packfile: allow reprepare_packed_git to handle arbitrary repositories
[git.git] / sha1_file.c
blob9c024cd957a3aa39635d0bcc67377dffbf2447ad
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "config.h"
11 #include "string-list.h"
12 #include "lockfile.h"
13 #include "delta.h"
14 #include "pack.h"
15 #include "blob.h"
16 #include "commit.h"
17 #include "run-command.h"
18 #include "tag.h"
19 #include "tree.h"
20 #include "tree-walk.h"
21 #include "refs.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "repository.h"
26 #include "streaming.h"
27 #include "dir.h"
28 #include "list.h"
29 #include "mergesort.h"
30 #include "quote.h"
31 #include "packfile.h"
32 #include "fetch-object.h"
33 #include "object-store.h"
35 const unsigned char null_sha1[GIT_MAX_RAWSZ];
36 const struct object_id null_oid;
37 const struct object_id empty_tree_oid = {
38 EMPTY_TREE_SHA1_BIN_LITERAL
40 const struct object_id empty_blob_oid = {
41 EMPTY_BLOB_SHA1_BIN_LITERAL
44 static void git_hash_sha1_init(void *ctx)
46 git_SHA1_Init((git_SHA_CTX *)ctx);
49 static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
51 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
54 static void git_hash_sha1_final(unsigned char *hash, void *ctx)
56 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
59 static void git_hash_unknown_init(void *ctx)
61 die("trying to init unknown hash");
64 static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
66 die("trying to update unknown hash");
69 static void git_hash_unknown_final(unsigned char *hash, void *ctx)
71 die("trying to finalize unknown hash");
74 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
76 NULL,
77 0x00000000,
81 git_hash_unknown_init,
82 git_hash_unknown_update,
83 git_hash_unknown_final,
84 NULL,
85 NULL,
88 "sha-1",
89 /* "sha1", big-endian */
90 0x73686131,
91 sizeof(git_SHA_CTX),
92 GIT_SHA1_RAWSZ,
93 GIT_SHA1_HEXSZ,
94 git_hash_sha1_init,
95 git_hash_sha1_update,
96 git_hash_sha1_final,
97 &empty_tree_oid,
98 &empty_blob_oid,
103 * This is meant to hold a *small* number of objects that you would
104 * want read_sha1_file() to be able to return, but yet you do not want
105 * to write them into the object store (e.g. a browse-only
106 * application).
108 static struct cached_object {
109 unsigned char sha1[20];
110 enum object_type type;
111 void *buf;
112 unsigned long size;
113 } *cached_objects;
114 static int cached_object_nr, cached_object_alloc;
116 static struct cached_object empty_tree = {
117 EMPTY_TREE_SHA1_BIN_LITERAL,
118 OBJ_TREE,
123 static struct cached_object *find_cached_object(const unsigned char *sha1)
125 int i;
126 struct cached_object *co = cached_objects;
128 for (i = 0; i < cached_object_nr; i++, co++) {
129 if (!hashcmp(co->sha1, sha1))
130 return co;
132 if (!hashcmp(sha1, empty_tree.sha1))
133 return &empty_tree;
134 return NULL;
138 static int get_conv_flags(unsigned flags)
140 if (flags & HASH_RENORMALIZE)
141 return CONV_EOL_RENORMALIZE;
142 else if (flags & HASH_WRITE_OBJECT)
143 return global_conv_flags_eol;
144 else
145 return 0;
149 int mkdir_in_gitdir(const char *path)
151 if (mkdir(path, 0777)) {
152 int saved_errno = errno;
153 struct stat st;
154 struct strbuf sb = STRBUF_INIT;
156 if (errno != EEXIST)
157 return -1;
159 * Are we looking at a path in a symlinked worktree
160 * whose original repository does not yet have it?
161 * e.g. .git/rr-cache pointing at its original
162 * repository in which the user hasn't performed any
163 * conflict resolution yet?
165 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
166 strbuf_readlink(&sb, path, st.st_size) ||
167 !is_absolute_path(sb.buf) ||
168 mkdir(sb.buf, 0777)) {
169 strbuf_release(&sb);
170 errno = saved_errno;
171 return -1;
173 strbuf_release(&sb);
175 return adjust_shared_perm(path);
178 enum scld_error safe_create_leading_directories(char *path)
180 char *next_component = path + offset_1st_component(path);
181 enum scld_error ret = SCLD_OK;
183 while (ret == SCLD_OK && next_component) {
184 struct stat st;
185 char *slash = next_component, slash_character;
187 while (*slash && !is_dir_sep(*slash))
188 slash++;
190 if (!*slash)
191 break;
193 next_component = slash + 1;
194 while (is_dir_sep(*next_component))
195 next_component++;
196 if (!*next_component)
197 break;
199 slash_character = *slash;
200 *slash = '\0';
201 if (!stat(path, &st)) {
202 /* path exists */
203 if (!S_ISDIR(st.st_mode)) {
204 errno = ENOTDIR;
205 ret = SCLD_EXISTS;
207 } else if (mkdir(path, 0777)) {
208 if (errno == EEXIST &&
209 !stat(path, &st) && S_ISDIR(st.st_mode))
210 ; /* somebody created it since we checked */
211 else if (errno == ENOENT)
213 * Either mkdir() failed because
214 * somebody just pruned the containing
215 * directory, or stat() failed because
216 * the file that was in our way was
217 * just removed. Either way, inform
218 * the caller that it might be worth
219 * trying again:
221 ret = SCLD_VANISHED;
222 else
223 ret = SCLD_FAILED;
224 } else if (adjust_shared_perm(path)) {
225 ret = SCLD_PERMS;
227 *slash = slash_character;
229 return ret;
232 enum scld_error safe_create_leading_directories_const(const char *path)
234 int save_errno;
235 /* path points to cache entries, so xstrdup before messing with it */
236 char *buf = xstrdup(path);
237 enum scld_error result = safe_create_leading_directories(buf);
239 save_errno = errno;
240 free(buf);
241 errno = save_errno;
242 return result;
245 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
248 * The number of times we will try to remove empty directories
249 * in the way of path. This is only 1 because if another
250 * process is racily creating directories that conflict with
251 * us, we don't want to fight against them.
253 int remove_directories_remaining = 1;
256 * The number of times that we will try to create the
257 * directories containing path. We are willing to attempt this
258 * more than once, because another process could be trying to
259 * clean up empty directories at the same time as we are
260 * trying to create them.
