756 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
757 if (!limit)
758 limit = SIZE_MAX;
760 if (length > limit)
761 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
762 (uintmax_t)length, (uintmax_t)limit);
765 void *xmmap_gently(void *start, size_t length,
766 int prot, int flags, int fd, off_t offset)
768 void *ret;
770 mmap_limit_check(length);
771 ret = mmap(start, length, prot, flags, fd, offset);
772 if (ret == MAP_FAILED) {
773 if (!length)
774 return NULL;
775 release_pack_memory(length);
776 ret = mmap(start, length, prot, flags, fd, offset);
778 return ret;
781 void *xmmap(void *start, size_t length,
782 int prot, int flags, int fd, off_t offset)
784 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
785 if (ret == MAP_FAILED)
786 die_errno("mmap failed");
787 return ret;
791 * With an in-core object data in "map", rehash it to make sure the
792 * object name actually matches "sha1" to detect object corruption.
793 * With "map" == NULL, try reading the object named with "sha1" using
794 * the streaming interface and rehash it to do the same.
796 int check_sha1_signature(const unsigned char *sha1, void *map,
797 unsigned long size, const char *type)
799 unsigned char real_sha1[20];
800 enum object_type obj_type;
801 struct git_istream *st;
802 git_SHA_CTX c;
803 char hdr[32];
804 int hdrlen;
806 if (map) {
807 hash_sha1_file(map, size, type, real_sha1);
808 return hashcmp(sha1, real_sha1) ? -1 : 0;
811 st = open_istream(sha1, &obj_type, &size, NULL);
812 if (!st)
813 return -1;
815 /* Generate the header */
816 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
818 /* Sha1.. */
819 git_SHA1_Init(&c);
820 git_SHA1_Update(&c, hdr, hdrlen);
821 for (;;) {
822 char buf[1024 * 16];
823 ssize_t readlen = read_istream(st, buf, sizeof(buf));
825 if (readlen < 0) {
826 close_istream(st);
827 return -1;
829 if (!readlen)
830 break;
831 git_SHA1_Update(&c, buf, readlen);
833 git_SHA1_Final(real_sha1, &c);
834 close_istream(st);
835 return hashcmp(sha1, real_sha1) ? -1 : 0;
838 int git_open_cloexec(const char *name, int flags)
840 int fd;
841 static int o_cloexec = O_CLOEXEC;
843 fd = open(name, flags | o_cloexec);
844 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
845 /* Try again w/o O_CLOEXEC: the kernel might not support it */
846 o_cloexec &= ~O_CLOEXEC;
847 fd = open(name, flags | o_cloexec);
850 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
852 static int fd_cloexec = FD_CLOEXEC;
854 if (!o_cloexec && 0 <= fd && fd_cloexec) {
855 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
856 int flags = fcntl(fd, F_GETFD);
857 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
858 fd_cloexec = 0;
861 #endif
862 return fd;
866 * Find "sha1" as a loose object in the local repository or in an alternate.
867 * Returns 0 on success, negative on failure.
869 * The "path" out-parameter will give the path of the object we found (if any).
870 * Note that it may point to static storage and is only valid until another
871 * call to sha1_file_name(), etc.
873 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
874 const char **path)
876 struct alternate_object_database *alt;
877 static struct strbuf buf = STRBUF_INIT;
879 strbuf_reset(&buf);
880 sha1_file_name(&buf, sha1);
881 *path = buf.buf;
883 if (!lstat(*path, st))
884 return 0;
886 prepare_alt_odb(the_repository);
887 errno = ENOENT;
888 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
889 *path = alt_sha1_path(alt, sha1);
890 if (!lstat(*path, st))
891 return 0;
894 return -1;
898 * Like stat_sha1_file(), but actually open the object and return the
899 * descriptor. See the caveats on the "path" parameter above.
901 static int open_sha1_file(const unsigned char *sha1, const char **path)
903 int fd;
904 struct alternate_object_database *alt;
905 int most_interesting_errno;
906 static struct strbuf buf = STRBUF_INIT;
908 strbuf_reset(&buf);
909 sha1_file_name(&buf, sha1);
910 *path = buf.buf;
912 fd = git_open(*path);
913 if (fd >= 0)
914 return fd;
915 most_interesting_errno = errno;
917 prepare_alt_odb(the_repository);
918 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
919 *path = alt_sha1_path(alt, sha1);
920 fd = git_open(*path);
921 if (fd >= 0)
922 return fd;
923 if (most_interesting_errno == ENOENT)
924 most_interesting_errno = errno;
926 errno = most_interesting_errno;
927 return -1;
931 * Map the loose object at "path" if it is not NULL, or the path found by
932 * searching for a loose object named "sha1".
