2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
31 const unsigned char null_sha1[20];
33 static unsigned int sha1_file_open_flag = O_NOATIME;
35 signed char hexval_table[256] = {
36 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
37 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
38 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
42 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
43 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
44 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
48 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
70 int get_sha1_hex(const char *hex, unsigned char *sha1)
73 for (i = 0; i < 20; i++) {
74 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
83 int safe_create_leading_directories(char *path)
92 pos = strchr(pos, '/');
96 if (!stat(path, &st)) {
98 if (!S_ISDIR(st.st_mode)) {
103 else if (mkdir(path, 0777)) {
107 else if (adjust_shared_perm(path)) {
116 char * sha1_to_hex(const unsigned char *sha1)
119 static char hexbuffer[4][50];
120 static const char hex[] = "0123456789abcdef";
121 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
124 for (i = 0; i < 20; i++) {
125 unsigned int val = *sha1++;
126 *buf++ = hex[val >> 4];
127 *buf++ = hex[val & 0xf];
134 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 for (i = 0; i < 20; i++) {
138 static char hex[] = "0123456789abcdef";
139 unsigned int val = sha1[i];
140 char *pos = pathbuf + i*2 + (i > 0);
141 *pos++ = hex[val >> 4];
142 *pos = hex[val & 0xf];
147 * NOTE! This returns a statically allocated buffer, so you have to be
148 * careful about using it. Do a "xstrdup()" if you need to save the
151 * Also note that this returns the location for creating. Reading
152 * SHA1 file can happen from any alternate directory listed in the
153 * DB_ENVIRONMENT environment variable if it is not found in
154 * the primary object database.
156 char *sha1_file_name(const unsigned char *sha1)
158 static char *name, *base;
161 const char *sha1_file_directory = get_object_directory();
162 int len = strlen(sha1_file_directory);
163 base = xmalloc(len + 60);
164 memcpy(base, sha1_file_directory, len);
165 memset(base+len, 0, 60);
168 name = base + len + 1;
170 fill_sha1_path(name, sha1);
174 char *sha1_pack_name(const unsigned char *sha1)
176 static const char hex[] = "0123456789abcdef";
177 static char *name, *base, *buf;
181 const char *sha1_file_directory = get_object_directory();
182 int len = strlen(sha1_file_directory);
183 base = xmalloc(len + 60);
184 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
185 name = base + len + 11;
190 for (i = 0; i < 20; i++) {
191 unsigned int val = *sha1++;
192 *buf++ = hex[val >> 4];
193 *buf++ = hex[val & 0xf];
199 char *sha1_pack_index_name(const unsigned char *sha1)
201 static const char hex[] = "0123456789abcdef";
202 static char *name, *base, *buf;
206 const char *sha1_file_directory = get_object_directory();
207 int len = strlen(sha1_file_directory);
208 base = xmalloc(len + 60);
209 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
210 name = base + len + 11;
215 for (i = 0; i < 20; i++) {
216 unsigned int val = *sha1++;
217 *buf++ = hex[val >> 4];
218 *buf++ = hex[val & 0xf];
224 struct alternate_object_database *alt_odb_list;
225 static struct alternate_object_database **alt_odb_tail;
227 static void read_info_alternates(const char * alternates, int depth);
230 * Prepare alternate object database registry.
232 * The variable alt_odb_list points at the list of struct
233 * alternate_object_database. The elements on this list come from
234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
236 * whose contents is similar to that environment variable but can be
237 * LF separated. Its base points at a statically allocated buffer that
238 * contains "/the/directory/corresponding/to/.git/objects/...", while
239 * its name points just after the slash at the end of ".git/objects/"
240 * in the example above, and has enough space to hold 40-byte hex
241 * SHA1, an extra slash for the first level indirection, and the
244 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
247 const char *objdir = get_object_directory();
248 struct alternate_object_database *ent;
249 struct alternate_object_database *alt;
250 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int entlen = pfxlen + 43;
255 if (*entry != '/' && relative_base) {
256 /* Relative alt-odb */
258 base_len = strlen(relative_base) + 1;
262 ent = xmalloc(sizeof(*ent) + entlen);
264 if (*entry != '/' && relative_base) {
265 memcpy(ent->base, relative_base, base_len - 1);
266 ent->base[base_len - 1] = '/';
267 memcpy(ent->base + base_len, entry, len);
270 memcpy(ent->base, entry, pfxlen);
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen + 3] = '/';
274 ent->base[pfxlen] = ent->base[entlen-1] = 0;
276 /* Detect cases where alternate disappeared */
277 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
278 error("object directory %s does not exist; "
279 "check .git/objects/info/alternates.",
285 /* Prevent the common mistake of listing the same
286 * thing twice, or object directory itself.
