pack: move unpack_object_header()
[git/gitweb.git] / sha1_file.c
blobccf225f055f8d1ee0f8a35c6362a058ec77b7efc
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[20];
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
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
46 * application).
48 static struct cached_object {
49 unsigned char sha1[20];
50 enum object_type type;
51 void *buf;
52 unsigned long size;
53 } *cached_objects;
54 static int cached_object_nr, cached_object_alloc;
56 static struct cached_object empty_tree = {
57 EMPTY_TREE_SHA1_BIN_LITERAL,
58 OBJ_TREE,
59 "",
63 static struct cached_object *find_cached_object(const unsigned char *sha1)
65 int i;
66 struct cached_object *co = cached_objects;
68 for (i = 0; i < cached_object_nr; i++, co++) {
69 if (!hashcmp(co->sha1, sha1))
70 return co;
72 if (!hashcmp(sha1, empty_tree.sha1))
73 return &empty_tree;
74 return NULL;
77 int mkdir_in_gitdir(const char *path)
79 if (mkdir(path, 0777)) {
80 int saved_errno = errno;
81 struct stat st;
82 struct strbuf sb = STRBUF_INIT;
84 if (errno != EEXIST)
85 return -1;
87 * Are we looking at a path in a symlinked worktree
88 * whose original repository does not yet have it?
89 * e.g. .git/rr-cache pointing at its original
90 * repository in which the user hasn't performed any
91 * conflict resolution yet?
93 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
94 strbuf_readlink(&sb, path, st.st_size) ||
95 !is_absolute_path(sb.buf) ||
96 mkdir(sb.buf, 0777)) {
97 strbuf_release(&sb);
98 errno = saved_errno;
99 return -1;
101 strbuf_release(&sb);
103 return adjust_shared_perm(path);
106 enum scld_error safe_create_leading_directories(char *path)
108 char *next_component = path + offset_1st_component(path);
109 enum scld_error ret = SCLD_OK;
111 while (ret == SCLD_OK && next_component) {
112 struct stat st;
113 char *slash = next_component, slash_character;
115 while (*slash && !is_dir_sep(*slash))
116 slash++;
118 if (!*slash)
119 break;
121 next_component = slash + 1;
122 while (is_dir_sep(*next_component))
123 next_component++;
124 if (!*next_component)
125 break;
127 slash_character = *slash;
128 *slash = '\0';
129 if (!stat(path, &st)) {
130 /* path exists */
131 if (!S_ISDIR(st.st_mode)) {
132 errno = ENOTDIR;
133 ret = SCLD_EXISTS;
135 } else if (mkdir(path, 0777)) {
136 if (errno == EEXIST &&
137 !stat(path, &st) && S_ISDIR(st.st_mode))
138 ; /* somebody created it since we checked */
139 else if (errno == ENOENT)
141 * Either mkdir() failed because
142 * somebody just pruned the containing
143 * directory, or stat() failed because
144 * the file that was in our way was
145 * just removed. Either way, inform
146 * the caller that it might be worth
147 * trying again:
149 ret = SCLD_VANISHED;
150 else
151 ret = SCLD_FAILED;
152 } else if (adjust_shared_perm(path)) {
153 ret = SCLD_PERMS;
155 *slash = slash_character;
157 return ret;
160 enum scld_error safe_create_leading_directories_const(const char *path)
162 int save_errno;
163 /* path points to cache entries, so xstrdup before messing with it */
164 char *buf = xstrdup(path);
165 enum scld_error result = safe_create_leading_directories(buf);
167 save_errno = errno;
168 free(buf);
169 errno = save_errno;
170 return result;
173 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
176 * The number of times we will try to remove empty directories
177 * in the way of path. This is only 1 because if another
178 * process is racily creating directories that conflict with
179 * us, we don't want to fight against them.
181 int remove_directories_remaining = 1;
184 * The number of times that we will try to create the
185 * directories containing path. We are willing to attempt this
186 * more than once, because another process could be trying to
187 * clean up empty directories at the same time as we are
188 * trying to create them.
190 int create_directories_remaining = 3;
192 /* A scratch copy of path, filled lazily if we need it: */
193 struct strbuf path_copy = STRBUF_INIT;
195 int ret, save_errno;
197 /* Sanity check: */
198 assert(*path);
200 retry_fn:
201 ret = fn(path, cb);
202 save_errno = errno;
203 if (!ret)
204 goto out;
206 if (errno == EISDIR && remove_directories_remaining-- > 0) {
208 * A directory is in the way. Maybe it is empty; try
209 * to remove it:
211 if (!path_copy.len)
212 strbuf_addstr(&path_copy, path);
214 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
215 goto retry_fn;
216 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
218 * Maybe the containing directory didn't exist, or
219 * maybe it was just deleted by a process that is
220 * racing with us to clean up empty directories. Try
221 * to create it:
223 enum scld_error scld_result;
225 if (!path_copy.len)
226 strbuf_addstr(&path_copy, path);
228 do {
229 scld_result = safe_create_leading_directories(path_copy.buf);
230 if (scld_result == SCLD_OK)
231 goto retry_fn;
232 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
235 out:
236 strbuf_release(&path_copy);
237 errno = save_errno;
238 return ret;
241 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
243 int i;
244 for (i = 0; i < 20; i++) {
245 static char hex[] = "0123456789abcdef";
246 unsigned int val = sha1[i];
247 strbuf_addch(buf, hex[val >> 4]);
248 strbuf_addch(buf, hex[val & 0xf]);
249 if (!i)
250 strbuf_addch(buf, '/');
254 const char *sha1_file_name(const unsigned char *sha1)
256 static struct strbuf buf = STRBUF_INIT;
258 strbuf_reset(&buf);
259 strbuf_addf(&buf, "%s/", get_object_directory());
261 fill_sha1_path(&buf, sha1);
262 return buf.buf;
265 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
267 strbuf_setlen(&alt->scratch, alt->base_len);
268 return &alt->scratch;
271 static const char *alt_sha1_path(struct alternate_object_database *alt,
272 const unsigned char *sha1)
274 struct strbuf *buf = alt_scratch_buf(alt);
275 fill_sha1_path(buf, sha1);
276 return buf->buf;
279 struct alternate_object_database *alt_odb_list;
280 static struct alternate_object_database **alt_odb_tail;
283 * Return non-zero iff the path is usable as an alternate object database.
285 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
287 struct alternate_object_database *alt;
289 /* Detect cases where alternate disappeared */
290 if (!is_directory(path->buf)) {
291 error("object directory %s does not exist; "
292 "check .git/objects/info/alternates.",
293 path->buf);
294 return 0;
298 * Prevent the common mistake of listing the same
299 * thing twice, or object directory itself.
301 for (alt = alt_odb_list; alt; alt = alt->next) {
302 if (!fspathcmp(path->buf, alt->path))
303 return 0;
305 if (!fspathcmp(path->buf, normalized_objdir))
306 return 0;
308 return 1;
312 * Prepare alternate object database registry.
314 * The variable alt_odb_list points at the list of struct
315 * alternate_object_database. The elements on this list come from
316 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
317 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
318 * whose contents is similar to that environment variable but can be
319 * LF separated. Its base points at a statically allocated buffer that
320 * contains "/the/directory/corresponding/to/.git/objects/...", while
321 * its name points just after the slash at the end of ".git/objects/"
322 * in the example above, and has enough space to hold 40-byte hex
323 * SHA1, an extra slash for the first level indirection, and the
324 * terminating NUL.
326 static void read_info_alternates(const char * relative_base, int depth);
327 static int link_alt_odb_entry(const char *entry, const char *relative_base,
328 int depth, const char *normalized_objdir)
330 struct alternate_object_database *ent;
331 struct strbuf pathbuf = STRBUF_INIT;
333 if (!is_absolute_path(entry) && relative_base) {
334 strbuf_realpath(&pathbuf, relative_base, 1);
335 strbuf_addch(&pathbuf, '/');
337 strbuf_addstr(&pathbuf, entry);
339 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
340 error("unable to normalize alternate object path: %s",
341 pathbuf.buf);
342 strbuf_release(&pathbuf);
343 return -1;
347 * The trailing slash after the directory name is given by
348 * this function at the end. Remove duplicates.
350 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
351 strbuf_setlen(&pathbuf, pathbuf.len - 1);
353 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
354 strbuf_release(&pathbuf);
355 return -1;
358 ent = alloc_alt_odb(pathbuf.buf);
360 /* add the alternate entry */
361 *alt_odb_tail = ent;
362 alt_odb_tail = &(ent->next);
363 ent->next = NULL;
365 /* recursively add alternates */
366 read_info_alternates(pathbuf.buf, depth + 1);
368 strbuf_release(&pathbuf);
369 return 0;
372 static const char *parse_alt_odb_entry(const char *string,
373 int sep,
374 struct strbuf *out)
376 const char *end;
378 strbuf_reset(out);
380 if (*string == '#') {
381 /* comment; consume up to next separator */
382 end = strchrnul(string, sep);
383 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
385 * quoted path; unquote_c_style has copied the
386 * data for us and set "end". Broken quoting (e.g.,
387 * an entry that doesn't end with a quote) falls
388 * back to the unquoted case below.
