Merge branch 'ma/shortlog-revparse'
[git.git] / sha1_file.c
blobad775495659ee09cf287965f2f7b3044bb1b04e2
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "config.h"
11 #include "string-list.h"
12 #include "lockfile.h"
13 #include "delta.h"
14 #include "pack.h"
15 #include "blob.h"
16 #include "commit.h"
17 #include "run-command.h"
18 #include "tag.h"
19 #include "tree.h"
20 #include "tree-walk.h"
21 #include "refs.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "streaming.h"
26 #include "dir.h"
27 #include "list.h"
28 #include "mergesort.h"
29 #include "quote.h"
30 #include "packfile.h"
31 #include "fetch-object.h"
33 const unsigned char null_sha1[GIT_MAX_RAWSZ];
34 const struct object_id null_oid;
35 const struct object_id empty_tree_oid = {
36 EMPTY_TREE_SHA1_BIN_LITERAL
38 const struct object_id empty_blob_oid = {
39 EMPTY_BLOB_SHA1_BIN_LITERAL
42 static void git_hash_sha1_init(git_hash_ctx *ctx)
44 git_SHA1_Init(&ctx->sha1);
47 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
49 git_SHA1_Update(&ctx->sha1, data, len);
52 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
54 git_SHA1_Final(hash, &ctx->sha1);
57 static void git_hash_unknown_init(git_hash_ctx *ctx)
59 die("trying to init unknown hash");
62 static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
64 die("trying to update unknown hash");
67 static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
69 die("trying to finalize unknown hash");
72 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
74 NULL,
75 0x00000000,
78 git_hash_unknown_init,
79 git_hash_unknown_update,
80 git_hash_unknown_final,
81 NULL,
82 NULL,
85 "sha-1",
86 /* "sha1", big-endian */
87 0x73686131,
88 GIT_SHA1_RAWSZ,
89 GIT_SHA1_HEXSZ,
90 git_hash_sha1_init,
91 git_hash_sha1_update,
92 git_hash_sha1_final,
93 &empty_tree_oid,
94 &empty_blob_oid,
99 * This is meant to hold a *small* number of objects that you would
100 * want read_sha1_file() to be able to return, but yet you do not want
101 * to write them into the object store (e.g. a browse-only
102 * application).
104 static struct cached_object {
105 unsigned char sha1[20];
106 enum object_type type;
107 void *buf;
108 unsigned long size;
109 } *cached_objects;
110 static int cached_object_nr, cached_object_alloc;
112 static struct cached_object empty_tree = {
113 EMPTY_TREE_SHA1_BIN_LITERAL,
114 OBJ_TREE,
119 static struct cached_object *find_cached_object(const unsigned char *sha1)
121 int i;
122 struct cached_object *co = cached_objects;
124 for (i = 0; i < cached_object_nr; i++, co++) {
125 if (!hashcmp(co->sha1, sha1))
126 return co;
128 if (!hashcmp(sha1, empty_tree.sha1))
129 return &empty_tree;
130 return NULL;
134 static int get_conv_flags(unsigned flags)
136 if (flags & HASH_RENORMALIZE)
137 return CONV_EOL_RENORMALIZE;
138 else if (flags & HASH_WRITE_OBJECT)
139 return global_conv_flags_eol;
140 else
141 return 0;
145 int mkdir_in_gitdir(const char *path)
147 if (mkdir(path, 0777)) {
148 int saved_errno = errno;
149 struct stat st;
150 struct strbuf sb = STRBUF_INIT;
152 if (errno != EEXIST)
153 return -1;
155 * Are we looking at a path in a symlinked worktree
156 * whose original repository does not yet have it?
157 * e.g. .git/rr-cache pointing at its original
158 * repository in which the user hasn't performed any
159 * conflict resolution yet?
161 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
162 strbuf_readlink(&sb, path, st.st_size) ||
163 !is_absolute_path(sb.buf) ||
164 mkdir(sb.buf, 0777)) {
165 strbuf_release(&sb);
166 errno = saved_errno;
167 return -1;
169 strbuf_release(&sb);
171 return adjust_shared_perm(path);
174 enum scld_error safe_create_leading_directories(char *path)
176 char *next_component = path + offset_1st_component(path);
177 enum scld_error ret = SCLD_OK;
179 while (ret == SCLD_OK && next_component) {
180 struct stat st;
181 char *slash = next_component, slash_character;
183 while (*slash && !is_dir_sep(*slash))
184 slash++;
186 if (!*slash)
187 break;
189 next_component = slash + 1;
190 while (is_dir_sep(*next_component))
191 next_component++;
192 if (!*next_component)
193 break;
195 slash_character = *slash;
196 *slash = '\0';
197 if (!stat(path, &st)) {
198 /* path exists */
199 if (!S_ISDIR(st.st_mode)) {
200 errno = ENOTDIR;
201 ret = SCLD_EXISTS;
203 } else if (mkdir(path, 0777)) {
204 if (errno == EEXIST &&
205 !stat(path, &st) && S_ISDIR(st.st_mode))
206 ; /* somebody created it since we checked */
207 else if (errno == ENOENT)
209 * Either mkdir() failed because
210 * somebody just pruned the containing
211 * directory, or stat() failed because
212 * the file that was in our way was
213 * just removed. Either way, inform
214 * the caller that it might be worth
215 * trying again:
217 ret = SCLD_VANISHED;
218 else
219 ret = SCLD_FAILED;
220 } else if (adjust_shared_perm(path)) {
221 ret = SCLD_PERMS;
223 *slash = slash_character;
225 return ret;
228 enum scld_error safe_create_leading_directories_const(const char *path)
230 int save_errno;
231 /* path points to cache entries, so xstrdup before messing with it */
232 char *buf = xstrdup(path);
233 enum scld_error result = safe_create_leading_directories(buf);
235 save_errno = errno;
236 free(buf);
237 errno = save_errno;
238 return result;
241 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
244 * The number of times we will try to remove empty directories
245 * in the way of path. This is only 1 because if another
246 * process is racily creating directories that conflict with
247 * us, we don't want to fight against them.
249 int remove_directories_remaining = 1;
252 * The number of times that we will try to create the
253 * directories containing path. We are willing to attempt this
254 * more than once, because another process could be trying to
255 * clean up empty directories at the same time as we are
256 * trying to create them.