934 static void *map_sha1_file_1(const char *path,
935 const unsigned char *sha1,
936 unsigned long *size)
938 void *map;
939 int fd;
941 if (path)
942 fd = git_open(path);
943 else
944 fd = open_sha1_file(sha1, &path);
945 map = NULL;
946 if (fd >= 0) {
947 struct stat st;
949 if (!fstat(fd, &st)) {
950 *size = xsize_t(st.st_size);
951 if (!*size) {
952 /* mmap() is forbidden on empty files */
953 error("object file %s is empty", path);
954 return NULL;
956 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
958 close(fd);
960 return map;
963 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
965 return map_sha1_file_1(NULL, sha1, size);
968 static int unpack_sha1_short_header(git_zstream *stream,
969 unsigned char *map, unsigned long mapsize,
970 void *buffer, unsigned long bufsiz)
972 /* Get the data stream */
973 memset(stream, 0, sizeof(*stream));
974 stream->next_in = map;
975 stream->avail_in = mapsize;
976 stream->next_out = buffer;
977 stream->avail_out = bufsiz;
979 git_inflate_init(stream);
980 return git_inflate(stream, 0);
983 int unpack_sha1_header(git_zstream *stream,
984 unsigned char *map, unsigned long mapsize,
985 void *buffer, unsigned long bufsiz)
987 int status = unpack_sha1_short_header(stream, map, mapsize,
988 buffer, bufsiz);
990 if (status < Z_OK)
991 return status;
993 /* Make sure we have the terminating NUL */
994 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
995 return -1;
996 return 0;
999 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1000 unsigned long mapsize, void *buffer,
1001 unsigned long bufsiz, struct strbuf *header)
1003 int status;
1005 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1006 if (status < Z_OK)
1007 return -1;
1010 * Check if entire header is unpacked in the first iteration.
1012 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1013 return 0;
1016 * buffer[0..bufsiz] was not large enough. Copy the partial
1017 * result out to header, and then append the result of further
1018 * reading the stream.
1020 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1021 stream->next_out = buffer;
1022 stream->avail_out = bufsiz;
1024 do {
1025 status = git_inflate(stream, 0);
1026 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1027 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1028 return 0;
1029 stream->next_out = buffer;
1030 stream->avail_out = bufsiz;
1031 } while (status != Z_STREAM_END);
1032 return -1;
1035 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1037 int bytes = strlen(buffer) + 1;
1038 unsigned char *buf = xmallocz(size);
1039 unsigned long n;
1040 int status = Z_OK;
1042 n = stream->total_out - bytes;
1043 if (n > size)
1044 n = size;
1045 memcpy(buf, (char *) buffer + bytes, n);
1046 bytes = n;
1047 if (bytes <= size) {
1049 * The above condition must be (bytes <= size), not
1050 * (bytes < size). In other words, even though we
1051 * expect no more output and set avail_out to zero,
1052 * the input zlib stream may have bytes that express
1053 * "this concludes the stream", and we *do* want to
1054 * eat that input.
1056 * Otherwise we would not be able to test that we
1057 * consumed all the input to reach the expected size;
1058 * we also want to check that zlib tells us that all
1059 * went well with status == Z_STREAM_END at the end.
1061 stream->next_out = buf + bytes;
1062 stream->avail_out = size - bytes;
1063 while (status == Z_OK)
1064 status = git_inflate(stream, Z_FINISH);
1066 if (status == Z_STREAM_END && !stream->avail_in) {
1067 git_inflate_end(stream);
1068 return buf;
1071 if (status < 0)
1072 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1073 else if (stream->avail_in)
1074 error("garbage at end of loose object '%s'",
1075 sha1_to_hex(sha1));
1076 free(buf);
1077 return NULL;
1081 * We used to just use "sscanf()", but that's actually way
1082 * too permissive for what we want to check. So do an anal
1083 * object header parse by hand.
1085 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1086 unsigned int flags)
1088 const char *type_buf = hdr;
1089 unsigned long size;
1090 int type, type_len = 0;
1093 * The type can be of any size but is followed by
1094 * a space.
1096 for (;;) {
1097 char c = *hdr++;
1098 if (!c)
1099 return -1;
1100 if (c == ' ')
1101 break;
1102 type_len++;
1105 type = type_from_string_gently(type_buf, type_len, 1);
1106 if (oi->typename)
1107 strbuf_add(oi->typename, type_buf, type_len);
1109 * Set type to 0 if its an unknown object and
1110 * we're obtaining the type using '--allow-unknown-type'
1111 * option.
1113 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1114 type = 0;
1115 else if (type < 0)
1116 die("invalid object type");
1117 if (oi->typep)
1118 *oi->typep = type;
1121 * The length must follow immediately, and be in canonical
1122 * decimal format (ie "010" is not valid).