288 for (alt = alt_odb_list; alt; alt = alt->next) {
289 if (!memcmp(ent->base, alt->base, pfxlen)) {
294 if (!memcmp(ent->base, objdir, pfxlen)) {
299 /* add the alternate entry */
301 alt_odb_tail = &(ent->next);
304 /* recursively add alternates */
305 read_info_alternates(ent->base, depth + 1);
307 ent->base[pfxlen] = '/';
312 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
313 const char *relative_base, int depth)
315 const char *cp, *last;
318 error("%s: ignoring alternate object stores, nesting too deep.",
326 if (cp < ep && *cp == '#') {
327 while (cp < ep && *cp != sep)
332 while (cp < ep && *cp != sep)
335 if ((*last != '/') && depth) {
336 error("%s: ignoring relative alternate object store %s",
337 relative_base, last);
339 link_alt_odb_entry(last, cp - last,
340 relative_base, depth);
343 while (cp < ep && *cp == sep)
349 static void read_info_alternates(const char * relative_base, int depth)
356 sprintf(path, "%s/info/alternates", relative_base);
357 fd = open(path, O_RDONLY);
360 if (fstat(fd, &st) || (st.st_size == 0)) {
364 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
367 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
369 munmap(map, st.st_size);
372 void prepare_alt_odb(void)
376 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
381 alt_odb_tail = &alt_odb_list;
382 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
384 read_info_alternates(get_object_directory(), 0);
387 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
389 char *name = sha1_file_name(sha1);
390 struct alternate_object_database *alt;
395 for (alt = alt_odb_list; alt; alt = alt->next) {
397 fill_sha1_path(name, sha1);
398 if (!stat(alt->base, st))
404 static unsigned int pack_used_ctr;
405 static unsigned int pack_mmap_calls;
406 static unsigned int peak_pack_open_windows;
407 static unsigned int pack_open_windows;
408 static size_t peak_pack_mapped;
409 static size_t pack_mapped;
410 static size_t page_size;
411 struct packed_git *packed_git;
416 "pack_report: getpagesize() = %10" SZ_FMT "\n"
417 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
418 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
420 packed_git_window_size,
423 "pack_report: pack_used_ctr = %10u\n"
424 "pack_report: pack_mmap_calls = %10u\n"
425 "pack_report: pack_open_windows = %10u / %10u\n"
426 "pack_report: pack_mapped = "
427 "%10" SZ_FMT " / %10" SZ_FMT "\n",
430 pack_open_windows, peak_pack_open_windows,
431 pack_mapped, peak_pack_mapped);
434 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
439 unsigned long idx_size;
441 int fd = open(path, O_RDONLY);
445 if (fstat(fd, &st)) {
449 idx_size = st.st_size;
450 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
455 *idx_size_ = idx_size;
457 /* check index map */
458 if (idx_size < 4*256 + 20 + 20)
459 return error("index file too small");
461 for (i = 0; i < 256; i++) {
462 unsigned int n = ntohl(index[i]);
464 return error("non-monotonic index");
470 * - 256 index entries 4 bytes each
471 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
472 * - 20-byte SHA1 of the packfile
473 * - 20-byte SHA1 file checksum
475 if (idx_size != 4*256 + nr * 24 + 20 + 20)
476 return error("wrong index file size");
481 static void scan_windows(struct packed_git *p,
482 struct packed_git **lru_p,
483 struct pack_window **lru_w,
484 struct pack_window **lru_l)
486 struct pack_window *w, *w_l;
488 for (w_l = NULL, w = p->windows; w; w = w->next) {
490 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
500 static int unuse_one_window(struct packed_git *current)
502 struct packed_git *p, *lru_p = NULL;
503 struct pack_window *lru_w = NULL, *lru_l = NULL;
506 scan_windows(current, &lru_p, &lru_w, &lru_l);
507 for (p = packed_git; p; p = p->next)
508 scan_windows(p, &lru_p, &lru_w, &lru_l);
510 munmap(lru_w->base, lru_w->len);
511 pack_mapped -= lru_w->len;
513 lru_l->next = lru_w->next;
515 lru_p->windows = lru_w->next;
516 if (!lru_p->windows && lru_p != current) {
517 close(lru_p->pack_fd);
528 void release_pack_memory(size_t need)
530 size_t cur = pack_mapped;
531 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
535 void unuse_pack(struct pack_window **w_cursor)
537 struct pack_window *w = *w_cursor;
544 static void open_packed_git(struct packed_git *p)
547 struct pack_header hdr;
548 unsigned char sha1[20];
549 unsigned char *idx_sha1;
552 p->pack_fd = open(p->pack_name, O_RDONLY);
553 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
554 die("packfile %s cannot be opened", p->pack_name);
556 /* If we created the struct before we had the pack we lack size. */
558 if (!S_ISREG(st.st_mode))
559 die("packfile %s not a regular file", p->pack_name);
560 p->pack_size = st.st_size;
561 } else if (p->pack_size != st.st_size)
562 die("packfile %s size changed", p->pack_name);
564 /* We leave these file descriptors open with sliding mmap;
565 * there is no point keeping them open across exec(), though.
