2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
33 static size_t sz_fmt(size_t s) { return s; }
36 const unsigned char null_sha1[20];
38 int safe_create_leading_directories(char *path)
40 char *pos = path + offset_1st_component(path);
44 pos = strchr(pos, '/');
52 if (!stat(path, &st)) {
54 if (!S_ISDIR(st.st_mode)) {
59 else if (mkdir(path, 0777)) {
63 else if (adjust_shared_perm(path)) {
72 int safe_create_leading_directories_const(const char *path)
74 /* path points to cache entries, so xstrdup before messing with it */
75 char *buf = xstrdup(path);
76 int result = safe_create_leading_directories(buf);
81 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
84 for (i = 0; i < 20; i++) {
85 static char hex[] = "0123456789abcdef";
86 unsigned int val = sha1[i];
87 char *pos = pathbuf + i*2 + (i > 0);
88 *pos++ = hex[val >> 4];
89 *pos = hex[val & 0xf];
94 * NOTE! This returns a statically allocated buffer, so you have to be
95 * careful about using it. Do an "xstrdup()" if you need to save the
98 * Also note that this returns the location for creating. Reading
99 * SHA1 file can happen from any alternate directory listed in the
100 * DB_ENVIRONMENT environment variable if it is not found in
101 * the primary object database.
103 char *sha1_file_name(const unsigned char *sha1)
105 static char buf[PATH_MAX];
109 objdir = get_object_directory();
110 len = strlen(objdir);
112 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
113 if (len + 43 > PATH_MAX)
114 die("insanely long object directory %s", objdir);
115 memcpy(buf, objdir, len);
119 fill_sha1_path(buf + len + 1, sha1);
123 static char *sha1_get_pack_name(const unsigned char *sha1,
124 char **name, char **base, const char *which)
126 static const char hex[] = "0123456789abcdef";
131 const char *sha1_file_directory = get_object_directory();
132 int len = strlen(sha1_file_directory);
133 *base = xmalloc(len + 60);
134 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
135 sha1_file_directory, which);
136 *name = *base + len + 11;
141 for (i = 0; i < 20; i++) {
142 unsigned int val = *sha1++;
143 *buf++ = hex[val >> 4];
144 *buf++ = hex[val & 0xf];
150 char *sha1_pack_name(const unsigned char *sha1)
152 static char *name, *base;
154 return sha1_get_pack_name(sha1, &name, &base, "pack");
157 char *sha1_pack_index_name(const unsigned char *sha1)
159 static char *name, *base;
161 return sha1_get_pack_name(sha1, &name, &base, "idx");
164 struct alternate_object_database *alt_odb_list;
165 static struct alternate_object_database **alt_odb_tail;
167 static void read_info_alternates(const char * alternates, int depth);
170 * Prepare alternate object database registry.
172 * The variable alt_odb_list points at the list of struct
173 * alternate_object_database. The elements on this list come from
174 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
175 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
176 * whose contents is similar to that environment variable but can be
177 * LF separated. Its base points at a statically allocated buffer that
178 * contains "/the/directory/corresponding/to/.git/objects/...", while
179 * its name points just after the slash at the end of ".git/objects/"
180 * in the example above, and has enough space to hold 40-byte hex
181 * SHA1, an extra slash for the first level indirection, and the
184 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
186 const char *objdir = get_object_directory();
187 struct alternate_object_database *ent;
188 struct alternate_object_database *alt;
189 /* 43 = 40-byte + 2 '/' + terminating NUL */
191 int entlen = pfxlen + 43;
194 if (!is_absolute_path(entry) && relative_base) {
195 /* Relative alt-odb */
197 base_len = strlen(relative_base) + 1;
201 ent = xmalloc(sizeof(*ent) + entlen);
203 if (!is_absolute_path(entry) && relative_base) {
204 memcpy(ent->base, relative_base, base_len - 1);
205 ent->base[base_len - 1] = '/';
206 memcpy(ent->base + base_len, entry, len);
209 memcpy(ent->base, entry, pfxlen);
211 ent->name = ent->base + pfxlen + 1;
212 ent->base[pfxlen + 3] = '/';
213 ent->base[pfxlen] = ent->base[entlen-1] = 0;
215 /* Detect cases where alternate disappeared */
216 if (!is_directory(ent->base)) {
217 error("object directory %s does not exist; "
218 "check .git/objects/info/alternates.",
224 /* Prevent the common mistake of listing the same
225 * thing twice, or object directory itself.
227 for (alt = alt_odb_list; alt; alt = alt->next) {
228 if (!memcmp(ent->base, alt->base, pfxlen)) {
233 if (!memcmp(ent->base, objdir, pfxlen)) {
238 /* add the alternate entry */
240 alt_odb_tail = &(ent->next);
243 /* recursively add alternates */
244 read_info_alternates(ent->base, depth + 1);
246 ent->base[pfxlen] = '/';
251 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
252 const char *relative_base, int depth)
254 const char *cp, *last;
257 error("%s: ignoring alternate object stores, nesting too deep.",
265 if (cp < ep && *cp == '#') {
266 while (cp < ep && *cp != sep)
271 while (cp < ep && *cp != sep)
274 if (!is_absolute_path(last) && depth) {
275 error("%s: ignoring relative alternate object store %s",
276 relative_base, last);
278 link_alt_odb_entry(last, cp - last,
279 relative_base, depth);
282 while (cp < ep && *cp == sep)
288 static void read_info_alternates(const char * relative_base, int depth)
293 const char alt_file_name[] = "info/alternates";
294 /* Given that relative_base is no longer than PATH_MAX,
295 ensure that "path" has enough space to append "/", the
296 file name, "info/alternates", and a trailing NUL. */
297 char path[PATH_MAX + 1 + sizeof alt_file_name];
300 sprintf(path, "%s/%s", relative_base, alt_file_name);
301 fd = open(path, O_RDONLY);
304 if (fstat(fd, &st) || (st.st_size == 0)) {
308 mapsz = xsize_t(st.st_size);
309 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
312 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
317 void add_to_alternates_file(const char *reference)
319 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
320 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
321 char *alt = mkpath("%s/objects\n", reference);
322 write_or_die(fd, alt, strlen(alt));
323 if (commit_lock_file(lock))
324 die("could not close alternates file");
326 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
329 void foreach_alt_odb(alt_odb_fn fn, void *cb)
331 struct alternate_object_database *ent;
334 for (ent = alt_odb_list; ent; ent = ent->next)
339 void prepare_alt_odb(void)
346 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
349 alt_odb_tail = &alt_odb_list;
350 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
352 read_info_alternates(get_object_directory(), 0);
355 static int has_loose_object_local(const unsigned char *sha1)
357 char *name = sha1_file_name(sha1);
358 return !access(name, F_OK);
