fmt-merge-msg: avoid leaking strbuf in shortlog()
[git/debian.git] / sha1_file.c
blobd77b915db699f6f58731f645d6bcc188843d9eb8
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 "string-list.h"
11 #include "lockfile.h"
12 #include "delta.h"
13 #include "pack.h"
14 #include "blob.h"
15 #include "commit.h"
16 #include "run-command.h"
17 #include "tag.h"
18 #include "tree.h"
19 #include "tree-walk.h"
20 #include "refs.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
25 #include "dir.h"
26 #include "mru.h"
27 #include "list.h"
28 #include "mergesort.h"
29 #include "quote.h"
31 #define SZ_FMT PRIuMAX
32 static inline uintmax_t sz_fmt(size_t s) { return s; }
34 const unsigned char null_sha1[20];
35 const struct object_id null_oid;
36 const struct object_id empty_tree_oid = {
37 EMPTY_TREE_SHA1_BIN_LITERAL
39 const struct object_id empty_blob_oid = {
40 EMPTY_BLOB_SHA1_BIN_LITERAL
44 * This is meant to hold a *small* number of objects that you would
45 * want read_sha1_file() to be able to return, but yet you do not want
46 * to write them into the object store (e.g. a browse-only
47 * application).
49 static struct cached_object {
50 unsigned char sha1[20];
51 enum object_type type;
52 void *buf;
53 unsigned long size;
54 } *cached_objects;
55 static int cached_object_nr, cached_object_alloc;
57 static struct cached_object empty_tree = {
58 EMPTY_TREE_SHA1_BIN_LITERAL,
59 OBJ_TREE,
60 "",
64 static struct cached_object *find_cached_object(const unsigned char *sha1)
66 int i;
67 struct cached_object *co = cached_objects;
69 for (i = 0; i < cached_object_nr; i++, co++) {
70 if (!hashcmp(co->sha1, sha1))
71 return co;
73 if (!hashcmp(sha1, empty_tree.sha1))
74 return &empty_tree;
75 return NULL;
78 int mkdir_in_gitdir(const char *path)
80 if (mkdir(path, 0777)) {
81 int saved_errno = errno;
82 struct stat st;
83 struct strbuf sb = STRBUF_INIT;
85 if (errno != EEXIST)
86 return -1;
88 * Are we looking at a path in a symlinked worktree
89 * whose original repository does not yet have it?
90 * e.g. .git/rr-cache pointing at its original
91 * repository in which the user hasn't performed any
92 * conflict resolution yet?
94 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
95 strbuf_readlink(&sb, path, st.st_size) ||
96 !is_absolute_path(sb.buf) ||
97 mkdir(sb.buf, 0777)) {
98 strbuf_release(&sb);
99 errno = saved_errno;
100 return -1;
102 strbuf_release(&sb);
104 return adjust_shared_perm(path);
107 enum scld_error safe_create_leading_directories(char *path)
109 char *next_component = path + offset_1st_component(path);
110 enum scld_error ret = SCLD_OK;
112 while (ret == SCLD_OK && next_component) {
113 struct stat st;
114 char *slash = next_component, slash_character;
116 while (*slash && !is_dir_sep(*slash))
117 slash++;
119 if (!*slash)
120 break;
122 next_component = slash + 1;
123 while (is_dir_sep(*next_component))
124 next_component++;
125 if (!*next_component)
126 break;
128 slash_character = *slash;
129 *slash = '\0';
130 if (!stat(path, &st)) {
131 /* path exists */
132 if (!S_ISDIR(st.st_mode))
133 ret = SCLD_EXISTS;
134 } else if (mkdir(path, 0777)) {
135 if (errno == EEXIST &&
136 !stat(path, &st) && S_ISDIR(st.st_mode))
137 ; /* somebody created it since we checked */
138 else if (errno == ENOENT)
140 * Either mkdir() failed because
141 * somebody just pruned the containing
142 * directory, or stat() failed because
143 * the file that was in our way was
144 * just removed. Either way, inform
145 * the caller that it might be worth
146 * trying again:
148 ret = SCLD_VANISHED;
149 else
150 ret = SCLD_FAILED;
151 } else if (adjust_shared_perm(path)) {
152 ret = SCLD_PERMS;
154 *slash = slash_character;
156 return ret;
159 enum scld_error safe_create_leading_directories_const(const char *path)
161 /* path points to cache entries, so xstrdup before messing with it */
162 char *buf = xstrdup(path);
163 enum scld_error result = safe_create_leading_directories(buf);
164 free(buf);
165 return result;
168 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
170 int i;
171 for (i = 0; i < 20; i++) {
172 static char hex[] = "0123456789abcdef";
173 unsigned int val = sha1[i];
174 strbuf_addch(buf, hex[val >> 4]);
175 strbuf_addch(buf, hex[val & 0xf]);
176 if (!i)
177 strbuf_addch(buf, '/');
181 const char *sha1_file_name(const unsigned char *sha1)
183 static struct strbuf buf = STRBUF_INIT;
185 strbuf_reset(&buf);
186 strbuf_addf(&buf, "%s/", get_object_directory());
188 fill_sha1_path(&buf, sha1);
189 return buf.buf;
192 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
194 strbuf_setlen(&alt->scratch, alt->base_len);
195 return &alt->scratch;
198 static const char *alt_sha1_path(struct alternate_object_database *alt,
199 const unsigned char *sha1)
201 struct strbuf *buf = alt_scratch_buf(alt);
202 fill_sha1_path(buf, sha1);
203 return buf->buf;
206 char *odb_pack_name(struct strbuf *buf,
207 const unsigned char *sha1,
208 const char *ext)
210 strbuf_reset(buf);
211 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
212 sha1_to_hex(sha1), ext);
213 return buf->buf;
216 char *sha1_pack_name(const unsigned char *sha1)
218 static struct strbuf buf = STRBUF_INIT;
219 return odb_pack_name(&buf, sha1, "pack");
222 char *sha1_pack_index_name(const unsigned char *sha1)
224 static struct strbuf buf = STRBUF_INIT;
225 return odb_pack_name(&buf, sha1, "idx");
228 struct alternate_object_database *alt_odb_list;
229 static struct alternate_object_database **alt_odb_tail;
232 * Return non-zero iff the path is usable as an alternate object database.
234 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
236 struct alternate_object_database *alt;
238 /* Detect cases where alternate disappeared */
239 if (!is_directory(path->buf)) {
240 error("object directory %s does not exist; "
241 "check .git/objects/info/alternates.",
242 path->buf);
243 return 0;
247 * Prevent the common mistake of listing the same
248 * thing twice, or object directory itself.
250 for (alt = alt_odb_list; alt; alt = alt->next) {
251 if (!fspathcmp(path->buf, alt->path))
252 return 0;
254 if (!fspathcmp(path->buf, normalized_objdir))
255 return 0;
257 return 1;
261 * Prepare alternate object database registry.
263 * The variable alt_odb_list points at the list of struct
264 * alternate_object_database. The elements on this list come from
265 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
266 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
267 * whose contents is similar to that environment variable but can be
268 * LF separated. Its base points at a statically allocated buffer that
269 * contains "/the/directory/corresponding/to/.git/objects/...", while
270 * its name points just after the slash at the end of ".git/objects/"
271 * in the example above, and has enough space to hold 40-byte hex
272 * SHA1, an extra slash for the first level indirection, and the
273 * terminating NUL.
275 static int link_alt_odb_entry(const char *entry, const char *relative_base,
276 int depth, const char *normalized_objdir)
278 struct alternate_object_database *ent;
279 struct strbuf pathbuf = STRBUF_INIT;
281 if (!is_absolute_path(entry) && relative_base) {
282 strbuf_realpath(&pathbuf, relative_base, 1);
283 strbuf_addch(&pathbuf, '/');
285 strbuf_addstr(&pathbuf, entry);
287 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
288 error("unable to normalize alternate object path: %s",
289 pathbuf.buf);
290 strbuf_release(&pathbuf);
291 return -1;
295 * The trailing slash after the directory name is given by
296 * this function at the end. Remove duplicates.
298 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
299 strbuf_setlen(&pathbuf, pathbuf.len - 1);
301 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
302 strbuf_release(&pathbuf);
303 return -1;
306 ent = alloc_alt_odb(pathbuf.buf);
308 /* add the alternate entry */
309 *alt_odb_tail = ent;
310 alt_odb_tail = &(ent->next);
311 ent->next = NULL;
313 /* recursively add alternates */
314 read_info_alternates(pathbuf.buf, depth + 1);
316 strbuf_release(&pathbuf);
317 return 0;
320 static const char *parse_alt_odb_entry(const char *string,
321 int sep,
322 struct strbuf *out)
324 const char *end;
326 strbuf_reset(out);
328 if (*string == '#') {
329 /* comment; consume up to next separator */
330 end = strchrnul(string, sep);
331 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
333 * quoted path; unquote_c_style has copied the
334 * data for us and set "end". Broken quoting (e.g.,
335 * an entry that doesn't end with a quote) falls
336 * back to the unquoted case below.
338 } else {
339 /* normal, unquoted path */
340 end = strchrnul(string, sep);
341 strbuf_add(out, string, end - string);
344 if (*end)
345 end++;
346 return end;
349 static void link_alt_odb_entries(const char *alt, int len, int sep,
350 const char *relative_base, int depth)
352 struct strbuf objdirbuf = STRBUF_INIT;
353 struct strbuf entry = STRBUF_INIT;
355 if (depth > 5) {
356 error("%s: ignoring alternate object stores, nesting too deep.",
357 relative_base);
358 return;
361 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
362 if (strbuf_normalize_path(&objdirbuf) < 0)
363 die("unable to normalize object directory: %s",
364 objdirbuf.buf);
366 while (*alt) {
367 alt = parse_alt_odb_entry(alt, sep, &entry);
368 if (!entry.len)
369 continue;
370 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
372 strbuf_release(&entry);
373 strbuf_release(&objdirbuf);
376 void read_info_alternates(const char * relative_base, int depth)
378 char *map;
379 size_t mapsz;
380 struct stat st;
381 char *path;
382 int fd;
384 path = xstrfmt("%s/info/alternates", relative_base);
385 fd = git_open(path);
386 free(path);
387 if (fd < 0)
388 return;
389 if (fstat(fd, &st) || (st.st_size == 0)) {
390 close(fd);
391 return;
393 mapsz = xsize_t(st.st_size);
394 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
395 close(fd);
397 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
399 munmap(map, mapsz);
402 struct alternate_object_database *alloc_alt_odb(const char *dir)
404 struct alternate_object_database *ent;
406 FLEX_ALLOC_STR(ent, path, dir);
407 strbuf_init(&ent->scratch, 0);
408 strbuf_addf(&ent->scratch, "%s/", dir);
409 ent->base_len = ent->scratch.len;
411 return ent;
414 void add_to_alternates_file(const char *reference)
416 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
417 char *alts = git_pathdup("objects/info/alternates");
418 FILE *in, *out;
420 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
421 out = fdopen_lock_file(lock, "w");
422 if (!out)
423 die_errno("unable to fdopen alternates lockfile");
425 in = fopen(alts, "r");
426 if (in) {
427 struct strbuf line = STRBUF_INIT;
428 int found = 0;
430 while (strbuf_getline(&line, in) != EOF) {
431 if (!strcmp(reference, line.buf)) {
432 found = 1;
433 break;
435 fprintf_or_die(out, "%s\n", line.buf);
438 strbuf_release(&line);
439 fclose(in);
441 if (found) {
442 rollback_lock_file(lock);
443 lock = NULL;
446 else if (errno != ENOENT)
447 die_errno("unable to read alternates file");
449 if (lock) {
450 fprintf_or_die(out, "%s\n", reference);
451 if (commit_lock_file(lock))
452 die_errno("unable to move new alternates file into place");
453 if (alt_odb_tail)
454 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
456 free(alts);
459 void add_to_alternates_memory(const char *reference)
462 * Make sure alternates are initialized, or else our entry may be
463 * overwritten when they are.
465 prepare_alt_odb();
467 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
471 * Compute the exact path an alternate is at and returns it. In case of
472 * error NULL is returned and the human readable error is added to `err`
473 * `path` may be relative and should point to $GITDIR.
474 * `err` must not be null.