390 } else {
391 /* normal, unquoted path */
392 end = strchrnul(string, sep);
393 strbuf_add(out, string, end - string);
396 if (*end)
397 end++;
398 return end;
401 static void link_alt_odb_entries(const char *alt, int len, int sep,
402 const char *relative_base, int depth)
404 struct strbuf objdirbuf = STRBUF_INIT;
405 struct strbuf entry = STRBUF_INIT;
407 if (depth > 5) {
408 error("%s: ignoring alternate object stores, nesting too deep.",
409 relative_base);
410 return;
413 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
414 if (strbuf_normalize_path(&objdirbuf) < 0)
415 die("unable to normalize object directory: %s",
416 objdirbuf.buf);
418 while (*alt) {
419 alt = parse_alt_odb_entry(alt, sep, &entry);
420 if (!entry.len)
421 continue;
422 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
424 strbuf_release(&entry);
425 strbuf_release(&objdirbuf);
428 static void read_info_alternates(const char * relative_base, int depth)
430 char *map;
431 size_t mapsz;
432 struct stat st;
433 char *path;
434 int fd;
436 path = xstrfmt("%s/info/alternates", relative_base);
437 fd = git_open(path);
438 free(path);
439 if (fd < 0)
440 return;
441 if (fstat(fd, &st) || (st.st_size == 0)) {
442 close(fd);
443 return;
445 mapsz = xsize_t(st.st_size);
446 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
447 close(fd);
449 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
451 munmap(map, mapsz);
454 struct alternate_object_database *alloc_alt_odb(const char *dir)
456 struct alternate_object_database *ent;
458 FLEX_ALLOC_STR(ent, path, dir);
459 strbuf_init(&ent->scratch, 0);
460 strbuf_addf(&ent->scratch, "%s/", dir);
461 ent->base_len = ent->scratch.len;
463 return ent;
466 void add_to_alternates_file(const char *reference)
468 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
469 char *alts = git_pathdup("objects/info/alternates");
470 FILE *in, *out;
472 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
473 out = fdopen_lock_file(lock, "w");
474 if (!out)
475 die_errno("unable to fdopen alternates lockfile");
477 in = fopen(alts, "r");
478 if (in) {
479 struct strbuf line = STRBUF_INIT;
480 int found = 0;
482 while (strbuf_getline(&line, in) != EOF) {
483 if (!strcmp(reference, line.buf)) {
484 found = 1;
485 break;
487 fprintf_or_die(out, "%s\n", line.buf);
490 strbuf_release(&line);
491 fclose(in);
493 if (found) {
494 rollback_lock_file(lock);
495 lock = NULL;
498 else if (errno != ENOENT)
499 die_errno("unable to read alternates file");
501 if (lock) {
502 fprintf_or_die(out, "%s\n", reference);
503 if (commit_lock_file(lock))
504 die_errno("unable to move new alternates file into place");
505 if (alt_odb_tail)
506 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
508 free(alts);
511 void add_to_alternates_memory(const char *reference)
514 * Make sure alternates are initialized, or else our entry may be
515 * overwritten when they are.
517 prepare_alt_odb();
519 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
523 * Compute the exact path an alternate is at and returns it. In case of
524 * error NULL is returned and the human readable error is added to `err`
525 * `path` may be relative and should point to $GITDIR.
526 * `err` must not be null.
528 char *compute_alternate_path(const char *path, struct strbuf *err)
530 char *ref_git = NULL;
531 const char *repo, *ref_git_s;
532 int seen_error = 0;
534 ref_git_s = real_path_if_valid(path);
535 if (!ref_git_s) {
536 seen_error = 1;
537 strbuf_addf(err, _("path '%s' does not exist"), path);
538 goto out;
539 } else
541 * Beware: read_gitfile(), real_path() and mkpath()
542 * return static buffer
544 ref_git = xstrdup(ref_git_s);
546 repo = read_gitfile(ref_git);
547 if (!repo)
548 repo = read_gitfile(mkpath("%s/.git", ref_git));
549 if (repo) {
550 free(ref_git);
551 ref_git = xstrdup(repo);
554 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
555 char *ref_git_git = mkpathdup("%s/.git", ref_git);
556 free(ref_git);
557 ref_git = ref_git_git;
558 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
559 struct strbuf sb = STRBUF_INIT;
560 seen_error = 1;
561 if (get_common_dir(&sb, ref_git)) {
562 strbuf_addf(err,
563 _("reference repository '%s' as a linked "
564 "checkout is not supported yet."),
565 path);
566 goto out;
569 strbuf_addf(err, _("reference repository '%s' is not a "
570 "local repository."), path);
571 goto out;
574 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
575 strbuf_addf(err, _("reference repository '%s' is shallow"),
576 path);
577 seen_error = 1;
578 goto out;
581 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
582 strbuf_addf(err,
583 _("reference repository '%s' is grafted"),
584 path);
585 seen_error = 1;
586 goto out;
589 out:
590 if (seen_error) {
591 FREE_AND_NULL(ref_git);
594 return ref_git;
597 int foreach_alt_odb(alt_odb_fn fn, void *cb)
599 struct alternate_object_database *ent;
600 int r = 0;
602 prepare_alt_odb();
603 for (ent = alt_odb_list; ent; ent = ent->next) {
604 r = fn(ent, cb);
605 if (r)
606 break;
608 return r;
611 void prepare_alt_odb(void)
613 const char *alt;
615 if (alt_odb_tail)
616 return;
618 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
619 if (!alt) alt = "";
621 alt_odb_tail = &alt_odb_list;
622 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
624 read_info_alternates(get_object_directory(), 0);
627 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
628 static int freshen_file(const char *fn)
630 struct utimbuf t;
631 t.actime = t.modtime = time(NULL);
632 return !utime(fn, &t);
636 * All of the check_and_freshen functions return 1 if the file exists and was
637 * freshened (if freshening was requested), 0 otherwise. If they return
638 * 0, you should not assume that it is safe to skip a write of the object (it
639 * either does not exist on disk, or has a stale mtime and may be subject to
640 * pruning).
642 int check_and_freshen_file(const char *fn, int freshen)
644 if (access(fn, F_OK))
645 return 0;
646 if (freshen && !freshen_file(fn))
647 return 0;
648 return 1;
651 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
653 return check_and_freshen_file(sha1_file_name(sha1), freshen);
656 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
658 struct alternate_object_database *alt;
659 prepare_alt_odb();
660 for (alt = alt_odb_list; alt; alt = alt->next) {
661 const char *path = alt_sha1_path(alt, sha1);
662 if (check_and_freshen_file(path, freshen))
663 return 1;
665 return 0;
668 static int check_and_freshen(const unsigned char *sha1, int freshen)
670 return check_and_freshen_local(sha1, freshen) ||
671 check_and_freshen_nonlocal(sha1, freshen);
674 int has_loose_object_nonlocal(const unsigned char *sha1)
676 return check_and_freshen_nonlocal(sha1, 0);
679 static int has_loose_object(const unsigned char *sha1)
681 return check_and_freshen(sha1, 0);
684 static void mmap_limit_check(size_t length)
686 static size_t limit = 0;
687 if (!limit) {
688 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
689 if (!limit)
690 limit = SIZE_MAX;
692 if (length > limit)
693 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
694 (uintmax_t)length, (uintmax_t)limit);
697 void *xmmap_gently(void *start, size_t length,
698 int prot, int flags, int fd, off_t offset)
700 void *ret;
702 mmap_limit_check(length);
703 ret = mmap(start, length, prot, flags, fd, offset);
704 if (ret == MAP_FAILED) {
705 if (!length)
706 return NULL;
707 release_pack_memory(length);
708 ret = mmap(start, length, prot, flags, fd, offset);
710 return ret;
713 void *xmmap(void *start, size_t length,
714 int prot, int flags, int fd, off_t offset)
716 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
717 if (ret == MAP_FAILED)
718 die_errno("mmap failed");
719 return ret;
722 static void mark_bad_packed_object(struct packed_git *p,
723 const unsigned char *sha1)
725 unsigned i;
726 for (i = 0; i < p->num_bad_objects; i++)
727 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
728 return;
729 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
730 st_mult(GIT_MAX_RAWSZ,
731 st_add(p->num_bad_objects, 1)));
732 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
733 p->num_bad_objects++;
736 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
738 struct packed_git *p;
739 unsigned i;
741 for (p = packed_git; p; p = p->next)
742 for (i = 0; i < p->num_bad_objects; i++)
743 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
744 return p;
745 return NULL;
749 * With an in-core object data in "map", rehash it to make sure the
750 * object name actually matches "sha1" to detect object corruption.
751 * With "map" == NULL, try reading the object named with "sha1" using
752 * the streaming interface and rehash it to do the same.
754 int check_sha1_signature(const unsigned char *sha1, void *map,
755 unsigned long size, const char *type)
757 unsigned char real_sha1[20];
758 enum object_type obj_type;
759 struct git_istream *st;
760 git_SHA_CTX c;
761 char hdr[32];
762 int hdrlen;
764 if (map) {
765 hash_sha1_file(map, size, type, real_sha1);
766 return hashcmp(sha1, real_sha1) ? -1 : 0;
769 st = open_istream(sha1, &obj_type, &size, NULL);
770 if (!st)
771 return -1;
773 /* Generate the header */
774 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
776 /* Sha1.. */
777 git_SHA1_Init(&c);
778 git_SHA1_Update(&c, hdr, hdrlen);
779 for (;;) {
780 char buf[1024 * 16];
781 ssize_t readlen = read_istream(st, buf, sizeof(buf));
783 if (readlen < 0) {
784 close_istream(st);
785 return -1;
787 if (!readlen)
788 break;
789 git_SHA1_Update(&c, buf, readlen);
791 git_SHA1_Final(real_sha1, &c);
792 close_istream(st);
793 return hashcmp(sha1, real_sha1) ? -1 : 0;
796 int git_open_cloexec(const char *name, int flags)
798 int fd;
799 static int o_cloexec = O_CLOEXEC;
801 fd = open(name, flags | o_cloexec);
802 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
803 /* Try again w/o O_CLOEXEC: the kernel might not support it */
804 o_cloexec &= ~O_CLOEXEC;
805 fd = open(name, flags | o_cloexec);
808 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
810 static int fd_cloexec = FD_CLOEXEC;
812 if (!o_cloexec && 0 <= fd && fd_cloexec) {
813 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
814 int flags = fcntl(fd, F_GETFD);
815 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
816 fd_cloexec = 0;
819 #endif
820 return fd;
824 * Find "sha1" as a loose object in the local repository or in an alternate.
825 * Returns 0 on success, negative on failure.