361 int has_loose_object_nonlocal(const unsigned char *sha1)
363 struct alternate_object_database *alt;
365 for (alt = alt_odb_list; alt; alt = alt->next) {
366 fill_sha1_path(alt->name, sha1);
367 if (!access(alt->base, F_OK))
373 static int has_loose_object(const unsigned char *sha1)
375 return has_loose_object_local(sha1) ||
376 has_loose_object_nonlocal(sha1);
379 static unsigned int pack_used_ctr;
380 static unsigned int pack_mmap_calls;
381 static unsigned int peak_pack_open_windows;
382 static unsigned int pack_open_windows;
383 static size_t peak_pack_mapped;
384 static size_t pack_mapped;
385 struct packed_git *packed_git;
387 void pack_report(void)
390 "pack_report: getpagesize() = %10" SZ_FMT "\n"
391 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
392 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
393 sz_fmt(getpagesize()),
394 sz_fmt(packed_git_window_size),
395 sz_fmt(packed_git_limit));
397 "pack_report: pack_used_ctr = %10u\n"
398 "pack_report: pack_mmap_calls = %10u\n"
399 "pack_report: pack_open_windows = %10u / %10u\n"
400 "pack_report: pack_mapped = "
401 "%10" SZ_FMT " / %10" SZ_FMT "\n",
404 pack_open_windows, peak_pack_open_windows,
405 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
408 static int check_packed_git_idx(const char *path, struct packed_git *p)
411 struct pack_idx_header *hdr;
413 uint32_t version, nr, i, *index;
414 int fd = open(path, O_RDONLY);
419 if (fstat(fd, &st)) {
423 idx_size = xsize_t(st.st_size);
424 if (idx_size < 4 * 256 + 20 + 20) {
426 return error("index file %s is too small", path);
428 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
432 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
433 version = ntohl(hdr->idx_version);
434 if (version < 2 || version > 2) {
435 munmap(idx_map, idx_size);
436 return error("index file %s is version %"PRIu32
437 " and is not supported by this binary"
438 " (try upgrading GIT to a newer version)",
447 index += 2; /* skip index header */
448 for (i = 0; i < 256; i++) {
449 uint32_t n = ntohl(index[i]);
451 munmap(idx_map, idx_size);
452 return error("non-monotonic index %s", path);
460 * - 256 index entries 4 bytes each
461 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
462 * - 20-byte SHA1 of the packfile
463 * - 20-byte SHA1 file checksum
465 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
466 munmap(idx_map, idx_size);
467 return error("wrong index v1 file size in %s", path);
469 } else if (version == 2) {
472 * - 8 bytes of header
473 * - 256 index entries 4 bytes each
474 * - 20-byte sha1 entry * nr
475 * - 4-byte crc entry * nr
476 * - 4-byte offset entry * nr
477 * - 20-byte SHA1 of the packfile
478 * - 20-byte SHA1 file checksum
479 * And after the 4-byte offset table might be a
480 * variable sized table containing 8-byte entries
481 * for offsets larger than 2^31.
483 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
484 unsigned long max_size = min_size;
486 max_size += (nr - 1)*8;
487 if (idx_size < min_size || idx_size > max_size) {
488 munmap(idx_map, idx_size);
489 return error("wrong index v2 file size in %s", path);
491 if (idx_size != min_size &&
493 * make sure we can deal with large pack offsets.
494 * 31-bit signed offset won't be enough, neither
495 * 32-bit unsigned one will be.
497 (sizeof(off_t) <= 4)) {
498 munmap(idx_map, idx_size);
499 return error("pack too large for current definition of off_t in %s", path);
503 p->index_version = version;
504 p->index_data = idx_map;
505 p->index_size = idx_size;
510 int open_pack_index(struct packed_git *p)
518 idx_name = xstrdup(p->pack_name);
519 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
520 ret = check_packed_git_idx(idx_name, p);
525 static void scan_windows(struct packed_git *p,
526 struct packed_git **lru_p,
527 struct pack_window **lru_w,
528 struct pack_window **lru_l)
530 struct pack_window *w, *w_l;
532 for (w_l = NULL, w = p->windows; w; w = w->next) {
534 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
544 static int unuse_one_window(struct packed_git *current, int keep_fd)
546 struct packed_git *p, *lru_p = NULL;
547 struct pack_window *lru_w = NULL, *lru_l = NULL;
550 scan_windows(current, &lru_p, &lru_w, &lru_l);
551 for (p = packed_git; p; p = p->next)
552 scan_windows(p, &lru_p, &lru_w, &lru_l);
554 munmap(lru_w->base, lru_w->len);
555 pack_mapped -= lru_w->len;
557 lru_l->next = lru_w->next;
559 lru_p->windows = lru_w->next;
560 if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
561 close(lru_p->pack_fd);
572 void release_pack_memory(size_t need, int fd)
574 size_t cur = pack_mapped;
575 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
579 void close_pack_windows(struct packed_git *p)
582 struct pack_window *w = p->windows;
585 die("pack '%s' still has open windows to it",
587 munmap(w->base, w->len);
588 pack_mapped -= w->len;
590 p->windows = w->next;
595 void unuse_pack(struct pack_window **w_cursor)
597 struct pack_window *w = *w_cursor;
605 * This is used by git-repack in case a newly created pack happens to
606 * contain the same set of objects as an existing one. In that case
607 * the resulting file might be different even if its name would be the
608 * same. It is best to close any reference to the old pack before it is
609 * replaced on disk. Of course no index pointers nor windows for given pack
610 * must subsist at this point. If ever objects from this pack are requested
611 * again, the new version of the pack will be reinitialized through
612 * reprepare_packed_git().
614 void free_pack_by_name(const char *pack_name)
616 struct packed_git *p, **pp = &packed_git;
620 if (strcmp(pack_name, p->pack_name) == 0) {
621 clear_delta_base_cache();
622 close_pack_windows(p);
623 if (p->pack_fd != -1)
626 munmap((void *)p->index_data, p->index_size);
627 free(p->bad_object_sha1);
637 * Do not call this directly as this leaks p->pack_fd on error return;
638 * call open_packed_git() instead.
640 static int open_packed_git_1(struct packed_git *p)
643 struct pack_header hdr;
644 unsigned char sha1[20];
645 unsigned char *idx_sha1;
648 if (!p->index_data && open_pack_index(p))
649 return error("packfile %s index unavailable", p->pack_name);
651 p->pack_fd = open(p->pack_name, O_RDONLY);
652 while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
653 p->pack_fd = open(p->pack_name, O_RDONLY);
654 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
657 /* If we created the struct before we had the pack we lack size. */
659 if (!S_ISREG(st.st_mode))
660 return error("packfile %s not a regular file", p->pack_name);
661 p->pack_size = st.st_size;
662 } else if (p->pack_size != st.st_size)
663 return error("packfile %s size changed", p->pack_name);
665 /* We leave these file descriptors open with sliding mmap;
666 * there is no point keeping them open across exec(), though.
