Integrate hash algorithm support with repo setup
[git.git] / sha1_file.c
bloba04389be71dc27f18106226d3c24624e4b62434b
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 "mru.h"
28 #include "list.h"
29 #include "mergesort.h"
30 #include "quote.h"
31 #include "packfile.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(void *ctx)
44 git_SHA1_Init((git_SHA_CTX *)ctx);
47 static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
49 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
52 static void git_hash_sha1_final(unsigned char *hash, void *ctx)
54 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
57 static void git_hash_unknown_init(void *ctx)
59 die("trying to init unknown hash");
62 static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
64 die("trying to update unknown hash");
67 static void git_hash_unknown_final(unsigned char *hash, void *ctx)
69 die("trying to finalize unknown hash");
72 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
74 NULL,
75 0x00000000,
79 git_hash_unknown_init,
80 git_hash_unknown_update,
81 git_hash_unknown_final,
82 NULL,
83 NULL,
86 "sha-1",
87 /* "sha1", big-endian */
88 0x73686131,
89 sizeof(git_SHA_CTX),
90 GIT_SHA1_RAWSZ,
91 GIT_SHA1_HEXSZ,
92 git_hash_sha1_init,
93 git_hash_sha1_update,
94 git_hash_sha1_final,
95 &empty_tree_oid,
96 &empty_blob_oid,
101 * This is meant to hold a *small* number of objects that you would
102 * want read_sha1_file() to be able to return, but yet you do not want
103 * to write them into the object store (e.g. a browse-only
104 * application).
106 static struct cached_object {
107 unsigned char sha1[20];
108 enum object_type type;
109 void *buf;
110 unsigned long size;
111 } *cached_objects;
112 static int cached_object_nr, cached_object_alloc;
114 static struct cached_object empty_tree = {
115 EMPTY_TREE_SHA1_BIN_LITERAL,
116 OBJ_TREE,
121 static struct cached_object *find_cached_object(const unsigned char *sha1)
123 int i;
124 struct cached_object *co = cached_objects;
126 for (i = 0; i < cached_object_nr; i++, co++) {
127 if (!hashcmp(co->sha1, sha1))
128 return co;
130 if (!hashcmp(sha1, empty_tree.sha1))
131 return &empty_tree;
132 return NULL;
135 int mkdir_in_gitdir(const char *path)
137 if (mkdir(path, 0777)) {
138 int saved_errno = errno;
139 struct stat st;
140 struct strbuf sb = STRBUF_INIT;
142 if (errno != EEXIST)
143 return -1;
145 * Are we looking at a path in a symlinked worktree
146 * whose original repository does not yet have it?
147 * e.g. .git/rr-cache pointing at its original
148 * repository in which the user hasn't performed any
149 * conflict resolution yet?
151 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
152 strbuf_readlink(&sb, path, st.st_size) ||
153 !is_absolute_path(sb.buf) ||
154 mkdir(sb.buf, 0777)) {
155 strbuf_release(&sb);
156 errno = saved_errno;
157 return -1;
159 strbuf_release(&sb);
161 return adjust_shared_perm(path);
164 enum scld_error safe_create_leading_directories(char *path)
166 char *next_component = path + offset_1st_component(path);
167 enum scld_error ret = SCLD_OK;
169 while (ret == SCLD_OK && next_component) {
170 struct stat st;
171 char *slash = next_component, slash_character;
173 while (*slash && !is_dir_sep(*slash))
174 slash++;
176 if (!*slash)
177 break;
179 next_component = slash + 1;
180 while (is_dir_sep(*next_component))
181 next_component++;
182 if (!*next_component)
183 break;
185 slash_character = *slash;
186 *slash = '\0';
187 if (!stat(path, &st)) {
188 /* path exists */
189 if (!S_ISDIR(st.st_mode)) {
190 errno = ENOTDIR;
191 ret = SCLD_EXISTS;
193 } else if (mkdir(path, 0777)) {
194 if (errno == EEXIST &&
195 !stat(path, &st) && S_ISDIR(st.st_mode))
196 ; /* somebody created it since we checked */
197 else if (errno == ENOENT)
199 * Either mkdir() failed because
200 * somebody just pruned the containing
201 * directory, or stat() failed because
202 * the file that was in our way was
203 * just removed. Either way, inform
204 * the caller that it might be worth
205 * trying again:
207 ret = SCLD_VANISHED;
208 else
209 ret = SCLD_FAILED;
210 } else if (adjust_shared_perm(path)) {
211 ret = SCLD_PERMS;
213 *slash = slash_character;
215 return ret;
218 enum scld_error safe_create_leading_directories_const(const char *path)
220 int save_errno;
221 /* path points to cache entries, so xstrdup before messing with it */
222 char *buf = xstrdup(path);
223 enum scld_error result = safe_create_leading_directories(buf);
225 save_errno = errno;
226 free(buf);
227 errno = save_errno;
228 return result;
231 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
234 * The number of times we will try to remove empty directories
235 * in the way of path. This is only 1 because if another
236 * process is racily creating directories that conflict with
237 * us, we don't want to fight against them.
239 int remove_directories_remaining = 1;
242 * The number of times that we will try to create the
243 * directories containing path. We are willing to attempt this
244 * more than once, because another process could be trying to
245 * clean up empty directories at the same time as we are
246 * trying to create them.
248 int create_directories_remaining = 3;
250 /* A scratch copy of path, filled lazily if we need it: */
251 struct strbuf path_copy = STRBUF_INIT;
253 int ret, save_errno;
255 /* Sanity check: */
256 assert(*path);
258 retry_fn:
259 ret = fn(path, cb);
260 save_errno = errno;
261 if (!ret)
262 goto out;
264 if (errno == EISDIR && remove_directories_remaining-- > 0) {
266 * A directory is in the way. Maybe it is empty; try
267 * to remove it:
269 if (!path_copy.len)
270 strbuf_addstr(&path_copy, path);
272 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
273 goto retry_fn;
274 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
276 * Maybe the containing directory didn't exist, or
277 * maybe it was just deleted by a process that is
278 * racing with us to clean up empty directories. Try
279 * to create it:
281 enum scld_error scld_result;
283 if (!path_copy.len)
284 strbuf_addstr(&path_copy, path);
286 do {
287 scld_result = safe_create_leading_directories(path_copy.buf);
288 if (scld_result == SCLD_OK)
289 goto retry_fn;
290 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
293 out:
294 strbuf_release(&path_copy);
295 errno = save_errno;
296 return ret;
299 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
301 int i;
302 for (i = 0; i < 20; i++) {
303 static char hex[] = "0123456789abcdef";
304 unsigned int val = sha1[i];
305 strbuf_addch(buf, hex[val >> 4]);
306 strbuf_addch(buf, hex[val & 0xf]);
307 if (!i)
308 strbuf_addch(buf, '/');
312 const char *sha1_file_name(const unsigned char *sha1)
314 static struct strbuf buf = STRBUF_INIT;
316 strbuf_reset(&buf);
317 strbuf_addf(&buf, "%s/", get_object_directory());
319 fill_sha1_path(&buf, sha1);
320 return buf.buf;
323 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
325 strbuf_setlen(&alt->scratch, alt->base_len);
326 return &alt->scratch;
329 static const char *alt_sha1_path(struct alternate_object_database *alt,
330 const unsigned char *sha1)
332 struct strbuf *buf = alt_scratch_buf(alt);
333 fill_sha1_path(buf, sha1);
334 return buf->buf;
337 struct alternate_object_database *alt_odb_list;
338 static struct alternate_object_database **alt_odb_tail;
341 * Return non-zero iff the path is usable as an alternate object database.