827 * The "path" out-parameter will give the path of the object we found (if any).
828 * Note that it may point to static storage and is only valid until another
829 * call to sha1_file_name(), etc.
831 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
832 const char **path)
834 struct alternate_object_database *alt;
836 *path = sha1_file_name(sha1);
837 if (!lstat(*path, st))
838 return 0;
840 prepare_alt_odb();
841 errno = ENOENT;
842 for (alt = alt_odb_list; alt; alt = alt->next) {
843 *path = alt_sha1_path(alt, sha1);
844 if (!lstat(*path, st))
845 return 0;
848 return -1;
852 * Like stat_sha1_file(), but actually open the object and return the
853 * descriptor. See the caveats on the "path" parameter above.
855 static int open_sha1_file(const unsigned char *sha1, const char **path)
857 int fd;
858 struct alternate_object_database *alt;
859 int most_interesting_errno;
861 *path = sha1_file_name(sha1);
862 fd = git_open(*path);
863 if (fd >= 0)
864 return fd;
865 most_interesting_errno = errno;
867 prepare_alt_odb();
868 for (alt = alt_odb_list; alt; alt = alt->next) {
869 *path = alt_sha1_path(alt, sha1);
870 fd = git_open(*path);
871 if (fd >= 0)
872 return fd;
873 if (most_interesting_errno == ENOENT)
874 most_interesting_errno = errno;
876 errno = most_interesting_errno;
877 return -1;
881 * Map the loose object at "path" if it is not NULL, or the path found by
882 * searching for a loose object named "sha1".
884 static void *map_sha1_file_1(const char *path,
885 const unsigned char *sha1,
886 unsigned long *size)
888 void *map;
889 int fd;
891 if (path)
892 fd = git_open(path);
893 else
894 fd = open_sha1_file(sha1, &path);
895 map = NULL;
896 if (fd >= 0) {
897 struct stat st;
899 if (!fstat(fd, &st)) {
900 *size = xsize_t(st.st_size);
901 if (!*size) {
902 /* mmap() is forbidden on empty files */
903 error("object file %s is empty", path);
904 return NULL;
906 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
908 close(fd);
910 return map;
913 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
915 return map_sha1_file_1(NULL, sha1, size);
918 static int unpack_sha1_short_header(git_zstream *stream,
919 unsigned char *map, unsigned long mapsize,
920 void *buffer, unsigned long bufsiz)
922 /* Get the data stream */
923 memset(stream, 0, sizeof(*stream));
924 stream->next_in = map;
925 stream->avail_in = mapsize;
926 stream->next_out = buffer;
927 stream->avail_out = bufsiz;
929 git_inflate_init(stream);
930 return git_inflate(stream, 0);
933 int unpack_sha1_header(git_zstream *stream,
934 unsigned char *map, unsigned long mapsize,
935 void *buffer, unsigned long bufsiz)
937 int status = unpack_sha1_short_header(stream, map, mapsize,
938 buffer, bufsiz);
940 if (status < Z_OK)
941 return status;
943 /* Make sure we have the terminating NUL */
944 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
945 return -1;
946 return 0;
949 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
950 unsigned long mapsize, void *buffer,
951 unsigned long bufsiz, struct strbuf *header)
953 int status;
955 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
956 if (status < Z_OK)
957 return -1;
960 * Check if entire header is unpacked in the first iteration.
962 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
963 return 0;
966 * buffer[0..bufsiz] was not large enough. Copy the partial
967 * result out to header, and then append the result of further
968 * reading the stream.
970 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
971 stream->next_out = buffer;
972 stream->avail_out = bufsiz;
974 do {
975 status = git_inflate(stream, 0);
976 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
977 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
978 return 0;
979 stream->next_out = buffer;
980 stream->avail_out = bufsiz;
981 } while (status != Z_STREAM_END);
982 return -1;
985 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
987 int bytes = strlen(buffer) + 1;
988 unsigned char *buf = xmallocz(size);
989 unsigned long n;
990 int status = Z_OK;
992 n = stream->total_out - bytes;
993 if (n > size)
994 n = size;
995 memcpy(buf, (char *) buffer + bytes, n);
996 bytes = n;
997 if (bytes <= size) {
999 * The above condition must be (bytes <= size), not
1000 * (bytes < size). In other words, even though we
1001 * expect no more output and set avail_out to zero,
1002 * the input zlib stream may have bytes that express
1003 * "this concludes the stream", and we *do* want to
1004 * eat that input.
1006 * Otherwise we would not be able to test that we
1007 * consumed all the input to reach the expected size;
1008 * we also want to check that zlib tells us that all
1009 * went well with status == Z_STREAM_END at the end.
1011 stream->next_out = buf + bytes;
1012 stream->avail_out = size - bytes;
1013 while (status == Z_OK)
1014 status = git_inflate(stream, Z_FINISH);
1016 if (status == Z_STREAM_END && !stream->avail_in) {
1017 git_inflate_end(stream);
1018 return buf;
1021 if (status < 0)
1022 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1023 else if (stream->avail_in)
1024 error("garbage at end of loose object '%s'",
1025 sha1_to_hex(sha1));
1026 free(buf);
1027 return NULL;
1031 * We used to just use "sscanf()", but that's actually way
1032 * too permissive for what we want to check. So do an anal
1033 * object header parse by hand.
1035 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1036 unsigned int flags)
1038 const char *type_buf = hdr;
1039 unsigned long size;
1040 int type, type_len = 0;
1043 * The type can be of any size but is followed by
1044 * a space.
1046 for (;;) {
1047 char c = *hdr++;
1048 if (!c)
1049 return -1;
1050 if (c == ' ')
1051 break;
1052 type_len++;
1055 type = type_from_string_gently(type_buf, type_len, 1);
1056 if (oi->typename)
1057 strbuf_add(oi->typename, type_buf, type_len);
1059 * Set type to 0 if its an unknown object and
1060 * we're obtaining the type using '--allow-unknown-type'
1061 * option.
1063 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1064 type = 0;
1065 else if (type < 0)
1066 die("invalid object type");
1067 if (oi->typep)
1068 *oi->typep = type;
1071 * The length must follow immediately, and be in canonical
1072 * decimal format (ie "010" is not valid).
1074 size = *hdr++ - '0';
1075 if (size > 9)
1076 return -1;
1077 if (size) {
1078 for (;;) {
1079 unsigned long c = *hdr - '0';
1080 if (c > 9)
1081 break;
1082 hdr++;
1083 size = size * 10 + c;
1087 if (oi->sizep)
1088 *oi->sizep = size;
1091 * The length must be followed by a zero byte
1093 return *hdr ? -1 : type;
1096 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1098 struct object_info oi = OBJECT_INFO_INIT;
1100 oi.sizep = sizep;
1101 return parse_sha1_header_extended(hdr, &oi, 0);
1104 static off_t get_delta_base(struct packed_git *p,
1105 struct pack_window **w_curs,
1106 off_t *curpos,
1107 enum object_type type,
1108 off_t delta_obj_offset)
1110 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1111 off_t base_offset;
1113 /* use_pack() assured us we have [base_info, base_info + 20)
1114 * as a range that we can look at without walking off the
1115 * end of the mapped window. Its actually the hash size
1116 * that is assured. An OFS_DELTA longer than the hash size
1117 * is stupid, as then a REF_DELTA would be smaller to store.
1119 if (type == OBJ_OFS_DELTA) {
1120 unsigned used = 0;
1121 unsigned char c = base_info[used++];
1122 base_offset = c & 127;
1123 while (c & 128) {
1124 base_offset += 1;
1125 if (!base_offset || MSB(base_offset, 7))
1126 return 0; /* overflow */
1127 c = base_info[used++];
1128 base_offset = (base_offset << 7) + (c & 127);
1130 base_offset = delta_obj_offset - base_offset;
1131 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1132 return 0; /* out of bound */
1133 *curpos += used;
1134 } else if (type == OBJ_REF_DELTA) {
1135 /* The base entry _must_ be in the same pack */
1136 base_offset = find_pack_entry_one(base_info, p);
1137 *curpos += 20;
1138 } else
1139 die("I am totally screwed");
1140 return base_offset;
1144 * Like get_delta_base above, but we return the sha1 instead of the pack
1145 * offset. This means it is cheaper for REF deltas (we do not have to do
1146 * the final object lookup), but more expensive for OFS deltas (we
1147 * have to load the revidx to convert the offset back into a sha1).