258 int create_directories_remaining = 3;
260 /* A scratch copy of path, filled lazily if we need it: */
261 struct strbuf path_copy = STRBUF_INIT;
263 int ret, save_errno;
265 /* Sanity check: */
266 assert(*path);
268 retry_fn:
269 ret = fn(path, cb);
270 save_errno = errno;
271 if (!ret)
272 goto out;
274 if (errno == EISDIR && remove_directories_remaining-- > 0) {
276 * A directory is in the way. Maybe it is empty; try
277 * to remove it:
279 if (!path_copy.len)
280 strbuf_addstr(&path_copy, path);
282 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
283 goto retry_fn;
284 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
286 * Maybe the containing directory didn't exist, or
287 * maybe it was just deleted by a process that is
288 * racing with us to clean up empty directories. Try
289 * to create it:
291 enum scld_error scld_result;
293 if (!path_copy.len)
294 strbuf_addstr(&path_copy, path);
296 do {
297 scld_result = safe_create_leading_directories(path_copy.buf);
298 if (scld_result == SCLD_OK)
299 goto retry_fn;
300 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
303 out:
304 strbuf_release(&path_copy);
305 errno = save_errno;
306 return ret;
309 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
311 int i;
312 for (i = 0; i < 20; i++) {
313 static char hex[] = "0123456789abcdef";
314 unsigned int val = sha1[i];
315 strbuf_addch(buf, hex[val >> 4]);
316 strbuf_addch(buf, hex[val & 0xf]);
317 if (!i)
318 strbuf_addch(buf, '/');
322 void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
324 strbuf_addstr(buf, get_object_directory());
325 strbuf_addch(buf, '/');
326 fill_sha1_path(buf, sha1);
329 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
331 strbuf_setlen(&alt->scratch, alt->base_len);
332 return &alt->scratch;
335 static const char *alt_sha1_path(struct alternate_object_database *alt,
336 const unsigned char *sha1)
338 struct strbuf *buf = alt_scratch_buf(alt);
339 fill_sha1_path(buf, sha1);
340 return buf->buf;
343 struct alternate_object_database *alt_odb_list;
344 static struct alternate_object_database **alt_odb_tail;
347 * Return non-zero iff the path is usable as an alternate object database.
349 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
351 struct alternate_object_database *alt;
353 /* Detect cases where alternate disappeared */
354 if (!is_directory(path->buf)) {
355 error("object directory %s does not exist; "
356 "check .git/objects/info/alternates.",
357 path->buf);
358 return 0;
362 * Prevent the common mistake of listing the same
363 * thing twice, or object directory itself.
365 for (alt = alt_odb_list; alt; alt = alt->next) {
366 if (!fspathcmp(path->buf, alt->path))
367 return 0;
369 if (!fspathcmp(path->buf, normalized_objdir))
370 return 0;
372 return 1;
376 * Prepare alternate object database registry.
378 * The variable alt_odb_list points at the list of struct
379 * alternate_object_database. The elements on this list come from
380 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
381 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
382 * whose contents is similar to that environment variable but can be
383 * LF separated. Its base points at a statically allocated buffer that
384 * contains "/the/directory/corresponding/to/.git/objects/...", while
385 * its name points just after the slash at the end of ".git/objects/"
386 * in the example above, and has enough space to hold 40-byte hex
387 * SHA1, an extra slash for the first level indirection, and the
388 * terminating NUL.
390 static void read_info_alternates(const char * relative_base, int depth);
391 static int link_alt_odb_entry(const char *entry, const char *relative_base,
392 int depth, const char *normalized_objdir)
394 struct alternate_object_database *ent;
395 struct strbuf pathbuf = STRBUF_INIT;
397 if (!is_absolute_path(entry) && relative_base) {
398 strbuf_realpath(&pathbuf, relative_base, 1);
399 strbuf_addch(&pathbuf, '/');
401 strbuf_addstr(&pathbuf, entry);
403 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
404 error("unable to normalize alternate object path: %s",
405 pathbuf.buf);
406 strbuf_release(&pathbuf);
407 return -1;
411 * The trailing slash after the directory name is given by
412 * this function at the end. Remove duplicates.
414 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
415 strbuf_setlen(&pathbuf, pathbuf.len - 1);
417 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
418 strbuf_release(&pathbuf);
419 return -1;
422 ent = alloc_alt_odb(pathbuf.buf);
424 /* add the alternate entry */
425 *alt_odb_tail = ent;
426 alt_odb_tail = &(ent->next);
427 ent->next = NULL;
429 /* recursively add alternates */
430 read_info_alternates(pathbuf.buf, depth + 1);
432 strbuf_release(&pathbuf);
433 return 0;
436 static const char *parse_alt_odb_entry(const char *string,
437 int sep,
438 struct strbuf *out)
440 const char *end;
442 strbuf_reset(out);
444 if (*string == '#') {
445 /* comment; consume up to next separator */
446 end = strchrnul(string, sep);
447 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
449 * quoted path; unquote_c_style has copied the
450 * data for us and set "end". Broken quoting (e.g.,
451 * an entry that doesn't end with a quote) falls
452 * back to the unquoted case below.
454 } else {
455 /* normal, unquoted path */
456 end = strchrnul(string, sep);
457 strbuf_add(out, string, end - string);
460 if (*end)
461 end++;
462 return end;
465 static void link_alt_odb_entries(const char *alt, int sep,
466 const char *relative_base, int depth)
468 struct strbuf objdirbuf = STRBUF_INIT;
469 struct strbuf entry = STRBUF_INIT;
471 if (!alt || !*alt)
472 return;
474 if (depth > 5) {
475 error("%s: ignoring alternate object stores, nesting too deep.",
476 relative_base);
477 return;
480 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
481 if (strbuf_normalize_path(&objdirbuf) < 0)
482 die("unable to normalize object directory: %s",
483 objdirbuf.buf);
485 while (*alt) {
486 alt = parse_alt_odb_entry(alt, sep, &entry);
487 if (!entry.len)
488 continue;
489 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
491 strbuf_release(&entry);
492 strbuf_release(&objdirbuf);
495 static void read_info_alternates(const char * relative_base, int depth)
497 char *path;
498 struct strbuf buf = STRBUF_INIT;
500 path = xstrfmt("%s/info/alternates", relative_base);
501 if (strbuf_read_file(&buf, path, 1024) < 0) {
502 warn_on_fopen_errors(path);
503 free(path);
504 return;
507 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
508 strbuf_release(&buf);
509 free(path);
512 struct alternate_object_database *alloc_alt_odb(const char *dir)
514 struct alternate_object_database *ent;
516 FLEX_ALLOC_STR(ent, path, dir);
517 strbuf_init(&ent->scratch, 0);
518 strbuf_addf(&ent->scratch, "%s/", dir);
519 ent->base_len = ent->scratch.len;
521 return ent;
524 void add_to_alternates_file(const char *reference)
526 struct lock_file lock = LOCK_INIT;
527 char *alts = git_pathdup("objects/info/alternates");
528 FILE *in, *out;
529 int found = 0;
531 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
532 out = fdopen_lock_file(&lock, "w");
533 if (!out)
534 die_errno("unable to fdopen alternates lockfile");
536 in = fopen(alts, "r");
537 if (in) {
538 struct strbuf line = STRBUF_INIT;
540 while (strbuf_getline(&line, in) != EOF) {
541 if (!strcmp(reference, line.buf)) {
542 found = 1;
543 break;
545 fprintf_or_die(out, "%s\n", line.buf);
548 strbuf_release(&line);
549 fclose(in);
551 else if (errno != ENOENT)
552 die_errno("unable to read alternates file");
554 if (found) {
555 rollback_lock_file(&lock);
556 } else {
557 fprintf_or_die(out, "%s\n", reference);
558 if (commit_lock_file(&lock))
559 die_errno("unable to move new alternates file into place");
560 if (alt_odb_tail)
561 link_alt_odb_entries(reference, '\n', NULL, 0);
563 free(alts);
566 void add_to_alternates_memory(const char *reference)
569 * Make sure alternates are initialized, or else our entry may be
570 * overwritten when they are.
572 prepare_alt_odb();
574 link_alt_odb_entries(reference, '\n', NULL, 0);
578 * Compute the exact path an alternate is at and returns it. In case of
579 * error NULL is returned and the human readable error is added to `err`
580 * `path` may be relative and should point to $GITDIR.