567 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
569 die("cannot determine file descriptor flags");
570 fd_flag |= FD_CLOEXEC;
571 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
572 die("cannot set FD_CLOEXEC");
574 /* Verify we recognize this pack file format. */
575 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
576 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
577 die("file %s is not a GIT packfile", p->pack_name);
578 if (!pack_version_ok(hdr.hdr_version))
579 die("packfile %s is version %u and not supported"
580 " (try upgrading GIT to a newer version)",
581 p->pack_name, ntohl(hdr.hdr_version));
583 /* Verify the pack matches its index. */
584 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
585 die("packfile %s claims to have %u objects"
586 " while index size indicates %u objects",
587 p->pack_name, ntohl(hdr.hdr_entries),
588 num_packed_objects(p));
589 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
590 die("end of packfile %s is unavailable", p->pack_name);
591 read_or_die(p->pack_fd, sha1, sizeof(sha1));
592 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
593 if (hashcmp(sha1, idx_sha1))
594 die("packfile %s does not match index", p->pack_name);
597 static int in_window(struct pack_window *win, unsigned long offset)
599 /* We must promise at least 20 bytes (one hash) after the
600 * offset is available from this window, otherwise the offset
601 * is not actually in this window and a different window (which
602 * has that one hash excess) must be used. This is to support
603 * the object header and delta base parsing routines below.
605 off_t win_off = win->offset;
606 return win_off <= offset
607 && (offset + 20) <= (win_off + win->len);
610 unsigned char* use_pack(struct packed_git *p,
611 struct pack_window **w_cursor,
612 unsigned long offset,
615 struct pack_window *win = *w_cursor;
617 if (p->pack_fd == -1)
620 /* Since packfiles end in a hash of their content and its
621 * pointless to ask for an offset into the middle of that
622 * hash, and the in_window function above wouldn't match
623 * don't allow an offset too close to the end of the file.
625 if (offset > (p->pack_size - 20))
626 die("offset beyond end of packfile (truncated pack?)");
628 if (!win || !in_window(win, offset)) {
631 for (win = p->windows; win; win = win->next) {
632 if (in_window(win, offset))
637 page_size = getpagesize();
638 win = xcalloc(1, sizeof(*win));
639 win->offset = (offset / page_size) * page_size;
640 win->len = p->pack_size - win->offset;
641 if (win->len > packed_git_window_size)
642 win->len = packed_git_window_size;
643 pack_mapped += win->len;
644 while (packed_git_limit < pack_mapped
645 && unuse_one_window(p))
647 win->base = xmmap(NULL, win->len,
648 PROT_READ, MAP_PRIVATE,
649 p->pack_fd, win->offset);
650 if (win->base == MAP_FAILED)
651 die("packfile %s cannot be mapped: %s",
656 if (pack_mapped > peak_pack_mapped)
657 peak_pack_mapped = pack_mapped;
658 if (pack_open_windows > peak_pack_open_windows)
659 peak_pack_open_windows = pack_open_windows;
660 win->next = p->windows;
664 if (win != *w_cursor) {
665 win->last_used = pack_used_ctr++;
669 offset -= win->offset;
671 *left = win->len - offset;
672 return win->base + offset;
675 struct packed_git *add_packed_git(char *path, int path_len, int local)
678 struct packed_git *p;
679 unsigned long idx_size;
681 unsigned char sha1[20];
683 if (check_packed_git_idx(path, &idx_size, &idx_map))
686 /* do we have a corresponding .pack file? */
687 strcpy(path + path_len - 4, ".pack");
688 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
689 munmap(idx_map, idx_size);
692 /* ok, it looks sane as far as we can check without
693 * actually mapping the pack file.
695 p = xmalloc(sizeof(*p) + path_len + 2);
696 strcpy(p->pack_name, path);
697 p->index_size = idx_size;
698 p->pack_size = st.st_size;
699 p->index_base = idx_map;
703 p->pack_local = local;
704 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
705 hashcpy(p->sha1, sha1);
709 struct packed_git *parse_pack_index(unsigned char *sha1)
711 char *path = sha1_pack_index_name(sha1);
712 return parse_pack_index_file(sha1, path);
715 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
717 struct packed_git *p;
718 unsigned long idx_size;
722 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
725 path = sha1_pack_name(sha1);
727 p = xmalloc(sizeof(*p) + strlen(path) + 2);
728 strcpy(p->pack_name, path);
729 p->index_size = idx_size;
731 p->index_base = idx_map;
735 hashcpy(p->sha1, sha1);
739 void install_packed_git(struct packed_git *pack)
741 pack->next = packed_git;
745 static void prepare_packed_git_one(char *objdir, int local)
752 sprintf(path, "%s/pack", objdir);
757 error("unable to open object pack directory: %s: %s",
758 path, strerror(errno));
762 while ((de = readdir(dir)) != NULL) {
763 int namelen = strlen(de->d_name);
764 struct packed_git *p;
766 if (!has_extension(de->d_name, ".idx"))
769 /* we have .idx. Is it a file we can map? */
770 strcpy(path + len, de->d_name);
771 for (p = packed_git; p; p = p->next) {
772 if (!memcmp(path, p->pack_name, len + namelen - 4))
777 p = add_packed_git(path, len + namelen, local);
780 p->next = packed_git;
786 static int prepare_packed_git_run_once = 0;
787 void prepare_packed_git(void)
789 struct alternate_object_database *alt;
791 if (prepare_packed_git_run_once)
793 prepare_packed_git_one(get_object_directory(), 1);
795 for (alt = alt_odb_list; alt; alt = alt->next) {
797 prepare_packed_git_one(alt->base, 0);
800 prepare_packed_git_run_once = 1;
803 void reprepare_packed_git(void)
805 prepare_packed_git_run_once = 0;
806 prepare_packed_git();
809 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
811 unsigned char real_sha1[20];
812 hash_sha1_file(map, size, type, real_sha1);
813 return hashcmp(sha1, real_sha1) ? -1 : 0;
816 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
821 char *filename = find_sha1_file(sha1, &st);
827 fd = open(filename, O_RDONLY | sha1_file_open_flag);
829 /* See if it works without O_NOATIME */
830 switch (sha1_file_open_flag) {
832 fd = open(filename, O_RDONLY);
840 /* If it failed once, it will probably fail again.