1149 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1150 struct pack_window **w_curs,
1151 off_t curpos,
1152 enum object_type type,
1153 off_t delta_obj_offset)
1155 if (type == OBJ_REF_DELTA) {
1156 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1157 return base;
1158 } else if (type == OBJ_OFS_DELTA) {
1159 struct revindex_entry *revidx;
1160 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1161 type, delta_obj_offset);
1163 if (!base_offset)
1164 return NULL;
1166 revidx = find_pack_revindex(p, base_offset);
1167 if (!revidx)
1168 return NULL;
1170 return nth_packed_object_sha1(p, revidx->nr);
1171 } else
1172 return NULL;
1175 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1177 int type;
1178 struct revindex_entry *revidx;
1179 const unsigned char *sha1;
1180 revidx = find_pack_revindex(p, obj_offset);
1181 if (!revidx)
1182 return OBJ_BAD;
1183 sha1 = nth_packed_object_sha1(p, revidx->nr);
1184 mark_bad_packed_object(p, sha1);
1185 type = sha1_object_info(sha1, NULL);
1186 if (type <= OBJ_NONE)
1187 return OBJ_BAD;
1188 return type;
1191 #define POI_STACK_PREALLOC 64
1193 static enum object_type packed_to_object_type(struct packed_git *p,
1194 off_t obj_offset,
1195 enum object_type type,
1196 struct pack_window **w_curs,
1197 off_t curpos)
1199 off_t small_poi_stack[POI_STACK_PREALLOC];
1200 off_t *poi_stack = small_poi_stack;
1201 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1203 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1204 off_t base_offset;
1205 unsigned long size;
1206 /* Push the object we're going to leave behind */
1207 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1208 poi_stack_alloc = alloc_nr(poi_stack_nr);
1209 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1210 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1211 } else {
1212 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1214 poi_stack[poi_stack_nr++] = obj_offset;
1215 /* If parsing the base offset fails, just unwind */
1216 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1217 if (!base_offset)
1218 goto unwind;
1219 curpos = obj_offset = base_offset;
1220 type = unpack_object_header(p, w_curs, &curpos, &size);
1221 if (type <= OBJ_NONE) {
1222 /* If getting the base itself fails, we first
1223 * retry the base, otherwise unwind */
1224 type = retry_bad_packed_offset(p, base_offset);
1225 if (type > OBJ_NONE)
1226 goto out;
1227 goto unwind;
1231 switch (type) {
1232 case OBJ_BAD:
1233 case OBJ_COMMIT:
1234 case OBJ_TREE:
1235 case OBJ_BLOB:
1236 case OBJ_TAG:
1237 break;
1238 default:
1239 error("unknown object type %i at offset %"PRIuMAX" in %s",
1240 type, (uintmax_t)obj_offset, p->pack_name);
1241 type = OBJ_BAD;
1244 out:
1245 if (poi_stack != small_poi_stack)
1246 free(poi_stack);
1247 return type;
1249 unwind:
1250 while (poi_stack_nr) {
1251 obj_offset = poi_stack[--poi_stack_nr];
1252 type = retry_bad_packed_offset(p, obj_offset);
1253 if (type > OBJ_NONE)
1254 goto out;
1256 type = OBJ_BAD;
1257 goto out;
1260 static struct hashmap delta_base_cache;
1261 static size_t delta_base_cached;
1263 static LIST_HEAD(delta_base_cache_lru);
1265 struct delta_base_cache_key {
1266 struct packed_git *p;
1267 off_t base_offset;
1270 struct delta_base_cache_entry {
1271 struct hashmap hash;
1272 struct delta_base_cache_key key;
1273 struct list_head lru;
1274 void *data;
1275 unsigned long size;
1276 enum object_type type;
1279 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1281 unsigned int hash;
1283 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1284 hash += (hash >> 8) + (hash >> 16);
1285 return hash;
1288 static struct delta_base_cache_entry *
1289 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1291 struct hashmap_entry entry;
1292 struct delta_base_cache_key key;
1294 if (!delta_base_cache.cmpfn)
1295 return NULL;
1297 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1298 key.p = p;
1299 key.base_offset = base_offset;
1300 return hashmap_get(&delta_base_cache, &entry, &key);
1303 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1304 const struct delta_base_cache_key *b)
1306 return a->p == b->p && a->base_offset == b->base_offset;
1309 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1310 const void *va, const void *vb,
1311 const void *vkey)
1313 const struct delta_base_cache_entry *a = va, *b = vb;
1314 const struct delta_base_cache_key *key = vkey;
1315 if (key)
1316 return !delta_base_cache_key_eq(&a->key, key);
1317 else
1318 return !delta_base_cache_key_eq(&a->key, &b->key);
1321 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1323 return !!get_delta_base_cache_entry(p, base_offset);
1327 * Remove the entry from the cache, but do _not_ free the associated
1328 * entry data. The caller takes ownership of the "data" buffer, and
1329 * should copy out any fields it wants before detaching.
1331 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1333 hashmap_remove(&delta_base_cache, ent, &ent->key);
1334 list_del(&ent->lru);
1335 delta_base_cached -= ent->size;
1336 free(ent);
1339 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1340 unsigned long *base_size, enum object_type *type)
1342 struct delta_base_cache_entry *ent;
1344 ent = get_delta_base_cache_entry(p, base_offset);
1345 if (!ent)
1346 return unpack_entry(p, base_offset, type, base_size);
1348 if (type)
1349 *type = ent->type;
1350 if (base_size)
1351 *base_size = ent->size;
1352 return xmemdupz(ent->data, ent->size);
1355 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1357 free(ent->data);
1358 detach_delta_base_cache_entry(ent);
1361 void clear_delta_base_cache(void)
1363 struct list_head *lru, *tmp;
1364 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1365 struct delta_base_cache_entry *entry =
1366 list_entry(lru, struct delta_base_cache_entry, lru);
1367 release_delta_base_cache(entry);
1371 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1372 void *base, unsigned long base_size, enum object_type type)
1374 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1375 struct list_head *lru, *tmp;
1377 delta_base_cached += base_size;
1379 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1380 struct delta_base_cache_entry *f =
1381 list_entry(lru, struct delta_base_cache_entry, lru);
1382 if (delta_base_cached <= delta_base_cache_limit)
1383 break;
1384 release_delta_base_cache(f);
1387 ent->key.p = p;
1388 ent->key.base_offset = base_offset;
1389 ent->type = type;
1390 ent->data = base;
1391 ent->size = base_size;
1392 list_add_tail(&ent->lru, &delta_base_cache_lru);
1394 if (!delta_base_cache.cmpfn)
1395 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1396 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1397 hashmap_add(&delta_base_cache, ent);
1400 int packed_object_info(struct packed_git *p, off_t obj_offset,
1401 struct object_info *oi)
1403 struct pack_window *w_curs = NULL;
1404 unsigned long size;
1405 off_t curpos = obj_offset;
1406 enum object_type type;
1409 * We always get the representation type, but only convert it to
1410 * a "real" type later if the caller is interested.
1412 if (oi->contentp) {
1413 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1414 &type);
1415 if (!*oi->contentp)
1416 type = OBJ_BAD;
1417 } else {
1418 type = unpack_object_header(p, &w_curs, &curpos, &size);
1421 if (!oi->contentp && oi->sizep) {
1422 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1423 off_t tmp_pos = curpos;
1424 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1425 type, obj_offset);
1426 if (!base_offset) {
1427 type = OBJ_BAD;
1428 goto out;
1430 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1431 if (*oi->sizep == 0) {
1432 type = OBJ_BAD;
1433 goto out;
1435 } else {
1436 *oi->sizep = size;
1440 if (oi->disk_sizep) {
1441 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1442 *oi->disk_sizep = revidx[1].offset - obj_offset;
1445 if (oi->typep || oi->typename) {
1446 enum object_type ptot;
1447 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
1448 curpos);
1449 if (oi->typep)
1450 *oi->typep = ptot;
1451 if (oi->typename) {
1452 const char *tn = typename(ptot);
1453 if (tn)
1454 strbuf_addstr(oi->typename, tn);
1456 if (ptot < 0) {
1457 type = OBJ_BAD;
1458 goto out;
1462 if (oi->delta_base_sha1) {
1463 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1464 const unsigned char *base;
1466 base = get_delta_base_sha1(p, &w_curs, curpos,
1467 type, obj_offset);
1468 if (!base) {
1469 type = OBJ_BAD;
1470 goto out;
1473 hashcpy(oi->delta_base_sha1, base);
1474 } else
1475 hashclr(oi->delta_base_sha1);
1478 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1479 OI_PACKED;
1481 out:
1482 unuse_pack(&w_curs);
1483 return type;
1486 static void *unpack_compressed_entry(struct packed_git *p,
1487 struct pack_window **w_curs,
1488 off_t curpos,
1489 unsigned long size)
1491 int st;
1492 git_zstream stream;
1493 unsigned char *buffer, *in;
1495 buffer = xmallocz_gently(size);
1496 if (!buffer)
1497 return NULL;
1498 memset(&stream, 0, sizeof(stream));
1499 stream.next_out = buffer;
1500 stream.avail_out = size + 1;
1502 git_inflate_init(&stream);
1503 do {
1504 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1505 stream.next_in = in;
1506 st = git_inflate(&stream, Z_FINISH);
1507 if (!stream.avail_out)
1508 break; /* the payload is larger than it should be */
1509 curpos += stream.next_in - in;
1510 } while (st == Z_OK || st == Z_BUF_ERROR);
1511 git_inflate_end(&stream);
1512 if ((st != Z_STREAM_END) || stream.total_out != size) {
1513 free(buffer);
1514 return NULL;
1517 return buffer;
1520 static void *read_object(const unsigned char *sha1, enum object_type *type,
1521 unsigned long *size);
1523 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1525 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1526 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1527 p->pack_name, (uintmax_t)obj_offset);
1530 int do_check_packed_object_crc;
1532 #define UNPACK_ENTRY_STACK_PREALLOC 64
1533 struct unpack_entry_stack_ent {
1534 off_t obj_offset;
1535 off_t curpos;
1536 unsigned long size;
1539 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1540 enum object_type *final_type, unsigned long *final_size)
1542 struct pack_window *w_curs = NULL;
1543 off_t curpos = obj_offset;
1544 void *data = NULL;
1545 unsigned long size;
1546 enum object_type type;
1547 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1548 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1549 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1550 int base_from_cache = 0;
1552 write_pack_access_log(p, obj_offset);
1554 /* PHASE 1: drill down to the innermost base object */
1555 for (;;) {
1556 off_t base_offset;
1557 int i;
1558 struct delta_base_cache_entry *ent;
1560 ent = get_delta_base_cache_entry(p, curpos);
1561 if (ent) {
1562 type = ent->type;
1563 data = ent->data;
1564 size = ent->size;
1565 detach_delta_base_cache_entry(ent);
1566 base_from_cache = 1;
1567 break;
1570 if (do_check_packed_object_crc && p->index_version > 1) {
1571 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1572 off_t len = revidx[1].offset - obj_offset;
1573 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1574 const unsigned char *sha1 =
1575 nth_packed_object_sha1(p, revidx->nr);
1576 error("bad packed object CRC for %s",
1577 sha1_to_hex(sha1));
1578 mark_bad_packed_object(p, sha1);
1579 data = NULL;
1580 goto out;
1584 type = unpack_object_header(p, &w_curs, &curpos, &size);
1585 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1586 break;
1588 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1589 if (!base_offset) {
1590 error("failed to validate delta base reference "
1591 "at offset %"PRIuMAX" from %s",
1592 (uintmax_t)curpos, p->pack_name);
1593 /* bail to phase 2, in hopes of recovery */
1594 data = NULL;
1595 break;
1598 /* push object, proceed to base */
1599 if (delta_stack_nr >= delta_stack_alloc
1600 && delta_stack == small_delta_stack) {
1601 delta_stack_alloc = alloc_nr(delta_stack_nr);
1602 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1603 memcpy(delta_stack, small_delta_stack,
1604 sizeof(*delta_stack)*delta_stack_nr);
1605 } else {
1606 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1608 i = delta_stack_nr++;
1609 delta_stack[i].obj_offset = obj_offset;
1610 delta_stack[i].curpos = curpos;
1611 delta_stack[i].size = size;
1613 curpos = obj_offset = base_offset;
1616 /* PHASE 2: handle the base */
1617 switch (type) {
1618 case OBJ_OFS_DELTA:
1619 case OBJ_REF_DELTA:
1620 if (data)
1621 die("BUG: unpack_entry: left loop at a valid delta");
1622 break;
1623 case OBJ_COMMIT:
1624 case OBJ_TREE:
1625 case OBJ_BLOB:
1626 case OBJ_TAG:
1627 if (!base_from_cache)
1628 data = unpack_compressed_entry(p, &w_curs, curpos, size);
1629 break;
1630 default:
1631 data = NULL;
1632 error("unknown object type %i at offset %"PRIuMAX" in %s",
1633 type, (uintmax_t)obj_offset, p->pack_name);
1636 /* PHASE 3: apply deltas in order */
1638 /* invariants:
1639 * 'data' holds the base data, or NULL if there was corruption
1641 while (delta_stack_nr) {
1642 void *delta_data;
1643 void *base = data;
1644 void *external_base = NULL;
1645 unsigned long delta_size, base_size = size;
1646 int i;
1648 data = NULL;
1650 if (base)
1651 add_delta_base_cache(p, obj_offset, base, base_size, type);
1653 if (!base) {
1655 * We're probably in deep shit, but let's try to fetch
1656 * the required base anyway from another pack or loose.
