pack: move approximate object count to object store
[git.git] / sha1_file.c
blob84b361c1255bc1ca5b487b6f1fbcb0e3c5d9f9c8
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 strbuf *path, const char *normalized_objdir)
352 struct alternate_object_database *alt;
354 /* Detect cases where alternate disappeared */
355 if (!is_directory(path->buf)) {
356 error("object directory %s does not exist; "
357 "check .git/objects/info/alternates.",
358 path->buf);
359 return 0;
363 * Prevent the common mistake of listing the same
364 * thing twice, or object directory itself.
366 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
367 if (!fspathcmp(path->buf, alt->path))
368 return 0;
370 if (!fspathcmp(path->buf, normalized_objdir))
371 return 0;
373 return 1;
377 * Prepare alternate object database registry.
379 * The variable alt_odb_list points at the list of struct
380 * alternate_object_database. The elements on this list come from
381 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
382 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
383 * whose contents is similar to that environment variable but can be
384 * LF separated. Its base points at a statically allocated buffer that
385 * contains "/the/directory/corresponding/to/.git/objects/...", while
386 * its name points just after the slash at the end of ".git/objects/"
387 * in the example above, and has enough space to hold 40-byte hex
388 * SHA1, an extra slash for the first level indirection, and the
389 * terminating NUL.
391 static void read_info_alternates(const char * relative_base, int depth);
392 static int link_alt_odb_entry(const char *entry, const char *relative_base,
393 int depth, const char *normalized_objdir)
395 struct alternate_object_database *ent;
396 struct strbuf pathbuf = STRBUF_INIT;
398 if (!is_absolute_path(entry) && relative_base) {
399 strbuf_realpath(&pathbuf, relative_base, 1);
400 strbuf_addch(&pathbuf, '/');
402 strbuf_addstr(&pathbuf, entry);
404 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
405 error("unable to normalize alternate object path: %s",
406 pathbuf.buf);
407 strbuf_release(&pathbuf);
408 return -1;
412 * The trailing slash after the directory name is given by
413 * this function at the end. Remove duplicates.
415 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
416 strbuf_setlen(&pathbuf, pathbuf.len - 1);
418 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
419 strbuf_release(&pathbuf);
420 return -1;
423 ent = alloc_alt_odb(pathbuf.buf);
425 /* add the alternate entry */
426 *the_repository->objects->alt_odb_tail = ent;
427 the_repository->objects->alt_odb_tail = &(ent->next);
428 ent->next = NULL;
430 /* recursively add alternates */
431 read_info_alternates(pathbuf.buf, depth + 1);
433 strbuf_release(&pathbuf);
434 return 0;
437 static const char *parse_alt_odb_entry(const char *string,
438 int sep,
439 struct strbuf *out)
441 const char *end;
443 strbuf_reset(out);
445 if (*string == '#') {
446 /* comment; consume up to next separator */
447 end = strchrnul(string, sep);
448 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
450 * quoted path; unquote_c_style has copied the
451 * data for us and set "end". Broken quoting (e.g.,
452 * an entry that doesn't end with a quote) falls
453 * back to the unquoted case below.
455 } else {
456 /* normal, unquoted path */
457 end = strchrnul(string, sep);
458 strbuf_add(out, string, end - string);
461 if (*end)
462 end++;
463 return end;
466 static void link_alt_odb_entries(const char *alt, int sep,
467 const char *relative_base, int depth)
469 struct strbuf objdirbuf = STRBUF_INIT;
470 struct strbuf entry = STRBUF_INIT;
472 if (!alt || !*alt)
473 return;
475 if (depth > 5) {
476 error("%s: ignoring alternate object stores, nesting too deep.",
477 relative_base);
478 return;
481 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
482 if (strbuf_normalize_path(&objdirbuf) < 0)
483 die("unable to normalize object directory: %s",
484 objdirbuf.buf);
486 while (*alt) {
487 alt = parse_alt_odb_entry(alt, sep, &entry);
488 if (!entry.len)
489 continue;
490 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
492 strbuf_release(&entry);
493 strbuf_release(&objdirbuf);
496 static void read_info_alternates(const char * relative_base, int depth)
498 char *path;
499 struct strbuf buf = STRBUF_INIT;
501 path = xstrfmt("%s/info/alternates", relative_base);
502 if (strbuf_read_file(&buf, path, 1024) < 0) {
503 warn_on_fopen_errors(path);
504 free(path);
505 return;
508 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
509 strbuf_release(&buf);
510 free(path);
513 struct alternate_object_database *alloc_alt_odb(const char *dir)
515 struct alternate_object_database *ent;
517 FLEX_ALLOC_STR(ent, path, dir);
518 strbuf_init(&ent->scratch, 0);
519 strbuf_addf(&ent->scratch, "%s/", dir);
520 ent->base_len = ent->scratch.len;
522 return ent;
525 void add_to_alternates_file(const char *reference)
527 struct lock_file lock = LOCK_INIT;
528 char *alts = git_pathdup("objects/info/alternates");
529 FILE *in, *out;
530 int found = 0;
532 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
533 out = fdopen_lock_file(&lock, "w");
534 if (!out)
535 die_errno("unable to fdopen alternates lockfile");
537 in = fopen(alts, "r");
538 if (in) {
539 struct strbuf line = STRBUF_INIT;
541 while (strbuf_getline(&line, in) != EOF) {
542 if (!strcmp(reference, line.buf)) {
543 found = 1;
544 break;
546 fprintf_or_die(out, "%s\n", line.buf);
549 strbuf_release(&line);
550 fclose(in);
552 else if (errno != ENOENT)
553 die_errno("unable to read alternates file");
555 if (found) {
556 rollback_lock_file(&lock);
557 } else {
558 fprintf_or_die(out, "%s\n", reference);
559 if (commit_lock_file(&lock))
560 die_errno("unable to move new alternates file into place");
561 if (the_repository->objects->alt_odb_tail)
562 link_alt_odb_entries(reference, '\n', NULL, 0);
564 free(alts);
567 void add_to_alternates_memory(const char *reference)
570 * Make sure alternates are initialized, or else our entry may be
571 * overwritten when they are.
573 prepare_alt_odb();
575 link_alt_odb_entries(reference, '\n', NULL, 0);
579 * Compute the exact path an alternate is at and returns it. In case of
580 * error NULL is returned and the human readable error is added to `err`
581 * `path` may be relative and should point to $GITDIR.