476 char *compute_alternate_path(const char *path, struct strbuf *err)
478 char *ref_git = NULL;
479 const char *repo, *ref_git_s;
480 int seen_error = 0;
482 ref_git_s = real_path_if_valid(path);
483 if (!ref_git_s) {
484 seen_error = 1;
485 strbuf_addf(err, _("path '%s' does not exist"), path);
486 goto out;
487 } else
489 * Beware: read_gitfile(), real_path() and mkpath()
490 * return static buffer
492 ref_git = xstrdup(ref_git_s);
494 repo = read_gitfile(ref_git);
495 if (!repo)
496 repo = read_gitfile(mkpath("%s/.git", ref_git));
497 if (repo) {
498 free(ref_git);
499 ref_git = xstrdup(repo);
502 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
503 char *ref_git_git = mkpathdup("%s/.git", ref_git);
504 free(ref_git);
505 ref_git = ref_git_git;
506 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
507 struct strbuf sb = STRBUF_INIT;
508 seen_error = 1;
509 if (get_common_dir(&sb, ref_git)) {
510 strbuf_addf(err,
511 _("reference repository '%s' as a linked "
512 "checkout is not supported yet."),
513 path);
514 goto out;
517 strbuf_addf(err, _("reference repository '%s' is not a "
518 "local repository."), path);
519 goto out;
522 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
523 strbuf_addf(err, _("reference repository '%s' is shallow"),
524 path);
525 seen_error = 1;
526 goto out;
529 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
530 strbuf_addf(err,
531 _("reference repository '%s' is grafted"),
532 path);
533 seen_error = 1;
534 goto out;
537 out:
538 if (seen_error) {
539 free(ref_git);
540 ref_git = NULL;
543 return ref_git;
546 int foreach_alt_odb(alt_odb_fn fn, void *cb)
548 struct alternate_object_database *ent;
549 int r = 0;
551 prepare_alt_odb();
552 for (ent = alt_odb_list; ent; ent = ent->next) {
553 r = fn(ent, cb);
554 if (r)
555 break;
557 return r;
560 void prepare_alt_odb(void)
562 const char *alt;
564 if (alt_odb_tail)
565 return;
567 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
568 if (!alt) alt = "";
570 alt_odb_tail = &alt_odb_list;
571 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
573 read_info_alternates(get_object_directory(), 0);
576 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
577 static int freshen_file(const char *fn)
579 struct utimbuf t;
580 t.actime = t.modtime = time(NULL);
581 return !utime(fn, &t);
585 * All of the check_and_freshen functions return 1 if the file exists and was
586 * freshened (if freshening was requested), 0 otherwise. If they return
587 * 0, you should not assume that it is safe to skip a write of the object (it
588 * either does not exist on disk, or has a stale mtime and may be subject to
589 * pruning).
591 static int check_and_freshen_file(const char *fn, int freshen)
593 if (access(fn, F_OK))
594 return 0;
595 if (freshen && !freshen_file(fn))
596 return 0;
597 return 1;
600 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
602 return check_and_freshen_file(sha1_file_name(sha1), freshen);
605 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
607 struct alternate_object_database *alt;
608 prepare_alt_odb();
609 for (alt = alt_odb_list; alt; alt = alt->next) {
610 const char *path = alt_sha1_path(alt, sha1);
611 if (check_and_freshen_file(path, freshen))
612 return 1;
614 return 0;
617 static int check_and_freshen(const unsigned char *sha1, int freshen)
619 return check_and_freshen_local(sha1, freshen) ||
620 check_and_freshen_nonlocal(sha1, freshen);
623 int has_loose_object_nonlocal(const unsigned char *sha1)
625 return check_and_freshen_nonlocal(sha1, 0);
628 static int has_loose_object(const unsigned char *sha1)
630 return check_and_freshen(sha1, 0);
633 static unsigned int pack_used_ctr;
634 static unsigned int pack_mmap_calls;
635 static unsigned int peak_pack_open_windows;
636 static unsigned int pack_open_windows;
637 static unsigned int pack_open_fds;
638 static unsigned int pack_max_fds;
639 static size_t peak_pack_mapped;
640 static size_t pack_mapped;
641 struct packed_git *packed_git;
643 static struct mru packed_git_mru_storage;
644 struct mru *packed_git_mru = &packed_git_mru_storage;
646 void pack_report(void)
648 fprintf(stderr,
649 "pack_report: getpagesize() = %10" SZ_FMT "\n"
650 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
651 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
652 sz_fmt(getpagesize()),
653 sz_fmt(packed_git_window_size),
654 sz_fmt(packed_git_limit));
655 fprintf(stderr,
656 "pack_report: pack_used_ctr = %10u\n"
657 "pack_report: pack_mmap_calls = %10u\n"
658 "pack_report: pack_open_windows = %10u / %10u\n"
659 "pack_report: pack_mapped = "
660 "%10" SZ_FMT " / %10" SZ_FMT "\n",
661 pack_used_ctr,
662 pack_mmap_calls,
663 pack_open_windows, peak_pack_open_windows,
664 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
668 * Open and mmap the index file at path, perform a couple of
669 * consistency checks, then record its information to p. Return 0 on
670 * success.
672 static int check_packed_git_idx(const char *path, struct packed_git *p)
674 void *idx_map;
675 struct pack_idx_header *hdr;
676 size_t idx_size;
677 uint32_t version, nr, i, *index;
678 int fd = git_open(path);
679 struct stat st;
681 if (fd < 0)
682 return -1;
683 if (fstat(fd, &st)) {
684 close(fd);
685 return -1;
687 idx_size = xsize_t(st.st_size);
688 if (idx_size < 4 * 256 + 20 + 20) {
689 close(fd);
690 return error("index file %s is too small", path);
692 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
693 close(fd);
695 hdr = idx_map;
696 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
697 version = ntohl(hdr->idx_version);
698 if (version < 2 || version > 2) {
699 munmap(idx_map, idx_size);
700 return error("index file %s is version %"PRIu32
701 " and is not supported by this binary"
702 " (try upgrading GIT to a newer version)",
703 path, version);
705 } else
706 version = 1;
708 nr = 0;
709 index = idx_map;
710 if (version > 1)
711 index += 2; /* skip index header */
712 for (i = 0; i < 256; i++) {
713 uint32_t n = ntohl(index[i]);
714 if (n < nr) {
715 munmap(idx_map, idx_size);
716 return error("non-monotonic index %s", path);
718 nr = n;
721 if (version == 1) {
723 * Total size:
724 * - 256 index entries 4 bytes each
725 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
726 * - 20-byte SHA1 of the packfile
727 * - 20-byte SHA1 file checksum
729 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
730 munmap(idx_map, idx_size);
731 return error("wrong index v1 file size in %s", path);
733 } else if (version == 2) {
735 * Minimum size:
736 * - 8 bytes of header
737 * - 256 index entries 4 bytes each
738 * - 20-byte sha1 entry * nr
739 * - 4-byte crc entry * nr
740 * - 4-byte offset entry * nr
741 * - 20-byte SHA1 of the packfile
742 * - 20-byte SHA1 file checksum
743 * And after the 4-byte offset table might be a
744 * variable sized table containing 8-byte entries
745 * for offsets larger than 2^31.
747 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
748 unsigned long max_size = min_size;
749 if (nr)
750 max_size += (nr - 1)*8;
751 if (idx_size < min_size || idx_size > max_size) {
752 munmap(idx_map, idx_size);
753 return error("wrong index v2 file size in %s", path);
755 if (idx_size != min_size &&
757 * make sure we can deal with large pack offsets.
758 * 31-bit signed offset won't be enough, neither
759 * 32-bit unsigned one will be.
761 (sizeof(off_t) <= 4)) {
762 munmap(idx_map, idx_size);
763 return error("pack too large for current definition of off_t in %s", path);
767 p->index_version = version;
768 p->index_data = idx_map;
769 p->index_size = idx_size;
770 p->num_objects = nr;
771 return 0;
774 int open_pack_index(struct packed_git *p)
776 char *idx_name;
777 size_t len;
778 int ret;
780 if (p->index_data)
781 return 0;
783 if (!strip_suffix(p->pack_name, ".pack", &len))
784 die("BUG: pack_name does not end in .pack");
785 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
786 ret = check_packed_git_idx(idx_name, p);
787 free(idx_name);
788 return ret;
791 static void scan_windows(struct packed_git *p,
792 struct packed_git **lru_p,
793 struct pack_window **lru_w,
794 struct pack_window **lru_l)
796 struct pack_window *w, *w_l;
798 for (w_l = NULL, w = p->windows; w; w = w->next) {
799 if (!w->inuse_cnt) {
800 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
801 *lru_p = p;
802 *lru_w = w;
803 *lru_l = w_l;
806 w_l = w;
810 static int unuse_one_window(struct packed_git *current)
812 struct packed_git *p, *lru_p = NULL;
813 struct pack_window *lru_w = NULL, *lru_l = NULL;
815 if (current)
816 scan_windows(current, &lru_p, &lru_w, &lru_l);
817 for (p = packed_git; p; p = p->next)
818 scan_windows(p, &lru_p, &lru_w, &lru_l);
819 if (lru_p) {
820 munmap(lru_w->base, lru_w->len);
821 pack_mapped -= lru_w->len;
822 if (lru_l)
823 lru_l->next = lru_w->next;
824 else
825 lru_p->windows = lru_w->next;
826 free(lru_w);
827 pack_open_windows--;
828 return 1;
830 return 0;
833 void release_pack_memory(size_t need)
835 size_t cur = pack_mapped;
836 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
837 ; /* nothing */
840 static void mmap_limit_check(size_t length)
842 static size_t limit = 0;
843 if (!limit) {
844 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
845 if (!limit)
846 limit = SIZE_MAX;
848 if (length > limit)
849 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
850 (uintmax_t)length, (uintmax_t)limit);
853 void *xmmap_gently(void *start, size_t length,
854 int prot, int flags, int fd, off_t offset)
856 void *ret;
858 mmap_limit_check(length);
859 ret = mmap(start, length, prot, flags, fd, offset);
860 if (ret == MAP_FAILED) {
861 if (!length)
862 return NULL;
863 release_pack_memory(length);
864 ret = mmap(start, length, prot, flags, fd, offset);
866 return ret;
869 void *xmmap(void *start, size_t length,
870 int prot, int flags, int fd, off_t offset)
872 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
873 if (ret == MAP_FAILED)
874 die_errno("mmap failed");
875 return ret;
878 void close_pack_windows(struct packed_git *p)
880 while (p->windows) {
881 struct pack_window *w = p->windows;
883 if (w->inuse_cnt)
884 die("pack '%s' still has open windows to it",
885 p->pack_name);
886 munmap(w->base, w->len);
887 pack_mapped -= w->len;
888 pack_open_windows--;
889 p->windows = w->next;
890 free(w);
894 static int close_pack_fd(struct packed_git *p)
896 if (p->pack_fd < 0)
897 return 0;
899 close(p->pack_fd);
900 pack_open_fds--;
901 p->pack_fd = -1;
903 return 1;
906 static void close_pack(struct packed_git *p)
908 close_pack_windows(p);
909 close_pack_fd(p);
910 close_pack_index(p);
913 void close_all_packs(void)
915 struct packed_git *p;
917 for (p = packed_git; p; p = p->next)
918 if (p->do_not_close)
919 die("BUG: want to close pack marked 'do-not-close'");
920 else
921 close_pack(p);
926 * The LRU pack is the one with the oldest MRU window, preferring packs
927 * with no used windows, or the oldest mtime if it has no windows allocated.
929 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
931 struct pack_window *w, *this_mru_w;
932 int has_windows_inuse = 0;
935 * Reject this pack if it has windows and the previously selected
936 * one does not. If this pack does not have windows, reject
937 * it if the pack file is newer than the previously selected one.
939 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
940 return;
942 for (w = this_mru_w = p->windows; w; w = w->next) {
944 * Reject this pack if any of its windows are in use,
945 * but the previously selected pack did not have any
946 * inuse windows. Otherwise, record that this pack
947 * has windows in use.
949 if (w->inuse_cnt) {
950 if (*accept_windows_inuse)
951 has_windows_inuse = 1;
952 else
953 return;
956 if (w->last_used > this_mru_w->last_used)
957 this_mru_w = w;
960 * Reject this pack if it has windows that have been
961 * used more recently than the previously selected pack.
962 * If the previously selected pack had windows inuse and
963 * we have not encountered a window in this pack that is
964 * inuse, skip this check since we prefer a pack with no
965 * inuse windows to one that has inuse windows.
967 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
968 this_mru_w->last_used > (*mru_w)->last_used)
969 return;
973 * Select this pack.
975 *mru_w = this_mru_w;
976 *lru_p = p;
977 *accept_windows_inuse = has_windows_inuse;
980 static int close_one_pack(void)
982 struct packed_git *p, *lru_p = NULL;
983 struct pack_window *mru_w = NULL;
984 int accept_windows_inuse = 1;
986 for (p = packed_git; p; p = p->next) {
987 if (p->pack_fd == -1)
988 continue;
989 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
992 if (lru_p)
993 return close_pack_fd(lru_p);
995 return 0;
998 void unuse_pack(struct pack_window **w_cursor)
1000 struct pack_window *w = *w_cursor;
1001 if (w) {
1002 w->inuse_cnt--;
1003 *w_cursor = NULL;
1007 void close_pack_index(struct packed_git *p)
1009 if (p->index_data) {
1010 munmap((void *)p->index_data, p->index_size);
1011 p->index_data = NULL;
1015 static unsigned int get_max_fd_limit(void)
1017 #ifdef RLIMIT_NOFILE
1019 struct rlimit lim;
1021 if (!getrlimit(RLIMIT_NOFILE, &lim))
1022 return lim.rlim_cur;
1024 #endif
1026 #ifdef _SC_OPEN_MAX
1028 long open_max = sysconf(_SC_OPEN_MAX);
1029 if (0 < open_max)
1030 return open_max;
1032 * Otherwise, we got -1 for one of the two
1033 * reasons:
1035 * (1) sysconf() did not understand _SC_OPEN_MAX
1036 * and signaled an error with -1; or
1037 * (2) sysconf() said there is no limit.