841 * Stop using O_NOATIME
843 sha1_file_open_flag = 0;
845 map = xmmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
851 int legacy_loose_object(unsigned char *map)
856 * Is it a zlib-compressed buffer? If so, the first byte
857 * must be 0x78 (15-bit window size, deflated), and the
858 * first 16-bit word is evenly divisible by 31
860 word = (map[0] << 8) + map[1];
861 if (map[0] == 0x78 && !(word % 31))
867 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
872 unsigned long used = 0;
875 *type = (c >> 4) & 7;
881 if (sizeof(long) * 8 <= shift)
884 size += (c & 0x7f) << shift;
891 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
893 unsigned long size, used;
894 static const char valid_loose_object_type[8] = {
896 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
897 0, /* "delta" and others are invalid in a loose object */
899 enum object_type type;
901 /* Get the data stream */
902 memset(stream, 0, sizeof(*stream));
903 stream->next_in = map;
904 stream->avail_in = mapsize;
905 stream->next_out = buffer;
906 stream->avail_out = bufsiz;
908 if (legacy_loose_object(map)) {
910 return inflate(stream, 0);
913 used = unpack_object_header_gently(map, mapsize, &type, &size);
914 if (!used || !valid_loose_object_type[type])
919 /* Set up the stream for the rest.. */
920 stream->next_in = map;
921 stream->avail_in = mapsize;
924 /* And generate the fake traditional header */
925 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
926 type_names[type], size);
930 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
932 int bytes = strlen(buffer) + 1;
933 unsigned char *buf = xmalloc(1+size);
936 n = stream->total_out - bytes;
939 memcpy(buf, (char *) buffer + bytes, n);
942 stream->next_out = buf + bytes;
943 stream->avail_out = size - bytes;
944 while (inflate(stream, Z_FINISH) == Z_OK)
953 * We used to just use "sscanf()", but that's actually way
954 * too permissive for what we want to check. So do an anal
955 * object header parse by hand.
957 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
963 * The type can be at most ten bytes (including the
964 * terminating '\0' that we add), and is followed by
979 * The length must follow immediately, and be in canonical
980 * decimal format (ie "010" is not valid).
987 unsigned long c = *hdr - '0';
991 size = size * 10 + c;
997 * The length must be followed by a zero byte
999 return *hdr ? -1 : 0;
1002 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
1008 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1009 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
1012 return unpack_sha1_rest(&stream, hdr, *size);
1015 static unsigned long get_delta_base(struct packed_git *p,
1016 struct pack_window **w_curs,
1017 unsigned long offset,
1018 enum object_type kind,
1019 unsigned long delta_obj_offset,
1020 unsigned long *base_obj_offset)
1022 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
1023 unsigned long base_offset;
1025 /* use_pack() assured us we have [base_info, base_info + 20)
1026 * as a range that we can look at without walking off the
1027 * end of the mapped window. Its actually the hash size
1028 * that is assured. An OFS_DELTA longer than the hash size
1029 * is stupid, as then a REF_DELTA would be smaller to store.