582 * `err` must not be null.
584 char *compute_alternate_path(const char *path, struct strbuf *err)
586 char *ref_git = NULL;
587 const char *repo, *ref_git_s;
588 int seen_error = 0;
590 ref_git_s = real_path_if_valid(path);
591 if (!ref_git_s) {
592 seen_error = 1;
593 strbuf_addf(err, _("path '%s' does not exist"), path);
594 goto out;
595 } else
597 * Beware: read_gitfile(), real_path() and mkpath()
598 * return static buffer
600 ref_git = xstrdup(ref_git_s);
602 repo = read_gitfile(ref_git);
603 if (!repo)
604 repo = read_gitfile(mkpath("%s/.git", ref_git));
605 if (repo) {
606 free(ref_git);
607 ref_git = xstrdup(repo);
610 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
611 char *ref_git_git = mkpathdup("%s/.git", ref_git);
612 free(ref_git);
613 ref_git = ref_git_git;
614 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
615 struct strbuf sb = STRBUF_INIT;
616 seen_error = 1;
617 if (get_common_dir(&sb, ref_git)) {
618 strbuf_addf(err,
619 _("reference repository '%s' as a linked "
620 "checkout is not supported yet."),
621 path);
622 goto out;
625 strbuf_addf(err, _("reference repository '%s' is not a "
626 "local repository."), path);
627 goto out;
630 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
631 strbuf_addf(err, _("reference repository '%s' is shallow"),
632 path);
633 seen_error = 1;
634 goto out;
637 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
638 strbuf_addf(err,
639 _("reference repository '%s' is grafted"),
640 path);
641 seen_error = 1;
642 goto out;
645 out:
646 if (seen_error) {
647 FREE_AND_NULL(ref_git);
650 return ref_git;
653 int foreach_alt_odb(alt_odb_fn fn, void *cb)
655 struct alternate_object_database *ent;
656 int r = 0;
658 prepare_alt_odb();
659 for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
660 r = fn(ent, cb);
661 if (r)
662 break;
664 return r;
667 void prepare_alt_odb(void)
669 if (the_repository->objects->alt_odb_tail)
670 return;
672 the_repository->objects->alt_odb_tail =
673 &the_repository->objects->alt_odb_list;
674 link_alt_odb_entries(the_repository->objects->alternate_db,
675 PATH_SEP, NULL, 0);
677 read_info_alternates(get_object_directory(), 0);
680 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
681 static int freshen_file(const char *fn)
683 struct utimbuf t;
684 t.actime = t.modtime = time(NULL);
685 return !utime(fn, &t);
689 * All of the check_and_freshen functions return 1 if the file exists and was
690 * freshened (if freshening was requested), 0 otherwise. If they return
691 * 0, you should not assume that it is safe to skip a write of the object (it
692 * either does not exist on disk, or has a stale mtime and may be subject to
693 * pruning).
695 int check_and_freshen_file(const char *fn, int freshen)
697 if (access(fn, F_OK))
698 return 0;
699 if (freshen && !freshen_file(fn))
700 return 0;
701 return 1;
704 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
706 static struct strbuf buf = STRBUF_INIT;
708 strbuf_reset(&buf);
709 sha1_file_name(&buf, sha1);
711 return check_and_freshen_file(buf.buf, freshen);
714 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
716 struct alternate_object_database *alt;
717 prepare_alt_odb();
718 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
719 const char *path = alt_sha1_path(alt, sha1);
720 if (check_and_freshen_file(path, freshen))
721 return 1;
723 return 0;
726 static int check_and_freshen(const unsigned char *sha1, int freshen)
728 return check_and_freshen_local(sha1, freshen) ||
729 check_and_freshen_nonlocal(sha1, freshen);
732 int has_loose_object_nonlocal(const unsigned char *sha1)
734 return check_and_freshen_nonlocal(sha1, 0);
737 static int has_loose_object(const unsigned char *sha1)
739 return check_and_freshen(sha1, 0);
742 static void mmap_limit_check(size_t length)
744 static size_t limit = 0;
745 if (!limit) {
746 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
747 if (!limit)
748 limit = SIZE_MAX;
750 if (length > limit)
751 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
752 (uintmax_t)length, (uintmax_t)limit);
755 void *xmmap_gently(void *start, size_t length,
756 int prot, int flags, int fd, off_t offset)
758 void *ret;
760 mmap_limit_check(length);
761 ret = mmap(start, length, prot, flags, fd, offset);
762 if (ret == MAP_FAILED) {
763 if (!length)
764 return NULL;
765 release_pack_memory(length);
766 ret = mmap(start, length, prot, flags, fd, offset);
768 return ret;
771 void *xmmap(void *start, size_t length,
772 int prot, int flags, int fd, off_t offset)
774 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
775 if (ret == MAP_FAILED)
776 die_errno("mmap failed");
777 return ret;
781 * With an in-core object data in "map", rehash it to make sure the
782 * object name actually matches "sha1" to detect object corruption.
783 * With "map" == NULL, try reading the object named with "sha1" using
784 * the streaming interface and rehash it to do the same.
786 int check_sha1_signature(const unsigned char *sha1, void *map,
787 unsigned long size, const char *type)
789 unsigned char real_sha1[20];
790 enum object_type obj_type;
791 struct git_istream *st;
792 git_SHA_CTX c;
793 char hdr[32];
794 int hdrlen;
796 if (map) {
797 hash_sha1_file(map, size, type, real_sha1);
798 return hashcmp(sha1, real_sha1) ? -1 : 0;
801 st = open_istream(sha1, &obj_type, &size, NULL);
802 if (!st)
803 return -1;
805 /* Generate the header */
806 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
808 /* Sha1.. */
809 git_SHA1_Init(&c);
810 git_SHA1_Update(&c, hdr, hdrlen);
811 for (;;) {
812 char buf[1024 * 16];
813 ssize_t readlen = read_istream(st, buf, sizeof(buf));
815 if (readlen < 0) {
816 close_istream(st);
817 return -1;
819 if (!readlen)
820 break;
821 git_SHA1_Update(&c, buf, readlen);
823 git_SHA1_Final(real_sha1, &c);
824 close_istream(st);
825 return hashcmp(sha1, real_sha1) ? -1 : 0;
828 int git_open_cloexec(const char *name, int flags)
830 int fd;
831 static int o_cloexec = O_CLOEXEC;
833 fd = open(name, flags | o_cloexec);
834 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
835 /* Try again w/o O_CLOEXEC: the kernel might not support it */
836 o_cloexec &= ~O_CLOEXEC;
837 fd = open(name, flags | o_cloexec);
840 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
842 static int fd_cloexec = FD_CLOEXEC;
844 if (!o_cloexec && 0 <= fd && fd_cloexec) {
845 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
846 int flags = fcntl(fd, F_GETFD);
847 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
848 fd_cloexec = 0;
851 #endif
852 return fd;
856 * Find "sha1" as a loose object in the local repository or in an alternate.