1124 size = *hdr++ - '0';
1125 if (size > 9)
1126 return -1;
1127 if (size) {
1128 for (;;) {
1129 unsigned long c = *hdr - '0';
1130 if (c > 9)
1131 break;
1132 hdr++;
1133 size = size * 10 + c;
1137 if (oi->sizep)
1138 *oi->sizep = size;
1141 * The length must be followed by a zero byte
1143 return *hdr ? -1 : type;
1146 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1148 struct object_info oi = OBJECT_INFO_INIT;
1150 oi.sizep = sizep;
1151 return parse_sha1_header_extended(hdr, &oi, 0);
1154 static int sha1_loose_object_info(const unsigned char *sha1,
1155 struct object_info *oi,
1156 int flags)
1158 int status = 0;
1159 unsigned long mapsize;
1160 void *map;
1161 git_zstream stream;
1162 char hdr[32];
1163 struct strbuf hdrbuf = STRBUF_INIT;
1164 unsigned long size_scratch;
1166 if (oi->delta_base_sha1)
1167 hashclr(oi->delta_base_sha1);
1170 * If we don't care about type or size, then we don't
1171 * need to look inside the object at all. Note that we
1172 * do not optimize out the stat call, even if the
1173 * caller doesn't care about the disk-size, since our
1174 * return value implicitly indicates whether the
1175 * object even exists.
1177 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1178 const char *path;
1179 struct stat st;
1180 if (stat_sha1_file(sha1, &st, &path) < 0)
1181 return -1;
1182 if (oi->disk_sizep)
1183 *oi->disk_sizep = st.st_size;
1184 return 0;
1187 map = map_sha1_file(sha1, &mapsize);
1188 if (!map)
1189 return -1;
1191 if (!oi->sizep)
1192 oi->sizep = &size_scratch;
1194 if (oi->disk_sizep)
1195 *oi->disk_sizep = mapsize;
1196 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1197 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1198 status = error("unable to unpack %s header with --allow-unknown-type",
1199 sha1_to_hex(sha1));
1200 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1201 status = error("unable to unpack %s header",
1202 sha1_to_hex(sha1));
1203 if (status < 0)
1204 ; /* Do nothing */
1205 else if (hdrbuf.len) {
1206 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1207 status = error("unable to parse %s header with --allow-unknown-type",
1208 sha1_to_hex(sha1));
1209 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1210 status = error("unable to parse %s header", sha1_to_hex(sha1));
1212 if (status >= 0 && oi->contentp) {
1213 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1214 *oi->sizep, sha1);
1215 if (!*oi->contentp) {
1216 git_inflate_end(&stream);
1217 status = -1;
1219 } else
1220 git_inflate_end(&stream);
1222 munmap(map, mapsize);
1223 if (status && oi->typep)
1224 *oi->typep = status;
1225 if (oi->sizep == &size_scratch)
1226 oi->sizep = NULL;
1227 strbuf_release(&hdrbuf);
1228 oi->whence = OI_LOOSE;
1229 return (status < 0) ? status : 0;
1232 int fetch_if_missing = 1;
1234 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1236 static struct object_info blank_oi = OBJECT_INFO_INIT;
1237 struct pack_entry e;
1238 int rtype;
1239 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1240 lookup_replace_object(sha1) :
1241 sha1;
1242 int already_retried = 0;
1244 if (is_null_sha1(real))
1245 return -1;
1247 if (!oi)
1248 oi = &blank_oi;
1250 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1251 struct cached_object *co = find_cached_object(real);
1252 if (co) {
1253 if (oi->typep)
1254 *(oi->typep) = co->type;
1255 if (oi->sizep)
1256 *(oi->sizep) = co->size;
1257 if (oi->disk_sizep)
1258 *(oi->disk_sizep) = 0;
1259 if (oi->delta_base_sha1)
1260 hashclr(oi->delta_base_sha1);
1261 if (oi->typename)
1262 strbuf_addstr(oi->typename, typename(co->type));
1263 if (oi->contentp)
1264 *oi->contentp = xmemdupz(co->buf, co->size);
1265 oi->whence = OI_CACHED;
1266 return 0;
1270 while (1) {
1271 if (find_pack_entry(real, &e))
1272 break;
1274 /* Most likely it's a loose object. */
1275 if (!sha1_loose_object_info(real, oi, flags))
1276 return 0;
1278 /* Not a loose object; someone else may have just packed it. */
1279 reprepare_packed_git();
1280 if (find_pack_entry(real, &e))
1281 break;
1283 /* Check if it is a missing object */
1284 if (fetch_if_missing && repository_format_partial_clone &&
1285 !already_retried) {
1287 * TODO Investigate haveing fetch_object() return
1288 * TODO error/success and stopping the music here.
1290 fetch_object(repository_format_partial_clone, real);
1291 already_retried = 1;
1292 continue;
1295 return -1;
1298 if (oi == &blank_oi)
1300 * We know that the caller doesn't actually need the
1301 * information below, so return early.