343 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
345 struct alternate_object_database *alt;
347 /* Detect cases where alternate disappeared */
348 if (!is_directory(path->buf)) {
349 error("object directory %s does not exist; "
350 "check .git/objects/info/alternates.",
351 path->buf);
352 return 0;
356 * Prevent the common mistake of listing the same
357 * thing twice, or object directory itself.
359 for (alt = alt_odb_list; alt; alt = alt->next) {
360 if (!fspathcmp(path->buf, alt->path))
361 return 0;
363 if (!fspathcmp(path->buf, normalized_objdir))
364 return 0;
366 return 1;
370 * Prepare alternate object database registry.
372 * The variable alt_odb_list points at the list of struct
373 * alternate_object_database. The elements on this list come from
374 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
375 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
376 * whose contents is similar to that environment variable but can be
377 * LF separated. Its base points at a statically allocated buffer that
378 * contains "/the/directory/corresponding/to/.git/objects/...", while
379 * its name points just after the slash at the end of ".git/objects/"
380 * in the example above, and has enough space to hold 40-byte hex
381 * SHA1, an extra slash for the first level indirection, and the
382 * terminating NUL.
384 static void read_info_alternates(const char * relative_base, int depth);
385 static int link_alt_odb_entry(const char *entry, const char *relative_base,
386 int depth, const char *normalized_objdir)
388 struct alternate_object_database *ent;
389 struct strbuf pathbuf = STRBUF_INIT;
391 if (!is_absolute_path(entry) && relative_base) {
392 strbuf_realpath(&pathbuf, relative_base, 1);
393 strbuf_addch(&pathbuf, '/');
395 strbuf_addstr(&pathbuf, entry);
397 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
398 error("unable to normalize alternate object path: %s",
399 pathbuf.buf);
400 strbuf_release(&pathbuf);
401 return -1;
405 * The trailing slash after the directory name is given by
406 * this function at the end. Remove duplicates.
408 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
409 strbuf_setlen(&pathbuf, pathbuf.len - 1);
411 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
412 strbuf_release(&pathbuf);
413 return -1;
416 ent = alloc_alt_odb(pathbuf.buf);
418 /* add the alternate entry */
419 *alt_odb_tail = ent;
420 alt_odb_tail = &(ent->next);
421 ent->next = NULL;
423 /* recursively add alternates */
424 read_info_alternates(pathbuf.buf, depth + 1);
426 strbuf_release(&pathbuf);
427 return 0;
430 static const char *parse_alt_odb_entry(const char *string,
431 int sep,
432 struct strbuf *out)
434 const char *end;
436 strbuf_reset(out);
438 if (*string == '#') {
439 /* comment; consume up to next separator */
440 end = strchrnul(string, sep);
441 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
443 * quoted path; unquote_c_style has copied the
444 * data for us and set "end". Broken quoting (e.g.,
445 * an entry that doesn't end with a quote) falls
446 * back to the unquoted case below.
448 } else {
449 /* normal, unquoted path */
450 end = strchrnul(string, sep);
451 strbuf_add(out, string, end - string);
454 if (*end)
455 end++;
456 return end;
459 static void link_alt_odb_entries(const char *alt, int sep,
460 const char *relative_base, int depth)
462 struct strbuf objdirbuf = STRBUF_INIT;
463 struct strbuf entry = STRBUF_INIT;
465 if (depth > 5) {
466 error("%s: ignoring alternate object stores, nesting too deep.",
467 relative_base);
468 return;
471 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
472 if (strbuf_normalize_path(&objdirbuf) < 0)
473 die("unable to normalize object directory: %s",
474 objdirbuf.buf);
476 while (*alt) {
477 alt = parse_alt_odb_entry(alt, sep, &entry);
478 if (!entry.len)
479 continue;
480 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
482 strbuf_release(&entry);
483 strbuf_release(&objdirbuf);
486 static void read_info_alternates(const char * relative_base, int depth)
488 char *path;
489 struct strbuf buf = STRBUF_INIT;
491 path = xstrfmt("%s/info/alternates", relative_base);
492 if (strbuf_read_file(&buf, path, 1024) < 0) {
493 warn_on_fopen_errors(path);
494 free(path);
495 return;
498 link_alt_odb_entries(buf.buf, '\n', relative_base, depth);
499 strbuf_release(&buf);
500 free(path);
503 struct alternate_object_database *alloc_alt_odb(const char *dir)
505 struct alternate_object_database *ent;
507 FLEX_ALLOC_STR(ent, path, dir);
508 strbuf_init(&ent->scratch, 0);
509 strbuf_addf(&ent->scratch, "%s/", dir);
510 ent->base_len = ent->scratch.len;
512 return ent;
515 void add_to_alternates_file(const char *reference)
517 struct lock_file lock = LOCK_INIT;
518 char *alts = git_pathdup("objects/info/alternates");
519 FILE *in, *out;
520 int found = 0;
522 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
523 out = fdopen_lock_file(&lock, "w");
524 if (!out)
525 die_errno("unable to fdopen alternates lockfile");
527 in = fopen(alts, "r");
528 if (in) {
529 struct strbuf line = STRBUF_INIT;
531 while (strbuf_getline(&line, in) != EOF) {
532 if (!strcmp(reference, line.buf)) {
533 found = 1;
534 break;
536 fprintf_or_die(out, "%s\n", line.buf);
539 strbuf_release(&line);
540 fclose(in);
542 else if (errno != ENOENT)
543 die_errno("unable to read alternates file");
545 if (found) {
546 rollback_lock_file(&lock);
547 } else {
548 fprintf_or_die(out, "%s\n", reference);
549 if (commit_lock_file(&lock))
550 die_errno("unable to move new alternates file into place");
551 if (alt_odb_tail)
552 link_alt_odb_entries(reference, '\n', NULL, 0);
554 free(alts);
557 void add_to_alternates_memory(const char *reference)
560 * Make sure alternates are initialized, or else our entry may be
561 * overwritten when they are.
563 prepare_alt_odb();
565 link_alt_odb_entries(reference, '\n', NULL, 0);
569 * Compute the exact path an alternate is at and returns it. In case of
570 * error NULL is returned and the human readable error is added to `err`
571 * `path` may be relative and should point to $GITDIR.
572 * `err` must not be null.