581 * `err` must not be null.
583 char *compute_alternate_path(const char *path, struct strbuf *err)
585 char *ref_git = NULL;
586 const char *repo, *ref_git_s;
587 int seen_error = 0;
589 ref_git_s = real_path_if_valid(path);
590 if (!ref_git_s) {
591 seen_error = 1;
592 strbuf_addf(err, _("path '%s' does not exist"), path);
593 goto out;
594 } else
596 * Beware: read_gitfile(), real_path() and mkpath()
597 * return static buffer
599 ref_git = xstrdup(ref_git_s);
601 repo = read_gitfile(ref_git);
602 if (!repo)
603 repo = read_gitfile(mkpath("%s/.git", ref_git));
604 if (repo) {
605 free(ref_git);
606 ref_git = xstrdup(repo);
609 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
610 char *ref_git_git = mkpathdup("%s/.git", ref_git);
611 free(ref_git);
612 ref_git = ref_git_git;
613 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
614 struct strbuf sb = STRBUF_INIT;
615 seen_error = 1;
616 if (get_common_dir(&sb, ref_git)) {
617 strbuf_addf(err,
618 _("reference repository '%s' as a linked "
619 "checkout is not supported yet."),
620 path);
621 goto out;
624 strbuf_addf(err, _("reference repository '%s' is not a "
625 "local repository."), path);
626 goto out;
629 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
630 strbuf_addf(err, _("reference repository '%s' is shallow"),
631 path);
632 seen_error = 1;
633 goto out;
636 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
637 strbuf_addf(err,
638 _("reference repository '%s' is grafted"),
639 path);
640 seen_error = 1;
641 goto out;
644 out:
645 if (seen_error) {
646 FREE_AND_NULL(ref_git);
649 return ref_git;
652 int foreach_alt_odb(alt_odb_fn fn, void *cb)
654 struct alternate_object_database *ent;
655 int r = 0;
657 prepare_alt_odb();
658 for (ent = alt_odb_list; ent; ent = ent->next) {
659 r = fn(ent, cb);
660 if (r)
661 break;
663 return r;
666 void prepare_alt_odb(void)
668 const char *alt;
670 if (alt_odb_tail)
671 return;
673 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
675 alt_odb_tail = &alt_odb_list;
676 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
678 read_info_alternates(get_object_directory(), 0);
681 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
682 static int freshen_file(const char *fn)
684 struct utimbuf t;
685 t.actime = t.modtime = time(NULL);
686 return !utime(fn, &t);
690 * All of the check_and_freshen functions return 1 if the file exists and was
691 * freshened (if freshening was requested), 0 otherwise. If they return
692 * 0, you should not assume that it is safe to skip a write of the object (it
693 * either does not exist on disk, or has a stale mtime and may be subject to
694 * pruning).
696 int check_and_freshen_file(const char *fn, int freshen)
698 if (access(fn, F_OK))
699 return 0;
700 if (freshen && !freshen_file(fn))
701 return 0;
702 return 1;
705 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
707 static struct strbuf buf = STRBUF_INIT;
709 strbuf_reset(&buf);
710 sha1_file_name(&buf, sha1);
712 return check_and_freshen_file(buf.buf, freshen);
715 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
717 struct alternate_object_database *alt;
718 prepare_alt_odb();
719 for (alt = alt_odb_list; alt; alt = alt->next) {
720 const char *path = alt_sha1_path(alt, sha1);
721 if (check_and_freshen_file(path, freshen))
722 return 1;
724 return 0;
727 static int check_and_freshen(const unsigned char *sha1, int freshen)
729 return check_and_freshen_local(sha1, freshen) ||
730 check_and_freshen_nonlocal(sha1, freshen);
733 int has_loose_object_nonlocal(const unsigned char *sha1)
735 return check_and_freshen_nonlocal(sha1, 0);
738 static int has_loose_object(const unsigned char *sha1)
740 return check_and_freshen(sha1, 0);
743 static void mmap_limit_check(size_t length)
745 static size_t limit = 0;
746 if (!limit) {
747 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
748 if (!limit)
749 limit = SIZE_MAX;
751 if (length > limit)
752 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
753 (uintmax_t)length, (uintmax_t)limit);
756 void *xmmap_gently(void *start, size_t length,
757 int prot, int flags, int fd, off_t offset)
759 void *ret;
761 mmap_limit_check(length);
762 ret = mmap(start, length, prot, flags, fd, offset);
763 if (ret == MAP_FAILED) {
764 if (!length)
765 return NULL;
766 release_pack_memory(length);
767 ret = mmap(start, length, prot, flags, fd, offset);
769 return ret;
772 void *xmmap(void *start, size_t length,
773 int prot, int flags, int fd, off_t offset)
775 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
776 if (ret == MAP_FAILED)
777 die_errno("mmap failed");
778 return ret;
782 * With an in-core object data in "map", rehash it to make sure the
783 * object name actually matches "sha1" to detect object corruption.
784 * With "map" == NULL, try reading the object named with "sha1" using
785 * the streaming interface and rehash it to do the same.
787 int check_sha1_signature(const unsigned char *sha1, void *map,
788 unsigned long size, const char *type)
790 struct object_id real_oid;
791 enum object_type obj_type;
792 struct git_istream *st;
793 git_hash_ctx c;
794 char hdr[32];
795 int hdrlen;
797 if (map) {
798 hash_object_file(map, size, type, &real_oid);
799 return hashcmp(sha1, real_oid.hash) ? -1 : 0;
802 st = open_istream(sha1, &obj_type, &size, NULL);
803 if (!st)
804 return -1;
806 /* Generate the header */
807 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
809 /* Sha1.. */
810 the_hash_algo->init_fn(&c);
811 the_hash_algo->update_fn(&c, hdr, hdrlen);
812 for (;;) {
813 char buf[1024 * 16];
814 ssize_t readlen = read_istream(st, buf, sizeof(buf));
816 if (readlen < 0) {
817 close_istream(st);
818 return -1;
820 if (!readlen)
821 break;
822 the_hash_algo->update_fn(&c, buf, readlen);
824 the_hash_algo->final_fn(real_oid.hash, &c);
825 close_istream(st);
826 return hashcmp(sha1, real_oid.hash) ? -1 : 0;
829 int git_open_cloexec(const char *name, int flags)
831 int fd;
832 static int o_cloexec = O_CLOEXEC;
834 fd = open(name, flags | o_cloexec);
835 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
836 /* Try again w/o O_CLOEXEC: the kernel might not support it */
837 o_cloexec &= ~O_CLOEXEC;
838 fd = open(name, flags | o_cloexec);
841 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
843 static int fd_cloexec = FD_CLOEXEC;
845 if (!o_cloexec && 0 <= fd && fd_cloexec) {
846 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
847 int flags = fcntl(fd, F_GETFD);
848 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
849 fd_cloexec = 0;
852 #endif
853 return fd;
857 * Find "sha1" as a loose object in the local repository or in an alternate.
858 * Returns 0 on success, negative on failure.
860 * The "path" out-parameter will give the path of the object we found (if any).
861 * Note that it may point to static storage and is only valid until another
862 * call to sha1_file_name(), etc.