1039 * We _could_ clear errno before calling sysconf() to
1040 * tell these two cases apart and return a huge number
1041 * in the latter case to let the caller cap it to a
1042 * value that is not so selfish, but letting the
1043 * fallback OPEN_MAX codepath take care of these cases
1044 * is a lot simpler.
1047 #endif
1049 #ifdef OPEN_MAX
1050 return OPEN_MAX;
1051 #else
1052 return 1; /* see the caller ;-) */
1053 #endif
1057 * Do not call this directly as this leaks p->pack_fd on error return;
1058 * call open_packed_git() instead.
1060 static int open_packed_git_1(struct packed_git *p)
1062 struct stat st;
1063 struct pack_header hdr;
1064 unsigned char sha1[20];
1065 unsigned char *idx_sha1;
1066 long fd_flag;
1068 if (!p->index_data && open_pack_index(p))
1069 return error("packfile %s index unavailable", p->pack_name);
1071 if (!pack_max_fds) {
1072 unsigned int max_fds = get_max_fd_limit();
1074 /* Save 3 for stdin/stdout/stderr, 22 for work */
1075 if (25 < max_fds)
1076 pack_max_fds = max_fds - 25;
1077 else
1078 pack_max_fds = 1;
1081 while (pack_max_fds <= pack_open_fds && close_one_pack())
1082 ; /* nothing */
1084 p->pack_fd = git_open(p->pack_name);
1085 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1086 return -1;
1087 pack_open_fds++;
1089 /* If we created the struct before we had the pack we lack size. */
1090 if (!p->pack_size) {
1091 if (!S_ISREG(st.st_mode))
1092 return error("packfile %s not a regular file", p->pack_name);
1093 p->pack_size = st.st_size;
1094 } else if (p->pack_size != st.st_size)
1095 return error("packfile %s size changed", p->pack_name);
1097 /* We leave these file descriptors open with sliding mmap;
1098 * there is no point keeping them open across exec(), though.
1100 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1101 if (fd_flag < 0)
1102 return error("cannot determine file descriptor flags");
1103 fd_flag |= FD_CLOEXEC;
1104 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1105 return error("cannot set FD_CLOEXEC");
1107 /* Verify we recognize this pack file format. */
1108 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1109 return error("file %s is far too short to be a packfile", p->pack_name);
1110 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1111 return error("file %s is not a GIT packfile", p->pack_name);
1112 if (!pack_version_ok(hdr.hdr_version))
1113 return error("packfile %s is version %"PRIu32" and not"
1114 " supported (try upgrading GIT to a newer version)",
1115 p->pack_name, ntohl(hdr.hdr_version));
1117 /* Verify the pack matches its index. */
1118 if (p->num_objects != ntohl(hdr.hdr_entries))
1119 return error("packfile %s claims to have %"PRIu32" objects"
1120 " while index indicates %"PRIu32" objects",
1121 p->pack_name, ntohl(hdr.hdr_entries),
1122 p->num_objects);
1123 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1124 return error("end of packfile %s is unavailable", p->pack_name);
1125 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1126 return error("packfile %s signature is unavailable", p->pack_name);
1127 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1128 if (hashcmp(sha1, idx_sha1))
1129 return error("packfile %s does not match index", p->pack_name);
1130 return 0;
1133 static int open_packed_git(struct packed_git *p)
1135 if (!open_packed_git_1(p))
1136 return 0;
1137 close_pack_fd(p);
1138 return -1;
1141 static int in_window(struct pack_window *win, off_t offset)
1143 /* We must promise at least 20 bytes (one hash) after the
1144 * offset is available from this window, otherwise the offset
1145 * is not actually in this window and a different window (which
1146 * has that one hash excess) must be used. This is to support
1147 * the object header and delta base parsing routines below.
1149 off_t win_off = win->offset;
1150 return win_off <= offset
1151 && (offset + 20) <= (win_off + win->len);
1154 unsigned char *use_pack(struct packed_git *p,
1155 struct pack_window **w_cursor,
1156 off_t offset,
1157 unsigned long *left)
1159 struct pack_window *win = *w_cursor;
1161 /* Since packfiles end in a hash of their content and it's
1162 * pointless to ask for an offset into the middle of that
1163 * hash, and the in_window function above wouldn't match
1164 * don't allow an offset too close to the end of the file.
1166 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1167 die("packfile %s cannot be accessed", p->pack_name);
1168 if (offset > (p->pack_size - 20))
1169 die("offset beyond end of packfile (truncated pack?)");
1170 if (offset < 0)
1171 die(_("offset before end of packfile (broken .idx?)"));
1173 if (!win || !in_window(win, offset)) {
1174 if (win)
1175 win->inuse_cnt--;
1176 for (win = p->windows; win; win = win->next) {
1177 if (in_window(win, offset))
1178 break;
1180 if (!win) {
1181 size_t window_align = packed_git_window_size / 2;
1182 off_t len;
1184 if (p->pack_fd == -1 && open_packed_git(p))
1185 die("packfile %s cannot be accessed", p->pack_name);
1187 win = xcalloc(1, sizeof(*win));
1188 win->offset = (offset / window_align) * window_align;
1189 len = p->pack_size - win->offset;
1190 if (len > packed_git_window_size)
1191 len = packed_git_window_size;
1192 win->len = (size_t)len;
1193 pack_mapped += win->len;
1194 while (packed_git_limit < pack_mapped
1195 && unuse_one_window(p))
1196 ; /* nothing */
1197 win->base = xmmap(NULL, win->len,
1198 PROT_READ, MAP_PRIVATE,
1199 p->pack_fd, win->offset);
1200 if (win->base == MAP_FAILED)
1201 die_errno("packfile %s cannot be mapped",
1202 p->pack_name);
1203 if (!win->offset && win->len == p->pack_size
1204 && !p->do_not_close)
1205 close_pack_fd(p);
1206 pack_mmap_calls++;
1207 pack_open_windows++;
1208 if (pack_mapped > peak_pack_mapped)
1209 peak_pack_mapped = pack_mapped;
1210 if (pack_open_windows > peak_pack_open_windows)
1211 peak_pack_open_windows = pack_open_windows;
1212 win->next = p->windows;
1213 p->windows = win;
1216 if (win != *w_cursor) {
1217 win->last_used = pack_used_ctr++;
1218 win->inuse_cnt++;
1219 *w_cursor = win;
1221 offset -= win->offset;
1222 if (left)
1223 *left = win->len - xsize_t(offset);
1224 return win->base + offset;
1227 static struct packed_git *alloc_packed_git(int extra)
1229 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1230 memset(p, 0, sizeof(*p));
1231 p->pack_fd = -1;
1232 return p;
1235 static void try_to_free_pack_memory(size_t size)
1237 release_pack_memory(size);
1240 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1242 static int have_set_try_to_free_routine;
1243 struct stat st;
1244 size_t alloc;
1245 struct packed_git *p;
1247 if (!have_set_try_to_free_routine) {
1248 have_set_try_to_free_routine = 1;
1249 set_try_to_free_routine(try_to_free_pack_memory);
1253 * Make sure a corresponding .pack file exists and that
1254 * the index looks sane.
1256 if (!strip_suffix_mem(path, &path_len, ".idx"))
1257 return NULL;
1260 * ".pack" is long enough to hold any suffix we're adding (and
1261 * the use xsnprintf double-checks that)
1263 alloc = st_add3(path_len, strlen(".pack"), 1);
1264 p = alloc_packed_git(alloc);
1265 memcpy(p->pack_name, path, path_len);
1267 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1268 if (!access(p->pack_name, F_OK))
1269 p->pack_keep = 1;
1271 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1272 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1273 free(p);
1274 return NULL;
1277 /* ok, it looks sane as far as we can check without
1278 * actually mapping the pack file.
1280 p->pack_size = st.st_size;
1281 p->pack_local = local;
1282 p->mtime = st.st_mtime;
1283 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1284 hashclr(p->sha1);
1285 return p;
1288 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1290 const char *path = sha1_pack_name(sha1);
1291 size_t alloc = st_add(strlen(path), 1);
1292 struct packed_git *p = alloc_packed_git(alloc);
1294 memcpy(p->pack_name, path, alloc); /* includes NUL */
1295 hashcpy(p->sha1, sha1);
1296 if (check_packed_git_idx(idx_path, p)) {
1297 free(p);
1298 return NULL;
1301 return p;
1304 void install_packed_git(struct packed_git *pack)
1306 if (pack->pack_fd != -1)
1307 pack_open_fds++;
1309 pack->next = packed_git;
1310 packed_git = pack;
1313 void (*report_garbage)(unsigned seen_bits, const char *path);
1315 static void report_helper(const struct string_list *list,
1316 int seen_bits, int first, int last)
1318 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1319 return;
1321 for (; first < last; first++)
1322 report_garbage(seen_bits, list->items[first].string);
1325 static void report_pack_garbage(struct string_list *list)
1327 int i, baselen = -1, first = 0, seen_bits = 0;
1329 if (!report_garbage)
1330 return;
1332 string_list_sort(list);
1334 for (i = 0; i < list->nr; i++) {
1335 const char *path = list->items[i].string;
1336 if (baselen != -1 &&
1337 strncmp(path, list->items[first].string, baselen)) {
1338 report_helper(list, seen_bits, first, i);
1339 baselen = -1;
1340 seen_bits = 0;
1342 if (baselen == -1) {
1343 const char *dot = strrchr(path, '.');
1344 if (!dot) {
1345 report_garbage(PACKDIR_FILE_GARBAGE, path);
1346 continue;
1348 baselen = dot - path + 1;
1349 first = i;
1351 if (!strcmp(path + baselen, "pack"))
1352 seen_bits |= 1;
1353 else if (!strcmp(path + baselen, "idx"))
1354 seen_bits |= 2;
1356 report_helper(list, seen_bits, first, list->nr);
1359 static void prepare_packed_git_one(char *objdir, int local)
1361 struct strbuf path = STRBUF_INIT;
1362 size_t dirnamelen;
1363 DIR *dir;
1364 struct dirent *de;
1365 struct string_list garbage = STRING_LIST_INIT_DUP;
1367 strbuf_addstr(&path, objdir);
1368 strbuf_addstr(&path, "/pack");
1369 dir = opendir(path.buf);
1370 if (!dir) {
1371 if (errno != ENOENT)
1372 error_errno("unable to open object pack directory: %s",
1373 path.buf);
1374 strbuf_release(&path);
1375 return;
1377 strbuf_addch(&path, '/');
1378 dirnamelen = path.len;
1379 while ((de = readdir(dir)) != NULL) {
1380 struct packed_git *p;
1381 size_t base_len;
1383 if (is_dot_or_dotdot(de->d_name))
1384 continue;
1386 strbuf_setlen(&path, dirnamelen);
1387 strbuf_addstr(&path, de->d_name);
1389 base_len = path.len;
1390 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1391 /* Don't reopen a pack we already have. */
1392 for (p = packed_git; p; p = p->next) {
1393 size_t len;
1394 if (strip_suffix(p->pack_name, ".pack", &len) &&
1395 len == base_len &&
1396 !memcmp(p->pack_name, path.buf, len))
1397 break;
1399 if (p == NULL &&
1401 * See if it really is a valid .idx file with
1402 * corresponding .pack file that we can map.
1404 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1405 install_packed_git(p);
1408 if (!report_garbage)
1409 continue;
1411 if (ends_with(de->d_name, ".idx") ||
1412 ends_with(de->d_name, ".pack") ||
1413 ends_with(de->d_name, ".bitmap") ||
1414 ends_with(de->d_name, ".keep"))
1415 string_list_append(&garbage, path.buf);
1416 else
1417 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1419 closedir(dir);
1420 report_pack_garbage(&garbage);
1421 string_list_clear(&garbage, 0);
1422 strbuf_release(&path);
1425 static int approximate_object_count_valid;
1428 * Give a fast, rough count of the number of objects in the repository. This
1429 * ignores loose objects completely. If you have a lot of them, then either
1430 * you should repack because your performance will be awful, or they are
1431 * all unreachable objects about to be pruned, in which case they're not really
1432 * interesting as a measure of repo size in the first place.
1434 unsigned long approximate_object_count(void)
1436 static unsigned long count;
1437 if (!approximate_object_count_valid) {
1438 struct packed_git *p;
1440 prepare_packed_git();
1441 count = 0;
1442 for (p = packed_git; p; p = p->next) {
1443 if (open_pack_index(p))
1444 continue;
1445 count += p->num_objects;
1448 return count;
1451 static void *get_next_packed_git(const void *p)
1453 return ((const struct packed_git *)p)->next;
1456 static void set_next_packed_git(void *p, void *next)
1458 ((struct packed_git *)p)->next = next;
1461 static int sort_pack(const void *a_, const void *b_)
1463 const struct packed_git *a = a_;
1464 const struct packed_git *b = b_;
1465 int st;
1468 * Local packs tend to contain objects specific to our
1469 * variant of the project than remote ones. In addition,
1470 * remote ones could be on a network mounted filesystem.
1471 * Favor local ones for these reasons.
1473 st = a->pack_local - b->pack_local;
1474 if (st)
1475 return -st;
1478 * Younger packs tend to contain more recent objects,
1479 * and more recent objects tend to get accessed more
1480 * often.