1657 * This is costly but should happen only in the presence
1658 * of a corrupted pack, and is better than failing outright.
1660 struct revindex_entry *revidx;
1661 const unsigned char *base_sha1;
1662 revidx = find_pack_revindex(p, obj_offset);
1663 if (revidx) {
1664 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1665 error("failed to read delta base object %s"
1666 " at offset %"PRIuMAX" from %s",
1667 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
1668 p->pack_name);
1669 mark_bad_packed_object(p, base_sha1);
1670 base = read_object(base_sha1, &type, &base_size);
1671 external_base = base;
1675 i = --delta_stack_nr;
1676 obj_offset = delta_stack[i].obj_offset;
1677 curpos = delta_stack[i].curpos;
1678 delta_size = delta_stack[i].size;
1680 if (!base)
1681 continue;
1683 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1685 if (!delta_data) {
1686 error("failed to unpack compressed delta "
1687 "at offset %"PRIuMAX" from %s",
1688 (uintmax_t)curpos, p->pack_name);
1689 data = NULL;
1690 free(external_base);
1691 continue;
1694 data = patch_delta(base, base_size,
1695 delta_data, delta_size,
1696 &size);
1699 * We could not apply the delta; warn the user, but keep going.
1700 * Our failure will be noticed either in the next iteration of
1701 * the loop, or if this is the final delta, in the caller when
1702 * we return NULL. Those code paths will take care of making
1703 * a more explicit warning and retrying with another copy of
1704 * the object.
1706 if (!data)
1707 error("failed to apply delta");
1709 free(delta_data);
1710 free(external_base);
1713 if (final_type)
1714 *final_type = type;
1715 if (final_size)
1716 *final_size = size;
1718 out:
1719 unuse_pack(&w_curs);
1721 if (delta_stack != small_delta_stack)
1722 free(delta_stack);
1724 return data;
1727 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1728 uint32_t n)
1730 const unsigned char *index = p->index_data;
1731 if (!index) {
1732 if (open_pack_index(p))
1733 return NULL;
1734 index = p->index_data;
1736 if (n >= p->num_objects)
1737 return NULL;
1738 index += 4 * 256;
1739 if (p->index_version == 1) {
1740 return index + 24 * n + 4;
1741 } else {
1742 index += 8;
1743 return index + 20 * n;
1747 const struct object_id *nth_packed_object_oid(struct object_id *oid,
1748 struct packed_git *p,
1749 uint32_t n)
1751 const unsigned char *hash = nth_packed_object_sha1(p, n);
1752 if (!hash)
1753 return NULL;
1754 hashcpy(oid->hash, hash);
1755 return oid;
1758 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1760 const unsigned char *ptr = vptr;
1761 const unsigned char *start = p->index_data;
1762 const unsigned char *end = start + p->index_size;
1763 if (ptr < start)
1764 die(_("offset before start of pack index for %s (corrupt index?)"),
1765 p->pack_name);
1766 /* No need to check for underflow; .idx files must be at least 8 bytes */
1767 if (ptr >= end - 8)
1768 die(_("offset beyond end of pack index for %s (truncated index?)"),
1769 p->pack_name);
1772 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1774 const unsigned char *index = p->index_data;
1775 index += 4 * 256;
1776 if (p->index_version == 1) {
1777 return ntohl(*((uint32_t *)(index + 24 * n)));
1778 } else {
1779 uint32_t off;
1780 index += 8 + p->num_objects * (20 + 4);
1781 off = ntohl(*((uint32_t *)(index + 4 * n)));
1782 if (!(off & 0x80000000))
1783 return off;
1784 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1785 check_pack_index_ptr(p, index);
1786 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1787 ntohl(*((uint32_t *)(index + 4)));
1791 off_t find_pack_entry_one(const unsigned char *sha1,
1792 struct packed_git *p)
1794 const uint32_t *level1_ofs = p->index_data;
1795 const unsigned char *index = p->index_data;
1796 unsigned hi, lo, stride;
1797 static int debug_lookup = -1;
1799 if (debug_lookup < 0)
1800 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1802 if (!index) {
1803 if (open_pack_index(p))
1804 return 0;
1805 level1_ofs = p->index_data;
1806 index = p->index_data;
1808 if (p->index_version > 1) {
1809 level1_ofs += 2;
1810 index += 8;
1812 index += 4 * 256;
1813 hi = ntohl(level1_ofs[*sha1]);
1814 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1815 if (p->index_version > 1) {
1816 stride = 20;
1817 } else {
1818 stride = 24;
1819 index += 4;
1822 if (debug_lookup)
1823 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1824 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1826 while (lo < hi) {
1827 unsigned mi = (lo + hi) / 2;
1828 int cmp = hashcmp(index + mi * stride, sha1);
1830 if (debug_lookup)
1831 printf("lo %u hi %u rg %u mi %u\n",
1832 lo, hi, hi - lo, mi);
1833 if (!cmp)
1834 return nth_packed_object_offset(p, mi);
1835 if (cmp > 0)
1836 hi = mi;
1837 else
1838 lo = mi+1;
1840 return 0;
1843 int is_pack_valid(struct packed_git *p)
1845 /* An already open pack is known to be valid. */
1846 if (p->pack_fd != -1)
1847 return 1;
1849 /* If the pack has one window completely covering the
1850 * file size, the pack is known to be valid even if
1851 * the descriptor is not currently open.
1853 if (p->windows) {
1854 struct pack_window *w = p->windows;
1856 if (!w->offset && w->len == p->pack_size)
1857 return 1;
1860 /* Force the pack to open to prove its valid. */
1861 return !open_packed_git(p);
1864 static int fill_pack_entry(const unsigned char *sha1,
1865 struct pack_entry *e,
1866 struct packed_git *p)
1868 off_t offset;
1870 if (p->num_bad_objects) {
1871 unsigned i;
1872 for (i = 0; i < p->num_bad_objects; i++)
1873 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1874 return 0;
1877 offset = find_pack_entry_one(sha1, p);
1878 if (!offset)
1879 return 0;
1882 * We are about to tell the caller where they can locate the
1883 * requested object. We better make sure the packfile is
1884 * still here and can be accessed before supplying that
1885 * answer, as it may have been deleted since the index was
1886 * loaded!
1888 if (!is_pack_valid(p))
1889 return 0;
1890 e->offset = offset;
1891 e->p = p;
1892 hashcpy(e->sha1, sha1);
1893 return 1;
1897 * Iff a pack file contains the object named by sha1, return true and
1898 * store its location to e.
1900 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1902 struct mru_entry *p;
1904 prepare_packed_git();
1905 if (!packed_git)
1906 return 0;
1908 for (p = packed_git_mru->head; p; p = p->next) {
1909 if (fill_pack_entry(sha1, e, p->item)) {
1910 mru_mark(packed_git_mru, p);
1911 return 1;
1914 return 0;
1917 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1918 struct packed_git *packs)
1920 struct packed_git *p;
1922 for (p = packs; p; p = p->next) {
1923 if (find_pack_entry_one(sha1, p))
1924 return p;
1926 return NULL;
1930 static int sha1_loose_object_info(const unsigned char *sha1,
1931 struct object_info *oi,
1932 int flags)
1934 int status = 0;
1935 unsigned long mapsize;
1936 void *map;
1937 git_zstream stream;
1938 char hdr[32];
1939 struct strbuf hdrbuf = STRBUF_INIT;
1940 unsigned long size_scratch;
1942 if (oi->delta_base_sha1)
1943 hashclr(oi->delta_base_sha1);
1946 * If we don't care about type or size, then we don't
1947 * need to look inside the object at all. Note that we
1948 * do not optimize out the stat call, even if the
1949 * caller doesn't care about the disk-size, since our
1950 * return value implicitly indicates whether the
1951 * object even exists.