574 char *compute_alternate_path(const char *path, struct strbuf *err)
576 char *ref_git = NULL;
577 const char *repo, *ref_git_s;
578 int seen_error = 0;
580 ref_git_s = real_path_if_valid(path);
581 if (!ref_git_s) {
582 seen_error = 1;
583 strbuf_addf(err, _("path '%s' does not exist"), path);
584 goto out;
585 } else
587 * Beware: read_gitfile(), real_path() and mkpath()
588 * return static buffer
590 ref_git = xstrdup(ref_git_s);
592 repo = read_gitfile(ref_git);
593 if (!repo)
594 repo = read_gitfile(mkpath("%s/.git", ref_git));
595 if (repo) {
596 free(ref_git);
597 ref_git = xstrdup(repo);
600 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
601 char *ref_git_git = mkpathdup("%s/.git", ref_git);
602 free(ref_git);
603 ref_git = ref_git_git;
604 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
605 struct strbuf sb = STRBUF_INIT;
606 seen_error = 1;
607 if (get_common_dir(&sb, ref_git)) {
608 strbuf_addf(err,
609 _("reference repository '%s' as a linked "
610 "checkout is not supported yet."),
611 path);
612 goto out;
615 strbuf_addf(err, _("reference repository '%s' is not a "
616 "local repository."), path);
617 goto out;
620 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
621 strbuf_addf(err, _("reference repository '%s' is shallow"),
622 path);
623 seen_error = 1;
624 goto out;
627 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
628 strbuf_addf(err,
629 _("reference repository '%s' is grafted"),
630 path);
631 seen_error = 1;
632 goto out;
635 out:
636 if (seen_error) {
637 FREE_AND_NULL(ref_git);
640 return ref_git;
643 int foreach_alt_odb(alt_odb_fn fn, void *cb)
645 struct alternate_object_database *ent;
646 int r = 0;
648 prepare_alt_odb();
649 for (ent = alt_odb_list; ent; ent = ent->next) {
650 r = fn(ent, cb);
651 if (r)
652 break;
654 return r;
657 void prepare_alt_odb(void)
659 const char *alt;
661 if (alt_odb_tail)
662 return;
664 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
665 if (!alt) alt = "";
667 alt_odb_tail = &alt_odb_list;
668 link_alt_odb_entries(alt, PATH_SEP, NULL, 0);
670 read_info_alternates(get_object_directory(), 0);
673 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
674 static int freshen_file(const char *fn)
676 struct utimbuf t;
677 t.actime = t.modtime = time(NULL);
678 return !utime(fn, &t);
682 * All of the check_and_freshen functions return 1 if the file exists and was
683 * freshened (if freshening was requested), 0 otherwise. If they return
684 * 0, you should not assume that it is safe to skip a write of the object (it
685 * either does not exist on disk, or has a stale mtime and may be subject to
686 * pruning).
688 int check_and_freshen_file(const char *fn, int freshen)
690 if (access(fn, F_OK))
691 return 0;
692 if (freshen && !freshen_file(fn))
693 return 0;
694 return 1;
697 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
699 return check_and_freshen_file(sha1_file_name(sha1), freshen);
702 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
704 struct alternate_object_database *alt;
705 prepare_alt_odb();
706 for (alt = alt_odb_list; alt; alt = alt->next) {
707 const char *path = alt_sha1_path(alt, sha1);
708 if (check_and_freshen_file(path, freshen))
709 return 1;
711 return 0;
714 static int check_and_freshen(const unsigned char *sha1, int freshen)
716 return check_and_freshen_local(sha1, freshen) ||
717 check_and_freshen_nonlocal(sha1, freshen);
720 int has_loose_object_nonlocal(const unsigned char *sha1)
722 return check_and_freshen_nonlocal(sha1, 0);
725 static int has_loose_object(const unsigned char *sha1)
727 return check_and_freshen(sha1, 0);
730 static void mmap_limit_check(size_t length)
732 static size_t limit = 0;
733 if (!limit) {
734 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
735 if (!limit)
736 limit = SIZE_MAX;
738 if (length > limit)
739 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
740 (uintmax_t)length, (uintmax_t)limit);
743 void *xmmap_gently(void *start, size_t length,
744 int prot, int flags, int fd, off_t offset)
746 void *ret;
748 mmap_limit_check(length);
749 ret = mmap(start, length, prot, flags, fd, offset);
750 if (ret == MAP_FAILED) {
751 if (!length)
752 return NULL;
753 release_pack_memory(length);
754 ret = mmap(start, length, prot, flags, fd, offset);
756 return ret;
759 void *xmmap(void *start, size_t length,
760 int prot, int flags, int fd, off_t offset)
762 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
763 if (ret == MAP_FAILED)
764 die_errno("mmap failed");
765 return ret;
769 * With an in-core object data in "map", rehash it to make sure the
770 * object name actually matches "sha1" to detect object corruption.
771 * With "map" == NULL, try reading the object named with "sha1" using
772 * the streaming interface and rehash it to do the same.
774 int check_sha1_signature(const unsigned char *sha1, void *map,
775 unsigned long size, const char *type)
777 unsigned char real_sha1[20];
778 enum object_type obj_type;
779 struct git_istream *st;
780 git_SHA_CTX c;
781 char hdr[32];
782 int hdrlen;
784 if (map) {
785 hash_sha1_file(map, size, type, real_sha1);
786 return hashcmp(sha1, real_sha1) ? -1 : 0;
789 st = open_istream(sha1, &obj_type, &size, NULL);
790 if (!st)
791 return -1;
793 /* Generate the header */
794 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
796 /* Sha1.. */
797 git_SHA1_Init(&c);
798 git_SHA1_Update(&c, hdr, hdrlen);
799 for (;;) {
800 char buf[1024 * 16];
801 ssize_t readlen = read_istream(st, buf, sizeof(buf));
803 if (readlen < 0) {
804 close_istream(st);
805 return -1;
807 if (!readlen)
808 break;
809 git_SHA1_Update(&c, buf, readlen);
811 git_SHA1_Final(real_sha1, &c);
812 close_istream(st);
813 return hashcmp(sha1, real_sha1) ? -1 : 0;
816 int git_open_cloexec(const char *name, int flags)
818 int fd;
819 static int o_cloexec = O_CLOEXEC;
821 fd = open(name, flags | o_cloexec);
822 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
823 /* Try again w/o O_CLOEXEC: the kernel might not support it */
824 o_cloexec &= ~O_CLOEXEC;
825 fd = open(name, flags | o_cloexec);
828 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
830 static int fd_cloexec = FD_CLOEXEC;
832 if (!o_cloexec && 0 <= fd && fd_cloexec) {
833 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
834 int flags = fcntl(fd, F_GETFD);
835 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
836 fd_cloexec = 0;
839 #endif
840 return fd;
844 * Find "sha1" as a loose object in the local repository or in an alternate.
845 * Returns 0 on success, negative on failure.
847 * The "path" out-parameter will give the path of the object we found (if any).
848 * Note that it may point to static storage and is only valid until another
849 * call to sha1_file_name(), etc.
851 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
852 const char **path)
854 struct alternate_object_database *alt;
856 *path = sha1_file_name(sha1);
857 if (!lstat(*path, st))
858 return 0;
860 prepare_alt_odb();
861 errno = ENOENT;
862 for (alt = alt_odb_list; alt; alt = alt->next) {
863 *path = alt_sha1_path(alt, sha1);
864 if (!lstat(*path, st))
865 return 0;
868 return -1;
872 * Like stat_sha1_file(), but actually open the object and return the
873 * descriptor. See the caveats on the "path" parameter above.