864 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
865 const char **path)
867 struct alternate_object_database *alt;
868 static struct strbuf buf = STRBUF_INIT;
870 strbuf_reset(&buf);
871 sha1_file_name(&buf, sha1);
872 *path = buf.buf;
874 if (!lstat(*path, st))
875 return 0;
877 prepare_alt_odb();
878 errno = ENOENT;
879 for (alt = alt_odb_list; alt; alt = alt->next) {
880 *path = alt_sha1_path(alt, sha1);
881 if (!lstat(*path, st))
882 return 0;
885 return -1;
889 * Like stat_sha1_file(), but actually open the object and return the
890 * descriptor. See the caveats on the "path" parameter above.
892 static int open_sha1_file(const unsigned char *sha1, const char **path)
894 int fd;
895 struct alternate_object_database *alt;
896 int most_interesting_errno;
897 static struct strbuf buf = STRBUF_INIT;
899 strbuf_reset(&buf);
900 sha1_file_name(&buf, sha1);
901 *path = buf.buf;
903 fd = git_open(*path);
904 if (fd >= 0)
905 return fd;
906 most_interesting_errno = errno;
908 prepare_alt_odb();
909 for (alt = alt_odb_list; alt; alt = alt->next) {
910 *path = alt_sha1_path(alt, sha1);
911 fd = git_open(*path);
912 if (fd >= 0)
913 return fd;
914 if (most_interesting_errno == ENOENT)
915 most_interesting_errno = errno;
917 errno = most_interesting_errno;
918 return -1;
922 * Map the loose object at "path" if it is not NULL, or the path found by
923 * searching for a loose object named "sha1".
925 static void *map_sha1_file_1(const char *path,
926 const unsigned char *sha1,
927 unsigned long *size)
929 void *map;
930 int fd;
932 if (path)
933 fd = git_open(path);
934 else
935 fd = open_sha1_file(sha1, &path);
936 map = NULL;
937 if (fd >= 0) {
938 struct stat st;
940 if (!fstat(fd, &st)) {
941 *size = xsize_t(st.st_size);
942 if (!*size) {
943 /* mmap() is forbidden on empty files */
944 error("object file %s is empty", path);
945 return NULL;
947 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
949 close(fd);
951 return map;
954 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
956 return map_sha1_file_1(NULL, sha1, size);
959 static int unpack_sha1_short_header(git_zstream *stream,
960 unsigned char *map, unsigned long mapsize,
961 void *buffer, unsigned long bufsiz)
963 /* Get the data stream */
964 memset(stream, 0, sizeof(*stream));
965 stream->next_in = map;
966 stream->avail_in = mapsize;
967 stream->next_out = buffer;
968 stream->avail_out = bufsiz;
970 git_inflate_init(stream);
971 return git_inflate(stream, 0);
974 int unpack_sha1_header(git_zstream *stream,
975 unsigned char *map, unsigned long mapsize,
976 void *buffer, unsigned long bufsiz)
978 int status = unpack_sha1_short_header(stream, map, mapsize,
979 buffer, bufsiz);
981 if (status < Z_OK)
982 return status;
984 /* Make sure we have the terminating NUL */
985 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
986 return -1;
987 return 0;
990 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
991 unsigned long mapsize, void *buffer,
992 unsigned long bufsiz, struct strbuf *header)
994 int status;
996 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
997 if (status < Z_OK)
998 return -1;
1001 * Check if entire header is unpacked in the first iteration.
1003 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1004 return 0;
1007 * buffer[0..bufsiz] was not large enough. Copy the partial
1008 * result out to header, and then append the result of further
1009 * reading the stream.
1011 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1012 stream->next_out = buffer;
1013 stream->avail_out = bufsiz;
1015 do {
1016 status = git_inflate(stream, 0);
1017 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1018 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1019 return 0;
1020 stream->next_out = buffer;
1021 stream->avail_out = bufsiz;
1022 } while (status != Z_STREAM_END);
1023 return -1;
1026 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1028 int bytes = strlen(buffer) + 1;
1029 unsigned char *buf = xmallocz(size);
1030 unsigned long n;
1031 int status = Z_OK;
1033 n = stream->total_out - bytes;
1034 if (n > size)
1035 n = size;
1036 memcpy(buf, (char *) buffer + bytes, n);
1037 bytes = n;
1038 if (bytes <= size) {
1040 * The above condition must be (bytes <= size), not
1041 * (bytes < size). In other words, even though we
1042 * expect no more output and set avail_out to zero,
1043 * the input zlib stream may have bytes that express
1044 * "this concludes the stream", and we *do* want to
1045 * eat that input.
1047 * Otherwise we would not be able to test that we
1048 * consumed all the input to reach the expected size;
1049 * we also want to check that zlib tells us that all
1050 * went well with status == Z_STREAM_END at the end.
1052 stream->next_out = buf + bytes;
1053 stream->avail_out = size - bytes;
1054 while (status == Z_OK)
1055 status = git_inflate(stream, Z_FINISH);
1057 if (status == Z_STREAM_END && !stream->avail_in) {
1058 git_inflate_end(stream);
1059 return buf;
1062 if (status < 0)
1063 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1064 else if (stream->avail_in)
1065 error("garbage at end of loose object '%s'",
1066 sha1_to_hex(sha1));
1067 free(buf);
1068 return NULL;
1072 * We used to just use "sscanf()", but that's actually way
1073 * too permissive for what we want to check. So do an anal
1074 * object header parse by hand.
1076 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1077 unsigned int flags)
1079 const char *type_buf = hdr;
1080 unsigned long size;
1081 int type, type_len = 0;
1084 * The type can be of any size but is followed by
1085 * a space.
1087 for (;;) {
1088 char c = *hdr++;
1089 if (!c)
1090 return -1;
1091 if (c == ' ')
1092 break;
1093 type_len++;
1096 type = type_from_string_gently(type_buf, type_len, 1);
1097 if (oi->type_name)
1098 strbuf_add(oi->type_name, type_buf, type_len);
1100 * Set type to 0 if its an unknown object and
1101 * we're obtaining the type using '--allow-unknown-type'
1102 * option.
1104 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1105 type = 0;
1106 else if (type < 0)
1107 die("invalid object type");
1108 if (oi->typep)
1109 *oi->typep = type;
1112 * The length must follow immediately, and be in canonical
1113 * decimal format (ie "010" is not valid).
1115 size = *hdr++ - '0';
1116 if (size > 9)
1117 return -1;
1118 if (size) {
1119 for (;;) {
1120 unsigned long c = *hdr - '0';
1121 if (c > 9)
1122 break;
1123 hdr++;
1124 size = size * 10 + c;
1128 if (oi->sizep)
1129 *oi->sizep = size;
1132 * The length must be followed by a zero byte
1134 return *hdr ? -1 : type;
1137 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1139 struct object_info oi = OBJECT_INFO_INIT;
1141 oi.sizep = sizep;
1142 return parse_sha1_header_extended(hdr, &oi, 0);
1145 static int sha1_loose_object_info(const unsigned char *sha1,
1146 struct object_info *oi,
1147 int flags)
1149 int status = 0;
1150 unsigned long mapsize;
1151 void *map;
1152 git_zstream stream;
1153 char hdr[32];
1154 struct strbuf hdrbuf = STRBUF_INIT;
1155 unsigned long size_scratch;
1157 if (oi->delta_base_sha1)
1158 hashclr(oi->delta_base_sha1);
1161 * If we don't care about type or size, then we don't
1162 * need to look inside the object at all. Note that we
1163 * do not optimize out the stat call, even if the
1164 * caller doesn't care about the disk-size, since our
1165 * return value implicitly indicates whether the
1166 * object even exists.