857 * Returns 0 on success, negative on failure.
859 * The "path" out-parameter will give the path of the object we found (if any).
860 * Note that it may point to static storage and is only valid until another
861 * call to sha1_file_name(), etc.
863 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
864 const char **path)
866 struct alternate_object_database *alt;
867 static struct strbuf buf = STRBUF_INIT;
869 strbuf_reset(&buf);
870 sha1_file_name(&buf, sha1);
871 *path = buf.buf;
873 if (!lstat(*path, st))
874 return 0;
876 prepare_alt_odb();
877 errno = ENOENT;
878 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
879 *path = alt_sha1_path(alt, sha1);
880 if (!lstat(*path, st))
881 return 0;
884 return -1;
888 * Like stat_sha1_file(), but actually open the object and return the
889 * descriptor. See the caveats on the "path" parameter above.
891 static int open_sha1_file(const unsigned char *sha1, const char **path)
893 int fd;
894 struct alternate_object_database *alt;
895 int most_interesting_errno;
896 static struct strbuf buf = STRBUF_INIT;
898 strbuf_reset(&buf);
899 sha1_file_name(&buf, sha1);
900 *path = buf.buf;
902 fd = git_open(*path);
903 if (fd >= 0)
904 return fd;
905 most_interesting_errno = errno;
907 prepare_alt_odb();
908 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
909 *path = alt_sha1_path(alt, sha1);
910 fd = git_open(*path);
911 if (fd >= 0)
912 return fd;
913 if (most_interesting_errno == ENOENT)
914 most_interesting_errno = errno;
916 errno = most_interesting_errno;
917 return -1;
921 * Map the loose object at "path" if it is not NULL, or the path found by
922 * searching for a loose object named "sha1".
924 static void *map_sha1_file_1(const char *path,
925 const unsigned char *sha1,
926 unsigned long *size)
928 void *map;
929 int fd;
931 if (path)
932 fd = git_open(path);
933 else
934 fd = open_sha1_file(sha1, &path);
935 map = NULL;
936 if (fd >= 0) {
937 struct stat st;
939 if (!fstat(fd, &st)) {
940 *size = xsize_t(st.st_size);
941 if (!*size) {
942 /* mmap() is forbidden on empty files */
943 error("object file %s is empty", path);
944 return NULL;
946 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
948 close(fd);
950 return map;
953 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
955 return map_sha1_file_1(NULL, sha1, size);
958 static int unpack_sha1_short_header(git_zstream *stream,
959 unsigned char *map, unsigned long mapsize,
960 void *buffer, unsigned long bufsiz)
962 /* Get the data stream */
963 memset(stream, 0, sizeof(*stream));
964 stream->next_in = map;
965 stream->avail_in = mapsize;
966 stream->next_out = buffer;
967 stream->avail_out = bufsiz;
969 git_inflate_init(stream);
970 return git_inflate(stream, 0);
973 int unpack_sha1_header(git_zstream *stream,
974 unsigned char *map, unsigned long mapsize,
975 void *buffer, unsigned long bufsiz)
977 int status = unpack_sha1_short_header(stream, map, mapsize,
978 buffer, bufsiz);
980 if (status < Z_OK)
981 return status;
983 /* Make sure we have the terminating NUL */
984 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
985 return -1;
986 return 0;
989 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
990 unsigned long mapsize, void *buffer,
991 unsigned long bufsiz, struct strbuf *header)
993 int status;
995 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
996 if (status < Z_OK)
997 return -1;
1000 * Check if entire header is unpacked in the first iteration.
1002 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1003 return 0;
1006 * buffer[0..bufsiz] was not large enough. Copy the partial
1007 * result out to header, and then append the result of further
1008 * reading the stream.
1010 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1011 stream->next_out = buffer;
1012 stream->avail_out = bufsiz;
1014 do {
1015 status = git_inflate(stream, 0);
1016 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1017 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1018 return 0;
1019 stream->next_out = buffer;
1020 stream->avail_out = bufsiz;
1021 } while (status != Z_STREAM_END);
1022 return -1;
1025 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1027 int bytes = strlen(buffer) + 1;
1028 unsigned char *buf = xmallocz(size);
1029 unsigned long n;
1030 int status = Z_OK;
1032 n = stream->total_out - bytes;
1033 if (n > size)
1034 n = size;
1035 memcpy(buf, (char *) buffer + bytes, n);
1036 bytes = n;
1037 if (bytes <= size) {
1039 * The above condition must be (bytes <= size), not
1040 * (bytes < size). In other words, even though we
1041 * expect no more output and set avail_out to zero,
1042 * the input zlib stream may have bytes that express
1043 * "this concludes the stream", and we *do* want to
1044 * eat that input.
1046 * Otherwise we would not be able to test that we
1047 * consumed all the input to reach the expected size;
1048 * we also want to check that zlib tells us that all
1049 * went well with status == Z_STREAM_END at the end.
1051 stream->next_out = buf + bytes;
1052 stream->avail_out = size - bytes;
1053 while (status == Z_OK)
1054 status = git_inflate(stream, Z_FINISH);
1056 if (status == Z_STREAM_END && !stream->avail_in) {
1057 git_inflate_end(stream);
1058 return buf;
1061 if (status < 0)
1062 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1063 else if (stream->avail_in)
1064 error("garbage at end of loose object '%s'",
1065 sha1_to_hex(sha1));
1066 free(buf);
1067 return NULL;
1071 * We used to just use "sscanf()", but that's actually way
1072 * too permissive for what we want to check. So do an anal
1073 * object header parse by hand.