1953 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1954 const char *path;
1955 struct stat st;
1956 if (stat_sha1_file(sha1, &st, &path) < 0)
1957 return -1;
1958 if (oi->disk_sizep)
1959 *oi->disk_sizep = st.st_size;
1960 return 0;
1963 map = map_sha1_file(sha1, &mapsize);
1964 if (!map)
1965 return -1;
1967 if (!oi->sizep)
1968 oi->sizep = &size_scratch;
1970 if (oi->disk_sizep)
1971 *oi->disk_sizep = mapsize;
1972 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1973 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1974 status = error("unable to unpack %s header with --allow-unknown-type",
1975 sha1_to_hex(sha1));
1976 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1977 status = error("unable to unpack %s header",
1978 sha1_to_hex(sha1));
1979 if (status < 0)
1980 ; /* Do nothing */
1981 else if (hdrbuf.len) {
1982 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1983 status = error("unable to parse %s header with --allow-unknown-type",
1984 sha1_to_hex(sha1));
1985 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1986 status = error("unable to parse %s header", sha1_to_hex(sha1));
1988 if (status >= 0 && oi->contentp)
1989 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1990 *oi->sizep, sha1);
1991 else
1992 git_inflate_end(&stream);
1994 munmap(map, mapsize);
1995 if (status && oi->typep)
1996 *oi->typep = status;
1997 if (oi->sizep == &size_scratch)
1998 oi->sizep = NULL;
1999 strbuf_release(&hdrbuf);
2000 oi->whence = OI_LOOSE;
2001 return (status < 0) ? status : 0;
2004 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2006 static struct object_info blank_oi = OBJECT_INFO_INIT;
2007 struct pack_entry e;
2008 int rtype;
2009 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2010 lookup_replace_object(sha1) :
2011 sha1;
2013 if (!oi)
2014 oi = &blank_oi;
2016 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2017 struct cached_object *co = find_cached_object(real);
2018 if (co) {
2019 if (oi->typep)
2020 *(oi->typep) = co->type;
2021 if (oi->sizep)
2022 *(oi->sizep) = co->size;
2023 if (oi->disk_sizep)
2024 *(oi->disk_sizep) = 0;
2025 if (oi->delta_base_sha1)
2026 hashclr(oi->delta_base_sha1);
2027 if (oi->typename)
2028 strbuf_addstr(oi->typename, typename(co->type));
2029 if (oi->contentp)
2030 *oi->contentp = xmemdupz(co->buf, co->size);
2031 oi->whence = OI_CACHED;
2032 return 0;
2036 if (!find_pack_entry(real, &e)) {
2037 /* Most likely it's a loose object. */
2038 if (!sha1_loose_object_info(real, oi, flags))
2039 return 0;
2041 /* Not a loose object; someone else may have just packed it. */
2042 if (flags & OBJECT_INFO_QUICK) {
2043 return -1;
2044 } else {
2045 reprepare_packed_git();
2046 if (!find_pack_entry(real, &e))
2047 return -1;
2051 if (oi == &blank_oi)
2053 * We know that the caller doesn't actually need the
2054 * information below, so return early.
2056 return 0;
2058 rtype = packed_object_info(e.p, e.offset, oi);
2059 if (rtype < 0) {
2060 mark_bad_packed_object(e.p, real);
2061 return sha1_object_info_extended(real, oi, 0);
2062 } else if (oi->whence == OI_PACKED) {
2063 oi->u.packed.offset = e.offset;
2064 oi->u.packed.pack = e.p;
2065 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2066 rtype == OBJ_OFS_DELTA);
2069 return 0;
2072 /* returns enum object_type or negative */
2073 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2075 enum object_type type;
2076 struct object_info oi = OBJECT_INFO_INIT;
2078 oi.typep = &type;
2079 oi.sizep = sizep;
2080 if (sha1_object_info_extended(sha1, &oi,
2081 OBJECT_INFO_LOOKUP_REPLACE) < 0)
2082 return -1;
2083 return type;
2086 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2087 unsigned char *sha1)
2089 struct cached_object *co;
2091 hash_sha1_file(buf, len, typename(type), sha1);
2092 if (has_sha1_file(sha1) || find_cached_object(sha1))
2093 return 0;
2094 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2095 co = &cached_objects[cached_object_nr++];
2096 co->size = len;
2097 co->type = type;
2098 co->buf = xmalloc(len);
2099 memcpy(co->buf, buf, len);
2100 hashcpy(co->sha1, sha1);
2101 return 0;
2104 static void *read_object(const unsigned char *sha1, enum object_type *type,
2105 unsigned long *size)
2107 struct object_info oi = OBJECT_INFO_INIT;
2108 void *content;
2109 oi.typep = type;
2110 oi.sizep = size;
2111 oi.contentp = &content;
2113 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
2114 return NULL;
2115 return content;
2119 * This function dies on corrupt objects; the callers who want to
2120 * deal with them should arrange to call read_object() and give error
2121 * messages themselves.
2123 void *read_sha1_file_extended(const unsigned char *sha1,
2124 enum object_type *type,
2125 unsigned long *size,
2126 int lookup_replace)
2128 void *data;
2129 const struct packed_git *p;
2130 const char *path;
2131 struct stat st;
2132 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
2133 : sha1;
2135 errno = 0;
2136 data = read_object(repl, type, size);
2137 if (data)
2138 return data;
2140 if (errno && errno != ENOENT)
2141 die_errno("failed to read object %s", sha1_to_hex(sha1));
2143 /* die if we replaced an object with one that does not exist */
2144 if (repl != sha1)
2145 die("replacement %s not found for %s",
2146 sha1_to_hex(repl), sha1_to_hex(sha1));
2148 if (!stat_sha1_file(repl, &st, &path))
2149 die("loose object %s (stored in %s) is corrupt",
2150 sha1_to_hex(repl), path);
2152 if ((p = has_packed_and_bad(repl)) != NULL)
2153 die("packed object %s (stored in %s) is corrupt",
2154 sha1_to_hex(repl), p->pack_name);
2156 return NULL;
2159 void *read_object_with_reference(const unsigned char *sha1,
2160 const char *required_type_name,
2161 unsigned long *size,
2162 unsigned char *actual_sha1_return)
2164 enum object_type type, required_type;
2165 void *buffer;
2166 unsigned long isize;
2167 unsigned char actual_sha1[20];
2169 required_type = type_from_string(required_type_name);
2170 hashcpy(actual_sha1, sha1);
2171 while (1) {
2172 int ref_length = -1;
2173 const char *ref_type = NULL;
2175 buffer = read_sha1_file(actual_sha1, &type, &isize);
2176 if (!buffer)
2177 return NULL;
2178 if (type == required_type) {
2179 *size = isize;
2180 if (actual_sha1_return)
2181 hashcpy(actual_sha1_return, actual_sha1);
2182 return buffer;
2184 /* Handle references */
2185 else if (type == OBJ_COMMIT)
2186 ref_type = "tree ";
2187 else if (type == OBJ_TAG)
2188 ref_type = "object ";
2189 else {
2190 free(buffer);
2191 return NULL;
2193 ref_length = strlen(ref_type);
2195 if (ref_length + 40 > isize ||
2196 memcmp(buffer, ref_type, ref_length) ||
2197 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2198 free(buffer);
2199 return NULL;
2201 free(buffer);
2202 /* Now we have the ID of the referred-to object in
2203 * actual_sha1. Check again. */
2207 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2208 const char *type, unsigned char *sha1,
2209 char *hdr, int *hdrlen)
2211 git_SHA_CTX c;
2213 /* Generate the header */
2214 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2216 /* Sha1.. */
2217 git_SHA1_Init(&c);
2218 git_SHA1_Update(&c, hdr, *hdrlen);
2219 git_SHA1_Update(&c, buf, len);
2220 git_SHA1_Final(sha1, &c);
2224 * Move the just written object into its final resting place.
2226 int finalize_object_file(const char *tmpfile, const char *filename)
2228 int ret = 0;
2230 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2231 goto try_rename;
2232 else if (link(tmpfile, filename))
2233 ret = errno;
2236 * Coda hack - coda doesn't like cross-directory links,
2237 * so we fall back to a rename, which will mean that it
2238 * won't be able to check collisions, but that's not a
2239 * big deal.
2241 * The same holds for FAT formatted media.
2243 * When this succeeds, we just return. We have nothing
2244 * left to unlink.
2246 if (ret && ret != EEXIST) {
2247 try_rename:
2248 if (!rename(tmpfile, filename))
2249 goto out;
2250 ret = errno;
2252 unlink_or_warn(tmpfile);
2253 if (ret) {
2254 if (ret != EEXIST) {
2255 return error_errno("unable to write sha1 filename %s", filename);
2257 /* FIXME!!! Collision check here ? */
2260 out:
2261 if (adjust_shared_perm(filename))
2262 return error("unable to set permission to '%s'", filename);
2263 return 0;
2266 static int write_buffer(int fd, const void *buf, size_t len)
2268 if (write_in_full(fd, buf, len) < 0)
2269 return error_errno("file write error");
2270 return 0;
2273 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2274 unsigned char *sha1)
2276 char hdr[32];
2277 int hdrlen = sizeof(hdr);
2278 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2279 return 0;
2282 /* Finalize a file on disk, and close it. */
2283 static void close_sha1_file(int fd)
2285 if (fsync_object_files)
2286 fsync_or_die(fd, "sha1 file");
2287 if (close(fd) != 0)
2288 die_errno("error when closing sha1 file");
2291 /* Size of directory component, including the ending '/' */
2292 static inline int directory_size(const char *filename)
2294 const char *s = strrchr(filename, '/');
2295 if (!s)
2296 return 0;
2297 return s - filename + 1;
2301 * This creates a temporary file in the same directory as the final
2302 * 'filename'
2304 * We want to avoid cross-directory filename renames, because those
2305 * can have problems on various filesystems (FAT, NFS, Coda).