1482 if (a->mtime < b->mtime)
1483 return 1;
1484 else if (a->mtime == b->mtime)
1485 return 0;
1486 return -1;
1489 static void rearrange_packed_git(void)
1491 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1492 set_next_packed_git, sort_pack);
1495 static void prepare_packed_git_mru(void)
1497 struct packed_git *p;
1499 mru_clear(packed_git_mru);
1500 for (p = packed_git; p; p = p->next)
1501 mru_append(packed_git_mru, p);
1504 static int prepare_packed_git_run_once = 0;
1505 void prepare_packed_git(void)
1507 struct alternate_object_database *alt;
1509 if (prepare_packed_git_run_once)
1510 return;
1511 prepare_packed_git_one(get_object_directory(), 1);
1512 prepare_alt_odb();
1513 for (alt = alt_odb_list; alt; alt = alt->next)
1514 prepare_packed_git_one(alt->path, 0);
1515 rearrange_packed_git();
1516 prepare_packed_git_mru();
1517 prepare_packed_git_run_once = 1;
1520 void reprepare_packed_git(void)
1522 approximate_object_count_valid = 0;
1523 prepare_packed_git_run_once = 0;
1524 prepare_packed_git();
1527 static void mark_bad_packed_object(struct packed_git *p,
1528 const unsigned char *sha1)
1530 unsigned i;
1531 for (i = 0; i < p->num_bad_objects; i++)
1532 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1533 return;
1534 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1535 st_mult(GIT_SHA1_RAWSZ,
1536 st_add(p->num_bad_objects, 1)));
1537 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1538 p->num_bad_objects++;
1541 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1543 struct packed_git *p;
1544 unsigned i;
1546 for (p = packed_git; p; p = p->next)
1547 for (i = 0; i < p->num_bad_objects; i++)
1548 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1549 return p;
1550 return NULL;
1554 * With an in-core object data in "map", rehash it to make sure the
1555 * object name actually matches "sha1" to detect object corruption.
1556 * With "map" == NULL, try reading the object named with "sha1" using
1557 * the streaming interface and rehash it to do the same.
1559 int check_sha1_signature(const unsigned char *sha1, void *map,
1560 unsigned long size, const char *type)
1562 unsigned char real_sha1[20];
1563 enum object_type obj_type;
1564 struct git_istream *st;
1565 git_SHA_CTX c;
1566 char hdr[32];
1567 int hdrlen;
1569 if (map) {
1570 hash_sha1_file(map, size, type, real_sha1);
1571 return hashcmp(sha1, real_sha1) ? -1 : 0;
1574 st = open_istream(sha1, &obj_type, &size, NULL);
1575 if (!st)
1576 return -1;
1578 /* Generate the header */
1579 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1581 /* Sha1.. */
1582 git_SHA1_Init(&c);
1583 git_SHA1_Update(&c, hdr, hdrlen);
1584 for (;;) {
1585 char buf[1024 * 16];
1586 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1588 if (readlen < 0) {
1589 close_istream(st);
1590 return -1;
1592 if (!readlen)
1593 break;
1594 git_SHA1_Update(&c, buf, readlen);
1596 git_SHA1_Final(real_sha1, &c);
1597 close_istream(st);
1598 return hashcmp(sha1, real_sha1) ? -1 : 0;
1601 int git_open_cloexec(const char *name, int flags)
1603 int fd;
1604 static int o_cloexec = O_CLOEXEC;
1606 fd = open(name, flags | o_cloexec);
1607 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1608 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1609 o_cloexec &= ~O_CLOEXEC;
1610 fd = open(name, flags | o_cloexec);
1613 #if defined(F_GETFL) && defined(F_SETFL) && defined(FD_CLOEXEC)
1615 static int fd_cloexec = FD_CLOEXEC;
1617 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1618 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1619 int flags = fcntl(fd, F_GETFL);
1620 if (fcntl(fd, F_SETFL, flags | fd_cloexec))
1621 fd_cloexec = 0;
1624 #endif
1625 return fd;
1629 * Find "sha1" as a loose object in the local repository or in an alternate.
1630 * Returns 0 on success, negative on failure.
1632 * The "path" out-parameter will give the path of the object we found (if any).
1633 * Note that it may point to static storage and is only valid until another
1634 * call to sha1_file_name(), etc.
1636 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1637 const char **path)
1639 struct alternate_object_database *alt;
1641 *path = sha1_file_name(sha1);
1642 if (!lstat(*path, st))
1643 return 0;
1645 prepare_alt_odb();
1646 errno = ENOENT;
1647 for (alt = alt_odb_list; alt; alt = alt->next) {
1648 *path = alt_sha1_path(alt, sha1);
1649 if (!lstat(*path, st))
1650 return 0;
1653 return -1;
1657 * Like stat_sha1_file(), but actually open the object and return the
1658 * descriptor. See the caveats on the "path" parameter above.
1660 static int open_sha1_file(const unsigned char *sha1, const char **path)
1662 int fd;
1663 struct alternate_object_database *alt;
1664 int most_interesting_errno;
1666 *path = sha1_file_name(sha1);
1667 fd = git_open(*path);
1668 if (fd >= 0)
1669 return fd;
1670 most_interesting_errno = errno;
1672 prepare_alt_odb();
1673 for (alt = alt_odb_list; alt; alt = alt->next) {
1674 *path = alt_sha1_path(alt, sha1);
1675 fd = git_open(*path);
1676 if (fd >= 0)
1677 return fd;
1678 if (most_interesting_errno == ENOENT)
1679 most_interesting_errno = errno;
1681 errno = most_interesting_errno;
1682 return -1;
1686 * Map the loose object at "path" if it is not NULL, or the path found by
1687 * searching for a loose object named "sha1".
1689 static void *map_sha1_file_1(const char *path,
1690 const unsigned char *sha1,
1691 unsigned long *size)
1693 void *map;
1694 int fd;
1696 if (path)
1697 fd = git_open(path);
1698 else
1699 fd = open_sha1_file(sha1, &path);
1700 map = NULL;
1701 if (fd >= 0) {
1702 struct stat st;
1704 if (!fstat(fd, &st)) {
1705 *size = xsize_t(st.st_size);
1706 if (!*size) {
1707 /* mmap() is forbidden on empty files */
1708 error("object file %s is empty", path);
1709 return NULL;
1711 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1713 close(fd);
1715 return map;
1718 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1720 return map_sha1_file_1(NULL, sha1, size);
1723 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1724 unsigned long len, enum object_type *type, unsigned long *sizep)
1726 unsigned shift;
1727 unsigned long size, c;
1728 unsigned long used = 0;
1730 c = buf[used++];
1731 *type = (c >> 4) & 7;
1732 size = c & 15;
1733 shift = 4;
1734 while (c & 0x80) {
1735 if (len <= used || bitsizeof(long) <= shift) {
1736 error("bad object header");
1737 size = used = 0;
1738 break;
1740 c = buf[used++];
1741 size += (c & 0x7f) << shift;
1742 shift += 7;
1744 *sizep = size;
1745 return used;
1748 static int unpack_sha1_short_header(git_zstream *stream,
1749 unsigned char *map, unsigned long mapsize,
1750 void *buffer, unsigned long bufsiz)
1752 /* Get the data stream */
1753 memset(stream, 0, sizeof(*stream));
1754 stream->next_in = map;
1755 stream->avail_in = mapsize;
1756 stream->next_out = buffer;
1757 stream->avail_out = bufsiz;
1759 git_inflate_init(stream);
1760 return git_inflate(stream, 0);
1763 int unpack_sha1_header(git_zstream *stream,
1764 unsigned char *map, unsigned long mapsize,
1765 void *buffer, unsigned long bufsiz)
1767 int status = unpack_sha1_short_header(stream, map, mapsize,
1768 buffer, bufsiz);
1770 if (status < Z_OK)
1771 return status;
1773 /* Make sure we have the terminating NUL */
1774 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1775 return -1;
1776 return 0;
1779 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1780 unsigned long mapsize, void *buffer,
1781 unsigned long bufsiz, struct strbuf *header)
1783 int status;
1785 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1786 if (status < Z_OK)
1787 return -1;
1790 * Check if entire header is unpacked in the first iteration.
1792 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1793 return 0;
1796 * buffer[0..bufsiz] was not large enough. Copy the partial
1797 * result out to header, and then append the result of further
1798 * reading the stream.
1800 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1801 stream->next_out = buffer;
1802 stream->avail_out = bufsiz;
1804 do {
1805 status = git_inflate(stream, 0);
1806 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1807 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1808 return 0;
1809 stream->next_out = buffer;
1810 stream->avail_out = bufsiz;
1811 } while (status != Z_STREAM_END);
1812 return -1;
1815 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1817 int bytes = strlen(buffer) + 1;
1818 unsigned char *buf = xmallocz(size);
1819 unsigned long n;
1820 int status = Z_OK;
1822 n = stream->total_out - bytes;
1823 if (n > size)
1824 n = size;
1825 memcpy(buf, (char *) buffer + bytes, n);
1826 bytes = n;
1827 if (bytes <= size) {
1829 * The above condition must be (bytes <= size), not
1830 * (bytes < size). In other words, even though we
1831 * expect no more output and set avail_out to zero,
1832 * the input zlib stream may have bytes that express
1833 * "this concludes the stream", and we *do* want to
1834 * eat that input.
1836 * Otherwise we would not be able to test that we
1837 * consumed all the input to reach the expected size;
1838 * we also want to check that zlib tells us that all
1839 * went well with status == Z_STREAM_END at the end.
1841 stream->next_out = buf + bytes;
1842 stream->avail_out = size - bytes;
1843 while (status == Z_OK)
1844 status = git_inflate(stream, Z_FINISH);
1846 if (status == Z_STREAM_END && !stream->avail_in) {
1847 git_inflate_end(stream);
1848 return buf;
1851 if (status < 0)
1852 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1853 else if (stream->avail_in)
1854 error("garbage at end of loose object '%s'",
1855 sha1_to_hex(sha1));
1856 free(buf);
1857 return NULL;
1861 * We used to just use "sscanf()", but that's actually way
1862 * too permissive for what we want to check. So do an anal
1863 * object header parse by hand.
1865 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1866 unsigned int flags)
1868 const char *type_buf = hdr;
1869 unsigned long size;
1870 int type, type_len = 0;
1873 * The type can be of any size but is followed by
1874 * a space.
1876 for (;;) {
1877 char c = *hdr++;
1878 if (!c)
1879 return -1;
1880 if (c == ' ')
1881 break;
1882 type_len++;
1885 type = type_from_string_gently(type_buf, type_len, 1);
1886 if (oi->typename)
1887 strbuf_add(oi->typename, type_buf, type_len);
1889 * Set type to 0 if its an unknown object and
1890 * we're obtaining the type using '--allow-unknown-type'
1891 * option.
1893 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1894 type = 0;
1895 else if (type < 0)
1896 die("invalid object type");
1897 if (oi->typep)
1898 *oi->typep = type;
1901 * The length must follow immediately, and be in canonical
1902 * decimal format (ie "010" is not valid).
1904 size = *hdr++ - '0';
1905 if (size > 9)
1906 return -1;
1907 if (size) {
1908 for (;;) {
1909 unsigned long c = *hdr - '0';
1910 if (c > 9)
1911 break;
1912 hdr++;
1913 size = size * 10 + c;
1917 if (oi->sizep)
1918 *oi->sizep = size;
1921 * The length must be followed by a zero byte
1923 return *hdr ? -1 : type;
1926 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1928 struct object_info oi = OBJECT_INFO_INIT;
1930 oi.sizep = sizep;
1931 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1934 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1936 int ret;
1937 git_zstream stream;
1938 char hdr[8192];
1940 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1941 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1942 return NULL;
1944 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1947 unsigned long get_size_from_delta(struct packed_git *p,
1948 struct pack_window **w_curs,
1949 off_t curpos)
1951 const unsigned char *data;
1952 unsigned char delta_head[20], *in;
1953 git_zstream stream;
1954 int st;
1956 memset(&stream, 0, sizeof(stream));
1957 stream.next_out = delta_head;
1958 stream.avail_out = sizeof(delta_head);
1960 git_inflate_init(&stream);
1961 do {
1962 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1963 stream.next_in = in;
1964 st = git_inflate(&stream, Z_FINISH);
1965 curpos += stream.next_in - in;
1966 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1967 stream.total_out < sizeof(delta_head));
1968 git_inflate_end(&stream);
1969 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1970 error("delta data unpack-initial failed");
1971 return 0;
1974 /* Examine the initial part of the delta to figure out
1975 * the result size.
1977 data = delta_head;
1979 /* ignore base size */
1980 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1982 /* Read the result size */
1983 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1986 static off_t get_delta_base(struct packed_git *p,
1987 struct pack_window **w_curs,
1988 off_t *curpos,
1989 enum object_type type,
1990 off_t delta_obj_offset)
1992 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1993 off_t base_offset;
1995 /* use_pack() assured us we have [base_info, base_info + 20)
1996 * as a range that we can look at without walking off the
1997 * end of the mapped window. Its actually the hash size
1998 * that is assured. An OFS_DELTA longer than the hash size
1999 * is stupid, as then a REF_DELTA would be smaller to store.