1075 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1076 unsigned int flags)
1078 const char *type_buf = hdr;
1079 unsigned long size;
1080 int type, type_len = 0;
1083 * The type can be of any size but is followed by
1084 * a space.
1086 for (;;) {
1087 char c = *hdr++;
1088 if (!c)
1089 return -1;
1090 if (c == ' ')
1091 break;
1092 type_len++;
1095 type = type_from_string_gently(type_buf, type_len, 1);
1096 if (oi->typename)
1097 strbuf_add(oi->typename, type_buf, type_len);
1099 * Set type to 0 if its an unknown object and
1100 * we're obtaining the type using '--allow-unknown-type'
1101 * option.
1103 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1104 type = 0;
1105 else if (type < 0)
1106 die("invalid object type");
1107 if (oi->typep)
1108 *oi->typep = type;
1111 * The length must follow immediately, and be in canonical
1112 * decimal format (ie "010" is not valid).
1114 size = *hdr++ - '0';
1115 if (size > 9)
1116 return -1;
1117 if (size) {
1118 for (;;) {
1119 unsigned long c = *hdr - '0';
1120 if (c > 9)
1121 break;
1122 hdr++;
1123 size = size * 10 + c;
1127 if (oi->sizep)
1128 *oi->sizep = size;
1131 * The length must be followed by a zero byte
1133 return *hdr ? -1 : type;
1136 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1138 struct object_info oi = OBJECT_INFO_INIT;
1140 oi.sizep = sizep;
1141 return parse_sha1_header_extended(hdr, &oi, 0);
1144 static int sha1_loose_object_info(const unsigned char *sha1,
1145 struct object_info *oi,
1146 int flags)
1148 int status = 0;
1149 unsigned long mapsize;
1150 void *map;
1151 git_zstream stream;
1152 char hdr[32];
1153 struct strbuf hdrbuf = STRBUF_INIT;
1154 unsigned long size_scratch;
1156 if (oi->delta_base_sha1)
1157 hashclr(oi->delta_base_sha1);
1160 * If we don't care about type or size, then we don't
1161 * need to look inside the object at all. Note that we
1162 * do not optimize out the stat call, even if the
1163 * caller doesn't care about the disk-size, since our
1164 * return value implicitly indicates whether the
1165 * object even exists.
1167 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1168 const char *path;
1169 struct stat st;
1170 if (stat_sha1_file(sha1, &st, &path) < 0)
1171 return -1;
1172 if (oi->disk_sizep)
1173 *oi->disk_sizep = st.st_size;
1174 return 0;
1177 map = map_sha1_file(sha1, &mapsize);
1178 if (!map)
1179 return -1;
1181 if (!oi->sizep)
1182 oi->sizep = &size_scratch;
1184 if (oi->disk_sizep)
1185 *oi->disk_sizep = mapsize;
1186 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1187 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1188 status = error("unable to unpack %s header with --allow-unknown-type",
1189 sha1_to_hex(sha1));
1190 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1191 status = error("unable to unpack %s header",
1192 sha1_to_hex(sha1));
1193 if (status < 0)
1194 ; /* Do nothing */
1195 else if (hdrbuf.len) {
1196 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1197 status = error("unable to parse %s header with --allow-unknown-type",
1198 sha1_to_hex(sha1));
1199 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1200 status = error("unable to parse %s header", sha1_to_hex(sha1));
1202 if (status >= 0 && oi->contentp) {
1203 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1204 *oi->sizep, sha1);
1205 if (!*oi->contentp) {
1206 git_inflate_end(&stream);
1207 status = -1;
1209 } else
1210 git_inflate_end(&stream);
1212 munmap(map, mapsize);
1213 if (status && oi->typep)
1214 *oi->typep = status;
1215 if (oi->sizep == &size_scratch)
1216 oi->sizep = NULL;
1217 strbuf_release(&hdrbuf);
1218 oi->whence = OI_LOOSE;
1219 return (status < 0) ? status : 0;
1222 int fetch_if_missing = 1;
1224 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1226 static struct object_info blank_oi = OBJECT_INFO_INIT;
1227 struct pack_entry e;
1228 int rtype;
1229 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1230 lookup_replace_object(sha1) :
1231 sha1;
1232 int already_retried = 0;
1234 if (is_null_sha1(real))
1235 return -1;
1237 if (!oi)
1238 oi = &blank_oi;
1240 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1241 struct cached_object *co = find_cached_object(real);
1242 if (co) {
1243 if (oi->typep)
1244 *(oi->typep) = co->type;
1245 if (oi->sizep)
1246 *(oi->sizep) = co->size;
1247 if (oi->disk_sizep)
1248 *(oi->disk_sizep) = 0;
1249 if (oi->delta_base_sha1)
1250 hashclr(oi->delta_base_sha1);
1251 if (oi->typename)
1252 strbuf_addstr(oi->typename, typename(co->type));
1253 if (oi->contentp)
1254 *oi->contentp = xmemdupz(co->buf, co->size);
1255 oi->whence = OI_CACHED;
1256 return 0;
1260 while (1) {
1261 if (find_pack_entry(real, &e))
1262 break;
1264 /* Most likely it's a loose object. */
1265 if (!sha1_loose_object_info(real, oi, flags))
1266 return 0;
1268 /* Not a loose object; someone else may have just packed it. */
1269 reprepare_packed_git();
1270 if (find_pack_entry(real, &e))
1271 break;
1273 /* Check if it is a missing object */
1274 if (fetch_if_missing && repository_format_partial_clone &&
1275 !already_retried) {
1277 * TODO Investigate haveing fetch_object() return
1278 * TODO error/success and stopping the music here.
1280 fetch_object(repository_format_partial_clone, real);
1281 already_retried = 1;
1282 continue;
1285 return -1;
1288 if (oi == &blank_oi)
1290 * We know that the caller doesn't actually need the
1291 * information below, so return early.