262 int create_directories_remaining = 3;
264 /* A scratch copy of path, filled lazily if we need it: */
265 struct strbuf path_copy = STRBUF_INIT;
267 int ret, save_errno;
269 /* Sanity check: */
270 assert(*path);
272 retry_fn:
273 ret = fn(path, cb);
274 save_errno = errno;
275 if (!ret)
276 goto out;
278 if (errno == EISDIR && remove_directories_remaining-- > 0) {
280 * A directory is in the way. Maybe it is empty; try
281 * to remove it:
283 if (!path_copy.len)
284 strbuf_addstr(&path_copy, path);
286 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
287 goto retry_fn;
288 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
290 * Maybe the containing directory didn't exist, or
291 * maybe it was just deleted by a process that is
292 * racing with us to clean up empty directories. Try
293 * to create it:
295 enum scld_error scld_result;
297 if (!path_copy.len)
298 strbuf_addstr(&path_copy, path);
300 do {
301 scld_result = safe_create_leading_directories(path_copy.buf);
302 if (scld_result == SCLD_OK)
303 goto retry_fn;
304 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
307 out:
308 strbuf_release(&path_copy);
309 errno = save_errno;
310 return ret;
313 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
315 int i;
316 for (i = 0; i < 20; i++) {
317 static char hex[] = "0123456789abcdef";
318 unsigned int val = sha1[i];
319 strbuf_addch(buf, hex[val >> 4]);
320 strbuf_addch(buf, hex[val & 0xf]);
321 if (!i)
322 strbuf_addch(buf, '/');
326 void sha1_file_name(struct repository *r, struct strbuf *buf, const unsigned char *sha1)
328 strbuf_addstr(buf, r->objects->objectdir);
329 strbuf_addch(buf, '/');
330 fill_sha1_path(buf, sha1);
333 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
335 strbuf_setlen(&alt->scratch, alt->base_len);
336 return &alt->scratch;
339 static const char *alt_sha1_path(struct alternate_object_database *alt,
340 const unsigned char *sha1)
342 struct strbuf *buf = alt_scratch_buf(alt);
343 fill_sha1_path(buf, sha1);
344 return buf->buf;
348 * Return non-zero iff the path is usable as an alternate object database.
350 static int alt_odb_usable(struct raw_object_store *o,
351 struct strbuf *path,
352 const char *normalized_objdir)
354 struct alternate_object_database *alt;
356 /* Detect cases where alternate disappeared */
357 if (!is_directory(path->buf)) {
358 error("object directory %s does not exist; "
359 "check .git/objects/info/alternates.",
360 path->buf);
361 return 0;
365 * Prevent the common mistake of listing the same
366 * thing twice, or object directory itself.
368 for (alt = o->alt_odb_list; alt; alt = alt->next) {
369 if (!fspathcmp(path->buf, alt->path))
370 return 0;
372 if (!fspathcmp(path->buf, normalized_objdir))
373 return 0;
375 return 1;
379 * Prepare alternate object database registry.
381 * The variable alt_odb_list points at the list of struct
382 * alternate_object_database. The elements on this list come from
383 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
384 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
385 * whose contents is similar to that environment variable but can be
386 * LF separated. Its base points at a statically allocated buffer that
387 * contains "/the/directory/corresponding/to/.git/objects/...", while
388 * its name points just after the slash at the end of ".git/objects/"
389 * in the example above, and has enough space to hold 40-byte hex
390 * SHA1, an extra slash for the first level indirection, and the
391 * terminating NUL.
393 static void read_info_alternates(struct repository *r,
394 const char *relative_base,
395 int depth);
396 static int link_alt_odb_entry(struct repository *r, const char *entry,
397 const char *relative_base, int depth, const char *normalized_objdir)
399 struct alternate_object_database *ent;
400 struct strbuf pathbuf = STRBUF_INIT;
402 if (!is_absolute_path(entry) && relative_base) {
403 strbuf_realpath(&pathbuf, relative_base, 1);
404 strbuf_addch(&pathbuf, '/');
406 strbuf_addstr(&pathbuf, entry);
408 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
409 error("unable to normalize alternate object path: %s",
410 pathbuf.buf);
411 strbuf_release(&pathbuf);
412 return -1;
416 * The trailing slash after the directory name is given by
417 * this function at the end. Remove duplicates.
419 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
420 strbuf_setlen(&pathbuf, pathbuf.len - 1);
422 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
423 strbuf_release(&pathbuf);
424 return -1;
427 ent = alloc_alt_odb(pathbuf.buf);
429 /* add the alternate entry */
430 *r->objects->alt_odb_tail = ent;
431 r->objects->alt_odb_tail = &(ent->next);
432 ent->next = NULL;
434 /* recursively add alternates */
435 read_info_alternates(r, pathbuf.buf, depth + 1);
437 strbuf_release(&pathbuf);
438 return 0;
441 static const char *parse_alt_odb_entry(const char *string,
442 int sep,
443 struct strbuf *out)
445 const char *end;
447 strbuf_reset(out);
449 if (*string == '#') {
450 /* comment; consume up to next separator */
451 end = strchrnul(string, sep);
452 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
454 * quoted path; unquote_c_style has copied the
455 * data for us and set "end". Broken quoting (e.g.,
456 * an entry that doesn't end with a quote) falls
457 * back to the unquoted case below.
459 } else {
460 /* normal, unquoted path */
461 end = strchrnul(string, sep);
462 strbuf_add(out, string, end - string);
465 if (*end)
466 end++;
467 return end;
470 static void link_alt_odb_entries(struct repository *r, const char *alt,
471 int sep, const char *relative_base, int depth)
473 struct strbuf objdirbuf = STRBUF_INIT;
474 struct strbuf entry = STRBUF_INIT;
476 if (!alt || !*alt)
477 return;
479 if (depth > 5) {
480 error("%s: ignoring alternate object stores, nesting too deep.",
481 relative_base);
482 return;
485 strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
486 if (strbuf_normalize_path(&objdirbuf) < 0)
487 die("unable to normalize object directory: %s",
488 objdirbuf.buf);
490 while (*alt) {
491 alt = parse_alt_odb_entry(alt, sep, &entry);
492 if (!entry.len)
493 continue;
494 link_alt_odb_entry(r, entry.buf,
495 relative_base, depth, objdirbuf.buf);
497 strbuf_release(&entry);
498 strbuf_release(&objdirbuf);
501 static void read_info_alternates(struct repository *r,
502 const char *relative_base,
503 int depth)
505 char *path;
506 struct strbuf buf = STRBUF_INIT;
508 path = xstrfmt("%s/info/alternates", relative_base);
509 if (strbuf_read_file(&buf, path, 1024) < 0) {
510 warn_on_fopen_errors(path);
511 free(path);
512 return;
515 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
516 strbuf_release(&buf);
517 free(path);
520 struct alternate_object_database *alloc_alt_odb(const char *dir)
522 struct alternate_object_database *ent;
524 FLEX_ALLOC_STR(ent, path, dir);
525 strbuf_init(&ent->scratch, 0);
526 strbuf_addf(&ent->scratch, "%s/", dir);
527 ent->base_len = ent->scratch.len;
529 return ent;
532 void add_to_alternates_file(const char *reference)
534 struct lock_file lock = LOCK_INIT;
535 char *alts = git_pathdup("objects/info/alternates");
536 FILE *in, *out;
537 int found = 0;
539 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
540 out = fdopen_lock_file(&lock, "w");
541 if (!out)
542 die_errno("unable to fdopen alternates lockfile");
544 in = fopen(alts, "r");
545 if (in) {
546 struct strbuf line = STRBUF_INIT;
548 while (strbuf_getline(&line, in) != EOF) {
549 if (!strcmp(reference, line.buf)) {
550 found = 1;
551 break;
553 fprintf_or_die(out, "%s\n", line.buf);
556 strbuf_release(&line);
557 fclose(in);
559 else if (errno != ENOENT)
560 die_errno("unable to read alternates file");
562 if (found) {
563 rollback_lock_file(&lock);
564 } else {
565 fprintf_or_die(out, "%s\n", reference);
566 if (commit_lock_file(&lock))
567 die_errno("unable to move new alternates file into place");
568 if (the_repository->objects->alt_odb_tail)
569 link_alt_odb_entries(the_repository, reference,
570 '\n', NULL, 0);
572 free(alts);
575 void add_to_alternates_memory(const char *reference)
578 * Make sure alternates are initialized, or else our entry may be
579 * overwritten when they are.