1168 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1169 const char *path;
1170 struct stat st;
1171 if (stat_sha1_file(sha1, &st, &path) < 0)
1172 return -1;
1173 if (oi->disk_sizep)
1174 *oi->disk_sizep = st.st_size;
1175 return 0;
1178 map = map_sha1_file(sha1, &mapsize);
1179 if (!map)
1180 return -1;
1182 if (!oi->sizep)
1183 oi->sizep = &size_scratch;
1185 if (oi->disk_sizep)
1186 *oi->disk_sizep = mapsize;
1187 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1188 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1189 status = error("unable to unpack %s header with --allow-unknown-type",
1190 sha1_to_hex(sha1));
1191 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1192 status = error("unable to unpack %s header",
1193 sha1_to_hex(sha1));
1194 if (status < 0)
1195 ; /* Do nothing */
1196 else if (hdrbuf.len) {
1197 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1198 status = error("unable to parse %s header with --allow-unknown-type",
1199 sha1_to_hex(sha1));
1200 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1201 status = error("unable to parse %s header", sha1_to_hex(sha1));
1203 if (status >= 0 && oi->contentp) {
1204 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1205 *oi->sizep, sha1);
1206 if (!*oi->contentp) {
1207 git_inflate_end(&stream);
1208 status = -1;
1210 } else
1211 git_inflate_end(&stream);
1213 munmap(map, mapsize);
1214 if (status && oi->typep)
1215 *oi->typep = status;
1216 if (oi->sizep == &size_scratch)
1217 oi->sizep = NULL;
1218 strbuf_release(&hdrbuf);
1219 oi->whence = OI_LOOSE;
1220 return (status < 0) ? status : 0;
1223 int fetch_if_missing = 1;
1225 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1227 static struct object_info blank_oi = OBJECT_INFO_INIT;
1228 struct pack_entry e;
1229 int rtype;
1230 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1231 lookup_replace_object(sha1) :
1232 sha1;
1233 int already_retried = 0;
1235 if (is_null_sha1(real))
1236 return -1;
1238 if (!oi)
1239 oi = &blank_oi;
1241 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1242 struct cached_object *co = find_cached_object(real);
1243 if (co) {
1244 if (oi->typep)
1245 *(oi->typep) = co->type;
1246 if (oi->sizep)
1247 *(oi->sizep) = co->size;
1248 if (oi->disk_sizep)
1249 *(oi->disk_sizep) = 0;
1250 if (oi->delta_base_sha1)
1251 hashclr(oi->delta_base_sha1);
1252 if (oi->type_name)
1253 strbuf_addstr(oi->type_name, type_name(co->type));
1254 if (oi->contentp)
1255 *oi->contentp = xmemdupz(co->buf, co->size);
1256 oi->whence = OI_CACHED;
1257 return 0;
1261 while (1) {
1262 if (find_pack_entry(real, &e))
1263 break;
1265 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1266 return -1;
1268 /* Most likely it's a loose object. */
1269 if (!sha1_loose_object_info(real, oi, flags))
1270 return 0;
1272 /* Not a loose object; someone else may have just packed it. */
1273 if (!(flags & OBJECT_INFO_QUICK)) {
1274 reprepare_packed_git();
1275 if (find_pack_entry(real, &e))
1276 break;
1279 /* Check if it is a missing object */
1280 if (fetch_if_missing && repository_format_partial_clone &&
1281 !already_retried) {
1283 * TODO Investigate haveing fetch_object() return
1284 * TODO error/success and stopping the music here.
1286 fetch_object(repository_format_partial_clone, real);
1287 already_retried = 1;
1288 continue;
1291 return -1;
1294 if (oi == &blank_oi)
1296 * We know that the caller doesn't actually need the
1297 * information below, so return early.
1299 return 0;
1300 rtype = packed_object_info(e.p, e.offset, oi);
1301 if (rtype < 0) {
1302 mark_bad_packed_object(e.p, real);
1303 return sha1_object_info_extended(real, oi, 0);
1304 } else if (oi->whence == OI_PACKED) {
1305 oi->u.packed.offset = e.offset;
1306 oi->u.packed.pack = e.p;
1307 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1308 rtype == OBJ_OFS_DELTA);
1311 return 0;
1314 /* returns enum object_type or negative */
1315 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1317 enum object_type type;
1318 struct object_info oi = OBJECT_INFO_INIT;
1320 oi.typep = &type;
1321 oi.sizep = sizep;
1322 if (sha1_object_info_extended(sha1, &oi,
1323 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1324 return -1;
1325 return type;
1328 static void *read_object(const unsigned char *sha1, enum object_type *type,
1329 unsigned long *size)
1331 struct object_info oi = OBJECT_INFO_INIT;
1332 void *content;
1333 oi.typep = type;
1334 oi.sizep = size;
1335 oi.contentp = &content;
1337 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1338 return NULL;
1339 return content;
1342 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1343 struct object_id *oid)
1345 struct cached_object *co;
1347 hash_object_file(buf, len, type_name(type), oid);
1348 if (has_sha1_file(oid->hash) || find_cached_object(oid->hash))
1349 return 0;
1350 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1351 co = &cached_objects[cached_object_nr++];
1352 co->size = len;
1353 co->type = type;
1354 co->buf = xmalloc(len);
1355 memcpy(co->buf, buf, len);
1356 hashcpy(co->sha1, oid->hash);
1357 return 0;
1361 * This function dies on corrupt objects; the callers who want to
1362 * deal with them should arrange to call read_object() and give error
1363 * messages themselves.
1365 void *read_sha1_file_extended(const unsigned char *sha1,
1366 enum object_type *type,
1367 unsigned long *size,
1368 int lookup_replace)
1370 void *data;
1371 const struct packed_git *p;
1372 const char *path;
1373 struct stat st;
1374 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1375 : sha1;
1377 errno = 0;
1378 data = read_object(repl, type, size);
1379 if (data)
1380 return data;
1382 if (errno && errno != ENOENT)
1383 die_errno("failed to read object %s", sha1_to_hex(sha1));
1385 /* die if we replaced an object with one that does not exist */
1386 if (repl != sha1)
1387 die("replacement %s not found for %s",
1388 sha1_to_hex(repl), sha1_to_hex(sha1));
1390 if (!stat_sha1_file(repl, &st, &path))
1391 die("loose object %s (stored in %s) is corrupt",
1392 sha1_to_hex(repl), path);
1394 if ((p = has_packed_and_bad(repl)) != NULL)
1395 die("packed object %s (stored in %s) is corrupt",
1396 sha1_to_hex(repl), p->pack_name);
1398 return NULL;
1401 void *read_object_with_reference(const unsigned char *sha1,
1402 const char *required_type_name,
1403 unsigned long *size,
1404 unsigned char *actual_sha1_return)
1406 enum object_type type, required_type;
1407 void *buffer;
1408 unsigned long isize;
1409 unsigned char actual_sha1[20];
1411 required_type = type_from_string(required_type_name);
1412 hashcpy(actual_sha1, sha1);
1413 while (1) {
1414 int ref_length = -1;
1415 const char *ref_type = NULL;
1417 buffer = read_sha1_file(actual_sha1, &type, &isize);
1418 if (!buffer)
1419 return NULL;
1420 if (type == required_type) {
1421 *size = isize;
1422 if (actual_sha1_return)
1423 hashcpy(actual_sha1_return, actual_sha1);
1424 return buffer;
1426 /* Handle references */
1427 else if (type == OBJ_COMMIT)
1428 ref_type = "tree ";
1429 else if (type == OBJ_TAG)
1430 ref_type = "object ";
1431 else {
1432 free(buffer);
1433 return NULL;
1435 ref_length = strlen(ref_type);
1437 if (ref_length + 40 > isize ||
1438 memcmp(buffer, ref_type, ref_length) ||
1439 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1440 free(buffer);
1441 return NULL;
1443 free(buffer);
1444 /* Now we have the ID of the referred-to object in
1445 * actual_sha1. Check again. */
1449 static void write_object_file_prepare(const void *buf, unsigned long len,
1450 const char *type, struct object_id *oid,
1451 char *hdr, int *hdrlen)
1453 git_hash_ctx c;
1455 /* Generate the header */
1456 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1458 /* Sha1.. */
1459 the_hash_algo->init_fn(&c);
1460 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1461 the_hash_algo->update_fn(&c, buf, len);
1462 the_hash_algo->final_fn(oid->hash, &c);
1466 * Move the just written object into its final resting place.