1293 return 0;
1294 rtype = packed_object_info(e.p, e.offset, oi);
1295 if (rtype < 0) {
1296 mark_bad_packed_object(e.p, real);
1297 return sha1_object_info_extended(real, oi, 0);
1298 } else if (oi->whence == OI_PACKED) {
1299 oi->u.packed.offset = e.offset;
1300 oi->u.packed.pack = e.p;
1301 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1302 rtype == OBJ_OFS_DELTA);
1305 return 0;
1308 /* returns enum object_type or negative */
1309 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1311 enum object_type type;
1312 struct object_info oi = OBJECT_INFO_INIT;
1314 oi.typep = &type;
1315 oi.sizep = sizep;
1316 if (sha1_object_info_extended(sha1, &oi,
1317 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1318 return -1;
1319 return type;
1322 static void *read_object(const unsigned char *sha1, enum object_type *type,
1323 unsigned long *size)
1325 struct object_info oi = OBJECT_INFO_INIT;
1326 void *content;
1327 oi.typep = type;
1328 oi.sizep = size;
1329 oi.contentp = &content;
1331 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1332 return NULL;
1333 return content;
1336 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1337 unsigned char *sha1)
1339 struct cached_object *co;
1341 hash_sha1_file(buf, len, typename(type), sha1);
1342 if (has_sha1_file(sha1) || find_cached_object(sha1))
1343 return 0;
1344 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1345 co = &cached_objects[cached_object_nr++];
1346 co->size = len;
1347 co->type = type;
1348 co->buf = xmalloc(len);
1349 memcpy(co->buf, buf, len);
1350 hashcpy(co->sha1, sha1);
1351 return 0;
1355 * This function dies on corrupt objects; the callers who want to
1356 * deal with them should arrange to call read_object() and give error
1357 * messages themselves.
1359 void *read_sha1_file_extended(const unsigned char *sha1,
1360 enum object_type *type,
1361 unsigned long *size,
1362 int lookup_replace)
1364 void *data;
1365 const struct packed_git *p;
1366 const char *path;
1367 struct stat st;
1368 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1369 : sha1;
1371 errno = 0;
1372 data = read_object(repl, type, size);
1373 if (data)
1374 return data;
1376 if (errno && errno != ENOENT)
1377 die_errno("failed to read object %s", sha1_to_hex(sha1));
1379 /* die if we replaced an object with one that does not exist */
1380 if (repl != sha1)
1381 die("replacement %s not found for %s",
1382 sha1_to_hex(repl), sha1_to_hex(sha1));
1384 if (!stat_sha1_file(repl, &st, &path))
1385 die("loose object %s (stored in %s) is corrupt",
1386 sha1_to_hex(repl), path);
1388 if ((p = has_packed_and_bad(repl)) != NULL)
1389 die("packed object %s (stored in %s) is corrupt",
1390 sha1_to_hex(repl), p->pack_name);
1392 return NULL;
1395 void *read_object_with_reference(const unsigned char *sha1,
1396 const char *required_type_name,
1397 unsigned long *size,
1398 unsigned char *actual_sha1_return)
1400 enum object_type type, required_type;
1401 void *buffer;
1402 unsigned long isize;
1403 unsigned char actual_sha1[20];
1405 required_type = type_from_string(required_type_name);
1406 hashcpy(actual_sha1, sha1);
1407 while (1) {
1408 int ref_length = -1;
1409 const char *ref_type = NULL;
1411 buffer = read_sha1_file(actual_sha1, &type, &isize);
1412 if (!buffer)
1413 return NULL;
1414 if (type == required_type) {
1415 *size = isize;
1416 if (actual_sha1_return)
1417 hashcpy(actual_sha1_return, actual_sha1);
1418 return buffer;
1420 /* Handle references */
1421 else if (type == OBJ_COMMIT)
1422 ref_type = "tree ";
1423 else if (type == OBJ_TAG)
1424 ref_type = "object ";
1425 else {
1426 free(buffer);
1427 return NULL;
1429 ref_length = strlen(ref_type);
1431 if (ref_length + 40 > isize ||
1432 memcmp(buffer, ref_type, ref_length) ||
1433 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1434 free(buffer);
1435 return NULL;
1437 free(buffer);
1438 /* Now we have the ID of the referred-to object in
1439 * actual_sha1. Check again. */
1443 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1444 const char *type, unsigned char *sha1,
1445 char *hdr, int *hdrlen)
1447 git_SHA_CTX c;
1449 /* Generate the header */
1450 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1452 /* Sha1.. */
1453 git_SHA1_Init(&c);
1454 git_SHA1_Update(&c, hdr, *hdrlen);
1455 git_SHA1_Update(&c, buf, len);
1456 git_SHA1_Final(sha1, &c);
1460 * Move the just written object into its final resting place.
1462 int finalize_object_file(const char *tmpfile, const char *filename)
1464 int ret = 0;
1466 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1467 goto try_rename;
1468 else if (link(tmpfile, filename))
1469 ret = errno;
1472 * Coda hack - coda doesn't like cross-directory links,
1473 * so we fall back to a rename, which will mean that it
1474 * won't be able to check collisions, but that's not a
1475 * big deal.
1477 * The same holds for FAT formatted media.
1479 * When this succeeds, we just return. We have nothing
1480 * left to unlink.
1482 if (ret && ret != EEXIST) {
1483 try_rename:
1484 if (!rename(tmpfile, filename))
1485 goto out;
1486 ret = errno;
1488 unlink_or_warn(tmpfile);
1489 if (ret) {
1490 if (ret != EEXIST) {
1491 return error_errno("unable to write sha1 filename %s", filename);
1493 /* FIXME!!! Collision check here ? */
1496 out:
1497 if (adjust_shared_perm(filename))
1498 return error("unable to set permission to '%s'", filename);
1499 return 0;
1502 static int write_buffer(int fd, const void *buf, size_t len)
1504 if (write_in_full(fd, buf, len) < 0)
1505 return error_errno("file write error");
1506 return 0;
1509 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1510 unsigned char *sha1)
1512 char hdr[32];
1513 int hdrlen = sizeof(hdr);
1514 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1515 return 0;
1518 /* Finalize a file on disk, and close it. */
1519 static void close_sha1_file(int fd)
1521 if (fsync_object_files)
1522 fsync_or_die(fd, "sha1 file");
1523 if (close(fd) != 0)
1524 die_errno("error when closing sha1 file");
1527 /* Size of directory component, including the ending '/' */
1528 static inline int directory_size(const char *filename)
1530 const char *s = strrchr(filename, '/');
1531 if (!s)
1532 return 0;
1533 return s - filename + 1;
1537 * This creates a temporary file in the same directory as the final
1538 * 'filename'
1540 * We want to avoid cross-directory filename renames, because those
1541 * can have problems on various filesystems (FAT, NFS, Coda).