1031 if (kind == OBJ_OFS_DELTA) {
1033 unsigned char c = base_info[used++];
1034 base_offset = c & 127;
1037 if (!base_offset || base_offset & ~(~0UL >> 7))
1038 die("offset value overflow for delta base object");
1039 c = base_info[used++];
1040 base_offset = (base_offset << 7) + (c & 127);
1042 base_offset = delta_obj_offset - base_offset;
1043 if (base_offset >= delta_obj_offset)
1044 die("delta base offset out of bound");
1046 } else if (kind == OBJ_REF_DELTA) {
1047 /* The base entry _must_ be in the same pack */
1048 base_offset = find_pack_entry_one(base_info, p);
1050 die("failed to find delta-pack base object %s",
1051 sha1_to_hex(base_info));
1054 die("I am totally screwed");
1055 *base_obj_offset = base_offset;
1059 /* forward declaration for a mutually recursive function */
1060 static int packed_object_info(struct packed_git *p, unsigned long offset,
1061 char *type, unsigned long *sizep);
1063 static int packed_delta_info(struct packed_git *p,
1064 struct pack_window **w_curs,
1065 unsigned long offset,
1066 enum object_type kind,
1067 unsigned long obj_offset,
1069 unsigned long *sizep)
1071 unsigned long base_offset;
1073 offset = get_delta_base(p, w_curs, offset, kind,
1074 obj_offset, &base_offset);
1076 /* We choose to only get the type of the base object and
1077 * ignore potentially corrupt pack file that expects the delta
1078 * based on a base with a wrong size. This saves tons of
1081 if (packed_object_info(p, base_offset, type, NULL))
1082 die("cannot get info for delta-pack base");
1085 const unsigned char *data;
1086 unsigned char delta_head[20], *in;
1087 unsigned long result_size;
1091 memset(&stream, 0, sizeof(stream));
1092 stream.next_out = delta_head;
1093 stream.avail_out = sizeof(delta_head);
1095 inflateInit(&stream);
1097 in = use_pack(p, w_curs, offset, &stream.avail_in);
1098 stream.next_in = in;
1099 st = inflate(&stream, Z_FINISH);
1100 offset += stream.next_in - in;
1101 } while ((st == Z_OK || st == Z_BUF_ERROR)
1102 && stream.total_out < sizeof(delta_head));
1103 inflateEnd(&stream);
1104 if ((st != Z_STREAM_END) &&
1105 stream.total_out != sizeof(delta_head))
1106 die("delta data unpack-initial failed");
1108 /* Examine the initial part of the delta to figure out
1113 /* ignore base size */
1114 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1116 /* Read the result size */
1117 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1118 *sizep = result_size;
1123 static unsigned long unpack_object_header(struct packed_git *p,
1124 struct pack_window **w_curs,
1125 unsigned long offset,
1126 enum object_type *type,
1127 unsigned long *sizep)
1129 unsigned char *base;
1133 /* use_pack() assures us we have [base, base + 20) available
1134 * as a range that we can look at at. (Its actually the hash
1135 * size that is assurred.) With our object header encoding
1136 * the maximum deflated object size is 2^137, which is just
1137 * insane, so we know won't exceed what we have been given.
1139 base = use_pack(p, w_curs, offset, &left);
1140 used = unpack_object_header_gently(base, left, type, sizep);
1142 die("object offset outside of pack file");
1144 return offset + used;
1147 void packed_object_info_detail(struct packed_git *p,
1148 unsigned long offset,
1150 unsigned long *size,
1151 unsigned long *store_size,
1152 unsigned int *delta_chain_length,
1153 unsigned char *base_sha1)
1155 struct pack_window *w_curs = NULL;
1156 unsigned long obj_offset, val;
1157 unsigned char *next_sha1;
1158 enum object_type kind;
1160 *delta_chain_length = 0;
1161 obj_offset = offset;
1162 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1167 die("pack %s contains unknown object type %d",
1168 p->pack_name, kind);
1173 strcpy(type, type_names[kind]);
1174 *store_size = 0; /* notyet */
1175 unuse_pack(&w_curs);
1178 get_delta_base(p, &w_curs, offset, kind,
1179 obj_offset, &offset);
1180 if (*delta_chain_length == 0) {
1181 /* TODO: find base_sha1 as pointed by offset */
1185 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1186 if (*delta_chain_length == 0)
1187 hashcpy(base_sha1, next_sha1);
1188 offset = find_pack_entry_one(next_sha1, p);
1191 obj_offset = offset;
1192 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1193 (*delta_chain_length)++;
1197 static int packed_object_info(struct packed_git *p, unsigned long offset,
1198 char *type, unsigned long *sizep)
1200 struct pack_window *w_curs = NULL;
1201 unsigned long size, obj_offset = offset;
1202 enum object_type kind;
1205 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1210 r = packed_delta_info(p, &w_curs, offset, kind,
1211 obj_offset, type, sizep);
1212 unuse_pack(&w_curs);
1218 strcpy(type, type_names[kind]);
1219 unuse_pack(&w_curs);
1222 die("pack %s contains unknown object type %d",
1223 p->pack_name, kind);
1230 static void *unpack_compressed_entry(struct packed_git *p,
1231 struct pack_window **w_curs,
1232 unsigned long offset,
1237 unsigned char *buffer, *in;
1239 buffer = xmalloc(size + 1);
1241 memset(&stream, 0, sizeof(stream));
1242 stream.next_out = buffer;
1243 stream.avail_out = size;
1245 inflateInit(&stream);
1247 in = use_pack(p, w_curs, offset, &stream.avail_in);
1248 stream.next_in = in;
1249 st = inflate(&stream, Z_FINISH);
1250 offset += stream.next_in - in;
1251 } while (st == Z_OK || st == Z_BUF_ERROR);
1252 inflateEnd(&stream);
1253 if ((st != Z_STREAM_END) || stream.total_out != size) {
1261 static void *unpack_delta_entry(struct packed_git *p,
1262 struct pack_window **w_curs,
1263 unsigned long offset,