875 static int open_sha1_file(const unsigned char *sha1, const char **path)
877 int fd;
878 struct alternate_object_database *alt;
879 int most_interesting_errno;
881 *path = sha1_file_name(sha1);
882 fd = git_open(*path);
883 if (fd >= 0)
884 return fd;
885 most_interesting_errno = errno;
887 prepare_alt_odb();
888 for (alt = alt_odb_list; alt; alt = alt->next) {
889 *path = alt_sha1_path(alt, sha1);
890 fd = git_open(*path);
891 if (fd >= 0)
892 return fd;
893 if (most_interesting_errno == ENOENT)
894 most_interesting_errno = errno;
896 errno = most_interesting_errno;
897 return -1;
901 * Map the loose object at "path" if it is not NULL, or the path found by
902 * searching for a loose object named "sha1".
904 static void *map_sha1_file_1(const char *path,
905 const unsigned char *sha1,
906 unsigned long *size)
908 void *map;
909 int fd;
911 if (path)
912 fd = git_open(path);
913 else
914 fd = open_sha1_file(sha1, &path);
915 map = NULL;
916 if (fd >= 0) {
917 struct stat st;
919 if (!fstat(fd, &st)) {
920 *size = xsize_t(st.st_size);
921 if (!*size) {
922 /* mmap() is forbidden on empty files */
923 error("object file %s is empty", path);
924 return NULL;
926 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
928 close(fd);
930 return map;
933 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
935 return map_sha1_file_1(NULL, sha1, size);
938 static int unpack_sha1_short_header(git_zstream *stream,
939 unsigned char *map, unsigned long mapsize,
940 void *buffer, unsigned long bufsiz)
942 /* Get the data stream */
943 memset(stream, 0, sizeof(*stream));
944 stream->next_in = map;
945 stream->avail_in = mapsize;
946 stream->next_out = buffer;
947 stream->avail_out = bufsiz;
949 git_inflate_init(stream);
950 return git_inflate(stream, 0);
953 int unpack_sha1_header(git_zstream *stream,
954 unsigned char *map, unsigned long mapsize,
955 void *buffer, unsigned long bufsiz)
957 int status = unpack_sha1_short_header(stream, map, mapsize,
958 buffer, bufsiz);
960 if (status < Z_OK)
961 return status;
963 /* Make sure we have the terminating NUL */
964 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
965 return -1;
966 return 0;
969 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
970 unsigned long mapsize, void *buffer,
971 unsigned long bufsiz, struct strbuf *header)
973 int status;
975 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
976 if (status < Z_OK)
977 return -1;
980 * Check if entire header is unpacked in the first iteration.
982 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
983 return 0;
986 * buffer[0..bufsiz] was not large enough. Copy the partial
987 * result out to header, and then append the result of further
988 * reading the stream.
990 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
991 stream->next_out = buffer;
992 stream->avail_out = bufsiz;
994 do {
995 status = git_inflate(stream, 0);
996 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
997 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
998 return 0;
999 stream->next_out = buffer;
1000 stream->avail_out = bufsiz;
1001 } while (status != Z_STREAM_END);
1002 return -1;
1005 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1007 int bytes = strlen(buffer) + 1;
1008 unsigned char *buf = xmallocz(size);
1009 unsigned long n;
1010 int status = Z_OK;
1012 n = stream->total_out - bytes;
1013 if (n > size)
1014 n = size;
1015 memcpy(buf, (char *) buffer + bytes, n);
1016 bytes = n;
1017 if (bytes <= size) {
1019 * The above condition must be (bytes <= size), not
1020 * (bytes < size). In other words, even though we
1021 * expect no more output and set avail_out to zero,
1022 * the input zlib stream may have bytes that express
1023 * "this concludes the stream", and we *do* want to
1024 * eat that input.
1026 * Otherwise we would not be able to test that we
1027 * consumed all the input to reach the expected size;
1028 * we also want to check that zlib tells us that all
1029 * went well with status == Z_STREAM_END at the end.
1031 stream->next_out = buf + bytes;
1032 stream->avail_out = size - bytes;
1033 while (status == Z_OK)
1034 status = git_inflate(stream, Z_FINISH);
1036 if (status == Z_STREAM_END && !stream->avail_in) {
1037 git_inflate_end(stream);
1038 return buf;
1041 if (status < 0)
1042 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1043 else if (stream->avail_in)
1044 error("garbage at end of loose object '%s'",
1045 sha1_to_hex(sha1));
1046 free(buf);
1047 return NULL;
1051 * We used to just use "sscanf()", but that's actually way
1052 * too permissive for what we want to check. So do an anal
1053 * object header parse by hand.
1055 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1056 unsigned int flags)
1058 const char *type_buf = hdr;
1059 unsigned long size;
1060 int type, type_len = 0;
1063 * The type can be of any size but is followed by
1064 * a space.
1066 for (;;) {
1067 char c = *hdr++;
1068 if (!c)
1069 return -1;
1070 if (c == ' ')
1071 break;
1072 type_len++;
1075 type = type_from_string_gently(type_buf, type_len, 1);
1076 if (oi->typename)
1077 strbuf_add(oi->typename, type_buf, type_len);
1079 * Set type to 0 if its an unknown object and
1080 * we're obtaining the type using '--allow-unknown-type'
1081 * option.
1083 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1084 type = 0;
1085 else if (type < 0)
1086 die("invalid object type");
1087 if (oi->typep)
1088 *oi->typep = type;
1091 * The length must follow immediately, and be in canonical
1092 * decimal format (ie "010" is not valid).
1094 size = *hdr++ - '0';
1095 if (size > 9)
1096 return -1;
1097 if (size) {
1098 for (;;) {
1099 unsigned long c = *hdr - '0';
1100 if (c > 9)
1101 break;
1102 hdr++;
1103 size = size * 10 + c;
1107 if (oi->sizep)
1108 *oi->sizep = size;
1111 * The length must be followed by a zero byte
1113 return *hdr ? -1 : type;
1116 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1118 struct object_info oi = OBJECT_INFO_INIT;
1120 oi.sizep = sizep;
1121 return parse_sha1_header_extended(hdr, &oi, 0);
1124 static int sha1_loose_object_info(const unsigned char *sha1,
1125 struct object_info *oi,
1126 int flags)
1128 int status = 0;
1129 unsigned long mapsize;
1130 void *map;
1131 git_zstream stream;
1132 char hdr[32];
1133 struct strbuf hdrbuf = STRBUF_INIT;
1134 unsigned long size_scratch;
1136 if (oi->delta_base_sha1)
1137 hashclr(oi->delta_base_sha1);
1140 * If we don't care about type or size, then we don't
1141 * need to look inside the object at all. Note that we
1142 * do not optimize out the stat call, even if the
1143 * caller doesn't care about the disk-size, since our
1144 * return value implicitly indicates whether the
1145 * object even exists.