1543 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1545 int fd, dirlen = directory_size(filename);
1547 strbuf_reset(tmp);
1548 strbuf_add(tmp, filename, dirlen);
1549 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1550 fd = git_mkstemp_mode(tmp->buf, 0444);
1551 if (fd < 0 && dirlen && errno == ENOENT) {
1553 * Make sure the directory exists; note that the contents
1554 * of the buffer are undefined after mkstemp returns an
1555 * error, so we have to rewrite the whole buffer from
1556 * scratch.
1558 strbuf_reset(tmp);
1559 strbuf_add(tmp, filename, dirlen - 1);
1560 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1561 return -1;
1562 if (adjust_shared_perm(tmp->buf))
1563 return -1;
1565 /* Try again */
1566 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1567 fd = git_mkstemp_mode(tmp->buf, 0444);
1569 return fd;
1572 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1573 const void *buf, unsigned long len, time_t mtime)
1575 int fd, ret;
1576 unsigned char compressed[4096];
1577 git_zstream stream;
1578 git_SHA_CTX c;
1579 unsigned char parano_sha1[20];
1580 static struct strbuf tmp_file = STRBUF_INIT;
1581 static struct strbuf filename = STRBUF_INIT;
1583 strbuf_reset(&filename);
1584 sha1_file_name(&filename, sha1);
1586 fd = create_tmpfile(&tmp_file, filename.buf);
1587 if (fd < 0) {
1588 if (errno == EACCES)
1589 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1590 else
1591 return error_errno("unable to create temporary file");
1594 /* Set it up */
1595 git_deflate_init(&stream, zlib_compression_level);
1596 stream.next_out = compressed;
1597 stream.avail_out = sizeof(compressed);
1598 git_SHA1_Init(&c);
1600 /* First header.. */
1601 stream.next_in = (unsigned char *)hdr;
1602 stream.avail_in = hdrlen;
1603 while (git_deflate(&stream, 0) == Z_OK)
1604 ; /* nothing */
1605 git_SHA1_Update(&c, hdr, hdrlen);
1607 /* Then the data itself.. */
1608 stream.next_in = (void *)buf;
1609 stream.avail_in = len;
1610 do {
1611 unsigned char *in0 = stream.next_in;
1612 ret = git_deflate(&stream, Z_FINISH);
1613 git_SHA1_Update(&c, in0, stream.next_in - in0);
1614 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1615 die("unable to write sha1 file");
1616 stream.next_out = compressed;
1617 stream.avail_out = sizeof(compressed);
1618 } while (ret == Z_OK);
1620 if (ret != Z_STREAM_END)
1621 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1622 ret = git_deflate_end_gently(&stream);
1623 if (ret != Z_OK)
1624 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1625 git_SHA1_Final(parano_sha1, &c);
1626 if (hashcmp(sha1, parano_sha1) != 0)
1627 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1629 close_sha1_file(fd);
1631 if (mtime) {
1632 struct utimbuf utb;
1633 utb.actime = mtime;
1634 utb.modtime = mtime;
1635 if (utime(tmp_file.buf, &utb) < 0)
1636 warning_errno("failed utime() on %s", tmp_file.buf);
1639 return finalize_object_file(tmp_file.buf, filename.buf);
1642 static int freshen_loose_object(const unsigned char *sha1)
1644 return check_and_freshen(sha1, 1);
1647 static int freshen_packed_object(const unsigned char *sha1)
1649 struct pack_entry e;
1650 if (!find_pack_entry(sha1, &e))
1651 return 0;
1652 if (e.p->freshened)
1653 return 1;
1654 if (!freshen_file(e.p->pack_name))
1655 return 0;
1656 e.p->freshened = 1;
1657 return 1;
1660 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1662 char hdr[32];
1663 int hdrlen = sizeof(hdr);
1665 /* Normally if we have it in the pack then we do not bother writing
1666 * it out into .git/objects/??/?{38} file.
1668 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1669 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1670 return 0;
1671 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1674 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1675 struct object_id *oid, unsigned flags)
1677 char *header;
1678 int hdrlen, status = 0;
1680 /* type string, SP, %lu of the length plus NUL must fit this */
1681 hdrlen = strlen(type) + 32;
1682 header = xmalloc(hdrlen);
1683 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1685 if (!(flags & HASH_WRITE_OBJECT))
1686 goto cleanup;
1687 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1688 goto cleanup;
1689 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1691 cleanup:
1692 free(header);
1693 return status;
1696 int force_object_loose(const unsigned char *sha1, time_t mtime)
1698 void *buf;
1699 unsigned long len;
1700 enum object_type type;
1701 char hdr[32];
1702 int hdrlen;
1703 int ret;
1705 if (has_loose_object(sha1))
1706 return 0;
1707 buf = read_object(sha1, &type, &len);
1708 if (!buf)
1709 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1710 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1711 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1712 free(buf);
1714 return ret;
1717 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1719 if (!startup_info->have_repository)
1720 return 0;
1721 return sha1_object_info_extended(sha1, NULL,
1722 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1725 int has_object_file(const struct object_id *oid)
1727 return has_sha1_file(oid->hash);
1730 int has_object_file_with_flags(const struct object_id *oid, int flags)
1732 return has_sha1_file_with_flags(oid->hash, flags);
1735 static void check_tree(const void *buf, size_t size)
1737 struct tree_desc desc;
1738 struct name_entry entry;
1740 init_tree_desc(&desc, buf, size);
1741 while (tree_entry(&desc, &entry))
1742 /* do nothing
1743 * tree_entry() will die() on malformed entries */
1747 static void check_commit(const void *buf, size_t size)
1749 struct commit c;
1750 memset(&c, 0, sizeof(c));
1751 if (parse_commit_buffer(&c, buf, size))
1752 die("corrupt commit");
1755 static void check_tag(const void *buf, size_t size)
1757 struct tag t;
1758 memset(&t, 0, sizeof(t));
1759 if (parse_tag_buffer(&t, buf, size))
1760 die("corrupt tag");
1763 static int index_mem(struct object_id *oid, void *buf, size_t size,
1764 enum object_type type,
1765 const char *path, unsigned flags)