668 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
670 return error("cannot determine file descriptor flags");
671 fd_flag |= FD_CLOEXEC;
672 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
673 return error("cannot set FD_CLOEXEC");
675 /* Verify we recognize this pack file format. */
676 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
677 return error("file %s is far too short to be a packfile", p->pack_name);
678 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
679 return error("file %s is not a GIT packfile", p->pack_name);
680 if (!pack_version_ok(hdr.hdr_version))
681 return error("packfile %s is version %"PRIu32" and not"
682 " supported (try upgrading GIT to a newer version)",
683 p->pack_name, ntohl(hdr.hdr_version));
685 /* Verify the pack matches its index. */
686 if (p->num_objects != ntohl(hdr.hdr_entries))
687 return error("packfile %s claims to have %"PRIu32" objects"
688 " while index indicates %"PRIu32" objects",
689 p->pack_name, ntohl(hdr.hdr_entries),
691 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
692 return error("end of packfile %s is unavailable", p->pack_name);
693 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
694 return error("packfile %s signature is unavailable", p->pack_name);
695 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
696 if (hashcmp(sha1, idx_sha1))
697 return error("packfile %s does not match index", p->pack_name);
701 static int open_packed_git(struct packed_git *p)
703 if (!open_packed_git_1(p))
705 if (p->pack_fd != -1) {
712 static int in_window(struct pack_window *win, off_t offset)
714 /* We must promise at least 20 bytes (one hash) after the
715 * offset is available from this window, otherwise the offset
716 * is not actually in this window and a different window (which
717 * has that one hash excess) must be used. This is to support
718 * the object header and delta base parsing routines below.
720 off_t win_off = win->offset;
721 return win_off <= offset
722 && (offset + 20) <= (win_off + win->len);
725 unsigned char *use_pack(struct packed_git *p,
726 struct pack_window **w_cursor,
730 struct pack_window *win = *w_cursor;
732 if (p->pack_fd == -1 && open_packed_git(p))
733 die("packfile %s cannot be accessed", p->pack_name);
735 /* Since packfiles end in a hash of their content and it's
736 * pointless to ask for an offset into the middle of that
737 * hash, and the in_window function above wouldn't match
738 * don't allow an offset too close to the end of the file.
740 if (offset > (p->pack_size - 20))
741 die("offset beyond end of packfile (truncated pack?)");
743 if (!win || !in_window(win, offset)) {
746 for (win = p->windows; win; win = win->next) {
747 if (in_window(win, offset))
751 size_t window_align = packed_git_window_size / 2;
753 win = xcalloc(1, sizeof(*win));
754 win->offset = (offset / window_align) * window_align;
755 len = p->pack_size - win->offset;
756 if (len > packed_git_window_size)
757 len = packed_git_window_size;
758 win->len = (size_t)len;
759 pack_mapped += win->len;
760 while (packed_git_limit < pack_mapped
761 && unuse_one_window(p, p->pack_fd))
763 win->base = xmmap(NULL, win->len,
764 PROT_READ, MAP_PRIVATE,
765 p->pack_fd, win->offset);
766 if (win->base == MAP_FAILED)
767 die("packfile %s cannot be mapped: %s",
772 if (pack_mapped > peak_pack_mapped)
773 peak_pack_mapped = pack_mapped;
774 if (pack_open_windows > peak_pack_open_windows)
775 peak_pack_open_windows = pack_open_windows;
776 win->next = p->windows;
780 if (win != *w_cursor) {
781 win->last_used = pack_used_ctr++;
785 offset -= win->offset;
787 *left = win->len - xsize_t(offset);
788 return win->base + offset;
791 static struct packed_git *alloc_packed_git(int extra)
793 struct packed_git *p = xmalloc(sizeof(*p) + extra);
794 memset(p, 0, sizeof(*p));
799 struct packed_git *add_packed_git(const char *path, int path_len, int local)
802 struct packed_git *p = alloc_packed_git(path_len + 2);
805 * Make sure a corresponding .pack file exists and that
806 * the index looks sane.
808 path_len -= strlen(".idx");
813 memcpy(p->pack_name, path, path_len);
815 strcpy(p->pack_name + path_len, ".keep");
816 if (!access(p->pack_name, F_OK))
819 strcpy(p->pack_name + path_len, ".pack");
820 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
825 /* ok, it looks sane as far as we can check without
826 * actually mapping the pack file.
828 p->pack_size = st.st_size;
829 p->pack_local = local;
830 p->mtime = st.st_mtime;
831 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
836 struct packed_git *parse_pack_index(unsigned char *sha1)
838 const char *idx_path = sha1_pack_index_name(sha1);
839 const char *path = sha1_pack_name(sha1);
840 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
842 strcpy(p->pack_name, path);
843 hashcpy(p->sha1, sha1);
844 if (check_packed_git_idx(idx_path, p)) {
852 void install_packed_git(struct packed_git *pack)
854 pack->next = packed_git;
858 static void prepare_packed_git_one(char *objdir, int local)
860 /* Ensure that this buffer is large enough so that we can
861 append "/pack/" without clobbering the stack even if
862 strlen(objdir) were PATH_MAX. */
863 char path[PATH_MAX + 1 + 4 + 1 + 1];
868 sprintf(path, "%s/pack", objdir);
871 while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
875 error("unable to open object pack directory: %s: %s",
876 path, strerror(errno));
880 while ((de = readdir(dir)) != NULL) {
881 int namelen = strlen(de->d_name);
882 struct packed_git *p;
884 if (!has_extension(de->d_name, ".idx"))
887 if (len + namelen + 1 > sizeof(path))
890 /* Don't reopen a pack we already have. */
891 strcpy(path + len, de->d_name);
892 for (p = packed_git; p; p = p->next) {
893 if (!memcmp(path, p->pack_name, len + namelen - 4))
898 /* See if it really is a valid .idx file with corresponding
899 * .pack file that we can map.
901 p = add_packed_git(path, len + namelen, local);
904 install_packed_git(p);
909 static int sort_pack(const void *a_, const void *b_)
911 struct packed_git *a = *((struct packed_git **)a_);
912 struct packed_git *b = *((struct packed_git **)b_);
916 * Local packs tend to contain objects specific to our
917 * variant of the project than remote ones. In addition,
918 * remote ones could be on a network mounted filesystem.
919 * Favor local ones for these reasons.