581 prepare_alt_odb(the_repository);
583 link_alt_odb_entries(the_repository, reference,
584 '\n', NULL, 0);
588 * Compute the exact path an alternate is at and returns it. In case of
589 * error NULL is returned and the human readable error is added to `err`
590 * `path` may be relative and should point to $GITDIR.
591 * `err` must not be null.
593 char *compute_alternate_path(const char *path, struct strbuf *err)
595 char *ref_git = NULL;
596 const char *repo, *ref_git_s;
597 int seen_error = 0;
599 ref_git_s = real_path_if_valid(path);
600 if (!ref_git_s) {
601 seen_error = 1;
602 strbuf_addf(err, _("path '%s' does not exist"), path);
603 goto out;
604 } else
606 * Beware: read_gitfile(), real_path() and mkpath()
607 * return static buffer
609 ref_git = xstrdup(ref_git_s);
611 repo = read_gitfile(ref_git);
612 if (!repo)
613 repo = read_gitfile(mkpath("%s/.git", ref_git));
614 if (repo) {
615 free(ref_git);
616 ref_git = xstrdup(repo);
619 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
620 char *ref_git_git = mkpathdup("%s/.git", ref_git);
621 free(ref_git);
622 ref_git = ref_git_git;
623 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
624 struct strbuf sb = STRBUF_INIT;
625 seen_error = 1;
626 if (get_common_dir(&sb, ref_git)) {
627 strbuf_addf(err,
628 _("reference repository '%s' as a linked "
629 "checkout is not supported yet."),
630 path);
631 goto out;
634 strbuf_addf(err, _("reference repository '%s' is not a "
635 "local repository."), path);
636 goto out;
639 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
640 strbuf_addf(err, _("reference repository '%s' is shallow"),
641 path);
642 seen_error = 1;
643 goto out;
646 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
647 strbuf_addf(err,
648 _("reference repository '%s' is grafted"),
649 path);
650 seen_error = 1;
651 goto out;
654 out:
655 if (seen_error) {
656 FREE_AND_NULL(ref_git);
659 return ref_git;
662 int foreach_alt_odb(alt_odb_fn fn, void *cb)
664 struct alternate_object_database *ent;
665 int r = 0;
667 prepare_alt_odb(the_repository);
668 for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
669 r = fn(ent, cb);
670 if (r)
671 break;
673 return r;
676 void prepare_alt_odb(struct repository *r)
678 if (r->objects->alt_odb_tail)
679 return;
681 r->objects->alt_odb_tail = &r->objects->alt_odb_list;
682 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
684 read_info_alternates(r, r->objects->objectdir, 0);
687 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
688 static int freshen_file(const char *fn)
690 struct utimbuf t;
691 t.actime = t.modtime = time(NULL);
692 return !utime(fn, &t);
696 * All of the check_and_freshen functions return 1 if the file exists and was
697 * freshened (if freshening was requested), 0 otherwise. If they return
698 * 0, you should not assume that it is safe to skip a write of the object (it
699 * either does not exist on disk, or has a stale mtime and may be subject to
700 * pruning).
702 int check_and_freshen_file(const char *fn, int freshen)
704 if (access(fn, F_OK))
705 return 0;
706 if (freshen && !freshen_file(fn))
707 return 0;
708 return 1;
711 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
713 static struct strbuf buf = STRBUF_INIT;
715 strbuf_reset(&buf);
716 sha1_file_name(the_repository, &buf, sha1);
718 return check_and_freshen_file(buf.buf, freshen);
721 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
723 struct alternate_object_database *alt;
724 prepare_alt_odb(the_repository);
725 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
726 const char *path = alt_sha1_path(alt, sha1);
727 if (check_and_freshen_file(path, freshen))
728 return 1;
730 return 0;
733 static int check_and_freshen(const unsigned char *sha1, int freshen)
735 return check_and_freshen_local(sha1, freshen) ||
736 check_and_freshen_nonlocal(sha1, freshen);
739 int has_loose_object_nonlocal(const unsigned char *sha1)
741 return check_and_freshen_nonlocal(sha1, 0);
744 static int has_loose_object(const unsigned char *sha1)
746 return check_and_freshen(sha1, 0);
749 static void mmap_limit_check(size_t length)
751 static size_t limit = 0;
752 if (!limit) {
753 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
754 if (!limit)
755 limit = SIZE_MAX;
757 if (length > limit)
758 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
759 (uintmax_t)length, (uintmax_t)limit);
762 void *xmmap_gently(void *start, size_t length,
763 int prot, int flags, int fd, off_t offset)
765 void *ret;
767 mmap_limit_check(length);
768 ret = mmap(start, length, prot, flags, fd, offset);
769 if (ret == MAP_FAILED) {
770 if (!length)
771 return NULL;
772 release_pack_memory(length);
773 ret = mmap(start, length, prot, flags, fd, offset);
775 return ret;
778 void *xmmap(void *start, size_t length,
779 int prot, int flags, int fd, off_t offset)
781 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
782 if (ret == MAP_FAILED)
783 die_errno("mmap failed");
784 return ret;
788 * With an in-core object data in "map", rehash it to make sure the
789 * object name actually matches "sha1" to detect object corruption.
790 * With "map" == NULL, try reading the object named with "sha1" using
791 * the streaming interface and rehash it to do the same.
793 int check_sha1_signature(const unsigned char *sha1, void *map,
794 unsigned long size, const char *type)
796 unsigned char real_sha1[20];
797 enum object_type obj_type;
798 struct git_istream *st;
799 git_SHA_CTX c;
800 char hdr[32];
801 int hdrlen;
803 if (map) {
804 hash_sha1_file(map, size, type, real_sha1);
805 return hashcmp(sha1, real_sha1) ? -1 : 0;
808 st = open_istream(sha1, &obj_type, &size, NULL);
809 if (!st)
810 return -1;
812 /* Generate the header */
813 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
815 /* Sha1.. */
816 git_SHA1_Init(&c);
817 git_SHA1_Update(&c, hdr, hdrlen);
818 for (;;) {
819 char buf[1024 * 16];
820 ssize_t readlen = read_istream(st, buf, sizeof(buf));
822 if (readlen < 0) {
823 close_istream(st);
824 return -1;
826 if (!readlen)
827 break;
828 git_SHA1_Update(&c, buf, readlen);
830 git_SHA1_Final(real_sha1, &c);
831 close_istream(st);
832 return hashcmp(sha1, real_sha1) ? -1 : 0;
835 int git_open_cloexec(const char *name, int flags)
837 int fd;
838 static int o_cloexec = O_CLOEXEC;
840 fd = open(name, flags | o_cloexec);
841 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
842 /* Try again w/o O_CLOEXEC: the kernel might not support it */
843 o_cloexec &= ~O_CLOEXEC;
844 fd = open(name, flags | o_cloexec);
847 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
849 static int fd_cloexec = FD_CLOEXEC;
851 if (!o_cloexec && 0 <= fd && fd_cloexec) {
852 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
853 int flags = fcntl(fd, F_GETFD);
854 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
855 fd_cloexec = 0;
858 #endif
859 return fd;
863 * Find "sha1" as a loose object in the local repository or in an alternate.
864 * Returns 0 on success, negative on failure.
866 * The "path" out-parameter will give the path of the object we found (if any).
867 * Note that it may point to static storage and is only valid until another
868 * call to sha1_file_name(), etc.