2001 if (type == OBJ_OFS_DELTA) {
2002 unsigned used = 0;
2003 unsigned char c = base_info[used++];
2004 base_offset = c & 127;
2005 while (c & 128) {
2006 base_offset += 1;
2007 if (!base_offset || MSB(base_offset, 7))
2008 return 0; /* overflow */
2009 c = base_info[used++];
2010 base_offset = (base_offset << 7) + (c & 127);
2012 base_offset = delta_obj_offset - base_offset;
2013 if (base_offset <= 0 || base_offset >= delta_obj_offset)
2014 return 0; /* out of bound */
2015 *curpos += used;
2016 } else if (type == OBJ_REF_DELTA) {
2017 /* The base entry _must_ be in the same pack */
2018 base_offset = find_pack_entry_one(base_info, p);
2019 *curpos += 20;
2020 } else
2021 die("I am totally screwed");
2022 return base_offset;
2026 * Like get_delta_base above, but we return the sha1 instead of the pack
2027 * offset. This means it is cheaper for REF deltas (we do not have to do
2028 * the final object lookup), but more expensive for OFS deltas (we
2029 * have to load the revidx to convert the offset back into a sha1).
2031 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2032 struct pack_window **w_curs,
2033 off_t curpos,
2034 enum object_type type,
2035 off_t delta_obj_offset)
2037 if (type == OBJ_REF_DELTA) {
2038 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2039 return base;
2040 } else if (type == OBJ_OFS_DELTA) {
2041 struct revindex_entry *revidx;
2042 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2043 type, delta_obj_offset);
2045 if (!base_offset)
2046 return NULL;
2048 revidx = find_pack_revindex(p, base_offset);
2049 if (!revidx)
2050 return NULL;
2052 return nth_packed_object_sha1(p, revidx->nr);
2053 } else
2054 return NULL;
2057 int unpack_object_header(struct packed_git *p,
2058 struct pack_window **w_curs,
2059 off_t *curpos,
2060 unsigned long *sizep)
2062 unsigned char *base;
2063 unsigned long left;
2064 unsigned long used;
2065 enum object_type type;
2067 /* use_pack() assures us we have [base, base + 20) available
2068 * as a range that we can look at. (Its actually the hash
2069 * size that is assured.) With our object header encoding
2070 * the maximum deflated object size is 2^137, which is just
2071 * insane, so we know won't exceed what we have been given.
2073 base = use_pack(p, w_curs, *curpos, &left);
2074 used = unpack_object_header_buffer(base, left, &type, sizep);
2075 if (!used) {
2076 type = OBJ_BAD;
2077 } else
2078 *curpos += used;
2080 return type;
2083 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2085 int type;
2086 struct revindex_entry *revidx;
2087 const unsigned char *sha1;
2088 revidx = find_pack_revindex(p, obj_offset);
2089 if (!revidx)
2090 return OBJ_BAD;
2091 sha1 = nth_packed_object_sha1(p, revidx->nr);
2092 mark_bad_packed_object(p, sha1);
2093 type = sha1_object_info(sha1, NULL);
2094 if (type <= OBJ_NONE)
2095 return OBJ_BAD;
2096 return type;
2099 #define POI_STACK_PREALLOC 64
2101 static enum object_type packed_to_object_type(struct packed_git *p,
2102 off_t obj_offset,
2103 enum object_type type,
2104 struct pack_window **w_curs,
2105 off_t curpos)
2107 off_t small_poi_stack[POI_STACK_PREALLOC];
2108 off_t *poi_stack = small_poi_stack;
2109 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2111 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2112 off_t base_offset;
2113 unsigned long size;
2114 /* Push the object we're going to leave behind */
2115 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2116 poi_stack_alloc = alloc_nr(poi_stack_nr);
2117 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2118 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2119 } else {
2120 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2122 poi_stack[poi_stack_nr++] = obj_offset;
2123 /* If parsing the base offset fails, just unwind */
2124 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2125 if (!base_offset)
2126 goto unwind;
2127 curpos = obj_offset = base_offset;
2128 type = unpack_object_header(p, w_curs, &curpos, &size);
2129 if (type <= OBJ_NONE) {
2130 /* If getting the base itself fails, we first
2131 * retry the base, otherwise unwind */
2132 type = retry_bad_packed_offset(p, base_offset);
2133 if (type > OBJ_NONE)
2134 goto out;
2135 goto unwind;
2139 switch (type) {
2140 case OBJ_BAD:
2141 case OBJ_COMMIT:
2142 case OBJ_TREE:
2143 case OBJ_BLOB:
2144 case OBJ_TAG:
2145 break;
2146 default:
2147 error("unknown object type %i at offset %"PRIuMAX" in %s",
2148 type, (uintmax_t)obj_offset, p->pack_name);
2149 type = OBJ_BAD;
2152 out:
2153 if (poi_stack != small_poi_stack)
2154 free(poi_stack);
2155 return type;
2157 unwind:
2158 while (poi_stack_nr) {
2159 obj_offset = poi_stack[--poi_stack_nr];
2160 type = retry_bad_packed_offset(p, obj_offset);
2161 if (type > OBJ_NONE)
2162 goto out;
2164 type = OBJ_BAD;
2165 goto out;
2168 int packed_object_info(struct packed_git *p, off_t obj_offset,
2169 struct object_info *oi)
2171 struct pack_window *w_curs = NULL;
2172 unsigned long size;
2173 off_t curpos = obj_offset;
2174 enum object_type type;
2177 * We always get the representation type, but only convert it to
2178 * a "real" type later if the caller is interested.
2180 type = unpack_object_header(p, &w_curs, &curpos, &size);
2182 if (oi->sizep) {
2183 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2184 off_t tmp_pos = curpos;
2185 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2186 type, obj_offset);
2187 if (!base_offset) {
2188 type = OBJ_BAD;
2189 goto out;
2191 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2192 if (*oi->sizep == 0) {
2193 type = OBJ_BAD;
2194 goto out;
2196 } else {
2197 *oi->sizep = size;
2201 if (oi->disk_sizep) {
2202 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2203 *oi->disk_sizep = revidx[1].offset - obj_offset;
2206 if (oi->typep) {
2207 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2208 if (*oi->typep < 0) {
2209 type = OBJ_BAD;
2210 goto out;
2214 if (oi->delta_base_sha1) {
2215 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2216 const unsigned char *base;
2218 base = get_delta_base_sha1(p, &w_curs, curpos,
2219 type, obj_offset);
2220 if (!base) {
2221 type = OBJ_BAD;
2222 goto out;
2225 hashcpy(oi->delta_base_sha1, base);
2226 } else
2227 hashclr(oi->delta_base_sha1);
2230 out:
2231 unuse_pack(&w_curs);
2232 return type;
2235 static void *unpack_compressed_entry(struct packed_git *p,
2236 struct pack_window **w_curs,
2237 off_t curpos,
2238 unsigned long size)
2240 int st;
2241 git_zstream stream;
2242 unsigned char *buffer, *in;
2244 buffer = xmallocz_gently(size);
2245 if (!buffer)
2246 return NULL;
2247 memset(&stream, 0, sizeof(stream));
2248 stream.next_out = buffer;
2249 stream.avail_out = size + 1;
2251 git_inflate_init(&stream);
2252 do {
2253 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2254 stream.next_in = in;
2255 st = git_inflate(&stream, Z_FINISH);
2256 if (!stream.avail_out)
2257 break; /* the payload is larger than it should be */
2258 curpos += stream.next_in - in;
2259 } while (st == Z_OK || st == Z_BUF_ERROR);
2260 git_inflate_end(&stream);
2261 if ((st != Z_STREAM_END) || stream.total_out != size) {
2262 free(buffer);
2263 return NULL;
2266 return buffer;
2269 static struct hashmap delta_base_cache;
2270 static size_t delta_base_cached;
2272 static LIST_HEAD(delta_base_cache_lru);
2274 struct delta_base_cache_key {
2275 struct packed_git *p;
2276 off_t base_offset;
2279 struct delta_base_cache_entry {
2280 struct hashmap hash;
2281 struct delta_base_cache_key key;
2282 struct list_head lru;
2283 void *data;
2284 unsigned long size;
2285 enum object_type type;
2288 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2290 unsigned int hash;
2292 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2293 hash += (hash >> 8) + (hash >> 16);
2294 return hash;
2297 static struct delta_base_cache_entry *
2298 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2300 struct hashmap_entry entry;
2301 struct delta_base_cache_key key;
2303 if (!delta_base_cache.cmpfn)
2304 return NULL;
2306 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2307 key.p = p;
2308 key.base_offset = base_offset;
2309 return hashmap_get(&delta_base_cache, &entry, &key);
2312 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2313 const struct delta_base_cache_key *b)
2315 return a->p == b->p && a->base_offset == b->base_offset;
2318 static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2319 const void *vkey)
2321 const struct delta_base_cache_entry *a = va, *b = vb;
2322 const struct delta_base_cache_key *key = vkey;
2323 if (key)
2324 return !delta_base_cache_key_eq(&a->key, key);
2325 else
2326 return !delta_base_cache_key_eq(&a->key, &b->key);
2329 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2331 return !!get_delta_base_cache_entry(p, base_offset);
2335 * Remove the entry from the cache, but do _not_ free the associated
2336 * entry data. The caller takes ownership of the "data" buffer, and
2337 * should copy out any fields it wants before detaching.
2339 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2341 hashmap_remove(&delta_base_cache, ent, &ent->key);
2342 list_del(&ent->lru);
2343 delta_base_cached -= ent->size;
2344 free(ent);
2347 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2348 unsigned long *base_size, enum object_type *type)
2350 struct delta_base_cache_entry *ent;
2352 ent = get_delta_base_cache_entry(p, base_offset);
2353 if (!ent)
2354 return unpack_entry(p, base_offset, type, base_size);
2356 *type = ent->type;
2357 *base_size = ent->size;
2358 return xmemdupz(ent->data, ent->size);
2361 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2363 free(ent->data);
2364 detach_delta_base_cache_entry(ent);
2367 void clear_delta_base_cache(void)
2369 struct list_head *lru, *tmp;
2370 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2371 struct delta_base_cache_entry *entry =
2372 list_entry(lru, struct delta_base_cache_entry, lru);
2373 release_delta_base_cache(entry);
2377 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2378 void *base, unsigned long base_size, enum object_type type)
2380 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2381 struct list_head *lru, *tmp;
2383 delta_base_cached += base_size;
2385 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2386 struct delta_base_cache_entry *f =
2387 list_entry(lru, struct delta_base_cache_entry, lru);
2388 if (delta_base_cached <= delta_base_cache_limit)
2389 break;
2390 release_delta_base_cache(f);
2393 ent->key.p = p;
2394 ent->key.base_offset = base_offset;
2395 ent->type = type;
2396 ent->data = base;
2397 ent->size = base_size;
2398 list_add_tail(&ent->lru, &delta_base_cache_lru);
2400 if (!delta_base_cache.cmpfn)
2401 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2402 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2403 hashmap_add(&delta_base_cache, ent);
2406 static void *read_object(const unsigned char *sha1, enum object_type *type,
2407 unsigned long *size);
2409 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2411 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2412 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2413 p->pack_name, (uintmax_t)obj_offset);
2416 int do_check_packed_object_crc;
2418 #define UNPACK_ENTRY_STACK_PREALLOC 64
2419 struct unpack_entry_stack_ent {
2420 off_t obj_offset;
2421 off_t curpos;
2422 unsigned long size;
2425 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2426 enum object_type *final_type, unsigned long *final_size)
2428 struct pack_window *w_curs = NULL;
2429 off_t curpos = obj_offset;
2430 void *data = NULL;
2431 unsigned long size;
2432 enum object_type type;
2433 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2434 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2435 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2436 int base_from_cache = 0;
2438 write_pack_access_log(p, obj_offset);
2440 /* PHASE 1: drill down to the innermost base object */
2441 for (;;) {
2442 off_t base_offset;
2443 int i;
2444 struct delta_base_cache_entry *ent;
2446 ent = get_delta_base_cache_entry(p, curpos);
2447 if (ent) {
2448 type = ent->type;
2449 data = ent->data;
2450 size = ent->size;
2451 detach_delta_base_cache_entry(ent);
2452 base_from_cache = 1;
2453 break;
2456 if (do_check_packed_object_crc && p->index_version > 1) {
2457 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2458 off_t len = revidx[1].offset - obj_offset;
2459 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2460 const unsigned char *sha1 =
2461 nth_packed_object_sha1(p, revidx->nr);
2462 error("bad packed object CRC for %s",
2463 sha1_to_hex(sha1));
2464 mark_bad_packed_object(p, sha1);
2465 unuse_pack(&w_curs);
2466 return NULL;
2470 type = unpack_object_header(p, &w_curs, &curpos, &size);
2471 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2472 break;
2474 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2475 if (!base_offset) {
2476 error("failed to validate delta base reference "
2477 "at offset %"PRIuMAX" from %s",
2478 (uintmax_t)curpos, p->pack_name);
2479 /* bail to phase 2, in hopes of recovery */
2480 data = NULL;
2481 break;
2484 /* push object, proceed to base */
2485 if (delta_stack_nr >= delta_stack_alloc
2486 && delta_stack == small_delta_stack) {
2487 delta_stack_alloc = alloc_nr(delta_stack_nr);
2488 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2489 memcpy(delta_stack, small_delta_stack,
2490 sizeof(*delta_stack)*delta_stack_nr);
2491 } else {
2492 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2494 i = delta_stack_nr++;
2495 delta_stack[i].obj_offset = obj_offset;
2496 delta_stack[i].curpos = curpos;
2497 delta_stack[i].size = size;
2499 curpos = obj_offset = base_offset;
2502 /* PHASE 2: handle the base */
2503 switch (type) {
2504 case OBJ_OFS_DELTA:
2505 case OBJ_REF_DELTA:
2506 if (data)
2507 die("BUG: unpack_entry: left loop at a valid delta");
2508 break;
2509 case OBJ_COMMIT:
2510 case OBJ_TREE:
2511 case OBJ_BLOB:
2512 case OBJ_TAG:
2513 if (!base_from_cache)
2514 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2515 break;
2516 default:
2517 data = NULL;
2518 error("unknown object type %i at offset %"PRIuMAX" in %s",
2519 type, (uintmax_t)obj_offset, p->pack_name);
2522 /* PHASE 3: apply deltas in order */
2524 /* invariants:
2525 * 'data' holds the base data, or NULL if there was corruption
2527 while (delta_stack_nr) {
2528 void *delta_data;
2529 void *base = data;
2530 void *external_base = NULL;
2531 unsigned long delta_size, base_size = size;
2532 int i;
2534 data = NULL;
2536 if (base)
2537 add_delta_base_cache(p, obj_offset, base, base_size, type);
2539 if (!base) {
2541 * We're probably in deep shit, but let's try to fetch
2542 * the required base anyway from another pack or loose.