1468 int finalize_object_file(const char *tmpfile, const char *filename)
1470 int ret = 0;
1472 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1473 goto try_rename;
1474 else if (link(tmpfile, filename))
1475 ret = errno;
1478 * Coda hack - coda doesn't like cross-directory links,
1479 * so we fall back to a rename, which will mean that it
1480 * won't be able to check collisions, but that's not a
1481 * big deal.
1483 * The same holds for FAT formatted media.
1485 * When this succeeds, we just return. We have nothing
1486 * left to unlink.
1488 if (ret && ret != EEXIST) {
1489 try_rename:
1490 if (!rename(tmpfile, filename))
1491 goto out;
1492 ret = errno;
1494 unlink_or_warn(tmpfile);
1495 if (ret) {
1496 if (ret != EEXIST) {
1497 return error_errno("unable to write sha1 filename %s", filename);
1499 /* FIXME!!! Collision check here ? */
1502 out:
1503 if (adjust_shared_perm(filename))
1504 return error("unable to set permission to '%s'", filename);
1505 return 0;
1508 static int write_buffer(int fd, const void *buf, size_t len)
1510 if (write_in_full(fd, buf, len) < 0)
1511 return error_errno("file write error");
1512 return 0;
1515 int hash_object_file(const void *buf, unsigned long len, const char *type,
1516 struct object_id *oid)
1518 char hdr[32];
1519 int hdrlen = sizeof(hdr);
1520 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1521 return 0;
1524 /* Finalize a file on disk, and close it. */
1525 static void close_sha1_file(int fd)
1527 if (fsync_object_files)
1528 fsync_or_die(fd, "sha1 file");
1529 if (close(fd) != 0)
1530 die_errno("error when closing sha1 file");
1533 /* Size of directory component, including the ending '/' */
1534 static inline int directory_size(const char *filename)
1536 const char *s = strrchr(filename, '/');
1537 if (!s)
1538 return 0;
1539 return s - filename + 1;
1543 * This creates a temporary file in the same directory as the final
1544 * 'filename'
1546 * We want to avoid cross-directory filename renames, because those
1547 * can have problems on various filesystems (FAT, NFS, Coda).
1549 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1551 int fd, dirlen = directory_size(filename);
1553 strbuf_reset(tmp);
1554 strbuf_add(tmp, filename, dirlen);
1555 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1556 fd = git_mkstemp_mode(tmp->buf, 0444);
1557 if (fd < 0 && dirlen && errno == ENOENT) {
1559 * Make sure the directory exists; note that the contents
1560 * of the buffer are undefined after mkstemp returns an
1561 * error, so we have to rewrite the whole buffer from
1562 * scratch.
1564 strbuf_reset(tmp);
1565 strbuf_add(tmp, filename, dirlen - 1);
1566 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1567 return -1;
1568 if (adjust_shared_perm(tmp->buf))
1569 return -1;
1571 /* Try again */
1572 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1573 fd = git_mkstemp_mode(tmp->buf, 0444);
1575 return fd;
1578 static int write_loose_object(const struct object_id *oid, char *hdr,
1579 int hdrlen, const void *buf, unsigned long len,
1580 time_t mtime)
1582 int fd, ret;
1583 unsigned char compressed[4096];
1584 git_zstream stream;
1585 git_hash_ctx c;
1586 struct object_id parano_oid;
1587 static struct strbuf tmp_file = STRBUF_INIT;
1588 static struct strbuf filename = STRBUF_INIT;
1590 strbuf_reset(&filename);
1591 sha1_file_name(&filename, oid->hash);
1593 fd = create_tmpfile(&tmp_file, filename.buf);
1594 if (fd < 0) {
1595 if (errno == EACCES)
1596 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1597 else
1598 return error_errno("unable to create temporary file");
1601 /* Set it up */
1602 git_deflate_init(&stream, zlib_compression_level);
1603 stream.next_out = compressed;
1604 stream.avail_out = sizeof(compressed);
1605 the_hash_algo->init_fn(&c);
1607 /* First header.. */
1608 stream.next_in = (unsigned char *)hdr;
1609 stream.avail_in = hdrlen;
1610 while (git_deflate(&stream, 0) == Z_OK)
1611 ; /* nothing */
1612 the_hash_algo->update_fn(&c, hdr, hdrlen);
1614 /* Then the data itself.. */
1615 stream.next_in = (void *)buf;
1616 stream.avail_in = len;
1617 do {
1618 unsigned char *in0 = stream.next_in;
1619 ret = git_deflate(&stream, Z_FINISH);
1620 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1621 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1622 die("unable to write sha1 file");
1623 stream.next_out = compressed;
1624 stream.avail_out = sizeof(compressed);
1625 } while (ret == Z_OK);
1627 if (ret != Z_STREAM_END)
1628 die("unable to deflate new object %s (%d)", oid_to_hex(oid),
1629 ret);
1630 ret = git_deflate_end_gently(&stream);
1631 if (ret != Z_OK)
1632 die("deflateEnd on object %s failed (%d)", oid_to_hex(oid),
1633 ret);
1634 the_hash_algo->final_fn(parano_oid.hash, &c);
1635 if (oidcmp(oid, &parano_oid) != 0)
1636 die("confused by unstable object source data for %s",
1637 oid_to_hex(oid));
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_object_file(const void *buf, unsigned long len, const char *type,
1671 struct object_id *oid)
1673 char hdr[32];
1674 int hdrlen = sizeof(hdr);
1676 /* Normally if we have it in the pack then we do not bother writing
1677 * it out into .git/objects/??/?{38} file.