870 static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
871 struct stat *st, const char **path)
873 struct alternate_object_database *alt;
874 static struct strbuf buf = STRBUF_INIT;
876 strbuf_reset(&buf);
877 sha1_file_name(r, &buf, sha1);
878 *path = buf.buf;
880 if (!lstat(*path, st))
881 return 0;
883 prepare_alt_odb(r);
884 errno = ENOENT;
885 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
886 *path = alt_sha1_path(alt, sha1);
887 if (!lstat(*path, st))
888 return 0;
891 return -1;
895 * Like stat_sha1_file(), but actually open the object and return the
896 * descriptor. See the caveats on the "path" parameter above.
898 static int open_sha1_file(struct repository *r,
899 const unsigned char *sha1, const char **path)
901 int fd;
902 struct alternate_object_database *alt;
903 int most_interesting_errno;
904 static struct strbuf buf = STRBUF_INIT;
906 strbuf_reset(&buf);
907 sha1_file_name(r, &buf, sha1);
908 *path = buf.buf;
910 fd = git_open(*path);
911 if (fd >= 0)
912 return fd;
913 most_interesting_errno = errno;
915 prepare_alt_odb(r);
916 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
917 *path = alt_sha1_path(alt, sha1);
918 fd = git_open(*path);
919 if (fd >= 0)
920 return fd;
921 if (most_interesting_errno == ENOENT)
922 most_interesting_errno = errno;
924 errno = most_interesting_errno;
925 return -1;
929 * Map the loose object at "path" if it is not NULL, or the path found by
930 * searching for a loose object named "sha1".
932 static void *map_sha1_file_1(struct repository *r, const char *path,
933 const unsigned char *sha1, unsigned long *size)
935 void *map;
936 int fd;
938 if (path)
939 fd = git_open(path);
940 else
941 fd = open_sha1_file(r, sha1, &path);
942 map = NULL;
943 if (fd >= 0) {
944 struct stat st;
946 if (!fstat(fd, &st)) {
947 *size = xsize_t(st.st_size);
948 if (!*size) {
949 /* mmap() is forbidden on empty files */
950 error("object file %s is empty", path);
951 return NULL;
953 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
955 close(fd);
957 return map;
960 void *map_sha1_file(struct repository *r,
961 const unsigned char *sha1, unsigned long *size)
963 return map_sha1_file_1(r, NULL, sha1, size);
966 static int unpack_sha1_short_header(git_zstream *stream,
967 unsigned char *map, unsigned long mapsize,
968 void *buffer, unsigned long bufsiz)
970 /* Get the data stream */
971 memset(stream, 0, sizeof(*stream));
972 stream->next_in = map;
973 stream->avail_in = mapsize;
974 stream->next_out = buffer;
975 stream->avail_out = bufsiz;
977 git_inflate_init(stream);
978 return git_inflate(stream, 0);
981 int unpack_sha1_header(git_zstream *stream,
982 unsigned char *map, unsigned long mapsize,
983 void *buffer, unsigned long bufsiz)
985 int status = unpack_sha1_short_header(stream, map, mapsize,
986 buffer, bufsiz);
988 if (status < Z_OK)
989 return status;
991 /* Make sure we have the terminating NUL */
992 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
993 return -1;
994 return 0;
997 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
998 unsigned long mapsize, void *buffer,
999 unsigned long bufsiz, struct strbuf *header)
1001 int status;
1003 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1004 if (status < Z_OK)
1005 return -1;
1008 * Check if entire header is unpacked in the first iteration.
1010 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1011 return 0;
1014 * buffer[0..bufsiz] was not large enough. Copy the partial
1015 * result out to header, and then append the result of further
1016 * reading the stream.
1018 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1019 stream->next_out = buffer;
1020 stream->avail_out = bufsiz;
1022 do {
1023 status = git_inflate(stream, 0);
1024 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1025 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1026 return 0;
1027 stream->next_out = buffer;
1028 stream->avail_out = bufsiz;
1029 } while (status != Z_STREAM_END);
1030 return -1;
1033 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1035 int bytes = strlen(buffer) + 1;
1036 unsigned char *buf = xmallocz(size);
1037 unsigned long n;
1038 int status = Z_OK;
1040 n = stream->total_out - bytes;
1041 if (n > size)
1042 n = size;
1043 memcpy(buf, (char *) buffer + bytes, n);
1044 bytes = n;
1045 if (bytes <= size) {
1047 * The above condition must be (bytes <= size), not
1048 * (bytes < size). In other words, even though we
1049 * expect no more output and set avail_out to zero,
1050 * the input zlib stream may have bytes that express
1051 * "this concludes the stream", and we *do* want to
1052 * eat that input.
1054 * Otherwise we would not be able to test that we
1055 * consumed all the input to reach the expected size;
1056 * we also want to check that zlib tells us that all
1057 * went well with status == Z_STREAM_END at the end.
1059 stream->next_out = buf + bytes;
1060 stream->avail_out = size - bytes;
1061 while (status == Z_OK)
1062 status = git_inflate(stream, Z_FINISH);
1064 if (status == Z_STREAM_END && !stream->avail_in) {
1065 git_inflate_end(stream);
1066 return buf;
1069 if (status < 0)
1070 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1071 else if (stream->avail_in)
1072 error("garbage at end of loose object '%s'",
1073 sha1_to_hex(sha1));
1074 free(buf);
1075 return NULL;
1079 * We used to just use "sscanf()", but that's actually way
1080 * too permissive for what we want to check. So do an anal
1081 * object header parse by hand.
1083 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1084 unsigned int flags)
1086 const char *type_buf = hdr;
1087 unsigned long size;
1088 int type, type_len = 0;
1091 * The type can be of any size but is followed by
1092 * a space.
1094 for (;;) {
1095 char c = *hdr++;
1096 if (!c)
1097 return -1;
1098 if (c == ' ')
1099 break;
1100 type_len++;
1103 type = type_from_string_gently(type_buf, type_len, 1);
1104 if (oi->typename)
1105 strbuf_add(oi->typename, type_buf, type_len);
1107 * Set type to 0 if its an unknown object and
1108 * we're obtaining the type using '--allow-unknown-type'
1109 * option.
1111 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1112 type = 0;
1113 else if (type < 0)
1114 die("invalid object type");
1115 if (oi->typep)
1116 *oi->typep = type;
1119 * The length must follow immediately, and be in canonical
1120 * decimal format (ie "010" is not valid).
1122 size = *hdr++ - '0';
1123 if (size > 9)
1124 return -1;
1125 if (size) {
1126 for (;;) {
1127 unsigned long c = *hdr - '0';
1128 if (c > 9)
1129 break;
1130 hdr++;
1131 size = size * 10 + c;
1135 if (oi->sizep)
1136 *oi->sizep = size;
1139 * The length must be followed by a zero byte
1141 return *hdr ? -1 : type;
1144 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1146 struct object_info oi = OBJECT_INFO_INIT;
1148 oi.sizep = sizep;
1149 return parse_sha1_header_extended(hdr, &oi, 0);
1152 static int sha1_loose_object_info(struct repository *r,
1153 const unsigned char *sha1,
1154 struct object_info *oi, int flags)
1156 int status = 0;
1157 unsigned long mapsize;
1158 void *map;
1159 git_zstream stream;
1160 char hdr[32];
1161 struct strbuf hdrbuf = STRBUF_INIT;
1162 unsigned long size_scratch;
1164 if (oi->delta_base_sha1)
1165 hashclr(oi->delta_base_sha1);
1168 * If we don't care about type or size, then we don't
1169 * need to look inside the object at all. Note that we
1170 * do not optimize out the stat call, even if the
1171 * caller doesn't care about the disk-size, since our
1172 * return value implicitly indicates whether the
1173 * object even exists.