1264 unsigned long delta_size,
1265 enum object_type kind,
1266 unsigned long obj_offset,
1268 unsigned long *sizep)
1270 void *delta_data, *result, *base;
1271 unsigned long result_size, base_size, base_offset;
1273 offset = get_delta_base(p, w_curs, offset, kind,
1274 obj_offset, &base_offset);
1275 base = unpack_entry(p, base_offset, type, &base_size);
1277 die("failed to read delta base object at %lu from %s",
1278 base_offset, p->pack_name);
1280 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1281 result = patch_delta(base, base_size,
1282 delta_data, delta_size,
1285 die("failed to apply delta");
1288 *sizep = result_size;
1292 void *unpack_entry(struct packed_git *p, unsigned long offset,
1293 char *type, unsigned long *sizep)
1295 struct pack_window *w_curs = NULL;
1296 unsigned long size, obj_offset = offset;
1297 enum object_type kind;
1300 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1304 retval = unpack_delta_entry(p, &w_curs, offset, size,
1305 kind, obj_offset, type, sizep);
1311 strcpy(type, type_names[kind]);
1313 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1316 die("unknown object type %i in %s", kind, p->pack_name);
1318 unuse_pack(&w_curs);
1322 int num_packed_objects(const struct packed_git *p)
1324 /* See check_packed_git_idx() */
1325 return (p->index_size - 20 - 20 - 4*256) / 24;
1328 int nth_packed_object_sha1(const struct packed_git *p, int n,
1329 unsigned char* sha1)
1331 void *index = p->index_base + 256;
1332 if (n < 0 || num_packed_objects(p) <= n)
1334 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1338 unsigned long find_pack_entry_one(const unsigned char *sha1,
1339 struct packed_git *p)
1341 unsigned int *level1_ofs = p->index_base;
1342 int hi = ntohl(level1_ofs[*sha1]);
1343 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1344 void *index = p->index_base + 256;
1347 int mi = (lo + hi) / 2;
1348 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1350 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1359 static int matches_pack_name(struct packed_git *p, const char *ig)
1361 const char *last_c, *c;
1363 if (!strcmp(p->pack_name, ig))
1366 for (c = p->pack_name, last_c = c; *c;)
1371 if (!strcmp(last_c, ig))
1377 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1379 struct packed_git *p;
1380 unsigned long offset;
1382 prepare_packed_git();
1384 for (p = packed_git; p; p = p->next) {
1385 if (ignore_packed) {
1387 for (ig = ignore_packed; *ig; ig++)
1388 if (!matches_pack_name(p, *ig))
1393 offset = find_pack_entry_one(sha1, p);
1397 hashcpy(e->sha1, sha1);
1404 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1405 struct packed_git *packs)
1407 struct packed_git *p;
1409 for (p = packs; p; p = p->next) {
1410 if (find_pack_entry_one(sha1, p))
1417 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1420 unsigned long mapsize, size;
1425 map = map_sha1_file(sha1, &mapsize);
1427 return error("unable to find %s", sha1_to_hex(sha1));
1428 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1429 status = error("unable to unpack %s header",
1431 if (parse_sha1_header(hdr, type, &size) < 0)
1432 status = error("unable to parse %s header", sha1_to_hex(sha1));
1438 inflateEnd(&stream);
1439 munmap(map, mapsize);
1443 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1445 struct pack_entry e;
1447 if (!find_pack_entry(sha1, &e, NULL)) {
1448 reprepare_packed_git();
1449 if (!find_pack_entry(sha1, &e, NULL))
1450 return sha1_loose_object_info(sha1, type, sizep);
1452 return packed_object_info(e.p, e.offset, type, sizep);
1455 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1457 struct pack_entry e;
1459 if (!find_pack_entry(sha1, &e, NULL)) {
1460 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1463 return unpack_entry(e.p, e.offset, type, size);
1466 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1468 unsigned long mapsize;
1470 struct pack_entry e;
1472 if (find_pack_entry(sha1, &e, NULL))
1473 return read_packed_sha1(sha1, type, size);
1474 map = map_sha1_file(sha1, &mapsize);
1476 buf = unpack_sha1_file(map, mapsize, type, size);
1477 munmap(map, mapsize);
1480 reprepare_packed_git();
1481 if (find_pack_entry(sha1, &e, NULL))
1482 return read_packed_sha1(sha1, type, size);
1486 void *read_object_with_reference(const unsigned char *sha1,
1487 const char *required_type,
1488 unsigned long *size,
1489 unsigned char *actual_sha1_return)
1493 unsigned long isize;
1494 unsigned char actual_sha1[20];
1496 hashcpy(actual_sha1, sha1);
1498 int ref_length = -1;
1499 const char *ref_type = NULL;
1501 buffer = read_sha1_file(actual_sha1, type, &isize);
1504 if (!strcmp(type, required_type)) {
1506 if (actual_sha1_return)
1507 hashcpy(actual_sha1_return, actual_sha1);
1510 /* Handle references */
1511 else if (!strcmp(type, commit_type))
1513 else if (!strcmp(type, tag_type))
1514 ref_type = "object ";
1519 ref_length = strlen(ref_type);
1521 if (memcmp(buffer, ref_type, ref_length) ||
1522 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1527 /* Now we have the ID of the referred-to object in
1528 * actual_sha1. Check again. */
1532 static void write_sha1_file_prepare(void *buf, unsigned long len,
1533 const char *type, unsigned char *sha1,
1534 unsigned char *hdr, int *hdrlen)
1538 /* Generate the header */
1539 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1543 SHA1_Update(&c, hdr, *hdrlen);
1544 SHA1_Update(&c, buf, len);
1545 SHA1_Final(sha1, &c);
1549 * Link the tempfile to the final place, possibly creating the
1550 * last directory level as you do so.