1767 int ret, re_allocated = 0;
1768 int write_object = flags & HASH_WRITE_OBJECT;
1770 if (!type)
1771 type = OBJ_BLOB;
1774 * Convert blobs to git internal format
1776 if ((type == OBJ_BLOB) && path) {
1777 struct strbuf nbuf = STRBUF_INIT;
1778 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1779 get_conv_flags(flags))) {
1780 buf = strbuf_detach(&nbuf, &size);
1781 re_allocated = 1;
1784 if (flags & HASH_FORMAT_CHECK) {
1785 if (type == OBJ_TREE)
1786 check_tree(buf, size);
1787 if (type == OBJ_COMMIT)
1788 check_commit(buf, size);
1789 if (type == OBJ_TAG)
1790 check_tag(buf, size);
1793 if (write_object)
1794 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1795 else
1796 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1797 if (re_allocated)
1798 free(buf);
1799 return ret;
1802 static int index_stream_convert_blob(struct object_id *oid, int fd,
1803 const char *path, unsigned flags)
1805 int ret;
1806 const int write_object = flags & HASH_WRITE_OBJECT;
1807 struct strbuf sbuf = STRBUF_INIT;
1809 assert(path);
1810 assert(would_convert_to_git_filter_fd(path));
1812 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1813 get_conv_flags(flags));
1815 if (write_object)
1816 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1817 oid->hash);
1818 else
1819 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1820 oid->hash);
1821 strbuf_release(&sbuf);
1822 return ret;
1825 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1826 const char *path, unsigned flags)
1828 struct strbuf sbuf = STRBUF_INIT;
1829 int ret;
1831 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1832 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1833 else
1834 ret = -1;
1835 strbuf_release(&sbuf);
1836 return ret;
1839 #define SMALL_FILE_SIZE (32*1024)
1841 static int index_core(struct object_id *oid, int fd, size_t size,
1842 enum object_type type, const char *path,
1843 unsigned flags)
1845 int ret;
1847 if (!size) {
1848 ret = index_mem(oid, "", size, type, path, flags);
1849 } else if (size <= SMALL_FILE_SIZE) {
1850 char *buf = xmalloc(size);
1851 ssize_t read_result = read_in_full(fd, buf, size);
1852 if (read_result < 0)
1853 ret = error_errno("read error while indexing %s",
1854 path ? path : "<unknown>");
1855 else if (read_result != size)
1856 ret = error("short read while indexing %s",
1857 path ? path : "<unknown>");
1858 else
1859 ret = index_mem(oid, buf, size, type, path, flags);
1860 free(buf);
1861 } else {
1862 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1863 ret = index_mem(oid, buf, size, type, path, flags);
1864 munmap(buf, size);
1866 return ret;
1870 * This creates one packfile per large blob unless bulk-checkin
1871 * machinery is "plugged".
1873 * This also bypasses the usual "convert-to-git" dance, and that is on
1874 * purpose. We could write a streaming version of the converting
1875 * functions and insert that before feeding the data to fast-import
1876 * (or equivalent in-core API described above). However, that is
1877 * somewhat complicated, as we do not know the size of the filter
1878 * result, which we need to know beforehand when writing a git object.
1879 * Since the primary motivation for trying to stream from the working
1880 * tree file and to avoid mmaping it in core is to deal with large
1881 * binary blobs, they generally do not want to get any conversion, and
1882 * callers should avoid this code path when filters are requested.
1884 static int index_stream(struct object_id *oid, int fd, size_t size,
1885 enum object_type type, const char *path,
1886 unsigned flags)
1888 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1891 int index_fd(struct object_id *oid, int fd, struct stat *st,
1892 enum object_type type, const char *path, unsigned flags)
1894 int ret;
1897 * Call xsize_t() only when needed to avoid potentially unnecessary
1898 * die() for large files.
1900 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1901 ret = index_stream_convert_blob(oid, fd, path, flags);
1902 else if (!S_ISREG(st->st_mode))
1903 ret = index_pipe(oid, fd, type, path, flags);
1904 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1905 (path && would_convert_to_git(&the_index, path)))
1906 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1907 flags);
1908 else
1909 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1910 flags);
1911 close(fd);
1912 return ret;
1915 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1917 int fd;
1918 struct strbuf sb = STRBUF_INIT;
1919 int rc = 0;
1921 switch (st->st_mode & S_IFMT) {
1922 case S_IFREG:
1923 fd = open(path, O_RDONLY);
1924 if (fd < 0)
1925 return error_errno("open(\"%s\")", path);
1926 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1927 return error("%s: failed to insert into database",
1928 path);
1929 break;
1930 case S_IFLNK:
1931 if (strbuf_readlink(&sb, path, st->st_size))
1932 return error_errno("readlink(\"%s\")", path);
1933 if (!(flags & HASH_WRITE_OBJECT))
1934 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1935 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1936 rc = error("%s: failed to insert into database", path);
1937 strbuf_release(&sb);
1938 break;
1939 case S_IFDIR:
1940 return resolve_gitlink_ref(path, "HEAD", oid);
1941 default:
1942 return error("%s: unsupported file type", path);
1944 return rc;
1947 int read_pack_header(int fd, struct pack_header *header)
1949 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1950 /* "eof before pack header was fully read" */
1951 return PH_ERROR_EOF;
1953 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1954 /* "protocol error (pack signature mismatch detected)" */
1955 return PH_ERROR_PACK_SIGNATURE;
1956 if (!pack_version_ok(header->hdr_version))
1957 /* "protocol error (pack version unsupported)" */
1958 return PH_ERROR_PROTOCOL;
1959 return 0;
1962 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1964 enum object_type type = sha1_object_info(sha1, NULL);
1965 if (type < 0)
1966 die("%s is not a valid object", sha1_to_hex(sha1));
1967 if (type != expect)