2307 static int create_tmpfile(struct strbuf *tmp, const char *filename)
2309 int fd, dirlen = directory_size(filename);
2311 strbuf_reset(tmp);
2312 strbuf_add(tmp, filename, dirlen);
2313 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
2314 fd = git_mkstemp_mode(tmp->buf, 0444);
2315 if (fd < 0 && dirlen && errno == ENOENT) {
2317 * Make sure the directory exists; note that the contents
2318 * of the buffer are undefined after mkstemp returns an
2319 * error, so we have to rewrite the whole buffer from
2320 * scratch.
2322 strbuf_reset(tmp);
2323 strbuf_add(tmp, filename, dirlen - 1);
2324 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
2325 return -1;
2326 if (adjust_shared_perm(tmp->buf))
2327 return -1;
2329 /* Try again */
2330 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
2331 fd = git_mkstemp_mode(tmp->buf, 0444);
2333 return fd;
2336 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2337 const void *buf, unsigned long len, time_t mtime)
2339 int fd, ret;
2340 unsigned char compressed[4096];
2341 git_zstream stream;
2342 git_SHA_CTX c;
2343 unsigned char parano_sha1[20];
2344 static struct strbuf tmp_file = STRBUF_INIT;
2345 const char *filename = sha1_file_name(sha1);
2347 fd = create_tmpfile(&tmp_file, filename);
2348 if (fd < 0) {
2349 if (errno == EACCES)
2350 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2351 else
2352 return error_errno("unable to create temporary file");
2355 /* Set it up */
2356 git_deflate_init(&stream, zlib_compression_level);
2357 stream.next_out = compressed;
2358 stream.avail_out = sizeof(compressed);
2359 git_SHA1_Init(&c);
2361 /* First header.. */
2362 stream.next_in = (unsigned char *)hdr;
2363 stream.avail_in = hdrlen;
2364 while (git_deflate(&stream, 0) == Z_OK)
2365 ; /* nothing */
2366 git_SHA1_Update(&c, hdr, hdrlen);
2368 /* Then the data itself.. */
2369 stream.next_in = (void *)buf;
2370 stream.avail_in = len;
2371 do {
2372 unsigned char *in0 = stream.next_in;
2373 ret = git_deflate(&stream, Z_FINISH);
2374 git_SHA1_Update(&c, in0, stream.next_in - in0);
2375 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2376 die("unable to write sha1 file");
2377 stream.next_out = compressed;
2378 stream.avail_out = sizeof(compressed);
2379 } while (ret == Z_OK);
2381 if (ret != Z_STREAM_END)
2382 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2383 ret = git_deflate_end_gently(&stream);
2384 if (ret != Z_OK)
2385 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2386 git_SHA1_Final(parano_sha1, &c);
2387 if (hashcmp(sha1, parano_sha1) != 0)
2388 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2390 close_sha1_file(fd);
2392 if (mtime) {
2393 struct utimbuf utb;
2394 utb.actime = mtime;
2395 utb.modtime = mtime;
2396 if (utime(tmp_file.buf, &utb) < 0)
2397 warning_errno("failed utime() on %s", tmp_file.buf);
2400 return finalize_object_file(tmp_file.buf, filename);
2403 static int freshen_loose_object(const unsigned char *sha1)
2405 return check_and_freshen(sha1, 1);
2408 static int freshen_packed_object(const unsigned char *sha1)
2410 struct pack_entry e;
2411 if (!find_pack_entry(sha1, &e))
2412 return 0;
2413 if (e.p->freshened)
2414 return 1;
2415 if (!freshen_file(e.p->pack_name))
2416 return 0;
2417 e.p->freshened = 1;
2418 return 1;
2421 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
2423 char hdr[32];
2424 int hdrlen = sizeof(hdr);
2426 /* Normally if we have it in the pack then we do not bother writing
2427 * it out into .git/objects/??/?{38} file.
2429 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2430 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2431 return 0;
2432 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2435 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
2436 unsigned char *sha1, unsigned flags)
2438 char *header;
2439 int hdrlen, status = 0;
2441 /* type string, SP, %lu of the length plus NUL must fit this */
2442 hdrlen = strlen(type) + 32;
2443 header = xmalloc(hdrlen);
2444 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
2446 if (!(flags & HASH_WRITE_OBJECT))
2447 goto cleanup;
2448 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2449 goto cleanup;
2450 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
2452 cleanup:
2453 free(header);
2454 return status;
2457 int force_object_loose(const unsigned char *sha1, time_t mtime)
2459 void *buf;
2460 unsigned long len;
2461 enum object_type type;
2462 char hdr[32];
2463 int hdrlen;
2464 int ret;
2466 if (has_loose_object(sha1))
2467 return 0;
2468 buf = read_object(sha1, &type, &len);
2469 if (!buf)
2470 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2471 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
2472 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2473 free(buf);
2475 return ret;
2478 int has_pack_index(const unsigned char *sha1)
2480 struct stat st;
2481 if (stat(sha1_pack_index_name(sha1), &st))
2482 return 0;
2483 return 1;
2486 int has_sha1_pack(const unsigned char *sha1)
2488 struct pack_entry e;
2489 return find_pack_entry(sha1, &e);
2492 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
2494 if (!startup_info->have_repository)
2495 return 0;
2496 return sha1_object_info_extended(sha1, NULL,
2497 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
2500 int has_object_file(const struct object_id *oid)
2502 return has_sha1_file(oid->hash);
2505 int has_object_file_with_flags(const struct object_id *oid, int flags)
2507 return has_sha1_file_with_flags(oid->hash, flags);
2510 static void check_tree(const void *buf, size_t size)
2512 struct tree_desc desc;
2513 struct name_entry entry;
2515 init_tree_desc(&desc, buf, size);
2516 while (tree_entry(&desc, &entry))
2517 /* do nothing
2518 * tree_entry() will die() on malformed entries */
2522 static void check_commit(const void *buf, size_t size)
2524 struct commit c;
2525 memset(&c, 0, sizeof(c));
2526 if (parse_commit_buffer(&c, buf, size))
2527 die("corrupt commit");
2530 static void check_tag(const void *buf, size_t size)
2532 struct tag t;
2533 memset(&t, 0, sizeof(t));
2534 if (parse_tag_buffer(&t, buf, size))
2535 die("corrupt tag");
2538 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2539 enum object_type type,
2540 const char *path, unsigned flags)
2542 int ret, re_allocated = 0;
2543 int write_object = flags & HASH_WRITE_OBJECT;
2545 if (!type)
2546 type = OBJ_BLOB;
2549 * Convert blobs to git internal format
2551 if ((type == OBJ_BLOB) && path) {
2552 struct strbuf nbuf = STRBUF_INIT;
2553 if (convert_to_git(&the_index, path, buf, size, &nbuf,
2554 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2555 buf = strbuf_detach(&nbuf, &size);
2556 re_allocated = 1;
2559 if (flags & HASH_FORMAT_CHECK) {
2560 if (type == OBJ_TREE)
2561 check_tree(buf, size);
2562 if (type == OBJ_COMMIT)
2563 check_commit(buf, size);
2564 if (type == OBJ_TAG)
2565 check_tag(buf, size);
2568 if (write_object)
2569 ret = write_sha1_file(buf, size, typename(type), sha1);
2570 else
2571 ret = hash_sha1_file(buf, size, typename(type), sha1);
2572 if (re_allocated)
2573 free(buf);
2574 return ret;
2577 static int index_stream_convert_blob(unsigned char *sha1, int fd,
2578 const char *path, unsigned flags)
2580 int ret;
2581 const int write_object = flags & HASH_WRITE_OBJECT;
2582 struct strbuf sbuf = STRBUF_INIT;
2584 assert(path);
2585 assert(would_convert_to_git_filter_fd(path));
2587 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
2588 write_object ? safe_crlf : SAFE_CRLF_FALSE);
2590 if (write_object)
2591 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2592 sha1);
2593 else
2594 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2595 sha1);
2596 strbuf_release(&sbuf);
2597 return ret;
2600 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2601 const char *path, unsigned flags)
2603 struct strbuf sbuf = STRBUF_INIT;
2604 int ret;
2606 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2607 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2608 else
2609 ret = -1;
2610 strbuf_release(&sbuf);
2611 return ret;
2614 #define SMALL_FILE_SIZE (32*1024)
2616 static int index_core(unsigned char *sha1, int fd, size_t size,
2617 enum object_type type, const char *path,
2618 unsigned flags)
2620 int ret;
2622 if (!size) {
2623 ret = index_mem(sha1, "", size, type, path, flags);
2624 } else if (size <= SMALL_FILE_SIZE) {
2625 char *buf = xmalloc(size);
2626 if (size == read_in_full(fd, buf, size))
2627 ret = index_mem(sha1, buf, size, type, path, flags);
2628 else
2629 ret = error_errno("short read");
2630 free(buf);
2631 } else {
2632 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2633 ret = index_mem(sha1, buf, size, type, path, flags);
2634 munmap(buf, size);
2636 return ret;
2640 * This creates one packfile per large blob unless bulk-checkin
2641 * machinery is "plugged".
2643 * This also bypasses the usual "convert-to-git" dance, and that is on
2644 * purpose. We could write a streaming version of the converting
2645 * functions and insert that before feeding the data to fast-import
2646 * (or equivalent in-core API described above). However, that is
2647 * somewhat complicated, as we do not know the size of the filter
2648 * result, which we need to know beforehand when writing a git object.
2649 * Since the primary motivation for trying to stream from the working
2650 * tree file and to avoid mmaping it in core is to deal with large
2651 * binary blobs, they generally do not want to get any conversion, and
2652 * callers should avoid this code path when filters are requested.
2654 static int index_stream(unsigned char *sha1, int fd, size_t size,
2655 enum object_type type, const char *path,
2656 unsigned flags)
2658 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2661 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2662 enum object_type type, const char *path, unsigned flags)
2664 int ret;
2667 * Call xsize_t() only when needed to avoid potentially unnecessary
2668 * die() for large files.