1147 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1148 const char *path;
1149 struct stat st;
1150 if (stat_sha1_file(sha1, &st, &path) < 0)
1151 return -1;
1152 if (oi->disk_sizep)
1153 *oi->disk_sizep = st.st_size;
1154 return 0;
1157 map = map_sha1_file(sha1, &mapsize);
1158 if (!map)
1159 return -1;
1161 if (!oi->sizep)
1162 oi->sizep = &size_scratch;
1164 if (oi->disk_sizep)
1165 *oi->disk_sizep = mapsize;
1166 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1167 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1168 status = error("unable to unpack %s header with --allow-unknown-type",
1169 sha1_to_hex(sha1));
1170 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1171 status = error("unable to unpack %s header",
1172 sha1_to_hex(sha1));
1173 if (status < 0)
1174 ; /* Do nothing */
1175 else if (hdrbuf.len) {
1176 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1177 status = error("unable to parse %s header with --allow-unknown-type",
1178 sha1_to_hex(sha1));
1179 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1180 status = error("unable to parse %s header", sha1_to_hex(sha1));
1182 if (status >= 0 && oi->contentp) {
1183 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1184 *oi->sizep, sha1);
1185 if (!*oi->contentp) {
1186 git_inflate_end(&stream);
1187 status = -1;
1189 } else
1190 git_inflate_end(&stream);
1192 munmap(map, mapsize);
1193 if (status && oi->typep)
1194 *oi->typep = status;
1195 if (oi->sizep == &size_scratch)
1196 oi->sizep = NULL;
1197 strbuf_release(&hdrbuf);
1198 oi->whence = OI_LOOSE;
1199 return (status < 0) ? status : 0;
1202 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1204 static struct object_info blank_oi = OBJECT_INFO_INIT;
1205 struct pack_entry e;
1206 int rtype;
1207 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1208 lookup_replace_object(sha1) :
1209 sha1;
1211 if (!oi)
1212 oi = &blank_oi;
1214 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1215 struct cached_object *co = find_cached_object(real);
1216 if (co) {
1217 if (oi->typep)
1218 *(oi->typep) = co->type;
1219 if (oi->sizep)
1220 *(oi->sizep) = co->size;
1221 if (oi->disk_sizep)
1222 *(oi->disk_sizep) = 0;
1223 if (oi->delta_base_sha1)
1224 hashclr(oi->delta_base_sha1);
1225 if (oi->typename)
1226 strbuf_addstr(oi->typename, typename(co->type));
1227 if (oi->contentp)
1228 *oi->contentp = xmemdupz(co->buf, co->size);
1229 oi->whence = OI_CACHED;
1230 return 0;
1234 if (!find_pack_entry(real, &e)) {
1235 /* Most likely it's a loose object. */
1236 if (!sha1_loose_object_info(real, oi, flags))
1237 return 0;
1239 /* Not a loose object; someone else may have just packed it. */
1240 if (flags & OBJECT_INFO_QUICK) {
1241 return -1;
1242 } else {
1243 reprepare_packed_git();
1244 if (!find_pack_entry(real, &e))
1245 return -1;
1249 if (oi == &blank_oi)
1251 * We know that the caller doesn't actually need the
1252 * information below, so return early.
1254 return 0;
1256 rtype = packed_object_info(e.p, e.offset, oi);
1257 if (rtype < 0) {
1258 mark_bad_packed_object(e.p, real);
1259 return sha1_object_info_extended(real, oi, 0);
1260 } else if (oi->whence == OI_PACKED) {
1261 oi->u.packed.offset = e.offset;
1262 oi->u.packed.pack = e.p;
1263 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1264 rtype == OBJ_OFS_DELTA);
1267 return 0;
1270 /* returns enum object_type or negative */
1271 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1273 enum object_type type;
1274 struct object_info oi = OBJECT_INFO_INIT;
1276 oi.typep = &type;
1277 oi.sizep = sizep;
1278 if (sha1_object_info_extended(sha1, &oi,
1279 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1280 return -1;
1281 return type;
1284 static void *read_object(const unsigned char *sha1, enum object_type *type,
1285 unsigned long *size)
1287 struct object_info oi = OBJECT_INFO_INIT;
1288 void *content;
1289 oi.typep = type;
1290 oi.sizep = size;
1291 oi.contentp = &content;
1293 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1294 return NULL;
1295 return content;
1298 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1299 unsigned char *sha1)
1301 struct cached_object *co;
1303 hash_sha1_file(buf, len, typename(type), sha1);
1304 if (has_sha1_file(sha1) || find_cached_object(sha1))
1305 return 0;
1306 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1307 co = &cached_objects[cached_object_nr++];
1308 co->size = len;
1309 co->type = type;
1310 co->buf = xmalloc(len);
1311 memcpy(co->buf, buf, len);
1312 hashcpy(co->sha1, sha1);
1313 return 0;
1317 * This function dies on corrupt objects; the callers who want to
1318 * deal with them should arrange to call read_object() and give error
1319 * messages themselves.
1321 void *read_sha1_file_extended(const unsigned char *sha1,
1322 enum object_type *type,
1323 unsigned long *size,
1324 int lookup_replace)
1326 void *data;
1327 const struct packed_git *p;
1328 const char *path;
1329 struct stat st;
1330 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1331 : sha1;
1333 errno = 0;
1334 data = read_object(repl, type, size);
1335 if (data)
1336 return data;
1338 if (errno && errno != ENOENT)
1339 die_errno("failed to read object %s", sha1_to_hex(sha1));
1341 /* die if we replaced an object with one that does not exist */
1342 if (repl != sha1)
1343 die("replacement %s not found for %s",
1344 sha1_to_hex(repl), sha1_to_hex(sha1));
1346 if (!stat_sha1_file(repl, &st, &path))
1347 die("loose object %s (stored in %s) is corrupt",
1348 sha1_to_hex(repl), path);
1350 if ((p = has_packed_and_bad(repl)) != NULL)
1351 die("packed object %s (stored in %s) is corrupt",
1352 sha1_to_hex(repl), p->pack_name);
1354 return NULL;
1357 void *read_object_with_reference(const unsigned char *sha1,
1358 const char *required_type_name,
1359 unsigned long *size,
1360 unsigned char *actual_sha1_return)
1362 enum object_type type, required_type;
1363 void *buffer;
1364 unsigned long isize;
1365 unsigned char actual_sha1[20];
1367 required_type = type_from_string(required_type_name);
1368 hashcpy(actual_sha1, sha1);
1369 while (1) {
1370 int ref_length = -1;
1371 const char *ref_type = NULL;
1373 buffer = read_sha1_file(actual_sha1, &type, &isize);
1374 if (!buffer)
1375 return NULL;
1376 if (type == required_type) {
1377 *size = isize;
1378 if (actual_sha1_return)
1379 hashcpy(actual_sha1_return, actual_sha1);
1380 return buffer;
1382 /* Handle references */
1383 else if (type == OBJ_COMMIT)
1384 ref_type = "tree ";
1385 else if (type == OBJ_TAG)
1386 ref_type = "object ";
1387 else {
1388 free(buffer);
1389 return NULL;
1391 ref_length = strlen(ref_type);
1393 if (ref_length + 40 > isize ||
1394 memcmp(buffer, ref_type, ref_length) ||
1395 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1396 free(buffer);
1397 return NULL;
1399 free(buffer);
1400 /* Now we have the ID of the referred-to object in
1401 * actual_sha1. Check again. */
1405 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1406 const char *type, unsigned char *sha1,
1407 char *hdr, int *hdrlen)
1409 git_SHA_CTX c;
1411 /* Generate the header */
1412 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1414 /* Sha1.. */
1415 git_SHA1_Init(&c);
1416 git_SHA1_Update(&c, hdr, *hdrlen);
1417 git_SHA1_Update(&c, buf, len);
1418 git_SHA1_Final(sha1, &c);
1422 * Move the just written object into its final resting place.