1968 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1969 typename(expect));
1972 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1973 struct strbuf *path,
1974 each_loose_object_fn obj_cb,
1975 each_loose_cruft_fn cruft_cb,
1976 each_loose_subdir_fn subdir_cb,
1977 void *data)
1979 size_t origlen, baselen;
1980 DIR *dir;
1981 struct dirent *de;
1982 int r = 0;
1983 struct object_id oid;
1985 if (subdir_nr > 0xff)
1986 BUG("invalid loose object subdirectory: %x", subdir_nr);
1988 origlen = path->len;
1989 strbuf_complete(path, '/');
1990 strbuf_addf(path, "%02x", subdir_nr);
1992 dir = opendir(path->buf);
1993 if (!dir) {
1994 if (errno != ENOENT)
1995 r = error_errno("unable to open %s", path->buf);
1996 strbuf_setlen(path, origlen);
1997 return r;
2000 oid.hash[0] = subdir_nr;
2001 strbuf_addch(path, '/');
2002 baselen = path->len;
2004 while ((de = readdir(dir))) {
2005 size_t namelen;
2006 if (is_dot_or_dotdot(de->d_name))
2007 continue;
2009 namelen = strlen(de->d_name);
2010 strbuf_setlen(path, baselen);
2011 strbuf_add(path, de->d_name, namelen);
2012 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2013 !hex_to_bytes(oid.hash + 1, de->d_name,
2014 GIT_SHA1_RAWSZ - 1)) {
2015 if (obj_cb) {
2016 r = obj_cb(&oid, path->buf, data);
2017 if (r)
2018 break;
2020 continue;
2023 if (cruft_cb) {
2024 r = cruft_cb(de->d_name, path->buf, data);
2025 if (r)
2026 break;
2029 closedir(dir);
2031 strbuf_setlen(path, baselen - 1);
2032 if (!r && subdir_cb)
2033 r = subdir_cb(subdir_nr, path->buf, data);
2035 strbuf_setlen(path, origlen);
2037 return r;
2040 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2041 each_loose_object_fn obj_cb,
2042 each_loose_cruft_fn cruft_cb,
2043 each_loose_subdir_fn subdir_cb,
2044 void *data)
2046 int r = 0;
2047 int i;
2049 for (i = 0; i < 256; i++) {
2050 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2051 subdir_cb, data);
2052 if (r)
2053 break;
2056 return r;
2059 int for_each_loose_file_in_objdir(const char *path,
2060 each_loose_object_fn obj_cb,
2061 each_loose_cruft_fn cruft_cb,
2062 each_loose_subdir_fn subdir_cb,
2063 void *data)
2065 struct strbuf buf = STRBUF_INIT;
2066 int r;
2068 strbuf_addstr(&buf, path);
2069 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2070 subdir_cb, data);
2071 strbuf_release(&buf);
2073 return r;
2076 struct loose_alt_odb_data {
2077 each_loose_object_fn *cb;
2078 void *data;
2081 static int loose_from_alt_odb(struct alternate_object_database *alt,
2082 void *vdata)
2084 struct loose_alt_odb_data *data = vdata;
2085 struct strbuf buf = STRBUF_INIT;
2086 int r;
2088 strbuf_addstr(&buf, alt->path);
2089 r = for_each_loose_file_in_objdir_buf(&buf,
2090 data->cb, NULL, NULL,
2091 data->data);
2092 strbuf_release(&buf);
2093 return r;
2096 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2098 struct loose_alt_odb_data alt;
2099 int r;
2101 r = for_each_loose_file_in_objdir(get_object_directory(),
2102 cb, NULL, NULL, data);
2103 if (r)
2104 return r;
2106 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2107 return 0;
2109 alt.cb = cb;
2110 alt.data = data;
2111 return foreach_alt_odb(loose_from_alt_odb, &alt);
2114 static int check_stream_sha1(git_zstream *stream,
2115 const char *hdr,
2116 unsigned long size,
2117 const char *path,
2118 const unsigned char *expected_sha1)
2120 git_SHA_CTX c;
2121 unsigned char real_sha1[GIT_MAX_RAWSZ];
2122 unsigned char buf[4096];
2123 unsigned long total_read;
2124 int status = Z_OK;
2126 git_SHA1_Init(&c);
2127 git_SHA1_Update(&c, hdr, stream->total_out);
2130 * We already read some bytes into hdr, but the ones up to the NUL
2131 * do not count against the object's content size.
2133 total_read = stream->total_out - strlen(hdr) - 1;
2136 * This size comparison must be "<=" to read the final zlib packets;
2137 * see the comment in unpack_sha1_rest for details.
2139 while (total_read <= size &&
2140 (status == Z_OK || status == Z_BUF_ERROR)) {
2141 stream->next_out = buf;
2142 stream->avail_out = sizeof(buf);
2143 if (size - total_read < stream->avail_out)
2144 stream->avail_out = size - total_read;
2145 status = git_inflate(stream, Z_FINISH);
2146 git_SHA1_Update(&c, buf, stream->next_out - buf);
2147 total_read += stream->next_out - buf;
2149 git_inflate_end(stream);
2151 if (status != Z_STREAM_END) {
2152 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2153 return -1;
2155 if (stream->avail_in) {
2156 error("garbage at end of loose object '%s'",
2157 sha1_to_hex(expected_sha1));
2158 return -1;
2161 git_SHA1_Final(real_sha1, &c);
2162 if (hashcmp(expected_sha1, real_sha1)) {
2163 error("sha1 mismatch for %s (expected %s)", path,
2164 sha1_to_hex(expected_sha1));
2165 return -1;
2168 return 0;
2171 int read_loose_object(const char *path,
2172 const unsigned char *expected_sha1,
2173 enum object_type *type,
2174 unsigned long *size,
2175 void **contents)
2177 int ret = -1;
2178 void *map = NULL;
2179 unsigned long mapsize;
2180 git_zstream stream;
2181 char hdr[32];
2183 *contents = NULL;
2185 map = map_sha1_file_1(path, NULL, &mapsize);
2186 if (!map) {
2187 error_errno("unable to mmap %s", path);
2188 goto out;
2191 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2192 error("unable to unpack header of %s", path);
2193 goto out;
2196 *type = parse_sha1_header(hdr, size);
2197 if (*type < 0) {
2198 error("unable to parse header of %s", path);
2199 git_inflate_end(&stream);
2200 goto out;
2203 if (*type == OBJ_BLOB) {
2204 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2205 goto out;
2206 } else {
2207 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2208 if (!*contents) {
2209 error("unable to unpack contents of %s", path);
2210 git_inflate_end(&stream);
2211 goto out;
2213 if (check_sha1_signature(expected_sha1, *contents,
2214 *size, typename(*type))) {
2215 error("sha1 mismatch for %s (expected %s)", path,
2216 sha1_to_hex(expected_sha1));
2217 free(*contents);
2218 goto out;
2222 ret = 0; /* everything checks out */
2224 out:
2225 if (map)
2226 munmap(map, mapsize);
2227 return ret;