1679 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1680 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1681 return 0;
1682 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1685 int hash_object_file_literally(const void *buf, unsigned long len,
1686 const char *type, struct object_id *oid,
1687 unsigned flags)
1689 char *header;
1690 int hdrlen, status = 0;
1692 /* type string, SP, %lu of the length plus NUL must fit this */
1693 hdrlen = strlen(type) + 32;
1694 header = xmalloc(hdrlen);
1695 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1697 if (!(flags & HASH_WRITE_OBJECT))
1698 goto cleanup;
1699 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1700 goto cleanup;
1701 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1703 cleanup:
1704 free(header);
1705 return status;
1708 int force_object_loose(const struct object_id *oid, time_t mtime)
1710 void *buf;
1711 unsigned long len;
1712 enum object_type type;
1713 char hdr[32];
1714 int hdrlen;
1715 int ret;
1717 if (has_loose_object(oid->hash))
1718 return 0;
1719 buf = read_object(oid->hash, &type, &len);
1720 if (!buf)
1721 return error("cannot read sha1_file for %s", oid_to_hex(oid));
1722 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1723 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1724 free(buf);
1726 return ret;
1729 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1731 if (!startup_info->have_repository)
1732 return 0;
1733 return sha1_object_info_extended(sha1, NULL,
1734 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1737 int has_object_file(const struct object_id *oid)
1739 return has_sha1_file(oid->hash);
1742 int has_object_file_with_flags(const struct object_id *oid, int flags)
1744 return has_sha1_file_with_flags(oid->hash, flags);
1747 static void check_tree(const void *buf, size_t size)
1749 struct tree_desc desc;
1750 struct name_entry entry;
1752 init_tree_desc(&desc, buf, size);
1753 while (tree_entry(&desc, &entry))
1754 /* do nothing
1755 * tree_entry() will die() on malformed entries */
1759 static void check_commit(const void *buf, size_t size)
1761 struct commit c;
1762 memset(&c, 0, sizeof(c));
1763 if (parse_commit_buffer(&c, buf, size))
1764 die("corrupt commit");
1767 static void check_tag(const void *buf, size_t size)
1769 struct tag t;
1770 memset(&t, 0, sizeof(t));
1771 if (parse_tag_buffer(&t, buf, size))
1772 die("corrupt tag");
1775 static int index_mem(struct object_id *oid, void *buf, size_t size,
1776 enum object_type type,
1777 const char *path, unsigned flags)
1779 int ret, re_allocated = 0;
1780 int write_object = flags & HASH_WRITE_OBJECT;
1782 if (!type)
1783 type = OBJ_BLOB;
1786 * Convert blobs to git internal format
1788 if ((type == OBJ_BLOB) && path) {
1789 struct strbuf nbuf = STRBUF_INIT;
1790 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1791 get_conv_flags(flags))) {
1792 buf = strbuf_detach(&nbuf, &size);
1793 re_allocated = 1;
1796 if (flags & HASH_FORMAT_CHECK) {
1797 if (type == OBJ_TREE)
1798 check_tree(buf, size);
1799 if (type == OBJ_COMMIT)
1800 check_commit(buf, size);
1801 if (type == OBJ_TAG)
1802 check_tag(buf, size);
1805 if (write_object)
1806 ret = write_object_file(buf, size, type_name(type), oid);
1807 else
1808 ret = hash_object_file(buf, size, type_name(type), oid);
1809 if (re_allocated)
1810 free(buf);
1811 return ret;
1814 static int index_stream_convert_blob(struct object_id *oid, int fd,
1815 const char *path, unsigned flags)
1817 int ret;
1818 const int write_object = flags & HASH_WRITE_OBJECT;
1819 struct strbuf sbuf = STRBUF_INIT;
1821 assert(path);
1822 assert(would_convert_to_git_filter_fd(path));
1824 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1825 get_conv_flags(flags));
1827 if (write_object)
1828 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1829 oid);
1830 else
1831 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1832 oid);
1833 strbuf_release(&sbuf);
1834 return ret;
1837 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1838 const char *path, unsigned flags)
1840 struct strbuf sbuf = STRBUF_INIT;
1841 int ret;
1843 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1844 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1845 else
1846 ret = -1;
1847 strbuf_release(&sbuf);
1848 return ret;
1851 #define SMALL_FILE_SIZE (32*1024)
1853 static int index_core(struct object_id *oid, int fd, size_t size,
1854 enum object_type type, const char *path,
1855 unsigned flags)
1857 int ret;
1859 if (!size) {
1860 ret = index_mem(oid, "", size, type, path, flags);
1861 } else if (size <= SMALL_FILE_SIZE) {
1862 char *buf = xmalloc(size);
1863 ssize_t read_result = read_in_full(fd, buf, size);
1864 if (read_result < 0)
1865 ret = error_errno("read error while indexing %s",
1866 path ? path : "<unknown>");
1867 else if (read_result != size)
1868 ret = error("short read while indexing %s",
1869 path ? path : "<unknown>");
1870 else
1871 ret = index_mem(oid, buf, size, type, path, flags);
1872 free(buf);
1873 } else {
1874 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1875 ret = index_mem(oid, buf, size, type, path, flags);
1876 munmap(buf, size);
1878 return ret;
1882 * This creates one packfile per large blob unless bulk-checkin
1883 * machinery is "plugged".
1885 * This also bypasses the usual "convert-to-git" dance, and that is on
1886 * purpose. We could write a streaming version of the converting
1887 * functions and insert that before feeding the data to fast-import
1888 * (or equivalent in-core API described above). However, that is
1889 * somewhat complicated, as we do not know the size of the filter
1890 * result, which we need to know beforehand when writing a git object.
1891 * Since the primary motivation for trying to stream from the working
1892 * tree file and to avoid mmaping it in core is to deal with large
1893 * binary blobs, they generally do not want to get any conversion, and
1894 * callers should avoid this code path when filters are requested.
1896 static int index_stream(struct object_id *oid, int fd, size_t size,
1897 enum object_type type, const char *path,
1898 unsigned flags)
1900 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1903 int index_fd(struct object_id *oid, int fd, struct stat *st,
1904 enum object_type type, const char *path, unsigned flags)
1906 int ret;
1909 * Call xsize_t() only when needed to avoid potentially unnecessary
1910 * die() for large files.