2670 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
2671 ret = index_stream_convert_blob(sha1, fd, path, flags);
2672 else if (!S_ISREG(st->st_mode))
2673 ret = index_pipe(sha1, fd, type, path, flags);
2674 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2675 (path && would_convert_to_git(&the_index, path)))
2676 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
2677 flags);
2678 else
2679 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
2680 flags);
2681 close(fd);
2682 return ret;
2685 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2687 int fd;
2688 struct strbuf sb = STRBUF_INIT;
2690 switch (st->st_mode & S_IFMT) {
2691 case S_IFREG:
2692 fd = open(path, O_RDONLY);
2693 if (fd < 0)
2694 return error_errno("open(\"%s\")", path);
2695 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2696 return error("%s: failed to insert into database",
2697 path);
2698 break;
2699 case S_IFLNK:
2700 if (strbuf_readlink(&sb, path, st->st_size))
2701 return error_errno("readlink(\"%s\")", path);
2702 if (!(flags & HASH_WRITE_OBJECT))
2703 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2704 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2705 return error("%s: failed to insert into database",
2706 path);
2707 strbuf_release(&sb);
2708 break;
2709 case S_IFDIR:
2710 return resolve_gitlink_ref(path, "HEAD", sha1);
2711 default:
2712 return error("%s: unsupported file type", path);
2714 return 0;
2717 int read_pack_header(int fd, struct pack_header *header)
2719 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2720 /* "eof before pack header was fully read" */
2721 return PH_ERROR_EOF;
2723 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2724 /* "protocol error (pack signature mismatch detected)" */
2725 return PH_ERROR_PACK_SIGNATURE;
2726 if (!pack_version_ok(header->hdr_version))
2727 /* "protocol error (pack version unsupported)" */
2728 return PH_ERROR_PROTOCOL;
2729 return 0;
2732 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2734 enum object_type type = sha1_object_info(sha1, NULL);
2735 if (type < 0)
2736 die("%s is not a valid object", sha1_to_hex(sha1));
2737 if (type != expect)
2738 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2739 typename(expect));
2742 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2743 struct strbuf *path,
2744 each_loose_object_fn obj_cb,
2745 each_loose_cruft_fn cruft_cb,
2746 each_loose_subdir_fn subdir_cb,
2747 void *data)
2749 size_t origlen, baselen;
2750 DIR *dir;
2751 struct dirent *de;
2752 int r = 0;
2754 if (subdir_nr > 0xff)
2755 BUG("invalid loose object subdirectory: %x", subdir_nr);
2757 origlen = path->len;
2758 strbuf_complete(path, '/');
2759 strbuf_addf(path, "%02x", subdir_nr);
2760 baselen = path->len;
2762 dir = opendir(path->buf);
2763 if (!dir) {
2764 if (errno != ENOENT)
2765 r = error_errno("unable to open %s", path->buf);
2766 strbuf_setlen(path, origlen);
2767 return r;
2770 while ((de = readdir(dir))) {
2771 if (is_dot_or_dotdot(de->d_name))
2772 continue;
2774 strbuf_setlen(path, baselen);
2775 strbuf_addf(path, "/%s", de->d_name);
2777 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
2778 char hex[GIT_MAX_HEXSZ+1];
2779 struct object_id oid;
2781 xsnprintf(hex, sizeof(hex), "%02x%s",
2782 subdir_nr, de->d_name);
2783 if (!get_oid_hex(hex, &oid)) {
2784 if (obj_cb) {
2785 r = obj_cb(&oid, path->buf, data);
2786 if (r)
2787 break;
2789 continue;
2793 if (cruft_cb) {
2794 r = cruft_cb(de->d_name, path->buf, data);
2795 if (r)
2796 break;
2799 closedir(dir);
2801 strbuf_setlen(path, baselen);
2802 if (!r && subdir_cb)
2803 r = subdir_cb(subdir_nr, path->buf, data);
2805 strbuf_setlen(path, origlen);
2807 return r;
2810 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2811 each_loose_object_fn obj_cb,
2812 each_loose_cruft_fn cruft_cb,
2813 each_loose_subdir_fn subdir_cb,
2814 void *data)
2816 int r = 0;
2817 int i;
2819 for (i = 0; i < 256; i++) {
2820 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2821 subdir_cb, data);
2822 if (r)
2823 break;
2826 return r;
2829 int for_each_loose_file_in_objdir(const char *path,
2830 each_loose_object_fn obj_cb,
2831 each_loose_cruft_fn cruft_cb,
2832 each_loose_subdir_fn subdir_cb,
2833 void *data)
2835 struct strbuf buf = STRBUF_INIT;
2836 int r;
2838 strbuf_addstr(&buf, path);
2839 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2840 subdir_cb, data);
2841 strbuf_release(&buf);
2843 return r;
2846 struct loose_alt_odb_data {
2847 each_loose_object_fn *cb;
2848 void *data;
2851 static int loose_from_alt_odb(struct alternate_object_database *alt,
2852 void *vdata)
2854 struct loose_alt_odb_data *data = vdata;
2855 struct strbuf buf = STRBUF_INIT;
2856 int r;
2858 strbuf_addstr(&buf, alt->path);
2859 r = for_each_loose_file_in_objdir_buf(&buf,
2860 data->cb, NULL, NULL,
2861 data->data);
2862 strbuf_release(&buf);
2863 return r;
2866 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2868 struct loose_alt_odb_data alt;
2869 int r;
2871 r = for_each_loose_file_in_objdir(get_object_directory(),
2872 cb, NULL, NULL, data);
2873 if (r)
2874 return r;
2876 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2877 return 0;
2879 alt.cb = cb;
2880 alt.data = data;
2881 return foreach_alt_odb(loose_from_alt_odb, &alt);
2884 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
2886 uint32_t i;
2887 int r = 0;
2889 for (i = 0; i < p->num_objects; i++) {
2890 struct object_id oid;
2892 if (!nth_packed_object_oid(&oid, p, i))
2893 return error("unable to get sha1 of object %u in %s",
2894 i, p->pack_name);
2896 r = cb(&oid, p, i, data);
2897 if (r)
2898 break;
2900 return r;
2903 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
2905 struct packed_git *p;
2906 int r = 0;
2907 int pack_errors = 0;
2909 prepare_packed_git();
2910 for (p = packed_git; p; p = p->next) {
2911 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2912 continue;
2913 if (open_pack_index(p)) {
2914 pack_errors = 1;
2915 continue;
2917 r = for_each_object_in_pack(p, cb, data);
2918 if (r)
2919 break;
2921 return r ? r : pack_errors;
2924 static int check_stream_sha1(git_zstream *stream,
2925 const char *hdr,
2926 unsigned long size,
2927 const char *path,
2928 const unsigned char *expected_sha1)
2930 git_SHA_CTX c;
2931 unsigned char real_sha1[GIT_MAX_RAWSZ];
2932 unsigned char buf[4096];
2933 unsigned long total_read;
2934 int status = Z_OK;
2936 git_SHA1_Init(&c);
2937 git_SHA1_Update(&c, hdr, stream->total_out);
2940 * We already read some bytes into hdr, but the ones up to the NUL
2941 * do not count against the object's content size.
2943 total_read = stream->total_out - strlen(hdr) - 1;
2946 * This size comparison must be "<=" to read the final zlib packets;
2947 * see the comment in unpack_sha1_rest for details.
2949 while (total_read <= size &&
2950 (status == Z_OK || status == Z_BUF_ERROR)) {
2951 stream->next_out = buf;
2952 stream->avail_out = sizeof(buf);
2953 if (size - total_read < stream->avail_out)
2954 stream->avail_out = size - total_read;
2955 status = git_inflate(stream, Z_FINISH);
2956 git_SHA1_Update(&c, buf, stream->next_out - buf);
2957 total_read += stream->next_out - buf;
2959 git_inflate_end(stream);
2961 if (status != Z_STREAM_END) {
2962 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2963 return -1;
2965 if (stream->avail_in) {
2966 error("garbage at end of loose object '%s'",
2967 sha1_to_hex(expected_sha1));
2968 return -1;
2971 git_SHA1_Final(real_sha1, &c);
2972 if (hashcmp(expected_sha1, real_sha1)) {
2973 error("sha1 mismatch for %s (expected %s)", path,
2974 sha1_to_hex(expected_sha1));
2975 return -1;
2978 return 0;
2981 int read_loose_object(const char *path,
2982 const unsigned char *expected_sha1,
2983 enum object_type *type,
2984 unsigned long *size,
2985 void **contents)
2987 int ret = -1;
2988 void *map = NULL;
2989 unsigned long mapsize;
2990 git_zstream stream;
2991 char hdr[32];
2993 *contents = NULL;
2995 map = map_sha1_file_1(path, NULL, &mapsize);
2996 if (!map) {
2997 error_errno("unable to mmap %s", path);
2998 goto out;
3001 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
3002 error("unable to unpack header of %s", path);
3003 goto out;
3006 *type = parse_sha1_header(hdr, size);
3007 if (*type < 0) {
3008 error("unable to parse header of %s", path);
3009 git_inflate_end(&stream);
3010 goto out;
3013 if (*type == OBJ_BLOB) {
3014 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
3015 goto out;
3016 } else {
3017 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
3018 if (!*contents) {
3019 error("unable to unpack contents of %s", path);
3020 git_inflate_end(&stream);
3021 goto out;
3023 if (check_sha1_signature(expected_sha1, *contents,
3024 *size, typename(*type))) {
3025 error("sha1 mismatch for %s (expected %s)", path,
3026 sha1_to_hex(expected_sha1));
3027 free(*contents);
3028 goto out;
3032 ret = 0; /* everything checks out */
3034 out:
3035 if (map)
3036 munmap(map, mapsize);
3037 return ret;