921 st = a->pack_local - b->pack_local;
926 * Younger packs tend to contain more recent objects,
927 * and more recent objects tend to get accessed more
930 if (a->mtime < b->mtime)
932 else if (a->mtime == b->mtime)
937 static void rearrange_packed_git(void)
939 struct packed_git **ary, *p;
942 for (n = 0, p = packed_git; p; p = p->next)
947 /* prepare an array of packed_git for easier sorting */
948 ary = xcalloc(n, sizeof(struct packed_git *));
949 for (n = 0, p = packed_git; p; p = p->next)
952 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
954 /* link them back again */
955 for (i = 0; i < n - 1; i++)
956 ary[i]->next = ary[i + 1];
957 ary[n - 1]->next = NULL;
963 static int prepare_packed_git_run_once = 0;
964 void prepare_packed_git(void)
966 struct alternate_object_database *alt;
968 if (prepare_packed_git_run_once)
970 prepare_packed_git_one(get_object_directory(), 1);
972 for (alt = alt_odb_list; alt; alt = alt->next) {
974 prepare_packed_git_one(alt->base, 0);
977 rearrange_packed_git();
978 prepare_packed_git_run_once = 1;
981 void reprepare_packed_git(void)
984 prepare_packed_git_run_once = 0;
985 prepare_packed_git();
988 static void mark_bad_packed_object(struct packed_git *p,
989 const unsigned char *sha1)
992 for (i = 0; i < p->num_bad_objects; i++)
993 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
995 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
996 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
997 p->num_bad_objects++;
1000 static int has_packed_and_bad(const unsigned char *sha1)
1002 struct packed_git *p;
1005 for (p = packed_git; p; p = p->next)
1006 for (i = 0; i < p->num_bad_objects; i++)
1007 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1012 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1014 unsigned char real_sha1[20];
1015 hash_sha1_file(map, size, type, real_sha1);
1016 return hashcmp(sha1, real_sha1) ? -1 : 0;
1019 static int git_open_noatime(const char *name)
1021 static int sha1_file_open_flag = O_NOATIME;
1022 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1024 /* Might the failure be due to O_NOATIME? */
1025 if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1026 fd = open(name, O_RDONLY);
1028 sha1_file_open_flag = 0;
1033 static int open_sha1_file(const unsigned char *sha1)
1036 char *name = sha1_file_name(sha1);
1037 struct alternate_object_database *alt;
1039 fd = git_open_noatime(name);
1045 for (alt = alt_odb_list; alt; alt = alt->next) {
1047 fill_sha1_path(name, sha1);
1048 fd = git_open_noatime(alt->base);
1055 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1060 fd = open_sha1_file(sha1);
1065 if (!fstat(fd, &st)) {
1066 *size = xsize_t(st.st_size);
1067 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1074 static int legacy_loose_object(unsigned char *map)
1079 * Is it a zlib-compressed buffer? If so, the first byte
1080 * must be 0x78 (15-bit window size, deflated), and the
1081 * first 16-bit word is evenly divisible by 31
1083 word = (map[0] << 8) + map[1];
1084 if (map[0] == 0x78 && !(word % 31))
1090 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1091 unsigned long len, enum object_type *type, unsigned long *sizep)
1094 unsigned long size, c;
1095 unsigned long used = 0;
1098 *type = (c >> 4) & 7;
1102 if (len <= used || bitsizeof(long) <= shift) {
1103 error("bad object header");
1107 size += (c & 0x7f) << shift;
1114 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1116 unsigned long size, used;
1117 static const char valid_loose_object_type[8] = {
1119 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1120 0, /* "delta" and others are invalid in a loose object */
1122 enum object_type type;
1124 /* Get the data stream */
1125 memset(stream, 0, sizeof(*stream));
1126 stream->next_in = map;
1127 stream->avail_in = mapsize;
1128 stream->next_out = buffer;
1129 stream->avail_out = bufsiz;
1131 if (legacy_loose_object(map)) {
1132 git_inflate_init(stream);
1133 return git_inflate(stream, 0);
1138 * There used to be a second loose object header format which
1139 * was meant to mimic the in-pack format, allowing for direct
1140 * copy of the object data. This format turned up not to be
1141 * really worth it and we don't write it any longer. But we
1142 * can still read it.
1144 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1145 if (!used || !valid_loose_object_type[type])
1150 /* Set up the stream for the rest.. */
1151 stream->next_in = map;
1152 stream->avail_in = mapsize;
1153 git_inflate_init(stream);
1155 /* And generate the fake traditional header */
1156 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1157 typename(type), size);
1161 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1163 int bytes = strlen(buffer) + 1;
1164 unsigned char *buf = xmallocz(size);
1168 n = stream->total_out - bytes;
1171 memcpy(buf, (char *) buffer + bytes, n);
1173 if (bytes <= size) {
1175 * The above condition must be (bytes <= size), not
1176 * (bytes < size). In other words, even though we
1177 * expect no more output and set avail_out to zer0,
1178 * the input zlib stream may have bytes that express
1179 * "this concludes the stream", and we *do* want to
1182 * Otherwise we would not be able to test that we
1183 * consumed all the input to reach the expected size;
1184 * we also want to check that zlib tells us that all
1185 * went well with status == Z_STREAM_END at the end.
1187 stream->next_out = buf + bytes;
1188 stream->avail_out = size - bytes;
1189 while (status == Z_OK)
1190 status = git_inflate(stream, Z_FINISH);
1192 if (status == Z_STREAM_END && !stream->avail_in) {
1193 git_inflate_end(stream);
1198 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1199 else if (stream->avail_in)
1200 error("garbage at end of loose object '%s'",
1207 * We used to just use "sscanf()", but that's actually way
1208 * too permissive for what we want to check. So do an anal
1209 * object header parse by hand.
1211 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1218 * The type can be at most ten bytes (including the
1219 * terminating '\0' that we add), and is followed by
1228 if (i >= sizeof(type))
1234 * The length must follow immediately, and be in canonical
1235 * decimal format (ie "010" is not valid).
1237 size = *hdr++ - '0';
1242 unsigned long c = *hdr - '0';
1246 size = size * 10 + c;
1252 * The length must be followed by a zero byte
1254 return *hdr ? -1 : type_from_string(type);
1257 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1263 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1264 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1267 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1270 unsigned long get_size_from_delta(struct packed_git *p,
1271 struct pack_window **w_curs,
1274 const unsigned char *data;
1275 unsigned char delta_head[20], *in;
1279 memset(&stream, 0, sizeof(stream));
1280 stream.next_out = delta_head;
1281 stream.avail_out = sizeof(delta_head);
1283 git_inflate_init(&stream);
1285 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1286 stream.next_in = in;
1287 st = git_inflate(&stream, Z_FINISH);
1288 curpos += stream.next_in - in;
1289 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1290 stream.total_out < sizeof(delta_head));
1291 git_inflate_end(&stream);
1292 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1293 error("delta data unpack-initial failed");
1297 /* Examine the initial part of the delta to figure out
1302 /* ignore base size */
1303 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1305 /* Read the result size */
1306 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1309 static off_t get_delta_base(struct packed_git *p,
1310 struct pack_window **w_curs,
1312 enum object_type type,
1313 off_t delta_obj_offset)
1315 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1318 /* use_pack() assured us we have [base_info, base_info + 20)
1319 * as a range that we can look at without walking off the
1320 * end of the mapped window. Its actually the hash size
1321 * that is assured. An OFS_DELTA longer than the hash size
1322 * is stupid, as then a REF_DELTA would be smaller to store.