1303 return 0;
1304 rtype = packed_object_info(e.p, e.offset, oi);
1305 if (rtype < 0) {
1306 mark_bad_packed_object(e.p, real);
1307 return sha1_object_info_extended(real, oi, 0);
1308 } else if (oi->whence == OI_PACKED) {
1309 oi->u.packed.offset = e.offset;
1310 oi->u.packed.pack = e.p;
1311 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1312 rtype == OBJ_OFS_DELTA);
1315 return 0;
1318 /* returns enum object_type or negative */
1319 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1321 enum object_type type;
1322 struct object_info oi = OBJECT_INFO_INIT;
1324 oi.typep = &type;
1325 oi.sizep = sizep;
1326 if (sha1_object_info_extended(sha1, &oi,
1327 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1328 return -1;
1329 return type;
1332 static void *read_object(const unsigned char *sha1, enum object_type *type,
1333 unsigned long *size)
1335 struct object_info oi = OBJECT_INFO_INIT;
1336 void *content;
1337 oi.typep = type;
1338 oi.sizep = size;
1339 oi.contentp = &content;
1341 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1342 return NULL;
1343 return content;
1346 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1347 unsigned char *sha1)
1349 struct cached_object *co;
1351 hash_sha1_file(buf, len, typename(type), sha1);
1352 if (has_sha1_file(sha1) || find_cached_object(sha1))
1353 return 0;
1354 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1355 co = &cached_objects[cached_object_nr++];
1356 co->size = len;
1357 co->type = type;
1358 co->buf = xmalloc(len);
1359 memcpy(co->buf, buf, len);
1360 hashcpy(co->sha1, sha1);
1361 return 0;
1365 * This function dies on corrupt objects; the callers who want to
1366 * deal with them should arrange to call read_object() and give error
1367 * messages themselves.
1369 void *read_sha1_file_extended(const unsigned char *sha1,
1370 enum object_type *type,
1371 unsigned long *size,
1372 int lookup_replace)
1374 void *data;
1375 const struct packed_git *p;
1376 const char *path;
1377 struct stat st;
1378 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1379 : sha1;
1381 errno = 0;
1382 data = read_object(repl, type, size);
1383 if (data)
1384 return data;
1386 if (errno && errno != ENOENT)
1387 die_errno("failed to read object %s", sha1_to_hex(sha1));
1389 /* die if we replaced an object with one that does not exist */
1390 if (repl != sha1)
1391 die("replacement %s not found for %s",
1392 sha1_to_hex(repl), sha1_to_hex(sha1));
1394 if (!stat_sha1_file(repl, &st, &path))
1395 die("loose object %s (stored in %s) is corrupt",
1396 sha1_to_hex(repl), path);
1398 if ((p = has_packed_and_bad(repl)) != NULL)
1399 die("packed object %s (stored in %s) is corrupt",
1400 sha1_to_hex(repl), p->pack_name);
1402 return NULL;
1405 void *read_object_with_reference(const unsigned char *sha1,
1406 const char *required_type_name,
1407 unsigned long *size,
1408 unsigned char *actual_sha1_return)
1410 enum object_type type, required_type;
1411 void *buffer;
1412 unsigned long isize;
1413 unsigned char actual_sha1[20];
1415 required_type = type_from_string(required_type_name);
1416 hashcpy(actual_sha1, sha1);
1417 while (1) {
1418 int ref_length = -1;
1419 const char *ref_type = NULL;
1421 buffer = read_sha1_file(actual_sha1, &type, &isize);
1422 if (!buffer)
1423 return NULL;
1424 if (type == required_type) {
1425 *size = isize;
1426 if (actual_sha1_return)
1427 hashcpy(actual_sha1_return, actual_sha1);
1428 return buffer;
1430 /* Handle references */
1431 else if (type == OBJ_COMMIT)
1432 ref_type = "tree ";
1433 else if (type == OBJ_TAG)
1434 ref_type = "object ";
1435 else {
1436 free(buffer);
1437 return NULL;
1439 ref_length = strlen(ref_type);
1441 if (ref_length + 40 > isize ||
1442 memcmp(buffer, ref_type, ref_length) ||
1443 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1444 free(buffer);
1445 return NULL;
1447 free(buffer);
1448 /* Now we have the ID of the referred-to object in
1449 * actual_sha1. Check again. */
1453 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1454 const char *type, unsigned char *sha1,
1455 char *hdr, int *hdrlen)
1457 git_SHA_CTX c;
1459 /* Generate the header */
1460 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1462 /* Sha1.. */
1463 git_SHA1_Init(&c);
1464 git_SHA1_Update(&c, hdr, *hdrlen);
1465 git_SHA1_Update(&c, buf, len);
1466 git_SHA1_Final(sha1, &c);
1470 * Move the just written object into its final resting place.
1472 int finalize_object_file(const char *tmpfile, const char *filename)
1474 int ret = 0;
1476 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1477 goto try_rename;
1478 else if (link(tmpfile, filename))
1479 ret = errno;
1482 * Coda hack - coda doesn't like cross-directory links,
1483 * so we fall back to a rename, which will mean that it
1484 * won't be able to check collisions, but that's not a
1485 * big deal.
1487 * The same holds for FAT formatted media.
1489 * When this succeeds, we just return. We have nothing
1490 * left to unlink.