1175 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1176 const char *path;
1177 struct stat st;
1178 if (stat_sha1_file(r, sha1, &st, &path) < 0)
1179 return -1;
1180 if (oi->disk_sizep)
1181 *oi->disk_sizep = st.st_size;
1182 return 0;
1185 map = map_sha1_file(r, sha1, &mapsize);
1186 if (!map)
1187 return -1;
1189 if (!oi->sizep)
1190 oi->sizep = &size_scratch;
1192 if (oi->disk_sizep)
1193 *oi->disk_sizep = mapsize;
1194 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1195 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1196 status = error("unable to unpack %s header with --allow-unknown-type",
1197 sha1_to_hex(sha1));
1198 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1199 status = error("unable to unpack %s header",
1200 sha1_to_hex(sha1));
1201 if (status < 0)
1202 ; /* Do nothing */
1203 else if (hdrbuf.len) {
1204 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1205 status = error("unable to parse %s header with --allow-unknown-type",
1206 sha1_to_hex(sha1));
1207 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1208 status = error("unable to parse %s header", sha1_to_hex(sha1));
1210 if (status >= 0 && oi->contentp) {
1211 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1212 *oi->sizep, sha1);
1213 if (!*oi->contentp) {
1214 git_inflate_end(&stream);
1215 status = -1;
1217 } else
1218 git_inflate_end(&stream);
1220 munmap(map, mapsize);
1221 if (status && oi->typep)
1222 *oi->typep = status;
1223 if (oi->sizep == &size_scratch)
1224 oi->sizep = NULL;
1225 strbuf_release(&hdrbuf);
1226 oi->whence = OI_LOOSE;
1227 return (status < 0) ? status : 0;
1230 int fetch_if_missing = 1;
1232 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1234 static struct object_info blank_oi = OBJECT_INFO_INIT;
1235 struct pack_entry e;
1236 int rtype;
1237 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1238 lookup_replace_object(sha1) :
1239 sha1;
1240 int already_retried = 0;
1242 if (is_null_sha1(real))
1243 return -1;
1245 if (!oi)
1246 oi = &blank_oi;
1248 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1249 struct cached_object *co = find_cached_object(real);
1250 if (co) {
1251 if (oi->typep)
1252 *(oi->typep) = co->type;
1253 if (oi->sizep)
1254 *(oi->sizep) = co->size;
1255 if (oi->disk_sizep)
1256 *(oi->disk_sizep) = 0;
1257 if (oi->delta_base_sha1)
1258 hashclr(oi->delta_base_sha1);
1259 if (oi->typename)
1260 strbuf_addstr(oi->typename, typename(co->type));
1261 if (oi->contentp)
1262 *oi->contentp = xmemdupz(co->buf, co->size);
1263 oi->whence = OI_CACHED;
1264 return 0;
1268 while (1) {
1269 if (find_pack_entry(real, &e))
1270 break;
1272 /* Most likely it's a loose object. */
1273 if (!sha1_loose_object_info(the_repository, real, oi, flags))
1274 return 0;
1276 /* Not a loose object; someone else may have just packed it. */
1277 reprepare_packed_git(the_repository);
1278 if (find_pack_entry(real, &e))
1279 break;
1281 /* Check if it is a missing object */
1282 if (fetch_if_missing && repository_format_partial_clone &&
1283 !already_retried) {
1285 * TODO Investigate haveing fetch_object() return
1286 * TODO error/success and stopping the music here.
1288 fetch_object(repository_format_partial_clone, real);
1289 already_retried = 1;
1290 continue;
1293 return -1;
1296 if (oi == &blank_oi)
1298 * We know that the caller doesn't actually need the
1299 * information below, so return early.
1301 return 0;
1302 rtype = packed_object_info(e.p, e.offset, oi);
1303 if (rtype < 0) {
1304 mark_bad_packed_object(e.p, real);
1305 return sha1_object_info_extended(real, oi, 0);
1306 } else if (oi->whence == OI_PACKED) {
1307 oi->u.packed.offset = e.offset;
1308 oi->u.packed.pack = e.p;
1309 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1310 rtype == OBJ_OFS_DELTA);
1313 return 0;
1316 /* returns enum object_type or negative */
1317 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1319 enum object_type type;
1320 struct object_info oi = OBJECT_INFO_INIT;
1322 oi.typep = &type;
1323 oi.sizep = sizep;
1324 if (sha1_object_info_extended(sha1, &oi,
1325 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1326 return -1;
1327 return type;
1330 static void *read_object(const unsigned char *sha1, enum object_type *type,
1331 unsigned long *size)
1333 struct object_info oi = OBJECT_INFO_INIT;
1334 void *content;
1335 oi.typep = type;
1336 oi.sizep = size;
1337 oi.contentp = &content;
1339 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1340 return NULL;
1341 return content;
1344 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1345 unsigned char *sha1)
1347 struct cached_object *co;
1349 hash_sha1_file(buf, len, typename(type), sha1);
1350 if (has_sha1_file(sha1) || find_cached_object(sha1))
1351 return 0;
1352 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1353 co = &cached_objects[cached_object_nr++];
1354 co->size = len;
1355 co->type = type;
1356 co->buf = xmalloc(len);
1357 memcpy(co->buf, buf, len);
1358 hashcpy(co->sha1, sha1);
1359 return 0;
1363 * This function dies on corrupt objects; the callers who want to
1364 * deal with them should arrange to call read_object() and give error
1365 * messages themselves.
1367 void *read_sha1_file_extended(const unsigned char *sha1,
1368 enum object_type *type,
1369 unsigned long *size,
1370 int lookup_replace)
1372 void *data;
1373 const struct packed_git *p;
1374 const char *path;
1375 struct stat st;
1376 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1377 : sha1;
1379 errno = 0;
1380 data = read_object(repl, type, size);
1381 if (data)
1382 return data;
1384 if (errno && errno != ENOENT)
1385 die_errno("failed to read object %s", sha1_to_hex(sha1));
1387 /* die if we replaced an object with one that does not exist */
1388 if (repl != sha1)
1389 die("replacement %s not found for %s",
1390 sha1_to_hex(repl), sha1_to_hex(sha1));
1392 if (!stat_sha1_file(the_repository, repl, &st, &path))
1393 die("loose object %s (stored in %s) is corrupt",
1394 sha1_to_hex(repl), path);
1396 if ((p = has_packed_and_bad(repl)) != NULL)
1397 die("packed object %s (stored in %s) is corrupt",
1398 sha1_to_hex(repl), p->pack_name);
1400 return NULL;
1403 void *read_object_with_reference(const unsigned char *sha1,
1404 const char *required_type_name,
1405 unsigned long *size,
1406 unsigned char *actual_sha1_return)
1408 enum object_type type, required_type;
1409 void *buffer;
1410 unsigned long isize;
1411 unsigned char actual_sha1[20];
1413 required_type = type_from_string(required_type_name);
1414 hashcpy(actual_sha1, sha1);
1415 while (1) {
1416 int ref_length = -1;
1417 const char *ref_type = NULL;
1419 buffer = read_sha1_file(actual_sha1, &type, &isize);
1420 if (!buffer)
1421 return NULL;
1422 if (type == required_type) {
1423 *size = isize;
1424 if (actual_sha1_return)
1425 hashcpy(actual_sha1_return, actual_sha1);
1426 return buffer;
1428 /* Handle references */
1429 else if (type == OBJ_COMMIT)
1430 ref_type = "tree ";
1431 else if (type == OBJ_TAG)
1432 ref_type = "object ";
1433 else {
1434 free(buffer);
1435 return NULL;
1437 ref_length = strlen(ref_type);
1439 if (ref_length + 40 > isize ||
1440 memcmp(buffer, ref_type, ref_length) ||
1441 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1442 free(buffer);
1443 return NULL;
1445 free(buffer);
1446 /* Now we have the ID of the referred-to object in
1447 * actual_sha1. Check again. */
1451 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1452 const char *type, unsigned char *sha1,
1453 char *hdr, int *hdrlen)
1455 git_SHA_CTX c;
1457 /* Generate the header */
1458 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1460 /* Sha1.. */
1461 git_SHA1_Init(&c);
1462 git_SHA1_Update(&c, hdr, *hdrlen);
1463 git_SHA1_Update(&c, buf, len);
1464 git_SHA1_Final(sha1, &c);
1468 * Move the just written object into its final resting place.