1324 if (type == OBJ_OFS_DELTA) {
1326 unsigned char c = base_info[used++];
1327 base_offset = c & 127;
1330 if (!base_offset || MSB(base_offset, 7))
1331 return 0; /* overflow */
1332 c = base_info[used++];
1333 base_offset = (base_offset << 7) + (c & 127);
1335 base_offset = delta_obj_offset - base_offset;
1336 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1337 return 0; /* out of bound */
1339 } else if (type == OBJ_REF_DELTA) {
1340 /* The base entry _must_ be in the same pack */
1341 base_offset = find_pack_entry_one(base_info, p);
1344 die("I am totally screwed");
1348 /* forward declaration for a mutually recursive function */
1349 static int packed_object_info(struct packed_git *p, off_t offset,
1350 unsigned long *sizep);
1352 static int packed_delta_info(struct packed_git *p,
1353 struct pack_window **w_curs,
1355 enum object_type type,
1357 unsigned long *sizep)
1361 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1364 type = packed_object_info(p, base_offset, NULL);
1365 if (type <= OBJ_NONE) {
1366 struct revindex_entry *revidx;
1367 const unsigned char *base_sha1;
1368 revidx = find_pack_revindex(p, base_offset);
1371 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1372 mark_bad_packed_object(p, base_sha1);
1373 type = sha1_object_info(base_sha1, NULL);
1374 if (type <= OBJ_NONE)
1378 /* We choose to only get the type of the base object and
1379 * ignore potentially corrupt pack file that expects the delta
1380 * based on a base with a wrong size. This saves tons of
1384 *sizep = get_size_from_delta(p, w_curs, curpos);
1392 static int unpack_object_header(struct packed_git *p,
1393 struct pack_window **w_curs,
1395 unsigned long *sizep)
1397 unsigned char *base;
1400 enum object_type type;
1402 /* use_pack() assures us we have [base, base + 20) available
1403 * as a range that we can look at at. (Its actually the hash
1404 * size that is assured.) With our object header encoding
1405 * the maximum deflated object size is 2^137, which is just
1406 * insane, so we know won't exceed what we have been given.
1408 base = use_pack(p, w_curs, *curpos, &left);
1409 used = unpack_object_header_buffer(base, left, &type, sizep);
1418 const char *packed_object_info_detail(struct packed_git *p,
1420 unsigned long *size,
1421 unsigned long *store_size,
1422 unsigned int *delta_chain_length,
1423 unsigned char *base_sha1)
1425 struct pack_window *w_curs = NULL;
1427 unsigned long dummy;
1428 unsigned char *next_sha1;
1429 enum object_type type;
1430 struct revindex_entry *revidx;
1432 *delta_chain_length = 0;
1433 curpos = obj_offset;
1434 type = unpack_object_header(p, &w_curs, &curpos, size);
1436 revidx = find_pack_revindex(p, obj_offset);
1437 *store_size = revidx[1].offset - obj_offset;
1442 die("pack %s contains unknown object type %d",
1443 p->pack_name, type);
1448 unuse_pack(&w_curs);
1449 return typename(type);
1451 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1453 die("pack %s contains bad delta base reference of type %s",
1454 p->pack_name, typename(type));
1455 if (*delta_chain_length == 0) {
1456 revidx = find_pack_revindex(p, obj_offset);
1457 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1461 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1462 if (*delta_chain_length == 0)
1463 hashcpy(base_sha1, next_sha1);
1464 obj_offset = find_pack_entry_one(next_sha1, p);
1467 (*delta_chain_length)++;
1468 curpos = obj_offset;
1469 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1473 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1474 unsigned long *sizep)
1476 struct pack_window *w_curs = NULL;
1478 off_t curpos = obj_offset;
1479 enum object_type type;
1481 type = unpack_object_header(p, &w_curs, &curpos, &size);
1486 type = packed_delta_info(p, &w_curs, curpos,
1487 type, obj_offset, sizep);
1497 error("unknown object type %i at offset %"PRIuMAX" in %s",
1498 type, (uintmax_t)obj_offset, p->pack_name);
1501 unuse_pack(&w_curs);
1505 static void *unpack_compressed_entry(struct packed_git *p,
1506 struct pack_window **w_curs,
1512 unsigned char *buffer, *in;
1514 buffer = xmallocz(size);
1515 memset(&stream, 0, sizeof(stream));
1516 stream.next_out = buffer;
1517 stream.avail_out = size + 1;
1519 git_inflate_init(&stream);
1521 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1522 stream.next_in = in;
1523 st = git_inflate(&stream, Z_FINISH);
1524 if (!stream.avail_out)
1525 break; /* the payload is larger than it should be */
1526 curpos += stream.next_in - in;
1527 } while (st == Z_OK || st == Z_BUF_ERROR);
1528 git_inflate_end(&stream);
1529 if ((st != Z_STREAM_END) || stream.total_out != size) {
1537 #define MAX_DELTA_CACHE (256)
1539 static size_t delta_base_cached;
1541 static struct delta_base_cache_lru_list {
1542 struct delta_base_cache_lru_list *prev;
1543 struct delta_base_cache_lru_list *next;
1544 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1546 static struct delta_base_cache_entry {
1547 struct delta_base_cache_lru_list lru;
1549 struct packed_git *p;
1552 enum object_type type;
1553 } delta_base_cache[MAX_DELTA_CACHE];
1555 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1559 hash = (unsigned long)p + (unsigned long)base_offset;
1560 hash += (hash >> 8) + (hash >> 16);
1561 return hash % MAX_DELTA_CACHE;
1564 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1565 unsigned long *base_size, enum object_type *type, int keep_cache)
1568 unsigned long hash = pack_entry_hash(p, base_offset);
1569 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1572 if (!ret || ent->p != p || ent->base_offset != base_offset)
1573 return unpack_entry(p, base_offset, type, base_size);
1577 ent->lru.next->prev = ent->lru.prev;
1578 ent->lru.prev->next = ent->lru.next;
1579 delta_base_cached -= ent->size;
1581 ret = xmemdupz(ent->data, ent->size);
1584 *base_size = ent->size;
1588 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1593 ent->lru.next->prev = ent->lru.prev;
1594 ent->lru.prev->next = ent->lru.next;
1595 delta_base_cached -= ent->size;
1599 void clear_delta_base_cache(void)
1602 for (p = 0; p < MAX_DELTA_CACHE; p++)
1603 release_delta_base_cache(&delta_base_cache[p]);
1606 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1607 void *base, unsigned long base_size, enum object_type type)
1609 unsigned long hash = pack_entry_hash(p, base_offset);
1610 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1611 struct delta_base_cache_lru_list *lru;
1613 release_delta_base_cache(ent);
1614 delta_base_cached += base_size;
1616 for (lru = delta_base_cache_lru.next;
1617 delta_base_cached > delta_base_cache_limit
1618 && lru != &delta_base_cache_lru;
1620 struct delta_base_cache_entry *f = (void *)lru;
1621 if (f->type == OBJ_BLOB)
1622 release_delta_base_cache(f);
1624 for (lru = delta_base_cache_lru.next;
1625 delta_base_cached > delta_base_cache_limit
1626 && lru != &delta_base_cache_lru;
1628 struct delta_base_cache_entry *f = (void *)lru;
1629 release_delta_base_cache(f);
1633 ent->base_offset = base_offset;
1636 ent->size = base_size;