1492 if (ret && ret != EEXIST) {
1493 try_rename:
1494 if (!rename(tmpfile, filename))
1495 goto out;
1496 ret = errno;
1498 unlink_or_warn(tmpfile);
1499 if (ret) {
1500 if (ret != EEXIST) {
1501 return error_errno("unable to write sha1 filename %s", filename);
1503 /* FIXME!!! Collision check here ? */
1506 out:
1507 if (adjust_shared_perm(filename))
1508 return error("unable to set permission to '%s'", filename);
1509 return 0;
1512 static int write_buffer(int fd, const void *buf, size_t len)
1514 if (write_in_full(fd, buf, len) < 0)
1515 return error_errno("file write error");
1516 return 0;
1519 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1520 unsigned char *sha1)
1522 char hdr[32];
1523 int hdrlen = sizeof(hdr);
1524 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1525 return 0;
1528 /* Finalize a file on disk, and close it. */
1529 static void close_sha1_file(int fd)
1531 if (fsync_object_files)
1532 fsync_or_die(fd, "sha1 file");
1533 if (close(fd) != 0)
1534 die_errno("error when closing sha1 file");
1537 /* Size of directory component, including the ending '/' */
1538 static inline int directory_size(const char *filename)
1540 const char *s = strrchr(filename, '/');
1541 if (!s)
1542 return 0;
1543 return s - filename + 1;
1547 * This creates a temporary file in the same directory as the final
1548 * 'filename'
1550 * We want to avoid cross-directory filename renames, because those
1551 * can have problems on various filesystems (FAT, NFS, Coda).
1553 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1555 int fd, dirlen = directory_size(filename);
1557 strbuf_reset(tmp);
1558 strbuf_add(tmp, filename, dirlen);
1559 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1560 fd = git_mkstemp_mode(tmp->buf, 0444);
1561 if (fd < 0 && dirlen && errno == ENOENT) {
1563 * Make sure the directory exists; note that the contents
1564 * of the buffer are undefined after mkstemp returns an
1565 * error, so we have to rewrite the whole buffer from
1566 * scratch.
1568 strbuf_reset(tmp);
1569 strbuf_add(tmp, filename, dirlen - 1);
1570 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1571 return -1;
1572 if (adjust_shared_perm(tmp->buf))
1573 return -1;
1575 /* Try again */
1576 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1577 fd = git_mkstemp_mode(tmp->buf, 0444);
1579 return fd;
1582 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1583 const void *buf, unsigned long len, time_t mtime)
1585 int fd, ret;
1586 unsigned char compressed[4096];
1587 git_zstream stream;
1588 git_SHA_CTX c;
1589 unsigned char parano_sha1[20];
1590 static struct strbuf tmp_file = STRBUF_INIT;
1591 static struct strbuf filename = STRBUF_INIT;
1593 strbuf_reset(&filename);
1594 sha1_file_name(&filename, sha1);
1596 fd = create_tmpfile(&tmp_file, filename.buf);
1597 if (fd < 0) {
1598 if (errno == EACCES)
1599 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1600 else
1601 return error_errno("unable to create temporary file");
1604 /* Set it up */
1605 git_deflate_init(&stream, zlib_compression_level);
1606 stream.next_out = compressed;
1607 stream.avail_out = sizeof(compressed);
1608 git_SHA1_Init(&c);
1610 /* First header.. */
1611 stream.next_in = (unsigned char *)hdr;
1612 stream.avail_in = hdrlen;
1613 while (git_deflate(&stream, 0) == Z_OK)
1614 ; /* nothing */
1615 git_SHA1_Update(&c, hdr, hdrlen);
1617 /* Then the data itself.. */
1618 stream.next_in = (void *)buf;
1619 stream.avail_in = len;
1620 do {
1621 unsigned char *in0 = stream.next_in;
1622 ret = git_deflate(&stream, Z_FINISH);
1623 git_SHA1_Update(&c, in0, stream.next_in - in0);
1624 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1625 die("unable to write sha1 file");
1626 stream.next_out = compressed;
1627 stream.avail_out = sizeof(compressed);
1628 } while (ret == Z_OK);
1630 if (ret != Z_STREAM_END)
1631 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1632 ret = git_deflate_end_gently(&stream);
1633 if (ret != Z_OK)
1634 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1635 git_SHA1_Final(parano_sha1, &c);
1636 if (hashcmp(sha1, parano_sha1) != 0)
1637 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1639 close_sha1_file(fd);
1641 if (mtime) {
1642 struct utimbuf utb;
1643 utb.actime = mtime;
1644 utb.modtime = mtime;
1645 if (utime(tmp_file.buf, &utb) < 0)
1646 warning_errno("failed utime() on %s", tmp_file.buf);
1649 return finalize_object_file(tmp_file.buf, filename.buf);
1652 static int freshen_loose_object(const unsigned char *sha1)
1654 return check_and_freshen(sha1, 1);
1657 static int freshen_packed_object(const unsigned char *sha1)
1659 struct pack_entry e;
1660 if (!find_pack_entry(sha1, &e))
1661 return 0;
1662 if (e.p->freshened)
1663 return 1;
1664 if (!freshen_file(e.p->pack_name))
1665 return 0;
1666 e.p->freshened = 1;
1667 return 1;
1670 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1672 char hdr[32];
1673 int hdrlen = sizeof(hdr);
1675 /* Normally if we have it in the pack then we do not bother writing
1676 * it out into .git/objects/??/?{38} file.