1424 int finalize_object_file(const char *tmpfile, const char *filename)
1426 int ret = 0;
1428 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1429 goto try_rename;
1430 else if (link(tmpfile, filename))
1431 ret = errno;
1434 * Coda hack - coda doesn't like cross-directory links,
1435 * so we fall back to a rename, which will mean that it
1436 * won't be able to check collisions, but that's not a
1437 * big deal.
1439 * The same holds for FAT formatted media.
1441 * When this succeeds, we just return. We have nothing
1442 * left to unlink.
1444 if (ret && ret != EEXIST) {
1445 try_rename:
1446 if (!rename(tmpfile, filename))
1447 goto out;
1448 ret = errno;
1450 unlink_or_warn(tmpfile);
1451 if (ret) {
1452 if (ret != EEXIST) {
1453 return error_errno("unable to write sha1 filename %s", filename);
1455 /* FIXME!!! Collision check here ? */
1458 out:
1459 if (adjust_shared_perm(filename))
1460 return error("unable to set permission to '%s'", filename);
1461 return 0;
1464 static int write_buffer(int fd, const void *buf, size_t len)
1466 if (write_in_full(fd, buf, len) < 0)
1467 return error_errno("file write error");
1468 return 0;
1471 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1472 unsigned char *sha1)
1474 char hdr[32];
1475 int hdrlen = sizeof(hdr);
1476 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1477 return 0;
1480 /* Finalize a file on disk, and close it. */
1481 static void close_sha1_file(int fd)
1483 if (fsync_object_files)
1484 fsync_or_die(fd, "sha1 file");
1485 if (close(fd) != 0)
1486 die_errno("error when closing sha1 file");
1489 /* Size of directory component, including the ending '/' */
1490 static inline int directory_size(const char *filename)
1492 const char *s = strrchr(filename, '/');
1493 if (!s)
1494 return 0;
1495 return s - filename + 1;
1499 * This creates a temporary file in the same directory as the final
1500 * 'filename'
1502 * We want to avoid cross-directory filename renames, because those
1503 * can have problems on various filesystems (FAT, NFS, Coda).
1505 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1507 int fd, dirlen = directory_size(filename);
1509 strbuf_reset(tmp);
1510 strbuf_add(tmp, filename, dirlen);
1511 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1512 fd = git_mkstemp_mode(tmp->buf, 0444);
1513 if (fd < 0 && dirlen && errno == ENOENT) {
1515 * Make sure the directory exists; note that the contents
1516 * of the buffer are undefined after mkstemp returns an
1517 * error, so we have to rewrite the whole buffer from
1518 * scratch.
1520 strbuf_reset(tmp);
1521 strbuf_add(tmp, filename, dirlen - 1);
1522 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1523 return -1;
1524 if (adjust_shared_perm(tmp->buf))
1525 return -1;
1527 /* Try again */
1528 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1529 fd = git_mkstemp_mode(tmp->buf, 0444);
1531 return fd;
1534 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1535 const void *buf, unsigned long len, time_t mtime)
1537 int fd, ret;
1538 unsigned char compressed[4096];
1539 git_zstream stream;
1540 git_SHA_CTX c;
1541 unsigned char parano_sha1[20];
1542 static struct strbuf tmp_file = STRBUF_INIT;
1543 const char *filename = sha1_file_name(sha1);
1545 fd = create_tmpfile(&tmp_file, filename);
1546 if (fd < 0) {
1547 if (errno == EACCES)
1548 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1549 else
1550 return error_errno("unable to create temporary file");
1553 /* Set it up */
1554 git_deflate_init(&stream, zlib_compression_level);
1555 stream.next_out = compressed;
1556 stream.avail_out = sizeof(compressed);
1557 git_SHA1_Init(&c);
1559 /* First header.. */
1560 stream.next_in = (unsigned char *)hdr;
1561 stream.avail_in = hdrlen;
1562 while (git_deflate(&stream, 0) == Z_OK)
1563 ; /* nothing */
1564 git_SHA1_Update(&c, hdr, hdrlen);
1566 /* Then the data itself.. */
1567 stream.next_in = (void *)buf;
1568 stream.avail_in = len;
1569 do {
1570 unsigned char *in0 = stream.next_in;
1571 ret = git_deflate(&stream, Z_FINISH);
1572 git_SHA1_Update(&c, in0, stream.next_in - in0);
1573 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1574 die("unable to write sha1 file");
1575 stream.next_out = compressed;
1576 stream.avail_out = sizeof(compressed);
1577 } while (ret == Z_OK);
1579 if (ret != Z_STREAM_END)
1580 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1581 ret = git_deflate_end_gently(&stream);
1582 if (ret != Z_OK)
1583 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1584 git_SHA1_Final(parano_sha1, &c);
1585 if (hashcmp(sha1, parano_sha1) != 0)
1586 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1588 close_sha1_file(fd);
1590 if (mtime) {
1591 struct utimbuf utb;
1592 utb.actime = mtime;
1593 utb.modtime = mtime;
1594 if (utime(tmp_file.buf, &utb) < 0)
1595 warning_errno("failed utime() on %s", tmp_file.buf);
1598 return finalize_object_file(tmp_file.buf, filename);
1601 static int freshen_loose_object(const unsigned char *sha1)
1603 return check_and_freshen(sha1, 1);
1606 static int freshen_packed_object(const unsigned char *sha1)
1608 struct pack_entry e;
1609 if (!find_pack_entry(sha1, &e))
1610 return 0;
1611 if (e.p->freshened)
1612 return 1;
1613 if (!freshen_file(e.p->pack_name))
1614 return 0;
1615 e.p->freshened = 1;
1616 return 1;
1619 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1621 char hdr[32];
1622 int hdrlen = sizeof(hdr);
1624 /* Normally if we have it in the pack then we do not bother writing
1625 * it out into .git/objects/??/?{38} file.