2543 * This is costly but should happen only in the presence
2544 * of a corrupted pack, and is better than failing outright.
2546 struct revindex_entry *revidx;
2547 const unsigned char *base_sha1;
2548 revidx = find_pack_revindex(p, obj_offset);
2549 if (revidx) {
2550 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2551 error("failed to read delta base object %s"
2552 " at offset %"PRIuMAX" from %s",
2553 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2554 p->pack_name);
2555 mark_bad_packed_object(p, base_sha1);
2556 base = read_object(base_sha1, &type, &base_size);
2557 external_base = base;
2561 i = --delta_stack_nr;
2562 obj_offset = delta_stack[i].obj_offset;
2563 curpos = delta_stack[i].curpos;
2564 delta_size = delta_stack[i].size;
2566 if (!base)
2567 continue;
2569 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2571 if (!delta_data) {
2572 error("failed to unpack compressed delta "
2573 "at offset %"PRIuMAX" from %s",
2574 (uintmax_t)curpos, p->pack_name);
2575 data = NULL;
2576 free(external_base);
2577 continue;
2580 data = patch_delta(base, base_size,
2581 delta_data, delta_size,
2582 &size);
2585 * We could not apply the delta; warn the user, but keep going.
2586 * Our failure will be noticed either in the next iteration of
2587 * the loop, or if this is the final delta, in the caller when
2588 * we return NULL. Those code paths will take care of making
2589 * a more explicit warning and retrying with another copy of
2590 * the object.
2592 if (!data)
2593 error("failed to apply delta");
2595 free(delta_data);
2596 free(external_base);
2599 *final_type = type;
2600 *final_size = size;
2602 unuse_pack(&w_curs);
2604 if (delta_stack != small_delta_stack)
2605 free(delta_stack);
2607 return data;
2610 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2611 uint32_t n)
2613 const unsigned char *index = p->index_data;
2614 if (!index) {
2615 if (open_pack_index(p))
2616 return NULL;
2617 index = p->index_data;
2619 if (n >= p->num_objects)
2620 return NULL;
2621 index += 4 * 256;
2622 if (p->index_version == 1) {
2623 return index + 24 * n + 4;
2624 } else {
2625 index += 8;
2626 return index + 20 * n;
2630 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2632 const unsigned char *ptr = vptr;
2633 const unsigned char *start = p->index_data;
2634 const unsigned char *end = start + p->index_size;
2635 if (ptr < start)
2636 die(_("offset before start of pack index for %s (corrupt index?)"),
2637 p->pack_name);
2638 /* No need to check for underflow; .idx files must be at least 8 bytes */
2639 if (ptr >= end - 8)
2640 die(_("offset beyond end of pack index for %s (truncated index?)"),
2641 p->pack_name);
2644 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2646 const unsigned char *index = p->index_data;
2647 index += 4 * 256;
2648 if (p->index_version == 1) {
2649 return ntohl(*((uint32_t *)(index + 24 * n)));
2650 } else {
2651 uint32_t off;
2652 index += 8 + p->num_objects * (20 + 4);
2653 off = ntohl(*((uint32_t *)(index + 4 * n)));
2654 if (!(off & 0x80000000))
2655 return off;
2656 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2657 check_pack_index_ptr(p, index);
2658 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2659 ntohl(*((uint32_t *)(index + 4)));
2663 off_t find_pack_entry_one(const unsigned char *sha1,
2664 struct packed_git *p)
2666 const uint32_t *level1_ofs = p->index_data;
2667 const unsigned char *index = p->index_data;
2668 unsigned hi, lo, stride;
2669 static int use_lookup = -1;
2670 static int debug_lookup = -1;
2672 if (debug_lookup < 0)
2673 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2675 if (!index) {
2676 if (open_pack_index(p))
2677 return 0;
2678 level1_ofs = p->index_data;
2679 index = p->index_data;
2681 if (p->index_version > 1) {
2682 level1_ofs += 2;
2683 index += 8;
2685 index += 4 * 256;
2686 hi = ntohl(level1_ofs[*sha1]);
2687 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2688 if (p->index_version > 1) {
2689 stride = 20;
2690 } else {
2691 stride = 24;
2692 index += 4;
2695 if (debug_lookup)
2696 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2697 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2699 if (use_lookup < 0)
2700 use_lookup = !!getenv("GIT_USE_LOOKUP");
2701 if (use_lookup) {
2702 int pos = sha1_entry_pos(index, stride, 0,
2703 lo, hi, p->num_objects, sha1);
2704 if (pos < 0)
2705 return 0;
2706 return nth_packed_object_offset(p, pos);
2709 do {
2710 unsigned mi = (lo + hi) / 2;
2711 int cmp = hashcmp(index + mi * stride, sha1);
2713 if (debug_lookup)
2714 printf("lo %u hi %u rg %u mi %u\n",
2715 lo, hi, hi - lo, mi);
2716 if (!cmp)
2717 return nth_packed_object_offset(p, mi);
2718 if (cmp > 0)
2719 hi = mi;
2720 else
2721 lo = mi+1;
2722 } while (lo < hi);
2723 return 0;
2726 int is_pack_valid(struct packed_git *p)
2728 /* An already open pack is known to be valid. */
2729 if (p->pack_fd != -1)
2730 return 1;
2732 /* If the pack has one window completely covering the
2733 * file size, the pack is known to be valid even if
2734 * the descriptor is not currently open.
2736 if (p->windows) {
2737 struct pack_window *w = p->windows;
2739 if (!w->offset && w->len == p->pack_size)
2740 return 1;
2743 /* Force the pack to open to prove its valid. */
2744 return !open_packed_git(p);
2747 static int fill_pack_entry(const unsigned char *sha1,
2748 struct pack_entry *e,
2749 struct packed_git *p)
2751 off_t offset;
2753 if (p->num_bad_objects) {
2754 unsigned i;
2755 for (i = 0; i < p->num_bad_objects; i++)
2756 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2757 return 0;
2760 offset = find_pack_entry_one(sha1, p);
2761 if (!offset)
2762 return 0;
2765 * We are about to tell the caller where they can locate the
2766 * requested object. We better make sure the packfile is
2767 * still here and can be accessed before supplying that
2768 * answer, as it may have been deleted since the index was
2769 * loaded!
2771 if (!is_pack_valid(p))
2772 return 0;
2773 e->offset = offset;
2774 e->p = p;
2775 hashcpy(e->sha1, sha1);
2776 return 1;
2780 * Iff a pack file contains the object named by sha1, return true and
2781 * store its location to e.
2783 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2785 struct mru_entry *p;
2787 prepare_packed_git();
2788 if (!packed_git)
2789 return 0;
2791 for (p = packed_git_mru->head; p; p = p->next) {
2792 if (fill_pack_entry(sha1, e, p->item)) {
2793 mru_mark(packed_git_mru, p);
2794 return 1;
2797 return 0;
2800 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2801 struct packed_git *packs)
2803 struct packed_git *p;
2805 for (p = packs; p; p = p->next) {
2806 if (find_pack_entry_one(sha1, p))
2807 return p;
2809 return NULL;
2813 static int sha1_loose_object_info(const unsigned char *sha1,
2814 struct object_info *oi,
2815 int flags)
2817 int status = 0;
2818 unsigned long mapsize;
2819 void *map;
2820 git_zstream stream;
2821 char hdr[32];
2822 struct strbuf hdrbuf = STRBUF_INIT;
2824 if (oi->delta_base_sha1)
2825 hashclr(oi->delta_base_sha1);
2828 * If we don't care about type or size, then we don't
2829 * need to look inside the object at all. Note that we
2830 * do not optimize out the stat call, even if the
2831 * caller doesn't care about the disk-size, since our
2832 * return value implicitly indicates whether the
2833 * object even exists.
2835 if (!oi->typep && !oi->typename && !oi->sizep) {
2836 const char *path;
2837 struct stat st;
2838 if (stat_sha1_file(sha1, &st, &path) < 0)
2839 return -1;
2840 if (oi->disk_sizep)
2841 *oi->disk_sizep = st.st_size;
2842 return 0;
2845 map = map_sha1_file(sha1, &mapsize);
2846 if (!map)
2847 return -1;
2848 if (oi->disk_sizep)
2849 *oi->disk_sizep = mapsize;
2850 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2851 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2852 status = error("unable to unpack %s header with --allow-unknown-type",
2853 sha1_to_hex(sha1));
2854 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2855 status = error("unable to unpack %s header",
2856 sha1_to_hex(sha1));
2857 if (status < 0)
2858 ; /* Do nothing */
2859 else if (hdrbuf.len) {
2860 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2861 status = error("unable to parse %s header with --allow-unknown-type",
2862 sha1_to_hex(sha1));
2863 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2864 status = error("unable to parse %s header", sha1_to_hex(sha1));
2865 git_inflate_end(&stream);
2866 munmap(map, mapsize);
2867 if (status && oi->typep)
2868 *oi->typep = status;
2869 strbuf_release(&hdrbuf);
2870 return 0;
2873 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2875 struct cached_object *co;
2876 struct pack_entry e;
2877 int rtype;
2878 enum object_type real_type;
2879 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2881 co = find_cached_object(real);
2882 if (co) {
2883 if (oi->typep)
2884 *(oi->typep) = co->type;
2885 if (oi->sizep)
2886 *(oi->sizep) = co->size;
2887 if (oi->disk_sizep)
2888 *(oi->disk_sizep) = 0;
2889 if (oi->delta_base_sha1)
2890 hashclr(oi->delta_base_sha1);
2891 if (oi->typename)
2892 strbuf_addstr(oi->typename, typename(co->type));
2893 oi->whence = OI_CACHED;
2894 return 0;
2897 if (!find_pack_entry(real, &e)) {
2898 /* Most likely it's a loose object. */
2899 if (!sha1_loose_object_info(real, oi, flags)) {
2900 oi->whence = OI_LOOSE;
2901 return 0;
2904 /* Not a loose object; someone else may have just packed it. */
2905 reprepare_packed_git();
2906 if (!find_pack_entry(real, &e))
2907 return -1;
2911 * packed_object_info() does not follow the delta chain to
2912 * find out the real type, unless it is given oi->typep.
2914 if (oi->typename && !oi->typep)
2915 oi->typep = &real_type;
2917 rtype = packed_object_info(e.p, e.offset, oi);
2918 if (rtype < 0) {
2919 mark_bad_packed_object(e.p, real);
2920 if (oi->typep == &real_type)
2921 oi->typep = NULL;
2922 return sha1_object_info_extended(real, oi, 0);
2923 } else if (in_delta_base_cache(e.p, e.offset)) {
2924 oi->whence = OI_DBCACHED;
2925 } else {
2926 oi->whence = OI_PACKED;
2927 oi->u.packed.offset = e.offset;
2928 oi->u.packed.pack = e.p;
2929 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2930 rtype == OBJ_OFS_DELTA);
2932 if (oi->typename)
2933 strbuf_addstr(oi->typename, typename(*oi->typep));
2934 if (oi->typep == &real_type)
2935 oi->typep = NULL;
2937 return 0;
2940 /* returns enum object_type or negative */
2941 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2943 enum object_type type;
2944 struct object_info oi = OBJECT_INFO_INIT;
2946 oi.typep = &type;
2947 oi.sizep = sizep;
2948 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2949 return -1;
2950 return type;
2953 static void *read_packed_sha1(const unsigned char *sha1,
2954 enum object_type *type, unsigned long *size)
2956 struct pack_entry e;
2957 void *data;
2959 if (!find_pack_entry(sha1, &e))
2960 return NULL;
2961 data = cache_or_unpack_entry(e.p, e.offset, size, type);
2962 if (!data) {
2964 * We're probably in deep shit, but let's try to fetch
2965 * the required object anyway from another pack or loose.