1637 ent->lru.next = &delta_base_cache_lru;
1638 ent->lru.prev = delta_base_cache_lru.prev;
1639 delta_base_cache_lru.prev->next = &ent->lru;
1640 delta_base_cache_lru.prev = &ent->lru;
1643 static void *read_object(const unsigned char *sha1, enum object_type *type,
1644 unsigned long *size);
1646 static void *unpack_delta_entry(struct packed_git *p,
1647 struct pack_window **w_curs,
1649 unsigned long delta_size,
1651 enum object_type *type,
1652 unsigned long *sizep)
1654 void *delta_data, *result, *base;
1655 unsigned long base_size;
1658 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1660 error("failed to validate delta base reference "
1661 "at offset %"PRIuMAX" from %s",
1662 (uintmax_t)curpos, p->pack_name);
1666 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1669 * We're probably in deep shit, but let's try to fetch
1670 * the required base anyway from another pack or loose.
1671 * This is costly but should happen only in the presence
1672 * of a corrupted pack, and is better than failing outright.
1674 struct revindex_entry *revidx;
1675 const unsigned char *base_sha1;
1676 revidx = find_pack_revindex(p, base_offset);
1679 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1680 error("failed to read delta base object %s"
1681 " at offset %"PRIuMAX" from %s",
1682 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1684 mark_bad_packed_object(p, base_sha1);
1685 base = read_object(base_sha1, type, &base_size);
1690 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1692 error("failed to unpack compressed delta "
1693 "at offset %"PRIuMAX" from %s",
1694 (uintmax_t)curpos, p->pack_name);
1698 result = patch_delta(base, base_size,
1699 delta_data, delta_size,
1702 die("failed to apply delta");
1704 add_delta_base_cache(p, base_offset, base, base_size, *type);
1708 int do_check_packed_object_crc;
1710 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1711 enum object_type *type, unsigned long *sizep)
1713 struct pack_window *w_curs = NULL;
1714 off_t curpos = obj_offset;
1717 if (do_check_packed_object_crc && p->index_version > 1) {
1718 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1719 unsigned long len = revidx[1].offset - obj_offset;
1720 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1721 const unsigned char *sha1 =
1722 nth_packed_object_sha1(p, revidx->nr);
1723 error("bad packed object CRC for %s",
1725 mark_bad_packed_object(p, sha1);
1726 unuse_pack(&w_curs);
1731 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1735 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1736 obj_offset, type, sizep);
1742 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1746 error("unknown object type %i at offset %"PRIuMAX" in %s",
1747 *type, (uintmax_t)obj_offset, p->pack_name);
1749 unuse_pack(&w_curs);
1753 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1756 const unsigned char *index = p->index_data;
1758 if (open_pack_index(p))
1760 index = p->index_data;
1762 if (n >= p->num_objects)
1765 if (p->index_version == 1) {
1766 return index + 24 * n + 4;
1769 return index + 20 * n;
1773 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1775 const unsigned char *index = p->index_data;
1777 if (p->index_version == 1) {
1778 return ntohl(*((uint32_t *)(index + 24 * n)));
1781 index += 8 + p->num_objects * (20 + 4);
1782 off = ntohl(*((uint32_t *)(index + 4 * n)));
1783 if (!(off & 0x80000000))
1785 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1786 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1787 ntohl(*((uint32_t *)(index + 4)));
1791 off_t find_pack_entry_one(const unsigned char *sha1,
1792 struct packed_git *p)
1794 const uint32_t *level1_ofs = p->index_data;
1795 const unsigned char *index = p->index_data;
1796 unsigned hi, lo, stride;
1797 static int use_lookup = -1;
1798 static int debug_lookup = -1;
1800 if (debug_lookup < 0)
1801 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1804 if (open_pack_index(p))
1806 level1_ofs = p->index_data;
1807 index = p->index_data;
1809 if (p->index_version > 1) {
1814 hi = ntohl(level1_ofs[*sha1]);
1815 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1816 if (p->index_version > 1) {
1824 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1825 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1828 use_lookup = !!getenv("GIT_USE_LOOKUP");
1830 int pos = sha1_entry_pos(index, stride, 0,
1831 lo, hi, p->num_objects, sha1);
1834 return nth_packed_object_offset(p, pos);
1838 unsigned mi = (lo + hi) / 2;
1839 int cmp = hashcmp(index + mi * stride, sha1);
1842 printf("lo %u hi %u rg %u mi %u\n",
1843 lo, hi, hi - lo, mi);
1845 return nth_packed_object_offset(p, mi);
1854 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1856 static struct packed_git *last_found = (void *)1;
1857 struct packed_git *p;
1860 prepare_packed_git();
1863 p = (last_found == (void *)1) ? packed_git : last_found;
1866 if (p->num_bad_objects) {
1868 for (i = 0; i < p->num_bad_objects; i++)
1869 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1873 offset = find_pack_entry_one(sha1, p);
1876 * We are about to tell the caller where they can
1877 * locate the requested object. We better make
1878 * sure the packfile is still here and can be
1879 * accessed before supplying that answer, as
1880 * it may have been deleted since the index
1883 if (p->pack_fd == -1 && open_packed_git(p)) {
1884 error("packfile %s cannot be accessed", p->pack_name);
1889 hashcpy(e->sha1, sha1);
1895 if (p == last_found)
1899 if (p == last_found)
1905 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1906 struct packed_git *packs)
1908 struct packed_git *p;
1910 for (p = packs; p; p = p->next) {
1911 if (find_pack_entry_one(sha1, p))
1918 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1921 unsigned long mapsize, size;
1926 map = map_sha1_file(sha1, &mapsize);
1928 return error("unable to find %s", sha1_to_hex(sha1));
1929 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1930 status = error("unable to unpack %s header",
1932 else if ((status = parse_sha1_header(hdr, &size)) < 0)
1933 status = error("unable to parse %s header", sha1_to_hex(sha1));
1936 git_inflate_end(&stream);
1937 munmap(map, mapsize);
1941 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1943 struct pack_entry e;
1946 if (!find_pack_entry(sha1, &e)) {
1947 /* Most likely it's a loose object. */
1948 status = sha1_loose_object_info(sha1, sizep);
1952 /* Not a loose object; someone else may have just packed it. */
1953 reprepare_packed_git();
1954 if (!find_pack_entry(sha1, &e))
1958 status = packed_object_info(e.p, e.offset, sizep);
1960 mark_bad_packed_object(e.p, sha1);
1961 status = sha1_object_info(sha1, sizep);
1967 static void *read_packed_sha1(const unsigned char *sha1,
1968 enum object_type *type, unsigned long *size)
1970 struct pack_entry e;
1973 if (!find_pack_entry(sha1, &e))
1975 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1978 * We're probably in deep shit, but let's try to fetch
1979 * the required object anyway from another pack or loose.