1627 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1628 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1629 return 0;
1630 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1633 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1634 struct object_id *oid, unsigned flags)
1636 char *header;
1637 int hdrlen, status = 0;
1639 /* type string, SP, %lu of the length plus NUL must fit this */
1640 hdrlen = strlen(type) + 32;
1641 header = xmalloc(hdrlen);
1642 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1644 if (!(flags & HASH_WRITE_OBJECT))
1645 goto cleanup;
1646 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1647 goto cleanup;
1648 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1650 cleanup:
1651 free(header);
1652 return status;
1655 int force_object_loose(const unsigned char *sha1, time_t mtime)
1657 void *buf;
1658 unsigned long len;
1659 enum object_type type;
1660 char hdr[32];
1661 int hdrlen;
1662 int ret;
1664 if (has_loose_object(sha1))
1665 return 0;
1666 buf = read_object(sha1, &type, &len);
1667 if (!buf)
1668 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1669 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1670 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1671 free(buf);
1673 return ret;
1676 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1678 if (!startup_info->have_repository)
1679 return 0;
1680 return sha1_object_info_extended(sha1, NULL,
1681 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1684 int has_object_file(const struct object_id *oid)
1686 return has_sha1_file(oid->hash);
1689 int has_object_file_with_flags(const struct object_id *oid, int flags)
1691 return has_sha1_file_with_flags(oid->hash, flags);
1694 static void check_tree(const void *buf, size_t size)
1696 struct tree_desc desc;
1697 struct name_entry entry;
1699 init_tree_desc(&desc, buf, size);
1700 while (tree_entry(&desc, &entry))
1701 /* do nothing
1702 * tree_entry() will die() on malformed entries */
1706 static void check_commit(const void *buf, size_t size)
1708 struct commit c;
1709 memset(&c, 0, sizeof(c));
1710 if (parse_commit_buffer(&c, buf, size))
1711 die("corrupt commit");
1714 static void check_tag(const void *buf, size_t size)
1716 struct tag t;
1717 memset(&t, 0, sizeof(t));
1718 if (parse_tag_buffer(&t, buf, size))
1719 die("corrupt tag");
1722 static int index_mem(struct object_id *oid, void *buf, size_t size,
1723 enum object_type type,
1724 const char *path, unsigned flags)
1726 int ret, re_allocated = 0;
1727 int write_object = flags & HASH_WRITE_OBJECT;
1729 if (!type)
1730 type = OBJ_BLOB;
1733 * Convert blobs to git internal format
1735 if ((type == OBJ_BLOB) && path) {
1736 struct strbuf nbuf = STRBUF_INIT;
1737 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1738 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
1739 buf = strbuf_detach(&nbuf, &size);
1740 re_allocated = 1;
1743 if (flags & HASH_FORMAT_CHECK) {
1744 if (type == OBJ_TREE)
1745 check_tree(buf, size);
1746 if (type == OBJ_COMMIT)
1747 check_commit(buf, size);
1748 if (type == OBJ_TAG)
1749 check_tag(buf, size);
1752 if (write_object)
1753 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1754 else
1755 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1756 if (re_allocated)
1757 free(buf);
1758 return ret;
1761 static int index_stream_convert_blob(struct object_id *oid, int fd,
1762 const char *path, unsigned flags)
1764 int ret;
1765 const int write_object = flags & HASH_WRITE_OBJECT;
1766 struct strbuf sbuf = STRBUF_INIT;
1768 assert(path);
1769 assert(would_convert_to_git_filter_fd(path));
1771 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1772 write_object ? safe_crlf : SAFE_CRLF_FALSE);
1774 if (write_object)
1775 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1776 oid->hash);
1777 else
1778 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1779 oid->hash);
1780 strbuf_release(&sbuf);
1781 return ret;
1784 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1785 const char *path, unsigned flags)
1787 struct strbuf sbuf = STRBUF_INIT;
1788 int ret;
1790 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1791 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1792 else
1793 ret = -1;
1794 strbuf_release(&sbuf);
1795 return ret;
1798 #define SMALL_FILE_SIZE (32*1024)
1800 static int index_core(struct object_id *oid, int fd, size_t size,
1801 enum object_type type, const char *path,
1802 unsigned flags)
1804 int ret;
1806 if (!size) {
1807 ret = index_mem(oid, "", size, type, path, flags);
1808 } else if (size <= SMALL_FILE_SIZE) {
1809 char *buf = xmalloc(size);
1810 ssize_t read_result = read_in_full(fd, buf, size);
1811 if (read_result < 0)
1812 ret = error_errno("read error while indexing %s",
1813 path ? path : "<unknown>");
1814 else if (read_result != size)
1815 ret = error("short read while indexing %s",
1816 path ? path : "<unknown>");
1817 else
1818 ret = index_mem(oid, buf, size, type, path, flags);
1819 free(buf);
1820 } else {
1821 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1822 ret = index_mem(oid, buf, size, type, path, flags);
1823 munmap(buf, size);
1825 return ret;
1829 * This creates one packfile per large blob unless bulk-checkin
1830 * machinery is "plugged".
1832 * This also bypasses the usual "convert-to-git" dance, and that is on
1833 * purpose. We could write a streaming version of the converting
1834 * functions and insert that before feeding the data to fast-import
1835 * (or equivalent in-core API described above). However, that is
1836 * somewhat complicated, as we do not know the size of the filter
1837 * result, which we need to know beforehand when writing a git object.
1838 * Since the primary motivation for trying to stream from the working
1839 * tree file and to avoid mmaping it in core is to deal with large
1840 * binary blobs, they generally do not want to get any conversion, and
1841 * callers should avoid this code path when filters are requested.
1843 static int index_stream(struct object_id *oid, int fd, size_t size,
1844 enum object_type type, const char *path,
1845 unsigned flags)
1847 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1850 int index_fd(struct object_id *oid, int fd, struct stat *st,
1851 enum object_type type, const char *path, unsigned flags)
1853 int ret;
1856 * Call xsize_t() only when needed to avoid potentially unnecessary
1857 * die() for large files.