1552 * Returns the errno on failure, 0 on success.
1554 static int link_temp_to_file(const char *tmpfile, const char *filename)
1559 if (!link(tmpfile, filename))
1563 * Try to mkdir the last path component if that failed.
1565 * Re-try the "link()" regardless of whether the mkdir
1566 * succeeds, since a race might mean that somebody
1570 dir = strrchr(filename, '/');
1573 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1578 if (!link(tmpfile, filename))
1586 * Move the just written object into its final resting place
1588 int move_temp_to_file(const char *tmpfile, const char *filename)
1590 int ret = link_temp_to_file(tmpfile, filename);
1593 * Coda hack - coda doesn't like cross-directory links,
1594 * so we fall back to a rename, which will mean that it
1595 * won't be able to check collisions, but that's not a
1598 * The same holds for FAT formatted media.
1600 * When this succeeds, we just return 0. We have nothing
1603 if (ret && ret != EEXIST) {
1604 if (!rename(tmpfile, filename))
1610 if (ret != EEXIST) {
1611 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1613 /* FIXME!!! Collision check here ? */
1619 static int write_buffer(int fd, const void *buf, size_t len)
1625 size = write_in_full(fd, buf, len);
1627 return error("file write: disk full");
1629 return error("file write error (%s)", strerror(errno));
1633 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1638 c = (type << 4) | (len & 15);
1651 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1655 if (use_legacy_headers) {
1656 while (deflate(stream, 0) == Z_OK)
1660 if (!strcmp(type, blob_type))
1661 obj_type = OBJ_BLOB;
1662 else if (!strcmp(type, tree_type))
1663 obj_type = OBJ_TREE;
1664 else if (!strcmp(type, commit_type))
1665 obj_type = OBJ_COMMIT;
1666 else if (!strcmp(type, tag_type))
1669 die("trying to generate bogus object of type '%s'", type);
1670 hdr = write_binary_header(stream->next_out, obj_type, len);
1671 stream->total_out = hdr;
1672 stream->next_out += hdr;
1673 stream->avail_out -= hdr;
1676 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1677 unsigned char *sha1)
1679 unsigned char hdr[50];
1681 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1685 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1688 unsigned char *compressed;
1690 unsigned char sha1[20];
1692 static char tmpfile[PATH_MAX];
1693 unsigned char hdr[50];
1696 /* Normally if we have it in the pack then we do not bother writing
1697 * it out into .git/objects/??/?{38} file.
1699 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1700 filename = sha1_file_name(sha1);
1702 hashcpy(returnsha1, sha1);
1703 if (has_sha1_file(sha1))
1705 fd = open(filename, O_RDONLY);
1708 * FIXME!!! We might do collision checking here, but we'd
1709 * need to uncompress the old file and check it. Later.
1715 if (errno != ENOENT) {
1716 return error("sha1 file %s: %s\n", filename, strerror(errno));
1719 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1721 fd = mkstemp(tmpfile);
1724 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1726 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1730 memset(&stream, 0, sizeof(stream));
1731 deflateInit(&stream, zlib_compression_level);
1732 size = 8 + deflateBound(&stream, len+hdrlen);
1733 compressed = xmalloc(size);
1736 stream.next_out = compressed;
1737 stream.avail_out = size;
1739 /* First header.. */
1740 stream.next_in = hdr;
1741 stream.avail_in = hdrlen;
1742 setup_object_header(&stream, type, len);
1744 /* Then the data itself.. */
1745 stream.next_in = buf;
1746 stream.avail_in = len;
1747 while (deflate(&stream, Z_FINISH) == Z_OK)
1749 deflateEnd(&stream);
1750 size = stream.total_out;
1752 if (write_buffer(fd, compressed, size) < 0)
1753 die("unable to write sha1 file");
1758 return move_temp_to_file(tmpfile, filename);
1762 * We need to unpack and recompress the object for writing
1763 * it out to a different file.