1470 int finalize_object_file(const char *tmpfile, const char *filename)
1472 int ret = 0;
1474 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1475 goto try_rename;
1476 else if (link(tmpfile, filename))
1477 ret = errno;
1480 * Coda hack - coda doesn't like cross-directory links,
1481 * so we fall back to a rename, which will mean that it
1482 * won't be able to check collisions, but that's not a
1483 * big deal.
1485 * The same holds for FAT formatted media.
1487 * When this succeeds, we just return. We have nothing
1488 * left to unlink.
1490 if (ret && ret != EEXIST) {
1491 try_rename:
1492 if (!rename(tmpfile, filename))
1493 goto out;
1494 ret = errno;
1496 unlink_or_warn(tmpfile);
1497 if (ret) {
1498 if (ret != EEXIST) {
1499 return error_errno("unable to write sha1 filename %s", filename);
1501 /* FIXME!!! Collision check here ? */
1504 out:
1505 if (adjust_shared_perm(filename))
1506 return error("unable to set permission to '%s'", filename);
1507 return 0;
1510 static int write_buffer(int fd, const void *buf, size_t len)
1512 if (write_in_full(fd, buf, len) < 0)
1513 return error_errno("file write error");
1514 return 0;
1517 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1518 unsigned char *sha1)
1520 char hdr[32];
1521 int hdrlen = sizeof(hdr);
1522 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1523 return 0;
1526 /* Finalize a file on disk, and close it. */
1527 static void close_sha1_file(int fd)
1529 if (fsync_object_files)
1530 fsync_or_die(fd, "sha1 file");
1531 if (close(fd) != 0)
1532 die_errno("error when closing sha1 file");
1535 /* Size of directory component, including the ending '/' */
1536 static inline int directory_size(const char *filename)
1538 const char *s = strrchr(filename, '/');
1539 if (!s)
1540 return 0;
1541 return s - filename + 1;
1545 * This creates a temporary file in the same directory as the final
1546 * 'filename'
1548 * We want to avoid cross-directory filename renames, because those
1549 * can have problems on various filesystems (FAT, NFS, Coda).
1551 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1553 int fd, dirlen = directory_size(filename);
1555 strbuf_reset(tmp);
1556 strbuf_add(tmp, filename, dirlen);
1557 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1558 fd = git_mkstemp_mode(tmp->buf, 0444);
1559 if (fd < 0 && dirlen && errno == ENOENT) {
1561 * Make sure the directory exists; note that the contents
1562 * of the buffer are undefined after mkstemp returns an
1563 * error, so we have to rewrite the whole buffer from
1564 * scratch.
1566 strbuf_reset(tmp);
1567 strbuf_add(tmp, filename, dirlen - 1);
1568 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1569 return -1;
1570 if (adjust_shared_perm(tmp->buf))
1571 return -1;
1573 /* Try again */
1574 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1575 fd = git_mkstemp_mode(tmp->buf, 0444);
1577 return fd;
1580 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1581 const void *buf, unsigned long len, time_t mtime)
1583 int fd, ret;
1584 unsigned char compressed[4096];
1585 git_zstream stream;
1586 git_SHA_CTX c;
1587 unsigned char parano_sha1[20];
1588 static struct strbuf tmp_file = STRBUF_INIT;
1589 static struct strbuf filename = STRBUF_INIT;
1591 strbuf_reset(&filename);
1592 sha1_file_name(the_repository, &filename, sha1);
1594 fd = create_tmpfile(&tmp_file, filename.buf);
1595 if (fd < 0) {
1596 if (errno == EACCES)
1597 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1598 else
1599 return error_errno("unable to create temporary file");
1602 /* Set it up */
1603 git_deflate_init(&stream, zlib_compression_level);
1604 stream.next_out = compressed;
1605 stream.avail_out = sizeof(compressed);
1606 git_SHA1_Init(&c);
1608 /* First header.. */
1609 stream.next_in = (unsigned char *)hdr;
1610 stream.avail_in = hdrlen;
1611 while (git_deflate(&stream, 0) == Z_OK)
1612 ; /* nothing */
1613 git_SHA1_Update(&c, hdr, hdrlen);
1615 /* Then the data itself.. */
1616 stream.next_in = (void *)buf;
1617 stream.avail_in = len;
1618 do {
1619 unsigned char *in0 = stream.next_in;
1620 ret = git_deflate(&stream, Z_FINISH);
1621 git_SHA1_Update(&c, in0, stream.next_in - in0);
1622 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1623 die("unable to write sha1 file");
1624 stream.next_out = compressed;
1625 stream.avail_out = sizeof(compressed);
1626 } while (ret == Z_OK);
1628 if (ret != Z_STREAM_END)
1629 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1630 ret = git_deflate_end_gently(&stream);
1631 if (ret != Z_OK)
1632 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1633 git_SHA1_Final(parano_sha1, &c);
1634 if (hashcmp(sha1, parano_sha1) != 0)
1635 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1637 close_sha1_file(fd);
1639 if (mtime) {
1640 struct utimbuf utb;
1641 utb.actime = mtime;
1642 utb.modtime = mtime;
1643 if (utime(tmp_file.buf, &utb) < 0)
1644 warning_errno("failed utime() on %s", tmp_file.buf);
1647 return finalize_object_file(tmp_file.buf, filename.buf);
1650 static int freshen_loose_object(const unsigned char *sha1)
1652 return check_and_freshen(sha1, 1);
1655 static int freshen_packed_object(const unsigned char *sha1)
1657 struct pack_entry e;
1658 if (!find_pack_entry(sha1, &e))
1659 return 0;
1660 if (e.p->freshened)
1661 return 1;
1662 if (!freshen_file(e.p->pack_name))
1663 return 0;
1664 e.p->freshened = 1;
1665 return 1;
1668 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1670 char hdr[32];
1671 int hdrlen = sizeof(hdr);
1673 /* Normally if we have it in the pack then we do not bother writing
1674 * it out into .git/objects/??/?{38} file.