1678 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1679 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1680 return 0;
1681 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1684 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1685 struct object_id *oid, unsigned flags)
1687 char *header;
1688 int hdrlen, status = 0;
1690 /* type string, SP, %lu of the length plus NUL must fit this */
1691 hdrlen = strlen(type) + 32;
1692 header = xmalloc(hdrlen);
1693 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1695 if (!(flags & HASH_WRITE_OBJECT))
1696 goto cleanup;
1697 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1698 goto cleanup;
1699 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1701 cleanup:
1702 free(header);
1703 return status;
1706 int force_object_loose(const unsigned char *sha1, time_t mtime)
1708 void *buf;
1709 unsigned long len;
1710 enum object_type type;
1711 char hdr[32];
1712 int hdrlen;
1713 int ret;
1715 if (has_loose_object(sha1))
1716 return 0;
1717 buf = read_object(sha1, &type, &len);
1718 if (!buf)
1719 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1720 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1721 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1722 free(buf);
1724 return ret;
1727 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1729 if (!startup_info->have_repository)
1730 return 0;
1731 return sha1_object_info_extended(sha1, NULL,
1732 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1735 int has_object_file(const struct object_id *oid)
1737 return has_sha1_file(oid->hash);
1740 int has_object_file_with_flags(const struct object_id *oid, int flags)
1742 return has_sha1_file_with_flags(oid->hash, flags);
1745 static void check_tree(const void *buf, size_t size)
1747 struct tree_desc desc;
1748 struct name_entry entry;
1750 init_tree_desc(&desc, buf, size);
1751 while (tree_entry(&desc, &entry))
1752 /* do nothing
1753 * tree_entry() will die() on malformed entries */
1757 static void check_commit(const void *buf, size_t size)
1759 struct commit c;
1760 memset(&c, 0, sizeof(c));
1761 if (parse_commit_buffer(&c, buf, size))
1762 die("corrupt commit");
1765 static void check_tag(const void *buf, size_t size)
1767 struct tag t;
1768 memset(&t, 0, sizeof(t));
1769 if (parse_tag_buffer(&t, buf, size))
1770 die("corrupt tag");
1773 static int index_mem(struct object_id *oid, void *buf, size_t size,
1774 enum object_type type,
1775 const char *path, unsigned flags)
1777 int ret, re_allocated = 0;
1778 int write_object = flags & HASH_WRITE_OBJECT;
1780 if (!type)
1781 type = OBJ_BLOB;
1784 * Convert blobs to git internal format
1786 if ((type == OBJ_BLOB) && path) {
1787 struct strbuf nbuf = STRBUF_INIT;
1788 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1789 get_conv_flags(flags))) {
1790 buf = strbuf_detach(&nbuf, &size);
1791 re_allocated = 1;
1794 if (flags & HASH_FORMAT_CHECK) {
1795 if (type == OBJ_TREE)
1796 check_tree(buf, size);
1797 if (type == OBJ_COMMIT)
1798 check_commit(buf, size);
1799 if (type == OBJ_TAG)
1800 check_tag(buf, size);
1803 if (write_object)
1804 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1805 else
1806 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1807 if (re_allocated)
1808 free(buf);
1809 return ret;
1812 static int index_stream_convert_blob(struct object_id *oid, int fd,
1813 const char *path, unsigned flags)
1815 int ret;
1816 const int write_object = flags & HASH_WRITE_OBJECT;
1817 struct strbuf sbuf = STRBUF_INIT;
1819 assert(path);
1820 assert(would_convert_to_git_filter_fd(path));
1822 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1823 get_conv_flags(flags));
1825 if (write_object)
1826 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1827 oid->hash);
1828 else
1829 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1830 oid->hash);
1831 strbuf_release(&sbuf);
1832 return ret;
1835 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1836 const char *path, unsigned flags)
1838 struct strbuf sbuf = STRBUF_INIT;
1839 int ret;
1841 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1842 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1843 else
1844 ret = -1;
1845 strbuf_release(&sbuf);
1846 return ret;
1849 #define SMALL_FILE_SIZE (32*1024)
1851 static int index_core(struct object_id *oid, int fd, size_t size,
1852 enum object_type type, const char *path,
1853 unsigned flags)
1855 int ret;
1857 if (!size) {
1858 ret = index_mem(oid, "", size, type, path, flags);
1859 } else if (size <= SMALL_FILE_SIZE) {
1860 char *buf = xmalloc(size);
1861 ssize_t read_result = read_in_full(fd, buf, size);
1862 if (read_result < 0)
1863 ret = error_errno("read error while indexing %s",
1864 path ? path : "<unknown>");
1865 else if (read_result != size)
1866 ret = error("short read while indexing %s",
1867 path ? path : "<unknown>");
1868 else
1869 ret = index_mem(oid, buf, size, type, path, flags);
1870 free(buf);
1871 } else {
1872 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1873 ret = index_mem(oid, buf, size, type, path, flags);
1874 munmap(buf, size);
1876 return ret;
1880 * This creates one packfile per large blob unless bulk-checkin
1881 * machinery is "plugged".