1980 * This should happen only in the presence of a corrupted
1981 * pack, and is better than failing outright.
1983 error("failed to read object %s at offset %"PRIuMAX" from %s",
1984 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1985 mark_bad_packed_object(e.p, sha1);
1986 data = read_object(sha1, type, size);
1992 * This is meant to hold a *small* number of objects that you would
1993 * want read_sha1_file() to be able to return, but yet you do not want
1994 * to write them into the object store (e.g. a browse-only
1997 static struct cached_object {
1998 unsigned char sha1[20];
1999 enum object_type type;
2003 static int cached_object_nr, cached_object_alloc;
2005 static struct cached_object empty_tree = {
2006 EMPTY_TREE_SHA1_BIN,
2012 static struct cached_object *find_cached_object(const unsigned char *sha1)
2015 struct cached_object *co = cached_objects;
2017 for (i = 0; i < cached_object_nr; i++, co++) {
2018 if (!hashcmp(co->sha1, sha1))
2021 if (!hashcmp(sha1, empty_tree.sha1))
2026 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2027 unsigned char *sha1)
2029 struct cached_object *co;
2031 hash_sha1_file(buf, len, typename(type), sha1);
2032 if (has_sha1_file(sha1) || find_cached_object(sha1))
2034 if (cached_object_alloc <= cached_object_nr) {
2035 cached_object_alloc = alloc_nr(cached_object_alloc);
2036 cached_objects = xrealloc(cached_objects,
2037 sizeof(*cached_objects) *
2038 cached_object_alloc);
2040 co = &cached_objects[cached_object_nr++];
2043 co->buf = xmalloc(len);
2044 memcpy(co->buf, buf, len);
2045 hashcpy(co->sha1, sha1);
2049 static void *read_object(const unsigned char *sha1, enum object_type *type,
2050 unsigned long *size)
2052 unsigned long mapsize;
2054 struct cached_object *co;
2056 co = find_cached_object(sha1);
2060 return xmemdupz(co->buf, co->size);
2063 buf = read_packed_sha1(sha1, type, size);
2066 map = map_sha1_file(sha1, &mapsize);
2068 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2069 munmap(map, mapsize);
2072 reprepare_packed_git();
2073 return read_packed_sha1(sha1, type, size);
2076 void *read_sha1_file_repl(const unsigned char *sha1,
2077 enum object_type *type,
2078 unsigned long *size,
2079 const unsigned char **replacement)
2081 const unsigned char *repl = lookup_replace_object(sha1);
2082 void *data = read_object(repl, type, size);
2084 /* die if we replaced an object with one that does not exist */
2085 if (!data && repl != sha1)
2086 die("replacement %s not found for %s",
2087 sha1_to_hex(repl), sha1_to_hex(sha1));
2089 /* legacy behavior is to die on corrupted objects */
2090 if (!data && (has_loose_object(repl) || has_packed_and_bad(repl)))
2091 die("object %s is corrupted", sha1_to_hex(repl));
2094 *replacement = repl;
2099 void *read_object_with_reference(const unsigned char *sha1,
2100 const char *required_type_name,
2101 unsigned long *size,
2102 unsigned char *actual_sha1_return)
2104 enum object_type type, required_type;
2106 unsigned long isize;
2107 unsigned char actual_sha1[20];
2109 required_type = type_from_string(required_type_name);
2110 hashcpy(actual_sha1, sha1);
2112 int ref_length = -1;
2113 const char *ref_type = NULL;
2115 buffer = read_sha1_file(actual_sha1, &type, &isize);
2118 if (type == required_type) {
2120 if (actual_sha1_return)
2121 hashcpy(actual_sha1_return, actual_sha1);
2124 /* Handle references */
2125 else if (type == OBJ_COMMIT)
2127 else if (type == OBJ_TAG)
2128 ref_type = "object ";
2133 ref_length = strlen(ref_type);
2135 if (ref_length + 40 > isize ||
2136 memcmp(buffer, ref_type, ref_length) ||
2137 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2142 /* Now we have the ID of the referred-to object in
2143 * actual_sha1. Check again. */
2147 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2148 const char *type, unsigned char *sha1,
2149 char *hdr, int *hdrlen)
2153 /* Generate the header */
2154 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2158 git_SHA1_Update(&c, hdr, *hdrlen);
2159 git_SHA1_Update(&c, buf, len);
2160 git_SHA1_Final(sha1, &c);
2164 * Move the just written object into its final resting place.
2165 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2166 * "moving" is only a part of what it does, when no patch between
2167 * master to pu changes the call sites of this function.
2169 int move_temp_to_file(const char *tmpfile, const char *filename)
2173 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2175 else if (link(tmpfile, filename))
2179 * Coda hack - coda doesn't like cross-directory links,
2180 * so we fall back to a rename, which will mean that it
2181 * won't be able to check collisions, but that's not a
2184 * The same holds for FAT formatted media.