1859 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1860 ret = index_stream_convert_blob(oid, fd, path, flags);
1861 else if (!S_ISREG(st->st_mode))
1862 ret = index_pipe(oid, fd, type, path, flags);
1863 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1864 (path && would_convert_to_git(&the_index, path)))
1865 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1866 flags);
1867 else
1868 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1869 flags);
1870 close(fd);
1871 return ret;
1874 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1876 int fd;
1877 struct strbuf sb = STRBUF_INIT;
1878 int rc = 0;
1880 switch (st->st_mode & S_IFMT) {
1881 case S_IFREG:
1882 fd = open(path, O_RDONLY);
1883 if (fd < 0)
1884 return error_errno("open(\"%s\")", path);
1885 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1886 return error("%s: failed to insert into database",
1887 path);
1888 break;
1889 case S_IFLNK:
1890 if (strbuf_readlink(&sb, path, st->st_size))
1891 return error_errno("readlink(\"%s\")", path);
1892 if (!(flags & HASH_WRITE_OBJECT))
1893 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1894 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1895 rc = error("%s: failed to insert into database", path);
1896 strbuf_release(&sb);
1897 break;
1898 case S_IFDIR:
1899 return resolve_gitlink_ref(path, "HEAD", oid);
1900 default:
1901 return error("%s: unsupported file type", path);
1903 return rc;
1906 int read_pack_header(int fd, struct pack_header *header)
1908 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1909 /* "eof before pack header was fully read" */
1910 return PH_ERROR_EOF;
1912 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1913 /* "protocol error (pack signature mismatch detected)" */
1914 return PH_ERROR_PACK_SIGNATURE;
1915 if (!pack_version_ok(header->hdr_version))
1916 /* "protocol error (pack version unsupported)" */
1917 return PH_ERROR_PROTOCOL;
1918 return 0;
1921 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1923 enum object_type type = sha1_object_info(sha1, NULL);
1924 if (type < 0)
1925 die("%s is not a valid object", sha1_to_hex(sha1));
1926 if (type != expect)
1927 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1928 typename(expect));
1931 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1932 struct strbuf *path,
1933 each_loose_object_fn obj_cb,
1934 each_loose_cruft_fn cruft_cb,
1935 each_loose_subdir_fn subdir_cb,
1936 void *data)
1938 size_t origlen, baselen;
1939 DIR *dir;
1940 struct dirent *de;
1941 int r = 0;
1942 struct object_id oid;
1944 if (subdir_nr > 0xff)
1945 BUG("invalid loose object subdirectory: %x", subdir_nr);
1947 origlen = path->len;
1948 strbuf_complete(path, '/');
1949 strbuf_addf(path, "%02x", subdir_nr);
1950 baselen = path->len;
1952 dir = opendir(path->buf);
1953 if (!dir) {
1954 if (errno != ENOENT)
1955 r = error_errno("unable to open %s", path->buf);
1956 strbuf_setlen(path, origlen);
1957 return r;
1960 oid.hash[0] = subdir_nr;
1962 while ((de = readdir(dir))) {
1963 if (is_dot_or_dotdot(de->d_name))
1964 continue;
1966 strbuf_setlen(path, baselen);
1967 strbuf_addf(path, "/%s", de->d_name);
1969 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2 &&
1970 !hex_to_bytes(oid.hash + 1, de->d_name,
1971 GIT_SHA1_RAWSZ - 1)) {
1972 if (obj_cb) {
1973 r = obj_cb(&oid, path->buf, data);
1974 if (r)
1975 break;
1977 continue;
1980 if (cruft_cb) {
1981 r = cruft_cb(de->d_name, path->buf, data);
1982 if (r)
1983 break;
1986 closedir(dir);
1988 strbuf_setlen(path, baselen);
1989 if (!r && subdir_cb)
1990 r = subdir_cb(subdir_nr, path->buf, data);
1992 strbuf_setlen(path, origlen);
1994 return r;
1997 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
1998 each_loose_object_fn obj_cb,
1999 each_loose_cruft_fn cruft_cb,
2000 each_loose_subdir_fn subdir_cb,
2001 void *data)
2003 int r = 0;
2004 int i;
2006 for (i = 0; i < 256; i++) {
2007 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2008 subdir_cb, data);
2009 if (r)
2010 break;
2013 return r;
2016 int for_each_loose_file_in_objdir(const char *path,
2017 each_loose_object_fn obj_cb,
2018 each_loose_cruft_fn cruft_cb,
2019 each_loose_subdir_fn subdir_cb,
2020 void *data)
2022 struct strbuf buf = STRBUF_INIT;
2023 int r;
2025 strbuf_addstr(&buf, path);
2026 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2027 subdir_cb, data);
2028 strbuf_release(&buf);
2030 return r;
2033 struct loose_alt_odb_data {
2034 each_loose_object_fn *cb;
2035 void *data;
2038 static int loose_from_alt_odb(struct alternate_object_database *alt,
2039 void *vdata)
2041 struct loose_alt_odb_data *data = vdata;
2042 struct strbuf buf = STRBUF_INIT;
2043 int r;
2045 strbuf_addstr(&buf, alt->path);
2046 r = for_each_loose_file_in_objdir_buf(&buf,
2047 data->cb, NULL, NULL,
2048 data->data);
2049 strbuf_release(&buf);
2050 return r;
2053 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2055 struct loose_alt_odb_data alt;
2056 int r;
2058 r = for_each_loose_file_in_objdir(get_object_directory(),
2059 cb, NULL, NULL, data);
2060 if (r)
2061 return r;
2063 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2064 return 0;
2066 alt.cb = cb;
2067 alt.data = data;
2068 return foreach_alt_odb(loose_from_alt_odb, &alt);
2071 static int check_stream_sha1(git_zstream *stream,
2072 const char *hdr,
2073 unsigned long size,
2074 const char *path,
2075 const unsigned char *expected_sha1)
2077 git_SHA_CTX c;
2078 unsigned char real_sha1[GIT_MAX_RAWSZ];
2079 unsigned char buf[4096];
2080 unsigned long total_read;
2081 int status = Z_OK;
2083 git_SHA1_Init(&c);
2084 git_SHA1_Update(&c, hdr, stream->total_out);
2087 * We already read some bytes into hdr, but the ones up to the NUL
2088 * do not count against the object's content size.
2090 total_read = stream->total_out - strlen(hdr) - 1;
2093 * This size comparison must be "<=" to read the final zlib packets;
2094 * see the comment in unpack_sha1_rest for details.
2096 while (total_read <= size &&
2097 (status == Z_OK || status == Z_BUF_ERROR)) {
2098 stream->next_out = buf;
2099 stream->avail_out = sizeof(buf);
2100 if (size - total_read < stream->avail_out)
2101 stream->avail_out = size - total_read;
2102 status = git_inflate(stream, Z_FINISH);
2103 git_SHA1_Update(&c, buf, stream->next_out - buf);
2104 total_read += stream->next_out - buf;
2106 git_inflate_end(stream);
2108 if (status != Z_STREAM_END) {
2109 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2110 return -1;
2112 if (stream->avail_in) {
2113 error("garbage at end of loose object '%s'",
2114 sha1_to_hex(expected_sha1));
2115 return -1;
2118 git_SHA1_Final(real_sha1, &c);
2119 if (hashcmp(expected_sha1, real_sha1)) {
2120 error("sha1 mismatch for %s (expected %s)", path,
2121 sha1_to_hex(expected_sha1));
2122 return -1;
2125 return 0;
2128 int read_loose_object(const char *path,
2129 const unsigned char *expected_sha1,
2130 enum object_type *type,
2131 unsigned long *size,
2132 void **contents)
2134 int ret = -1;
2135 void *map = NULL;
2136 unsigned long mapsize;
2137 git_zstream stream;
2138 char hdr[32];
2140 *contents = NULL;
2142 map = map_sha1_file_1(path, NULL, &mapsize);
2143 if (!map) {
2144 error_errno("unable to mmap %s", path);
2145 goto out;
2148 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2149 error("unable to unpack header of %s", path);
2150 goto out;
2153 *type = parse_sha1_header(hdr, size);
2154 if (*type < 0) {
2155 error("unable to parse header of %s", path);
2156 git_inflate_end(&stream);
2157 goto out;
2160 if (*type == OBJ_BLOB) {
2161 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2162 goto out;
2163 } else {
2164 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2165 if (!*contents) {
2166 error("unable to unpack contents of %s", path);
2167 git_inflate_end(&stream);
2168 goto out;
2170 if (check_sha1_signature(expected_sha1, *contents,
2171 *size, typename(*type))) {
2172 error("sha1 mismatch for %s (expected %s)", path,
2173 sha1_to_hex(expected_sha1));
2174 free(*contents);
2175 goto out;
2179 ret = 0; /* everything checks out */
2181 out:
2182 if (map)
2183 munmap(map, mapsize);
2184 return ret;