2966 * This should happen only in the presence of a corrupted
2967 * pack, and is better than failing outright.
2969 error("failed to read object %s at offset %"PRIuMAX" from %s",
2970 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2971 mark_bad_packed_object(e.p, sha1);
2972 data = read_object(sha1, type, size);
2974 return data;
2977 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2978 unsigned char *sha1)
2980 struct cached_object *co;
2982 hash_sha1_file(buf, len, typename(type), sha1);
2983 if (has_sha1_file(sha1) || find_cached_object(sha1))
2984 return 0;
2985 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2986 co = &cached_objects[cached_object_nr++];
2987 co->size = len;
2988 co->type = type;
2989 co->buf = xmalloc(len);
2990 memcpy(co->buf, buf, len);
2991 hashcpy(co->sha1, sha1);
2992 return 0;
2995 static void *read_object(const unsigned char *sha1, enum object_type *type,
2996 unsigned long *size)
2998 unsigned long mapsize;
2999 void *map, *buf;
3000 struct cached_object *co;
3002 co = find_cached_object(sha1);
3003 if (co) {
3004 *type = co->type;
3005 *size = co->size;
3006 return xmemdupz(co->buf, co->size);
3009 buf = read_packed_sha1(sha1, type, size);
3010 if (buf)
3011 return buf;
3012 map = map_sha1_file(sha1, &mapsize);
3013 if (map) {
3014 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
3015 munmap(map, mapsize);
3016 return buf;
3018 reprepare_packed_git();
3019 return read_packed_sha1(sha1, type, size);
3023 * This function dies on corrupt objects; the callers who want to
3024 * deal with them should arrange to call read_object() and give error
3025 * messages themselves.
3027 void *read_sha1_file_extended(const unsigned char *sha1,
3028 enum object_type *type,
3029 unsigned long *size,
3030 unsigned flag)
3032 void *data;
3033 const struct packed_git *p;
3034 const char *path;
3035 struct stat st;
3036 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
3038 errno = 0;
3039 data = read_object(repl, type, size);
3040 if (data)
3041 return data;
3043 if (errno && errno != ENOENT)
3044 die_errno("failed to read object %s", sha1_to_hex(sha1));
3046 /* die if we replaced an object with one that does not exist */
3047 if (repl != sha1)
3048 die("replacement %s not found for %s",
3049 sha1_to_hex(repl), sha1_to_hex(sha1));
3051 if (!stat_sha1_file(repl, &st, &path))
3052 die("loose object %s (stored in %s) is corrupt",
3053 sha1_to_hex(repl), path);
3055 if ((p = has_packed_and_bad(repl)) != NULL)
3056 die("packed object %s (stored in %s) is corrupt",
3057 sha1_to_hex(repl), p->pack_name);
3059 return NULL;
3062 void *read_object_with_reference(const unsigned char *sha1,
3063 const char *required_type_name,
3064 unsigned long *size,
3065 unsigned char *actual_sha1_return)
3067 enum object_type type, required_type;
3068 void *buffer;
3069 unsigned long isize;
3070 unsigned char actual_sha1[20];
3072 required_type = type_from_string(required_type_name);
3073 hashcpy(actual_sha1, sha1);
3074 while (1) {
3075 int ref_length = -1;
3076 const char *ref_type = NULL;
3078 buffer = read_sha1_file(actual_sha1, &type, &isize);
3079 if (!buffer)
3080 return NULL;
3081 if (type == required_type) {
3082 *size = isize;
3083 if (actual_sha1_return)
3084 hashcpy(actual_sha1_return, actual_sha1);
3085 return buffer;
3087 /* Handle references */
3088 else if (type == OBJ_COMMIT)
3089 ref_type = "tree ";
3090 else if (type == OBJ_TAG)
3091 ref_type = "object ";
3092 else {
3093 free(buffer);
3094 return NULL;
3096 ref_length = strlen(ref_type);
3098 if (ref_length + 40 > isize ||
3099 memcmp(buffer, ref_type, ref_length) ||
3100 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3101 free(buffer);
3102 return NULL;
3104 free(buffer);
3105 /* Now we have the ID of the referred-to object in
3106 * actual_sha1. Check again. */
3110 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3111 const char *type, unsigned char *sha1,
3112 char *hdr, int *hdrlen)
3114 git_SHA_CTX c;
3116 /* Generate the header */
3117 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3119 /* Sha1.. */
3120 git_SHA1_Init(&c);
3121 git_SHA1_Update(&c, hdr, *hdrlen);
3122 git_SHA1_Update(&c, buf, len);
3123 git_SHA1_Final(sha1, &c);
3127 * Move the just written object into its final resting place.
3129 int finalize_object_file(const char *tmpfile, const char *filename)
3131 int ret = 0;
3133 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3134 goto try_rename;
3135 else if (link(tmpfile, filename))
3136 ret = errno;
3139 * Coda hack - coda doesn't like cross-directory links,
3140 * so we fall back to a rename, which will mean that it
3141 * won't be able to check collisions, but that's not a
3142 * big deal.
3144 * The same holds for FAT formatted media.
3146 * When this succeeds, we just return. We have nothing
3147 * left to unlink.
3149 if (ret && ret != EEXIST) {
3150 try_rename:
3151 if (!rename(tmpfile, filename))
3152 goto out;
3153 ret = errno;
3155 unlink_or_warn(tmpfile);
3156 if (ret) {
3157 if (ret != EEXIST) {
3158 return error_errno("unable to write sha1 filename %s", filename);
3160 /* FIXME!!! Collision check here ? */
3163 out:
3164 if (adjust_shared_perm(filename))
3165 return error("unable to set permission to '%s'", filename);
3166 return 0;
3169 static int write_buffer(int fd, const void *buf, size_t len)
3171 if (write_in_full(fd, buf, len) < 0)
3172 return error_errno("file write error");
3173 return 0;
3176 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3177 unsigned char *sha1)
3179 char hdr[32];
3180 int hdrlen = sizeof(hdr);
3181 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3182 return 0;
3185 /* Finalize a file on disk, and close it. */
3186 static void close_sha1_file(int fd)
3188 if (fsync_object_files)
3189 fsync_or_die(fd, "sha1 file");
3190 if (close(fd) != 0)
3191 die_errno("error when closing sha1 file");
3194 /* Size of directory component, including the ending '/' */
3195 static inline int directory_size(const char *filename)
3197 const char *s = strrchr(filename, '/');
3198 if (!s)
3199 return 0;
3200 return s - filename + 1;
3204 * This creates a temporary file in the same directory as the final
3205 * 'filename'
3207 * We want to avoid cross-directory filename renames, because those
3208 * can have problems on various filesystems (FAT, NFS, Coda).
3210 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3212 int fd, dirlen = directory_size(filename);
3214 strbuf_reset(tmp);
3215 strbuf_add(tmp, filename, dirlen);
3216 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3217 fd = git_mkstemp_mode(tmp->buf, 0444);
3218 if (fd < 0 && dirlen && errno == ENOENT) {
3220 * Make sure the directory exists; note that the contents
3221 * of the buffer are undefined after mkstemp returns an
3222 * error, so we have to rewrite the whole buffer from
3223 * scratch.
3225 strbuf_reset(tmp);
3226 strbuf_add(tmp, filename, dirlen - 1);
3227 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3228 return -1;
3229 if (adjust_shared_perm(tmp->buf))
3230 return -1;
3232 /* Try again */
3233 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3234 fd = git_mkstemp_mode(tmp->buf, 0444);
3236 return fd;
3239 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3240 const void *buf, unsigned long len, time_t mtime)
3242 int fd, ret;
3243 unsigned char compressed[4096];
3244 git_zstream stream;
3245 git_SHA_CTX c;
3246 unsigned char parano_sha1[20];
3247 static struct strbuf tmp_file = STRBUF_INIT;
3248 const char *filename = sha1_file_name(sha1);
3250 fd = create_tmpfile(&tmp_file, filename);
3251 if (fd < 0) {
3252 if (errno == EACCES)
3253 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3254 else
3255 return error_errno("unable to create temporary file");
3258 /* Set it up */
3259 git_deflate_init(&stream, zlib_compression_level);
3260 stream.next_out = compressed;
3261 stream.avail_out = sizeof(compressed);
3262 git_SHA1_Init(&c);
3264 /* First header.. */
3265 stream.next_in = (unsigned char *)hdr;
3266 stream.avail_in = hdrlen;
3267 while (git_deflate(&stream, 0) == Z_OK)
3268 ; /* nothing */
3269 git_SHA1_Update(&c, hdr, hdrlen);
3271 /* Then the data itself.. */
3272 stream.next_in = (void *)buf;
3273 stream.avail_in = len;
3274 do {
3275 unsigned char *in0 = stream.next_in;
3276 ret = git_deflate(&stream, Z_FINISH);
3277 git_SHA1_Update(&c, in0, stream.next_in - in0);
3278 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3279 die("unable to write sha1 file");
3280 stream.next_out = compressed;
3281 stream.avail_out = sizeof(compressed);
3282 } while (ret == Z_OK);
3284 if (ret != Z_STREAM_END)
3285 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3286 ret = git_deflate_end_gently(&stream);
3287 if (ret != Z_OK)
3288 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3289 git_SHA1_Final(parano_sha1, &c);
3290 if (hashcmp(sha1, parano_sha1) != 0)
3291 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3293 close_sha1_file(fd);
3295 if (mtime) {
3296 struct utimbuf utb;
3297 utb.actime = mtime;
3298 utb.modtime = mtime;
3299 if (utime(tmp_file.buf, &utb) < 0)
3300 warning_errno("failed utime() on %s", tmp_file.buf);
3303 return finalize_object_file(tmp_file.buf, filename);
3306 static int freshen_loose_object(const unsigned char *sha1)
3308 return check_and_freshen(sha1, 1);
3311 static int freshen_packed_object(const unsigned char *sha1)
3313 struct pack_entry e;
3314 if (!find_pack_entry(sha1, &e))
3315 return 0;
3316 if (e.p->freshened)
3317 return 1;
3318 if (!freshen_file(e.p->pack_name))
3319 return 0;
3320 e.p->freshened = 1;
3321 return 1;
3324 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3326 char hdr[32];
3327 int hdrlen = sizeof(hdr);
3329 /* Normally if we have it in the pack then we do not bother writing
3330 * it out into .git/objects/??/?{38} file.