2186 * When this succeeds, we just return. We have nothing
2189 if (ret && ret != EEXIST) {
2191 if (!rename(tmpfile, filename))
2195 unlink_or_warn(tmpfile);
2197 if (ret != EEXIST) {
2198 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2200 /* FIXME!!! Collision check here ? */
2204 if (adjust_shared_perm(filename))
2205 return error("unable to set permission to '%s'", filename);
2209 static int write_buffer(int fd, const void *buf, size_t len)
2211 if (write_in_full(fd, buf, len) < 0)
2212 return error("file write error (%s)", strerror(errno));
2216 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2217 unsigned char *sha1)
2221 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2225 /* Finalize a file on disk, and close it. */
2226 static void close_sha1_file(int fd)
2228 if (fsync_object_files)
2229 fsync_or_die(fd, "sha1 file");
2231 die_errno("error when closing sha1 file");
2234 /* Size of directory component, including the ending '/' */
2235 static inline int directory_size(const char *filename)
2237 const char *s = strrchr(filename, '/');
2240 return s - filename + 1;
2244 * This creates a temporary file in the same directory as the final
2247 * We want to avoid cross-directory filename renames, because those
2248 * can have problems on various filesystems (FAT, NFS, Coda).
2250 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2252 int fd, dirlen = directory_size(filename);
2254 if (dirlen + 20 > bufsiz) {
2255 errno = ENAMETOOLONG;
2258 memcpy(buffer, filename, dirlen);
2259 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2260 fd = git_mkstemp_mode(buffer, 0444);
2261 if (fd < 0 && dirlen && errno == ENOENT) {
2262 /* Make sure the directory exists */
2263 memcpy(buffer, filename, dirlen);
2264 buffer[dirlen-1] = 0;
2265 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2269 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2270 fd = git_mkstemp_mode(buffer, 0444);
2275 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2276 const void *buf, unsigned long len, time_t mtime)
2279 unsigned char compressed[4096];
2282 unsigned char parano_sha1[20];
2284 static char tmpfile[PATH_MAX];
2286 filename = sha1_file_name(sha1);
2287 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2288 while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2289 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2291 if (errno == EACCES)
2292 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2294 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2298 memset(&stream, 0, sizeof(stream));
2299 deflateInit(&stream, zlib_compression_level);
2300 stream.next_out = compressed;
2301 stream.avail_out = sizeof(compressed);
2304 /* First header.. */
2305 stream.next_in = (unsigned char *)hdr;
2306 stream.avail_in = hdrlen;
2307 while (deflate(&stream, 0) == Z_OK)
2309 git_SHA1_Update(&c, hdr, hdrlen);
2311 /* Then the data itself.. */
2312 stream.next_in = (void *)buf;
2313 stream.avail_in = len;
2315 unsigned char *in0 = stream.next_in;
2316 ret = deflate(&stream, Z_FINISH);
2317 git_SHA1_Update(&c, in0, stream.next_in - in0);
2318 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2319 die("unable to write sha1 file");
2320 stream.next_out = compressed;
2321 stream.avail_out = sizeof(compressed);
2322 } while (ret == Z_OK);
2324 if (ret != Z_STREAM_END)
2325 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2326 ret = deflateEnd(&stream);
2328 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2329 git_SHA1_Final(parano_sha1, &c);
2330 if (hashcmp(sha1, parano_sha1) != 0)
2331 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2333 close_sha1_file(fd);
2338 utb.modtime = mtime;
2339 if (utime(tmpfile, &utb) < 0)
2340 warning("failed utime() on %s: %s",
2341 tmpfile, strerror(errno));
2344 return move_temp_to_file(tmpfile, filename);
2347 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2349 unsigned char sha1[20];
2353 /* Normally if we have it in the pack then we do not bother writing
2354 * it out into .git/objects/??/?{38} file.
2356 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2358 hashcpy(returnsha1, sha1);
2359 if (has_sha1_file(sha1))
2361 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2364 int force_object_loose(const unsigned char *sha1, time_t mtime)
2368 enum object_type type;
2373 if (has_loose_object(sha1))
2375 buf = read_packed_sha1(sha1, &type, &len);
2377 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2378 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2379 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2385 int has_pack_index(const unsigned char *sha1)
2388 if (stat(sha1_pack_index_name(sha1), &st))
2393 int has_sha1_pack(const unsigned char *sha1)
2395 struct pack_entry e;
2396 return find_pack_entry(sha1, &e);
2399 int has_sha1_file(const unsigned char *sha1)
2401 struct pack_entry e;
2403 if (find_pack_entry(sha1, &e))
2405 return has_loose_object(sha1);
2408 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2409 int write_object, enum object_type type, const char *path)
2411 int ret, re_allocated = 0;
2417 * Convert blobs to git internal format
2419 if ((type == OBJ_BLOB) && path) {
2420 struct strbuf nbuf = STRBUF_INIT;
2421 if (convert_to_git(path, buf, size, &nbuf,
2422 write_object ? safe_crlf : 0)) {
2423 buf = strbuf_detach(&nbuf, &size);
2429 ret = write_sha1_file(buf, size, typename(type), sha1);
2431 ret = hash_sha1_file(buf, size, typename(type), sha1);
2437 #define SMALL_FILE_SIZE (32*1024)
2439 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2440 enum object_type type, const char *path)
2443 size_t size = xsize_t(st->st_size);
2445 if (!S_ISREG(st->st_mode)) {
2446 struct strbuf sbuf = STRBUF_INIT;
2447 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2448 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2452 strbuf_release(&sbuf);
2454 ret = index_mem(sha1, NULL, size, write_object, type, path);
2455 } else if (size <= SMALL_FILE_SIZE) {
2456 char *buf = xmalloc(size);
2457 if (size == read_in_full(fd, buf, size))
2458 ret = index_mem(sha1, buf, size, write_object, type,
2461 ret = error("short read %s", strerror(errno));
2464 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2465 ret = index_mem(sha1, buf, size, write_object, type, path);
2472 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2475 struct strbuf sb = STRBUF_INIT;
2477 switch (st->st_mode & S_IFMT) {
2479 fd = open(path, O_RDONLY);
2481 return error("open(\"%s\"): %s", path,
2483 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2484 return error("%s: failed to insert into database",
2488 if (strbuf_readlink(&sb, path, st->st_size)) {
2489 char *errstr = strerror(errno);
2490 return error("readlink(\"%s\"): %s", path,
2494 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2495 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2496 return error("%s: failed to insert into database",
2498 strbuf_release(&sb);
2501 return resolve_gitlink_ref(path, "HEAD", sha1);
2503 return error("%s: unsupported file type", path);
2508 int read_pack_header(int fd, struct pack_header *header)
2510 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2511 /* "eof before pack header was fully read" */
2512 return PH_ERROR_EOF;
2514 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2515 /* "protocol error (pack signature mismatch detected)" */
2516 return PH_ERROR_PACK_SIGNATURE;
2517 if (!pack_version_ok(header->hdr_version))
2518 /* "protocol error (pack version unsupported)" */
2519 return PH_ERROR_PROTOCOL;