1883 * This also bypasses the usual "convert-to-git" dance, and that is on
1884 * purpose. We could write a streaming version of the converting
1885 * functions and insert that before feeding the data to fast-import
1886 * (or equivalent in-core API described above). However, that is
1887 * somewhat complicated, as we do not know the size of the filter
1888 * result, which we need to know beforehand when writing a git object.
1889 * Since the primary motivation for trying to stream from the working
1890 * tree file and to avoid mmaping it in core is to deal with large
1891 * binary blobs, they generally do not want to get any conversion, and
1892 * callers should avoid this code path when filters are requested.
1894 static int index_stream(struct object_id *oid, int fd, size_t size,
1895 enum object_type type, const char *path,
1896 unsigned flags)
1898 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1901 int index_fd(struct object_id *oid, int fd, struct stat *st,
1902 enum object_type type, const char *path, unsigned flags)
1904 int ret;
1907 * Call xsize_t() only when needed to avoid potentially unnecessary
1908 * die() for large files.
1910 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1911 ret = index_stream_convert_blob(oid, fd, path, flags);
1912 else if (!S_ISREG(st->st_mode))
1913 ret = index_pipe(oid, fd, type, path, flags);
1914 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1915 (path && would_convert_to_git(&the_index, path)))
1916 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1917 flags);
1918 else
1919 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1920 flags);
1921 close(fd);
1922 return ret;
1925 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1927 int fd;
1928 struct strbuf sb = STRBUF_INIT;
1929 int rc = 0;
1931 switch (st->st_mode & S_IFMT) {
1932 case S_IFREG:
1933 fd = open(path, O_RDONLY);
1934 if (fd < 0)
1935 return error_errno("open(\"%s\")", path);
1936 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1937 return error("%s: failed to insert into database",
1938 path);
1939 break;
1940 case S_IFLNK:
1941 if (strbuf_readlink(&sb, path, st->st_size))
1942 return error_errno("readlink(\"%s\")", path);
1943 if (!(flags & HASH_WRITE_OBJECT))
1944 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1945 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1946 rc = error("%s: failed to insert into database", path);
1947 strbuf_release(&sb);
1948 break;
1949 case S_IFDIR:
1950 return resolve_gitlink_ref(path, "HEAD", oid);
1951 default:
1952 return error("%s: unsupported file type", path);
1954 return rc;
1957 int read_pack_header(int fd, struct pack_header *header)
1959 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1960 /* "eof before pack header was fully read" */
1961 return PH_ERROR_EOF;
1963 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1964 /* "protocol error (pack signature mismatch detected)" */
1965 return PH_ERROR_PACK_SIGNATURE;
1966 if (!pack_version_ok(header->hdr_version))
1967 /* "protocol error (pack version unsupported)" */
1968 return PH_ERROR_PROTOCOL;
1969 return 0;
1972 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1974 enum object_type type = sha1_object_info(sha1, NULL);
1975 if (type < 0)
1976 die("%s is not a valid object", sha1_to_hex(sha1));
1977 if (type != expect)
1978 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1979 typename(expect));
1982 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1983 struct strbuf *path,
1984 each_loose_object_fn obj_cb,
1985 each_loose_cruft_fn cruft_cb,
1986 each_loose_subdir_fn subdir_cb,
1987 void *data)
1989 size_t origlen, baselen;
1990 DIR *dir;
1991 struct dirent *de;
1992 int r = 0;
1993 struct object_id oid;
1995 if (subdir_nr > 0xff)
1996 BUG("invalid loose object subdirectory: %x", subdir_nr);
1998 origlen = path->len;
1999 strbuf_complete(path, '/');
2000 strbuf_addf(path, "%02x", subdir_nr);
2002 dir = opendir(path->buf);
2003 if (!dir) {
2004 if (errno != ENOENT)
2005 r = error_errno("unable to open %s", path->buf);
2006 strbuf_setlen(path, origlen);
2007 return r;
2010 oid.hash[0] = subdir_nr;
2011 strbuf_addch(path, '/');
2012 baselen = path->len;
2014 while ((de = readdir(dir))) {
2015 size_t namelen;
2016 if (is_dot_or_dotdot(de->d_name))
2017 continue;
2019 namelen = strlen(de->d_name);
2020 strbuf_setlen(path, baselen);
2021 strbuf_add(path, de->d_name, namelen);
2022 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2023 !hex_to_bytes(oid.hash + 1, de->d_name,
2024 GIT_SHA1_RAWSZ - 1)) {
2025 if (obj_cb) {
2026 r = obj_cb(&oid, path->buf, data);
2027 if (r)
2028 break;
2030 continue;
2033 if (cruft_cb) {
2034 r = cruft_cb(de->d_name, path->buf, data);