1912 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1913 ret = index_stream_convert_blob(oid, fd, path, flags);
1914 else if (!S_ISREG(st->st_mode))
1915 ret = index_pipe(oid, fd, type, path, flags);
1916 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1917 (path && would_convert_to_git(&the_index, path)))
1918 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1919 flags);
1920 else
1921 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1922 flags);
1923 close(fd);
1924 return ret;
1927 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1929 int fd;
1930 struct strbuf sb = STRBUF_INIT;
1931 int rc = 0;
1933 switch (st->st_mode & S_IFMT) {
1934 case S_IFREG:
1935 fd = open(path, O_RDONLY);
1936 if (fd < 0)
1937 return error_errno("open(\"%s\")", path);
1938 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1939 return error("%s: failed to insert into database",
1940 path);
1941 break;
1942 case S_IFLNK:
1943 if (strbuf_readlink(&sb, path, st->st_size))
1944 return error_errno("readlink(\"%s\")", path);
1945 if (!(flags & HASH_WRITE_OBJECT))
1946 hash_object_file(sb.buf, sb.len, blob_type, oid);
1947 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
1948 rc = error("%s: failed to insert into database", path);
1949 strbuf_release(&sb);
1950 break;
1951 case S_IFDIR:
1952 return resolve_gitlink_ref(path, "HEAD", oid);
1953 default:
1954 return error("%s: unsupported file type", path);
1956 return rc;
1959 int read_pack_header(int fd, struct pack_header *header)
1961 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1962 /* "eof before pack header was fully read" */
1963 return PH_ERROR_EOF;
1965 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1966 /* "protocol error (pack signature mismatch detected)" */
1967 return PH_ERROR_PACK_SIGNATURE;
1968 if (!pack_version_ok(header->hdr_version))
1969 /* "protocol error (pack version unsupported)" */
1970 return PH_ERROR_PROTOCOL;
1971 return 0;
1974 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1976 enum object_type type = sha1_object_info(sha1, NULL);
1977 if (type < 0)
1978 die("%s is not a valid object", sha1_to_hex(sha1));
1979 if (type != expect)
1980 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1981 type_name(expect));
1984 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1985 struct strbuf *path,
1986 each_loose_object_fn obj_cb,
1987 each_loose_cruft_fn cruft_cb,
1988 each_loose_subdir_fn subdir_cb,
1989 void *data)
1991 size_t origlen, baselen;
1992 DIR *dir;
1993 struct dirent *de;
1994 int r = 0;
1995 struct object_id oid;
1997 if (subdir_nr > 0xff)
1998 BUG("invalid loose object subdirectory: %x", subdir_nr);
2000 origlen = path->len;
2001 strbuf_complete(path, '/');
2002 strbuf_addf(path, "%02x", subdir_nr);
2004 dir = opendir(path->buf);
2005 if (!dir) {
2006 if (errno != ENOENT)
2007 r = error_errno("unable to open %s", path->buf);
2008 strbuf_setlen(path, origlen);
2009 return r;
2012 oid.hash[0] = subdir_nr;
2013 strbuf_addch(path, '/');
2014 baselen = path->len;
2016 while ((de = readdir(dir))) {
2017 size_t namelen;
2018 if (is_dot_or_dotdot(de->d_name))
2019 continue;
2021 namelen = strlen(de->d_name);
2022 strbuf_setlen(path, baselen);
2023 strbuf_add(path, de->d_name, namelen);
2024 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2025 !hex_to_bytes(oid.hash + 1, de->d_name,
2026 GIT_SHA1_RAWSZ - 1)) {
2027 if (obj_cb) {
2028 r = obj_cb(&oid, path->buf, data);
2029 if (r)
2030 break;
2032 continue;
2035 if (cruft_cb) {
2036 r = cruft_cb(de->d_name, path->buf, data);
2037 if (r)
2038 break;
2041 closedir(dir);
2043 strbuf_setlen(path, baselen - 1);
2044 if (!r && subdir_cb)
2045 r = subdir_cb(subdir_nr, path->buf, data);
2047 strbuf_setlen(path, origlen);
2049 return r;
2052 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2053 each_loose_object_fn obj_cb,
2054 each_loose_cruft_fn cruft_cb,
2055 each_loose_subdir_fn subdir_cb,
2056 void *data)
2058 int r = 0;
2059 int i;
2061 for (i = 0; i < 256; i++) {
2062 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2063 subdir_cb, data);
2064 if (r)
2065 break;
2068 return r;
2071 int for_each_loose_file_in_objdir(const char *path,
2072 each_loose_object_fn obj_cb,
2073 each_loose_cruft_fn cruft_cb,
2074 each_loose_subdir_fn subdir_cb,
2075 void *data)
2077 struct strbuf buf = STRBUF_INIT;
2078 int r;
2080 strbuf_addstr(&buf, path);
2081 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2082 subdir_cb, data);
2083 strbuf_release(&buf);
2085 return r;
2088 struct loose_alt_odb_data {
2089 each_loose_object_fn *cb;
2090 void *data;
2093 static int loose_from_alt_odb(struct alternate_object_database *alt,
2094 void *vdata)
2096 struct loose_alt_odb_data *data = vdata;
2097 struct strbuf buf = STRBUF_INIT;
2098 int r;
2100 strbuf_addstr(&buf, alt->path);
2101 r = for_each_loose_file_in_objdir_buf(&buf,
2102 data->cb, NULL, NULL,
2103 data->data);
2104 strbuf_release(&buf);
2105 return r;
2108 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2110 struct loose_alt_odb_data alt;
2111 int r;
2113 r = for_each_loose_file_in_objdir(get_object_directory(),
2114 cb, NULL, NULL, data);
2115 if (r)
2116 return r;
2118 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2119 return 0;
2121 alt.cb = cb;
2122 alt.data = data;
2123 return foreach_alt_odb(loose_from_alt_odb, &alt);
2126 static int check_stream_sha1(git_zstream *stream,
2127 const char *hdr,
2128 unsigned long size,
2129 const char *path,
2130 const unsigned char *expected_sha1)
2132 git_hash_ctx c;
2133 unsigned char real_sha1[GIT_MAX_RAWSZ];
2134 unsigned char buf[4096];
2135 unsigned long total_read;
2136 int status = Z_OK;
2138 the_hash_algo->init_fn(&c);
2139 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2142 * We already read some bytes into hdr, but the ones up to the NUL
2143 * do not count against the object's content size.
2145 total_read = stream->total_out - strlen(hdr) - 1;
2148 * This size comparison must be "<=" to read the final zlib packets;
2149 * see the comment in unpack_sha1_rest for details.
2151 while (total_read <= size &&
2152 (status == Z_OK || status == Z_BUF_ERROR)) {
2153 stream->next_out = buf;
2154 stream->avail_out = sizeof(buf);
2155 if (size - total_read < stream->avail_out)
2156 stream->avail_out = size - total_read;
2157 status = git_inflate(stream, Z_FINISH);
2158 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2159 total_read += stream->next_out - buf;
2161 git_inflate_end(stream);
2163 if (status != Z_STREAM_END) {
2164 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2165 return -1;
2167 if (stream->avail_in) {
2168 error("garbage at end of loose object '%s'",
2169 sha1_to_hex(expected_sha1));
2170 return -1;
2173 the_hash_algo->final_fn(real_sha1, &c);
2174 if (hashcmp(expected_sha1, real_sha1)) {
2175 error("sha1 mismatch for %s (expected %s)", path,
2176 sha1_to_hex(expected_sha1));
2177 return -1;
2180 return 0;
2183 int read_loose_object(const char *path,
2184 const unsigned char *expected_sha1,
2185 enum object_type *type,
2186 unsigned long *size,
2187 void **contents)
2189 int ret = -1;
2190 void *map = NULL;
2191 unsigned long mapsize;
2192 git_zstream stream;
2193 char hdr[32];
2195 *contents = NULL;
2197 map = map_sha1_file_1(path, NULL, &mapsize);
2198 if (!map) {
2199 error_errno("unable to mmap %s", path);
2200 goto out;
2203 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2204 error("unable to unpack header of %s", path);
2205 goto out;
2208 *type = parse_sha1_header(hdr, size);
2209 if (*type < 0) {
2210 error("unable to parse header of %s", path);
2211 git_inflate_end(&stream);
2212 goto out;
2215 if (*type == OBJ_BLOB) {
2216 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2217 goto out;
2218 } else {
2219 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2220 if (!*contents) {
2221 error("unable to unpack contents of %s", path);
2222 git_inflate_end(&stream);
2223 goto out;
2225 if (check_sha1_signature(expected_sha1, *contents,
2226 *size, type_name(*type))) {
2227 error("sha1 mismatch for %s (expected %s)", path,
2228 sha1_to_hex(expected_sha1));
2229 free(*contents);
2230 goto out;
2234 ret = 0; /* everything checks out */
2236 out:
2237 if (map)
2238 munmap(map, mapsize);
2239 return ret;