1676 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1677 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1678 return 0;
1679 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1682 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1683 struct object_id *oid, unsigned flags)
1685 char *header;
1686 int hdrlen, status = 0;
1688 /* type string, SP, %lu of the length plus NUL must fit this */
1689 hdrlen = strlen(type) + 32;
1690 header = xmalloc(hdrlen);
1691 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1693 if (!(flags & HASH_WRITE_OBJECT))
1694 goto cleanup;
1695 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1696 goto cleanup;
1697 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1699 cleanup:
1700 free(header);
1701 return status;
1704 int force_object_loose(const unsigned char *sha1, time_t mtime)
1706 void *buf;
1707 unsigned long len;
1708 enum object_type type;
1709 char hdr[32];
1710 int hdrlen;
1711 int ret;
1713 if (has_loose_object(sha1))
1714 return 0;
1715 buf = read_object(sha1, &type, &len);
1716 if (!buf)
1717 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1718 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1719 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1720 free(buf);
1722 return ret;
1725 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1727 if (!startup_info->have_repository)
1728 return 0;
1729 return sha1_object_info_extended(sha1, NULL,
1730 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1733 int has_object_file(const struct object_id *oid)
1735 return has_sha1_file(oid->hash);
1738 int has_object_file_with_flags(const struct object_id *oid, int flags)
1740 return has_sha1_file_with_flags(oid->hash, flags);
1743 static void check_tree(const void *buf, size_t size)
1745 struct tree_desc desc;
1746 struct name_entry entry;
1748 init_tree_desc(&desc, buf, size);
1749 while (tree_entry(&desc, &entry))
1750 /* do nothing
1751 * tree_entry() will die() on malformed entries */
1755 static void check_commit(const void *buf, size_t size)
1757 struct commit c;
1758 memset(&c, 0, sizeof(c));
1759 if (parse_commit_buffer(&c, buf, size))
1760 die("corrupt commit");
1763 static void check_tag(const void *buf, size_t size)
1765 struct tag t;
1766 memset(&t, 0, sizeof(t));
1767 if (parse_tag_buffer(&t, buf, size))
1768 die("corrupt tag");
1771 static int index_mem(struct object_id *oid, void *buf, size_t size,
1772 enum object_type type,
1773 const char *path, unsigned flags)
1775 int ret, re_allocated = 0;
1776 int write_object = flags & HASH_WRITE_OBJECT;
1778 if (!type)
1779 type = OBJ_BLOB;
1782 * Convert blobs to git internal format
1784 if ((type == OBJ_BLOB) && path) {
1785 struct strbuf nbuf = STRBUF_INIT;
1786 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1787 get_conv_flags(flags))) {
1788 buf = strbuf_detach(&nbuf, &size);
1789 re_allocated = 1;
1792 if (flags & HASH_FORMAT_CHECK) {
1793 if (type == OBJ_TREE)
1794 check_tree(buf, size);
1795 if (type == OBJ_COMMIT)
1796 check_commit(buf, size);
1797 if (type == OBJ_TAG)
1798 check_tag(buf, size);
1801 if (write_object)
1802 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1803 else
1804 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1805 if (re_allocated)
1806 free(buf);
1807 return ret;
1810 static int index_stream_convert_blob(struct object_id *oid, int fd,
1811 const char *path, unsigned flags)
1813 int ret;
1814 const int write_object = flags & HASH_WRITE_OBJECT;
1815 struct strbuf sbuf = STRBUF_INIT;
1817 assert(path);
1818 assert(would_convert_to_git_filter_fd(path));
1820 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1821 get_conv_flags(flags));
1823 if (write_object)
1824 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1825 oid->hash);
1826 else
1827 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1828 oid->hash);
1829 strbuf_release(&sbuf);
1830 return ret;
1833 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1834 const char *path, unsigned flags)
1836 struct strbuf sbuf = STRBUF_INIT;
1837 int ret;
1839 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1840 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1841 else
1842 ret = -1;
1843 strbuf_release(&sbuf);
1844 return ret;
1847 #define SMALL_FILE_SIZE (32*1024)
1849 static int index_core(struct object_id *oid, int fd, size_t size,
1850 enum object_type type, const char *path,
1851 unsigned flags)
1853 int ret;
1855 if (!size) {
1856 ret = index_mem(oid, "", size, type, path, flags);
1857 } else if (size <= SMALL_FILE_SIZE) {
1858 char *buf = xmalloc(size);
1859 ssize_t read_result = read_in_full(fd, buf, size);
1860 if (read_result < 0)
1861 ret = error_errno("read error while indexing %s",
1862 path ? path : "<unknown>");
1863 else if (read_result != size)
1864 ret = error("short read while indexing %s",
1865 path ? path : "<unknown>");
1866 else
1867 ret = index_mem(oid, buf, size, type, path, flags);
1868 free(buf);
1869 } else {
1870 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1871 ret = index_mem(oid, buf, size, type, path, flags);
1872 munmap(buf, size);
1874 return ret;
1878 * This creates one packfile per large blob unless bulk-checkin
1879 * machinery is "plugged".
1881 * This also bypasses the usual "convert-to-git" dance, and that is on
1882 * purpose. We could write a streaming version of the converting
1883 * functions and insert that before feeding the data to fast-import
1884 * (or equivalent in-core API described above). However, that is
1885 * somewhat complicated, as we do not know the size of the filter
1886 * result, which we need to know beforehand when writing a git object.
1887 * Since the primary motivation for trying to stream from the working
1888 * tree file and to avoid mmaping it in core is to deal with large
1889 * binary blobs, they generally do not want to get any conversion, and
1890 * callers should avoid this code path when filters are requested.
1892 static int index_stream(struct object_id *oid, int fd, size_t size,
1893 enum object_type type, const char *path,
1894 unsigned flags)
1896 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1899 int index_fd(struct object_id *oid, int fd, struct stat *st,
1900 enum object_type type, const char *path, unsigned flags)
1902 int ret;
1905 * Call xsize_t() only when needed to avoid potentially unnecessary
1906 * die() for large files.