3332 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3333 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3334 return 0;
3335 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3338 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3339 unsigned char *sha1, unsigned flags)
3341 char *header;
3342 int hdrlen, status = 0;
3344 /* type string, SP, %lu of the length plus NUL must fit this */
3345 hdrlen = strlen(type) + 32;
3346 header = xmalloc(hdrlen);
3347 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3349 if (!(flags & HASH_WRITE_OBJECT))
3350 goto cleanup;
3351 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3352 goto cleanup;
3353 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3355 cleanup:
3356 free(header);
3357 return status;
3360 int force_object_loose(const unsigned char *sha1, time_t mtime)
3362 void *buf;
3363 unsigned long len;
3364 enum object_type type;
3365 char hdr[32];
3366 int hdrlen;
3367 int ret;
3369 if (has_loose_object(sha1))
3370 return 0;
3371 buf = read_packed_sha1(sha1, &type, &len);
3372 if (!buf)
3373 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3374 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3375 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3376 free(buf);
3378 return ret;
3381 int has_pack_index(const unsigned char *sha1)
3383 struct stat st;
3384 if (stat(sha1_pack_index_name(sha1), &st))
3385 return 0;
3386 return 1;
3389 int has_sha1_pack(const unsigned char *sha1)
3391 struct pack_entry e;
3392 return find_pack_entry(sha1, &e);
3395 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3397 struct pack_entry e;
3399 if (find_pack_entry(sha1, &e))
3400 return 1;
3401 if (has_loose_object(sha1))
3402 return 1;
3403 if (flags & HAS_SHA1_QUICK)
3404 return 0;
3405 reprepare_packed_git();
3406 return find_pack_entry(sha1, &e);
3409 int has_object_file(const struct object_id *oid)
3411 return has_sha1_file(oid->hash);
3414 int has_object_file_with_flags(const struct object_id *oid, int flags)
3416 return has_sha1_file_with_flags(oid->hash, flags);
3419 static void check_tree(const void *buf, size_t size)
3421 struct tree_desc desc;
3422 struct name_entry entry;
3424 init_tree_desc(&desc, buf, size);
3425 while (tree_entry(&desc, &entry))
3426 /* do nothing
3427 * tree_entry() will die() on malformed entries */
3431 static void check_commit(const void *buf, size_t size)
3433 struct commit c;
3434 memset(&c, 0, sizeof(c));
3435 if (parse_commit_buffer(&c, buf, size))
3436 die("corrupt commit");
3439 static void check_tag(const void *buf, size_t size)
3441 struct tag t;
3442 memset(&t, 0, sizeof(t));
3443 if (parse_tag_buffer(&t, buf, size))
3444 die("corrupt tag");
3447 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3448 enum object_type type,
3449 const char *path, unsigned flags)
3451 int ret, re_allocated = 0;
3452 int write_object = flags & HASH_WRITE_OBJECT;
3454 if (!type)
3455 type = OBJ_BLOB;
3458 * Convert blobs to git internal format
3460 if ((type == OBJ_BLOB) && path) {
3461 struct strbuf nbuf = STRBUF_INIT;
3462 if (convert_to_git(path, buf, size, &nbuf,
3463 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3464 buf = strbuf_detach(&nbuf, &size);
3465 re_allocated = 1;
3468 if (flags & HASH_FORMAT_CHECK) {
3469 if (type == OBJ_TREE)
3470 check_tree(buf, size);
3471 if (type == OBJ_COMMIT)
3472 check_commit(buf, size);
3473 if (type == OBJ_TAG)
3474 check_tag(buf, size);
3477 if (write_object)
3478 ret = write_sha1_file(buf, size, typename(type), sha1);
3479 else
3480 ret = hash_sha1_file(buf, size, typename(type), sha1);
3481 if (re_allocated)
3482 free(buf);
3483 return ret;
3486 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3487 const char *path, unsigned flags)
3489 int ret;
3490 const int write_object = flags & HASH_WRITE_OBJECT;
3491 struct strbuf sbuf = STRBUF_INIT;
3493 assert(path);
3494 assert(would_convert_to_git_filter_fd(path));
3496 convert_to_git_filter_fd(path, fd, &sbuf,
3497 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3499 if (write_object)
3500 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3501 sha1);
3502 else
3503 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3504 sha1);
3505 strbuf_release(&sbuf);
3506 return ret;
3509 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3510 const char *path, unsigned flags)
3512 struct strbuf sbuf = STRBUF_INIT;
3513 int ret;
3515 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3516 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3517 else
3518 ret = -1;
3519 strbuf_release(&sbuf);
3520 return ret;
3523 #define SMALL_FILE_SIZE (32*1024)
3525 static int index_core(unsigned char *sha1, int fd, size_t size,
3526 enum object_type type, const char *path,
3527 unsigned flags)
3529 int ret;
3531 if (!size) {
3532 ret = index_mem(sha1, "", size, type, path, flags);
3533 } else if (size <= SMALL_FILE_SIZE) {
3534 char *buf = xmalloc(size);
3535 if (size == read_in_full(fd, buf, size))
3536 ret = index_mem(sha1, buf, size, type, path, flags);
3537 else
3538 ret = error_errno("short read");
3539 free(buf);
3540 } else {
3541 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3542 ret = index_mem(sha1, buf, size, type, path, flags);
3543 munmap(buf, size);
3545 return ret;
3549 * This creates one packfile per large blob unless bulk-checkin
3550 * machinery is "plugged".
3552 * This also bypasses the usual "convert-to-git" dance, and that is on
3553 * purpose. We could write a streaming version of the converting
3554 * functions and insert that before feeding the data to fast-import
3555 * (or equivalent in-core API described above). However, that is
3556 * somewhat complicated, as we do not know the size of the filter
3557 * result, which we need to know beforehand when writing a git object.
3558 * Since the primary motivation for trying to stream from the working
3559 * tree file and to avoid mmaping it in core is to deal with large
3560 * binary blobs, they generally do not want to get any conversion, and
3561 * callers should avoid this code path when filters are requested.
3563 static int index_stream(unsigned char *sha1, int fd, size_t size,
3564 enum object_type type, const char *path,
3565 unsigned flags)
3567 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3570 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3571 enum object_type type, const char *path, unsigned flags)
3573 int ret;
3576 * Call xsize_t() only when needed to avoid potentially unnecessary
3577 * die() for large files.
3579 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3580 ret = index_stream_convert_blob(sha1, fd, path, flags);
3581 else if (!S_ISREG(st->st_mode))
3582 ret = index_pipe(sha1, fd, type, path, flags);
3583 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3584 (path && would_convert_to_git(path)))
3585 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3586 flags);
3587 else
3588 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3589 flags);
3590 close(fd);
3591 return ret;
3594 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3596 int fd;
3597 struct strbuf sb = STRBUF_INIT;
3599 switch (st->st_mode & S_IFMT) {
3600 case S_IFREG:
3601 fd = open(path, O_RDONLY);
3602 if (fd < 0)
3603 return error_errno("open(\"%s\")", path);
3604 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3605 return error("%s: failed to insert into database",
3606 path);
3607 break;
3608 case S_IFLNK:
3609 if (strbuf_readlink(&sb, path, st->st_size))
3610 return error_errno("readlink(\"%s\")", path);
3611 if (!(flags & HASH_WRITE_OBJECT))
3612 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3613 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3614 return error("%s: failed to insert into database",
3615 path);
3616 strbuf_release(&sb);
3617 break;
3618 case S_IFDIR:
3619 return resolve_gitlink_ref(path, "HEAD", sha1);
3620 default:
3621 return error("%s: unsupported file type", path);
3623 return 0;
3626 int read_pack_header(int fd, struct pack_header *header)
3628 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3629 /* "eof before pack header was fully read" */
3630 return PH_ERROR_EOF;
3632 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3633 /* "protocol error (pack signature mismatch detected)" */
3634 return PH_ERROR_PACK_SIGNATURE;
3635 if (!pack_version_ok(header->hdr_version))
3636 /* "protocol error (pack version unsupported)" */
3637 return PH_ERROR_PROTOCOL;
3638 return 0;
3641 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3643 enum object_type type = sha1_object_info(sha1, NULL);
3644 if (type < 0)
3645 die("%s is not a valid object", sha1_to_hex(sha1));
3646 if (type != expect)
3647 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3648 typename(expect));
3651 static int for_each_file_in_obj_subdir(int subdir_nr,
3652 struct strbuf *path,
3653 each_loose_object_fn obj_cb,
3654 each_loose_cruft_fn cruft_cb,
3655 each_loose_subdir_fn subdir_cb,
3656 void *data)
3658 size_t baselen = path->len;
3659 DIR *dir = opendir(path->buf);
3660 struct dirent *de;
3661 int r = 0;
3663 if (!dir) {
3664 if (errno == ENOENT)
3665 return 0;
3666 return error_errno("unable to open %s", path->buf);
3669 while ((de = readdir(dir))) {
3670 if (is_dot_or_dotdot(de->d_name))
3671 continue;
3673 strbuf_setlen(path, baselen);
3674 strbuf_addf(path, "/%s", de->d_name);
3676 if (strlen(de->d_name) == 38) {
3677 char hex[41];
3678 unsigned char sha1[20];
3680 snprintf(hex, sizeof(hex), "%02x%s",
3681 subdir_nr, de->d_name);
3682 if (!get_sha1_hex(hex, sha1)) {
3683 if (obj_cb) {
3684 r = obj_cb(sha1, path->buf, data);
3685 if (r)
3686 break;
3688 continue;
3692 if (cruft_cb) {
3693 r = cruft_cb(de->d_name, path->buf, data);
3694 if (r)
3695 break;
3698 closedir(dir);
3700 strbuf_setlen(path, baselen);
3701 if (!r && subdir_cb)
3702 r = subdir_cb(subdir_nr, path->buf, data);
3704 return r;
3707 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3708 each_loose_object_fn obj_cb,
3709 each_loose_cruft_fn cruft_cb,
3710 each_loose_subdir_fn subdir_cb,
3711 void *data)
3713 size_t baselen = path->len;
3714 int r = 0;
3715 int i;
3717 for (i = 0; i < 256; i++) {
3718 strbuf_addf(path, "/%02x", i);
3719 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3720 subdir_cb, data);
3721 strbuf_setlen(path, baselen);
3722 if (r)
3723 break;
3726 return r;
3729 int for_each_loose_file_in_objdir(const char *path,
3730 each_loose_object_fn obj_cb,
3731 each_loose_cruft_fn cruft_cb,
3732 each_loose_subdir_fn subdir_cb,
3733 void *data)
3735 struct strbuf buf = STRBUF_INIT;
3736 int r;
3738 strbuf_addstr(&buf, path);
3739 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3740 subdir_cb, data);
3741 strbuf_release(&buf);
3743 return r;
3746 struct loose_alt_odb_data {
3747 each_loose_object_fn *cb;
3748 void *data;
3751 static int loose_from_alt_odb(struct alternate_object_database *alt,
3752 void *vdata)
3754 struct loose_alt_odb_data *data = vdata;
3755 struct strbuf buf = STRBUF_INIT;
3756 int r;
3758 strbuf_addstr(&buf, alt->path);
3759 r = for_each_loose_file_in_objdir_buf(&buf,
3760 data->cb, NULL, NULL,
3761 data->data);
3762 strbuf_release(&buf);
3763 return r;
3766 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3768 struct loose_alt_odb_data alt;
3769 int r;
3771 r = for_each_loose_file_in_objdir(get_object_directory(),
3772 cb, NULL, NULL, data);
3773 if (r)
3774 return r;
3776 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3777 return 0;
3779 alt.cb = cb;
3780 alt.data = data;
3781 return foreach_alt_odb(loose_from_alt_odb, &alt);
3784 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3786 uint32_t i;
3787 int r = 0;
3789 for (i = 0; i < p->num_objects; i++) {
3790 const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3792 if (!sha1)
3793 return error("unable to get sha1 of object %u in %s",
3794 i, p->pack_name);
3796 r = cb(sha1, p, i, data);
3797 if (r)
3798 break;
3800 return r;
3803 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3805 struct packed_git *p;
3806 int r = 0;
3807 int pack_errors = 0;
3809 prepare_packed_git();
3810 for (p = packed_git; p; p = p->next) {
3811 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3812 continue;
3813 if (open_pack_index(p)) {
3814 pack_errors = 1;
3815 continue;
3817 r = for_each_object_in_pack(p, cb, data);
3818 if (r)
3819 break;
3821 return r ? r : pack_errors;
3824 static int check_stream_sha1(git_zstream *stream,
3825 const char *hdr,
3826 unsigned long size,
3827 const char *path,
3828 const unsigned char *expected_sha1)
3830 git_SHA_CTX c;
3831 unsigned char real_sha1[GIT_SHA1_RAWSZ];
3832 unsigned char buf[4096];
3833 unsigned long total_read;
3834 int status = Z_OK;
3836 git_SHA1_Init(&c);
3837 git_SHA1_Update(&c, hdr, stream->total_out);
3840 * We already read some bytes into hdr, but the ones up to the NUL
3841 * do not count against the object's content size.
3843 total_read = stream->total_out - strlen(hdr) - 1;
3846 * This size comparison must be "<=" to read the final zlib packets;
3847 * see the comment in unpack_sha1_rest for details.
3849 while (total_read <= size &&
3850 (status == Z_OK || status == Z_BUF_ERROR)) {
3851 stream->next_out = buf;
3852 stream->avail_out = sizeof(buf);
3853 if (size - total_read < stream->avail_out)
3854 stream->avail_out = size - total_read;
3855 status = git_inflate(stream, Z_FINISH);
3856 git_SHA1_Update(&c, buf, stream->next_out - buf);
3857 total_read += stream->next_out - buf;
3859 git_inflate_end(stream);
3861 if (status != Z_STREAM_END) {
3862 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
3863 return -1;
3865 if (stream->avail_in) {
3866 error("garbage at end of loose object '%s'",
3867 sha1_to_hex(expected_sha1));
3868 return -1;
3871 git_SHA1_Final(real_sha1, &c);
3872 if (hashcmp(expected_sha1, real_sha1)) {
3873 error("sha1 mismatch for %s (expected %s)", path,
3874 sha1_to_hex(expected_sha1));
3875 return -1;
3878 return 0;
3881 int read_loose_object(const char *path,
3882 const unsigned char *expected_sha1,
3883 enum object_type *type,
3884 unsigned long *size,
3885 void **contents)
3887 int ret = -1;
3888 int fd = -1;
3889 void *map = NULL;
3890 unsigned long mapsize;
3891 git_zstream stream;
3892 char hdr[32];
3894 *contents = NULL;
3896 map = map_sha1_file_1(path, NULL, &mapsize);
3897 if (!map) {
3898 error_errno("unable to mmap %s", path);
3899 goto out;
3902 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
3903 error("unable to unpack header of %s", path);
3904 goto out;
3907 *type = parse_sha1_header(hdr, size);
3908 if (*type < 0) {
3909 error("unable to parse header of %s", path);
3910 git_inflate_end(&stream);
3911 goto out;
3914 if (*type == OBJ_BLOB) {
3915 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
3916 goto out;
3917 } else {
3918 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
3919 if (!*contents) {
3920 error("unable to unpack contents of %s", path);
3921 git_inflate_end(&stream);
3922 goto out;
3924 if (check_sha1_signature(expected_sha1, *contents,
3925 *size, typename(*type))) {
3926 error("sha1 mismatch for %s (expected %s)", path,
3927 sha1_to_hex(expected_sha1));
3928 free(*contents);
3929 goto out;
3933 ret = 0; /* everything checks out */
3935 out:
3936 if (map)
3937 munmap(map, mapsize);
3938 if (fd >= 0)
3939 close(fd);
3940 return ret;