2035 if (r)
2036 break;
2039 closedir(dir);
2041 strbuf_setlen(path, baselen - 1);
2042 if (!r && subdir_cb)
2043 r = subdir_cb(subdir_nr, path->buf, data);
2045 strbuf_setlen(path, origlen);
2047 return r;
2050 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2051 each_loose_object_fn obj_cb,
2052 each_loose_cruft_fn cruft_cb,
2053 each_loose_subdir_fn subdir_cb,
2054 void *data)
2056 int r = 0;
2057 int i;
2059 for (i = 0; i < 256; i++) {
2060 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2061 subdir_cb, data);
2062 if (r)
2063 break;
2066 return r;
2069 int for_each_loose_file_in_objdir(const char *path,
2070 each_loose_object_fn obj_cb,
2071 each_loose_cruft_fn cruft_cb,
2072 each_loose_subdir_fn subdir_cb,
2073 void *data)
2075 struct strbuf buf = STRBUF_INIT;
2076 int r;
2078 strbuf_addstr(&buf, path);
2079 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2080 subdir_cb, data);
2081 strbuf_release(&buf);
2083 return r;
2086 struct loose_alt_odb_data {
2087 each_loose_object_fn *cb;
2088 void *data;
2091 static int loose_from_alt_odb(struct alternate_object_database *alt,
2092 void *vdata)
2094 struct loose_alt_odb_data *data = vdata;
2095 struct strbuf buf = STRBUF_INIT;
2096 int r;
2098 strbuf_addstr(&buf, alt->path);
2099 r = for_each_loose_file_in_objdir_buf(&buf,
2100 data->cb, NULL, NULL,
2101 data->data);
2102 strbuf_release(&buf);
2103 return r;
2106 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2108 struct loose_alt_odb_data alt;
2109 int r;
2111 r = for_each_loose_file_in_objdir(get_object_directory(),
2112 cb, NULL, NULL, data);
2113 if (r)
2114 return r;
2116 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2117 return 0;
2119 alt.cb = cb;
2120 alt.data = data;
2121 return foreach_alt_odb(loose_from_alt_odb, &alt);
2124 static int check_stream_sha1(git_zstream *stream,
2125 const char *hdr,
2126 unsigned long size,
2127 const char *path,
2128 const unsigned char *expected_sha1)
2130 git_SHA_CTX c;
2131 unsigned char real_sha1[GIT_MAX_RAWSZ];
2132 unsigned char buf[4096];
2133 unsigned long total_read;
2134 int status = Z_OK;
2136 git_SHA1_Init(&c);
2137 git_SHA1_Update(&c, hdr, stream->total_out);
2140 * We already read some bytes into hdr, but the ones up to the NUL
2141 * do not count against the object's content size.
2143 total_read = stream->total_out - strlen(hdr) - 1;
2146 * This size comparison must be "<=" to read the final zlib packets;
2147 * see the comment in unpack_sha1_rest for details.
2149 while (total_read <= size &&
2150 (status == Z_OK || status == Z_BUF_ERROR)) {
2151 stream->next_out = buf;
2152 stream->avail_out = sizeof(buf);
2153 if (size - total_read < stream->avail_out)
2154 stream->avail_out = size - total_read;
2155 status = git_inflate(stream, Z_FINISH);
2156 git_SHA1_Update(&c, buf, stream->next_out - buf);
2157 total_read += stream->next_out - buf;
2159 git_inflate_end(stream);
2161 if (status != Z_STREAM_END) {
2162 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2163 return -1;
2165 if (stream->avail_in) {
2166 error("garbage at end of loose object '%s'",
2167 sha1_to_hex(expected_sha1));
2168 return -1;
2171 git_SHA1_Final(real_sha1, &c);
2172 if (hashcmp(expected_sha1, real_sha1)) {
2173 error("sha1 mismatch for %s (expected %s)", path,
2174 sha1_to_hex(expected_sha1));
2175 return -1;
2178 return 0;
2181 int read_loose_object(const char *path,
2182 const unsigned char *expected_sha1,
2183 enum object_type *type,
2184 unsigned long *size,
2185 void **contents)
2187 int ret = -1;
2188 void *map = NULL;
2189 unsigned long mapsize;
2190 git_zstream stream;
2191 char hdr[32];
2193 *contents = NULL;
2195 map = map_sha1_file_1(path, NULL, &mapsize);
2196 if (!map) {
2197 error_errno("unable to mmap %s", path);
2198 goto out;
2201 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2202 error("unable to unpack header of %s", path);
2203 goto out;
2206 *type = parse_sha1_header(hdr, size);
2207 if (*type < 0) {
2208 error("unable to parse header of %s", path);
2209 git_inflate_end(&stream);
2210 goto out;
2213 if (*type == OBJ_BLOB) {
2214 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2215 goto out;
2216 } else {
2217 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2218 if (!*contents) {
2219 error("unable to unpack contents of %s", path);
2220 git_inflate_end(&stream);
2221 goto out;
2223 if (check_sha1_signature(expected_sha1, *contents,
2224 *size, typename(*type))) {
2225 error("sha1 mismatch for %s (expected %s)", path,
2226 sha1_to_hex(expected_sha1));
2227 free(*contents);
2228 goto out;
2232 ret = 0; /* everything checks out */
2234 out:
2235 if (map)
2236 munmap(map, mapsize);
2237 return ret;