1908 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1909 ret = index_stream_convert_blob(oid, fd, path, flags);
1910 else if (!S_ISREG(st->st_mode))
1911 ret = index_pipe(oid, fd, type, path, flags);
1912 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1913 (path && would_convert_to_git(&the_index, path)))
1914 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1915 flags);
1916 else
1917 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1918 flags);
1919 close(fd);
1920 return ret;
1923 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1925 int fd;
1926 struct strbuf sb = STRBUF_INIT;
1927 int rc = 0;
1929 switch (st->st_mode & S_IFMT) {
1930 case S_IFREG:
1931 fd = open(path, O_RDONLY);
1932 if (fd < 0)
1933 return error_errno("open(\"%s\")", path);
1934 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1935 return error("%s: failed to insert into database",
1936 path);
1937 break;
1938 case S_IFLNK:
1939 if (strbuf_readlink(&sb, path, st->st_size))
1940 return error_errno("readlink(\"%s\")", path);
1941 if (!(flags & HASH_WRITE_OBJECT))
1942 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1943 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1944 rc = error("%s: failed to insert into database", path);
1945 strbuf_release(&sb);
1946 break;
1947 case S_IFDIR:
1948 return resolve_gitlink_ref(path, "HEAD", oid);
1949 default:
1950 return error("%s: unsupported file type", path);
1952 return rc;
1955 int read_pack_header(int fd, struct pack_header *header)
1957 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1958 /* "eof before pack header was fully read" */
1959 return PH_ERROR_EOF;
1961 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1962 /* "protocol error (pack signature mismatch detected)" */
1963 return PH_ERROR_PACK_SIGNATURE;
1964 if (!pack_version_ok(header->hdr_version))
1965 /* "protocol error (pack version unsupported)" */
1966 return PH_ERROR_PROTOCOL;
1967 return 0;
1970 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1972 enum object_type type = sha1_object_info(sha1, NULL);
1973 if (type < 0)
1974 die("%s is not a valid object", sha1_to_hex(sha1));
1975 if (type != expect)
1976 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1977 typename(expect));
1980 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1981 struct strbuf *path,
1982 each_loose_object_fn obj_cb,
1983 each_loose_cruft_fn cruft_cb,
1984 each_loose_subdir_fn subdir_cb,
1985 void *data)
1987 size_t origlen, baselen;
1988 DIR *dir;
1989 struct dirent *de;
1990 int r = 0;
1991 struct object_id oid;
1993 if (subdir_nr > 0xff)
1994 BUG("invalid loose object subdirectory: %x", subdir_nr);
1996 origlen = path->len;
1997 strbuf_complete(path, '/');
1998 strbuf_addf(path, "%02x", subdir_nr);
2000 dir = opendir(path->buf);
2001 if (!dir) {
2002 if (errno != ENOENT)
2003 r = error_errno("unable to open %s", path->buf);
2004 strbuf_setlen(path, origlen);
2005 return r;
2008 oid.hash[0] = subdir_nr;
2009 strbuf_addch(path, '/');
2010 baselen = path->len;
2012 while ((de = readdir(dir))) {
2013 size_t namelen;
2014 if (is_dot_or_dotdot(de->d_name))
2015 continue;
2017 namelen = strlen(de->d_name);
2018 strbuf_setlen(path, baselen);
2019 strbuf_add(path, de->d_name, namelen);
2020 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2021 !hex_to_bytes(oid.hash + 1, de->d_name,
2022 GIT_SHA1_RAWSZ - 1)) {
2023 if (obj_cb) {
2024 r = obj_cb(&oid, path->buf, data);
2025 if (r)
2026 break;
2028 continue;
2031 if (cruft_cb) {
2032 r = cruft_cb(de->d_name, path->buf, data);
2033 if (r)
2034 break;
2037 closedir(dir);
2039 strbuf_setlen(path, baselen - 1);
2040 if (!r && subdir_cb)
2041 r = subdir_cb(subdir_nr, path->buf, data);
2043 strbuf_setlen(path, origlen);
2045 return r;
2048 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2049 each_loose_object_fn obj_cb,
2050 each_loose_cruft_fn cruft_cb,
2051 each_loose_subdir_fn subdir_cb,
2052 void *data)
2054 int r = 0;
2055 int i;
2057 for (i = 0; i < 256; i++) {
2058 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2059 subdir_cb, data);
2060 if (r)
2061 break;
2064 return r;
2067 int for_each_loose_file_in_objdir(const char *path,
2068 each_loose_object_fn obj_cb,
2069 each_loose_cruft_fn cruft_cb,
2070 each_loose_subdir_fn subdir_cb,
2071 void *data)
2073 struct strbuf buf = STRBUF_INIT;
2074 int r;
2076 strbuf_addstr(&buf, path);
2077 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2078 subdir_cb, data);
2079 strbuf_release(&buf);
2081 return r;
2084 struct loose_alt_odb_data {
2085 each_loose_object_fn *cb;
2086 void *data;
2089 static int loose_from_alt_odb(struct alternate_object_database *alt,
2090 void *vdata)
2092 struct loose_alt_odb_data *data = vdata;
2093 struct strbuf buf = STRBUF_INIT;
2094 int r;
2096 strbuf_addstr(&buf, alt->path);
2097 r = for_each_loose_file_in_objdir_buf(&buf,
2098 data->cb, NULL, NULL,
2099 data->data);
2100 strbuf_release(&buf);
2101 return r;
2104 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2106 struct loose_alt_odb_data alt;
2107 int r;
2109 r = for_each_loose_file_in_objdir(get_object_directory(),
2110 cb, NULL, NULL, data);
2111 if (r)
2112 return r;
2114 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2115 return 0;
2117 alt.cb = cb;
2118 alt.data = data;
2119 return foreach_alt_odb(loose_from_alt_odb, &alt);
2122 static int check_stream_sha1(git_zstream *stream,
2123 const char *hdr,
2124 unsigned long size,
2125 const char *path,
2126 const unsigned char *expected_sha1)
2128 git_SHA_CTX c;
2129 unsigned char real_sha1[GIT_MAX_RAWSZ];
2130 unsigned char buf[4096];
2131 unsigned long total_read;
2132 int status = Z_OK;
2134 git_SHA1_Init(&c);
2135 git_SHA1_Update(&c, hdr, stream->total_out);
2138 * We already read some bytes into hdr, but the ones up to the NUL
2139 * do not count against the object's content size.
2141 total_read = stream->total_out - strlen(hdr) - 1;
2144 * This size comparison must be "<=" to read the final zlib packets;
2145 * see the comment in unpack_sha1_rest for details.
2147 while (total_read <= size &&
2148 (status == Z_OK || status == Z_BUF_ERROR)) {
2149 stream->next_out = buf;
2150 stream->avail_out = sizeof(buf);
2151 if (size - total_read < stream->avail_out)
2152 stream->avail_out = size - total_read;
2153 status = git_inflate(stream, Z_FINISH);
2154 git_SHA1_Update(&c, buf, stream->next_out - buf);
2155 total_read += stream->next_out - buf;
2157 git_inflate_end(stream);
2159 if (status != Z_STREAM_END) {
2160 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2161 return -1;
2163 if (stream->avail_in) {
2164 error("garbage at end of loose object '%s'",
2165 sha1_to_hex(expected_sha1));
2166 return -1;
2169 git_SHA1_Final(real_sha1, &c);
2170 if (hashcmp(expected_sha1, real_sha1)) {
2171 error("sha1 mismatch for %s (expected %s)", path,
2172 sha1_to_hex(expected_sha1));
2173 return -1;
2176 return 0;
2179 int read_loose_object(const char *path,
2180 const unsigned char *expected_sha1,
2181 enum object_type *type,
2182 unsigned long *size,
2183 void **contents)
2185 int ret = -1;
2186 void *map = NULL;
2187 unsigned long mapsize;
2188 git_zstream stream;
2189 char hdr[32];
2191 *contents = NULL;
2193 map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2194 if (!map) {
2195 error_errno("unable to mmap %s", path);
2196 goto out;
2199 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2200 error("unable to unpack header of %s", path);
2201 goto out;
2204 *type = parse_sha1_header(hdr, size);
2205 if (*type < 0) {
2206 error("unable to parse header of %s", path);
2207 git_inflate_end(&stream);
2208 goto out;
2211 if (*type == OBJ_BLOB) {
2212 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2213 goto out;
2214 } else {
2215 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2216 if (!*contents) {
2217 error("unable to unpack contents of %s", path);
2218 git_inflate_end(&stream);
2219 goto out;
2221 if (check_sha1_signature(expected_sha1, *contents,
2222 *size, typename(*type))) {
2223 error("sha1 mismatch for %s (expected %s)", path,
2224 sha1_to_hex(expected_sha1));
2225 free(*contents);
2226 goto out;
2230 ret = 0; /* everything checks out */
2232 out:
2233 if (map)
2234 munmap(map, mapsize);
2235 return ret;