1765 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1769 unsigned char *unpacked;
1776 /* need to unpack and recompress it by itself */
1777 unpacked = read_packed_sha1(sha1, type, &len);
1779 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1782 memset(&stream, 0, sizeof(stream));
1783 deflateInit(&stream, zlib_compression_level);
1784 size = deflateBound(&stream, len + hdrlen);
1785 buf = xmalloc(size);
1788 stream.next_out = buf;
1789 stream.avail_out = size;
1791 /* First header.. */
1792 stream.next_in = (void *)hdr;
1793 stream.avail_in = hdrlen;
1794 while (deflate(&stream, 0) == Z_OK)
1797 /* Then the data itself.. */
1798 stream.next_in = unpacked;
1799 stream.avail_in = len;
1800 while (deflate(&stream, Z_FINISH) == Z_OK)
1802 deflateEnd(&stream);
1805 *objsize = stream.total_out;
1809 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1812 unsigned long objsize;
1813 void *buf = map_sha1_file(sha1, &objsize);
1816 retval = write_buffer(fd, buf, objsize);
1817 munmap(buf, objsize);
1821 buf = repack_object(sha1, &objsize);
1822 retval = write_buffer(fd, buf, objsize);
1827 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1828 size_t bufsize, size_t *bufposn)
1830 char tmpfile[PATH_MAX];
1833 unsigned char real_sha1[20];
1834 unsigned char discard[4096];
1838 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1840 local = mkstemp(tmpfile);
1843 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1845 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1848 memset(&stream, 0, sizeof(stream));
1850 inflateInit(&stream);
1857 stream.avail_in = *bufposn;
1858 stream.next_in = (unsigned char *) buffer;
1860 stream.next_out = discard;
1861 stream.avail_out = sizeof(discard);
1862 ret = inflate(&stream, Z_SYNC_FLUSH);
1863 SHA1_Update(&c, discard, sizeof(discard) -
1865 } while (stream.avail_in && ret == Z_OK);
1866 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1867 die("unable to write sha1 file");
1868 memmove(buffer, buffer + *bufposn - stream.avail_in,
1870 *bufposn = stream.avail_in;
1874 size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
1879 return error("Connection closed?");
1880 perror("Reading from connection");
1885 inflateEnd(&stream);
1888 SHA1_Final(real_sha1, &c);
1889 if (ret != Z_STREAM_END) {
1891 return error("File %s corrupted", sha1_to_hex(sha1));
1893 if (hashcmp(sha1, real_sha1)) {
1895 return error("File %s has bad hash", sha1_to_hex(sha1));
1898 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1901 int has_pack_index(const unsigned char *sha1)
1904 if (stat(sha1_pack_index_name(sha1), &st))
1909 int has_pack_file(const unsigned char *sha1)
1912 if (stat(sha1_pack_name(sha1), &st))
1917 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1919 struct pack_entry e;
1920 return find_pack_entry(sha1, &e, ignore_packed);
1923 int has_sha1_file(const unsigned char *sha1)
1926 struct pack_entry e;
1928 if (find_pack_entry(sha1, &e, NULL))
1930 return find_sha1_file(sha1, &st) ? 1 : 0;
1934 * reads from fd as long as possible into a supplied buffer of size bytes.
1935 * If necessary the buffer's size is increased using realloc()
1937 * returns 0 if anything went fine and -1 otherwise
1939 * NOTE: both buf and size may change, but even when -1 is returned
1940 * you still have to free() it yourself.
1942 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1944 char* buf = *return_buf;
1945 unsigned long size = *return_size;
1947 unsigned long off = 0;
1950 iret = xread(fd, buf + off, size - off);
1955 buf = xrealloc(buf, size);
1968 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1970 unsigned long size = 4096;
1971 char *buf = xmalloc(size);
1974 if (read_pipe(fd, &buf, &size)) {
1982 ret = write_sha1_file(buf, size, type, sha1);
1984 ret = hash_sha1_file(buf, size, type, sha1);
1989 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1991 unsigned long size = st->st_size;
1997 buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2003 ret = write_sha1_file(buf, size, type, sha1);
2005 ret = hash_sha1_file(buf, size, type, sha1);
2011 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2016 switch (st->st_mode & S_IFMT) {
2018 fd = open(path, O_RDONLY);
2020 return error("open(\"%s\"): %s", path,
2022 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
2023 return error("%s: failed to insert into database",
2027 target = xmalloc(st->st_size+1);
2028 if (readlink(path, target, st->st_size+1) != st->st_size) {
2029 char *errstr = strerror(errno);
2031 return error("readlink(\"%s\"): %s", path,
2035 hash_sha1_file(target, st->st_size, blob_type, sha1);
2036 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
2037 return error("%s: failed to insert into database",
2042 return error("%s: unsupported file type", path);